2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "block/module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
51 #include "block/coroutines.h"
54 #include <sys/ioctl.h>
55 #include <sys/queue.h>
65 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
67 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
68 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
70 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
71 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
73 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
74 QLIST_HEAD_INITIALIZER(bdrv_drivers);
76 static BlockDriverState *bdrv_open_inherit(const char *filename,
77 const char *reference,
78 QDict *options, int flags,
79 BlockDriverState *parent,
80 const BdrvChildClass *child_class,
81 BdrvChildRole child_role,
84 /* If non-zero, use only whitelisted block drivers */
85 static int use_bdrv_whitelist;
88 static int is_windows_drive_prefix(const char *filename)
90 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
91 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
95 int is_windows_drive(const char *filename)
97 if (is_windows_drive_prefix(filename) &&
100 if (strstart(filename, "\\\\.\\", NULL) ||
101 strstart(filename, "//./", NULL))
107 size_t bdrv_opt_mem_align(BlockDriverState *bs)
109 if (!bs || !bs->drv) {
110 /* page size or 4k (hdd sector size) should be on the safe side */
111 return MAX(4096, qemu_real_host_page_size);
114 return bs->bl.opt_mem_alignment;
117 size_t bdrv_min_mem_align(BlockDriverState *bs)
119 if (!bs || !bs->drv) {
120 /* page size or 4k (hdd sector size) should be on the safe side */
121 return MAX(4096, qemu_real_host_page_size);
124 return bs->bl.min_mem_alignment;
127 /* check if the path starts with "<protocol>:" */
128 int path_has_protocol(const char *path)
133 if (is_windows_drive(path) ||
134 is_windows_drive_prefix(path)) {
137 p = path + strcspn(path, ":/\\");
139 p = path + strcspn(path, ":/");
145 int path_is_absolute(const char *path)
148 /* specific case for names like: "\\.\d:" */
149 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
152 return (*path == '/' || *path == '\\');
154 return (*path == '/');
158 /* if filename is absolute, just return its duplicate. Otherwise, build a
159 path to it by considering it is relative to base_path. URL are
161 char *path_combine(const char *base_path, const char *filename)
163 const char *protocol_stripped = NULL;
168 if (path_is_absolute(filename)) {
169 return g_strdup(filename);
172 if (path_has_protocol(base_path)) {
173 protocol_stripped = strchr(base_path, ':');
174 if (protocol_stripped) {
178 p = protocol_stripped ?: base_path;
180 p1 = strrchr(base_path, '/');
184 p2 = strrchr(base_path, '\\');
185 if (!p1 || p2 > p1) {
200 result = g_malloc(len + strlen(filename) + 1);
201 memcpy(result, base_path, len);
202 strcpy(result + len, filename);
208 * Helper function for bdrv_parse_filename() implementations to remove optional
209 * protocol prefixes (especially "file:") from a filename and for putting the
210 * stripped filename into the options QDict if there is such a prefix.
212 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
215 if (strstart(filename, prefix, &filename)) {
216 /* Stripping the explicit protocol prefix may result in a protocol
217 * prefix being (wrongly) detected (if the filename contains a colon) */
218 if (path_has_protocol(filename)) {
219 QString *fat_filename;
221 /* This means there is some colon before the first slash; therefore,
222 * this cannot be an absolute path */
223 assert(!path_is_absolute(filename));
225 /* And we can thus fix the protocol detection issue by prefixing it
227 fat_filename = qstring_from_str("./");
228 qstring_append(fat_filename, filename);
230 assert(!path_has_protocol(qstring_get_str(fat_filename)));
232 qdict_put(options, "filename", fat_filename);
234 /* If no protocol prefix was detected, we can use the shortened
236 qdict_put_str(options, "filename", filename);
242 /* Returns whether the image file is opened as read-only. Note that this can
243 * return false and writing to the image file is still not possible because the
244 * image is inactivated. */
245 bool bdrv_is_read_only(BlockDriverState *bs)
247 return bs->read_only;
250 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
251 bool ignore_allow_rdw, Error **errp)
253 /* Do not set read_only if copy_on_read is enabled */
254 if (bs->copy_on_read && read_only) {
255 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
256 bdrv_get_device_or_node_name(bs));
260 /* Do not clear read_only if it is prohibited */
261 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
264 error_setg(errp, "Node '%s' is read only",
265 bdrv_get_device_or_node_name(bs));
273 * Called by a driver that can only provide a read-only image.
275 * Returns 0 if the node is already read-only or it could switch the node to
276 * read-only because BDRV_O_AUTO_RDONLY is set.
278 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
279 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
280 * is not NULL, it is used as the error message for the Error object.
282 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
287 if (!(bs->open_flags & BDRV_O_RDWR)) {
290 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
294 ret = bdrv_can_set_read_only(bs, true, false, NULL);
299 bs->read_only = true;
300 bs->open_flags &= ~BDRV_O_RDWR;
305 error_setg(errp, "%s", errmsg ?: "Image is read-only");
310 * If @backing is empty, this function returns NULL without setting
311 * @errp. In all other cases, NULL will only be returned with @errp
314 * Therefore, a return value of NULL without @errp set means that
315 * there is no backing file; if @errp is set, there is one but its
316 * absolute filename cannot be generated.
318 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
322 if (backing[0] == '\0') {
324 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
325 return g_strdup(backing);
326 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
327 error_setg(errp, "Cannot use relative backing file names for '%s'",
331 return path_combine(backed, backing);
336 * If @filename is empty or NULL, this function returns NULL without
337 * setting @errp. In all other cases, NULL will only be returned with
340 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
341 const char *filename, Error **errp)
343 char *dir, *full_name;
345 if (!filename || filename[0] == '\0') {
347 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
348 return g_strdup(filename);
351 dir = bdrv_dirname(relative_to, errp);
356 full_name = g_strconcat(dir, filename, NULL);
361 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
363 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
366 void bdrv_register(BlockDriver *bdrv)
368 assert(bdrv->format_name);
369 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
372 BlockDriverState *bdrv_new(void)
374 BlockDriverState *bs;
377 bs = g_new0(BlockDriverState, 1);
378 QLIST_INIT(&bs->dirty_bitmaps);
379 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
380 QLIST_INIT(&bs->op_blockers[i]);
382 notifier_with_return_list_init(&bs->before_write_notifiers);
383 qemu_co_mutex_init(&bs->reqs_lock);
384 qemu_mutex_init(&bs->dirty_bitmap_mutex);
386 bs->aio_context = qemu_get_aio_context();
388 qemu_co_queue_init(&bs->flush_queue);
390 for (i = 0; i < bdrv_drain_all_count; i++) {
391 bdrv_drained_begin(bs);
394 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
399 static BlockDriver *bdrv_do_find_format(const char *format_name)
403 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
404 if (!strcmp(drv1->format_name, format_name)) {
412 BlockDriver *bdrv_find_format(const char *format_name)
417 drv1 = bdrv_do_find_format(format_name);
422 /* The driver isn't registered, maybe we need to load a module */
423 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
424 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
425 block_module_load_one(block_driver_modules[i].library_name);
430 return bdrv_do_find_format(format_name);
433 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
435 static const char *whitelist_rw[] = {
436 CONFIG_BDRV_RW_WHITELIST
439 static const char *whitelist_ro[] = {
440 CONFIG_BDRV_RO_WHITELIST
445 if (!whitelist_rw[0] && !whitelist_ro[0]) {
446 return 1; /* no whitelist, anything goes */
449 for (p = whitelist_rw; *p; p++) {
450 if (!strcmp(format_name, *p)) {
455 for (p = whitelist_ro; *p; p++) {
456 if (!strcmp(format_name, *p)) {
464 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
466 return bdrv_format_is_whitelisted(drv->format_name, read_only);
469 bool bdrv_uses_whitelist(void)
471 return use_bdrv_whitelist;
474 typedef struct CreateCo {
482 static void coroutine_fn bdrv_create_co_entry(void *opaque)
484 Error *local_err = NULL;
487 CreateCo *cco = opaque;
490 ret = cco->drv->bdrv_co_create_opts(cco->drv,
491 cco->filename, cco->opts, &local_err);
492 error_propagate(&cco->err, local_err);
496 int bdrv_create(BlockDriver *drv, const char* filename,
497 QemuOpts *opts, Error **errp)
504 .filename = g_strdup(filename),
510 if (!drv->bdrv_co_create_opts) {
511 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
516 if (qemu_in_coroutine()) {
517 /* Fast-path if already in coroutine context */
518 bdrv_create_co_entry(&cco);
520 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
521 qemu_coroutine_enter(co);
522 while (cco.ret == NOT_DONE) {
523 aio_poll(qemu_get_aio_context(), true);
530 error_propagate(errp, cco.err);
532 error_setg_errno(errp, -ret, "Could not create image");
537 g_free(cco.filename);
542 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
543 * least the given @minimum_size.
545 * On success, return @blk's actual length.
546 * Otherwise, return -errno.
548 static int64_t create_file_fallback_truncate(BlockBackend *blk,
549 int64_t minimum_size, Error **errp)
551 Error *local_err = NULL;
555 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
557 if (ret < 0 && ret != -ENOTSUP) {
558 error_propagate(errp, local_err);
562 size = blk_getlength(blk);
564 error_free(local_err);
565 error_setg_errno(errp, -size,
566 "Failed to inquire the new image file's length");
570 if (size < minimum_size) {
571 /* Need to grow the image, but we failed to do that */
572 error_propagate(errp, local_err);
576 error_free(local_err);
583 * Helper function for bdrv_create_file_fallback(): Zero the first
584 * sector to remove any potentially pre-existing image header.
586 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
587 int64_t current_size,
590 int64_t bytes_to_clear;
593 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
594 if (bytes_to_clear) {
595 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
597 error_setg_errno(errp, -ret,
598 "Failed to clear the new image's first sector");
607 * Simple implementation of bdrv_co_create_opts for protocol drivers
608 * which only support creation via opening a file
609 * (usually existing raw storage device)
611 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
612 const char *filename,
620 PreallocMode prealloc;
621 Error *local_err = NULL;
624 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
625 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
626 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
627 PREALLOC_MODE_OFF, &local_err);
630 error_propagate(errp, local_err);
634 if (prealloc != PREALLOC_MODE_OFF) {
635 error_setg(errp, "Unsupported preallocation mode '%s'",
636 PreallocMode_str(prealloc));
640 options = qdict_new();
641 qdict_put_str(options, "driver", drv->format_name);
643 blk = blk_new_open(filename, NULL, options,
644 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
646 error_prepend(errp, "Protocol driver '%s' does not support image "
647 "creation, and opening the image failed: ",
652 size = create_file_fallback_truncate(blk, size, errp);
658 ret = create_file_fallback_zero_first_sector(blk, size, errp);
669 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
673 drv = bdrv_find_protocol(filename, true, errp);
678 return bdrv_create(drv, filename, opts, errp);
681 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
683 Error *local_err = NULL;
689 error_setg(errp, "Block node '%s' is not opened", bs->filename);
693 if (!bs->drv->bdrv_co_delete_file) {
694 error_setg(errp, "Driver '%s' does not support image deletion",
695 bs->drv->format_name);
699 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
701 error_propagate(errp, local_err);
708 * Try to get @bs's logical and physical block size.
709 * On success, store them in @bsz struct and return 0.
710 * On failure return -errno.
711 * @bs must not be empty.
713 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
715 BlockDriver *drv = bs->drv;
716 BlockDriverState *filtered = bdrv_filter_bs(bs);
718 if (drv && drv->bdrv_probe_blocksizes) {
719 return drv->bdrv_probe_blocksizes(bs, bsz);
720 } else if (filtered) {
721 return bdrv_probe_blocksizes(filtered, bsz);
728 * Try to get @bs's geometry (cyls, heads, sectors).
729 * On success, store them in @geo struct and return 0.
730 * On failure return -errno.
731 * @bs must not be empty.
733 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
735 BlockDriver *drv = bs->drv;
736 BlockDriverState *filtered = bdrv_filter_bs(bs);
738 if (drv && drv->bdrv_probe_geometry) {
739 return drv->bdrv_probe_geometry(bs, geo);
740 } else if (filtered) {
741 return bdrv_probe_geometry(filtered, geo);
748 * Create a uniquely-named empty temporary file.
749 * Return 0 upon success, otherwise a negative errno value.
751 int get_tmp_filename(char *filename, int size)
754 char temp_dir[MAX_PATH];
755 /* GetTempFileName requires that its output buffer (4th param)
756 have length MAX_PATH or greater. */
757 assert(size >= MAX_PATH);
758 return (GetTempPath(MAX_PATH, temp_dir)
759 && GetTempFileName(temp_dir, "qem", 0, filename)
760 ? 0 : -GetLastError());
764 tmpdir = getenv("TMPDIR");
768 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
771 fd = mkstemp(filename);
775 if (close(fd) != 0) {
784 * Detect host devices. By convention, /dev/cdrom[N] is always
785 * recognized as a host CDROM.
787 static BlockDriver *find_hdev_driver(const char *filename)
789 int score_max = 0, score;
790 BlockDriver *drv = NULL, *d;
792 QLIST_FOREACH(d, &bdrv_drivers, list) {
793 if (d->bdrv_probe_device) {
794 score = d->bdrv_probe_device(filename);
795 if (score > score_max) {
805 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
809 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
810 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
818 BlockDriver *bdrv_find_protocol(const char *filename,
819 bool allow_protocol_prefix,
828 /* TODO Drivers without bdrv_file_open must be specified explicitly */
831 * XXX(hch): we really should not let host device detection
832 * override an explicit protocol specification, but moving this
833 * later breaks access to device names with colons in them.
834 * Thanks to the brain-dead persistent naming schemes on udev-
835 * based Linux systems those actually are quite common.
837 drv1 = find_hdev_driver(filename);
842 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
846 p = strchr(filename, ':');
849 if (len > sizeof(protocol) - 1)
850 len = sizeof(protocol) - 1;
851 memcpy(protocol, filename, len);
852 protocol[len] = '\0';
854 drv1 = bdrv_do_find_protocol(protocol);
859 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
860 if (block_driver_modules[i].protocol_name &&
861 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
862 block_module_load_one(block_driver_modules[i].library_name);
867 drv1 = bdrv_do_find_protocol(protocol);
869 error_setg(errp, "Unknown protocol '%s'", protocol);
875 * Guess image format by probing its contents.
876 * This is not a good idea when your image is raw (CVE-2008-2004), but
877 * we do it anyway for backward compatibility.
879 * @buf contains the image's first @buf_size bytes.
880 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
881 * but can be smaller if the image file is smaller)
882 * @filename is its filename.
884 * For all block drivers, call the bdrv_probe() method to get its
886 * Return the first block driver with the highest probing score.
888 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
889 const char *filename)
891 int score_max = 0, score;
892 BlockDriver *drv = NULL, *d;
894 QLIST_FOREACH(d, &bdrv_drivers, list) {
896 score = d->bdrv_probe(buf, buf_size, filename);
897 if (score > score_max) {
907 static int find_image_format(BlockBackend *file, const char *filename,
908 BlockDriver **pdrv, Error **errp)
911 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
914 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
915 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
920 ret = blk_pread(file, 0, buf, sizeof(buf));
922 error_setg_errno(errp, -ret, "Could not read image for determining its "
928 drv = bdrv_probe_all(buf, ret, filename);
930 error_setg(errp, "Could not determine image format: No compatible "
939 * Set the current 'total_sectors' value
940 * Return 0 on success, -errno on error.
942 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
944 BlockDriver *drv = bs->drv;
950 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
954 /* query actual device if possible, otherwise just trust the hint */
955 if (drv->bdrv_getlength) {
956 int64_t length = drv->bdrv_getlength(bs);
960 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
963 bs->total_sectors = hint;
968 * Combines a QDict of new block driver @options with any missing options taken
969 * from @old_options, so that leaving out an option defaults to its old value.
971 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
974 if (bs->drv && bs->drv->bdrv_join_options) {
975 bs->drv->bdrv_join_options(options, old_options);
977 qdict_join(options, old_options, false);
981 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
985 Error *local_err = NULL;
986 char *value = qemu_opt_get_del(opts, "detect-zeroes");
987 BlockdevDetectZeroesOptions detect_zeroes =
988 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
989 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
992 error_propagate(errp, local_err);
993 return detect_zeroes;
996 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
997 !(open_flags & BDRV_O_UNMAP))
999 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1000 "without setting discard operation to unmap");
1003 return detect_zeroes;
1007 * Set open flags for aio engine
1009 * Return 0 on success, -1 if the engine specified is invalid
1011 int bdrv_parse_aio(const char *mode, int *flags)
1013 if (!strcmp(mode, "threads")) {
1014 /* do nothing, default */
1015 } else if (!strcmp(mode, "native")) {
1016 *flags |= BDRV_O_NATIVE_AIO;
1017 #ifdef CONFIG_LINUX_IO_URING
1018 } else if (!strcmp(mode, "io_uring")) {
1019 *flags |= BDRV_O_IO_URING;
1029 * Set open flags for a given discard mode
1031 * Return 0 on success, -1 if the discard mode was invalid.
1033 int bdrv_parse_discard_flags(const char *mode, int *flags)
1035 *flags &= ~BDRV_O_UNMAP;
1037 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1039 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1040 *flags |= BDRV_O_UNMAP;
1049 * Set open flags for a given cache mode
1051 * Return 0 on success, -1 if the cache mode was invalid.
1053 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1055 *flags &= ~BDRV_O_CACHE_MASK;
1057 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1058 *writethrough = false;
1059 *flags |= BDRV_O_NOCACHE;
1060 } else if (!strcmp(mode, "directsync")) {
1061 *writethrough = true;
1062 *flags |= BDRV_O_NOCACHE;
1063 } else if (!strcmp(mode, "writeback")) {
1064 *writethrough = false;
1065 } else if (!strcmp(mode, "unsafe")) {
1066 *writethrough = false;
1067 *flags |= BDRV_O_NO_FLUSH;
1068 } else if (!strcmp(mode, "writethrough")) {
1069 *writethrough = true;
1077 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1079 BlockDriverState *parent = c->opaque;
1080 return g_strdup(bdrv_get_device_or_node_name(parent));
1083 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1085 BlockDriverState *bs = child->opaque;
1086 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1089 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1091 BlockDriverState *bs = child->opaque;
1092 return bdrv_drain_poll(bs, false, NULL, false);
1095 static void bdrv_child_cb_drained_end(BdrvChild *child,
1096 int *drained_end_counter)
1098 BlockDriverState *bs = child->opaque;
1099 bdrv_drained_end_no_poll(bs, drained_end_counter);
1102 static int bdrv_child_cb_inactivate(BdrvChild *child)
1104 BlockDriverState *bs = child->opaque;
1105 assert(bs->open_flags & BDRV_O_INACTIVE);
1109 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1110 GSList **ignore, Error **errp)
1112 BlockDriverState *bs = child->opaque;
1113 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1116 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1119 BlockDriverState *bs = child->opaque;
1120 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1124 * Returns the options and flags that a temporary snapshot should get, based on
1125 * the originally requested flags (the originally requested image will have
1126 * flags like a backing file)
1128 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1129 int parent_flags, QDict *parent_options)
1131 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1133 /* For temporary files, unconditional cache=unsafe is fine */
1134 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1135 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1137 /* Copy the read-only and discard options from the parent */
1138 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1139 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1141 /* aio=native doesn't work for cache.direct=off, so disable it for the
1142 * temporary snapshot */
1143 *child_flags &= ~BDRV_O_NATIVE_AIO;
1146 static void bdrv_backing_attach(BdrvChild *c)
1148 BlockDriverState *parent = c->opaque;
1149 BlockDriverState *backing_hd = c->bs;
1151 assert(!parent->backing_blocker);
1152 error_setg(&parent->backing_blocker,
1153 "node is used as backing hd of '%s'",
1154 bdrv_get_device_or_node_name(parent));
1156 bdrv_refresh_filename(backing_hd);
1158 parent->open_flags &= ~BDRV_O_NO_BACKING;
1160 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1161 /* Otherwise we won't be able to commit or stream */
1162 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1163 parent->backing_blocker);
1164 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1165 parent->backing_blocker);
1167 * We do backup in 3 ways:
1169 * The target bs is new opened, and the source is top BDS
1170 * 2. blockdev backup
1171 * Both the source and the target are top BDSes.
1172 * 3. internal backup(used for block replication)
1173 * Both the source and the target are backing file
1175 * In case 1 and 2, neither the source nor the target is the backing file.
1176 * In case 3, we will block the top BDS, so there is only one block job
1177 * for the top BDS and its backing chain.
1179 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1180 parent->backing_blocker);
1181 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1182 parent->backing_blocker);
1185 static void bdrv_backing_detach(BdrvChild *c)
1187 BlockDriverState *parent = c->opaque;
1189 assert(parent->backing_blocker);
1190 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1191 error_free(parent->backing_blocker);
1192 parent->backing_blocker = NULL;
1195 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1196 const char *filename, Error **errp)
1198 BlockDriverState *parent = c->opaque;
1199 bool read_only = bdrv_is_read_only(parent);
1203 ret = bdrv_reopen_set_read_only(parent, false, errp);
1209 ret = bdrv_change_backing_file(parent, filename,
1210 base->drv ? base->drv->format_name : "",
1213 error_setg_errno(errp, -ret, "Could not update backing file link");
1217 bdrv_reopen_set_read_only(parent, true, NULL);
1224 * Returns the options and flags that a generic child of a BDS should
1225 * get, based on the given options and flags for the parent BDS.
1227 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1228 int *child_flags, QDict *child_options,
1229 int parent_flags, QDict *parent_options)
1231 int flags = parent_flags;
1234 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1235 * Generally, the question to answer is: Should this child be
1236 * format-probed by default?
1240 * Pure and non-filtered data children of non-format nodes should
1241 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1242 * set). This only affects a very limited set of drivers (namely
1243 * quorum and blkverify when this comment was written).
1244 * Force-clear BDRV_O_PROTOCOL then.
1246 if (!parent_is_format &&
1247 (role & BDRV_CHILD_DATA) &&
1248 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1250 flags &= ~BDRV_O_PROTOCOL;
1254 * All children of format nodes (except for COW children) and all
1255 * metadata children in general should never be format-probed.
1256 * Force-set BDRV_O_PROTOCOL then.
1258 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1259 (role & BDRV_CHILD_METADATA))
1261 flags |= BDRV_O_PROTOCOL;
1265 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1268 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1269 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1270 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1272 if (role & BDRV_CHILD_COW) {
1273 /* backing files are opened read-only by default */
1274 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1275 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1277 /* Inherit the read-only option from the parent if it's not set */
1278 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1279 qdict_copy_default(child_options, parent_options,
1280 BDRV_OPT_AUTO_READ_ONLY);
1284 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1285 * can default to enable it on lower layers regardless of the
1288 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1290 /* Clear flags that only apply to the top layer */
1291 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1293 if (role & BDRV_CHILD_METADATA) {
1294 flags &= ~BDRV_O_NO_IO;
1296 if (role & BDRV_CHILD_COW) {
1297 flags &= ~BDRV_O_TEMPORARY;
1300 *child_flags = flags;
1303 static void bdrv_child_cb_attach(BdrvChild *child)
1305 BlockDriverState *bs = child->opaque;
1307 if (child->role & BDRV_CHILD_COW) {
1308 bdrv_backing_attach(child);
1311 bdrv_apply_subtree_drain(child, bs);
1314 static void bdrv_child_cb_detach(BdrvChild *child)
1316 BlockDriverState *bs = child->opaque;
1318 if (child->role & BDRV_CHILD_COW) {
1319 bdrv_backing_detach(child);
1322 bdrv_unapply_subtree_drain(child, bs);
1325 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1326 const char *filename, Error **errp)
1328 if (c->role & BDRV_CHILD_COW) {
1329 return bdrv_backing_update_filename(c, base, filename, errp);
1334 const BdrvChildClass child_of_bds = {
1335 .parent_is_bds = true,
1336 .get_parent_desc = bdrv_child_get_parent_desc,
1337 .inherit_options = bdrv_inherited_options,
1338 .drained_begin = bdrv_child_cb_drained_begin,
1339 .drained_poll = bdrv_child_cb_drained_poll,
1340 .drained_end = bdrv_child_cb_drained_end,
1341 .attach = bdrv_child_cb_attach,
1342 .detach = bdrv_child_cb_detach,
1343 .inactivate = bdrv_child_cb_inactivate,
1344 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1345 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1346 .update_filename = bdrv_child_cb_update_filename,
1349 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1351 int open_flags = flags;
1354 * Clear flags that are internal to the block layer before opening the
1357 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1362 static void update_flags_from_options(int *flags, QemuOpts *opts)
1364 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1366 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1367 *flags |= BDRV_O_NO_FLUSH;
1370 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1371 *flags |= BDRV_O_NOCACHE;
1374 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1375 *flags |= BDRV_O_RDWR;
1378 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1379 *flags |= BDRV_O_AUTO_RDONLY;
1383 static void update_options_from_flags(QDict *options, int flags)
1385 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1386 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1388 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1389 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1390 flags & BDRV_O_NO_FLUSH);
1392 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1393 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1395 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1396 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1397 flags & BDRV_O_AUTO_RDONLY);
1401 static void bdrv_assign_node_name(BlockDriverState *bs,
1402 const char *node_name,
1405 char *gen_node_name = NULL;
1408 node_name = gen_node_name = id_generate(ID_BLOCK);
1409 } else if (!id_wellformed(node_name)) {
1411 * Check for empty string or invalid characters, but not if it is
1412 * generated (generated names use characters not available to the user)
1414 error_setg(errp, "Invalid node name");
1418 /* takes care of avoiding namespaces collisions */
1419 if (blk_by_name(node_name)) {
1420 error_setg(errp, "node-name=%s is conflicting with a device id",
1425 /* takes care of avoiding duplicates node names */
1426 if (bdrv_find_node(node_name)) {
1427 error_setg(errp, "Duplicate node name");
1431 /* Make sure that the node name isn't truncated */
1432 if (strlen(node_name) >= sizeof(bs->node_name)) {
1433 error_setg(errp, "Node name too long");
1437 /* copy node name into the bs and insert it into the graph list */
1438 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1439 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1441 g_free(gen_node_name);
1444 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1445 const char *node_name, QDict *options,
1446 int open_flags, Error **errp)
1448 Error *local_err = NULL;
1451 bdrv_assign_node_name(bs, node_name, &local_err);
1453 error_propagate(errp, local_err);
1458 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1459 bs->opaque = g_malloc0(drv->instance_size);
1461 if (drv->bdrv_file_open) {
1462 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1463 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1464 } else if (drv->bdrv_open) {
1465 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1472 error_propagate(errp, local_err);
1473 } else if (bs->filename[0]) {
1474 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1476 error_setg_errno(errp, -ret, "Could not open image");
1481 ret = refresh_total_sectors(bs, bs->total_sectors);
1483 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1487 bdrv_refresh_limits(bs, &local_err);
1489 error_propagate(errp, local_err);
1493 assert(bdrv_opt_mem_align(bs) != 0);
1494 assert(bdrv_min_mem_align(bs) != 0);
1495 assert(is_power_of_2(bs->bl.request_alignment));
1497 for (i = 0; i < bs->quiesce_counter; i++) {
1498 if (drv->bdrv_co_drain_begin) {
1499 drv->bdrv_co_drain_begin(bs);
1506 if (bs->file != NULL) {
1507 bdrv_unref_child(bs, bs->file);
1515 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1516 int flags, Error **errp)
1518 BlockDriverState *bs;
1522 bs->open_flags = flags;
1523 bs->explicit_options = qdict_new();
1524 bs->options = qdict_new();
1527 update_options_from_flags(bs->options, flags);
1529 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1531 qobject_unref(bs->explicit_options);
1532 bs->explicit_options = NULL;
1533 qobject_unref(bs->options);
1542 QemuOptsList bdrv_runtime_opts = {
1543 .name = "bdrv_common",
1544 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1547 .name = "node-name",
1548 .type = QEMU_OPT_STRING,
1549 .help = "Node name of the block device node",
1553 .type = QEMU_OPT_STRING,
1554 .help = "Block driver to use for the node",
1557 .name = BDRV_OPT_CACHE_DIRECT,
1558 .type = QEMU_OPT_BOOL,
1559 .help = "Bypass software writeback cache on the host",
1562 .name = BDRV_OPT_CACHE_NO_FLUSH,
1563 .type = QEMU_OPT_BOOL,
1564 .help = "Ignore flush requests",
1567 .name = BDRV_OPT_READ_ONLY,
1568 .type = QEMU_OPT_BOOL,
1569 .help = "Node is opened in read-only mode",
1572 .name = BDRV_OPT_AUTO_READ_ONLY,
1573 .type = QEMU_OPT_BOOL,
1574 .help = "Node can become read-only if opening read-write fails",
1577 .name = "detect-zeroes",
1578 .type = QEMU_OPT_STRING,
1579 .help = "try to optimize zero writes (off, on, unmap)",
1582 .name = BDRV_OPT_DISCARD,
1583 .type = QEMU_OPT_STRING,
1584 .help = "discard operation (ignore/off, unmap/on)",
1587 .name = BDRV_OPT_FORCE_SHARE,
1588 .type = QEMU_OPT_BOOL,
1589 .help = "always accept other writers (default: off)",
1591 { /* end of list */ }
1595 QemuOptsList bdrv_create_opts_simple = {
1596 .name = "simple-create-opts",
1597 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1600 .name = BLOCK_OPT_SIZE,
1601 .type = QEMU_OPT_SIZE,
1602 .help = "Virtual disk size"
1605 .name = BLOCK_OPT_PREALLOC,
1606 .type = QEMU_OPT_STRING,
1607 .help = "Preallocation mode (allowed values: off)"
1609 { /* end of list */ }
1614 * Common part for opening disk images and files
1616 * Removes all processed options from *options.
1618 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1619 QDict *options, Error **errp)
1621 int ret, open_flags;
1622 const char *filename;
1623 const char *driver_name = NULL;
1624 const char *node_name = NULL;
1625 const char *discard;
1628 Error *local_err = NULL;
1630 assert(bs->file == NULL);
1631 assert(options != NULL && bs->options != options);
1633 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1634 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1639 update_flags_from_options(&bs->open_flags, opts);
1641 driver_name = qemu_opt_get(opts, "driver");
1642 drv = bdrv_find_format(driver_name);
1643 assert(drv != NULL);
1645 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1647 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1649 BDRV_OPT_FORCE_SHARE
1650 "=on can only be used with read-only images");
1656 bdrv_refresh_filename(blk_bs(file));
1657 filename = blk_bs(file)->filename;
1660 * Caution: while qdict_get_try_str() is fine, getting
1661 * non-string types would require more care. When @options
1662 * come from -blockdev or blockdev_add, its members are typed
1663 * according to the QAPI schema, but when they come from
1664 * -drive, they're all QString.
1666 filename = qdict_get_try_str(options, "filename");
1669 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1670 error_setg(errp, "The '%s' block driver requires a file name",
1676 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1679 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1681 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1682 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1683 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1689 !bs->read_only && bdrv_is_whitelisted(drv, true)
1690 ? "Driver '%s' can only be used for read-only devices"
1691 : "Driver '%s' is not whitelisted",
1697 /* bdrv_new() and bdrv_close() make it so */
1698 assert(qatomic_read(&bs->copy_on_read) == 0);
1700 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1701 if (!bs->read_only) {
1702 bdrv_enable_copy_on_read(bs);
1704 error_setg(errp, "Can't use copy-on-read on read-only device");
1710 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1711 if (discard != NULL) {
1712 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1713 error_setg(errp, "Invalid discard option");
1720 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1722 error_propagate(errp, local_err);
1727 if (filename != NULL) {
1728 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1730 bs->filename[0] = '\0';
1732 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1734 /* Open the image, either directly or using a protocol */
1735 open_flags = bdrv_open_flags(bs, bs->open_flags);
1736 node_name = qemu_opt_get(opts, "node-name");
1738 assert(!drv->bdrv_file_open || file == NULL);
1739 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1744 qemu_opts_del(opts);
1748 qemu_opts_del(opts);
1752 static QDict *parse_json_filename(const char *filename, Error **errp)
1754 QObject *options_obj;
1758 ret = strstart(filename, "json:", &filename);
1761 options_obj = qobject_from_json(filename, errp);
1763 error_prepend(errp, "Could not parse the JSON options: ");
1767 options = qobject_to(QDict, options_obj);
1769 qobject_unref(options_obj);
1770 error_setg(errp, "Invalid JSON object given");
1774 qdict_flatten(options);
1779 static void parse_json_protocol(QDict *options, const char **pfilename,
1782 QDict *json_options;
1783 Error *local_err = NULL;
1785 /* Parse json: pseudo-protocol */
1786 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1790 json_options = parse_json_filename(*pfilename, &local_err);
1792 error_propagate(errp, local_err);
1796 /* Options given in the filename have lower priority than options
1797 * specified directly */
1798 qdict_join(options, json_options, false);
1799 qobject_unref(json_options);
1804 * Fills in default options for opening images and converts the legacy
1805 * filename/flags pair to option QDict entries.
1806 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1807 * block driver has been specified explicitly.
1809 static int bdrv_fill_options(QDict **options, const char *filename,
1810 int *flags, Error **errp)
1812 const char *drvname;
1813 bool protocol = *flags & BDRV_O_PROTOCOL;
1814 bool parse_filename = false;
1815 BlockDriver *drv = NULL;
1816 Error *local_err = NULL;
1819 * Caution: while qdict_get_try_str() is fine, getting non-string
1820 * types would require more care. When @options come from
1821 * -blockdev or blockdev_add, its members are typed according to
1822 * the QAPI schema, but when they come from -drive, they're all
1825 drvname = qdict_get_try_str(*options, "driver");
1827 drv = bdrv_find_format(drvname);
1829 error_setg(errp, "Unknown driver '%s'", drvname);
1832 /* If the user has explicitly specified the driver, this choice should
1833 * override the BDRV_O_PROTOCOL flag */
1834 protocol = drv->bdrv_file_open;
1838 *flags |= BDRV_O_PROTOCOL;
1840 *flags &= ~BDRV_O_PROTOCOL;
1843 /* Translate cache options from flags into options */
1844 update_options_from_flags(*options, *flags);
1846 /* Fetch the file name from the options QDict if necessary */
1847 if (protocol && filename) {
1848 if (!qdict_haskey(*options, "filename")) {
1849 qdict_put_str(*options, "filename", filename);
1850 parse_filename = true;
1852 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1858 /* Find the right block driver */
1859 /* See cautionary note on accessing @options above */
1860 filename = qdict_get_try_str(*options, "filename");
1862 if (!drvname && protocol) {
1864 drv = bdrv_find_protocol(filename, parse_filename, errp);
1869 drvname = drv->format_name;
1870 qdict_put_str(*options, "driver", drvname);
1872 error_setg(errp, "Must specify either driver or file");
1877 assert(drv || !protocol);
1879 /* Driver-specific filename parsing */
1880 if (drv && drv->bdrv_parse_filename && parse_filename) {
1881 drv->bdrv_parse_filename(filename, *options, &local_err);
1883 error_propagate(errp, local_err);
1887 if (!drv->bdrv_needs_filename) {
1888 qdict_del(*options, "filename");
1895 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1896 uint64_t perm, uint64_t shared,
1897 GSList *ignore_children,
1898 bool *tighten_restrictions, Error **errp);
1899 static void bdrv_child_abort_perm_update(BdrvChild *c);
1900 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1902 typedef struct BlockReopenQueueEntry {
1905 BDRVReopenState state;
1906 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1907 } BlockReopenQueueEntry;
1910 * Return the flags that @bs will have after the reopens in @q have
1911 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1912 * return the current flags.
1914 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1916 BlockReopenQueueEntry *entry;
1919 QTAILQ_FOREACH(entry, q, entry) {
1920 if (entry->state.bs == bs) {
1921 return entry->state.flags;
1926 return bs->open_flags;
1929 /* Returns whether the image file can be written to after the reopen queue @q
1930 * has been successfully applied, or right now if @q is NULL. */
1931 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1932 BlockReopenQueue *q)
1934 int flags = bdrv_reopen_get_flags(q, bs);
1936 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1940 * Return whether the BDS can be written to. This is not necessarily
1941 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1942 * be written to but do not count as read-only images.
1944 bool bdrv_is_writable(BlockDriverState *bs)
1946 return bdrv_is_writable_after_reopen(bs, NULL);
1949 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1950 BdrvChild *c, BdrvChildRole role,
1951 BlockReopenQueue *reopen_queue,
1952 uint64_t parent_perm, uint64_t parent_shared,
1953 uint64_t *nperm, uint64_t *nshared)
1955 assert(bs->drv && bs->drv->bdrv_child_perm);
1956 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1957 parent_perm, parent_shared,
1959 /* TODO Take force_share from reopen_queue */
1960 if (child_bs && child_bs->force_share) {
1961 *nshared = BLK_PERM_ALL;
1966 * Check whether permissions on this node can be changed in a way that
1967 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1968 * permissions of all its parents. This involves checking whether all necessary
1969 * permission changes to child nodes can be performed.
1971 * Will set *tighten_restrictions to true if and only if new permissions have to
1972 * be taken or currently shared permissions are to be unshared. Otherwise,
1973 * errors are not fatal as long as the caller accepts that the restrictions
1974 * remain tighter than they need to be. The caller still has to abort the
1976 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1977 * encounter fatal errors even though no restrictions are to be tightened. For
1978 * example, changing a node from RW to RO will fail if the WRITE permission is
1981 * A call to this function must always be followed by a call to bdrv_set_perm()
1982 * or bdrv_abort_perm_update().
1984 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1985 uint64_t cumulative_perms,
1986 uint64_t cumulative_shared_perms,
1987 GSList *ignore_children,
1988 bool *tighten_restrictions, Error **errp)
1990 BlockDriver *drv = bs->drv;
1994 assert(!q || !tighten_restrictions);
1996 if (tighten_restrictions) {
1997 uint64_t current_perms, current_shared;
1998 uint64_t added_perms, removed_shared_perms;
2000 bdrv_get_cumulative_perm(bs, ¤t_perms, ¤t_shared);
2002 added_perms = cumulative_perms & ~current_perms;
2003 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2005 *tighten_restrictions = added_perms || removed_shared_perms;
2008 /* Write permissions never work with read-only images */
2009 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2010 !bdrv_is_writable_after_reopen(bs, q))
2012 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2013 error_setg(errp, "Block node is read-only");
2015 uint64_t current_perms, current_shared;
2016 bdrv_get_cumulative_perm(bs, ¤t_perms, ¤t_shared);
2017 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2018 error_setg(errp, "Cannot make block node read-only, there is "
2021 error_setg(errp, "Cannot make block node read-only and create "
2030 * Unaligned requests will automatically be aligned to bl.request_alignment
2031 * and without RESIZE we can't extend requests to write to space beyond the
2032 * end of the image, so it's required that the image size is aligned.
2034 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2035 !(cumulative_perms & BLK_PERM_RESIZE))
2037 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2038 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2039 "Image size is not a multiple of request "
2045 /* Check this node */
2050 if (drv->bdrv_check_perm) {
2051 return drv->bdrv_check_perm(bs, cumulative_perms,
2052 cumulative_shared_perms, errp);
2055 /* Drivers that never have children can omit .bdrv_child_perm() */
2056 if (!drv->bdrv_child_perm) {
2057 assert(QLIST_EMPTY(&bs->children));
2061 /* Check all children */
2062 QLIST_FOREACH(c, &bs->children, next) {
2063 uint64_t cur_perm, cur_shared;
2064 bool child_tighten_restr;
2066 bdrv_child_perm(bs, c->bs, c, c->role, q,
2067 cumulative_perms, cumulative_shared_perms,
2068 &cur_perm, &cur_shared);
2069 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2070 tighten_restrictions ? &child_tighten_restr
2073 if (tighten_restrictions) {
2074 *tighten_restrictions |= child_tighten_restr;
2085 * Notifies drivers that after a previous bdrv_check_perm() call, the
2086 * permission update is not performed and any preparations made for it (e.g.
2087 * taken file locks) need to be undone.
2089 * This function recursively notifies all child nodes.
2091 static void bdrv_abort_perm_update(BlockDriverState *bs)
2093 BlockDriver *drv = bs->drv;
2100 if (drv->bdrv_abort_perm_update) {
2101 drv->bdrv_abort_perm_update(bs);
2104 QLIST_FOREACH(c, &bs->children, next) {
2105 bdrv_child_abort_perm_update(c);
2109 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2110 uint64_t cumulative_shared_perms)
2112 BlockDriver *drv = bs->drv;
2119 /* Update this node */
2120 if (drv->bdrv_set_perm) {
2121 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2124 /* Drivers that never have children can omit .bdrv_child_perm() */
2125 if (!drv->bdrv_child_perm) {
2126 assert(QLIST_EMPTY(&bs->children));
2130 /* Update all children */
2131 QLIST_FOREACH(c, &bs->children, next) {
2132 uint64_t cur_perm, cur_shared;
2133 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2134 cumulative_perms, cumulative_shared_perms,
2135 &cur_perm, &cur_shared);
2136 bdrv_child_set_perm(c, cur_perm, cur_shared);
2140 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2141 uint64_t *shared_perm)
2144 uint64_t cumulative_perms = 0;
2145 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2147 QLIST_FOREACH(c, &bs->parents, next_parent) {
2148 cumulative_perms |= c->perm;
2149 cumulative_shared_perms &= c->shared_perm;
2152 *perm = cumulative_perms;
2153 *shared_perm = cumulative_shared_perms;
2156 static char *bdrv_child_user_desc(BdrvChild *c)
2158 if (c->klass->get_parent_desc) {
2159 return c->klass->get_parent_desc(c);
2162 return g_strdup("another user");
2165 char *bdrv_perm_names(uint64_t perm)
2171 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2172 { BLK_PERM_WRITE, "write" },
2173 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2174 { BLK_PERM_RESIZE, "resize" },
2175 { BLK_PERM_GRAPH_MOD, "change children" },
2179 GString *result = g_string_sized_new(30);
2180 struct perm_name *p;
2182 for (p = permissions; p->name; p++) {
2183 if (perm & p->perm) {
2184 if (result->len > 0) {
2185 g_string_append(result, ", ");
2187 g_string_append(result, p->name);
2191 return g_string_free(result, FALSE);
2195 * Checks whether a new reference to @bs can be added if the new user requires
2196 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2197 * set, the BdrvChild objects in this list are ignored in the calculations;
2198 * this allows checking permission updates for an existing reference.
2200 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2202 * Needs to be followed by a call to either bdrv_set_perm() or
2203 * bdrv_abort_perm_update(). */
2204 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2205 uint64_t new_used_perm,
2206 uint64_t new_shared_perm,
2207 GSList *ignore_children,
2208 bool *tighten_restrictions,
2212 uint64_t cumulative_perms = new_used_perm;
2213 uint64_t cumulative_shared_perms = new_shared_perm;
2215 assert(!q || !tighten_restrictions);
2217 /* There is no reason why anyone couldn't tolerate write_unchanged */
2218 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2220 QLIST_FOREACH(c, &bs->parents, next_parent) {
2221 if (g_slist_find(ignore_children, c)) {
2225 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2226 char *user = bdrv_child_user_desc(c);
2227 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2229 if (tighten_restrictions) {
2230 *tighten_restrictions = true;
2233 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2235 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2241 if ((c->perm & new_shared_perm) != c->perm) {
2242 char *user = bdrv_child_user_desc(c);
2243 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2245 if (tighten_restrictions) {
2246 *tighten_restrictions = true;
2249 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2251 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2257 cumulative_perms |= c->perm;
2258 cumulative_shared_perms &= c->shared_perm;
2261 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2262 ignore_children, tighten_restrictions, errp);
2265 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2266 * bdrv_child_abort_perm_update(). */
2267 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2268 uint64_t perm, uint64_t shared,
2269 GSList *ignore_children,
2270 bool *tighten_restrictions, Error **errp)
2274 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2275 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2276 tighten_restrictions, errp);
2277 g_slist_free(ignore_children);
2283 if (!c->has_backup_perm) {
2284 c->has_backup_perm = true;
2285 c->backup_perm = c->perm;
2286 c->backup_shared_perm = c->shared_perm;
2289 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2290 * same child twice during check_perm procedure
2294 c->shared_perm = shared;
2299 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2301 uint64_t cumulative_perms, cumulative_shared_perms;
2303 c->has_backup_perm = false;
2306 c->shared_perm = shared;
2308 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2309 &cumulative_shared_perms);
2310 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2313 static void bdrv_child_abort_perm_update(BdrvChild *c)
2315 if (c->has_backup_perm) {
2316 c->perm = c->backup_perm;
2317 c->shared_perm = c->backup_shared_perm;
2318 c->has_backup_perm = false;
2321 bdrv_abort_perm_update(c->bs);
2324 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2327 Error *local_err = NULL;
2329 bool tighten_restrictions;
2331 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2332 &tighten_restrictions, &local_err);
2334 bdrv_child_abort_perm_update(c);
2335 if (tighten_restrictions) {
2336 error_propagate(errp, local_err);
2339 * Our caller may intend to only loosen restrictions and
2340 * does not expect this function to fail. Errors are not
2341 * fatal in such a case, so we can just hide them from our
2344 error_free(local_err);
2350 bdrv_child_set_perm(c, perm, shared);
2355 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2357 uint64_t parent_perms, parent_shared;
2358 uint64_t perms, shared;
2360 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2361 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2362 parent_perms, parent_shared, &perms, &shared);
2364 return bdrv_child_try_set_perm(c, perms, shared, errp);
2368 * Default implementation for .bdrv_child_perm() for block filters:
2369 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2372 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2374 BlockReopenQueue *reopen_queue,
2375 uint64_t perm, uint64_t shared,
2376 uint64_t *nperm, uint64_t *nshared)
2378 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2379 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2382 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2384 BlockReopenQueue *reopen_queue,
2385 uint64_t perm, uint64_t shared,
2386 uint64_t *nperm, uint64_t *nshared)
2388 assert(role & BDRV_CHILD_COW);
2391 * We want consistent read from backing files if the parent needs it.
2392 * No other operations are performed on backing files.
2394 perm &= BLK_PERM_CONSISTENT_READ;
2397 * If the parent can deal with changing data, we're okay with a
2398 * writable and resizable backing file.
2399 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2401 if (shared & BLK_PERM_WRITE) {
2402 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2407 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2408 BLK_PERM_WRITE_UNCHANGED;
2410 if (bs->open_flags & BDRV_O_INACTIVE) {
2411 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2418 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2420 BlockReopenQueue *reopen_queue,
2421 uint64_t perm, uint64_t shared,
2422 uint64_t *nperm, uint64_t *nshared)
2426 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2428 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2431 * Apart from the modifications below, the same permissions are
2432 * forwarded and left alone as for filters
2434 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2435 perm, shared, &perm, &shared);
2437 if (role & BDRV_CHILD_METADATA) {
2438 /* Format drivers may touch metadata even if the guest doesn't write */
2439 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2440 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2444 * bs->file always needs to be consistent because of the
2445 * metadata. We can never allow other users to resize or write
2448 if (!(flags & BDRV_O_NO_IO)) {
2449 perm |= BLK_PERM_CONSISTENT_READ;
2451 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2454 if (role & BDRV_CHILD_DATA) {
2456 * Technically, everything in this block is a subset of the
2457 * BDRV_CHILD_METADATA path taken above, and so this could
2458 * be an "else if" branch. However, that is not obvious, and
2459 * this function is not performance critical, therefore we let
2460 * this be an independent "if".
2464 * We cannot allow other users to resize the file because the
2465 * format driver might have some assumptions about the size
2466 * (e.g. because it is stored in metadata, or because the file
2467 * is split into fixed-size data files).
2469 shared &= ~BLK_PERM_RESIZE;
2472 * WRITE_UNCHANGED often cannot be performed as such on the
2473 * data file. For example, the qcow2 driver may still need to
2474 * write copied clusters on copy-on-read.
2476 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2477 perm |= BLK_PERM_WRITE;
2481 * If the data file is written to, the format driver may
2482 * expect to be able to resize it by writing beyond the EOF.
2484 if (perm & BLK_PERM_WRITE) {
2485 perm |= BLK_PERM_RESIZE;
2489 if (bs->open_flags & BDRV_O_INACTIVE) {
2490 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2497 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2498 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2499 uint64_t perm, uint64_t shared,
2500 uint64_t *nperm, uint64_t *nshared)
2502 if (role & BDRV_CHILD_FILTERED) {
2503 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2505 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2506 perm, shared, nperm, nshared);
2507 } else if (role & BDRV_CHILD_COW) {
2508 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2509 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2510 perm, shared, nperm, nshared);
2511 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2512 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2513 perm, shared, nperm, nshared);
2515 g_assert_not_reached();
2519 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2521 static const uint64_t permissions[] = {
2522 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2523 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2524 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2525 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2526 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2529 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2530 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2532 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2534 return permissions[qapi_perm];
2537 static void bdrv_replace_child_noperm(BdrvChild *child,
2538 BlockDriverState *new_bs)
2540 BlockDriverState *old_bs = child->bs;
2541 int new_bs_quiesce_counter;
2544 assert(!child->frozen);
2546 if (old_bs && new_bs) {
2547 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2550 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2551 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2554 * If the new child node is drained but the old one was not, flush
2555 * all outstanding requests to the old child node.
2557 while (drain_saldo > 0 && child->klass->drained_begin) {
2558 bdrv_parent_drained_begin_single(child, true);
2563 /* Detach first so that the recursive drain sections coming from @child
2564 * are already gone and we only end the drain sections that came from
2566 if (child->klass->detach) {
2567 child->klass->detach(child);
2569 QLIST_REMOVE(child, next_parent);
2575 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2578 * Detaching the old node may have led to the new node's
2579 * quiesce_counter having been decreased. Not a problem, we
2580 * just need to recognize this here and then invoke
2581 * drained_end appropriately more often.
2583 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2584 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2586 /* Attach only after starting new drained sections, so that recursive
2587 * drain sections coming from @child don't get an extra .drained_begin
2589 if (child->klass->attach) {
2590 child->klass->attach(child);
2595 * If the old child node was drained but the new one is not, allow
2596 * requests to come in only after the new node has been attached.
2598 while (drain_saldo < 0 && child->klass->drained_end) {
2599 bdrv_parent_drained_end_single(child);
2605 * Updates @child to change its reference to point to @new_bs, including
2606 * checking and applying the necessary permission updates both to the old node
2609 * NULL is passed as @new_bs for removing the reference before freeing @child.
2611 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2612 * function uses bdrv_set_perm() to update the permissions according to the new
2613 * reference that @new_bs gets.
2615 * Callers must ensure that child->frozen is false.
2617 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2619 BlockDriverState *old_bs = child->bs;
2620 uint64_t perm, shared_perm;
2622 /* Asserts that child->frozen == false */
2623 bdrv_replace_child_noperm(child, new_bs);
2626 * Start with the new node's permissions. If @new_bs is a (direct
2627 * or indirect) child of @old_bs, we must complete the permission
2628 * update on @new_bs before we loosen the restrictions on @old_bs.
2629 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2630 * updating the permissions of @new_bs, and thus not purely loosen
2634 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2635 bdrv_set_perm(new_bs, perm, shared_perm);
2639 /* Update permissions for old node. This is guaranteed to succeed
2640 * because we're just taking a parent away, so we're loosening
2642 bool tighten_restrictions;
2645 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2646 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2647 &tighten_restrictions, NULL);
2648 assert(tighten_restrictions == false);
2650 /* We only tried to loosen restrictions, so errors are not fatal */
2651 bdrv_abort_perm_update(old_bs);
2653 bdrv_set_perm(old_bs, perm, shared_perm);
2656 /* When the parent requiring a non-default AioContext is removed, the
2657 * node moves back to the main AioContext */
2658 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2663 * This function steals the reference to child_bs from the caller.
2664 * That reference is later dropped by bdrv_root_unref_child().
2666 * On failure NULL is returned, errp is set and the reference to
2667 * child_bs is also dropped.
2669 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2670 * (unless @child_bs is already in @ctx).
2672 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2673 const char *child_name,
2674 const BdrvChildClass *child_class,
2675 BdrvChildRole child_role,
2677 uint64_t perm, uint64_t shared_perm,
2678 void *opaque, Error **errp)
2681 Error *local_err = NULL;
2684 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2687 bdrv_abort_perm_update(child_bs);
2688 bdrv_unref(child_bs);
2692 child = g_new(BdrvChild, 1);
2693 *child = (BdrvChild) {
2695 .name = g_strdup(child_name),
2696 .klass = child_class,
2699 .shared_perm = shared_perm,
2703 /* If the AioContexts don't match, first try to move the subtree of
2704 * child_bs into the AioContext of the new parent. If this doesn't work,
2705 * try moving the parent into the AioContext of child_bs instead. */
2706 if (bdrv_get_aio_context(child_bs) != ctx) {
2707 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2708 if (ret < 0 && child_class->can_set_aio_ctx) {
2709 GSList *ignore = g_slist_prepend(NULL, child);
2710 ctx = bdrv_get_aio_context(child_bs);
2711 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2712 error_free(local_err);
2714 g_slist_free(ignore);
2715 ignore = g_slist_prepend(NULL, child);
2716 child_class->set_aio_ctx(child, ctx, &ignore);
2718 g_slist_free(ignore);
2721 error_propagate(errp, local_err);
2723 bdrv_abort_perm_update(child_bs);
2724 bdrv_unref(child_bs);
2729 /* This performs the matching bdrv_set_perm() for the above check. */
2730 bdrv_replace_child(child, child_bs);
2736 * This function transfers the reference to child_bs from the caller
2737 * to parent_bs. That reference is later dropped by parent_bs on
2738 * bdrv_close() or if someone calls bdrv_unref_child().
2740 * On failure NULL is returned, errp is set and the reference to
2741 * child_bs is also dropped.
2743 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2744 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2746 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2747 BlockDriverState *child_bs,
2748 const char *child_name,
2749 const BdrvChildClass *child_class,
2750 BdrvChildRole child_role,
2754 uint64_t perm, shared_perm;
2756 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2758 assert(parent_bs->drv);
2759 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2760 perm, shared_perm, &perm, &shared_perm);
2762 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2763 child_role, bdrv_get_aio_context(parent_bs),
2764 perm, shared_perm, parent_bs, errp);
2765 if (child == NULL) {
2769 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2773 static void bdrv_detach_child(BdrvChild *child)
2775 QLIST_SAFE_REMOVE(child, next);
2777 bdrv_replace_child(child, NULL);
2779 g_free(child->name);
2783 /* Callers must ensure that child->frozen is false. */
2784 void bdrv_root_unref_child(BdrvChild *child)
2786 BlockDriverState *child_bs;
2788 child_bs = child->bs;
2789 bdrv_detach_child(child);
2790 bdrv_unref(child_bs);
2794 * Clear all inherits_from pointers from children and grandchildren of
2795 * @root that point to @root, where necessary.
2797 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2801 if (child->bs->inherits_from == root) {
2803 * Remove inherits_from only when the last reference between root and
2804 * child->bs goes away.
2806 QLIST_FOREACH(c, &root->children, next) {
2807 if (c != child && c->bs == child->bs) {
2812 child->bs->inherits_from = NULL;
2816 QLIST_FOREACH(c, &child->bs->children, next) {
2817 bdrv_unset_inherits_from(root, c);
2821 /* Callers must ensure that child->frozen is false. */
2822 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2824 if (child == NULL) {
2828 bdrv_unset_inherits_from(parent, child);
2829 bdrv_root_unref_child(child);
2833 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2836 QLIST_FOREACH(c, &bs->parents, next_parent) {
2837 if (c->klass->change_media) {
2838 c->klass->change_media(c, load);
2843 /* Return true if you can reach parent going through child->inherits_from
2844 * recursively. If parent or child are NULL, return false */
2845 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2846 BlockDriverState *parent)
2848 while (child && child != parent) {
2849 child = child->inherits_from;
2852 return child != NULL;
2856 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2857 * mostly used for COW backing children (role = COW), but also for
2858 * filtered children (role = FILTERED | PRIMARY).
2860 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2862 if (bs->drv && bs->drv->is_filter) {
2863 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2865 return BDRV_CHILD_COW;
2870 * Sets the bs->backing link of a BDS. A new reference is created; callers
2871 * which don't need their own reference any more must call bdrv_unref().
2873 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2876 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2877 bdrv_inherits_from_recursive(backing_hd, bs);
2879 if (bdrv_is_backing_chain_frozen(bs, child_bs(bs->backing), errp)) {
2884 bdrv_ref(backing_hd);
2888 /* Cannot be frozen, we checked that above */
2889 bdrv_unref_child(bs, bs->backing);
2897 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2898 bdrv_backing_role(bs), errp);
2899 /* If backing_hd was already part of bs's backing chain, and
2900 * inherits_from pointed recursively to bs then let's update it to
2901 * point directly to bs (else it will become NULL). */
2902 if (bs->backing && update_inherits_from) {
2903 backing_hd->inherits_from = bs;
2907 bdrv_refresh_limits(bs, NULL);
2911 * Opens the backing file for a BlockDriverState if not yet open
2913 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2914 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2915 * itself, all options starting with "${bdref_key}." are considered part of the
2918 * TODO Can this be unified with bdrv_open_image()?
2920 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2921 const char *bdref_key, Error **errp)
2923 char *backing_filename = NULL;
2924 char *bdref_key_dot;
2925 const char *reference = NULL;
2927 bool implicit_backing = false;
2928 BlockDriverState *backing_hd;
2930 QDict *tmp_parent_options = NULL;
2931 Error *local_err = NULL;
2933 if (bs->backing != NULL) {
2937 /* NULL means an empty set of options */
2938 if (parent_options == NULL) {
2939 tmp_parent_options = qdict_new();
2940 parent_options = tmp_parent_options;
2943 bs->open_flags &= ~BDRV_O_NO_BACKING;
2945 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2946 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2947 g_free(bdref_key_dot);
2950 * Caution: while qdict_get_try_str() is fine, getting non-string
2951 * types would require more care. When @parent_options come from
2952 * -blockdev or blockdev_add, its members are typed according to
2953 * the QAPI schema, but when they come from -drive, they're all
2956 reference = qdict_get_try_str(parent_options, bdref_key);
2957 if (reference || qdict_haskey(options, "file.filename")) {
2958 /* keep backing_filename NULL */
2959 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2960 qobject_unref(options);
2963 if (qdict_size(options) == 0) {
2964 /* If the user specifies options that do not modify the
2965 * backing file's behavior, we might still consider it the
2966 * implicit backing file. But it's easier this way, and
2967 * just specifying some of the backing BDS's options is
2968 * only possible with -drive anyway (otherwise the QAPI
2969 * schema forces the user to specify everything). */
2970 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2973 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2976 error_propagate(errp, local_err);
2977 qobject_unref(options);
2982 if (!bs->drv || !bs->drv->supports_backing) {
2984 error_setg(errp, "Driver doesn't support backing files");
2985 qobject_unref(options);
2990 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2991 qdict_put_str(options, "driver", bs->backing_format);
2994 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2995 &child_of_bds, bdrv_backing_role(bs), errp);
2997 bs->open_flags |= BDRV_O_NO_BACKING;
2998 error_prepend(errp, "Could not open backing file: ");
3003 if (implicit_backing) {
3004 bdrv_refresh_filename(backing_hd);
3005 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3006 backing_hd->filename);
3009 /* Hook up the backing file link; drop our reference, bs owns the
3010 * backing_hd reference now */
3011 bdrv_set_backing_hd(bs, backing_hd, &local_err);
3012 bdrv_unref(backing_hd);
3014 error_propagate(errp, local_err);
3019 qdict_del(parent_options, bdref_key);
3022 g_free(backing_filename);
3023 qobject_unref(tmp_parent_options);
3027 static BlockDriverState *
3028 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3029 BlockDriverState *parent, const BdrvChildClass *child_class,
3030 BdrvChildRole child_role, bool allow_none, Error **errp)
3032 BlockDriverState *bs = NULL;
3033 QDict *image_options;
3034 char *bdref_key_dot;
3035 const char *reference;
3037 assert(child_class != NULL);
3039 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3040 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3041 g_free(bdref_key_dot);
3044 * Caution: while qdict_get_try_str() is fine, getting non-string
3045 * types would require more care. When @options come from
3046 * -blockdev or blockdev_add, its members are typed according to
3047 * the QAPI schema, but when they come from -drive, they're all
3050 reference = qdict_get_try_str(options, bdref_key);
3051 if (!filename && !reference && !qdict_size(image_options)) {
3053 error_setg(errp, "A block device must be specified for \"%s\"",
3056 qobject_unref(image_options);
3060 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3061 parent, child_class, child_role, errp);
3067 qdict_del(options, bdref_key);
3072 * Opens a disk image whose options are given as BlockdevRef in another block
3075 * If allow_none is true, no image will be opened if filename is false and no
3076 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3078 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3079 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3080 * itself, all options starting with "${bdref_key}." are considered part of the
3083 * The BlockdevRef will be removed from the options QDict.
3085 BdrvChild *bdrv_open_child(const char *filename,
3086 QDict *options, const char *bdref_key,
3087 BlockDriverState *parent,
3088 const BdrvChildClass *child_class,
3089 BdrvChildRole child_role,
3090 bool allow_none, Error **errp)
3092 BlockDriverState *bs;
3094 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3095 child_role, allow_none, errp);
3100 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3105 * TODO Future callers may need to specify parent/child_class in order for
3106 * option inheritance to work. Existing callers use it for the root node.
3108 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3110 BlockDriverState *bs = NULL;
3111 QObject *obj = NULL;
3112 QDict *qdict = NULL;
3113 const char *reference = NULL;
3116 if (ref->type == QTYPE_QSTRING) {
3117 reference = ref->u.reference;
3119 BlockdevOptions *options = &ref->u.definition;
3120 assert(ref->type == QTYPE_QDICT);
3122 v = qobject_output_visitor_new(&obj);
3123 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3124 visit_complete(v, &obj);
3126 qdict = qobject_to(QDict, obj);
3127 qdict_flatten(qdict);
3129 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3130 * compatibility with other callers) rather than what we want as the
3131 * real defaults. Apply the defaults here instead. */
3132 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3133 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3134 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3135 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3139 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3146 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3148 QDict *snapshot_options,
3151 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3152 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3154 QemuOpts *opts = NULL;
3155 BlockDriverState *bs_snapshot = NULL;
3156 Error *local_err = NULL;
3159 /* if snapshot, we create a temporary backing file and open it
3160 instead of opening 'filename' directly */
3162 /* Get the required size from the image */
3163 total_size = bdrv_getlength(bs);
3164 if (total_size < 0) {
3165 error_setg_errno(errp, -total_size, "Could not get image size");
3169 /* Create the temporary image */
3170 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3172 error_setg_errno(errp, -ret, "Could not get temporary filename");
3176 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3178 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3179 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3180 qemu_opts_del(opts);
3182 error_prepend(errp, "Could not create temporary overlay '%s': ",
3187 /* Prepare options QDict for the temporary file */
3188 qdict_put_str(snapshot_options, "file.driver", "file");
3189 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3190 qdict_put_str(snapshot_options, "driver", "qcow2");
3192 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3193 snapshot_options = NULL;
3198 /* bdrv_append() consumes a strong reference to bs_snapshot
3199 * (i.e. it will call bdrv_unref() on it) even on error, so in
3200 * order to be able to return one, we have to increase
3201 * bs_snapshot's refcount here */
3202 bdrv_ref(bs_snapshot);
3203 bdrv_append(bs_snapshot, bs, &local_err);
3205 error_propagate(errp, local_err);
3211 qobject_unref(snapshot_options);
3212 g_free(tmp_filename);
3217 * Opens a disk image (raw, qcow2, vmdk, ...)
3219 * options is a QDict of options to pass to the block drivers, or NULL for an
3220 * empty set of options. The reference to the QDict belongs to the block layer
3221 * after the call (even on failure), so if the caller intends to reuse the
3222 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3224 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3225 * If it is not NULL, the referenced BDS will be reused.
3227 * The reference parameter may be used to specify an existing block device which
3228 * should be opened. If specified, neither options nor a filename may be given,
3229 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3231 static BlockDriverState *bdrv_open_inherit(const char *filename,
3232 const char *reference,
3233 QDict *options, int flags,
3234 BlockDriverState *parent,
3235 const BdrvChildClass *child_class,
3236 BdrvChildRole child_role,
3240 BlockBackend *file = NULL;
3241 BlockDriverState *bs;
3242 BlockDriver *drv = NULL;
3244 const char *drvname;
3245 const char *backing;
3246 Error *local_err = NULL;
3247 QDict *snapshot_options = NULL;
3248 int snapshot_flags = 0;
3250 assert(!child_class || !flags);
3251 assert(!child_class == !parent);
3254 bool options_non_empty = options ? qdict_size(options) : false;
3255 qobject_unref(options);
3257 if (filename || options_non_empty) {
3258 error_setg(errp, "Cannot reference an existing block device with "
3259 "additional options or a new filename");
3263 bs = bdrv_lookup_bs(reference, reference, errp);
3274 /* NULL means an empty set of options */
3275 if (options == NULL) {
3276 options = qdict_new();
3279 /* json: syntax counts as explicit options, as if in the QDict */
3280 parse_json_protocol(options, &filename, &local_err);
3285 bs->explicit_options = qdict_clone_shallow(options);
3288 bool parent_is_format;
3291 parent_is_format = parent->drv->is_format;
3294 * parent->drv is not set yet because this node is opened for
3295 * (potential) format probing. That means that @parent is going
3296 * to be a format node.
3298 parent_is_format = true;
3301 bs->inherits_from = parent;
3302 child_class->inherit_options(child_role, parent_is_format,
3304 parent->open_flags, parent->options);
3307 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3313 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3314 * Caution: getting a boolean member of @options requires care.
3315 * When @options come from -blockdev or blockdev_add, members are
3316 * typed according to the QAPI schema, but when they come from
3317 * -drive, they're all QString.
3319 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3320 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3321 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3323 flags &= ~BDRV_O_RDWR;
3326 if (flags & BDRV_O_SNAPSHOT) {
3327 snapshot_options = qdict_new();
3328 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3330 /* Let bdrv_backing_options() override "read-only" */
3331 qdict_del(options, BDRV_OPT_READ_ONLY);
3332 bdrv_inherited_options(BDRV_CHILD_COW, true,
3333 &flags, options, flags, options);
3336 bs->open_flags = flags;
3337 bs->options = options;
3338 options = qdict_clone_shallow(options);
3340 /* Find the right image format driver */
3341 /* See cautionary note on accessing @options above */
3342 drvname = qdict_get_try_str(options, "driver");
3344 drv = bdrv_find_format(drvname);
3346 error_setg(errp, "Unknown driver: '%s'", drvname);
3351 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3353 /* See cautionary note on accessing @options above */
3354 backing = qdict_get_try_str(options, "backing");
3355 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3356 (backing && *backing == '\0'))
3359 warn_report("Use of \"backing\": \"\" is deprecated; "
3360 "use \"backing\": null instead");
3362 flags |= BDRV_O_NO_BACKING;
3363 qdict_del(bs->explicit_options, "backing");
3364 qdict_del(bs->options, "backing");
3365 qdict_del(options, "backing");
3368 /* Open image file without format layer. This BlockBackend is only used for
3369 * probing, the block drivers will do their own bdrv_open_child() for the
3370 * same BDS, which is why we put the node name back into options. */
3371 if ((flags & BDRV_O_PROTOCOL) == 0) {
3372 BlockDriverState *file_bs;
3374 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3375 &child_of_bds, BDRV_CHILD_IMAGE,
3380 if (file_bs != NULL) {
3381 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3382 * looking at the header to guess the image format. This works even
3383 * in cases where a guest would not see a consistent state. */
3384 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3385 blk_insert_bs(file, file_bs, &local_err);
3386 bdrv_unref(file_bs);
3391 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3395 /* Image format probing */
3398 ret = find_image_format(file, filename, &drv, &local_err);
3403 * This option update would logically belong in bdrv_fill_options(),
3404 * but we first need to open bs->file for the probing to work, while
3405 * opening bs->file already requires the (mostly) final set of options
3406 * so that cache mode etc. can be inherited.
3408 * Adding the driver later is somewhat ugly, but it's not an option
3409 * that would ever be inherited, so it's correct. We just need to make
3410 * sure to update both bs->options (which has the full effective
3411 * options for bs) and options (which has file.* already removed).
3413 qdict_put_str(bs->options, "driver", drv->format_name);
3414 qdict_put_str(options, "driver", drv->format_name);
3416 error_setg(errp, "Must specify either driver or file");
3420 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3421 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3422 /* file must be NULL if a protocol BDS is about to be created
3423 * (the inverse results in an error message from bdrv_open_common()) */
3424 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3426 /* Open the image */
3427 ret = bdrv_open_common(bs, file, options, &local_err);
3437 /* If there is a backing file, use it */
3438 if ((flags & BDRV_O_NO_BACKING) == 0) {
3439 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3441 goto close_and_fail;
3445 /* Remove all children options and references
3446 * from bs->options and bs->explicit_options */
3447 QLIST_FOREACH(child, &bs->children, next) {
3448 char *child_key_dot;
3449 child_key_dot = g_strdup_printf("%s.", child->name);
3450 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3451 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3452 qdict_del(bs->explicit_options, child->name);
3453 qdict_del(bs->options, child->name);
3454 g_free(child_key_dot);
3457 /* Check if any unknown options were used */
3458 if (qdict_size(options) != 0) {
3459 const QDictEntry *entry = qdict_first(options);
3460 if (flags & BDRV_O_PROTOCOL) {
3461 error_setg(errp, "Block protocol '%s' doesn't support the option "
3462 "'%s'", drv->format_name, entry->key);
3465 "Block format '%s' does not support the option '%s'",
3466 drv->format_name, entry->key);
3469 goto close_and_fail;
3472 bdrv_parent_cb_change_media(bs, true);
3474 qobject_unref(options);
3477 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3478 * temporary snapshot afterwards. */
3479 if (snapshot_flags) {
3480 BlockDriverState *snapshot_bs;
3481 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3482 snapshot_options, &local_err);
3483 snapshot_options = NULL;
3485 goto close_and_fail;
3487 /* We are not going to return bs but the overlay on top of it
3488 * (snapshot_bs); thus, we have to drop the strong reference to bs
3489 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3490 * though, because the overlay still has a reference to it. */
3499 qobject_unref(snapshot_options);
3500 qobject_unref(bs->explicit_options);
3501 qobject_unref(bs->options);
3502 qobject_unref(options);
3504 bs->explicit_options = NULL;
3506 error_propagate(errp, local_err);
3511 qobject_unref(snapshot_options);
3512 qobject_unref(options);
3513 error_propagate(errp, local_err);
3517 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3518 QDict *options, int flags, Error **errp)
3520 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3524 /* Return true if the NULL-terminated @list contains @str */
3525 static bool is_str_in_list(const char *str, const char *const *list)
3529 for (i = 0; list[i] != NULL; i++) {
3530 if (!strcmp(str, list[i])) {
3539 * Check that every option set in @bs->options is also set in
3542 * Options listed in the common_options list and in
3543 * @bs->drv->mutable_opts are skipped.
3545 * Return 0 on success, otherwise return -EINVAL and set @errp.
3547 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3548 const QDict *new_opts, Error **errp)
3550 const QDictEntry *e;
3551 /* These options are common to all block drivers and are handled
3552 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3553 const char *const common_options[] = {
3554 "node-name", "discard", "cache.direct", "cache.no-flush",
3555 "read-only", "auto-read-only", "detect-zeroes", NULL
3558 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3559 if (!qdict_haskey(new_opts, e->key) &&
3560 !is_str_in_list(e->key, common_options) &&
3561 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3562 error_setg(errp, "Option '%s' cannot be reset "
3563 "to its default value", e->key);
3572 * Returns true if @child can be reached recursively from @bs
3574 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3575 BlockDriverState *child)
3583 QLIST_FOREACH(c, &bs->children, next) {
3584 if (bdrv_recurse_has_child(c->bs, child)) {
3593 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3594 * reopen of multiple devices.
3596 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3597 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3598 * be created and initialized. This newly created BlockReopenQueue should be
3599 * passed back in for subsequent calls that are intended to be of the same
3602 * bs is the BlockDriverState to add to the reopen queue.
3604 * options contains the changed options for the associated bs
3605 * (the BlockReopenQueue takes ownership)
3607 * flags contains the open flags for the associated bs
3609 * returns a pointer to bs_queue, which is either the newly allocated
3610 * bs_queue, or the existing bs_queue being used.
3612 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3614 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3615 BlockDriverState *bs,
3617 const BdrvChildClass *klass,
3619 bool parent_is_format,
3620 QDict *parent_options,
3626 BlockReopenQueueEntry *bs_entry;
3628 QDict *old_options, *explicit_options, *options_copy;
3632 /* Make sure that the caller remembered to use a drained section. This is
3633 * important to avoid graph changes between the recursive queuing here and
3634 * bdrv_reopen_multiple(). */
3635 assert(bs->quiesce_counter > 0);
3637 if (bs_queue == NULL) {
3638 bs_queue = g_new0(BlockReopenQueue, 1);
3639 QTAILQ_INIT(bs_queue);
3643 options = qdict_new();
3646 /* Check if this BlockDriverState is already in the queue */
3647 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3648 if (bs == bs_entry->state.bs) {
3654 * Precedence of options:
3655 * 1. Explicitly passed in options (highest)
3656 * 2. Retained from explicitly set options of bs
3657 * 3. Inherited from parent node
3658 * 4. Retained from effective options of bs
3661 /* Old explicitly set values (don't overwrite by inherited value) */
3662 if (bs_entry || keep_old_opts) {
3663 old_options = qdict_clone_shallow(bs_entry ?
3664 bs_entry->state.explicit_options :
3665 bs->explicit_options);
3666 bdrv_join_options(bs, options, old_options);
3667 qobject_unref(old_options);
3670 explicit_options = qdict_clone_shallow(options);
3672 /* Inherit from parent node */
3673 if (parent_options) {
3675 klass->inherit_options(role, parent_is_format, &flags, options,
3676 parent_flags, parent_options);
3678 flags = bdrv_get_flags(bs);
3681 if (keep_old_opts) {
3682 /* Old values are used for options that aren't set yet */
3683 old_options = qdict_clone_shallow(bs->options);
3684 bdrv_join_options(bs, options, old_options);
3685 qobject_unref(old_options);
3688 /* We have the final set of options so let's update the flags */
3689 options_copy = qdict_clone_shallow(options);
3690 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3691 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3692 update_flags_from_options(&flags, opts);
3693 qemu_opts_del(opts);
3694 qobject_unref(options_copy);
3696 /* bdrv_open_inherit() sets and clears some additional flags internally */
3697 flags &= ~BDRV_O_PROTOCOL;
3698 if (flags & BDRV_O_RDWR) {
3699 flags |= BDRV_O_ALLOW_RDWR;
3703 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3704 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3706 qobject_unref(bs_entry->state.options);
3707 qobject_unref(bs_entry->state.explicit_options);
3710 bs_entry->state.bs = bs;
3711 bs_entry->state.options = options;
3712 bs_entry->state.explicit_options = explicit_options;
3713 bs_entry->state.flags = flags;
3715 /* This needs to be overwritten in bdrv_reopen_prepare() */
3716 bs_entry->state.perm = UINT64_MAX;
3717 bs_entry->state.shared_perm = 0;
3720 * If keep_old_opts is false then it means that unspecified
3721 * options must be reset to their original value. We don't allow
3722 * resetting 'backing' but we need to know if the option is
3723 * missing in order to decide if we have to return an error.
3725 if (!keep_old_opts) {
3726 bs_entry->state.backing_missing =
3727 !qdict_haskey(options, "backing") &&
3728 !qdict_haskey(options, "backing.driver");
3731 QLIST_FOREACH(child, &bs->children, next) {
3732 QDict *new_child_options = NULL;
3733 bool child_keep_old = keep_old_opts;
3735 /* reopen can only change the options of block devices that were
3736 * implicitly created and inherited options. For other (referenced)
3737 * block devices, a syntax like "backing.foo" results in an error. */
3738 if (child->bs->inherits_from != bs) {
3742 /* Check if the options contain a child reference */
3743 if (qdict_haskey(options, child->name)) {
3744 const char *childref = qdict_get_try_str(options, child->name);
3746 * The current child must not be reopened if the child
3747 * reference is null or points to a different node.
3749 if (g_strcmp0(childref, child->bs->node_name)) {
3753 * If the child reference points to the current child then
3754 * reopen it with its existing set of options (note that
3755 * it can still inherit new options from the parent).
3757 child_keep_old = true;
3759 /* Extract child options ("child-name.*") */
3760 char *child_key_dot = g_strdup_printf("%s.", child->name);
3761 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3762 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3763 g_free(child_key_dot);
3766 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3767 child->klass, child->role, bs->drv->is_format,
3768 options, flags, child_keep_old);
3774 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3775 BlockDriverState *bs,
3776 QDict *options, bool keep_old_opts)
3778 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
3779 NULL, 0, keep_old_opts);
3783 * Reopen multiple BlockDriverStates atomically & transactionally.
3785 * The queue passed in (bs_queue) must have been built up previous
3786 * via bdrv_reopen_queue().
3788 * Reopens all BDS specified in the queue, with the appropriate
3789 * flags. All devices are prepared for reopen, and failure of any
3790 * device will cause all device changes to be abandoned, and intermediate
3793 * If all devices prepare successfully, then the changes are committed
3796 * All affected nodes must be drained between bdrv_reopen_queue() and
3797 * bdrv_reopen_multiple().
3799 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
3802 BlockReopenQueueEntry *bs_entry, *next;
3804 assert(bs_queue != NULL);
3806 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3807 assert(bs_entry->state.bs->quiesce_counter > 0);
3808 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
3811 bs_entry->prepared = true;
3814 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3815 BDRVReopenState *state = &bs_entry->state;
3816 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
3817 state->shared_perm, NULL, NULL, errp);
3821 /* Check if new_backing_bs would accept the new permissions */
3822 if (state->replace_backing_bs && state->new_backing_bs) {
3823 uint64_t nperm, nshared;
3824 bdrv_child_perm(state->bs, state->new_backing_bs,
3825 NULL, bdrv_backing_role(state->bs),
3826 bs_queue, state->perm, state->shared_perm,
3828 ret = bdrv_check_update_perm(state->new_backing_bs, NULL,
3829 nperm, nshared, NULL, NULL, errp);
3834 bs_entry->perms_checked = true;
3838 * If we reach this point, we have success and just need to apply the
3841 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3842 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3843 * children are usually goes after parents in reopen-queue, so go from last
3846 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3847 bdrv_reopen_commit(&bs_entry->state);
3852 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3853 BDRVReopenState *state = &bs_entry->state;
3855 if (!bs_entry->perms_checked) {
3860 bdrv_set_perm(state->bs, state->perm, state->shared_perm);
3862 bdrv_abort_perm_update(state->bs);
3863 if (state->replace_backing_bs && state->new_backing_bs) {
3864 bdrv_abort_perm_update(state->new_backing_bs);
3870 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3871 BlockDriverState *bs = bs_entry->state.bs;
3873 if (bs->drv->bdrv_reopen_commit_post)
3874 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
3878 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3880 if (bs_entry->prepared) {
3881 bdrv_reopen_abort(&bs_entry->state);
3883 qobject_unref(bs_entry->state.explicit_options);
3884 qobject_unref(bs_entry->state.options);
3886 if (bs_entry->state.new_backing_bs) {
3887 bdrv_unref(bs_entry->state.new_backing_bs);
3896 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3900 BlockReopenQueue *queue;
3901 QDict *opts = qdict_new();
3903 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3905 bdrv_subtree_drained_begin(bs);
3906 queue = bdrv_reopen_queue(NULL, bs, opts, true);
3907 ret = bdrv_reopen_multiple(queue, errp);
3908 bdrv_subtree_drained_end(bs);
3913 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3916 BlockReopenQueueEntry *entry;
3918 QTAILQ_FOREACH(entry, q, entry) {
3919 BlockDriverState *bs = entry->state.bs;
3922 QLIST_FOREACH(child, &bs->children, next) {
3932 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3933 uint64_t *perm, uint64_t *shared)
3936 BlockReopenQueueEntry *parent;
3937 uint64_t cumulative_perms = 0;
3938 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3940 QLIST_FOREACH(c, &bs->parents, next_parent) {
3941 parent = find_parent_in_reopen_queue(q, c);
3943 cumulative_perms |= c->perm;
3944 cumulative_shared_perms &= c->shared_perm;
3946 uint64_t nperm, nshared;
3948 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3949 parent->state.perm, parent->state.shared_perm,
3952 cumulative_perms |= nperm;
3953 cumulative_shared_perms &= nshared;
3956 *perm = cumulative_perms;
3957 *shared = cumulative_shared_perms;
3960 static bool bdrv_reopen_can_attach(BlockDriverState *parent,
3962 BlockDriverState *new_child,
3965 AioContext *parent_ctx = bdrv_get_aio_context(parent);
3966 AioContext *child_ctx = bdrv_get_aio_context(new_child);
3970 ignore = g_slist_prepend(NULL, child);
3971 ret = bdrv_can_set_aio_context(new_child, parent_ctx, &ignore, NULL);
3972 g_slist_free(ignore);
3977 ignore = g_slist_prepend(NULL, child);
3978 ret = bdrv_can_set_aio_context(parent, child_ctx, &ignore, errp);
3979 g_slist_free(ignore);
3984 * Take a BDRVReopenState and check if the value of 'backing' in the
3985 * reopen_state->options QDict is valid or not.
3987 * If 'backing' is missing from the QDict then return 0.
3989 * If 'backing' contains the node name of the backing file of
3990 * reopen_state->bs then return 0.
3992 * If 'backing' contains a different node name (or is null) then check
3993 * whether the current backing file can be replaced with the new one.
3994 * If that's the case then reopen_state->replace_backing_bs is set to
3995 * true and reopen_state->new_backing_bs contains a pointer to the new
3996 * backing BlockDriverState (or NULL).
3998 * Return 0 on success, otherwise return < 0 and set @errp.
4000 static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
4003 BlockDriverState *bs = reopen_state->bs;
4004 BlockDriverState *overlay_bs, *below_bs, *new_backing_bs;
4008 value = qdict_get(reopen_state->options, "backing");
4009 if (value == NULL) {
4013 switch (qobject_type(value)) {
4015 new_backing_bs = NULL;
4018 str = qobject_get_try_str(value);
4019 new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
4020 if (new_backing_bs == NULL) {
4022 } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
4023 error_setg(errp, "Making '%s' a backing file of '%s' "
4024 "would create a cycle", str, bs->node_name);
4029 /* 'backing' does not allow any other data type */
4030 g_assert_not_reached();
4034 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4035 * bdrv_reopen_commit() won't fail.
4037 if (new_backing_bs) {
4038 if (!bdrv_reopen_can_attach(bs, bs->backing, new_backing_bs, errp)) {
4044 * Ensure that @bs can really handle backing files, because we are
4045 * about to give it one (or swap the existing one)
4047 if (bs->drv->is_filter) {
4048 /* Filters always have a file or a backing child */
4050 error_setg(errp, "'%s' is a %s filter node that does not support a "
4051 "backing child", bs->node_name, bs->drv->format_name);
4054 } else if (!bs->drv->supports_backing) {
4055 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
4056 "files", bs->drv->format_name, bs->node_name);
4061 * Find the "actual" backing file by skipping all links that point
4062 * to an implicit node, if any (e.g. a commit filter node).
4063 * We cannot use any of the bdrv_skip_*() functions here because
4064 * those return the first explicit node, while we are looking for
4068 for (below_bs = bdrv_filter_or_cow_bs(overlay_bs);
4069 below_bs && below_bs->implicit;
4070 below_bs = bdrv_filter_or_cow_bs(overlay_bs))
4072 overlay_bs = below_bs;
4075 /* If we want to replace the backing file we need some extra checks */
4076 if (new_backing_bs != bdrv_filter_or_cow_bs(overlay_bs)) {
4077 /* Check for implicit nodes between bs and its backing file */
4078 if (bs != overlay_bs) {
4079 error_setg(errp, "Cannot change backing link if '%s' has "
4080 "an implicit backing file", bs->node_name);
4084 * Check if the backing link that we want to replace is frozen.
4086 * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
4087 * because we know that overlay_bs == bs, and that @bs
4088 * either is a filter that uses ->backing or a COW format BDS
4089 * with bs->drv->supports_backing == true.
4091 if (bdrv_is_backing_chain_frozen(overlay_bs,
4092 child_bs(overlay_bs->backing), errp))
4096 reopen_state->replace_backing_bs = true;
4097 if (new_backing_bs) {
4098 bdrv_ref(new_backing_bs);
4099 reopen_state->new_backing_bs = new_backing_bs;
4107 * Prepares a BlockDriverState for reopen. All changes are staged in the
4108 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4109 * the block driver layer .bdrv_reopen_prepare()
4111 * bs is the BlockDriverState to reopen
4112 * flags are the new open flags
4113 * queue is the reopen queue
4115 * Returns 0 on success, non-zero on error. On error errp will be set
4118 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4119 * It is the responsibility of the caller to then call the abort() or
4120 * commit() for any other BDS that have been left in a prepare() state
4123 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
4128 Error *local_err = NULL;
4131 QDict *orig_reopen_opts;
4132 char *discard = NULL;
4134 bool drv_prepared = false;
4136 assert(reopen_state != NULL);
4137 assert(reopen_state->bs->drv != NULL);
4138 drv = reopen_state->bs->drv;
4140 /* This function and each driver's bdrv_reopen_prepare() remove
4141 * entries from reopen_state->options as they are processed, so
4142 * we need to make a copy of the original QDict. */
4143 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4145 /* Process generic block layer options */
4146 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4147 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4152 /* This was already called in bdrv_reopen_queue_child() so the flags
4153 * are up-to-date. This time we simply want to remove the options from
4154 * QemuOpts in order to indicate that they have been processed. */
4155 old_flags = reopen_state->flags;
4156 update_flags_from_options(&reopen_state->flags, opts);
4157 assert(old_flags == reopen_state->flags);
4159 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4160 if (discard != NULL) {
4161 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4162 error_setg(errp, "Invalid discard option");
4168 reopen_state->detect_zeroes =
4169 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4171 error_propagate(errp, local_err);
4176 /* All other options (including node-name and driver) must be unchanged.
4177 * Put them back into the QDict, so that they are checked at the end
4178 * of this function. */
4179 qemu_opts_to_qdict(opts, reopen_state->options);
4181 /* If we are to stay read-only, do not allow permission change
4182 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4183 * not set, or if the BDS still has copy_on_read enabled */
4184 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4185 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4187 error_propagate(errp, local_err);
4191 /* Calculate required permissions after reopening */
4192 bdrv_reopen_perm(queue, reopen_state->bs,
4193 &reopen_state->perm, &reopen_state->shared_perm);
4195 ret = bdrv_flush(reopen_state->bs);
4197 error_setg_errno(errp, -ret, "Error flushing drive");
4201 if (drv->bdrv_reopen_prepare) {
4203 * If a driver-specific option is missing, it means that we
4204 * should reset it to its default value.
4205 * But not all options allow that, so we need to check it first.
4207 ret = bdrv_reset_options_allowed(reopen_state->bs,
4208 reopen_state->options, errp);
4213 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4215 if (local_err != NULL) {
4216 error_propagate(errp, local_err);
4218 bdrv_refresh_filename(reopen_state->bs);
4219 error_setg(errp, "failed while preparing to reopen image '%s'",
4220 reopen_state->bs->filename);
4225 /* It is currently mandatory to have a bdrv_reopen_prepare()
4226 * handler for each supported drv. */
4227 error_setg(errp, "Block format '%s' used by node '%s' "
4228 "does not support reopening files", drv->format_name,
4229 bdrv_get_device_or_node_name(reopen_state->bs));
4234 drv_prepared = true;
4237 * We must provide the 'backing' option if the BDS has a backing
4238 * file or if the image file has a backing file name as part of
4239 * its metadata. Otherwise the 'backing' option can be omitted.
4241 if (drv->supports_backing && reopen_state->backing_missing &&
4242 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4243 error_setg(errp, "backing is missing for '%s'",
4244 reopen_state->bs->node_name);
4250 * Allow changing the 'backing' option. The new value can be
4251 * either a reference to an existing node (using its node name)
4252 * or NULL to simply detach the current backing file.
4254 ret = bdrv_reopen_parse_backing(reopen_state, errp);
4258 qdict_del(reopen_state->options, "backing");
4260 /* Options that are not handled are only okay if they are unchanged
4261 * compared to the old state. It is expected that some options are only
4262 * used for the initial open, but not reopen (e.g. filename) */
4263 if (qdict_size(reopen_state->options)) {
4264 const QDictEntry *entry = qdict_first(reopen_state->options);
4267 QObject *new = entry->value;
4268 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4270 /* Allow child references (child_name=node_name) as long as they
4271 * point to the current child (i.e. everything stays the same). */
4272 if (qobject_type(new) == QTYPE_QSTRING) {
4274 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4275 if (!strcmp(child->name, entry->key)) {
4281 const char *str = qobject_get_try_str(new);
4282 if (!strcmp(child->bs->node_name, str)) {
4283 continue; /* Found child with this name, skip option */
4289 * TODO: When using -drive to specify blockdev options, all values
4290 * will be strings; however, when using -blockdev, blockdev-add or
4291 * filenames using the json:{} pseudo-protocol, they will be
4293 * In contrast, reopening options are (currently) always strings
4294 * (because you can only specify them through qemu-io; all other
4295 * callers do not specify any options).
4296 * Therefore, when using anything other than -drive to create a BDS,
4297 * this cannot detect non-string options as unchanged, because
4298 * qobject_is_equal() always returns false for objects of different
4299 * type. In the future, this should be remedied by correctly typing
4300 * all options. For now, this is not too big of an issue because
4301 * the user can simply omit options which cannot be changed anyway,
4302 * so they will stay unchanged.
4304 if (!qobject_is_equal(new, old)) {
4305 error_setg(errp, "Cannot change the option '%s'", entry->key);
4309 } while ((entry = qdict_next(reopen_state->options, entry)));
4314 /* Restore the original reopen_state->options QDict */
4315 qobject_unref(reopen_state->options);
4316 reopen_state->options = qobject_ref(orig_reopen_opts);
4319 if (ret < 0 && drv_prepared) {
4320 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4321 * call drv->bdrv_reopen_abort() before signaling an error
4322 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4323 * when the respective bdrv_reopen_prepare() has failed) */
4324 if (drv->bdrv_reopen_abort) {
4325 drv->bdrv_reopen_abort(reopen_state);
4328 qemu_opts_del(opts);
4329 qobject_unref(orig_reopen_opts);
4335 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4336 * makes them final by swapping the staging BlockDriverState contents into
4337 * the active BlockDriverState contents.
4339 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4342 BlockDriverState *bs;
4345 assert(reopen_state != NULL);
4346 bs = reopen_state->bs;
4348 assert(drv != NULL);
4350 /* If there are any driver level actions to take */
4351 if (drv->bdrv_reopen_commit) {
4352 drv->bdrv_reopen_commit(reopen_state);
4355 /* set BDS specific flags now */
4356 qobject_unref(bs->explicit_options);
4357 qobject_unref(bs->options);
4359 bs->explicit_options = reopen_state->explicit_options;
4360 bs->options = reopen_state->options;
4361 bs->open_flags = reopen_state->flags;
4362 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
4363 bs->detect_zeroes = reopen_state->detect_zeroes;
4365 if (reopen_state->replace_backing_bs) {
4366 qdict_del(bs->explicit_options, "backing");
4367 qdict_del(bs->options, "backing");
4370 /* Remove child references from bs->options and bs->explicit_options.
4371 * Child options were already removed in bdrv_reopen_queue_child() */
4372 QLIST_FOREACH(child, &bs->children, next) {
4373 qdict_del(bs->explicit_options, child->name);
4374 qdict_del(bs->options, child->name);
4378 * Change the backing file if a new one was specified. We do this
4379 * after updating bs->options, so bdrv_refresh_filename() (called
4380 * from bdrv_set_backing_hd()) has the new values.
4382 if (reopen_state->replace_backing_bs) {
4383 BlockDriverState *old_backing_bs = child_bs(bs->backing);
4384 assert(!old_backing_bs || !old_backing_bs->implicit);
4385 /* Abort the permission update on the backing bs we're detaching */
4386 if (old_backing_bs) {
4387 bdrv_abort_perm_update(old_backing_bs);
4389 bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort);
4392 bdrv_refresh_limits(bs, NULL);
4396 * Abort the reopen, and delete and free the staged changes in
4399 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4403 assert(reopen_state != NULL);
4404 drv = reopen_state->bs->drv;
4405 assert(drv != NULL);
4407 if (drv->bdrv_reopen_abort) {
4408 drv->bdrv_reopen_abort(reopen_state);
4413 static void bdrv_close(BlockDriverState *bs)
4415 BdrvAioNotifier *ban, *ban_next;
4416 BdrvChild *child, *next;
4418 assert(!bs->refcnt);
4420 bdrv_drained_begin(bs); /* complete I/O */
4422 bdrv_drain(bs); /* in case flush left pending I/O */
4425 if (bs->drv->bdrv_close) {
4426 /* Must unfreeze all children, so bdrv_unref_child() works */
4427 bs->drv->bdrv_close(bs);
4432 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4433 bdrv_unref_child(bs, child);
4440 qatomic_set(&bs->copy_on_read, 0);
4441 bs->backing_file[0] = '\0';
4442 bs->backing_format[0] = '\0';
4443 bs->total_sectors = 0;
4444 bs->encrypted = false;
4446 qobject_unref(bs->options);
4447 qobject_unref(bs->explicit_options);
4449 bs->explicit_options = NULL;
4450 qobject_unref(bs->full_open_options);
4451 bs->full_open_options = NULL;
4453 bdrv_release_named_dirty_bitmaps(bs);
4454 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4456 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4459 QLIST_INIT(&bs->aio_notifiers);
4460 bdrv_drained_end(bs);
4463 void bdrv_close_all(void)
4465 assert(job_next(NULL) == NULL);
4466 blk_exp_close_all();
4468 /* Drop references from requests still in flight, such as canceled block
4469 * jobs whose AIO context has not been polled yet */
4472 blk_remove_all_bs();
4473 blockdev_close_all_bdrv_states();
4475 assert(QTAILQ_EMPTY(&all_bdrv_states));
4478 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4484 if (c->klass->stay_at_node) {
4488 /* If the child @c belongs to the BDS @to, replacing the current
4489 * c->bs by @to would mean to create a loop.
4491 * Such a case occurs when appending a BDS to a backing chain.
4492 * For instance, imagine the following chain:
4494 * guest device -> node A -> further backing chain...
4496 * Now we create a new BDS B which we want to put on top of this
4497 * chain, so we first attach A as its backing node:
4502 * guest device -> node A -> further backing chain...
4504 * Finally we want to replace A by B. When doing that, we want to
4505 * replace all pointers to A by pointers to B -- except for the
4506 * pointer from B because (1) that would create a loop, and (2)
4507 * that pointer should simply stay intact:
4509 * guest device -> node B
4512 * node A -> further backing chain...
4514 * In general, when replacing a node A (c->bs) by a node B (@to),
4515 * if A is a child of B, that means we cannot replace A by B there
4516 * because that would create a loop. Silently detaching A from B
4517 * is also not really an option. So overall just leaving A in
4518 * place there is the most sensible choice.
4520 * We would also create a loop in any cases where @c is only
4521 * indirectly referenced by @to. Prevent this by returning false
4522 * if @c is found (by breadth-first search) anywhere in the whole
4527 found = g_hash_table_new(NULL, NULL);
4528 g_hash_table_add(found, to);
4529 queue = g_queue_new();
4530 g_queue_push_tail(queue, to);
4532 while (!g_queue_is_empty(queue)) {
4533 BlockDriverState *v = g_queue_pop_head(queue);
4536 QLIST_FOREACH(c2, &v->children, next) {
4542 if (g_hash_table_contains(found, c2->bs)) {
4546 g_queue_push_tail(queue, c2->bs);
4547 g_hash_table_add(found, c2->bs);
4551 g_queue_free(queue);
4552 g_hash_table_destroy(found);
4557 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
4560 BdrvChild *c, *next;
4561 GSList *list = NULL, *p;
4562 uint64_t perm = 0, shared = BLK_PERM_ALL;
4565 /* Make sure that @from doesn't go away until we have successfully attached
4566 * all of its parents to @to. */
4569 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4570 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
4571 bdrv_drained_begin(from);
4573 /* Put all parents into @list and calculate their cumulative permissions */
4574 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4575 assert(c->bs == from);
4576 if (!should_update_child(c, to)) {
4580 error_setg(errp, "Cannot change '%s' link to '%s'",
4581 c->name, from->node_name);
4584 list = g_slist_prepend(list, c);
4586 shared &= c->shared_perm;
4589 /* Check whether the required permissions can be granted on @to, ignoring
4590 * all BdrvChild in @list so that they can't block themselves. */
4591 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, NULL, errp);
4593 bdrv_abort_perm_update(to);
4597 /* Now actually perform the change. We performed the permission check for
4598 * all elements of @list at once, so set the permissions all at once at the
4600 for (p = list; p != NULL; p = p->next) {
4604 bdrv_replace_child_noperm(c, to);
4608 bdrv_get_cumulative_perm(to, &perm, &shared);
4609 bdrv_set_perm(to, perm, shared);
4613 bdrv_drained_end(from);
4618 * Add new bs contents at the top of an image chain while the chain is
4619 * live, while keeping required fields on the top layer.
4621 * This will modify the BlockDriverState fields, and swap contents
4622 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4624 * bs_new must not be attached to a BlockBackend.
4626 * This function does not create any image files.
4628 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4629 * that's what the callers commonly need. bs_new will be referenced by the old
4630 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4631 * reference of its own, it must call bdrv_ref().
4633 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
4636 Error *local_err = NULL;
4638 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
4640 error_propagate(errp, local_err);
4644 bdrv_replace_node(bs_top, bs_new, &local_err);
4646 error_propagate(errp, local_err);
4647 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
4651 /* bs_new is now referenced by its new parents, we don't need the
4652 * additional reference any more. */
4657 static void bdrv_delete(BlockDriverState *bs)
4659 assert(bdrv_op_blocker_is_empty(bs));
4660 assert(!bs->refcnt);
4662 /* remove from list, if necessary */
4663 if (bs->node_name[0] != '\0') {
4664 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
4666 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
4674 * Run consistency checks on an image
4676 * Returns 0 if the check could be completed (it doesn't mean that the image is
4677 * free of errors) or -errno when an internal error occurred. The results of the
4678 * check are stored in res.
4680 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
4681 BdrvCheckResult *res, BdrvCheckMode fix)
4683 if (bs->drv == NULL) {
4686 if (bs->drv->bdrv_co_check == NULL) {
4690 memset(res, 0, sizeof(*res));
4691 return bs->drv->bdrv_co_check(bs, res, fix);
4697 * -EINVAL - backing format specified, but no file
4698 * -ENOSPC - can't update the backing file because no space is left in the
4700 * -ENOTSUP - format driver doesn't support changing the backing file
4702 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
4703 const char *backing_fmt, bool warn)
4705 BlockDriver *drv = bs->drv;
4712 /* Backing file format doesn't make sense without a backing file */
4713 if (backing_fmt && !backing_file) {
4717 if (warn && backing_file && !backing_fmt) {
4718 warn_report("Deprecated use of backing file without explicit "
4719 "backing format, use of this image requires "
4720 "potentially unsafe format probing");
4723 if (drv->bdrv_change_backing_file != NULL) {
4724 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
4730 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
4731 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
4732 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
4733 backing_file ?: "");
4739 * Finds the first non-filter node above bs in the chain between
4740 * active and bs. The returned node is either an immediate parent of
4741 * bs, or there are only filter nodes between the two.
4743 * Returns NULL if bs is not found in active's image chain,
4744 * or if active == bs.
4746 * Returns the bottommost base image if bs == NULL.
4748 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
4749 BlockDriverState *bs)
4751 bs = bdrv_skip_filters(bs);
4752 active = bdrv_skip_filters(active);
4755 BlockDriverState *next = bdrv_backing_chain_next(active);
4765 /* Given a BDS, searches for the base layer. */
4766 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4768 return bdrv_find_overlay(bs, NULL);
4772 * Return true if at least one of the COW (backing) and filter links
4773 * between @bs and @base is frozen. @errp is set if that's the case.
4774 * @base must be reachable from @bs, or NULL.
4776 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
4779 BlockDriverState *i;
4782 for (i = bs; i != base; i = child_bs(child)) {
4783 child = bdrv_filter_or_cow_child(i);
4785 if (child && child->frozen) {
4786 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4787 child->name, i->node_name, child->bs->node_name);
4796 * Freeze all COW (backing) and filter links between @bs and @base.
4797 * If any of the links is already frozen the operation is aborted and
4798 * none of the links are modified.
4799 * @base must be reachable from @bs, or NULL.
4800 * Returns 0 on success. On failure returns < 0 and sets @errp.
4802 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4805 BlockDriverState *i;
4808 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4812 for (i = bs; i != base; i = child_bs(child)) {
4813 child = bdrv_filter_or_cow_child(i);
4814 if (child && child->bs->never_freeze) {
4815 error_setg(errp, "Cannot freeze '%s' link to '%s'",
4816 child->name, child->bs->node_name);
4821 for (i = bs; i != base; i = child_bs(child)) {
4822 child = bdrv_filter_or_cow_child(i);
4824 child->frozen = true;
4832 * Unfreeze all COW (backing) and filter links between @bs and @base.
4833 * The caller must ensure that all links are frozen before using this
4835 * @base must be reachable from @bs, or NULL.
4837 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4839 BlockDriverState *i;
4842 for (i = bs; i != base; i = child_bs(child)) {
4843 child = bdrv_filter_or_cow_child(i);
4845 assert(child->frozen);
4846 child->frozen = false;
4852 * Drops images above 'base' up to and including 'top', and sets the image
4853 * above 'top' to have base as its backing file.
4855 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4856 * information in 'bs' can be properly updated.
4858 * E.g., this will convert the following chain:
4859 * bottom <- base <- intermediate <- top <- active
4863 * bottom <- base <- active
4865 * It is allowed for bottom==base, in which case it converts:
4867 * base <- intermediate <- top <- active
4873 * If backing_file_str is non-NULL, it will be used when modifying top's
4874 * overlay image metadata.
4877 * if active == top, that is considered an error
4880 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4881 const char *backing_file_str)
4883 BlockDriverState *explicit_top = top;
4884 bool update_inherits_from;
4885 BdrvChild *c, *next;
4886 Error *local_err = NULL;
4890 bdrv_subtree_drained_begin(top);
4892 if (!top->drv || !base->drv) {
4896 /* Make sure that base is in the backing chain of top */
4897 if (!bdrv_chain_contains(top, base)) {
4901 /* This function changes all links that point to top and makes
4902 * them point to base. Check that none of them is frozen. */
4903 QLIST_FOREACH(c, &top->parents, next_parent) {
4909 /* If 'base' recursively inherits from 'top' then we should set
4910 * base->inherits_from to top->inherits_from after 'top' and all
4911 * other intermediate nodes have been dropped.
4912 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4913 * it because no one inherits from it. We use explicit_top for that. */
4914 explicit_top = bdrv_skip_implicit_filters(explicit_top);
4915 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4917 /* success - we can delete the intermediate states, and link top->base */
4918 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4919 * we've figured out how they should work. */
4920 if (!backing_file_str) {
4921 bdrv_refresh_filename(base);
4922 backing_file_str = base->filename;
4925 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
4926 /* Check whether we are allowed to switch c from top to base */
4927 GSList *ignore_children = g_slist_prepend(NULL, c);
4928 ret = bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
4929 ignore_children, NULL, &local_err);
4930 g_slist_free(ignore_children);
4932 error_report_err(local_err);
4936 /* If so, update the backing file path in the image file */
4937 if (c->klass->update_filename) {
4938 ret = c->klass->update_filename(c, base, backing_file_str,
4941 bdrv_abort_perm_update(base);
4942 error_report_err(local_err);
4948 * Do the actual switch in the in-memory graph.
4949 * Completes bdrv_check_update_perm() transaction internally.
4950 * c->frozen is false, we have checked that above.
4953 bdrv_replace_child(c, base);
4957 if (update_inherits_from) {
4958 base->inherits_from = explicit_top->inherits_from;
4963 bdrv_subtree_drained_end(top);
4969 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
4970 * sums the size of all data-bearing children. (This excludes backing
4973 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
4976 int64_t child_size, sum = 0;
4978 QLIST_FOREACH(child, &bs->children, next) {
4979 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
4980 BDRV_CHILD_FILTERED))
4982 child_size = bdrv_get_allocated_file_size(child->bs);
4983 if (child_size < 0) {
4994 * Length of a allocated file in bytes. Sparse files are counted by actual
4995 * allocated space. Return < 0 if error or unknown.
4997 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
4999 BlockDriver *drv = bs->drv;
5003 if (drv->bdrv_get_allocated_file_size) {
5004 return drv->bdrv_get_allocated_file_size(bs);
5007 if (drv->bdrv_file_open) {
5009 * Protocol drivers default to -ENOTSUP (most of their data is
5010 * not stored in any of their children (if they even have any),
5011 * so there is no generic way to figure it out).
5014 } else if (drv->is_filter) {
5015 /* Filter drivers default to the size of their filtered child */
5016 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5018 /* Other drivers default to summing their children's sizes */
5019 return bdrv_sum_allocated_file_size(bs);
5025 * @drv: Format driver
5026 * @opts: Creation options for new image
5027 * @in_bs: Existing image containing data for new image (may be NULL)
5028 * @errp: Error object
5029 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5032 * Calculate file size required to create a new image.
5034 * If @in_bs is given then space for allocated clusters and zero clusters
5035 * from that image are included in the calculation. If @opts contains a
5036 * backing file that is shared by @in_bs then backing clusters may be omitted
5037 * from the calculation.
5039 * If @in_bs is NULL then the calculation includes no allocated clusters
5040 * unless a preallocation option is given in @opts.
5042 * Note that @in_bs may use a different BlockDriver from @drv.
5044 * If an error occurs the @errp pointer is set.
5046 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5047 BlockDriverState *in_bs, Error **errp)
5049 if (!drv->bdrv_measure) {
5050 error_setg(errp, "Block driver '%s' does not support size measurement",
5055 return drv->bdrv_measure(opts, in_bs, errp);
5059 * Return number of sectors on success, -errno on error.
5061 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5063 BlockDriver *drv = bs->drv;
5068 if (drv->has_variable_length) {
5069 int ret = refresh_total_sectors(bs, bs->total_sectors);
5074 return bs->total_sectors;
5078 * Return length in bytes on success, -errno on error.
5079 * The length is always a multiple of BDRV_SECTOR_SIZE.
5081 int64_t bdrv_getlength(BlockDriverState *bs)
5083 int64_t ret = bdrv_nb_sectors(bs);
5085 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
5086 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
5089 /* return 0 as number of sectors if no device present or error */
5090 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5092 int64_t nb_sectors = bdrv_nb_sectors(bs);
5094 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5097 bool bdrv_is_sg(BlockDriverState *bs)
5103 * Return whether the given node supports compressed writes.
5105 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5107 BlockDriverState *filtered;
5109 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5113 filtered = bdrv_filter_bs(bs);
5116 * Filters can only forward compressed writes, so we have to
5119 return bdrv_supports_compressed_writes(filtered);
5125 const char *bdrv_get_format_name(BlockDriverState *bs)
5127 return bs->drv ? bs->drv->format_name : NULL;
5130 static int qsort_strcmp(const void *a, const void *b)
5132 return strcmp(*(char *const *)a, *(char *const *)b);
5135 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5136 void *opaque, bool read_only)
5141 const char **formats = NULL;
5143 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5144 if (drv->format_name) {
5148 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5152 while (formats && i && !found) {
5153 found = !strcmp(formats[--i], drv->format_name);
5157 formats = g_renew(const char *, formats, count + 1);
5158 formats[count++] = drv->format_name;
5163 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5164 const char *format_name = block_driver_modules[i].format_name;
5170 if (use_bdrv_whitelist &&
5171 !bdrv_format_is_whitelisted(format_name, read_only)) {
5175 while (formats && j && !found) {
5176 found = !strcmp(formats[--j], format_name);
5180 formats = g_renew(const char *, formats, count + 1);
5181 formats[count++] = format_name;
5186 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5188 for (i = 0; i < count; i++) {
5189 it(opaque, formats[i]);
5195 /* This function is to find a node in the bs graph */
5196 BlockDriverState *bdrv_find_node(const char *node_name)
5198 BlockDriverState *bs;
5202 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5203 if (!strcmp(node_name, bs->node_name)) {
5210 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5211 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5214 BlockDeviceInfoList *list, *entry;
5215 BlockDriverState *bs;
5218 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5219 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5221 qapi_free_BlockDeviceInfoList(list);
5224 entry = g_malloc0(sizeof(*entry));
5225 entry->value = info;
5233 #define QAPI_LIST_ADD(list, element) do { \
5234 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5235 _tmp->value = (element); \
5236 _tmp->next = (list); \
5240 typedef struct XDbgBlockGraphConstructor {
5241 XDbgBlockGraph *graph;
5242 GHashTable *graph_nodes;
5243 } XDbgBlockGraphConstructor;
5245 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5247 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5249 gr->graph = g_new0(XDbgBlockGraph, 1);
5250 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5255 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5257 XDbgBlockGraph *graph = gr->graph;
5259 g_hash_table_destroy(gr->graph_nodes);
5265 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5267 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5274 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5275 * answer of g_hash_table_lookup.
5277 ret = g_hash_table_size(gr->graph_nodes) + 1;
5278 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5283 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5284 XDbgBlockGraphNodeType type, const char *name)
5286 XDbgBlockGraphNode *n;
5288 n = g_new0(XDbgBlockGraphNode, 1);
5290 n->id = xdbg_graph_node_num(gr, node);
5292 n->name = g_strdup(name);
5294 QAPI_LIST_ADD(gr->graph->nodes, n);
5297 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5298 const BdrvChild *child)
5300 BlockPermission qapi_perm;
5301 XDbgBlockGraphEdge *edge;
5303 edge = g_new0(XDbgBlockGraphEdge, 1);
5305 edge->parent = xdbg_graph_node_num(gr, parent);
5306 edge->child = xdbg_graph_node_num(gr, child->bs);
5307 edge->name = g_strdup(child->name);
5309 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5310 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5312 if (flag & child->perm) {
5313 QAPI_LIST_ADD(edge->perm, qapi_perm);
5315 if (flag & child->shared_perm) {
5316 QAPI_LIST_ADD(edge->shared_perm, qapi_perm);
5320 QAPI_LIST_ADD(gr->graph->edges, edge);
5324 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5328 BlockDriverState *bs;
5330 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5332 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5333 char *allocated_name = NULL;
5334 const char *name = blk_name(blk);
5337 name = allocated_name = blk_get_attached_dev_id(blk);
5339 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5341 g_free(allocated_name);
5342 if (blk_root(blk)) {
5343 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5347 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5350 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5352 for (el = job->nodes; el; el = el->next) {
5353 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5357 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5358 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5360 QLIST_FOREACH(child, &bs->children, next) {
5361 xdbg_graph_add_edge(gr, bs, child);
5365 return xdbg_graph_finalize(gr);
5368 BlockDriverState *bdrv_lookup_bs(const char *device,
5369 const char *node_name,
5373 BlockDriverState *bs;
5376 blk = blk_by_name(device);
5381 error_setg(errp, "Device '%s' has no medium", device);
5389 bs = bdrv_find_node(node_name);
5396 error_setg(errp, "Cannot find device=%s nor node_name=%s",
5397 device ? device : "",
5398 node_name ? node_name : "");
5402 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5403 * return false. If either argument is NULL, return false. */
5404 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
5406 while (top && top != base) {
5407 top = bdrv_filter_or_cow_bs(top);
5413 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
5416 return QTAILQ_FIRST(&graph_bdrv_states);
5418 return QTAILQ_NEXT(bs, node_list);
5421 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5424 return QTAILQ_FIRST(&all_bdrv_states);
5426 return QTAILQ_NEXT(bs, bs_list);
5429 const char *bdrv_get_node_name(const BlockDriverState *bs)
5431 return bs->node_name;
5434 const char *bdrv_get_parent_name(const BlockDriverState *bs)
5439 /* If multiple parents have a name, just pick the first one. */
5440 QLIST_FOREACH(c, &bs->parents, next_parent) {
5441 if (c->klass->get_name) {
5442 name = c->klass->get_name(c);
5443 if (name && *name) {
5452 /* TODO check what callers really want: bs->node_name or blk_name() */
5453 const char *bdrv_get_device_name(const BlockDriverState *bs)
5455 return bdrv_get_parent_name(bs) ?: "";
5458 /* This can be used to identify nodes that might not have a device
5459 * name associated. Since node and device names live in the same
5460 * namespace, the result is unambiguous. The exception is if both are
5461 * absent, then this returns an empty (non-null) string. */
5462 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
5464 return bdrv_get_parent_name(bs) ?: bs->node_name;
5467 int bdrv_get_flags(BlockDriverState *bs)
5469 return bs->open_flags;
5472 int bdrv_has_zero_init_1(BlockDriverState *bs)
5477 int bdrv_has_zero_init(BlockDriverState *bs)
5479 BlockDriverState *filtered;
5485 /* If BS is a copy on write image, it is initialized to
5486 the contents of the base image, which may not be zeroes. */
5487 if (bdrv_cow_child(bs)) {
5490 if (bs->drv->bdrv_has_zero_init) {
5491 return bs->drv->bdrv_has_zero_init(bs);
5494 filtered = bdrv_filter_bs(bs);
5496 return bdrv_has_zero_init(filtered);
5503 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
5505 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5509 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
5512 void bdrv_get_backing_filename(BlockDriverState *bs,
5513 char *filename, int filename_size)
5515 pstrcpy(filename, filename_size, bs->backing_file);
5518 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5520 BlockDriver *drv = bs->drv;
5521 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5525 if (!drv->bdrv_get_info) {
5526 BlockDriverState *filtered = bdrv_filter_bs(bs);
5528 return bdrv_get_info(filtered, bdi);
5532 memset(bdi, 0, sizeof(*bdi));
5533 return drv->bdrv_get_info(bs, bdi);
5536 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
5539 BlockDriver *drv = bs->drv;
5540 if (drv && drv->bdrv_get_specific_info) {
5541 return drv->bdrv_get_specific_info(bs, errp);
5546 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
5548 BlockDriver *drv = bs->drv;
5549 if (!drv || !drv->bdrv_get_specific_stats) {
5552 return drv->bdrv_get_specific_stats(bs);
5555 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
5557 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
5561 bs->drv->bdrv_debug_event(bs, event);
5564 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
5566 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
5567 bs = bdrv_primary_bs(bs);
5570 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
5571 assert(bs->drv->bdrv_debug_remove_breakpoint);
5578 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
5581 bs = bdrv_find_debug_node(bs);
5583 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
5589 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
5591 bs = bdrv_find_debug_node(bs);
5593 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
5599 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
5601 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
5602 bs = bdrv_primary_bs(bs);
5605 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
5606 return bs->drv->bdrv_debug_resume(bs, tag);
5612 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
5614 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
5615 bs = bdrv_primary_bs(bs);
5618 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
5619 return bs->drv->bdrv_debug_is_suspended(bs, tag);
5625 /* backing_file can either be relative, or absolute, or a protocol. If it is
5626 * relative, it must be relative to the chain. So, passing in bs->filename
5627 * from a BDS as backing_file should not be done, as that may be relative to
5628 * the CWD rather than the chain. */
5629 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
5630 const char *backing_file)
5632 char *filename_full = NULL;
5633 char *backing_file_full = NULL;
5634 char *filename_tmp = NULL;
5635 int is_protocol = 0;
5636 bool filenames_refreshed = false;
5637 BlockDriverState *curr_bs = NULL;
5638 BlockDriverState *retval = NULL;
5639 BlockDriverState *bs_below;
5641 if (!bs || !bs->drv || !backing_file) {
5645 filename_full = g_malloc(PATH_MAX);
5646 backing_file_full = g_malloc(PATH_MAX);
5648 is_protocol = path_has_protocol(backing_file);
5651 * Being largely a legacy function, skip any filters here
5652 * (because filters do not have normal filenames, so they cannot
5653 * match anyway; and allowing json:{} filenames is a bit out of
5656 for (curr_bs = bdrv_skip_filters(bs);
5657 bdrv_cow_child(curr_bs) != NULL;
5660 bs_below = bdrv_backing_chain_next(curr_bs);
5662 if (bdrv_backing_overridden(curr_bs)) {
5664 * If the backing file was overridden, we can only compare
5665 * directly against the backing node's filename.
5668 if (!filenames_refreshed) {
5670 * This will automatically refresh all of the
5671 * filenames in the rest of the backing chain, so we
5672 * only need to do this once.
5674 bdrv_refresh_filename(bs_below);
5675 filenames_refreshed = true;
5678 if (strcmp(backing_file, bs_below->filename) == 0) {
5682 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
5684 * If either of the filename paths is actually a protocol, then
5685 * compare unmodified paths; otherwise make paths relative.
5687 char *backing_file_full_ret;
5689 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
5693 /* Also check against the full backing filename for the image */
5694 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
5696 if (backing_file_full_ret) {
5697 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
5698 g_free(backing_file_full_ret);
5705 /* If not an absolute filename path, make it relative to the current
5706 * image's filename path */
5707 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
5709 /* We are going to compare canonicalized absolute pathnames */
5710 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
5711 g_free(filename_tmp);
5714 g_free(filename_tmp);
5716 /* We need to make sure the backing filename we are comparing against
5717 * is relative to the current image filename (or absolute) */
5718 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
5719 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
5720 g_free(filename_tmp);
5723 g_free(filename_tmp);
5725 if (strcmp(backing_file_full, filename_full) == 0) {
5732 g_free(filename_full);
5733 g_free(backing_file_full);
5737 void bdrv_init(void)
5739 module_call_init(MODULE_INIT_BLOCK);
5742 void bdrv_init_with_whitelist(void)
5744 use_bdrv_whitelist = 1;
5748 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
5750 BdrvChild *child, *parent;
5751 uint64_t perm, shared_perm;
5752 Error *local_err = NULL;
5754 BdrvDirtyBitmap *bm;
5760 QLIST_FOREACH(child, &bs->children, next) {
5761 bdrv_co_invalidate_cache(child->bs, &local_err);
5763 error_propagate(errp, local_err);
5769 * Update permissions, they may differ for inactive nodes.
5771 * Note that the required permissions of inactive images are always a
5772 * subset of the permissions required after activating the image. This
5773 * allows us to just get the permissions upfront without restricting
5774 * drv->bdrv_invalidate_cache().
5776 * It also means that in error cases, we don't have to try and revert to
5777 * the old permissions (which is an operation that could fail, too). We can
5778 * just keep the extended permissions for the next time that an activation
5779 * of the image is tried.
5781 if (bs->open_flags & BDRV_O_INACTIVE) {
5782 bs->open_flags &= ~BDRV_O_INACTIVE;
5783 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5784 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, NULL, errp);
5786 bs->open_flags |= BDRV_O_INACTIVE;
5789 bdrv_set_perm(bs, perm, shared_perm);
5791 if (bs->drv->bdrv_co_invalidate_cache) {
5792 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
5794 bs->open_flags |= BDRV_O_INACTIVE;
5795 error_propagate(errp, local_err);
5800 FOR_EACH_DIRTY_BITMAP(bs, bm) {
5801 bdrv_dirty_bitmap_skip_store(bm, false);
5804 ret = refresh_total_sectors(bs, bs->total_sectors);
5806 bs->open_flags |= BDRV_O_INACTIVE;
5807 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5812 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5813 if (parent->klass->activate) {
5814 parent->klass->activate(parent, &local_err);
5816 bs->open_flags |= BDRV_O_INACTIVE;
5817 error_propagate(errp, local_err);
5826 void bdrv_invalidate_cache_all(Error **errp)
5828 BlockDriverState *bs;
5829 BdrvNextIterator it;
5831 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5832 AioContext *aio_context = bdrv_get_aio_context(bs);
5835 aio_context_acquire(aio_context);
5836 ret = bdrv_invalidate_cache(bs, errp);
5837 aio_context_release(aio_context);
5839 bdrv_next_cleanup(&it);
5845 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5849 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5850 if (parent->klass->parent_is_bds) {
5851 BlockDriverState *parent_bs = parent->opaque;
5852 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5861 static int bdrv_inactivate_recurse(BlockDriverState *bs)
5863 BdrvChild *child, *parent;
5864 bool tighten_restrictions;
5865 uint64_t perm, shared_perm;
5872 /* Make sure that we don't inactivate a child before its parent.
5873 * It will be covered by recursion from the yet active parent. */
5874 if (bdrv_has_bds_parent(bs, true)) {
5878 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5880 /* Inactivate this node */
5881 if (bs->drv->bdrv_inactivate) {
5882 ret = bs->drv->bdrv_inactivate(bs);
5888 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5889 if (parent->klass->inactivate) {
5890 ret = parent->klass->inactivate(parent);
5897 bs->open_flags |= BDRV_O_INACTIVE;
5899 /* Update permissions, they may differ for inactive nodes */
5900 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5901 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL,
5902 &tighten_restrictions, NULL);
5903 assert(tighten_restrictions == false);
5905 /* We only tried to loosen restrictions, so errors are not fatal */
5906 bdrv_abort_perm_update(bs);
5908 bdrv_set_perm(bs, perm, shared_perm);
5912 /* Recursively inactivate children */
5913 QLIST_FOREACH(child, &bs->children, next) {
5914 ret = bdrv_inactivate_recurse(child->bs);
5923 int bdrv_inactivate_all(void)
5925 BlockDriverState *bs = NULL;
5926 BdrvNextIterator it;
5928 GSList *aio_ctxs = NULL, *ctx;
5930 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5931 AioContext *aio_context = bdrv_get_aio_context(bs);
5933 if (!g_slist_find(aio_ctxs, aio_context)) {
5934 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5935 aio_context_acquire(aio_context);
5939 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5940 /* Nodes with BDS parents are covered by recursion from the last
5941 * parent that gets inactivated. Don't inactivate them a second
5942 * time if that has already happened. */
5943 if (bdrv_has_bds_parent(bs, false)) {
5946 ret = bdrv_inactivate_recurse(bs);
5948 bdrv_next_cleanup(&it);
5954 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5955 AioContext *aio_context = ctx->data;
5956 aio_context_release(aio_context);
5958 g_slist_free(aio_ctxs);
5963 /**************************************************************/
5964 /* removable device support */
5967 * Return TRUE if the media is present
5969 bool bdrv_is_inserted(BlockDriverState *bs)
5971 BlockDriver *drv = bs->drv;
5977 if (drv->bdrv_is_inserted) {
5978 return drv->bdrv_is_inserted(bs);
5980 QLIST_FOREACH(child, &bs->children, next) {
5981 if (!bdrv_is_inserted(child->bs)) {
5989 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5991 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5993 BlockDriver *drv = bs->drv;
5995 if (drv && drv->bdrv_eject) {
5996 drv->bdrv_eject(bs, eject_flag);
6001 * Lock or unlock the media (if it is locked, the user won't be able
6002 * to eject it manually).
6004 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6006 BlockDriver *drv = bs->drv;
6008 trace_bdrv_lock_medium(bs, locked);
6010 if (drv && drv->bdrv_lock_medium) {
6011 drv->bdrv_lock_medium(bs, locked);
6015 /* Get a reference to bs */
6016 void bdrv_ref(BlockDriverState *bs)
6021 /* Release a previously grabbed reference to bs.
6022 * If after releasing, reference count is zero, the BlockDriverState is
6024 void bdrv_unref(BlockDriverState *bs)
6029 assert(bs->refcnt > 0);
6030 if (--bs->refcnt == 0) {
6035 struct BdrvOpBlocker {
6037 QLIST_ENTRY(BdrvOpBlocker) list;
6040 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6042 BdrvOpBlocker *blocker;
6043 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6044 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6045 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6046 error_propagate_prepend(errp, error_copy(blocker->reason),
6047 "Node '%s' is busy: ",
6048 bdrv_get_device_or_node_name(bs));
6054 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6056 BdrvOpBlocker *blocker;
6057 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6059 blocker = g_new0(BdrvOpBlocker, 1);
6060 blocker->reason = reason;
6061 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6064 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6066 BdrvOpBlocker *blocker, *next;
6067 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6068 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6069 if (blocker->reason == reason) {
6070 QLIST_REMOVE(blocker, list);
6076 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6079 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6080 bdrv_op_block(bs, i, reason);
6084 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6087 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6088 bdrv_op_unblock(bs, i, reason);
6092 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6096 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6097 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6104 void bdrv_img_create(const char *filename, const char *fmt,
6105 const char *base_filename, const char *base_fmt,
6106 char *options, uint64_t img_size, int flags, bool quiet,
6109 QemuOptsList *create_opts = NULL;
6110 QemuOpts *opts = NULL;
6111 const char *backing_fmt, *backing_file;
6113 BlockDriver *drv, *proto_drv;
6114 Error *local_err = NULL;
6117 /* Find driver and parse its options */
6118 drv = bdrv_find_format(fmt);
6120 error_setg(errp, "Unknown file format '%s'", fmt);
6124 proto_drv = bdrv_find_protocol(filename, true, errp);
6129 if (!drv->create_opts) {
6130 error_setg(errp, "Format driver '%s' does not support image creation",
6135 if (!proto_drv->create_opts) {
6136 error_setg(errp, "Protocol driver '%s' does not support image creation",
6137 proto_drv->format_name);
6141 /* Create parameter list */
6142 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6143 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6145 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6147 /* Parse -o options */
6149 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6154 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6155 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6156 } else if (img_size != UINT64_C(-1)) {
6157 error_setg(errp, "The image size must be specified only once");
6161 if (base_filename) {
6162 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6164 error_setg(errp, "Backing file not supported for file format '%s'",
6171 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
6172 error_setg(errp, "Backing file format not supported for file "
6173 "format '%s'", fmt);
6178 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6180 if (!strcmp(filename, backing_file)) {
6181 error_setg(errp, "Error: Trying to create an image with the "
6182 "same filename as the backing file");
6185 if (backing_file[0] == '\0') {
6186 error_setg(errp, "Expected backing file name, got empty string");
6191 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
6193 /* The size for the image must always be specified, unless we have a backing
6194 * file and we have not been forbidden from opening it. */
6195 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6196 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6197 BlockDriverState *bs;
6200 QDict *backing_options = NULL;
6203 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6208 assert(full_backing);
6210 /* backing files always opened read-only */
6212 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
6214 backing_options = qdict_new();
6216 qdict_put_str(backing_options, "driver", backing_fmt);
6218 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
6220 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6222 g_free(full_backing);
6224 error_append_hint(&local_err, "Could not open backing image.\n");
6228 warn_report("Deprecated use of backing file without explicit "
6229 "backing format (detected format of %s)",
6230 bs->drv->format_name);
6231 if (bs->drv != &bdrv_raw) {
6233 * A probe of raw deserves the most attention:
6234 * leaving the backing format out of the image
6235 * will ensure bs->probed is set (ensuring we
6236 * don't accidentally commit into the backing
6237 * file), and allow more spots to warn the users
6238 * to fix their toolchain when opening this image
6239 * later. For other images, we can safely record
6240 * the format that we probed.
6242 backing_fmt = bs->drv->format_name;
6243 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, backing_fmt,
6248 /* Opened BS, have no size */
6249 size = bdrv_getlength(bs);
6251 error_setg_errno(errp, -size, "Could not get size of '%s'",
6256 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
6260 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6261 } else if (backing_file && !backing_fmt) {
6262 warn_report("Deprecated use of unopened backing file without "
6263 "explicit backing format, use of this image requires "
6264 "potentially unsafe format probing");
6268 error_setg(errp, "Image creation needs a size parameter");
6273 printf("Formatting '%s', fmt=%s ", filename, fmt);
6274 qemu_opts_print(opts, " ");
6279 ret = bdrv_create(drv, filename, opts, &local_err);
6281 if (ret == -EFBIG) {
6282 /* This is generally a better message than whatever the driver would
6283 * deliver (especially because of the cluster_size_hint), since that
6284 * is most probably not much different from "image too large". */
6285 const char *cluster_size_hint = "";
6286 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
6287 cluster_size_hint = " (try using a larger cluster size)";
6289 error_setg(errp, "The image size is too large for file format '%s'"
6290 "%s", fmt, cluster_size_hint);
6291 error_free(local_err);
6296 qemu_opts_del(opts);
6297 qemu_opts_free(create_opts);
6298 error_propagate(errp, local_err);
6301 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6303 return bs ? bs->aio_context : qemu_get_aio_context();
6306 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
6308 Coroutine *self = qemu_coroutine_self();
6309 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
6310 AioContext *new_ctx;
6313 * Increase bs->in_flight to ensure that this operation is completed before
6314 * moving the node to a different AioContext. Read new_ctx only afterwards.
6316 bdrv_inc_in_flight(bs);
6318 new_ctx = bdrv_get_aio_context(bs);
6319 aio_co_reschedule_self(new_ctx);
6323 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
6325 aio_co_reschedule_self(old_ctx);
6326 bdrv_dec_in_flight(bs);
6329 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6331 aio_co_enter(bdrv_get_aio_context(bs), co);
6334 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6336 QLIST_REMOVE(ban, list);
6340 static void bdrv_detach_aio_context(BlockDriverState *bs)
6342 BdrvAioNotifier *baf, *baf_tmp;
6344 assert(!bs->walking_aio_notifiers);
6345 bs->walking_aio_notifiers = true;
6346 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6348 bdrv_do_remove_aio_context_notifier(baf);
6350 baf->detach_aio_context(baf->opaque);
6353 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6354 * remove remaining aio notifiers if we aren't called again.
6356 bs->walking_aio_notifiers = false;
6358 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
6359 bs->drv->bdrv_detach_aio_context(bs);
6362 if (bs->quiesce_counter) {
6363 aio_enable_external(bs->aio_context);
6365 bs->aio_context = NULL;
6368 static void bdrv_attach_aio_context(BlockDriverState *bs,
6369 AioContext *new_context)
6371 BdrvAioNotifier *ban, *ban_tmp;
6373 if (bs->quiesce_counter) {
6374 aio_disable_external(new_context);
6377 bs->aio_context = new_context;
6379 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
6380 bs->drv->bdrv_attach_aio_context(bs, new_context);
6383 assert(!bs->walking_aio_notifiers);
6384 bs->walking_aio_notifiers = true;
6385 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6387 bdrv_do_remove_aio_context_notifier(ban);
6389 ban->attached_aio_context(new_context, ban->opaque);
6392 bs->walking_aio_notifiers = false;
6396 * Changes the AioContext used for fd handlers, timers, and BHs by this
6397 * BlockDriverState and all its children and parents.
6399 * Must be called from the main AioContext.
6401 * The caller must own the AioContext lock for the old AioContext of bs, but it
6402 * must not own the AioContext lock for new_context (unless new_context is the
6403 * same as the current context of bs).
6405 * @ignore will accumulate all visited BdrvChild object. The caller is
6406 * responsible for freeing the list afterwards.
6408 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6409 AioContext *new_context, GSList **ignore)
6411 AioContext *old_context = bdrv_get_aio_context(bs);
6414 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6416 if (old_context == new_context) {
6420 bdrv_drained_begin(bs);
6422 QLIST_FOREACH(child, &bs->children, next) {
6423 if (g_slist_find(*ignore, child)) {
6426 *ignore = g_slist_prepend(*ignore, child);
6427 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6429 QLIST_FOREACH(child, &bs->parents, next_parent) {
6430 if (g_slist_find(*ignore, child)) {
6433 assert(child->klass->set_aio_ctx);
6434 *ignore = g_slist_prepend(*ignore, child);
6435 child->klass->set_aio_ctx(child, new_context, ignore);
6438 bdrv_detach_aio_context(bs);
6440 /* Acquire the new context, if necessary */
6441 if (qemu_get_aio_context() != new_context) {
6442 aio_context_acquire(new_context);
6445 bdrv_attach_aio_context(bs, new_context);
6448 * If this function was recursively called from
6449 * bdrv_set_aio_context_ignore(), there may be nodes in the
6450 * subtree that have not yet been moved to the new AioContext.
6451 * Release the old one so bdrv_drained_end() can poll them.
6453 if (qemu_get_aio_context() != old_context) {
6454 aio_context_release(old_context);
6457 bdrv_drained_end(bs);
6459 if (qemu_get_aio_context() != old_context) {
6460 aio_context_acquire(old_context);
6462 if (qemu_get_aio_context() != new_context) {
6463 aio_context_release(new_context);
6467 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6468 GSList **ignore, Error **errp)
6470 if (g_slist_find(*ignore, c)) {
6473 *ignore = g_slist_prepend(*ignore, c);
6476 * A BdrvChildClass that doesn't handle AioContext changes cannot
6477 * tolerate any AioContext changes
6479 if (!c->klass->can_set_aio_ctx) {
6480 char *user = bdrv_child_user_desc(c);
6481 error_setg(errp, "Changing iothreads is not supported by %s", user);
6485 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
6486 assert(!errp || *errp);
6492 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6493 GSList **ignore, Error **errp)
6495 if (g_slist_find(*ignore, c)) {
6498 *ignore = g_slist_prepend(*ignore, c);
6499 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6502 /* @ignore will accumulate all visited BdrvChild object. The caller is
6503 * responsible for freeing the list afterwards. */
6504 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6505 GSList **ignore, Error **errp)
6509 if (bdrv_get_aio_context(bs) == ctx) {
6513 QLIST_FOREACH(c, &bs->parents, next_parent) {
6514 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
6518 QLIST_FOREACH(c, &bs->children, next) {
6519 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
6527 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6528 BdrvChild *ignore_child, Error **errp)
6533 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6534 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
6535 g_slist_free(ignore);
6541 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6542 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
6543 g_slist_free(ignore);
6548 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6551 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
6554 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
6555 void (*attached_aio_context)(AioContext *new_context, void *opaque),
6556 void (*detach_aio_context)(void *opaque), void *opaque)
6558 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
6559 *ban = (BdrvAioNotifier){
6560 .attached_aio_context = attached_aio_context,
6561 .detach_aio_context = detach_aio_context,
6565 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
6568 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
6569 void (*attached_aio_context)(AioContext *,
6571 void (*detach_aio_context)(void *),
6574 BdrvAioNotifier *ban, *ban_next;
6576 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
6577 if (ban->attached_aio_context == attached_aio_context &&
6578 ban->detach_aio_context == detach_aio_context &&
6579 ban->opaque == opaque &&
6580 ban->deleted == false)
6582 if (bs->walking_aio_notifiers) {
6583 ban->deleted = true;
6585 bdrv_do_remove_aio_context_notifier(ban);
6594 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
6595 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
6600 error_setg(errp, "Node is ejected");
6603 if (!bs->drv->bdrv_amend_options) {
6604 error_setg(errp, "Block driver '%s' does not support option amendment",
6605 bs->drv->format_name);
6608 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
6609 cb_opaque, force, errp);
6613 * This function checks whether the given @to_replace is allowed to be
6614 * replaced by a node that always shows the same data as @bs. This is
6615 * used for example to verify whether the mirror job can replace
6616 * @to_replace by the target mirrored from @bs.
6617 * To be replaceable, @bs and @to_replace may either be guaranteed to
6618 * always show the same data (because they are only connected through
6619 * filters), or some driver may allow replacing one of its children
6620 * because it can guarantee that this child's data is not visible at
6621 * all (for example, for dissenting quorum children that have no other
6624 bool bdrv_recurse_can_replace(BlockDriverState *bs,
6625 BlockDriverState *to_replace)
6627 BlockDriverState *filtered;
6629 if (!bs || !bs->drv) {
6633 if (bs == to_replace) {
6637 /* See what the driver can do */
6638 if (bs->drv->bdrv_recurse_can_replace) {
6639 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
6642 /* For filters without an own implementation, we can recurse on our own */
6643 filtered = bdrv_filter_bs(bs);
6645 return bdrv_recurse_can_replace(filtered, to_replace);
6653 * Check whether the given @node_name can be replaced by a node that
6654 * has the same data as @parent_bs. If so, return @node_name's BDS;
6657 * @node_name must be a (recursive) *child of @parent_bs (or this
6658 * function will return NULL).
6660 * The result (whether the node can be replaced or not) is only valid
6661 * for as long as no graph or permission changes occur.
6663 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
6664 const char *node_name, Error **errp)
6666 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
6667 AioContext *aio_context;
6669 if (!to_replace_bs) {
6670 error_setg(errp, "Node name '%s' not found", node_name);
6674 aio_context = bdrv_get_aio_context(to_replace_bs);
6675 aio_context_acquire(aio_context);
6677 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
6678 to_replace_bs = NULL;
6682 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6683 * most non filter in order to prevent data corruption.
6684 * Another benefit is that this tests exclude backing files which are
6685 * blocked by the backing blockers.
6687 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
6688 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
6689 "because it cannot be guaranteed that doing so would not "
6690 "lead to an abrupt change of visible data",
6691 node_name, parent_bs->node_name);
6692 to_replace_bs = NULL;
6697 aio_context_release(aio_context);
6698 return to_replace_bs;
6702 * Iterates through the list of runtime option keys that are said to
6703 * be "strong" for a BDS. An option is called "strong" if it changes
6704 * a BDS's data. For example, the null block driver's "size" and
6705 * "read-zeroes" options are strong, but its "latency-ns" option is
6708 * If a key returned by this function ends with a dot, all options
6709 * starting with that prefix are strong.
6711 static const char *const *strong_options(BlockDriverState *bs,
6712 const char *const *curopt)
6714 static const char *const global_options[] = {
6715 "driver", "filename", NULL
6719 return &global_options[0];
6723 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
6724 curopt = bs->drv->strong_runtime_opts;
6727 return (curopt && *curopt) ? curopt : NULL;
6731 * Copies all strong runtime options from bs->options to the given
6732 * QDict. The set of strong option keys is determined by invoking
6735 * Returns true iff any strong option was present in bs->options (and
6736 * thus copied to the target QDict) with the exception of "filename"
6737 * and "driver". The caller is expected to use this value to decide
6738 * whether the existence of strong options prevents the generation of
6741 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
6743 bool found_any = false;
6744 const char *const *option_name = NULL;
6750 while ((option_name = strong_options(bs, option_name))) {
6751 bool option_given = false;
6753 assert(strlen(*option_name) > 0);
6754 if ((*option_name)[strlen(*option_name) - 1] != '.') {
6755 QObject *entry = qdict_get(bs->options, *option_name);
6760 qdict_put_obj(d, *option_name, qobject_ref(entry));
6761 option_given = true;
6763 const QDictEntry *entry;
6764 for (entry = qdict_first(bs->options); entry;
6765 entry = qdict_next(bs->options, entry))
6767 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
6768 qdict_put_obj(d, qdict_entry_key(entry),
6769 qobject_ref(qdict_entry_value(entry)));
6770 option_given = true;
6775 /* While "driver" and "filename" need to be included in a JSON filename,
6776 * their existence does not prohibit generation of a plain filename. */
6777 if (!found_any && option_given &&
6778 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
6784 if (!qdict_haskey(d, "driver")) {
6785 /* Drivers created with bdrv_new_open_driver() may not have a
6786 * @driver option. Add it here. */
6787 qdict_put_str(d, "driver", bs->drv->format_name);
6793 /* Note: This function may return false positives; it may return true
6794 * even if opening the backing file specified by bs's image header
6795 * would result in exactly bs->backing. */
6796 bool bdrv_backing_overridden(BlockDriverState *bs)
6799 return strcmp(bs->auto_backing_file,
6800 bs->backing->bs->filename);
6802 /* No backing BDS, so if the image header reports any backing
6803 * file, it must have been suppressed */
6804 return bs->auto_backing_file[0] != '\0';
6808 /* Updates the following BDS fields:
6809 * - exact_filename: A filename which may be used for opening a block device
6810 * which (mostly) equals the given BDS (even without any
6811 * other options; so reading and writing must return the same
6812 * results, but caching etc. may be different)
6813 * - full_open_options: Options which, when given when opening a block device
6814 * (without a filename), result in a BDS (mostly)
6815 * equalling the given one
6816 * - filename: If exact_filename is set, it is copied here. Otherwise,
6817 * full_open_options is converted to a JSON object, prefixed with
6818 * "json:" (for use through the JSON pseudo protocol) and put here.
6820 void bdrv_refresh_filename(BlockDriverState *bs)
6822 BlockDriver *drv = bs->drv;
6824 BlockDriverState *primary_child_bs;
6826 bool backing_overridden;
6827 bool generate_json_filename; /* Whether our default implementation should
6828 fill exact_filename (false) or not (true) */
6834 /* This BDS's file name may depend on any of its children's file names, so
6835 * refresh those first */
6836 QLIST_FOREACH(child, &bs->children, next) {
6837 bdrv_refresh_filename(child->bs);
6841 /* For implicit nodes, just copy everything from the single child */
6842 child = QLIST_FIRST(&bs->children);
6843 assert(QLIST_NEXT(child, next) == NULL);
6845 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
6846 child->bs->exact_filename);
6847 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
6849 qobject_unref(bs->full_open_options);
6850 bs->full_open_options = qobject_ref(child->bs->full_open_options);
6855 backing_overridden = bdrv_backing_overridden(bs);
6857 if (bs->open_flags & BDRV_O_NO_IO) {
6858 /* Without I/O, the backing file does not change anything.
6859 * Therefore, in such a case (primarily qemu-img), we can
6860 * pretend the backing file has not been overridden even if
6861 * it technically has been. */
6862 backing_overridden = false;
6865 /* Gather the options QDict */
6867 generate_json_filename = append_strong_runtime_options(opts, bs);
6868 generate_json_filename |= backing_overridden;
6870 if (drv->bdrv_gather_child_options) {
6871 /* Some block drivers may not want to present all of their children's
6872 * options, or name them differently from BdrvChild.name */
6873 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
6875 QLIST_FOREACH(child, &bs->children, next) {
6876 if (child == bs->backing && !backing_overridden) {
6877 /* We can skip the backing BDS if it has not been overridden */
6881 qdict_put(opts, child->name,
6882 qobject_ref(child->bs->full_open_options));
6885 if (backing_overridden && !bs->backing) {
6886 /* Force no backing file */
6887 qdict_put_null(opts, "backing");
6891 qobject_unref(bs->full_open_options);
6892 bs->full_open_options = opts;
6894 primary_child_bs = bdrv_primary_bs(bs);
6896 if (drv->bdrv_refresh_filename) {
6897 /* Obsolete information is of no use here, so drop the old file name
6898 * information before refreshing it */
6899 bs->exact_filename[0] = '\0';
6901 drv->bdrv_refresh_filename(bs);
6902 } else if (primary_child_bs) {
6904 * Try to reconstruct valid information from the underlying
6905 * file -- this only works for format nodes (filter nodes
6906 * cannot be probed and as such must be selected by the user
6907 * either through an options dict, or through a special
6908 * filename which the filter driver must construct in its
6909 * .bdrv_refresh_filename() implementation).
6912 bs->exact_filename[0] = '\0';
6915 * We can use the underlying file's filename if:
6916 * - it has a filename,
6917 * - the current BDS is not a filter,
6918 * - the file is a protocol BDS, and
6919 * - opening that file (as this BDS's format) will automatically create
6920 * the BDS tree we have right now, that is:
6921 * - the user did not significantly change this BDS's behavior with
6922 * some explicit (strong) options
6923 * - no non-file child of this BDS has been overridden by the user
6924 * Both of these conditions are represented by generate_json_filename.
6926 if (primary_child_bs->exact_filename[0] &&
6927 primary_child_bs->drv->bdrv_file_open &&
6928 !drv->is_filter && !generate_json_filename)
6930 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
6934 if (bs->exact_filename[0]) {
6935 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
6937 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
6938 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
6939 qstring_get_str(json)) >= sizeof(bs->filename)) {
6940 /* Give user a hint if we truncated things. */
6941 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
6943 qobject_unref(json);
6947 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
6949 BlockDriver *drv = bs->drv;
6950 BlockDriverState *child_bs;
6953 error_setg(errp, "Node '%s' is ejected", bs->node_name);
6957 if (drv->bdrv_dirname) {
6958 return drv->bdrv_dirname(bs, errp);
6961 child_bs = bdrv_primary_bs(bs);
6963 return bdrv_dirname(child_bs, errp);
6966 bdrv_refresh_filename(bs);
6967 if (bs->exact_filename[0] != '\0') {
6968 return path_combine(bs->exact_filename, "");
6971 error_setg(errp, "Cannot generate a base directory for %s nodes",
6977 * Hot add/remove a BDS's child. So the user can take a child offline when
6978 * it is broken and take a new child online
6980 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
6984 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
6985 error_setg(errp, "The node %s does not support adding a child",
6986 bdrv_get_device_or_node_name(parent_bs));
6990 if (!QLIST_EMPTY(&child_bs->parents)) {
6991 error_setg(errp, "The node %s already has a parent",
6992 child_bs->node_name);
6996 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
6999 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7003 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7004 error_setg(errp, "The node %s does not support removing a child",
7005 bdrv_get_device_or_node_name(parent_bs));
7009 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7016 error_setg(errp, "The node %s does not have a child named %s",
7017 bdrv_get_device_or_node_name(parent_bs),
7018 bdrv_get_device_or_node_name(child->bs));
7022 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7025 int bdrv_make_empty(BdrvChild *c, Error **errp)
7027 BlockDriver *drv = c->bs->drv;
7030 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7032 if (!drv->bdrv_make_empty) {
7033 error_setg(errp, "%s does not support emptying nodes",
7038 ret = drv->bdrv_make_empty(c->bs);
7040 error_setg_errno(errp, -ret, "Failed to empty %s",
7049 * Return the child that @bs acts as an overlay for, and from which data may be
7050 * copied in COW or COR operations. Usually this is the backing file.
7052 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7054 if (!bs || !bs->drv) {
7058 if (bs->drv->is_filter) {
7066 assert(bs->backing->role & BDRV_CHILD_COW);
7071 * If @bs acts as a filter for exactly one of its children, return
7074 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7078 if (!bs || !bs->drv) {
7082 if (!bs->drv->is_filter) {
7086 /* Only one of @backing or @file may be used */
7087 assert(!(bs->backing && bs->file));
7089 c = bs->backing ?: bs->file;
7094 assert(c->role & BDRV_CHILD_FILTERED);
7099 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7100 * whichever is non-NULL.
7102 * Return NULL if both are NULL.
7104 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7106 BdrvChild *cow_child = bdrv_cow_child(bs);
7107 BdrvChild *filter_child = bdrv_filter_child(bs);
7109 /* Filter nodes cannot have COW backing files */
7110 assert(!(cow_child && filter_child));
7112 return cow_child ?: filter_child;
7116 * Return the primary child of this node: For filters, that is the
7117 * filtered child. For other nodes, that is usually the child storing
7119 * (A generally more helpful description is that this is (usually) the
7120 * child that has the same filename as @bs.)
7122 * Drivers do not necessarily have a primary child; for example quorum
7125 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
7127 BdrvChild *c, *found = NULL;
7129 QLIST_FOREACH(c, &bs->children, next) {
7130 if (c->role & BDRV_CHILD_PRIMARY) {
7139 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
7140 bool stop_on_explicit_filter)
7148 while (!(stop_on_explicit_filter && !bs->implicit)) {
7149 c = bdrv_filter_child(bs);
7152 * A filter that is embedded in a working block graph must
7153 * have a child. Assert this here so this function does
7154 * not return a filter node that is not expected by the
7157 assert(!bs->drv || !bs->drv->is_filter);
7163 * Note that this treats nodes with bs->drv == NULL as not being
7164 * filters (bs->drv == NULL should be replaced by something else
7166 * The advantage of this behavior is that this function will thus
7167 * always return a non-NULL value (given a non-NULL @bs).
7174 * Return the first BDS that has not been added implicitly or that
7175 * does not have a filtered child down the chain starting from @bs
7176 * (including @bs itself).
7178 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
7180 return bdrv_do_skip_filters(bs, true);
7184 * Return the first BDS that does not have a filtered child down the
7185 * chain starting from @bs (including @bs itself).
7187 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
7189 return bdrv_do_skip_filters(bs, false);
7193 * For a backing chain, return the first non-filter backing image of
7194 * the first non-filter image.
7196 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
7198 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));