2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
66 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
69 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
72 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
73 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
75 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
76 const char *reference
,
77 QDict
*options
, int flags
,
78 BlockDriverState
*parent
,
79 const BdrvChildClass
*child_class
,
80 BdrvChildRole child_role
,
83 /* TODO: Remove when no longer needed */
84 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
85 int *child_flags
, QDict
*child_options
,
86 int parent_flags
, QDict
*parent_options
);
87 static void bdrv_child_cb_attach(BdrvChild
*child
);
88 static void bdrv_child_cb_detach(BdrvChild
*child
);
90 /* If non-zero, use only whitelisted block drivers */
91 static int use_bdrv_whitelist
;
94 static int is_windows_drive_prefix(const char *filename
)
96 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
97 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
101 int is_windows_drive(const char *filename
)
103 if (is_windows_drive_prefix(filename
) &&
106 if (strstart(filename
, "\\\\.\\", NULL
) ||
107 strstart(filename
, "//./", NULL
))
113 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
115 if (!bs
|| !bs
->drv
) {
116 /* page size or 4k (hdd sector size) should be on the safe side */
117 return MAX(4096, qemu_real_host_page_size
);
120 return bs
->bl
.opt_mem_alignment
;
123 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
125 if (!bs
|| !bs
->drv
) {
126 /* page size or 4k (hdd sector size) should be on the safe side */
127 return MAX(4096, qemu_real_host_page_size
);
130 return bs
->bl
.min_mem_alignment
;
133 /* check if the path starts with "<protocol>:" */
134 int path_has_protocol(const char *path
)
139 if (is_windows_drive(path
) ||
140 is_windows_drive_prefix(path
)) {
143 p
= path
+ strcspn(path
, ":/\\");
145 p
= path
+ strcspn(path
, ":/");
151 int path_is_absolute(const char *path
)
154 /* specific case for names like: "\\.\d:" */
155 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
158 return (*path
== '/' || *path
== '\\');
160 return (*path
== '/');
164 /* if filename is absolute, just return its duplicate. Otherwise, build a
165 path to it by considering it is relative to base_path. URL are
167 char *path_combine(const char *base_path
, const char *filename
)
169 const char *protocol_stripped
= NULL
;
174 if (path_is_absolute(filename
)) {
175 return g_strdup(filename
);
178 if (path_has_protocol(base_path
)) {
179 protocol_stripped
= strchr(base_path
, ':');
180 if (protocol_stripped
) {
184 p
= protocol_stripped
?: base_path
;
186 p1
= strrchr(base_path
, '/');
190 p2
= strrchr(base_path
, '\\');
191 if (!p1
|| p2
> p1
) {
206 result
= g_malloc(len
+ strlen(filename
) + 1);
207 memcpy(result
, base_path
, len
);
208 strcpy(result
+ len
, filename
);
214 * Helper function for bdrv_parse_filename() implementations to remove optional
215 * protocol prefixes (especially "file:") from a filename and for putting the
216 * stripped filename into the options QDict if there is such a prefix.
218 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
221 if (strstart(filename
, prefix
, &filename
)) {
222 /* Stripping the explicit protocol prefix may result in a protocol
223 * prefix being (wrongly) detected (if the filename contains a colon) */
224 if (path_has_protocol(filename
)) {
225 QString
*fat_filename
;
227 /* This means there is some colon before the first slash; therefore,
228 * this cannot be an absolute path */
229 assert(!path_is_absolute(filename
));
231 /* And we can thus fix the protocol detection issue by prefixing it
233 fat_filename
= qstring_from_str("./");
234 qstring_append(fat_filename
, filename
);
236 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
238 qdict_put(options
, "filename", fat_filename
);
240 /* If no protocol prefix was detected, we can use the shortened
242 qdict_put_str(options
, "filename", filename
);
248 /* Returns whether the image file is opened as read-only. Note that this can
249 * return false and writing to the image file is still not possible because the
250 * image is inactivated. */
251 bool bdrv_is_read_only(BlockDriverState
*bs
)
253 return bs
->read_only
;
256 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
257 bool ignore_allow_rdw
, Error
**errp
)
259 /* Do not set read_only if copy_on_read is enabled */
260 if (bs
->copy_on_read
&& read_only
) {
261 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
262 bdrv_get_device_or_node_name(bs
));
266 /* Do not clear read_only if it is prohibited */
267 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
270 error_setg(errp
, "Node '%s' is read only",
271 bdrv_get_device_or_node_name(bs
));
279 * Called by a driver that can only provide a read-only image.
281 * Returns 0 if the node is already read-only or it could switch the node to
282 * read-only because BDRV_O_AUTO_RDONLY is set.
284 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
285 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
286 * is not NULL, it is used as the error message for the Error object.
288 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
293 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
296 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
300 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
305 bs
->read_only
= true;
306 bs
->open_flags
&= ~BDRV_O_RDWR
;
311 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
316 * If @backing is empty, this function returns NULL without setting
317 * @errp. In all other cases, NULL will only be returned with @errp
320 * Therefore, a return value of NULL without @errp set means that
321 * there is no backing file; if @errp is set, there is one but its
322 * absolute filename cannot be generated.
324 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
328 if (backing
[0] == '\0') {
330 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
331 return g_strdup(backing
);
332 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
333 error_setg(errp
, "Cannot use relative backing file names for '%s'",
337 return path_combine(backed
, backing
);
342 * If @filename is empty or NULL, this function returns NULL without
343 * setting @errp. In all other cases, NULL will only be returned with
346 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
347 const char *filename
, Error
**errp
)
349 char *dir
, *full_name
;
351 if (!filename
|| filename
[0] == '\0') {
353 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
354 return g_strdup(filename
);
357 dir
= bdrv_dirname(relative_to
, errp
);
362 full_name
= g_strconcat(dir
, filename
, NULL
);
367 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
369 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
372 void bdrv_register(BlockDriver
*bdrv
)
374 assert(bdrv
->format_name
);
375 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
378 BlockDriverState
*bdrv_new(void)
380 BlockDriverState
*bs
;
383 bs
= g_new0(BlockDriverState
, 1);
384 QLIST_INIT(&bs
->dirty_bitmaps
);
385 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
386 QLIST_INIT(&bs
->op_blockers
[i
]);
388 notifier_with_return_list_init(&bs
->before_write_notifiers
);
389 qemu_co_mutex_init(&bs
->reqs_lock
);
390 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
392 bs
->aio_context
= qemu_get_aio_context();
394 qemu_co_queue_init(&bs
->flush_queue
);
396 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
397 bdrv_drained_begin(bs
);
400 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
405 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
409 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
410 if (!strcmp(drv1
->format_name
, format_name
)) {
418 BlockDriver
*bdrv_find_format(const char *format_name
)
423 drv1
= bdrv_do_find_format(format_name
);
428 /* The driver isn't registered, maybe we need to load a module */
429 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
430 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
431 block_module_load_one(block_driver_modules
[i
].library_name
);
436 return bdrv_do_find_format(format_name
);
439 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
441 static const char *whitelist_rw
[] = {
442 CONFIG_BDRV_RW_WHITELIST
444 static const char *whitelist_ro
[] = {
445 CONFIG_BDRV_RO_WHITELIST
449 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
450 return 1; /* no whitelist, anything goes */
453 for (p
= whitelist_rw
; *p
; p
++) {
454 if (!strcmp(format_name
, *p
)) {
459 for (p
= whitelist_ro
; *p
; p
++) {
460 if (!strcmp(format_name
, *p
)) {
468 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
470 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
473 bool bdrv_uses_whitelist(void)
475 return use_bdrv_whitelist
;
478 typedef struct CreateCo
{
486 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
488 Error
*local_err
= NULL
;
491 CreateCo
*cco
= opaque
;
494 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
495 cco
->filename
, cco
->opts
, &local_err
);
496 error_propagate(&cco
->err
, local_err
);
500 int bdrv_create(BlockDriver
*drv
, const char* filename
,
501 QemuOpts
*opts
, Error
**errp
)
508 .filename
= g_strdup(filename
),
514 if (!drv
->bdrv_co_create_opts
) {
515 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
520 if (qemu_in_coroutine()) {
521 /* Fast-path if already in coroutine context */
522 bdrv_create_co_entry(&cco
);
524 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
525 qemu_coroutine_enter(co
);
526 while (cco
.ret
== NOT_DONE
) {
527 aio_poll(qemu_get_aio_context(), true);
534 error_propagate(errp
, cco
.err
);
536 error_setg_errno(errp
, -ret
, "Could not create image");
541 g_free(cco
.filename
);
546 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
547 * least the given @minimum_size.
549 * On success, return @blk's actual length.
550 * Otherwise, return -errno.
552 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
553 int64_t minimum_size
, Error
**errp
)
555 Error
*local_err
= NULL
;
559 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
561 if (ret
< 0 && ret
!= -ENOTSUP
) {
562 error_propagate(errp
, local_err
);
566 size
= blk_getlength(blk
);
568 error_free(local_err
);
569 error_setg_errno(errp
, -size
,
570 "Failed to inquire the new image file's length");
574 if (size
< minimum_size
) {
575 /* Need to grow the image, but we failed to do that */
576 error_propagate(errp
, local_err
);
580 error_free(local_err
);
587 * Helper function for bdrv_create_file_fallback(): Zero the first
588 * sector to remove any potentially pre-existing image header.
590 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
591 int64_t current_size
,
594 int64_t bytes_to_clear
;
597 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
598 if (bytes_to_clear
) {
599 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
601 error_setg_errno(errp
, -ret
,
602 "Failed to clear the new image's first sector");
611 * Simple implementation of bdrv_co_create_opts for protocol drivers
612 * which only support creation via opening a file
613 * (usually existing raw storage device)
615 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
616 const char *filename
,
624 PreallocMode prealloc
;
625 Error
*local_err
= NULL
;
628 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
629 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
630 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
631 PREALLOC_MODE_OFF
, &local_err
);
634 error_propagate(errp
, local_err
);
638 if (prealloc
!= PREALLOC_MODE_OFF
) {
639 error_setg(errp
, "Unsupported preallocation mode '%s'",
640 PreallocMode_str(prealloc
));
644 options
= qdict_new();
645 qdict_put_str(options
, "driver", drv
->format_name
);
647 blk
= blk_new_open(filename
, NULL
, options
,
648 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
650 error_prepend(errp
, "Protocol driver '%s' does not support image "
651 "creation, and opening the image failed: ",
656 size
= create_file_fallback_truncate(blk
, size
, errp
);
662 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
673 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
677 drv
= bdrv_find_protocol(filename
, true, errp
);
682 return bdrv_create(drv
, filename
, opts
, errp
);
685 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
687 Error
*local_err
= NULL
;
693 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
697 if (!bs
->drv
->bdrv_co_delete_file
) {
698 error_setg(errp
, "Driver '%s' does not support image deletion",
699 bs
->drv
->format_name
);
703 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
705 error_propagate(errp
, local_err
);
712 * Try to get @bs's logical and physical block size.
713 * On success, store them in @bsz struct and return 0.
714 * On failure return -errno.
715 * @bs must not be empty.
717 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
719 BlockDriver
*drv
= bs
->drv
;
721 if (drv
&& drv
->bdrv_probe_blocksizes
) {
722 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
723 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
724 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
731 * Try to get @bs's geometry (cyls, heads, sectors).
732 * On success, store them in @geo struct and return 0.
733 * On failure return -errno.
734 * @bs must not be empty.
736 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
738 BlockDriver
*drv
= bs
->drv
;
740 if (drv
&& drv
->bdrv_probe_geometry
) {
741 return drv
->bdrv_probe_geometry(bs
, geo
);
742 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
743 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
750 * Create a uniquely-named empty temporary file.
751 * Return 0 upon success, otherwise a negative errno value.
753 int get_tmp_filename(char *filename
, int size
)
756 char temp_dir
[MAX_PATH
];
757 /* GetTempFileName requires that its output buffer (4th param)
758 have length MAX_PATH or greater. */
759 assert(size
>= MAX_PATH
);
760 return (GetTempPath(MAX_PATH
, temp_dir
)
761 && GetTempFileName(temp_dir
, "qem", 0, filename
)
762 ? 0 : -GetLastError());
766 tmpdir
= getenv("TMPDIR");
770 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
773 fd
= mkstemp(filename
);
777 if (close(fd
) != 0) {
786 * Detect host devices. By convention, /dev/cdrom[N] is always
787 * recognized as a host CDROM.
789 static BlockDriver
*find_hdev_driver(const char *filename
)
791 int score_max
= 0, score
;
792 BlockDriver
*drv
= NULL
, *d
;
794 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
795 if (d
->bdrv_probe_device
) {
796 score
= d
->bdrv_probe_device(filename
);
797 if (score
> score_max
) {
807 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
811 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
812 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
820 BlockDriver
*bdrv_find_protocol(const char *filename
,
821 bool allow_protocol_prefix
,
830 /* TODO Drivers without bdrv_file_open must be specified explicitly */
833 * XXX(hch): we really should not let host device detection
834 * override an explicit protocol specification, but moving this
835 * later breaks access to device names with colons in them.
836 * Thanks to the brain-dead persistent naming schemes on udev-
837 * based Linux systems those actually are quite common.
839 drv1
= find_hdev_driver(filename
);
844 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
848 p
= strchr(filename
, ':');
851 if (len
> sizeof(protocol
) - 1)
852 len
= sizeof(protocol
) - 1;
853 memcpy(protocol
, filename
, len
);
854 protocol
[len
] = '\0';
856 drv1
= bdrv_do_find_protocol(protocol
);
861 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
862 if (block_driver_modules
[i
].protocol_name
&&
863 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
864 block_module_load_one(block_driver_modules
[i
].library_name
);
869 drv1
= bdrv_do_find_protocol(protocol
);
871 error_setg(errp
, "Unknown protocol '%s'", protocol
);
877 * Guess image format by probing its contents.
878 * This is not a good idea when your image is raw (CVE-2008-2004), but
879 * we do it anyway for backward compatibility.
881 * @buf contains the image's first @buf_size bytes.
882 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
883 * but can be smaller if the image file is smaller)
884 * @filename is its filename.
886 * For all block drivers, call the bdrv_probe() method to get its
888 * Return the first block driver with the highest probing score.
890 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
891 const char *filename
)
893 int score_max
= 0, score
;
894 BlockDriver
*drv
= NULL
, *d
;
896 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
898 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
899 if (score
> score_max
) {
909 static int find_image_format(BlockBackend
*file
, const char *filename
,
910 BlockDriver
**pdrv
, Error
**errp
)
913 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
916 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
917 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
922 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
924 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
930 drv
= bdrv_probe_all(buf
, ret
, filename
);
932 error_setg(errp
, "Could not determine image format: No compatible "
941 * Set the current 'total_sectors' value
942 * Return 0 on success, -errno on error.
944 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
946 BlockDriver
*drv
= bs
->drv
;
952 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
956 /* query actual device if possible, otherwise just trust the hint */
957 if (drv
->bdrv_getlength
) {
958 int64_t length
= drv
->bdrv_getlength(bs
);
962 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
965 bs
->total_sectors
= hint
;
970 * Combines a QDict of new block driver @options with any missing options taken
971 * from @old_options, so that leaving out an option defaults to its old value.
973 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
976 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
977 bs
->drv
->bdrv_join_options(options
, old_options
);
979 qdict_join(options
, old_options
, false);
983 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
987 Error
*local_err
= NULL
;
988 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
989 BlockdevDetectZeroesOptions detect_zeroes
=
990 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
991 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
994 error_propagate(errp
, local_err
);
995 return detect_zeroes
;
998 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
999 !(open_flags
& BDRV_O_UNMAP
))
1001 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1002 "without setting discard operation to unmap");
1005 return detect_zeroes
;
1009 * Set open flags for aio engine
1011 * Return 0 on success, -1 if the engine specified is invalid
1013 int bdrv_parse_aio(const char *mode
, int *flags
)
1015 if (!strcmp(mode
, "threads")) {
1016 /* do nothing, default */
1017 } else if (!strcmp(mode
, "native")) {
1018 *flags
|= BDRV_O_NATIVE_AIO
;
1019 #ifdef CONFIG_LINUX_IO_URING
1020 } else if (!strcmp(mode
, "io_uring")) {
1021 *flags
|= BDRV_O_IO_URING
;
1031 * Set open flags for a given discard mode
1033 * Return 0 on success, -1 if the discard mode was invalid.
1035 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1037 *flags
&= ~BDRV_O_UNMAP
;
1039 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1041 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1042 *flags
|= BDRV_O_UNMAP
;
1051 * Set open flags for a given cache mode
1053 * Return 0 on success, -1 if the cache mode was invalid.
1055 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1057 *flags
&= ~BDRV_O_CACHE_MASK
;
1059 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1060 *writethrough
= false;
1061 *flags
|= BDRV_O_NOCACHE
;
1062 } else if (!strcmp(mode
, "directsync")) {
1063 *writethrough
= true;
1064 *flags
|= BDRV_O_NOCACHE
;
1065 } else if (!strcmp(mode
, "writeback")) {
1066 *writethrough
= false;
1067 } else if (!strcmp(mode
, "unsafe")) {
1068 *writethrough
= false;
1069 *flags
|= BDRV_O_NO_FLUSH
;
1070 } else if (!strcmp(mode
, "writethrough")) {
1071 *writethrough
= true;
1079 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1081 BlockDriverState
*parent
= c
->opaque
;
1082 return g_strdup(bdrv_get_device_or_node_name(parent
));
1085 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1087 BlockDriverState
*bs
= child
->opaque
;
1088 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1091 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1093 BlockDriverState
*bs
= child
->opaque
;
1094 return bdrv_drain_poll(bs
, false, NULL
, false);
1097 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1098 int *drained_end_counter
)
1100 BlockDriverState
*bs
= child
->opaque
;
1101 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1104 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1106 BlockDriverState
*bs
= child
->opaque
;
1107 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1111 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1112 GSList
**ignore
, Error
**errp
)
1114 BlockDriverState
*bs
= child
->opaque
;
1115 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1118 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1121 BlockDriverState
*bs
= child
->opaque
;
1122 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1126 * Returns the options and flags that a temporary snapshot should get, based on
1127 * the originally requested flags (the originally requested image will have
1128 * flags like a backing file)
1130 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1131 int parent_flags
, QDict
*parent_options
)
1133 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1135 /* For temporary files, unconditional cache=unsafe is fine */
1136 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1137 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1139 /* Copy the read-only and discard options from the parent */
1140 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1141 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1143 /* aio=native doesn't work for cache.direct=off, so disable it for the
1144 * temporary snapshot */
1145 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1149 * Returns the options and flags that bs->file should get if a protocol driver
1150 * is expected, based on the given options and flags for the parent BDS
1152 static void bdrv_protocol_options(BdrvChildRole role
, bool parent_is_format
,
1153 int *child_flags
, QDict
*child_options
,
1154 int parent_flags
, QDict
*parent_options
)
1156 bdrv_inherited_options(BDRV_CHILD_IMAGE
, true,
1157 child_flags
, child_options
,
1158 parent_flags
, parent_options
);
1161 const BdrvChildClass child_file
= {
1162 .parent_is_bds
= true,
1163 .get_parent_desc
= bdrv_child_get_parent_desc
,
1164 .inherit_options
= bdrv_protocol_options
,
1165 .drained_begin
= bdrv_child_cb_drained_begin
,
1166 .drained_poll
= bdrv_child_cb_drained_poll
,
1167 .drained_end
= bdrv_child_cb_drained_end
,
1168 .attach
= bdrv_child_cb_attach
,
1169 .detach
= bdrv_child_cb_detach
,
1170 .inactivate
= bdrv_child_cb_inactivate
,
1171 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1172 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1176 * Returns the options and flags that bs->file should get if the use of formats
1177 * (and not only protocols) is permitted for it, based on the given options and
1178 * flags for the parent BDS
1180 static void bdrv_inherited_fmt_options(BdrvChildRole role
,
1181 bool parent_is_format
,
1182 int *child_flags
, QDict
*child_options
,
1183 int parent_flags
, QDict
*parent_options
)
1185 bdrv_inherited_options(BDRV_CHILD_DATA
, false,
1186 child_flags
, child_options
,
1187 parent_flags
, parent_options
);
1190 const BdrvChildClass child_format
= {
1191 .parent_is_bds
= true,
1192 .get_parent_desc
= bdrv_child_get_parent_desc
,
1193 .inherit_options
= bdrv_inherited_fmt_options
,
1194 .drained_begin
= bdrv_child_cb_drained_begin
,
1195 .drained_poll
= bdrv_child_cb_drained_poll
,
1196 .drained_end
= bdrv_child_cb_drained_end
,
1197 .attach
= bdrv_child_cb_attach
,
1198 .detach
= bdrv_child_cb_detach
,
1199 .inactivate
= bdrv_child_cb_inactivate
,
1200 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1201 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1204 static void bdrv_backing_attach(BdrvChild
*c
)
1206 BlockDriverState
*parent
= c
->opaque
;
1207 BlockDriverState
*backing_hd
= c
->bs
;
1209 assert(!parent
->backing_blocker
);
1210 error_setg(&parent
->backing_blocker
,
1211 "node is used as backing hd of '%s'",
1212 bdrv_get_device_or_node_name(parent
));
1214 bdrv_refresh_filename(backing_hd
);
1216 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1217 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1218 backing_hd
->filename
);
1219 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1220 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1222 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1223 /* Otherwise we won't be able to commit or stream */
1224 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1225 parent
->backing_blocker
);
1226 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1227 parent
->backing_blocker
);
1229 * We do backup in 3 ways:
1231 * The target bs is new opened, and the source is top BDS
1232 * 2. blockdev backup
1233 * Both the source and the target are top BDSes.
1234 * 3. internal backup(used for block replication)
1235 * Both the source and the target are backing file
1237 * In case 1 and 2, neither the source nor the target is the backing file.
1238 * In case 3, we will block the top BDS, so there is only one block job
1239 * for the top BDS and its backing chain.
1241 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1242 parent
->backing_blocker
);
1243 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1244 parent
->backing_blocker
);
1247 /* XXX: Will be removed along with child_backing */
1248 static void bdrv_child_cb_attach_backing(BdrvChild
*c
)
1250 if (!(c
->role
& BDRV_CHILD_COW
)) {
1251 bdrv_backing_attach(c
);
1253 bdrv_child_cb_attach(c
);
1256 static void bdrv_backing_detach(BdrvChild
*c
)
1258 BlockDriverState
*parent
= c
->opaque
;
1260 assert(parent
->backing_blocker
);
1261 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1262 error_free(parent
->backing_blocker
);
1263 parent
->backing_blocker
= NULL
;
1266 /* XXX: Will be removed along with child_backing */
1267 static void bdrv_child_cb_detach_backing(BdrvChild
*c
)
1269 if (!(c
->role
& BDRV_CHILD_COW
)) {
1270 bdrv_backing_detach(c
);
1272 bdrv_child_cb_detach(c
);
1276 * Returns the options and flags that bs->backing should get, based on the
1277 * given options and flags for the parent BDS
1279 static void bdrv_backing_options(BdrvChildRole role
, bool parent_is_format
,
1280 int *child_flags
, QDict
*child_options
,
1281 int parent_flags
, QDict
*parent_options
)
1283 bdrv_inherited_options(BDRV_CHILD_COW
, true,
1284 child_flags
, child_options
,
1285 parent_flags
, parent_options
);
1288 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1289 const char *filename
, Error
**errp
)
1291 BlockDriverState
*parent
= c
->opaque
;
1292 bool read_only
= bdrv_is_read_only(parent
);
1296 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1302 ret
= bdrv_change_backing_file(parent
, filename
,
1303 base
->drv
? base
->drv
->format_name
: "");
1305 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1309 bdrv_reopen_set_read_only(parent
, true, NULL
);
1315 const BdrvChildClass child_backing
= {
1316 .parent_is_bds
= true,
1317 .get_parent_desc
= bdrv_child_get_parent_desc
,
1318 .attach
= bdrv_child_cb_attach_backing
,
1319 .detach
= bdrv_child_cb_detach_backing
,
1320 .inherit_options
= bdrv_backing_options
,
1321 .drained_begin
= bdrv_child_cb_drained_begin
,
1322 .drained_poll
= bdrv_child_cb_drained_poll
,
1323 .drained_end
= bdrv_child_cb_drained_end
,
1324 .inactivate
= bdrv_child_cb_inactivate
,
1325 .update_filename
= bdrv_backing_update_filename
,
1326 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1327 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1331 * Returns the options and flags that a generic child of a BDS should
1332 * get, based on the given options and flags for the parent BDS.
1334 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1335 int *child_flags
, QDict
*child_options
,
1336 int parent_flags
, QDict
*parent_options
)
1338 int flags
= parent_flags
;
1341 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1342 * Generally, the question to answer is: Should this child be
1343 * format-probed by default?
1347 * Pure and non-filtered data children of non-format nodes should
1348 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1349 * set). This only affects a very limited set of drivers (namely
1350 * quorum and blkverify when this comment was written).
1351 * Force-clear BDRV_O_PROTOCOL then.
1353 if (!parent_is_format
&&
1354 (role
& BDRV_CHILD_DATA
) &&
1355 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1357 flags
&= ~BDRV_O_PROTOCOL
;
1361 * All children of format nodes (except for COW children) and all
1362 * metadata children in general should never be format-probed.
1363 * Force-set BDRV_O_PROTOCOL then.
1365 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1366 (role
& BDRV_CHILD_METADATA
))
1368 flags
|= BDRV_O_PROTOCOL
;
1372 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1375 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1376 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1377 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1379 if (role
& BDRV_CHILD_COW
) {
1380 /* backing files are opened read-only by default */
1381 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1382 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1384 /* Inherit the read-only option from the parent if it's not set */
1385 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1386 qdict_copy_default(child_options
, parent_options
,
1387 BDRV_OPT_AUTO_READ_ONLY
);
1391 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1392 * can default to enable it on lower layers regardless of the
1395 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1397 /* Clear flags that only apply to the top layer */
1398 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1400 if (role
& BDRV_CHILD_METADATA
) {
1401 flags
&= ~BDRV_O_NO_IO
;
1403 if (role
& BDRV_CHILD_COW
) {
1404 flags
&= ~BDRV_O_TEMPORARY
;
1407 *child_flags
= flags
;
1410 static void bdrv_child_cb_attach(BdrvChild
*child
)
1412 BlockDriverState
*bs
= child
->opaque
;
1414 if (child
->role
& BDRV_CHILD_COW
) {
1415 bdrv_backing_attach(child
);
1418 bdrv_apply_subtree_drain(child
, bs
);
1421 static void bdrv_child_cb_detach(BdrvChild
*child
)
1423 BlockDriverState
*bs
= child
->opaque
;
1425 if (child
->role
& BDRV_CHILD_COW
) {
1426 bdrv_backing_detach(child
);
1429 bdrv_unapply_subtree_drain(child
, bs
);
1432 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1434 int open_flags
= flags
;
1437 * Clear flags that are internal to the block layer before opening the
1440 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1445 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1447 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1449 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1450 *flags
|= BDRV_O_NO_FLUSH
;
1453 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1454 *flags
|= BDRV_O_NOCACHE
;
1457 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1458 *flags
|= BDRV_O_RDWR
;
1461 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1462 *flags
|= BDRV_O_AUTO_RDONLY
;
1466 static void update_options_from_flags(QDict
*options
, int flags
)
1468 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1469 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1471 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1472 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1473 flags
& BDRV_O_NO_FLUSH
);
1475 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1476 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1478 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1479 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1480 flags
& BDRV_O_AUTO_RDONLY
);
1484 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1485 const char *node_name
,
1488 char *gen_node_name
= NULL
;
1491 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1492 } else if (!id_wellformed(node_name
)) {
1494 * Check for empty string or invalid characters, but not if it is
1495 * generated (generated names use characters not available to the user)
1497 error_setg(errp
, "Invalid node name");
1501 /* takes care of avoiding namespaces collisions */
1502 if (blk_by_name(node_name
)) {
1503 error_setg(errp
, "node-name=%s is conflicting with a device id",
1508 /* takes care of avoiding duplicates node names */
1509 if (bdrv_find_node(node_name
)) {
1510 error_setg(errp
, "Duplicate node name");
1514 /* Make sure that the node name isn't truncated */
1515 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1516 error_setg(errp
, "Node name too long");
1520 /* copy node name into the bs and insert it into the graph list */
1521 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1522 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1524 g_free(gen_node_name
);
1527 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1528 const char *node_name
, QDict
*options
,
1529 int open_flags
, Error
**errp
)
1531 Error
*local_err
= NULL
;
1534 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1536 error_propagate(errp
, local_err
);
1541 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1542 bs
->opaque
= g_malloc0(drv
->instance_size
);
1544 if (drv
->bdrv_file_open
) {
1545 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1546 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1547 } else if (drv
->bdrv_open
) {
1548 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1555 error_propagate(errp
, local_err
);
1556 } else if (bs
->filename
[0]) {
1557 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1559 error_setg_errno(errp
, -ret
, "Could not open image");
1564 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1566 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1570 bdrv_refresh_limits(bs
, &local_err
);
1572 error_propagate(errp
, local_err
);
1576 assert(bdrv_opt_mem_align(bs
) != 0);
1577 assert(bdrv_min_mem_align(bs
) != 0);
1578 assert(is_power_of_2(bs
->bl
.request_alignment
));
1580 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1581 if (drv
->bdrv_co_drain_begin
) {
1582 drv
->bdrv_co_drain_begin(bs
);
1589 if (bs
->file
!= NULL
) {
1590 bdrv_unref_child(bs
, bs
->file
);
1598 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1599 int flags
, Error
**errp
)
1601 BlockDriverState
*bs
;
1605 bs
->open_flags
= flags
;
1606 bs
->explicit_options
= qdict_new();
1607 bs
->options
= qdict_new();
1610 update_options_from_flags(bs
->options
, flags
);
1612 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1614 qobject_unref(bs
->explicit_options
);
1615 bs
->explicit_options
= NULL
;
1616 qobject_unref(bs
->options
);
1625 QemuOptsList bdrv_runtime_opts
= {
1626 .name
= "bdrv_common",
1627 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1630 .name
= "node-name",
1631 .type
= QEMU_OPT_STRING
,
1632 .help
= "Node name of the block device node",
1636 .type
= QEMU_OPT_STRING
,
1637 .help
= "Block driver to use for the node",
1640 .name
= BDRV_OPT_CACHE_DIRECT
,
1641 .type
= QEMU_OPT_BOOL
,
1642 .help
= "Bypass software writeback cache on the host",
1645 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1646 .type
= QEMU_OPT_BOOL
,
1647 .help
= "Ignore flush requests",
1650 .name
= BDRV_OPT_READ_ONLY
,
1651 .type
= QEMU_OPT_BOOL
,
1652 .help
= "Node is opened in read-only mode",
1655 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1656 .type
= QEMU_OPT_BOOL
,
1657 .help
= "Node can become read-only if opening read-write fails",
1660 .name
= "detect-zeroes",
1661 .type
= QEMU_OPT_STRING
,
1662 .help
= "try to optimize zero writes (off, on, unmap)",
1665 .name
= BDRV_OPT_DISCARD
,
1666 .type
= QEMU_OPT_STRING
,
1667 .help
= "discard operation (ignore/off, unmap/on)",
1670 .name
= BDRV_OPT_FORCE_SHARE
,
1671 .type
= QEMU_OPT_BOOL
,
1672 .help
= "always accept other writers (default: off)",
1674 { /* end of list */ }
1678 QemuOptsList bdrv_create_opts_simple
= {
1679 .name
= "simple-create-opts",
1680 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1683 .name
= BLOCK_OPT_SIZE
,
1684 .type
= QEMU_OPT_SIZE
,
1685 .help
= "Virtual disk size"
1688 .name
= BLOCK_OPT_PREALLOC
,
1689 .type
= QEMU_OPT_STRING
,
1690 .help
= "Preallocation mode (allowed values: off)"
1692 { /* end of list */ }
1697 * Common part for opening disk images and files
1699 * Removes all processed options from *options.
1701 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1702 QDict
*options
, Error
**errp
)
1704 int ret
, open_flags
;
1705 const char *filename
;
1706 const char *driver_name
= NULL
;
1707 const char *node_name
= NULL
;
1708 const char *discard
;
1711 Error
*local_err
= NULL
;
1713 assert(bs
->file
== NULL
);
1714 assert(options
!= NULL
&& bs
->options
!= options
);
1716 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1717 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1719 error_propagate(errp
, local_err
);
1724 update_flags_from_options(&bs
->open_flags
, opts
);
1726 driver_name
= qemu_opt_get(opts
, "driver");
1727 drv
= bdrv_find_format(driver_name
);
1728 assert(drv
!= NULL
);
1730 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1732 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1734 BDRV_OPT_FORCE_SHARE
1735 "=on can only be used with read-only images");
1741 bdrv_refresh_filename(blk_bs(file
));
1742 filename
= blk_bs(file
)->filename
;
1745 * Caution: while qdict_get_try_str() is fine, getting
1746 * non-string types would require more care. When @options
1747 * come from -blockdev or blockdev_add, its members are typed
1748 * according to the QAPI schema, but when they come from
1749 * -drive, they're all QString.
1751 filename
= qdict_get_try_str(options
, "filename");
1754 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1755 error_setg(errp
, "The '%s' block driver requires a file name",
1761 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1764 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1766 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1767 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1768 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1774 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1775 ? "Driver '%s' can only be used for read-only devices"
1776 : "Driver '%s' is not whitelisted",
1782 /* bdrv_new() and bdrv_close() make it so */
1783 assert(atomic_read(&bs
->copy_on_read
) == 0);
1785 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1786 if (!bs
->read_only
) {
1787 bdrv_enable_copy_on_read(bs
);
1789 error_setg(errp
, "Can't use copy-on-read on read-only device");
1795 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1796 if (discard
!= NULL
) {
1797 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1798 error_setg(errp
, "Invalid discard option");
1805 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1807 error_propagate(errp
, local_err
);
1812 if (filename
!= NULL
) {
1813 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1815 bs
->filename
[0] = '\0';
1817 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1819 /* Open the image, either directly or using a protocol */
1820 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1821 node_name
= qemu_opt_get(opts
, "node-name");
1823 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1824 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1829 qemu_opts_del(opts
);
1833 qemu_opts_del(opts
);
1837 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1839 QObject
*options_obj
;
1843 ret
= strstart(filename
, "json:", &filename
);
1846 options_obj
= qobject_from_json(filename
, errp
);
1848 error_prepend(errp
, "Could not parse the JSON options: ");
1852 options
= qobject_to(QDict
, options_obj
);
1854 qobject_unref(options_obj
);
1855 error_setg(errp
, "Invalid JSON object given");
1859 qdict_flatten(options
);
1864 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1867 QDict
*json_options
;
1868 Error
*local_err
= NULL
;
1870 /* Parse json: pseudo-protocol */
1871 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1875 json_options
= parse_json_filename(*pfilename
, &local_err
);
1877 error_propagate(errp
, local_err
);
1881 /* Options given in the filename have lower priority than options
1882 * specified directly */
1883 qdict_join(options
, json_options
, false);
1884 qobject_unref(json_options
);
1889 * Fills in default options for opening images and converts the legacy
1890 * filename/flags pair to option QDict entries.
1891 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1892 * block driver has been specified explicitly.
1894 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1895 int *flags
, Error
**errp
)
1897 const char *drvname
;
1898 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1899 bool parse_filename
= false;
1900 BlockDriver
*drv
= NULL
;
1901 Error
*local_err
= NULL
;
1904 * Caution: while qdict_get_try_str() is fine, getting non-string
1905 * types would require more care. When @options come from
1906 * -blockdev or blockdev_add, its members are typed according to
1907 * the QAPI schema, but when they come from -drive, they're all
1910 drvname
= qdict_get_try_str(*options
, "driver");
1912 drv
= bdrv_find_format(drvname
);
1914 error_setg(errp
, "Unknown driver '%s'", drvname
);
1917 /* If the user has explicitly specified the driver, this choice should
1918 * override the BDRV_O_PROTOCOL flag */
1919 protocol
= drv
->bdrv_file_open
;
1923 *flags
|= BDRV_O_PROTOCOL
;
1925 *flags
&= ~BDRV_O_PROTOCOL
;
1928 /* Translate cache options from flags into options */
1929 update_options_from_flags(*options
, *flags
);
1931 /* Fetch the file name from the options QDict if necessary */
1932 if (protocol
&& filename
) {
1933 if (!qdict_haskey(*options
, "filename")) {
1934 qdict_put_str(*options
, "filename", filename
);
1935 parse_filename
= true;
1937 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1943 /* Find the right block driver */
1944 /* See cautionary note on accessing @options above */
1945 filename
= qdict_get_try_str(*options
, "filename");
1947 if (!drvname
&& protocol
) {
1949 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1954 drvname
= drv
->format_name
;
1955 qdict_put_str(*options
, "driver", drvname
);
1957 error_setg(errp
, "Must specify either driver or file");
1962 assert(drv
|| !protocol
);
1964 /* Driver-specific filename parsing */
1965 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1966 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1968 error_propagate(errp
, local_err
);
1972 if (!drv
->bdrv_needs_filename
) {
1973 qdict_del(*options
, "filename");
1980 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1981 uint64_t perm
, uint64_t shared
,
1982 GSList
*ignore_children
,
1983 bool *tighten_restrictions
, Error
**errp
);
1984 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1985 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1987 typedef struct BlockReopenQueueEntry
{
1990 BDRVReopenState state
;
1991 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1992 } BlockReopenQueueEntry
;
1995 * Return the flags that @bs will have after the reopens in @q have
1996 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1997 * return the current flags.
1999 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2001 BlockReopenQueueEntry
*entry
;
2004 QTAILQ_FOREACH(entry
, q
, entry
) {
2005 if (entry
->state
.bs
== bs
) {
2006 return entry
->state
.flags
;
2011 return bs
->open_flags
;
2014 /* Returns whether the image file can be written to after the reopen queue @q
2015 * has been successfully applied, or right now if @q is NULL. */
2016 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2017 BlockReopenQueue
*q
)
2019 int flags
= bdrv_reopen_get_flags(q
, bs
);
2021 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2025 * Return whether the BDS can be written to. This is not necessarily
2026 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2027 * be written to but do not count as read-only images.
2029 bool bdrv_is_writable(BlockDriverState
*bs
)
2031 return bdrv_is_writable_after_reopen(bs
, NULL
);
2034 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2035 BdrvChild
*c
, const BdrvChildClass
*child_class
,
2036 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2037 uint64_t parent_perm
, uint64_t parent_shared
,
2038 uint64_t *nperm
, uint64_t *nshared
)
2040 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2041 bs
->drv
->bdrv_child_perm(bs
, c
, child_class
, role
, reopen_queue
,
2042 parent_perm
, parent_shared
,
2044 /* TODO Take force_share from reopen_queue */
2045 if (child_bs
&& child_bs
->force_share
) {
2046 *nshared
= BLK_PERM_ALL
;
2051 * Check whether permissions on this node can be changed in a way that
2052 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
2053 * permissions of all its parents. This involves checking whether all necessary
2054 * permission changes to child nodes can be performed.
2056 * Will set *tighten_restrictions to true if and only if new permissions have to
2057 * be taken or currently shared permissions are to be unshared. Otherwise,
2058 * errors are not fatal as long as the caller accepts that the restrictions
2059 * remain tighter than they need to be. The caller still has to abort the
2061 * @tighten_restrictions cannot be used together with @q: When reopening, we may
2062 * encounter fatal errors even though no restrictions are to be tightened. For
2063 * example, changing a node from RW to RO will fail if the WRITE permission is
2066 * A call to this function must always be followed by a call to bdrv_set_perm()
2067 * or bdrv_abort_perm_update().
2069 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2070 uint64_t cumulative_perms
,
2071 uint64_t cumulative_shared_perms
,
2072 GSList
*ignore_children
,
2073 bool *tighten_restrictions
, Error
**errp
)
2075 BlockDriver
*drv
= bs
->drv
;
2079 assert(!q
|| !tighten_restrictions
);
2081 if (tighten_restrictions
) {
2082 uint64_t current_perms
, current_shared
;
2083 uint64_t added_perms
, removed_shared_perms
;
2085 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2087 added_perms
= cumulative_perms
& ~current_perms
;
2088 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2090 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2093 /* Write permissions never work with read-only images */
2094 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2095 !bdrv_is_writable_after_reopen(bs
, q
))
2097 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2098 error_setg(errp
, "Block node is read-only");
2100 uint64_t current_perms
, current_shared
;
2101 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2102 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2103 error_setg(errp
, "Cannot make block node read-only, there is "
2106 error_setg(errp
, "Cannot make block node read-only and create "
2114 /* Check this node */
2119 if (drv
->bdrv_check_perm
) {
2120 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2121 cumulative_shared_perms
, errp
);
2124 /* Drivers that never have children can omit .bdrv_child_perm() */
2125 if (!drv
->bdrv_child_perm
) {
2126 assert(QLIST_EMPTY(&bs
->children
));
2130 /* Check all children */
2131 QLIST_FOREACH(c
, &bs
->children
, next
) {
2132 uint64_t cur_perm
, cur_shared
;
2133 bool child_tighten_restr
;
2135 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, c
->role
, q
,
2136 cumulative_perms
, cumulative_shared_perms
,
2137 &cur_perm
, &cur_shared
);
2138 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2139 tighten_restrictions
? &child_tighten_restr
2142 if (tighten_restrictions
) {
2143 *tighten_restrictions
|= child_tighten_restr
;
2154 * Notifies drivers that after a previous bdrv_check_perm() call, the
2155 * permission update is not performed and any preparations made for it (e.g.
2156 * taken file locks) need to be undone.
2158 * This function recursively notifies all child nodes.
2160 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2162 BlockDriver
*drv
= bs
->drv
;
2169 if (drv
->bdrv_abort_perm_update
) {
2170 drv
->bdrv_abort_perm_update(bs
);
2173 QLIST_FOREACH(c
, &bs
->children
, next
) {
2174 bdrv_child_abort_perm_update(c
);
2178 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2179 uint64_t cumulative_shared_perms
)
2181 BlockDriver
*drv
= bs
->drv
;
2188 /* Update this node */
2189 if (drv
->bdrv_set_perm
) {
2190 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2193 /* Drivers that never have children can omit .bdrv_child_perm() */
2194 if (!drv
->bdrv_child_perm
) {
2195 assert(QLIST_EMPTY(&bs
->children
));
2199 /* Update all children */
2200 QLIST_FOREACH(c
, &bs
->children
, next
) {
2201 uint64_t cur_perm
, cur_shared
;
2202 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, c
->role
, NULL
,
2203 cumulative_perms
, cumulative_shared_perms
,
2204 &cur_perm
, &cur_shared
);
2205 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2209 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2210 uint64_t *shared_perm
)
2213 uint64_t cumulative_perms
= 0;
2214 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2216 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2217 cumulative_perms
|= c
->perm
;
2218 cumulative_shared_perms
&= c
->shared_perm
;
2221 *perm
= cumulative_perms
;
2222 *shared_perm
= cumulative_shared_perms
;
2225 static char *bdrv_child_user_desc(BdrvChild
*c
)
2227 if (c
->klass
->get_parent_desc
) {
2228 return c
->klass
->get_parent_desc(c
);
2231 return g_strdup("another user");
2234 char *bdrv_perm_names(uint64_t perm
)
2240 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2241 { BLK_PERM_WRITE
, "write" },
2242 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2243 { BLK_PERM_RESIZE
, "resize" },
2244 { BLK_PERM_GRAPH_MOD
, "change children" },
2248 GString
*result
= g_string_sized_new(30);
2249 struct perm_name
*p
;
2251 for (p
= permissions
; p
->name
; p
++) {
2252 if (perm
& p
->perm
) {
2253 if (result
->len
> 0) {
2254 g_string_append(result
, ", ");
2256 g_string_append(result
, p
->name
);
2260 return g_string_free(result
, FALSE
);
2264 * Checks whether a new reference to @bs can be added if the new user requires
2265 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2266 * set, the BdrvChild objects in this list are ignored in the calculations;
2267 * this allows checking permission updates for an existing reference.
2269 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2271 * Needs to be followed by a call to either bdrv_set_perm() or
2272 * bdrv_abort_perm_update(). */
2273 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2274 uint64_t new_used_perm
,
2275 uint64_t new_shared_perm
,
2276 GSList
*ignore_children
,
2277 bool *tighten_restrictions
,
2281 uint64_t cumulative_perms
= new_used_perm
;
2282 uint64_t cumulative_shared_perms
= new_shared_perm
;
2284 assert(!q
|| !tighten_restrictions
);
2286 /* There is no reason why anyone couldn't tolerate write_unchanged */
2287 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2289 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2290 if (g_slist_find(ignore_children
, c
)) {
2294 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2295 char *user
= bdrv_child_user_desc(c
);
2296 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2298 if (tighten_restrictions
) {
2299 *tighten_restrictions
= true;
2302 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2304 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2310 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2311 char *user
= bdrv_child_user_desc(c
);
2312 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2314 if (tighten_restrictions
) {
2315 *tighten_restrictions
= true;
2318 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2320 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2326 cumulative_perms
|= c
->perm
;
2327 cumulative_shared_perms
&= c
->shared_perm
;
2330 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2331 ignore_children
, tighten_restrictions
, errp
);
2334 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2335 * bdrv_child_abort_perm_update(). */
2336 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2337 uint64_t perm
, uint64_t shared
,
2338 GSList
*ignore_children
,
2339 bool *tighten_restrictions
, Error
**errp
)
2343 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2344 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2345 tighten_restrictions
, errp
);
2346 g_slist_free(ignore_children
);
2352 if (!c
->has_backup_perm
) {
2353 c
->has_backup_perm
= true;
2354 c
->backup_perm
= c
->perm
;
2355 c
->backup_shared_perm
= c
->shared_perm
;
2358 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2359 * same child twice during check_perm procedure
2363 c
->shared_perm
= shared
;
2368 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2370 uint64_t cumulative_perms
, cumulative_shared_perms
;
2372 c
->has_backup_perm
= false;
2375 c
->shared_perm
= shared
;
2377 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2378 &cumulative_shared_perms
);
2379 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2382 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2384 if (c
->has_backup_perm
) {
2385 c
->perm
= c
->backup_perm
;
2386 c
->shared_perm
= c
->backup_shared_perm
;
2387 c
->has_backup_perm
= false;
2390 bdrv_abort_perm_update(c
->bs
);
2393 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2396 Error
*local_err
= NULL
;
2398 bool tighten_restrictions
;
2400 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2401 &tighten_restrictions
, &local_err
);
2403 bdrv_child_abort_perm_update(c
);
2404 if (tighten_restrictions
) {
2405 error_propagate(errp
, local_err
);
2408 * Our caller may intend to only loosen restrictions and
2409 * does not expect this function to fail. Errors are not
2410 * fatal in such a case, so we can just hide them from our
2413 error_free(local_err
);
2419 bdrv_child_set_perm(c
, perm
, shared
);
2424 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2426 uint64_t parent_perms
, parent_shared
;
2427 uint64_t perms
, shared
;
2429 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2430 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, c
->role
, NULL
,
2431 parent_perms
, parent_shared
, &perms
, &shared
);
2433 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2436 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2437 const BdrvChildClass
*child_class
,
2439 BlockReopenQueue
*reopen_queue
,
2440 uint64_t perm
, uint64_t shared
,
2441 uint64_t *nperm
, uint64_t *nshared
)
2443 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2444 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2447 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2448 const BdrvChildClass
*child_class
,
2450 BlockReopenQueue
*reopen_queue
,
2451 uint64_t perm
, uint64_t shared
,
2452 uint64_t *nperm
, uint64_t *nshared
)
2454 bool backing
= (child_class
== &child_backing
);
2455 assert(child_class
== &child_backing
|| child_class
== &child_file
);
2458 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2460 /* Apart from the modifications below, the same permissions are
2461 * forwarded and left alone as for filters */
2462 bdrv_filter_default_perms(bs
, c
, child_class
, role
, reopen_queue
,
2463 perm
, shared
, &perm
, &shared
);
2465 /* Format drivers may touch metadata even if the guest doesn't write */
2466 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2467 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2470 /* bs->file always needs to be consistent because of the metadata. We
2471 * can never allow other users to resize or write to it. */
2472 if (!(flags
& BDRV_O_NO_IO
)) {
2473 perm
|= BLK_PERM_CONSISTENT_READ
;
2475 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2477 /* We want consistent read from backing files if the parent needs it.
2478 * No other operations are performed on backing files. */
2479 perm
&= BLK_PERM_CONSISTENT_READ
;
2481 /* If the parent can deal with changing data, we're okay with a
2482 * writable and resizable backing file. */
2483 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2484 if (shared
& BLK_PERM_WRITE
) {
2485 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2490 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2491 BLK_PERM_WRITE_UNCHANGED
;
2494 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2495 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2502 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2504 static const uint64_t permissions
[] = {
2505 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2506 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2507 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2508 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2509 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2512 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2513 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2515 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2517 return permissions
[qapi_perm
];
2520 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2521 BlockDriverState
*new_bs
)
2523 BlockDriverState
*old_bs
= child
->bs
;
2524 int new_bs_quiesce_counter
;
2527 assert(!child
->frozen
);
2529 if (old_bs
&& new_bs
) {
2530 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2533 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2534 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2537 * If the new child node is drained but the old one was not, flush
2538 * all outstanding requests to the old child node.
2540 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2541 bdrv_parent_drained_begin_single(child
, true);
2546 /* Detach first so that the recursive drain sections coming from @child
2547 * are already gone and we only end the drain sections that came from
2549 if (child
->klass
->detach
) {
2550 child
->klass
->detach(child
);
2552 QLIST_REMOVE(child
, next_parent
);
2558 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2561 * Detaching the old node may have led to the new node's
2562 * quiesce_counter having been decreased. Not a problem, we
2563 * just need to recognize this here and then invoke
2564 * drained_end appropriately more often.
2566 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2567 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2569 /* Attach only after starting new drained sections, so that recursive
2570 * drain sections coming from @child don't get an extra .drained_begin
2572 if (child
->klass
->attach
) {
2573 child
->klass
->attach(child
);
2578 * If the old child node was drained but the new one is not, allow
2579 * requests to come in only after the new node has been attached.
2581 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2582 bdrv_parent_drained_end_single(child
);
2588 * Updates @child to change its reference to point to @new_bs, including
2589 * checking and applying the necessary permisson updates both to the old node
2592 * NULL is passed as @new_bs for removing the reference before freeing @child.
2594 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2595 * function uses bdrv_set_perm() to update the permissions according to the new
2596 * reference that @new_bs gets.
2598 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2600 BlockDriverState
*old_bs
= child
->bs
;
2601 uint64_t perm
, shared_perm
;
2603 bdrv_replace_child_noperm(child
, new_bs
);
2606 * Start with the new node's permissions. If @new_bs is a (direct
2607 * or indirect) child of @old_bs, we must complete the permission
2608 * update on @new_bs before we loosen the restrictions on @old_bs.
2609 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2610 * updating the permissions of @new_bs, and thus not purely loosen
2614 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2615 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2619 /* Update permissions for old node. This is guaranteed to succeed
2620 * because we're just taking a parent away, so we're loosening
2622 bool tighten_restrictions
;
2625 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2626 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2627 &tighten_restrictions
, NULL
);
2628 assert(tighten_restrictions
== false);
2630 /* We only tried to loosen restrictions, so errors are not fatal */
2631 bdrv_abort_perm_update(old_bs
);
2633 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2636 /* When the parent requiring a non-default AioContext is removed, the
2637 * node moves back to the main AioContext */
2638 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2643 * This function steals the reference to child_bs from the caller.
2644 * That reference is later dropped by bdrv_root_unref_child().
2646 * On failure NULL is returned, errp is set and the reference to
2647 * child_bs is also dropped.
2649 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2650 * (unless @child_bs is already in @ctx).
2652 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2653 const char *child_name
,
2654 const BdrvChildClass
*child_class
,
2655 BdrvChildRole child_role
,
2657 uint64_t perm
, uint64_t shared_perm
,
2658 void *opaque
, Error
**errp
)
2661 Error
*local_err
= NULL
;
2664 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2667 bdrv_abort_perm_update(child_bs
);
2668 bdrv_unref(child_bs
);
2672 child
= g_new(BdrvChild
, 1);
2673 *child
= (BdrvChild
) {
2675 .name
= g_strdup(child_name
),
2676 .klass
= child_class
,
2679 .shared_perm
= shared_perm
,
2683 /* If the AioContexts don't match, first try to move the subtree of
2684 * child_bs into the AioContext of the new parent. If this doesn't work,
2685 * try moving the parent into the AioContext of child_bs instead. */
2686 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2687 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2688 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2689 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2690 ctx
= bdrv_get_aio_context(child_bs
);
2691 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2692 error_free(local_err
);
2694 g_slist_free(ignore
);
2695 ignore
= g_slist_prepend(NULL
, child
);
2696 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2698 g_slist_free(ignore
);
2701 error_propagate(errp
, local_err
);
2703 bdrv_abort_perm_update(child_bs
);
2704 bdrv_unref(child_bs
);
2709 /* This performs the matching bdrv_set_perm() for the above check. */
2710 bdrv_replace_child(child
, child_bs
);
2716 * This function transfers the reference to child_bs from the caller
2717 * to parent_bs. That reference is later dropped by parent_bs on
2718 * bdrv_close() or if someone calls bdrv_unref_child().
2720 * On failure NULL is returned, errp is set and the reference to
2721 * child_bs is also dropped.
2723 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2724 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2726 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2727 BlockDriverState
*child_bs
,
2728 const char *child_name
,
2729 const BdrvChildClass
*child_class
,
2730 BdrvChildRole child_role
,
2734 uint64_t perm
, shared_perm
;
2736 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2738 assert(parent_bs
->drv
);
2739 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_class
, child_role
, NULL
,
2740 perm
, shared_perm
, &perm
, &shared_perm
);
2742 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2743 child_role
, bdrv_get_aio_context(parent_bs
),
2744 perm
, shared_perm
, parent_bs
, errp
);
2745 if (child
== NULL
) {
2749 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2753 static void bdrv_detach_child(BdrvChild
*child
)
2755 QLIST_SAFE_REMOVE(child
, next
);
2757 bdrv_replace_child(child
, NULL
);
2759 g_free(child
->name
);
2763 void bdrv_root_unref_child(BdrvChild
*child
)
2765 BlockDriverState
*child_bs
;
2767 child_bs
= child
->bs
;
2768 bdrv_detach_child(child
);
2769 bdrv_unref(child_bs
);
2773 * Clear all inherits_from pointers from children and grandchildren of
2774 * @root that point to @root, where necessary.
2776 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2780 if (child
->bs
->inherits_from
== root
) {
2782 * Remove inherits_from only when the last reference between root and
2783 * child->bs goes away.
2785 QLIST_FOREACH(c
, &root
->children
, next
) {
2786 if (c
!= child
&& c
->bs
== child
->bs
) {
2791 child
->bs
->inherits_from
= NULL
;
2795 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2796 bdrv_unset_inherits_from(root
, c
);
2800 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2802 if (child
== NULL
) {
2806 bdrv_unset_inherits_from(parent
, child
);
2807 bdrv_root_unref_child(child
);
2811 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2814 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2815 if (c
->klass
->change_media
) {
2816 c
->klass
->change_media(c
, load
);
2821 /* Return true if you can reach parent going through child->inherits_from
2822 * recursively. If parent or child are NULL, return false */
2823 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2824 BlockDriverState
*parent
)
2826 while (child
&& child
!= parent
) {
2827 child
= child
->inherits_from
;
2830 return child
!= NULL
;
2834 * Sets the backing file link of a BDS. A new reference is created; callers
2835 * which don't need their own reference any more must call bdrv_unref().
2837 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2840 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2841 bdrv_inherits_from_recursive(backing_hd
, bs
);
2843 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2848 bdrv_ref(backing_hd
);
2852 bdrv_unref_child(bs
, bs
->backing
);
2860 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2862 /* If backing_hd was already part of bs's backing chain, and
2863 * inherits_from pointed recursively to bs then let's update it to
2864 * point directly to bs (else it will become NULL). */
2865 if (bs
->backing
&& update_inherits_from
) {
2866 backing_hd
->inherits_from
= bs
;
2870 bdrv_refresh_limits(bs
, NULL
);
2874 * Opens the backing file for a BlockDriverState if not yet open
2876 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2877 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2878 * itself, all options starting with "${bdref_key}." are considered part of the
2881 * TODO Can this be unified with bdrv_open_image()?
2883 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2884 const char *bdref_key
, Error
**errp
)
2886 char *backing_filename
= NULL
;
2887 char *bdref_key_dot
;
2888 const char *reference
= NULL
;
2890 bool implicit_backing
= false;
2891 BlockDriverState
*backing_hd
;
2893 QDict
*tmp_parent_options
= NULL
;
2894 Error
*local_err
= NULL
;
2896 if (bs
->backing
!= NULL
) {
2900 /* NULL means an empty set of options */
2901 if (parent_options
== NULL
) {
2902 tmp_parent_options
= qdict_new();
2903 parent_options
= tmp_parent_options
;
2906 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2908 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2909 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2910 g_free(bdref_key_dot
);
2913 * Caution: while qdict_get_try_str() is fine, getting non-string
2914 * types would require more care. When @parent_options come from
2915 * -blockdev or blockdev_add, its members are typed according to
2916 * the QAPI schema, but when they come from -drive, they're all
2919 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2920 if (reference
|| qdict_haskey(options
, "file.filename")) {
2921 /* keep backing_filename NULL */
2922 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2923 qobject_unref(options
);
2926 if (qdict_size(options
) == 0) {
2927 /* If the user specifies options that do not modify the
2928 * backing file's behavior, we might still consider it the
2929 * implicit backing file. But it's easier this way, and
2930 * just specifying some of the backing BDS's options is
2931 * only possible with -drive anyway (otherwise the QAPI
2932 * schema forces the user to specify everything). */
2933 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2936 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2939 error_propagate(errp
, local_err
);
2940 qobject_unref(options
);
2945 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2947 error_setg(errp
, "Driver doesn't support backing files");
2948 qobject_unref(options
);
2953 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2954 qdict_put_str(options
, "driver", bs
->backing_format
);
2957 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2958 &child_backing
, 0, errp
);
2960 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2961 error_prepend(errp
, "Could not open backing file: ");
2966 if (implicit_backing
) {
2967 bdrv_refresh_filename(backing_hd
);
2968 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2969 backing_hd
->filename
);
2972 /* Hook up the backing file link; drop our reference, bs owns the
2973 * backing_hd reference now */
2974 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2975 bdrv_unref(backing_hd
);
2977 error_propagate(errp
, local_err
);
2982 qdict_del(parent_options
, bdref_key
);
2985 g_free(backing_filename
);
2986 qobject_unref(tmp_parent_options
);
2990 static BlockDriverState
*
2991 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2992 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
2993 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
2995 BlockDriverState
*bs
= NULL
;
2996 QDict
*image_options
;
2997 char *bdref_key_dot
;
2998 const char *reference
;
3000 assert(child_class
!= NULL
);
3002 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3003 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3004 g_free(bdref_key_dot
);
3007 * Caution: while qdict_get_try_str() is fine, getting non-string
3008 * types would require more care. When @options come from
3009 * -blockdev or blockdev_add, its members are typed according to
3010 * the QAPI schema, but when they come from -drive, they're all
3013 reference
= qdict_get_try_str(options
, bdref_key
);
3014 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3016 error_setg(errp
, "A block device must be specified for \"%s\"",
3019 qobject_unref(image_options
);
3023 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3024 parent
, child_class
, child_role
, errp
);
3030 qdict_del(options
, bdref_key
);
3035 * Opens a disk image whose options are given as BlockdevRef in another block
3038 * If allow_none is true, no image will be opened if filename is false and no
3039 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3041 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3042 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3043 * itself, all options starting with "${bdref_key}." are considered part of the
3046 * The BlockdevRef will be removed from the options QDict.
3048 BdrvChild
*bdrv_open_child(const char *filename
,
3049 QDict
*options
, const char *bdref_key
,
3050 BlockDriverState
*parent
,
3051 const BdrvChildClass
*child_class
,
3052 BdrvChildRole child_role
,
3053 bool allow_none
, Error
**errp
)
3055 BlockDriverState
*bs
;
3057 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3058 child_role
, allow_none
, errp
);
3063 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3068 * TODO Future callers may need to specify parent/child_class in order for
3069 * option inheritance to work. Existing callers use it for the root node.
3071 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3073 BlockDriverState
*bs
= NULL
;
3074 QObject
*obj
= NULL
;
3075 QDict
*qdict
= NULL
;
3076 const char *reference
= NULL
;
3079 if (ref
->type
== QTYPE_QSTRING
) {
3080 reference
= ref
->u
.reference
;
3082 BlockdevOptions
*options
= &ref
->u
.definition
;
3083 assert(ref
->type
== QTYPE_QDICT
);
3085 v
= qobject_output_visitor_new(&obj
);
3086 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3087 visit_complete(v
, &obj
);
3089 qdict
= qobject_to(QDict
, obj
);
3090 qdict_flatten(qdict
);
3092 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3093 * compatibility with other callers) rather than what we want as the
3094 * real defaults. Apply the defaults here instead. */
3095 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3096 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3097 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3098 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3102 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3109 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3111 QDict
*snapshot_options
,
3114 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3115 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3117 QemuOpts
*opts
= NULL
;
3118 BlockDriverState
*bs_snapshot
= NULL
;
3119 Error
*local_err
= NULL
;
3122 /* if snapshot, we create a temporary backing file and open it
3123 instead of opening 'filename' directly */
3125 /* Get the required size from the image */
3126 total_size
= bdrv_getlength(bs
);
3127 if (total_size
< 0) {
3128 error_setg_errno(errp
, -total_size
, "Could not get image size");
3132 /* Create the temporary image */
3133 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3135 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3139 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3141 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3142 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3143 qemu_opts_del(opts
);
3145 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3150 /* Prepare options QDict for the temporary file */
3151 qdict_put_str(snapshot_options
, "file.driver", "file");
3152 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3153 qdict_put_str(snapshot_options
, "driver", "qcow2");
3155 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3156 snapshot_options
= NULL
;
3161 /* bdrv_append() consumes a strong reference to bs_snapshot
3162 * (i.e. it will call bdrv_unref() on it) even on error, so in
3163 * order to be able to return one, we have to increase
3164 * bs_snapshot's refcount here */
3165 bdrv_ref(bs_snapshot
);
3166 bdrv_append(bs_snapshot
, bs
, &local_err
);
3168 error_propagate(errp
, local_err
);
3174 qobject_unref(snapshot_options
);
3175 g_free(tmp_filename
);
3180 * Opens a disk image (raw, qcow2, vmdk, ...)
3182 * options is a QDict of options to pass to the block drivers, or NULL for an
3183 * empty set of options. The reference to the QDict belongs to the block layer
3184 * after the call (even on failure), so if the caller intends to reuse the
3185 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3187 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3188 * If it is not NULL, the referenced BDS will be reused.
3190 * The reference parameter may be used to specify an existing block device which
3191 * should be opened. If specified, neither options nor a filename may be given,
3192 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3194 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3195 const char *reference
,
3196 QDict
*options
, int flags
,
3197 BlockDriverState
*parent
,
3198 const BdrvChildClass
*child_class
,
3199 BdrvChildRole child_role
,
3203 BlockBackend
*file
= NULL
;
3204 BlockDriverState
*bs
;
3205 BlockDriver
*drv
= NULL
;
3207 const char *drvname
;
3208 const char *backing
;
3209 Error
*local_err
= NULL
;
3210 QDict
*snapshot_options
= NULL
;
3211 int snapshot_flags
= 0;
3213 assert(!child_class
|| !flags
);
3214 assert(!child_class
== !parent
);
3217 bool options_non_empty
= options
? qdict_size(options
) : false;
3218 qobject_unref(options
);
3220 if (filename
|| options_non_empty
) {
3221 error_setg(errp
, "Cannot reference an existing block device with "
3222 "additional options or a new filename");
3226 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3237 /* NULL means an empty set of options */
3238 if (options
== NULL
) {
3239 options
= qdict_new();
3242 /* json: syntax counts as explicit options, as if in the QDict */
3243 parse_json_protocol(options
, &filename
, &local_err
);
3248 bs
->explicit_options
= qdict_clone_shallow(options
);
3251 bool parent_is_format
;
3254 parent_is_format
= parent
->drv
->is_format
;
3257 * parent->drv is not set yet because this node is opened for
3258 * (potential) format probing. That means that @parent is going
3259 * to be a format node.
3261 parent_is_format
= true;
3264 bs
->inherits_from
= parent
;
3265 child_class
->inherit_options(child_role
, parent_is_format
,
3267 parent
->open_flags
, parent
->options
);
3270 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3276 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3277 * Caution: getting a boolean member of @options requires care.
3278 * When @options come from -blockdev or blockdev_add, members are
3279 * typed according to the QAPI schema, but when they come from
3280 * -drive, they're all QString.
3282 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3283 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3284 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3286 flags
&= ~BDRV_O_RDWR
;
3289 if (flags
& BDRV_O_SNAPSHOT
) {
3290 snapshot_options
= qdict_new();
3291 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3293 /* Let bdrv_backing_options() override "read-only" */
3294 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3295 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3296 &flags
, options
, flags
, options
);
3299 bs
->open_flags
= flags
;
3300 bs
->options
= options
;
3301 options
= qdict_clone_shallow(options
);
3303 /* Find the right image format driver */
3304 /* See cautionary note on accessing @options above */
3305 drvname
= qdict_get_try_str(options
, "driver");
3307 drv
= bdrv_find_format(drvname
);
3309 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3314 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3316 /* See cautionary note on accessing @options above */
3317 backing
= qdict_get_try_str(options
, "backing");
3318 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3319 (backing
&& *backing
== '\0'))
3322 warn_report("Use of \"backing\": \"\" is deprecated; "
3323 "use \"backing\": null instead");
3325 flags
|= BDRV_O_NO_BACKING
;
3326 qdict_del(bs
->explicit_options
, "backing");
3327 qdict_del(bs
->options
, "backing");
3328 qdict_del(options
, "backing");
3331 /* Open image file without format layer. This BlockBackend is only used for
3332 * probing, the block drivers will do their own bdrv_open_child() for the
3333 * same BDS, which is why we put the node name back into options. */
3334 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3335 BlockDriverState
*file_bs
;
3337 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3338 &child_file
, 0, true, &local_err
);
3342 if (file_bs
!= NULL
) {
3343 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3344 * looking at the header to guess the image format. This works even
3345 * in cases where a guest would not see a consistent state. */
3346 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3347 blk_insert_bs(file
, file_bs
, &local_err
);
3348 bdrv_unref(file_bs
);
3353 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3357 /* Image format probing */
3360 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3365 * This option update would logically belong in bdrv_fill_options(),
3366 * but we first need to open bs->file for the probing to work, while
3367 * opening bs->file already requires the (mostly) final set of options
3368 * so that cache mode etc. can be inherited.
3370 * Adding the driver later is somewhat ugly, but it's not an option
3371 * that would ever be inherited, so it's correct. We just need to make
3372 * sure to update both bs->options (which has the full effective
3373 * options for bs) and options (which has file.* already removed).
3375 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3376 qdict_put_str(options
, "driver", drv
->format_name
);
3378 error_setg(errp
, "Must specify either driver or file");
3382 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3383 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3384 /* file must be NULL if a protocol BDS is about to be created
3385 * (the inverse results in an error message from bdrv_open_common()) */
3386 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3388 /* Open the image */
3389 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3399 /* If there is a backing file, use it */
3400 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3401 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3403 goto close_and_fail
;
3407 /* Remove all children options and references
3408 * from bs->options and bs->explicit_options */
3409 QLIST_FOREACH(child
, &bs
->children
, next
) {
3410 char *child_key_dot
;
3411 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3412 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3413 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3414 qdict_del(bs
->explicit_options
, child
->name
);
3415 qdict_del(bs
->options
, child
->name
);
3416 g_free(child_key_dot
);
3419 /* Check if any unknown options were used */
3420 if (qdict_size(options
) != 0) {
3421 const QDictEntry
*entry
= qdict_first(options
);
3422 if (flags
& BDRV_O_PROTOCOL
) {
3423 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3424 "'%s'", drv
->format_name
, entry
->key
);
3427 "Block format '%s' does not support the option '%s'",
3428 drv
->format_name
, entry
->key
);
3431 goto close_and_fail
;
3434 bdrv_parent_cb_change_media(bs
, true);
3436 qobject_unref(options
);
3439 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3440 * temporary snapshot afterwards. */
3441 if (snapshot_flags
) {
3442 BlockDriverState
*snapshot_bs
;
3443 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3444 snapshot_options
, &local_err
);
3445 snapshot_options
= NULL
;
3447 goto close_and_fail
;
3449 /* We are not going to return bs but the overlay on top of it
3450 * (snapshot_bs); thus, we have to drop the strong reference to bs
3451 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3452 * though, because the overlay still has a reference to it. */
3461 qobject_unref(snapshot_options
);
3462 qobject_unref(bs
->explicit_options
);
3463 qobject_unref(bs
->options
);
3464 qobject_unref(options
);
3466 bs
->explicit_options
= NULL
;
3468 error_propagate(errp
, local_err
);
3473 qobject_unref(snapshot_options
);
3474 qobject_unref(options
);
3475 error_propagate(errp
, local_err
);
3479 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3480 QDict
*options
, int flags
, Error
**errp
)
3482 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3486 /* Return true if the NULL-terminated @list contains @str */
3487 static bool is_str_in_list(const char *str
, const char *const *list
)
3491 for (i
= 0; list
[i
] != NULL
; i
++) {
3492 if (!strcmp(str
, list
[i
])) {
3501 * Check that every option set in @bs->options is also set in
3504 * Options listed in the common_options list and in
3505 * @bs->drv->mutable_opts are skipped.
3507 * Return 0 on success, otherwise return -EINVAL and set @errp.
3509 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3510 const QDict
*new_opts
, Error
**errp
)
3512 const QDictEntry
*e
;
3513 /* These options are common to all block drivers and are handled
3514 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3515 const char *const common_options
[] = {
3516 "node-name", "discard", "cache.direct", "cache.no-flush",
3517 "read-only", "auto-read-only", "detect-zeroes", NULL
3520 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3521 if (!qdict_haskey(new_opts
, e
->key
) &&
3522 !is_str_in_list(e
->key
, common_options
) &&
3523 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3524 error_setg(errp
, "Option '%s' cannot be reset "
3525 "to its default value", e
->key
);
3534 * Returns true if @child can be reached recursively from @bs
3536 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3537 BlockDriverState
*child
)
3545 QLIST_FOREACH(c
, &bs
->children
, next
) {
3546 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3555 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3556 * reopen of multiple devices.
3558 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3559 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3560 * be created and initialized. This newly created BlockReopenQueue should be
3561 * passed back in for subsequent calls that are intended to be of the same
3564 * bs is the BlockDriverState to add to the reopen queue.
3566 * options contains the changed options for the associated bs
3567 * (the BlockReopenQueue takes ownership)
3569 * flags contains the open flags for the associated bs
3571 * returns a pointer to bs_queue, which is either the newly allocated
3572 * bs_queue, or the existing bs_queue being used.
3574 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3576 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3577 BlockDriverState
*bs
,
3579 const BdrvChildClass
*klass
,
3581 bool parent_is_format
,
3582 QDict
*parent_options
,
3588 BlockReopenQueueEntry
*bs_entry
;
3590 QDict
*old_options
, *explicit_options
, *options_copy
;
3594 /* Make sure that the caller remembered to use a drained section. This is
3595 * important to avoid graph changes between the recursive queuing here and
3596 * bdrv_reopen_multiple(). */
3597 assert(bs
->quiesce_counter
> 0);
3599 if (bs_queue
== NULL
) {
3600 bs_queue
= g_new0(BlockReopenQueue
, 1);
3601 QTAILQ_INIT(bs_queue
);
3605 options
= qdict_new();
3608 /* Check if this BlockDriverState is already in the queue */
3609 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3610 if (bs
== bs_entry
->state
.bs
) {
3616 * Precedence of options:
3617 * 1. Explicitly passed in options (highest)
3618 * 2. Retained from explicitly set options of bs
3619 * 3. Inherited from parent node
3620 * 4. Retained from effective options of bs
3623 /* Old explicitly set values (don't overwrite by inherited value) */
3624 if (bs_entry
|| keep_old_opts
) {
3625 old_options
= qdict_clone_shallow(bs_entry
?
3626 bs_entry
->state
.explicit_options
:
3627 bs
->explicit_options
);
3628 bdrv_join_options(bs
, options
, old_options
);
3629 qobject_unref(old_options
);
3632 explicit_options
= qdict_clone_shallow(options
);
3634 /* Inherit from parent node */
3635 if (parent_options
) {
3637 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
3638 parent_flags
, parent_options
);
3640 flags
= bdrv_get_flags(bs
);
3643 if (keep_old_opts
) {
3644 /* Old values are used for options that aren't set yet */
3645 old_options
= qdict_clone_shallow(bs
->options
);
3646 bdrv_join_options(bs
, options
, old_options
);
3647 qobject_unref(old_options
);
3650 /* We have the final set of options so let's update the flags */
3651 options_copy
= qdict_clone_shallow(options
);
3652 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3653 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3654 update_flags_from_options(&flags
, opts
);
3655 qemu_opts_del(opts
);
3656 qobject_unref(options_copy
);
3658 /* bdrv_open_inherit() sets and clears some additional flags internally */
3659 flags
&= ~BDRV_O_PROTOCOL
;
3660 if (flags
& BDRV_O_RDWR
) {
3661 flags
|= BDRV_O_ALLOW_RDWR
;
3665 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3666 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3668 qobject_unref(bs_entry
->state
.options
);
3669 qobject_unref(bs_entry
->state
.explicit_options
);
3672 bs_entry
->state
.bs
= bs
;
3673 bs_entry
->state
.options
= options
;
3674 bs_entry
->state
.explicit_options
= explicit_options
;
3675 bs_entry
->state
.flags
= flags
;
3677 /* This needs to be overwritten in bdrv_reopen_prepare() */
3678 bs_entry
->state
.perm
= UINT64_MAX
;
3679 bs_entry
->state
.shared_perm
= 0;
3682 * If keep_old_opts is false then it means that unspecified
3683 * options must be reset to their original value. We don't allow
3684 * resetting 'backing' but we need to know if the option is
3685 * missing in order to decide if we have to return an error.
3687 if (!keep_old_opts
) {
3688 bs_entry
->state
.backing_missing
=
3689 !qdict_haskey(options
, "backing") &&
3690 !qdict_haskey(options
, "backing.driver");
3693 QLIST_FOREACH(child
, &bs
->children
, next
) {
3694 QDict
*new_child_options
= NULL
;
3695 bool child_keep_old
= keep_old_opts
;
3697 /* reopen can only change the options of block devices that were
3698 * implicitly created and inherited options. For other (referenced)
3699 * block devices, a syntax like "backing.foo" results in an error. */
3700 if (child
->bs
->inherits_from
!= bs
) {
3704 /* Check if the options contain a child reference */
3705 if (qdict_haskey(options
, child
->name
)) {
3706 const char *childref
= qdict_get_try_str(options
, child
->name
);
3708 * The current child must not be reopened if the child
3709 * reference is null or points to a different node.
3711 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3715 * If the child reference points to the current child then
3716 * reopen it with its existing set of options (note that
3717 * it can still inherit new options from the parent).
3719 child_keep_old
= true;
3721 /* Extract child options ("child-name.*") */
3722 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3723 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3724 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3725 g_free(child_key_dot
);
3728 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3729 child
->klass
, child
->role
, bs
->drv
->is_format
,
3730 options
, flags
, child_keep_old
);
3736 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3737 BlockDriverState
*bs
,
3738 QDict
*options
, bool keep_old_opts
)
3740 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
3741 NULL
, 0, keep_old_opts
);
3745 * Reopen multiple BlockDriverStates atomically & transactionally.
3747 * The queue passed in (bs_queue) must have been built up previous
3748 * via bdrv_reopen_queue().
3750 * Reopens all BDS specified in the queue, with the appropriate
3751 * flags. All devices are prepared for reopen, and failure of any
3752 * device will cause all device changes to be abandoned, and intermediate
3755 * If all devices prepare successfully, then the changes are committed
3758 * All affected nodes must be drained between bdrv_reopen_queue() and
3759 * bdrv_reopen_multiple().
3761 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3764 BlockReopenQueueEntry
*bs_entry
, *next
;
3766 assert(bs_queue
!= NULL
);
3768 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3769 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3770 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3773 bs_entry
->prepared
= true;
3776 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3777 BDRVReopenState
*state
= &bs_entry
->state
;
3778 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3779 state
->shared_perm
, NULL
, NULL
, errp
);
3783 /* Check if new_backing_bs would accept the new permissions */
3784 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3785 uint64_t nperm
, nshared
;
3786 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3787 NULL
, &child_backing
, 0, bs_queue
,
3788 state
->perm
, state
->shared_perm
,
3790 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3791 nperm
, nshared
, NULL
, NULL
, errp
);
3796 bs_entry
->perms_checked
= true;
3800 * If we reach this point, we have success and just need to apply the
3803 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3804 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3805 * children are usually goes after parents in reopen-queue, so go from last
3808 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3809 bdrv_reopen_commit(&bs_entry
->state
);
3814 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3815 BDRVReopenState
*state
= &bs_entry
->state
;
3817 if (!bs_entry
->perms_checked
) {
3822 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3824 bdrv_abort_perm_update(state
->bs
);
3825 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3826 bdrv_abort_perm_update(state
->new_backing_bs
);
3832 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3833 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3835 if (bs
->drv
->bdrv_reopen_commit_post
)
3836 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3840 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3842 if (bs_entry
->prepared
) {
3843 bdrv_reopen_abort(&bs_entry
->state
);
3845 qobject_unref(bs_entry
->state
.explicit_options
);
3846 qobject_unref(bs_entry
->state
.options
);
3848 if (bs_entry
->state
.new_backing_bs
) {
3849 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3858 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3862 BlockReopenQueue
*queue
;
3863 QDict
*opts
= qdict_new();
3865 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3867 bdrv_subtree_drained_begin(bs
);
3868 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3869 ret
= bdrv_reopen_multiple(queue
, errp
);
3870 bdrv_subtree_drained_end(bs
);
3875 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3878 BlockReopenQueueEntry
*entry
;
3880 QTAILQ_FOREACH(entry
, q
, entry
) {
3881 BlockDriverState
*bs
= entry
->state
.bs
;
3884 QLIST_FOREACH(child
, &bs
->children
, next
) {
3894 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3895 uint64_t *perm
, uint64_t *shared
)
3898 BlockReopenQueueEntry
*parent
;
3899 uint64_t cumulative_perms
= 0;
3900 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3902 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3903 parent
= find_parent_in_reopen_queue(q
, c
);
3905 cumulative_perms
|= c
->perm
;
3906 cumulative_shared_perms
&= c
->shared_perm
;
3908 uint64_t nperm
, nshared
;
3910 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->klass
, c
->role
, q
,
3911 parent
->state
.perm
, parent
->state
.shared_perm
,
3914 cumulative_perms
|= nperm
;
3915 cumulative_shared_perms
&= nshared
;
3918 *perm
= cumulative_perms
;
3919 *shared
= cumulative_shared_perms
;
3922 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3924 BlockDriverState
*new_child
,
3927 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3928 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3932 ignore
= g_slist_prepend(NULL
, child
);
3933 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3934 g_slist_free(ignore
);
3939 ignore
= g_slist_prepend(NULL
, child
);
3940 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3941 g_slist_free(ignore
);
3946 * Take a BDRVReopenState and check if the value of 'backing' in the
3947 * reopen_state->options QDict is valid or not.
3949 * If 'backing' is missing from the QDict then return 0.
3951 * If 'backing' contains the node name of the backing file of
3952 * reopen_state->bs then return 0.
3954 * If 'backing' contains a different node name (or is null) then check
3955 * whether the current backing file can be replaced with the new one.
3956 * If that's the case then reopen_state->replace_backing_bs is set to
3957 * true and reopen_state->new_backing_bs contains a pointer to the new
3958 * backing BlockDriverState (or NULL).
3960 * Return 0 on success, otherwise return < 0 and set @errp.
3962 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3965 BlockDriverState
*bs
= reopen_state
->bs
;
3966 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3970 value
= qdict_get(reopen_state
->options
, "backing");
3971 if (value
== NULL
) {
3975 switch (qobject_type(value
)) {
3977 new_backing_bs
= NULL
;
3980 str
= qobject_get_try_str(value
);
3981 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3982 if (new_backing_bs
== NULL
) {
3984 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3985 error_setg(errp
, "Making '%s' a backing file of '%s' "
3986 "would create a cycle", str
, bs
->node_name
);
3991 /* 'backing' does not allow any other data type */
3992 g_assert_not_reached();
3996 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
3997 * bdrv_reopen_commit() won't fail.
3999 if (new_backing_bs
) {
4000 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4006 * Find the "actual" backing file by skipping all links that point
4007 * to an implicit node, if any (e.g. a commit filter node).
4010 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
4011 overlay_bs
= backing_bs(overlay_bs
);
4014 /* If we want to replace the backing file we need some extra checks */
4015 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
4016 /* Check for implicit nodes between bs and its backing file */
4017 if (bs
!= overlay_bs
) {
4018 error_setg(errp
, "Cannot change backing link if '%s' has "
4019 "an implicit backing file", bs
->node_name
);
4022 /* Check if the backing link that we want to replace is frozen */
4023 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
4027 reopen_state
->replace_backing_bs
= true;
4028 if (new_backing_bs
) {
4029 bdrv_ref(new_backing_bs
);
4030 reopen_state
->new_backing_bs
= new_backing_bs
;
4038 * Prepares a BlockDriverState for reopen. All changes are staged in the
4039 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4040 * the block driver layer .bdrv_reopen_prepare()
4042 * bs is the BlockDriverState to reopen
4043 * flags are the new open flags
4044 * queue is the reopen queue
4046 * Returns 0 on success, non-zero on error. On error errp will be set
4049 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4050 * It is the responsibility of the caller to then call the abort() or
4051 * commit() for any other BDS that have been left in a prepare() state
4054 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4059 Error
*local_err
= NULL
;
4062 QDict
*orig_reopen_opts
;
4063 char *discard
= NULL
;
4065 bool drv_prepared
= false;
4067 assert(reopen_state
!= NULL
);
4068 assert(reopen_state
->bs
->drv
!= NULL
);
4069 drv
= reopen_state
->bs
->drv
;
4071 /* This function and each driver's bdrv_reopen_prepare() remove
4072 * entries from reopen_state->options as they are processed, so
4073 * we need to make a copy of the original QDict. */
4074 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4076 /* Process generic block layer options */
4077 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4078 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
4080 error_propagate(errp
, local_err
);
4085 /* This was already called in bdrv_reopen_queue_child() so the flags
4086 * are up-to-date. This time we simply want to remove the options from
4087 * QemuOpts in order to indicate that they have been processed. */
4088 old_flags
= reopen_state
->flags
;
4089 update_flags_from_options(&reopen_state
->flags
, opts
);
4090 assert(old_flags
== reopen_state
->flags
);
4092 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4093 if (discard
!= NULL
) {
4094 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4095 error_setg(errp
, "Invalid discard option");
4101 reopen_state
->detect_zeroes
=
4102 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4104 error_propagate(errp
, local_err
);
4109 /* All other options (including node-name and driver) must be unchanged.
4110 * Put them back into the QDict, so that they are checked at the end
4111 * of this function. */
4112 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4114 /* If we are to stay read-only, do not allow permission change
4115 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4116 * not set, or if the BDS still has copy_on_read enabled */
4117 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4118 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4120 error_propagate(errp
, local_err
);
4124 /* Calculate required permissions after reopening */
4125 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4126 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4128 ret
= bdrv_flush(reopen_state
->bs
);
4130 error_setg_errno(errp
, -ret
, "Error flushing drive");
4134 if (drv
->bdrv_reopen_prepare
) {
4136 * If a driver-specific option is missing, it means that we
4137 * should reset it to its default value.
4138 * But not all options allow that, so we need to check it first.
4140 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4141 reopen_state
->options
, errp
);
4146 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4148 if (local_err
!= NULL
) {
4149 error_propagate(errp
, local_err
);
4151 bdrv_refresh_filename(reopen_state
->bs
);
4152 error_setg(errp
, "failed while preparing to reopen image '%s'",
4153 reopen_state
->bs
->filename
);
4158 /* It is currently mandatory to have a bdrv_reopen_prepare()
4159 * handler for each supported drv. */
4160 error_setg(errp
, "Block format '%s' used by node '%s' "
4161 "does not support reopening files", drv
->format_name
,
4162 bdrv_get_device_or_node_name(reopen_state
->bs
));
4167 drv_prepared
= true;
4170 * We must provide the 'backing' option if the BDS has a backing
4171 * file or if the image file has a backing file name as part of
4172 * its metadata. Otherwise the 'backing' option can be omitted.
4174 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4175 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4176 error_setg(errp
, "backing is missing for '%s'",
4177 reopen_state
->bs
->node_name
);
4183 * Allow changing the 'backing' option. The new value can be
4184 * either a reference to an existing node (using its node name)
4185 * or NULL to simply detach the current backing file.
4187 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4191 qdict_del(reopen_state
->options
, "backing");
4193 /* Options that are not handled are only okay if they are unchanged
4194 * compared to the old state. It is expected that some options are only
4195 * used for the initial open, but not reopen (e.g. filename) */
4196 if (qdict_size(reopen_state
->options
)) {
4197 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4200 QObject
*new = entry
->value
;
4201 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4203 /* Allow child references (child_name=node_name) as long as they
4204 * point to the current child (i.e. everything stays the same). */
4205 if (qobject_type(new) == QTYPE_QSTRING
) {
4207 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4208 if (!strcmp(child
->name
, entry
->key
)) {
4214 const char *str
= qobject_get_try_str(new);
4215 if (!strcmp(child
->bs
->node_name
, str
)) {
4216 continue; /* Found child with this name, skip option */
4222 * TODO: When using -drive to specify blockdev options, all values
4223 * will be strings; however, when using -blockdev, blockdev-add or
4224 * filenames using the json:{} pseudo-protocol, they will be
4226 * In contrast, reopening options are (currently) always strings
4227 * (because you can only specify them through qemu-io; all other
4228 * callers do not specify any options).
4229 * Therefore, when using anything other than -drive to create a BDS,
4230 * this cannot detect non-string options as unchanged, because
4231 * qobject_is_equal() always returns false for objects of different
4232 * type. In the future, this should be remedied by correctly typing
4233 * all options. For now, this is not too big of an issue because
4234 * the user can simply omit options which cannot be changed anyway,
4235 * so they will stay unchanged.
4237 if (!qobject_is_equal(new, old
)) {
4238 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4242 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4247 /* Restore the original reopen_state->options QDict */
4248 qobject_unref(reopen_state
->options
);
4249 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4252 if (ret
< 0 && drv_prepared
) {
4253 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4254 * call drv->bdrv_reopen_abort() before signaling an error
4255 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4256 * when the respective bdrv_reopen_prepare() has failed) */
4257 if (drv
->bdrv_reopen_abort
) {
4258 drv
->bdrv_reopen_abort(reopen_state
);
4261 qemu_opts_del(opts
);
4262 qobject_unref(orig_reopen_opts
);
4268 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4269 * makes them final by swapping the staging BlockDriverState contents into
4270 * the active BlockDriverState contents.
4272 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4275 BlockDriverState
*bs
;
4278 assert(reopen_state
!= NULL
);
4279 bs
= reopen_state
->bs
;
4281 assert(drv
!= NULL
);
4283 /* If there are any driver level actions to take */
4284 if (drv
->bdrv_reopen_commit
) {
4285 drv
->bdrv_reopen_commit(reopen_state
);
4288 /* set BDS specific flags now */
4289 qobject_unref(bs
->explicit_options
);
4290 qobject_unref(bs
->options
);
4292 bs
->explicit_options
= reopen_state
->explicit_options
;
4293 bs
->options
= reopen_state
->options
;
4294 bs
->open_flags
= reopen_state
->flags
;
4295 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4296 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4298 if (reopen_state
->replace_backing_bs
) {
4299 qdict_del(bs
->explicit_options
, "backing");
4300 qdict_del(bs
->options
, "backing");
4303 /* Remove child references from bs->options and bs->explicit_options.
4304 * Child options were already removed in bdrv_reopen_queue_child() */
4305 QLIST_FOREACH(child
, &bs
->children
, next
) {
4306 qdict_del(bs
->explicit_options
, child
->name
);
4307 qdict_del(bs
->options
, child
->name
);
4311 * Change the backing file if a new one was specified. We do this
4312 * after updating bs->options, so bdrv_refresh_filename() (called
4313 * from bdrv_set_backing_hd()) has the new values.
4315 if (reopen_state
->replace_backing_bs
) {
4316 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4317 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4318 /* Abort the permission update on the backing bs we're detaching */
4319 if (old_backing_bs
) {
4320 bdrv_abort_perm_update(old_backing_bs
);
4322 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4325 bdrv_refresh_limits(bs
, NULL
);
4329 * Abort the reopen, and delete and free the staged changes in
4332 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4336 assert(reopen_state
!= NULL
);
4337 drv
= reopen_state
->bs
->drv
;
4338 assert(drv
!= NULL
);
4340 if (drv
->bdrv_reopen_abort
) {
4341 drv
->bdrv_reopen_abort(reopen_state
);
4346 static void bdrv_close(BlockDriverState
*bs
)
4348 BdrvAioNotifier
*ban
, *ban_next
;
4349 BdrvChild
*child
, *next
;
4351 assert(!bs
->refcnt
);
4353 bdrv_drained_begin(bs
); /* complete I/O */
4355 bdrv_drain(bs
); /* in case flush left pending I/O */
4358 if (bs
->drv
->bdrv_close
) {
4359 bs
->drv
->bdrv_close(bs
);
4364 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4365 bdrv_unref_child(bs
, child
);
4372 atomic_set(&bs
->copy_on_read
, 0);
4373 bs
->backing_file
[0] = '\0';
4374 bs
->backing_format
[0] = '\0';
4375 bs
->total_sectors
= 0;
4376 bs
->encrypted
= false;
4378 qobject_unref(bs
->options
);
4379 qobject_unref(bs
->explicit_options
);
4381 bs
->explicit_options
= NULL
;
4382 qobject_unref(bs
->full_open_options
);
4383 bs
->full_open_options
= NULL
;
4385 bdrv_release_named_dirty_bitmaps(bs
);
4386 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4388 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4391 QLIST_INIT(&bs
->aio_notifiers
);
4392 bdrv_drained_end(bs
);
4395 void bdrv_close_all(void)
4397 assert(job_next(NULL
) == NULL
);
4398 nbd_export_close_all();
4400 /* Drop references from requests still in flight, such as canceled block
4401 * jobs whose AIO context has not been polled yet */
4404 blk_remove_all_bs();
4405 blockdev_close_all_bdrv_states();
4407 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4410 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4416 if (c
->klass
->stay_at_node
) {
4420 /* If the child @c belongs to the BDS @to, replacing the current
4421 * c->bs by @to would mean to create a loop.
4423 * Such a case occurs when appending a BDS to a backing chain.
4424 * For instance, imagine the following chain:
4426 * guest device -> node A -> further backing chain...
4428 * Now we create a new BDS B which we want to put on top of this
4429 * chain, so we first attach A as its backing node:
4434 * guest device -> node A -> further backing chain...
4436 * Finally we want to replace A by B. When doing that, we want to
4437 * replace all pointers to A by pointers to B -- except for the
4438 * pointer from B because (1) that would create a loop, and (2)
4439 * that pointer should simply stay intact:
4441 * guest device -> node B
4444 * node A -> further backing chain...
4446 * In general, when replacing a node A (c->bs) by a node B (@to),
4447 * if A is a child of B, that means we cannot replace A by B there
4448 * because that would create a loop. Silently detaching A from B
4449 * is also not really an option. So overall just leaving A in
4450 * place there is the most sensible choice.
4452 * We would also create a loop in any cases where @c is only
4453 * indirectly referenced by @to. Prevent this by returning false
4454 * if @c is found (by breadth-first search) anywhere in the whole
4459 found
= g_hash_table_new(NULL
, NULL
);
4460 g_hash_table_add(found
, to
);
4461 queue
= g_queue_new();
4462 g_queue_push_tail(queue
, to
);
4464 while (!g_queue_is_empty(queue
)) {
4465 BlockDriverState
*v
= g_queue_pop_head(queue
);
4468 QLIST_FOREACH(c2
, &v
->children
, next
) {
4474 if (g_hash_table_contains(found
, c2
->bs
)) {
4478 g_queue_push_tail(queue
, c2
->bs
);
4479 g_hash_table_add(found
, c2
->bs
);
4483 g_queue_free(queue
);
4484 g_hash_table_destroy(found
);
4489 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4492 BdrvChild
*c
, *next
;
4493 GSList
*list
= NULL
, *p
;
4494 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4497 /* Make sure that @from doesn't go away until we have successfully attached
4498 * all of its parents to @to. */
4501 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4502 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4503 bdrv_drained_begin(from
);
4505 /* Put all parents into @list and calculate their cumulative permissions */
4506 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4507 assert(c
->bs
== from
);
4508 if (!should_update_child(c
, to
)) {
4512 error_setg(errp
, "Cannot change '%s' link to '%s'",
4513 c
->name
, from
->node_name
);
4516 list
= g_slist_prepend(list
, c
);
4518 shared
&= c
->shared_perm
;
4521 /* Check whether the required permissions can be granted on @to, ignoring
4522 * all BdrvChild in @list so that they can't block themselves. */
4523 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4525 bdrv_abort_perm_update(to
);
4529 /* Now actually perform the change. We performed the permission check for
4530 * all elements of @list at once, so set the permissions all at once at the
4532 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4536 bdrv_replace_child_noperm(c
, to
);
4540 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4541 bdrv_set_perm(to
, perm
, shared
);
4545 bdrv_drained_end(from
);
4550 * Add new bs contents at the top of an image chain while the chain is
4551 * live, while keeping required fields on the top layer.
4553 * This will modify the BlockDriverState fields, and swap contents
4554 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4556 * bs_new must not be attached to a BlockBackend.
4558 * This function does not create any image files.
4560 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4561 * that's what the callers commonly need. bs_new will be referenced by the old
4562 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4563 * reference of its own, it must call bdrv_ref().
4565 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4568 Error
*local_err
= NULL
;
4570 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4572 error_propagate(errp
, local_err
);
4576 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4578 error_propagate(errp
, local_err
);
4579 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4583 /* bs_new is now referenced by its new parents, we don't need the
4584 * additional reference any more. */
4589 static void bdrv_delete(BlockDriverState
*bs
)
4591 assert(bdrv_op_blocker_is_empty(bs
));
4592 assert(!bs
->refcnt
);
4594 /* remove from list, if necessary */
4595 if (bs
->node_name
[0] != '\0') {
4596 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4598 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4606 * Run consistency checks on an image
4608 * Returns 0 if the check could be completed (it doesn't mean that the image is
4609 * free of errors) or -errno when an internal error occurred. The results of the
4610 * check are stored in res.
4612 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4613 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4615 if (bs
->drv
== NULL
) {
4618 if (bs
->drv
->bdrv_co_check
== NULL
) {
4622 memset(res
, 0, sizeof(*res
));
4623 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4626 typedef struct CheckCo
{
4627 BlockDriverState
*bs
;
4628 BdrvCheckResult
*res
;
4633 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4635 CheckCo
*cco
= opaque
;
4636 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4640 int bdrv_check(BlockDriverState
*bs
,
4641 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4647 .ret
= -EINPROGRESS
,
4651 if (qemu_in_coroutine()) {
4652 /* Fast-path if already in coroutine context */
4653 bdrv_check_co_entry(&cco
);
4655 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4656 bdrv_coroutine_enter(bs
, co
);
4657 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4666 * -EINVAL - backing format specified, but no file
4667 * -ENOSPC - can't update the backing file because no space is left in the
4669 * -ENOTSUP - format driver doesn't support changing the backing file
4671 int bdrv_change_backing_file(BlockDriverState
*bs
,
4672 const char *backing_file
, const char *backing_fmt
)
4674 BlockDriver
*drv
= bs
->drv
;
4681 /* Backing file format doesn't make sense without a backing file */
4682 if (backing_fmt
&& !backing_file
) {
4686 if (drv
->bdrv_change_backing_file
!= NULL
) {
4687 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4693 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4694 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4695 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4696 backing_file
?: "");
4702 * Finds the image layer in the chain that has 'bs' as its backing file.
4704 * active is the current topmost image.
4706 * Returns NULL if bs is not found in active's image chain,
4707 * or if active == bs.
4709 * Returns the bottommost base image if bs == NULL.
4711 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4712 BlockDriverState
*bs
)
4714 while (active
&& bs
!= backing_bs(active
)) {
4715 active
= backing_bs(active
);
4721 /* Given a BDS, searches for the base layer. */
4722 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4724 return bdrv_find_overlay(bs
, NULL
);
4728 * Return true if at least one of the backing links between @bs and
4729 * @base is frozen. @errp is set if that's the case.
4730 * @base must be reachable from @bs, or NULL.
4732 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4735 BlockDriverState
*i
;
4737 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4738 if (i
->backing
&& i
->backing
->frozen
) {
4739 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4740 i
->backing
->name
, i
->node_name
,
4741 backing_bs(i
)->node_name
);
4750 * Freeze all backing links between @bs and @base.
4751 * If any of the links is already frozen the operation is aborted and
4752 * none of the links are modified.
4753 * @base must be reachable from @bs, or NULL.
4754 * Returns 0 on success. On failure returns < 0 and sets @errp.
4756 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4759 BlockDriverState
*i
;
4761 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4765 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4766 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4767 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4768 i
->backing
->name
, backing_bs(i
)->node_name
);
4773 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4775 i
->backing
->frozen
= true;
4783 * Unfreeze all backing links between @bs and @base. The caller must
4784 * ensure that all links are frozen before using this function.
4785 * @base must be reachable from @bs, or NULL.
4787 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4789 BlockDriverState
*i
;
4791 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4793 assert(i
->backing
->frozen
);
4794 i
->backing
->frozen
= false;
4800 * Drops images above 'base' up to and including 'top', and sets the image
4801 * above 'top' to have base as its backing file.
4803 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4804 * information in 'bs' can be properly updated.
4806 * E.g., this will convert the following chain:
4807 * bottom <- base <- intermediate <- top <- active
4811 * bottom <- base <- active
4813 * It is allowed for bottom==base, in which case it converts:
4815 * base <- intermediate <- top <- active
4821 * If backing_file_str is non-NULL, it will be used when modifying top's
4822 * overlay image metadata.
4825 * if active == top, that is considered an error
4828 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4829 const char *backing_file_str
)
4831 BlockDriverState
*explicit_top
= top
;
4832 bool update_inherits_from
;
4833 BdrvChild
*c
, *next
;
4834 Error
*local_err
= NULL
;
4838 bdrv_subtree_drained_begin(top
);
4840 if (!top
->drv
|| !base
->drv
) {
4844 /* Make sure that base is in the backing chain of top */
4845 if (!bdrv_chain_contains(top
, base
)) {
4849 /* This function changes all links that point to top and makes
4850 * them point to base. Check that none of them is frozen. */
4851 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4857 /* If 'base' recursively inherits from 'top' then we should set
4858 * base->inherits_from to top->inherits_from after 'top' and all
4859 * other intermediate nodes have been dropped.
4860 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4861 * it because no one inherits from it. We use explicit_top for that. */
4862 while (explicit_top
&& explicit_top
->implicit
) {
4863 explicit_top
= backing_bs(explicit_top
);
4865 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4867 /* success - we can delete the intermediate states, and link top->base */
4868 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4869 * we've figured out how they should work. */
4870 if (!backing_file_str
) {
4871 bdrv_refresh_filename(base
);
4872 backing_file_str
= base
->filename
;
4875 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4876 /* Check whether we are allowed to switch c from top to base */
4877 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4878 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4879 ignore_children
, NULL
, &local_err
);
4880 g_slist_free(ignore_children
);
4882 error_report_err(local_err
);
4886 /* If so, update the backing file path in the image file */
4887 if (c
->klass
->update_filename
) {
4888 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
4891 bdrv_abort_perm_update(base
);
4892 error_report_err(local_err
);
4897 /* Do the actual switch in the in-memory graph.
4898 * Completes bdrv_check_update_perm() transaction internally. */
4900 bdrv_replace_child(c
, base
);
4904 if (update_inherits_from
) {
4905 base
->inherits_from
= explicit_top
->inherits_from
;
4910 bdrv_subtree_drained_end(top
);
4916 * Length of a allocated file in bytes. Sparse files are counted by actual
4917 * allocated space. Return < 0 if error or unknown.
4919 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4921 BlockDriver
*drv
= bs
->drv
;
4925 if (drv
->bdrv_get_allocated_file_size
) {
4926 return drv
->bdrv_get_allocated_file_size(bs
);
4929 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4936 * @drv: Format driver
4937 * @opts: Creation options for new image
4938 * @in_bs: Existing image containing data for new image (may be NULL)
4939 * @errp: Error object
4940 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4943 * Calculate file size required to create a new image.
4945 * If @in_bs is given then space for allocated clusters and zero clusters
4946 * from that image are included in the calculation. If @opts contains a
4947 * backing file that is shared by @in_bs then backing clusters may be omitted
4948 * from the calculation.
4950 * If @in_bs is NULL then the calculation includes no allocated clusters
4951 * unless a preallocation option is given in @opts.
4953 * Note that @in_bs may use a different BlockDriver from @drv.
4955 * If an error occurs the @errp pointer is set.
4957 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4958 BlockDriverState
*in_bs
, Error
**errp
)
4960 if (!drv
->bdrv_measure
) {
4961 error_setg(errp
, "Block driver '%s' does not support size measurement",
4966 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4970 * Return number of sectors on success, -errno on error.
4972 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4974 BlockDriver
*drv
= bs
->drv
;
4979 if (drv
->has_variable_length
) {
4980 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4985 return bs
->total_sectors
;
4989 * Return length in bytes on success, -errno on error.
4990 * The length is always a multiple of BDRV_SECTOR_SIZE.
4992 int64_t bdrv_getlength(BlockDriverState
*bs
)
4994 int64_t ret
= bdrv_nb_sectors(bs
);
4996 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4997 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
5000 /* return 0 as number of sectors if no device present or error */
5001 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5003 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5005 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5008 bool bdrv_is_sg(BlockDriverState
*bs
)
5013 bool bdrv_is_encrypted(BlockDriverState
*bs
)
5015 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
5018 return bs
->encrypted
;
5021 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5023 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5026 static int qsort_strcmp(const void *a
, const void *b
)
5028 return strcmp(*(char *const *)a
, *(char *const *)b
);
5031 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5032 void *opaque
, bool read_only
)
5037 const char **formats
= NULL
;
5039 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5040 if (drv
->format_name
) {
5044 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5048 while (formats
&& i
&& !found
) {
5049 found
= !strcmp(formats
[--i
], drv
->format_name
);
5053 formats
= g_renew(const char *, formats
, count
+ 1);
5054 formats
[count
++] = drv
->format_name
;
5059 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5060 const char *format_name
= block_driver_modules
[i
].format_name
;
5066 if (use_bdrv_whitelist
&&
5067 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5071 while (formats
&& j
&& !found
) {
5072 found
= !strcmp(formats
[--j
], format_name
);
5076 formats
= g_renew(const char *, formats
, count
+ 1);
5077 formats
[count
++] = format_name
;
5082 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5084 for (i
= 0; i
< count
; i
++) {
5085 it(opaque
, formats
[i
]);
5091 /* This function is to find a node in the bs graph */
5092 BlockDriverState
*bdrv_find_node(const char *node_name
)
5094 BlockDriverState
*bs
;
5098 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5099 if (!strcmp(node_name
, bs
->node_name
)) {
5106 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5107 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5110 BlockDeviceInfoList
*list
, *entry
;
5111 BlockDriverState
*bs
;
5114 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5115 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5117 qapi_free_BlockDeviceInfoList(list
);
5120 entry
= g_malloc0(sizeof(*entry
));
5121 entry
->value
= info
;
5129 #define QAPI_LIST_ADD(list, element) do { \
5130 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5131 _tmp->value = (element); \
5132 _tmp->next = (list); \
5136 typedef struct XDbgBlockGraphConstructor
{
5137 XDbgBlockGraph
*graph
;
5138 GHashTable
*graph_nodes
;
5139 } XDbgBlockGraphConstructor
;
5141 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5143 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5145 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5146 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5151 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5153 XDbgBlockGraph
*graph
= gr
->graph
;
5155 g_hash_table_destroy(gr
->graph_nodes
);
5161 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5163 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5170 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5171 * answer of g_hash_table_lookup.
5173 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5174 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5179 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5180 XDbgBlockGraphNodeType type
, const char *name
)
5182 XDbgBlockGraphNode
*n
;
5184 n
= g_new0(XDbgBlockGraphNode
, 1);
5186 n
->id
= xdbg_graph_node_num(gr
, node
);
5188 n
->name
= g_strdup(name
);
5190 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5193 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5194 const BdrvChild
*child
)
5196 BlockPermission qapi_perm
;
5197 XDbgBlockGraphEdge
*edge
;
5199 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5201 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5202 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5203 edge
->name
= g_strdup(child
->name
);
5205 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5206 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5208 if (flag
& child
->perm
) {
5209 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5211 if (flag
& child
->shared_perm
) {
5212 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5216 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5220 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5224 BlockDriverState
*bs
;
5226 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5228 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5229 char *allocated_name
= NULL
;
5230 const char *name
= blk_name(blk
);
5233 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5235 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5237 g_free(allocated_name
);
5238 if (blk_root(blk
)) {
5239 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5243 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5246 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5248 for (el
= job
->nodes
; el
; el
= el
->next
) {
5249 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5253 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5254 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5256 QLIST_FOREACH(child
, &bs
->children
, next
) {
5257 xdbg_graph_add_edge(gr
, bs
, child
);
5261 return xdbg_graph_finalize(gr
);
5264 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5265 const char *node_name
,
5269 BlockDriverState
*bs
;
5272 blk
= blk_by_name(device
);
5277 error_setg(errp
, "Device '%s' has no medium", device
);
5285 bs
= bdrv_find_node(node_name
);
5292 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5293 device
? device
: "",
5294 node_name
? node_name
: "");
5298 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5299 * return false. If either argument is NULL, return false. */
5300 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5302 while (top
&& top
!= base
) {
5303 top
= backing_bs(top
);
5309 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5312 return QTAILQ_FIRST(&graph_bdrv_states
);
5314 return QTAILQ_NEXT(bs
, node_list
);
5317 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5320 return QTAILQ_FIRST(&all_bdrv_states
);
5322 return QTAILQ_NEXT(bs
, bs_list
);
5325 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5327 return bs
->node_name
;
5330 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5335 /* If multiple parents have a name, just pick the first one. */
5336 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5337 if (c
->klass
->get_name
) {
5338 name
= c
->klass
->get_name(c
);
5339 if (name
&& *name
) {
5348 /* TODO check what callers really want: bs->node_name or blk_name() */
5349 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5351 return bdrv_get_parent_name(bs
) ?: "";
5354 /* This can be used to identify nodes that might not have a device
5355 * name associated. Since node and device names live in the same
5356 * namespace, the result is unambiguous. The exception is if both are
5357 * absent, then this returns an empty (non-null) string. */
5358 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5360 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5363 int bdrv_get_flags(BlockDriverState
*bs
)
5365 return bs
->open_flags
;
5368 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5373 int bdrv_has_zero_init(BlockDriverState
*bs
)
5379 /* If BS is a copy on write image, it is initialized to
5380 the contents of the base image, which may not be zeroes. */
5384 if (bs
->drv
->bdrv_has_zero_init
) {
5385 return bs
->drv
->bdrv_has_zero_init(bs
);
5387 if (bs
->file
&& bs
->drv
->is_filter
) {
5388 return bdrv_has_zero_init(bs
->file
->bs
);
5395 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5397 BlockDriverInfo bdi
;
5403 if (bdrv_get_info(bs
, &bdi
) == 0) {
5404 return bdi
.unallocated_blocks_are_zero
;
5410 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5412 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5416 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5419 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5420 char *filename
, int filename_size
)
5422 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5425 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5427 BlockDriver
*drv
= bs
->drv
;
5428 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5432 if (!drv
->bdrv_get_info
) {
5433 if (bs
->file
&& drv
->is_filter
) {
5434 return bdrv_get_info(bs
->file
->bs
, bdi
);
5438 memset(bdi
, 0, sizeof(*bdi
));
5439 return drv
->bdrv_get_info(bs
, bdi
);
5442 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5445 BlockDriver
*drv
= bs
->drv
;
5446 if (drv
&& drv
->bdrv_get_specific_info
) {
5447 return drv
->bdrv_get_specific_info(bs
, errp
);
5452 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5454 BlockDriver
*drv
= bs
->drv
;
5455 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5458 return drv
->bdrv_get_specific_stats(bs
);
5461 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5463 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5467 bs
->drv
->bdrv_debug_event(bs
, event
);
5470 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5472 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5478 if (bs
->drv
->is_filter
&& bs
->backing
) {
5479 bs
= bs
->backing
->bs
;
5486 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5487 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5494 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5497 bs
= bdrv_find_debug_node(bs
);
5499 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5505 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5507 bs
= bdrv_find_debug_node(bs
);
5509 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5515 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5517 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5518 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5521 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5522 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5528 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5530 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5531 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5534 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5535 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5541 /* backing_file can either be relative, or absolute, or a protocol. If it is
5542 * relative, it must be relative to the chain. So, passing in bs->filename
5543 * from a BDS as backing_file should not be done, as that may be relative to
5544 * the CWD rather than the chain. */
5545 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5546 const char *backing_file
)
5548 char *filename_full
= NULL
;
5549 char *backing_file_full
= NULL
;
5550 char *filename_tmp
= NULL
;
5551 int is_protocol
= 0;
5552 BlockDriverState
*curr_bs
= NULL
;
5553 BlockDriverState
*retval
= NULL
;
5555 if (!bs
|| !bs
->drv
|| !backing_file
) {
5559 filename_full
= g_malloc(PATH_MAX
);
5560 backing_file_full
= g_malloc(PATH_MAX
);
5562 is_protocol
= path_has_protocol(backing_file
);
5564 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5566 /* If either of the filename paths is actually a protocol, then
5567 * compare unmodified paths; otherwise make paths relative */
5568 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5569 char *backing_file_full_ret
;
5571 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5572 retval
= curr_bs
->backing
->bs
;
5575 /* Also check against the full backing filename for the image */
5576 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5578 if (backing_file_full_ret
) {
5579 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5580 g_free(backing_file_full_ret
);
5582 retval
= curr_bs
->backing
->bs
;
5587 /* If not an absolute filename path, make it relative to the current
5588 * image's filename path */
5589 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5591 /* We are going to compare canonicalized absolute pathnames */
5592 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5593 g_free(filename_tmp
);
5596 g_free(filename_tmp
);
5598 /* We need to make sure the backing filename we are comparing against
5599 * is relative to the current image filename (or absolute) */
5600 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5601 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5602 g_free(filename_tmp
);
5605 g_free(filename_tmp
);
5607 if (strcmp(backing_file_full
, filename_full
) == 0) {
5608 retval
= curr_bs
->backing
->bs
;
5614 g_free(filename_full
);
5615 g_free(backing_file_full
);
5619 void bdrv_init(void)
5621 module_call_init(MODULE_INIT_BLOCK
);
5624 void bdrv_init_with_whitelist(void)
5626 use_bdrv_whitelist
= 1;
5630 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5633 BdrvChild
*child
, *parent
;
5634 uint64_t perm
, shared_perm
;
5635 Error
*local_err
= NULL
;
5637 BdrvDirtyBitmap
*bm
;
5643 QLIST_FOREACH(child
, &bs
->children
, next
) {
5644 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5646 error_propagate(errp
, local_err
);
5652 * Update permissions, they may differ for inactive nodes.
5654 * Note that the required permissions of inactive images are always a
5655 * subset of the permissions required after activating the image. This
5656 * allows us to just get the permissions upfront without restricting
5657 * drv->bdrv_invalidate_cache().
5659 * It also means that in error cases, we don't have to try and revert to
5660 * the old permissions (which is an operation that could fail, too). We can
5661 * just keep the extended permissions for the next time that an activation
5662 * of the image is tried.
5664 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5665 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5666 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5667 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5669 bs
->open_flags
|= BDRV_O_INACTIVE
;
5670 error_propagate(errp
, local_err
);
5673 bdrv_set_perm(bs
, perm
, shared_perm
);
5675 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5676 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5678 bs
->open_flags
|= BDRV_O_INACTIVE
;
5679 error_propagate(errp
, local_err
);
5684 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5685 bdrv_dirty_bitmap_skip_store(bm
, false);
5688 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5690 bs
->open_flags
|= BDRV_O_INACTIVE
;
5691 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5696 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5697 if (parent
->klass
->activate
) {
5698 parent
->klass
->activate(parent
, &local_err
);
5700 bs
->open_flags
|= BDRV_O_INACTIVE
;
5701 error_propagate(errp
, local_err
);
5708 typedef struct InvalidateCacheCo
{
5709 BlockDriverState
*bs
;
5712 } InvalidateCacheCo
;
5714 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5716 InvalidateCacheCo
*ico
= opaque
;
5717 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5722 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5725 InvalidateCacheCo ico
= {
5731 if (qemu_in_coroutine()) {
5732 /* Fast-path if already in coroutine context */
5733 bdrv_invalidate_cache_co_entry(&ico
);
5735 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5736 bdrv_coroutine_enter(bs
, co
);
5737 BDRV_POLL_WHILE(bs
, !ico
.done
);
5741 void bdrv_invalidate_cache_all(Error
**errp
)
5743 BlockDriverState
*bs
;
5744 Error
*local_err
= NULL
;
5745 BdrvNextIterator it
;
5747 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5748 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5750 aio_context_acquire(aio_context
);
5751 bdrv_invalidate_cache(bs
, &local_err
);
5752 aio_context_release(aio_context
);
5754 error_propagate(errp
, local_err
);
5755 bdrv_next_cleanup(&it
);
5761 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5765 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5766 if (parent
->klass
->parent_is_bds
) {
5767 BlockDriverState
*parent_bs
= parent
->opaque
;
5768 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5777 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5779 BdrvChild
*child
, *parent
;
5780 bool tighten_restrictions
;
5781 uint64_t perm
, shared_perm
;
5788 /* Make sure that we don't inactivate a child before its parent.
5789 * It will be covered by recursion from the yet active parent. */
5790 if (bdrv_has_bds_parent(bs
, true)) {
5794 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5796 /* Inactivate this node */
5797 if (bs
->drv
->bdrv_inactivate
) {
5798 ret
= bs
->drv
->bdrv_inactivate(bs
);
5804 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5805 if (parent
->klass
->inactivate
) {
5806 ret
= parent
->klass
->inactivate(parent
);
5813 bs
->open_flags
|= BDRV_O_INACTIVE
;
5815 /* Update permissions, they may differ for inactive nodes */
5816 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5817 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5818 &tighten_restrictions
, NULL
);
5819 assert(tighten_restrictions
== false);
5821 /* We only tried to loosen restrictions, so errors are not fatal */
5822 bdrv_abort_perm_update(bs
);
5824 bdrv_set_perm(bs
, perm
, shared_perm
);
5828 /* Recursively inactivate children */
5829 QLIST_FOREACH(child
, &bs
->children
, next
) {
5830 ret
= bdrv_inactivate_recurse(child
->bs
);
5839 int bdrv_inactivate_all(void)
5841 BlockDriverState
*bs
= NULL
;
5842 BdrvNextIterator it
;
5844 GSList
*aio_ctxs
= NULL
, *ctx
;
5846 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5847 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5849 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5850 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5851 aio_context_acquire(aio_context
);
5855 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5856 /* Nodes with BDS parents are covered by recursion from the last
5857 * parent that gets inactivated. Don't inactivate them a second
5858 * time if that has already happened. */
5859 if (bdrv_has_bds_parent(bs
, false)) {
5862 ret
= bdrv_inactivate_recurse(bs
);
5864 bdrv_next_cleanup(&it
);
5870 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5871 AioContext
*aio_context
= ctx
->data
;
5872 aio_context_release(aio_context
);
5874 g_slist_free(aio_ctxs
);
5879 /**************************************************************/
5880 /* removable device support */
5883 * Return TRUE if the media is present
5885 bool bdrv_is_inserted(BlockDriverState
*bs
)
5887 BlockDriver
*drv
= bs
->drv
;
5893 if (drv
->bdrv_is_inserted
) {
5894 return drv
->bdrv_is_inserted(bs
);
5896 QLIST_FOREACH(child
, &bs
->children
, next
) {
5897 if (!bdrv_is_inserted(child
->bs
)) {
5905 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5907 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5909 BlockDriver
*drv
= bs
->drv
;
5911 if (drv
&& drv
->bdrv_eject
) {
5912 drv
->bdrv_eject(bs
, eject_flag
);
5917 * Lock or unlock the media (if it is locked, the user won't be able
5918 * to eject it manually).
5920 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5922 BlockDriver
*drv
= bs
->drv
;
5924 trace_bdrv_lock_medium(bs
, locked
);
5926 if (drv
&& drv
->bdrv_lock_medium
) {
5927 drv
->bdrv_lock_medium(bs
, locked
);
5931 /* Get a reference to bs */
5932 void bdrv_ref(BlockDriverState
*bs
)
5937 /* Release a previously grabbed reference to bs.
5938 * If after releasing, reference count is zero, the BlockDriverState is
5940 void bdrv_unref(BlockDriverState
*bs
)
5945 assert(bs
->refcnt
> 0);
5946 if (--bs
->refcnt
== 0) {
5951 struct BdrvOpBlocker
{
5953 QLIST_ENTRY(BdrvOpBlocker
) list
;
5956 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5958 BdrvOpBlocker
*blocker
;
5959 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5960 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5961 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5962 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5963 "Node '%s' is busy: ",
5964 bdrv_get_device_or_node_name(bs
));
5970 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5972 BdrvOpBlocker
*blocker
;
5973 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5975 blocker
= g_new0(BdrvOpBlocker
, 1);
5976 blocker
->reason
= reason
;
5977 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5980 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5982 BdrvOpBlocker
*blocker
, *next
;
5983 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5984 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5985 if (blocker
->reason
== reason
) {
5986 QLIST_REMOVE(blocker
, list
);
5992 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5995 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5996 bdrv_op_block(bs
, i
, reason
);
6000 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6003 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6004 bdrv_op_unblock(bs
, i
, reason
);
6008 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6012 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6013 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6020 void bdrv_img_create(const char *filename
, const char *fmt
,
6021 const char *base_filename
, const char *base_fmt
,
6022 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6025 QemuOptsList
*create_opts
= NULL
;
6026 QemuOpts
*opts
= NULL
;
6027 const char *backing_fmt
, *backing_file
;
6029 BlockDriver
*drv
, *proto_drv
;
6030 Error
*local_err
= NULL
;
6033 /* Find driver and parse its options */
6034 drv
= bdrv_find_format(fmt
);
6036 error_setg(errp
, "Unknown file format '%s'", fmt
);
6040 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6045 if (!drv
->create_opts
) {
6046 error_setg(errp
, "Format driver '%s' does not support image creation",
6051 if (!proto_drv
->create_opts
) {
6052 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6053 proto_drv
->format_name
);
6057 /* Create parameter list */
6058 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6059 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6061 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6063 /* Parse -o options */
6065 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
6071 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6072 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6073 } else if (img_size
!= UINT64_C(-1)) {
6074 error_setg(errp
, "The image size must be specified only once");
6078 if (base_filename
) {
6079 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
6081 error_setg(errp
, "Backing file not supported for file format '%s'",
6088 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
6090 error_setg(errp
, "Backing file format not supported for file "
6091 "format '%s'", fmt
);
6096 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6098 if (!strcmp(filename
, backing_file
)) {
6099 error_setg(errp
, "Error: Trying to create an image with the "
6100 "same filename as the backing file");
6105 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6107 /* The size for the image must always be specified, unless we have a backing
6108 * file and we have not been forbidden from opening it. */
6109 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6110 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6111 BlockDriverState
*bs
;
6114 QDict
*backing_options
= NULL
;
6117 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6122 assert(full_backing
);
6124 /* backing files always opened read-only */
6126 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6128 backing_options
= qdict_new();
6130 qdict_put_str(backing_options
, "driver", backing_fmt
);
6132 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6134 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6136 g_free(full_backing
);
6137 if (!bs
&& size
!= -1) {
6138 /* Couldn't open BS, but we have a size, so it's nonfatal */
6139 warn_reportf_err(local_err
,
6140 "Could not verify backing image. "
6141 "This may become an error in future versions.\n");
6144 /* Couldn't open bs, do not have size */
6145 error_append_hint(&local_err
,
6146 "Could not open backing image to determine size.\n");
6150 /* Opened BS, have no size */
6151 size
= bdrv_getlength(bs
);
6153 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6158 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6162 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6165 error_setg(errp
, "Image creation needs a size parameter");
6170 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6171 qemu_opts_print(opts
, " ");
6175 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6177 if (ret
== -EFBIG
) {
6178 /* This is generally a better message than whatever the driver would
6179 * deliver (especially because of the cluster_size_hint), since that
6180 * is most probably not much different from "image too large". */
6181 const char *cluster_size_hint
= "";
6182 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6183 cluster_size_hint
= " (try using a larger cluster size)";
6185 error_setg(errp
, "The image size is too large for file format '%s'"
6186 "%s", fmt
, cluster_size_hint
);
6187 error_free(local_err
);
6192 qemu_opts_del(opts
);
6193 qemu_opts_free(create_opts
);
6194 error_propagate(errp
, local_err
);
6197 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6199 return bs
? bs
->aio_context
: qemu_get_aio_context();
6202 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6204 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6207 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6209 QLIST_REMOVE(ban
, list
);
6213 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6215 BdrvAioNotifier
*baf
, *baf_tmp
;
6217 assert(!bs
->walking_aio_notifiers
);
6218 bs
->walking_aio_notifiers
= true;
6219 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6221 bdrv_do_remove_aio_context_notifier(baf
);
6223 baf
->detach_aio_context(baf
->opaque
);
6226 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6227 * remove remaining aio notifiers if we aren't called again.
6229 bs
->walking_aio_notifiers
= false;
6231 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6232 bs
->drv
->bdrv_detach_aio_context(bs
);
6235 if (bs
->quiesce_counter
) {
6236 aio_enable_external(bs
->aio_context
);
6238 bs
->aio_context
= NULL
;
6241 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6242 AioContext
*new_context
)
6244 BdrvAioNotifier
*ban
, *ban_tmp
;
6246 if (bs
->quiesce_counter
) {
6247 aio_disable_external(new_context
);
6250 bs
->aio_context
= new_context
;
6252 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6253 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6256 assert(!bs
->walking_aio_notifiers
);
6257 bs
->walking_aio_notifiers
= true;
6258 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6260 bdrv_do_remove_aio_context_notifier(ban
);
6262 ban
->attached_aio_context(new_context
, ban
->opaque
);
6265 bs
->walking_aio_notifiers
= false;
6269 * Changes the AioContext used for fd handlers, timers, and BHs by this
6270 * BlockDriverState and all its children and parents.
6272 * Must be called from the main AioContext.
6274 * The caller must own the AioContext lock for the old AioContext of bs, but it
6275 * must not own the AioContext lock for new_context (unless new_context is the
6276 * same as the current context of bs).
6278 * @ignore will accumulate all visited BdrvChild object. The caller is
6279 * responsible for freeing the list afterwards.
6281 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6282 AioContext
*new_context
, GSList
**ignore
)
6284 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6287 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6289 if (old_context
== new_context
) {
6293 bdrv_drained_begin(bs
);
6295 QLIST_FOREACH(child
, &bs
->children
, next
) {
6296 if (g_slist_find(*ignore
, child
)) {
6299 *ignore
= g_slist_prepend(*ignore
, child
);
6300 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6302 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6303 if (g_slist_find(*ignore
, child
)) {
6306 assert(child
->klass
->set_aio_ctx
);
6307 *ignore
= g_slist_prepend(*ignore
, child
);
6308 child
->klass
->set_aio_ctx(child
, new_context
, ignore
);
6311 bdrv_detach_aio_context(bs
);
6313 /* Acquire the new context, if necessary */
6314 if (qemu_get_aio_context() != new_context
) {
6315 aio_context_acquire(new_context
);
6318 bdrv_attach_aio_context(bs
, new_context
);
6321 * If this function was recursively called from
6322 * bdrv_set_aio_context_ignore(), there may be nodes in the
6323 * subtree that have not yet been moved to the new AioContext.
6324 * Release the old one so bdrv_drained_end() can poll them.
6326 if (qemu_get_aio_context() != old_context
) {
6327 aio_context_release(old_context
);
6330 bdrv_drained_end(bs
);
6332 if (qemu_get_aio_context() != old_context
) {
6333 aio_context_acquire(old_context
);
6335 if (qemu_get_aio_context() != new_context
) {
6336 aio_context_release(new_context
);
6340 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6341 GSList
**ignore
, Error
**errp
)
6343 if (g_slist_find(*ignore
, c
)) {
6346 *ignore
= g_slist_prepend(*ignore
, c
);
6349 * A BdrvChildClass that doesn't handle AioContext changes cannot
6350 * tolerate any AioContext changes
6352 if (!c
->klass
->can_set_aio_ctx
) {
6353 char *user
= bdrv_child_user_desc(c
);
6354 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6358 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6359 assert(!errp
|| *errp
);
6365 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6366 GSList
**ignore
, Error
**errp
)
6368 if (g_slist_find(*ignore
, c
)) {
6371 *ignore
= g_slist_prepend(*ignore
, c
);
6372 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6375 /* @ignore will accumulate all visited BdrvChild object. The caller is
6376 * responsible for freeing the list afterwards. */
6377 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6378 GSList
**ignore
, Error
**errp
)
6382 if (bdrv_get_aio_context(bs
) == ctx
) {
6386 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6387 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6391 QLIST_FOREACH(c
, &bs
->children
, next
) {
6392 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6400 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6401 BdrvChild
*ignore_child
, Error
**errp
)
6406 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6407 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6408 g_slist_free(ignore
);
6414 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6415 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6416 g_slist_free(ignore
);
6421 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6424 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6427 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6428 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6429 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6431 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6432 *ban
= (BdrvAioNotifier
){
6433 .attached_aio_context
= attached_aio_context
,
6434 .detach_aio_context
= detach_aio_context
,
6438 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6441 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6442 void (*attached_aio_context
)(AioContext
*,
6444 void (*detach_aio_context
)(void *),
6447 BdrvAioNotifier
*ban
, *ban_next
;
6449 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6450 if (ban
->attached_aio_context
== attached_aio_context
&&
6451 ban
->detach_aio_context
== detach_aio_context
&&
6452 ban
->opaque
== opaque
&&
6453 ban
->deleted
== false)
6455 if (bs
->walking_aio_notifiers
) {
6456 ban
->deleted
= true;
6458 bdrv_do_remove_aio_context_notifier(ban
);
6467 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6468 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6472 error_setg(errp
, "Node is ejected");
6475 if (!bs
->drv
->bdrv_amend_options
) {
6476 error_setg(errp
, "Block driver '%s' does not support option amendment",
6477 bs
->drv
->format_name
);
6480 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6484 * This function checks whether the given @to_replace is allowed to be
6485 * replaced by a node that always shows the same data as @bs. This is
6486 * used for example to verify whether the mirror job can replace
6487 * @to_replace by the target mirrored from @bs.
6488 * To be replaceable, @bs and @to_replace may either be guaranteed to
6489 * always show the same data (because they are only connected through
6490 * filters), or some driver may allow replacing one of its children
6491 * because it can guarantee that this child's data is not visible at
6492 * all (for example, for dissenting quorum children that have no other
6495 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6496 BlockDriverState
*to_replace
)
6498 if (!bs
|| !bs
->drv
) {
6502 if (bs
== to_replace
) {
6506 /* See what the driver can do */
6507 if (bs
->drv
->bdrv_recurse_can_replace
) {
6508 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6511 /* For filters without an own implementation, we can recurse on our own */
6512 if (bs
->drv
->is_filter
) {
6513 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6514 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6522 * Check whether the given @node_name can be replaced by a node that
6523 * has the same data as @parent_bs. If so, return @node_name's BDS;
6526 * @node_name must be a (recursive) *child of @parent_bs (or this
6527 * function will return NULL).
6529 * The result (whether the node can be replaced or not) is only valid
6530 * for as long as no graph or permission changes occur.
6532 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6533 const char *node_name
, Error
**errp
)
6535 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6536 AioContext
*aio_context
;
6538 if (!to_replace_bs
) {
6539 error_setg(errp
, "Node name '%s' not found", node_name
);
6543 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6544 aio_context_acquire(aio_context
);
6546 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6547 to_replace_bs
= NULL
;
6551 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6552 * most non filter in order to prevent data corruption.
6553 * Another benefit is that this tests exclude backing files which are
6554 * blocked by the backing blockers.
6556 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6557 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6558 "because it cannot be guaranteed that doing so would not "
6559 "lead to an abrupt change of visible data",
6560 node_name
, parent_bs
->node_name
);
6561 to_replace_bs
= NULL
;
6566 aio_context_release(aio_context
);
6567 return to_replace_bs
;
6571 * Iterates through the list of runtime option keys that are said to
6572 * be "strong" for a BDS. An option is called "strong" if it changes
6573 * a BDS's data. For example, the null block driver's "size" and
6574 * "read-zeroes" options are strong, but its "latency-ns" option is
6577 * If a key returned by this function ends with a dot, all options
6578 * starting with that prefix are strong.
6580 static const char *const *strong_options(BlockDriverState
*bs
,
6581 const char *const *curopt
)
6583 static const char *const global_options
[] = {
6584 "driver", "filename", NULL
6588 return &global_options
[0];
6592 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6593 curopt
= bs
->drv
->strong_runtime_opts
;
6596 return (curopt
&& *curopt
) ? curopt
: NULL
;
6600 * Copies all strong runtime options from bs->options to the given
6601 * QDict. The set of strong option keys is determined by invoking
6604 * Returns true iff any strong option was present in bs->options (and
6605 * thus copied to the target QDict) with the exception of "filename"
6606 * and "driver". The caller is expected to use this value to decide
6607 * whether the existence of strong options prevents the generation of
6610 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6612 bool found_any
= false;
6613 const char *const *option_name
= NULL
;
6619 while ((option_name
= strong_options(bs
, option_name
))) {
6620 bool option_given
= false;
6622 assert(strlen(*option_name
) > 0);
6623 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6624 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6629 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6630 option_given
= true;
6632 const QDictEntry
*entry
;
6633 for (entry
= qdict_first(bs
->options
); entry
;
6634 entry
= qdict_next(bs
->options
, entry
))
6636 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6637 qdict_put_obj(d
, qdict_entry_key(entry
),
6638 qobject_ref(qdict_entry_value(entry
)));
6639 option_given
= true;
6644 /* While "driver" and "filename" need to be included in a JSON filename,
6645 * their existence does not prohibit generation of a plain filename. */
6646 if (!found_any
&& option_given
&&
6647 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6653 if (!qdict_haskey(d
, "driver")) {
6654 /* Drivers created with bdrv_new_open_driver() may not have a
6655 * @driver option. Add it here. */
6656 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6662 /* Note: This function may return false positives; it may return true
6663 * even if opening the backing file specified by bs's image header
6664 * would result in exactly bs->backing. */
6665 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6668 return strcmp(bs
->auto_backing_file
,
6669 bs
->backing
->bs
->filename
);
6671 /* No backing BDS, so if the image header reports any backing
6672 * file, it must have been suppressed */
6673 return bs
->auto_backing_file
[0] != '\0';
6677 /* Updates the following BDS fields:
6678 * - exact_filename: A filename which may be used for opening a block device
6679 * which (mostly) equals the given BDS (even without any
6680 * other options; so reading and writing must return the same
6681 * results, but caching etc. may be different)
6682 * - full_open_options: Options which, when given when opening a block device
6683 * (without a filename), result in a BDS (mostly)
6684 * equalling the given one
6685 * - filename: If exact_filename is set, it is copied here. Otherwise,
6686 * full_open_options is converted to a JSON object, prefixed with
6687 * "json:" (for use through the JSON pseudo protocol) and put here.
6689 void bdrv_refresh_filename(BlockDriverState
*bs
)
6691 BlockDriver
*drv
= bs
->drv
;
6694 bool backing_overridden
;
6695 bool generate_json_filename
; /* Whether our default implementation should
6696 fill exact_filename (false) or not (true) */
6702 /* This BDS's file name may depend on any of its children's file names, so
6703 * refresh those first */
6704 QLIST_FOREACH(child
, &bs
->children
, next
) {
6705 bdrv_refresh_filename(child
->bs
);
6709 /* For implicit nodes, just copy everything from the single child */
6710 child
= QLIST_FIRST(&bs
->children
);
6711 assert(QLIST_NEXT(child
, next
) == NULL
);
6713 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6714 child
->bs
->exact_filename
);
6715 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6717 qobject_unref(bs
->full_open_options
);
6718 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6723 backing_overridden
= bdrv_backing_overridden(bs
);
6725 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6726 /* Without I/O, the backing file does not change anything.
6727 * Therefore, in such a case (primarily qemu-img), we can
6728 * pretend the backing file has not been overridden even if
6729 * it technically has been. */
6730 backing_overridden
= false;
6733 /* Gather the options QDict */
6735 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6736 generate_json_filename
|= backing_overridden
;
6738 if (drv
->bdrv_gather_child_options
) {
6739 /* Some block drivers may not want to present all of their children's
6740 * options, or name them differently from BdrvChild.name */
6741 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6743 QLIST_FOREACH(child
, &bs
->children
, next
) {
6744 if (child
->klass
== &child_backing
&& !backing_overridden
) {
6745 /* We can skip the backing BDS if it has not been overridden */
6749 qdict_put(opts
, child
->name
,
6750 qobject_ref(child
->bs
->full_open_options
));
6753 if (backing_overridden
&& !bs
->backing
) {
6754 /* Force no backing file */
6755 qdict_put_null(opts
, "backing");
6759 qobject_unref(bs
->full_open_options
);
6760 bs
->full_open_options
= opts
;
6762 if (drv
->bdrv_refresh_filename
) {
6763 /* Obsolete information is of no use here, so drop the old file name
6764 * information before refreshing it */
6765 bs
->exact_filename
[0] = '\0';
6767 drv
->bdrv_refresh_filename(bs
);
6768 } else if (bs
->file
) {
6769 /* Try to reconstruct valid information from the underlying file */
6771 bs
->exact_filename
[0] = '\0';
6774 * We can use the underlying file's filename if:
6775 * - it has a filename,
6776 * - the file is a protocol BDS, and
6777 * - opening that file (as this BDS's format) will automatically create
6778 * the BDS tree we have right now, that is:
6779 * - the user did not significantly change this BDS's behavior with
6780 * some explicit (strong) options
6781 * - no non-file child of this BDS has been overridden by the user
6782 * Both of these conditions are represented by generate_json_filename.
6784 if (bs
->file
->bs
->exact_filename
[0] &&
6785 bs
->file
->bs
->drv
->bdrv_file_open
&&
6786 !generate_json_filename
)
6788 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6792 if (bs
->exact_filename
[0]) {
6793 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6795 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6796 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6797 qstring_get_str(json
));
6798 qobject_unref(json
);
6802 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6804 BlockDriver
*drv
= bs
->drv
;
6807 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6811 if (drv
->bdrv_dirname
) {
6812 return drv
->bdrv_dirname(bs
, errp
);
6816 return bdrv_dirname(bs
->file
->bs
, errp
);
6819 bdrv_refresh_filename(bs
);
6820 if (bs
->exact_filename
[0] != '\0') {
6821 return path_combine(bs
->exact_filename
, "");
6824 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6830 * Hot add/remove a BDS's child. So the user can take a child offline when
6831 * it is broken and take a new child online
6833 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6837 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6838 error_setg(errp
, "The node %s does not support adding a child",
6839 bdrv_get_device_or_node_name(parent_bs
));
6843 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6844 error_setg(errp
, "The node %s already has a parent",
6845 child_bs
->node_name
);
6849 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6852 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6856 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6857 error_setg(errp
, "The node %s does not support removing a child",
6858 bdrv_get_device_or_node_name(parent_bs
));
6862 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6869 error_setg(errp
, "The node %s does not have a child named %s",
6870 bdrv_get_device_or_node_name(parent_bs
),
6871 bdrv_get_device_or_node_name(child
->bs
));
6875 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6878 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
6880 BlockDriver
*drv
= c
->bs
->drv
;
6883 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
6885 if (!drv
->bdrv_make_empty
) {
6886 error_setg(errp
, "%s does not support emptying nodes",
6891 ret
= drv
->bdrv_make_empty(c
->bs
);
6893 error_setg_errno(errp
, -ret
, "Failed to empty %s",