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
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
45 #include "qapi/util.h"
48 #include <sys/ioctl.h>
49 #include <sys/queue.h>
59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
61 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
64 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
65 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
67 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
68 QLIST_HEAD_INITIALIZER(bdrv_drivers);
70 static BlockDriverState *bdrv_open_inherit(const char *filename,
71 const char *reference,
72 QDict *options, int flags,
73 BlockDriverState *parent,
74 const BdrvChildRole *child_role,
77 /* If non-zero, use only whitelisted block drivers */
78 static int use_bdrv_whitelist;
81 static int is_windows_drive_prefix(const char *filename)
83 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
84 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
88 int is_windows_drive(const char *filename)
90 if (is_windows_drive_prefix(filename) &&
93 if (strstart(filename, "\\\\.\\", NULL) ||
94 strstart(filename, "//./", NULL))
100 size_t bdrv_opt_mem_align(BlockDriverState *bs)
102 if (!bs || !bs->drv) {
103 /* page size or 4k (hdd sector size) should be on the safe side */
104 return MAX(4096, getpagesize());
107 return bs->bl.opt_mem_alignment;
110 size_t bdrv_min_mem_align(BlockDriverState *bs)
112 if (!bs || !bs->drv) {
113 /* page size or 4k (hdd sector size) should be on the safe side */
114 return MAX(4096, getpagesize());
117 return bs->bl.min_mem_alignment;
120 /* check if the path starts with "<protocol>:" */
121 int path_has_protocol(const char *path)
126 if (is_windows_drive(path) ||
127 is_windows_drive_prefix(path)) {
130 p = path + strcspn(path, ":/\\");
132 p = path + strcspn(path, ":/");
138 int path_is_absolute(const char *path)
141 /* specific case for names like: "\\.\d:" */
142 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
145 return (*path == '/' || *path == '\\');
147 return (*path == '/');
151 /* if filename is absolute, just copy it to dest. Otherwise, build a
152 path to it by considering it is relative to base_path. URL are
154 void path_combine(char *dest, int dest_size,
155 const char *base_path,
156 const char *filename)
163 if (path_is_absolute(filename)) {
164 pstrcpy(dest, dest_size, filename);
166 p = strchr(base_path, ':');
171 p1 = strrchr(base_path, '/');
175 p2 = strrchr(base_path, '\\');
187 if (len > dest_size - 1)
189 memcpy(dest, base_path, len);
191 pstrcat(dest, dest_size, filename);
195 bool bdrv_is_read_only(BlockDriverState *bs)
197 return bs->read_only;
200 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
202 /* Do not set read_only if copy_on_read is enabled */
203 if (bs->copy_on_read && read_only) {
204 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
205 bdrv_get_device_or_node_name(bs));
209 /* Do not clear read_only if it is prohibited */
210 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR)) {
211 error_setg(errp, "Node '%s' is read only",
212 bdrv_get_device_or_node_name(bs));
219 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
223 ret = bdrv_can_set_read_only(bs, read_only, errp);
228 bs->read_only = read_only;
232 void bdrv_get_full_backing_filename_from_filename(const char *backed,
234 char *dest, size_t sz,
237 if (backing[0] == '\0' || path_has_protocol(backing) ||
238 path_is_absolute(backing))
240 pstrcpy(dest, sz, backing);
241 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
242 error_setg(errp, "Cannot use relative backing file names for '%s'",
245 path_combine(dest, sz, backed, backing);
249 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
252 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
254 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
258 void bdrv_register(BlockDriver *bdrv)
260 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
263 BlockDriverState *bdrv_new(void)
265 BlockDriverState *bs;
268 bs = g_new0(BlockDriverState, 1);
269 QLIST_INIT(&bs->dirty_bitmaps);
270 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
271 QLIST_INIT(&bs->op_blockers[i]);
273 notifier_with_return_list_init(&bs->before_write_notifiers);
275 bs->aio_context = qemu_get_aio_context();
277 qemu_co_queue_init(&bs->flush_queue);
279 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
284 static BlockDriver *bdrv_do_find_format(const char *format_name)
288 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
289 if (!strcmp(drv1->format_name, format_name)) {
297 BlockDriver *bdrv_find_format(const char *format_name)
302 drv1 = bdrv_do_find_format(format_name);
307 /* The driver isn't registered, maybe we need to load a module */
308 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
309 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
310 block_module_load_one(block_driver_modules[i].library_name);
315 return bdrv_do_find_format(format_name);
318 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
320 static const char *whitelist_rw[] = {
321 CONFIG_BDRV_RW_WHITELIST
323 static const char *whitelist_ro[] = {
324 CONFIG_BDRV_RO_WHITELIST
328 if (!whitelist_rw[0] && !whitelist_ro[0]) {
329 return 1; /* no whitelist, anything goes */
332 for (p = whitelist_rw; *p; p++) {
333 if (!strcmp(drv->format_name, *p)) {
338 for (p = whitelist_ro; *p; p++) {
339 if (!strcmp(drv->format_name, *p)) {
347 bool bdrv_uses_whitelist(void)
349 return use_bdrv_whitelist;
352 typedef struct CreateCo {
360 static void coroutine_fn bdrv_create_co_entry(void *opaque)
362 Error *local_err = NULL;
365 CreateCo *cco = opaque;
368 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
369 error_propagate(&cco->err, local_err);
373 int bdrv_create(BlockDriver *drv, const char* filename,
374 QemuOpts *opts, Error **errp)
381 .filename = g_strdup(filename),
387 if (!drv->bdrv_create) {
388 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
393 if (qemu_in_coroutine()) {
394 /* Fast-path if already in coroutine context */
395 bdrv_create_co_entry(&cco);
397 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
398 qemu_coroutine_enter(co);
399 while (cco.ret == NOT_DONE) {
400 aio_poll(qemu_get_aio_context(), true);
407 error_propagate(errp, cco.err);
409 error_setg_errno(errp, -ret, "Could not create image");
414 g_free(cco.filename);
418 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
421 Error *local_err = NULL;
424 drv = bdrv_find_protocol(filename, true, errp);
429 ret = bdrv_create(drv, filename, opts, &local_err);
430 error_propagate(errp, local_err);
435 * Try to get @bs's logical and physical block size.
436 * On success, store them in @bsz struct and return 0.
437 * On failure return -errno.
438 * @bs must not be empty.
440 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
442 BlockDriver *drv = bs->drv;
444 if (drv && drv->bdrv_probe_blocksizes) {
445 return drv->bdrv_probe_blocksizes(bs, bsz);
452 * Try to get @bs's geometry (cyls, heads, sectors).
453 * On success, store them in @geo struct and return 0.
454 * On failure return -errno.
455 * @bs must not be empty.
457 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
459 BlockDriver *drv = bs->drv;
461 if (drv && drv->bdrv_probe_geometry) {
462 return drv->bdrv_probe_geometry(bs, geo);
469 * Create a uniquely-named empty temporary file.
470 * Return 0 upon success, otherwise a negative errno value.
472 int get_tmp_filename(char *filename, int size)
475 char temp_dir[MAX_PATH];
476 /* GetTempFileName requires that its output buffer (4th param)
477 have length MAX_PATH or greater. */
478 assert(size >= MAX_PATH);
479 return (GetTempPath(MAX_PATH, temp_dir)
480 && GetTempFileName(temp_dir, "qem", 0, filename)
481 ? 0 : -GetLastError());
485 tmpdir = getenv("TMPDIR");
489 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
492 fd = mkstemp(filename);
496 if (close(fd) != 0) {
505 * Detect host devices. By convention, /dev/cdrom[N] is always
506 * recognized as a host CDROM.
508 static BlockDriver *find_hdev_driver(const char *filename)
510 int score_max = 0, score;
511 BlockDriver *drv = NULL, *d;
513 QLIST_FOREACH(d, &bdrv_drivers, list) {
514 if (d->bdrv_probe_device) {
515 score = d->bdrv_probe_device(filename);
516 if (score > score_max) {
526 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
530 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
531 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
539 BlockDriver *bdrv_find_protocol(const char *filename,
540 bool allow_protocol_prefix,
549 /* TODO Drivers without bdrv_file_open must be specified explicitly */
552 * XXX(hch): we really should not let host device detection
553 * override an explicit protocol specification, but moving this
554 * later breaks access to device names with colons in them.
555 * Thanks to the brain-dead persistent naming schemes on udev-
556 * based Linux systems those actually are quite common.
558 drv1 = find_hdev_driver(filename);
563 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
567 p = strchr(filename, ':');
570 if (len > sizeof(protocol) - 1)
571 len = sizeof(protocol) - 1;
572 memcpy(protocol, filename, len);
573 protocol[len] = '\0';
575 drv1 = bdrv_do_find_protocol(protocol);
580 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
581 if (block_driver_modules[i].protocol_name &&
582 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
583 block_module_load_one(block_driver_modules[i].library_name);
588 drv1 = bdrv_do_find_protocol(protocol);
590 error_setg(errp, "Unknown protocol '%s'", protocol);
596 * Guess image format by probing its contents.
597 * This is not a good idea when your image is raw (CVE-2008-2004), but
598 * we do it anyway for backward compatibility.
600 * @buf contains the image's first @buf_size bytes.
601 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
602 * but can be smaller if the image file is smaller)
603 * @filename is its filename.
605 * For all block drivers, call the bdrv_probe() method to get its
607 * Return the first block driver with the highest probing score.
609 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
610 const char *filename)
612 int score_max = 0, score;
613 BlockDriver *drv = NULL, *d;
615 QLIST_FOREACH(d, &bdrv_drivers, list) {
617 score = d->bdrv_probe(buf, buf_size, filename);
618 if (score > score_max) {
628 static int find_image_format(BlockBackend *file, const char *filename,
629 BlockDriver **pdrv, Error **errp)
632 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
635 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
636 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
641 ret = blk_pread(file, 0, buf, sizeof(buf));
643 error_setg_errno(errp, -ret, "Could not read image for determining its "
649 drv = bdrv_probe_all(buf, ret, filename);
651 error_setg(errp, "Could not determine image format: No compatible "
660 * Set the current 'total_sectors' value
661 * Return 0 on success, -errno on error.
663 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
665 BlockDriver *drv = bs->drv;
667 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
671 /* query actual device if possible, otherwise just trust the hint */
672 if (drv->bdrv_getlength) {
673 int64_t length = drv->bdrv_getlength(bs);
677 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
680 bs->total_sectors = hint;
685 * Combines a QDict of new block driver @options with any missing options taken
686 * from @old_options, so that leaving out an option defaults to its old value.
688 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
691 if (bs->drv && bs->drv->bdrv_join_options) {
692 bs->drv->bdrv_join_options(options, old_options);
694 qdict_join(options, old_options, false);
699 * Set open flags for a given discard mode
701 * Return 0 on success, -1 if the discard mode was invalid.
703 int bdrv_parse_discard_flags(const char *mode, int *flags)
705 *flags &= ~BDRV_O_UNMAP;
707 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
709 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
710 *flags |= BDRV_O_UNMAP;
719 * Set open flags for a given cache mode
721 * Return 0 on success, -1 if the cache mode was invalid.
723 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
725 *flags &= ~BDRV_O_CACHE_MASK;
727 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
728 *writethrough = false;
729 *flags |= BDRV_O_NOCACHE;
730 } else if (!strcmp(mode, "directsync")) {
731 *writethrough = true;
732 *flags |= BDRV_O_NOCACHE;
733 } else if (!strcmp(mode, "writeback")) {
734 *writethrough = false;
735 } else if (!strcmp(mode, "unsafe")) {
736 *writethrough = false;
737 *flags |= BDRV_O_NO_FLUSH;
738 } else if (!strcmp(mode, "writethrough")) {
739 *writethrough = true;
747 static char *bdrv_child_get_parent_desc(BdrvChild *c)
749 BlockDriverState *parent = c->opaque;
750 return g_strdup(bdrv_get_device_or_node_name(parent));
753 static void bdrv_child_cb_drained_begin(BdrvChild *child)
755 BlockDriverState *bs = child->opaque;
756 bdrv_drained_begin(bs);
759 static void bdrv_child_cb_drained_end(BdrvChild *child)
761 BlockDriverState *bs = child->opaque;
762 bdrv_drained_end(bs);
766 * Returns the options and flags that a temporary snapshot should get, based on
767 * the originally requested flags (the originally requested image will have
768 * flags like a backing file)
770 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
771 int parent_flags, QDict *parent_options)
773 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
775 /* For temporary files, unconditional cache=unsafe is fine */
776 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
777 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
779 /* Copy the read-only option from the parent */
780 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
782 /* aio=native doesn't work for cache.direct=off, so disable it for the
783 * temporary snapshot */
784 *child_flags &= ~BDRV_O_NATIVE_AIO;
788 * Returns the options and flags that bs->file should get if a protocol driver
789 * is expected, based on the given options and flags for the parent BDS
791 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
792 int parent_flags, QDict *parent_options)
794 int flags = parent_flags;
796 /* Enable protocol handling, disable format probing for bs->file */
797 flags |= BDRV_O_PROTOCOL;
799 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
801 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
802 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
804 /* Inherit the read-only option from the parent if it's not set */
805 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
807 /* Our block drivers take care to send flushes and respect unmap policy,
808 * so we can default to enable both on lower layers regardless of the
809 * corresponding parent options. */
810 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
812 /* Clear flags that only apply to the top layer */
813 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
816 *child_flags = flags;
819 const BdrvChildRole child_file = {
820 .get_parent_desc = bdrv_child_get_parent_desc,
821 .inherit_options = bdrv_inherited_options,
822 .drained_begin = bdrv_child_cb_drained_begin,
823 .drained_end = bdrv_child_cb_drained_end,
827 * Returns the options and flags that bs->file should get if the use of formats
828 * (and not only protocols) is permitted for it, based on the given options and
829 * flags for the parent BDS
831 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
832 int parent_flags, QDict *parent_options)
834 child_file.inherit_options(child_flags, child_options,
835 parent_flags, parent_options);
837 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
840 const BdrvChildRole child_format = {
841 .get_parent_desc = bdrv_child_get_parent_desc,
842 .inherit_options = bdrv_inherited_fmt_options,
843 .drained_begin = bdrv_child_cb_drained_begin,
844 .drained_end = bdrv_child_cb_drained_end,
847 static void bdrv_backing_attach(BdrvChild *c)
849 BlockDriverState *parent = c->opaque;
850 BlockDriverState *backing_hd = c->bs;
852 assert(!parent->backing_blocker);
853 error_setg(&parent->backing_blocker,
854 "node is used as backing hd of '%s'",
855 bdrv_get_device_or_node_name(parent));
857 parent->open_flags &= ~BDRV_O_NO_BACKING;
858 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
859 backing_hd->filename);
860 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
861 backing_hd->drv ? backing_hd->drv->format_name : "");
863 bdrv_op_block_all(backing_hd, parent->backing_blocker);
864 /* Otherwise we won't be able to commit or stream */
865 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
866 parent->backing_blocker);
867 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
868 parent->backing_blocker);
870 * We do backup in 3 ways:
872 * The target bs is new opened, and the source is top BDS
874 * Both the source and the target are top BDSes.
875 * 3. internal backup(used for block replication)
876 * Both the source and the target are backing file
878 * In case 1 and 2, neither the source nor the target is the backing file.
879 * In case 3, we will block the top BDS, so there is only one block job
880 * for the top BDS and its backing chain.
882 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
883 parent->backing_blocker);
884 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
885 parent->backing_blocker);
888 static void bdrv_backing_detach(BdrvChild *c)
890 BlockDriverState *parent = c->opaque;
892 assert(parent->backing_blocker);
893 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
894 error_free(parent->backing_blocker);
895 parent->backing_blocker = NULL;
899 * Returns the options and flags that bs->backing should get, based on the
900 * given options and flags for the parent BDS
902 static void bdrv_backing_options(int *child_flags, QDict *child_options,
903 int parent_flags, QDict *parent_options)
905 int flags = parent_flags;
907 /* The cache mode is inherited unmodified for backing files; except WCE,
908 * which is only applied on the top level (BlockBackend) */
909 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
910 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
912 /* backing files always opened read-only */
913 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
914 flags &= ~BDRV_O_COPY_ON_READ;
916 /* snapshot=on is handled on the top layer */
917 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
919 *child_flags = flags;
922 const BdrvChildRole child_backing = {
923 .get_parent_desc = bdrv_child_get_parent_desc,
924 .attach = bdrv_backing_attach,
925 .detach = bdrv_backing_detach,
926 .inherit_options = bdrv_backing_options,
927 .drained_begin = bdrv_child_cb_drained_begin,
928 .drained_end = bdrv_child_cb_drained_end,
931 static int bdrv_open_flags(BlockDriverState *bs, int flags)
933 int open_flags = flags;
936 * Clear flags that are internal to the block layer before opening the
939 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
942 * Snapshots should be writable.
944 if (flags & BDRV_O_TEMPORARY) {
945 open_flags |= BDRV_O_RDWR;
951 static void update_flags_from_options(int *flags, QemuOpts *opts)
953 *flags &= ~BDRV_O_CACHE_MASK;
955 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
956 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
957 *flags |= BDRV_O_NO_FLUSH;
960 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
961 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
962 *flags |= BDRV_O_NOCACHE;
965 *flags &= ~BDRV_O_RDWR;
967 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
968 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
969 *flags |= BDRV_O_RDWR;
974 static void update_options_from_flags(QDict *options, int flags)
976 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
977 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
979 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
980 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
981 flags & BDRV_O_NO_FLUSH);
983 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
984 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
988 static void bdrv_assign_node_name(BlockDriverState *bs,
989 const char *node_name,
992 char *gen_node_name = NULL;
995 node_name = gen_node_name = id_generate(ID_BLOCK);
996 } else if (!id_wellformed(node_name)) {
998 * Check for empty string or invalid characters, but not if it is
999 * generated (generated names use characters not available to the user)
1001 error_setg(errp, "Invalid node name");
1005 /* takes care of avoiding namespaces collisions */
1006 if (blk_by_name(node_name)) {
1007 error_setg(errp, "node-name=%s is conflicting with a device id",
1012 /* takes care of avoiding duplicates node names */
1013 if (bdrv_find_node(node_name)) {
1014 error_setg(errp, "Duplicate node name");
1018 /* copy node name into the bs and insert it into the graph list */
1019 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1020 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1022 g_free(gen_node_name);
1025 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1026 const char *node_name, QDict *options,
1027 int open_flags, Error **errp)
1029 Error *local_err = NULL;
1032 bdrv_assign_node_name(bs, node_name, &local_err);
1034 error_propagate(errp, local_err);
1039 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1040 bs->opaque = g_malloc0(drv->instance_size);
1042 if (drv->bdrv_file_open) {
1043 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1044 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1045 } else if (drv->bdrv_open) {
1046 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1053 error_propagate(errp, local_err);
1054 } else if (bs->filename[0]) {
1055 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1057 error_setg_errno(errp, -ret, "Could not open image");
1062 ret = refresh_total_sectors(bs, bs->total_sectors);
1064 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1068 bdrv_refresh_limits(bs, &local_err);
1070 error_propagate(errp, local_err);
1075 assert(bdrv_opt_mem_align(bs) != 0);
1076 assert(bdrv_min_mem_align(bs) != 0);
1077 assert(is_power_of_2(bs->bl.request_alignment));
1082 /* FIXME Close bs first if already opened*/
1089 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1090 int flags, Error **errp)
1092 BlockDriverState *bs;
1096 bs->open_flags = flags;
1097 bs->explicit_options = qdict_new();
1098 bs->options = qdict_new();
1101 update_options_from_flags(bs->options, flags);
1103 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1105 QDECREF(bs->explicit_options);
1106 QDECREF(bs->options);
1114 QemuOptsList bdrv_runtime_opts = {
1115 .name = "bdrv_common",
1116 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1119 .name = "node-name",
1120 .type = QEMU_OPT_STRING,
1121 .help = "Node name of the block device node",
1125 .type = QEMU_OPT_STRING,
1126 .help = "Block driver to use for the node",
1129 .name = BDRV_OPT_CACHE_DIRECT,
1130 .type = QEMU_OPT_BOOL,
1131 .help = "Bypass software writeback cache on the host",
1134 .name = BDRV_OPT_CACHE_NO_FLUSH,
1135 .type = QEMU_OPT_BOOL,
1136 .help = "Ignore flush requests",
1139 .name = BDRV_OPT_READ_ONLY,
1140 .type = QEMU_OPT_BOOL,
1141 .help = "Node is opened in read-only mode",
1144 .name = "detect-zeroes",
1145 .type = QEMU_OPT_STRING,
1146 .help = "try to optimize zero writes (off, on, unmap)",
1150 .type = QEMU_OPT_STRING,
1151 .help = "discard operation (ignore/off, unmap/on)",
1153 { /* end of list */ }
1158 * Common part for opening disk images and files
1160 * Removes all processed options from *options.
1162 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1163 QDict *options, Error **errp)
1165 int ret, open_flags;
1166 const char *filename;
1167 const char *driver_name = NULL;
1168 const char *node_name = NULL;
1169 const char *discard;
1170 const char *detect_zeroes;
1173 Error *local_err = NULL;
1175 assert(bs->file == NULL);
1176 assert(options != NULL && bs->options != options);
1178 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1179 qemu_opts_absorb_qdict(opts, options, &local_err);
1181 error_propagate(errp, local_err);
1186 update_flags_from_options(&bs->open_flags, opts);
1188 driver_name = qemu_opt_get(opts, "driver");
1189 drv = bdrv_find_format(driver_name);
1190 assert(drv != NULL);
1193 filename = blk_bs(file)->filename;
1196 * Caution: while qdict_get_try_str() is fine, getting
1197 * non-string types would require more care. When @options
1198 * come from -blockdev or blockdev_add, its members are typed
1199 * according to the QAPI schema, but when they come from
1200 * -drive, they're all QString.
1202 filename = qdict_get_try_str(options, "filename");
1205 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1206 error_setg(errp, "The '%s' block driver requires a file name",
1212 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1215 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1217 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1219 !bs->read_only && bdrv_is_whitelisted(drv, true)
1220 ? "Driver '%s' can only be used for read-only devices"
1221 : "Driver '%s' is not whitelisted",
1227 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
1228 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1229 if (!bs->read_only) {
1230 bdrv_enable_copy_on_read(bs);
1232 error_setg(errp, "Can't use copy-on-read on read-only device");
1238 discard = qemu_opt_get(opts, "discard");
1239 if (discard != NULL) {
1240 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1241 error_setg(errp, "Invalid discard option");
1247 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1248 if (detect_zeroes) {
1249 BlockdevDetectZeroesOptions value =
1250 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1252 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1253 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1256 error_propagate(errp, local_err);
1261 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1262 !(bs->open_flags & BDRV_O_UNMAP))
1264 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1265 "without setting discard operation to unmap");
1270 bs->detect_zeroes = value;
1273 if (filename != NULL) {
1274 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1276 bs->filename[0] = '\0';
1278 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1280 /* Open the image, either directly or using a protocol */
1281 open_flags = bdrv_open_flags(bs, bs->open_flags);
1282 node_name = qemu_opt_get(opts, "node-name");
1284 assert(!drv->bdrv_file_open || file == NULL);
1285 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1290 qemu_opts_del(opts);
1294 qemu_opts_del(opts);
1298 static QDict *parse_json_filename(const char *filename, Error **errp)
1300 QObject *options_obj;
1304 ret = strstart(filename, "json:", &filename);
1307 options_obj = qobject_from_json(filename, errp);
1309 /* Work around qobject_from_json() lossage TODO fix that */
1310 if (errp && !*errp) {
1311 error_setg(errp, "Could not parse the JSON options");
1314 error_prepend(errp, "Could not parse the JSON options: ");
1318 options = qobject_to_qdict(options_obj);
1320 qobject_decref(options_obj);
1321 error_setg(errp, "Invalid JSON object given");
1325 qdict_flatten(options);
1330 static void parse_json_protocol(QDict *options, const char **pfilename,
1333 QDict *json_options;
1334 Error *local_err = NULL;
1336 /* Parse json: pseudo-protocol */
1337 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1341 json_options = parse_json_filename(*pfilename, &local_err);
1343 error_propagate(errp, local_err);
1347 /* Options given in the filename have lower priority than options
1348 * specified directly */
1349 qdict_join(options, json_options, false);
1350 QDECREF(json_options);
1355 * Fills in default options for opening images and converts the legacy
1356 * filename/flags pair to option QDict entries.
1357 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1358 * block driver has been specified explicitly.
1360 static int bdrv_fill_options(QDict **options, const char *filename,
1361 int *flags, Error **errp)
1363 const char *drvname;
1364 bool protocol = *flags & BDRV_O_PROTOCOL;
1365 bool parse_filename = false;
1366 BlockDriver *drv = NULL;
1367 Error *local_err = NULL;
1370 * Caution: while qdict_get_try_str() is fine, getting non-string
1371 * types would require more care. When @options come from
1372 * -blockdev or blockdev_add, its members are typed according to
1373 * the QAPI schema, but when they come from -drive, they're all
1376 drvname = qdict_get_try_str(*options, "driver");
1378 drv = bdrv_find_format(drvname);
1380 error_setg(errp, "Unknown driver '%s'", drvname);
1383 /* If the user has explicitly specified the driver, this choice should
1384 * override the BDRV_O_PROTOCOL flag */
1385 protocol = drv->bdrv_file_open;
1389 *flags |= BDRV_O_PROTOCOL;
1391 *flags &= ~BDRV_O_PROTOCOL;
1394 /* Translate cache options from flags into options */
1395 update_options_from_flags(*options, *flags);
1397 /* Fetch the file name from the options QDict if necessary */
1398 if (protocol && filename) {
1399 if (!qdict_haskey(*options, "filename")) {
1400 qdict_put_str(*options, "filename", filename);
1401 parse_filename = true;
1403 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1409 /* Find the right block driver */
1410 /* See cautionary note on accessing @options above */
1411 filename = qdict_get_try_str(*options, "filename");
1413 if (!drvname && protocol) {
1415 drv = bdrv_find_protocol(filename, parse_filename, errp);
1420 drvname = drv->format_name;
1421 qdict_put_str(*options, "driver", drvname);
1423 error_setg(errp, "Must specify either driver or file");
1428 assert(drv || !protocol);
1430 /* Driver-specific filename parsing */
1431 if (drv && drv->bdrv_parse_filename && parse_filename) {
1432 drv->bdrv_parse_filename(filename, *options, &local_err);
1434 error_propagate(errp, local_err);
1438 if (!drv->bdrv_needs_filename) {
1439 qdict_del(*options, "filename");
1446 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1447 GSList *ignore_children, Error **errp);
1448 static void bdrv_child_abort_perm_update(BdrvChild *c);
1449 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1452 * Check whether permissions on this node can be changed in a way that
1453 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1454 * permissions of all its parents. This involves checking whether all necessary
1455 * permission changes to child nodes can be performed.
1457 * A call to this function must always be followed by a call to bdrv_set_perm()
1458 * or bdrv_abort_perm_update().
1460 static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1461 uint64_t cumulative_shared_perms,
1462 GSList *ignore_children, Error **errp)
1464 BlockDriver *drv = bs->drv;
1468 /* Write permissions never work with read-only images */
1469 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1470 bdrv_is_read_only(bs))
1472 error_setg(errp, "Block node is read-only");
1476 /* Check this node */
1481 if (drv->bdrv_check_perm) {
1482 return drv->bdrv_check_perm(bs, cumulative_perms,
1483 cumulative_shared_perms, errp);
1486 /* Drivers that never have children can omit .bdrv_child_perm() */
1487 if (!drv->bdrv_child_perm) {
1488 assert(QLIST_EMPTY(&bs->children));
1492 /* Check all children */
1493 QLIST_FOREACH(c, &bs->children, next) {
1494 uint64_t cur_perm, cur_shared;
1495 drv->bdrv_child_perm(bs, c, c->role,
1496 cumulative_perms, cumulative_shared_perms,
1497 &cur_perm, &cur_shared);
1498 ret = bdrv_child_check_perm(c, cur_perm, cur_shared, ignore_children,
1509 * Notifies drivers that after a previous bdrv_check_perm() call, the
1510 * permission update is not performed and any preparations made for it (e.g.
1511 * taken file locks) need to be undone.
1513 * This function recursively notifies all child nodes.
1515 static void bdrv_abort_perm_update(BlockDriverState *bs)
1517 BlockDriver *drv = bs->drv;
1524 if (drv->bdrv_abort_perm_update) {
1525 drv->bdrv_abort_perm_update(bs);
1528 QLIST_FOREACH(c, &bs->children, next) {
1529 bdrv_child_abort_perm_update(c);
1533 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1534 uint64_t cumulative_shared_perms)
1536 BlockDriver *drv = bs->drv;
1543 /* Update this node */
1544 if (drv->bdrv_set_perm) {
1545 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1548 /* Drivers that never have children can omit .bdrv_child_perm() */
1549 if (!drv->bdrv_child_perm) {
1550 assert(QLIST_EMPTY(&bs->children));
1554 /* Update all children */
1555 QLIST_FOREACH(c, &bs->children, next) {
1556 uint64_t cur_perm, cur_shared;
1557 drv->bdrv_child_perm(bs, c, c->role,
1558 cumulative_perms, cumulative_shared_perms,
1559 &cur_perm, &cur_shared);
1560 bdrv_child_set_perm(c, cur_perm, cur_shared);
1564 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1565 uint64_t *shared_perm)
1568 uint64_t cumulative_perms = 0;
1569 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1571 QLIST_FOREACH(c, &bs->parents, next_parent) {
1572 cumulative_perms |= c->perm;
1573 cumulative_shared_perms &= c->shared_perm;
1576 *perm = cumulative_perms;
1577 *shared_perm = cumulative_shared_perms;
1580 static char *bdrv_child_user_desc(BdrvChild *c)
1582 if (c->role->get_parent_desc) {
1583 return c->role->get_parent_desc(c);
1586 return g_strdup("another user");
1589 static char *bdrv_perm_names(uint64_t perm)
1595 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1596 { BLK_PERM_WRITE, "write" },
1597 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1598 { BLK_PERM_RESIZE, "resize" },
1599 { BLK_PERM_GRAPH_MOD, "change children" },
1603 char *result = g_strdup("");
1604 struct perm_name *p;
1606 for (p = permissions; p->name; p++) {
1607 if (perm & p->perm) {
1609 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1618 * Checks whether a new reference to @bs can be added if the new user requires
1619 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1620 * set, the BdrvChild objects in this list are ignored in the calculations;
1621 * this allows checking permission updates for an existing reference.
1623 * Needs to be followed by a call to either bdrv_set_perm() or
1624 * bdrv_abort_perm_update(). */
1625 static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm,
1626 uint64_t new_shared_perm,
1627 GSList *ignore_children, Error **errp)
1630 uint64_t cumulative_perms = new_used_perm;
1631 uint64_t cumulative_shared_perms = new_shared_perm;
1633 /* There is no reason why anyone couldn't tolerate write_unchanged */
1634 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1636 QLIST_FOREACH(c, &bs->parents, next_parent) {
1637 if (g_slist_find(ignore_children, c)) {
1641 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1642 char *user = bdrv_child_user_desc(c);
1643 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1644 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1646 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1652 if ((c->perm & new_shared_perm) != c->perm) {
1653 char *user = bdrv_child_user_desc(c);
1654 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1655 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1657 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1663 cumulative_perms |= c->perm;
1664 cumulative_shared_perms &= c->shared_perm;
1667 return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms,
1668 ignore_children, errp);
1671 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1672 * bdrv_child_abort_perm_update(). */
1673 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1674 GSList *ignore_children, Error **errp)
1678 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1679 ret = bdrv_check_update_perm(c->bs, perm, shared, ignore_children, errp);
1680 g_slist_free(ignore_children);
1685 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1687 uint64_t cumulative_perms, cumulative_shared_perms;
1690 c->shared_perm = shared;
1692 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1693 &cumulative_shared_perms);
1694 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1697 static void bdrv_child_abort_perm_update(BdrvChild *c)
1699 bdrv_abort_perm_update(c->bs);
1702 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1707 ret = bdrv_child_check_perm(c, perm, shared, NULL, errp);
1709 bdrv_child_abort_perm_update(c);
1713 bdrv_child_set_perm(c, perm, shared);
1718 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1720 | BLK_PERM_WRITE_UNCHANGED \
1722 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1724 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1725 const BdrvChildRole *role,
1726 uint64_t perm, uint64_t shared,
1727 uint64_t *nperm, uint64_t *nshared)
1730 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1731 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1735 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1736 (c->perm & DEFAULT_PERM_UNCHANGED);
1737 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1738 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1741 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1742 const BdrvChildRole *role,
1743 uint64_t perm, uint64_t shared,
1744 uint64_t *nperm, uint64_t *nshared)
1746 bool backing = (role == &child_backing);
1747 assert(role == &child_backing || role == &child_file);
1750 /* Apart from the modifications below, the same permissions are
1751 * forwarded and left alone as for filters */
1752 bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared);
1754 /* Format drivers may touch metadata even if the guest doesn't write */
1755 if (!bdrv_is_read_only(bs)) {
1756 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1759 /* bs->file always needs to be consistent because of the metadata. We
1760 * can never allow other users to resize or write to it. */
1761 perm |= BLK_PERM_CONSISTENT_READ;
1762 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1764 /* We want consistent read from backing files if the parent needs it.
1765 * No other operations are performed on backing files. */
1766 perm &= BLK_PERM_CONSISTENT_READ;
1768 /* If the parent can deal with changing data, we're okay with a
1769 * writable and resizable backing file. */
1770 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1771 if (shared & BLK_PERM_WRITE) {
1772 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1777 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1778 BLK_PERM_WRITE_UNCHANGED;
1785 static void bdrv_replace_child_noperm(BdrvChild *child,
1786 BlockDriverState *new_bs)
1788 BlockDriverState *old_bs = child->bs;
1790 if (old_bs && new_bs) {
1791 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1794 if (old_bs->quiesce_counter && child->role->drained_end) {
1795 child->role->drained_end(child);
1797 if (child->role->detach) {
1798 child->role->detach(child);
1800 QLIST_REMOVE(child, next_parent);
1806 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1807 if (new_bs->quiesce_counter && child->role->drained_begin) {
1808 child->role->drained_begin(child);
1811 if (child->role->attach) {
1812 child->role->attach(child);
1818 * Updates @child to change its reference to point to @new_bs, including
1819 * checking and applying the necessary permisson updates both to the old node
1822 * NULL is passed as @new_bs for removing the reference before freeing @child.
1824 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1825 * function uses bdrv_set_perm() to update the permissions according to the new
1826 * reference that @new_bs gets.
1828 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1830 BlockDriverState *old_bs = child->bs;
1831 uint64_t perm, shared_perm;
1834 /* Update permissions for old node. This is guaranteed to succeed
1835 * because we're just taking a parent away, so we're loosening
1837 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1838 bdrv_check_perm(old_bs, perm, shared_perm, NULL, &error_abort);
1839 bdrv_set_perm(old_bs, perm, shared_perm);
1842 bdrv_replace_child_noperm(child, new_bs);
1845 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
1846 bdrv_set_perm(new_bs, perm, shared_perm);
1850 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1851 const char *child_name,
1852 const BdrvChildRole *child_role,
1853 uint64_t perm, uint64_t shared_perm,
1854 void *opaque, Error **errp)
1859 ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp);
1861 bdrv_abort_perm_update(child_bs);
1865 child = g_new(BdrvChild, 1);
1866 *child = (BdrvChild) {
1868 .name = g_strdup(child_name),
1871 .shared_perm = shared_perm,
1875 /* This performs the matching bdrv_set_perm() for the above check. */
1876 bdrv_replace_child(child, child_bs);
1881 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1882 BlockDriverState *child_bs,
1883 const char *child_name,
1884 const BdrvChildRole *child_role,
1888 uint64_t perm, shared_perm;
1890 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
1892 assert(parent_bs->drv);
1893 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
1894 parent_bs->drv->bdrv_child_perm(parent_bs, NULL, child_role,
1895 perm, shared_perm, &perm, &shared_perm);
1897 child = bdrv_root_attach_child(child_bs, child_name, child_role,
1898 perm, shared_perm, parent_bs, errp);
1899 if (child == NULL) {
1903 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1907 static void bdrv_detach_child(BdrvChild *child)
1909 if (child->next.le_prev) {
1910 QLIST_REMOVE(child, next);
1911 child->next.le_prev = NULL;
1914 bdrv_replace_child(child, NULL);
1916 g_free(child->name);
1920 void bdrv_root_unref_child(BdrvChild *child)
1922 BlockDriverState *child_bs;
1924 child_bs = child->bs;
1925 bdrv_detach_child(child);
1926 bdrv_unref(child_bs);
1929 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1931 if (child == NULL) {
1935 if (child->bs->inherits_from == parent) {
1938 /* Remove inherits_from only when the last reference between parent and
1939 * child->bs goes away. */
1940 QLIST_FOREACH(c, &parent->children, next) {
1941 if (c != child && c->bs == child->bs) {
1946 child->bs->inherits_from = NULL;
1950 bdrv_root_unref_child(child);
1954 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1957 QLIST_FOREACH(c, &bs->parents, next_parent) {
1958 if (c->role->change_media) {
1959 c->role->change_media(c, load);
1964 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1967 QLIST_FOREACH(c, &bs->parents, next_parent) {
1968 if (c->role->resize) {
1975 * Sets the backing file link of a BDS. A new reference is created; callers
1976 * which don't need their own reference any more must call bdrv_unref().
1978 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
1982 bdrv_ref(backing_hd);
1986 bdrv_unref_child(bs, bs->backing);
1994 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
1997 bdrv_unref(backing_hd);
2000 bdrv_refresh_filename(bs);
2003 bdrv_refresh_limits(bs, NULL);
2007 * Opens the backing file for a BlockDriverState if not yet open
2009 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2010 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2011 * itself, all options starting with "${bdref_key}." are considered part of the
2014 * TODO Can this be unified with bdrv_open_image()?
2016 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2017 const char *bdref_key, Error **errp)
2019 char *backing_filename = g_malloc0(PATH_MAX);
2020 char *bdref_key_dot;
2021 const char *reference = NULL;
2023 BlockDriverState *backing_hd;
2025 QDict *tmp_parent_options = NULL;
2026 Error *local_err = NULL;
2028 if (bs->backing != NULL) {
2032 /* NULL means an empty set of options */
2033 if (parent_options == NULL) {
2034 tmp_parent_options = qdict_new();
2035 parent_options = tmp_parent_options;
2038 bs->open_flags &= ~BDRV_O_NO_BACKING;
2040 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2041 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2042 g_free(bdref_key_dot);
2045 * Caution: while qdict_get_try_str() is fine, getting non-string
2046 * types would require more care. When @parent_options come from
2047 * -blockdev or blockdev_add, its members are typed according to
2048 * the QAPI schema, but when they come from -drive, they're all
2051 reference = qdict_get_try_str(parent_options, bdref_key);
2052 if (reference || qdict_haskey(options, "file.filename")) {
2053 backing_filename[0] = '\0';
2054 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2058 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2062 error_propagate(errp, local_err);
2068 if (!bs->drv || !bs->drv->supports_backing) {
2070 error_setg(errp, "Driver doesn't support backing files");
2075 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2076 qdict_put_str(options, "driver", bs->backing_format);
2079 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2080 reference, options, 0, bs, &child_backing,
2083 bs->open_flags |= BDRV_O_NO_BACKING;
2084 error_prepend(errp, "Could not open backing file: ");
2089 /* Hook up the backing file link; drop our reference, bs owns the
2090 * backing_hd reference now */
2091 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2092 bdrv_unref(backing_hd);
2094 error_propagate(errp, local_err);
2099 qdict_del(parent_options, bdref_key);
2102 g_free(backing_filename);
2103 QDECREF(tmp_parent_options);
2107 static BlockDriverState *
2108 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2109 BlockDriverState *parent, const BdrvChildRole *child_role,
2110 bool allow_none, Error **errp)
2112 BlockDriverState *bs = NULL;
2113 QDict *image_options;
2114 char *bdref_key_dot;
2115 const char *reference;
2117 assert(child_role != NULL);
2119 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2120 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2121 g_free(bdref_key_dot);
2124 * Caution: while qdict_get_try_str() is fine, getting non-string
2125 * types would require more care. When @options come from
2126 * -blockdev or blockdev_add, its members are typed according to
2127 * the QAPI schema, but when they come from -drive, they're all
2130 reference = qdict_get_try_str(options, bdref_key);
2131 if (!filename && !reference && !qdict_size(image_options)) {
2133 error_setg(errp, "A block device must be specified for \"%s\"",
2136 QDECREF(image_options);
2140 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2141 parent, child_role, errp);
2147 qdict_del(options, bdref_key);
2152 * Opens a disk image whose options are given as BlockdevRef in another block
2155 * If allow_none is true, no image will be opened if filename is false and no
2156 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2158 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2159 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2160 * itself, all options starting with "${bdref_key}." are considered part of the
2163 * The BlockdevRef will be removed from the options QDict.
2165 BdrvChild *bdrv_open_child(const char *filename,
2166 QDict *options, const char *bdref_key,
2167 BlockDriverState *parent,
2168 const BdrvChildRole *child_role,
2169 bool allow_none, Error **errp)
2172 BlockDriverState *bs;
2174 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2180 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2189 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2191 QDict *snapshot_options,
2194 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2195 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2197 QemuOpts *opts = NULL;
2198 BlockDriverState *bs_snapshot;
2199 Error *local_err = NULL;
2202 /* if snapshot, we create a temporary backing file and open it
2203 instead of opening 'filename' directly */
2205 /* Get the required size from the image */
2206 total_size = bdrv_getlength(bs);
2207 if (total_size < 0) {
2208 error_setg_errno(errp, -total_size, "Could not get image size");
2212 /* Create the temporary image */
2213 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2215 error_setg_errno(errp, -ret, "Could not get temporary filename");
2219 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2221 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2222 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2223 qemu_opts_del(opts);
2225 error_prepend(errp, "Could not create temporary overlay '%s': ",
2230 /* Prepare options QDict for the temporary file */
2231 qdict_put_str(snapshot_options, "file.driver", "file");
2232 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2233 qdict_put_str(snapshot_options, "driver", "qcow2");
2235 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2236 snapshot_options = NULL;
2242 /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will
2243 * call bdrv_unref() on it), so in order to be able to return one, we have
2244 * to increase bs_snapshot's refcount here */
2245 bdrv_ref(bs_snapshot);
2246 bdrv_append(bs_snapshot, bs, &local_err);
2248 error_propagate(errp, local_err);
2253 g_free(tmp_filename);
2257 QDECREF(snapshot_options);
2258 g_free(tmp_filename);
2263 * Opens a disk image (raw, qcow2, vmdk, ...)
2265 * options is a QDict of options to pass to the block drivers, or NULL for an
2266 * empty set of options. The reference to the QDict belongs to the block layer
2267 * after the call (even on failure), so if the caller intends to reuse the
2268 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2270 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2271 * If it is not NULL, the referenced BDS will be reused.
2273 * The reference parameter may be used to specify an existing block device which
2274 * should be opened. If specified, neither options nor a filename may be given,
2275 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2277 static BlockDriverState *bdrv_open_inherit(const char *filename,
2278 const char *reference,
2279 QDict *options, int flags,
2280 BlockDriverState *parent,
2281 const BdrvChildRole *child_role,
2285 BlockBackend *file = NULL;
2286 BlockDriverState *bs;
2287 BlockDriver *drv = NULL;
2288 const char *drvname;
2289 const char *backing;
2290 Error *local_err = NULL;
2291 QDict *snapshot_options = NULL;
2292 int snapshot_flags = 0;
2294 assert(!child_role || !flags);
2295 assert(!child_role == !parent);
2298 bool options_non_empty = options ? qdict_size(options) : false;
2301 if (filename || options_non_empty) {
2302 error_setg(errp, "Cannot reference an existing block device with "
2303 "additional options or a new filename");
2307 bs = bdrv_lookup_bs(reference, reference, errp);
2318 /* NULL means an empty set of options */
2319 if (options == NULL) {
2320 options = qdict_new();
2323 /* json: syntax counts as explicit options, as if in the QDict */
2324 parse_json_protocol(options, &filename, &local_err);
2329 bs->explicit_options = qdict_clone_shallow(options);
2332 bs->inherits_from = parent;
2333 child_role->inherit_options(&flags, options,
2334 parent->open_flags, parent->options);
2337 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2343 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2344 * Caution: getting a boolean member of @options requires care.
2345 * When @options come from -blockdev or blockdev_add, members are
2346 * typed according to the QAPI schema, but when they come from
2347 * -drive, they're all QString.
2349 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2350 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2351 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2353 flags &= ~BDRV_O_RDWR;
2356 if (flags & BDRV_O_SNAPSHOT) {
2357 snapshot_options = qdict_new();
2358 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2360 /* Let bdrv_backing_options() override "read-only" */
2361 qdict_del(options, BDRV_OPT_READ_ONLY);
2362 bdrv_backing_options(&flags, options, flags, options);
2365 bs->open_flags = flags;
2366 bs->options = options;
2367 options = qdict_clone_shallow(options);
2369 /* Find the right image format driver */
2370 /* See cautionary note on accessing @options above */
2371 drvname = qdict_get_try_str(options, "driver");
2373 drv = bdrv_find_format(drvname);
2375 error_setg(errp, "Unknown driver: '%s'", drvname);
2380 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2382 /* See cautionary note on accessing @options above */
2383 backing = qdict_get_try_str(options, "backing");
2384 if (backing && *backing == '\0') {
2385 flags |= BDRV_O_NO_BACKING;
2386 qdict_del(options, "backing");
2389 /* Open image file without format layer. This BlockBackend is only used for
2390 * probing, the block drivers will do their own bdrv_open_child() for the
2391 * same BDS, which is why we put the node name back into options. */
2392 if ((flags & BDRV_O_PROTOCOL) == 0) {
2393 BlockDriverState *file_bs;
2395 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2396 &child_file, true, &local_err);
2400 if (file_bs != NULL) {
2401 file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
2402 blk_insert_bs(file, file_bs, &local_err);
2403 bdrv_unref(file_bs);
2408 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2412 /* Image format probing */
2415 ret = find_image_format(file, filename, &drv, &local_err);
2420 * This option update would logically belong in bdrv_fill_options(),
2421 * but we first need to open bs->file for the probing to work, while
2422 * opening bs->file already requires the (mostly) final set of options
2423 * so that cache mode etc. can be inherited.
2425 * Adding the driver later is somewhat ugly, but it's not an option
2426 * that would ever be inherited, so it's correct. We just need to make
2427 * sure to update both bs->options (which has the full effective
2428 * options for bs) and options (which has file.* already removed).
2430 qdict_put_str(bs->options, "driver", drv->format_name);
2431 qdict_put_str(options, "driver", drv->format_name);
2433 error_setg(errp, "Must specify either driver or file");
2437 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2438 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2439 /* file must be NULL if a protocol BDS is about to be created
2440 * (the inverse results in an error message from bdrv_open_common()) */
2441 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2443 /* Open the image */
2444 ret = bdrv_open_common(bs, file, options, &local_err);
2454 /* If there is a backing file, use it */
2455 if ((flags & BDRV_O_NO_BACKING) == 0) {
2456 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2458 goto close_and_fail;
2462 bdrv_refresh_filename(bs);
2464 /* Check if any unknown options were used */
2465 if (qdict_size(options) != 0) {
2466 const QDictEntry *entry = qdict_first(options);
2467 if (flags & BDRV_O_PROTOCOL) {
2468 error_setg(errp, "Block protocol '%s' doesn't support the option "
2469 "'%s'", drv->format_name, entry->key);
2472 "Block format '%s' does not support the option '%s'",
2473 drv->format_name, entry->key);
2476 goto close_and_fail;
2479 if (!bdrv_key_required(bs)) {
2480 bdrv_parent_cb_change_media(bs, true);
2481 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
2482 && !runstate_check(RUN_STATE_INMIGRATE)
2483 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
2485 "Guest must be stopped for opening of encrypted image");
2486 goto close_and_fail;
2491 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2492 * temporary snapshot afterwards. */
2493 if (snapshot_flags) {
2494 BlockDriverState *snapshot_bs;
2495 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2496 snapshot_options, &local_err);
2497 snapshot_options = NULL;
2499 goto close_and_fail;
2501 /* We are not going to return bs but the overlay on top of it
2502 * (snapshot_bs); thus, we have to drop the strong reference to bs
2503 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2504 * though, because the overlay still has a reference to it. */
2513 if (bs->file != NULL) {
2514 bdrv_unref_child(bs, bs->file);
2516 QDECREF(snapshot_options);
2517 QDECREF(bs->explicit_options);
2518 QDECREF(bs->options);
2522 error_propagate(errp, local_err);
2527 QDECREF(snapshot_options);
2529 error_propagate(errp, local_err);
2533 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2534 QDict *options, int flags, Error **errp)
2536 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2540 typedef struct BlockReopenQueueEntry {
2542 BDRVReopenState state;
2543 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2544 } BlockReopenQueueEntry;
2547 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2548 * reopen of multiple devices.
2550 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2551 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2552 * be created and initialized. This newly created BlockReopenQueue should be
2553 * passed back in for subsequent calls that are intended to be of the same
2556 * bs is the BlockDriverState to add to the reopen queue.
2558 * options contains the changed options for the associated bs
2559 * (the BlockReopenQueue takes ownership)
2561 * flags contains the open flags for the associated bs
2563 * returns a pointer to bs_queue, which is either the newly allocated
2564 * bs_queue, or the existing bs_queue being used.
2567 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2568 BlockDriverState *bs,
2571 const BdrvChildRole *role,
2572 QDict *parent_options,
2577 BlockReopenQueueEntry *bs_entry;
2579 QDict *old_options, *explicit_options;
2581 if (bs_queue == NULL) {
2582 bs_queue = g_new0(BlockReopenQueue, 1);
2583 QSIMPLEQ_INIT(bs_queue);
2587 options = qdict_new();
2590 /* Check if this BlockDriverState is already in the queue */
2591 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2592 if (bs == bs_entry->state.bs) {
2598 * Precedence of options:
2599 * 1. Explicitly passed in options (highest)
2600 * 2. Set in flags (only for top level)
2601 * 3. Retained from explicitly set options of bs
2602 * 4. Inherited from parent node
2603 * 5. Retained from effective options of bs
2606 if (!parent_options) {
2608 * Any setting represented by flags is always updated. If the
2609 * corresponding QDict option is set, it takes precedence. Otherwise
2610 * the flag is translated into a QDict option. The old setting of bs is
2613 update_options_from_flags(options, flags);
2616 /* Old explicitly set values (don't overwrite by inherited value) */
2618 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2620 old_options = qdict_clone_shallow(bs->explicit_options);
2622 bdrv_join_options(bs, options, old_options);
2623 QDECREF(old_options);
2625 explicit_options = qdict_clone_shallow(options);
2627 /* Inherit from parent node */
2628 if (parent_options) {
2630 role->inherit_options(&flags, options, parent_flags, parent_options);
2633 /* Old values are used for options that aren't set yet */
2634 old_options = qdict_clone_shallow(bs->options);
2635 bdrv_join_options(bs, options, old_options);
2636 QDECREF(old_options);
2638 /* bdrv_open() masks this flag out */
2639 flags &= ~BDRV_O_PROTOCOL;
2641 QLIST_FOREACH(child, &bs->children, next) {
2642 QDict *new_child_options;
2643 char *child_key_dot;
2645 /* reopen can only change the options of block devices that were
2646 * implicitly created and inherited options. For other (referenced)
2647 * block devices, a syntax like "backing.foo" results in an error. */
2648 if (child->bs->inherits_from != bs) {
2652 child_key_dot = g_strdup_printf("%s.", child->name);
2653 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2654 g_free(child_key_dot);
2656 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2657 child->role, options, flags);
2661 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2662 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2664 QDECREF(bs_entry->state.options);
2665 QDECREF(bs_entry->state.explicit_options);
2668 bs_entry->state.bs = bs;
2669 bs_entry->state.options = options;
2670 bs_entry->state.explicit_options = explicit_options;
2671 bs_entry->state.flags = flags;
2676 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2677 BlockDriverState *bs,
2678 QDict *options, int flags)
2680 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2685 * Reopen multiple BlockDriverStates atomically & transactionally.
2687 * The queue passed in (bs_queue) must have been built up previous
2688 * via bdrv_reopen_queue().
2690 * Reopens all BDS specified in the queue, with the appropriate
2691 * flags. All devices are prepared for reopen, and failure of any
2692 * device will cause all device changes to be abandonded, and intermediate
2695 * If all devices prepare successfully, then the changes are committed
2699 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2702 BlockReopenQueueEntry *bs_entry, *next;
2703 Error *local_err = NULL;
2705 assert(bs_queue != NULL);
2707 aio_context_release(ctx);
2708 bdrv_drain_all_begin();
2709 aio_context_acquire(ctx);
2711 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2712 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2713 error_propagate(errp, local_err);
2716 bs_entry->prepared = true;
2719 /* If we reach this point, we have success and just need to apply the
2722 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2723 bdrv_reopen_commit(&bs_entry->state);
2729 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2730 if (ret && bs_entry->prepared) {
2731 bdrv_reopen_abort(&bs_entry->state);
2733 QDECREF(bs_entry->state.explicit_options);
2735 QDECREF(bs_entry->state.options);
2740 bdrv_drain_all_end();
2746 /* Reopen a single BlockDriverState with the specified flags. */
2747 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2750 Error *local_err = NULL;
2751 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2753 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2754 if (local_err != NULL) {
2755 error_propagate(errp, local_err);
2762 * Prepares a BlockDriverState for reopen. All changes are staged in the
2763 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2764 * the block driver layer .bdrv_reopen_prepare()
2766 * bs is the BlockDriverState to reopen
2767 * flags are the new open flags
2768 * queue is the reopen queue
2770 * Returns 0 on success, non-zero on error. On error errp will be set
2773 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2774 * It is the responsibility of the caller to then call the abort() or
2775 * commit() for any other BDS that have been left in a prepare() state
2778 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2782 Error *local_err = NULL;
2788 assert(reopen_state != NULL);
2789 assert(reopen_state->bs->drv != NULL);
2790 drv = reopen_state->bs->drv;
2792 /* Process generic block layer options */
2793 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2794 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2796 error_propagate(errp, local_err);
2801 update_flags_from_options(&reopen_state->flags, opts);
2803 /* node-name and driver must be unchanged. Put them back into the QDict, so
2804 * that they are checked at the end of this function. */
2805 value = qemu_opt_get(opts, "node-name");
2807 qdict_put_str(reopen_state->options, "node-name", value);
2810 value = qemu_opt_get(opts, "driver");
2812 qdict_put_str(reopen_state->options, "driver", value);
2815 /* If we are to stay read-only, do not allow permission change
2816 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
2817 * not set, or if the BDS still has copy_on_read enabled */
2818 read_only = !(reopen_state->flags & BDRV_O_RDWR);
2819 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, &local_err);
2821 error_propagate(errp, local_err);
2826 ret = bdrv_flush(reopen_state->bs);
2828 error_setg_errno(errp, -ret, "Error flushing drive");
2832 if (drv->bdrv_reopen_prepare) {
2833 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2835 if (local_err != NULL) {
2836 error_propagate(errp, local_err);
2838 error_setg(errp, "failed while preparing to reopen image '%s'",
2839 reopen_state->bs->filename);
2844 /* It is currently mandatory to have a bdrv_reopen_prepare()
2845 * handler for each supported drv. */
2846 error_setg(errp, "Block format '%s' used by node '%s' "
2847 "does not support reopening files", drv->format_name,
2848 bdrv_get_device_or_node_name(reopen_state->bs));
2853 /* Options that are not handled are only okay if they are unchanged
2854 * compared to the old state. It is expected that some options are only
2855 * used for the initial open, but not reopen (e.g. filename) */
2856 if (qdict_size(reopen_state->options)) {
2857 const QDictEntry *entry = qdict_first(reopen_state->options);
2860 QString *new_obj = qobject_to_qstring(entry->value);
2861 const char *new = qstring_get_str(new_obj);
2863 * Caution: while qdict_get_try_str() is fine, getting
2864 * non-string types would require more care. When
2865 * bs->options come from -blockdev or blockdev_add, its
2866 * members are typed according to the QAPI schema, but
2867 * when they come from -drive, they're all QString.
2869 const char *old = qdict_get_try_str(reopen_state->bs->options,
2872 if (!old || strcmp(new, old)) {
2873 error_setg(errp, "Cannot change the option '%s'", entry->key);
2877 } while ((entry = qdict_next(reopen_state->options, entry)));
2883 qemu_opts_del(opts);
2888 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2889 * makes them final by swapping the staging BlockDriverState contents into
2890 * the active BlockDriverState contents.
2892 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2896 assert(reopen_state != NULL);
2897 drv = reopen_state->bs->drv;
2898 assert(drv != NULL);
2900 /* If there are any driver level actions to take */
2901 if (drv->bdrv_reopen_commit) {
2902 drv->bdrv_reopen_commit(reopen_state);
2905 /* set BDS specific flags now */
2906 QDECREF(reopen_state->bs->explicit_options);
2908 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2909 reopen_state->bs->open_flags = reopen_state->flags;
2910 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2912 bdrv_refresh_limits(reopen_state->bs, NULL);
2916 * Abort the reopen, and delete and free the staged changes in
2919 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2923 assert(reopen_state != NULL);
2924 drv = reopen_state->bs->drv;
2925 assert(drv != NULL);
2927 if (drv->bdrv_reopen_abort) {
2928 drv->bdrv_reopen_abort(reopen_state);
2931 QDECREF(reopen_state->explicit_options);
2935 static void bdrv_close(BlockDriverState *bs)
2937 BdrvAioNotifier *ban, *ban_next;
2940 assert(!bs->refcnt);
2942 bdrv_drained_begin(bs); /* complete I/O */
2944 bdrv_drain(bs); /* in case flush left pending I/O */
2946 bdrv_release_named_dirty_bitmaps(bs);
2947 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2950 BdrvChild *child, *next;
2952 bs->drv->bdrv_close(bs);
2955 bdrv_set_backing_hd(bs, NULL, &error_abort);
2957 if (bs->file != NULL) {
2958 bdrv_unref_child(bs, bs->file);
2962 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2963 /* TODO Remove bdrv_unref() from drivers' close function and use
2964 * bdrv_unref_child() here */
2965 if (child->bs->inherits_from == bs) {
2966 child->bs->inherits_from = NULL;
2968 bdrv_detach_child(child);
2973 bs->copy_on_read = 0;
2974 bs->backing_file[0] = '\0';
2975 bs->backing_format[0] = '\0';
2976 bs->total_sectors = 0;
2977 bs->encrypted = false;
2978 bs->valid_key = false;
2980 QDECREF(bs->options);
2981 QDECREF(bs->explicit_options);
2983 QDECREF(bs->full_open_options);
2984 bs->full_open_options = NULL;
2987 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2990 QLIST_INIT(&bs->aio_notifiers);
2991 bdrv_drained_end(bs);
2994 void bdrv_close_all(void)
2996 block_job_cancel_sync_all();
2997 nbd_export_close_all();
2999 /* Drop references from requests still in flight, such as canceled block
3000 * jobs whose AIO context has not been polled yet */
3003 blk_remove_all_bs();
3004 blockdev_close_all_bdrv_states();
3006 assert(QTAILQ_EMPTY(&all_bdrv_states));
3009 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3013 if (c->role->stay_at_node) {
3017 if (c->role == &child_backing) {
3018 /* If @from is a backing file of @to, ignore the child to avoid
3019 * creating a loop. We only want to change the pointer of other
3021 QLIST_FOREACH(to_c, &to->children, next) {
3034 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3037 BdrvChild *c, *next;
3038 GSList *list = NULL, *p;
3039 uint64_t old_perm, old_shared;
3040 uint64_t perm = 0, shared = BLK_PERM_ALL;
3043 assert(!atomic_read(&from->in_flight));
3044 assert(!atomic_read(&to->in_flight));
3046 /* Make sure that @from doesn't go away until we have successfully attached
3047 * all of its parents to @to. */
3050 /* Put all parents into @list and calculate their cumulative permissions */
3051 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3052 if (!should_update_child(c, to)) {
3055 list = g_slist_prepend(list, c);
3057 shared &= c->shared_perm;
3060 /* Check whether the required permissions can be granted on @to, ignoring
3061 * all BdrvChild in @list so that they can't block themselves. */
3062 ret = bdrv_check_update_perm(to, perm, shared, list, errp);
3064 bdrv_abort_perm_update(to);
3068 /* Now actually perform the change. We performed the permission check for
3069 * all elements of @list at once, so set the permissions all at once at the
3071 for (p = list; p != NULL; p = p->next) {
3075 bdrv_replace_child_noperm(c, to);
3079 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3080 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3088 * Add new bs contents at the top of an image chain while the chain is
3089 * live, while keeping required fields on the top layer.
3091 * This will modify the BlockDriverState fields, and swap contents
3092 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3094 * bs_new must not be attached to a BlockBackend.
3096 * This function does not create any image files.
3098 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3099 * that's what the callers commonly need. bs_new will be referenced by the old
3100 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3101 * reference of its own, it must call bdrv_ref().
3103 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3106 Error *local_err = NULL;
3108 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3110 error_propagate(errp, local_err);
3114 bdrv_replace_node(bs_top, bs_new, &local_err);
3116 error_propagate(errp, local_err);
3117 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3121 /* bs_new is now referenced by its new parents, we don't need the
3122 * additional reference any more. */
3127 static void bdrv_delete(BlockDriverState *bs)
3130 assert(bdrv_op_blocker_is_empty(bs));
3131 assert(!bs->refcnt);
3135 /* remove from list, if necessary */
3136 if (bs->node_name[0] != '\0') {
3137 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3139 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3145 * Run consistency checks on an image
3147 * Returns 0 if the check could be completed (it doesn't mean that the image is
3148 * free of errors) or -errno when an internal error occurred. The results of the
3149 * check are stored in res.
3151 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3153 if (bs->drv == NULL) {
3156 if (bs->drv->bdrv_check == NULL) {
3160 memset(res, 0, sizeof(*res));
3161 return bs->drv->bdrv_check(bs, res, fix);
3167 * -EINVAL - backing format specified, but no file
3168 * -ENOSPC - can't update the backing file because no space is left in the
3170 * -ENOTSUP - format driver doesn't support changing the backing file
3172 int bdrv_change_backing_file(BlockDriverState *bs,
3173 const char *backing_file, const char *backing_fmt)
3175 BlockDriver *drv = bs->drv;
3178 /* Backing file format doesn't make sense without a backing file */
3179 if (backing_fmt && !backing_file) {
3183 if (drv->bdrv_change_backing_file != NULL) {
3184 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3190 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3191 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3197 * Finds the image layer in the chain that has 'bs' as its backing file.
3199 * active is the current topmost image.
3201 * Returns NULL if bs is not found in active's image chain,
3202 * or if active == bs.
3204 * Returns the bottommost base image if bs == NULL.
3206 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3207 BlockDriverState *bs)
3209 while (active && bs != backing_bs(active)) {
3210 active = backing_bs(active);
3216 /* Given a BDS, searches for the base layer. */
3217 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3219 return bdrv_find_overlay(bs, NULL);
3223 * Drops images above 'base' up to and including 'top', and sets the image
3224 * above 'top' to have base as its backing file.
3226 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3227 * information in 'bs' can be properly updated.
3229 * E.g., this will convert the following chain:
3230 * bottom <- base <- intermediate <- top <- active
3234 * bottom <- base <- active
3236 * It is allowed for bottom==base, in which case it converts:
3238 * base <- intermediate <- top <- active
3244 * If backing_file_str is non-NULL, it will be used when modifying top's
3245 * overlay image metadata.
3248 * if active == top, that is considered an error
3251 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
3252 BlockDriverState *base, const char *backing_file_str)
3254 BlockDriverState *new_top_bs = NULL;
3255 Error *local_err = NULL;
3258 if (!top->drv || !base->drv) {
3262 new_top_bs = bdrv_find_overlay(active, top);
3264 if (new_top_bs == NULL) {
3265 /* we could not find the image above 'top', this is an error */
3269 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3270 * to do, no intermediate images */
3271 if (backing_bs(new_top_bs) == base) {
3276 /* Make sure that base is in the backing chain of top */
3277 if (!bdrv_chain_contains(top, base)) {
3281 /* success - we can delete the intermediate states, and link top->base */
3282 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3283 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
3284 base->drv ? base->drv->format_name : "");
3289 bdrv_set_backing_hd(new_top_bs, base, &local_err);
3292 error_report_err(local_err);
3302 * Truncate file to 'offset' bytes (needed only for file protocols)
3304 int bdrv_truncate(BdrvChild *child, int64_t offset, Error **errp)
3306 BlockDriverState *bs = child->bs;
3307 BlockDriver *drv = bs->drv;
3310 assert(child->perm & BLK_PERM_RESIZE);
3313 error_setg(errp, "No medium inserted");
3316 if (!drv->bdrv_truncate) {
3317 error_setg(errp, "Image format driver does not support resize");
3320 if (bs->read_only) {
3321 error_setg(errp, "Image is read-only");
3325 assert(!(bs->open_flags & BDRV_O_INACTIVE));
3327 ret = drv->bdrv_truncate(bs, offset, errp);
3329 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3330 bdrv_dirty_bitmap_truncate(bs);
3331 bdrv_parent_cb_resize(bs);
3338 * Length of a allocated file in bytes. Sparse files are counted by actual
3339 * allocated space. Return < 0 if error or unknown.
3341 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3343 BlockDriver *drv = bs->drv;
3347 if (drv->bdrv_get_allocated_file_size) {
3348 return drv->bdrv_get_allocated_file_size(bs);
3351 return bdrv_get_allocated_file_size(bs->file->bs);
3357 * Return number of sectors on success, -errno on error.
3359 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3361 BlockDriver *drv = bs->drv;
3366 if (drv->has_variable_length) {
3367 int ret = refresh_total_sectors(bs, bs->total_sectors);
3372 return bs->total_sectors;
3376 * Return length in bytes on success, -errno on error.
3377 * The length is always a multiple of BDRV_SECTOR_SIZE.
3379 int64_t bdrv_getlength(BlockDriverState *bs)
3381 int64_t ret = bdrv_nb_sectors(bs);
3383 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3384 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3387 /* return 0 as number of sectors if no device present or error */
3388 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3390 int64_t nb_sectors = bdrv_nb_sectors(bs);
3392 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3395 bool bdrv_is_sg(BlockDriverState *bs)
3400 bool bdrv_is_encrypted(BlockDriverState *bs)
3402 if (bs->backing && bs->backing->bs->encrypted) {
3405 return bs->encrypted;
3408 bool bdrv_key_required(BlockDriverState *bs)
3410 BdrvChild *backing = bs->backing;
3412 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
3415 return (bs->encrypted && !bs->valid_key);
3418 int bdrv_set_key(BlockDriverState *bs, const char *key)
3421 if (bs->backing && bs->backing->bs->encrypted) {
3422 ret = bdrv_set_key(bs->backing->bs, key);
3428 if (!bs->encrypted) {
3430 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3433 ret = bs->drv->bdrv_set_key(bs, key);
3435 bs->valid_key = false;
3436 } else if (!bs->valid_key) {
3437 /* call the change callback now, we skipped it on open */
3438 bs->valid_key = true;
3439 bdrv_parent_cb_change_media(bs, true);
3445 * Provide an encryption key for @bs.
3446 * If @key is non-null:
3447 * If @bs is not encrypted, fail.
3448 * Else if the key is invalid, fail.
3449 * Else set @bs's key to @key, replacing the existing key, if any.
3451 * If @bs is encrypted and still lacks a key, fail.
3453 * On failure, store an error object through @errp if non-null.
3455 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
3458 if (!bdrv_is_encrypted(bs)) {
3459 error_setg(errp, "Node '%s' is not encrypted",
3460 bdrv_get_device_or_node_name(bs));
3461 } else if (bdrv_set_key(bs, key) < 0) {
3462 error_setg(errp, QERR_INVALID_PASSWORD);
3465 if (bdrv_key_required(bs)) {
3466 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
3467 "'%s' (%s) is encrypted",
3468 bdrv_get_device_or_node_name(bs),
3469 bdrv_get_encrypted_filename(bs));
3474 const char *bdrv_get_format_name(BlockDriverState *bs)
3476 return bs->drv ? bs->drv->format_name : NULL;
3479 static int qsort_strcmp(const void *a, const void *b)
3481 return strcmp(*(char *const *)a, *(char *const *)b);
3484 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3490 const char **formats = NULL;
3492 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3493 if (drv->format_name) {
3496 while (formats && i && !found) {
3497 found = !strcmp(formats[--i], drv->format_name);
3501 formats = g_renew(const char *, formats, count + 1);
3502 formats[count++] = drv->format_name;
3507 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3508 const char *format_name = block_driver_modules[i].format_name;
3514 while (formats && j && !found) {
3515 found = !strcmp(formats[--j], format_name);
3519 formats = g_renew(const char *, formats, count + 1);
3520 formats[count++] = format_name;
3525 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3527 for (i = 0; i < count; i++) {
3528 it(opaque, formats[i]);
3534 /* This function is to find a node in the bs graph */
3535 BlockDriverState *bdrv_find_node(const char *node_name)
3537 BlockDriverState *bs;
3541 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3542 if (!strcmp(node_name, bs->node_name)) {
3549 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3550 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3552 BlockDeviceInfoList *list, *entry;
3553 BlockDriverState *bs;
3556 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3557 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3559 qapi_free_BlockDeviceInfoList(list);
3562 entry = g_malloc0(sizeof(*entry));
3563 entry->value = info;
3571 BlockDriverState *bdrv_lookup_bs(const char *device,
3572 const char *node_name,
3576 BlockDriverState *bs;
3579 blk = blk_by_name(device);
3584 error_setg(errp, "Device '%s' has no medium", device);
3592 bs = bdrv_find_node(node_name);
3599 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3600 device ? device : "",
3601 node_name ? node_name : "");
3605 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3606 * return false. If either argument is NULL, return false. */
3607 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3609 while (top && top != base) {
3610 top = backing_bs(top);
3616 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3619 return QTAILQ_FIRST(&graph_bdrv_states);
3621 return QTAILQ_NEXT(bs, node_list);
3624 const char *bdrv_get_node_name(const BlockDriverState *bs)
3626 return bs->node_name;
3629 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3634 /* If multiple parents have a name, just pick the first one. */
3635 QLIST_FOREACH(c, &bs->parents, next_parent) {
3636 if (c->role->get_name) {
3637 name = c->role->get_name(c);
3638 if (name && *name) {
3647 /* TODO check what callers really want: bs->node_name or blk_name() */
3648 const char *bdrv_get_device_name(const BlockDriverState *bs)
3650 return bdrv_get_parent_name(bs) ?: "";
3653 /* This can be used to identify nodes that might not have a device
3654 * name associated. Since node and device names live in the same
3655 * namespace, the result is unambiguous. The exception is if both are
3656 * absent, then this returns an empty (non-null) string. */
3657 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3659 return bdrv_get_parent_name(bs) ?: bs->node_name;
3662 int bdrv_get_flags(BlockDriverState *bs)
3664 return bs->open_flags;
3667 int bdrv_has_zero_init_1(BlockDriverState *bs)
3672 int bdrv_has_zero_init(BlockDriverState *bs)
3676 /* If BS is a copy on write image, it is initialized to
3677 the contents of the base image, which may not be zeroes. */
3681 if (bs->drv->bdrv_has_zero_init) {
3682 return bs->drv->bdrv_has_zero_init(bs);
3689 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3691 BlockDriverInfo bdi;
3697 if (bdrv_get_info(bs, &bdi) == 0) {
3698 return bdi.unallocated_blocks_are_zero;
3704 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3706 BlockDriverInfo bdi;
3708 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3712 if (bdrv_get_info(bs, &bdi) == 0) {
3713 return bdi.can_write_zeroes_with_unmap;
3719 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3721 if (bs->backing && bs->backing->bs->encrypted)
3722 return bs->backing_file;
3723 else if (bs->encrypted)
3724 return bs->filename;
3729 void bdrv_get_backing_filename(BlockDriverState *bs,
3730 char *filename, int filename_size)
3732 pstrcpy(filename, filename_size, bs->backing_file);
3735 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3737 BlockDriver *drv = bs->drv;
3740 if (!drv->bdrv_get_info)
3742 memset(bdi, 0, sizeof(*bdi));
3743 return drv->bdrv_get_info(bs, bdi);
3746 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3748 BlockDriver *drv = bs->drv;
3749 if (drv && drv->bdrv_get_specific_info) {
3750 return drv->bdrv_get_specific_info(bs);
3755 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3757 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3761 bs->drv->bdrv_debug_event(bs, event);
3764 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3767 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3768 bs = bs->file ? bs->file->bs : NULL;
3771 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3772 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3778 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3780 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3781 bs = bs->file ? bs->file->bs : NULL;
3784 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3785 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3791 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3793 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3794 bs = bs->file ? bs->file->bs : NULL;
3797 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3798 return bs->drv->bdrv_debug_resume(bs, tag);
3804 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3806 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3807 bs = bs->file ? bs->file->bs : NULL;
3810 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3811 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3817 /* backing_file can either be relative, or absolute, or a protocol. If it is
3818 * relative, it must be relative to the chain. So, passing in bs->filename
3819 * from a BDS as backing_file should not be done, as that may be relative to
3820 * the CWD rather than the chain. */
3821 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3822 const char *backing_file)
3824 char *filename_full = NULL;
3825 char *backing_file_full = NULL;
3826 char *filename_tmp = NULL;
3827 int is_protocol = 0;
3828 BlockDriverState *curr_bs = NULL;
3829 BlockDriverState *retval = NULL;
3830 Error *local_error = NULL;
3832 if (!bs || !bs->drv || !backing_file) {
3836 filename_full = g_malloc(PATH_MAX);
3837 backing_file_full = g_malloc(PATH_MAX);
3838 filename_tmp = g_malloc(PATH_MAX);
3840 is_protocol = path_has_protocol(backing_file);
3842 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3844 /* If either of the filename paths is actually a protocol, then
3845 * compare unmodified paths; otherwise make paths relative */
3846 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3847 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3848 retval = curr_bs->backing->bs;
3851 /* Also check against the full backing filename for the image */
3852 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
3854 if (local_error == NULL) {
3855 if (strcmp(backing_file, backing_file_full) == 0) {
3856 retval = curr_bs->backing->bs;
3860 error_free(local_error);
3864 /* If not an absolute filename path, make it relative to the current
3865 * image's filename path */
3866 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3869 /* We are going to compare absolute pathnames */
3870 if (!realpath(filename_tmp, filename_full)) {
3874 /* We need to make sure the backing filename we are comparing against
3875 * is relative to the current image filename (or absolute) */
3876 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3877 curr_bs->backing_file);
3879 if (!realpath(filename_tmp, backing_file_full)) {
3883 if (strcmp(backing_file_full, filename_full) == 0) {
3884 retval = curr_bs->backing->bs;
3890 g_free(filename_full);
3891 g_free(backing_file_full);
3892 g_free(filename_tmp);
3896 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3906 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3909 void bdrv_init(void)
3911 module_call_init(MODULE_INIT_BLOCK);
3914 void bdrv_init_with_whitelist(void)
3916 use_bdrv_whitelist = 1;
3920 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3923 Error *local_err = NULL;
3930 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3934 QLIST_FOREACH(child, &bs->children, next) {
3935 bdrv_invalidate_cache(child->bs, &local_err);
3937 error_propagate(errp, local_err);
3942 bs->open_flags &= ~BDRV_O_INACTIVE;
3943 if (bs->drv->bdrv_invalidate_cache) {
3944 bs->drv->bdrv_invalidate_cache(bs, &local_err);
3946 bs->open_flags |= BDRV_O_INACTIVE;
3947 error_propagate(errp, local_err);
3952 ret = refresh_total_sectors(bs, bs->total_sectors);
3954 bs->open_flags |= BDRV_O_INACTIVE;
3955 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3960 void bdrv_invalidate_cache_all(Error **errp)
3962 BlockDriverState *bs;
3963 Error *local_err = NULL;
3964 BdrvNextIterator it;
3966 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3967 AioContext *aio_context = bdrv_get_aio_context(bs);
3969 aio_context_acquire(aio_context);
3970 bdrv_invalidate_cache(bs, &local_err);
3971 aio_context_release(aio_context);
3973 error_propagate(errp, local_err);
3979 static int bdrv_inactivate_recurse(BlockDriverState *bs,
3985 if (!setting_flag && bs->drv->bdrv_inactivate) {
3986 ret = bs->drv->bdrv_inactivate(bs);
3992 QLIST_FOREACH(child, &bs->children, next) {
3993 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4000 bs->open_flags |= BDRV_O_INACTIVE;
4005 int bdrv_inactivate_all(void)
4007 BlockDriverState *bs = NULL;
4008 BdrvNextIterator it;
4012 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4013 aio_context_acquire(bdrv_get_aio_context(bs));
4016 /* We do two passes of inactivation. The first pass calls to drivers'
4017 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4018 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4020 for (pass = 0; pass < 2; pass++) {
4021 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4022 ret = bdrv_inactivate_recurse(bs, pass);
4030 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4031 aio_context_release(bdrv_get_aio_context(bs));
4037 /**************************************************************/
4038 /* removable device support */
4041 * Return TRUE if the media is present
4043 bool bdrv_is_inserted(BlockDriverState *bs)
4045 BlockDriver *drv = bs->drv;
4051 if (drv->bdrv_is_inserted) {
4052 return drv->bdrv_is_inserted(bs);
4054 QLIST_FOREACH(child, &bs->children, next) {
4055 if (!bdrv_is_inserted(child->bs)) {
4063 * Return whether the media changed since the last call to this
4064 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4066 int bdrv_media_changed(BlockDriverState *bs)
4068 BlockDriver *drv = bs->drv;
4070 if (drv && drv->bdrv_media_changed) {
4071 return drv->bdrv_media_changed(bs);
4077 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4079 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4081 BlockDriver *drv = bs->drv;
4083 if (drv && drv->bdrv_eject) {
4084 drv->bdrv_eject(bs, eject_flag);
4089 * Lock or unlock the media (if it is locked, the user won't be able
4090 * to eject it manually).
4092 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4094 BlockDriver *drv = bs->drv;
4096 trace_bdrv_lock_medium(bs, locked);
4098 if (drv && drv->bdrv_lock_medium) {
4099 drv->bdrv_lock_medium(bs, locked);
4103 /* Get a reference to bs */
4104 void bdrv_ref(BlockDriverState *bs)
4109 /* Release a previously grabbed reference to bs.
4110 * If after releasing, reference count is zero, the BlockDriverState is
4112 void bdrv_unref(BlockDriverState *bs)
4117 assert(bs->refcnt > 0);
4118 if (--bs->refcnt == 0) {
4123 struct BdrvOpBlocker {
4125 QLIST_ENTRY(BdrvOpBlocker) list;
4128 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4130 BdrvOpBlocker *blocker;
4131 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4132 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4133 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4135 *errp = error_copy(blocker->reason);
4136 error_prepend(errp, "Node '%s' is busy: ",
4137 bdrv_get_device_or_node_name(bs));
4144 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4146 BdrvOpBlocker *blocker;
4147 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4149 blocker = g_new0(BdrvOpBlocker, 1);
4150 blocker->reason = reason;
4151 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4154 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4156 BdrvOpBlocker *blocker, *next;
4157 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4158 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4159 if (blocker->reason == reason) {
4160 QLIST_REMOVE(blocker, list);
4166 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4169 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4170 bdrv_op_block(bs, i, reason);
4174 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4177 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4178 bdrv_op_unblock(bs, i, reason);
4182 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4186 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4187 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4194 void bdrv_img_create(const char *filename, const char *fmt,
4195 const char *base_filename, const char *base_fmt,
4196 char *options, uint64_t img_size, int flags, bool quiet,
4199 QemuOptsList *create_opts = NULL;
4200 QemuOpts *opts = NULL;
4201 const char *backing_fmt, *backing_file;
4203 BlockDriver *drv, *proto_drv;
4204 Error *local_err = NULL;
4207 /* Find driver and parse its options */
4208 drv = bdrv_find_format(fmt);
4210 error_setg(errp, "Unknown file format '%s'", fmt);
4214 proto_drv = bdrv_find_protocol(filename, true, errp);
4219 if (!drv->create_opts) {
4220 error_setg(errp, "Format driver '%s' does not support image creation",
4225 if (!proto_drv->create_opts) {
4226 error_setg(errp, "Protocol driver '%s' does not support image creation",
4227 proto_drv->format_name);
4231 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4232 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4234 /* Create parameter list with default values */
4235 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4236 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4238 /* Parse -o options */
4240 qemu_opts_do_parse(opts, options, NULL, &local_err);
4242 error_report_err(local_err);
4244 error_setg(errp, "Invalid options for file format '%s'", fmt);
4249 if (base_filename) {
4250 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4252 error_setg(errp, "Backing file not supported for file format '%s'",
4259 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4261 error_setg(errp, "Backing file format not supported for file "
4262 "format '%s'", fmt);
4267 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4269 if (!strcmp(filename, backing_file)) {
4270 error_setg(errp, "Error: Trying to create an image with the "
4271 "same filename as the backing file");
4276 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4278 // The size for the image must always be specified, with one exception:
4279 // If we are using a backing file, we can obtain the size from there
4280 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4283 BlockDriverState *bs;
4284 char *full_backing = g_new0(char, PATH_MAX);
4287 QDict *backing_options = NULL;
4289 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4290 full_backing, PATH_MAX,
4293 g_free(full_backing);
4297 /* backing files always opened read-only */
4299 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4302 backing_options = qdict_new();
4303 qdict_put_str(backing_options, "driver", backing_fmt);
4306 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4308 g_free(full_backing);
4312 size = bdrv_getlength(bs);
4314 error_setg_errno(errp, -size, "Could not get size of '%s'",
4320 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4324 error_setg(errp, "Image creation needs a size parameter");
4330 printf("Formatting '%s', fmt=%s ", filename, fmt);
4331 qemu_opts_print(opts, " ");
4335 ret = bdrv_create(drv, filename, opts, &local_err);
4337 if (ret == -EFBIG) {
4338 /* This is generally a better message than whatever the driver would
4339 * deliver (especially because of the cluster_size_hint), since that
4340 * is most probably not much different from "image too large". */
4341 const char *cluster_size_hint = "";
4342 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4343 cluster_size_hint = " (try using a larger cluster size)";
4345 error_setg(errp, "The image size is too large for file format '%s'"
4346 "%s", fmt, cluster_size_hint);
4347 error_free(local_err);
4352 qemu_opts_del(opts);
4353 qemu_opts_free(create_opts);
4354 error_propagate(errp, local_err);
4357 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4359 return bs->aio_context;
4362 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4364 aio_co_enter(bdrv_get_aio_context(bs), co);
4367 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4369 QLIST_REMOVE(ban, list);
4373 void bdrv_detach_aio_context(BlockDriverState *bs)
4375 BdrvAioNotifier *baf, *baf_tmp;
4382 assert(!bs->walking_aio_notifiers);
4383 bs->walking_aio_notifiers = true;
4384 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4386 bdrv_do_remove_aio_context_notifier(baf);
4388 baf->detach_aio_context(baf->opaque);
4391 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4392 * remove remaining aio notifiers if we aren't called again.
4394 bs->walking_aio_notifiers = false;
4396 if (bs->drv->bdrv_detach_aio_context) {
4397 bs->drv->bdrv_detach_aio_context(bs);
4399 QLIST_FOREACH(child, &bs->children, next) {
4400 bdrv_detach_aio_context(child->bs);
4403 bs->aio_context = NULL;
4406 void bdrv_attach_aio_context(BlockDriverState *bs,
4407 AioContext *new_context)
4409 BdrvAioNotifier *ban, *ban_tmp;
4416 bs->aio_context = new_context;
4418 QLIST_FOREACH(child, &bs->children, next) {
4419 bdrv_attach_aio_context(child->bs, new_context);
4421 if (bs->drv->bdrv_attach_aio_context) {
4422 bs->drv->bdrv_attach_aio_context(bs, new_context);
4425 assert(!bs->walking_aio_notifiers);
4426 bs->walking_aio_notifiers = true;
4427 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4429 bdrv_do_remove_aio_context_notifier(ban);
4431 ban->attached_aio_context(new_context, ban->opaque);
4434 bs->walking_aio_notifiers = false;
4437 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4439 AioContext *ctx = bdrv_get_aio_context(bs);
4441 aio_disable_external(ctx);
4442 bdrv_parent_drained_begin(bs);
4443 bdrv_drain(bs); /* ensure there are no in-flight requests */
4445 while (aio_poll(ctx, false)) {
4446 /* wait for all bottom halves to execute */
4449 bdrv_detach_aio_context(bs);
4451 /* This function executes in the old AioContext so acquire the new one in
4452 * case it runs in a different thread.
4454 aio_context_acquire(new_context);
4455 bdrv_attach_aio_context(bs, new_context);
4456 bdrv_parent_drained_end(bs);
4457 aio_enable_external(ctx);
4458 aio_context_release(new_context);
4461 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4462 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4463 void (*detach_aio_context)(void *opaque), void *opaque)
4465 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4466 *ban = (BdrvAioNotifier){
4467 .attached_aio_context = attached_aio_context,
4468 .detach_aio_context = detach_aio_context,
4472 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4475 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4476 void (*attached_aio_context)(AioContext *,
4478 void (*detach_aio_context)(void *),
4481 BdrvAioNotifier *ban, *ban_next;
4483 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4484 if (ban->attached_aio_context == attached_aio_context &&
4485 ban->detach_aio_context == detach_aio_context &&
4486 ban->opaque == opaque &&
4487 ban->deleted == false)
4489 if (bs->walking_aio_notifiers) {
4490 ban->deleted = true;
4492 bdrv_do_remove_aio_context_notifier(ban);
4501 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4502 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4504 if (!bs->drv->bdrv_amend_options) {
4507 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
4510 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4511 * of block filter and by bdrv_is_first_non_filter.
4512 * It is used to test if the given bs is the candidate or recurse more in the
4515 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
4516 BlockDriverState *candidate)
4518 /* return false if basic checks fails */
4519 if (!bs || !bs->drv) {
4523 /* the code reached a non block filter driver -> check if the bs is
4524 * the same as the candidate. It's the recursion termination condition.
4526 if (!bs->drv->is_filter) {
4527 return bs == candidate;
4529 /* Down this path the driver is a block filter driver */
4531 /* If the block filter recursion method is defined use it to recurse down
4534 if (bs->drv->bdrv_recurse_is_first_non_filter) {
4535 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
4538 /* the driver is a block filter but don't allow to recurse -> return false
4543 /* This function checks if the candidate is the first non filter bs down it's
4544 * bs chain. Since we don't have pointers to parents it explore all bs chains
4545 * from the top. Some filters can choose not to pass down the recursion.
4547 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
4549 BlockDriverState *bs;
4550 BdrvNextIterator it;
4552 /* walk down the bs forest recursively */
4553 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4556 /* try to recurse in this top level bs */
4557 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
4559 /* candidate is the first non filter */
4568 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
4569 const char *node_name, Error **errp)
4571 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
4572 AioContext *aio_context;
4574 if (!to_replace_bs) {
4575 error_setg(errp, "Node name '%s' not found", node_name);
4579 aio_context = bdrv_get_aio_context(to_replace_bs);
4580 aio_context_acquire(aio_context);
4582 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
4583 to_replace_bs = NULL;
4587 /* We don't want arbitrary node of the BDS chain to be replaced only the top
4588 * most non filter in order to prevent data corruption.
4589 * Another benefit is that this tests exclude backing files which are
4590 * blocked by the backing blockers.
4592 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
4593 error_setg(errp, "Only top most non filter can be replaced");
4594 to_replace_bs = NULL;
4599 aio_context_release(aio_context);
4600 return to_replace_bs;
4603 static bool append_open_options(QDict *d, BlockDriverState *bs)
4605 const QDictEntry *entry;
4608 bool found_any = false;
4611 for (entry = qdict_first(bs->options); entry;
4612 entry = qdict_next(bs->options, entry))
4614 /* Exclude options for children */
4615 QLIST_FOREACH(child, &bs->children, next) {
4616 if (strstart(qdict_entry_key(entry), child->name, &p)
4617 && (!*p || *p == '.'))
4626 /* And exclude all non-driver-specific options */
4627 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4628 if (!strcmp(qdict_entry_key(entry), desc->name)) {
4636 qobject_incref(qdict_entry_value(entry));
4637 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
4644 /* Updates the following BDS fields:
4645 * - exact_filename: A filename which may be used for opening a block device
4646 * which (mostly) equals the given BDS (even without any
4647 * other options; so reading and writing must return the same
4648 * results, but caching etc. may be different)
4649 * - full_open_options: Options which, when given when opening a block device
4650 * (without a filename), result in a BDS (mostly)
4651 * equalling the given one
4652 * - filename: If exact_filename is set, it is copied here. Otherwise,
4653 * full_open_options is converted to a JSON object, prefixed with
4654 * "json:" (for use through the JSON pseudo protocol) and put here.
4656 void bdrv_refresh_filename(BlockDriverState *bs)
4658 BlockDriver *drv = bs->drv;
4665 /* This BDS's file name will most probably depend on its file's name, so
4666 * refresh that first */
4668 bdrv_refresh_filename(bs->file->bs);
4671 if (drv->bdrv_refresh_filename) {
4672 /* Obsolete information is of no use here, so drop the old file name
4673 * information before refreshing it */
4674 bs->exact_filename[0] = '\0';
4675 if (bs->full_open_options) {
4676 QDECREF(bs->full_open_options);
4677 bs->full_open_options = NULL;
4681 append_open_options(opts, bs);
4682 drv->bdrv_refresh_filename(bs, opts);
4684 } else if (bs->file) {
4685 /* Try to reconstruct valid information from the underlying file */
4686 bool has_open_options;
4688 bs->exact_filename[0] = '\0';
4689 if (bs->full_open_options) {
4690 QDECREF(bs->full_open_options);
4691 bs->full_open_options = NULL;
4695 has_open_options = append_open_options(opts, bs);
4697 /* If no specific options have been given for this BDS, the filename of
4698 * the underlying file should suffice for this one as well */
4699 if (bs->file->bs->exact_filename[0] && !has_open_options) {
4700 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
4702 /* Reconstructing the full options QDict is simple for most format block
4703 * drivers, as long as the full options are known for the underlying
4704 * file BDS. The full options QDict of that file BDS should somehow
4705 * contain a representation of the filename, therefore the following
4706 * suffices without querying the (exact_)filename of this BDS. */
4707 if (bs->file->bs->full_open_options) {
4708 qdict_put_str(opts, "driver", drv->format_name);
4709 QINCREF(bs->file->bs->full_open_options);
4710 qdict_put(opts, "file", bs->file->bs->full_open_options);
4712 bs->full_open_options = opts;
4716 } else if (!bs->full_open_options && qdict_size(bs->options)) {
4717 /* There is no underlying file BDS (at least referenced by BDS.file),
4718 * so the full options QDict should be equal to the options given
4719 * specifically for this block device when it was opened (plus the
4720 * driver specification).
4721 * Because those options don't change, there is no need to update
4722 * full_open_options when it's already set. */
4725 append_open_options(opts, bs);
4726 qdict_put_str(opts, "driver", drv->format_name);
4728 if (bs->exact_filename[0]) {
4729 /* This may not work for all block protocol drivers (some may
4730 * require this filename to be parsed), but we have to find some
4731 * default solution here, so just include it. If some block driver
4732 * does not support pure options without any filename at all or
4733 * needs some special format of the options QDict, it needs to
4734 * implement the driver-specific bdrv_refresh_filename() function.
4736 qdict_put_str(opts, "filename", bs->exact_filename);
4739 bs->full_open_options = opts;
4742 if (bs->exact_filename[0]) {
4743 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
4744 } else if (bs->full_open_options) {
4745 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
4746 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
4747 qstring_get_str(json));
4753 * Hot add/remove a BDS's child. So the user can take a child offline when
4754 * it is broken and take a new child online
4756 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
4760 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
4761 error_setg(errp, "The node %s does not support adding a child",
4762 bdrv_get_device_or_node_name(parent_bs));
4766 if (!QLIST_EMPTY(&child_bs->parents)) {
4767 error_setg(errp, "The node %s already has a parent",
4768 child_bs->node_name);
4772 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
4775 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
4779 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
4780 error_setg(errp, "The node %s does not support removing a child",
4781 bdrv_get_device_or_node_name(parent_bs));
4785 QLIST_FOREACH(tmp, &parent_bs->children, next) {
4792 error_setg(errp, "The node %s does not have a child named %s",
4793 bdrv_get_device_or_node_name(parent_bs),
4794 bdrv_get_device_or_node_name(child->bs));
4798 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);