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/fuse.h"
30 #include "block/nbd.h"
31 #include "block/qdict.h"
32 #include "qemu/error-report.h"
33 #include "block/module_block.h"
34 #include "qemu/main-loop.h"
35 #include "qemu/module.h"
36 #include "qapi/error.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qjson.h"
39 #include "qapi/qmp/qnull.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qapi/qobject-output-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "sysemu/block-backend.h"
44 #include "sysemu/sysemu.h"
45 #include "qemu/notify.h"
46 #include "qemu/option.h"
47 #include "qemu/coroutine.h"
48 #include "block/qapi.h"
49 #include "qemu/timer.h"
50 #include "qemu/cutils.h"
52 #include "block/coroutines.h"
55 #include <sys/ioctl.h>
56 #include <sys/queue.h>
66 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
68 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
71 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
72 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
74 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
75 QLIST_HEAD_INITIALIZER(bdrv_drivers);
77 static BlockDriverState *bdrv_open_inherit(const char *filename,
78 const char *reference,
79 QDict *options, int flags,
80 BlockDriverState *parent,
81 const BdrvChildClass *child_class,
82 BdrvChildRole child_role,
85 /* If non-zero, use only whitelisted block drivers */
86 static int use_bdrv_whitelist;
89 static int is_windows_drive_prefix(const char *filename)
91 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
92 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
96 int is_windows_drive(const char *filename)
98 if (is_windows_drive_prefix(filename) &&
101 if (strstart(filename, "\\\\.\\", NULL) ||
102 strstart(filename, "//./", NULL))
108 size_t bdrv_opt_mem_align(BlockDriverState *bs)
110 if (!bs || !bs->drv) {
111 /* page size or 4k (hdd sector size) should be on the safe side */
112 return MAX(4096, qemu_real_host_page_size);
115 return bs->bl.opt_mem_alignment;
118 size_t bdrv_min_mem_align(BlockDriverState *bs)
120 if (!bs || !bs->drv) {
121 /* page size or 4k (hdd sector size) should be on the safe side */
122 return MAX(4096, qemu_real_host_page_size);
125 return bs->bl.min_mem_alignment;
128 /* check if the path starts with "<protocol>:" */
129 int path_has_protocol(const char *path)
134 if (is_windows_drive(path) ||
135 is_windows_drive_prefix(path)) {
138 p = path + strcspn(path, ":/\\");
140 p = path + strcspn(path, ":/");
146 int path_is_absolute(const char *path)
149 /* specific case for names like: "\\.\d:" */
150 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
153 return (*path == '/' || *path == '\\');
155 return (*path == '/');
159 /* if filename is absolute, just return its duplicate. Otherwise, build a
160 path to it by considering it is relative to base_path. URL are
162 char *path_combine(const char *base_path, const char *filename)
164 const char *protocol_stripped = NULL;
169 if (path_is_absolute(filename)) {
170 return g_strdup(filename);
173 if (path_has_protocol(base_path)) {
174 protocol_stripped = strchr(base_path, ':');
175 if (protocol_stripped) {
179 p = protocol_stripped ?: base_path;
181 p1 = strrchr(base_path, '/');
185 p2 = strrchr(base_path, '\\');
186 if (!p1 || p2 > p1) {
201 result = g_malloc(len + strlen(filename) + 1);
202 memcpy(result, base_path, len);
203 strcpy(result + len, filename);
209 * Helper function for bdrv_parse_filename() implementations to remove optional
210 * protocol prefixes (especially "file:") from a filename and for putting the
211 * stripped filename into the options QDict if there is such a prefix.
213 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
216 if (strstart(filename, prefix, &filename)) {
217 /* Stripping the explicit protocol prefix may result in a protocol
218 * prefix being (wrongly) detected (if the filename contains a colon) */
219 if (path_has_protocol(filename)) {
220 GString *fat_filename;
222 /* This means there is some colon before the first slash; therefore,
223 * this cannot be an absolute path */
224 assert(!path_is_absolute(filename));
226 /* And we can thus fix the protocol detection issue by prefixing it
228 fat_filename = g_string_new("./");
229 g_string_append(fat_filename, filename);
231 assert(!path_has_protocol(fat_filename->str));
233 qdict_put(options, "filename",
234 qstring_from_gstring(fat_filename));
236 /* If no protocol prefix was detected, we can use the shortened
238 qdict_put_str(options, "filename", filename);
244 /* Returns whether the image file is opened as read-only. Note that this can
245 * return false and writing to the image file is still not possible because the
246 * image is inactivated. */
247 bool bdrv_is_read_only(BlockDriverState *bs)
249 return bs->read_only;
252 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
253 bool ignore_allow_rdw, Error **errp)
255 /* Do not set read_only if copy_on_read is enabled */
256 if (bs->copy_on_read && read_only) {
257 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
258 bdrv_get_device_or_node_name(bs));
262 /* Do not clear read_only if it is prohibited */
263 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
266 error_setg(errp, "Node '%s' is read only",
267 bdrv_get_device_or_node_name(bs));
275 * Called by a driver that can only provide a read-only image.
277 * Returns 0 if the node is already read-only or it could switch the node to
278 * read-only because BDRV_O_AUTO_RDONLY is set.
280 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
281 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
282 * is not NULL, it is used as the error message for the Error object.
284 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
289 if (!(bs->open_flags & BDRV_O_RDWR)) {
292 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
296 ret = bdrv_can_set_read_only(bs, true, false, NULL);
301 bs->read_only = true;
302 bs->open_flags &= ~BDRV_O_RDWR;
307 error_setg(errp, "%s", errmsg ?: "Image is read-only");
312 * If @backing is empty, this function returns NULL without setting
313 * @errp. In all other cases, NULL will only be returned with @errp
316 * Therefore, a return value of NULL without @errp set means that
317 * there is no backing file; if @errp is set, there is one but its
318 * absolute filename cannot be generated.
320 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
324 if (backing[0] == '\0') {
326 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
327 return g_strdup(backing);
328 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
329 error_setg(errp, "Cannot use relative backing file names for '%s'",
333 return path_combine(backed, backing);
338 * If @filename is empty or NULL, this function returns NULL without
339 * setting @errp. In all other cases, NULL will only be returned with
342 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
343 const char *filename, Error **errp)
345 char *dir, *full_name;
347 if (!filename || filename[0] == '\0') {
349 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
350 return g_strdup(filename);
353 dir = bdrv_dirname(relative_to, errp);
358 full_name = g_strconcat(dir, filename, NULL);
363 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
365 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
368 void bdrv_register(BlockDriver *bdrv)
370 assert(bdrv->format_name);
371 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
374 BlockDriverState *bdrv_new(void)
376 BlockDriverState *bs;
379 bs = g_new0(BlockDriverState, 1);
380 QLIST_INIT(&bs->dirty_bitmaps);
381 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
382 QLIST_INIT(&bs->op_blockers[i]);
384 notifier_with_return_list_init(&bs->before_write_notifiers);
385 qemu_co_mutex_init(&bs->reqs_lock);
386 qemu_mutex_init(&bs->dirty_bitmap_mutex);
388 bs->aio_context = qemu_get_aio_context();
390 qemu_co_queue_init(&bs->flush_queue);
392 for (i = 0; i < bdrv_drain_all_count; i++) {
393 bdrv_drained_begin(bs);
396 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
401 static BlockDriver *bdrv_do_find_format(const char *format_name)
405 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
406 if (!strcmp(drv1->format_name, format_name)) {
414 BlockDriver *bdrv_find_format(const char *format_name)
419 drv1 = bdrv_do_find_format(format_name);
424 /* The driver isn't registered, maybe we need to load a module */
425 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
426 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
427 block_module_load_one(block_driver_modules[i].library_name);
432 return bdrv_do_find_format(format_name);
435 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
437 static const char *whitelist_rw[] = {
438 CONFIG_BDRV_RW_WHITELIST
441 static const char *whitelist_ro[] = {
442 CONFIG_BDRV_RO_WHITELIST
447 if (!whitelist_rw[0] && !whitelist_ro[0]) {
448 return 1; /* no whitelist, anything goes */
451 for (p = whitelist_rw; *p; p++) {
452 if (!strcmp(format_name, *p)) {
457 for (p = whitelist_ro; *p; p++) {
458 if (!strcmp(format_name, *p)) {
466 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
468 return bdrv_format_is_whitelisted(drv->format_name, read_only);
471 bool bdrv_uses_whitelist(void)
473 return use_bdrv_whitelist;
476 typedef struct CreateCo {
484 static void coroutine_fn bdrv_create_co_entry(void *opaque)
486 Error *local_err = NULL;
489 CreateCo *cco = opaque;
492 ret = cco->drv->bdrv_co_create_opts(cco->drv,
493 cco->filename, cco->opts, &local_err);
494 error_propagate(&cco->err, local_err);
498 int bdrv_create(BlockDriver *drv, const char* filename,
499 QemuOpts *opts, Error **errp)
506 .filename = g_strdup(filename),
512 if (!drv->bdrv_co_create_opts) {
513 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
518 if (qemu_in_coroutine()) {
519 /* Fast-path if already in coroutine context */
520 bdrv_create_co_entry(&cco);
522 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
523 qemu_coroutine_enter(co);
524 while (cco.ret == NOT_DONE) {
525 aio_poll(qemu_get_aio_context(), true);
532 error_propagate(errp, cco.err);
534 error_setg_errno(errp, -ret, "Could not create image");
539 g_free(cco.filename);
544 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
545 * least the given @minimum_size.
547 * On success, return @blk's actual length.
548 * Otherwise, return -errno.
550 static int64_t create_file_fallback_truncate(BlockBackend *blk,
551 int64_t minimum_size, Error **errp)
553 Error *local_err = NULL;
557 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
559 if (ret < 0 && ret != -ENOTSUP) {
560 error_propagate(errp, local_err);
564 size = blk_getlength(blk);
566 error_free(local_err);
567 error_setg_errno(errp, -size,
568 "Failed to inquire the new image file's length");
572 if (size < minimum_size) {
573 /* Need to grow the image, but we failed to do that */
574 error_propagate(errp, local_err);
578 error_free(local_err);
585 * Helper function for bdrv_create_file_fallback(): Zero the first
586 * sector to remove any potentially pre-existing image header.
588 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
589 int64_t current_size,
592 int64_t bytes_to_clear;
595 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
596 if (bytes_to_clear) {
597 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
599 error_setg_errno(errp, -ret,
600 "Failed to clear the new image's first sector");
609 * Simple implementation of bdrv_co_create_opts for protocol drivers
610 * which only support creation via opening a file
611 * (usually existing raw storage device)
613 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
614 const char *filename,
622 PreallocMode prealloc;
623 Error *local_err = NULL;
626 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
627 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
628 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
629 PREALLOC_MODE_OFF, &local_err);
632 error_propagate(errp, local_err);
636 if (prealloc != PREALLOC_MODE_OFF) {
637 error_setg(errp, "Unsupported preallocation mode '%s'",
638 PreallocMode_str(prealloc));
642 options = qdict_new();
643 qdict_put_str(options, "driver", drv->format_name);
645 blk = blk_new_open(filename, NULL, options,
646 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
648 error_prepend(errp, "Protocol driver '%s' does not support image "
649 "creation, and opening the image failed: ",
654 size = create_file_fallback_truncate(blk, size, errp);
660 ret = create_file_fallback_zero_first_sector(blk, size, errp);
671 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
675 drv = bdrv_find_protocol(filename, true, errp);
680 return bdrv_create(drv, filename, opts, errp);
683 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
685 Error *local_err = NULL;
691 error_setg(errp, "Block node '%s' is not opened", bs->filename);
695 if (!bs->drv->bdrv_co_delete_file) {
696 error_setg(errp, "Driver '%s' does not support image deletion",
697 bs->drv->format_name);
701 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
703 error_propagate(errp, local_err);
710 * Try to get @bs's logical and physical block size.
711 * On success, store them in @bsz struct and return 0.
712 * On failure return -errno.
713 * @bs must not be empty.
715 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
717 BlockDriver *drv = bs->drv;
718 BlockDriverState *filtered = bdrv_filter_bs(bs);
720 if (drv && drv->bdrv_probe_blocksizes) {
721 return drv->bdrv_probe_blocksizes(bs, bsz);
722 } else if (filtered) {
723 return bdrv_probe_blocksizes(filtered, bsz);
730 * Try to get @bs's geometry (cyls, heads, sectors).
731 * On success, store them in @geo struct and return 0.
732 * On failure return -errno.
733 * @bs must not be empty.
735 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
737 BlockDriver *drv = bs->drv;
738 BlockDriverState *filtered = bdrv_filter_bs(bs);
740 if (drv && drv->bdrv_probe_geometry) {
741 return drv->bdrv_probe_geometry(bs, geo);
742 } else if (filtered) {
743 return bdrv_probe_geometry(filtered, geo);
750 * Create a uniquely-named empty temporary file.
751 * Return 0 upon success, otherwise a negative errno value.
753 int get_tmp_filename(char *filename, int size)
756 char temp_dir[MAX_PATH];
757 /* GetTempFileName requires that its output buffer (4th param)
758 have length MAX_PATH or greater. */
759 assert(size >= MAX_PATH);
760 return (GetTempPath(MAX_PATH, temp_dir)
761 && GetTempFileName(temp_dir, "qem", 0, filename)
762 ? 0 : -GetLastError());
766 tmpdir = getenv("TMPDIR");
770 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
773 fd = mkstemp(filename);
777 if (close(fd) != 0) {
786 * Detect host devices. By convention, /dev/cdrom[N] is always
787 * recognized as a host CDROM.
789 static BlockDriver *find_hdev_driver(const char *filename)
791 int score_max = 0, score;
792 BlockDriver *drv = NULL, *d;
794 QLIST_FOREACH(d, &bdrv_drivers, list) {
795 if (d->bdrv_probe_device) {
796 score = d->bdrv_probe_device(filename);
797 if (score > score_max) {
807 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
811 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
812 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
820 BlockDriver *bdrv_find_protocol(const char *filename,
821 bool allow_protocol_prefix,
830 /* TODO Drivers without bdrv_file_open must be specified explicitly */
833 * XXX(hch): we really should not let host device detection
834 * override an explicit protocol specification, but moving this
835 * later breaks access to device names with colons in them.
836 * Thanks to the brain-dead persistent naming schemes on udev-
837 * based Linux systems those actually are quite common.
839 drv1 = find_hdev_driver(filename);
844 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
848 p = strchr(filename, ':');
851 if (len > sizeof(protocol) - 1)
852 len = sizeof(protocol) - 1;
853 memcpy(protocol, filename, len);
854 protocol[len] = '\0';
856 drv1 = bdrv_do_find_protocol(protocol);
861 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
862 if (block_driver_modules[i].protocol_name &&
863 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
864 block_module_load_one(block_driver_modules[i].library_name);
869 drv1 = bdrv_do_find_protocol(protocol);
871 error_setg(errp, "Unknown protocol '%s'", protocol);
877 * Guess image format by probing its contents.
878 * This is not a good idea when your image is raw (CVE-2008-2004), but
879 * we do it anyway for backward compatibility.
881 * @buf contains the image's first @buf_size bytes.
882 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
883 * but can be smaller if the image file is smaller)
884 * @filename is its filename.
886 * For all block drivers, call the bdrv_probe() method to get its
888 * Return the first block driver with the highest probing score.
890 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
891 const char *filename)
893 int score_max = 0, score;
894 BlockDriver *drv = NULL, *d;
896 QLIST_FOREACH(d, &bdrv_drivers, list) {
898 score = d->bdrv_probe(buf, buf_size, filename);
899 if (score > score_max) {
909 static int find_image_format(BlockBackend *file, const char *filename,
910 BlockDriver **pdrv, Error **errp)
913 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
916 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
917 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
922 ret = blk_pread(file, 0, buf, sizeof(buf));
924 error_setg_errno(errp, -ret, "Could not read image for determining its "
930 drv = bdrv_probe_all(buf, ret, filename);
932 error_setg(errp, "Could not determine image format: No compatible "
941 * Set the current 'total_sectors' value
942 * Return 0 on success, -errno on error.
944 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
946 BlockDriver *drv = bs->drv;
952 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
956 /* query actual device if possible, otherwise just trust the hint */
957 if (drv->bdrv_getlength) {
958 int64_t length = drv->bdrv_getlength(bs);
962 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
965 bs->total_sectors = hint;
967 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
975 * Combines a QDict of new block driver @options with any missing options taken
976 * from @old_options, so that leaving out an option defaults to its old value.
978 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
981 if (bs->drv && bs->drv->bdrv_join_options) {
982 bs->drv->bdrv_join_options(options, old_options);
984 qdict_join(options, old_options, false);
988 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
992 Error *local_err = NULL;
993 char *value = qemu_opt_get_del(opts, "detect-zeroes");
994 BlockdevDetectZeroesOptions detect_zeroes =
995 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
996 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
999 error_propagate(errp, local_err);
1000 return detect_zeroes;
1003 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1004 !(open_flags & BDRV_O_UNMAP))
1006 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1007 "without setting discard operation to unmap");
1010 return detect_zeroes;
1014 * Set open flags for aio engine
1016 * Return 0 on success, -1 if the engine specified is invalid
1018 int bdrv_parse_aio(const char *mode, int *flags)
1020 if (!strcmp(mode, "threads")) {
1021 /* do nothing, default */
1022 } else if (!strcmp(mode, "native")) {
1023 *flags |= BDRV_O_NATIVE_AIO;
1024 #ifdef CONFIG_LINUX_IO_URING
1025 } else if (!strcmp(mode, "io_uring")) {
1026 *flags |= BDRV_O_IO_URING;
1036 * Set open flags for a given discard mode
1038 * Return 0 on success, -1 if the discard mode was invalid.
1040 int bdrv_parse_discard_flags(const char *mode, int *flags)
1042 *flags &= ~BDRV_O_UNMAP;
1044 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1046 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1047 *flags |= BDRV_O_UNMAP;
1056 * Set open flags for a given cache mode
1058 * Return 0 on success, -1 if the cache mode was invalid.
1060 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1062 *flags &= ~BDRV_O_CACHE_MASK;
1064 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1065 *writethrough = false;
1066 *flags |= BDRV_O_NOCACHE;
1067 } else if (!strcmp(mode, "directsync")) {
1068 *writethrough = true;
1069 *flags |= BDRV_O_NOCACHE;
1070 } else if (!strcmp(mode, "writeback")) {
1071 *writethrough = false;
1072 } else if (!strcmp(mode, "unsafe")) {
1073 *writethrough = false;
1074 *flags |= BDRV_O_NO_FLUSH;
1075 } else if (!strcmp(mode, "writethrough")) {
1076 *writethrough = true;
1084 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1086 BlockDriverState *parent = c->opaque;
1087 return g_strdup(bdrv_get_device_or_node_name(parent));
1090 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1092 BlockDriverState *bs = child->opaque;
1093 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1096 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1098 BlockDriverState *bs = child->opaque;
1099 return bdrv_drain_poll(bs, false, NULL, false);
1102 static void bdrv_child_cb_drained_end(BdrvChild *child,
1103 int *drained_end_counter)
1105 BlockDriverState *bs = child->opaque;
1106 bdrv_drained_end_no_poll(bs, drained_end_counter);
1109 static int bdrv_child_cb_inactivate(BdrvChild *child)
1111 BlockDriverState *bs = child->opaque;
1112 assert(bs->open_flags & BDRV_O_INACTIVE);
1116 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1117 GSList **ignore, Error **errp)
1119 BlockDriverState *bs = child->opaque;
1120 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1123 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1126 BlockDriverState *bs = child->opaque;
1127 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1131 * Returns the options and flags that a temporary snapshot should get, based on
1132 * the originally requested flags (the originally requested image will have
1133 * flags like a backing file)
1135 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1136 int parent_flags, QDict *parent_options)
1138 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1140 /* For temporary files, unconditional cache=unsafe is fine */
1141 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1142 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1144 /* Copy the read-only and discard options from the parent */
1145 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1146 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1148 /* aio=native doesn't work for cache.direct=off, so disable it for the
1149 * temporary snapshot */
1150 *child_flags &= ~BDRV_O_NATIVE_AIO;
1153 static void bdrv_backing_attach(BdrvChild *c)
1155 BlockDriverState *parent = c->opaque;
1156 BlockDriverState *backing_hd = c->bs;
1158 assert(!parent->backing_blocker);
1159 error_setg(&parent->backing_blocker,
1160 "node is used as backing hd of '%s'",
1161 bdrv_get_device_or_node_name(parent));
1163 bdrv_refresh_filename(backing_hd);
1165 parent->open_flags &= ~BDRV_O_NO_BACKING;
1167 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1168 /* Otherwise we won't be able to commit or stream */
1169 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1170 parent->backing_blocker);
1171 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1172 parent->backing_blocker);
1174 * We do backup in 3 ways:
1176 * The target bs is new opened, and the source is top BDS
1177 * 2. blockdev backup
1178 * Both the source and the target are top BDSes.
1179 * 3. internal backup(used for block replication)
1180 * Both the source and the target are backing file
1182 * In case 1 and 2, neither the source nor the target is the backing file.
1183 * In case 3, we will block the top BDS, so there is only one block job
1184 * for the top BDS and its backing chain.
1186 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1187 parent->backing_blocker);
1188 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1189 parent->backing_blocker);
1192 static void bdrv_backing_detach(BdrvChild *c)
1194 BlockDriverState *parent = c->opaque;
1196 assert(parent->backing_blocker);
1197 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1198 error_free(parent->backing_blocker);
1199 parent->backing_blocker = NULL;
1202 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1203 const char *filename, Error **errp)
1205 BlockDriverState *parent = c->opaque;
1206 bool read_only = bdrv_is_read_only(parent);
1210 ret = bdrv_reopen_set_read_only(parent, false, errp);
1216 ret = bdrv_change_backing_file(parent, filename,
1217 base->drv ? base->drv->format_name : "",
1220 error_setg_errno(errp, -ret, "Could not update backing file link");
1224 bdrv_reopen_set_read_only(parent, true, NULL);
1231 * Returns the options and flags that a generic child of a BDS should
1232 * get, based on the given options and flags for the parent BDS.
1234 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1235 int *child_flags, QDict *child_options,
1236 int parent_flags, QDict *parent_options)
1238 int flags = parent_flags;
1241 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1242 * Generally, the question to answer is: Should this child be
1243 * format-probed by default?
1247 * Pure and non-filtered data children of non-format nodes should
1248 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1249 * set). This only affects a very limited set of drivers (namely
1250 * quorum and blkverify when this comment was written).
1251 * Force-clear BDRV_O_PROTOCOL then.
1253 if (!parent_is_format &&
1254 (role & BDRV_CHILD_DATA) &&
1255 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1257 flags &= ~BDRV_O_PROTOCOL;
1261 * All children of format nodes (except for COW children) and all
1262 * metadata children in general should never be format-probed.
1263 * Force-set BDRV_O_PROTOCOL then.
1265 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1266 (role & BDRV_CHILD_METADATA))
1268 flags |= BDRV_O_PROTOCOL;
1272 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1275 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1276 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1277 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1279 if (role & BDRV_CHILD_COW) {
1280 /* backing files are opened read-only by default */
1281 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1282 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1284 /* Inherit the read-only option from the parent if it's not set */
1285 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1286 qdict_copy_default(child_options, parent_options,
1287 BDRV_OPT_AUTO_READ_ONLY);
1291 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1292 * can default to enable it on lower layers regardless of the
1295 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1297 /* Clear flags that only apply to the top layer */
1298 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1300 if (role & BDRV_CHILD_METADATA) {
1301 flags &= ~BDRV_O_NO_IO;
1303 if (role & BDRV_CHILD_COW) {
1304 flags &= ~BDRV_O_TEMPORARY;
1307 *child_flags = flags;
1310 static void bdrv_child_cb_attach(BdrvChild *child)
1312 BlockDriverState *bs = child->opaque;
1314 if (child->role & BDRV_CHILD_COW) {
1315 bdrv_backing_attach(child);
1318 bdrv_apply_subtree_drain(child, bs);
1321 static void bdrv_child_cb_detach(BdrvChild *child)
1323 BlockDriverState *bs = child->opaque;
1325 if (child->role & BDRV_CHILD_COW) {
1326 bdrv_backing_detach(child);
1329 bdrv_unapply_subtree_drain(child, bs);
1332 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1333 const char *filename, Error **errp)
1335 if (c->role & BDRV_CHILD_COW) {
1336 return bdrv_backing_update_filename(c, base, filename, errp);
1341 const BdrvChildClass child_of_bds = {
1342 .parent_is_bds = true,
1343 .get_parent_desc = bdrv_child_get_parent_desc,
1344 .inherit_options = bdrv_inherited_options,
1345 .drained_begin = bdrv_child_cb_drained_begin,
1346 .drained_poll = bdrv_child_cb_drained_poll,
1347 .drained_end = bdrv_child_cb_drained_end,
1348 .attach = bdrv_child_cb_attach,
1349 .detach = bdrv_child_cb_detach,
1350 .inactivate = bdrv_child_cb_inactivate,
1351 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1352 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1353 .update_filename = bdrv_child_cb_update_filename,
1356 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1358 int open_flags = flags;
1361 * Clear flags that are internal to the block layer before opening the
1364 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1369 static void update_flags_from_options(int *flags, QemuOpts *opts)
1371 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1373 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1374 *flags |= BDRV_O_NO_FLUSH;
1377 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1378 *flags |= BDRV_O_NOCACHE;
1381 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1382 *flags |= BDRV_O_RDWR;
1385 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1386 *flags |= BDRV_O_AUTO_RDONLY;
1390 static void update_options_from_flags(QDict *options, int flags)
1392 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1393 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1395 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1396 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1397 flags & BDRV_O_NO_FLUSH);
1399 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1400 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1402 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1403 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1404 flags & BDRV_O_AUTO_RDONLY);
1408 static void bdrv_assign_node_name(BlockDriverState *bs,
1409 const char *node_name,
1412 char *gen_node_name = NULL;
1415 node_name = gen_node_name = id_generate(ID_BLOCK);
1416 } else if (!id_wellformed(node_name)) {
1418 * Check for empty string or invalid characters, but not if it is
1419 * generated (generated names use characters not available to the user)
1421 error_setg(errp, "Invalid node name");
1425 /* takes care of avoiding namespaces collisions */
1426 if (blk_by_name(node_name)) {
1427 error_setg(errp, "node-name=%s is conflicting with a device id",
1432 /* takes care of avoiding duplicates node names */
1433 if (bdrv_find_node(node_name)) {
1434 error_setg(errp, "Duplicate node name");
1438 /* Make sure that the node name isn't truncated */
1439 if (strlen(node_name) >= sizeof(bs->node_name)) {
1440 error_setg(errp, "Node name too long");
1444 /* copy node name into the bs and insert it into the graph list */
1445 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1446 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1448 g_free(gen_node_name);
1451 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1452 const char *node_name, QDict *options,
1453 int open_flags, Error **errp)
1455 Error *local_err = NULL;
1458 bdrv_assign_node_name(bs, node_name, &local_err);
1460 error_propagate(errp, local_err);
1465 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1466 bs->opaque = g_malloc0(drv->instance_size);
1468 if (drv->bdrv_file_open) {
1469 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1470 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1471 } else if (drv->bdrv_open) {
1472 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1479 error_propagate(errp, local_err);
1480 } else if (bs->filename[0]) {
1481 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1483 error_setg_errno(errp, -ret, "Could not open image");
1488 ret = refresh_total_sectors(bs, bs->total_sectors);
1490 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1494 bdrv_refresh_limits(bs, &local_err);
1496 error_propagate(errp, local_err);
1500 assert(bdrv_opt_mem_align(bs) != 0);
1501 assert(bdrv_min_mem_align(bs) != 0);
1502 assert(is_power_of_2(bs->bl.request_alignment));
1504 for (i = 0; i < bs->quiesce_counter; i++) {
1505 if (drv->bdrv_co_drain_begin) {
1506 drv->bdrv_co_drain_begin(bs);
1513 if (bs->file != NULL) {
1514 bdrv_unref_child(bs, bs->file);
1522 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1523 int flags, Error **errp)
1525 BlockDriverState *bs;
1529 bs->open_flags = flags;
1530 bs->explicit_options = qdict_new();
1531 bs->options = qdict_new();
1534 update_options_from_flags(bs->options, flags);
1536 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1538 qobject_unref(bs->explicit_options);
1539 bs->explicit_options = NULL;
1540 qobject_unref(bs->options);
1549 QemuOptsList bdrv_runtime_opts = {
1550 .name = "bdrv_common",
1551 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1554 .name = "node-name",
1555 .type = QEMU_OPT_STRING,
1556 .help = "Node name of the block device node",
1560 .type = QEMU_OPT_STRING,
1561 .help = "Block driver to use for the node",
1564 .name = BDRV_OPT_CACHE_DIRECT,
1565 .type = QEMU_OPT_BOOL,
1566 .help = "Bypass software writeback cache on the host",
1569 .name = BDRV_OPT_CACHE_NO_FLUSH,
1570 .type = QEMU_OPT_BOOL,
1571 .help = "Ignore flush requests",
1574 .name = BDRV_OPT_READ_ONLY,
1575 .type = QEMU_OPT_BOOL,
1576 .help = "Node is opened in read-only mode",
1579 .name = BDRV_OPT_AUTO_READ_ONLY,
1580 .type = QEMU_OPT_BOOL,
1581 .help = "Node can become read-only if opening read-write fails",
1584 .name = "detect-zeroes",
1585 .type = QEMU_OPT_STRING,
1586 .help = "try to optimize zero writes (off, on, unmap)",
1589 .name = BDRV_OPT_DISCARD,
1590 .type = QEMU_OPT_STRING,
1591 .help = "discard operation (ignore/off, unmap/on)",
1594 .name = BDRV_OPT_FORCE_SHARE,
1595 .type = QEMU_OPT_BOOL,
1596 .help = "always accept other writers (default: off)",
1598 { /* end of list */ }
1602 QemuOptsList bdrv_create_opts_simple = {
1603 .name = "simple-create-opts",
1604 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1607 .name = BLOCK_OPT_SIZE,
1608 .type = QEMU_OPT_SIZE,
1609 .help = "Virtual disk size"
1612 .name = BLOCK_OPT_PREALLOC,
1613 .type = QEMU_OPT_STRING,
1614 .help = "Preallocation mode (allowed values: off)"
1616 { /* end of list */ }
1621 * Common part for opening disk images and files
1623 * Removes all processed options from *options.
1625 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1626 QDict *options, Error **errp)
1628 int ret, open_flags;
1629 const char *filename;
1630 const char *driver_name = NULL;
1631 const char *node_name = NULL;
1632 const char *discard;
1635 Error *local_err = NULL;
1637 assert(bs->file == NULL);
1638 assert(options != NULL && bs->options != options);
1640 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1641 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1646 update_flags_from_options(&bs->open_flags, opts);
1648 driver_name = qemu_opt_get(opts, "driver");
1649 drv = bdrv_find_format(driver_name);
1650 assert(drv != NULL);
1652 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1654 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1656 BDRV_OPT_FORCE_SHARE
1657 "=on can only be used with read-only images");
1663 bdrv_refresh_filename(blk_bs(file));
1664 filename = blk_bs(file)->filename;
1667 * Caution: while qdict_get_try_str() is fine, getting
1668 * non-string types would require more care. When @options
1669 * come from -blockdev or blockdev_add, its members are typed
1670 * according to the QAPI schema, but when they come from
1671 * -drive, they're all QString.
1673 filename = qdict_get_try_str(options, "filename");
1676 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1677 error_setg(errp, "The '%s' block driver requires a file name",
1683 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1686 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1688 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1689 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1690 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1696 !bs->read_only && bdrv_is_whitelisted(drv, true)
1697 ? "Driver '%s' can only be used for read-only devices"
1698 : "Driver '%s' is not whitelisted",
1704 /* bdrv_new() and bdrv_close() make it so */
1705 assert(qatomic_read(&bs->copy_on_read) == 0);
1707 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1708 if (!bs->read_only) {
1709 bdrv_enable_copy_on_read(bs);
1711 error_setg(errp, "Can't use copy-on-read on read-only device");
1717 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1718 if (discard != NULL) {
1719 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1720 error_setg(errp, "Invalid discard option");
1727 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1729 error_propagate(errp, local_err);
1734 if (filename != NULL) {
1735 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1737 bs->filename[0] = '\0';
1739 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1741 /* Open the image, either directly or using a protocol */
1742 open_flags = bdrv_open_flags(bs, bs->open_flags);
1743 node_name = qemu_opt_get(opts, "node-name");
1745 assert(!drv->bdrv_file_open || file == NULL);
1746 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1751 qemu_opts_del(opts);
1755 qemu_opts_del(opts);
1759 static QDict *parse_json_filename(const char *filename, Error **errp)
1761 QObject *options_obj;
1765 ret = strstart(filename, "json:", &filename);
1768 options_obj = qobject_from_json(filename, errp);
1770 error_prepend(errp, "Could not parse the JSON options: ");
1774 options = qobject_to(QDict, options_obj);
1776 qobject_unref(options_obj);
1777 error_setg(errp, "Invalid JSON object given");
1781 qdict_flatten(options);
1786 static void parse_json_protocol(QDict *options, const char **pfilename,
1789 QDict *json_options;
1790 Error *local_err = NULL;
1792 /* Parse json: pseudo-protocol */
1793 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1797 json_options = parse_json_filename(*pfilename, &local_err);
1799 error_propagate(errp, local_err);
1803 /* Options given in the filename have lower priority than options
1804 * specified directly */
1805 qdict_join(options, json_options, false);
1806 qobject_unref(json_options);
1811 * Fills in default options for opening images and converts the legacy
1812 * filename/flags pair to option QDict entries.
1813 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1814 * block driver has been specified explicitly.
1816 static int bdrv_fill_options(QDict **options, const char *filename,
1817 int *flags, Error **errp)
1819 const char *drvname;
1820 bool protocol = *flags & BDRV_O_PROTOCOL;
1821 bool parse_filename = false;
1822 BlockDriver *drv = NULL;
1823 Error *local_err = NULL;
1826 * Caution: while qdict_get_try_str() is fine, getting non-string
1827 * types would require more care. When @options come from
1828 * -blockdev or blockdev_add, its members are typed according to
1829 * the QAPI schema, but when they come from -drive, they're all
1832 drvname = qdict_get_try_str(*options, "driver");
1834 drv = bdrv_find_format(drvname);
1836 error_setg(errp, "Unknown driver '%s'", drvname);
1839 /* If the user has explicitly specified the driver, this choice should
1840 * override the BDRV_O_PROTOCOL flag */
1841 protocol = drv->bdrv_file_open;
1845 *flags |= BDRV_O_PROTOCOL;
1847 *flags &= ~BDRV_O_PROTOCOL;
1850 /* Translate cache options from flags into options */
1851 update_options_from_flags(*options, *flags);
1853 /* Fetch the file name from the options QDict if necessary */
1854 if (protocol && filename) {
1855 if (!qdict_haskey(*options, "filename")) {
1856 qdict_put_str(*options, "filename", filename);
1857 parse_filename = true;
1859 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1865 /* Find the right block driver */
1866 /* See cautionary note on accessing @options above */
1867 filename = qdict_get_try_str(*options, "filename");
1869 if (!drvname && protocol) {
1871 drv = bdrv_find_protocol(filename, parse_filename, errp);
1876 drvname = drv->format_name;
1877 qdict_put_str(*options, "driver", drvname);
1879 error_setg(errp, "Must specify either driver or file");
1884 assert(drv || !protocol);
1886 /* Driver-specific filename parsing */
1887 if (drv && drv->bdrv_parse_filename && parse_filename) {
1888 drv->bdrv_parse_filename(filename, *options, &local_err);
1890 error_propagate(errp, local_err);
1894 if (!drv->bdrv_needs_filename) {
1895 qdict_del(*options, "filename");
1902 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1903 uint64_t perm, uint64_t shared,
1904 GSList *ignore_children, Error **errp);
1905 static void bdrv_child_abort_perm_update(BdrvChild *c);
1906 static void bdrv_child_set_perm(BdrvChild *c);
1908 typedef struct BlockReopenQueueEntry {
1911 BDRVReopenState state;
1912 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1913 } BlockReopenQueueEntry;
1916 * Return the flags that @bs will have after the reopens in @q have
1917 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1918 * return the current flags.
1920 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1922 BlockReopenQueueEntry *entry;
1925 QTAILQ_FOREACH(entry, q, entry) {
1926 if (entry->state.bs == bs) {
1927 return entry->state.flags;
1932 return bs->open_flags;
1935 /* Returns whether the image file can be written to after the reopen queue @q
1936 * has been successfully applied, or right now if @q is NULL. */
1937 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1938 BlockReopenQueue *q)
1940 int flags = bdrv_reopen_get_flags(q, bs);
1942 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1946 * Return whether the BDS can be written to. This is not necessarily
1947 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1948 * be written to but do not count as read-only images.
1950 bool bdrv_is_writable(BlockDriverState *bs)
1952 return bdrv_is_writable_after_reopen(bs, NULL);
1955 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1956 BdrvChild *c, BdrvChildRole role,
1957 BlockReopenQueue *reopen_queue,
1958 uint64_t parent_perm, uint64_t parent_shared,
1959 uint64_t *nperm, uint64_t *nshared)
1961 assert(bs->drv && bs->drv->bdrv_child_perm);
1962 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1963 parent_perm, parent_shared,
1965 /* TODO Take force_share from reopen_queue */
1966 if (child_bs && child_bs->force_share) {
1967 *nshared = BLK_PERM_ALL;
1972 * Check whether permissions on this node can be changed in a way that
1973 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1974 * permissions of all its parents. This involves checking whether all necessary
1975 * permission changes to child nodes can be performed.
1977 * A call to this function must always be followed by a call to bdrv_set_perm()
1978 * or bdrv_abort_perm_update().
1980 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1981 uint64_t cumulative_perms,
1982 uint64_t cumulative_shared_perms,
1983 GSList *ignore_children, Error **errp)
1985 BlockDriver *drv = bs->drv;
1989 /* Write permissions never work with read-only images */
1990 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1991 !bdrv_is_writable_after_reopen(bs, q))
1993 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
1994 error_setg(errp, "Block node is read-only");
1996 uint64_t current_perms, current_shared;
1997 bdrv_get_cumulative_perm(bs, ¤t_perms, ¤t_shared);
1998 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
1999 error_setg(errp, "Cannot make block node read-only, there is "
2002 error_setg(errp, "Cannot make block node read-only and create "
2011 * Unaligned requests will automatically be aligned to bl.request_alignment
2012 * and without RESIZE we can't extend requests to write to space beyond the
2013 * end of the image, so it's required that the image size is aligned.
2015 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2016 !(cumulative_perms & BLK_PERM_RESIZE))
2018 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2019 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2020 "Image size is not a multiple of request "
2026 /* Check this node */
2031 if (drv->bdrv_check_perm) {
2032 ret = drv->bdrv_check_perm(bs, cumulative_perms,
2033 cumulative_shared_perms, errp);
2039 /* Drivers that never have children can omit .bdrv_child_perm() */
2040 if (!drv->bdrv_child_perm) {
2041 assert(QLIST_EMPTY(&bs->children));
2045 /* Check all children */
2046 QLIST_FOREACH(c, &bs->children, next) {
2047 uint64_t cur_perm, cur_shared;
2049 bdrv_child_perm(bs, c->bs, c, c->role, q,
2050 cumulative_perms, cumulative_shared_perms,
2051 &cur_perm, &cur_shared);
2052 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2063 * Notifies drivers that after a previous bdrv_check_perm() call, the
2064 * permission update is not performed and any preparations made for it (e.g.
2065 * taken file locks) need to be undone.
2067 * This function recursively notifies all child nodes.
2069 static void bdrv_abort_perm_update(BlockDriverState *bs)
2071 BlockDriver *drv = bs->drv;
2078 if (drv->bdrv_abort_perm_update) {
2079 drv->bdrv_abort_perm_update(bs);
2082 QLIST_FOREACH(c, &bs->children, next) {
2083 bdrv_child_abort_perm_update(c);
2087 static void bdrv_set_perm(BlockDriverState *bs)
2089 uint64_t cumulative_perms, cumulative_shared_perms;
2090 BlockDriver *drv = bs->drv;
2097 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2099 /* Update this node */
2100 if (drv->bdrv_set_perm) {
2101 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2104 /* Drivers that never have children can omit .bdrv_child_perm() */
2105 if (!drv->bdrv_child_perm) {
2106 assert(QLIST_EMPTY(&bs->children));
2110 /* Update all children */
2111 QLIST_FOREACH(c, &bs->children, next) {
2112 bdrv_child_set_perm(c);
2116 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2117 uint64_t *shared_perm)
2120 uint64_t cumulative_perms = 0;
2121 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2123 QLIST_FOREACH(c, &bs->parents, next_parent) {
2124 cumulative_perms |= c->perm;
2125 cumulative_shared_perms &= c->shared_perm;
2128 *perm = cumulative_perms;
2129 *shared_perm = cumulative_shared_perms;
2132 static char *bdrv_child_user_desc(BdrvChild *c)
2134 if (c->klass->get_parent_desc) {
2135 return c->klass->get_parent_desc(c);
2138 return g_strdup("another user");
2141 char *bdrv_perm_names(uint64_t perm)
2147 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2148 { BLK_PERM_WRITE, "write" },
2149 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2150 { BLK_PERM_RESIZE, "resize" },
2151 { BLK_PERM_GRAPH_MOD, "change children" },
2155 GString *result = g_string_sized_new(30);
2156 struct perm_name *p;
2158 for (p = permissions; p->name; p++) {
2159 if (perm & p->perm) {
2160 if (result->len > 0) {
2161 g_string_append(result, ", ");
2163 g_string_append(result, p->name);
2167 return g_string_free(result, FALSE);
2171 * Checks whether a new reference to @bs can be added if the new user requires
2172 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2173 * set, the BdrvChild objects in this list are ignored in the calculations;
2174 * this allows checking permission updates for an existing reference.
2176 * Needs to be followed by a call to either bdrv_set_perm() or
2177 * bdrv_abort_perm_update(). */
2178 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2179 uint64_t new_used_perm,
2180 uint64_t new_shared_perm,
2181 GSList *ignore_children,
2185 uint64_t cumulative_perms = new_used_perm;
2186 uint64_t cumulative_shared_perms = new_shared_perm;
2189 /* There is no reason why anyone couldn't tolerate write_unchanged */
2190 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2192 QLIST_FOREACH(c, &bs->parents, next_parent) {
2193 if (g_slist_find(ignore_children, c)) {
2197 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2198 char *user = bdrv_child_user_desc(c);
2199 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2201 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2203 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2209 if ((c->perm & new_shared_perm) != c->perm) {
2210 char *user = bdrv_child_user_desc(c);
2211 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2213 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2215 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2221 cumulative_perms |= c->perm;
2222 cumulative_shared_perms &= c->shared_perm;
2225 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2226 ignore_children, errp);
2229 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2230 * bdrv_child_abort_perm_update(). */
2231 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2232 uint64_t perm, uint64_t shared,
2233 GSList *ignore_children, Error **errp)
2237 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2238 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
2239 g_slist_free(ignore_children);
2245 if (!c->has_backup_perm) {
2246 c->has_backup_perm = true;
2247 c->backup_perm = c->perm;
2248 c->backup_shared_perm = c->shared_perm;
2251 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2252 * same child twice during check_perm procedure
2256 c->shared_perm = shared;
2261 static void bdrv_child_set_perm(BdrvChild *c)
2263 c->has_backup_perm = false;
2265 bdrv_set_perm(c->bs);
2268 static void bdrv_child_abort_perm_update(BdrvChild *c)
2270 if (c->has_backup_perm) {
2271 c->perm = c->backup_perm;
2272 c->shared_perm = c->backup_shared_perm;
2273 c->has_backup_perm = false;
2276 bdrv_abort_perm_update(c->bs);
2279 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2282 uint64_t perm, shared_perm;
2284 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
2285 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, errp);
2287 bdrv_abort_perm_update(bs);
2295 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2298 Error *local_err = NULL;
2301 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, &local_err);
2303 bdrv_child_abort_perm_update(c);
2304 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2305 /* tighten permissions */
2306 error_propagate(errp, local_err);
2309 * Our caller may intend to only loosen restrictions and
2310 * does not expect this function to fail. Errors are not
2311 * fatal in such a case, so we can just hide them from our
2314 error_free(local_err);
2320 bdrv_child_set_perm(c);
2325 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2327 uint64_t parent_perms, parent_shared;
2328 uint64_t perms, shared;
2330 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2331 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2332 parent_perms, parent_shared, &perms, &shared);
2334 return bdrv_child_try_set_perm(c, perms, shared, errp);
2338 * Default implementation for .bdrv_child_perm() for block filters:
2339 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2342 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2344 BlockReopenQueue *reopen_queue,
2345 uint64_t perm, uint64_t shared,
2346 uint64_t *nperm, uint64_t *nshared)
2348 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2349 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2352 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2354 BlockReopenQueue *reopen_queue,
2355 uint64_t perm, uint64_t shared,
2356 uint64_t *nperm, uint64_t *nshared)
2358 assert(role & BDRV_CHILD_COW);
2361 * We want consistent read from backing files if the parent needs it.
2362 * No other operations are performed on backing files.
2364 perm &= BLK_PERM_CONSISTENT_READ;
2367 * If the parent can deal with changing data, we're okay with a
2368 * writable and resizable backing file.
2369 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2371 if (shared & BLK_PERM_WRITE) {
2372 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2377 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2378 BLK_PERM_WRITE_UNCHANGED;
2380 if (bs->open_flags & BDRV_O_INACTIVE) {
2381 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2388 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2390 BlockReopenQueue *reopen_queue,
2391 uint64_t perm, uint64_t shared,
2392 uint64_t *nperm, uint64_t *nshared)
2396 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2398 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2401 * Apart from the modifications below, the same permissions are
2402 * forwarded and left alone as for filters
2404 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2405 perm, shared, &perm, &shared);
2407 if (role & BDRV_CHILD_METADATA) {
2408 /* Format drivers may touch metadata even if the guest doesn't write */
2409 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2410 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2414 * bs->file always needs to be consistent because of the
2415 * metadata. We can never allow other users to resize or write
2418 if (!(flags & BDRV_O_NO_IO)) {
2419 perm |= BLK_PERM_CONSISTENT_READ;
2421 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2424 if (role & BDRV_CHILD_DATA) {
2426 * Technically, everything in this block is a subset of the
2427 * BDRV_CHILD_METADATA path taken above, and so this could
2428 * be an "else if" branch. However, that is not obvious, and
2429 * this function is not performance critical, therefore we let
2430 * this be an independent "if".
2434 * We cannot allow other users to resize the file because the
2435 * format driver might have some assumptions about the size
2436 * (e.g. because it is stored in metadata, or because the file
2437 * is split into fixed-size data files).
2439 shared &= ~BLK_PERM_RESIZE;
2442 * WRITE_UNCHANGED often cannot be performed as such on the
2443 * data file. For example, the qcow2 driver may still need to
2444 * write copied clusters on copy-on-read.
2446 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2447 perm |= BLK_PERM_WRITE;
2451 * If the data file is written to, the format driver may
2452 * expect to be able to resize it by writing beyond the EOF.
2454 if (perm & BLK_PERM_WRITE) {
2455 perm |= BLK_PERM_RESIZE;
2459 if (bs->open_flags & BDRV_O_INACTIVE) {
2460 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2467 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2468 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2469 uint64_t perm, uint64_t shared,
2470 uint64_t *nperm, uint64_t *nshared)
2472 if (role & BDRV_CHILD_FILTERED) {
2473 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2475 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2476 perm, shared, nperm, nshared);
2477 } else if (role & BDRV_CHILD_COW) {
2478 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2479 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2480 perm, shared, nperm, nshared);
2481 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2482 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2483 perm, shared, nperm, nshared);
2485 g_assert_not_reached();
2489 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2491 static const uint64_t permissions[] = {
2492 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2493 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2494 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2495 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2496 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2499 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2500 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2502 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2504 return permissions[qapi_perm];
2507 static void bdrv_replace_child_noperm(BdrvChild *child,
2508 BlockDriverState *new_bs)
2510 BlockDriverState *old_bs = child->bs;
2511 int new_bs_quiesce_counter;
2514 assert(!child->frozen);
2516 if (old_bs && new_bs) {
2517 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2520 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2521 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2524 * If the new child node is drained but the old one was not, flush
2525 * all outstanding requests to the old child node.
2527 while (drain_saldo > 0 && child->klass->drained_begin) {
2528 bdrv_parent_drained_begin_single(child, true);
2533 /* Detach first so that the recursive drain sections coming from @child
2534 * are already gone and we only end the drain sections that came from
2536 if (child->klass->detach) {
2537 child->klass->detach(child);
2539 QLIST_REMOVE(child, next_parent);
2545 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2548 * Detaching the old node may have led to the new node's
2549 * quiesce_counter having been decreased. Not a problem, we
2550 * just need to recognize this here and then invoke
2551 * drained_end appropriately more often.
2553 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2554 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2556 /* Attach only after starting new drained sections, so that recursive
2557 * drain sections coming from @child don't get an extra .drained_begin
2559 if (child->klass->attach) {
2560 child->klass->attach(child);
2565 * If the old child node was drained but the new one is not, allow
2566 * requests to come in only after the new node has been attached.
2568 while (drain_saldo < 0 && child->klass->drained_end) {
2569 bdrv_parent_drained_end_single(child);
2575 * Updates @child to change its reference to point to @new_bs, including
2576 * checking and applying the necessary permission updates both to the old node
2579 * NULL is passed as @new_bs for removing the reference before freeing @child.
2581 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2582 * function uses bdrv_set_perm() to update the permissions according to the new
2583 * reference that @new_bs gets.
2585 * Callers must ensure that child->frozen is false.
2587 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2589 BlockDriverState *old_bs = child->bs;
2591 /* Asserts that child->frozen == false */
2592 bdrv_replace_child_noperm(child, new_bs);
2595 * Start with the new node's permissions. If @new_bs is a (direct
2596 * or indirect) child of @old_bs, we must complete the permission
2597 * update on @new_bs before we loosen the restrictions on @old_bs.
2598 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2599 * updating the permissions of @new_bs, and thus not purely loosen
2603 bdrv_set_perm(new_bs);
2608 * Update permissions for old node. We're just taking a parent away, so
2609 * we're loosening restrictions. Errors of permission update are not
2610 * fatal in this case, ignore them.
2612 bdrv_refresh_perms(old_bs, NULL);
2614 /* When the parent requiring a non-default AioContext is removed, the
2615 * node moves back to the main AioContext */
2616 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2621 * This function steals the reference to child_bs from the caller.
2622 * That reference is later dropped by bdrv_root_unref_child().
2624 * On failure NULL is returned, errp is set and the reference to
2625 * child_bs is also dropped.
2627 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2628 * (unless @child_bs is already in @ctx).
2630 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2631 const char *child_name,
2632 const BdrvChildClass *child_class,
2633 BdrvChildRole child_role,
2635 uint64_t perm, uint64_t shared_perm,
2636 void *opaque, Error **errp)
2639 Error *local_err = NULL;
2642 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2644 bdrv_abort_perm_update(child_bs);
2645 bdrv_unref(child_bs);
2649 child = g_new(BdrvChild, 1);
2650 *child = (BdrvChild) {
2652 .name = g_strdup(child_name),
2653 .klass = child_class,
2656 .shared_perm = shared_perm,
2660 /* If the AioContexts don't match, first try to move the subtree of
2661 * child_bs into the AioContext of the new parent. If this doesn't work,
2662 * try moving the parent into the AioContext of child_bs instead. */
2663 if (bdrv_get_aio_context(child_bs) != ctx) {
2664 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2665 if (ret < 0 && child_class->can_set_aio_ctx) {
2666 GSList *ignore = g_slist_prepend(NULL, child);
2667 ctx = bdrv_get_aio_context(child_bs);
2668 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2669 error_free(local_err);
2671 g_slist_free(ignore);
2672 ignore = g_slist_prepend(NULL, child);
2673 child_class->set_aio_ctx(child, ctx, &ignore);
2675 g_slist_free(ignore);
2678 error_propagate(errp, local_err);
2680 bdrv_abort_perm_update(child_bs);
2681 bdrv_unref(child_bs);
2686 /* This performs the matching bdrv_set_perm() for the above check. */
2687 bdrv_replace_child(child, child_bs);
2693 * This function transfers the reference to child_bs from the caller
2694 * to parent_bs. That reference is later dropped by parent_bs on
2695 * bdrv_close() or if someone calls bdrv_unref_child().
2697 * On failure NULL is returned, errp is set and the reference to
2698 * child_bs is also dropped.
2700 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2701 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2703 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2704 BlockDriverState *child_bs,
2705 const char *child_name,
2706 const BdrvChildClass *child_class,
2707 BdrvChildRole child_role,
2711 uint64_t perm, shared_perm;
2713 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2715 assert(parent_bs->drv);
2716 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2717 perm, shared_perm, &perm, &shared_perm);
2719 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2720 child_role, bdrv_get_aio_context(parent_bs),
2721 perm, shared_perm, parent_bs, errp);
2722 if (child == NULL) {
2726 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2730 static void bdrv_detach_child(BdrvChild *child)
2732 QLIST_SAFE_REMOVE(child, next);
2734 bdrv_replace_child(child, NULL);
2736 g_free(child->name);
2740 /* Callers must ensure that child->frozen is false. */
2741 void bdrv_root_unref_child(BdrvChild *child)
2743 BlockDriverState *child_bs;
2745 child_bs = child->bs;
2746 bdrv_detach_child(child);
2747 bdrv_unref(child_bs);
2751 * Clear all inherits_from pointers from children and grandchildren of
2752 * @root that point to @root, where necessary.
2754 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2758 if (child->bs->inherits_from == root) {
2760 * Remove inherits_from only when the last reference between root and
2761 * child->bs goes away.
2763 QLIST_FOREACH(c, &root->children, next) {
2764 if (c != child && c->bs == child->bs) {
2769 child->bs->inherits_from = NULL;
2773 QLIST_FOREACH(c, &child->bs->children, next) {
2774 bdrv_unset_inherits_from(root, c);
2778 /* Callers must ensure that child->frozen is false. */
2779 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2781 if (child == NULL) {
2785 bdrv_unset_inherits_from(parent, child);
2786 bdrv_root_unref_child(child);
2790 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2793 QLIST_FOREACH(c, &bs->parents, next_parent) {
2794 if (c->klass->change_media) {
2795 c->klass->change_media(c, load);
2800 /* Return true if you can reach parent going through child->inherits_from
2801 * recursively. If parent or child are NULL, return false */
2802 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2803 BlockDriverState *parent)
2805 while (child && child != parent) {
2806 child = child->inherits_from;
2809 return child != NULL;
2813 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2814 * mostly used for COW backing children (role = COW), but also for
2815 * filtered children (role = FILTERED | PRIMARY).
2817 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2819 if (bs->drv && bs->drv->is_filter) {
2820 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2822 return BDRV_CHILD_COW;
2827 * Sets the bs->backing link of a BDS. A new reference is created; callers
2828 * which don't need their own reference any more must call bdrv_unref().
2830 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2833 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2834 bdrv_inherits_from_recursive(backing_hd, bs);
2836 if (bdrv_is_backing_chain_frozen(bs, child_bs(bs->backing), errp)) {
2841 bdrv_ref(backing_hd);
2845 /* Cannot be frozen, we checked that above */
2846 bdrv_unref_child(bs, bs->backing);
2854 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2855 bdrv_backing_role(bs), errp);
2856 /* If backing_hd was already part of bs's backing chain, and
2857 * inherits_from pointed recursively to bs then let's update it to
2858 * point directly to bs (else it will become NULL). */
2859 if (bs->backing && update_inherits_from) {
2860 backing_hd->inherits_from = bs;
2864 bdrv_refresh_limits(bs, NULL);
2868 * Opens the backing file for a BlockDriverState if not yet open
2870 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2871 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2872 * itself, all options starting with "${bdref_key}." are considered part of the
2875 * TODO Can this be unified with bdrv_open_image()?
2877 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2878 const char *bdref_key, Error **errp)
2880 char *backing_filename = NULL;
2881 char *bdref_key_dot;
2882 const char *reference = NULL;
2884 bool implicit_backing = false;
2885 BlockDriverState *backing_hd;
2887 QDict *tmp_parent_options = NULL;
2888 Error *local_err = NULL;
2890 if (bs->backing != NULL) {
2894 /* NULL means an empty set of options */
2895 if (parent_options == NULL) {
2896 tmp_parent_options = qdict_new();
2897 parent_options = tmp_parent_options;
2900 bs->open_flags &= ~BDRV_O_NO_BACKING;
2902 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2903 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2904 g_free(bdref_key_dot);
2907 * Caution: while qdict_get_try_str() is fine, getting non-string
2908 * types would require more care. When @parent_options come from
2909 * -blockdev or blockdev_add, its members are typed according to
2910 * the QAPI schema, but when they come from -drive, they're all
2913 reference = qdict_get_try_str(parent_options, bdref_key);
2914 if (reference || qdict_haskey(options, "file.filename")) {
2915 /* keep backing_filename NULL */
2916 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2917 qobject_unref(options);
2920 if (qdict_size(options) == 0) {
2921 /* If the user specifies options that do not modify the
2922 * backing file's behavior, we might still consider it the
2923 * implicit backing file. But it's easier this way, and
2924 * just specifying some of the backing BDS's options is
2925 * only possible with -drive anyway (otherwise the QAPI
2926 * schema forces the user to specify everything). */
2927 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2930 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2933 error_propagate(errp, local_err);
2934 qobject_unref(options);
2939 if (!bs->drv || !bs->drv->supports_backing) {
2941 error_setg(errp, "Driver doesn't support backing files");
2942 qobject_unref(options);
2947 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2948 qdict_put_str(options, "driver", bs->backing_format);
2951 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2952 &child_of_bds, bdrv_backing_role(bs), errp);
2954 bs->open_flags |= BDRV_O_NO_BACKING;
2955 error_prepend(errp, "Could not open backing file: ");
2960 if (implicit_backing) {
2961 bdrv_refresh_filename(backing_hd);
2962 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2963 backing_hd->filename);
2966 /* Hook up the backing file link; drop our reference, bs owns the
2967 * backing_hd reference now */
2968 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2969 bdrv_unref(backing_hd);
2971 error_propagate(errp, local_err);
2976 qdict_del(parent_options, bdref_key);
2979 g_free(backing_filename);
2980 qobject_unref(tmp_parent_options);
2984 static BlockDriverState *
2985 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2986 BlockDriverState *parent, const BdrvChildClass *child_class,
2987 BdrvChildRole child_role, bool allow_none, Error **errp)
2989 BlockDriverState *bs = NULL;
2990 QDict *image_options;
2991 char *bdref_key_dot;
2992 const char *reference;
2994 assert(child_class != NULL);
2996 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2997 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2998 g_free(bdref_key_dot);
3001 * Caution: while qdict_get_try_str() is fine, getting non-string
3002 * types would require more care. When @options come from
3003 * -blockdev or blockdev_add, its members are typed according to
3004 * the QAPI schema, but when they come from -drive, they're all
3007 reference = qdict_get_try_str(options, bdref_key);
3008 if (!filename && !reference && !qdict_size(image_options)) {
3010 error_setg(errp, "A block device must be specified for \"%s\"",
3013 qobject_unref(image_options);
3017 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3018 parent, child_class, child_role, errp);
3024 qdict_del(options, bdref_key);
3029 * Opens a disk image whose options are given as BlockdevRef in another block
3032 * If allow_none is true, no image will be opened if filename is false and no
3033 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3035 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3036 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3037 * itself, all options starting with "${bdref_key}." are considered part of the
3040 * The BlockdevRef will be removed from the options QDict.
3042 BdrvChild *bdrv_open_child(const char *filename,
3043 QDict *options, const char *bdref_key,
3044 BlockDriverState *parent,
3045 const BdrvChildClass *child_class,
3046 BdrvChildRole child_role,
3047 bool allow_none, Error **errp)
3049 BlockDriverState *bs;
3051 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3052 child_role, allow_none, errp);
3057 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3062 * TODO Future callers may need to specify parent/child_class in order for
3063 * option inheritance to work. Existing callers use it for the root node.
3065 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3067 BlockDriverState *bs = NULL;
3068 QObject *obj = NULL;
3069 QDict *qdict = NULL;
3070 const char *reference = NULL;
3073 if (ref->type == QTYPE_QSTRING) {
3074 reference = ref->u.reference;
3076 BlockdevOptions *options = &ref->u.definition;
3077 assert(ref->type == QTYPE_QDICT);
3079 v = qobject_output_visitor_new(&obj);
3080 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3081 visit_complete(v, &obj);
3083 qdict = qobject_to(QDict, obj);
3084 qdict_flatten(qdict);
3086 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3087 * compatibility with other callers) rather than what we want as the
3088 * real defaults. Apply the defaults here instead. */
3089 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3090 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3091 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3092 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3096 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3103 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3105 QDict *snapshot_options,
3108 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3109 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3111 QemuOpts *opts = NULL;
3112 BlockDriverState *bs_snapshot = NULL;
3113 Error *local_err = NULL;
3116 /* if snapshot, we create a temporary backing file and open it
3117 instead of opening 'filename' directly */
3119 /* Get the required size from the image */
3120 total_size = bdrv_getlength(bs);
3121 if (total_size < 0) {
3122 error_setg_errno(errp, -total_size, "Could not get image size");
3126 /* Create the temporary image */
3127 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3129 error_setg_errno(errp, -ret, "Could not get temporary filename");
3133 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3135 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3136 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3137 qemu_opts_del(opts);
3139 error_prepend(errp, "Could not create temporary overlay '%s': ",
3144 /* Prepare options QDict for the temporary file */
3145 qdict_put_str(snapshot_options, "file.driver", "file");
3146 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3147 qdict_put_str(snapshot_options, "driver", "qcow2");
3149 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3150 snapshot_options = NULL;
3155 /* bdrv_append() consumes a strong reference to bs_snapshot
3156 * (i.e. it will call bdrv_unref() on it) even on error, so in
3157 * order to be able to return one, we have to increase
3158 * bs_snapshot's refcount here */
3159 bdrv_ref(bs_snapshot);
3160 bdrv_append(bs_snapshot, bs, &local_err);
3162 error_propagate(errp, local_err);
3168 qobject_unref(snapshot_options);
3169 g_free(tmp_filename);
3174 * Opens a disk image (raw, qcow2, vmdk, ...)
3176 * options is a QDict of options to pass to the block drivers, or NULL for an
3177 * empty set of options. The reference to the QDict belongs to the block layer
3178 * after the call (even on failure), so if the caller intends to reuse the
3179 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3181 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3182 * If it is not NULL, the referenced BDS will be reused.
3184 * The reference parameter may be used to specify an existing block device which
3185 * should be opened. If specified, neither options nor a filename may be given,
3186 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3188 static BlockDriverState *bdrv_open_inherit(const char *filename,
3189 const char *reference,
3190 QDict *options, int flags,
3191 BlockDriverState *parent,
3192 const BdrvChildClass *child_class,
3193 BdrvChildRole child_role,
3197 BlockBackend *file = NULL;
3198 BlockDriverState *bs;
3199 BlockDriver *drv = NULL;
3201 const char *drvname;
3202 const char *backing;
3203 Error *local_err = NULL;
3204 QDict *snapshot_options = NULL;
3205 int snapshot_flags = 0;
3207 assert(!child_class || !flags);
3208 assert(!child_class == !parent);
3211 bool options_non_empty = options ? qdict_size(options) : false;
3212 qobject_unref(options);
3214 if (filename || options_non_empty) {
3215 error_setg(errp, "Cannot reference an existing block device with "
3216 "additional options or a new filename");
3220 bs = bdrv_lookup_bs(reference, reference, errp);
3231 /* NULL means an empty set of options */
3232 if (options == NULL) {
3233 options = qdict_new();
3236 /* json: syntax counts as explicit options, as if in the QDict */
3237 parse_json_protocol(options, &filename, &local_err);
3242 bs->explicit_options = qdict_clone_shallow(options);
3245 bool parent_is_format;
3248 parent_is_format = parent->drv->is_format;
3251 * parent->drv is not set yet because this node is opened for
3252 * (potential) format probing. That means that @parent is going
3253 * to be a format node.
3255 parent_is_format = true;
3258 bs->inherits_from = parent;
3259 child_class->inherit_options(child_role, parent_is_format,
3261 parent->open_flags, parent->options);
3264 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3270 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3271 * Caution: getting a boolean member of @options requires care.
3272 * When @options come from -blockdev or blockdev_add, members are
3273 * typed according to the QAPI schema, but when they come from
3274 * -drive, they're all QString.
3276 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3277 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3278 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3280 flags &= ~BDRV_O_RDWR;
3283 if (flags & BDRV_O_SNAPSHOT) {
3284 snapshot_options = qdict_new();
3285 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3287 /* Let bdrv_backing_options() override "read-only" */
3288 qdict_del(options, BDRV_OPT_READ_ONLY);
3289 bdrv_inherited_options(BDRV_CHILD_COW, true,
3290 &flags, options, flags, options);
3293 bs->open_flags = flags;
3294 bs->options = options;
3295 options = qdict_clone_shallow(options);
3297 /* Find the right image format driver */
3298 /* See cautionary note on accessing @options above */
3299 drvname = qdict_get_try_str(options, "driver");
3301 drv = bdrv_find_format(drvname);
3303 error_setg(errp, "Unknown driver: '%s'", drvname);
3308 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3310 /* See cautionary note on accessing @options above */
3311 backing = qdict_get_try_str(options, "backing");
3312 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3313 (backing && *backing == '\0'))
3316 warn_report("Use of \"backing\": \"\" is deprecated; "
3317 "use \"backing\": null instead");
3319 flags |= BDRV_O_NO_BACKING;
3320 qdict_del(bs->explicit_options, "backing");
3321 qdict_del(bs->options, "backing");
3322 qdict_del(options, "backing");
3325 /* Open image file without format layer. This BlockBackend is only used for
3326 * probing, the block drivers will do their own bdrv_open_child() for the
3327 * same BDS, which is why we put the node name back into options. */
3328 if ((flags & BDRV_O_PROTOCOL) == 0) {
3329 BlockDriverState *file_bs;
3331 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3332 &child_of_bds, BDRV_CHILD_IMAGE,
3337 if (file_bs != NULL) {
3338 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3339 * looking at the header to guess the image format. This works even
3340 * in cases where a guest would not see a consistent state. */
3341 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3342 blk_insert_bs(file, file_bs, &local_err);
3343 bdrv_unref(file_bs);
3348 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3352 /* Image format probing */
3355 ret = find_image_format(file, filename, &drv, &local_err);
3360 * This option update would logically belong in bdrv_fill_options(),
3361 * but we first need to open bs->file for the probing to work, while
3362 * opening bs->file already requires the (mostly) final set of options
3363 * so that cache mode etc. can be inherited.
3365 * Adding the driver later is somewhat ugly, but it's not an option
3366 * that would ever be inherited, so it's correct. We just need to make
3367 * sure to update both bs->options (which has the full effective
3368 * options for bs) and options (which has file.* already removed).
3370 qdict_put_str(bs->options, "driver", drv->format_name);
3371 qdict_put_str(options, "driver", drv->format_name);
3373 error_setg(errp, "Must specify either driver or file");
3377 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3378 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3379 /* file must be NULL if a protocol BDS is about to be created
3380 * (the inverse results in an error message from bdrv_open_common()) */
3381 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3383 /* Open the image */
3384 ret = bdrv_open_common(bs, file, options, &local_err);
3394 /* If there is a backing file, use it */
3395 if ((flags & BDRV_O_NO_BACKING) == 0) {
3396 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3398 goto close_and_fail;
3402 /* Remove all children options and references
3403 * from bs->options and bs->explicit_options */
3404 QLIST_FOREACH(child, &bs->children, next) {
3405 char *child_key_dot;
3406 child_key_dot = g_strdup_printf("%s.", child->name);
3407 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3408 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3409 qdict_del(bs->explicit_options, child->name);
3410 qdict_del(bs->options, child->name);
3411 g_free(child_key_dot);
3414 /* Check if any unknown options were used */
3415 if (qdict_size(options) != 0) {
3416 const QDictEntry *entry = qdict_first(options);
3417 if (flags & BDRV_O_PROTOCOL) {
3418 error_setg(errp, "Block protocol '%s' doesn't support the option "
3419 "'%s'", drv->format_name, entry->key);
3422 "Block format '%s' does not support the option '%s'",
3423 drv->format_name, entry->key);
3426 goto close_and_fail;
3429 bdrv_parent_cb_change_media(bs, true);
3431 qobject_unref(options);
3434 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3435 * temporary snapshot afterwards. */
3436 if (snapshot_flags) {
3437 BlockDriverState *snapshot_bs;
3438 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3439 snapshot_options, &local_err);
3440 snapshot_options = NULL;
3442 goto close_and_fail;
3444 /* We are not going to return bs but the overlay on top of it
3445 * (snapshot_bs); thus, we have to drop the strong reference to bs
3446 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3447 * though, because the overlay still has a reference to it. */
3456 qobject_unref(snapshot_options);
3457 qobject_unref(bs->explicit_options);
3458 qobject_unref(bs->options);
3459 qobject_unref(options);
3461 bs->explicit_options = NULL;
3463 error_propagate(errp, local_err);
3468 qobject_unref(snapshot_options);
3469 qobject_unref(options);
3470 error_propagate(errp, local_err);
3474 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3475 QDict *options, int flags, Error **errp)
3477 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3481 /* Return true if the NULL-terminated @list contains @str */
3482 static bool is_str_in_list(const char *str, const char *const *list)
3486 for (i = 0; list[i] != NULL; i++) {
3487 if (!strcmp(str, list[i])) {
3496 * Check that every option set in @bs->options is also set in
3499 * Options listed in the common_options list and in
3500 * @bs->drv->mutable_opts are skipped.
3502 * Return 0 on success, otherwise return -EINVAL and set @errp.
3504 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3505 const QDict *new_opts, Error **errp)
3507 const QDictEntry *e;
3508 /* These options are common to all block drivers and are handled
3509 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3510 const char *const common_options[] = {
3511 "node-name", "discard", "cache.direct", "cache.no-flush",
3512 "read-only", "auto-read-only", "detect-zeroes", NULL
3515 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3516 if (!qdict_haskey(new_opts, e->key) &&
3517 !is_str_in_list(e->key, common_options) &&
3518 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3519 error_setg(errp, "Option '%s' cannot be reset "
3520 "to its default value", e->key);
3529 * Returns true if @child can be reached recursively from @bs
3531 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3532 BlockDriverState *child)
3540 QLIST_FOREACH(c, &bs->children, next) {
3541 if (bdrv_recurse_has_child(c->bs, child)) {
3550 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3551 * reopen of multiple devices.
3553 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3554 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3555 * be created and initialized. This newly created BlockReopenQueue should be
3556 * passed back in for subsequent calls that are intended to be of the same
3559 * bs is the BlockDriverState to add to the reopen queue.
3561 * options contains the changed options for the associated bs
3562 * (the BlockReopenQueue takes ownership)
3564 * flags contains the open flags for the associated bs
3566 * returns a pointer to bs_queue, which is either the newly allocated
3567 * bs_queue, or the existing bs_queue being used.
3569 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3571 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3572 BlockDriverState *bs,
3574 const BdrvChildClass *klass,
3576 bool parent_is_format,
3577 QDict *parent_options,
3583 BlockReopenQueueEntry *bs_entry;
3585 QDict *old_options, *explicit_options, *options_copy;
3589 /* Make sure that the caller remembered to use a drained section. This is
3590 * important to avoid graph changes between the recursive queuing here and
3591 * bdrv_reopen_multiple(). */
3592 assert(bs->quiesce_counter > 0);
3594 if (bs_queue == NULL) {
3595 bs_queue = g_new0(BlockReopenQueue, 1);
3596 QTAILQ_INIT(bs_queue);
3600 options = qdict_new();
3603 /* Check if this BlockDriverState is already in the queue */
3604 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3605 if (bs == bs_entry->state.bs) {
3611 * Precedence of options:
3612 * 1. Explicitly passed in options (highest)
3613 * 2. Retained from explicitly set options of bs
3614 * 3. Inherited from parent node
3615 * 4. Retained from effective options of bs
3618 /* Old explicitly set values (don't overwrite by inherited value) */
3619 if (bs_entry || keep_old_opts) {
3620 old_options = qdict_clone_shallow(bs_entry ?
3621 bs_entry->state.explicit_options :
3622 bs->explicit_options);
3623 bdrv_join_options(bs, options, old_options);
3624 qobject_unref(old_options);
3627 explicit_options = qdict_clone_shallow(options);
3629 /* Inherit from parent node */
3630 if (parent_options) {
3632 klass->inherit_options(role, parent_is_format, &flags, options,
3633 parent_flags, parent_options);
3635 flags = bdrv_get_flags(bs);
3638 if (keep_old_opts) {
3639 /* Old values are used for options that aren't set yet */
3640 old_options = qdict_clone_shallow(bs->options);
3641 bdrv_join_options(bs, options, old_options);
3642 qobject_unref(old_options);
3645 /* We have the final set of options so let's update the flags */
3646 options_copy = qdict_clone_shallow(options);
3647 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3648 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3649 update_flags_from_options(&flags, opts);
3650 qemu_opts_del(opts);
3651 qobject_unref(options_copy);
3653 /* bdrv_open_inherit() sets and clears some additional flags internally */
3654 flags &= ~BDRV_O_PROTOCOL;
3655 if (flags & BDRV_O_RDWR) {
3656 flags |= BDRV_O_ALLOW_RDWR;
3660 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3661 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3663 qobject_unref(bs_entry->state.options);
3664 qobject_unref(bs_entry->state.explicit_options);
3667 bs_entry->state.bs = bs;
3668 bs_entry->state.options = options;
3669 bs_entry->state.explicit_options = explicit_options;
3670 bs_entry->state.flags = flags;
3672 /* This needs to be overwritten in bdrv_reopen_prepare() */
3673 bs_entry->state.perm = UINT64_MAX;
3674 bs_entry->state.shared_perm = 0;
3677 * If keep_old_opts is false then it means that unspecified
3678 * options must be reset to their original value. We don't allow
3679 * resetting 'backing' but we need to know if the option is
3680 * missing in order to decide if we have to return an error.
3682 if (!keep_old_opts) {
3683 bs_entry->state.backing_missing =
3684 !qdict_haskey(options, "backing") &&
3685 !qdict_haskey(options, "backing.driver");
3688 QLIST_FOREACH(child, &bs->children, next) {
3689 QDict *new_child_options = NULL;
3690 bool child_keep_old = keep_old_opts;
3692 /* reopen can only change the options of block devices that were
3693 * implicitly created and inherited options. For other (referenced)
3694 * block devices, a syntax like "backing.foo" results in an error. */
3695 if (child->bs->inherits_from != bs) {
3699 /* Check if the options contain a child reference */
3700 if (qdict_haskey(options, child->name)) {
3701 const char *childref = qdict_get_try_str(options, child->name);
3703 * The current child must not be reopened if the child
3704 * reference is null or points to a different node.
3706 if (g_strcmp0(childref, child->bs->node_name)) {
3710 * If the child reference points to the current child then
3711 * reopen it with its existing set of options (note that
3712 * it can still inherit new options from the parent).
3714 child_keep_old = true;
3716 /* Extract child options ("child-name.*") */
3717 char *child_key_dot = g_strdup_printf("%s.", child->name);
3718 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3719 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3720 g_free(child_key_dot);
3723 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3724 child->klass, child->role, bs->drv->is_format,
3725 options, flags, child_keep_old);
3731 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3732 BlockDriverState *bs,
3733 QDict *options, bool keep_old_opts)
3735 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
3736 NULL, 0, keep_old_opts);
3740 * Reopen multiple BlockDriverStates atomically & transactionally.
3742 * The queue passed in (bs_queue) must have been built up previous
3743 * via bdrv_reopen_queue().
3745 * Reopens all BDS specified in the queue, with the appropriate
3746 * flags. All devices are prepared for reopen, and failure of any
3747 * device will cause all device changes to be abandoned, and intermediate
3750 * If all devices prepare successfully, then the changes are committed
3753 * All affected nodes must be drained between bdrv_reopen_queue() and
3754 * bdrv_reopen_multiple().
3756 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
3759 BlockReopenQueueEntry *bs_entry, *next;
3761 assert(bs_queue != NULL);
3763 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3764 assert(bs_entry->state.bs->quiesce_counter > 0);
3765 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
3768 bs_entry->prepared = true;
3771 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3772 BDRVReopenState *state = &bs_entry->state;
3773 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
3774 state->shared_perm, NULL, errp);
3778 /* Check if new_backing_bs would accept the new permissions */
3779 if (state->replace_backing_bs && state->new_backing_bs) {
3780 uint64_t nperm, nshared;
3781 bdrv_child_perm(state->bs, state->new_backing_bs,
3782 NULL, bdrv_backing_role(state->bs),
3783 bs_queue, state->perm, state->shared_perm,
3785 ret = bdrv_check_update_perm(state->new_backing_bs, NULL,
3786 nperm, nshared, NULL, errp);
3791 bs_entry->perms_checked = true;
3795 * If we reach this point, we have success and just need to apply the
3798 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3799 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3800 * children are usually goes after parents in reopen-queue, so go from last
3803 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3804 bdrv_reopen_commit(&bs_entry->state);
3809 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3810 BDRVReopenState *state = &bs_entry->state;
3812 if (!bs_entry->perms_checked) {
3817 uint64_t perm, shared;
3819 bdrv_get_cumulative_perm(state->bs, &perm, &shared);
3820 assert(perm == state->perm);
3821 assert(shared == state->shared_perm);
3823 bdrv_set_perm(state->bs);
3825 bdrv_abort_perm_update(state->bs);
3826 if (state->replace_backing_bs && state->new_backing_bs) {
3827 bdrv_abort_perm_update(state->new_backing_bs);
3833 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3834 BlockDriverState *bs = bs_entry->state.bs;
3836 if (bs->drv->bdrv_reopen_commit_post)
3837 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
3841 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3843 if (bs_entry->prepared) {
3844 bdrv_reopen_abort(&bs_entry->state);
3846 qobject_unref(bs_entry->state.explicit_options);
3847 qobject_unref(bs_entry->state.options);
3849 if (bs_entry->state.new_backing_bs) {
3850 bdrv_unref(bs_entry->state.new_backing_bs);
3859 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3863 BlockReopenQueue *queue;
3864 QDict *opts = qdict_new();
3866 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3868 bdrv_subtree_drained_begin(bs);
3869 queue = bdrv_reopen_queue(NULL, bs, opts, true);
3870 ret = bdrv_reopen_multiple(queue, errp);
3871 bdrv_subtree_drained_end(bs);
3876 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3879 BlockReopenQueueEntry *entry;
3881 QTAILQ_FOREACH(entry, q, entry) {
3882 BlockDriverState *bs = entry->state.bs;
3885 QLIST_FOREACH(child, &bs->children, next) {
3895 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3896 uint64_t *perm, uint64_t *shared)
3899 BlockReopenQueueEntry *parent;
3900 uint64_t cumulative_perms = 0;
3901 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3903 QLIST_FOREACH(c, &bs->parents, next_parent) {
3904 parent = find_parent_in_reopen_queue(q, c);
3906 cumulative_perms |= c->perm;
3907 cumulative_shared_perms &= c->shared_perm;
3909 uint64_t nperm, nshared;
3911 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3912 parent->state.perm, parent->state.shared_perm,
3915 cumulative_perms |= nperm;
3916 cumulative_shared_perms &= nshared;
3919 *perm = cumulative_perms;
3920 *shared = cumulative_shared_perms;
3923 static bool bdrv_reopen_can_attach(BlockDriverState *parent,
3925 BlockDriverState *new_child,
3928 AioContext *parent_ctx = bdrv_get_aio_context(parent);
3929 AioContext *child_ctx = bdrv_get_aio_context(new_child);
3933 ignore = g_slist_prepend(NULL, child);
3934 ret = bdrv_can_set_aio_context(new_child, parent_ctx, &ignore, NULL);
3935 g_slist_free(ignore);
3940 ignore = g_slist_prepend(NULL, child);
3941 ret = bdrv_can_set_aio_context(parent, child_ctx, &ignore, errp);
3942 g_slist_free(ignore);
3947 * Take a BDRVReopenState and check if the value of 'backing' in the
3948 * reopen_state->options QDict is valid or not.
3950 * If 'backing' is missing from the QDict then return 0.
3952 * If 'backing' contains the node name of the backing file of
3953 * reopen_state->bs then return 0.
3955 * If 'backing' contains a different node name (or is null) then check
3956 * whether the current backing file can be replaced with the new one.
3957 * If that's the case then reopen_state->replace_backing_bs is set to
3958 * true and reopen_state->new_backing_bs contains a pointer to the new
3959 * backing BlockDriverState (or NULL).
3961 * Return 0 on success, otherwise return < 0 and set @errp.
3963 static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
3966 BlockDriverState *bs = reopen_state->bs;
3967 BlockDriverState *overlay_bs, *below_bs, *new_backing_bs;
3971 value = qdict_get(reopen_state->options, "backing");
3972 if (value == NULL) {
3976 switch (qobject_type(value)) {
3978 new_backing_bs = NULL;
3981 str = qstring_get_str(qobject_to(QString, value));
3982 new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
3983 if (new_backing_bs == NULL) {
3985 } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
3986 error_setg(errp, "Making '%s' a backing file of '%s' "
3987 "would create a cycle", str, bs->node_name);
3992 /* 'backing' does not allow any other data type */
3993 g_assert_not_reached();
3997 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
3998 * bdrv_reopen_commit() won't fail.
4000 if (new_backing_bs) {
4001 if (!bdrv_reopen_can_attach(bs, bs->backing, new_backing_bs, errp)) {
4007 * Ensure that @bs can really handle backing files, because we are
4008 * about to give it one (or swap the existing one)
4010 if (bs->drv->is_filter) {
4011 /* Filters always have a file or a backing child */
4013 error_setg(errp, "'%s' is a %s filter node that does not support a "
4014 "backing child", bs->node_name, bs->drv->format_name);
4017 } else if (!bs->drv->supports_backing) {
4018 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
4019 "files", bs->drv->format_name, bs->node_name);
4024 * Find the "actual" backing file by skipping all links that point
4025 * to an implicit node, if any (e.g. a commit filter node).
4026 * We cannot use any of the bdrv_skip_*() functions here because
4027 * those return the first explicit node, while we are looking for
4031 for (below_bs = bdrv_filter_or_cow_bs(overlay_bs);
4032 below_bs && below_bs->implicit;
4033 below_bs = bdrv_filter_or_cow_bs(overlay_bs))
4035 overlay_bs = below_bs;
4038 /* If we want to replace the backing file we need some extra checks */
4039 if (new_backing_bs != bdrv_filter_or_cow_bs(overlay_bs)) {
4040 /* Check for implicit nodes between bs and its backing file */
4041 if (bs != overlay_bs) {
4042 error_setg(errp, "Cannot change backing link if '%s' has "
4043 "an implicit backing file", bs->node_name);
4047 * Check if the backing link that we want to replace is frozen.
4049 * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
4050 * because we know that overlay_bs == bs, and that @bs
4051 * either is a filter that uses ->backing or a COW format BDS
4052 * with bs->drv->supports_backing == true.
4054 if (bdrv_is_backing_chain_frozen(overlay_bs,
4055 child_bs(overlay_bs->backing), errp))
4059 reopen_state->replace_backing_bs = true;
4060 if (new_backing_bs) {
4061 bdrv_ref(new_backing_bs);
4062 reopen_state->new_backing_bs = new_backing_bs;
4070 * Prepares a BlockDriverState for reopen. All changes are staged in the
4071 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4072 * the block driver layer .bdrv_reopen_prepare()
4074 * bs is the BlockDriverState to reopen
4075 * flags are the new open flags
4076 * queue is the reopen queue
4078 * Returns 0 on success, non-zero on error. On error errp will be set
4081 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4082 * It is the responsibility of the caller to then call the abort() or
4083 * commit() for any other BDS that have been left in a prepare() state
4086 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
4091 Error *local_err = NULL;
4094 QDict *orig_reopen_opts;
4095 char *discard = NULL;
4097 bool drv_prepared = false;
4099 assert(reopen_state != NULL);
4100 assert(reopen_state->bs->drv != NULL);
4101 drv = reopen_state->bs->drv;
4103 /* This function and each driver's bdrv_reopen_prepare() remove
4104 * entries from reopen_state->options as they are processed, so
4105 * we need to make a copy of the original QDict. */
4106 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4108 /* Process generic block layer options */
4109 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4110 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4115 /* This was already called in bdrv_reopen_queue_child() so the flags
4116 * are up-to-date. This time we simply want to remove the options from
4117 * QemuOpts in order to indicate that they have been processed. */
4118 old_flags = reopen_state->flags;
4119 update_flags_from_options(&reopen_state->flags, opts);
4120 assert(old_flags == reopen_state->flags);
4122 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4123 if (discard != NULL) {
4124 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4125 error_setg(errp, "Invalid discard option");
4131 reopen_state->detect_zeroes =
4132 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4134 error_propagate(errp, local_err);
4139 /* All other options (including node-name and driver) must be unchanged.
4140 * Put them back into the QDict, so that they are checked at the end
4141 * of this function. */
4142 qemu_opts_to_qdict(opts, reopen_state->options);
4144 /* If we are to stay read-only, do not allow permission change
4145 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4146 * not set, or if the BDS still has copy_on_read enabled */
4147 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4148 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4150 error_propagate(errp, local_err);
4154 /* Calculate required permissions after reopening */
4155 bdrv_reopen_perm(queue, reopen_state->bs,
4156 &reopen_state->perm, &reopen_state->shared_perm);
4158 ret = bdrv_flush(reopen_state->bs);
4160 error_setg_errno(errp, -ret, "Error flushing drive");
4164 if (drv->bdrv_reopen_prepare) {
4166 * If a driver-specific option is missing, it means that we
4167 * should reset it to its default value.
4168 * But not all options allow that, so we need to check it first.
4170 ret = bdrv_reset_options_allowed(reopen_state->bs,
4171 reopen_state->options, errp);
4176 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4178 if (local_err != NULL) {
4179 error_propagate(errp, local_err);
4181 bdrv_refresh_filename(reopen_state->bs);
4182 error_setg(errp, "failed while preparing to reopen image '%s'",
4183 reopen_state->bs->filename);
4188 /* It is currently mandatory to have a bdrv_reopen_prepare()
4189 * handler for each supported drv. */
4190 error_setg(errp, "Block format '%s' used by node '%s' "
4191 "does not support reopening files", drv->format_name,
4192 bdrv_get_device_or_node_name(reopen_state->bs));
4197 drv_prepared = true;
4200 * We must provide the 'backing' option if the BDS has a backing
4201 * file or if the image file has a backing file name as part of
4202 * its metadata. Otherwise the 'backing' option can be omitted.
4204 if (drv->supports_backing && reopen_state->backing_missing &&
4205 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4206 error_setg(errp, "backing is missing for '%s'",
4207 reopen_state->bs->node_name);
4213 * Allow changing the 'backing' option. The new value can be
4214 * either a reference to an existing node (using its node name)
4215 * or NULL to simply detach the current backing file.
4217 ret = bdrv_reopen_parse_backing(reopen_state, errp);
4221 qdict_del(reopen_state->options, "backing");
4223 /* Options that are not handled are only okay if they are unchanged
4224 * compared to the old state. It is expected that some options are only
4225 * used for the initial open, but not reopen (e.g. filename) */
4226 if (qdict_size(reopen_state->options)) {
4227 const QDictEntry *entry = qdict_first(reopen_state->options);
4230 QObject *new = entry->value;
4231 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4233 /* Allow child references (child_name=node_name) as long as they
4234 * point to the current child (i.e. everything stays the same). */
4235 if (qobject_type(new) == QTYPE_QSTRING) {
4237 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4238 if (!strcmp(child->name, entry->key)) {
4244 if (!strcmp(child->bs->node_name,
4245 qstring_get_str(qobject_to(QString, new)))) {
4246 continue; /* Found child with this name, skip option */
4252 * TODO: When using -drive to specify blockdev options, all values
4253 * will be strings; however, when using -blockdev, blockdev-add or
4254 * filenames using the json:{} pseudo-protocol, they will be
4256 * In contrast, reopening options are (currently) always strings
4257 * (because you can only specify them through qemu-io; all other
4258 * callers do not specify any options).
4259 * Therefore, when using anything other than -drive to create a BDS,
4260 * this cannot detect non-string options as unchanged, because
4261 * qobject_is_equal() always returns false for objects of different
4262 * type. In the future, this should be remedied by correctly typing
4263 * all options. For now, this is not too big of an issue because
4264 * the user can simply omit options which cannot be changed anyway,
4265 * so they will stay unchanged.
4267 if (!qobject_is_equal(new, old)) {
4268 error_setg(errp, "Cannot change the option '%s'", entry->key);
4272 } while ((entry = qdict_next(reopen_state->options, entry)));
4277 /* Restore the original reopen_state->options QDict */
4278 qobject_unref(reopen_state->options);
4279 reopen_state->options = qobject_ref(orig_reopen_opts);
4282 if (ret < 0 && drv_prepared) {
4283 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4284 * call drv->bdrv_reopen_abort() before signaling an error
4285 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4286 * when the respective bdrv_reopen_prepare() has failed) */
4287 if (drv->bdrv_reopen_abort) {
4288 drv->bdrv_reopen_abort(reopen_state);
4291 qemu_opts_del(opts);
4292 qobject_unref(orig_reopen_opts);
4298 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4299 * makes them final by swapping the staging BlockDriverState contents into
4300 * the active BlockDriverState contents.
4302 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4305 BlockDriverState *bs;
4308 assert(reopen_state != NULL);
4309 bs = reopen_state->bs;
4311 assert(drv != NULL);
4313 /* If there are any driver level actions to take */
4314 if (drv->bdrv_reopen_commit) {
4315 drv->bdrv_reopen_commit(reopen_state);
4318 /* set BDS specific flags now */
4319 qobject_unref(bs->explicit_options);
4320 qobject_unref(bs->options);
4322 bs->explicit_options = reopen_state->explicit_options;
4323 bs->options = reopen_state->options;
4324 bs->open_flags = reopen_state->flags;
4325 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
4326 bs->detect_zeroes = reopen_state->detect_zeroes;
4328 if (reopen_state->replace_backing_bs) {
4329 qdict_del(bs->explicit_options, "backing");
4330 qdict_del(bs->options, "backing");
4333 /* Remove child references from bs->options and bs->explicit_options.
4334 * Child options were already removed in bdrv_reopen_queue_child() */
4335 QLIST_FOREACH(child, &bs->children, next) {
4336 qdict_del(bs->explicit_options, child->name);
4337 qdict_del(bs->options, child->name);
4341 * Change the backing file if a new one was specified. We do this
4342 * after updating bs->options, so bdrv_refresh_filename() (called
4343 * from bdrv_set_backing_hd()) has the new values.
4345 if (reopen_state->replace_backing_bs) {
4346 BlockDriverState *old_backing_bs = child_bs(bs->backing);
4347 assert(!old_backing_bs || !old_backing_bs->implicit);
4348 /* Abort the permission update on the backing bs we're detaching */
4349 if (old_backing_bs) {
4350 bdrv_abort_perm_update(old_backing_bs);
4352 bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort);
4355 bdrv_refresh_limits(bs, NULL);
4359 * Abort the reopen, and delete and free the staged changes in
4362 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4366 assert(reopen_state != NULL);
4367 drv = reopen_state->bs->drv;
4368 assert(drv != NULL);
4370 if (drv->bdrv_reopen_abort) {
4371 drv->bdrv_reopen_abort(reopen_state);
4376 static void bdrv_close(BlockDriverState *bs)
4378 BdrvAioNotifier *ban, *ban_next;
4379 BdrvChild *child, *next;
4381 assert(!bs->refcnt);
4383 bdrv_drained_begin(bs); /* complete I/O */
4385 bdrv_drain(bs); /* in case flush left pending I/O */
4388 if (bs->drv->bdrv_close) {
4389 /* Must unfreeze all children, so bdrv_unref_child() works */
4390 bs->drv->bdrv_close(bs);
4395 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4396 bdrv_unref_child(bs, child);
4403 qatomic_set(&bs->copy_on_read, 0);
4404 bs->backing_file[0] = '\0';
4405 bs->backing_format[0] = '\0';
4406 bs->total_sectors = 0;
4407 bs->encrypted = false;
4409 qobject_unref(bs->options);
4410 qobject_unref(bs->explicit_options);
4412 bs->explicit_options = NULL;
4413 qobject_unref(bs->full_open_options);
4414 bs->full_open_options = NULL;
4416 bdrv_release_named_dirty_bitmaps(bs);
4417 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4419 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4422 QLIST_INIT(&bs->aio_notifiers);
4423 bdrv_drained_end(bs);
4426 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4427 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4430 if (bs->quiesce_counter) {
4431 bdrv_drain_all_end_quiesce(bs);
4435 void bdrv_close_all(void)
4437 assert(job_next(NULL) == NULL);
4438 blk_exp_close_all();
4440 /* Drop references from requests still in flight, such as canceled block
4441 * jobs whose AIO context has not been polled yet */
4444 blk_remove_all_bs();
4445 blockdev_close_all_bdrv_states();
4447 assert(QTAILQ_EMPTY(&all_bdrv_states));
4450 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4456 if (c->klass->stay_at_node) {
4460 /* If the child @c belongs to the BDS @to, replacing the current
4461 * c->bs by @to would mean to create a loop.
4463 * Such a case occurs when appending a BDS to a backing chain.
4464 * For instance, imagine the following chain:
4466 * guest device -> node A -> further backing chain...
4468 * Now we create a new BDS B which we want to put on top of this
4469 * chain, so we first attach A as its backing node:
4474 * guest device -> node A -> further backing chain...
4476 * Finally we want to replace A by B. When doing that, we want to
4477 * replace all pointers to A by pointers to B -- except for the
4478 * pointer from B because (1) that would create a loop, and (2)
4479 * that pointer should simply stay intact:
4481 * guest device -> node B
4484 * node A -> further backing chain...
4486 * In general, when replacing a node A (c->bs) by a node B (@to),
4487 * if A is a child of B, that means we cannot replace A by B there
4488 * because that would create a loop. Silently detaching A from B
4489 * is also not really an option. So overall just leaving A in
4490 * place there is the most sensible choice.
4492 * We would also create a loop in any cases where @c is only
4493 * indirectly referenced by @to. Prevent this by returning false
4494 * if @c is found (by breadth-first search) anywhere in the whole
4499 found = g_hash_table_new(NULL, NULL);
4500 g_hash_table_add(found, to);
4501 queue = g_queue_new();
4502 g_queue_push_tail(queue, to);
4504 while (!g_queue_is_empty(queue)) {
4505 BlockDriverState *v = g_queue_pop_head(queue);
4508 QLIST_FOREACH(c2, &v->children, next) {
4514 if (g_hash_table_contains(found, c2->bs)) {
4518 g_queue_push_tail(queue, c2->bs);
4519 g_hash_table_add(found, c2->bs);
4523 g_queue_free(queue);
4524 g_hash_table_destroy(found);
4530 * With auto_skip=true bdrv_replace_node_common skips updating from parents
4531 * if it creates a parent-child relation loop or if parent is block-job.
4533 * With auto_skip=false the error is returned if from has a parent which should
4536 static void bdrv_replace_node_common(BlockDriverState *from,
4537 BlockDriverState *to,
4538 bool auto_skip, Error **errp)
4540 BdrvChild *c, *next;
4541 GSList *list = NULL, *p;
4542 uint64_t perm = 0, shared = BLK_PERM_ALL;
4545 /* Make sure that @from doesn't go away until we have successfully attached
4546 * all of its parents to @to. */
4549 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4550 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
4551 bdrv_drained_begin(from);
4553 /* Put all parents into @list and calculate their cumulative permissions */
4554 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4555 assert(c->bs == from);
4556 if (!should_update_child(c, to)) {
4560 error_setg(errp, "Should not change '%s' link to '%s'",
4561 c->name, from->node_name);
4565 error_setg(errp, "Cannot change '%s' link to '%s'",
4566 c->name, from->node_name);
4569 list = g_slist_prepend(list, c);
4571 shared &= c->shared_perm;
4574 /* Check whether the required permissions can be granted on @to, ignoring
4575 * all BdrvChild in @list so that they can't block themselves. */
4576 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
4578 bdrv_abort_perm_update(to);
4582 /* Now actually perform the change. We performed the permission check for
4583 * all elements of @list at once, so set the permissions all at once at the
4585 for (p = list; p != NULL; p = p->next) {
4589 bdrv_replace_child_noperm(c, to);
4597 bdrv_drained_end(from);
4601 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
4604 return bdrv_replace_node_common(from, to, true, errp);
4608 * Add new bs contents at the top of an image chain while the chain is
4609 * live, while keeping required fields on the top layer.
4611 * This will modify the BlockDriverState fields, and swap contents
4612 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4614 * bs_new must not be attached to a BlockBackend.
4616 * This function does not create any image files.
4618 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4619 * that's what the callers commonly need. bs_new will be referenced by the old
4620 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4621 * reference of its own, it must call bdrv_ref().
4623 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
4626 Error *local_err = NULL;
4628 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
4630 error_propagate(errp, local_err);
4634 bdrv_replace_node(bs_top, bs_new, &local_err);
4636 error_propagate(errp, local_err);
4637 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
4641 /* bs_new is now referenced by its new parents, we don't need the
4642 * additional reference any more. */
4647 static void bdrv_delete(BlockDriverState *bs)
4649 assert(bdrv_op_blocker_is_empty(bs));
4650 assert(!bs->refcnt);
4652 /* remove from list, if necessary */
4653 if (bs->node_name[0] != '\0') {
4654 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
4656 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
4664 * Run consistency checks on an image
4666 * Returns 0 if the check could be completed (it doesn't mean that the image is
4667 * free of errors) or -errno when an internal error occurred. The results of the
4668 * check are stored in res.
4670 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
4671 BdrvCheckResult *res, BdrvCheckMode fix)
4673 if (bs->drv == NULL) {
4676 if (bs->drv->bdrv_co_check == NULL) {
4680 memset(res, 0, sizeof(*res));
4681 return bs->drv->bdrv_co_check(bs, res, fix);
4687 * -EINVAL - backing format specified, but no file
4688 * -ENOSPC - can't update the backing file because no space is left in the
4690 * -ENOTSUP - format driver doesn't support changing the backing file
4692 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
4693 const char *backing_fmt, bool warn)
4695 BlockDriver *drv = bs->drv;
4702 /* Backing file format doesn't make sense without a backing file */
4703 if (backing_fmt && !backing_file) {
4707 if (warn && backing_file && !backing_fmt) {
4708 warn_report("Deprecated use of backing file without explicit "
4709 "backing format, use of this image requires "
4710 "potentially unsafe format probing");
4713 if (drv->bdrv_change_backing_file != NULL) {
4714 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
4720 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
4721 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
4722 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
4723 backing_file ?: "");
4729 * Finds the first non-filter node above bs in the chain between
4730 * active and bs. The returned node is either an immediate parent of
4731 * bs, or there are only filter nodes between the two.
4733 * Returns NULL if bs is not found in active's image chain,
4734 * or if active == bs.
4736 * Returns the bottommost base image if bs == NULL.
4738 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
4739 BlockDriverState *bs)
4741 bs = bdrv_skip_filters(bs);
4742 active = bdrv_skip_filters(active);
4745 BlockDriverState *next = bdrv_backing_chain_next(active);
4755 /* Given a BDS, searches for the base layer. */
4756 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4758 return bdrv_find_overlay(bs, NULL);
4762 * Return true if at least one of the COW (backing) and filter links
4763 * between @bs and @base is frozen. @errp is set if that's the case.
4764 * @base must be reachable from @bs, or NULL.
4766 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
4769 BlockDriverState *i;
4772 for (i = bs; i != base; i = child_bs(child)) {
4773 child = bdrv_filter_or_cow_child(i);
4775 if (child && child->frozen) {
4776 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4777 child->name, i->node_name, child->bs->node_name);
4786 * Freeze all COW (backing) and filter links between @bs and @base.
4787 * If any of the links is already frozen the operation is aborted and
4788 * none of the links are modified.
4789 * @base must be reachable from @bs, or NULL.
4790 * Returns 0 on success. On failure returns < 0 and sets @errp.
4792 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4795 BlockDriverState *i;
4798 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4802 for (i = bs; i != base; i = child_bs(child)) {
4803 child = bdrv_filter_or_cow_child(i);
4804 if (child && child->bs->never_freeze) {
4805 error_setg(errp, "Cannot freeze '%s' link to '%s'",
4806 child->name, child->bs->node_name);
4811 for (i = bs; i != base; i = child_bs(child)) {
4812 child = bdrv_filter_or_cow_child(i);
4814 child->frozen = true;
4822 * Unfreeze all COW (backing) and filter links between @bs and @base.
4823 * The caller must ensure that all links are frozen before using this
4825 * @base must be reachable from @bs, or NULL.
4827 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4829 BlockDriverState *i;
4832 for (i = bs; i != base; i = child_bs(child)) {
4833 child = bdrv_filter_or_cow_child(i);
4835 assert(child->frozen);
4836 child->frozen = false;
4842 * Drops images above 'base' up to and including 'top', and sets the image
4843 * above 'top' to have base as its backing file.
4845 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4846 * information in 'bs' can be properly updated.
4848 * E.g., this will convert the following chain:
4849 * bottom <- base <- intermediate <- top <- active
4853 * bottom <- base <- active
4855 * It is allowed for bottom==base, in which case it converts:
4857 * base <- intermediate <- top <- active
4863 * If backing_file_str is non-NULL, it will be used when modifying top's
4864 * overlay image metadata.
4867 * if active == top, that is considered an error
4870 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4871 const char *backing_file_str)
4873 BlockDriverState *explicit_top = top;
4874 bool update_inherits_from;
4876 Error *local_err = NULL;
4878 g_autoptr(GSList) updated_children = NULL;
4882 bdrv_subtree_drained_begin(top);
4884 if (!top->drv || !base->drv) {
4888 /* Make sure that base is in the backing chain of top */
4889 if (!bdrv_chain_contains(top, base)) {
4893 /* If 'base' recursively inherits from 'top' then we should set
4894 * base->inherits_from to top->inherits_from after 'top' and all
4895 * other intermediate nodes have been dropped.
4896 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4897 * it because no one inherits from it. We use explicit_top for that. */
4898 explicit_top = bdrv_skip_implicit_filters(explicit_top);
4899 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4901 /* success - we can delete the intermediate states, and link top->base */
4902 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4903 * we've figured out how they should work. */
4904 if (!backing_file_str) {
4905 bdrv_refresh_filename(base);
4906 backing_file_str = base->filename;
4909 QLIST_FOREACH(c, &top->parents, next_parent) {
4910 updated_children = g_slist_prepend(updated_children, c);
4913 bdrv_replace_node_common(top, base, false, &local_err);
4915 error_report_err(local_err);
4919 for (p = updated_children; p; p = p->next) {
4922 if (c->klass->update_filename) {
4923 ret = c->klass->update_filename(c, base, backing_file_str,
4927 * TODO: Actually, we want to rollback all previous iterations
4928 * of this loop, and (which is almost impossible) previous
4929 * bdrv_replace_node()...
4931 * Note, that c->klass->update_filename may lead to permission
4932 * update, so it's a bad idea to call it inside permission
4933 * update transaction of bdrv_replace_node.
4935 error_report_err(local_err);
4941 if (update_inherits_from) {
4942 base->inherits_from = explicit_top->inherits_from;
4947 bdrv_subtree_drained_end(top);
4953 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
4954 * sums the size of all data-bearing children. (This excludes backing
4957 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
4960 int64_t child_size, sum = 0;
4962 QLIST_FOREACH(child, &bs->children, next) {
4963 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
4964 BDRV_CHILD_FILTERED))
4966 child_size = bdrv_get_allocated_file_size(child->bs);
4967 if (child_size < 0) {
4978 * Length of a allocated file in bytes. Sparse files are counted by actual
4979 * allocated space. Return < 0 if error or unknown.
4981 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
4983 BlockDriver *drv = bs->drv;
4987 if (drv->bdrv_get_allocated_file_size) {
4988 return drv->bdrv_get_allocated_file_size(bs);
4991 if (drv->bdrv_file_open) {
4993 * Protocol drivers default to -ENOTSUP (most of their data is
4994 * not stored in any of their children (if they even have any),
4995 * so there is no generic way to figure it out).
4998 } else if (drv->is_filter) {
4999 /* Filter drivers default to the size of their filtered child */
5000 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5002 /* Other drivers default to summing their children's sizes */
5003 return bdrv_sum_allocated_file_size(bs);
5009 * @drv: Format driver
5010 * @opts: Creation options for new image
5011 * @in_bs: Existing image containing data for new image (may be NULL)
5012 * @errp: Error object
5013 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5016 * Calculate file size required to create a new image.
5018 * If @in_bs is given then space for allocated clusters and zero clusters
5019 * from that image are included in the calculation. If @opts contains a
5020 * backing file that is shared by @in_bs then backing clusters may be omitted
5021 * from the calculation.
5023 * If @in_bs is NULL then the calculation includes no allocated clusters
5024 * unless a preallocation option is given in @opts.
5026 * Note that @in_bs may use a different BlockDriver from @drv.
5028 * If an error occurs the @errp pointer is set.
5030 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5031 BlockDriverState *in_bs, Error **errp)
5033 if (!drv->bdrv_measure) {
5034 error_setg(errp, "Block driver '%s' does not support size measurement",
5039 return drv->bdrv_measure(opts, in_bs, errp);
5043 * Return number of sectors on success, -errno on error.
5045 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5047 BlockDriver *drv = bs->drv;
5052 if (drv->has_variable_length) {
5053 int ret = refresh_total_sectors(bs, bs->total_sectors);
5058 return bs->total_sectors;
5062 * Return length in bytes on success, -errno on error.
5063 * The length is always a multiple of BDRV_SECTOR_SIZE.
5065 int64_t bdrv_getlength(BlockDriverState *bs)
5067 int64_t ret = bdrv_nb_sectors(bs);
5072 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5075 return ret * BDRV_SECTOR_SIZE;
5078 /* return 0 as number of sectors if no device present or error */
5079 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5081 int64_t nb_sectors = bdrv_nb_sectors(bs);
5083 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5086 bool bdrv_is_sg(BlockDriverState *bs)
5092 * Return whether the given node supports compressed writes.
5094 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5096 BlockDriverState *filtered;
5098 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5102 filtered = bdrv_filter_bs(bs);
5105 * Filters can only forward compressed writes, so we have to
5108 return bdrv_supports_compressed_writes(filtered);
5114 const char *bdrv_get_format_name(BlockDriverState *bs)
5116 return bs->drv ? bs->drv->format_name : NULL;
5119 static int qsort_strcmp(const void *a, const void *b)
5121 return strcmp(*(char *const *)a, *(char *const *)b);
5124 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5125 void *opaque, bool read_only)
5130 const char **formats = NULL;
5132 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5133 if (drv->format_name) {
5137 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5141 while (formats && i && !found) {
5142 found = !strcmp(formats[--i], drv->format_name);
5146 formats = g_renew(const char *, formats, count + 1);
5147 formats[count++] = drv->format_name;
5152 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5153 const char *format_name = block_driver_modules[i].format_name;
5159 if (use_bdrv_whitelist &&
5160 !bdrv_format_is_whitelisted(format_name, read_only)) {
5164 while (formats && j && !found) {
5165 found = !strcmp(formats[--j], format_name);
5169 formats = g_renew(const char *, formats, count + 1);
5170 formats[count++] = format_name;
5175 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5177 for (i = 0; i < count; i++) {
5178 it(opaque, formats[i]);
5184 /* This function is to find a node in the bs graph */
5185 BlockDriverState *bdrv_find_node(const char *node_name)
5187 BlockDriverState *bs;
5191 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5192 if (!strcmp(node_name, bs->node_name)) {
5199 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5200 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5203 BlockDeviceInfoList *list;
5204 BlockDriverState *bs;
5207 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5208 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5210 qapi_free_BlockDeviceInfoList(list);
5213 QAPI_LIST_PREPEND(list, info);
5219 typedef struct XDbgBlockGraphConstructor {
5220 XDbgBlockGraph *graph;
5221 GHashTable *graph_nodes;
5222 } XDbgBlockGraphConstructor;
5224 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5226 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5228 gr->graph = g_new0(XDbgBlockGraph, 1);
5229 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5234 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5236 XDbgBlockGraph *graph = gr->graph;
5238 g_hash_table_destroy(gr->graph_nodes);
5244 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5246 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5253 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5254 * answer of g_hash_table_lookup.
5256 ret = g_hash_table_size(gr->graph_nodes) + 1;
5257 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5262 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5263 XDbgBlockGraphNodeType type, const char *name)
5265 XDbgBlockGraphNode *n;
5267 n = g_new0(XDbgBlockGraphNode, 1);
5269 n->id = xdbg_graph_node_num(gr, node);
5271 n->name = g_strdup(name);
5273 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5276 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5277 const BdrvChild *child)
5279 BlockPermission qapi_perm;
5280 XDbgBlockGraphEdge *edge;
5282 edge = g_new0(XDbgBlockGraphEdge, 1);
5284 edge->parent = xdbg_graph_node_num(gr, parent);
5285 edge->child = xdbg_graph_node_num(gr, child->bs);
5286 edge->name = g_strdup(child->name);
5288 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5289 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5291 if (flag & child->perm) {
5292 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5294 if (flag & child->shared_perm) {
5295 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5299 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5303 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5307 BlockDriverState *bs;
5309 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5311 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5312 char *allocated_name = NULL;
5313 const char *name = blk_name(blk);
5316 name = allocated_name = blk_get_attached_dev_id(blk);
5318 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5320 g_free(allocated_name);
5321 if (blk_root(blk)) {
5322 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5326 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5329 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5331 for (el = job->nodes; el; el = el->next) {
5332 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5336 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5337 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5339 QLIST_FOREACH(child, &bs->children, next) {
5340 xdbg_graph_add_edge(gr, bs, child);
5344 return xdbg_graph_finalize(gr);
5347 BlockDriverState *bdrv_lookup_bs(const char *device,
5348 const char *node_name,
5352 BlockDriverState *bs;
5355 blk = blk_by_name(device);
5360 error_setg(errp, "Device '%s' has no medium", device);
5368 bs = bdrv_find_node(node_name);
5375 error_setg(errp, "Cannot find device=%s nor node_name=%s",
5376 device ? device : "",
5377 node_name ? node_name : "");
5381 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5382 * return false. If either argument is NULL, return false. */
5383 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
5385 while (top && top != base) {
5386 top = bdrv_filter_or_cow_bs(top);
5392 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
5395 return QTAILQ_FIRST(&graph_bdrv_states);
5397 return QTAILQ_NEXT(bs, node_list);
5400 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5403 return QTAILQ_FIRST(&all_bdrv_states);
5405 return QTAILQ_NEXT(bs, bs_list);
5408 const char *bdrv_get_node_name(const BlockDriverState *bs)
5410 return bs->node_name;
5413 const char *bdrv_get_parent_name(const BlockDriverState *bs)
5418 /* If multiple parents have a name, just pick the first one. */
5419 QLIST_FOREACH(c, &bs->parents, next_parent) {
5420 if (c->klass->get_name) {
5421 name = c->klass->get_name(c);
5422 if (name && *name) {
5431 /* TODO check what callers really want: bs->node_name or blk_name() */
5432 const char *bdrv_get_device_name(const BlockDriverState *bs)
5434 return bdrv_get_parent_name(bs) ?: "";
5437 /* This can be used to identify nodes that might not have a device
5438 * name associated. Since node and device names live in the same
5439 * namespace, the result is unambiguous. The exception is if both are
5440 * absent, then this returns an empty (non-null) string. */
5441 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
5443 return bdrv_get_parent_name(bs) ?: bs->node_name;
5446 int bdrv_get_flags(BlockDriverState *bs)
5448 return bs->open_flags;
5451 int bdrv_has_zero_init_1(BlockDriverState *bs)
5456 int bdrv_has_zero_init(BlockDriverState *bs)
5458 BlockDriverState *filtered;
5464 /* If BS is a copy on write image, it is initialized to
5465 the contents of the base image, which may not be zeroes. */
5466 if (bdrv_cow_child(bs)) {
5469 if (bs->drv->bdrv_has_zero_init) {
5470 return bs->drv->bdrv_has_zero_init(bs);
5473 filtered = bdrv_filter_bs(bs);
5475 return bdrv_has_zero_init(filtered);
5482 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
5484 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5488 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
5491 void bdrv_get_backing_filename(BlockDriverState *bs,
5492 char *filename, int filename_size)
5494 pstrcpy(filename, filename_size, bs->backing_file);
5497 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5500 BlockDriver *drv = bs->drv;
5501 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5505 if (!drv->bdrv_get_info) {
5506 BlockDriverState *filtered = bdrv_filter_bs(bs);
5508 return bdrv_get_info(filtered, bdi);
5512 memset(bdi, 0, sizeof(*bdi));
5513 ret = drv->bdrv_get_info(bs, bdi);
5518 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
5525 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
5528 BlockDriver *drv = bs->drv;
5529 if (drv && drv->bdrv_get_specific_info) {
5530 return drv->bdrv_get_specific_info(bs, errp);
5535 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
5537 BlockDriver *drv = bs->drv;
5538 if (!drv || !drv->bdrv_get_specific_stats) {
5541 return drv->bdrv_get_specific_stats(bs);
5544 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
5546 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
5550 bs->drv->bdrv_debug_event(bs, event);
5553 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
5555 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
5556 bs = bdrv_primary_bs(bs);
5559 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
5560 assert(bs->drv->bdrv_debug_remove_breakpoint);
5567 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
5570 bs = bdrv_find_debug_node(bs);
5572 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
5578 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
5580 bs = bdrv_find_debug_node(bs);
5582 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
5588 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
5590 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
5591 bs = bdrv_primary_bs(bs);
5594 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
5595 return bs->drv->bdrv_debug_resume(bs, tag);
5601 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
5603 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
5604 bs = bdrv_primary_bs(bs);
5607 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
5608 return bs->drv->bdrv_debug_is_suspended(bs, tag);
5614 /* backing_file can either be relative, or absolute, or a protocol. If it is
5615 * relative, it must be relative to the chain. So, passing in bs->filename
5616 * from a BDS as backing_file should not be done, as that may be relative to
5617 * the CWD rather than the chain. */
5618 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
5619 const char *backing_file)
5621 char *filename_full = NULL;
5622 char *backing_file_full = NULL;
5623 char *filename_tmp = NULL;
5624 int is_protocol = 0;
5625 bool filenames_refreshed = false;
5626 BlockDriverState *curr_bs = NULL;
5627 BlockDriverState *retval = NULL;
5628 BlockDriverState *bs_below;
5630 if (!bs || !bs->drv || !backing_file) {
5634 filename_full = g_malloc(PATH_MAX);
5635 backing_file_full = g_malloc(PATH_MAX);
5637 is_protocol = path_has_protocol(backing_file);
5640 * Being largely a legacy function, skip any filters here
5641 * (because filters do not have normal filenames, so they cannot
5642 * match anyway; and allowing json:{} filenames is a bit out of
5645 for (curr_bs = bdrv_skip_filters(bs);
5646 bdrv_cow_child(curr_bs) != NULL;
5649 bs_below = bdrv_backing_chain_next(curr_bs);
5651 if (bdrv_backing_overridden(curr_bs)) {
5653 * If the backing file was overridden, we can only compare
5654 * directly against the backing node's filename.
5657 if (!filenames_refreshed) {
5659 * This will automatically refresh all of the
5660 * filenames in the rest of the backing chain, so we
5661 * only need to do this once.
5663 bdrv_refresh_filename(bs_below);
5664 filenames_refreshed = true;
5667 if (strcmp(backing_file, bs_below->filename) == 0) {
5671 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
5673 * If either of the filename paths is actually a protocol, then
5674 * compare unmodified paths; otherwise make paths relative.
5676 char *backing_file_full_ret;
5678 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
5682 /* Also check against the full backing filename for the image */
5683 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
5685 if (backing_file_full_ret) {
5686 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
5687 g_free(backing_file_full_ret);
5694 /* If not an absolute filename path, make it relative to the current
5695 * image's filename path */
5696 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
5698 /* We are going to compare canonicalized absolute pathnames */
5699 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
5700 g_free(filename_tmp);
5703 g_free(filename_tmp);
5705 /* We need to make sure the backing filename we are comparing against
5706 * is relative to the current image filename (or absolute) */
5707 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
5708 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
5709 g_free(filename_tmp);
5712 g_free(filename_tmp);
5714 if (strcmp(backing_file_full, filename_full) == 0) {
5721 g_free(filename_full);
5722 g_free(backing_file_full);
5726 void bdrv_init(void)
5728 module_call_init(MODULE_INIT_BLOCK);
5731 void bdrv_init_with_whitelist(void)
5733 use_bdrv_whitelist = 1;
5737 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
5739 BdrvChild *child, *parent;
5740 Error *local_err = NULL;
5742 BdrvDirtyBitmap *bm;
5748 QLIST_FOREACH(child, &bs->children, next) {
5749 bdrv_co_invalidate_cache(child->bs, &local_err);
5751 error_propagate(errp, local_err);
5757 * Update permissions, they may differ for inactive nodes.
5759 * Note that the required permissions of inactive images are always a
5760 * subset of the permissions required after activating the image. This
5761 * allows us to just get the permissions upfront without restricting
5762 * drv->bdrv_invalidate_cache().
5764 * It also means that in error cases, we don't have to try and revert to
5765 * the old permissions (which is an operation that could fail, too). We can
5766 * just keep the extended permissions for the next time that an activation
5767 * of the image is tried.
5769 if (bs->open_flags & BDRV_O_INACTIVE) {
5770 bs->open_flags &= ~BDRV_O_INACTIVE;
5771 ret = bdrv_refresh_perms(bs, errp);
5773 bs->open_flags |= BDRV_O_INACTIVE;
5777 if (bs->drv->bdrv_co_invalidate_cache) {
5778 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
5780 bs->open_flags |= BDRV_O_INACTIVE;
5781 error_propagate(errp, local_err);
5786 FOR_EACH_DIRTY_BITMAP(bs, bm) {
5787 bdrv_dirty_bitmap_skip_store(bm, false);
5790 ret = refresh_total_sectors(bs, bs->total_sectors);
5792 bs->open_flags |= BDRV_O_INACTIVE;
5793 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5798 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5799 if (parent->klass->activate) {
5800 parent->klass->activate(parent, &local_err);
5802 bs->open_flags |= BDRV_O_INACTIVE;
5803 error_propagate(errp, local_err);
5812 void bdrv_invalidate_cache_all(Error **errp)
5814 BlockDriverState *bs;
5815 BdrvNextIterator it;
5817 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5818 AioContext *aio_context = bdrv_get_aio_context(bs);
5821 aio_context_acquire(aio_context);
5822 ret = bdrv_invalidate_cache(bs, errp);
5823 aio_context_release(aio_context);
5825 bdrv_next_cleanup(&it);
5831 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5835 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5836 if (parent->klass->parent_is_bds) {
5837 BlockDriverState *parent_bs = parent->opaque;
5838 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5847 static int bdrv_inactivate_recurse(BlockDriverState *bs)
5849 BdrvChild *child, *parent;
5856 /* Make sure that we don't inactivate a child before its parent.
5857 * It will be covered by recursion from the yet active parent. */
5858 if (bdrv_has_bds_parent(bs, true)) {
5862 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5864 /* Inactivate this node */
5865 if (bs->drv->bdrv_inactivate) {
5866 ret = bs->drv->bdrv_inactivate(bs);
5872 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5873 if (parent->klass->inactivate) {
5874 ret = parent->klass->inactivate(parent);
5881 bs->open_flags |= BDRV_O_INACTIVE;
5884 * Update permissions, they may differ for inactive nodes.
5885 * We only tried to loosen restrictions, so errors are not fatal, ignore
5888 bdrv_refresh_perms(bs, NULL);
5890 /* Recursively inactivate children */
5891 QLIST_FOREACH(child, &bs->children, next) {
5892 ret = bdrv_inactivate_recurse(child->bs);
5901 int bdrv_inactivate_all(void)
5903 BlockDriverState *bs = NULL;
5904 BdrvNextIterator it;
5906 GSList *aio_ctxs = NULL, *ctx;
5908 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5909 AioContext *aio_context = bdrv_get_aio_context(bs);
5911 if (!g_slist_find(aio_ctxs, aio_context)) {
5912 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5913 aio_context_acquire(aio_context);
5917 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5918 /* Nodes with BDS parents are covered by recursion from the last
5919 * parent that gets inactivated. Don't inactivate them a second
5920 * time if that has already happened. */
5921 if (bdrv_has_bds_parent(bs, false)) {
5924 ret = bdrv_inactivate_recurse(bs);
5926 bdrv_next_cleanup(&it);
5932 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5933 AioContext *aio_context = ctx->data;
5934 aio_context_release(aio_context);
5936 g_slist_free(aio_ctxs);
5941 /**************************************************************/
5942 /* removable device support */
5945 * Return TRUE if the media is present
5947 bool bdrv_is_inserted(BlockDriverState *bs)
5949 BlockDriver *drv = bs->drv;
5955 if (drv->bdrv_is_inserted) {
5956 return drv->bdrv_is_inserted(bs);
5958 QLIST_FOREACH(child, &bs->children, next) {
5959 if (!bdrv_is_inserted(child->bs)) {
5967 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5969 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5971 BlockDriver *drv = bs->drv;
5973 if (drv && drv->bdrv_eject) {
5974 drv->bdrv_eject(bs, eject_flag);
5979 * Lock or unlock the media (if it is locked, the user won't be able
5980 * to eject it manually).
5982 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5984 BlockDriver *drv = bs->drv;
5986 trace_bdrv_lock_medium(bs, locked);
5988 if (drv && drv->bdrv_lock_medium) {
5989 drv->bdrv_lock_medium(bs, locked);
5993 /* Get a reference to bs */
5994 void bdrv_ref(BlockDriverState *bs)
5999 /* Release a previously grabbed reference to bs.
6000 * If after releasing, reference count is zero, the BlockDriverState is
6002 void bdrv_unref(BlockDriverState *bs)
6007 assert(bs->refcnt > 0);
6008 if (--bs->refcnt == 0) {
6013 struct BdrvOpBlocker {
6015 QLIST_ENTRY(BdrvOpBlocker) list;
6018 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6020 BdrvOpBlocker *blocker;
6021 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6022 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6023 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6024 error_propagate_prepend(errp, error_copy(blocker->reason),
6025 "Node '%s' is busy: ",
6026 bdrv_get_device_or_node_name(bs));
6032 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6034 BdrvOpBlocker *blocker;
6035 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6037 blocker = g_new0(BdrvOpBlocker, 1);
6038 blocker->reason = reason;
6039 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6042 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6044 BdrvOpBlocker *blocker, *next;
6045 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6046 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6047 if (blocker->reason == reason) {
6048 QLIST_REMOVE(blocker, list);
6054 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6057 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6058 bdrv_op_block(bs, i, reason);
6062 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6065 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6066 bdrv_op_unblock(bs, i, reason);
6070 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6074 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6075 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6082 void bdrv_img_create(const char *filename, const char *fmt,
6083 const char *base_filename, const char *base_fmt,
6084 char *options, uint64_t img_size, int flags, bool quiet,
6087 QemuOptsList *create_opts = NULL;
6088 QemuOpts *opts = NULL;
6089 const char *backing_fmt, *backing_file;
6091 BlockDriver *drv, *proto_drv;
6092 Error *local_err = NULL;
6095 /* Find driver and parse its options */
6096 drv = bdrv_find_format(fmt);
6098 error_setg(errp, "Unknown file format '%s'", fmt);
6102 proto_drv = bdrv_find_protocol(filename, true, errp);
6107 if (!drv->create_opts) {
6108 error_setg(errp, "Format driver '%s' does not support image creation",
6113 if (!proto_drv->create_opts) {
6114 error_setg(errp, "Protocol driver '%s' does not support image creation",
6115 proto_drv->format_name);
6119 /* Create parameter list */
6120 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6121 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6123 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6125 /* Parse -o options */
6127 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6132 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6133 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6134 } else if (img_size != UINT64_C(-1)) {
6135 error_setg(errp, "The image size must be specified only once");
6139 if (base_filename) {
6140 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6142 error_setg(errp, "Backing file not supported for file format '%s'",
6149 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
6150 error_setg(errp, "Backing file format not supported for file "
6151 "format '%s'", fmt);
6156 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6158 if (!strcmp(filename, backing_file)) {
6159 error_setg(errp, "Error: Trying to create an image with the "
6160 "same filename as the backing file");
6163 if (backing_file[0] == '\0') {
6164 error_setg(errp, "Expected backing file name, got empty string");
6169 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
6171 /* The size for the image must always be specified, unless we have a backing
6172 * file and we have not been forbidden from opening it. */
6173 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6174 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6175 BlockDriverState *bs;
6178 QDict *backing_options = NULL;
6181 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6186 assert(full_backing);
6188 /* backing files always opened read-only */
6190 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
6192 backing_options = qdict_new();
6194 qdict_put_str(backing_options, "driver", backing_fmt);
6196 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
6198 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6200 g_free(full_backing);
6202 error_append_hint(&local_err, "Could not open backing image.\n");
6206 warn_report("Deprecated use of backing file without explicit "
6207 "backing format (detected format of %s)",
6208 bs->drv->format_name);
6209 if (bs->drv != &bdrv_raw) {
6211 * A probe of raw deserves the most attention:
6212 * leaving the backing format out of the image
6213 * will ensure bs->probed is set (ensuring we
6214 * don't accidentally commit into the backing
6215 * file), and allow more spots to warn the users
6216 * to fix their toolchain when opening this image
6217 * later. For other images, we can safely record
6218 * the format that we probed.
6220 backing_fmt = bs->drv->format_name;
6221 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, backing_fmt,
6226 /* Opened BS, have no size */
6227 size = bdrv_getlength(bs);
6229 error_setg_errno(errp, -size, "Could not get size of '%s'",
6234 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
6238 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6239 } else if (backing_file && !backing_fmt) {
6240 warn_report("Deprecated use of unopened backing file without "
6241 "explicit backing format, use of this image requires "
6242 "potentially unsafe format probing");
6246 error_setg(errp, "Image creation needs a size parameter");
6251 printf("Formatting '%s', fmt=%s ", filename, fmt);
6252 qemu_opts_print(opts, " ");
6257 ret = bdrv_create(drv, filename, opts, &local_err);
6259 if (ret == -EFBIG) {
6260 /* This is generally a better message than whatever the driver would
6261 * deliver (especially because of the cluster_size_hint), since that
6262 * is most probably not much different from "image too large". */
6263 const char *cluster_size_hint = "";
6264 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
6265 cluster_size_hint = " (try using a larger cluster size)";
6267 error_setg(errp, "The image size is too large for file format '%s'"
6268 "%s", fmt, cluster_size_hint);
6269 error_free(local_err);
6274 qemu_opts_del(opts);
6275 qemu_opts_free(create_opts);
6276 error_propagate(errp, local_err);
6279 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6281 return bs ? bs->aio_context : qemu_get_aio_context();
6284 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
6286 Coroutine *self = qemu_coroutine_self();
6287 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
6288 AioContext *new_ctx;
6291 * Increase bs->in_flight to ensure that this operation is completed before
6292 * moving the node to a different AioContext. Read new_ctx only afterwards.
6294 bdrv_inc_in_flight(bs);
6296 new_ctx = bdrv_get_aio_context(bs);
6297 aio_co_reschedule_self(new_ctx);
6301 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
6303 aio_co_reschedule_self(old_ctx);
6304 bdrv_dec_in_flight(bs);
6307 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
6309 AioContext *ctx = bdrv_get_aio_context(bs);
6311 /* In the main thread, bs->aio_context won't change concurrently */
6312 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6315 * We're in coroutine context, so we already hold the lock of the main
6316 * loop AioContext. Don't lock it twice to avoid deadlocks.
6318 assert(qemu_in_coroutine());
6319 if (ctx != qemu_get_aio_context()) {
6320 aio_context_acquire(ctx);
6324 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
6326 AioContext *ctx = bdrv_get_aio_context(bs);
6328 assert(qemu_in_coroutine());
6329 if (ctx != qemu_get_aio_context()) {
6330 aio_context_release(ctx);
6334 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6336 aio_co_enter(bdrv_get_aio_context(bs), co);
6339 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6341 QLIST_REMOVE(ban, list);
6345 static void bdrv_detach_aio_context(BlockDriverState *bs)
6347 BdrvAioNotifier *baf, *baf_tmp;
6349 assert(!bs->walking_aio_notifiers);
6350 bs->walking_aio_notifiers = true;
6351 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6353 bdrv_do_remove_aio_context_notifier(baf);
6355 baf->detach_aio_context(baf->opaque);
6358 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6359 * remove remaining aio notifiers if we aren't called again.
6361 bs->walking_aio_notifiers = false;
6363 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
6364 bs->drv->bdrv_detach_aio_context(bs);
6367 if (bs->quiesce_counter) {
6368 aio_enable_external(bs->aio_context);
6370 bs->aio_context = NULL;
6373 static void bdrv_attach_aio_context(BlockDriverState *bs,
6374 AioContext *new_context)
6376 BdrvAioNotifier *ban, *ban_tmp;
6378 if (bs->quiesce_counter) {
6379 aio_disable_external(new_context);
6382 bs->aio_context = new_context;
6384 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
6385 bs->drv->bdrv_attach_aio_context(bs, new_context);
6388 assert(!bs->walking_aio_notifiers);
6389 bs->walking_aio_notifiers = true;
6390 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6392 bdrv_do_remove_aio_context_notifier(ban);
6394 ban->attached_aio_context(new_context, ban->opaque);
6397 bs->walking_aio_notifiers = false;
6401 * Changes the AioContext used for fd handlers, timers, and BHs by this
6402 * BlockDriverState and all its children and parents.
6404 * Must be called from the main AioContext.
6406 * The caller must own the AioContext lock for the old AioContext of bs, but it
6407 * must not own the AioContext lock for new_context (unless new_context is the
6408 * same as the current context of bs).
6410 * @ignore will accumulate all visited BdrvChild object. The caller is
6411 * responsible for freeing the list afterwards.
6413 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6414 AioContext *new_context, GSList **ignore)
6416 AioContext *old_context = bdrv_get_aio_context(bs);
6419 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6421 if (old_context == new_context) {
6425 bdrv_drained_begin(bs);
6427 QLIST_FOREACH(child, &bs->children, next) {
6428 if (g_slist_find(*ignore, child)) {
6431 *ignore = g_slist_prepend(*ignore, child);
6432 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6434 QLIST_FOREACH(child, &bs->parents, next_parent) {
6435 if (g_slist_find(*ignore, child)) {
6438 assert(child->klass->set_aio_ctx);
6439 *ignore = g_slist_prepend(*ignore, child);
6440 child->klass->set_aio_ctx(child, new_context, ignore);
6443 bdrv_detach_aio_context(bs);
6445 /* Acquire the new context, if necessary */
6446 if (qemu_get_aio_context() != new_context) {
6447 aio_context_acquire(new_context);
6450 bdrv_attach_aio_context(bs, new_context);
6453 * If this function was recursively called from
6454 * bdrv_set_aio_context_ignore(), there may be nodes in the
6455 * subtree that have not yet been moved to the new AioContext.
6456 * Release the old one so bdrv_drained_end() can poll them.
6458 if (qemu_get_aio_context() != old_context) {
6459 aio_context_release(old_context);
6462 bdrv_drained_end(bs);
6464 if (qemu_get_aio_context() != old_context) {
6465 aio_context_acquire(old_context);
6467 if (qemu_get_aio_context() != new_context) {
6468 aio_context_release(new_context);
6472 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6473 GSList **ignore, Error **errp)
6475 if (g_slist_find(*ignore, c)) {
6478 *ignore = g_slist_prepend(*ignore, c);
6481 * A BdrvChildClass that doesn't handle AioContext changes cannot
6482 * tolerate any AioContext changes
6484 if (!c->klass->can_set_aio_ctx) {
6485 char *user = bdrv_child_user_desc(c);
6486 error_setg(errp, "Changing iothreads is not supported by %s", user);
6490 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
6491 assert(!errp || *errp);
6497 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6498 GSList **ignore, Error **errp)
6500 if (g_slist_find(*ignore, c)) {
6503 *ignore = g_slist_prepend(*ignore, c);
6504 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6507 /* @ignore will accumulate all visited BdrvChild object. The caller is
6508 * responsible for freeing the list afterwards. */
6509 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6510 GSList **ignore, Error **errp)
6514 if (bdrv_get_aio_context(bs) == ctx) {
6518 QLIST_FOREACH(c, &bs->parents, next_parent) {
6519 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
6523 QLIST_FOREACH(c, &bs->children, next) {
6524 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
6532 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6533 BdrvChild *ignore_child, Error **errp)
6538 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6539 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
6540 g_slist_free(ignore);
6546 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6547 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
6548 g_slist_free(ignore);
6553 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6556 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
6559 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
6560 void (*attached_aio_context)(AioContext *new_context, void *opaque),
6561 void (*detach_aio_context)(void *opaque), void *opaque)
6563 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
6564 *ban = (BdrvAioNotifier){
6565 .attached_aio_context = attached_aio_context,
6566 .detach_aio_context = detach_aio_context,
6570 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
6573 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
6574 void (*attached_aio_context)(AioContext *,
6576 void (*detach_aio_context)(void *),
6579 BdrvAioNotifier *ban, *ban_next;
6581 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
6582 if (ban->attached_aio_context == attached_aio_context &&
6583 ban->detach_aio_context == detach_aio_context &&
6584 ban->opaque == opaque &&
6585 ban->deleted == false)
6587 if (bs->walking_aio_notifiers) {
6588 ban->deleted = true;
6590 bdrv_do_remove_aio_context_notifier(ban);
6599 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
6600 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
6605 error_setg(errp, "Node is ejected");
6608 if (!bs->drv->bdrv_amend_options) {
6609 error_setg(errp, "Block driver '%s' does not support option amendment",
6610 bs->drv->format_name);
6613 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
6614 cb_opaque, force, errp);
6618 * This function checks whether the given @to_replace is allowed to be
6619 * replaced by a node that always shows the same data as @bs. This is
6620 * used for example to verify whether the mirror job can replace
6621 * @to_replace by the target mirrored from @bs.
6622 * To be replaceable, @bs and @to_replace may either be guaranteed to
6623 * always show the same data (because they are only connected through
6624 * filters), or some driver may allow replacing one of its children
6625 * because it can guarantee that this child's data is not visible at
6626 * all (for example, for dissenting quorum children that have no other
6629 bool bdrv_recurse_can_replace(BlockDriverState *bs,
6630 BlockDriverState *to_replace)
6632 BlockDriverState *filtered;
6634 if (!bs || !bs->drv) {
6638 if (bs == to_replace) {
6642 /* See what the driver can do */
6643 if (bs->drv->bdrv_recurse_can_replace) {
6644 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
6647 /* For filters without an own implementation, we can recurse on our own */
6648 filtered = bdrv_filter_bs(bs);
6650 return bdrv_recurse_can_replace(filtered, to_replace);
6658 * Check whether the given @node_name can be replaced by a node that
6659 * has the same data as @parent_bs. If so, return @node_name's BDS;
6662 * @node_name must be a (recursive) *child of @parent_bs (or this
6663 * function will return NULL).
6665 * The result (whether the node can be replaced or not) is only valid
6666 * for as long as no graph or permission changes occur.
6668 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
6669 const char *node_name, Error **errp)
6671 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
6672 AioContext *aio_context;
6674 if (!to_replace_bs) {
6675 error_setg(errp, "Node name '%s' not found", node_name);
6679 aio_context = bdrv_get_aio_context(to_replace_bs);
6680 aio_context_acquire(aio_context);
6682 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
6683 to_replace_bs = NULL;
6687 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6688 * most non filter in order to prevent data corruption.
6689 * Another benefit is that this tests exclude backing files which are
6690 * blocked by the backing blockers.
6692 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
6693 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
6694 "because it cannot be guaranteed that doing so would not "
6695 "lead to an abrupt change of visible data",
6696 node_name, parent_bs->node_name);
6697 to_replace_bs = NULL;
6702 aio_context_release(aio_context);
6703 return to_replace_bs;
6707 * Iterates through the list of runtime option keys that are said to
6708 * be "strong" for a BDS. An option is called "strong" if it changes
6709 * a BDS's data. For example, the null block driver's "size" and
6710 * "read-zeroes" options are strong, but its "latency-ns" option is
6713 * If a key returned by this function ends with a dot, all options
6714 * starting with that prefix are strong.
6716 static const char *const *strong_options(BlockDriverState *bs,
6717 const char *const *curopt)
6719 static const char *const global_options[] = {
6720 "driver", "filename", NULL
6724 return &global_options[0];
6728 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
6729 curopt = bs->drv->strong_runtime_opts;
6732 return (curopt && *curopt) ? curopt : NULL;
6736 * Copies all strong runtime options from bs->options to the given
6737 * QDict. The set of strong option keys is determined by invoking
6740 * Returns true iff any strong option was present in bs->options (and
6741 * thus copied to the target QDict) with the exception of "filename"
6742 * and "driver". The caller is expected to use this value to decide
6743 * whether the existence of strong options prevents the generation of
6746 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
6748 bool found_any = false;
6749 const char *const *option_name = NULL;
6755 while ((option_name = strong_options(bs, option_name))) {
6756 bool option_given = false;
6758 assert(strlen(*option_name) > 0);
6759 if ((*option_name)[strlen(*option_name) - 1] != '.') {
6760 QObject *entry = qdict_get(bs->options, *option_name);
6765 qdict_put_obj(d, *option_name, qobject_ref(entry));
6766 option_given = true;
6768 const QDictEntry *entry;
6769 for (entry = qdict_first(bs->options); entry;
6770 entry = qdict_next(bs->options, entry))
6772 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
6773 qdict_put_obj(d, qdict_entry_key(entry),
6774 qobject_ref(qdict_entry_value(entry)));
6775 option_given = true;
6780 /* While "driver" and "filename" need to be included in a JSON filename,
6781 * their existence does not prohibit generation of a plain filename. */
6782 if (!found_any && option_given &&
6783 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
6789 if (!qdict_haskey(d, "driver")) {
6790 /* Drivers created with bdrv_new_open_driver() may not have a
6791 * @driver option. Add it here. */
6792 qdict_put_str(d, "driver", bs->drv->format_name);
6798 /* Note: This function may return false positives; it may return true
6799 * even if opening the backing file specified by bs's image header
6800 * would result in exactly bs->backing. */
6801 bool bdrv_backing_overridden(BlockDriverState *bs)
6804 return strcmp(bs->auto_backing_file,
6805 bs->backing->bs->filename);
6807 /* No backing BDS, so if the image header reports any backing
6808 * file, it must have been suppressed */
6809 return bs->auto_backing_file[0] != '\0';
6813 /* Updates the following BDS fields:
6814 * - exact_filename: A filename which may be used for opening a block device
6815 * which (mostly) equals the given BDS (even without any
6816 * other options; so reading and writing must return the same
6817 * results, but caching etc. may be different)
6818 * - full_open_options: Options which, when given when opening a block device
6819 * (without a filename), result in a BDS (mostly)
6820 * equalling the given one
6821 * - filename: If exact_filename is set, it is copied here. Otherwise,
6822 * full_open_options is converted to a JSON object, prefixed with
6823 * "json:" (for use through the JSON pseudo protocol) and put here.
6825 void bdrv_refresh_filename(BlockDriverState *bs)
6827 BlockDriver *drv = bs->drv;
6829 BlockDriverState *primary_child_bs;
6831 bool backing_overridden;
6832 bool generate_json_filename; /* Whether our default implementation should
6833 fill exact_filename (false) or not (true) */
6839 /* This BDS's file name may depend on any of its children's file names, so
6840 * refresh those first */
6841 QLIST_FOREACH(child, &bs->children, next) {
6842 bdrv_refresh_filename(child->bs);
6846 /* For implicit nodes, just copy everything from the single child */
6847 child = QLIST_FIRST(&bs->children);
6848 assert(QLIST_NEXT(child, next) == NULL);
6850 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
6851 child->bs->exact_filename);
6852 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
6854 qobject_unref(bs->full_open_options);
6855 bs->full_open_options = qobject_ref(child->bs->full_open_options);
6860 backing_overridden = bdrv_backing_overridden(bs);
6862 if (bs->open_flags & BDRV_O_NO_IO) {
6863 /* Without I/O, the backing file does not change anything.
6864 * Therefore, in such a case (primarily qemu-img), we can
6865 * pretend the backing file has not been overridden even if
6866 * it technically has been. */
6867 backing_overridden = false;
6870 /* Gather the options QDict */
6872 generate_json_filename = append_strong_runtime_options(opts, bs);
6873 generate_json_filename |= backing_overridden;
6875 if (drv->bdrv_gather_child_options) {
6876 /* Some block drivers may not want to present all of their children's
6877 * options, or name them differently from BdrvChild.name */
6878 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
6880 QLIST_FOREACH(child, &bs->children, next) {
6881 if (child == bs->backing && !backing_overridden) {
6882 /* We can skip the backing BDS if it has not been overridden */
6886 qdict_put(opts, child->name,
6887 qobject_ref(child->bs->full_open_options));
6890 if (backing_overridden && !bs->backing) {
6891 /* Force no backing file */
6892 qdict_put_null(opts, "backing");
6896 qobject_unref(bs->full_open_options);
6897 bs->full_open_options = opts;
6899 primary_child_bs = bdrv_primary_bs(bs);
6901 if (drv->bdrv_refresh_filename) {
6902 /* Obsolete information is of no use here, so drop the old file name
6903 * information before refreshing it */
6904 bs->exact_filename[0] = '\0';
6906 drv->bdrv_refresh_filename(bs);
6907 } else if (primary_child_bs) {
6909 * Try to reconstruct valid information from the underlying
6910 * file -- this only works for format nodes (filter nodes
6911 * cannot be probed and as such must be selected by the user
6912 * either through an options dict, or through a special
6913 * filename which the filter driver must construct in its
6914 * .bdrv_refresh_filename() implementation).
6917 bs->exact_filename[0] = '\0';
6920 * We can use the underlying file's filename if:
6921 * - it has a filename,
6922 * - the current BDS is not a filter,
6923 * - the file is a protocol BDS, and
6924 * - opening that file (as this BDS's format) will automatically create
6925 * the BDS tree we have right now, that is:
6926 * - the user did not significantly change this BDS's behavior with
6927 * some explicit (strong) options
6928 * - no non-file child of this BDS has been overridden by the user
6929 * Both of these conditions are represented by generate_json_filename.
6931 if (primary_child_bs->exact_filename[0] &&
6932 primary_child_bs->drv->bdrv_file_open &&
6933 !drv->is_filter && !generate_json_filename)
6935 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
6939 if (bs->exact_filename[0]) {
6940 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
6942 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
6943 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
6944 json->str) >= sizeof(bs->filename)) {
6945 /* Give user a hint if we truncated things. */
6946 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
6948 g_string_free(json, true);
6952 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
6954 BlockDriver *drv = bs->drv;
6955 BlockDriverState *child_bs;
6958 error_setg(errp, "Node '%s' is ejected", bs->node_name);
6962 if (drv->bdrv_dirname) {
6963 return drv->bdrv_dirname(bs, errp);
6966 child_bs = bdrv_primary_bs(bs);
6968 return bdrv_dirname(child_bs, errp);
6971 bdrv_refresh_filename(bs);
6972 if (bs->exact_filename[0] != '\0') {
6973 return path_combine(bs->exact_filename, "");
6976 error_setg(errp, "Cannot generate a base directory for %s nodes",
6982 * Hot add/remove a BDS's child. So the user can take a child offline when
6983 * it is broken and take a new child online
6985 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
6989 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
6990 error_setg(errp, "The node %s does not support adding a child",
6991 bdrv_get_device_or_node_name(parent_bs));
6995 if (!QLIST_EMPTY(&child_bs->parents)) {
6996 error_setg(errp, "The node %s already has a parent",
6997 child_bs->node_name);
7001 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7004 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7008 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7009 error_setg(errp, "The node %s does not support removing a child",
7010 bdrv_get_device_or_node_name(parent_bs));
7014 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7021 error_setg(errp, "The node %s does not have a child named %s",
7022 bdrv_get_device_or_node_name(parent_bs),
7023 bdrv_get_device_or_node_name(child->bs));
7027 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7030 int bdrv_make_empty(BdrvChild *c, Error **errp)
7032 BlockDriver *drv = c->bs->drv;
7035 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7037 if (!drv->bdrv_make_empty) {
7038 error_setg(errp, "%s does not support emptying nodes",
7043 ret = drv->bdrv_make_empty(c->bs);
7045 error_setg_errno(errp, -ret, "Failed to empty %s",
7054 * Return the child that @bs acts as an overlay for, and from which data may be
7055 * copied in COW or COR operations. Usually this is the backing file.
7057 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7059 if (!bs || !bs->drv) {
7063 if (bs->drv->is_filter) {
7071 assert(bs->backing->role & BDRV_CHILD_COW);
7076 * If @bs acts as a filter for exactly one of its children, return
7079 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7083 if (!bs || !bs->drv) {
7087 if (!bs->drv->is_filter) {
7091 /* Only one of @backing or @file may be used */
7092 assert(!(bs->backing && bs->file));
7094 c = bs->backing ?: bs->file;
7099 assert(c->role & BDRV_CHILD_FILTERED);
7104 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7105 * whichever is non-NULL.
7107 * Return NULL if both are NULL.
7109 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7111 BdrvChild *cow_child = bdrv_cow_child(bs);
7112 BdrvChild *filter_child = bdrv_filter_child(bs);
7114 /* Filter nodes cannot have COW backing files */
7115 assert(!(cow_child && filter_child));
7117 return cow_child ?: filter_child;
7121 * Return the primary child of this node: For filters, that is the
7122 * filtered child. For other nodes, that is usually the child storing
7124 * (A generally more helpful description is that this is (usually) the
7125 * child that has the same filename as @bs.)
7127 * Drivers do not necessarily have a primary child; for example quorum
7130 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
7132 BdrvChild *c, *found = NULL;
7134 QLIST_FOREACH(c, &bs->children, next) {
7135 if (c->role & BDRV_CHILD_PRIMARY) {
7144 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
7145 bool stop_on_explicit_filter)
7153 while (!(stop_on_explicit_filter && !bs->implicit)) {
7154 c = bdrv_filter_child(bs);
7157 * A filter that is embedded in a working block graph must
7158 * have a child. Assert this here so this function does
7159 * not return a filter node that is not expected by the
7162 assert(!bs->drv || !bs->drv->is_filter);
7168 * Note that this treats nodes with bs->drv == NULL as not being
7169 * filters (bs->drv == NULL should be replaced by something else
7171 * The advantage of this behavior is that this function will thus
7172 * always return a non-NULL value (given a non-NULL @bs).
7179 * Return the first BDS that has not been added implicitly or that
7180 * does not have a filtered child down the chain starting from @bs
7181 * (including @bs itself).
7183 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
7185 return bdrv_do_skip_filters(bs, true);
7189 * Return the first BDS that does not have a filtered child down the
7190 * chain starting from @bs (including @bs itself).
7192 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
7194 return bdrv_do_skip_filters(bs, false);
7198 * For a backing chain, return the first non-filter backing image of
7199 * the first non-filter image.
7201 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
7203 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));