2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
53 #include "qemu/range.h"
55 #include "block/coroutines.h"
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
81 QLIST_HEAD_INITIALIZER(bdrv_drivers);
83 static BlockDriverState *bdrv_open_inherit(const char *filename,
84 const char *reference,
85 QDict *options, int flags,
86 BlockDriverState *parent,
87 const BdrvChildClass *child_class,
88 BdrvChildRole child_role,
91 static bool bdrv_recurse_has_child(BlockDriverState *bs,
92 BlockDriverState *child);
94 static void GRAPH_WRLOCK
95 bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs);
97 static void GRAPH_WRLOCK
98 bdrv_remove_child(BdrvChild *child, Transaction *tran);
100 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
101 BlockReopenQueue *queue,
102 Transaction *change_child_tran, Error **errp);
103 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
104 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
106 static bool bdrv_backing_overridden(BlockDriverState *bs);
108 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
109 GHashTable *visited, Transaction *tran,
112 /* If non-zero, use only whitelisted block drivers */
113 static int use_bdrv_whitelist;
116 static int is_windows_drive_prefix(const char *filename)
118 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
119 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
123 int is_windows_drive(const char *filename)
125 if (is_windows_drive_prefix(filename) &&
128 if (strstart(filename, "\\\\.\\", NULL) ||
129 strstart(filename, "//./", NULL))
135 size_t bdrv_opt_mem_align(BlockDriverState *bs)
137 if (!bs || !bs->drv) {
138 /* page size or 4k (hdd sector size) should be on the safe side */
139 return MAX(4096, qemu_real_host_page_size());
143 return bs->bl.opt_mem_alignment;
146 size_t bdrv_min_mem_align(BlockDriverState *bs)
148 if (!bs || !bs->drv) {
149 /* page size or 4k (hdd sector size) should be on the safe side */
150 return MAX(4096, qemu_real_host_page_size());
154 return bs->bl.min_mem_alignment;
157 /* check if the path starts with "<protocol>:" */
158 int path_has_protocol(const char *path)
163 if (is_windows_drive(path) ||
164 is_windows_drive_prefix(path)) {
167 p = path + strcspn(path, ":/\\");
169 p = path + strcspn(path, ":/");
175 int path_is_absolute(const char *path)
178 /* specific case for names like: "\\.\d:" */
179 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
182 return (*path == '/' || *path == '\\');
184 return (*path == '/');
188 /* if filename is absolute, just return its duplicate. Otherwise, build a
189 path to it by considering it is relative to base_path. URL are
191 char *path_combine(const char *base_path, const char *filename)
193 const char *protocol_stripped = NULL;
198 if (path_is_absolute(filename)) {
199 return g_strdup(filename);
202 if (path_has_protocol(base_path)) {
203 protocol_stripped = strchr(base_path, ':');
204 if (protocol_stripped) {
208 p = protocol_stripped ?: base_path;
210 p1 = strrchr(base_path, '/');
214 p2 = strrchr(base_path, '\\');
215 if (!p1 || p2 > p1) {
230 result = g_malloc(len + strlen(filename) + 1);
231 memcpy(result, base_path, len);
232 strcpy(result + len, filename);
238 * Helper function for bdrv_parse_filename() implementations to remove optional
239 * protocol prefixes (especially "file:") from a filename and for putting the
240 * stripped filename into the options QDict if there is such a prefix.
242 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
245 if (strstart(filename, prefix, &filename)) {
246 /* Stripping the explicit protocol prefix may result in a protocol
247 * prefix being (wrongly) detected (if the filename contains a colon) */
248 if (path_has_protocol(filename)) {
249 GString *fat_filename;
251 /* This means there is some colon before the first slash; therefore,
252 * this cannot be an absolute path */
253 assert(!path_is_absolute(filename));
255 /* And we can thus fix the protocol detection issue by prefixing it
257 fat_filename = g_string_new("./");
258 g_string_append(fat_filename, filename);
260 assert(!path_has_protocol(fat_filename->str));
262 qdict_put(options, "filename",
263 qstring_from_gstring(fat_filename));
265 /* If no protocol prefix was detected, we can use the shortened
267 qdict_put_str(options, "filename", filename);
273 /* Returns whether the image file is opened as read-only. Note that this can
274 * return false and writing to the image file is still not possible because the
275 * image is inactivated. */
276 bool bdrv_is_read_only(BlockDriverState *bs)
279 return !(bs->open_flags & BDRV_O_RDWR);
282 static int GRAPH_RDLOCK
283 bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
284 bool ignore_allow_rdw, Error **errp)
288 /* Do not set read_only if copy_on_read is enabled */
289 if (bs->copy_on_read && read_only) {
290 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
291 bdrv_get_device_or_node_name(bs));
295 /* Do not clear read_only if it is prohibited */
296 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
299 error_setg(errp, "Node '%s' is read only",
300 bdrv_get_device_or_node_name(bs));
308 * Called by a driver that can only provide a read-only image.
310 * Returns 0 if the node is already read-only or it could switch the node to
311 * read-only because BDRV_O_AUTO_RDONLY is set.
313 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
314 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
315 * is not NULL, it is used as the error message for the Error object.
317 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
323 if (!(bs->open_flags & BDRV_O_RDWR)) {
326 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
330 ret = bdrv_can_set_read_only(bs, true, false, NULL);
335 bs->open_flags &= ~BDRV_O_RDWR;
340 error_setg(errp, "%s", errmsg ?: "Image is read-only");
345 * If @backing is empty, this function returns NULL without setting
346 * @errp. In all other cases, NULL will only be returned with @errp
349 * Therefore, a return value of NULL without @errp set means that
350 * there is no backing file; if @errp is set, there is one but its
351 * absolute filename cannot be generated.
353 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
357 if (backing[0] == '\0') {
359 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
360 return g_strdup(backing);
361 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
362 error_setg(errp, "Cannot use relative backing file names for '%s'",
366 return path_combine(backed, backing);
371 * If @filename is empty or NULL, this function returns NULL without
372 * setting @errp. In all other cases, NULL will only be returned with
375 static char * GRAPH_RDLOCK
376 bdrv_make_absolute_filename(BlockDriverState *relative_to,
377 const char *filename, Error **errp)
379 char *dir, *full_name;
381 if (!filename || filename[0] == '\0') {
383 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
384 return g_strdup(filename);
387 dir = bdrv_dirname(relative_to, errp);
392 full_name = g_strconcat(dir, filename, NULL);
397 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
400 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
403 void bdrv_register(BlockDriver *bdrv)
405 assert(bdrv->format_name);
407 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
410 BlockDriverState *bdrv_new(void)
412 BlockDriverState *bs;
417 bs = g_new0(BlockDriverState, 1);
418 QLIST_INIT(&bs->dirty_bitmaps);
419 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
420 QLIST_INIT(&bs->op_blockers[i]);
422 qemu_mutex_init(&bs->reqs_lock);
423 qemu_mutex_init(&bs->dirty_bitmap_mutex);
425 bs->aio_context = qemu_get_aio_context();
427 qemu_co_queue_init(&bs->flush_queue);
429 qemu_co_mutex_init(&bs->bsc_modify_lock);
430 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
432 for (i = 0; i < bdrv_drain_all_count; i++) {
433 bdrv_drained_begin(bs);
436 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
441 static BlockDriver *bdrv_do_find_format(const char *format_name)
446 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
447 if (!strcmp(drv1->format_name, format_name)) {
455 BlockDriver *bdrv_find_format(const char *format_name)
462 drv1 = bdrv_do_find_format(format_name);
467 /* The driver isn't registered, maybe we need to load a module */
468 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
469 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
470 Error *local_err = NULL;
471 int rv = block_module_load(block_driver_modules[i].library_name,
474 return bdrv_do_find_format(format_name);
476 error_report_err(local_err);
484 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
486 static const char *whitelist_rw[] = {
487 CONFIG_BDRV_RW_WHITELIST
490 static const char *whitelist_ro[] = {
491 CONFIG_BDRV_RO_WHITELIST
496 if (!whitelist_rw[0] && !whitelist_ro[0]) {
497 return 1; /* no whitelist, anything goes */
500 for (p = whitelist_rw; *p; p++) {
501 if (!strcmp(format_name, *p)) {
506 for (p = whitelist_ro; *p; p++) {
507 if (!strcmp(format_name, *p)) {
515 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
518 return bdrv_format_is_whitelisted(drv->format_name, read_only);
521 bool bdrv_uses_whitelist(void)
523 return use_bdrv_whitelist;
526 typedef struct CreateCo {
534 int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
535 QemuOpts *opts, Error **errp)
541 if (!drv->bdrv_co_create_opts) {
542 error_setg(errp, "Driver '%s' does not support image creation",
547 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
548 if (ret < 0 && !*errp) {
549 error_setg_errno(errp, -ret, "Could not create image");
556 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
557 * least the given @minimum_size.
559 * On success, return @blk's actual length.
560 * Otherwise, return -errno.
562 static int64_t coroutine_fn GRAPH_UNLOCKED
563 create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
566 Error *local_err = NULL;
572 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
574 if (ret < 0 && ret != -ENOTSUP) {
575 error_propagate(errp, local_err);
579 size = blk_co_getlength(blk);
581 error_free(local_err);
582 error_setg_errno(errp, -size,
583 "Failed to inquire the new image file's length");
587 if (size < minimum_size) {
588 /* Need to grow the image, but we failed to do that */
589 error_propagate(errp, local_err);
593 error_free(local_err);
600 * Helper function for bdrv_create_file_fallback(): Zero the first
601 * sector to remove any potentially pre-existing image header.
603 static int coroutine_fn
604 create_file_fallback_zero_first_sector(BlockBackend *blk,
605 int64_t current_size,
608 int64_t bytes_to_clear;
613 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
614 if (bytes_to_clear) {
615 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
617 error_setg_errno(errp, -ret,
618 "Failed to clear the new image's first sector");
627 * Simple implementation of bdrv_co_create_opts for protocol drivers
628 * which only support creation via opening a file
629 * (usually existing raw storage device)
631 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
632 const char *filename,
640 PreallocMode prealloc;
641 Error *local_err = NULL;
646 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
647 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
648 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
649 PREALLOC_MODE_OFF, &local_err);
652 error_propagate(errp, local_err);
656 if (prealloc != PREALLOC_MODE_OFF) {
657 error_setg(errp, "Unsupported preallocation mode '%s'",
658 PreallocMode_str(prealloc));
662 options = qdict_new();
663 qdict_put_str(options, "driver", drv->format_name);
665 blk = blk_co_new_open(filename, NULL, options,
666 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
668 error_prepend(errp, "Protocol driver '%s' does not support creating "
669 "new images, so an existing image must be selected as "
670 "the target; however, opening the given target as an "
671 "existing image failed: ",
676 size = create_file_fallback_truncate(blk, size, errp);
682 ret = create_file_fallback_zero_first_sector(blk, size, errp);
693 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
696 QemuOpts *protocol_opts;
703 drv = bdrv_find_protocol(filename, true, errp);
708 if (!drv->create_opts) {
709 error_setg(errp, "Driver '%s' does not support image creation",
715 * 'opts' contains a QemuOptsList with a combination of format and protocol
718 * The format properly removes its options, but the default values remain
719 * in 'opts->list'. So if the protocol has options with the same name
720 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
721 * of the format, since for overlapping options, the format wins.
723 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
724 * only the set options, and then convert it back to QemuOpts, using the
725 * create_opts of the protocol. So the new QemuOpts, will contain only the
728 qdict = qemu_opts_to_qdict(opts, NULL);
729 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
730 if (protocol_opts == NULL) {
735 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
737 qemu_opts_del(protocol_opts);
738 qobject_unref(qdict);
742 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
744 Error *local_err = NULL;
749 assert_bdrv_graph_readable();
752 error_setg(errp, "Block node '%s' is not opened", bs->filename);
756 if (!bs->drv->bdrv_co_delete_file) {
757 error_setg(errp, "Driver '%s' does not support image deletion",
758 bs->drv->format_name);
762 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
764 error_propagate(errp, local_err);
770 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
772 Error *local_err = NULL;
780 ret = bdrv_co_delete_file(bs, &local_err);
782 * ENOTSUP will happen if the block driver doesn't support
783 * the 'bdrv_co_delete_file' interface. This is a predictable
784 * scenario and shouldn't be reported back to the user.
786 if (ret == -ENOTSUP) {
787 error_free(local_err);
788 } else if (ret < 0) {
789 error_report_err(local_err);
794 * Try to get @bs's logical and physical block size.
795 * On success, store them in @bsz struct and return 0.
796 * On failure return -errno.
797 * @bs must not be empty.
799 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
801 BlockDriver *drv = bs->drv;
802 BlockDriverState *filtered = bdrv_filter_bs(bs);
805 if (drv && drv->bdrv_probe_blocksizes) {
806 return drv->bdrv_probe_blocksizes(bs, bsz);
807 } else if (filtered) {
808 return bdrv_probe_blocksizes(filtered, bsz);
815 * Try to get @bs's geometry (cyls, heads, sectors).
816 * On success, store them in @geo struct and return 0.
817 * On failure return -errno.
818 * @bs must not be empty.
820 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
822 BlockDriver *drv = bs->drv;
823 BlockDriverState *filtered;
826 GRAPH_RDLOCK_GUARD_MAINLOOP();
828 if (drv && drv->bdrv_probe_geometry) {
829 return drv->bdrv_probe_geometry(bs, geo);
832 filtered = bdrv_filter_bs(bs);
834 return bdrv_probe_geometry(filtered, geo);
841 * Create a uniquely-named empty temporary file.
842 * Return the actual file name used upon success, otherwise NULL.
843 * This string should be freed with g_free() when not needed any longer.
845 * Note: creating a temporary file for the caller to (re)open is
846 * inherently racy. Use g_file_open_tmp() instead whenever practical.
848 char *create_tmp_file(Error **errp)
852 g_autofree char *filename = NULL;
854 tmpdir = g_get_tmp_dir();
857 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
859 * This function is used to create temporary disk images (like -snapshot),
860 * so the files can become very large. /tmp is often a tmpfs where as
861 * /var/tmp is usually on a disk, so more appropriate for disk images.
863 if (!g_strcmp0(tmpdir, "/tmp")) {
868 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
869 fd = g_mkstemp(filename);
871 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
877 return g_steal_pointer(&filename);
881 * Detect host devices. By convention, /dev/cdrom[N] is always
882 * recognized as a host CDROM.
884 static BlockDriver *find_hdev_driver(const char *filename)
886 int score_max = 0, score;
887 BlockDriver *drv = NULL, *d;
890 QLIST_FOREACH(d, &bdrv_drivers, list) {
891 if (d->bdrv_probe_device) {
892 score = d->bdrv_probe_device(filename);
893 if (score > score_max) {
903 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
908 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
909 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
917 BlockDriver *bdrv_find_protocol(const char *filename,
918 bool allow_protocol_prefix,
928 /* TODO Drivers without bdrv_file_open must be specified explicitly */
931 * XXX(hch): we really should not let host device detection
932 * override an explicit protocol specification, but moving this
933 * later breaks access to device names with colons in them.
934 * Thanks to the brain-dead persistent naming schemes on udev-
935 * based Linux systems those actually are quite common.
937 drv1 = find_hdev_driver(filename);
942 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
946 p = strchr(filename, ':');
949 if (len > sizeof(protocol) - 1)
950 len = sizeof(protocol) - 1;
951 memcpy(protocol, filename, len);
952 protocol[len] = '\0';
954 drv1 = bdrv_do_find_protocol(protocol);
959 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
960 if (block_driver_modules[i].protocol_name &&
961 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
962 int rv = block_module_load(block_driver_modules[i].library_name, errp);
964 drv1 = bdrv_do_find_protocol(protocol);
973 error_setg(errp, "Unknown protocol '%s'", protocol);
979 * Guess image format by probing its contents.
980 * This is not a good idea when your image is raw (CVE-2008-2004), but
981 * we do it anyway for backward compatibility.
983 * @buf contains the image's first @buf_size bytes.
984 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
985 * but can be smaller if the image file is smaller)
986 * @filename is its filename.
988 * For all block drivers, call the bdrv_probe() method to get its
990 * Return the first block driver with the highest probing score.
992 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
993 const char *filename)
995 int score_max = 0, score;
996 BlockDriver *drv = NULL, *d;
999 QLIST_FOREACH(d, &bdrv_drivers, list) {
1000 if (d->bdrv_probe) {
1001 score = d->bdrv_probe(buf, buf_size, filename);
1002 if (score > score_max) {
1012 static int find_image_format(BlockBackend *file, const char *filename,
1013 BlockDriver **pdrv, Error **errp)
1016 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1019 GLOBAL_STATE_CODE();
1021 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1022 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1027 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1029 error_setg_errno(errp, -ret, "Could not read image for determining its "
1035 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1037 error_setg(errp, "Could not determine image format: No compatible "
1048 * Set the current 'total_sectors' value
1049 * Return 0 on success, -errno on error.
1051 int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1054 BlockDriver *drv = bs->drv;
1056 assert_bdrv_graph_readable();
1062 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1066 /* query actual device if possible, otherwise just trust the hint */
1067 if (drv->bdrv_co_getlength) {
1068 int64_t length = drv->bdrv_co_getlength(bs);
1072 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1075 bs->total_sectors = hint;
1077 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1085 * Combines a QDict of new block driver @options with any missing options taken
1086 * from @old_options, so that leaving out an option defaults to its old value.
1088 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1091 GLOBAL_STATE_CODE();
1092 if (bs->drv && bs->drv->bdrv_join_options) {
1093 bs->drv->bdrv_join_options(options, old_options);
1095 qdict_join(options, old_options, false);
1099 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1103 Error *local_err = NULL;
1104 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1105 BlockdevDetectZeroesOptions detect_zeroes =
1106 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1107 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1108 GLOBAL_STATE_CODE();
1111 error_propagate(errp, local_err);
1112 return detect_zeroes;
1115 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1116 !(open_flags & BDRV_O_UNMAP))
1118 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1119 "without setting discard operation to unmap");
1122 return detect_zeroes;
1126 * Set open flags for aio engine
1128 * Return 0 on success, -1 if the engine specified is invalid
1130 int bdrv_parse_aio(const char *mode, int *flags)
1132 if (!strcmp(mode, "threads")) {
1133 /* do nothing, default */
1134 } else if (!strcmp(mode, "native")) {
1135 *flags |= BDRV_O_NATIVE_AIO;
1136 #ifdef CONFIG_LINUX_IO_URING
1137 } else if (!strcmp(mode, "io_uring")) {
1138 *flags |= BDRV_O_IO_URING;
1148 * Set open flags for a given discard mode
1150 * Return 0 on success, -1 if the discard mode was invalid.
1152 int bdrv_parse_discard_flags(const char *mode, int *flags)
1154 *flags &= ~BDRV_O_UNMAP;
1156 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1158 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1159 *flags |= BDRV_O_UNMAP;
1168 * Set open flags for a given cache mode
1170 * Return 0 on success, -1 if the cache mode was invalid.
1172 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1174 *flags &= ~BDRV_O_CACHE_MASK;
1176 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1177 *writethrough = false;
1178 *flags |= BDRV_O_NOCACHE;
1179 } else if (!strcmp(mode, "directsync")) {
1180 *writethrough = true;
1181 *flags |= BDRV_O_NOCACHE;
1182 } else if (!strcmp(mode, "writeback")) {
1183 *writethrough = false;
1184 } else if (!strcmp(mode, "unsafe")) {
1185 *writethrough = false;
1186 *flags |= BDRV_O_NO_FLUSH;
1187 } else if (!strcmp(mode, "writethrough")) {
1188 *writethrough = true;
1196 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1198 BlockDriverState *parent = c->opaque;
1199 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1202 static void GRAPH_RDLOCK bdrv_child_cb_drained_begin(BdrvChild *child)
1204 BlockDriverState *bs = child->opaque;
1205 bdrv_do_drained_begin_quiesce(bs, NULL);
1208 static bool GRAPH_RDLOCK bdrv_child_cb_drained_poll(BdrvChild *child)
1210 BlockDriverState *bs = child->opaque;
1211 return bdrv_drain_poll(bs, NULL, false);
1214 static void GRAPH_RDLOCK bdrv_child_cb_drained_end(BdrvChild *child)
1216 BlockDriverState *bs = child->opaque;
1217 bdrv_drained_end(bs);
1220 static int bdrv_child_cb_inactivate(BdrvChild *child)
1222 BlockDriverState *bs = child->opaque;
1223 GLOBAL_STATE_CODE();
1224 assert(bs->open_flags & BDRV_O_INACTIVE);
1228 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1229 GHashTable *visited, Transaction *tran,
1232 BlockDriverState *bs = child->opaque;
1233 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1237 * Returns the options and flags that a temporary snapshot should get, based on
1238 * the originally requested flags (the originally requested image will have
1239 * flags like a backing file)
1241 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1242 int parent_flags, QDict *parent_options)
1244 GLOBAL_STATE_CODE();
1245 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1247 /* For temporary files, unconditional cache=unsafe is fine */
1248 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1249 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1251 /* Copy the read-only and discard options from the parent */
1252 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1253 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1255 /* aio=native doesn't work for cache.direct=off, so disable it for the
1256 * temporary snapshot */
1257 *child_flags &= ~BDRV_O_NATIVE_AIO;
1260 static void GRAPH_WRLOCK bdrv_backing_attach(BdrvChild *c)
1262 BlockDriverState *parent = c->opaque;
1263 BlockDriverState *backing_hd = c->bs;
1265 GLOBAL_STATE_CODE();
1266 assert(!parent->backing_blocker);
1267 error_setg(&parent->backing_blocker,
1268 "node is used as backing hd of '%s'",
1269 bdrv_get_device_or_node_name(parent));
1271 bdrv_refresh_filename(backing_hd);
1273 parent->open_flags &= ~BDRV_O_NO_BACKING;
1275 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1276 /* Otherwise we won't be able to commit or stream */
1277 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1278 parent->backing_blocker);
1279 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1280 parent->backing_blocker);
1282 * We do backup in 3 ways:
1284 * The target bs is new opened, and the source is top BDS
1285 * 2. blockdev backup
1286 * Both the source and the target are top BDSes.
1287 * 3. internal backup(used for block replication)
1288 * Both the source and the target are backing file
1290 * In case 1 and 2, neither the source nor the target is the backing file.
1291 * In case 3, we will block the top BDS, so there is only one block job
1292 * for the top BDS and its backing chain.
1294 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1295 parent->backing_blocker);
1296 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1297 parent->backing_blocker);
1300 static void bdrv_backing_detach(BdrvChild *c)
1302 BlockDriverState *parent = c->opaque;
1304 GLOBAL_STATE_CODE();
1305 assert(parent->backing_blocker);
1306 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1307 error_free(parent->backing_blocker);
1308 parent->backing_blocker = NULL;
1311 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1312 const char *filename, Error **errp)
1314 BlockDriverState *parent = c->opaque;
1315 bool read_only = bdrv_is_read_only(parent);
1317 GLOBAL_STATE_CODE();
1320 ret = bdrv_reopen_set_read_only(parent, false, errp);
1326 ret = bdrv_change_backing_file(parent, filename,
1327 base->drv ? base->drv->format_name : "",
1330 error_setg_errno(errp, -ret, "Could not update backing file link");
1334 bdrv_reopen_set_read_only(parent, true, NULL);
1341 * Returns the options and flags that a generic child of a BDS should
1342 * get, based on the given options and flags for the parent BDS.
1344 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1345 int *child_flags, QDict *child_options,
1346 int parent_flags, QDict *parent_options)
1348 int flags = parent_flags;
1349 GLOBAL_STATE_CODE();
1352 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1353 * Generally, the question to answer is: Should this child be
1354 * format-probed by default?
1358 * Pure and non-filtered data children of non-format nodes should
1359 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1360 * set). This only affects a very limited set of drivers (namely
1361 * quorum and blkverify when this comment was written).
1362 * Force-clear BDRV_O_PROTOCOL then.
1364 if (!parent_is_format &&
1365 (role & BDRV_CHILD_DATA) &&
1366 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1368 flags &= ~BDRV_O_PROTOCOL;
1372 * All children of format nodes (except for COW children) and all
1373 * metadata children in general should never be format-probed.
1374 * Force-set BDRV_O_PROTOCOL then.
1376 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1377 (role & BDRV_CHILD_METADATA))
1379 flags |= BDRV_O_PROTOCOL;
1383 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1386 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1387 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1388 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1390 if (role & BDRV_CHILD_COW) {
1391 /* backing files are opened read-only by default */
1392 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1393 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1395 /* Inherit the read-only option from the parent if it's not set */
1396 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1397 qdict_copy_default(child_options, parent_options,
1398 BDRV_OPT_AUTO_READ_ONLY);
1402 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1403 * can default to enable it on lower layers regardless of the
1406 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1408 /* Clear flags that only apply to the top layer */
1409 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1411 if (role & BDRV_CHILD_METADATA) {
1412 flags &= ~BDRV_O_NO_IO;
1414 if (role & BDRV_CHILD_COW) {
1415 flags &= ~BDRV_O_TEMPORARY;
1418 *child_flags = flags;
1421 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1423 BlockDriverState *bs = child->opaque;
1425 assert_bdrv_graph_writable();
1426 QLIST_INSERT_HEAD(&bs->children, child, next);
1427 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1429 * Here we handle filters and block/raw-format.c when it behave like
1430 * filter. They generally have a single PRIMARY child, which is also the
1431 * FILTERED child, and that they may have multiple more children, which
1432 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1433 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1434 * into bs->backing on exceptional cases; and bs->backing will be
1437 assert(!(child->role & BDRV_CHILD_COW));
1438 if (child->role & BDRV_CHILD_PRIMARY) {
1439 assert(child->role & BDRV_CHILD_FILTERED);
1440 assert(!bs->backing);
1443 if (bs->drv->filtered_child_is_backing) {
1444 bs->backing = child;
1449 assert(!(child->role & BDRV_CHILD_FILTERED));
1451 } else if (child->role & BDRV_CHILD_COW) {
1452 assert(bs->drv->supports_backing);
1453 assert(!(child->role & BDRV_CHILD_PRIMARY));
1454 assert(!bs->backing);
1455 bs->backing = child;
1456 bdrv_backing_attach(child);
1457 } else if (child->role & BDRV_CHILD_PRIMARY) {
1463 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1465 BlockDriverState *bs = child->opaque;
1467 if (child->role & BDRV_CHILD_COW) {
1468 bdrv_backing_detach(child);
1471 assert_bdrv_graph_writable();
1472 QLIST_REMOVE(child, next);
1473 if (child == bs->backing) {
1474 assert(child != bs->file);
1476 } else if (child == bs->file) {
1481 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1482 const char *filename, Error **errp)
1484 if (c->role & BDRV_CHILD_COW) {
1485 return bdrv_backing_update_filename(c, base, filename, errp);
1490 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1492 BlockDriverState *bs = c->opaque;
1495 return bdrv_get_aio_context(bs);
1498 const BdrvChildClass child_of_bds = {
1499 .parent_is_bds = true,
1500 .get_parent_desc = bdrv_child_get_parent_desc,
1501 .inherit_options = bdrv_inherited_options,
1502 .drained_begin = bdrv_child_cb_drained_begin,
1503 .drained_poll = bdrv_child_cb_drained_poll,
1504 .drained_end = bdrv_child_cb_drained_end,
1505 .attach = bdrv_child_cb_attach,
1506 .detach = bdrv_child_cb_detach,
1507 .inactivate = bdrv_child_cb_inactivate,
1508 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1509 .update_filename = bdrv_child_cb_update_filename,
1510 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1513 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1516 return c->klass->get_parent_aio_context(c);
1519 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1521 int open_flags = flags;
1522 GLOBAL_STATE_CODE();
1525 * Clear flags that are internal to the block layer before opening the
1528 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1533 static void update_flags_from_options(int *flags, QemuOpts *opts)
1535 GLOBAL_STATE_CODE();
1537 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1539 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1540 *flags |= BDRV_O_NO_FLUSH;
1543 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1544 *flags |= BDRV_O_NOCACHE;
1547 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1548 *flags |= BDRV_O_RDWR;
1551 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1552 *flags |= BDRV_O_AUTO_RDONLY;
1556 static void update_options_from_flags(QDict *options, int flags)
1558 GLOBAL_STATE_CODE();
1559 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1560 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1562 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1563 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1564 flags & BDRV_O_NO_FLUSH);
1566 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1567 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1569 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1570 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1571 flags & BDRV_O_AUTO_RDONLY);
1575 static void bdrv_assign_node_name(BlockDriverState *bs,
1576 const char *node_name,
1579 char *gen_node_name = NULL;
1580 GLOBAL_STATE_CODE();
1583 node_name = gen_node_name = id_generate(ID_BLOCK);
1584 } else if (!id_wellformed(node_name)) {
1586 * Check for empty string or invalid characters, but not if it is
1587 * generated (generated names use characters not available to the user)
1589 error_setg(errp, "Invalid node-name: '%s'", node_name);
1593 /* takes care of avoiding namespaces collisions */
1594 if (blk_by_name(node_name)) {
1595 error_setg(errp, "node-name=%s is conflicting with a device id",
1600 /* takes care of avoiding duplicates node names */
1601 if (bdrv_find_node(node_name)) {
1602 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1606 /* Make sure that the node name isn't truncated */
1607 if (strlen(node_name) >= sizeof(bs->node_name)) {
1608 error_setg(errp, "Node name too long");
1612 /* copy node name into the bs and insert it into the graph list */
1613 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1614 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1616 g_free(gen_node_name);
1620 * The caller must always hold @bs AioContext lock, because this function calls
1621 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1624 static int no_coroutine_fn GRAPH_UNLOCKED
1625 bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1626 QDict *options, int open_flags, Error **errp)
1629 Error *local_err = NULL;
1631 GLOBAL_STATE_CODE();
1633 bdrv_assign_node_name(bs, node_name, &local_err);
1635 error_propagate(errp, local_err);
1640 bs->opaque = g_malloc0(drv->instance_size);
1642 if (drv->bdrv_file_open) {
1643 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1644 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1645 } else if (drv->bdrv_open) {
1646 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1653 error_propagate(errp, local_err);
1654 } else if (bs->filename[0]) {
1655 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1657 error_setg_errno(errp, -ret, "Could not open image");
1662 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1663 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1666 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1667 * drivers that pass read/write requests through to a child the trouble of
1668 * declaring support explicitly.
1670 * Drivers must not propagate this flag accidentally when they initiate I/O
1671 * to a bounce buffer. That case should be rare though.
1673 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1674 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1676 /* Get the context after .bdrv_open, it can change the context */
1677 ctx = bdrv_get_aio_context(bs);
1678 aio_context_acquire(ctx);
1680 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
1682 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1683 aio_context_release(ctx);
1687 bdrv_graph_rdlock_main_loop();
1688 bdrv_refresh_limits(bs, NULL, &local_err);
1689 bdrv_graph_rdunlock_main_loop();
1690 aio_context_release(ctx);
1693 error_propagate(errp, local_err);
1697 assert(bdrv_opt_mem_align(bs) != 0);
1698 assert(bdrv_min_mem_align(bs) != 0);
1699 assert(is_power_of_2(bs->bl.request_alignment));
1701 for (i = 0; i < bs->quiesce_counter; i++) {
1702 if (drv->bdrv_drain_begin) {
1703 drv->bdrv_drain_begin(bs);
1710 if (bs->file != NULL) {
1711 bdrv_graph_wrlock(NULL);
1712 bdrv_unref_child(bs, bs->file);
1713 bdrv_graph_wrunlock();
1722 * Create and open a block node.
1724 * @options is a QDict of options to pass to the block drivers, or NULL for an
1725 * empty set of options. The reference to the QDict belongs to the block layer
1726 * after the call (even on failure), so if the caller intends to reuse the
1727 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1729 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1730 const char *node_name,
1731 QDict *options, int flags,
1734 BlockDriverState *bs;
1737 GLOBAL_STATE_CODE();
1740 bs->open_flags = flags;
1741 bs->options = options ?: qdict_new();
1742 bs->explicit_options = qdict_clone_shallow(bs->options);
1745 update_options_from_flags(bs->options, flags);
1747 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1749 qobject_unref(bs->explicit_options);
1750 bs->explicit_options = NULL;
1751 qobject_unref(bs->options);
1760 /* Create and open a block node. */
1761 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1762 int flags, Error **errp)
1764 GLOBAL_STATE_CODE();
1765 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1768 QemuOptsList bdrv_runtime_opts = {
1769 .name = "bdrv_common",
1770 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1773 .name = "node-name",
1774 .type = QEMU_OPT_STRING,
1775 .help = "Node name of the block device node",
1779 .type = QEMU_OPT_STRING,
1780 .help = "Block driver to use for the node",
1783 .name = BDRV_OPT_CACHE_DIRECT,
1784 .type = QEMU_OPT_BOOL,
1785 .help = "Bypass software writeback cache on the host",
1788 .name = BDRV_OPT_CACHE_NO_FLUSH,
1789 .type = QEMU_OPT_BOOL,
1790 .help = "Ignore flush requests",
1793 .name = BDRV_OPT_READ_ONLY,
1794 .type = QEMU_OPT_BOOL,
1795 .help = "Node is opened in read-only mode",
1798 .name = BDRV_OPT_AUTO_READ_ONLY,
1799 .type = QEMU_OPT_BOOL,
1800 .help = "Node can become read-only if opening read-write fails",
1803 .name = "detect-zeroes",
1804 .type = QEMU_OPT_STRING,
1805 .help = "try to optimize zero writes (off, on, unmap)",
1808 .name = BDRV_OPT_DISCARD,
1809 .type = QEMU_OPT_STRING,
1810 .help = "discard operation (ignore/off, unmap/on)",
1813 .name = BDRV_OPT_FORCE_SHARE,
1814 .type = QEMU_OPT_BOOL,
1815 .help = "always accept other writers (default: off)",
1817 { /* end of list */ }
1821 QemuOptsList bdrv_create_opts_simple = {
1822 .name = "simple-create-opts",
1823 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1826 .name = BLOCK_OPT_SIZE,
1827 .type = QEMU_OPT_SIZE,
1828 .help = "Virtual disk size"
1831 .name = BLOCK_OPT_PREALLOC,
1832 .type = QEMU_OPT_STRING,
1833 .help = "Preallocation mode (allowed values: off)"
1835 { /* end of list */ }
1840 * Common part for opening disk images and files
1842 * Removes all processed options from *options.
1844 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1845 QDict *options, Error **errp)
1847 int ret, open_flags;
1848 const char *filename;
1849 const char *driver_name = NULL;
1850 const char *node_name = NULL;
1851 const char *discard;
1854 Error *local_err = NULL;
1857 assert(bs->file == NULL);
1858 assert(options != NULL && bs->options != options);
1859 GLOBAL_STATE_CODE();
1861 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1862 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1867 update_flags_from_options(&bs->open_flags, opts);
1869 driver_name = qemu_opt_get(opts, "driver");
1870 drv = bdrv_find_format(driver_name);
1871 assert(drv != NULL);
1873 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1875 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1877 BDRV_OPT_FORCE_SHARE
1878 "=on can only be used with read-only images");
1884 bdrv_graph_rdlock_main_loop();
1885 bdrv_refresh_filename(blk_bs(file));
1886 bdrv_graph_rdunlock_main_loop();
1888 filename = blk_bs(file)->filename;
1891 * Caution: while qdict_get_try_str() is fine, getting
1892 * non-string types would require more care. When @options
1893 * come from -blockdev or blockdev_add, its members are typed
1894 * according to the QAPI schema, but when they come from
1895 * -drive, they're all QString.
1897 filename = qdict_get_try_str(options, "filename");
1900 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1901 error_setg(errp, "The '%s' block driver requires a file name",
1907 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1910 ro = bdrv_is_read_only(bs);
1912 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1913 if (!ro && bdrv_is_whitelisted(drv, true)) {
1914 bdrv_graph_rdlock_main_loop();
1915 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1916 bdrv_graph_rdunlock_main_loop();
1922 !ro && bdrv_is_whitelisted(drv, true)
1923 ? "Driver '%s' can only be used for read-only devices"
1924 : "Driver '%s' is not whitelisted",
1930 /* bdrv_new() and bdrv_close() make it so */
1931 assert(qatomic_read(&bs->copy_on_read) == 0);
1933 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1935 bdrv_enable_copy_on_read(bs);
1937 error_setg(errp, "Can't use copy-on-read on read-only device");
1943 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1944 if (discard != NULL) {
1945 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1946 error_setg(errp, "Invalid discard option");
1953 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1955 error_propagate(errp, local_err);
1960 if (filename != NULL) {
1961 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1963 bs->filename[0] = '\0';
1965 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1967 /* Open the image, either directly or using a protocol */
1968 open_flags = bdrv_open_flags(bs, bs->open_flags);
1969 node_name = qemu_opt_get(opts, "node-name");
1971 assert(!drv->bdrv_file_open || file == NULL);
1972 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1977 qemu_opts_del(opts);
1981 qemu_opts_del(opts);
1985 static QDict *parse_json_filename(const char *filename, Error **errp)
1987 QObject *options_obj;
1990 GLOBAL_STATE_CODE();
1992 ret = strstart(filename, "json:", &filename);
1995 options_obj = qobject_from_json(filename, errp);
1997 error_prepend(errp, "Could not parse the JSON options: ");
2001 options = qobject_to(QDict, options_obj);
2003 qobject_unref(options_obj);
2004 error_setg(errp, "Invalid JSON object given");
2008 qdict_flatten(options);
2013 static void parse_json_protocol(QDict *options, const char **pfilename,
2016 QDict *json_options;
2017 Error *local_err = NULL;
2018 GLOBAL_STATE_CODE();
2020 /* Parse json: pseudo-protocol */
2021 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2025 json_options = parse_json_filename(*pfilename, &local_err);
2027 error_propagate(errp, local_err);
2031 /* Options given in the filename have lower priority than options
2032 * specified directly */
2033 qdict_join(options, json_options, false);
2034 qobject_unref(json_options);
2039 * Fills in default options for opening images and converts the legacy
2040 * filename/flags pair to option QDict entries.
2041 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2042 * block driver has been specified explicitly.
2044 static int bdrv_fill_options(QDict **options, const char *filename,
2045 int *flags, Error **errp)
2047 const char *drvname;
2048 bool protocol = *flags & BDRV_O_PROTOCOL;
2049 bool parse_filename = false;
2050 BlockDriver *drv = NULL;
2051 Error *local_err = NULL;
2053 GLOBAL_STATE_CODE();
2056 * Caution: while qdict_get_try_str() is fine, getting non-string
2057 * types would require more care. When @options come from
2058 * -blockdev or blockdev_add, its members are typed according to
2059 * the QAPI schema, but when they come from -drive, they're all
2062 drvname = qdict_get_try_str(*options, "driver");
2064 drv = bdrv_find_format(drvname);
2066 error_setg(errp, "Unknown driver '%s'", drvname);
2069 /* If the user has explicitly specified the driver, this choice should
2070 * override the BDRV_O_PROTOCOL flag */
2071 protocol = drv->bdrv_file_open;
2075 *flags |= BDRV_O_PROTOCOL;
2077 *flags &= ~BDRV_O_PROTOCOL;
2080 /* Translate cache options from flags into options */
2081 update_options_from_flags(*options, *flags);
2083 /* Fetch the file name from the options QDict if necessary */
2084 if (protocol && filename) {
2085 if (!qdict_haskey(*options, "filename")) {
2086 qdict_put_str(*options, "filename", filename);
2087 parse_filename = true;
2089 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2095 /* Find the right block driver */
2096 /* See cautionary note on accessing @options above */
2097 filename = qdict_get_try_str(*options, "filename");
2099 if (!drvname && protocol) {
2101 drv = bdrv_find_protocol(filename, parse_filename, errp);
2106 drvname = drv->format_name;
2107 qdict_put_str(*options, "driver", drvname);
2109 error_setg(errp, "Must specify either driver or file");
2114 assert(drv || !protocol);
2116 /* Driver-specific filename parsing */
2117 if (drv && drv->bdrv_parse_filename && parse_filename) {
2118 drv->bdrv_parse_filename(filename, *options, &local_err);
2120 error_propagate(errp, local_err);
2124 if (!drv->bdrv_needs_filename) {
2125 qdict_del(*options, "filename");
2132 typedef struct BlockReopenQueueEntry {
2134 BDRVReopenState state;
2135 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2136 } BlockReopenQueueEntry;
2139 * Return the flags that @bs will have after the reopens in @q have
2140 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2141 * return the current flags.
2143 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2145 BlockReopenQueueEntry *entry;
2148 QTAILQ_FOREACH(entry, q, entry) {
2149 if (entry->state.bs == bs) {
2150 return entry->state.flags;
2155 return bs->open_flags;
2158 /* Returns whether the image file can be written to after the reopen queue @q
2159 * has been successfully applied, or right now if @q is NULL. */
2160 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2161 BlockReopenQueue *q)
2163 int flags = bdrv_reopen_get_flags(q, bs);
2165 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2169 * Return whether the BDS can be written to. This is not necessarily
2170 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2171 * be written to but do not count as read-only images.
2173 bool bdrv_is_writable(BlockDriverState *bs)
2176 return bdrv_is_writable_after_reopen(bs, NULL);
2179 static char *bdrv_child_user_desc(BdrvChild *c)
2181 GLOBAL_STATE_CODE();
2182 return c->klass->get_parent_desc(c);
2186 * Check that @a allows everything that @b needs. @a and @b must reference same
2189 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2191 const char *child_bs_name;
2192 g_autofree char *a_user = NULL;
2193 g_autofree char *b_user = NULL;
2194 g_autofree char *perms = NULL;
2197 assert(a->bs == b->bs);
2198 GLOBAL_STATE_CODE();
2200 if ((b->perm & a->shared_perm) == b->perm) {
2204 child_bs_name = bdrv_get_node_name(b->bs);
2205 a_user = bdrv_child_user_desc(a);
2206 b_user = bdrv_child_user_desc(b);
2207 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2209 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2210 "both required by %s (uses node '%s' as '%s' child) and "
2211 "unshared by %s (uses node '%s' as '%s' child).",
2212 child_bs_name, perms,
2213 b_user, child_bs_name, b->name,
2214 a_user, child_bs_name, a->name);
2219 static bool GRAPH_RDLOCK
2220 bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2223 GLOBAL_STATE_CODE();
2226 * During the loop we'll look at each pair twice. That's correct because
2227 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2230 QLIST_FOREACH(a, &bs->parents, next_parent) {
2231 QLIST_FOREACH(b, &bs->parents, next_parent) {
2236 if (!bdrv_a_allow_b(a, b, errp)) {
2245 static void GRAPH_RDLOCK
2246 bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2247 BdrvChild *c, BdrvChildRole role,
2248 BlockReopenQueue *reopen_queue,
2249 uint64_t parent_perm, uint64_t parent_shared,
2250 uint64_t *nperm, uint64_t *nshared)
2252 assert(bs->drv && bs->drv->bdrv_child_perm);
2253 GLOBAL_STATE_CODE();
2254 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2255 parent_perm, parent_shared,
2257 /* TODO Take force_share from reopen_queue */
2258 if (child_bs && child_bs->force_share) {
2259 *nshared = BLK_PERM_ALL;
2264 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2265 * nodes that are already in the @list, of course) so that final list is
2266 * topologically sorted. Return the result (GSList @list object is updated, so
2267 * don't use old reference after function call).
2269 * On function start @list must be already topologically sorted and for any node
2270 * in the @list the whole subtree of the node must be in the @list as well. The
2271 * simplest way to satisfy this criteria: use only result of
2272 * bdrv_topological_dfs() or NULL as @list parameter.
2274 static GSList * GRAPH_RDLOCK
2275 bdrv_topological_dfs(GSList *list, GHashTable *found, BlockDriverState *bs)
2278 g_autoptr(GHashTable) local_found = NULL;
2280 GLOBAL_STATE_CODE();
2284 found = local_found = g_hash_table_new(NULL, NULL);
2287 if (g_hash_table_contains(found, bs)) {
2290 g_hash_table_add(found, bs);
2292 QLIST_FOREACH(child, &bs->children, next) {
2293 list = bdrv_topological_dfs(list, found, child->bs);
2296 return g_slist_prepend(list, bs);
2299 typedef struct BdrvChildSetPermState {
2302 uint64_t old_shared_perm;
2303 } BdrvChildSetPermState;
2305 static void bdrv_child_set_perm_abort(void *opaque)
2307 BdrvChildSetPermState *s = opaque;
2309 GLOBAL_STATE_CODE();
2311 s->child->perm = s->old_perm;
2312 s->child->shared_perm = s->old_shared_perm;
2315 static TransactionActionDrv bdrv_child_set_pem_drv = {
2316 .abort = bdrv_child_set_perm_abort,
2320 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2321 uint64_t shared, Transaction *tran)
2323 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2324 GLOBAL_STATE_CODE();
2326 *s = (BdrvChildSetPermState) {
2328 .old_perm = c->perm,
2329 .old_shared_perm = c->shared_perm,
2333 c->shared_perm = shared;
2335 tran_add(tran, &bdrv_child_set_pem_drv, s);
2338 static void GRAPH_RDLOCK bdrv_drv_set_perm_commit(void *opaque)
2340 BlockDriverState *bs = opaque;
2341 uint64_t cumulative_perms, cumulative_shared_perms;
2342 GLOBAL_STATE_CODE();
2344 if (bs->drv->bdrv_set_perm) {
2345 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2346 &cumulative_shared_perms);
2347 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2351 static void GRAPH_RDLOCK bdrv_drv_set_perm_abort(void *opaque)
2353 BlockDriverState *bs = opaque;
2354 GLOBAL_STATE_CODE();
2356 if (bs->drv->bdrv_abort_perm_update) {
2357 bs->drv->bdrv_abort_perm_update(bs);
2361 TransactionActionDrv bdrv_drv_set_perm_drv = {
2362 .abort = bdrv_drv_set_perm_abort,
2363 .commit = bdrv_drv_set_perm_commit,
2367 * After calling this function, the transaction @tran may only be completed
2368 * while holding a reader lock for the graph.
2370 static int GRAPH_RDLOCK
2371 bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared_perm,
2372 Transaction *tran, Error **errp)
2374 GLOBAL_STATE_CODE();
2379 if (bs->drv->bdrv_check_perm) {
2380 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2387 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2393 typedef struct BdrvReplaceChildState {
2395 BlockDriverState *old_bs;
2396 } BdrvReplaceChildState;
2398 static void GRAPH_WRLOCK bdrv_replace_child_commit(void *opaque)
2400 BdrvReplaceChildState *s = opaque;
2401 GLOBAL_STATE_CODE();
2403 bdrv_schedule_unref(s->old_bs);
2406 static void GRAPH_WRLOCK bdrv_replace_child_abort(void *opaque)
2408 BdrvReplaceChildState *s = opaque;
2409 BlockDriverState *new_bs = s->child->bs;
2411 GLOBAL_STATE_CODE();
2412 assert_bdrv_graph_writable();
2414 /* old_bs reference is transparently moved from @s to @s->child */
2415 if (!s->child->bs) {
2417 * The parents were undrained when removing old_bs from the child. New
2418 * requests can't have been made, though, because the child was empty.
2420 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2421 * undraining the parent in the first place. Once this is done, having
2422 * new_bs drained when calling bdrv_replace_child_tran() is not a
2423 * requirement any more.
2425 bdrv_parent_drained_begin_single(s->child);
2426 assert(!bdrv_parent_drained_poll_single(s->child));
2428 assert(s->child->quiesced_parent);
2429 bdrv_replace_child_noperm(s->child, s->old_bs);
2434 static TransactionActionDrv bdrv_replace_child_drv = {
2435 .commit = bdrv_replace_child_commit,
2436 .abort = bdrv_replace_child_abort,
2441 * bdrv_replace_child_tran
2443 * Note: real unref of old_bs is done only on commit.
2445 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2446 * kept drained until the transaction is completed.
2448 * After calling this function, the transaction @tran may only be completed
2449 * while holding a writer lock for the graph.
2451 * The function doesn't update permissions, caller is responsible for this.
2453 static void GRAPH_WRLOCK
2454 bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2457 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2459 assert(child->quiesced_parent);
2460 assert(!new_bs || new_bs->quiesce_counter);
2462 *s = (BdrvReplaceChildState) {
2464 .old_bs = child->bs,
2466 tran_add(tran, &bdrv_replace_child_drv, s);
2472 bdrv_replace_child_noperm(child, new_bs);
2473 /* old_bs reference is transparently moved from @child to @s */
2477 * Refresh permissions in @bs subtree. The function is intended to be called
2478 * after some graph modification that was done without permission update.
2480 * After calling this function, the transaction @tran may only be completed
2481 * while holding a reader lock for the graph.
2483 static int GRAPH_RDLOCK
2484 bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2485 Transaction *tran, Error **errp)
2487 BlockDriver *drv = bs->drv;
2490 uint64_t cumulative_perms, cumulative_shared_perms;
2491 GLOBAL_STATE_CODE();
2493 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2495 /* Write permissions never work with read-only images */
2496 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2497 !bdrv_is_writable_after_reopen(bs, q))
2499 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2500 error_setg(errp, "Block node is read-only");
2502 error_setg(errp, "Read-only block node '%s' cannot support "
2503 "read-write users", bdrv_get_node_name(bs));
2510 * Unaligned requests will automatically be aligned to bl.request_alignment
2511 * and without RESIZE we can't extend requests to write to space beyond the
2512 * end of the image, so it's required that the image size is aligned.
2514 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2515 !(cumulative_perms & BLK_PERM_RESIZE))
2517 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2518 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2519 "Image size is not a multiple of request "
2525 /* Check this node */
2530 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2536 /* Drivers that never have children can omit .bdrv_child_perm() */
2537 if (!drv->bdrv_child_perm) {
2538 assert(QLIST_EMPTY(&bs->children));
2542 /* Check all children */
2543 QLIST_FOREACH(c, &bs->children, next) {
2544 uint64_t cur_perm, cur_shared;
2546 bdrv_child_perm(bs, c->bs, c, c->role, q,
2547 cumulative_perms, cumulative_shared_perms,
2548 &cur_perm, &cur_shared);
2549 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2556 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2557 * a topologically sorted subgraph.
2559 * After calling this function, the transaction @tran may only be completed
2560 * while holding a reader lock for the graph.
2562 static int GRAPH_RDLOCK
2563 bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2567 BlockDriverState *bs;
2568 GLOBAL_STATE_CODE();
2570 for ( ; list; list = list->next) {
2573 if (bdrv_parent_perms_conflict(bs, errp)) {
2577 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2587 * @list is any list of nodes. List is completed by all subtrees and
2588 * topologically sorted. It's not a problem if some node occurs in the @list
2591 * After calling this function, the transaction @tran may only be completed
2592 * while holding a reader lock for the graph.
2594 static int GRAPH_RDLOCK
2595 bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2598 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2599 g_autoptr(GSList) refresh_list = NULL;
2601 for ( ; list; list = list->next) {
2602 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2605 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2608 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2609 uint64_t *shared_perm)
2612 uint64_t cumulative_perms = 0;
2613 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2615 GLOBAL_STATE_CODE();
2617 QLIST_FOREACH(c, &bs->parents, next_parent) {
2618 cumulative_perms |= c->perm;
2619 cumulative_shared_perms &= c->shared_perm;
2622 *perm = cumulative_perms;
2623 *shared_perm = cumulative_shared_perms;
2626 char *bdrv_perm_names(uint64_t perm)
2632 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2633 { BLK_PERM_WRITE, "write" },
2634 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2635 { BLK_PERM_RESIZE, "resize" },
2639 GString *result = g_string_sized_new(30);
2640 struct perm_name *p;
2642 for (p = permissions; p->name; p++) {
2643 if (perm & p->perm) {
2644 if (result->len > 0) {
2645 g_string_append(result, ", ");
2647 g_string_append(result, p->name);
2651 return g_string_free(result, FALSE);
2656 * @tran is allowed to be NULL. In this case no rollback is possible.
2658 * After calling this function, the transaction @tran may only be completed
2659 * while holding a reader lock for the graph.
2661 static int GRAPH_RDLOCK
2662 bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, Error **errp)
2665 Transaction *local_tran = NULL;
2666 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2667 GLOBAL_STATE_CODE();
2670 tran = local_tran = tran_new();
2673 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2676 tran_finalize(local_tran, ret);
2682 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2685 Error *local_err = NULL;
2686 Transaction *tran = tran_new();
2689 GLOBAL_STATE_CODE();
2691 bdrv_child_set_perm(c, perm, shared, tran);
2693 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2695 tran_finalize(tran, ret);
2698 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2699 /* tighten permissions */
2700 error_propagate(errp, local_err);
2703 * Our caller may intend to only loosen restrictions and
2704 * does not expect this function to fail. Errors are not
2705 * fatal in such a case, so we can just hide them from our
2708 error_free(local_err);
2716 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2718 uint64_t parent_perms, parent_shared;
2719 uint64_t perms, shared;
2721 GLOBAL_STATE_CODE();
2723 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2724 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2725 parent_perms, parent_shared, &perms, &shared);
2727 return bdrv_child_try_set_perm(c, perms, shared, errp);
2731 * Default implementation for .bdrv_child_perm() for block filters:
2732 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2735 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2737 BlockReopenQueue *reopen_queue,
2738 uint64_t perm, uint64_t shared,
2739 uint64_t *nperm, uint64_t *nshared)
2741 GLOBAL_STATE_CODE();
2742 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2743 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2746 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2748 BlockReopenQueue *reopen_queue,
2749 uint64_t perm, uint64_t shared,
2750 uint64_t *nperm, uint64_t *nshared)
2752 assert(role & BDRV_CHILD_COW);
2753 GLOBAL_STATE_CODE();
2756 * We want consistent read from backing files if the parent needs it.
2757 * No other operations are performed on backing files.
2759 perm &= BLK_PERM_CONSISTENT_READ;
2762 * If the parent can deal with changing data, we're okay with a
2763 * writable and resizable backing file.
2764 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2766 if (shared & BLK_PERM_WRITE) {
2767 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2772 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2774 if (bs->open_flags & BDRV_O_INACTIVE) {
2775 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2782 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2784 BlockReopenQueue *reopen_queue,
2785 uint64_t perm, uint64_t shared,
2786 uint64_t *nperm, uint64_t *nshared)
2790 GLOBAL_STATE_CODE();
2791 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2793 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2796 * Apart from the modifications below, the same permissions are
2797 * forwarded and left alone as for filters
2799 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2800 perm, shared, &perm, &shared);
2802 if (role & BDRV_CHILD_METADATA) {
2803 /* Format drivers may touch metadata even if the guest doesn't write */
2804 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2805 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2809 * bs->file always needs to be consistent because of the
2810 * metadata. We can never allow other users to resize or write
2813 if (!(flags & BDRV_O_NO_IO)) {
2814 perm |= BLK_PERM_CONSISTENT_READ;
2816 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2819 if (role & BDRV_CHILD_DATA) {
2821 * Technically, everything in this block is a subset of the
2822 * BDRV_CHILD_METADATA path taken above, and so this could
2823 * be an "else if" branch. However, that is not obvious, and
2824 * this function is not performance critical, therefore we let
2825 * this be an independent "if".
2829 * We cannot allow other users to resize the file because the
2830 * format driver might have some assumptions about the size
2831 * (e.g. because it is stored in metadata, or because the file
2832 * is split into fixed-size data files).
2834 shared &= ~BLK_PERM_RESIZE;
2837 * WRITE_UNCHANGED often cannot be performed as such on the
2838 * data file. For example, the qcow2 driver may still need to
2839 * write copied clusters on copy-on-read.
2841 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2842 perm |= BLK_PERM_WRITE;
2846 * If the data file is written to, the format driver may
2847 * expect to be able to resize it by writing beyond the EOF.
2849 if (perm & BLK_PERM_WRITE) {
2850 perm |= BLK_PERM_RESIZE;
2854 if (bs->open_flags & BDRV_O_INACTIVE) {
2855 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2862 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2863 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2864 uint64_t perm, uint64_t shared,
2865 uint64_t *nperm, uint64_t *nshared)
2867 GLOBAL_STATE_CODE();
2868 if (role & BDRV_CHILD_FILTERED) {
2869 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2871 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2872 perm, shared, nperm, nshared);
2873 } else if (role & BDRV_CHILD_COW) {
2874 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2875 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2876 perm, shared, nperm, nshared);
2877 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2878 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2879 perm, shared, nperm, nshared);
2881 g_assert_not_reached();
2885 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2887 static const uint64_t permissions[] = {
2888 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2889 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2890 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2891 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2894 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2895 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2897 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2899 return permissions[qapi_perm];
2903 * Replaces the node that a BdrvChild points to without updating permissions.
2905 * If @new_bs is non-NULL, the parent of @child must already be drained through
2906 * @child and the caller must hold the AioContext lock for @new_bs.
2908 static void GRAPH_WRLOCK
2909 bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs)
2911 BlockDriverState *old_bs = child->bs;
2912 int new_bs_quiesce_counter;
2914 assert(!child->frozen);
2917 * If we want to change the BdrvChild to point to a drained node as its new
2918 * child->bs, we need to make sure that its new parent is drained, too. In
2919 * other words, either child->quiesce_parent must already be true or we must
2920 * be able to set it and keep the parent's quiesce_counter consistent with
2921 * that, but without polling or starting new requests (this function
2922 * guarantees that it doesn't poll, and starting new requests would be
2923 * against the invariants of drain sections).
2925 * To keep things simple, we pick the first option (child->quiesce_parent
2926 * must already be true). We also generalise the rule a bit to make it
2927 * easier to verify in callers and more likely to be covered in test cases:
2928 * The parent must be quiesced through this child even if new_bs isn't
2929 * currently drained.
2931 * The only exception is for callers that always pass new_bs == NULL. In
2932 * this case, we obviously never need to consider the case of a drained
2933 * new_bs, so we can keep the callers simpler by allowing them not to drain
2936 assert(!new_bs || child->quiesced_parent);
2937 assert(old_bs != new_bs);
2938 GLOBAL_STATE_CODE();
2940 if (old_bs && new_bs) {
2941 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2945 if (child->klass->detach) {
2946 child->klass->detach(child);
2948 QLIST_REMOVE(child, next_parent);
2954 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2955 if (child->klass->attach) {
2956 child->klass->attach(child);
2961 * If the parent was drained through this BdrvChild previously, but new_bs
2962 * is not drained, allow requests to come in only after the new node has
2965 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2966 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2967 bdrv_parent_drained_end_single(child);
2972 * Free the given @child.
2974 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2975 * unused (i.e. not in a children list).
2977 static void bdrv_child_free(BdrvChild *child)
2980 GLOBAL_STATE_CODE();
2981 GRAPH_RDLOCK_GUARD_MAINLOOP();
2983 assert(!child->next.le_prev); /* not in children list */
2985 g_free(child->name);
2989 typedef struct BdrvAttachChildCommonState {
2991 AioContext *old_parent_ctx;
2992 AioContext *old_child_ctx;
2993 } BdrvAttachChildCommonState;
2995 static void GRAPH_WRLOCK bdrv_attach_child_common_abort(void *opaque)
2997 BdrvAttachChildCommonState *s = opaque;
2998 BlockDriverState *bs = s->child->bs;
3000 GLOBAL_STATE_CODE();
3001 assert_bdrv_graph_writable();
3003 bdrv_replace_child_noperm(s->child, NULL);
3005 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
3006 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
3009 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
3011 GHashTable *visited;
3016 /* No need to visit `child`, because it has been detached already */
3017 visited = g_hash_table_new(NULL, NULL);
3018 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
3019 visited, tran, &error_abort);
3020 g_hash_table_destroy(visited);
3022 /* transaction is supposed to always succeed */
3023 assert(ret == true);
3027 bdrv_schedule_unref(bs);
3028 bdrv_child_free(s->child);
3031 static TransactionActionDrv bdrv_attach_child_common_drv = {
3032 .abort = bdrv_attach_child_common_abort,
3037 * Common part of attaching bdrv child to bs or to blk or to job
3039 * Function doesn't update permissions, caller is responsible for this.
3041 * After calling this function, the transaction @tran may only be completed
3042 * while holding a writer lock for the graph.
3044 * Returns new created child.
3046 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3047 * @child_bs can move to a different AioContext in this function. Callers must
3048 * make sure that their AioContext locking is still correct after this.
3050 static BdrvChild * GRAPH_WRLOCK
3051 bdrv_attach_child_common(BlockDriverState *child_bs,
3052 const char *child_name,
3053 const BdrvChildClass *child_class,
3054 BdrvChildRole child_role,
3055 uint64_t perm, uint64_t shared_perm,
3057 Transaction *tran, Error **errp)
3059 BdrvChild *new_child;
3060 AioContext *parent_ctx, *new_child_ctx;
3061 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3063 assert(child_class->get_parent_desc);
3064 GLOBAL_STATE_CODE();
3066 new_child = g_new(BdrvChild, 1);
3067 *new_child = (BdrvChild) {
3069 .name = g_strdup(child_name),
3070 .klass = child_class,
3073 .shared_perm = shared_perm,
3078 * If the AioContexts don't match, first try to move the subtree of
3079 * child_bs into the AioContext of the new parent. If this doesn't work,
3080 * try moving the parent into the AioContext of child_bs instead.
3082 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3083 if (child_ctx != parent_ctx) {
3084 Error *local_err = NULL;
3085 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3088 if (ret < 0 && child_class->change_aio_ctx) {
3089 Transaction *aio_ctx_tran = tran_new();
3090 GHashTable *visited = g_hash_table_new(NULL, NULL);
3093 g_hash_table_add(visited, new_child);
3094 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3095 visited, aio_ctx_tran,
3097 if (ret_child == true) {
3098 error_free(local_err);
3101 tran_finalize(aio_ctx_tran, ret_child == true ? 0 : -1);
3102 g_hash_table_destroy(visited);
3106 error_propagate(errp, local_err);
3107 bdrv_child_free(new_child);
3112 new_child_ctx = bdrv_get_aio_context(child_bs);
3113 if (new_child_ctx != child_ctx) {
3114 aio_context_release(child_ctx);
3115 aio_context_acquire(new_child_ctx);
3120 * Let every new BdrvChild start with a drained parent. Inserting the child
3121 * in the graph with bdrv_replace_child_noperm() will undrain it if
3122 * @child_bs is not drained.
3124 * The child was only just created and is not yet visible in global state
3125 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3126 * could have sent requests and polling is not necessary.
3128 * Note that this means that the parent isn't fully drained yet, we only
3129 * stop new requests from coming in. This is fine, we don't care about the
3130 * old requests here, they are not for this child. If another place enters a
3131 * drain section for the same parent, but wants it to be fully quiesced, it
3132 * will not run most of the the code in .drained_begin() again (which is not
3133 * a problem, we already did this), but it will still poll until the parent
3134 * is fully quiesced, so it will not be negatively affected either.
3136 bdrv_parent_drained_begin_single(new_child);
3137 bdrv_replace_child_noperm(new_child, child_bs);
3139 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3140 *s = (BdrvAttachChildCommonState) {
3142 .old_parent_ctx = parent_ctx,
3143 .old_child_ctx = child_ctx,
3145 tran_add(tran, &bdrv_attach_child_common_drv, s);
3147 if (new_child_ctx != child_ctx) {
3148 aio_context_release(new_child_ctx);
3149 aio_context_acquire(child_ctx);
3156 * Function doesn't update permissions, caller is responsible for this.
3158 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3159 * @child_bs can move to a different AioContext in this function. Callers must
3160 * make sure that their AioContext locking is still correct after this.
3162 * After calling this function, the transaction @tran may only be completed
3163 * while holding a writer lock for the graph.
3165 static BdrvChild * GRAPH_WRLOCK
3166 bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3167 BlockDriverState *child_bs,
3168 const char *child_name,
3169 const BdrvChildClass *child_class,
3170 BdrvChildRole child_role,
3174 uint64_t perm, shared_perm;
3176 assert(parent_bs->drv);
3177 GLOBAL_STATE_CODE();
3179 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3180 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3181 child_bs->node_name, child_name, parent_bs->node_name);
3185 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3186 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3187 perm, shared_perm, &perm, &shared_perm);
3189 return bdrv_attach_child_common(child_bs, child_name, child_class,
3190 child_role, perm, shared_perm, parent_bs,
3195 * This function steals the reference to child_bs from the caller.
3196 * That reference is later dropped by bdrv_root_unref_child().
3198 * On failure NULL is returned, errp is set and the reference to
3199 * child_bs is also dropped.
3201 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3202 * (unless @child_bs is already in @ctx).
3204 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3205 const char *child_name,
3206 const BdrvChildClass *child_class,
3207 BdrvChildRole child_role,
3208 uint64_t perm, uint64_t shared_perm,
3209 void *opaque, Error **errp)
3213 Transaction *tran = tran_new();
3215 GLOBAL_STATE_CODE();
3217 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3218 child_role, perm, shared_perm, opaque,
3225 ret = bdrv_refresh_perms(child_bs, tran, errp);
3228 tran_finalize(tran, ret);
3230 bdrv_schedule_unref(child_bs);
3232 return ret < 0 ? NULL : child;
3236 * This function transfers the reference to child_bs from the caller
3237 * to parent_bs. That reference is later dropped by parent_bs on
3238 * bdrv_close() or if someone calls bdrv_unref_child().
3240 * On failure NULL is returned, errp is set and the reference to
3241 * child_bs is also dropped.
3243 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3244 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3246 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3247 BlockDriverState *child_bs,
3248 const char *child_name,
3249 const BdrvChildClass *child_class,
3250 BdrvChildRole child_role,
3255 Transaction *tran = tran_new();
3257 GLOBAL_STATE_CODE();
3259 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3260 child_class, child_role, tran, errp);
3266 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3272 tran_finalize(tran, ret);
3274 bdrv_schedule_unref(child_bs);
3276 return ret < 0 ? NULL : child;
3279 /* Callers must ensure that child->frozen is false. */
3280 void bdrv_root_unref_child(BdrvChild *child)
3282 BlockDriverState *child_bs = child->bs;
3284 GLOBAL_STATE_CODE();
3285 bdrv_replace_child_noperm(child, NULL);
3286 bdrv_child_free(child);
3290 * Update permissions for old node. We're just taking a parent away, so
3291 * we're loosening restrictions. Errors of permission update are not
3292 * fatal in this case, ignore them.
3294 bdrv_refresh_perms(child_bs, NULL, NULL);
3297 * When the parent requiring a non-default AioContext is removed, the
3298 * node moves back to the main AioContext
3300 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3304 bdrv_schedule_unref(child_bs);
3307 typedef struct BdrvSetInheritsFrom {
3308 BlockDriverState *bs;
3309 BlockDriverState *old_inherits_from;
3310 } BdrvSetInheritsFrom;
3312 static void bdrv_set_inherits_from_abort(void *opaque)
3314 BdrvSetInheritsFrom *s = opaque;
3316 s->bs->inherits_from = s->old_inherits_from;
3319 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3320 .abort = bdrv_set_inherits_from_abort,
3324 /* @tran is allowed to be NULL. In this case no rollback is possible */
3325 static void bdrv_set_inherits_from(BlockDriverState *bs,
3326 BlockDriverState *new_inherits_from,
3330 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3332 *s = (BdrvSetInheritsFrom) {
3334 .old_inherits_from = bs->inherits_from,
3337 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3340 bs->inherits_from = new_inherits_from;
3344 * Clear all inherits_from pointers from children and grandchildren of
3345 * @root that point to @root, where necessary.
3346 * @tran is allowed to be NULL. In this case no rollback is possible
3348 static void GRAPH_WRLOCK
3349 bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3354 if (child->bs->inherits_from == root) {
3356 * Remove inherits_from only when the last reference between root and
3357 * child->bs goes away.
3359 QLIST_FOREACH(c, &root->children, next) {
3360 if (c != child && c->bs == child->bs) {
3365 bdrv_set_inherits_from(child->bs, NULL, tran);
3369 QLIST_FOREACH(c, &child->bs->children, next) {
3370 bdrv_unset_inherits_from(root, c, tran);
3374 /* Callers must ensure that child->frozen is false. */
3375 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3377 GLOBAL_STATE_CODE();
3378 if (child == NULL) {
3382 bdrv_unset_inherits_from(parent, child, NULL);
3383 bdrv_root_unref_child(child);
3387 static void GRAPH_RDLOCK
3388 bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3391 GLOBAL_STATE_CODE();
3392 QLIST_FOREACH(c, &bs->parents, next_parent) {
3393 if (c->klass->change_media) {
3394 c->klass->change_media(c, load);
3399 /* Return true if you can reach parent going through child->inherits_from
3400 * recursively. If parent or child are NULL, return false */
3401 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3402 BlockDriverState *parent)
3404 while (child && child != parent) {
3405 child = child->inherits_from;
3408 return child != NULL;
3412 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3413 * mostly used for COW backing children (role = COW), but also for
3414 * filtered children (role = FILTERED | PRIMARY).
3416 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3418 if (bs->drv && bs->drv->is_filter) {
3419 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3421 return BDRV_CHILD_COW;
3426 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3427 * callers which don't need their own reference any more must call bdrv_unref().
3429 * If the respective child is already present (i.e. we're detaching a node),
3430 * that child node must be drained.
3432 * Function doesn't update permissions, caller is responsible for this.
3434 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3435 * @child_bs can move to a different AioContext in this function. Callers must
3436 * make sure that their AioContext locking is still correct after this.
3438 * After calling this function, the transaction @tran may only be completed
3439 * while holding a writer lock for the graph.
3441 static int GRAPH_WRLOCK
3442 bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3443 BlockDriverState *child_bs,
3445 Transaction *tran, Error **errp)
3447 bool update_inherits_from =
3448 bdrv_inherits_from_recursive(child_bs, parent_bs);
3449 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3452 GLOBAL_STATE_CODE();
3454 if (!parent_bs->drv) {
3456 * Node without drv is an object without a class :/. TODO: finally fix
3457 * qcow2 driver to never clear bs->drv and implement format corruption
3458 * handling in other way.
3460 error_setg(errp, "Node corrupted");
3464 if (child && child->frozen) {
3465 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3466 child->name, parent_bs->node_name, child->bs->node_name);
3470 if (is_backing && !parent_bs->drv->is_filter &&
3471 !parent_bs->drv->supports_backing)
3473 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3474 "files", parent_bs->drv->format_name, parent_bs->node_name);
3478 if (parent_bs->drv->is_filter) {
3479 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3480 } else if (is_backing) {
3481 role = BDRV_CHILD_COW;
3484 * We only can use same role as it is in existing child. We don't have
3485 * infrastructure to determine role of file child in generic way
3488 error_setg(errp, "Cannot set file child to format node without "
3496 assert(child->bs->quiesce_counter);
3497 bdrv_unset_inherits_from(parent_bs, child, tran);
3498 bdrv_remove_child(child, tran);
3505 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3506 is_backing ? "backing" : "file",
3507 &child_of_bds, role,
3515 * If inherits_from pointed recursively to bs then let's update it to
3516 * point directly to bs (else it will become NULL).
3518 if (update_inherits_from) {
3519 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3523 bdrv_refresh_limits(parent_bs, tran, NULL);
3529 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3530 * @backing_hd can move to a different AioContext in this function. Callers must
3531 * make sure that their AioContext locking is still correct after this.
3533 * If a backing child is already present (i.e. we're detaching a node), that
3534 * child node must be drained.
3536 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3537 BlockDriverState *backing_hd,
3541 Transaction *tran = tran_new();
3543 GLOBAL_STATE_CODE();
3544 assert(bs->quiesce_counter > 0);
3546 assert(bs->backing->bs->quiesce_counter > 0);
3549 ret = bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3554 ret = bdrv_refresh_perms(bs, tran, errp);
3556 tran_finalize(tran, ret);
3560 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3563 BlockDriverState *drain_bs = bs->backing ? bs->backing->bs : bs;
3565 GLOBAL_STATE_CODE();
3568 bdrv_drained_begin(drain_bs);
3569 bdrv_graph_wrlock(backing_hd);
3570 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3571 bdrv_graph_wrunlock();
3572 bdrv_drained_end(drain_bs);
3573 bdrv_unref(drain_bs);
3579 * Opens the backing file for a BlockDriverState if not yet open
3581 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3582 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3583 * itself, all options starting with "${bdref_key}." are considered part of the
3586 * The caller must hold the main AioContext lock.
3588 * TODO Can this be unified with bdrv_open_image()?
3590 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3591 const char *bdref_key, Error **errp)
3593 char *backing_filename = NULL;
3594 char *bdref_key_dot;
3595 const char *reference = NULL;
3597 bool implicit_backing = false;
3598 BlockDriverState *backing_hd;
3599 AioContext *backing_hd_ctx;
3601 QDict *tmp_parent_options = NULL;
3602 Error *local_err = NULL;
3604 GLOBAL_STATE_CODE();
3606 if (bs->backing != NULL) {
3610 /* NULL means an empty set of options */
3611 if (parent_options == NULL) {
3612 tmp_parent_options = qdict_new();
3613 parent_options = tmp_parent_options;
3616 bs->open_flags &= ~BDRV_O_NO_BACKING;
3618 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3619 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3620 g_free(bdref_key_dot);
3623 * Caution: while qdict_get_try_str() is fine, getting non-string
3624 * types would require more care. When @parent_options come from
3625 * -blockdev or blockdev_add, its members are typed according to
3626 * the QAPI schema, but when they come from -drive, they're all
3629 reference = qdict_get_try_str(parent_options, bdref_key);
3630 if (reference || qdict_haskey(options, "file.filename")) {
3631 /* keep backing_filename NULL */
3632 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3633 qobject_unref(options);
3636 if (qdict_size(options) == 0) {
3637 /* If the user specifies options that do not modify the
3638 * backing file's behavior, we might still consider it the
3639 * implicit backing file. But it's easier this way, and
3640 * just specifying some of the backing BDS's options is
3641 * only possible with -drive anyway (otherwise the QAPI
3642 * schema forces the user to specify everything). */
3643 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3646 bdrv_graph_rdlock_main_loop();
3647 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3648 bdrv_graph_rdunlock_main_loop();
3652 error_propagate(errp, local_err);
3653 qobject_unref(options);
3658 if (!bs->drv || !bs->drv->supports_backing) {
3660 error_setg(errp, "Driver doesn't support backing files");
3661 qobject_unref(options);
3666 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3667 qdict_put_str(options, "driver", bs->backing_format);
3670 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3671 &child_of_bds, bdrv_backing_role(bs), errp);
3673 bs->open_flags |= BDRV_O_NO_BACKING;
3674 error_prepend(errp, "Could not open backing file: ");
3679 if (implicit_backing) {
3680 bdrv_graph_rdlock_main_loop();
3681 bdrv_refresh_filename(backing_hd);
3682 bdrv_graph_rdunlock_main_loop();
3683 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3684 backing_hd->filename);
3687 /* Hook up the backing file link; drop our reference, bs owns the
3688 * backing_hd reference now */
3689 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3690 aio_context_acquire(backing_hd_ctx);
3691 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3692 bdrv_unref(backing_hd);
3693 aio_context_release(backing_hd_ctx);
3699 qdict_del(parent_options, bdref_key);
3702 g_free(backing_filename);
3703 qobject_unref(tmp_parent_options);
3707 static BlockDriverState *
3708 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3709 BlockDriverState *parent, const BdrvChildClass *child_class,
3710 BdrvChildRole child_role, bool allow_none, Error **errp)
3712 BlockDriverState *bs = NULL;
3713 QDict *image_options;
3714 char *bdref_key_dot;
3715 const char *reference;
3717 assert(child_class != NULL);
3719 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3720 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3721 g_free(bdref_key_dot);
3724 * Caution: while qdict_get_try_str() is fine, getting non-string
3725 * types would require more care. When @options come from
3726 * -blockdev or blockdev_add, its members are typed according to
3727 * the QAPI schema, but when they come from -drive, they're all
3730 reference = qdict_get_try_str(options, bdref_key);
3731 if (!filename && !reference && !qdict_size(image_options)) {
3733 error_setg(errp, "A block device must be specified for \"%s\"",
3736 qobject_unref(image_options);
3740 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3741 parent, child_class, child_role, errp);
3747 qdict_del(options, bdref_key);
3752 * Opens a disk image whose options are given as BlockdevRef in another block
3755 * If allow_none is true, no image will be opened if filename is false and no
3756 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3758 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3759 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3760 * itself, all options starting with "${bdref_key}." are considered part of the
3763 * The BlockdevRef will be removed from the options QDict.
3765 * The caller must hold the lock of the main AioContext and no other AioContext.
3766 * @parent can move to a different AioContext in this function. Callers must
3767 * make sure that their AioContext locking is still correct after this.
3769 BdrvChild *bdrv_open_child(const char *filename,
3770 QDict *options, const char *bdref_key,
3771 BlockDriverState *parent,
3772 const BdrvChildClass *child_class,
3773 BdrvChildRole child_role,
3774 bool allow_none, Error **errp)
3776 BlockDriverState *bs;
3780 GLOBAL_STATE_CODE();
3782 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3783 child_role, allow_none, errp);
3788 bdrv_graph_wrlock(NULL);
3789 ctx = bdrv_get_aio_context(bs);
3790 aio_context_acquire(ctx);
3791 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3793 aio_context_release(ctx);
3794 bdrv_graph_wrunlock();
3800 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3802 * The caller must hold the lock of the main AioContext and no other AioContext.
3803 * @parent can move to a different AioContext in this function. Callers must
3804 * make sure that their AioContext locking is still correct after this.
3806 int bdrv_open_file_child(const char *filename,
3807 QDict *options, const char *bdref_key,
3808 BlockDriverState *parent, Error **errp)
3812 /* commit_top and mirror_top don't use this function */
3813 assert(!parent->drv->filtered_child_is_backing);
3814 role = parent->drv->is_filter ?
3815 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3817 if (!bdrv_open_child(filename, options, bdref_key, parent,
3818 &child_of_bds, role, false, errp))
3827 * TODO Future callers may need to specify parent/child_class in order for
3828 * option inheritance to work. Existing callers use it for the root node.
3830 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3832 BlockDriverState *bs = NULL;
3833 QObject *obj = NULL;
3834 QDict *qdict = NULL;
3835 const char *reference = NULL;
3838 GLOBAL_STATE_CODE();
3840 if (ref->type == QTYPE_QSTRING) {
3841 reference = ref->u.reference;
3843 BlockdevOptions *options = &ref->u.definition;
3844 assert(ref->type == QTYPE_QDICT);
3846 v = qobject_output_visitor_new(&obj);
3847 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3848 visit_complete(v, &obj);
3850 qdict = qobject_to(QDict, obj);
3851 qdict_flatten(qdict);
3853 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3854 * compatibility with other callers) rather than what we want as the
3855 * real defaults. Apply the defaults here instead. */
3856 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3857 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3858 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3859 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3863 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3870 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3872 QDict *snapshot_options,
3875 g_autofree char *tmp_filename = NULL;
3877 QemuOpts *opts = NULL;
3878 BlockDriverState *bs_snapshot = NULL;
3879 AioContext *ctx = bdrv_get_aio_context(bs);
3882 GLOBAL_STATE_CODE();
3884 /* if snapshot, we create a temporary backing file and open it
3885 instead of opening 'filename' directly */
3887 /* Get the required size from the image */
3888 aio_context_acquire(ctx);
3889 total_size = bdrv_getlength(bs);
3890 aio_context_release(ctx);
3892 if (total_size < 0) {
3893 error_setg_errno(errp, -total_size, "Could not get image size");
3897 /* Create the temporary image */
3898 tmp_filename = create_tmp_file(errp);
3899 if (!tmp_filename) {
3903 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3905 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3906 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3907 qemu_opts_del(opts);
3909 error_prepend(errp, "Could not create temporary overlay '%s': ",
3914 /* Prepare options QDict for the temporary file */
3915 qdict_put_str(snapshot_options, "file.driver", "file");
3916 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3917 qdict_put_str(snapshot_options, "driver", "qcow2");
3919 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3920 snapshot_options = NULL;
3925 aio_context_acquire(ctx);
3926 ret = bdrv_append(bs_snapshot, bs, errp);
3927 aio_context_release(ctx);
3935 qobject_unref(snapshot_options);
3940 * Opens a disk image (raw, qcow2, vmdk, ...)
3942 * options is a QDict of options to pass to the block drivers, or NULL for an
3943 * empty set of options. The reference to the QDict belongs to the block layer
3944 * after the call (even on failure), so if the caller intends to reuse the
3945 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3947 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3948 * If it is not NULL, the referenced BDS will be reused.
3950 * The reference parameter may be used to specify an existing block device which
3951 * should be opened. If specified, neither options nor a filename may be given,
3952 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3954 * The caller must always hold the main AioContext lock.
3956 static BlockDriverState * no_coroutine_fn
3957 bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3958 int flags, BlockDriverState *parent,
3959 const BdrvChildClass *child_class, BdrvChildRole child_role,
3963 BlockBackend *file = NULL;
3964 BlockDriverState *bs;
3965 BlockDriver *drv = NULL;
3967 const char *drvname;
3968 const char *backing;
3969 Error *local_err = NULL;
3970 QDict *snapshot_options = NULL;
3971 int snapshot_flags = 0;
3972 AioContext *ctx = qemu_get_aio_context();
3974 assert(!child_class || !flags);
3975 assert(!child_class == !parent);
3976 GLOBAL_STATE_CODE();
3977 assert(!qemu_in_coroutine());
3979 /* TODO We'll eventually have to take a writer lock in this function */
3980 GRAPH_RDLOCK_GUARD_MAINLOOP();
3983 bool options_non_empty = options ? qdict_size(options) : false;
3984 qobject_unref(options);
3986 if (filename || options_non_empty) {
3987 error_setg(errp, "Cannot reference an existing block device with "
3988 "additional options or a new filename");
3992 bs = bdrv_lookup_bs(reference, reference, errp);
4003 /* NULL means an empty set of options */
4004 if (options == NULL) {
4005 options = qdict_new();
4008 /* json: syntax counts as explicit options, as if in the QDict */
4009 parse_json_protocol(options, &filename, &local_err);
4014 bs->explicit_options = qdict_clone_shallow(options);
4017 bool parent_is_format;
4020 parent_is_format = parent->drv->is_format;
4023 * parent->drv is not set yet because this node is opened for
4024 * (potential) format probing. That means that @parent is going
4025 * to be a format node.
4027 parent_is_format = true;
4030 bs->inherits_from = parent;
4031 child_class->inherit_options(child_role, parent_is_format,
4033 parent->open_flags, parent->options);
4036 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
4042 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4043 * Caution: getting a boolean member of @options requires care.
4044 * When @options come from -blockdev or blockdev_add, members are
4045 * typed according to the QAPI schema, but when they come from
4046 * -drive, they're all QString.
4048 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
4049 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
4050 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
4052 flags &= ~BDRV_O_RDWR;
4055 if (flags & BDRV_O_SNAPSHOT) {
4056 snapshot_options = qdict_new();
4057 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
4059 /* Let bdrv_backing_options() override "read-only" */
4060 qdict_del(options, BDRV_OPT_READ_ONLY);
4061 bdrv_inherited_options(BDRV_CHILD_COW, true,
4062 &flags, options, flags, options);
4065 bs->open_flags = flags;
4066 bs->options = options;
4067 options = qdict_clone_shallow(options);
4069 /* Find the right image format driver */
4070 /* See cautionary note on accessing @options above */
4071 drvname = qdict_get_try_str(options, "driver");
4073 drv = bdrv_find_format(drvname);
4075 error_setg(errp, "Unknown driver: '%s'", drvname);
4080 assert(drvname || !(flags & BDRV_O_PROTOCOL));
4082 /* See cautionary note on accessing @options above */
4083 backing = qdict_get_try_str(options, "backing");
4084 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4085 (backing && *backing == '\0'))
4088 warn_report("Use of \"backing\": \"\" is deprecated; "
4089 "use \"backing\": null instead");
4091 flags |= BDRV_O_NO_BACKING;
4092 qdict_del(bs->explicit_options, "backing");
4093 qdict_del(bs->options, "backing");
4094 qdict_del(options, "backing");
4097 /* Open image file without format layer. This BlockBackend is only used for
4098 * probing, the block drivers will do their own bdrv_open_child() for the
4099 * same BDS, which is why we put the node name back into options. */
4100 if ((flags & BDRV_O_PROTOCOL) == 0) {
4101 BlockDriverState *file_bs;
4103 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
4104 &child_of_bds, BDRV_CHILD_IMAGE,
4109 if (file_bs != NULL) {
4110 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4111 * looking at the header to guess the image format. This works even
4112 * in cases where a guest would not see a consistent state. */
4113 ctx = bdrv_get_aio_context(file_bs);
4114 aio_context_acquire(ctx);
4115 file = blk_new(ctx, 0, BLK_PERM_ALL);
4116 blk_insert_bs(file, file_bs, &local_err);
4117 bdrv_unref(file_bs);
4118 aio_context_release(ctx);
4124 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4128 /* Image format probing */
4131 ret = find_image_format(file, filename, &drv, &local_err);
4136 * This option update would logically belong in bdrv_fill_options(),
4137 * but we first need to open bs->file for the probing to work, while
4138 * opening bs->file already requires the (mostly) final set of options
4139 * so that cache mode etc. can be inherited.
4141 * Adding the driver later is somewhat ugly, but it's not an option
4142 * that would ever be inherited, so it's correct. We just need to make
4143 * sure to update both bs->options (which has the full effective
4144 * options for bs) and options (which has file.* already removed).
4146 qdict_put_str(bs->options, "driver", drv->format_name);
4147 qdict_put_str(options, "driver", drv->format_name);
4149 error_setg(errp, "Must specify either driver or file");
4153 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4154 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4155 /* file must be NULL if a protocol BDS is about to be created
4156 * (the inverse results in an error message from bdrv_open_common()) */
4157 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4159 /* Open the image */
4160 ret = bdrv_open_common(bs, file, options, &local_err);
4165 /* The AioContext could have changed during bdrv_open_common() */
4166 ctx = bdrv_get_aio_context(bs);
4169 aio_context_acquire(ctx);
4171 aio_context_release(ctx);
4175 /* If there is a backing file, use it */
4176 if ((flags & BDRV_O_NO_BACKING) == 0) {
4177 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4179 goto close_and_fail;
4183 /* Remove all children options and references
4184 * from bs->options and bs->explicit_options */
4185 QLIST_FOREACH(child, &bs->children, next) {
4186 char *child_key_dot;
4187 child_key_dot = g_strdup_printf("%s.", child->name);
4188 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4189 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4190 qdict_del(bs->explicit_options, child->name);
4191 qdict_del(bs->options, child->name);
4192 g_free(child_key_dot);
4195 /* Check if any unknown options were used */
4196 if (qdict_size(options) != 0) {
4197 const QDictEntry *entry = qdict_first(options);
4198 if (flags & BDRV_O_PROTOCOL) {
4199 error_setg(errp, "Block protocol '%s' doesn't support the option "
4200 "'%s'", drv->format_name, entry->key);
4203 "Block format '%s' does not support the option '%s'",
4204 drv->format_name, entry->key);
4207 goto close_and_fail;
4210 bdrv_parent_cb_change_media(bs, true);
4212 qobject_unref(options);
4215 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4216 * temporary snapshot afterwards. */
4217 if (snapshot_flags) {
4218 BlockDriverState *snapshot_bs;
4219 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4220 snapshot_options, &local_err);
4221 snapshot_options = NULL;
4223 goto close_and_fail;
4225 /* We are not going to return bs but the overlay on top of it
4226 * (snapshot_bs); thus, we have to drop the strong reference to bs
4227 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4228 * though, because the overlay still has a reference to it. */
4229 aio_context_acquire(ctx);
4231 aio_context_release(ctx);
4238 aio_context_acquire(ctx);
4240 qobject_unref(snapshot_options);
4241 qobject_unref(bs->explicit_options);
4242 qobject_unref(bs->options);
4243 qobject_unref(options);
4245 bs->explicit_options = NULL;
4247 aio_context_release(ctx);
4248 error_propagate(errp, local_err);
4252 aio_context_acquire(ctx);
4254 aio_context_release(ctx);
4255 qobject_unref(snapshot_options);
4256 qobject_unref(options);
4257 error_propagate(errp, local_err);
4261 /* The caller must always hold the main AioContext lock. */
4262 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4263 QDict *options, int flags, Error **errp)
4265 GLOBAL_STATE_CODE();
4267 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4271 /* Return true if the NULL-terminated @list contains @str */
4272 static bool is_str_in_list(const char *str, const char *const *list)
4276 for (i = 0; list[i] != NULL; i++) {
4277 if (!strcmp(str, list[i])) {
4286 * Check that every option set in @bs->options is also set in
4289 * Options listed in the common_options list and in
4290 * @bs->drv->mutable_opts are skipped.
4292 * Return 0 on success, otherwise return -EINVAL and set @errp.
4294 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4295 const QDict *new_opts, Error **errp)
4297 const QDictEntry *e;
4298 /* These options are common to all block drivers and are handled
4299 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4300 const char *const common_options[] = {
4301 "node-name", "discard", "cache.direct", "cache.no-flush",
4302 "read-only", "auto-read-only", "detect-zeroes", NULL
4305 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4306 if (!qdict_haskey(new_opts, e->key) &&
4307 !is_str_in_list(e->key, common_options) &&
4308 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4309 error_setg(errp, "Option '%s' cannot be reset "
4310 "to its default value", e->key);
4319 * Returns true if @child can be reached recursively from @bs
4321 static bool GRAPH_RDLOCK
4322 bdrv_recurse_has_child(BlockDriverState *bs, BlockDriverState *child)
4330 QLIST_FOREACH(c, &bs->children, next) {
4331 if (bdrv_recurse_has_child(c->bs, child)) {
4340 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4341 * reopen of multiple devices.
4343 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4344 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4345 * be created and initialized. This newly created BlockReopenQueue should be
4346 * passed back in for subsequent calls that are intended to be of the same
4349 * bs is the BlockDriverState to add to the reopen queue.
4351 * options contains the changed options for the associated bs
4352 * (the BlockReopenQueue takes ownership)
4354 * flags contains the open flags for the associated bs
4356 * returns a pointer to bs_queue, which is either the newly allocated
4357 * bs_queue, or the existing bs_queue being used.
4359 * bs is drained here and undrained by bdrv_reopen_queue_free().
4361 * To be called with bs->aio_context locked.
4363 static BlockReopenQueue * GRAPH_RDLOCK
4364 bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, BlockDriverState *bs,
4365 QDict *options, const BdrvChildClass *klass,
4366 BdrvChildRole role, bool parent_is_format,
4367 QDict *parent_options, int parent_flags,
4372 BlockReopenQueueEntry *bs_entry;
4374 QDict *old_options, *explicit_options, *options_copy;
4378 GLOBAL_STATE_CODE();
4381 * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that
4382 * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok
4385 bdrv_drained_begin(bs);
4387 if (bs_queue == NULL) {
4388 bs_queue = g_new0(BlockReopenQueue, 1);
4389 QTAILQ_INIT(bs_queue);
4393 options = qdict_new();
4396 /* Check if this BlockDriverState is already in the queue */
4397 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4398 if (bs == bs_entry->state.bs) {
4404 * Precedence of options:
4405 * 1. Explicitly passed in options (highest)
4406 * 2. Retained from explicitly set options of bs
4407 * 3. Inherited from parent node
4408 * 4. Retained from effective options of bs
4411 /* Old explicitly set values (don't overwrite by inherited value) */
4412 if (bs_entry || keep_old_opts) {
4413 old_options = qdict_clone_shallow(bs_entry ?
4414 bs_entry->state.explicit_options :
4415 bs->explicit_options);
4416 bdrv_join_options(bs, options, old_options);
4417 qobject_unref(old_options);
4420 explicit_options = qdict_clone_shallow(options);
4422 /* Inherit from parent node */
4423 if (parent_options) {
4425 klass->inherit_options(role, parent_is_format, &flags, options,
4426 parent_flags, parent_options);
4428 flags = bdrv_get_flags(bs);
4431 if (keep_old_opts) {
4432 /* Old values are used for options that aren't set yet */
4433 old_options = qdict_clone_shallow(bs->options);
4434 bdrv_join_options(bs, options, old_options);
4435 qobject_unref(old_options);
4438 /* We have the final set of options so let's update the flags */
4439 options_copy = qdict_clone_shallow(options);
4440 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4441 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4442 update_flags_from_options(&flags, opts);
4443 qemu_opts_del(opts);
4444 qobject_unref(options_copy);
4446 /* bdrv_open_inherit() sets and clears some additional flags internally */
4447 flags &= ~BDRV_O_PROTOCOL;
4448 if (flags & BDRV_O_RDWR) {
4449 flags |= BDRV_O_ALLOW_RDWR;
4453 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4454 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4456 qobject_unref(bs_entry->state.options);
4457 qobject_unref(bs_entry->state.explicit_options);
4460 bs_entry->state.bs = bs;
4461 bs_entry->state.options = options;
4462 bs_entry->state.explicit_options = explicit_options;
4463 bs_entry->state.flags = flags;
4466 * If keep_old_opts is false then it means that unspecified
4467 * options must be reset to their original value. We don't allow
4468 * resetting 'backing' but we need to know if the option is
4469 * missing in order to decide if we have to return an error.
4471 if (!keep_old_opts) {
4472 bs_entry->state.backing_missing =
4473 !qdict_haskey(options, "backing") &&
4474 !qdict_haskey(options, "backing.driver");
4477 QLIST_FOREACH(child, &bs->children, next) {
4478 QDict *new_child_options = NULL;
4479 bool child_keep_old = keep_old_opts;
4481 /* reopen can only change the options of block devices that were
4482 * implicitly created and inherited options. For other (referenced)
4483 * block devices, a syntax like "backing.foo" results in an error. */
4484 if (child->bs->inherits_from != bs) {
4488 /* Check if the options contain a child reference */
4489 if (qdict_haskey(options, child->name)) {
4490 const char *childref = qdict_get_try_str(options, child->name);
4492 * The current child must not be reopened if the child
4493 * reference is null or points to a different node.
4495 if (g_strcmp0(childref, child->bs->node_name)) {
4499 * If the child reference points to the current child then
4500 * reopen it with its existing set of options (note that
4501 * it can still inherit new options from the parent).
4503 child_keep_old = true;
4505 /* Extract child options ("child-name.*") */
4506 char *child_key_dot = g_strdup_printf("%s.", child->name);
4507 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4508 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4509 g_free(child_key_dot);
4512 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4513 child->klass, child->role, bs->drv->is_format,
4514 options, flags, child_keep_old);
4520 /* To be called with bs->aio_context locked */
4521 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4522 BlockDriverState *bs,
4523 QDict *options, bool keep_old_opts)
4525 GLOBAL_STATE_CODE();
4526 GRAPH_RDLOCK_GUARD_MAINLOOP();
4528 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4529 NULL, 0, keep_old_opts);
4532 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4534 GLOBAL_STATE_CODE();
4536 BlockReopenQueueEntry *bs_entry, *next;
4537 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4538 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4540 aio_context_acquire(ctx);
4541 bdrv_drained_end(bs_entry->state.bs);
4542 aio_context_release(ctx);
4544 qobject_unref(bs_entry->state.explicit_options);
4545 qobject_unref(bs_entry->state.options);
4553 * Reopen multiple BlockDriverStates atomically & transactionally.
4555 * The queue passed in (bs_queue) must have been built up previous
4556 * via bdrv_reopen_queue().
4558 * Reopens all BDS specified in the queue, with the appropriate
4559 * flags. All devices are prepared for reopen, and failure of any
4560 * device will cause all device changes to be abandoned, and intermediate
4563 * If all devices prepare successfully, then the changes are committed
4566 * All affected nodes must be drained between bdrv_reopen_queue() and
4567 * bdrv_reopen_multiple().
4569 * To be called from the main thread, with all other AioContexts unlocked.
4571 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4574 BlockReopenQueueEntry *bs_entry, *next;
4576 Transaction *tran = tran_new();
4577 g_autoptr(GSList) refresh_list = NULL;
4579 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4580 assert(bs_queue != NULL);
4581 GLOBAL_STATE_CODE();
4583 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4584 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4585 aio_context_acquire(ctx);
4586 ret = bdrv_flush(bs_entry->state.bs);
4587 aio_context_release(ctx);
4589 error_setg_errno(errp, -ret, "Error flushing drive");
4594 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4595 assert(bs_entry->state.bs->quiesce_counter > 0);
4596 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4597 aio_context_acquire(ctx);
4598 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4599 aio_context_release(ctx);
4603 bs_entry->prepared = true;
4606 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4607 BDRVReopenState *state = &bs_entry->state;
4609 refresh_list = g_slist_prepend(refresh_list, state->bs);
4610 if (state->old_backing_bs) {
4611 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4613 if (state->old_file_bs) {
4614 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4619 * Note that file-posix driver rely on permission update done during reopen
4620 * (even if no permission changed), because it wants "new" permissions for
4621 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4622 * in raw_reopen_prepare() which is called with "old" permissions.
4624 bdrv_graph_rdlock_main_loop();
4625 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4626 bdrv_graph_rdunlock_main_loop();
4633 * If we reach this point, we have success and just need to apply the
4636 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4637 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4638 * children are usually goes after parents in reopen-queue, so go from last
4641 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4642 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4643 aio_context_acquire(ctx);
4644 bdrv_reopen_commit(&bs_entry->state);
4645 aio_context_release(ctx);
4648 bdrv_graph_wrlock(NULL);
4650 bdrv_graph_wrunlock();
4652 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4653 BlockDriverState *bs = bs_entry->state.bs;
4655 if (bs->drv->bdrv_reopen_commit_post) {
4656 ctx = bdrv_get_aio_context(bs);
4657 aio_context_acquire(ctx);
4658 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4659 aio_context_release(ctx);
4667 bdrv_graph_wrlock(NULL);
4669 bdrv_graph_wrunlock();
4671 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4672 if (bs_entry->prepared) {
4673 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4674 aio_context_acquire(ctx);
4675 bdrv_reopen_abort(&bs_entry->state);
4676 aio_context_release(ctx);
4681 bdrv_reopen_queue_free(bs_queue);
4686 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4689 AioContext *ctx = bdrv_get_aio_context(bs);
4690 BlockReopenQueue *queue;
4693 GLOBAL_STATE_CODE();
4695 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4697 if (ctx != qemu_get_aio_context()) {
4698 aio_context_release(ctx);
4700 ret = bdrv_reopen_multiple(queue, errp);
4702 if (ctx != qemu_get_aio_context()) {
4703 aio_context_acquire(ctx);
4709 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4712 QDict *opts = qdict_new();
4714 GLOBAL_STATE_CODE();
4716 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4718 return bdrv_reopen(bs, opts, true, errp);
4722 * Take a BDRVReopenState and check if the value of 'backing' in the
4723 * reopen_state->options QDict is valid or not.
4725 * If 'backing' is missing from the QDict then return 0.
4727 * If 'backing' contains the node name of the backing file of
4728 * reopen_state->bs then return 0.
4730 * If 'backing' contains a different node name (or is null) then check
4731 * whether the current backing file can be replaced with the new one.
4732 * If that's the case then reopen_state->replace_backing_bs is set to
4733 * true and reopen_state->new_backing_bs contains a pointer to the new
4734 * backing BlockDriverState (or NULL).
4736 * After calling this function, the transaction @tran may only be completed
4737 * while holding a writer lock for the graph.
4739 * Return 0 on success, otherwise return < 0 and set @errp.
4741 * The caller must hold the AioContext lock of @reopen_state->bs.
4742 * @reopen_state->bs can move to a different AioContext in this function.
4743 * Callers must make sure that their AioContext locking is still correct after
4746 static int GRAPH_UNLOCKED
4747 bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4748 bool is_backing, Transaction *tran,
4751 BlockDriverState *bs = reopen_state->bs;
4752 BlockDriverState *new_child_bs;
4753 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4755 const char *child_name = is_backing ? "backing" : "file";
4758 AioContext *ctx, *old_ctx;
4762 GLOBAL_STATE_CODE();
4764 value = qdict_get(reopen_state->options, child_name);
4765 if (value == NULL) {
4769 bdrv_graph_rdlock_main_loop();
4771 switch (qobject_type(value)) {
4773 assert(is_backing); /* The 'file' option does not allow a null value */
4774 new_child_bs = NULL;
4777 str = qstring_get_str(qobject_to(QString, value));
4778 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4779 if (new_child_bs == NULL) {
4784 has_child = bdrv_recurse_has_child(new_child_bs, bs);
4786 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4787 "cycle", str, child_name, bs->node_name);
4794 * The options QDict has been flattened, so 'backing' and 'file'
4795 * do not allow any other data type here.
4797 g_assert_not_reached();
4800 if (old_child_bs == new_child_bs) {
4806 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4811 if (old_child_bs->implicit) {
4812 error_setg(errp, "Cannot replace implicit %s child of %s",
4813 child_name, bs->node_name);
4819 if (bs->drv->is_filter && !old_child_bs) {
4821 * Filters always have a file or a backing child, so we are trying to
4822 * change wrong child
4824 error_setg(errp, "'%s' is a %s filter node that does not support a "
4825 "%s child", bs->node_name, bs->drv->format_name, child_name);
4831 reopen_state->old_backing_bs = old_child_bs;
4833 reopen_state->old_file_bs = old_child_bs;
4837 bdrv_ref(old_child_bs);
4838 bdrv_drained_begin(old_child_bs);
4841 old_ctx = bdrv_get_aio_context(bs);
4842 ctx = bdrv_get_aio_context(new_child_bs);
4843 if (old_ctx != ctx) {
4844 aio_context_release(old_ctx);
4845 aio_context_acquire(ctx);
4848 bdrv_graph_rdunlock_main_loop();
4849 bdrv_graph_wrlock(new_child_bs);
4851 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4854 bdrv_graph_wrunlock();
4856 if (old_ctx != ctx) {
4857 aio_context_release(ctx);
4858 aio_context_acquire(old_ctx);
4862 bdrv_drained_end(old_child_bs);
4863 bdrv_unref(old_child_bs);
4869 bdrv_graph_rdunlock_main_loop();
4874 * Prepares a BlockDriverState for reopen. All changes are staged in the
4875 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4876 * the block driver layer .bdrv_reopen_prepare()
4878 * bs is the BlockDriverState to reopen
4879 * flags are the new open flags
4880 * queue is the reopen queue
4882 * Returns 0 on success, non-zero on error. On error errp will be set
4885 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4886 * It is the responsibility of the caller to then call the abort() or
4887 * commit() for any other BDS that have been left in a prepare() state
4889 * The caller must hold the AioContext lock of @reopen_state->bs.
4891 * After calling this function, the transaction @change_child_tran may only be
4892 * completed while holding a writer lock for the graph.
4894 static int GRAPH_UNLOCKED
4895 bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
4896 Transaction *change_child_tran, Error **errp)
4900 Error *local_err = NULL;
4903 QDict *orig_reopen_opts;
4904 char *discard = NULL;
4906 bool drv_prepared = false;
4908 assert(reopen_state != NULL);
4909 assert(reopen_state->bs->drv != NULL);
4910 GLOBAL_STATE_CODE();
4911 drv = reopen_state->bs->drv;
4913 /* This function and each driver's bdrv_reopen_prepare() remove
4914 * entries from reopen_state->options as they are processed, so
4915 * we need to make a copy of the original QDict. */
4916 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4918 /* Process generic block layer options */
4919 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4920 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4925 /* This was already called in bdrv_reopen_queue_child() so the flags
4926 * are up-to-date. This time we simply want to remove the options from
4927 * QemuOpts in order to indicate that they have been processed. */
4928 old_flags = reopen_state->flags;
4929 update_flags_from_options(&reopen_state->flags, opts);
4930 assert(old_flags == reopen_state->flags);
4932 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4933 if (discard != NULL) {
4934 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4935 error_setg(errp, "Invalid discard option");
4941 reopen_state->detect_zeroes =
4942 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4944 error_propagate(errp, local_err);
4949 /* All other options (including node-name and driver) must be unchanged.
4950 * Put them back into the QDict, so that they are checked at the end
4951 * of this function. */
4952 qemu_opts_to_qdict(opts, reopen_state->options);
4954 /* If we are to stay read-only, do not allow permission change
4955 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4956 * not set, or if the BDS still has copy_on_read enabled */
4957 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4959 bdrv_graph_rdlock_main_loop();
4960 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4961 bdrv_graph_rdunlock_main_loop();
4963 error_propagate(errp, local_err);
4967 if (drv->bdrv_reopen_prepare) {
4969 * If a driver-specific option is missing, it means that we
4970 * should reset it to its default value.
4971 * But not all options allow that, so we need to check it first.
4973 ret = bdrv_reset_options_allowed(reopen_state->bs,
4974 reopen_state->options, errp);
4979 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4981 if (local_err != NULL) {
4982 error_propagate(errp, local_err);
4984 bdrv_graph_rdlock_main_loop();
4985 bdrv_refresh_filename(reopen_state->bs);
4986 bdrv_graph_rdunlock_main_loop();
4987 error_setg(errp, "failed while preparing to reopen image '%s'",
4988 reopen_state->bs->filename);
4993 /* It is currently mandatory to have a bdrv_reopen_prepare()
4994 * handler for each supported drv. */
4995 bdrv_graph_rdlock_main_loop();
4996 error_setg(errp, "Block format '%s' used by node '%s' "
4997 "does not support reopening files", drv->format_name,
4998 bdrv_get_device_or_node_name(reopen_state->bs));
4999 bdrv_graph_rdunlock_main_loop();
5004 drv_prepared = true;
5007 * We must provide the 'backing' option if the BDS has a backing
5008 * file or if the image file has a backing file name as part of
5009 * its metadata. Otherwise the 'backing' option can be omitted.
5011 if (drv->supports_backing && reopen_state->backing_missing &&
5012 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
5013 error_setg(errp, "backing is missing for '%s'",
5014 reopen_state->bs->node_name);
5020 * Allow changing the 'backing' option. The new value can be
5021 * either a reference to an existing node (using its node name)
5022 * or NULL to simply detach the current backing file.
5024 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
5025 change_child_tran, errp);
5029 qdict_del(reopen_state->options, "backing");
5031 /* Allow changing the 'file' option. In this case NULL is not allowed */
5032 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
5033 change_child_tran, errp);
5037 qdict_del(reopen_state->options, "file");
5039 /* Options that are not handled are only okay if they are unchanged
5040 * compared to the old state. It is expected that some options are only
5041 * used for the initial open, but not reopen (e.g. filename) */
5042 if (qdict_size(reopen_state->options)) {
5043 const QDictEntry *entry = qdict_first(reopen_state->options);
5045 GRAPH_RDLOCK_GUARD_MAINLOOP();
5048 QObject *new = entry->value;
5049 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
5051 /* Allow child references (child_name=node_name) as long as they
5052 * point to the current child (i.e. everything stays the same). */
5053 if (qobject_type(new) == QTYPE_QSTRING) {
5055 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
5056 if (!strcmp(child->name, entry->key)) {
5062 if (!strcmp(child->bs->node_name,
5063 qstring_get_str(qobject_to(QString, new)))) {
5064 continue; /* Found child with this name, skip option */
5070 * TODO: When using -drive to specify blockdev options, all values
5071 * will be strings; however, when using -blockdev, blockdev-add or
5072 * filenames using the json:{} pseudo-protocol, they will be
5074 * In contrast, reopening options are (currently) always strings
5075 * (because you can only specify them through qemu-io; all other
5076 * callers do not specify any options).
5077 * Therefore, when using anything other than -drive to create a BDS,
5078 * this cannot detect non-string options as unchanged, because
5079 * qobject_is_equal() always returns false for objects of different
5080 * type. In the future, this should be remedied by correctly typing
5081 * all options. For now, this is not too big of an issue because
5082 * the user can simply omit options which cannot be changed anyway,
5083 * so they will stay unchanged.
5085 if (!qobject_is_equal(new, old)) {
5086 error_setg(errp, "Cannot change the option '%s'", entry->key);
5090 } while ((entry = qdict_next(reopen_state->options, entry)));
5095 /* Restore the original reopen_state->options QDict */
5096 qobject_unref(reopen_state->options);
5097 reopen_state->options = qobject_ref(orig_reopen_opts);
5100 if (ret < 0 && drv_prepared) {
5101 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5102 * call drv->bdrv_reopen_abort() before signaling an error
5103 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5104 * when the respective bdrv_reopen_prepare() has failed) */
5105 if (drv->bdrv_reopen_abort) {
5106 drv->bdrv_reopen_abort(reopen_state);
5109 qemu_opts_del(opts);
5110 qobject_unref(orig_reopen_opts);
5116 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5117 * makes them final by swapping the staging BlockDriverState contents into
5118 * the active BlockDriverState contents.
5120 static void GRAPH_UNLOCKED bdrv_reopen_commit(BDRVReopenState *reopen_state)
5123 BlockDriverState *bs;
5126 assert(reopen_state != NULL);
5127 bs = reopen_state->bs;
5129 assert(drv != NULL);
5130 GLOBAL_STATE_CODE();
5132 /* If there are any driver level actions to take */
5133 if (drv->bdrv_reopen_commit) {
5134 drv->bdrv_reopen_commit(reopen_state);
5137 GRAPH_RDLOCK_GUARD_MAINLOOP();
5139 /* set BDS specific flags now */
5140 qobject_unref(bs->explicit_options);
5141 qobject_unref(bs->options);
5142 qobject_ref(reopen_state->explicit_options);
5143 qobject_ref(reopen_state->options);
5145 bs->explicit_options = reopen_state->explicit_options;
5146 bs->options = reopen_state->options;
5147 bs->open_flags = reopen_state->flags;
5148 bs->detect_zeroes = reopen_state->detect_zeroes;
5150 /* Remove child references from bs->options and bs->explicit_options.
5151 * Child options were already removed in bdrv_reopen_queue_child() */
5152 QLIST_FOREACH(child, &bs->children, next) {
5153 qdict_del(bs->explicit_options, child->name);
5154 qdict_del(bs->options, child->name);
5156 /* backing is probably removed, so it's not handled by previous loop */
5157 qdict_del(bs->explicit_options, "backing");
5158 qdict_del(bs->options, "backing");
5160 bdrv_refresh_limits(bs, NULL, NULL);
5161 bdrv_refresh_total_sectors(bs, bs->total_sectors);
5165 * Abort the reopen, and delete and free the staged changes in
5168 static void GRAPH_UNLOCKED bdrv_reopen_abort(BDRVReopenState *reopen_state)
5172 assert(reopen_state != NULL);
5173 drv = reopen_state->bs->drv;
5174 assert(drv != NULL);
5175 GLOBAL_STATE_CODE();
5177 if (drv->bdrv_reopen_abort) {
5178 drv->bdrv_reopen_abort(reopen_state);
5183 static void bdrv_close(BlockDriverState *bs)
5185 BdrvAioNotifier *ban, *ban_next;
5186 BdrvChild *child, *next;
5188 GLOBAL_STATE_CODE();
5189 assert(!bs->refcnt);
5191 bdrv_drained_begin(bs); /* complete I/O */
5193 bdrv_drain(bs); /* in case flush left pending I/O */
5196 if (bs->drv->bdrv_close) {
5197 /* Must unfreeze all children, so bdrv_unref_child() works */
5198 bs->drv->bdrv_close(bs);
5203 bdrv_graph_wrlock(bs);
5204 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
5205 bdrv_unref_child(bs, child);
5207 bdrv_graph_wrunlock();
5209 assert(!bs->backing);
5213 qatomic_set(&bs->copy_on_read, 0);
5214 bs->backing_file[0] = '\0';
5215 bs->backing_format[0] = '\0';
5216 bs->total_sectors = 0;
5217 bs->encrypted = false;
5219 qobject_unref(bs->options);
5220 qobject_unref(bs->explicit_options);
5222 bs->explicit_options = NULL;
5223 qobject_unref(bs->full_open_options);
5224 bs->full_open_options = NULL;
5225 g_free(bs->block_status_cache);
5226 bs->block_status_cache = NULL;
5228 bdrv_release_named_dirty_bitmaps(bs);
5229 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5231 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5234 QLIST_INIT(&bs->aio_notifiers);
5235 bdrv_drained_end(bs);
5238 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5239 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5242 if (bs->quiesce_counter) {
5243 bdrv_drain_all_end_quiesce(bs);
5247 void bdrv_close_all(void)
5249 GLOBAL_STATE_CODE();
5250 assert(job_next(NULL) == NULL);
5252 /* Drop references from requests still in flight, such as canceled block
5253 * jobs whose AIO context has not been polled yet */
5256 blk_remove_all_bs();
5257 blockdev_close_all_bdrv_states();
5259 assert(QTAILQ_EMPTY(&all_bdrv_states));
5262 static bool GRAPH_RDLOCK should_update_child(BdrvChild *c, BlockDriverState *to)
5268 if (c->klass->stay_at_node) {
5272 /* If the child @c belongs to the BDS @to, replacing the current
5273 * c->bs by @to would mean to create a loop.
5275 * Such a case occurs when appending a BDS to a backing chain.
5276 * For instance, imagine the following chain:
5278 * guest device -> node A -> further backing chain...
5280 * Now we create a new BDS B which we want to put on top of this
5281 * chain, so we first attach A as its backing node:
5286 * guest device -> node A -> further backing chain...
5288 * Finally we want to replace A by B. When doing that, we want to
5289 * replace all pointers to A by pointers to B -- except for the
5290 * pointer from B because (1) that would create a loop, and (2)
5291 * that pointer should simply stay intact:
5293 * guest device -> node B
5296 * node A -> further backing chain...
5298 * In general, when replacing a node A (c->bs) by a node B (@to),
5299 * if A is a child of B, that means we cannot replace A by B there
5300 * because that would create a loop. Silently detaching A from B
5301 * is also not really an option. So overall just leaving A in
5302 * place there is the most sensible choice.
5304 * We would also create a loop in any cases where @c is only
5305 * indirectly referenced by @to. Prevent this by returning false
5306 * if @c is found (by breadth-first search) anywhere in the whole
5311 found = g_hash_table_new(NULL, NULL);
5312 g_hash_table_add(found, to);
5313 queue = g_queue_new();
5314 g_queue_push_tail(queue, to);
5316 while (!g_queue_is_empty(queue)) {
5317 BlockDriverState *v = g_queue_pop_head(queue);
5320 QLIST_FOREACH(c2, &v->children, next) {
5326 if (g_hash_table_contains(found, c2->bs)) {
5330 g_queue_push_tail(queue, c2->bs);
5331 g_hash_table_add(found, c2->bs);
5335 g_queue_free(queue);
5336 g_hash_table_destroy(found);
5341 static void bdrv_remove_child_commit(void *opaque)
5343 GLOBAL_STATE_CODE();
5344 bdrv_child_free(opaque);
5347 static TransactionActionDrv bdrv_remove_child_drv = {
5348 .commit = bdrv_remove_child_commit,
5352 * Function doesn't update permissions, caller is responsible for this.
5354 * @child->bs (if non-NULL) must be drained.
5356 * After calling this function, the transaction @tran may only be completed
5357 * while holding a writer lock for the graph.
5359 static void GRAPH_WRLOCK bdrv_remove_child(BdrvChild *child, Transaction *tran)
5366 assert(child->quiesced_parent);
5367 bdrv_replace_child_tran(child, NULL, tran);
5370 tran_add(tran, &bdrv_remove_child_drv, child);
5374 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5375 * until the transaction is completed.
5377 * After calling this function, the transaction @tran may only be completed
5378 * while holding a writer lock for the graph.
5380 static int GRAPH_WRLOCK
5381 bdrv_replace_node_noperm(BlockDriverState *from,
5382 BlockDriverState *to,
5383 bool auto_skip, Transaction *tran,
5386 BdrvChild *c, *next;
5388 GLOBAL_STATE_CODE();
5390 assert(from->quiesce_counter);
5391 assert(to->quiesce_counter);
5393 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5394 assert(c->bs == from);
5395 if (!should_update_child(c, to)) {
5399 error_setg(errp, "Should not change '%s' link to '%s'",
5400 c->name, from->node_name);
5404 error_setg(errp, "Cannot change '%s' link to '%s'",
5405 c->name, from->node_name);
5408 bdrv_replace_child_tran(c, to, tran);
5415 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5416 * if it creates a parent-child relation loop or if parent is block-job.
5418 * With auto_skip=false the error is returned if from has a parent which should
5421 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5422 * case backing link of the cow-parent of @to is removed.
5424 static int bdrv_replace_node_common(BlockDriverState *from,
5425 BlockDriverState *to,
5426 bool auto_skip, bool detach_subchain,
5429 Transaction *tran = tran_new();
5430 g_autoptr(GSList) refresh_list = NULL;
5431 BlockDriverState *to_cow_parent = NULL;
5434 GLOBAL_STATE_CODE();
5436 /* Make sure that @from doesn't go away until we have successfully attached
5437 * all of its parents to @to. */
5440 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5441 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5442 bdrv_drained_begin(from);
5443 bdrv_drained_begin(to);
5445 bdrv_graph_wrlock(to);
5447 if (detach_subchain) {
5448 assert(bdrv_chain_contains(from, to));
5450 for (to_cow_parent = from;
5451 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5452 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5459 * Do the replacement without permission update.
5460 * Replacement may influence the permissions, we should calculate new
5461 * permissions based on new graph. If we fail, we'll roll-back the
5464 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5469 if (detach_subchain) {
5470 /* to_cow_parent is already drained because from is drained */
5471 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5474 refresh_list = g_slist_prepend(refresh_list, to);
5475 refresh_list = g_slist_prepend(refresh_list, from);
5477 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5485 tran_finalize(tran, ret);
5486 bdrv_graph_wrunlock();
5488 bdrv_drained_end(to);
5489 bdrv_drained_end(from);
5495 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5498 GLOBAL_STATE_CODE();
5500 return bdrv_replace_node_common(from, to, true, false, errp);
5503 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5505 BlockDriverState *child_bs;
5507 GLOBAL_STATE_CODE();
5508 bdrv_graph_rdlock_main_loop();
5509 child_bs = bdrv_filter_or_cow_bs(bs);
5510 bdrv_graph_rdunlock_main_loop();
5512 return bdrv_replace_node_common(bs, child_bs, true, true, errp);
5516 * Add new bs contents at the top of an image chain while the chain is
5517 * live, while keeping required fields on the top layer.
5519 * This will modify the BlockDriverState fields, and swap contents
5520 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5522 * bs_new must not be attached to a BlockBackend and must not have backing
5525 * This function does not create any image files.
5527 * The caller must hold the AioContext lock for @bs_top.
5529 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5534 Transaction *tran = tran_new();
5535 AioContext *old_context, *new_context = NULL;
5537 GLOBAL_STATE_CODE();
5539 assert(!bs_new->backing);
5541 old_context = bdrv_get_aio_context(bs_top);
5542 bdrv_drained_begin(bs_top);
5545 * bdrv_drained_begin() requires that only the AioContext of the drained
5546 * node is locked, and at this point it can still differ from the AioContext
5549 new_context = bdrv_get_aio_context(bs_new);
5550 aio_context_release(old_context);
5551 aio_context_acquire(new_context);
5552 bdrv_drained_begin(bs_new);
5553 aio_context_release(new_context);
5554 aio_context_acquire(old_context);
5557 bdrv_graph_wrlock(bs_top);
5559 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5560 &child_of_bds, bdrv_backing_role(bs_new),
5568 * bdrv_attach_child_noperm could change the AioContext of bs_top and
5569 * bs_new, but at least they are in the same AioContext now. This is the
5570 * AioContext that we need to lock for the rest of the function.
5572 new_context = bdrv_get_aio_context(bs_top);
5574 if (old_context != new_context) {
5575 aio_context_release(old_context);
5576 aio_context_acquire(new_context);
5579 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5584 ret = bdrv_refresh_perms(bs_new, tran, errp);
5586 tran_finalize(tran, ret);
5588 bdrv_refresh_limits(bs_top, NULL, NULL);
5589 bdrv_graph_wrunlock();
5591 bdrv_drained_end(bs_top);
5592 bdrv_drained_end(bs_new);
5594 if (new_context && old_context != new_context) {
5595 aio_context_release(new_context);
5596 aio_context_acquire(old_context);
5602 /* Not for empty child */
5603 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5607 Transaction *tran = tran_new();
5608 g_autoptr(GSList) refresh_list = NULL;
5609 BlockDriverState *old_bs = child->bs;
5611 GLOBAL_STATE_CODE();
5614 bdrv_drained_begin(old_bs);
5615 bdrv_drained_begin(new_bs);
5616 bdrv_graph_wrlock(new_bs);
5618 bdrv_replace_child_tran(child, new_bs, tran);
5620 refresh_list = g_slist_prepend(refresh_list, old_bs);
5621 refresh_list = g_slist_prepend(refresh_list, new_bs);
5623 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5625 tran_finalize(tran, ret);
5627 bdrv_graph_wrunlock();
5628 bdrv_drained_end(old_bs);
5629 bdrv_drained_end(new_bs);
5635 static void bdrv_delete(BlockDriverState *bs)
5637 assert(bdrv_op_blocker_is_empty(bs));
5638 assert(!bs->refcnt);
5639 GLOBAL_STATE_CODE();
5641 /* remove from list, if necessary */
5642 if (bs->node_name[0] != '\0') {
5643 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5645 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5649 qemu_mutex_destroy(&bs->reqs_lock);
5656 * Replace @bs by newly created block node.
5658 * @options is a QDict of options to pass to the block drivers, or NULL for an
5659 * empty set of options. The reference to the QDict belongs to the block layer
5660 * after the call (even on failure), so if the caller intends to reuse the
5661 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5663 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5664 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5665 * different AioContext.
5667 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5668 int flags, Error **errp)
5672 AioContext *ctx = bdrv_get_aio_context(bs);
5673 BlockDriverState *new_node_bs = NULL;
5674 const char *drvname, *node_name;
5677 drvname = qdict_get_try_str(options, "driver");
5679 error_setg(errp, "driver is not specified");
5683 drv = bdrv_find_format(drvname);
5685 error_setg(errp, "Unknown driver: '%s'", drvname);
5689 node_name = qdict_get_try_str(options, "node-name");
5691 GLOBAL_STATE_CODE();
5693 aio_context_release(ctx);
5694 aio_context_acquire(qemu_get_aio_context());
5695 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5697 aio_context_release(qemu_get_aio_context());
5698 aio_context_acquire(ctx);
5699 assert(bdrv_get_aio_context(bs) == ctx);
5701 options = NULL; /* bdrv_new_open_driver() eats options */
5703 error_prepend(errp, "Could not create node: ");
5707 bdrv_drained_begin(bs);
5708 ret = bdrv_replace_node(bs, new_node_bs, errp);
5709 bdrv_drained_end(bs);
5712 error_prepend(errp, "Could not replace node: ");
5719 qobject_unref(options);
5720 bdrv_unref(new_node_bs);
5725 * Run consistency checks on an image
5727 * Returns 0 if the check could be completed (it doesn't mean that the image is
5728 * free of errors) or -errno when an internal error occurred. The results of the
5729 * check are stored in res.
5731 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5732 BdrvCheckResult *res, BdrvCheckMode fix)
5735 assert_bdrv_graph_readable();
5736 if (bs->drv == NULL) {
5739 if (bs->drv->bdrv_co_check == NULL) {
5743 memset(res, 0, sizeof(*res));
5744 return bs->drv->bdrv_co_check(bs, res, fix);
5750 * -EINVAL - backing format specified, but no file
5751 * -ENOSPC - can't update the backing file because no space is left in the
5753 * -ENOTSUP - format driver doesn't support changing the backing file
5755 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5756 const char *backing_fmt, bool require)
5758 BlockDriver *drv = bs->drv;
5761 GLOBAL_STATE_CODE();
5767 /* Backing file format doesn't make sense without a backing file */
5768 if (backing_fmt && !backing_file) {
5772 if (require && backing_file && !backing_fmt) {
5776 if (drv->bdrv_change_backing_file != NULL) {
5777 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5783 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5784 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5785 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5786 backing_file ?: "");
5792 * Finds the first non-filter node above bs in the chain between
5793 * active and bs. The returned node is either an immediate parent of
5794 * bs, or there are only filter nodes between the two.
5796 * Returns NULL if bs is not found in active's image chain,
5797 * or if active == bs.
5799 * Returns the bottommost base image if bs == NULL.
5801 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5802 BlockDriverState *bs)
5805 GLOBAL_STATE_CODE();
5807 bs = bdrv_skip_filters(bs);
5808 active = bdrv_skip_filters(active);
5811 BlockDriverState *next = bdrv_backing_chain_next(active);
5821 /* Given a BDS, searches for the base layer. */
5822 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5824 GLOBAL_STATE_CODE();
5826 return bdrv_find_overlay(bs, NULL);
5830 * Return true if at least one of the COW (backing) and filter links
5831 * between @bs and @base is frozen. @errp is set if that's the case.
5832 * @base must be reachable from @bs, or NULL.
5834 static bool GRAPH_RDLOCK
5835 bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5838 BlockDriverState *i;
5841 GLOBAL_STATE_CODE();
5843 for (i = bs; i != base; i = child_bs(child)) {
5844 child = bdrv_filter_or_cow_child(i);
5846 if (child && child->frozen) {
5847 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5848 child->name, i->node_name, child->bs->node_name);
5857 * Freeze all COW (backing) and filter links between @bs and @base.
5858 * If any of the links is already frozen the operation is aborted and
5859 * none of the links are modified.
5860 * @base must be reachable from @bs, or NULL.
5861 * Returns 0 on success. On failure returns < 0 and sets @errp.
5863 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5866 BlockDriverState *i;
5869 GLOBAL_STATE_CODE();
5871 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5875 for (i = bs; i != base; i = child_bs(child)) {
5876 child = bdrv_filter_or_cow_child(i);
5877 if (child && child->bs->never_freeze) {
5878 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5879 child->name, child->bs->node_name);
5884 for (i = bs; i != base; i = child_bs(child)) {
5885 child = bdrv_filter_or_cow_child(i);
5887 child->frozen = true;
5895 * Unfreeze all COW (backing) and filter links between @bs and @base.
5896 * The caller must ensure that all links are frozen before using this
5898 * @base must be reachable from @bs, or NULL.
5900 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5902 BlockDriverState *i;
5905 GLOBAL_STATE_CODE();
5907 for (i = bs; i != base; i = child_bs(child)) {
5908 child = bdrv_filter_or_cow_child(i);
5910 assert(child->frozen);
5911 child->frozen = false;
5917 * Drops images above 'base' up to and including 'top', and sets the image
5918 * above 'top' to have base as its backing file.
5920 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5921 * information in 'bs' can be properly updated.
5923 * E.g., this will convert the following chain:
5924 * bottom <- base <- intermediate <- top <- active
5928 * bottom <- base <- active
5930 * It is allowed for bottom==base, in which case it converts:
5932 * base <- intermediate <- top <- active
5938 * If backing_file_str is non-NULL, it will be used when modifying top's
5939 * overlay image metadata.
5942 * if active == top, that is considered an error
5945 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5946 const char *backing_file_str)
5948 BlockDriverState *explicit_top = top;
5949 bool update_inherits_from;
5951 Error *local_err = NULL;
5953 g_autoptr(GSList) updated_children = NULL;
5956 GLOBAL_STATE_CODE();
5959 bdrv_drained_begin(base);
5960 bdrv_graph_rdlock_main_loop();
5962 if (!top->drv || !base->drv) {
5966 /* Make sure that base is in the backing chain of top */
5967 if (!bdrv_chain_contains(top, base)) {
5971 /* If 'base' recursively inherits from 'top' then we should set
5972 * base->inherits_from to top->inherits_from after 'top' and all
5973 * other intermediate nodes have been dropped.
5974 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5975 * it because no one inherits from it. We use explicit_top for that. */
5976 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5977 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5979 /* success - we can delete the intermediate states, and link top->base */
5980 if (!backing_file_str) {
5981 bdrv_refresh_filename(base);
5982 backing_file_str = base->filename;
5985 QLIST_FOREACH(c, &top->parents, next_parent) {
5986 updated_children = g_slist_prepend(updated_children, c);
5990 * It seems correct to pass detach_subchain=true here, but it triggers
5991 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5992 * another drained section, which modify the graph (for example, removing
5993 * the child, which we keep in updated_children list). So, it's a TODO.
5995 * Note, bug triggered if pass detach_subchain=true here and run
5996 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5999 bdrv_replace_node_common(top, base, false, false, &local_err);
6001 error_report_err(local_err);
6005 for (p = updated_children; p; p = p->next) {
6008 if (c->klass->update_filename) {
6009 ret = c->klass->update_filename(c, base, backing_file_str,
6013 * TODO: Actually, we want to rollback all previous iterations
6014 * of this loop, and (which is almost impossible) previous
6015 * bdrv_replace_node()...
6017 * Note, that c->klass->update_filename may lead to permission
6018 * update, so it's a bad idea to call it inside permission
6019 * update transaction of bdrv_replace_node.
6021 error_report_err(local_err);
6027 if (update_inherits_from) {
6028 base->inherits_from = explicit_top->inherits_from;
6033 bdrv_graph_rdunlock_main_loop();
6034 bdrv_drained_end(base);
6040 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
6041 * sums the size of all data-bearing children. (This excludes backing
6044 static int64_t coroutine_fn GRAPH_RDLOCK
6045 bdrv_sum_allocated_file_size(BlockDriverState *bs)
6048 int64_t child_size, sum = 0;
6050 QLIST_FOREACH(child, &bs->children, next) {
6051 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
6052 BDRV_CHILD_FILTERED))
6054 child_size = bdrv_co_get_allocated_file_size(child->bs);
6055 if (child_size < 0) {
6066 * Length of a allocated file in bytes. Sparse files are counted by actual
6067 * allocated space. Return < 0 if error or unknown.
6069 int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
6071 BlockDriver *drv = bs->drv;
6073 assert_bdrv_graph_readable();
6078 if (drv->bdrv_co_get_allocated_file_size) {
6079 return drv->bdrv_co_get_allocated_file_size(bs);
6082 if (drv->bdrv_file_open) {
6084 * Protocol drivers default to -ENOTSUP (most of their data is
6085 * not stored in any of their children (if they even have any),
6086 * so there is no generic way to figure it out).
6089 } else if (drv->is_filter) {
6090 /* Filter drivers default to the size of their filtered child */
6091 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
6093 /* Other drivers default to summing their children's sizes */
6094 return bdrv_sum_allocated_file_size(bs);
6100 * @drv: Format driver
6101 * @opts: Creation options for new image
6102 * @in_bs: Existing image containing data for new image (may be NULL)
6103 * @errp: Error object
6104 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6107 * Calculate file size required to create a new image.
6109 * If @in_bs is given then space for allocated clusters and zero clusters
6110 * from that image are included in the calculation. If @opts contains a
6111 * backing file that is shared by @in_bs then backing clusters may be omitted
6112 * from the calculation.
6114 * If @in_bs is NULL then the calculation includes no allocated clusters
6115 * unless a preallocation option is given in @opts.
6117 * Note that @in_bs may use a different BlockDriver from @drv.
6119 * If an error occurs the @errp pointer is set.
6121 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
6122 BlockDriverState *in_bs, Error **errp)
6125 if (!drv->bdrv_measure) {
6126 error_setg(errp, "Block driver '%s' does not support size measurement",
6131 return drv->bdrv_measure(opts, in_bs, errp);
6135 * Return number of sectors on success, -errno on error.
6137 int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
6139 BlockDriver *drv = bs->drv;
6141 assert_bdrv_graph_readable();
6146 if (bs->bl.has_variable_length) {
6147 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
6152 return bs->total_sectors;
6156 * This wrapper is written by hand because this function is in the hot I/O path,
6157 * via blk_get_geometry.
6159 int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
6161 BlockDriver *drv = bs->drv;
6167 if (bs->bl.has_variable_length) {
6168 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6174 return bs->total_sectors;
6178 * Return length in bytes on success, -errno on error.
6179 * The length is always a multiple of BDRV_SECTOR_SIZE.
6181 int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
6185 assert_bdrv_graph_readable();
6187 ret = bdrv_co_nb_sectors(bs);
6191 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6194 return ret * BDRV_SECTOR_SIZE;
6197 bool bdrv_is_sg(BlockDriverState *bs)
6204 * Return whether the given node supports compressed writes.
6206 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6208 BlockDriverState *filtered;
6211 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6215 filtered = bdrv_filter_bs(bs);
6218 * Filters can only forward compressed writes, so we have to
6221 return bdrv_supports_compressed_writes(filtered);
6227 const char *bdrv_get_format_name(BlockDriverState *bs)
6230 return bs->drv ? bs->drv->format_name : NULL;
6233 static int qsort_strcmp(const void *a, const void *b)
6235 return strcmp(*(char *const *)a, *(char *const *)b);
6238 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
6239 void *opaque, bool read_only)
6244 const char **formats = NULL;
6246 GLOBAL_STATE_CODE();
6248 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6249 if (drv->format_name) {
6252 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6257 while (formats && i && !found) {
6258 found = !strcmp(formats[--i], drv->format_name);
6262 formats = g_renew(const char *, formats, count + 1);
6263 formats[count++] = drv->format_name;
6268 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6269 const char *format_name = block_driver_modules[i].format_name;
6275 if (use_bdrv_whitelist &&
6276 !bdrv_format_is_whitelisted(format_name, read_only)) {
6280 while (formats && j && !found) {
6281 found = !strcmp(formats[--j], format_name);
6285 formats = g_renew(const char *, formats, count + 1);
6286 formats[count++] = format_name;
6291 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6293 for (i = 0; i < count; i++) {
6294 it(opaque, formats[i]);
6300 /* This function is to find a node in the bs graph */
6301 BlockDriverState *bdrv_find_node(const char *node_name)
6303 BlockDriverState *bs;
6306 GLOBAL_STATE_CODE();
6308 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6309 if (!strcmp(node_name, bs->node_name)) {
6316 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6317 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6320 BlockDeviceInfoList *list;
6321 BlockDriverState *bs;
6323 GLOBAL_STATE_CODE();
6324 GRAPH_RDLOCK_GUARD_MAINLOOP();
6327 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6328 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6330 qapi_free_BlockDeviceInfoList(list);
6333 QAPI_LIST_PREPEND(list, info);
6339 typedef struct XDbgBlockGraphConstructor {
6340 XDbgBlockGraph *graph;
6341 GHashTable *graph_nodes;
6342 } XDbgBlockGraphConstructor;
6344 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6346 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6348 gr->graph = g_new0(XDbgBlockGraph, 1);
6349 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6354 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6356 XDbgBlockGraph *graph = gr->graph;
6358 g_hash_table_destroy(gr->graph_nodes);
6364 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6366 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6373 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6374 * answer of g_hash_table_lookup.
6376 ret = g_hash_table_size(gr->graph_nodes) + 1;
6377 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6382 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6383 XDbgBlockGraphNodeType type, const char *name)
6385 XDbgBlockGraphNode *n;
6387 n = g_new0(XDbgBlockGraphNode, 1);
6389 n->id = xdbg_graph_node_num(gr, node);
6391 n->name = g_strdup(name);
6393 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6396 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6397 const BdrvChild *child)
6399 BlockPermission qapi_perm;
6400 XDbgBlockGraphEdge *edge;
6401 GLOBAL_STATE_CODE();
6403 edge = g_new0(XDbgBlockGraphEdge, 1);
6405 edge->parent = xdbg_graph_node_num(gr, parent);
6406 edge->child = xdbg_graph_node_num(gr, child->bs);
6407 edge->name = g_strdup(child->name);
6409 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6410 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6412 if (flag & child->perm) {
6413 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6415 if (flag & child->shared_perm) {
6416 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6420 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6424 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6428 BlockDriverState *bs;
6430 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6432 GLOBAL_STATE_CODE();
6434 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6435 char *allocated_name = NULL;
6436 const char *name = blk_name(blk);
6439 name = allocated_name = blk_get_attached_dev_id(blk);
6441 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6443 g_free(allocated_name);
6444 if (blk_root(blk)) {
6445 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6449 WITH_JOB_LOCK_GUARD() {
6450 for (job = block_job_next_locked(NULL); job;
6451 job = block_job_next_locked(job)) {
6454 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6456 for (el = job->nodes; el; el = el->next) {
6457 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6462 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6463 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6465 QLIST_FOREACH(child, &bs->children, next) {
6466 xdbg_graph_add_edge(gr, bs, child);
6470 return xdbg_graph_finalize(gr);
6473 BlockDriverState *bdrv_lookup_bs(const char *device,
6474 const char *node_name,
6478 BlockDriverState *bs;
6480 GLOBAL_STATE_CODE();
6483 blk = blk_by_name(device);
6488 error_setg(errp, "Device '%s' has no medium", device);
6496 bs = bdrv_find_node(node_name);
6503 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6504 device ? device : "",
6505 node_name ? node_name : "");
6509 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6510 * return false. If either argument is NULL, return false. */
6511 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6514 GLOBAL_STATE_CODE();
6516 while (top && top != base) {
6517 top = bdrv_filter_or_cow_bs(top);
6523 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6525 GLOBAL_STATE_CODE();
6527 return QTAILQ_FIRST(&graph_bdrv_states);
6529 return QTAILQ_NEXT(bs, node_list);
6532 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6534 GLOBAL_STATE_CODE();
6536 return QTAILQ_FIRST(&all_bdrv_states);
6538 return QTAILQ_NEXT(bs, bs_list);
6541 const char *bdrv_get_node_name(const BlockDriverState *bs)
6544 return bs->node_name;
6547 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6553 /* If multiple parents have a name, just pick the first one. */
6554 QLIST_FOREACH(c, &bs->parents, next_parent) {
6555 if (c->klass->get_name) {
6556 name = c->klass->get_name(c);
6557 if (name && *name) {
6566 /* TODO check what callers really want: bs->node_name or blk_name() */
6567 const char *bdrv_get_device_name(const BlockDriverState *bs)
6570 return bdrv_get_parent_name(bs) ?: "";
6573 /* This can be used to identify nodes that might not have a device
6574 * name associated. Since node and device names live in the same
6575 * namespace, the result is unambiguous. The exception is if both are
6576 * absent, then this returns an empty (non-null) string. */
6577 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6580 return bdrv_get_parent_name(bs) ?: bs->node_name;
6583 int bdrv_get_flags(BlockDriverState *bs)
6586 return bs->open_flags;
6589 int bdrv_has_zero_init_1(BlockDriverState *bs)
6591 GLOBAL_STATE_CODE();
6595 int coroutine_mixed_fn bdrv_has_zero_init(BlockDriverState *bs)
6597 BlockDriverState *filtered;
6598 GLOBAL_STATE_CODE();
6604 /* If BS is a copy on write image, it is initialized to
6605 the contents of the base image, which may not be zeroes. */
6606 if (bdrv_cow_child(bs)) {
6609 if (bs->drv->bdrv_has_zero_init) {
6610 return bs->drv->bdrv_has_zero_init(bs);
6613 filtered = bdrv_filter_bs(bs);
6615 return bdrv_has_zero_init(filtered);
6622 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6625 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6629 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6632 void bdrv_get_backing_filename(BlockDriverState *bs,
6633 char *filename, int filename_size)
6636 pstrcpy(filename, filename_size, bs->backing_file);
6639 int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6642 BlockDriver *drv = bs->drv;
6644 assert_bdrv_graph_readable();
6646 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6650 if (!drv->bdrv_co_get_info) {
6651 BlockDriverState *filtered = bdrv_filter_bs(bs);
6653 return bdrv_co_get_info(filtered, bdi);
6657 memset(bdi, 0, sizeof(*bdi));
6658 ret = drv->bdrv_co_get_info(bs, bdi);
6659 if (bdi->subcluster_size == 0) {
6661 * If the driver left this unset, subclusters are not supported.
6662 * Then it is safe to treat each cluster as having only one subcluster.
6664 bdi->subcluster_size = bdi->cluster_size;
6670 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6677 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6680 BlockDriver *drv = bs->drv;
6682 if (drv && drv->bdrv_get_specific_info) {
6683 return drv->bdrv_get_specific_info(bs, errp);
6688 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6690 BlockDriver *drv = bs->drv;
6692 if (!drv || !drv->bdrv_get_specific_stats) {
6695 return drv->bdrv_get_specific_stats(bs);
6698 void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6701 assert_bdrv_graph_readable();
6703 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
6707 bs->drv->bdrv_co_debug_event(bs, event);
6710 static BlockDriverState * GRAPH_RDLOCK
6711 bdrv_find_debug_node(BlockDriverState *bs)
6713 GLOBAL_STATE_CODE();
6714 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6715 bs = bdrv_primary_bs(bs);
6718 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6719 assert(bs->drv->bdrv_debug_remove_breakpoint);
6726 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6729 GLOBAL_STATE_CODE();
6730 GRAPH_RDLOCK_GUARD_MAINLOOP();
6732 bs = bdrv_find_debug_node(bs);
6734 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6740 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6742 GLOBAL_STATE_CODE();
6743 GRAPH_RDLOCK_GUARD_MAINLOOP();
6745 bs = bdrv_find_debug_node(bs);
6747 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6753 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6755 GLOBAL_STATE_CODE();
6756 GRAPH_RDLOCK_GUARD_MAINLOOP();
6758 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6759 bs = bdrv_primary_bs(bs);
6762 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6763 return bs->drv->bdrv_debug_resume(bs, tag);
6769 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6771 GLOBAL_STATE_CODE();
6772 GRAPH_RDLOCK_GUARD_MAINLOOP();
6774 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6775 bs = bdrv_primary_bs(bs);
6778 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6779 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6785 /* backing_file can either be relative, or absolute, or a protocol. If it is
6786 * relative, it must be relative to the chain. So, passing in bs->filename
6787 * from a BDS as backing_file should not be done, as that may be relative to
6788 * the CWD rather than the chain. */
6789 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6790 const char *backing_file)
6792 char *filename_full = NULL;
6793 char *backing_file_full = NULL;
6794 char *filename_tmp = NULL;
6795 int is_protocol = 0;
6796 bool filenames_refreshed = false;
6797 BlockDriverState *curr_bs = NULL;
6798 BlockDriverState *retval = NULL;
6799 BlockDriverState *bs_below;
6801 GLOBAL_STATE_CODE();
6802 GRAPH_RDLOCK_GUARD_MAINLOOP();
6804 if (!bs || !bs->drv || !backing_file) {
6808 filename_full = g_malloc(PATH_MAX);
6809 backing_file_full = g_malloc(PATH_MAX);
6811 is_protocol = path_has_protocol(backing_file);
6814 * Being largely a legacy function, skip any filters here
6815 * (because filters do not have normal filenames, so they cannot
6816 * match anyway; and allowing json:{} filenames is a bit out of
6819 for (curr_bs = bdrv_skip_filters(bs);
6820 bdrv_cow_child(curr_bs) != NULL;
6823 bs_below = bdrv_backing_chain_next(curr_bs);
6825 if (bdrv_backing_overridden(curr_bs)) {
6827 * If the backing file was overridden, we can only compare
6828 * directly against the backing node's filename.
6831 if (!filenames_refreshed) {
6833 * This will automatically refresh all of the
6834 * filenames in the rest of the backing chain, so we
6835 * only need to do this once.
6837 bdrv_refresh_filename(bs_below);
6838 filenames_refreshed = true;
6841 if (strcmp(backing_file, bs_below->filename) == 0) {
6845 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6847 * If either of the filename paths is actually a protocol, then
6848 * compare unmodified paths; otherwise make paths relative.
6850 char *backing_file_full_ret;
6852 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6856 /* Also check against the full backing filename for the image */
6857 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6859 if (backing_file_full_ret) {
6860 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6861 g_free(backing_file_full_ret);
6868 /* If not an absolute filename path, make it relative to the current
6869 * image's filename path */
6870 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6872 /* We are going to compare canonicalized absolute pathnames */
6873 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6874 g_free(filename_tmp);
6877 g_free(filename_tmp);
6879 /* We need to make sure the backing filename we are comparing against
6880 * is relative to the current image filename (or absolute) */
6881 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6882 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6883 g_free(filename_tmp);
6886 g_free(filename_tmp);
6888 if (strcmp(backing_file_full, filename_full) == 0) {
6895 g_free(filename_full);
6896 g_free(backing_file_full);
6900 void bdrv_init(void)
6902 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6903 use_bdrv_whitelist = 1;
6905 module_call_init(MODULE_INIT_BLOCK);
6908 void bdrv_init_with_whitelist(void)
6910 use_bdrv_whitelist = 1;
6914 int bdrv_activate(BlockDriverState *bs, Error **errp)
6916 BdrvChild *child, *parent;
6917 Error *local_err = NULL;
6919 BdrvDirtyBitmap *bm;
6921 GLOBAL_STATE_CODE();
6922 GRAPH_RDLOCK_GUARD_MAINLOOP();
6928 QLIST_FOREACH(child, &bs->children, next) {
6929 bdrv_activate(child->bs, &local_err);
6931 error_propagate(errp, local_err);
6937 * Update permissions, they may differ for inactive nodes.
6939 * Note that the required permissions of inactive images are always a
6940 * subset of the permissions required after activating the image. This
6941 * allows us to just get the permissions upfront without restricting
6942 * bdrv_co_invalidate_cache().
6944 * It also means that in error cases, we don't have to try and revert to
6945 * the old permissions (which is an operation that could fail, too). We can
6946 * just keep the extended permissions for the next time that an activation
6947 * of the image is tried.
6949 if (bs->open_flags & BDRV_O_INACTIVE) {
6950 bs->open_flags &= ~BDRV_O_INACTIVE;
6951 ret = bdrv_refresh_perms(bs, NULL, errp);
6953 bs->open_flags |= BDRV_O_INACTIVE;
6957 ret = bdrv_invalidate_cache(bs, errp);
6959 bs->open_flags |= BDRV_O_INACTIVE;
6963 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6964 bdrv_dirty_bitmap_skip_store(bm, false);
6967 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6969 bs->open_flags |= BDRV_O_INACTIVE;
6970 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6975 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6976 if (parent->klass->activate) {
6977 parent->klass->activate(parent, &local_err);
6979 bs->open_flags |= BDRV_O_INACTIVE;
6980 error_propagate(errp, local_err);
6989 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6991 Error *local_err = NULL;
6994 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6995 assert_bdrv_graph_readable();
6997 if (bs->drv->bdrv_co_invalidate_cache) {
6998 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
7000 error_propagate(errp, local_err);
7008 void bdrv_activate_all(Error **errp)
7010 BlockDriverState *bs;
7011 BdrvNextIterator it;
7013 GLOBAL_STATE_CODE();
7014 GRAPH_RDLOCK_GUARD_MAINLOOP();
7016 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
7017 AioContext *aio_context = bdrv_get_aio_context(bs);
7020 aio_context_acquire(aio_context);
7021 ret = bdrv_activate(bs, errp);
7022 aio_context_release(aio_context);
7024 bdrv_next_cleanup(&it);
7030 static bool GRAPH_RDLOCK
7031 bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
7034 GLOBAL_STATE_CODE();
7036 QLIST_FOREACH(parent, &bs->parents, next_parent) {
7037 if (parent->klass->parent_is_bds) {
7038 BlockDriverState *parent_bs = parent->opaque;
7039 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
7048 static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
7050 BdrvChild *child, *parent;
7052 uint64_t cumulative_perms, cumulative_shared_perms;
7054 GLOBAL_STATE_CODE();
7060 /* Make sure that we don't inactivate a child before its parent.
7061 * It will be covered by recursion from the yet active parent. */
7062 if (bdrv_has_bds_parent(bs, true)) {
7066 assert(!(bs->open_flags & BDRV_O_INACTIVE));
7068 /* Inactivate this node */
7069 if (bs->drv->bdrv_inactivate) {
7070 ret = bs->drv->bdrv_inactivate(bs);
7076 QLIST_FOREACH(parent, &bs->parents, next_parent) {
7077 if (parent->klass->inactivate) {
7078 ret = parent->klass->inactivate(parent);
7085 bdrv_get_cumulative_perm(bs, &cumulative_perms,
7086 &cumulative_shared_perms);
7087 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
7088 /* Our inactive parents still need write access. Inactivation failed. */
7092 bs->open_flags |= BDRV_O_INACTIVE;
7095 * Update permissions, they may differ for inactive nodes.
7096 * We only tried to loosen restrictions, so errors are not fatal, ignore
7099 bdrv_refresh_perms(bs, NULL, NULL);
7101 /* Recursively inactivate children */
7102 QLIST_FOREACH(child, &bs->children, next) {
7103 ret = bdrv_inactivate_recurse(child->bs);
7112 int bdrv_inactivate_all(void)
7114 BlockDriverState *bs = NULL;
7115 BdrvNextIterator it;
7117 GSList *aio_ctxs = NULL, *ctx;
7119 GLOBAL_STATE_CODE();
7120 GRAPH_RDLOCK_GUARD_MAINLOOP();
7122 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
7123 AioContext *aio_context = bdrv_get_aio_context(bs);
7125 if (!g_slist_find(aio_ctxs, aio_context)) {
7126 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
7127 aio_context_acquire(aio_context);
7131 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
7132 /* Nodes with BDS parents are covered by recursion from the last
7133 * parent that gets inactivated. Don't inactivate them a second
7134 * time if that has already happened. */
7135 if (bdrv_has_bds_parent(bs, false)) {
7138 ret = bdrv_inactivate_recurse(bs);
7140 bdrv_next_cleanup(&it);
7146 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
7147 AioContext *aio_context = ctx->data;
7148 aio_context_release(aio_context);
7150 g_slist_free(aio_ctxs);
7155 /**************************************************************/
7156 /* removable device support */
7159 * Return TRUE if the media is present
7161 bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
7163 BlockDriver *drv = bs->drv;
7166 assert_bdrv_graph_readable();
7171 if (drv->bdrv_co_is_inserted) {
7172 return drv->bdrv_co_is_inserted(bs);
7174 QLIST_FOREACH(child, &bs->children, next) {
7175 if (!bdrv_co_is_inserted(child->bs)) {
7183 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7185 void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
7187 BlockDriver *drv = bs->drv;
7189 assert_bdrv_graph_readable();
7191 if (drv && drv->bdrv_co_eject) {
7192 drv->bdrv_co_eject(bs, eject_flag);
7197 * Lock or unlock the media (if it is locked, the user won't be able
7198 * to eject it manually).
7200 void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
7202 BlockDriver *drv = bs->drv;
7204 assert_bdrv_graph_readable();
7205 trace_bdrv_lock_medium(bs, locked);
7207 if (drv && drv->bdrv_co_lock_medium) {
7208 drv->bdrv_co_lock_medium(bs, locked);
7212 /* Get a reference to bs */
7213 void bdrv_ref(BlockDriverState *bs)
7215 GLOBAL_STATE_CODE();
7219 /* Release a previously grabbed reference to bs.
7220 * If after releasing, reference count is zero, the BlockDriverState is
7222 void bdrv_unref(BlockDriverState *bs)
7224 GLOBAL_STATE_CODE();
7228 assert(bs->refcnt > 0);
7229 if (--bs->refcnt == 0) {
7235 * Release a BlockDriverState reference while holding the graph write lock.
7237 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7238 * because bdrv_close() both involves polling and taking the graph lock
7239 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7240 * possibly closing @bs until the graph lock is released.
7242 void bdrv_schedule_unref(BlockDriverState *bs)
7247 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7248 (QEMUBHFunc *) bdrv_unref, bs);
7251 struct BdrvOpBlocker {
7253 QLIST_ENTRY(BdrvOpBlocker) list;
7256 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7258 BdrvOpBlocker *blocker;
7259 GLOBAL_STATE_CODE();
7261 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7262 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7263 blocker = QLIST_FIRST(&bs->op_blockers[op]);
7264 error_propagate_prepend(errp, error_copy(blocker->reason),
7265 "Node '%s' is busy: ",
7266 bdrv_get_device_or_node_name(bs));
7272 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7274 BdrvOpBlocker *blocker;
7275 GLOBAL_STATE_CODE();
7276 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7278 blocker = g_new0(BdrvOpBlocker, 1);
7279 blocker->reason = reason;
7280 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7283 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7285 BdrvOpBlocker *blocker, *next;
7286 GLOBAL_STATE_CODE();
7287 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7288 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7289 if (blocker->reason == reason) {
7290 QLIST_REMOVE(blocker, list);
7296 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7299 GLOBAL_STATE_CODE();
7300 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7301 bdrv_op_block(bs, i, reason);
7305 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7308 GLOBAL_STATE_CODE();
7309 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7310 bdrv_op_unblock(bs, i, reason);
7314 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7317 GLOBAL_STATE_CODE();
7318 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7319 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7327 * Must not be called while holding the lock of an AioContext other than the
7330 void bdrv_img_create(const char *filename, const char *fmt,
7331 const char *base_filename, const char *base_fmt,
7332 char *options, uint64_t img_size, int flags, bool quiet,
7335 QemuOptsList *create_opts = NULL;
7336 QemuOpts *opts = NULL;
7337 const char *backing_fmt, *backing_file;
7339 BlockDriver *drv, *proto_drv;
7340 Error *local_err = NULL;
7343 GLOBAL_STATE_CODE();
7345 /* Find driver and parse its options */
7346 drv = bdrv_find_format(fmt);
7348 error_setg(errp, "Unknown file format '%s'", fmt);
7352 proto_drv = bdrv_find_protocol(filename, true, errp);
7357 if (!drv->create_opts) {
7358 error_setg(errp, "Format driver '%s' does not support image creation",
7363 if (!proto_drv->create_opts) {
7364 error_setg(errp, "Protocol driver '%s' does not support image creation",
7365 proto_drv->format_name);
7369 aio_context_acquire(qemu_get_aio_context());
7371 /* Create parameter list */
7372 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7373 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7375 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7377 /* Parse -o options */
7379 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7384 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7385 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7386 } else if (img_size != UINT64_C(-1)) {
7387 error_setg(errp, "The image size must be specified only once");
7391 if (base_filename) {
7392 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7394 error_setg(errp, "Backing file not supported for file format '%s'",
7401 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7402 error_setg(errp, "Backing file format not supported for file "
7403 "format '%s'", fmt);
7408 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7410 if (!strcmp(filename, backing_file)) {
7411 error_setg(errp, "Error: Trying to create an image with the "
7412 "same filename as the backing file");
7415 if (backing_file[0] == '\0') {
7416 error_setg(errp, "Expected backing file name, got empty string");
7421 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7423 /* The size for the image must always be specified, unless we have a backing
7424 * file and we have not been forbidden from opening it. */
7425 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7426 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7427 BlockDriverState *bs;
7430 QDict *backing_options = NULL;
7433 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7438 assert(full_backing);
7441 * No need to do I/O here, which allows us to open encrypted
7442 * backing images without needing the secret
7445 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7446 back_flags |= BDRV_O_NO_IO;
7448 backing_options = qdict_new();
7450 qdict_put_str(backing_options, "driver", backing_fmt);
7452 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7454 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7456 g_free(full_backing);
7458 error_append_hint(&local_err, "Could not open backing image.\n");
7462 error_setg(&local_err,
7463 "Backing file specified without backing format");
7464 error_append_hint(&local_err, "Detected format of %s.\n",
7465 bs->drv->format_name);
7469 /* Opened BS, have no size */
7470 size = bdrv_getlength(bs);
7472 error_setg_errno(errp, -size, "Could not get size of '%s'",
7477 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7481 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7482 } else if (backing_file && !backing_fmt) {
7483 error_setg(&local_err,
7484 "Backing file specified without backing format");
7489 error_setg(errp, "Image creation needs a size parameter");
7494 printf("Formatting '%s', fmt=%s ", filename, fmt);
7495 qemu_opts_print(opts, " ");
7500 ret = bdrv_create(drv, filename, opts, &local_err);
7502 if (ret == -EFBIG) {
7503 /* This is generally a better message than whatever the driver would
7504 * deliver (especially because of the cluster_size_hint), since that
7505 * is most probably not much different from "image too large". */
7506 const char *cluster_size_hint = "";
7507 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7508 cluster_size_hint = " (try using a larger cluster size)";
7510 error_setg(errp, "The image size is too large for file format '%s'"
7511 "%s", fmt, cluster_size_hint);
7512 error_free(local_err);
7517 qemu_opts_del(opts);
7518 qemu_opts_free(create_opts);
7519 error_propagate(errp, local_err);
7520 aio_context_release(qemu_get_aio_context());
7523 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7526 return bs ? bs->aio_context : qemu_get_aio_context();
7529 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7531 Coroutine *self = qemu_coroutine_self();
7532 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7533 AioContext *new_ctx;
7537 * Increase bs->in_flight to ensure that this operation is completed before
7538 * moving the node to a different AioContext. Read new_ctx only afterwards.
7540 bdrv_inc_in_flight(bs);
7542 new_ctx = bdrv_get_aio_context(bs);
7543 aio_co_reschedule_self(new_ctx);
7547 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7550 aio_co_reschedule_self(old_ctx);
7551 bdrv_dec_in_flight(bs);
7554 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7556 AioContext *ctx = bdrv_get_aio_context(bs);
7558 /* In the main thread, bs->aio_context won't change concurrently */
7559 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7562 * We're in coroutine context, so we already hold the lock of the main
7563 * loop AioContext. Don't lock it twice to avoid deadlocks.
7565 assert(qemu_in_coroutine());
7566 if (ctx != qemu_get_aio_context()) {
7567 aio_context_acquire(ctx);
7571 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7573 AioContext *ctx = bdrv_get_aio_context(bs);
7575 assert(qemu_in_coroutine());
7576 if (ctx != qemu_get_aio_context()) {
7577 aio_context_release(ctx);
7581 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7583 GLOBAL_STATE_CODE();
7584 QLIST_REMOVE(ban, list);
7588 static void bdrv_detach_aio_context(BlockDriverState *bs)
7590 BdrvAioNotifier *baf, *baf_tmp;
7592 assert(!bs->walking_aio_notifiers);
7593 GLOBAL_STATE_CODE();
7594 bs->walking_aio_notifiers = true;
7595 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7597 bdrv_do_remove_aio_context_notifier(baf);
7599 baf->detach_aio_context(baf->opaque);
7602 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7603 * remove remaining aio notifiers if we aren't called again.
7605 bs->walking_aio_notifiers = false;
7607 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7608 bs->drv->bdrv_detach_aio_context(bs);
7611 bs->aio_context = NULL;
7614 static void bdrv_attach_aio_context(BlockDriverState *bs,
7615 AioContext *new_context)
7617 BdrvAioNotifier *ban, *ban_tmp;
7618 GLOBAL_STATE_CODE();
7620 bs->aio_context = new_context;
7622 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7623 bs->drv->bdrv_attach_aio_context(bs, new_context);
7626 assert(!bs->walking_aio_notifiers);
7627 bs->walking_aio_notifiers = true;
7628 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7630 bdrv_do_remove_aio_context_notifier(ban);
7632 ban->attached_aio_context(new_context, ban->opaque);
7635 bs->walking_aio_notifiers = false;
7638 typedef struct BdrvStateSetAioContext {
7639 AioContext *new_ctx;
7640 BlockDriverState *bs;
7641 } BdrvStateSetAioContext;
7643 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7644 GHashTable *visited,
7648 GLOBAL_STATE_CODE();
7649 if (g_hash_table_contains(visited, c)) {
7652 g_hash_table_add(visited, c);
7655 * A BdrvChildClass that doesn't handle AioContext changes cannot
7656 * tolerate any AioContext changes
7658 if (!c->klass->change_aio_ctx) {
7659 char *user = bdrv_child_user_desc(c);
7660 error_setg(errp, "Changing iothreads is not supported by %s", user);
7664 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7665 assert(!errp || *errp);
7671 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7672 GHashTable *visited, Transaction *tran,
7675 GLOBAL_STATE_CODE();
7676 if (g_hash_table_contains(visited, c)) {
7679 g_hash_table_add(visited, c);
7680 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7683 static void bdrv_set_aio_context_clean(void *opaque)
7685 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7686 BlockDriverState *bs = (BlockDriverState *) state->bs;
7688 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7689 bdrv_drained_end(bs);
7694 static void bdrv_set_aio_context_commit(void *opaque)
7696 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7697 BlockDriverState *bs = (BlockDriverState *) state->bs;
7698 AioContext *new_context = state->new_ctx;
7699 AioContext *old_context = bdrv_get_aio_context(bs);
7702 * Take the old AioContex when detaching it from bs.
7703 * At this point, new_context lock is already acquired, and we are now
7704 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7705 * does not call AIO_POLL_WHILE().
7707 if (old_context != qemu_get_aio_context()) {
7708 aio_context_acquire(old_context);
7710 bdrv_detach_aio_context(bs);
7711 if (old_context != qemu_get_aio_context()) {
7712 aio_context_release(old_context);
7714 bdrv_attach_aio_context(bs, new_context);
7717 static TransactionActionDrv set_aio_context = {
7718 .commit = bdrv_set_aio_context_commit,
7719 .clean = bdrv_set_aio_context_clean,
7723 * Changes the AioContext used for fd handlers, timers, and BHs by this
7724 * BlockDriverState and all its children and parents.
7726 * Must be called from the main AioContext.
7728 * The caller must own the AioContext lock for the old AioContext of bs, but it
7729 * must not own the AioContext lock for new_context (unless new_context is the
7730 * same as the current context of bs).
7732 * @visited will accumulate all visited BdrvChild objects. The caller is
7733 * responsible for freeing the list afterwards.
7735 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7736 GHashTable *visited, Transaction *tran,
7740 BdrvStateSetAioContext *state;
7742 GLOBAL_STATE_CODE();
7744 if (bdrv_get_aio_context(bs) == ctx) {
7748 bdrv_graph_rdlock_main_loop();
7749 QLIST_FOREACH(c, &bs->parents, next_parent) {
7750 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7751 bdrv_graph_rdunlock_main_loop();
7756 QLIST_FOREACH(c, &bs->children, next) {
7757 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7758 bdrv_graph_rdunlock_main_loop();
7762 bdrv_graph_rdunlock_main_loop();
7764 state = g_new(BdrvStateSetAioContext, 1);
7765 *state = (BdrvStateSetAioContext) {
7770 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7771 bdrv_drained_begin(bs);
7773 tran_add(tran, &set_aio_context, state);
7779 * Change bs's and recursively all of its parents' and children's AioContext
7780 * to the given new context, returning an error if that isn't possible.
7782 * If ignore_child is not NULL, that child (and its subgraph) will not
7785 * This function still requires the caller to take the bs current
7786 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7787 * assumes the lock is always held if bs is in another AioContext.
7788 * For the same reason, it temporarily also holds the new AioContext, since
7789 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7790 * Therefore the new AioContext lock must not be taken by the caller.
7792 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7793 BdrvChild *ignore_child, Error **errp)
7796 GHashTable *visited;
7798 AioContext *old_context = bdrv_get_aio_context(bs);
7799 GLOBAL_STATE_CODE();
7802 * Recursion phase: go through all nodes of the graph.
7803 * Take care of checking that all nodes support changing AioContext
7804 * and drain them, building a linear list of callbacks to run if everything
7805 * is successful (the transaction itself).
7808 visited = g_hash_table_new(NULL, NULL);
7810 g_hash_table_add(visited, ignore_child);
7812 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7813 g_hash_table_destroy(visited);
7816 * Linear phase: go through all callbacks collected in the transaction.
7817 * Run all callbacks collected in the recursion to switch all nodes
7818 * AioContext lock (transaction commit), or undo all changes done in the
7819 * recursion (transaction abort).
7823 /* Just run clean() callbacks. No AioContext changed. */
7829 * Release old AioContext, it won't be needed anymore, as all
7830 * bdrv_drained_begin() have been called already.
7832 if (qemu_get_aio_context() != old_context) {
7833 aio_context_release(old_context);
7837 * Acquire new AioContext since bdrv_drained_end() is going to be called
7838 * after we switched all nodes in the new AioContext, and the function
7839 * assumes that the lock of the bs is always taken.
7841 if (qemu_get_aio_context() != ctx) {
7842 aio_context_acquire(ctx);
7847 if (qemu_get_aio_context() != ctx) {
7848 aio_context_release(ctx);
7851 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7852 if (qemu_get_aio_context() != old_context) {
7853 aio_context_acquire(old_context);
7859 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7860 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7861 void (*detach_aio_context)(void *opaque), void *opaque)
7863 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7864 *ban = (BdrvAioNotifier){
7865 .attached_aio_context = attached_aio_context,
7866 .detach_aio_context = detach_aio_context,
7869 GLOBAL_STATE_CODE();
7871 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7874 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7875 void (*attached_aio_context)(AioContext *,
7877 void (*detach_aio_context)(void *),
7880 BdrvAioNotifier *ban, *ban_next;
7881 GLOBAL_STATE_CODE();
7883 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7884 if (ban->attached_aio_context == attached_aio_context &&
7885 ban->detach_aio_context == detach_aio_context &&
7886 ban->opaque == opaque &&
7887 ban->deleted == false)
7889 if (bs->walking_aio_notifiers) {
7890 ban->deleted = true;
7892 bdrv_do_remove_aio_context_notifier(ban);
7901 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7902 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7906 GLOBAL_STATE_CODE();
7908 error_setg(errp, "Node is ejected");
7911 if (!bs->drv->bdrv_amend_options) {
7912 error_setg(errp, "Block driver '%s' does not support option amendment",
7913 bs->drv->format_name);
7916 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7917 cb_opaque, force, errp);
7921 * This function checks whether the given @to_replace is allowed to be
7922 * replaced by a node that always shows the same data as @bs. This is
7923 * used for example to verify whether the mirror job can replace
7924 * @to_replace by the target mirrored from @bs.
7925 * To be replaceable, @bs and @to_replace may either be guaranteed to
7926 * always show the same data (because they are only connected through
7927 * filters), or some driver may allow replacing one of its children
7928 * because it can guarantee that this child's data is not visible at
7929 * all (for example, for dissenting quorum children that have no other
7932 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7933 BlockDriverState *to_replace)
7935 BlockDriverState *filtered;
7937 GLOBAL_STATE_CODE();
7939 if (!bs || !bs->drv) {
7943 if (bs == to_replace) {
7947 /* See what the driver can do */
7948 if (bs->drv->bdrv_recurse_can_replace) {
7949 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7952 /* For filters without an own implementation, we can recurse on our own */
7953 filtered = bdrv_filter_bs(bs);
7955 return bdrv_recurse_can_replace(filtered, to_replace);
7963 * Check whether the given @node_name can be replaced by a node that
7964 * has the same data as @parent_bs. If so, return @node_name's BDS;
7967 * @node_name must be a (recursive) *child of @parent_bs (or this
7968 * function will return NULL).
7970 * The result (whether the node can be replaced or not) is only valid
7971 * for as long as no graph or permission changes occur.
7973 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7974 const char *node_name, Error **errp)
7976 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7977 AioContext *aio_context;
7979 GLOBAL_STATE_CODE();
7981 if (!to_replace_bs) {
7982 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7986 aio_context = bdrv_get_aio_context(to_replace_bs);
7987 aio_context_acquire(aio_context);
7989 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7990 to_replace_bs = NULL;
7994 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7995 * most non filter in order to prevent data corruption.
7996 * Another benefit is that this tests exclude backing files which are
7997 * blocked by the backing blockers.
7999 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
8000 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
8001 "because it cannot be guaranteed that doing so would not "
8002 "lead to an abrupt change of visible data",
8003 node_name, parent_bs->node_name);
8004 to_replace_bs = NULL;
8009 aio_context_release(aio_context);
8010 return to_replace_bs;
8014 * Iterates through the list of runtime option keys that are said to
8015 * be "strong" for a BDS. An option is called "strong" if it changes
8016 * a BDS's data. For example, the null block driver's "size" and
8017 * "read-zeroes" options are strong, but its "latency-ns" option is
8020 * If a key returned by this function ends with a dot, all options
8021 * starting with that prefix are strong.
8023 static const char *const *strong_options(BlockDriverState *bs,
8024 const char *const *curopt)
8026 static const char *const global_options[] = {
8027 "driver", "filename", NULL
8031 return &global_options[0];
8035 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
8036 curopt = bs->drv->strong_runtime_opts;
8039 return (curopt && *curopt) ? curopt : NULL;
8043 * Copies all strong runtime options from bs->options to the given
8044 * QDict. The set of strong option keys is determined by invoking
8047 * Returns true iff any strong option was present in bs->options (and
8048 * thus copied to the target QDict) with the exception of "filename"
8049 * and "driver". The caller is expected to use this value to decide
8050 * whether the existence of strong options prevents the generation of
8053 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
8055 bool found_any = false;
8056 const char *const *option_name = NULL;
8062 while ((option_name = strong_options(bs, option_name))) {
8063 bool option_given = false;
8065 assert(strlen(*option_name) > 0);
8066 if ((*option_name)[strlen(*option_name) - 1] != '.') {
8067 QObject *entry = qdict_get(bs->options, *option_name);
8072 qdict_put_obj(d, *option_name, qobject_ref(entry));
8073 option_given = true;
8075 const QDictEntry *entry;
8076 for (entry = qdict_first(bs->options); entry;
8077 entry = qdict_next(bs->options, entry))
8079 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
8080 qdict_put_obj(d, qdict_entry_key(entry),
8081 qobject_ref(qdict_entry_value(entry)));
8082 option_given = true;
8087 /* While "driver" and "filename" need to be included in a JSON filename,
8088 * their existence does not prohibit generation of a plain filename. */
8089 if (!found_any && option_given &&
8090 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
8096 if (!qdict_haskey(d, "driver")) {
8097 /* Drivers created with bdrv_new_open_driver() may not have a
8098 * @driver option. Add it here. */
8099 qdict_put_str(d, "driver", bs->drv->format_name);
8105 /* Note: This function may return false positives; it may return true
8106 * even if opening the backing file specified by bs's image header
8107 * would result in exactly bs->backing. */
8108 static bool bdrv_backing_overridden(BlockDriverState *bs)
8110 GLOBAL_STATE_CODE();
8112 return strcmp(bs->auto_backing_file,
8113 bs->backing->bs->filename);
8115 /* No backing BDS, so if the image header reports any backing
8116 * file, it must have been suppressed */
8117 return bs->auto_backing_file[0] != '\0';
8121 /* Updates the following BDS fields:
8122 * - exact_filename: A filename which may be used for opening a block device
8123 * which (mostly) equals the given BDS (even without any
8124 * other options; so reading and writing must return the same
8125 * results, but caching etc. may be different)
8126 * - full_open_options: Options which, when given when opening a block device
8127 * (without a filename), result in a BDS (mostly)
8128 * equalling the given one
8129 * - filename: If exact_filename is set, it is copied here. Otherwise,
8130 * full_open_options is converted to a JSON object, prefixed with
8131 * "json:" (for use through the JSON pseudo protocol) and put here.
8133 void bdrv_refresh_filename(BlockDriverState *bs)
8135 BlockDriver *drv = bs->drv;
8137 BlockDriverState *primary_child_bs;
8139 bool backing_overridden;
8140 bool generate_json_filename; /* Whether our default implementation should
8141 fill exact_filename (false) or not (true) */
8143 GLOBAL_STATE_CODE();
8149 /* This BDS's file name may depend on any of its children's file names, so
8150 * refresh those first */
8151 QLIST_FOREACH(child, &bs->children, next) {
8152 bdrv_refresh_filename(child->bs);
8156 /* For implicit nodes, just copy everything from the single child */
8157 child = QLIST_FIRST(&bs->children);
8158 assert(QLIST_NEXT(child, next) == NULL);
8160 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
8161 child->bs->exact_filename);
8162 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
8164 qobject_unref(bs->full_open_options);
8165 bs->full_open_options = qobject_ref(child->bs->full_open_options);
8170 backing_overridden = bdrv_backing_overridden(bs);
8172 if (bs->open_flags & BDRV_O_NO_IO) {
8173 /* Without I/O, the backing file does not change anything.
8174 * Therefore, in such a case (primarily qemu-img), we can
8175 * pretend the backing file has not been overridden even if
8176 * it technically has been. */
8177 backing_overridden = false;
8180 /* Gather the options QDict */
8182 generate_json_filename = append_strong_runtime_options(opts, bs);
8183 generate_json_filename |= backing_overridden;
8185 if (drv->bdrv_gather_child_options) {
8186 /* Some block drivers may not want to present all of their children's
8187 * options, or name them differently from BdrvChild.name */
8188 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
8190 QLIST_FOREACH(child, &bs->children, next) {
8191 if (child == bs->backing && !backing_overridden) {
8192 /* We can skip the backing BDS if it has not been overridden */
8196 qdict_put(opts, child->name,
8197 qobject_ref(child->bs->full_open_options));
8200 if (backing_overridden && !bs->backing) {
8201 /* Force no backing file */
8202 qdict_put_null(opts, "backing");
8206 qobject_unref(bs->full_open_options);
8207 bs->full_open_options = opts;
8209 primary_child_bs = bdrv_primary_bs(bs);
8211 if (drv->bdrv_refresh_filename) {
8212 /* Obsolete information is of no use here, so drop the old file name
8213 * information before refreshing it */
8214 bs->exact_filename[0] = '\0';
8216 drv->bdrv_refresh_filename(bs);
8217 } else if (primary_child_bs) {
8219 * Try to reconstruct valid information from the underlying
8220 * file -- this only works for format nodes (filter nodes
8221 * cannot be probed and as such must be selected by the user
8222 * either through an options dict, or through a special
8223 * filename which the filter driver must construct in its
8224 * .bdrv_refresh_filename() implementation).
8227 bs->exact_filename[0] = '\0';
8230 * We can use the underlying file's filename if:
8231 * - it has a filename,
8232 * - the current BDS is not a filter,
8233 * - the file is a protocol BDS, and
8234 * - opening that file (as this BDS's format) will automatically create
8235 * the BDS tree we have right now, that is:
8236 * - the user did not significantly change this BDS's behavior with
8237 * some explicit (strong) options
8238 * - no non-file child of this BDS has been overridden by the user
8239 * Both of these conditions are represented by generate_json_filename.
8241 if (primary_child_bs->exact_filename[0] &&
8242 primary_child_bs->drv->bdrv_file_open &&
8243 !drv->is_filter && !generate_json_filename)
8245 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
8249 if (bs->exact_filename[0]) {
8250 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
8252 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
8253 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
8254 json->str) >= sizeof(bs->filename)) {
8255 /* Give user a hint if we truncated things. */
8256 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8258 g_string_free(json, true);
8262 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8264 BlockDriver *drv = bs->drv;
8265 BlockDriverState *child_bs;
8267 GLOBAL_STATE_CODE();
8270 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8274 if (drv->bdrv_dirname) {
8275 return drv->bdrv_dirname(bs, errp);
8278 child_bs = bdrv_primary_bs(bs);
8280 return bdrv_dirname(child_bs, errp);
8283 bdrv_refresh_filename(bs);
8284 if (bs->exact_filename[0] != '\0') {
8285 return path_combine(bs->exact_filename, "");
8288 error_setg(errp, "Cannot generate a base directory for %s nodes",
8294 * Hot add/remove a BDS's child. So the user can take a child offline when
8295 * it is broken and take a new child online
8297 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8300 GLOBAL_STATE_CODE();
8301 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8302 error_setg(errp, "The node %s does not support adding a child",
8303 bdrv_get_device_or_node_name(parent_bs));
8308 * Non-zoned block drivers do not follow zoned storage constraints
8309 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8310 * drivers in a graph.
8312 if (!parent_bs->drv->supports_zoned_children &&
8313 child_bs->bl.zoned == BLK_Z_HM) {
8315 * The host-aware model allows zoned storage constraints and random
8316 * write. Allow mixing host-aware and non-zoned drivers. Using
8317 * host-aware device as a regular device.
8319 error_setg(errp, "Cannot add a %s child to a %s parent",
8320 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8321 parent_bs->drv->supports_zoned_children ?
8322 "support zoned children" : "not support zoned children");
8326 if (!QLIST_EMPTY(&child_bs->parents)) {
8327 error_setg(errp, "The node %s already has a parent",
8328 child_bs->node_name);
8332 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8335 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8339 GLOBAL_STATE_CODE();
8340 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8341 error_setg(errp, "The node %s does not support removing a child",
8342 bdrv_get_device_or_node_name(parent_bs));
8346 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8353 error_setg(errp, "The node %s does not have a child named %s",
8354 bdrv_get_device_or_node_name(parent_bs),
8355 bdrv_get_device_or_node_name(child->bs));
8359 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8362 int bdrv_make_empty(BdrvChild *c, Error **errp)
8364 BlockDriver *drv = c->bs->drv;
8367 GLOBAL_STATE_CODE();
8368 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8370 if (!drv->bdrv_make_empty) {
8371 error_setg(errp, "%s does not support emptying nodes",
8376 ret = drv->bdrv_make_empty(c->bs);
8378 error_setg_errno(errp, -ret, "Failed to empty %s",
8387 * Return the child that @bs acts as an overlay for, and from which data may be
8388 * copied in COW or COR operations. Usually this is the backing file.
8390 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8394 if (!bs || !bs->drv) {
8398 if (bs->drv->is_filter) {
8406 assert(bs->backing->role & BDRV_CHILD_COW);
8411 * If @bs acts as a filter for exactly one of its children, return
8414 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8419 if (!bs || !bs->drv) {
8423 if (!bs->drv->is_filter) {
8427 /* Only one of @backing or @file may be used */
8428 assert(!(bs->backing && bs->file));
8430 c = bs->backing ?: bs->file;
8435 assert(c->role & BDRV_CHILD_FILTERED);
8440 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8441 * whichever is non-NULL.
8443 * Return NULL if both are NULL.
8445 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8447 BdrvChild *cow_child = bdrv_cow_child(bs);
8448 BdrvChild *filter_child = bdrv_filter_child(bs);
8451 /* Filter nodes cannot have COW backing files */
8452 assert(!(cow_child && filter_child));
8454 return cow_child ?: filter_child;
8458 * Return the primary child of this node: For filters, that is the
8459 * filtered child. For other nodes, that is usually the child storing
8461 * (A generally more helpful description is that this is (usually) the
8462 * child that has the same filename as @bs.)
8464 * Drivers do not necessarily have a primary child; for example quorum
8467 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8469 BdrvChild *c, *found = NULL;
8472 QLIST_FOREACH(c, &bs->children, next) {
8473 if (c->role & BDRV_CHILD_PRIMARY) {
8482 static BlockDriverState * GRAPH_RDLOCK
8483 bdrv_do_skip_filters(BlockDriverState *bs, bool stop_on_explicit_filter)
8491 while (!(stop_on_explicit_filter && !bs->implicit)) {
8492 c = bdrv_filter_child(bs);
8495 * A filter that is embedded in a working block graph must
8496 * have a child. Assert this here so this function does
8497 * not return a filter node that is not expected by the
8500 assert(!bs->drv || !bs->drv->is_filter);
8506 * Note that this treats nodes with bs->drv == NULL as not being
8507 * filters (bs->drv == NULL should be replaced by something else
8509 * The advantage of this behavior is that this function will thus
8510 * always return a non-NULL value (given a non-NULL @bs).
8517 * Return the first BDS that has not been added implicitly or that
8518 * does not have a filtered child down the chain starting from @bs
8519 * (including @bs itself).
8521 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8523 GLOBAL_STATE_CODE();
8524 return bdrv_do_skip_filters(bs, true);
8528 * Return the first BDS that does not have a filtered child down the
8529 * chain starting from @bs (including @bs itself).
8531 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8534 return bdrv_do_skip_filters(bs, false);
8538 * For a backing chain, return the first non-filter backing image of
8539 * the first non-filter image.
8541 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8544 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8548 * Check whether [offset, offset + bytes) overlaps with the cached
8549 * block-status data region.
8551 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8552 * which is what bdrv_bsc_is_data()'s interface needs.
8553 * Otherwise, *pnum is not touched.
8555 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8556 int64_t offset, int64_t bytes,
8559 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8563 qatomic_read(&bsc->valid) &&
8564 ranges_overlap(offset, bytes, bsc->data_start,
8565 bsc->data_end - bsc->data_start);
8567 if (overlaps && pnum) {
8568 *pnum = bsc->data_end - offset;
8575 * See block_int.h for this function's documentation.
8577 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8580 RCU_READ_LOCK_GUARD();
8581 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8585 * See block_int.h for this function's documentation.
8587 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8588 int64_t offset, int64_t bytes)
8591 RCU_READ_LOCK_GUARD();
8593 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8594 qatomic_set(&bs->block_status_cache->valid, false);
8599 * See block_int.h for this function's documentation.
8601 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8603 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8604 BdrvBlockStatusCache *old_bsc;
8607 *new_bsc = (BdrvBlockStatusCache) {
8609 .data_start = offset,
8610 .data_end = offset + bytes,
8613 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8615 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8616 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8618 g_free_rcu(old_bsc, rcu);