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 bdrv_replace_child_noperm(BdrvChild *child,
95 BlockDriverState *new_bs);
96 static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
98 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
99 BlockReopenQueue *queue,
100 Transaction *change_child_tran, Error **errp);
101 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
102 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
104 static bool bdrv_backing_overridden(BlockDriverState *bs);
106 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
107 GHashTable *visited, Transaction *tran,
110 /* If non-zero, use only whitelisted block drivers */
111 static int use_bdrv_whitelist;
114 static int is_windows_drive_prefix(const char *filename)
116 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
117 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
121 int is_windows_drive(const char *filename)
123 if (is_windows_drive_prefix(filename) &&
126 if (strstart(filename, "\\\\.\\", NULL) ||
127 strstart(filename, "//./", NULL))
133 size_t bdrv_opt_mem_align(BlockDriverState *bs)
135 if (!bs || !bs->drv) {
136 /* page size or 4k (hdd sector size) should be on the safe side */
137 return MAX(4096, qemu_real_host_page_size());
141 return bs->bl.opt_mem_alignment;
144 size_t bdrv_min_mem_align(BlockDriverState *bs)
146 if (!bs || !bs->drv) {
147 /* page size or 4k (hdd sector size) should be on the safe side */
148 return MAX(4096, qemu_real_host_page_size());
152 return bs->bl.min_mem_alignment;
155 /* check if the path starts with "<protocol>:" */
156 int path_has_protocol(const char *path)
161 if (is_windows_drive(path) ||
162 is_windows_drive_prefix(path)) {
165 p = path + strcspn(path, ":/\\");
167 p = path + strcspn(path, ":/");
173 int path_is_absolute(const char *path)
176 /* specific case for names like: "\\.\d:" */
177 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
180 return (*path == '/' || *path == '\\');
182 return (*path == '/');
186 /* if filename is absolute, just return its duplicate. Otherwise, build a
187 path to it by considering it is relative to base_path. URL are
189 char *path_combine(const char *base_path, const char *filename)
191 const char *protocol_stripped = NULL;
196 if (path_is_absolute(filename)) {
197 return g_strdup(filename);
200 if (path_has_protocol(base_path)) {
201 protocol_stripped = strchr(base_path, ':');
202 if (protocol_stripped) {
206 p = protocol_stripped ?: base_path;
208 p1 = strrchr(base_path, '/');
212 p2 = strrchr(base_path, '\\');
213 if (!p1 || p2 > p1) {
228 result = g_malloc(len + strlen(filename) + 1);
229 memcpy(result, base_path, len);
230 strcpy(result + len, filename);
236 * Helper function for bdrv_parse_filename() implementations to remove optional
237 * protocol prefixes (especially "file:") from a filename and for putting the
238 * stripped filename into the options QDict if there is such a prefix.
240 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
243 if (strstart(filename, prefix, &filename)) {
244 /* Stripping the explicit protocol prefix may result in a protocol
245 * prefix being (wrongly) detected (if the filename contains a colon) */
246 if (path_has_protocol(filename)) {
247 GString *fat_filename;
249 /* This means there is some colon before the first slash; therefore,
250 * this cannot be an absolute path */
251 assert(!path_is_absolute(filename));
253 /* And we can thus fix the protocol detection issue by prefixing it
255 fat_filename = g_string_new("./");
256 g_string_append(fat_filename, filename);
258 assert(!path_has_protocol(fat_filename->str));
260 qdict_put(options, "filename",
261 qstring_from_gstring(fat_filename));
263 /* If no protocol prefix was detected, we can use the shortened
265 qdict_put_str(options, "filename", filename);
271 /* Returns whether the image file is opened as read-only. Note that this can
272 * return false and writing to the image file is still not possible because the
273 * image is inactivated. */
274 bool bdrv_is_read_only(BlockDriverState *bs)
277 return !(bs->open_flags & BDRV_O_RDWR);
280 static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
281 bool ignore_allow_rdw, Error **errp)
285 /* Do not set read_only if copy_on_read is enabled */
286 if (bs->copy_on_read && read_only) {
287 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
288 bdrv_get_device_or_node_name(bs));
292 /* Do not clear read_only if it is prohibited */
293 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
296 error_setg(errp, "Node '%s' is read only",
297 bdrv_get_device_or_node_name(bs));
305 * Called by a driver that can only provide a read-only image.
307 * Returns 0 if the node is already read-only or it could switch the node to
308 * read-only because BDRV_O_AUTO_RDONLY is set.
310 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
311 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
312 * is not NULL, it is used as the error message for the Error object.
314 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
320 if (!(bs->open_flags & BDRV_O_RDWR)) {
323 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
327 ret = bdrv_can_set_read_only(bs, true, false, NULL);
332 bs->open_flags &= ~BDRV_O_RDWR;
337 error_setg(errp, "%s", errmsg ?: "Image is read-only");
342 * If @backing is empty, this function returns NULL without setting
343 * @errp. In all other cases, NULL will only be returned with @errp
346 * Therefore, a return value of NULL without @errp set means that
347 * there is no backing file; if @errp is set, there is one but its
348 * absolute filename cannot be generated.
350 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
354 if (backing[0] == '\0') {
356 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
357 return g_strdup(backing);
358 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
359 error_setg(errp, "Cannot use relative backing file names for '%s'",
363 return path_combine(backed, backing);
368 * If @filename is empty or NULL, this function returns NULL without
369 * setting @errp. In all other cases, NULL will only be returned with
372 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
373 const char *filename, Error **errp)
375 char *dir, *full_name;
377 if (!filename || filename[0] == '\0') {
379 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
380 return g_strdup(filename);
383 dir = bdrv_dirname(relative_to, errp);
388 full_name = g_strconcat(dir, filename, NULL);
393 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
396 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
399 void bdrv_register(BlockDriver *bdrv)
401 assert(bdrv->format_name);
403 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
406 BlockDriverState *bdrv_new(void)
408 BlockDriverState *bs;
413 bs = g_new0(BlockDriverState, 1);
414 QLIST_INIT(&bs->dirty_bitmaps);
415 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
416 QLIST_INIT(&bs->op_blockers[i]);
418 qemu_co_mutex_init(&bs->reqs_lock);
419 qemu_mutex_init(&bs->dirty_bitmap_mutex);
421 bs->aio_context = qemu_get_aio_context();
423 qemu_co_queue_init(&bs->flush_queue);
425 qemu_co_mutex_init(&bs->bsc_modify_lock);
426 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
428 for (i = 0; i < bdrv_drain_all_count; i++) {
429 bdrv_drained_begin(bs);
432 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
437 static BlockDriver *bdrv_do_find_format(const char *format_name)
442 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
443 if (!strcmp(drv1->format_name, format_name)) {
451 BlockDriver *bdrv_find_format(const char *format_name)
458 drv1 = bdrv_do_find_format(format_name);
463 /* The driver isn't registered, maybe we need to load a module */
464 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
465 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
466 Error *local_err = NULL;
467 int rv = block_module_load(block_driver_modules[i].library_name,
470 return bdrv_do_find_format(format_name);
472 error_report_err(local_err);
480 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
482 static const char *whitelist_rw[] = {
483 CONFIG_BDRV_RW_WHITELIST
486 static const char *whitelist_ro[] = {
487 CONFIG_BDRV_RO_WHITELIST
492 if (!whitelist_rw[0] && !whitelist_ro[0]) {
493 return 1; /* no whitelist, anything goes */
496 for (p = whitelist_rw; *p; p++) {
497 if (!strcmp(format_name, *p)) {
502 for (p = whitelist_ro; *p; p++) {
503 if (!strcmp(format_name, *p)) {
511 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
514 return bdrv_format_is_whitelisted(drv->format_name, read_only);
517 bool bdrv_uses_whitelist(void)
519 return use_bdrv_whitelist;
522 typedef struct CreateCo {
530 int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
531 QemuOpts *opts, Error **errp)
536 assert_bdrv_graph_readable();
538 if (!drv->bdrv_co_create_opts) {
539 error_setg(errp, "Driver '%s' does not support image creation",
544 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
545 if (ret < 0 && !*errp) {
546 error_setg_errno(errp, -ret, "Could not create image");
553 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
554 * least the given @minimum_size.
556 * On success, return @blk's actual length.
557 * Otherwise, return -errno.
559 static int64_t create_file_fallback_truncate(BlockBackend *blk,
560 int64_t minimum_size, Error **errp)
562 Error *local_err = NULL;
568 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
570 if (ret < 0 && ret != -ENOTSUP) {
571 error_propagate(errp, local_err);
575 size = blk_getlength(blk);
577 error_free(local_err);
578 error_setg_errno(errp, -size,
579 "Failed to inquire the new image file's length");
583 if (size < minimum_size) {
584 /* Need to grow the image, but we failed to do that */
585 error_propagate(errp, local_err);
589 error_free(local_err);
596 * Helper function for bdrv_create_file_fallback(): Zero the first
597 * sector to remove any potentially pre-existing image header.
599 static int coroutine_fn
600 create_file_fallback_zero_first_sector(BlockBackend *blk,
601 int64_t current_size,
604 int64_t bytes_to_clear;
609 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
610 if (bytes_to_clear) {
611 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
613 error_setg_errno(errp, -ret,
614 "Failed to clear the new image's first sector");
623 * Simple implementation of bdrv_co_create_opts for protocol drivers
624 * which only support creation via opening a file
625 * (usually existing raw storage device)
627 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
628 const char *filename,
636 PreallocMode prealloc;
637 Error *local_err = NULL;
642 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
643 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
644 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
645 PREALLOC_MODE_OFF, &local_err);
648 error_propagate(errp, local_err);
652 if (prealloc != PREALLOC_MODE_OFF) {
653 error_setg(errp, "Unsupported preallocation mode '%s'",
654 PreallocMode_str(prealloc));
658 options = qdict_new();
659 qdict_put_str(options, "driver", drv->format_name);
661 blk = blk_co_new_open(filename, NULL, options,
662 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
664 error_prepend(errp, "Protocol driver '%s' does not support image "
665 "creation, and opening the image failed: ",
670 size = create_file_fallback_truncate(blk, size, errp);
676 ret = create_file_fallback_zero_first_sector(blk, size, errp);
687 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
690 QemuOpts *protocol_opts;
697 drv = bdrv_find_protocol(filename, true, errp);
702 if (!drv->create_opts) {
703 error_setg(errp, "Driver '%s' does not support image creation",
709 * 'opts' contains a QemuOptsList with a combination of format and protocol
712 * The format properly removes its options, but the default values remain
713 * in 'opts->list'. So if the protocol has options with the same name
714 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
715 * of the format, since for overlapping options, the format wins.
717 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
718 * only the set options, and then convert it back to QemuOpts, using the
719 * create_opts of the protocol. So the new QemuOpts, will contain only the
722 qdict = qemu_opts_to_qdict(opts, NULL);
723 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
724 if (protocol_opts == NULL) {
729 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
731 qemu_opts_del(protocol_opts);
732 qobject_unref(qdict);
736 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
738 Error *local_err = NULL;
743 assert_bdrv_graph_readable();
746 error_setg(errp, "Block node '%s' is not opened", bs->filename);
750 if (!bs->drv->bdrv_co_delete_file) {
751 error_setg(errp, "Driver '%s' does not support image deletion",
752 bs->drv->format_name);
756 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
758 error_propagate(errp, local_err);
764 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
766 Error *local_err = NULL;
774 ret = bdrv_co_delete_file(bs, &local_err);
776 * ENOTSUP will happen if the block driver doesn't support
777 * the 'bdrv_co_delete_file' interface. This is a predictable
778 * scenario and shouldn't be reported back to the user.
780 if (ret == -ENOTSUP) {
781 error_free(local_err);
782 } else if (ret < 0) {
783 error_report_err(local_err);
788 * Try to get @bs's logical and physical block size.
789 * On success, store them in @bsz struct and return 0.
790 * On failure return -errno.
791 * @bs must not be empty.
793 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
795 BlockDriver *drv = bs->drv;
796 BlockDriverState *filtered = bdrv_filter_bs(bs);
799 if (drv && drv->bdrv_probe_blocksizes) {
800 return drv->bdrv_probe_blocksizes(bs, bsz);
801 } else if (filtered) {
802 return bdrv_probe_blocksizes(filtered, bsz);
809 * Try to get @bs's geometry (cyls, heads, sectors).
810 * On success, store them in @geo struct and return 0.
811 * On failure return -errno.
812 * @bs must not be empty.
814 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
816 BlockDriver *drv = bs->drv;
817 BlockDriverState *filtered = bdrv_filter_bs(bs);
820 if (drv && drv->bdrv_probe_geometry) {
821 return drv->bdrv_probe_geometry(bs, geo);
822 } else if (filtered) {
823 return bdrv_probe_geometry(filtered, geo);
830 * Create a uniquely-named empty temporary file.
831 * Return the actual file name used upon success, otherwise NULL.
832 * This string should be freed with g_free() when not needed any longer.
834 * Note: creating a temporary file for the caller to (re)open is
835 * inherently racy. Use g_file_open_tmp() instead whenever practical.
837 char *create_tmp_file(Error **errp)
841 g_autofree char *filename = NULL;
843 tmpdir = g_get_tmp_dir();
846 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
848 * This function is used to create temporary disk images (like -snapshot),
849 * so the files can become very large. /tmp is often a tmpfs where as
850 * /var/tmp is usually on a disk, so more appropriate for disk images.
852 if (!g_strcmp0(tmpdir, "/tmp")) {
857 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
858 fd = g_mkstemp(filename);
860 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
866 return g_steal_pointer(&filename);
870 * Detect host devices. By convention, /dev/cdrom[N] is always
871 * recognized as a host CDROM.
873 static BlockDriver *find_hdev_driver(const char *filename)
875 int score_max = 0, score;
876 BlockDriver *drv = NULL, *d;
879 QLIST_FOREACH(d, &bdrv_drivers, list) {
880 if (d->bdrv_probe_device) {
881 score = d->bdrv_probe_device(filename);
882 if (score > score_max) {
892 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
897 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
898 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
906 BlockDriver *bdrv_find_protocol(const char *filename,
907 bool allow_protocol_prefix,
917 /* TODO Drivers without bdrv_file_open must be specified explicitly */
920 * XXX(hch): we really should not let host device detection
921 * override an explicit protocol specification, but moving this
922 * later breaks access to device names with colons in them.
923 * Thanks to the brain-dead persistent naming schemes on udev-
924 * based Linux systems those actually are quite common.
926 drv1 = find_hdev_driver(filename);
931 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
935 p = strchr(filename, ':');
938 if (len > sizeof(protocol) - 1)
939 len = sizeof(protocol) - 1;
940 memcpy(protocol, filename, len);
941 protocol[len] = '\0';
943 drv1 = bdrv_do_find_protocol(protocol);
948 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
949 if (block_driver_modules[i].protocol_name &&
950 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
951 int rv = block_module_load(block_driver_modules[i].library_name, errp);
953 drv1 = bdrv_do_find_protocol(protocol);
962 error_setg(errp, "Unknown protocol '%s'", protocol);
968 * Guess image format by probing its contents.
969 * This is not a good idea when your image is raw (CVE-2008-2004), but
970 * we do it anyway for backward compatibility.
972 * @buf contains the image's first @buf_size bytes.
973 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
974 * but can be smaller if the image file is smaller)
975 * @filename is its filename.
977 * For all block drivers, call the bdrv_probe() method to get its
979 * Return the first block driver with the highest probing score.
981 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
982 const char *filename)
984 int score_max = 0, score;
985 BlockDriver *drv = NULL, *d;
988 QLIST_FOREACH(d, &bdrv_drivers, list) {
990 score = d->bdrv_probe(buf, buf_size, filename);
991 if (score > score_max) {
1001 static int find_image_format(BlockBackend *file, const char *filename,
1002 BlockDriver **pdrv, Error **errp)
1005 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1008 GLOBAL_STATE_CODE();
1010 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1011 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1016 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1018 error_setg_errno(errp, -ret, "Could not read image for determining its "
1024 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1026 error_setg(errp, "Could not determine image format: No compatible "
1037 * Set the current 'total_sectors' value
1038 * Return 0 on success, -errno on error.
1040 int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1043 BlockDriver *drv = bs->drv;
1045 assert_bdrv_graph_readable();
1051 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1055 /* query actual device if possible, otherwise just trust the hint */
1056 if (drv->bdrv_co_getlength) {
1057 int64_t length = drv->bdrv_co_getlength(bs);
1061 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1064 bs->total_sectors = hint;
1066 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1074 * Combines a QDict of new block driver @options with any missing options taken
1075 * from @old_options, so that leaving out an option defaults to its old value.
1077 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1080 GLOBAL_STATE_CODE();
1081 if (bs->drv && bs->drv->bdrv_join_options) {
1082 bs->drv->bdrv_join_options(options, old_options);
1084 qdict_join(options, old_options, false);
1088 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1092 Error *local_err = NULL;
1093 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1094 BlockdevDetectZeroesOptions detect_zeroes =
1095 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1096 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1097 GLOBAL_STATE_CODE();
1100 error_propagate(errp, local_err);
1101 return detect_zeroes;
1104 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1105 !(open_flags & BDRV_O_UNMAP))
1107 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1108 "without setting discard operation to unmap");
1111 return detect_zeroes;
1115 * Set open flags for aio engine
1117 * Return 0 on success, -1 if the engine specified is invalid
1119 int bdrv_parse_aio(const char *mode, int *flags)
1121 if (!strcmp(mode, "threads")) {
1122 /* do nothing, default */
1123 } else if (!strcmp(mode, "native")) {
1124 *flags |= BDRV_O_NATIVE_AIO;
1125 #ifdef CONFIG_LINUX_IO_URING
1126 } else if (!strcmp(mode, "io_uring")) {
1127 *flags |= BDRV_O_IO_URING;
1137 * Set open flags for a given discard mode
1139 * Return 0 on success, -1 if the discard mode was invalid.
1141 int bdrv_parse_discard_flags(const char *mode, int *flags)
1143 *flags &= ~BDRV_O_UNMAP;
1145 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1147 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1148 *flags |= BDRV_O_UNMAP;
1157 * Set open flags for a given cache mode
1159 * Return 0 on success, -1 if the cache mode was invalid.
1161 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1163 *flags &= ~BDRV_O_CACHE_MASK;
1165 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1166 *writethrough = false;
1167 *flags |= BDRV_O_NOCACHE;
1168 } else if (!strcmp(mode, "directsync")) {
1169 *writethrough = true;
1170 *flags |= BDRV_O_NOCACHE;
1171 } else if (!strcmp(mode, "writeback")) {
1172 *writethrough = false;
1173 } else if (!strcmp(mode, "unsafe")) {
1174 *writethrough = false;
1175 *flags |= BDRV_O_NO_FLUSH;
1176 } else if (!strcmp(mode, "writethrough")) {
1177 *writethrough = true;
1185 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1187 BlockDriverState *parent = c->opaque;
1188 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1191 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1193 BlockDriverState *bs = child->opaque;
1194 bdrv_do_drained_begin_quiesce(bs, NULL);
1197 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1199 BlockDriverState *bs = child->opaque;
1200 return bdrv_drain_poll(bs, NULL, false);
1203 static void bdrv_child_cb_drained_end(BdrvChild *child)
1205 BlockDriverState *bs = child->opaque;
1206 bdrv_drained_end(bs);
1209 static int bdrv_child_cb_inactivate(BdrvChild *child)
1211 BlockDriverState *bs = child->opaque;
1212 GLOBAL_STATE_CODE();
1213 assert(bs->open_flags & BDRV_O_INACTIVE);
1217 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1218 GHashTable *visited, Transaction *tran,
1221 BlockDriverState *bs = child->opaque;
1222 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1226 * Returns the options and flags that a temporary snapshot should get, based on
1227 * the originally requested flags (the originally requested image will have
1228 * flags like a backing file)
1230 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1231 int parent_flags, QDict *parent_options)
1233 GLOBAL_STATE_CODE();
1234 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1236 /* For temporary files, unconditional cache=unsafe is fine */
1237 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1238 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1240 /* Copy the read-only and discard options from the parent */
1241 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1242 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1244 /* aio=native doesn't work for cache.direct=off, so disable it for the
1245 * temporary snapshot */
1246 *child_flags &= ~BDRV_O_NATIVE_AIO;
1249 static void bdrv_backing_attach(BdrvChild *c)
1251 BlockDriverState *parent = c->opaque;
1252 BlockDriverState *backing_hd = c->bs;
1254 GLOBAL_STATE_CODE();
1255 assert(!parent->backing_blocker);
1256 error_setg(&parent->backing_blocker,
1257 "node is used as backing hd of '%s'",
1258 bdrv_get_device_or_node_name(parent));
1260 bdrv_refresh_filename(backing_hd);
1262 parent->open_flags &= ~BDRV_O_NO_BACKING;
1264 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1265 /* Otherwise we won't be able to commit or stream */
1266 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1267 parent->backing_blocker);
1268 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1269 parent->backing_blocker);
1271 * We do backup in 3 ways:
1273 * The target bs is new opened, and the source is top BDS
1274 * 2. blockdev backup
1275 * Both the source and the target are top BDSes.
1276 * 3. internal backup(used for block replication)
1277 * Both the source and the target are backing file
1279 * In case 1 and 2, neither the source nor the target is the backing file.
1280 * In case 3, we will block the top BDS, so there is only one block job
1281 * for the top BDS and its backing chain.
1283 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1284 parent->backing_blocker);
1285 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1286 parent->backing_blocker);
1289 static void bdrv_backing_detach(BdrvChild *c)
1291 BlockDriverState *parent = c->opaque;
1293 GLOBAL_STATE_CODE();
1294 assert(parent->backing_blocker);
1295 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1296 error_free(parent->backing_blocker);
1297 parent->backing_blocker = NULL;
1300 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1301 const char *filename, Error **errp)
1303 BlockDriverState *parent = c->opaque;
1304 bool read_only = bdrv_is_read_only(parent);
1306 GLOBAL_STATE_CODE();
1309 ret = bdrv_reopen_set_read_only(parent, false, errp);
1315 ret = bdrv_change_backing_file(parent, filename,
1316 base->drv ? base->drv->format_name : "",
1319 error_setg_errno(errp, -ret, "Could not update backing file link");
1323 bdrv_reopen_set_read_only(parent, true, NULL);
1330 * Returns the options and flags that a generic child of a BDS should
1331 * get, based on the given options and flags for the parent BDS.
1333 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1334 int *child_flags, QDict *child_options,
1335 int parent_flags, QDict *parent_options)
1337 int flags = parent_flags;
1338 GLOBAL_STATE_CODE();
1341 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1342 * Generally, the question to answer is: Should this child be
1343 * format-probed by default?
1347 * Pure and non-filtered data children of non-format nodes should
1348 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1349 * set). This only affects a very limited set of drivers (namely
1350 * quorum and blkverify when this comment was written).
1351 * Force-clear BDRV_O_PROTOCOL then.
1353 if (!parent_is_format &&
1354 (role & BDRV_CHILD_DATA) &&
1355 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1357 flags &= ~BDRV_O_PROTOCOL;
1361 * All children of format nodes (except for COW children) and all
1362 * metadata children in general should never be format-probed.
1363 * Force-set BDRV_O_PROTOCOL then.
1365 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1366 (role & BDRV_CHILD_METADATA))
1368 flags |= BDRV_O_PROTOCOL;
1372 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1375 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1376 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1377 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1379 if (role & BDRV_CHILD_COW) {
1380 /* backing files are opened read-only by default */
1381 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1382 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1384 /* Inherit the read-only option from the parent if it's not set */
1385 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1386 qdict_copy_default(child_options, parent_options,
1387 BDRV_OPT_AUTO_READ_ONLY);
1391 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1392 * can default to enable it on lower layers regardless of the
1395 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1397 /* Clear flags that only apply to the top layer */
1398 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1400 if (role & BDRV_CHILD_METADATA) {
1401 flags &= ~BDRV_O_NO_IO;
1403 if (role & BDRV_CHILD_COW) {
1404 flags &= ~BDRV_O_TEMPORARY;
1407 *child_flags = flags;
1410 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1412 BlockDriverState *bs = child->opaque;
1414 assert_bdrv_graph_writable();
1415 QLIST_INSERT_HEAD(&bs->children, child, next);
1416 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1418 * Here we handle filters and block/raw-format.c when it behave like
1419 * filter. They generally have a single PRIMARY child, which is also the
1420 * FILTERED child, and that they may have multiple more children, which
1421 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1422 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1423 * into bs->backing on exceptional cases; and bs->backing will be
1426 assert(!(child->role & BDRV_CHILD_COW));
1427 if (child->role & BDRV_CHILD_PRIMARY) {
1428 assert(child->role & BDRV_CHILD_FILTERED);
1429 assert(!bs->backing);
1432 if (bs->drv->filtered_child_is_backing) {
1433 bs->backing = child;
1438 assert(!(child->role & BDRV_CHILD_FILTERED));
1440 } else if (child->role & BDRV_CHILD_COW) {
1441 assert(bs->drv->supports_backing);
1442 assert(!(child->role & BDRV_CHILD_PRIMARY));
1443 assert(!bs->backing);
1444 bs->backing = child;
1445 bdrv_backing_attach(child);
1446 } else if (child->role & BDRV_CHILD_PRIMARY) {
1452 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1454 BlockDriverState *bs = child->opaque;
1456 if (child->role & BDRV_CHILD_COW) {
1457 bdrv_backing_detach(child);
1460 assert_bdrv_graph_writable();
1461 QLIST_REMOVE(child, next);
1462 if (child == bs->backing) {
1463 assert(child != bs->file);
1465 } else if (child == bs->file) {
1470 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1471 const char *filename, Error **errp)
1473 if (c->role & BDRV_CHILD_COW) {
1474 return bdrv_backing_update_filename(c, base, filename, errp);
1479 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1481 BlockDriverState *bs = c->opaque;
1484 return bdrv_get_aio_context(bs);
1487 const BdrvChildClass child_of_bds = {
1488 .parent_is_bds = true,
1489 .get_parent_desc = bdrv_child_get_parent_desc,
1490 .inherit_options = bdrv_inherited_options,
1491 .drained_begin = bdrv_child_cb_drained_begin,
1492 .drained_poll = bdrv_child_cb_drained_poll,
1493 .drained_end = bdrv_child_cb_drained_end,
1494 .attach = bdrv_child_cb_attach,
1495 .detach = bdrv_child_cb_detach,
1496 .inactivate = bdrv_child_cb_inactivate,
1497 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1498 .update_filename = bdrv_child_cb_update_filename,
1499 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1502 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1505 return c->klass->get_parent_aio_context(c);
1508 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1510 int open_flags = flags;
1511 GLOBAL_STATE_CODE();
1514 * Clear flags that are internal to the block layer before opening the
1517 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1522 static void update_flags_from_options(int *flags, QemuOpts *opts)
1524 GLOBAL_STATE_CODE();
1526 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1528 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1529 *flags |= BDRV_O_NO_FLUSH;
1532 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1533 *flags |= BDRV_O_NOCACHE;
1536 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1537 *flags |= BDRV_O_RDWR;
1540 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1541 *flags |= BDRV_O_AUTO_RDONLY;
1545 static void update_options_from_flags(QDict *options, int flags)
1547 GLOBAL_STATE_CODE();
1548 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1549 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1551 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1552 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1553 flags & BDRV_O_NO_FLUSH);
1555 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1556 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1558 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1559 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1560 flags & BDRV_O_AUTO_RDONLY);
1564 static void bdrv_assign_node_name(BlockDriverState *bs,
1565 const char *node_name,
1568 char *gen_node_name = NULL;
1569 GLOBAL_STATE_CODE();
1572 node_name = gen_node_name = id_generate(ID_BLOCK);
1573 } else if (!id_wellformed(node_name)) {
1575 * Check for empty string or invalid characters, but not if it is
1576 * generated (generated names use characters not available to the user)
1578 error_setg(errp, "Invalid node-name: '%s'", node_name);
1582 /* takes care of avoiding namespaces collisions */
1583 if (blk_by_name(node_name)) {
1584 error_setg(errp, "node-name=%s is conflicting with a device id",
1589 /* takes care of avoiding duplicates node names */
1590 if (bdrv_find_node(node_name)) {
1591 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1595 /* Make sure that the node name isn't truncated */
1596 if (strlen(node_name) >= sizeof(bs->node_name)) {
1597 error_setg(errp, "Node name too long");
1601 /* copy node name into the bs and insert it into the graph list */
1602 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1603 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1605 g_free(gen_node_name);
1609 * The caller must always hold @bs AioContext lock, because this function calls
1610 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1613 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1614 const char *node_name, QDict *options,
1615 int open_flags, Error **errp)
1617 Error *local_err = NULL;
1619 GLOBAL_STATE_CODE();
1621 bdrv_assign_node_name(bs, node_name, &local_err);
1623 error_propagate(errp, local_err);
1628 bs->opaque = g_malloc0(drv->instance_size);
1630 if (drv->bdrv_file_open) {
1631 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1632 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1633 } else if (drv->bdrv_open) {
1634 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1641 error_propagate(errp, local_err);
1642 } else if (bs->filename[0]) {
1643 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1645 error_setg_errno(errp, -ret, "Could not open image");
1650 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1651 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1654 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1655 * drivers that pass read/write requests through to a child the trouble of
1656 * declaring support explicitly.
1658 * Drivers must not propagate this flag accidentally when they initiate I/O
1659 * to a bounce buffer. That case should be rare though.
1661 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1662 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1664 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
1666 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1670 bdrv_refresh_limits(bs, NULL, &local_err);
1672 error_propagate(errp, local_err);
1676 assert(bdrv_opt_mem_align(bs) != 0);
1677 assert(bdrv_min_mem_align(bs) != 0);
1678 assert(is_power_of_2(bs->bl.request_alignment));
1680 for (i = 0; i < bs->quiesce_counter; i++) {
1681 if (drv->bdrv_drain_begin) {
1682 drv->bdrv_drain_begin(bs);
1689 if (bs->file != NULL) {
1690 bdrv_unref_child(bs, bs->file);
1699 * Create and open a block node.
1701 * @options is a QDict of options to pass to the block drivers, or NULL for an
1702 * empty set of options. The reference to the QDict belongs to the block layer
1703 * after the call (even on failure), so if the caller intends to reuse the
1704 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1706 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1707 const char *node_name,
1708 QDict *options, int flags,
1711 BlockDriverState *bs;
1714 GLOBAL_STATE_CODE();
1717 bs->open_flags = flags;
1718 bs->options = options ?: qdict_new();
1719 bs->explicit_options = qdict_clone_shallow(bs->options);
1722 update_options_from_flags(bs->options, flags);
1724 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1726 qobject_unref(bs->explicit_options);
1727 bs->explicit_options = NULL;
1728 qobject_unref(bs->options);
1737 /* Create and open a block node. */
1738 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1739 int flags, Error **errp)
1741 GLOBAL_STATE_CODE();
1742 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1745 QemuOptsList bdrv_runtime_opts = {
1746 .name = "bdrv_common",
1747 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1750 .name = "node-name",
1751 .type = QEMU_OPT_STRING,
1752 .help = "Node name of the block device node",
1756 .type = QEMU_OPT_STRING,
1757 .help = "Block driver to use for the node",
1760 .name = BDRV_OPT_CACHE_DIRECT,
1761 .type = QEMU_OPT_BOOL,
1762 .help = "Bypass software writeback cache on the host",
1765 .name = BDRV_OPT_CACHE_NO_FLUSH,
1766 .type = QEMU_OPT_BOOL,
1767 .help = "Ignore flush requests",
1770 .name = BDRV_OPT_READ_ONLY,
1771 .type = QEMU_OPT_BOOL,
1772 .help = "Node is opened in read-only mode",
1775 .name = BDRV_OPT_AUTO_READ_ONLY,
1776 .type = QEMU_OPT_BOOL,
1777 .help = "Node can become read-only if opening read-write fails",
1780 .name = "detect-zeroes",
1781 .type = QEMU_OPT_STRING,
1782 .help = "try to optimize zero writes (off, on, unmap)",
1785 .name = BDRV_OPT_DISCARD,
1786 .type = QEMU_OPT_STRING,
1787 .help = "discard operation (ignore/off, unmap/on)",
1790 .name = BDRV_OPT_FORCE_SHARE,
1791 .type = QEMU_OPT_BOOL,
1792 .help = "always accept other writers (default: off)",
1794 { /* end of list */ }
1798 QemuOptsList bdrv_create_opts_simple = {
1799 .name = "simple-create-opts",
1800 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1803 .name = BLOCK_OPT_SIZE,
1804 .type = QEMU_OPT_SIZE,
1805 .help = "Virtual disk size"
1808 .name = BLOCK_OPT_PREALLOC,
1809 .type = QEMU_OPT_STRING,
1810 .help = "Preallocation mode (allowed values: off)"
1812 { /* end of list */ }
1817 * Common part for opening disk images and files
1819 * Removes all processed options from *options.
1821 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1822 QDict *options, Error **errp)
1824 int ret, open_flags;
1825 const char *filename;
1826 const char *driver_name = NULL;
1827 const char *node_name = NULL;
1828 const char *discard;
1831 Error *local_err = NULL;
1834 assert(bs->file == NULL);
1835 assert(options != NULL && bs->options != options);
1836 GLOBAL_STATE_CODE();
1838 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1839 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1844 update_flags_from_options(&bs->open_flags, opts);
1846 driver_name = qemu_opt_get(opts, "driver");
1847 drv = bdrv_find_format(driver_name);
1848 assert(drv != NULL);
1850 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1852 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1854 BDRV_OPT_FORCE_SHARE
1855 "=on can only be used with read-only images");
1861 bdrv_refresh_filename(blk_bs(file));
1862 filename = blk_bs(file)->filename;
1865 * Caution: while qdict_get_try_str() is fine, getting
1866 * non-string types would require more care. When @options
1867 * come from -blockdev or blockdev_add, its members are typed
1868 * according to the QAPI schema, but when they come from
1869 * -drive, they're all QString.
1871 filename = qdict_get_try_str(options, "filename");
1874 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1875 error_setg(errp, "The '%s' block driver requires a file name",
1881 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1884 ro = bdrv_is_read_only(bs);
1886 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1887 if (!ro && bdrv_is_whitelisted(drv, true)) {
1888 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1894 !ro && bdrv_is_whitelisted(drv, true)
1895 ? "Driver '%s' can only be used for read-only devices"
1896 : "Driver '%s' is not whitelisted",
1902 /* bdrv_new() and bdrv_close() make it so */
1903 assert(qatomic_read(&bs->copy_on_read) == 0);
1905 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1907 bdrv_enable_copy_on_read(bs);
1909 error_setg(errp, "Can't use copy-on-read on read-only device");
1915 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1916 if (discard != NULL) {
1917 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1918 error_setg(errp, "Invalid discard option");
1925 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1927 error_propagate(errp, local_err);
1932 if (filename != NULL) {
1933 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1935 bs->filename[0] = '\0';
1937 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1939 /* Open the image, either directly or using a protocol */
1940 open_flags = bdrv_open_flags(bs, bs->open_flags);
1941 node_name = qemu_opt_get(opts, "node-name");
1943 assert(!drv->bdrv_file_open || file == NULL);
1944 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1949 qemu_opts_del(opts);
1953 qemu_opts_del(opts);
1957 static QDict *parse_json_filename(const char *filename, Error **errp)
1959 QObject *options_obj;
1962 GLOBAL_STATE_CODE();
1964 ret = strstart(filename, "json:", &filename);
1967 options_obj = qobject_from_json(filename, errp);
1969 error_prepend(errp, "Could not parse the JSON options: ");
1973 options = qobject_to(QDict, options_obj);
1975 qobject_unref(options_obj);
1976 error_setg(errp, "Invalid JSON object given");
1980 qdict_flatten(options);
1985 static void parse_json_protocol(QDict *options, const char **pfilename,
1988 QDict *json_options;
1989 Error *local_err = NULL;
1990 GLOBAL_STATE_CODE();
1992 /* Parse json: pseudo-protocol */
1993 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1997 json_options = parse_json_filename(*pfilename, &local_err);
1999 error_propagate(errp, local_err);
2003 /* Options given in the filename have lower priority than options
2004 * specified directly */
2005 qdict_join(options, json_options, false);
2006 qobject_unref(json_options);
2011 * Fills in default options for opening images and converts the legacy
2012 * filename/flags pair to option QDict entries.
2013 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2014 * block driver has been specified explicitly.
2016 static int bdrv_fill_options(QDict **options, const char *filename,
2017 int *flags, Error **errp)
2019 const char *drvname;
2020 bool protocol = *flags & BDRV_O_PROTOCOL;
2021 bool parse_filename = false;
2022 BlockDriver *drv = NULL;
2023 Error *local_err = NULL;
2025 GLOBAL_STATE_CODE();
2028 * Caution: while qdict_get_try_str() is fine, getting non-string
2029 * types would require more care. When @options come from
2030 * -blockdev or blockdev_add, its members are typed according to
2031 * the QAPI schema, but when they come from -drive, they're all
2034 drvname = qdict_get_try_str(*options, "driver");
2036 drv = bdrv_find_format(drvname);
2038 error_setg(errp, "Unknown driver '%s'", drvname);
2041 /* If the user has explicitly specified the driver, this choice should
2042 * override the BDRV_O_PROTOCOL flag */
2043 protocol = drv->bdrv_file_open;
2047 *flags |= BDRV_O_PROTOCOL;
2049 *flags &= ~BDRV_O_PROTOCOL;
2052 /* Translate cache options from flags into options */
2053 update_options_from_flags(*options, *flags);
2055 /* Fetch the file name from the options QDict if necessary */
2056 if (protocol && filename) {
2057 if (!qdict_haskey(*options, "filename")) {
2058 qdict_put_str(*options, "filename", filename);
2059 parse_filename = true;
2061 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2067 /* Find the right block driver */
2068 /* See cautionary note on accessing @options above */
2069 filename = qdict_get_try_str(*options, "filename");
2071 if (!drvname && protocol) {
2073 drv = bdrv_find_protocol(filename, parse_filename, errp);
2078 drvname = drv->format_name;
2079 qdict_put_str(*options, "driver", drvname);
2081 error_setg(errp, "Must specify either driver or file");
2086 assert(drv || !protocol);
2088 /* Driver-specific filename parsing */
2089 if (drv && drv->bdrv_parse_filename && parse_filename) {
2090 drv->bdrv_parse_filename(filename, *options, &local_err);
2092 error_propagate(errp, local_err);
2096 if (!drv->bdrv_needs_filename) {
2097 qdict_del(*options, "filename");
2104 typedef struct BlockReopenQueueEntry {
2107 BDRVReopenState state;
2108 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2109 } BlockReopenQueueEntry;
2112 * Return the flags that @bs will have after the reopens in @q have
2113 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2114 * return the current flags.
2116 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2118 BlockReopenQueueEntry *entry;
2121 QTAILQ_FOREACH(entry, q, entry) {
2122 if (entry->state.bs == bs) {
2123 return entry->state.flags;
2128 return bs->open_flags;
2131 /* Returns whether the image file can be written to after the reopen queue @q
2132 * has been successfully applied, or right now if @q is NULL. */
2133 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2134 BlockReopenQueue *q)
2136 int flags = bdrv_reopen_get_flags(q, bs);
2138 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2142 * Return whether the BDS can be written to. This is not necessarily
2143 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2144 * be written to but do not count as read-only images.
2146 bool bdrv_is_writable(BlockDriverState *bs)
2149 return bdrv_is_writable_after_reopen(bs, NULL);
2152 static char *bdrv_child_user_desc(BdrvChild *c)
2154 GLOBAL_STATE_CODE();
2155 return c->klass->get_parent_desc(c);
2159 * Check that @a allows everything that @b needs. @a and @b must reference same
2162 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2164 const char *child_bs_name;
2165 g_autofree char *a_user = NULL;
2166 g_autofree char *b_user = NULL;
2167 g_autofree char *perms = NULL;
2170 assert(a->bs == b->bs);
2171 GLOBAL_STATE_CODE();
2173 if ((b->perm & a->shared_perm) == b->perm) {
2177 child_bs_name = bdrv_get_node_name(b->bs);
2178 a_user = bdrv_child_user_desc(a);
2179 b_user = bdrv_child_user_desc(b);
2180 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2182 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2183 "both required by %s (uses node '%s' as '%s' child) and "
2184 "unshared by %s (uses node '%s' as '%s' child).",
2185 child_bs_name, perms,
2186 b_user, child_bs_name, b->name,
2187 a_user, child_bs_name, a->name);
2192 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2195 GLOBAL_STATE_CODE();
2198 * During the loop we'll look at each pair twice. That's correct because
2199 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2202 QLIST_FOREACH(a, &bs->parents, next_parent) {
2203 QLIST_FOREACH(b, &bs->parents, next_parent) {
2208 if (!bdrv_a_allow_b(a, b, errp)) {
2217 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2218 BdrvChild *c, BdrvChildRole role,
2219 BlockReopenQueue *reopen_queue,
2220 uint64_t parent_perm, uint64_t parent_shared,
2221 uint64_t *nperm, uint64_t *nshared)
2223 assert(bs->drv && bs->drv->bdrv_child_perm);
2224 GLOBAL_STATE_CODE();
2225 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2226 parent_perm, parent_shared,
2228 /* TODO Take force_share from reopen_queue */
2229 if (child_bs && child_bs->force_share) {
2230 *nshared = BLK_PERM_ALL;
2235 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2236 * nodes that are already in the @list, of course) so that final list is
2237 * topologically sorted. Return the result (GSList @list object is updated, so
2238 * don't use old reference after function call).
2240 * On function start @list must be already topologically sorted and for any node
2241 * in the @list the whole subtree of the node must be in the @list as well. The
2242 * simplest way to satisfy this criteria: use only result of
2243 * bdrv_topological_dfs() or NULL as @list parameter.
2245 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2246 BlockDriverState *bs)
2249 g_autoptr(GHashTable) local_found = NULL;
2251 GLOBAL_STATE_CODE();
2255 found = local_found = g_hash_table_new(NULL, NULL);
2258 if (g_hash_table_contains(found, bs)) {
2261 g_hash_table_add(found, bs);
2263 QLIST_FOREACH(child, &bs->children, next) {
2264 list = bdrv_topological_dfs(list, found, child->bs);
2267 return g_slist_prepend(list, bs);
2270 typedef struct BdrvChildSetPermState {
2273 uint64_t old_shared_perm;
2274 } BdrvChildSetPermState;
2276 static void bdrv_child_set_perm_abort(void *opaque)
2278 BdrvChildSetPermState *s = opaque;
2280 GLOBAL_STATE_CODE();
2282 s->child->perm = s->old_perm;
2283 s->child->shared_perm = s->old_shared_perm;
2286 static TransactionActionDrv bdrv_child_set_pem_drv = {
2287 .abort = bdrv_child_set_perm_abort,
2291 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2292 uint64_t shared, Transaction *tran)
2294 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2295 GLOBAL_STATE_CODE();
2297 *s = (BdrvChildSetPermState) {
2299 .old_perm = c->perm,
2300 .old_shared_perm = c->shared_perm,
2304 c->shared_perm = shared;
2306 tran_add(tran, &bdrv_child_set_pem_drv, s);
2309 static void bdrv_drv_set_perm_commit(void *opaque)
2311 BlockDriverState *bs = opaque;
2312 uint64_t cumulative_perms, cumulative_shared_perms;
2313 GLOBAL_STATE_CODE();
2315 if (bs->drv->bdrv_set_perm) {
2316 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2317 &cumulative_shared_perms);
2318 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2322 static void bdrv_drv_set_perm_abort(void *opaque)
2324 BlockDriverState *bs = opaque;
2325 GLOBAL_STATE_CODE();
2327 if (bs->drv->bdrv_abort_perm_update) {
2328 bs->drv->bdrv_abort_perm_update(bs);
2332 TransactionActionDrv bdrv_drv_set_perm_drv = {
2333 .abort = bdrv_drv_set_perm_abort,
2334 .commit = bdrv_drv_set_perm_commit,
2337 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2338 uint64_t shared_perm, Transaction *tran,
2341 GLOBAL_STATE_CODE();
2346 if (bs->drv->bdrv_check_perm) {
2347 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2354 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2360 typedef struct BdrvReplaceChildState {
2362 BlockDriverState *old_bs;
2363 } BdrvReplaceChildState;
2365 static void bdrv_replace_child_commit(void *opaque)
2367 BdrvReplaceChildState *s = opaque;
2368 GLOBAL_STATE_CODE();
2370 bdrv_unref(s->old_bs);
2373 static void bdrv_replace_child_abort(void *opaque)
2375 BdrvReplaceChildState *s = opaque;
2376 BlockDriverState *new_bs = s->child->bs;
2378 GLOBAL_STATE_CODE();
2379 /* old_bs reference is transparently moved from @s to @s->child */
2380 if (!s->child->bs) {
2382 * The parents were undrained when removing old_bs from the child. New
2383 * requests can't have been made, though, because the child was empty.
2385 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2386 * undraining the parent in the first place. Once this is done, having
2387 * new_bs drained when calling bdrv_replace_child_tran() is not a
2388 * requirement any more.
2390 bdrv_parent_drained_begin_single(s->child);
2391 assert(!bdrv_parent_drained_poll_single(s->child));
2393 assert(s->child->quiesced_parent);
2394 bdrv_replace_child_noperm(s->child, s->old_bs);
2398 static TransactionActionDrv bdrv_replace_child_drv = {
2399 .commit = bdrv_replace_child_commit,
2400 .abort = bdrv_replace_child_abort,
2405 * bdrv_replace_child_tran
2407 * Note: real unref of old_bs is done only on commit.
2409 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2410 * kept drained until the transaction is completed.
2412 * The function doesn't update permissions, caller is responsible for this.
2414 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2417 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2419 assert(child->quiesced_parent);
2420 assert(!new_bs || new_bs->quiesce_counter);
2422 *s = (BdrvReplaceChildState) {
2424 .old_bs = child->bs,
2426 tran_add(tran, &bdrv_replace_child_drv, s);
2431 bdrv_replace_child_noperm(child, new_bs);
2432 /* old_bs reference is transparently moved from @child to @s */
2436 * Refresh permissions in @bs subtree. The function is intended to be called
2437 * after some graph modification that was done without permission update.
2439 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2440 Transaction *tran, Error **errp)
2442 BlockDriver *drv = bs->drv;
2445 uint64_t cumulative_perms, cumulative_shared_perms;
2446 GLOBAL_STATE_CODE();
2448 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2450 /* Write permissions never work with read-only images */
2451 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2452 !bdrv_is_writable_after_reopen(bs, q))
2454 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2455 error_setg(errp, "Block node is read-only");
2457 error_setg(errp, "Read-only block node '%s' cannot support "
2458 "read-write users", bdrv_get_node_name(bs));
2465 * Unaligned requests will automatically be aligned to bl.request_alignment
2466 * and without RESIZE we can't extend requests to write to space beyond the
2467 * end of the image, so it's required that the image size is aligned.
2469 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2470 !(cumulative_perms & BLK_PERM_RESIZE))
2472 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2473 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2474 "Image size is not a multiple of request "
2480 /* Check this node */
2485 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2491 /* Drivers that never have children can omit .bdrv_child_perm() */
2492 if (!drv->bdrv_child_perm) {
2493 assert(QLIST_EMPTY(&bs->children));
2497 /* Check all children */
2498 QLIST_FOREACH(c, &bs->children, next) {
2499 uint64_t cur_perm, cur_shared;
2501 bdrv_child_perm(bs, c->bs, c, c->role, q,
2502 cumulative_perms, cumulative_shared_perms,
2503 &cur_perm, &cur_shared);
2504 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2511 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2512 * a topologically sorted subgraph.
2514 static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2515 Transaction *tran, Error **errp)
2518 BlockDriverState *bs;
2519 GLOBAL_STATE_CODE();
2521 for ( ; list; list = list->next) {
2524 if (bdrv_parent_perms_conflict(bs, errp)) {
2528 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2538 * @list is any list of nodes. List is completed by all subtrees and
2539 * topologically sorted. It's not a problem if some node occurs in the @list
2542 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2543 Transaction *tran, Error **errp)
2545 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2546 g_autoptr(GSList) refresh_list = NULL;
2548 for ( ; list; list = list->next) {
2549 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2552 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2555 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2556 uint64_t *shared_perm)
2559 uint64_t cumulative_perms = 0;
2560 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2562 GLOBAL_STATE_CODE();
2564 QLIST_FOREACH(c, &bs->parents, next_parent) {
2565 cumulative_perms |= c->perm;
2566 cumulative_shared_perms &= c->shared_perm;
2569 *perm = cumulative_perms;
2570 *shared_perm = cumulative_shared_perms;
2573 char *bdrv_perm_names(uint64_t perm)
2579 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2580 { BLK_PERM_WRITE, "write" },
2581 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2582 { BLK_PERM_RESIZE, "resize" },
2586 GString *result = g_string_sized_new(30);
2587 struct perm_name *p;
2589 for (p = permissions; p->name; p++) {
2590 if (perm & p->perm) {
2591 if (result->len > 0) {
2592 g_string_append(result, ", ");
2594 g_string_append(result, p->name);
2598 return g_string_free(result, FALSE);
2602 /* @tran is allowed to be NULL. In this case no rollback is possible */
2603 static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2607 Transaction *local_tran = NULL;
2608 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2609 GLOBAL_STATE_CODE();
2612 tran = local_tran = tran_new();
2615 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2618 tran_finalize(local_tran, ret);
2624 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2627 Error *local_err = NULL;
2628 Transaction *tran = tran_new();
2631 GLOBAL_STATE_CODE();
2633 bdrv_child_set_perm(c, perm, shared, tran);
2635 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2637 tran_finalize(tran, ret);
2640 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2641 /* tighten permissions */
2642 error_propagate(errp, local_err);
2645 * Our caller may intend to only loosen restrictions and
2646 * does not expect this function to fail. Errors are not
2647 * fatal in such a case, so we can just hide them from our
2650 error_free(local_err);
2658 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2660 uint64_t parent_perms, parent_shared;
2661 uint64_t perms, shared;
2663 GLOBAL_STATE_CODE();
2665 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2666 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2667 parent_perms, parent_shared, &perms, &shared);
2669 return bdrv_child_try_set_perm(c, perms, shared, errp);
2673 * Default implementation for .bdrv_child_perm() for block filters:
2674 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2677 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2679 BlockReopenQueue *reopen_queue,
2680 uint64_t perm, uint64_t shared,
2681 uint64_t *nperm, uint64_t *nshared)
2683 GLOBAL_STATE_CODE();
2684 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2685 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2688 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2690 BlockReopenQueue *reopen_queue,
2691 uint64_t perm, uint64_t shared,
2692 uint64_t *nperm, uint64_t *nshared)
2694 assert(role & BDRV_CHILD_COW);
2695 GLOBAL_STATE_CODE();
2698 * We want consistent read from backing files if the parent needs it.
2699 * No other operations are performed on backing files.
2701 perm &= BLK_PERM_CONSISTENT_READ;
2704 * If the parent can deal with changing data, we're okay with a
2705 * writable and resizable backing file.
2706 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2708 if (shared & BLK_PERM_WRITE) {
2709 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2714 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2716 if (bs->open_flags & BDRV_O_INACTIVE) {
2717 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2724 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2726 BlockReopenQueue *reopen_queue,
2727 uint64_t perm, uint64_t shared,
2728 uint64_t *nperm, uint64_t *nshared)
2732 GLOBAL_STATE_CODE();
2733 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2735 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2738 * Apart from the modifications below, the same permissions are
2739 * forwarded and left alone as for filters
2741 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2742 perm, shared, &perm, &shared);
2744 if (role & BDRV_CHILD_METADATA) {
2745 /* Format drivers may touch metadata even if the guest doesn't write */
2746 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2747 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2751 * bs->file always needs to be consistent because of the
2752 * metadata. We can never allow other users to resize or write
2755 if (!(flags & BDRV_O_NO_IO)) {
2756 perm |= BLK_PERM_CONSISTENT_READ;
2758 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2761 if (role & BDRV_CHILD_DATA) {
2763 * Technically, everything in this block is a subset of the
2764 * BDRV_CHILD_METADATA path taken above, and so this could
2765 * be an "else if" branch. However, that is not obvious, and
2766 * this function is not performance critical, therefore we let
2767 * this be an independent "if".
2771 * We cannot allow other users to resize the file because the
2772 * format driver might have some assumptions about the size
2773 * (e.g. because it is stored in metadata, or because the file
2774 * is split into fixed-size data files).
2776 shared &= ~BLK_PERM_RESIZE;
2779 * WRITE_UNCHANGED often cannot be performed as such on the
2780 * data file. For example, the qcow2 driver may still need to
2781 * write copied clusters on copy-on-read.
2783 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2784 perm |= BLK_PERM_WRITE;
2788 * If the data file is written to, the format driver may
2789 * expect to be able to resize it by writing beyond the EOF.
2791 if (perm & BLK_PERM_WRITE) {
2792 perm |= BLK_PERM_RESIZE;
2796 if (bs->open_flags & BDRV_O_INACTIVE) {
2797 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2804 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2805 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2806 uint64_t perm, uint64_t shared,
2807 uint64_t *nperm, uint64_t *nshared)
2809 GLOBAL_STATE_CODE();
2810 if (role & BDRV_CHILD_FILTERED) {
2811 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2813 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2814 perm, shared, nperm, nshared);
2815 } else if (role & BDRV_CHILD_COW) {
2816 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2817 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2818 perm, shared, nperm, nshared);
2819 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2820 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2821 perm, shared, nperm, nshared);
2823 g_assert_not_reached();
2827 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2829 static const uint64_t permissions[] = {
2830 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2831 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2832 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2833 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2836 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2837 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2839 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2841 return permissions[qapi_perm];
2845 * Replaces the node that a BdrvChild points to without updating permissions.
2847 * If @new_bs is non-NULL, the parent of @child must already be drained through
2850 static void bdrv_replace_child_noperm(BdrvChild *child,
2851 BlockDriverState *new_bs)
2853 BlockDriverState *old_bs = child->bs;
2854 int new_bs_quiesce_counter;
2856 assert(!child->frozen);
2859 * If we want to change the BdrvChild to point to a drained node as its new
2860 * child->bs, we need to make sure that its new parent is drained, too. In
2861 * other words, either child->quiesce_parent must already be true or we must
2862 * be able to set it and keep the parent's quiesce_counter consistent with
2863 * that, but without polling or starting new requests (this function
2864 * guarantees that it doesn't poll, and starting new requests would be
2865 * against the invariants of drain sections).
2867 * To keep things simple, we pick the first option (child->quiesce_parent
2868 * must already be true). We also generalise the rule a bit to make it
2869 * easier to verify in callers and more likely to be covered in test cases:
2870 * The parent must be quiesced through this child even if new_bs isn't
2871 * currently drained.
2873 * The only exception is for callers that always pass new_bs == NULL. In
2874 * this case, we obviously never need to consider the case of a drained
2875 * new_bs, so we can keep the callers simpler by allowing them not to drain
2878 assert(!new_bs || child->quiesced_parent);
2879 assert(old_bs != new_bs);
2880 GLOBAL_STATE_CODE();
2882 if (old_bs && new_bs) {
2883 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2886 /* TODO Pull this up into the callers to avoid polling here */
2887 bdrv_graph_wrlock();
2889 if (child->klass->detach) {
2890 child->klass->detach(child);
2892 QLIST_REMOVE(child, next_parent);
2898 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2899 if (child->klass->attach) {
2900 child->klass->attach(child);
2903 bdrv_graph_wrunlock();
2906 * If the parent was drained through this BdrvChild previously, but new_bs
2907 * is not drained, allow requests to come in only after the new node has
2910 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2911 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2912 bdrv_parent_drained_end_single(child);
2917 * Free the given @child.
2919 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2920 * unused (i.e. not in a children list).
2922 static void bdrv_child_free(BdrvChild *child)
2925 GLOBAL_STATE_CODE();
2926 assert(!child->next.le_prev); /* not in children list */
2928 g_free(child->name);
2932 typedef struct BdrvAttachChildCommonState {
2934 AioContext *old_parent_ctx;
2935 AioContext *old_child_ctx;
2936 } BdrvAttachChildCommonState;
2938 static void bdrv_attach_child_common_abort(void *opaque)
2940 BdrvAttachChildCommonState *s = opaque;
2941 BlockDriverState *bs = s->child->bs;
2943 GLOBAL_STATE_CODE();
2944 bdrv_replace_child_noperm(s->child, NULL);
2946 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2947 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2950 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2952 GHashTable *visited;
2957 /* No need to visit `child`, because it has been detached already */
2958 visited = g_hash_table_new(NULL, NULL);
2959 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2960 visited, tran, &error_abort);
2961 g_hash_table_destroy(visited);
2963 /* transaction is supposed to always succeed */
2964 assert(ret == true);
2969 bdrv_child_free(s->child);
2972 static TransactionActionDrv bdrv_attach_child_common_drv = {
2973 .abort = bdrv_attach_child_common_abort,
2978 * Common part of attaching bdrv child to bs or to blk or to job
2980 * Function doesn't update permissions, caller is responsible for this.
2982 * Returns new created child.
2984 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2985 const char *child_name,
2986 const BdrvChildClass *child_class,
2987 BdrvChildRole child_role,
2988 uint64_t perm, uint64_t shared_perm,
2990 Transaction *tran, Error **errp)
2992 BdrvChild *new_child;
2993 AioContext *parent_ctx;
2994 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2996 assert(child_class->get_parent_desc);
2997 GLOBAL_STATE_CODE();
2999 new_child = g_new(BdrvChild, 1);
3000 *new_child = (BdrvChild) {
3002 .name = g_strdup(child_name),
3003 .klass = child_class,
3006 .shared_perm = shared_perm,
3011 * If the AioContexts don't match, first try to move the subtree of
3012 * child_bs into the AioContext of the new parent. If this doesn't work,
3013 * try moving the parent into the AioContext of child_bs instead.
3015 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3016 if (child_ctx != parent_ctx) {
3017 Error *local_err = NULL;
3018 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3021 if (ret < 0 && child_class->change_aio_ctx) {
3022 Transaction *tran = tran_new();
3023 GHashTable *visited = g_hash_table_new(NULL, NULL);
3026 g_hash_table_add(visited, new_child);
3027 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3028 visited, tran, NULL);
3029 if (ret_child == true) {
3030 error_free(local_err);
3033 tran_finalize(tran, ret_child == true ? 0 : -1);
3034 g_hash_table_destroy(visited);
3038 error_propagate(errp, local_err);
3039 bdrv_child_free(new_child);
3046 * Let every new BdrvChild start with a drained parent. Inserting the child
3047 * in the graph with bdrv_replace_child_noperm() will undrain it if
3048 * @child_bs is not drained.
3050 * The child was only just created and is not yet visible in global state
3051 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3052 * could have sent requests and polling is not necessary.
3054 * Note that this means that the parent isn't fully drained yet, we only
3055 * stop new requests from coming in. This is fine, we don't care about the
3056 * old requests here, they are not for this child. If another place enters a
3057 * drain section for the same parent, but wants it to be fully quiesced, it
3058 * will not run most of the the code in .drained_begin() again (which is not
3059 * a problem, we already did this), but it will still poll until the parent
3060 * is fully quiesced, so it will not be negatively affected either.
3062 bdrv_parent_drained_begin_single(new_child);
3063 bdrv_replace_child_noperm(new_child, child_bs);
3065 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3066 *s = (BdrvAttachChildCommonState) {
3068 .old_parent_ctx = parent_ctx,
3069 .old_child_ctx = child_ctx,
3071 tran_add(tran, &bdrv_attach_child_common_drv, s);
3077 * Function doesn't update permissions, caller is responsible for this.
3079 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3080 BlockDriverState *child_bs,
3081 const char *child_name,
3082 const BdrvChildClass *child_class,
3083 BdrvChildRole child_role,
3087 uint64_t perm, shared_perm;
3089 assert(parent_bs->drv);
3090 GLOBAL_STATE_CODE();
3092 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3093 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3094 child_bs->node_name, child_name, parent_bs->node_name);
3098 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3099 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3100 perm, shared_perm, &perm, &shared_perm);
3102 return bdrv_attach_child_common(child_bs, child_name, child_class,
3103 child_role, perm, shared_perm, parent_bs,
3108 * This function steals the reference to child_bs from the caller.
3109 * That reference is later dropped by bdrv_root_unref_child().
3111 * On failure NULL is returned, errp is set and the reference to
3112 * child_bs is also dropped.
3114 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3115 * (unless @child_bs is already in @ctx).
3117 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3118 const char *child_name,
3119 const BdrvChildClass *child_class,
3120 BdrvChildRole child_role,
3121 uint64_t perm, uint64_t shared_perm,
3122 void *opaque, Error **errp)
3126 Transaction *tran = tran_new();
3128 GLOBAL_STATE_CODE();
3130 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3131 child_role, perm, shared_perm, opaque,
3138 ret = bdrv_refresh_perms(child_bs, tran, errp);
3141 tran_finalize(tran, ret);
3143 bdrv_unref(child_bs);
3145 return ret < 0 ? NULL : child;
3149 * This function transfers the reference to child_bs from the caller
3150 * to parent_bs. That reference is later dropped by parent_bs on
3151 * bdrv_close() or if someone calls bdrv_unref_child().
3153 * On failure NULL is returned, errp is set and the reference to
3154 * child_bs is also dropped.
3156 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3157 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3159 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3160 BlockDriverState *child_bs,
3161 const char *child_name,
3162 const BdrvChildClass *child_class,
3163 BdrvChildRole child_role,
3168 Transaction *tran = tran_new();
3170 GLOBAL_STATE_CODE();
3172 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3173 child_class, child_role, tran, errp);
3179 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3185 tran_finalize(tran, ret);
3187 bdrv_unref(child_bs);
3189 return ret < 0 ? NULL : child;
3192 /* Callers must ensure that child->frozen is false. */
3193 void bdrv_root_unref_child(BdrvChild *child)
3195 BlockDriverState *child_bs = child->bs;
3197 GLOBAL_STATE_CODE();
3198 bdrv_replace_child_noperm(child, NULL);
3199 bdrv_child_free(child);
3203 * Update permissions for old node. We're just taking a parent away, so
3204 * we're loosening restrictions. Errors of permission update are not
3205 * fatal in this case, ignore them.
3207 bdrv_refresh_perms(child_bs, NULL, NULL);
3210 * When the parent requiring a non-default AioContext is removed, the
3211 * node moves back to the main AioContext
3213 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3217 bdrv_unref(child_bs);
3220 typedef struct BdrvSetInheritsFrom {
3221 BlockDriverState *bs;
3222 BlockDriverState *old_inherits_from;
3223 } BdrvSetInheritsFrom;
3225 static void bdrv_set_inherits_from_abort(void *opaque)
3227 BdrvSetInheritsFrom *s = opaque;
3229 s->bs->inherits_from = s->old_inherits_from;
3232 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3233 .abort = bdrv_set_inherits_from_abort,
3237 /* @tran is allowed to be NULL. In this case no rollback is possible */
3238 static void bdrv_set_inherits_from(BlockDriverState *bs,
3239 BlockDriverState *new_inherits_from,
3243 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3245 *s = (BdrvSetInheritsFrom) {
3247 .old_inherits_from = bs->inherits_from,
3250 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3253 bs->inherits_from = new_inherits_from;
3257 * Clear all inherits_from pointers from children and grandchildren of
3258 * @root that point to @root, where necessary.
3259 * @tran is allowed to be NULL. In this case no rollback is possible
3261 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3266 if (child->bs->inherits_from == root) {
3268 * Remove inherits_from only when the last reference between root and
3269 * child->bs goes away.
3271 QLIST_FOREACH(c, &root->children, next) {
3272 if (c != child && c->bs == child->bs) {
3277 bdrv_set_inherits_from(child->bs, NULL, tran);
3281 QLIST_FOREACH(c, &child->bs->children, next) {
3282 bdrv_unset_inherits_from(root, c, tran);
3286 /* Callers must ensure that child->frozen is false. */
3287 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3289 GLOBAL_STATE_CODE();
3290 if (child == NULL) {
3294 bdrv_unset_inherits_from(parent, child, NULL);
3295 bdrv_root_unref_child(child);
3299 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3302 GLOBAL_STATE_CODE();
3303 QLIST_FOREACH(c, &bs->parents, next_parent) {
3304 if (c->klass->change_media) {
3305 c->klass->change_media(c, load);
3310 /* Return true if you can reach parent going through child->inherits_from
3311 * recursively. If parent or child are NULL, return false */
3312 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3313 BlockDriverState *parent)
3315 while (child && child != parent) {
3316 child = child->inherits_from;
3319 return child != NULL;
3323 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3324 * mostly used for COW backing children (role = COW), but also for
3325 * filtered children (role = FILTERED | PRIMARY).
3327 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3329 if (bs->drv && bs->drv->is_filter) {
3330 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3332 return BDRV_CHILD_COW;
3337 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3338 * callers which don't need their own reference any more must call bdrv_unref().
3340 * Function doesn't update permissions, caller is responsible for this.
3342 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3343 BlockDriverState *child_bs,
3345 Transaction *tran, Error **errp)
3347 bool update_inherits_from =
3348 bdrv_inherits_from_recursive(child_bs, parent_bs);
3349 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3352 GLOBAL_STATE_CODE();
3354 if (!parent_bs->drv) {
3356 * Node without drv is an object without a class :/. TODO: finally fix
3357 * qcow2 driver to never clear bs->drv and implement format corruption
3358 * handling in other way.
3360 error_setg(errp, "Node corrupted");
3364 if (child && child->frozen) {
3365 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3366 child->name, parent_bs->node_name, child->bs->node_name);
3370 if (is_backing && !parent_bs->drv->is_filter &&
3371 !parent_bs->drv->supports_backing)
3373 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3374 "files", parent_bs->drv->format_name, parent_bs->node_name);
3378 if (parent_bs->drv->is_filter) {
3379 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3380 } else if (is_backing) {
3381 role = BDRV_CHILD_COW;
3384 * We only can use same role as it is in existing child. We don't have
3385 * infrastructure to determine role of file child in generic way
3388 error_setg(errp, "Cannot set file child to format node without "
3396 bdrv_unset_inherits_from(parent_bs, child, tran);
3397 bdrv_remove_child(child, tran);
3404 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3405 is_backing ? "backing" : "file",
3406 &child_of_bds, role,
3414 * If inherits_from pointed recursively to bs then let's update it to
3415 * point directly to bs (else it will become NULL).
3417 if (update_inherits_from) {
3418 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3422 bdrv_refresh_limits(parent_bs, tran, NULL);
3427 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3428 BlockDriverState *backing_hd,
3429 Transaction *tran, Error **errp)
3431 GLOBAL_STATE_CODE();
3432 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3435 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3436 BlockDriverState *backing_hd,
3440 Transaction *tran = tran_new();
3442 GLOBAL_STATE_CODE();
3443 assert(bs->quiesce_counter > 0);
3445 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3450 ret = bdrv_refresh_perms(bs, tran, errp);
3452 tran_finalize(tran, ret);
3456 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3460 GLOBAL_STATE_CODE();
3462 bdrv_drained_begin(bs);
3463 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3464 bdrv_drained_end(bs);
3470 * Opens the backing file for a BlockDriverState if not yet open
3472 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3473 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3474 * itself, all options starting with "${bdref_key}." are considered part of the
3477 * TODO Can this be unified with bdrv_open_image()?
3479 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3480 const char *bdref_key, Error **errp)
3482 char *backing_filename = NULL;
3483 char *bdref_key_dot;
3484 const char *reference = NULL;
3486 bool implicit_backing = false;
3487 BlockDriverState *backing_hd;
3489 QDict *tmp_parent_options = NULL;
3490 Error *local_err = NULL;
3492 GLOBAL_STATE_CODE();
3494 if (bs->backing != NULL) {
3498 /* NULL means an empty set of options */
3499 if (parent_options == NULL) {
3500 tmp_parent_options = qdict_new();
3501 parent_options = tmp_parent_options;
3504 bs->open_flags &= ~BDRV_O_NO_BACKING;
3506 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3507 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3508 g_free(bdref_key_dot);
3511 * Caution: while qdict_get_try_str() is fine, getting non-string
3512 * types would require more care. When @parent_options come from
3513 * -blockdev or blockdev_add, its members are typed according to
3514 * the QAPI schema, but when they come from -drive, they're all
3517 reference = qdict_get_try_str(parent_options, bdref_key);
3518 if (reference || qdict_haskey(options, "file.filename")) {
3519 /* keep backing_filename NULL */
3520 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3521 qobject_unref(options);
3524 if (qdict_size(options) == 0) {
3525 /* If the user specifies options that do not modify the
3526 * backing file's behavior, we might still consider it the
3527 * implicit backing file. But it's easier this way, and
3528 * just specifying some of the backing BDS's options is
3529 * only possible with -drive anyway (otherwise the QAPI
3530 * schema forces the user to specify everything). */
3531 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3534 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3537 error_propagate(errp, local_err);
3538 qobject_unref(options);
3543 if (!bs->drv || !bs->drv->supports_backing) {
3545 error_setg(errp, "Driver doesn't support backing files");
3546 qobject_unref(options);
3551 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3552 qdict_put_str(options, "driver", bs->backing_format);
3555 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3556 &child_of_bds, bdrv_backing_role(bs), errp);
3558 bs->open_flags |= BDRV_O_NO_BACKING;
3559 error_prepend(errp, "Could not open backing file: ");
3564 if (implicit_backing) {
3565 bdrv_refresh_filename(backing_hd);
3566 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3567 backing_hd->filename);
3570 /* Hook up the backing file link; drop our reference, bs owns the
3571 * backing_hd reference now */
3572 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3573 bdrv_unref(backing_hd);
3578 qdict_del(parent_options, bdref_key);
3581 g_free(backing_filename);
3582 qobject_unref(tmp_parent_options);
3586 static BlockDriverState *
3587 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3588 BlockDriverState *parent, const BdrvChildClass *child_class,
3589 BdrvChildRole child_role, bool allow_none, Error **errp)
3591 BlockDriverState *bs = NULL;
3592 QDict *image_options;
3593 char *bdref_key_dot;
3594 const char *reference;
3596 assert(child_class != NULL);
3598 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3599 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3600 g_free(bdref_key_dot);
3603 * Caution: while qdict_get_try_str() is fine, getting non-string
3604 * types would require more care. When @options come from
3605 * -blockdev or blockdev_add, its members are typed according to
3606 * the QAPI schema, but when they come from -drive, they're all
3609 reference = qdict_get_try_str(options, bdref_key);
3610 if (!filename && !reference && !qdict_size(image_options)) {
3612 error_setg(errp, "A block device must be specified for \"%s\"",
3615 qobject_unref(image_options);
3619 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3620 parent, child_class, child_role, errp);
3626 qdict_del(options, bdref_key);
3631 * Opens a disk image whose options are given as BlockdevRef in another block
3634 * If allow_none is true, no image will be opened if filename is false and no
3635 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3637 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3638 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3639 * itself, all options starting with "${bdref_key}." are considered part of the
3642 * The BlockdevRef will be removed from the options QDict.
3644 BdrvChild *bdrv_open_child(const char *filename,
3645 QDict *options, const char *bdref_key,
3646 BlockDriverState *parent,
3647 const BdrvChildClass *child_class,
3648 BdrvChildRole child_role,
3649 bool allow_none, Error **errp)
3651 BlockDriverState *bs;
3653 GLOBAL_STATE_CODE();
3655 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3656 child_role, allow_none, errp);
3661 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3666 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3668 int bdrv_open_file_child(const char *filename,
3669 QDict *options, const char *bdref_key,
3670 BlockDriverState *parent, Error **errp)
3674 /* commit_top and mirror_top don't use this function */
3675 assert(!parent->drv->filtered_child_is_backing);
3676 role = parent->drv->is_filter ?
3677 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3679 if (!bdrv_open_child(filename, options, bdref_key, parent,
3680 &child_of_bds, role, false, errp))
3689 * TODO Future callers may need to specify parent/child_class in order for
3690 * option inheritance to work. Existing callers use it for the root node.
3692 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3694 BlockDriverState *bs = NULL;
3695 QObject *obj = NULL;
3696 QDict *qdict = NULL;
3697 const char *reference = NULL;
3700 GLOBAL_STATE_CODE();
3702 if (ref->type == QTYPE_QSTRING) {
3703 reference = ref->u.reference;
3705 BlockdevOptions *options = &ref->u.definition;
3706 assert(ref->type == QTYPE_QDICT);
3708 v = qobject_output_visitor_new(&obj);
3709 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3710 visit_complete(v, &obj);
3712 qdict = qobject_to(QDict, obj);
3713 qdict_flatten(qdict);
3715 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3716 * compatibility with other callers) rather than what we want as the
3717 * real defaults. Apply the defaults here instead. */
3718 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3719 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3720 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3721 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3725 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3732 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3734 QDict *snapshot_options,
3737 g_autofree char *tmp_filename = NULL;
3739 QemuOpts *opts = NULL;
3740 BlockDriverState *bs_snapshot = NULL;
3743 GLOBAL_STATE_CODE();
3745 /* if snapshot, we create a temporary backing file and open it
3746 instead of opening 'filename' directly */
3748 /* Get the required size from the image */
3749 total_size = bdrv_getlength(bs);
3750 if (total_size < 0) {
3751 error_setg_errno(errp, -total_size, "Could not get image size");
3755 /* Create the temporary image */
3756 tmp_filename = create_tmp_file(errp);
3757 if (!tmp_filename) {
3761 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3763 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3764 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3765 qemu_opts_del(opts);
3767 error_prepend(errp, "Could not create temporary overlay '%s': ",
3772 /* Prepare options QDict for the temporary file */
3773 qdict_put_str(snapshot_options, "file.driver", "file");
3774 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3775 qdict_put_str(snapshot_options, "driver", "qcow2");
3777 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3778 snapshot_options = NULL;
3783 ret = bdrv_append(bs_snapshot, bs, errp);
3790 qobject_unref(snapshot_options);
3795 * Opens a disk image (raw, qcow2, vmdk, ...)
3797 * options is a QDict of options to pass to the block drivers, or NULL for an
3798 * empty set of options. The reference to the QDict belongs to the block layer
3799 * after the call (even on failure), so if the caller intends to reuse the
3800 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3802 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3803 * If it is not NULL, the referenced BDS will be reused.
3805 * The reference parameter may be used to specify an existing block device which
3806 * should be opened. If specified, neither options nor a filename may be given,
3807 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3809 * The caller must always hold @filename AioContext lock, because this
3810 * function eventually calls bdrv_refresh_total_sectors() which polls
3811 * when called from non-coroutine context.
3813 static BlockDriverState * no_coroutine_fn
3814 bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3815 int flags, BlockDriverState *parent,
3816 const BdrvChildClass *child_class, BdrvChildRole child_role,
3820 BlockBackend *file = NULL;
3821 BlockDriverState *bs;
3822 BlockDriver *drv = NULL;
3824 const char *drvname;
3825 const char *backing;
3826 Error *local_err = NULL;
3827 QDict *snapshot_options = NULL;
3828 int snapshot_flags = 0;
3830 assert(!child_class || !flags);
3831 assert(!child_class == !parent);
3832 GLOBAL_STATE_CODE();
3833 assert(!qemu_in_coroutine());
3836 bool options_non_empty = options ? qdict_size(options) : false;
3837 qobject_unref(options);
3839 if (filename || options_non_empty) {
3840 error_setg(errp, "Cannot reference an existing block device with "
3841 "additional options or a new filename");
3845 bs = bdrv_lookup_bs(reference, reference, errp);
3856 /* NULL means an empty set of options */
3857 if (options == NULL) {
3858 options = qdict_new();
3861 /* json: syntax counts as explicit options, as if in the QDict */
3862 parse_json_protocol(options, &filename, &local_err);
3867 bs->explicit_options = qdict_clone_shallow(options);
3870 bool parent_is_format;
3873 parent_is_format = parent->drv->is_format;
3876 * parent->drv is not set yet because this node is opened for
3877 * (potential) format probing. That means that @parent is going
3878 * to be a format node.
3880 parent_is_format = true;
3883 bs->inherits_from = parent;
3884 child_class->inherit_options(child_role, parent_is_format,
3886 parent->open_flags, parent->options);
3889 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3895 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3896 * Caution: getting a boolean member of @options requires care.
3897 * When @options come from -blockdev or blockdev_add, members are
3898 * typed according to the QAPI schema, but when they come from
3899 * -drive, they're all QString.
3901 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3902 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3903 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3905 flags &= ~BDRV_O_RDWR;
3908 if (flags & BDRV_O_SNAPSHOT) {
3909 snapshot_options = qdict_new();
3910 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3912 /* Let bdrv_backing_options() override "read-only" */
3913 qdict_del(options, BDRV_OPT_READ_ONLY);
3914 bdrv_inherited_options(BDRV_CHILD_COW, true,
3915 &flags, options, flags, options);
3918 bs->open_flags = flags;
3919 bs->options = options;
3920 options = qdict_clone_shallow(options);
3922 /* Find the right image format driver */
3923 /* See cautionary note on accessing @options above */
3924 drvname = qdict_get_try_str(options, "driver");
3926 drv = bdrv_find_format(drvname);
3928 error_setg(errp, "Unknown driver: '%s'", drvname);
3933 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3935 /* See cautionary note on accessing @options above */
3936 backing = qdict_get_try_str(options, "backing");
3937 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3938 (backing && *backing == '\0'))
3941 warn_report("Use of \"backing\": \"\" is deprecated; "
3942 "use \"backing\": null instead");
3944 flags |= BDRV_O_NO_BACKING;
3945 qdict_del(bs->explicit_options, "backing");
3946 qdict_del(bs->options, "backing");
3947 qdict_del(options, "backing");
3950 /* Open image file without format layer. This BlockBackend is only used for
3951 * probing, the block drivers will do their own bdrv_open_child() for the
3952 * same BDS, which is why we put the node name back into options. */
3953 if ((flags & BDRV_O_PROTOCOL) == 0) {
3954 BlockDriverState *file_bs;
3956 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3957 &child_of_bds, BDRV_CHILD_IMAGE,
3962 if (file_bs != NULL) {
3963 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3964 * looking at the header to guess the image format. This works even
3965 * in cases where a guest would not see a consistent state. */
3966 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3967 blk_insert_bs(file, file_bs, &local_err);
3968 bdrv_unref(file_bs);
3973 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3977 /* Image format probing */
3980 ret = find_image_format(file, filename, &drv, &local_err);
3985 * This option update would logically belong in bdrv_fill_options(),
3986 * but we first need to open bs->file for the probing to work, while
3987 * opening bs->file already requires the (mostly) final set of options
3988 * so that cache mode etc. can be inherited.
3990 * Adding the driver later is somewhat ugly, but it's not an option
3991 * that would ever be inherited, so it's correct. We just need to make
3992 * sure to update both bs->options (which has the full effective
3993 * options for bs) and options (which has file.* already removed).
3995 qdict_put_str(bs->options, "driver", drv->format_name);
3996 qdict_put_str(options, "driver", drv->format_name);
3998 error_setg(errp, "Must specify either driver or file");
4002 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4003 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4004 /* file must be NULL if a protocol BDS is about to be created
4005 * (the inverse results in an error message from bdrv_open_common()) */
4006 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4008 /* Open the image */
4009 ret = bdrv_open_common(bs, file, options, &local_err);
4019 /* If there is a backing file, use it */
4020 if ((flags & BDRV_O_NO_BACKING) == 0) {
4021 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4023 goto close_and_fail;
4027 /* Remove all children options and references
4028 * from bs->options and bs->explicit_options */
4029 QLIST_FOREACH(child, &bs->children, next) {
4030 char *child_key_dot;
4031 child_key_dot = g_strdup_printf("%s.", child->name);
4032 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4033 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4034 qdict_del(bs->explicit_options, child->name);
4035 qdict_del(bs->options, child->name);
4036 g_free(child_key_dot);
4039 /* Check if any unknown options were used */
4040 if (qdict_size(options) != 0) {
4041 const QDictEntry *entry = qdict_first(options);
4042 if (flags & BDRV_O_PROTOCOL) {
4043 error_setg(errp, "Block protocol '%s' doesn't support the option "
4044 "'%s'", drv->format_name, entry->key);
4047 "Block format '%s' does not support the option '%s'",
4048 drv->format_name, entry->key);
4051 goto close_and_fail;
4054 bdrv_parent_cb_change_media(bs, true);
4056 qobject_unref(options);
4059 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4060 * temporary snapshot afterwards. */
4061 if (snapshot_flags) {
4062 BlockDriverState *snapshot_bs;
4063 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4064 snapshot_options, &local_err);
4065 snapshot_options = NULL;
4067 goto close_and_fail;
4069 /* We are not going to return bs but the overlay on top of it
4070 * (snapshot_bs); thus, we have to drop the strong reference to bs
4071 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4072 * though, because the overlay still has a reference to it. */
4081 qobject_unref(snapshot_options);
4082 qobject_unref(bs->explicit_options);
4083 qobject_unref(bs->options);
4084 qobject_unref(options);
4086 bs->explicit_options = NULL;
4088 error_propagate(errp, local_err);
4093 qobject_unref(snapshot_options);
4094 qobject_unref(options);
4095 error_propagate(errp, local_err);
4100 * The caller must always hold @filename AioContext lock, because this
4101 * function eventually calls bdrv_refresh_total_sectors() which polls
4102 * when called from non-coroutine context.
4104 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4105 QDict *options, int flags, Error **errp)
4107 GLOBAL_STATE_CODE();
4109 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4113 /* Return true if the NULL-terminated @list contains @str */
4114 static bool is_str_in_list(const char *str, const char *const *list)
4118 for (i = 0; list[i] != NULL; i++) {
4119 if (!strcmp(str, list[i])) {
4128 * Check that every option set in @bs->options is also set in
4131 * Options listed in the common_options list and in
4132 * @bs->drv->mutable_opts are skipped.
4134 * Return 0 on success, otherwise return -EINVAL and set @errp.
4136 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4137 const QDict *new_opts, Error **errp)
4139 const QDictEntry *e;
4140 /* These options are common to all block drivers and are handled
4141 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4142 const char *const common_options[] = {
4143 "node-name", "discard", "cache.direct", "cache.no-flush",
4144 "read-only", "auto-read-only", "detect-zeroes", NULL
4147 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4148 if (!qdict_haskey(new_opts, e->key) &&
4149 !is_str_in_list(e->key, common_options) &&
4150 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4151 error_setg(errp, "Option '%s' cannot be reset "
4152 "to its default value", e->key);
4161 * Returns true if @child can be reached recursively from @bs
4163 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4164 BlockDriverState *child)
4172 QLIST_FOREACH(c, &bs->children, next) {
4173 if (bdrv_recurse_has_child(c->bs, child)) {
4182 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4183 * reopen of multiple devices.
4185 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4186 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4187 * be created and initialized. This newly created BlockReopenQueue should be
4188 * passed back in for subsequent calls that are intended to be of the same
4191 * bs is the BlockDriverState to add to the reopen queue.
4193 * options contains the changed options for the associated bs
4194 * (the BlockReopenQueue takes ownership)
4196 * flags contains the open flags for the associated bs
4198 * returns a pointer to bs_queue, which is either the newly allocated
4199 * bs_queue, or the existing bs_queue being used.
4201 * bs is drained here and undrained by bdrv_reopen_queue_free().
4203 * To be called with bs->aio_context locked.
4205 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4206 BlockDriverState *bs,
4208 const BdrvChildClass *klass,
4210 bool parent_is_format,
4211 QDict *parent_options,
4217 BlockReopenQueueEntry *bs_entry;
4219 QDict *old_options, *explicit_options, *options_copy;
4223 GLOBAL_STATE_CODE();
4225 bdrv_drained_begin(bs);
4227 if (bs_queue == NULL) {
4228 bs_queue = g_new0(BlockReopenQueue, 1);
4229 QTAILQ_INIT(bs_queue);
4233 options = qdict_new();
4236 /* Check if this BlockDriverState is already in the queue */
4237 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4238 if (bs == bs_entry->state.bs) {
4244 * Precedence of options:
4245 * 1. Explicitly passed in options (highest)
4246 * 2. Retained from explicitly set options of bs
4247 * 3. Inherited from parent node
4248 * 4. Retained from effective options of bs
4251 /* Old explicitly set values (don't overwrite by inherited value) */
4252 if (bs_entry || keep_old_opts) {
4253 old_options = qdict_clone_shallow(bs_entry ?
4254 bs_entry->state.explicit_options :
4255 bs->explicit_options);
4256 bdrv_join_options(bs, options, old_options);
4257 qobject_unref(old_options);
4260 explicit_options = qdict_clone_shallow(options);
4262 /* Inherit from parent node */
4263 if (parent_options) {
4265 klass->inherit_options(role, parent_is_format, &flags, options,
4266 parent_flags, parent_options);
4268 flags = bdrv_get_flags(bs);
4271 if (keep_old_opts) {
4272 /* Old values are used for options that aren't set yet */
4273 old_options = qdict_clone_shallow(bs->options);
4274 bdrv_join_options(bs, options, old_options);
4275 qobject_unref(old_options);
4278 /* We have the final set of options so let's update the flags */
4279 options_copy = qdict_clone_shallow(options);
4280 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4281 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4282 update_flags_from_options(&flags, opts);
4283 qemu_opts_del(opts);
4284 qobject_unref(options_copy);
4286 /* bdrv_open_inherit() sets and clears some additional flags internally */
4287 flags &= ~BDRV_O_PROTOCOL;
4288 if (flags & BDRV_O_RDWR) {
4289 flags |= BDRV_O_ALLOW_RDWR;
4293 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4294 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4296 qobject_unref(bs_entry->state.options);
4297 qobject_unref(bs_entry->state.explicit_options);
4300 bs_entry->state.bs = bs;
4301 bs_entry->state.options = options;
4302 bs_entry->state.explicit_options = explicit_options;
4303 bs_entry->state.flags = flags;
4306 * If keep_old_opts is false then it means that unspecified
4307 * options must be reset to their original value. We don't allow
4308 * resetting 'backing' but we need to know if the option is
4309 * missing in order to decide if we have to return an error.
4311 if (!keep_old_opts) {
4312 bs_entry->state.backing_missing =
4313 !qdict_haskey(options, "backing") &&
4314 !qdict_haskey(options, "backing.driver");
4317 QLIST_FOREACH(child, &bs->children, next) {
4318 QDict *new_child_options = NULL;
4319 bool child_keep_old = keep_old_opts;
4321 /* reopen can only change the options of block devices that were
4322 * implicitly created and inherited options. For other (referenced)
4323 * block devices, a syntax like "backing.foo" results in an error. */
4324 if (child->bs->inherits_from != bs) {
4328 /* Check if the options contain a child reference */
4329 if (qdict_haskey(options, child->name)) {
4330 const char *childref = qdict_get_try_str(options, child->name);
4332 * The current child must not be reopened if the child
4333 * reference is null or points to a different node.
4335 if (g_strcmp0(childref, child->bs->node_name)) {
4339 * If the child reference points to the current child then
4340 * reopen it with its existing set of options (note that
4341 * it can still inherit new options from the parent).
4343 child_keep_old = true;
4345 /* Extract child options ("child-name.*") */
4346 char *child_key_dot = g_strdup_printf("%s.", child->name);
4347 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4348 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4349 g_free(child_key_dot);
4352 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4353 child->klass, child->role, bs->drv->is_format,
4354 options, flags, child_keep_old);
4360 /* To be called with bs->aio_context locked */
4361 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4362 BlockDriverState *bs,
4363 QDict *options, bool keep_old_opts)
4365 GLOBAL_STATE_CODE();
4367 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4368 NULL, 0, keep_old_opts);
4371 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4373 GLOBAL_STATE_CODE();
4375 BlockReopenQueueEntry *bs_entry, *next;
4376 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4377 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4379 aio_context_acquire(ctx);
4380 bdrv_drained_end(bs_entry->state.bs);
4381 aio_context_release(ctx);
4383 qobject_unref(bs_entry->state.explicit_options);
4384 qobject_unref(bs_entry->state.options);
4392 * Reopen multiple BlockDriverStates atomically & transactionally.
4394 * The queue passed in (bs_queue) must have been built up previous
4395 * via bdrv_reopen_queue().
4397 * Reopens all BDS specified in the queue, with the appropriate
4398 * flags. All devices are prepared for reopen, and failure of any
4399 * device will cause all device changes to be abandoned, and intermediate
4402 * If all devices prepare successfully, then the changes are committed
4405 * All affected nodes must be drained between bdrv_reopen_queue() and
4406 * bdrv_reopen_multiple().
4408 * To be called from the main thread, with all other AioContexts unlocked.
4410 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4413 BlockReopenQueueEntry *bs_entry, *next;
4415 Transaction *tran = tran_new();
4416 g_autoptr(GSList) refresh_list = NULL;
4418 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4419 assert(bs_queue != NULL);
4420 GLOBAL_STATE_CODE();
4422 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4423 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4424 aio_context_acquire(ctx);
4425 ret = bdrv_flush(bs_entry->state.bs);
4426 aio_context_release(ctx);
4428 error_setg_errno(errp, -ret, "Error flushing drive");
4433 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4434 assert(bs_entry->state.bs->quiesce_counter > 0);
4435 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4436 aio_context_acquire(ctx);
4437 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4438 aio_context_release(ctx);
4442 bs_entry->prepared = true;
4445 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4446 BDRVReopenState *state = &bs_entry->state;
4448 refresh_list = g_slist_prepend(refresh_list, state->bs);
4449 if (state->old_backing_bs) {
4450 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4452 if (state->old_file_bs) {
4453 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4458 * Note that file-posix driver rely on permission update done during reopen
4459 * (even if no permission changed), because it wants "new" permissions for
4460 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4461 * in raw_reopen_prepare() which is called with "old" permissions.
4463 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4469 * If we reach this point, we have success and just need to apply the
4472 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4473 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4474 * children are usually goes after parents in reopen-queue, so go from last
4477 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4478 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4479 aio_context_acquire(ctx);
4480 bdrv_reopen_commit(&bs_entry->state);
4481 aio_context_release(ctx);
4486 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4487 BlockDriverState *bs = bs_entry->state.bs;
4489 if (bs->drv->bdrv_reopen_commit_post) {
4490 ctx = bdrv_get_aio_context(bs);
4491 aio_context_acquire(ctx);
4492 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4493 aio_context_release(ctx);
4502 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4503 if (bs_entry->prepared) {
4504 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4505 aio_context_acquire(ctx);
4506 bdrv_reopen_abort(&bs_entry->state);
4507 aio_context_release(ctx);
4512 bdrv_reopen_queue_free(bs_queue);
4517 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4520 AioContext *ctx = bdrv_get_aio_context(bs);
4521 BlockReopenQueue *queue;
4524 GLOBAL_STATE_CODE();
4526 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4528 if (ctx != qemu_get_aio_context()) {
4529 aio_context_release(ctx);
4531 ret = bdrv_reopen_multiple(queue, errp);
4533 if (ctx != qemu_get_aio_context()) {
4534 aio_context_acquire(ctx);
4540 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4543 QDict *opts = qdict_new();
4545 GLOBAL_STATE_CODE();
4547 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4549 return bdrv_reopen(bs, opts, true, errp);
4553 * Take a BDRVReopenState and check if the value of 'backing' in the
4554 * reopen_state->options QDict is valid or not.
4556 * If 'backing' is missing from the QDict then return 0.
4558 * If 'backing' contains the node name of the backing file of
4559 * reopen_state->bs then return 0.
4561 * If 'backing' contains a different node name (or is null) then check
4562 * whether the current backing file can be replaced with the new one.
4563 * If that's the case then reopen_state->replace_backing_bs is set to
4564 * true and reopen_state->new_backing_bs contains a pointer to the new
4565 * backing BlockDriverState (or NULL).
4567 * Return 0 on success, otherwise return < 0 and set @errp.
4569 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4570 bool is_backing, Transaction *tran,
4573 BlockDriverState *bs = reopen_state->bs;
4574 BlockDriverState *new_child_bs;
4575 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4577 const char *child_name = is_backing ? "backing" : "file";
4581 GLOBAL_STATE_CODE();
4583 value = qdict_get(reopen_state->options, child_name);
4584 if (value == NULL) {
4588 switch (qobject_type(value)) {
4590 assert(is_backing); /* The 'file' option does not allow a null value */
4591 new_child_bs = NULL;
4594 str = qstring_get_str(qobject_to(QString, value));
4595 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4596 if (new_child_bs == NULL) {
4598 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4599 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4600 "cycle", str, child_name, bs->node_name);
4606 * The options QDict has been flattened, so 'backing' and 'file'
4607 * do not allow any other data type here.
4609 g_assert_not_reached();
4612 if (old_child_bs == new_child_bs) {
4617 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4621 if (old_child_bs->implicit) {
4622 error_setg(errp, "Cannot replace implicit %s child of %s",
4623 child_name, bs->node_name);
4628 if (bs->drv->is_filter && !old_child_bs) {
4630 * Filters always have a file or a backing child, so we are trying to
4631 * change wrong child
4633 error_setg(errp, "'%s' is a %s filter node that does not support a "
4634 "%s child", bs->node_name, bs->drv->format_name, child_name);
4639 reopen_state->old_backing_bs = old_child_bs;
4641 reopen_state->old_file_bs = old_child_bs;
4644 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4649 * Prepares a BlockDriverState for reopen. All changes are staged in the
4650 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4651 * the block driver layer .bdrv_reopen_prepare()
4653 * bs is the BlockDriverState to reopen
4654 * flags are the new open flags
4655 * queue is the reopen queue
4657 * Returns 0 on success, non-zero on error. On error errp will be set
4660 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4661 * It is the responsibility of the caller to then call the abort() or
4662 * commit() for any other BDS that have been left in a prepare() state
4665 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4666 BlockReopenQueue *queue,
4667 Transaction *change_child_tran, Error **errp)
4671 Error *local_err = NULL;
4674 QDict *orig_reopen_opts;
4675 char *discard = NULL;
4677 bool drv_prepared = false;
4679 assert(reopen_state != NULL);
4680 assert(reopen_state->bs->drv != NULL);
4681 GLOBAL_STATE_CODE();
4682 drv = reopen_state->bs->drv;
4684 /* This function and each driver's bdrv_reopen_prepare() remove
4685 * entries from reopen_state->options as they are processed, so
4686 * we need to make a copy of the original QDict. */
4687 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4689 /* Process generic block layer options */
4690 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4691 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4696 /* This was already called in bdrv_reopen_queue_child() so the flags
4697 * are up-to-date. This time we simply want to remove the options from
4698 * QemuOpts in order to indicate that they have been processed. */
4699 old_flags = reopen_state->flags;
4700 update_flags_from_options(&reopen_state->flags, opts);
4701 assert(old_flags == reopen_state->flags);
4703 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4704 if (discard != NULL) {
4705 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4706 error_setg(errp, "Invalid discard option");
4712 reopen_state->detect_zeroes =
4713 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4715 error_propagate(errp, local_err);
4720 /* All other options (including node-name and driver) must be unchanged.
4721 * Put them back into the QDict, so that they are checked at the end
4722 * of this function. */
4723 qemu_opts_to_qdict(opts, reopen_state->options);
4725 /* If we are to stay read-only, do not allow permission change
4726 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4727 * not set, or if the BDS still has copy_on_read enabled */
4728 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4729 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4731 error_propagate(errp, local_err);
4735 if (drv->bdrv_reopen_prepare) {
4737 * If a driver-specific option is missing, it means that we
4738 * should reset it to its default value.
4739 * But not all options allow that, so we need to check it first.
4741 ret = bdrv_reset_options_allowed(reopen_state->bs,
4742 reopen_state->options, errp);
4747 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4749 if (local_err != NULL) {
4750 error_propagate(errp, local_err);
4752 bdrv_refresh_filename(reopen_state->bs);
4753 error_setg(errp, "failed while preparing to reopen image '%s'",
4754 reopen_state->bs->filename);
4759 /* It is currently mandatory to have a bdrv_reopen_prepare()
4760 * handler for each supported drv. */
4761 error_setg(errp, "Block format '%s' used by node '%s' "
4762 "does not support reopening files", drv->format_name,
4763 bdrv_get_device_or_node_name(reopen_state->bs));
4768 drv_prepared = true;
4771 * We must provide the 'backing' option if the BDS has a backing
4772 * file or if the image file has a backing file name as part of
4773 * its metadata. Otherwise the 'backing' option can be omitted.
4775 if (drv->supports_backing && reopen_state->backing_missing &&
4776 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4777 error_setg(errp, "backing is missing for '%s'",
4778 reopen_state->bs->node_name);
4784 * Allow changing the 'backing' option. The new value can be
4785 * either a reference to an existing node (using its node name)
4786 * or NULL to simply detach the current backing file.
4788 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4789 change_child_tran, errp);
4793 qdict_del(reopen_state->options, "backing");
4795 /* Allow changing the 'file' option. In this case NULL is not allowed */
4796 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4797 change_child_tran, errp);
4801 qdict_del(reopen_state->options, "file");
4803 /* Options that are not handled are only okay if they are unchanged
4804 * compared to the old state. It is expected that some options are only
4805 * used for the initial open, but not reopen (e.g. filename) */
4806 if (qdict_size(reopen_state->options)) {
4807 const QDictEntry *entry = qdict_first(reopen_state->options);
4810 QObject *new = entry->value;
4811 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4813 /* Allow child references (child_name=node_name) as long as they
4814 * point to the current child (i.e. everything stays the same). */
4815 if (qobject_type(new) == QTYPE_QSTRING) {
4817 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4818 if (!strcmp(child->name, entry->key)) {
4824 if (!strcmp(child->bs->node_name,
4825 qstring_get_str(qobject_to(QString, new)))) {
4826 continue; /* Found child with this name, skip option */
4832 * TODO: When using -drive to specify blockdev options, all values
4833 * will be strings; however, when using -blockdev, blockdev-add or
4834 * filenames using the json:{} pseudo-protocol, they will be
4836 * In contrast, reopening options are (currently) always strings
4837 * (because you can only specify them through qemu-io; all other
4838 * callers do not specify any options).
4839 * Therefore, when using anything other than -drive to create a BDS,
4840 * this cannot detect non-string options as unchanged, because
4841 * qobject_is_equal() always returns false for objects of different
4842 * type. In the future, this should be remedied by correctly typing
4843 * all options. For now, this is not too big of an issue because
4844 * the user can simply omit options which cannot be changed anyway,
4845 * so they will stay unchanged.
4847 if (!qobject_is_equal(new, old)) {
4848 error_setg(errp, "Cannot change the option '%s'", entry->key);
4852 } while ((entry = qdict_next(reopen_state->options, entry)));
4857 /* Restore the original reopen_state->options QDict */
4858 qobject_unref(reopen_state->options);
4859 reopen_state->options = qobject_ref(orig_reopen_opts);
4862 if (ret < 0 && drv_prepared) {
4863 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4864 * call drv->bdrv_reopen_abort() before signaling an error
4865 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4866 * when the respective bdrv_reopen_prepare() has failed) */
4867 if (drv->bdrv_reopen_abort) {
4868 drv->bdrv_reopen_abort(reopen_state);
4871 qemu_opts_del(opts);
4872 qobject_unref(orig_reopen_opts);
4878 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4879 * makes them final by swapping the staging BlockDriverState contents into
4880 * the active BlockDriverState contents.
4882 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4885 BlockDriverState *bs;
4888 assert(reopen_state != NULL);
4889 bs = reopen_state->bs;
4891 assert(drv != NULL);
4892 GLOBAL_STATE_CODE();
4894 /* If there are any driver level actions to take */
4895 if (drv->bdrv_reopen_commit) {
4896 drv->bdrv_reopen_commit(reopen_state);
4899 /* set BDS specific flags now */
4900 qobject_unref(bs->explicit_options);
4901 qobject_unref(bs->options);
4902 qobject_ref(reopen_state->explicit_options);
4903 qobject_ref(reopen_state->options);
4905 bs->explicit_options = reopen_state->explicit_options;
4906 bs->options = reopen_state->options;
4907 bs->open_flags = reopen_state->flags;
4908 bs->detect_zeroes = reopen_state->detect_zeroes;
4910 /* Remove child references from bs->options and bs->explicit_options.
4911 * Child options were already removed in bdrv_reopen_queue_child() */
4912 QLIST_FOREACH(child, &bs->children, next) {
4913 qdict_del(bs->explicit_options, child->name);
4914 qdict_del(bs->options, child->name);
4916 /* backing is probably removed, so it's not handled by previous loop */
4917 qdict_del(bs->explicit_options, "backing");
4918 qdict_del(bs->options, "backing");
4920 bdrv_refresh_limits(bs, NULL, NULL);
4921 bdrv_refresh_total_sectors(bs, bs->total_sectors);
4925 * Abort the reopen, and delete and free the staged changes in
4928 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4932 assert(reopen_state != NULL);
4933 drv = reopen_state->bs->drv;
4934 assert(drv != NULL);
4935 GLOBAL_STATE_CODE();
4937 if (drv->bdrv_reopen_abort) {
4938 drv->bdrv_reopen_abort(reopen_state);
4943 static void bdrv_close(BlockDriverState *bs)
4945 BdrvAioNotifier *ban, *ban_next;
4946 BdrvChild *child, *next;
4948 GLOBAL_STATE_CODE();
4949 assert(!bs->refcnt);
4951 bdrv_drained_begin(bs); /* complete I/O */
4953 bdrv_drain(bs); /* in case flush left pending I/O */
4956 if (bs->drv->bdrv_close) {
4957 /* Must unfreeze all children, so bdrv_unref_child() works */
4958 bs->drv->bdrv_close(bs);
4963 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4964 bdrv_unref_child(bs, child);
4967 assert(!bs->backing);
4971 qatomic_set(&bs->copy_on_read, 0);
4972 bs->backing_file[0] = '\0';
4973 bs->backing_format[0] = '\0';
4974 bs->total_sectors = 0;
4975 bs->encrypted = false;
4977 qobject_unref(bs->options);
4978 qobject_unref(bs->explicit_options);
4980 bs->explicit_options = NULL;
4981 qobject_unref(bs->full_open_options);
4982 bs->full_open_options = NULL;
4983 g_free(bs->block_status_cache);
4984 bs->block_status_cache = NULL;
4986 bdrv_release_named_dirty_bitmaps(bs);
4987 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4989 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4992 QLIST_INIT(&bs->aio_notifiers);
4993 bdrv_drained_end(bs);
4996 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4997 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5000 if (bs->quiesce_counter) {
5001 bdrv_drain_all_end_quiesce(bs);
5005 void bdrv_close_all(void)
5007 GLOBAL_STATE_CODE();
5008 assert(job_next(NULL) == NULL);
5010 /* Drop references from requests still in flight, such as canceled block
5011 * jobs whose AIO context has not been polled yet */
5014 blk_remove_all_bs();
5015 blockdev_close_all_bdrv_states();
5017 assert(QTAILQ_EMPTY(&all_bdrv_states));
5020 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
5026 if (c->klass->stay_at_node) {
5030 /* If the child @c belongs to the BDS @to, replacing the current
5031 * c->bs by @to would mean to create a loop.
5033 * Such a case occurs when appending a BDS to a backing chain.
5034 * For instance, imagine the following chain:
5036 * guest device -> node A -> further backing chain...
5038 * Now we create a new BDS B which we want to put on top of this
5039 * chain, so we first attach A as its backing node:
5044 * guest device -> node A -> further backing chain...
5046 * Finally we want to replace A by B. When doing that, we want to
5047 * replace all pointers to A by pointers to B -- except for the
5048 * pointer from B because (1) that would create a loop, and (2)
5049 * that pointer should simply stay intact:
5051 * guest device -> node B
5054 * node A -> further backing chain...
5056 * In general, when replacing a node A (c->bs) by a node B (@to),
5057 * if A is a child of B, that means we cannot replace A by B there
5058 * because that would create a loop. Silently detaching A from B
5059 * is also not really an option. So overall just leaving A in
5060 * place there is the most sensible choice.
5062 * We would also create a loop in any cases where @c is only
5063 * indirectly referenced by @to. Prevent this by returning false
5064 * if @c is found (by breadth-first search) anywhere in the whole
5069 found = g_hash_table_new(NULL, NULL);
5070 g_hash_table_add(found, to);
5071 queue = g_queue_new();
5072 g_queue_push_tail(queue, to);
5074 while (!g_queue_is_empty(queue)) {
5075 BlockDriverState *v = g_queue_pop_head(queue);
5078 QLIST_FOREACH(c2, &v->children, next) {
5084 if (g_hash_table_contains(found, c2->bs)) {
5088 g_queue_push_tail(queue, c2->bs);
5089 g_hash_table_add(found, c2->bs);
5093 g_queue_free(queue);
5094 g_hash_table_destroy(found);
5099 static void bdrv_remove_child_commit(void *opaque)
5101 GLOBAL_STATE_CODE();
5102 bdrv_child_free(opaque);
5105 static TransactionActionDrv bdrv_remove_child_drv = {
5106 .commit = bdrv_remove_child_commit,
5109 /* Function doesn't update permissions, caller is responsible for this. */
5110 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5117 BlockDriverState *bs = child->bs;
5118 bdrv_drained_begin(bs);
5119 bdrv_replace_child_tran(child, NULL, tran);
5120 bdrv_drained_end(bs);
5123 tran_add(tran, &bdrv_remove_child_drv, child);
5126 static void undrain_on_clean_cb(void *opaque)
5128 bdrv_drained_end(opaque);
5131 static TransactionActionDrv undrain_on_clean = {
5132 .clean = undrain_on_clean_cb,
5135 static int bdrv_replace_node_noperm(BlockDriverState *from,
5136 BlockDriverState *to,
5137 bool auto_skip, Transaction *tran,
5140 BdrvChild *c, *next;
5142 GLOBAL_STATE_CODE();
5144 bdrv_drained_begin(from);
5145 bdrv_drained_begin(to);
5146 tran_add(tran, &undrain_on_clean, from);
5147 tran_add(tran, &undrain_on_clean, to);
5149 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5150 assert(c->bs == from);
5151 if (!should_update_child(c, to)) {
5155 error_setg(errp, "Should not change '%s' link to '%s'",
5156 c->name, from->node_name);
5160 error_setg(errp, "Cannot change '%s' link to '%s'",
5161 c->name, from->node_name);
5164 bdrv_replace_child_tran(c, to, tran);
5171 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5172 * if it creates a parent-child relation loop or if parent is block-job.
5174 * With auto_skip=false the error is returned if from has a parent which should
5177 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5178 * case backing link of the cow-parent of @to is removed.
5180 static int bdrv_replace_node_common(BlockDriverState *from,
5181 BlockDriverState *to,
5182 bool auto_skip, bool detach_subchain,
5185 Transaction *tran = tran_new();
5186 g_autoptr(GSList) refresh_list = NULL;
5187 BlockDriverState *to_cow_parent = NULL;
5190 GLOBAL_STATE_CODE();
5192 if (detach_subchain) {
5193 assert(bdrv_chain_contains(from, to));
5195 for (to_cow_parent = from;
5196 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5197 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5203 /* Make sure that @from doesn't go away until we have successfully attached
5204 * all of its parents to @to. */
5207 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5208 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5209 bdrv_drained_begin(from);
5212 * Do the replacement without permission update.
5213 * Replacement may influence the permissions, we should calculate new
5214 * permissions based on new graph. If we fail, we'll roll-back the
5217 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5222 if (detach_subchain) {
5223 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5226 refresh_list = g_slist_prepend(refresh_list, to);
5227 refresh_list = g_slist_prepend(refresh_list, from);
5229 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5237 tran_finalize(tran, ret);
5239 bdrv_drained_end(from);
5245 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5248 GLOBAL_STATE_CODE();
5250 return bdrv_replace_node_common(from, to, true, false, errp);
5253 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5255 GLOBAL_STATE_CODE();
5257 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5262 * Add new bs contents at the top of an image chain while the chain is
5263 * live, while keeping required fields on the top layer.
5265 * This will modify the BlockDriverState fields, and swap contents
5266 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5268 * bs_new must not be attached to a BlockBackend and must not have backing
5271 * This function does not create any image files.
5273 * The caller must hold the AioContext lock for @bs_top.
5275 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5280 Transaction *tran = tran_new();
5281 AioContext *old_context, *new_context = NULL;
5283 GLOBAL_STATE_CODE();
5285 assert(!bs_new->backing);
5287 old_context = bdrv_get_aio_context(bs_top);
5289 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5290 &child_of_bds, bdrv_backing_role(bs_new),
5298 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5299 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5300 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5301 * that assumes the new lock is taken.
5303 new_context = bdrv_get_aio_context(bs_top);
5305 if (old_context != new_context) {
5306 aio_context_release(old_context);
5307 aio_context_acquire(new_context);
5310 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5315 ret = bdrv_refresh_perms(bs_new, tran, errp);
5317 tran_finalize(tran, ret);
5319 bdrv_refresh_limits(bs_top, NULL, NULL);
5321 if (new_context && old_context != new_context) {
5322 aio_context_release(new_context);
5323 aio_context_acquire(old_context);
5329 /* Not for empty child */
5330 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5334 Transaction *tran = tran_new();
5335 g_autoptr(GSList) refresh_list = NULL;
5336 BlockDriverState *old_bs = child->bs;
5338 GLOBAL_STATE_CODE();
5341 bdrv_drained_begin(old_bs);
5342 bdrv_drained_begin(new_bs);
5344 bdrv_replace_child_tran(child, new_bs, tran);
5346 refresh_list = g_slist_prepend(refresh_list, old_bs);
5347 refresh_list = g_slist_prepend(refresh_list, new_bs);
5349 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5351 tran_finalize(tran, ret);
5353 bdrv_drained_end(old_bs);
5354 bdrv_drained_end(new_bs);
5360 static void bdrv_delete(BlockDriverState *bs)
5362 assert(bdrv_op_blocker_is_empty(bs));
5363 assert(!bs->refcnt);
5364 GLOBAL_STATE_CODE();
5366 /* remove from list, if necessary */
5367 if (bs->node_name[0] != '\0') {
5368 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5370 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5379 * Replace @bs by newly created block node.
5381 * @options is a QDict of options to pass to the block drivers, or NULL for an
5382 * empty set of options. The reference to the QDict belongs to the block layer
5383 * after the call (even on failure), so if the caller intends to reuse the
5384 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5386 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5387 int flags, Error **errp)
5391 BlockDriverState *new_node_bs = NULL;
5392 const char *drvname, *node_name;
5395 drvname = qdict_get_try_str(options, "driver");
5397 error_setg(errp, "driver is not specified");
5401 drv = bdrv_find_format(drvname);
5403 error_setg(errp, "Unknown driver: '%s'", drvname);
5407 node_name = qdict_get_try_str(options, "node-name");
5409 GLOBAL_STATE_CODE();
5411 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5413 options = NULL; /* bdrv_new_open_driver() eats options */
5415 error_prepend(errp, "Could not create node: ");
5419 bdrv_drained_begin(bs);
5420 ret = bdrv_replace_node(bs, new_node_bs, errp);
5421 bdrv_drained_end(bs);
5424 error_prepend(errp, "Could not replace node: ");
5431 qobject_unref(options);
5432 bdrv_unref(new_node_bs);
5437 * Run consistency checks on an image
5439 * Returns 0 if the check could be completed (it doesn't mean that the image is
5440 * free of errors) or -errno when an internal error occurred. The results of the
5441 * check are stored in res.
5443 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5444 BdrvCheckResult *res, BdrvCheckMode fix)
5447 assert_bdrv_graph_readable();
5448 if (bs->drv == NULL) {
5451 if (bs->drv->bdrv_co_check == NULL) {
5455 memset(res, 0, sizeof(*res));
5456 return bs->drv->bdrv_co_check(bs, res, fix);
5462 * -EINVAL - backing format specified, but no file
5463 * -ENOSPC - can't update the backing file because no space is left in the
5465 * -ENOTSUP - format driver doesn't support changing the backing file
5467 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5468 const char *backing_fmt, bool require)
5470 BlockDriver *drv = bs->drv;
5473 GLOBAL_STATE_CODE();
5479 /* Backing file format doesn't make sense without a backing file */
5480 if (backing_fmt && !backing_file) {
5484 if (require && backing_file && !backing_fmt) {
5488 if (drv->bdrv_change_backing_file != NULL) {
5489 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5495 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5496 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5497 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5498 backing_file ?: "");
5504 * Finds the first non-filter node above bs in the chain between
5505 * active and bs. The returned node is either an immediate parent of
5506 * bs, or there are only filter nodes between the two.
5508 * Returns NULL if bs is not found in active's image chain,
5509 * or if active == bs.
5511 * Returns the bottommost base image if bs == NULL.
5513 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5514 BlockDriverState *bs)
5517 GLOBAL_STATE_CODE();
5519 bs = bdrv_skip_filters(bs);
5520 active = bdrv_skip_filters(active);
5523 BlockDriverState *next = bdrv_backing_chain_next(active);
5533 /* Given a BDS, searches for the base layer. */
5534 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5536 GLOBAL_STATE_CODE();
5538 return bdrv_find_overlay(bs, NULL);
5542 * Return true if at least one of the COW (backing) and filter links
5543 * between @bs and @base is frozen. @errp is set if that's the case.
5544 * @base must be reachable from @bs, or NULL.
5546 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5549 BlockDriverState *i;
5552 GLOBAL_STATE_CODE();
5554 for (i = bs; i != base; i = child_bs(child)) {
5555 child = bdrv_filter_or_cow_child(i);
5557 if (child && child->frozen) {
5558 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5559 child->name, i->node_name, child->bs->node_name);
5568 * Freeze all COW (backing) and filter links between @bs and @base.
5569 * If any of the links is already frozen the operation is aborted and
5570 * none of the links are modified.
5571 * @base must be reachable from @bs, or NULL.
5572 * Returns 0 on success. On failure returns < 0 and sets @errp.
5574 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5577 BlockDriverState *i;
5580 GLOBAL_STATE_CODE();
5582 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5586 for (i = bs; i != base; i = child_bs(child)) {
5587 child = bdrv_filter_or_cow_child(i);
5588 if (child && child->bs->never_freeze) {
5589 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5590 child->name, child->bs->node_name);
5595 for (i = bs; i != base; i = child_bs(child)) {
5596 child = bdrv_filter_or_cow_child(i);
5598 child->frozen = true;
5606 * Unfreeze all COW (backing) and filter links between @bs and @base.
5607 * The caller must ensure that all links are frozen before using this
5609 * @base must be reachable from @bs, or NULL.
5611 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5613 BlockDriverState *i;
5616 GLOBAL_STATE_CODE();
5618 for (i = bs; i != base; i = child_bs(child)) {
5619 child = bdrv_filter_or_cow_child(i);
5621 assert(child->frozen);
5622 child->frozen = false;
5628 * Drops images above 'base' up to and including 'top', and sets the image
5629 * above 'top' to have base as its backing file.
5631 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5632 * information in 'bs' can be properly updated.
5634 * E.g., this will convert the following chain:
5635 * bottom <- base <- intermediate <- top <- active
5639 * bottom <- base <- active
5641 * It is allowed for bottom==base, in which case it converts:
5643 * base <- intermediate <- top <- active
5649 * If backing_file_str is non-NULL, it will be used when modifying top's
5650 * overlay image metadata.
5653 * if active == top, that is considered an error
5656 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5657 const char *backing_file_str)
5659 BlockDriverState *explicit_top = top;
5660 bool update_inherits_from;
5662 Error *local_err = NULL;
5664 g_autoptr(GSList) updated_children = NULL;
5667 GLOBAL_STATE_CODE();
5670 bdrv_drained_begin(base);
5672 if (!top->drv || !base->drv) {
5676 /* Make sure that base is in the backing chain of top */
5677 if (!bdrv_chain_contains(top, base)) {
5681 /* If 'base' recursively inherits from 'top' then we should set
5682 * base->inherits_from to top->inherits_from after 'top' and all
5683 * other intermediate nodes have been dropped.
5684 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5685 * it because no one inherits from it. We use explicit_top for that. */
5686 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5687 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5689 /* success - we can delete the intermediate states, and link top->base */
5690 if (!backing_file_str) {
5691 bdrv_refresh_filename(base);
5692 backing_file_str = base->filename;
5695 QLIST_FOREACH(c, &top->parents, next_parent) {
5696 updated_children = g_slist_prepend(updated_children, c);
5700 * It seems correct to pass detach_subchain=true here, but it triggers
5701 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5702 * another drained section, which modify the graph (for example, removing
5703 * the child, which we keep in updated_children list). So, it's a TODO.
5705 * Note, bug triggered if pass detach_subchain=true here and run
5706 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5709 bdrv_replace_node_common(top, base, false, false, &local_err);
5711 error_report_err(local_err);
5715 for (p = updated_children; p; p = p->next) {
5718 if (c->klass->update_filename) {
5719 ret = c->klass->update_filename(c, base, backing_file_str,
5723 * TODO: Actually, we want to rollback all previous iterations
5724 * of this loop, and (which is almost impossible) previous
5725 * bdrv_replace_node()...
5727 * Note, that c->klass->update_filename may lead to permission
5728 * update, so it's a bad idea to call it inside permission
5729 * update transaction of bdrv_replace_node.
5731 error_report_err(local_err);
5737 if (update_inherits_from) {
5738 base->inherits_from = explicit_top->inherits_from;
5743 bdrv_drained_end(base);
5749 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5750 * sums the size of all data-bearing children. (This excludes backing
5753 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5756 int64_t child_size, sum = 0;
5758 QLIST_FOREACH(child, &bs->children, next) {
5759 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5760 BDRV_CHILD_FILTERED))
5762 child_size = bdrv_co_get_allocated_file_size(child->bs);
5763 if (child_size < 0) {
5774 * Length of a allocated file in bytes. Sparse files are counted by actual
5775 * allocated space. Return < 0 if error or unknown.
5777 int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
5779 BlockDriver *drv = bs->drv;
5785 if (drv->bdrv_co_get_allocated_file_size) {
5786 return drv->bdrv_co_get_allocated_file_size(bs);
5789 if (drv->bdrv_file_open) {
5791 * Protocol drivers default to -ENOTSUP (most of their data is
5792 * not stored in any of their children (if they even have any),
5793 * so there is no generic way to figure it out).
5796 } else if (drv->is_filter) {
5797 /* Filter drivers default to the size of their filtered child */
5798 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
5800 /* Other drivers default to summing their children's sizes */
5801 return bdrv_sum_allocated_file_size(bs);
5807 * @drv: Format driver
5808 * @opts: Creation options for new image
5809 * @in_bs: Existing image containing data for new image (may be NULL)
5810 * @errp: Error object
5811 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5814 * Calculate file size required to create a new image.
5816 * If @in_bs is given then space for allocated clusters and zero clusters
5817 * from that image are included in the calculation. If @opts contains a
5818 * backing file that is shared by @in_bs then backing clusters may be omitted
5819 * from the calculation.
5821 * If @in_bs is NULL then the calculation includes no allocated clusters
5822 * unless a preallocation option is given in @opts.
5824 * Note that @in_bs may use a different BlockDriver from @drv.
5826 * If an error occurs the @errp pointer is set.
5828 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5829 BlockDriverState *in_bs, Error **errp)
5832 if (!drv->bdrv_measure) {
5833 error_setg(errp, "Block driver '%s' does not support size measurement",
5838 return drv->bdrv_measure(opts, in_bs, errp);
5842 * Return number of sectors on success, -errno on error.
5844 int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
5846 BlockDriver *drv = bs->drv;
5848 assert_bdrv_graph_readable();
5853 if (bs->bl.has_variable_length) {
5854 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
5859 return bs->total_sectors;
5863 * Return length in bytes on success, -errno on error.
5864 * The length is always a multiple of BDRV_SECTOR_SIZE.
5866 int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
5870 assert_bdrv_graph_readable();
5872 ret = bdrv_co_nb_sectors(bs);
5876 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5879 return ret * BDRV_SECTOR_SIZE;
5882 bool bdrv_is_sg(BlockDriverState *bs)
5889 * Return whether the given node supports compressed writes.
5891 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5893 BlockDriverState *filtered;
5896 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5900 filtered = bdrv_filter_bs(bs);
5903 * Filters can only forward compressed writes, so we have to
5906 return bdrv_supports_compressed_writes(filtered);
5912 const char *bdrv_get_format_name(BlockDriverState *bs)
5915 return bs->drv ? bs->drv->format_name : NULL;
5918 static int qsort_strcmp(const void *a, const void *b)
5920 return strcmp(*(char *const *)a, *(char *const *)b);
5923 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5924 void *opaque, bool read_only)
5929 const char **formats = NULL;
5931 GLOBAL_STATE_CODE();
5933 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5934 if (drv->format_name) {
5938 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5942 while (formats && i && !found) {
5943 found = !strcmp(formats[--i], drv->format_name);
5947 formats = g_renew(const char *, formats, count + 1);
5948 formats[count++] = drv->format_name;
5953 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5954 const char *format_name = block_driver_modules[i].format_name;
5960 if (use_bdrv_whitelist &&
5961 !bdrv_format_is_whitelisted(format_name, read_only)) {
5965 while (formats && j && !found) {
5966 found = !strcmp(formats[--j], format_name);
5970 formats = g_renew(const char *, formats, count + 1);
5971 formats[count++] = format_name;
5976 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5978 for (i = 0; i < count; i++) {
5979 it(opaque, formats[i]);
5985 /* This function is to find a node in the bs graph */
5986 BlockDriverState *bdrv_find_node(const char *node_name)
5988 BlockDriverState *bs;
5991 GLOBAL_STATE_CODE();
5993 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5994 if (!strcmp(node_name, bs->node_name)) {
6001 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6002 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6005 BlockDeviceInfoList *list;
6006 BlockDriverState *bs;
6008 GLOBAL_STATE_CODE();
6011 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6012 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6014 qapi_free_BlockDeviceInfoList(list);
6017 QAPI_LIST_PREPEND(list, info);
6023 typedef struct XDbgBlockGraphConstructor {
6024 XDbgBlockGraph *graph;
6025 GHashTable *graph_nodes;
6026 } XDbgBlockGraphConstructor;
6028 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6030 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6032 gr->graph = g_new0(XDbgBlockGraph, 1);
6033 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6038 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6040 XDbgBlockGraph *graph = gr->graph;
6042 g_hash_table_destroy(gr->graph_nodes);
6048 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6050 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6057 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6058 * answer of g_hash_table_lookup.
6060 ret = g_hash_table_size(gr->graph_nodes) + 1;
6061 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6066 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6067 XDbgBlockGraphNodeType type, const char *name)
6069 XDbgBlockGraphNode *n;
6071 n = g_new0(XDbgBlockGraphNode, 1);
6073 n->id = xdbg_graph_node_num(gr, node);
6075 n->name = g_strdup(name);
6077 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6080 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6081 const BdrvChild *child)
6083 BlockPermission qapi_perm;
6084 XDbgBlockGraphEdge *edge;
6085 GLOBAL_STATE_CODE();
6087 edge = g_new0(XDbgBlockGraphEdge, 1);
6089 edge->parent = xdbg_graph_node_num(gr, parent);
6090 edge->child = xdbg_graph_node_num(gr, child->bs);
6091 edge->name = g_strdup(child->name);
6093 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6094 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6096 if (flag & child->perm) {
6097 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6099 if (flag & child->shared_perm) {
6100 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6104 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6108 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6112 BlockDriverState *bs;
6114 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6116 GLOBAL_STATE_CODE();
6118 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6119 char *allocated_name = NULL;
6120 const char *name = blk_name(blk);
6123 name = allocated_name = blk_get_attached_dev_id(blk);
6125 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6127 g_free(allocated_name);
6128 if (blk_root(blk)) {
6129 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6133 WITH_JOB_LOCK_GUARD() {
6134 for (job = block_job_next_locked(NULL); job;
6135 job = block_job_next_locked(job)) {
6138 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6140 for (el = job->nodes; el; el = el->next) {
6141 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6146 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6147 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6149 QLIST_FOREACH(child, &bs->children, next) {
6150 xdbg_graph_add_edge(gr, bs, child);
6154 return xdbg_graph_finalize(gr);
6157 BlockDriverState *bdrv_lookup_bs(const char *device,
6158 const char *node_name,
6162 BlockDriverState *bs;
6164 GLOBAL_STATE_CODE();
6167 blk = blk_by_name(device);
6172 error_setg(errp, "Device '%s' has no medium", device);
6180 bs = bdrv_find_node(node_name);
6187 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6188 device ? device : "",
6189 node_name ? node_name : "");
6193 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6194 * return false. If either argument is NULL, return false. */
6195 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6198 GLOBAL_STATE_CODE();
6200 while (top && top != base) {
6201 top = bdrv_filter_or_cow_bs(top);
6207 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6209 GLOBAL_STATE_CODE();
6211 return QTAILQ_FIRST(&graph_bdrv_states);
6213 return QTAILQ_NEXT(bs, node_list);
6216 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6218 GLOBAL_STATE_CODE();
6220 return QTAILQ_FIRST(&all_bdrv_states);
6222 return QTAILQ_NEXT(bs, bs_list);
6225 const char *bdrv_get_node_name(const BlockDriverState *bs)
6228 return bs->node_name;
6231 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6237 /* If multiple parents have a name, just pick the first one. */
6238 QLIST_FOREACH(c, &bs->parents, next_parent) {
6239 if (c->klass->get_name) {
6240 name = c->klass->get_name(c);
6241 if (name && *name) {
6250 /* TODO check what callers really want: bs->node_name or blk_name() */
6251 const char *bdrv_get_device_name(const BlockDriverState *bs)
6254 return bdrv_get_parent_name(bs) ?: "";
6257 /* This can be used to identify nodes that might not have a device
6258 * name associated. Since node and device names live in the same
6259 * namespace, the result is unambiguous. The exception is if both are
6260 * absent, then this returns an empty (non-null) string. */
6261 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6264 return bdrv_get_parent_name(bs) ?: bs->node_name;
6267 int bdrv_get_flags(BlockDriverState *bs)
6270 return bs->open_flags;
6273 int bdrv_has_zero_init_1(BlockDriverState *bs)
6275 GLOBAL_STATE_CODE();
6279 int bdrv_has_zero_init(BlockDriverState *bs)
6281 BlockDriverState *filtered;
6282 GLOBAL_STATE_CODE();
6288 /* If BS is a copy on write image, it is initialized to
6289 the contents of the base image, which may not be zeroes. */
6290 if (bdrv_cow_child(bs)) {
6293 if (bs->drv->bdrv_has_zero_init) {
6294 return bs->drv->bdrv_has_zero_init(bs);
6297 filtered = bdrv_filter_bs(bs);
6299 return bdrv_has_zero_init(filtered);
6306 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6309 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6313 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6316 void bdrv_get_backing_filename(BlockDriverState *bs,
6317 char *filename, int filename_size)
6320 pstrcpy(filename, filename_size, bs->backing_file);
6323 int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6326 BlockDriver *drv = bs->drv;
6328 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6332 if (!drv->bdrv_co_get_info) {
6333 BlockDriverState *filtered = bdrv_filter_bs(bs);
6335 return bdrv_co_get_info(filtered, bdi);
6339 memset(bdi, 0, sizeof(*bdi));
6340 ret = drv->bdrv_co_get_info(bs, bdi);
6345 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6352 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6355 BlockDriver *drv = bs->drv;
6357 if (drv && drv->bdrv_get_specific_info) {
6358 return drv->bdrv_get_specific_info(bs, errp);
6363 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6365 BlockDriver *drv = bs->drv;
6367 if (!drv || !drv->bdrv_get_specific_stats) {
6370 return drv->bdrv_get_specific_stats(bs);
6373 void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6376 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
6380 bs->drv->bdrv_co_debug_event(bs, event);
6383 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6385 GLOBAL_STATE_CODE();
6386 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6387 bs = bdrv_primary_bs(bs);
6390 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6391 assert(bs->drv->bdrv_debug_remove_breakpoint);
6398 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6401 GLOBAL_STATE_CODE();
6402 bs = bdrv_find_debug_node(bs);
6404 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6410 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6412 GLOBAL_STATE_CODE();
6413 bs = bdrv_find_debug_node(bs);
6415 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6421 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6423 GLOBAL_STATE_CODE();
6424 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6425 bs = bdrv_primary_bs(bs);
6428 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6429 return bs->drv->bdrv_debug_resume(bs, tag);
6435 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6437 GLOBAL_STATE_CODE();
6438 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6439 bs = bdrv_primary_bs(bs);
6442 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6443 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6449 /* backing_file can either be relative, or absolute, or a protocol. If it is
6450 * relative, it must be relative to the chain. So, passing in bs->filename
6451 * from a BDS as backing_file should not be done, as that may be relative to
6452 * the CWD rather than the chain. */
6453 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6454 const char *backing_file)
6456 char *filename_full = NULL;
6457 char *backing_file_full = NULL;
6458 char *filename_tmp = NULL;
6459 int is_protocol = 0;
6460 bool filenames_refreshed = false;
6461 BlockDriverState *curr_bs = NULL;
6462 BlockDriverState *retval = NULL;
6463 BlockDriverState *bs_below;
6465 GLOBAL_STATE_CODE();
6467 if (!bs || !bs->drv || !backing_file) {
6471 filename_full = g_malloc(PATH_MAX);
6472 backing_file_full = g_malloc(PATH_MAX);
6474 is_protocol = path_has_protocol(backing_file);
6477 * Being largely a legacy function, skip any filters here
6478 * (because filters do not have normal filenames, so they cannot
6479 * match anyway; and allowing json:{} filenames is a bit out of
6482 for (curr_bs = bdrv_skip_filters(bs);
6483 bdrv_cow_child(curr_bs) != NULL;
6486 bs_below = bdrv_backing_chain_next(curr_bs);
6488 if (bdrv_backing_overridden(curr_bs)) {
6490 * If the backing file was overridden, we can only compare
6491 * directly against the backing node's filename.
6494 if (!filenames_refreshed) {
6496 * This will automatically refresh all of the
6497 * filenames in the rest of the backing chain, so we
6498 * only need to do this once.
6500 bdrv_refresh_filename(bs_below);
6501 filenames_refreshed = true;
6504 if (strcmp(backing_file, bs_below->filename) == 0) {
6508 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6510 * If either of the filename paths is actually a protocol, then
6511 * compare unmodified paths; otherwise make paths relative.
6513 char *backing_file_full_ret;
6515 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6519 /* Also check against the full backing filename for the image */
6520 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6522 if (backing_file_full_ret) {
6523 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6524 g_free(backing_file_full_ret);
6531 /* If not an absolute filename path, make it relative to the current
6532 * image's filename path */
6533 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6535 /* We are going to compare canonicalized absolute pathnames */
6536 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6537 g_free(filename_tmp);
6540 g_free(filename_tmp);
6542 /* We need to make sure the backing filename we are comparing against
6543 * is relative to the current image filename (or absolute) */
6544 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6545 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6546 g_free(filename_tmp);
6549 g_free(filename_tmp);
6551 if (strcmp(backing_file_full, filename_full) == 0) {
6558 g_free(filename_full);
6559 g_free(backing_file_full);
6563 void bdrv_init(void)
6565 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6566 use_bdrv_whitelist = 1;
6568 module_call_init(MODULE_INIT_BLOCK);
6571 void bdrv_init_with_whitelist(void)
6573 use_bdrv_whitelist = 1;
6577 int bdrv_activate(BlockDriverState *bs, Error **errp)
6579 BdrvChild *child, *parent;
6580 Error *local_err = NULL;
6582 BdrvDirtyBitmap *bm;
6584 GLOBAL_STATE_CODE();
6590 QLIST_FOREACH(child, &bs->children, next) {
6591 bdrv_activate(child->bs, &local_err);
6593 error_propagate(errp, local_err);
6599 * Update permissions, they may differ for inactive nodes.
6601 * Note that the required permissions of inactive images are always a
6602 * subset of the permissions required after activating the image. This
6603 * allows us to just get the permissions upfront without restricting
6604 * bdrv_co_invalidate_cache().
6606 * It also means that in error cases, we don't have to try and revert to
6607 * the old permissions (which is an operation that could fail, too). We can
6608 * just keep the extended permissions for the next time that an activation
6609 * of the image is tried.
6611 if (bs->open_flags & BDRV_O_INACTIVE) {
6612 bs->open_flags &= ~BDRV_O_INACTIVE;
6613 ret = bdrv_refresh_perms(bs, NULL, errp);
6615 bs->open_flags |= BDRV_O_INACTIVE;
6619 ret = bdrv_invalidate_cache(bs, errp);
6621 bs->open_flags |= BDRV_O_INACTIVE;
6625 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6626 bdrv_dirty_bitmap_skip_store(bm, false);
6629 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6631 bs->open_flags |= BDRV_O_INACTIVE;
6632 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6637 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6638 if (parent->klass->activate) {
6639 parent->klass->activate(parent, &local_err);
6641 bs->open_flags |= BDRV_O_INACTIVE;
6642 error_propagate(errp, local_err);
6651 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6653 Error *local_err = NULL;
6656 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6657 assert_bdrv_graph_readable();
6659 if (bs->drv->bdrv_co_invalidate_cache) {
6660 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6662 error_propagate(errp, local_err);
6670 void bdrv_activate_all(Error **errp)
6672 BlockDriverState *bs;
6673 BdrvNextIterator it;
6675 GLOBAL_STATE_CODE();
6677 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6678 AioContext *aio_context = bdrv_get_aio_context(bs);
6681 aio_context_acquire(aio_context);
6682 ret = bdrv_activate(bs, errp);
6683 aio_context_release(aio_context);
6685 bdrv_next_cleanup(&it);
6691 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6694 GLOBAL_STATE_CODE();
6696 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6697 if (parent->klass->parent_is_bds) {
6698 BlockDriverState *parent_bs = parent->opaque;
6699 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6708 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6710 BdrvChild *child, *parent;
6712 uint64_t cumulative_perms, cumulative_shared_perms;
6714 GLOBAL_STATE_CODE();
6720 /* Make sure that we don't inactivate a child before its parent.
6721 * It will be covered by recursion from the yet active parent. */
6722 if (bdrv_has_bds_parent(bs, true)) {
6726 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6728 /* Inactivate this node */
6729 if (bs->drv->bdrv_inactivate) {
6730 ret = bs->drv->bdrv_inactivate(bs);
6736 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6737 if (parent->klass->inactivate) {
6738 ret = parent->klass->inactivate(parent);
6745 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6746 &cumulative_shared_perms);
6747 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6748 /* Our inactive parents still need write access. Inactivation failed. */
6752 bs->open_flags |= BDRV_O_INACTIVE;
6755 * Update permissions, they may differ for inactive nodes.
6756 * We only tried to loosen restrictions, so errors are not fatal, ignore
6759 bdrv_refresh_perms(bs, NULL, NULL);
6761 /* Recursively inactivate children */
6762 QLIST_FOREACH(child, &bs->children, next) {
6763 ret = bdrv_inactivate_recurse(child->bs);
6772 int bdrv_inactivate_all(void)
6774 BlockDriverState *bs = NULL;
6775 BdrvNextIterator it;
6777 GSList *aio_ctxs = NULL, *ctx;
6779 GLOBAL_STATE_CODE();
6781 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6782 AioContext *aio_context = bdrv_get_aio_context(bs);
6784 if (!g_slist_find(aio_ctxs, aio_context)) {
6785 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6786 aio_context_acquire(aio_context);
6790 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6791 /* Nodes with BDS parents are covered by recursion from the last
6792 * parent that gets inactivated. Don't inactivate them a second
6793 * time if that has already happened. */
6794 if (bdrv_has_bds_parent(bs, false)) {
6797 ret = bdrv_inactivate_recurse(bs);
6799 bdrv_next_cleanup(&it);
6805 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6806 AioContext *aio_context = ctx->data;
6807 aio_context_release(aio_context);
6809 g_slist_free(aio_ctxs);
6814 /**************************************************************/
6815 /* removable device support */
6818 * Return TRUE if the media is present
6820 bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
6822 BlockDriver *drv = bs->drv;
6825 assert_bdrv_graph_readable();
6830 if (drv->bdrv_co_is_inserted) {
6831 return drv->bdrv_co_is_inserted(bs);
6833 QLIST_FOREACH(child, &bs->children, next) {
6834 if (!bdrv_co_is_inserted(child->bs)) {
6842 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6844 void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
6846 BlockDriver *drv = bs->drv;
6848 assert_bdrv_graph_readable();
6850 if (drv && drv->bdrv_co_eject) {
6851 drv->bdrv_co_eject(bs, eject_flag);
6856 * Lock or unlock the media (if it is locked, the user won't be able
6857 * to eject it manually).
6859 void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
6861 BlockDriver *drv = bs->drv;
6863 assert_bdrv_graph_readable();
6864 trace_bdrv_lock_medium(bs, locked);
6866 if (drv && drv->bdrv_co_lock_medium) {
6867 drv->bdrv_co_lock_medium(bs, locked);
6871 /* Get a reference to bs */
6872 void bdrv_ref(BlockDriverState *bs)
6874 GLOBAL_STATE_CODE();
6878 /* Release a previously grabbed reference to bs.
6879 * If after releasing, reference count is zero, the BlockDriverState is
6881 void bdrv_unref(BlockDriverState *bs)
6883 GLOBAL_STATE_CODE();
6887 assert(bs->refcnt > 0);
6888 if (--bs->refcnt == 0) {
6893 struct BdrvOpBlocker {
6895 QLIST_ENTRY(BdrvOpBlocker) list;
6898 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6900 BdrvOpBlocker *blocker;
6901 GLOBAL_STATE_CODE();
6902 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6903 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6904 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6905 error_propagate_prepend(errp, error_copy(blocker->reason),
6906 "Node '%s' is busy: ",
6907 bdrv_get_device_or_node_name(bs));
6913 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6915 BdrvOpBlocker *blocker;
6916 GLOBAL_STATE_CODE();
6917 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6919 blocker = g_new0(BdrvOpBlocker, 1);
6920 blocker->reason = reason;
6921 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6924 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6926 BdrvOpBlocker *blocker, *next;
6927 GLOBAL_STATE_CODE();
6928 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6929 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6930 if (blocker->reason == reason) {
6931 QLIST_REMOVE(blocker, list);
6937 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6940 GLOBAL_STATE_CODE();
6941 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6942 bdrv_op_block(bs, i, reason);
6946 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6949 GLOBAL_STATE_CODE();
6950 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6951 bdrv_op_unblock(bs, i, reason);
6955 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6958 GLOBAL_STATE_CODE();
6959 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6960 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6968 * Must not be called while holding the lock of an AioContext other than the
6971 void bdrv_img_create(const char *filename, const char *fmt,
6972 const char *base_filename, const char *base_fmt,
6973 char *options, uint64_t img_size, int flags, bool quiet,
6976 QemuOptsList *create_opts = NULL;
6977 QemuOpts *opts = NULL;
6978 const char *backing_fmt, *backing_file;
6980 BlockDriver *drv, *proto_drv;
6981 Error *local_err = NULL;
6984 GLOBAL_STATE_CODE();
6986 /* Find driver and parse its options */
6987 drv = bdrv_find_format(fmt);
6989 error_setg(errp, "Unknown file format '%s'", fmt);
6993 proto_drv = bdrv_find_protocol(filename, true, errp);
6998 if (!drv->create_opts) {
6999 error_setg(errp, "Format driver '%s' does not support image creation",
7004 if (!proto_drv->create_opts) {
7005 error_setg(errp, "Protocol driver '%s' does not support image creation",
7006 proto_drv->format_name);
7010 /* Create parameter list */
7011 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7012 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7014 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7016 /* Parse -o options */
7018 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7023 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7024 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7025 } else if (img_size != UINT64_C(-1)) {
7026 error_setg(errp, "The image size must be specified only once");
7030 if (base_filename) {
7031 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7033 error_setg(errp, "Backing file not supported for file format '%s'",
7040 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7041 error_setg(errp, "Backing file format not supported for file "
7042 "format '%s'", fmt);
7047 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7049 if (!strcmp(filename, backing_file)) {
7050 error_setg(errp, "Error: Trying to create an image with the "
7051 "same filename as the backing file");
7054 if (backing_file[0] == '\0') {
7055 error_setg(errp, "Expected backing file name, got empty string");
7060 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7062 /* The size for the image must always be specified, unless we have a backing
7063 * file and we have not been forbidden from opening it. */
7064 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7065 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7066 BlockDriverState *bs;
7069 QDict *backing_options = NULL;
7072 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7077 assert(full_backing);
7080 * No need to do I/O here, which allows us to open encrypted
7081 * backing images without needing the secret
7084 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7085 back_flags |= BDRV_O_NO_IO;
7087 backing_options = qdict_new();
7089 qdict_put_str(backing_options, "driver", backing_fmt);
7091 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7093 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7095 g_free(full_backing);
7097 error_append_hint(&local_err, "Could not open backing image.\n");
7101 error_setg(&local_err,
7102 "Backing file specified without backing format");
7103 error_append_hint(&local_err, "Detected format of %s.",
7104 bs->drv->format_name);
7108 /* Opened BS, have no size */
7109 size = bdrv_getlength(bs);
7111 error_setg_errno(errp, -size, "Could not get size of '%s'",
7116 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7120 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7121 } else if (backing_file && !backing_fmt) {
7122 error_setg(&local_err,
7123 "Backing file specified without backing format");
7128 error_setg(errp, "Image creation needs a size parameter");
7133 printf("Formatting '%s', fmt=%s ", filename, fmt);
7134 qemu_opts_print(opts, " ");
7139 ret = bdrv_create(drv, filename, opts, &local_err);
7141 if (ret == -EFBIG) {
7142 /* This is generally a better message than whatever the driver would
7143 * deliver (especially because of the cluster_size_hint), since that
7144 * is most probably not much different from "image too large". */
7145 const char *cluster_size_hint = "";
7146 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7147 cluster_size_hint = " (try using a larger cluster size)";
7149 error_setg(errp, "The image size is too large for file format '%s'"
7150 "%s", fmt, cluster_size_hint);
7151 error_free(local_err);
7156 qemu_opts_del(opts);
7157 qemu_opts_free(create_opts);
7158 error_propagate(errp, local_err);
7161 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7164 return bs ? bs->aio_context : qemu_get_aio_context();
7167 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7169 Coroutine *self = qemu_coroutine_self();
7170 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7171 AioContext *new_ctx;
7175 * Increase bs->in_flight to ensure that this operation is completed before
7176 * moving the node to a different AioContext. Read new_ctx only afterwards.
7178 bdrv_inc_in_flight(bs);
7180 new_ctx = bdrv_get_aio_context(bs);
7181 aio_co_reschedule_self(new_ctx);
7185 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7188 aio_co_reschedule_self(old_ctx);
7189 bdrv_dec_in_flight(bs);
7192 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7194 AioContext *ctx = bdrv_get_aio_context(bs);
7196 /* In the main thread, bs->aio_context won't change concurrently */
7197 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7200 * We're in coroutine context, so we already hold the lock of the main
7201 * loop AioContext. Don't lock it twice to avoid deadlocks.
7203 assert(qemu_in_coroutine());
7204 if (ctx != qemu_get_aio_context()) {
7205 aio_context_acquire(ctx);
7209 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7211 AioContext *ctx = bdrv_get_aio_context(bs);
7213 assert(qemu_in_coroutine());
7214 if (ctx != qemu_get_aio_context()) {
7215 aio_context_release(ctx);
7219 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7221 GLOBAL_STATE_CODE();
7222 QLIST_REMOVE(ban, list);
7226 static void bdrv_detach_aio_context(BlockDriverState *bs)
7228 BdrvAioNotifier *baf, *baf_tmp;
7230 assert(!bs->walking_aio_notifiers);
7231 GLOBAL_STATE_CODE();
7232 bs->walking_aio_notifiers = true;
7233 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7235 bdrv_do_remove_aio_context_notifier(baf);
7237 baf->detach_aio_context(baf->opaque);
7240 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7241 * remove remaining aio notifiers if we aren't called again.
7243 bs->walking_aio_notifiers = false;
7245 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7246 bs->drv->bdrv_detach_aio_context(bs);
7249 if (bs->quiesce_counter) {
7250 aio_enable_external(bs->aio_context);
7252 bs->aio_context = NULL;
7255 static void bdrv_attach_aio_context(BlockDriverState *bs,
7256 AioContext *new_context)
7258 BdrvAioNotifier *ban, *ban_tmp;
7259 GLOBAL_STATE_CODE();
7261 if (bs->quiesce_counter) {
7262 aio_disable_external(new_context);
7265 bs->aio_context = new_context;
7267 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7268 bs->drv->bdrv_attach_aio_context(bs, new_context);
7271 assert(!bs->walking_aio_notifiers);
7272 bs->walking_aio_notifiers = true;
7273 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7275 bdrv_do_remove_aio_context_notifier(ban);
7277 ban->attached_aio_context(new_context, ban->opaque);
7280 bs->walking_aio_notifiers = false;
7283 typedef struct BdrvStateSetAioContext {
7284 AioContext *new_ctx;
7285 BlockDriverState *bs;
7286 } BdrvStateSetAioContext;
7288 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7289 GHashTable *visited,
7293 GLOBAL_STATE_CODE();
7294 if (g_hash_table_contains(visited, c)) {
7297 g_hash_table_add(visited, c);
7300 * A BdrvChildClass that doesn't handle AioContext changes cannot
7301 * tolerate any AioContext changes
7303 if (!c->klass->change_aio_ctx) {
7304 char *user = bdrv_child_user_desc(c);
7305 error_setg(errp, "Changing iothreads is not supported by %s", user);
7309 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7310 assert(!errp || *errp);
7316 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7317 GHashTable *visited, Transaction *tran,
7320 GLOBAL_STATE_CODE();
7321 if (g_hash_table_contains(visited, c)) {
7324 g_hash_table_add(visited, c);
7325 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7328 static void bdrv_set_aio_context_clean(void *opaque)
7330 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7331 BlockDriverState *bs = (BlockDriverState *) state->bs;
7333 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7334 bdrv_drained_end(bs);
7339 static void bdrv_set_aio_context_commit(void *opaque)
7341 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7342 BlockDriverState *bs = (BlockDriverState *) state->bs;
7343 AioContext *new_context = state->new_ctx;
7344 AioContext *old_context = bdrv_get_aio_context(bs);
7347 * Take the old AioContex when detaching it from bs.
7348 * At this point, new_context lock is already acquired, and we are now
7349 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7350 * does not call AIO_POLL_WHILE().
7352 if (old_context != qemu_get_aio_context()) {
7353 aio_context_acquire(old_context);
7355 bdrv_detach_aio_context(bs);
7356 if (old_context != qemu_get_aio_context()) {
7357 aio_context_release(old_context);
7359 bdrv_attach_aio_context(bs, new_context);
7362 static TransactionActionDrv set_aio_context = {
7363 .commit = bdrv_set_aio_context_commit,
7364 .clean = bdrv_set_aio_context_clean,
7368 * Changes the AioContext used for fd handlers, timers, and BHs by this
7369 * BlockDriverState and all its children and parents.
7371 * Must be called from the main AioContext.
7373 * The caller must own the AioContext lock for the old AioContext of bs, but it
7374 * must not own the AioContext lock for new_context (unless new_context is the
7375 * same as the current context of bs).
7377 * @visited will accumulate all visited BdrvChild objects. The caller is
7378 * responsible for freeing the list afterwards.
7380 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7381 GHashTable *visited, Transaction *tran,
7385 BdrvStateSetAioContext *state;
7387 GLOBAL_STATE_CODE();
7389 if (bdrv_get_aio_context(bs) == ctx) {
7393 QLIST_FOREACH(c, &bs->parents, next_parent) {
7394 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7399 QLIST_FOREACH(c, &bs->children, next) {
7400 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7405 state = g_new(BdrvStateSetAioContext, 1);
7406 *state = (BdrvStateSetAioContext) {
7411 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7412 bdrv_drained_begin(bs);
7414 tran_add(tran, &set_aio_context, state);
7420 * Change bs's and recursively all of its parents' and children's AioContext
7421 * to the given new context, returning an error if that isn't possible.
7423 * If ignore_child is not NULL, that child (and its subgraph) will not
7426 * This function still requires the caller to take the bs current
7427 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7428 * assumes the lock is always held if bs is in another AioContext.
7429 * For the same reason, it temporarily also holds the new AioContext, since
7430 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7431 * Therefore the new AioContext lock must not be taken by the caller.
7433 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7434 BdrvChild *ignore_child, Error **errp)
7437 GHashTable *visited;
7439 AioContext *old_context = bdrv_get_aio_context(bs);
7440 GLOBAL_STATE_CODE();
7443 * Recursion phase: go through all nodes of the graph.
7444 * Take care of checking that all nodes support changing AioContext
7445 * and drain them, builing a linear list of callbacks to run if everything
7446 * is successful (the transaction itself).
7449 visited = g_hash_table_new(NULL, NULL);
7451 g_hash_table_add(visited, ignore_child);
7453 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7454 g_hash_table_destroy(visited);
7457 * Linear phase: go through all callbacks collected in the transaction.
7458 * Run all callbacks collected in the recursion to switch all nodes
7459 * AioContext lock (transaction commit), or undo all changes done in the
7460 * recursion (transaction abort).
7464 /* Just run clean() callbacks. No AioContext changed. */
7470 * Release old AioContext, it won't be needed anymore, as all
7471 * bdrv_drained_begin() have been called already.
7473 if (qemu_get_aio_context() != old_context) {
7474 aio_context_release(old_context);
7478 * Acquire new AioContext since bdrv_drained_end() is going to be called
7479 * after we switched all nodes in the new AioContext, and the function
7480 * assumes that the lock of the bs is always taken.
7482 if (qemu_get_aio_context() != ctx) {
7483 aio_context_acquire(ctx);
7488 if (qemu_get_aio_context() != ctx) {
7489 aio_context_release(ctx);
7492 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7493 if (qemu_get_aio_context() != old_context) {
7494 aio_context_acquire(old_context);
7500 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7501 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7502 void (*detach_aio_context)(void *opaque), void *opaque)
7504 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7505 *ban = (BdrvAioNotifier){
7506 .attached_aio_context = attached_aio_context,
7507 .detach_aio_context = detach_aio_context,
7510 GLOBAL_STATE_CODE();
7512 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7515 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7516 void (*attached_aio_context)(AioContext *,
7518 void (*detach_aio_context)(void *),
7521 BdrvAioNotifier *ban, *ban_next;
7522 GLOBAL_STATE_CODE();
7524 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7525 if (ban->attached_aio_context == attached_aio_context &&
7526 ban->detach_aio_context == detach_aio_context &&
7527 ban->opaque == opaque &&
7528 ban->deleted == false)
7530 if (bs->walking_aio_notifiers) {
7531 ban->deleted = true;
7533 bdrv_do_remove_aio_context_notifier(ban);
7542 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7543 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7547 GLOBAL_STATE_CODE();
7549 error_setg(errp, "Node is ejected");
7552 if (!bs->drv->bdrv_amend_options) {
7553 error_setg(errp, "Block driver '%s' does not support option amendment",
7554 bs->drv->format_name);
7557 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7558 cb_opaque, force, errp);
7562 * This function checks whether the given @to_replace is allowed to be
7563 * replaced by a node that always shows the same data as @bs. This is
7564 * used for example to verify whether the mirror job can replace
7565 * @to_replace by the target mirrored from @bs.
7566 * To be replaceable, @bs and @to_replace may either be guaranteed to
7567 * always show the same data (because they are only connected through
7568 * filters), or some driver may allow replacing one of its children
7569 * because it can guarantee that this child's data is not visible at
7570 * all (for example, for dissenting quorum children that have no other
7573 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7574 BlockDriverState *to_replace)
7576 BlockDriverState *filtered;
7578 GLOBAL_STATE_CODE();
7580 if (!bs || !bs->drv) {
7584 if (bs == to_replace) {
7588 /* See what the driver can do */
7589 if (bs->drv->bdrv_recurse_can_replace) {
7590 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7593 /* For filters without an own implementation, we can recurse on our own */
7594 filtered = bdrv_filter_bs(bs);
7596 return bdrv_recurse_can_replace(filtered, to_replace);
7604 * Check whether the given @node_name can be replaced by a node that
7605 * has the same data as @parent_bs. If so, return @node_name's BDS;
7608 * @node_name must be a (recursive) *child of @parent_bs (or this
7609 * function will return NULL).
7611 * The result (whether the node can be replaced or not) is only valid
7612 * for as long as no graph or permission changes occur.
7614 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7615 const char *node_name, Error **errp)
7617 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7618 AioContext *aio_context;
7620 GLOBAL_STATE_CODE();
7622 if (!to_replace_bs) {
7623 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7627 aio_context = bdrv_get_aio_context(to_replace_bs);
7628 aio_context_acquire(aio_context);
7630 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7631 to_replace_bs = NULL;
7635 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7636 * most non filter in order to prevent data corruption.
7637 * Another benefit is that this tests exclude backing files which are
7638 * blocked by the backing blockers.
7640 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7641 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7642 "because it cannot be guaranteed that doing so would not "
7643 "lead to an abrupt change of visible data",
7644 node_name, parent_bs->node_name);
7645 to_replace_bs = NULL;
7650 aio_context_release(aio_context);
7651 return to_replace_bs;
7655 * Iterates through the list of runtime option keys that are said to
7656 * be "strong" for a BDS. An option is called "strong" if it changes
7657 * a BDS's data. For example, the null block driver's "size" and
7658 * "read-zeroes" options are strong, but its "latency-ns" option is
7661 * If a key returned by this function ends with a dot, all options
7662 * starting with that prefix are strong.
7664 static const char *const *strong_options(BlockDriverState *bs,
7665 const char *const *curopt)
7667 static const char *const global_options[] = {
7668 "driver", "filename", NULL
7672 return &global_options[0];
7676 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7677 curopt = bs->drv->strong_runtime_opts;
7680 return (curopt && *curopt) ? curopt : NULL;
7684 * Copies all strong runtime options from bs->options to the given
7685 * QDict. The set of strong option keys is determined by invoking
7688 * Returns true iff any strong option was present in bs->options (and
7689 * thus copied to the target QDict) with the exception of "filename"
7690 * and "driver". The caller is expected to use this value to decide
7691 * whether the existence of strong options prevents the generation of
7694 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7696 bool found_any = false;
7697 const char *const *option_name = NULL;
7703 while ((option_name = strong_options(bs, option_name))) {
7704 bool option_given = false;
7706 assert(strlen(*option_name) > 0);
7707 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7708 QObject *entry = qdict_get(bs->options, *option_name);
7713 qdict_put_obj(d, *option_name, qobject_ref(entry));
7714 option_given = true;
7716 const QDictEntry *entry;
7717 for (entry = qdict_first(bs->options); entry;
7718 entry = qdict_next(bs->options, entry))
7720 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7721 qdict_put_obj(d, qdict_entry_key(entry),
7722 qobject_ref(qdict_entry_value(entry)));
7723 option_given = true;
7728 /* While "driver" and "filename" need to be included in a JSON filename,
7729 * their existence does not prohibit generation of a plain filename. */
7730 if (!found_any && option_given &&
7731 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7737 if (!qdict_haskey(d, "driver")) {
7738 /* Drivers created with bdrv_new_open_driver() may not have a
7739 * @driver option. Add it here. */
7740 qdict_put_str(d, "driver", bs->drv->format_name);
7746 /* Note: This function may return false positives; it may return true
7747 * even if opening the backing file specified by bs's image header
7748 * would result in exactly bs->backing. */
7749 static bool bdrv_backing_overridden(BlockDriverState *bs)
7751 GLOBAL_STATE_CODE();
7753 return strcmp(bs->auto_backing_file,
7754 bs->backing->bs->filename);
7756 /* No backing BDS, so if the image header reports any backing
7757 * file, it must have been suppressed */
7758 return bs->auto_backing_file[0] != '\0';
7762 /* Updates the following BDS fields:
7763 * - exact_filename: A filename which may be used for opening a block device
7764 * which (mostly) equals the given BDS (even without any
7765 * other options; so reading and writing must return the same
7766 * results, but caching etc. may be different)
7767 * - full_open_options: Options which, when given when opening a block device
7768 * (without a filename), result in a BDS (mostly)
7769 * equalling the given one
7770 * - filename: If exact_filename is set, it is copied here. Otherwise,
7771 * full_open_options is converted to a JSON object, prefixed with
7772 * "json:" (for use through the JSON pseudo protocol) and put here.
7774 void bdrv_refresh_filename(BlockDriverState *bs)
7776 BlockDriver *drv = bs->drv;
7778 BlockDriverState *primary_child_bs;
7780 bool backing_overridden;
7781 bool generate_json_filename; /* Whether our default implementation should
7782 fill exact_filename (false) or not (true) */
7784 GLOBAL_STATE_CODE();
7790 /* This BDS's file name may depend on any of its children's file names, so
7791 * refresh those first */
7792 QLIST_FOREACH(child, &bs->children, next) {
7793 bdrv_refresh_filename(child->bs);
7797 /* For implicit nodes, just copy everything from the single child */
7798 child = QLIST_FIRST(&bs->children);
7799 assert(QLIST_NEXT(child, next) == NULL);
7801 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7802 child->bs->exact_filename);
7803 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7805 qobject_unref(bs->full_open_options);
7806 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7811 backing_overridden = bdrv_backing_overridden(bs);
7813 if (bs->open_flags & BDRV_O_NO_IO) {
7814 /* Without I/O, the backing file does not change anything.
7815 * Therefore, in such a case (primarily qemu-img), we can
7816 * pretend the backing file has not been overridden even if
7817 * it technically has been. */
7818 backing_overridden = false;
7821 /* Gather the options QDict */
7823 generate_json_filename = append_strong_runtime_options(opts, bs);
7824 generate_json_filename |= backing_overridden;
7826 if (drv->bdrv_gather_child_options) {
7827 /* Some block drivers may not want to present all of their children's
7828 * options, or name them differently from BdrvChild.name */
7829 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7831 QLIST_FOREACH(child, &bs->children, next) {
7832 if (child == bs->backing && !backing_overridden) {
7833 /* We can skip the backing BDS if it has not been overridden */
7837 qdict_put(opts, child->name,
7838 qobject_ref(child->bs->full_open_options));
7841 if (backing_overridden && !bs->backing) {
7842 /* Force no backing file */
7843 qdict_put_null(opts, "backing");
7847 qobject_unref(bs->full_open_options);
7848 bs->full_open_options = opts;
7850 primary_child_bs = bdrv_primary_bs(bs);
7852 if (drv->bdrv_refresh_filename) {
7853 /* Obsolete information is of no use here, so drop the old file name
7854 * information before refreshing it */
7855 bs->exact_filename[0] = '\0';
7857 drv->bdrv_refresh_filename(bs);
7858 } else if (primary_child_bs) {
7860 * Try to reconstruct valid information from the underlying
7861 * file -- this only works for format nodes (filter nodes
7862 * cannot be probed and as such must be selected by the user
7863 * either through an options dict, or through a special
7864 * filename which the filter driver must construct in its
7865 * .bdrv_refresh_filename() implementation).
7868 bs->exact_filename[0] = '\0';
7871 * We can use the underlying file's filename if:
7872 * - it has a filename,
7873 * - the current BDS is not a filter,
7874 * - the file is a protocol BDS, and
7875 * - opening that file (as this BDS's format) will automatically create
7876 * the BDS tree we have right now, that is:
7877 * - the user did not significantly change this BDS's behavior with
7878 * some explicit (strong) options
7879 * - no non-file child of this BDS has been overridden by the user
7880 * Both of these conditions are represented by generate_json_filename.
7882 if (primary_child_bs->exact_filename[0] &&
7883 primary_child_bs->drv->bdrv_file_open &&
7884 !drv->is_filter && !generate_json_filename)
7886 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7890 if (bs->exact_filename[0]) {
7891 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7893 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7894 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7895 json->str) >= sizeof(bs->filename)) {
7896 /* Give user a hint if we truncated things. */
7897 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7899 g_string_free(json, true);
7903 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7905 BlockDriver *drv = bs->drv;
7906 BlockDriverState *child_bs;
7908 GLOBAL_STATE_CODE();
7911 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7915 if (drv->bdrv_dirname) {
7916 return drv->bdrv_dirname(bs, errp);
7919 child_bs = bdrv_primary_bs(bs);
7921 return bdrv_dirname(child_bs, errp);
7924 bdrv_refresh_filename(bs);
7925 if (bs->exact_filename[0] != '\0') {
7926 return path_combine(bs->exact_filename, "");
7929 error_setg(errp, "Cannot generate a base directory for %s nodes",
7935 * Hot add/remove a BDS's child. So the user can take a child offline when
7936 * it is broken and take a new child online
7938 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7941 GLOBAL_STATE_CODE();
7942 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7943 error_setg(errp, "The node %s does not support adding a child",
7944 bdrv_get_device_or_node_name(parent_bs));
7948 if (!QLIST_EMPTY(&child_bs->parents)) {
7949 error_setg(errp, "The node %s already has a parent",
7950 child_bs->node_name);
7954 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7957 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7961 GLOBAL_STATE_CODE();
7962 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7963 error_setg(errp, "The node %s does not support removing a child",
7964 bdrv_get_device_or_node_name(parent_bs));
7968 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7975 error_setg(errp, "The node %s does not have a child named %s",
7976 bdrv_get_device_or_node_name(parent_bs),
7977 bdrv_get_device_or_node_name(child->bs));
7981 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7984 int bdrv_make_empty(BdrvChild *c, Error **errp)
7986 BlockDriver *drv = c->bs->drv;
7989 GLOBAL_STATE_CODE();
7990 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7992 if (!drv->bdrv_make_empty) {
7993 error_setg(errp, "%s does not support emptying nodes",
7998 ret = drv->bdrv_make_empty(c->bs);
8000 error_setg_errno(errp, -ret, "Failed to empty %s",
8009 * Return the child that @bs acts as an overlay for, and from which data may be
8010 * copied in COW or COR operations. Usually this is the backing file.
8012 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8016 if (!bs || !bs->drv) {
8020 if (bs->drv->is_filter) {
8028 assert(bs->backing->role & BDRV_CHILD_COW);
8033 * If @bs acts as a filter for exactly one of its children, return
8036 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8041 if (!bs || !bs->drv) {
8045 if (!bs->drv->is_filter) {
8049 /* Only one of @backing or @file may be used */
8050 assert(!(bs->backing && bs->file));
8052 c = bs->backing ?: bs->file;
8057 assert(c->role & BDRV_CHILD_FILTERED);
8062 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8063 * whichever is non-NULL.
8065 * Return NULL if both are NULL.
8067 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8069 BdrvChild *cow_child = bdrv_cow_child(bs);
8070 BdrvChild *filter_child = bdrv_filter_child(bs);
8073 /* Filter nodes cannot have COW backing files */
8074 assert(!(cow_child && filter_child));
8076 return cow_child ?: filter_child;
8080 * Return the primary child of this node: For filters, that is the
8081 * filtered child. For other nodes, that is usually the child storing
8083 * (A generally more helpful description is that this is (usually) the
8084 * child that has the same filename as @bs.)
8086 * Drivers do not necessarily have a primary child; for example quorum
8089 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8091 BdrvChild *c, *found = NULL;
8094 QLIST_FOREACH(c, &bs->children, next) {
8095 if (c->role & BDRV_CHILD_PRIMARY) {
8104 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8105 bool stop_on_explicit_filter)
8113 while (!(stop_on_explicit_filter && !bs->implicit)) {
8114 c = bdrv_filter_child(bs);
8117 * A filter that is embedded in a working block graph must
8118 * have a child. Assert this here so this function does
8119 * not return a filter node that is not expected by the
8122 assert(!bs->drv || !bs->drv->is_filter);
8128 * Note that this treats nodes with bs->drv == NULL as not being
8129 * filters (bs->drv == NULL should be replaced by something else
8131 * The advantage of this behavior is that this function will thus
8132 * always return a non-NULL value (given a non-NULL @bs).
8139 * Return the first BDS that has not been added implicitly or that
8140 * does not have a filtered child down the chain starting from @bs
8141 * (including @bs itself).
8143 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8145 GLOBAL_STATE_CODE();
8146 return bdrv_do_skip_filters(bs, true);
8150 * Return the first BDS that does not have a filtered child down the
8151 * chain starting from @bs (including @bs itself).
8153 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8156 return bdrv_do_skip_filters(bs, false);
8160 * For a backing chain, return the first non-filter backing image of
8161 * the first non-filter image.
8163 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8166 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8170 * Check whether [offset, offset + bytes) overlaps with the cached
8171 * block-status data region.
8173 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8174 * which is what bdrv_bsc_is_data()'s interface needs.
8175 * Otherwise, *pnum is not touched.
8177 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8178 int64_t offset, int64_t bytes,
8181 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8185 qatomic_read(&bsc->valid) &&
8186 ranges_overlap(offset, bytes, bsc->data_start,
8187 bsc->data_end - bsc->data_start);
8189 if (overlaps && pnum) {
8190 *pnum = bsc->data_end - offset;
8197 * See block_int.h for this function's documentation.
8199 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8202 RCU_READ_LOCK_GUARD();
8203 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8207 * See block_int.h for this function's documentation.
8209 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8210 int64_t offset, int64_t bytes)
8213 RCU_READ_LOCK_GUARD();
8215 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8216 qatomic_set(&bs->block_status_cache->valid, false);
8221 * See block_int.h for this function's documentation.
8223 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8225 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8226 BdrvBlockStatusCache *old_bsc;
8229 *new_bsc = (BdrvBlockStatusCache) {
8231 .data_start = offset,
8232 .data_end = offset + bytes,
8235 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8237 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8238 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8240 g_free_rcu(old_bsc, rcu);