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_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)
537 if (!drv->bdrv_co_create_opts) {
538 error_setg(errp, "Driver '%s' does not support image creation",
543 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
544 if (ret < 0 && !*errp) {
545 error_setg_errno(errp, -ret, "Could not create image");
552 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
553 * least the given @minimum_size.
555 * On success, return @blk's actual length.
556 * Otherwise, return -errno.
558 static int64_t coroutine_fn GRAPH_UNLOCKED
559 create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
562 Error *local_err = NULL;
568 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
570 if (ret < 0 && ret != -ENOTSUP) {
571 error_propagate(errp, local_err);
575 size = blk_co_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 creating "
665 "new images, so an existing image must be selected as "
666 "the target; however, opening the given target as an "
667 "existing image failed: ",
672 size = create_file_fallback_truncate(blk, size, errp);
678 ret = create_file_fallback_zero_first_sector(blk, size, errp);
689 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
692 QemuOpts *protocol_opts;
699 drv = bdrv_find_protocol(filename, true, errp);
704 if (!drv->create_opts) {
705 error_setg(errp, "Driver '%s' does not support image creation",
711 * 'opts' contains a QemuOptsList with a combination of format and protocol
714 * The format properly removes its options, but the default values remain
715 * in 'opts->list'. So if the protocol has options with the same name
716 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
717 * of the format, since for overlapping options, the format wins.
719 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
720 * only the set options, and then convert it back to QemuOpts, using the
721 * create_opts of the protocol. So the new QemuOpts, will contain only the
724 qdict = qemu_opts_to_qdict(opts, NULL);
725 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
726 if (protocol_opts == NULL) {
731 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
733 qemu_opts_del(protocol_opts);
734 qobject_unref(qdict);
738 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
740 Error *local_err = NULL;
745 assert_bdrv_graph_readable();
748 error_setg(errp, "Block node '%s' is not opened", bs->filename);
752 if (!bs->drv->bdrv_co_delete_file) {
753 error_setg(errp, "Driver '%s' does not support image deletion",
754 bs->drv->format_name);
758 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
760 error_propagate(errp, local_err);
766 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
768 Error *local_err = NULL;
776 ret = bdrv_co_delete_file(bs, &local_err);
778 * ENOTSUP will happen if the block driver doesn't support
779 * the 'bdrv_co_delete_file' interface. This is a predictable
780 * scenario and shouldn't be reported back to the user.
782 if (ret == -ENOTSUP) {
783 error_free(local_err);
784 } else if (ret < 0) {
785 error_report_err(local_err);
790 * Try to get @bs's logical and physical block size.
791 * On success, store them in @bsz struct and return 0.
792 * On failure return -errno.
793 * @bs must not be empty.
795 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
797 BlockDriver *drv = bs->drv;
798 BlockDriverState *filtered = bdrv_filter_bs(bs);
801 if (drv && drv->bdrv_probe_blocksizes) {
802 return drv->bdrv_probe_blocksizes(bs, bsz);
803 } else if (filtered) {
804 return bdrv_probe_blocksizes(filtered, bsz);
811 * Try to get @bs's geometry (cyls, heads, sectors).
812 * On success, store them in @geo struct and return 0.
813 * On failure return -errno.
814 * @bs must not be empty.
816 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
818 BlockDriver *drv = bs->drv;
819 BlockDriverState *filtered = bdrv_filter_bs(bs);
822 if (drv && drv->bdrv_probe_geometry) {
823 return drv->bdrv_probe_geometry(bs, geo);
824 } else if (filtered) {
825 return bdrv_probe_geometry(filtered, geo);
832 * Create a uniquely-named empty temporary file.
833 * Return the actual file name used upon success, otherwise NULL.
834 * This string should be freed with g_free() when not needed any longer.
836 * Note: creating a temporary file for the caller to (re)open is
837 * inherently racy. Use g_file_open_tmp() instead whenever practical.
839 char *create_tmp_file(Error **errp)
843 g_autofree char *filename = NULL;
845 tmpdir = g_get_tmp_dir();
848 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
850 * This function is used to create temporary disk images (like -snapshot),
851 * so the files can become very large. /tmp is often a tmpfs where as
852 * /var/tmp is usually on a disk, so more appropriate for disk images.
854 if (!g_strcmp0(tmpdir, "/tmp")) {
859 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
860 fd = g_mkstemp(filename);
862 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
868 return g_steal_pointer(&filename);
872 * Detect host devices. By convention, /dev/cdrom[N] is always
873 * recognized as a host CDROM.
875 static BlockDriver *find_hdev_driver(const char *filename)
877 int score_max = 0, score;
878 BlockDriver *drv = NULL, *d;
881 QLIST_FOREACH(d, &bdrv_drivers, list) {
882 if (d->bdrv_probe_device) {
883 score = d->bdrv_probe_device(filename);
884 if (score > score_max) {
894 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
899 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
900 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
908 BlockDriver *bdrv_find_protocol(const char *filename,
909 bool allow_protocol_prefix,
919 /* TODO Drivers without bdrv_file_open must be specified explicitly */
922 * XXX(hch): we really should not let host device detection
923 * override an explicit protocol specification, but moving this
924 * later breaks access to device names with colons in them.
925 * Thanks to the brain-dead persistent naming schemes on udev-
926 * based Linux systems those actually are quite common.
928 drv1 = find_hdev_driver(filename);
933 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
937 p = strchr(filename, ':');
940 if (len > sizeof(protocol) - 1)
941 len = sizeof(protocol) - 1;
942 memcpy(protocol, filename, len);
943 protocol[len] = '\0';
945 drv1 = bdrv_do_find_protocol(protocol);
950 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
951 if (block_driver_modules[i].protocol_name &&
952 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
953 int rv = block_module_load(block_driver_modules[i].library_name, errp);
955 drv1 = bdrv_do_find_protocol(protocol);
964 error_setg(errp, "Unknown protocol '%s'", protocol);
970 * Guess image format by probing its contents.
971 * This is not a good idea when your image is raw (CVE-2008-2004), but
972 * we do it anyway for backward compatibility.
974 * @buf contains the image's first @buf_size bytes.
975 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
976 * but can be smaller if the image file is smaller)
977 * @filename is its filename.
979 * For all block drivers, call the bdrv_probe() method to get its
981 * Return the first block driver with the highest probing score.
983 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
984 const char *filename)
986 int score_max = 0, score;
987 BlockDriver *drv = NULL, *d;
990 QLIST_FOREACH(d, &bdrv_drivers, list) {
992 score = d->bdrv_probe(buf, buf_size, filename);
993 if (score > score_max) {
1003 static int find_image_format(BlockBackend *file, const char *filename,
1004 BlockDriver **pdrv, Error **errp)
1007 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1010 GLOBAL_STATE_CODE();
1012 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1013 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1018 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1020 error_setg_errno(errp, -ret, "Could not read image for determining its "
1026 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1028 error_setg(errp, "Could not determine image format: No compatible "
1039 * Set the current 'total_sectors' value
1040 * Return 0 on success, -errno on error.
1042 int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1045 BlockDriver *drv = bs->drv;
1047 assert_bdrv_graph_readable();
1053 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1057 /* query actual device if possible, otherwise just trust the hint */
1058 if (drv->bdrv_co_getlength) {
1059 int64_t length = drv->bdrv_co_getlength(bs);
1063 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1066 bs->total_sectors = hint;
1068 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1076 * Combines a QDict of new block driver @options with any missing options taken
1077 * from @old_options, so that leaving out an option defaults to its old value.
1079 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1082 GLOBAL_STATE_CODE();
1083 if (bs->drv && bs->drv->bdrv_join_options) {
1084 bs->drv->bdrv_join_options(options, old_options);
1086 qdict_join(options, old_options, false);
1090 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1094 Error *local_err = NULL;
1095 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1096 BlockdevDetectZeroesOptions detect_zeroes =
1097 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1098 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1099 GLOBAL_STATE_CODE();
1102 error_propagate(errp, local_err);
1103 return detect_zeroes;
1106 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1107 !(open_flags & BDRV_O_UNMAP))
1109 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1110 "without setting discard operation to unmap");
1113 return detect_zeroes;
1117 * Set open flags for aio engine
1119 * Return 0 on success, -1 if the engine specified is invalid
1121 int bdrv_parse_aio(const char *mode, int *flags)
1123 if (!strcmp(mode, "threads")) {
1124 /* do nothing, default */
1125 } else if (!strcmp(mode, "native")) {
1126 *flags |= BDRV_O_NATIVE_AIO;
1127 #ifdef CONFIG_LINUX_IO_URING
1128 } else if (!strcmp(mode, "io_uring")) {
1129 *flags |= BDRV_O_IO_URING;
1139 * Set open flags for a given discard mode
1141 * Return 0 on success, -1 if the discard mode was invalid.
1143 int bdrv_parse_discard_flags(const char *mode, int *flags)
1145 *flags &= ~BDRV_O_UNMAP;
1147 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1149 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1150 *flags |= BDRV_O_UNMAP;
1159 * Set open flags for a given cache mode
1161 * Return 0 on success, -1 if the cache mode was invalid.
1163 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1165 *flags &= ~BDRV_O_CACHE_MASK;
1167 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1168 *writethrough = false;
1169 *flags |= BDRV_O_NOCACHE;
1170 } else if (!strcmp(mode, "directsync")) {
1171 *writethrough = true;
1172 *flags |= BDRV_O_NOCACHE;
1173 } else if (!strcmp(mode, "writeback")) {
1174 *writethrough = false;
1175 } else if (!strcmp(mode, "unsafe")) {
1176 *writethrough = false;
1177 *flags |= BDRV_O_NO_FLUSH;
1178 } else if (!strcmp(mode, "writethrough")) {
1179 *writethrough = true;
1187 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1189 BlockDriverState *parent = c->opaque;
1190 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1193 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1195 BlockDriverState *bs = child->opaque;
1196 bdrv_do_drained_begin_quiesce(bs, NULL);
1199 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1201 BlockDriverState *bs = child->opaque;
1202 return bdrv_drain_poll(bs, NULL, false);
1205 static void bdrv_child_cb_drained_end(BdrvChild *child)
1207 BlockDriverState *bs = child->opaque;
1208 bdrv_drained_end(bs);
1211 static int bdrv_child_cb_inactivate(BdrvChild *child)
1213 BlockDriverState *bs = child->opaque;
1214 GLOBAL_STATE_CODE();
1215 assert(bs->open_flags & BDRV_O_INACTIVE);
1219 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1220 GHashTable *visited, Transaction *tran,
1223 BlockDriverState *bs = child->opaque;
1224 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1228 * Returns the options and flags that a temporary snapshot should get, based on
1229 * the originally requested flags (the originally requested image will have
1230 * flags like a backing file)
1232 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1233 int parent_flags, QDict *parent_options)
1235 GLOBAL_STATE_CODE();
1236 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1238 /* For temporary files, unconditional cache=unsafe is fine */
1239 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1240 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1242 /* Copy the read-only and discard options from the parent */
1243 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1244 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1246 /* aio=native doesn't work for cache.direct=off, so disable it for the
1247 * temporary snapshot */
1248 *child_flags &= ~BDRV_O_NATIVE_AIO;
1251 static void bdrv_backing_attach(BdrvChild *c)
1253 BlockDriverState *parent = c->opaque;
1254 BlockDriverState *backing_hd = c->bs;
1256 GLOBAL_STATE_CODE();
1257 assert(!parent->backing_blocker);
1258 error_setg(&parent->backing_blocker,
1259 "node is used as backing hd of '%s'",
1260 bdrv_get_device_or_node_name(parent));
1262 bdrv_refresh_filename(backing_hd);
1264 parent->open_flags &= ~BDRV_O_NO_BACKING;
1266 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1267 /* Otherwise we won't be able to commit or stream */
1268 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1269 parent->backing_blocker);
1270 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1271 parent->backing_blocker);
1273 * We do backup in 3 ways:
1275 * The target bs is new opened, and the source is top BDS
1276 * 2. blockdev backup
1277 * Both the source and the target are top BDSes.
1278 * 3. internal backup(used for block replication)
1279 * Both the source and the target are backing file
1281 * In case 1 and 2, neither the source nor the target is the backing file.
1282 * In case 3, we will block the top BDS, so there is only one block job
1283 * for the top BDS and its backing chain.
1285 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1286 parent->backing_blocker);
1287 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1288 parent->backing_blocker);
1291 static void bdrv_backing_detach(BdrvChild *c)
1293 BlockDriverState *parent = c->opaque;
1295 GLOBAL_STATE_CODE();
1296 assert(parent->backing_blocker);
1297 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1298 error_free(parent->backing_blocker);
1299 parent->backing_blocker = NULL;
1302 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1303 const char *filename, Error **errp)
1305 BlockDriverState *parent = c->opaque;
1306 bool read_only = bdrv_is_read_only(parent);
1308 GLOBAL_STATE_CODE();
1311 ret = bdrv_reopen_set_read_only(parent, false, errp);
1317 ret = bdrv_change_backing_file(parent, filename,
1318 base->drv ? base->drv->format_name : "",
1321 error_setg_errno(errp, -ret, "Could not update backing file link");
1325 bdrv_reopen_set_read_only(parent, true, NULL);
1332 * Returns the options and flags that a generic child of a BDS should
1333 * get, based on the given options and flags for the parent BDS.
1335 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1336 int *child_flags, QDict *child_options,
1337 int parent_flags, QDict *parent_options)
1339 int flags = parent_flags;
1340 GLOBAL_STATE_CODE();
1343 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1344 * Generally, the question to answer is: Should this child be
1345 * format-probed by default?
1349 * Pure and non-filtered data children of non-format nodes should
1350 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1351 * set). This only affects a very limited set of drivers (namely
1352 * quorum and blkverify when this comment was written).
1353 * Force-clear BDRV_O_PROTOCOL then.
1355 if (!parent_is_format &&
1356 (role & BDRV_CHILD_DATA) &&
1357 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1359 flags &= ~BDRV_O_PROTOCOL;
1363 * All children of format nodes (except for COW children) and all
1364 * metadata children in general should never be format-probed.
1365 * Force-set BDRV_O_PROTOCOL then.
1367 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1368 (role & BDRV_CHILD_METADATA))
1370 flags |= BDRV_O_PROTOCOL;
1374 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1377 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1378 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1379 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1381 if (role & BDRV_CHILD_COW) {
1382 /* backing files are opened read-only by default */
1383 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1384 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1386 /* Inherit the read-only option from the parent if it's not set */
1387 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1388 qdict_copy_default(child_options, parent_options,
1389 BDRV_OPT_AUTO_READ_ONLY);
1393 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1394 * can default to enable it on lower layers regardless of the
1397 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1399 /* Clear flags that only apply to the top layer */
1400 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1402 if (role & BDRV_CHILD_METADATA) {
1403 flags &= ~BDRV_O_NO_IO;
1405 if (role & BDRV_CHILD_COW) {
1406 flags &= ~BDRV_O_TEMPORARY;
1409 *child_flags = flags;
1412 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1414 BlockDriverState *bs = child->opaque;
1416 assert_bdrv_graph_writable();
1417 QLIST_INSERT_HEAD(&bs->children, child, next);
1418 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1420 * Here we handle filters and block/raw-format.c when it behave like
1421 * filter. They generally have a single PRIMARY child, which is also the
1422 * FILTERED child, and that they may have multiple more children, which
1423 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1424 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1425 * into bs->backing on exceptional cases; and bs->backing will be
1428 assert(!(child->role & BDRV_CHILD_COW));
1429 if (child->role & BDRV_CHILD_PRIMARY) {
1430 assert(child->role & BDRV_CHILD_FILTERED);
1431 assert(!bs->backing);
1434 if (bs->drv->filtered_child_is_backing) {
1435 bs->backing = child;
1440 assert(!(child->role & BDRV_CHILD_FILTERED));
1442 } else if (child->role & BDRV_CHILD_COW) {
1443 assert(bs->drv->supports_backing);
1444 assert(!(child->role & BDRV_CHILD_PRIMARY));
1445 assert(!bs->backing);
1446 bs->backing = child;
1447 bdrv_backing_attach(child);
1448 } else if (child->role & BDRV_CHILD_PRIMARY) {
1454 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1456 BlockDriverState *bs = child->opaque;
1458 if (child->role & BDRV_CHILD_COW) {
1459 bdrv_backing_detach(child);
1462 assert_bdrv_graph_writable();
1463 QLIST_REMOVE(child, next);
1464 if (child == bs->backing) {
1465 assert(child != bs->file);
1467 } else if (child == bs->file) {
1472 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1473 const char *filename, Error **errp)
1475 if (c->role & BDRV_CHILD_COW) {
1476 return bdrv_backing_update_filename(c, base, filename, errp);
1481 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1483 BlockDriverState *bs = c->opaque;
1486 return bdrv_get_aio_context(bs);
1489 const BdrvChildClass child_of_bds = {
1490 .parent_is_bds = true,
1491 .get_parent_desc = bdrv_child_get_parent_desc,
1492 .inherit_options = bdrv_inherited_options,
1493 .drained_begin = bdrv_child_cb_drained_begin,
1494 .drained_poll = bdrv_child_cb_drained_poll,
1495 .drained_end = bdrv_child_cb_drained_end,
1496 .attach = bdrv_child_cb_attach,
1497 .detach = bdrv_child_cb_detach,
1498 .inactivate = bdrv_child_cb_inactivate,
1499 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1500 .update_filename = bdrv_child_cb_update_filename,
1501 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1504 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1507 return c->klass->get_parent_aio_context(c);
1510 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1512 int open_flags = flags;
1513 GLOBAL_STATE_CODE();
1516 * Clear flags that are internal to the block layer before opening the
1519 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1524 static void update_flags_from_options(int *flags, QemuOpts *opts)
1526 GLOBAL_STATE_CODE();
1528 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1530 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1531 *flags |= BDRV_O_NO_FLUSH;
1534 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1535 *flags |= BDRV_O_NOCACHE;
1538 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1539 *flags |= BDRV_O_RDWR;
1542 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1543 *flags |= BDRV_O_AUTO_RDONLY;
1547 static void update_options_from_flags(QDict *options, int flags)
1549 GLOBAL_STATE_CODE();
1550 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1551 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1553 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1554 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1555 flags & BDRV_O_NO_FLUSH);
1557 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1558 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1560 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1561 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1562 flags & BDRV_O_AUTO_RDONLY);
1566 static void bdrv_assign_node_name(BlockDriverState *bs,
1567 const char *node_name,
1570 char *gen_node_name = NULL;
1571 GLOBAL_STATE_CODE();
1574 node_name = gen_node_name = id_generate(ID_BLOCK);
1575 } else if (!id_wellformed(node_name)) {
1577 * Check for empty string or invalid characters, but not if it is
1578 * generated (generated names use characters not available to the user)
1580 error_setg(errp, "Invalid node-name: '%s'", node_name);
1584 /* takes care of avoiding namespaces collisions */
1585 if (blk_by_name(node_name)) {
1586 error_setg(errp, "node-name=%s is conflicting with a device id",
1591 /* takes care of avoiding duplicates node names */
1592 if (bdrv_find_node(node_name)) {
1593 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1597 /* Make sure that the node name isn't truncated */
1598 if (strlen(node_name) >= sizeof(bs->node_name)) {
1599 error_setg(errp, "Node name too long");
1603 /* copy node name into the bs and insert it into the graph list */
1604 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1605 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1607 g_free(gen_node_name);
1611 * The caller must always hold @bs AioContext lock, because this function calls
1612 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1615 static int no_coroutine_fn GRAPH_UNLOCKED
1616 bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1617 QDict *options, int open_flags, Error **errp)
1620 Error *local_err = NULL;
1622 GLOBAL_STATE_CODE();
1624 bdrv_assign_node_name(bs, node_name, &local_err);
1626 error_propagate(errp, local_err);
1631 bs->opaque = g_malloc0(drv->instance_size);
1633 if (drv->bdrv_file_open) {
1634 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1635 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1636 } else if (drv->bdrv_open) {
1637 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1644 error_propagate(errp, local_err);
1645 } else if (bs->filename[0]) {
1646 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1648 error_setg_errno(errp, -ret, "Could not open image");
1653 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1654 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1657 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1658 * drivers that pass read/write requests through to a child the trouble of
1659 * declaring support explicitly.
1661 * Drivers must not propagate this flag accidentally when they initiate I/O
1662 * to a bounce buffer. That case should be rare though.
1664 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1665 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1667 /* Get the context after .bdrv_open, it can change the context */
1668 ctx = bdrv_get_aio_context(bs);
1669 aio_context_acquire(ctx);
1671 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
1673 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1674 aio_context_release(ctx);
1678 bdrv_graph_rdlock_main_loop();
1679 bdrv_refresh_limits(bs, NULL, &local_err);
1680 bdrv_graph_rdunlock_main_loop();
1681 aio_context_release(ctx);
1684 error_propagate(errp, local_err);
1688 assert(bdrv_opt_mem_align(bs) != 0);
1689 assert(bdrv_min_mem_align(bs) != 0);
1690 assert(is_power_of_2(bs->bl.request_alignment));
1692 for (i = 0; i < bs->quiesce_counter; i++) {
1693 if (drv->bdrv_drain_begin) {
1694 drv->bdrv_drain_begin(bs);
1701 if (bs->file != NULL) {
1702 bdrv_unref_child(bs, bs->file);
1711 * Create and open a block node.
1713 * @options is a QDict of options to pass to the block drivers, or NULL for an
1714 * empty set of options. The reference to the QDict belongs to the block layer
1715 * after the call (even on failure), so if the caller intends to reuse the
1716 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1718 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1719 const char *node_name,
1720 QDict *options, int flags,
1723 BlockDriverState *bs;
1726 GLOBAL_STATE_CODE();
1729 bs->open_flags = flags;
1730 bs->options = options ?: qdict_new();
1731 bs->explicit_options = qdict_clone_shallow(bs->options);
1734 update_options_from_flags(bs->options, flags);
1736 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1738 qobject_unref(bs->explicit_options);
1739 bs->explicit_options = NULL;
1740 qobject_unref(bs->options);
1749 /* Create and open a block node. */
1750 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1751 int flags, Error **errp)
1753 GLOBAL_STATE_CODE();
1754 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1757 QemuOptsList bdrv_runtime_opts = {
1758 .name = "bdrv_common",
1759 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1762 .name = "node-name",
1763 .type = QEMU_OPT_STRING,
1764 .help = "Node name of the block device node",
1768 .type = QEMU_OPT_STRING,
1769 .help = "Block driver to use for the node",
1772 .name = BDRV_OPT_CACHE_DIRECT,
1773 .type = QEMU_OPT_BOOL,
1774 .help = "Bypass software writeback cache on the host",
1777 .name = BDRV_OPT_CACHE_NO_FLUSH,
1778 .type = QEMU_OPT_BOOL,
1779 .help = "Ignore flush requests",
1782 .name = BDRV_OPT_READ_ONLY,
1783 .type = QEMU_OPT_BOOL,
1784 .help = "Node is opened in read-only mode",
1787 .name = BDRV_OPT_AUTO_READ_ONLY,
1788 .type = QEMU_OPT_BOOL,
1789 .help = "Node can become read-only if opening read-write fails",
1792 .name = "detect-zeroes",
1793 .type = QEMU_OPT_STRING,
1794 .help = "try to optimize zero writes (off, on, unmap)",
1797 .name = BDRV_OPT_DISCARD,
1798 .type = QEMU_OPT_STRING,
1799 .help = "discard operation (ignore/off, unmap/on)",
1802 .name = BDRV_OPT_FORCE_SHARE,
1803 .type = QEMU_OPT_BOOL,
1804 .help = "always accept other writers (default: off)",
1806 { /* end of list */ }
1810 QemuOptsList bdrv_create_opts_simple = {
1811 .name = "simple-create-opts",
1812 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1815 .name = BLOCK_OPT_SIZE,
1816 .type = QEMU_OPT_SIZE,
1817 .help = "Virtual disk size"
1820 .name = BLOCK_OPT_PREALLOC,
1821 .type = QEMU_OPT_STRING,
1822 .help = "Preallocation mode (allowed values: off)"
1824 { /* end of list */ }
1829 * Common part for opening disk images and files
1831 * Removes all processed options from *options.
1833 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1834 QDict *options, Error **errp)
1836 int ret, open_flags;
1837 const char *filename;
1838 const char *driver_name = NULL;
1839 const char *node_name = NULL;
1840 const char *discard;
1843 Error *local_err = NULL;
1846 assert(bs->file == NULL);
1847 assert(options != NULL && bs->options != options);
1848 GLOBAL_STATE_CODE();
1850 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1851 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1856 update_flags_from_options(&bs->open_flags, opts);
1858 driver_name = qemu_opt_get(opts, "driver");
1859 drv = bdrv_find_format(driver_name);
1860 assert(drv != NULL);
1862 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1864 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1866 BDRV_OPT_FORCE_SHARE
1867 "=on can only be used with read-only images");
1873 bdrv_refresh_filename(blk_bs(file));
1874 filename = blk_bs(file)->filename;
1877 * Caution: while qdict_get_try_str() is fine, getting
1878 * non-string types would require more care. When @options
1879 * come from -blockdev or blockdev_add, its members are typed
1880 * according to the QAPI schema, but when they come from
1881 * -drive, they're all QString.
1883 filename = qdict_get_try_str(options, "filename");
1886 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1887 error_setg(errp, "The '%s' block driver requires a file name",
1893 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1896 ro = bdrv_is_read_only(bs);
1898 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1899 if (!ro && bdrv_is_whitelisted(drv, true)) {
1900 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1906 !ro && bdrv_is_whitelisted(drv, true)
1907 ? "Driver '%s' can only be used for read-only devices"
1908 : "Driver '%s' is not whitelisted",
1914 /* bdrv_new() and bdrv_close() make it so */
1915 assert(qatomic_read(&bs->copy_on_read) == 0);
1917 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1919 bdrv_enable_copy_on_read(bs);
1921 error_setg(errp, "Can't use copy-on-read on read-only device");
1927 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1928 if (discard != NULL) {
1929 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1930 error_setg(errp, "Invalid discard option");
1937 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1939 error_propagate(errp, local_err);
1944 if (filename != NULL) {
1945 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1947 bs->filename[0] = '\0';
1949 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1951 /* Open the image, either directly or using a protocol */
1952 open_flags = bdrv_open_flags(bs, bs->open_flags);
1953 node_name = qemu_opt_get(opts, "node-name");
1955 assert(!drv->bdrv_file_open || file == NULL);
1956 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1961 qemu_opts_del(opts);
1965 qemu_opts_del(opts);
1969 static QDict *parse_json_filename(const char *filename, Error **errp)
1971 QObject *options_obj;
1974 GLOBAL_STATE_CODE();
1976 ret = strstart(filename, "json:", &filename);
1979 options_obj = qobject_from_json(filename, errp);
1981 error_prepend(errp, "Could not parse the JSON options: ");
1985 options = qobject_to(QDict, options_obj);
1987 qobject_unref(options_obj);
1988 error_setg(errp, "Invalid JSON object given");
1992 qdict_flatten(options);
1997 static void parse_json_protocol(QDict *options, const char **pfilename,
2000 QDict *json_options;
2001 Error *local_err = NULL;
2002 GLOBAL_STATE_CODE();
2004 /* Parse json: pseudo-protocol */
2005 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2009 json_options = parse_json_filename(*pfilename, &local_err);
2011 error_propagate(errp, local_err);
2015 /* Options given in the filename have lower priority than options
2016 * specified directly */
2017 qdict_join(options, json_options, false);
2018 qobject_unref(json_options);
2023 * Fills in default options for opening images and converts the legacy
2024 * filename/flags pair to option QDict entries.
2025 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2026 * block driver has been specified explicitly.
2028 static int bdrv_fill_options(QDict **options, const char *filename,
2029 int *flags, Error **errp)
2031 const char *drvname;
2032 bool protocol = *flags & BDRV_O_PROTOCOL;
2033 bool parse_filename = false;
2034 BlockDriver *drv = NULL;
2035 Error *local_err = NULL;
2037 GLOBAL_STATE_CODE();
2040 * Caution: while qdict_get_try_str() is fine, getting non-string
2041 * types would require more care. When @options come from
2042 * -blockdev or blockdev_add, its members are typed according to
2043 * the QAPI schema, but when they come from -drive, they're all
2046 drvname = qdict_get_try_str(*options, "driver");
2048 drv = bdrv_find_format(drvname);
2050 error_setg(errp, "Unknown driver '%s'", drvname);
2053 /* If the user has explicitly specified the driver, this choice should
2054 * override the BDRV_O_PROTOCOL flag */
2055 protocol = drv->bdrv_file_open;
2059 *flags |= BDRV_O_PROTOCOL;
2061 *flags &= ~BDRV_O_PROTOCOL;
2064 /* Translate cache options from flags into options */
2065 update_options_from_flags(*options, *flags);
2067 /* Fetch the file name from the options QDict if necessary */
2068 if (protocol && filename) {
2069 if (!qdict_haskey(*options, "filename")) {
2070 qdict_put_str(*options, "filename", filename);
2071 parse_filename = true;
2073 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2079 /* Find the right block driver */
2080 /* See cautionary note on accessing @options above */
2081 filename = qdict_get_try_str(*options, "filename");
2083 if (!drvname && protocol) {
2085 drv = bdrv_find_protocol(filename, parse_filename, errp);
2090 drvname = drv->format_name;
2091 qdict_put_str(*options, "driver", drvname);
2093 error_setg(errp, "Must specify either driver or file");
2098 assert(drv || !protocol);
2100 /* Driver-specific filename parsing */
2101 if (drv && drv->bdrv_parse_filename && parse_filename) {
2102 drv->bdrv_parse_filename(filename, *options, &local_err);
2104 error_propagate(errp, local_err);
2108 if (!drv->bdrv_needs_filename) {
2109 qdict_del(*options, "filename");
2116 typedef struct BlockReopenQueueEntry {
2118 BDRVReopenState state;
2119 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2120 } BlockReopenQueueEntry;
2123 * Return the flags that @bs will have after the reopens in @q have
2124 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2125 * return the current flags.
2127 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2129 BlockReopenQueueEntry *entry;
2132 QTAILQ_FOREACH(entry, q, entry) {
2133 if (entry->state.bs == bs) {
2134 return entry->state.flags;
2139 return bs->open_flags;
2142 /* Returns whether the image file can be written to after the reopen queue @q
2143 * has been successfully applied, or right now if @q is NULL. */
2144 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2145 BlockReopenQueue *q)
2147 int flags = bdrv_reopen_get_flags(q, bs);
2149 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2153 * Return whether the BDS can be written to. This is not necessarily
2154 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2155 * be written to but do not count as read-only images.
2157 bool bdrv_is_writable(BlockDriverState *bs)
2160 return bdrv_is_writable_after_reopen(bs, NULL);
2163 static char *bdrv_child_user_desc(BdrvChild *c)
2165 GLOBAL_STATE_CODE();
2166 return c->klass->get_parent_desc(c);
2170 * Check that @a allows everything that @b needs. @a and @b must reference same
2173 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2175 const char *child_bs_name;
2176 g_autofree char *a_user = NULL;
2177 g_autofree char *b_user = NULL;
2178 g_autofree char *perms = NULL;
2181 assert(a->bs == b->bs);
2182 GLOBAL_STATE_CODE();
2184 if ((b->perm & a->shared_perm) == b->perm) {
2188 child_bs_name = bdrv_get_node_name(b->bs);
2189 a_user = bdrv_child_user_desc(a);
2190 b_user = bdrv_child_user_desc(b);
2191 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2193 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2194 "both required by %s (uses node '%s' as '%s' child) and "
2195 "unshared by %s (uses node '%s' as '%s' child).",
2196 child_bs_name, perms,
2197 b_user, child_bs_name, b->name,
2198 a_user, child_bs_name, a->name);
2203 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2206 GLOBAL_STATE_CODE();
2209 * During the loop we'll look at each pair twice. That's correct because
2210 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2213 QLIST_FOREACH(a, &bs->parents, next_parent) {
2214 QLIST_FOREACH(b, &bs->parents, next_parent) {
2219 if (!bdrv_a_allow_b(a, b, errp)) {
2228 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2229 BdrvChild *c, BdrvChildRole role,
2230 BlockReopenQueue *reopen_queue,
2231 uint64_t parent_perm, uint64_t parent_shared,
2232 uint64_t *nperm, uint64_t *nshared)
2234 assert(bs->drv && bs->drv->bdrv_child_perm);
2235 GLOBAL_STATE_CODE();
2236 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2237 parent_perm, parent_shared,
2239 /* TODO Take force_share from reopen_queue */
2240 if (child_bs && child_bs->force_share) {
2241 *nshared = BLK_PERM_ALL;
2246 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2247 * nodes that are already in the @list, of course) so that final list is
2248 * topologically sorted. Return the result (GSList @list object is updated, so
2249 * don't use old reference after function call).
2251 * On function start @list must be already topologically sorted and for any node
2252 * in the @list the whole subtree of the node must be in the @list as well. The
2253 * simplest way to satisfy this criteria: use only result of
2254 * bdrv_topological_dfs() or NULL as @list parameter.
2256 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2257 BlockDriverState *bs)
2260 g_autoptr(GHashTable) local_found = NULL;
2262 GLOBAL_STATE_CODE();
2266 found = local_found = g_hash_table_new(NULL, NULL);
2269 if (g_hash_table_contains(found, bs)) {
2272 g_hash_table_add(found, bs);
2274 QLIST_FOREACH(child, &bs->children, next) {
2275 list = bdrv_topological_dfs(list, found, child->bs);
2278 return g_slist_prepend(list, bs);
2281 typedef struct BdrvChildSetPermState {
2284 uint64_t old_shared_perm;
2285 } BdrvChildSetPermState;
2287 static void bdrv_child_set_perm_abort(void *opaque)
2289 BdrvChildSetPermState *s = opaque;
2291 GLOBAL_STATE_CODE();
2293 s->child->perm = s->old_perm;
2294 s->child->shared_perm = s->old_shared_perm;
2297 static TransactionActionDrv bdrv_child_set_pem_drv = {
2298 .abort = bdrv_child_set_perm_abort,
2302 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2303 uint64_t shared, Transaction *tran)
2305 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2306 GLOBAL_STATE_CODE();
2308 *s = (BdrvChildSetPermState) {
2310 .old_perm = c->perm,
2311 .old_shared_perm = c->shared_perm,
2315 c->shared_perm = shared;
2317 tran_add(tran, &bdrv_child_set_pem_drv, s);
2320 static void bdrv_drv_set_perm_commit(void *opaque)
2322 BlockDriverState *bs = opaque;
2323 uint64_t cumulative_perms, cumulative_shared_perms;
2324 GLOBAL_STATE_CODE();
2326 if (bs->drv->bdrv_set_perm) {
2327 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2328 &cumulative_shared_perms);
2329 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2333 static void bdrv_drv_set_perm_abort(void *opaque)
2335 BlockDriverState *bs = opaque;
2336 GLOBAL_STATE_CODE();
2338 if (bs->drv->bdrv_abort_perm_update) {
2339 bs->drv->bdrv_abort_perm_update(bs);
2343 TransactionActionDrv bdrv_drv_set_perm_drv = {
2344 .abort = bdrv_drv_set_perm_abort,
2345 .commit = bdrv_drv_set_perm_commit,
2348 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2349 uint64_t shared_perm, Transaction *tran,
2352 GLOBAL_STATE_CODE();
2357 if (bs->drv->bdrv_check_perm) {
2358 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2365 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2371 typedef struct BdrvReplaceChildState {
2373 BlockDriverState *old_bs;
2374 } BdrvReplaceChildState;
2376 static void bdrv_replace_child_commit(void *opaque)
2378 BdrvReplaceChildState *s = opaque;
2379 GLOBAL_STATE_CODE();
2381 bdrv_unref(s->old_bs);
2384 static void bdrv_replace_child_abort(void *opaque)
2386 BdrvReplaceChildState *s = opaque;
2387 BlockDriverState *new_bs = s->child->bs;
2389 GLOBAL_STATE_CODE();
2390 /* old_bs reference is transparently moved from @s to @s->child */
2391 if (!s->child->bs) {
2393 * The parents were undrained when removing old_bs from the child. New
2394 * requests can't have been made, though, because the child was empty.
2396 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2397 * undraining the parent in the first place. Once this is done, having
2398 * new_bs drained when calling bdrv_replace_child_tran() is not a
2399 * requirement any more.
2401 bdrv_parent_drained_begin_single(s->child);
2402 assert(!bdrv_parent_drained_poll_single(s->child));
2404 assert(s->child->quiesced_parent);
2405 bdrv_replace_child_noperm(s->child, s->old_bs);
2409 static TransactionActionDrv bdrv_replace_child_drv = {
2410 .commit = bdrv_replace_child_commit,
2411 .abort = bdrv_replace_child_abort,
2416 * bdrv_replace_child_tran
2418 * Note: real unref of old_bs is done only on commit.
2420 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2421 * kept drained until the transaction is completed.
2423 * The function doesn't update permissions, caller is responsible for this.
2425 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2428 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2430 assert(child->quiesced_parent);
2431 assert(!new_bs || new_bs->quiesce_counter);
2433 *s = (BdrvReplaceChildState) {
2435 .old_bs = child->bs,
2437 tran_add(tran, &bdrv_replace_child_drv, s);
2442 bdrv_replace_child_noperm(child, new_bs);
2443 /* old_bs reference is transparently moved from @child to @s */
2447 * Refresh permissions in @bs subtree. The function is intended to be called
2448 * after some graph modification that was done without permission update.
2450 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2451 Transaction *tran, Error **errp)
2453 BlockDriver *drv = bs->drv;
2456 uint64_t cumulative_perms, cumulative_shared_perms;
2457 GLOBAL_STATE_CODE();
2459 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2461 /* Write permissions never work with read-only images */
2462 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2463 !bdrv_is_writable_after_reopen(bs, q))
2465 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2466 error_setg(errp, "Block node is read-only");
2468 error_setg(errp, "Read-only block node '%s' cannot support "
2469 "read-write users", bdrv_get_node_name(bs));
2476 * Unaligned requests will automatically be aligned to bl.request_alignment
2477 * and without RESIZE we can't extend requests to write to space beyond the
2478 * end of the image, so it's required that the image size is aligned.
2480 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2481 !(cumulative_perms & BLK_PERM_RESIZE))
2483 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2484 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2485 "Image size is not a multiple of request "
2491 /* Check this node */
2496 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2502 /* Drivers that never have children can omit .bdrv_child_perm() */
2503 if (!drv->bdrv_child_perm) {
2504 assert(QLIST_EMPTY(&bs->children));
2508 /* Check all children */
2509 QLIST_FOREACH(c, &bs->children, next) {
2510 uint64_t cur_perm, cur_shared;
2512 bdrv_child_perm(bs, c->bs, c, c->role, q,
2513 cumulative_perms, cumulative_shared_perms,
2514 &cur_perm, &cur_shared);
2515 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2522 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2523 * a topologically sorted subgraph.
2525 static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2526 Transaction *tran, Error **errp)
2529 BlockDriverState *bs;
2530 GLOBAL_STATE_CODE();
2532 for ( ; list; list = list->next) {
2535 if (bdrv_parent_perms_conflict(bs, errp)) {
2539 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2549 * @list is any list of nodes. List is completed by all subtrees and
2550 * topologically sorted. It's not a problem if some node occurs in the @list
2553 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2554 Transaction *tran, Error **errp)
2556 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2557 g_autoptr(GSList) refresh_list = NULL;
2559 for ( ; list; list = list->next) {
2560 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2563 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2566 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2567 uint64_t *shared_perm)
2570 uint64_t cumulative_perms = 0;
2571 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2573 GLOBAL_STATE_CODE();
2575 QLIST_FOREACH(c, &bs->parents, next_parent) {
2576 cumulative_perms |= c->perm;
2577 cumulative_shared_perms &= c->shared_perm;
2580 *perm = cumulative_perms;
2581 *shared_perm = cumulative_shared_perms;
2584 char *bdrv_perm_names(uint64_t perm)
2590 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2591 { BLK_PERM_WRITE, "write" },
2592 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2593 { BLK_PERM_RESIZE, "resize" },
2597 GString *result = g_string_sized_new(30);
2598 struct perm_name *p;
2600 for (p = permissions; p->name; p++) {
2601 if (perm & p->perm) {
2602 if (result->len > 0) {
2603 g_string_append(result, ", ");
2605 g_string_append(result, p->name);
2609 return g_string_free(result, FALSE);
2613 /* @tran is allowed to be NULL. In this case no rollback is possible */
2614 static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2618 Transaction *local_tran = NULL;
2619 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2620 GLOBAL_STATE_CODE();
2623 tran = local_tran = tran_new();
2626 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2629 tran_finalize(local_tran, ret);
2635 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2638 Error *local_err = NULL;
2639 Transaction *tran = tran_new();
2642 GLOBAL_STATE_CODE();
2644 bdrv_child_set_perm(c, perm, shared, tran);
2646 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2648 tran_finalize(tran, ret);
2651 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2652 /* tighten permissions */
2653 error_propagate(errp, local_err);
2656 * Our caller may intend to only loosen restrictions and
2657 * does not expect this function to fail. Errors are not
2658 * fatal in such a case, so we can just hide them from our
2661 error_free(local_err);
2669 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2671 uint64_t parent_perms, parent_shared;
2672 uint64_t perms, shared;
2674 GLOBAL_STATE_CODE();
2676 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2677 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2678 parent_perms, parent_shared, &perms, &shared);
2680 return bdrv_child_try_set_perm(c, perms, shared, errp);
2684 * Default implementation for .bdrv_child_perm() for block filters:
2685 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2688 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2690 BlockReopenQueue *reopen_queue,
2691 uint64_t perm, uint64_t shared,
2692 uint64_t *nperm, uint64_t *nshared)
2694 GLOBAL_STATE_CODE();
2695 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2696 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2699 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2701 BlockReopenQueue *reopen_queue,
2702 uint64_t perm, uint64_t shared,
2703 uint64_t *nperm, uint64_t *nshared)
2705 assert(role & BDRV_CHILD_COW);
2706 GLOBAL_STATE_CODE();
2709 * We want consistent read from backing files if the parent needs it.
2710 * No other operations are performed on backing files.
2712 perm &= BLK_PERM_CONSISTENT_READ;
2715 * If the parent can deal with changing data, we're okay with a
2716 * writable and resizable backing file.
2717 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2719 if (shared & BLK_PERM_WRITE) {
2720 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2725 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2727 if (bs->open_flags & BDRV_O_INACTIVE) {
2728 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2735 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2737 BlockReopenQueue *reopen_queue,
2738 uint64_t perm, uint64_t shared,
2739 uint64_t *nperm, uint64_t *nshared)
2743 GLOBAL_STATE_CODE();
2744 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2746 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2749 * Apart from the modifications below, the same permissions are
2750 * forwarded and left alone as for filters
2752 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2753 perm, shared, &perm, &shared);
2755 if (role & BDRV_CHILD_METADATA) {
2756 /* Format drivers may touch metadata even if the guest doesn't write */
2757 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2758 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2762 * bs->file always needs to be consistent because of the
2763 * metadata. We can never allow other users to resize or write
2766 if (!(flags & BDRV_O_NO_IO)) {
2767 perm |= BLK_PERM_CONSISTENT_READ;
2769 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2772 if (role & BDRV_CHILD_DATA) {
2774 * Technically, everything in this block is a subset of the
2775 * BDRV_CHILD_METADATA path taken above, and so this could
2776 * be an "else if" branch. However, that is not obvious, and
2777 * this function is not performance critical, therefore we let
2778 * this be an independent "if".
2782 * We cannot allow other users to resize the file because the
2783 * format driver might have some assumptions about the size
2784 * (e.g. because it is stored in metadata, or because the file
2785 * is split into fixed-size data files).
2787 shared &= ~BLK_PERM_RESIZE;
2790 * WRITE_UNCHANGED often cannot be performed as such on the
2791 * data file. For example, the qcow2 driver may still need to
2792 * write copied clusters on copy-on-read.
2794 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2795 perm |= BLK_PERM_WRITE;
2799 * If the data file is written to, the format driver may
2800 * expect to be able to resize it by writing beyond the EOF.
2802 if (perm & BLK_PERM_WRITE) {
2803 perm |= BLK_PERM_RESIZE;
2807 if (bs->open_flags & BDRV_O_INACTIVE) {
2808 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2815 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2816 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2817 uint64_t perm, uint64_t shared,
2818 uint64_t *nperm, uint64_t *nshared)
2820 GLOBAL_STATE_CODE();
2821 if (role & BDRV_CHILD_FILTERED) {
2822 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2824 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2825 perm, shared, nperm, nshared);
2826 } else if (role & BDRV_CHILD_COW) {
2827 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2828 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2829 perm, shared, nperm, nshared);
2830 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2831 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2832 perm, shared, nperm, nshared);
2834 g_assert_not_reached();
2838 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2840 static const uint64_t permissions[] = {
2841 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2842 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2843 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2844 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2847 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2848 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2850 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2852 return permissions[qapi_perm];
2856 * Replaces the node that a BdrvChild points to without updating permissions.
2858 * If @new_bs is non-NULL, the parent of @child must already be drained through
2859 * @child and the caller must hold the AioContext lock for @new_bs.
2861 static void bdrv_replace_child_noperm(BdrvChild *child,
2862 BlockDriverState *new_bs)
2864 BlockDriverState *old_bs = child->bs;
2865 int new_bs_quiesce_counter;
2867 assert(!child->frozen);
2870 * If we want to change the BdrvChild to point to a drained node as its new
2871 * child->bs, we need to make sure that its new parent is drained, too. In
2872 * other words, either child->quiesce_parent must already be true or we must
2873 * be able to set it and keep the parent's quiesce_counter consistent with
2874 * that, but without polling or starting new requests (this function
2875 * guarantees that it doesn't poll, and starting new requests would be
2876 * against the invariants of drain sections).
2878 * To keep things simple, we pick the first option (child->quiesce_parent
2879 * must already be true). We also generalise the rule a bit to make it
2880 * easier to verify in callers and more likely to be covered in test cases:
2881 * The parent must be quiesced through this child even if new_bs isn't
2882 * currently drained.
2884 * The only exception is for callers that always pass new_bs == NULL. In
2885 * this case, we obviously never need to consider the case of a drained
2886 * new_bs, so we can keep the callers simpler by allowing them not to drain
2889 assert(!new_bs || child->quiesced_parent);
2890 assert(old_bs != new_bs);
2891 GLOBAL_STATE_CODE();
2893 if (old_bs && new_bs) {
2894 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2897 /* TODO Pull this up into the callers to avoid polling here */
2898 bdrv_graph_wrlock(new_bs);
2900 if (child->klass->detach) {
2901 child->klass->detach(child);
2903 QLIST_REMOVE(child, next_parent);
2909 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2910 if (child->klass->attach) {
2911 child->klass->attach(child);
2914 bdrv_graph_wrunlock();
2917 * If the parent was drained through this BdrvChild previously, but new_bs
2918 * is not drained, allow requests to come in only after the new node has
2921 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2922 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2923 bdrv_parent_drained_end_single(child);
2928 * Free the given @child.
2930 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2931 * unused (i.e. not in a children list).
2933 static void bdrv_child_free(BdrvChild *child)
2936 GLOBAL_STATE_CODE();
2937 assert(!child->next.le_prev); /* not in children list */
2939 g_free(child->name);
2943 typedef struct BdrvAttachChildCommonState {
2945 AioContext *old_parent_ctx;
2946 AioContext *old_child_ctx;
2947 } BdrvAttachChildCommonState;
2949 static void bdrv_attach_child_common_abort(void *opaque)
2951 BdrvAttachChildCommonState *s = opaque;
2952 BlockDriverState *bs = s->child->bs;
2954 GLOBAL_STATE_CODE();
2955 bdrv_replace_child_noperm(s->child, NULL);
2957 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2958 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2961 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2963 GHashTable *visited;
2968 /* No need to visit `child`, because it has been detached already */
2969 visited = g_hash_table_new(NULL, NULL);
2970 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2971 visited, tran, &error_abort);
2972 g_hash_table_destroy(visited);
2974 /* transaction is supposed to always succeed */
2975 assert(ret == true);
2980 bdrv_child_free(s->child);
2983 static TransactionActionDrv bdrv_attach_child_common_drv = {
2984 .abort = bdrv_attach_child_common_abort,
2989 * Common part of attaching bdrv child to bs or to blk or to job
2991 * Function doesn't update permissions, caller is responsible for this.
2993 * Returns new created child.
2995 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
2996 * @child_bs can move to a different AioContext in this function. Callers must
2997 * make sure that their AioContext locking is still correct after this.
2999 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
3000 const char *child_name,
3001 const BdrvChildClass *child_class,
3002 BdrvChildRole child_role,
3003 uint64_t perm, uint64_t shared_perm,
3005 Transaction *tran, Error **errp)
3007 BdrvChild *new_child;
3008 AioContext *parent_ctx, *new_child_ctx;
3009 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3011 assert(child_class->get_parent_desc);
3012 GLOBAL_STATE_CODE();
3014 new_child = g_new(BdrvChild, 1);
3015 *new_child = (BdrvChild) {
3017 .name = g_strdup(child_name),
3018 .klass = child_class,
3021 .shared_perm = shared_perm,
3026 * If the AioContexts don't match, first try to move the subtree of
3027 * child_bs into the AioContext of the new parent. If this doesn't work,
3028 * try moving the parent into the AioContext of child_bs instead.
3030 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3031 if (child_ctx != parent_ctx) {
3032 Error *local_err = NULL;
3033 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3036 if (ret < 0 && child_class->change_aio_ctx) {
3037 Transaction *tran = tran_new();
3038 GHashTable *visited = g_hash_table_new(NULL, NULL);
3041 g_hash_table_add(visited, new_child);
3042 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3043 visited, tran, NULL);
3044 if (ret_child == true) {
3045 error_free(local_err);
3048 tran_finalize(tran, ret_child == true ? 0 : -1);
3049 g_hash_table_destroy(visited);
3053 error_propagate(errp, local_err);
3054 bdrv_child_free(new_child);
3059 new_child_ctx = bdrv_get_aio_context(child_bs);
3060 if (new_child_ctx != child_ctx) {
3061 aio_context_release(child_ctx);
3062 aio_context_acquire(new_child_ctx);
3067 * Let every new BdrvChild start with a drained parent. Inserting the child
3068 * in the graph with bdrv_replace_child_noperm() will undrain it if
3069 * @child_bs is not drained.
3071 * The child was only just created and is not yet visible in global state
3072 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3073 * could have sent requests and polling is not necessary.
3075 * Note that this means that the parent isn't fully drained yet, we only
3076 * stop new requests from coming in. This is fine, we don't care about the
3077 * old requests here, they are not for this child. If another place enters a
3078 * drain section for the same parent, but wants it to be fully quiesced, it
3079 * will not run most of the the code in .drained_begin() again (which is not
3080 * a problem, we already did this), but it will still poll until the parent
3081 * is fully quiesced, so it will not be negatively affected either.
3083 bdrv_parent_drained_begin_single(new_child);
3084 bdrv_replace_child_noperm(new_child, child_bs);
3086 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3087 *s = (BdrvAttachChildCommonState) {
3089 .old_parent_ctx = parent_ctx,
3090 .old_child_ctx = child_ctx,
3092 tran_add(tran, &bdrv_attach_child_common_drv, s);
3094 if (new_child_ctx != child_ctx) {
3095 aio_context_release(new_child_ctx);
3096 aio_context_acquire(child_ctx);
3103 * Function doesn't update permissions, caller is responsible for this.
3105 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3106 * @child_bs can move to a different AioContext in this function. Callers must
3107 * make sure that their AioContext locking is still correct after this.
3109 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3110 BlockDriverState *child_bs,
3111 const char *child_name,
3112 const BdrvChildClass *child_class,
3113 BdrvChildRole child_role,
3117 uint64_t perm, shared_perm;
3119 assert(parent_bs->drv);
3120 GLOBAL_STATE_CODE();
3122 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3123 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3124 child_bs->node_name, child_name, parent_bs->node_name);
3128 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3129 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3130 perm, shared_perm, &perm, &shared_perm);
3132 return bdrv_attach_child_common(child_bs, child_name, child_class,
3133 child_role, perm, shared_perm, parent_bs,
3138 * This function steals the reference to child_bs from the caller.
3139 * That reference is later dropped by bdrv_root_unref_child().
3141 * On failure NULL is returned, errp is set and the reference to
3142 * child_bs is also dropped.
3144 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3145 * (unless @child_bs is already in @ctx).
3147 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3148 const char *child_name,
3149 const BdrvChildClass *child_class,
3150 BdrvChildRole child_role,
3151 uint64_t perm, uint64_t shared_perm,
3152 void *opaque, Error **errp)
3156 Transaction *tran = tran_new();
3158 GLOBAL_STATE_CODE();
3160 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3161 child_role, perm, shared_perm, opaque,
3168 ret = bdrv_refresh_perms(child_bs, tran, errp);
3171 tran_finalize(tran, ret);
3173 bdrv_unref(child_bs);
3175 return ret < 0 ? NULL : child;
3179 * This function transfers the reference to child_bs from the caller
3180 * to parent_bs. That reference is later dropped by parent_bs on
3181 * bdrv_close() or if someone calls bdrv_unref_child().
3183 * On failure NULL is returned, errp is set and the reference to
3184 * child_bs is also dropped.
3186 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3187 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3189 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3190 BlockDriverState *child_bs,
3191 const char *child_name,
3192 const BdrvChildClass *child_class,
3193 BdrvChildRole child_role,
3198 Transaction *tran = tran_new();
3200 GLOBAL_STATE_CODE();
3202 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3203 child_class, child_role, tran, errp);
3209 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3215 tran_finalize(tran, ret);
3217 bdrv_unref(child_bs);
3219 return ret < 0 ? NULL : child;
3222 /* Callers must ensure that child->frozen is false. */
3223 void bdrv_root_unref_child(BdrvChild *child)
3225 BlockDriverState *child_bs = child->bs;
3227 GLOBAL_STATE_CODE();
3228 bdrv_replace_child_noperm(child, NULL);
3229 bdrv_child_free(child);
3233 * Update permissions for old node. We're just taking a parent away, so
3234 * we're loosening restrictions. Errors of permission update are not
3235 * fatal in this case, ignore them.
3237 bdrv_refresh_perms(child_bs, NULL, NULL);
3240 * When the parent requiring a non-default AioContext is removed, the
3241 * node moves back to the main AioContext
3243 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3247 bdrv_unref(child_bs);
3250 typedef struct BdrvSetInheritsFrom {
3251 BlockDriverState *bs;
3252 BlockDriverState *old_inherits_from;
3253 } BdrvSetInheritsFrom;
3255 static void bdrv_set_inherits_from_abort(void *opaque)
3257 BdrvSetInheritsFrom *s = opaque;
3259 s->bs->inherits_from = s->old_inherits_from;
3262 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3263 .abort = bdrv_set_inherits_from_abort,
3267 /* @tran is allowed to be NULL. In this case no rollback is possible */
3268 static void bdrv_set_inherits_from(BlockDriverState *bs,
3269 BlockDriverState *new_inherits_from,
3273 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3275 *s = (BdrvSetInheritsFrom) {
3277 .old_inherits_from = bs->inherits_from,
3280 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3283 bs->inherits_from = new_inherits_from;
3287 * Clear all inherits_from pointers from children and grandchildren of
3288 * @root that point to @root, where necessary.
3289 * @tran is allowed to be NULL. In this case no rollback is possible
3291 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3296 if (child->bs->inherits_from == root) {
3298 * Remove inherits_from only when the last reference between root and
3299 * child->bs goes away.
3301 QLIST_FOREACH(c, &root->children, next) {
3302 if (c != child && c->bs == child->bs) {
3307 bdrv_set_inherits_from(child->bs, NULL, tran);
3311 QLIST_FOREACH(c, &child->bs->children, next) {
3312 bdrv_unset_inherits_from(root, c, tran);
3316 /* Callers must ensure that child->frozen is false. */
3317 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3319 GLOBAL_STATE_CODE();
3320 if (child == NULL) {
3324 bdrv_unset_inherits_from(parent, child, NULL);
3325 bdrv_root_unref_child(child);
3329 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3332 GLOBAL_STATE_CODE();
3333 QLIST_FOREACH(c, &bs->parents, next_parent) {
3334 if (c->klass->change_media) {
3335 c->klass->change_media(c, load);
3340 /* Return true if you can reach parent going through child->inherits_from
3341 * recursively. If parent or child are NULL, return false */
3342 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3343 BlockDriverState *parent)
3345 while (child && child != parent) {
3346 child = child->inherits_from;
3349 return child != NULL;
3353 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3354 * mostly used for COW backing children (role = COW), but also for
3355 * filtered children (role = FILTERED | PRIMARY).
3357 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3359 if (bs->drv && bs->drv->is_filter) {
3360 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3362 return BDRV_CHILD_COW;
3367 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3368 * callers which don't need their own reference any more must call bdrv_unref().
3370 * Function doesn't update permissions, caller is responsible for this.
3372 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3373 * @child_bs can move to a different AioContext in this function. Callers must
3374 * make sure that their AioContext locking is still correct after this.
3376 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3377 BlockDriverState *child_bs,
3379 Transaction *tran, Error **errp)
3381 bool update_inherits_from =
3382 bdrv_inherits_from_recursive(child_bs, parent_bs);
3383 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3386 GLOBAL_STATE_CODE();
3388 if (!parent_bs->drv) {
3390 * Node without drv is an object without a class :/. TODO: finally fix
3391 * qcow2 driver to never clear bs->drv and implement format corruption
3392 * handling in other way.
3394 error_setg(errp, "Node corrupted");
3398 if (child && child->frozen) {
3399 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3400 child->name, parent_bs->node_name, child->bs->node_name);
3404 if (is_backing && !parent_bs->drv->is_filter &&
3405 !parent_bs->drv->supports_backing)
3407 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3408 "files", parent_bs->drv->format_name, parent_bs->node_name);
3412 if (parent_bs->drv->is_filter) {
3413 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3414 } else if (is_backing) {
3415 role = BDRV_CHILD_COW;
3418 * We only can use same role as it is in existing child. We don't have
3419 * infrastructure to determine role of file child in generic way
3422 error_setg(errp, "Cannot set file child to format node without "
3430 bdrv_unset_inherits_from(parent_bs, child, tran);
3431 bdrv_remove_child(child, tran);
3438 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3439 is_backing ? "backing" : "file",
3440 &child_of_bds, role,
3448 * If inherits_from pointed recursively to bs then let's update it to
3449 * point directly to bs (else it will become NULL).
3451 if (update_inherits_from) {
3452 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3456 bdrv_graph_rdlock_main_loop();
3457 bdrv_refresh_limits(parent_bs, tran, NULL);
3458 bdrv_graph_rdunlock_main_loop();
3464 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3465 * @backing_hd can move to a different AioContext in this function. Callers must
3466 * make sure that their AioContext locking is still correct after this.
3468 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3469 BlockDriverState *backing_hd,
3470 Transaction *tran, Error **errp)
3472 GLOBAL_STATE_CODE();
3473 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3476 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3477 BlockDriverState *backing_hd,
3481 Transaction *tran = tran_new();
3483 GLOBAL_STATE_CODE();
3484 assert(bs->quiesce_counter > 0);
3486 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3491 ret = bdrv_refresh_perms(bs, tran, errp);
3493 tran_finalize(tran, ret);
3497 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3501 GLOBAL_STATE_CODE();
3503 bdrv_drained_begin(bs);
3504 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3505 bdrv_drained_end(bs);
3511 * Opens the backing file for a BlockDriverState if not yet open
3513 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3514 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3515 * itself, all options starting with "${bdref_key}." are considered part of the
3518 * The caller must hold the main AioContext lock.
3520 * TODO Can this be unified with bdrv_open_image()?
3522 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3523 const char *bdref_key, Error **errp)
3525 char *backing_filename = NULL;
3526 char *bdref_key_dot;
3527 const char *reference = NULL;
3529 bool implicit_backing = false;
3530 BlockDriverState *backing_hd;
3531 AioContext *backing_hd_ctx;
3533 QDict *tmp_parent_options = NULL;
3534 Error *local_err = NULL;
3536 GLOBAL_STATE_CODE();
3538 if (bs->backing != NULL) {
3542 /* NULL means an empty set of options */
3543 if (parent_options == NULL) {
3544 tmp_parent_options = qdict_new();
3545 parent_options = tmp_parent_options;
3548 bs->open_flags &= ~BDRV_O_NO_BACKING;
3550 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3551 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3552 g_free(bdref_key_dot);
3555 * Caution: while qdict_get_try_str() is fine, getting non-string
3556 * types would require more care. When @parent_options come from
3557 * -blockdev or blockdev_add, its members are typed according to
3558 * the QAPI schema, but when they come from -drive, they're all
3561 reference = qdict_get_try_str(parent_options, bdref_key);
3562 if (reference || qdict_haskey(options, "file.filename")) {
3563 /* keep backing_filename NULL */
3564 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3565 qobject_unref(options);
3568 if (qdict_size(options) == 0) {
3569 /* If the user specifies options that do not modify the
3570 * backing file's behavior, we might still consider it the
3571 * implicit backing file. But it's easier this way, and
3572 * just specifying some of the backing BDS's options is
3573 * only possible with -drive anyway (otherwise the QAPI
3574 * schema forces the user to specify everything). */
3575 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3578 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3581 error_propagate(errp, local_err);
3582 qobject_unref(options);
3587 if (!bs->drv || !bs->drv->supports_backing) {
3589 error_setg(errp, "Driver doesn't support backing files");
3590 qobject_unref(options);
3595 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3596 qdict_put_str(options, "driver", bs->backing_format);
3599 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3600 &child_of_bds, bdrv_backing_role(bs), errp);
3602 bs->open_flags |= BDRV_O_NO_BACKING;
3603 error_prepend(errp, "Could not open backing file: ");
3608 if (implicit_backing) {
3609 bdrv_refresh_filename(backing_hd);
3610 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3611 backing_hd->filename);
3614 /* Hook up the backing file link; drop our reference, bs owns the
3615 * backing_hd reference now */
3616 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3617 aio_context_acquire(backing_hd_ctx);
3618 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3619 bdrv_unref(backing_hd);
3620 aio_context_release(backing_hd_ctx);
3626 qdict_del(parent_options, bdref_key);
3629 g_free(backing_filename);
3630 qobject_unref(tmp_parent_options);
3634 static BlockDriverState *
3635 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3636 BlockDriverState *parent, const BdrvChildClass *child_class,
3637 BdrvChildRole child_role, bool allow_none, Error **errp)
3639 BlockDriverState *bs = NULL;
3640 QDict *image_options;
3641 char *bdref_key_dot;
3642 const char *reference;
3644 assert(child_class != NULL);
3646 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3647 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3648 g_free(bdref_key_dot);
3651 * Caution: while qdict_get_try_str() is fine, getting non-string
3652 * types would require more care. When @options come from
3653 * -blockdev or blockdev_add, its members are typed according to
3654 * the QAPI schema, but when they come from -drive, they're all
3657 reference = qdict_get_try_str(options, bdref_key);
3658 if (!filename && !reference && !qdict_size(image_options)) {
3660 error_setg(errp, "A block device must be specified for \"%s\"",
3663 qobject_unref(image_options);
3667 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3668 parent, child_class, child_role, errp);
3674 qdict_del(options, bdref_key);
3679 * Opens a disk image whose options are given as BlockdevRef in another block
3682 * If allow_none is true, no image will be opened if filename is false and no
3683 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3685 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3686 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3687 * itself, all options starting with "${bdref_key}." are considered part of the
3690 * The BlockdevRef will be removed from the options QDict.
3692 * The caller must hold the lock of the main AioContext and no other AioContext.
3693 * @parent can move to a different AioContext in this function. Callers must
3694 * make sure that their AioContext locking is still correct after this.
3696 BdrvChild *bdrv_open_child(const char *filename,
3697 QDict *options, const char *bdref_key,
3698 BlockDriverState *parent,
3699 const BdrvChildClass *child_class,
3700 BdrvChildRole child_role,
3701 bool allow_none, Error **errp)
3703 BlockDriverState *bs;
3707 GLOBAL_STATE_CODE();
3709 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3710 child_role, allow_none, errp);
3715 ctx = bdrv_get_aio_context(bs);
3716 aio_context_acquire(ctx);
3717 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3719 aio_context_release(ctx);
3725 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3727 * The caller must hold the lock of the main AioContext and no other AioContext.
3728 * @parent can move to a different AioContext in this function. Callers must
3729 * make sure that their AioContext locking is still correct after this.
3731 int bdrv_open_file_child(const char *filename,
3732 QDict *options, const char *bdref_key,
3733 BlockDriverState *parent, Error **errp)
3737 /* commit_top and mirror_top don't use this function */
3738 assert(!parent->drv->filtered_child_is_backing);
3739 role = parent->drv->is_filter ?
3740 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3742 if (!bdrv_open_child(filename, options, bdref_key, parent,
3743 &child_of_bds, role, false, errp))
3752 * TODO Future callers may need to specify parent/child_class in order for
3753 * option inheritance to work. Existing callers use it for the root node.
3755 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3757 BlockDriverState *bs = NULL;
3758 QObject *obj = NULL;
3759 QDict *qdict = NULL;
3760 const char *reference = NULL;
3763 GLOBAL_STATE_CODE();
3765 if (ref->type == QTYPE_QSTRING) {
3766 reference = ref->u.reference;
3768 BlockdevOptions *options = &ref->u.definition;
3769 assert(ref->type == QTYPE_QDICT);
3771 v = qobject_output_visitor_new(&obj);
3772 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3773 visit_complete(v, &obj);
3775 qdict = qobject_to(QDict, obj);
3776 qdict_flatten(qdict);
3778 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3779 * compatibility with other callers) rather than what we want as the
3780 * real defaults. Apply the defaults here instead. */
3781 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3782 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3783 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3784 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3788 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3795 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3797 QDict *snapshot_options,
3800 g_autofree char *tmp_filename = NULL;
3802 QemuOpts *opts = NULL;
3803 BlockDriverState *bs_snapshot = NULL;
3804 AioContext *ctx = bdrv_get_aio_context(bs);
3807 GLOBAL_STATE_CODE();
3809 /* if snapshot, we create a temporary backing file and open it
3810 instead of opening 'filename' directly */
3812 /* Get the required size from the image */
3813 aio_context_acquire(ctx);
3814 total_size = bdrv_getlength(bs);
3815 aio_context_release(ctx);
3817 if (total_size < 0) {
3818 error_setg_errno(errp, -total_size, "Could not get image size");
3822 /* Create the temporary image */
3823 tmp_filename = create_tmp_file(errp);
3824 if (!tmp_filename) {
3828 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3830 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3831 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3832 qemu_opts_del(opts);
3834 error_prepend(errp, "Could not create temporary overlay '%s': ",
3839 /* Prepare options QDict for the temporary file */
3840 qdict_put_str(snapshot_options, "file.driver", "file");
3841 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3842 qdict_put_str(snapshot_options, "driver", "qcow2");
3844 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3845 snapshot_options = NULL;
3850 aio_context_acquire(ctx);
3851 ret = bdrv_append(bs_snapshot, bs, errp);
3852 aio_context_release(ctx);
3860 qobject_unref(snapshot_options);
3865 * Opens a disk image (raw, qcow2, vmdk, ...)
3867 * options is a QDict of options to pass to the block drivers, or NULL for an
3868 * empty set of options. The reference to the QDict belongs to the block layer
3869 * after the call (even on failure), so if the caller intends to reuse the
3870 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3872 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3873 * If it is not NULL, the referenced BDS will be reused.
3875 * The reference parameter may be used to specify an existing block device which
3876 * should be opened. If specified, neither options nor a filename may be given,
3877 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3879 * The caller must always hold the main AioContext lock.
3881 static BlockDriverState * no_coroutine_fn
3882 bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3883 int flags, BlockDriverState *parent,
3884 const BdrvChildClass *child_class, BdrvChildRole child_role,
3888 BlockBackend *file = NULL;
3889 BlockDriverState *bs;
3890 BlockDriver *drv = NULL;
3892 const char *drvname;
3893 const char *backing;
3894 Error *local_err = NULL;
3895 QDict *snapshot_options = NULL;
3896 int snapshot_flags = 0;
3897 AioContext *ctx = qemu_get_aio_context();
3899 assert(!child_class || !flags);
3900 assert(!child_class == !parent);
3901 GLOBAL_STATE_CODE();
3902 assert(!qemu_in_coroutine());
3905 bool options_non_empty = options ? qdict_size(options) : false;
3906 qobject_unref(options);
3908 if (filename || options_non_empty) {
3909 error_setg(errp, "Cannot reference an existing block device with "
3910 "additional options or a new filename");
3914 bs = bdrv_lookup_bs(reference, reference, errp);
3925 /* NULL means an empty set of options */
3926 if (options == NULL) {
3927 options = qdict_new();
3930 /* json: syntax counts as explicit options, as if in the QDict */
3931 parse_json_protocol(options, &filename, &local_err);
3936 bs->explicit_options = qdict_clone_shallow(options);
3939 bool parent_is_format;
3942 parent_is_format = parent->drv->is_format;
3945 * parent->drv is not set yet because this node is opened for
3946 * (potential) format probing. That means that @parent is going
3947 * to be a format node.
3949 parent_is_format = true;
3952 bs->inherits_from = parent;
3953 child_class->inherit_options(child_role, parent_is_format,
3955 parent->open_flags, parent->options);
3958 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3964 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3965 * Caution: getting a boolean member of @options requires care.
3966 * When @options come from -blockdev or blockdev_add, members are
3967 * typed according to the QAPI schema, but when they come from
3968 * -drive, they're all QString.
3970 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3971 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3972 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3974 flags &= ~BDRV_O_RDWR;
3977 if (flags & BDRV_O_SNAPSHOT) {
3978 snapshot_options = qdict_new();
3979 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3981 /* Let bdrv_backing_options() override "read-only" */
3982 qdict_del(options, BDRV_OPT_READ_ONLY);
3983 bdrv_inherited_options(BDRV_CHILD_COW, true,
3984 &flags, options, flags, options);
3987 bs->open_flags = flags;
3988 bs->options = options;
3989 options = qdict_clone_shallow(options);
3991 /* Find the right image format driver */
3992 /* See cautionary note on accessing @options above */
3993 drvname = qdict_get_try_str(options, "driver");
3995 drv = bdrv_find_format(drvname);
3997 error_setg(errp, "Unknown driver: '%s'", drvname);
4002 assert(drvname || !(flags & BDRV_O_PROTOCOL));
4004 /* See cautionary note on accessing @options above */
4005 backing = qdict_get_try_str(options, "backing");
4006 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4007 (backing && *backing == '\0'))
4010 warn_report("Use of \"backing\": \"\" is deprecated; "
4011 "use \"backing\": null instead");
4013 flags |= BDRV_O_NO_BACKING;
4014 qdict_del(bs->explicit_options, "backing");
4015 qdict_del(bs->options, "backing");
4016 qdict_del(options, "backing");
4019 /* Open image file without format layer. This BlockBackend is only used for
4020 * probing, the block drivers will do their own bdrv_open_child() for the
4021 * same BDS, which is why we put the node name back into options. */
4022 if ((flags & BDRV_O_PROTOCOL) == 0) {
4023 BlockDriverState *file_bs;
4025 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
4026 &child_of_bds, BDRV_CHILD_IMAGE,
4031 if (file_bs != NULL) {
4032 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4033 * looking at the header to guess the image format. This works even
4034 * in cases where a guest would not see a consistent state. */
4035 ctx = bdrv_get_aio_context(file_bs);
4036 aio_context_acquire(ctx);
4037 file = blk_new(ctx, 0, BLK_PERM_ALL);
4038 blk_insert_bs(file, file_bs, &local_err);
4039 bdrv_unref(file_bs);
4040 aio_context_release(ctx);
4046 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4050 /* Image format probing */
4053 ret = find_image_format(file, filename, &drv, &local_err);
4058 * This option update would logically belong in bdrv_fill_options(),
4059 * but we first need to open bs->file for the probing to work, while
4060 * opening bs->file already requires the (mostly) final set of options
4061 * so that cache mode etc. can be inherited.
4063 * Adding the driver later is somewhat ugly, but it's not an option
4064 * that would ever be inherited, so it's correct. We just need to make
4065 * sure to update both bs->options (which has the full effective
4066 * options for bs) and options (which has file.* already removed).
4068 qdict_put_str(bs->options, "driver", drv->format_name);
4069 qdict_put_str(options, "driver", drv->format_name);
4071 error_setg(errp, "Must specify either driver or file");
4075 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4076 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4077 /* file must be NULL if a protocol BDS is about to be created
4078 * (the inverse results in an error message from bdrv_open_common()) */
4079 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4081 /* Open the image */
4082 ret = bdrv_open_common(bs, file, options, &local_err);
4087 /* The AioContext could have changed during bdrv_open_common() */
4088 ctx = bdrv_get_aio_context(bs);
4091 aio_context_acquire(ctx);
4093 aio_context_release(ctx);
4097 /* If there is a backing file, use it */
4098 if ((flags & BDRV_O_NO_BACKING) == 0) {
4099 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4101 goto close_and_fail;
4105 /* Remove all children options and references
4106 * from bs->options and bs->explicit_options */
4107 QLIST_FOREACH(child, &bs->children, next) {
4108 char *child_key_dot;
4109 child_key_dot = g_strdup_printf("%s.", child->name);
4110 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4111 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4112 qdict_del(bs->explicit_options, child->name);
4113 qdict_del(bs->options, child->name);
4114 g_free(child_key_dot);
4117 /* Check if any unknown options were used */
4118 if (qdict_size(options) != 0) {
4119 const QDictEntry *entry = qdict_first(options);
4120 if (flags & BDRV_O_PROTOCOL) {
4121 error_setg(errp, "Block protocol '%s' doesn't support the option "
4122 "'%s'", drv->format_name, entry->key);
4125 "Block format '%s' does not support the option '%s'",
4126 drv->format_name, entry->key);
4129 goto close_and_fail;
4132 bdrv_parent_cb_change_media(bs, true);
4134 qobject_unref(options);
4137 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4138 * temporary snapshot afterwards. */
4139 if (snapshot_flags) {
4140 BlockDriverState *snapshot_bs;
4141 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4142 snapshot_options, &local_err);
4143 snapshot_options = NULL;
4145 goto close_and_fail;
4147 /* We are not going to return bs but the overlay on top of it
4148 * (snapshot_bs); thus, we have to drop the strong reference to bs
4149 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4150 * though, because the overlay still has a reference to it. */
4151 aio_context_acquire(ctx);
4153 aio_context_release(ctx);
4160 aio_context_acquire(ctx);
4162 qobject_unref(snapshot_options);
4163 qobject_unref(bs->explicit_options);
4164 qobject_unref(bs->options);
4165 qobject_unref(options);
4167 bs->explicit_options = NULL;
4169 aio_context_release(ctx);
4170 error_propagate(errp, local_err);
4174 aio_context_acquire(ctx);
4176 aio_context_release(ctx);
4177 qobject_unref(snapshot_options);
4178 qobject_unref(options);
4179 error_propagate(errp, local_err);
4183 /* The caller must always hold the main AioContext lock. */
4184 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4185 QDict *options, int flags, Error **errp)
4187 GLOBAL_STATE_CODE();
4189 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4193 /* Return true if the NULL-terminated @list contains @str */
4194 static bool is_str_in_list(const char *str, const char *const *list)
4198 for (i = 0; list[i] != NULL; i++) {
4199 if (!strcmp(str, list[i])) {
4208 * Check that every option set in @bs->options is also set in
4211 * Options listed in the common_options list and in
4212 * @bs->drv->mutable_opts are skipped.
4214 * Return 0 on success, otherwise return -EINVAL and set @errp.
4216 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4217 const QDict *new_opts, Error **errp)
4219 const QDictEntry *e;
4220 /* These options are common to all block drivers and are handled
4221 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4222 const char *const common_options[] = {
4223 "node-name", "discard", "cache.direct", "cache.no-flush",
4224 "read-only", "auto-read-only", "detect-zeroes", NULL
4227 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4228 if (!qdict_haskey(new_opts, e->key) &&
4229 !is_str_in_list(e->key, common_options) &&
4230 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4231 error_setg(errp, "Option '%s' cannot be reset "
4232 "to its default value", e->key);
4241 * Returns true if @child can be reached recursively from @bs
4243 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4244 BlockDriverState *child)
4252 QLIST_FOREACH(c, &bs->children, next) {
4253 if (bdrv_recurse_has_child(c->bs, child)) {
4262 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4263 * reopen of multiple devices.
4265 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4266 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4267 * be created and initialized. This newly created BlockReopenQueue should be
4268 * passed back in for subsequent calls that are intended to be of the same
4271 * bs is the BlockDriverState to add to the reopen queue.
4273 * options contains the changed options for the associated bs
4274 * (the BlockReopenQueue takes ownership)
4276 * flags contains the open flags for the associated bs
4278 * returns a pointer to bs_queue, which is either the newly allocated
4279 * bs_queue, or the existing bs_queue being used.
4281 * bs is drained here and undrained by bdrv_reopen_queue_free().
4283 * To be called with bs->aio_context locked.
4285 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4286 BlockDriverState *bs,
4288 const BdrvChildClass *klass,
4290 bool parent_is_format,
4291 QDict *parent_options,
4297 BlockReopenQueueEntry *bs_entry;
4299 QDict *old_options, *explicit_options, *options_copy;
4303 GLOBAL_STATE_CODE();
4305 bdrv_drained_begin(bs);
4307 if (bs_queue == NULL) {
4308 bs_queue = g_new0(BlockReopenQueue, 1);
4309 QTAILQ_INIT(bs_queue);
4313 options = qdict_new();
4316 /* Check if this BlockDriverState is already in the queue */
4317 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4318 if (bs == bs_entry->state.bs) {
4324 * Precedence of options:
4325 * 1. Explicitly passed in options (highest)
4326 * 2. Retained from explicitly set options of bs
4327 * 3. Inherited from parent node
4328 * 4. Retained from effective options of bs
4331 /* Old explicitly set values (don't overwrite by inherited value) */
4332 if (bs_entry || keep_old_opts) {
4333 old_options = qdict_clone_shallow(bs_entry ?
4334 bs_entry->state.explicit_options :
4335 bs->explicit_options);
4336 bdrv_join_options(bs, options, old_options);
4337 qobject_unref(old_options);
4340 explicit_options = qdict_clone_shallow(options);
4342 /* Inherit from parent node */
4343 if (parent_options) {
4345 klass->inherit_options(role, parent_is_format, &flags, options,
4346 parent_flags, parent_options);
4348 flags = bdrv_get_flags(bs);
4351 if (keep_old_opts) {
4352 /* Old values are used for options that aren't set yet */
4353 old_options = qdict_clone_shallow(bs->options);
4354 bdrv_join_options(bs, options, old_options);
4355 qobject_unref(old_options);
4358 /* We have the final set of options so let's update the flags */
4359 options_copy = qdict_clone_shallow(options);
4360 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4361 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4362 update_flags_from_options(&flags, opts);
4363 qemu_opts_del(opts);
4364 qobject_unref(options_copy);
4366 /* bdrv_open_inherit() sets and clears some additional flags internally */
4367 flags &= ~BDRV_O_PROTOCOL;
4368 if (flags & BDRV_O_RDWR) {
4369 flags |= BDRV_O_ALLOW_RDWR;
4373 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4374 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4376 qobject_unref(bs_entry->state.options);
4377 qobject_unref(bs_entry->state.explicit_options);
4380 bs_entry->state.bs = bs;
4381 bs_entry->state.options = options;
4382 bs_entry->state.explicit_options = explicit_options;
4383 bs_entry->state.flags = flags;
4386 * If keep_old_opts is false then it means that unspecified
4387 * options must be reset to their original value. We don't allow
4388 * resetting 'backing' but we need to know if the option is
4389 * missing in order to decide if we have to return an error.
4391 if (!keep_old_opts) {
4392 bs_entry->state.backing_missing =
4393 !qdict_haskey(options, "backing") &&
4394 !qdict_haskey(options, "backing.driver");
4397 QLIST_FOREACH(child, &bs->children, next) {
4398 QDict *new_child_options = NULL;
4399 bool child_keep_old = keep_old_opts;
4401 /* reopen can only change the options of block devices that were
4402 * implicitly created and inherited options. For other (referenced)
4403 * block devices, a syntax like "backing.foo" results in an error. */
4404 if (child->bs->inherits_from != bs) {
4408 /* Check if the options contain a child reference */
4409 if (qdict_haskey(options, child->name)) {
4410 const char *childref = qdict_get_try_str(options, child->name);
4412 * The current child must not be reopened if the child
4413 * reference is null or points to a different node.
4415 if (g_strcmp0(childref, child->bs->node_name)) {
4419 * If the child reference points to the current child then
4420 * reopen it with its existing set of options (note that
4421 * it can still inherit new options from the parent).
4423 child_keep_old = true;
4425 /* Extract child options ("child-name.*") */
4426 char *child_key_dot = g_strdup_printf("%s.", child->name);
4427 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4428 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4429 g_free(child_key_dot);
4432 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4433 child->klass, child->role, bs->drv->is_format,
4434 options, flags, child_keep_old);
4440 /* To be called with bs->aio_context locked */
4441 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4442 BlockDriverState *bs,
4443 QDict *options, bool keep_old_opts)
4445 GLOBAL_STATE_CODE();
4447 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4448 NULL, 0, keep_old_opts);
4451 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4453 GLOBAL_STATE_CODE();
4455 BlockReopenQueueEntry *bs_entry, *next;
4456 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4457 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4459 aio_context_acquire(ctx);
4460 bdrv_drained_end(bs_entry->state.bs);
4461 aio_context_release(ctx);
4463 qobject_unref(bs_entry->state.explicit_options);
4464 qobject_unref(bs_entry->state.options);
4472 * Reopen multiple BlockDriverStates atomically & transactionally.
4474 * The queue passed in (bs_queue) must have been built up previous
4475 * via bdrv_reopen_queue().
4477 * Reopens all BDS specified in the queue, with the appropriate
4478 * flags. All devices are prepared for reopen, and failure of any
4479 * device will cause all device changes to be abandoned, and intermediate
4482 * If all devices prepare successfully, then the changes are committed
4485 * All affected nodes must be drained between bdrv_reopen_queue() and
4486 * bdrv_reopen_multiple().
4488 * To be called from the main thread, with all other AioContexts unlocked.
4490 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4493 BlockReopenQueueEntry *bs_entry, *next;
4495 Transaction *tran = tran_new();
4496 g_autoptr(GSList) refresh_list = NULL;
4498 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4499 assert(bs_queue != NULL);
4500 GLOBAL_STATE_CODE();
4502 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4503 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4504 aio_context_acquire(ctx);
4505 ret = bdrv_flush(bs_entry->state.bs);
4506 aio_context_release(ctx);
4508 error_setg_errno(errp, -ret, "Error flushing drive");
4513 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4514 assert(bs_entry->state.bs->quiesce_counter > 0);
4515 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4516 aio_context_acquire(ctx);
4517 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4518 aio_context_release(ctx);
4522 bs_entry->prepared = true;
4525 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4526 BDRVReopenState *state = &bs_entry->state;
4528 refresh_list = g_slist_prepend(refresh_list, state->bs);
4529 if (state->old_backing_bs) {
4530 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4532 if (state->old_file_bs) {
4533 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4538 * Note that file-posix driver rely on permission update done during reopen
4539 * (even if no permission changed), because it wants "new" permissions for
4540 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4541 * in raw_reopen_prepare() which is called with "old" permissions.
4543 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4549 * If we reach this point, we have success and just need to apply the
4552 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4553 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4554 * children are usually goes after parents in reopen-queue, so go from last
4557 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4558 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4559 aio_context_acquire(ctx);
4560 bdrv_reopen_commit(&bs_entry->state);
4561 aio_context_release(ctx);
4566 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4567 BlockDriverState *bs = bs_entry->state.bs;
4569 if (bs->drv->bdrv_reopen_commit_post) {
4570 ctx = bdrv_get_aio_context(bs);
4571 aio_context_acquire(ctx);
4572 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4573 aio_context_release(ctx);
4582 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4583 if (bs_entry->prepared) {
4584 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4585 aio_context_acquire(ctx);
4586 bdrv_reopen_abort(&bs_entry->state);
4587 aio_context_release(ctx);
4592 bdrv_reopen_queue_free(bs_queue);
4597 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4600 AioContext *ctx = bdrv_get_aio_context(bs);
4601 BlockReopenQueue *queue;
4604 GLOBAL_STATE_CODE();
4606 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4608 if (ctx != qemu_get_aio_context()) {
4609 aio_context_release(ctx);
4611 ret = bdrv_reopen_multiple(queue, errp);
4613 if (ctx != qemu_get_aio_context()) {
4614 aio_context_acquire(ctx);
4620 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4623 QDict *opts = qdict_new();
4625 GLOBAL_STATE_CODE();
4627 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4629 return bdrv_reopen(bs, opts, true, errp);
4633 * Take a BDRVReopenState and check if the value of 'backing' in the
4634 * reopen_state->options QDict is valid or not.
4636 * If 'backing' is missing from the QDict then return 0.
4638 * If 'backing' contains the node name of the backing file of
4639 * reopen_state->bs then return 0.
4641 * If 'backing' contains a different node name (or is null) then check
4642 * whether the current backing file can be replaced with the new one.
4643 * If that's the case then reopen_state->replace_backing_bs is set to
4644 * true and reopen_state->new_backing_bs contains a pointer to the new
4645 * backing BlockDriverState (or NULL).
4647 * Return 0 on success, otherwise return < 0 and set @errp.
4649 * The caller must hold the AioContext lock of @reopen_state->bs.
4650 * @reopen_state->bs can move to a different AioContext in this function.
4651 * Callers must make sure that their AioContext locking is still correct after
4654 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4655 bool is_backing, Transaction *tran,
4658 BlockDriverState *bs = reopen_state->bs;
4659 BlockDriverState *new_child_bs;
4660 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4662 const char *child_name = is_backing ? "backing" : "file";
4665 AioContext *ctx, *old_ctx;
4668 GLOBAL_STATE_CODE();
4670 value = qdict_get(reopen_state->options, child_name);
4671 if (value == NULL) {
4675 switch (qobject_type(value)) {
4677 assert(is_backing); /* The 'file' option does not allow a null value */
4678 new_child_bs = NULL;
4681 str = qstring_get_str(qobject_to(QString, value));
4682 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4683 if (new_child_bs == NULL) {
4685 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4686 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4687 "cycle", str, child_name, bs->node_name);
4693 * The options QDict has been flattened, so 'backing' and 'file'
4694 * do not allow any other data type here.
4696 g_assert_not_reached();
4699 if (old_child_bs == new_child_bs) {
4704 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4708 if (old_child_bs->implicit) {
4709 error_setg(errp, "Cannot replace implicit %s child of %s",
4710 child_name, bs->node_name);
4715 if (bs->drv->is_filter && !old_child_bs) {
4717 * Filters always have a file or a backing child, so we are trying to
4718 * change wrong child
4720 error_setg(errp, "'%s' is a %s filter node that does not support a "
4721 "%s child", bs->node_name, bs->drv->format_name, child_name);
4726 reopen_state->old_backing_bs = old_child_bs;
4728 reopen_state->old_file_bs = old_child_bs;
4731 old_ctx = bdrv_get_aio_context(bs);
4732 ctx = bdrv_get_aio_context(new_child_bs);
4733 if (old_ctx != ctx) {
4734 aio_context_release(old_ctx);
4735 aio_context_acquire(ctx);
4738 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4741 if (old_ctx != ctx) {
4742 aio_context_release(ctx);
4743 aio_context_acquire(old_ctx);
4750 * Prepares a BlockDriverState for reopen. All changes are staged in the
4751 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4752 * the block driver layer .bdrv_reopen_prepare()
4754 * bs is the BlockDriverState to reopen
4755 * flags are the new open flags
4756 * queue is the reopen queue
4758 * Returns 0 on success, non-zero on error. On error errp will be set
4761 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4762 * It is the responsibility of the caller to then call the abort() or
4763 * commit() for any other BDS that have been left in a prepare() state
4765 * The caller must hold the AioContext lock of @reopen_state->bs.
4767 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4768 BlockReopenQueue *queue,
4769 Transaction *change_child_tran, Error **errp)
4773 Error *local_err = NULL;
4776 QDict *orig_reopen_opts;
4777 char *discard = NULL;
4779 bool drv_prepared = false;
4781 assert(reopen_state != NULL);
4782 assert(reopen_state->bs->drv != NULL);
4783 GLOBAL_STATE_CODE();
4784 drv = reopen_state->bs->drv;
4786 /* This function and each driver's bdrv_reopen_prepare() remove
4787 * entries from reopen_state->options as they are processed, so
4788 * we need to make a copy of the original QDict. */
4789 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4791 /* Process generic block layer options */
4792 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4793 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4798 /* This was already called in bdrv_reopen_queue_child() so the flags
4799 * are up-to-date. This time we simply want to remove the options from
4800 * QemuOpts in order to indicate that they have been processed. */
4801 old_flags = reopen_state->flags;
4802 update_flags_from_options(&reopen_state->flags, opts);
4803 assert(old_flags == reopen_state->flags);
4805 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4806 if (discard != NULL) {
4807 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4808 error_setg(errp, "Invalid discard option");
4814 reopen_state->detect_zeroes =
4815 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4817 error_propagate(errp, local_err);
4822 /* All other options (including node-name and driver) must be unchanged.
4823 * Put them back into the QDict, so that they are checked at the end
4824 * of this function. */
4825 qemu_opts_to_qdict(opts, reopen_state->options);
4827 /* If we are to stay read-only, do not allow permission change
4828 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4829 * not set, or if the BDS still has copy_on_read enabled */
4830 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4831 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4833 error_propagate(errp, local_err);
4837 if (drv->bdrv_reopen_prepare) {
4839 * If a driver-specific option is missing, it means that we
4840 * should reset it to its default value.
4841 * But not all options allow that, so we need to check it first.
4843 ret = bdrv_reset_options_allowed(reopen_state->bs,
4844 reopen_state->options, errp);
4849 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4851 if (local_err != NULL) {
4852 error_propagate(errp, local_err);
4854 bdrv_refresh_filename(reopen_state->bs);
4855 error_setg(errp, "failed while preparing to reopen image '%s'",
4856 reopen_state->bs->filename);
4861 /* It is currently mandatory to have a bdrv_reopen_prepare()
4862 * handler for each supported drv. */
4863 error_setg(errp, "Block format '%s' used by node '%s' "
4864 "does not support reopening files", drv->format_name,
4865 bdrv_get_device_or_node_name(reopen_state->bs));
4870 drv_prepared = true;
4873 * We must provide the 'backing' option if the BDS has a backing
4874 * file or if the image file has a backing file name as part of
4875 * its metadata. Otherwise the 'backing' option can be omitted.
4877 if (drv->supports_backing && reopen_state->backing_missing &&
4878 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4879 error_setg(errp, "backing is missing for '%s'",
4880 reopen_state->bs->node_name);
4886 * Allow changing the 'backing' option. The new value can be
4887 * either a reference to an existing node (using its node name)
4888 * or NULL to simply detach the current backing file.
4890 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4891 change_child_tran, errp);
4895 qdict_del(reopen_state->options, "backing");
4897 /* Allow changing the 'file' option. In this case NULL is not allowed */
4898 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4899 change_child_tran, errp);
4903 qdict_del(reopen_state->options, "file");
4905 /* Options that are not handled are only okay if they are unchanged
4906 * compared to the old state. It is expected that some options are only
4907 * used for the initial open, but not reopen (e.g. filename) */
4908 if (qdict_size(reopen_state->options)) {
4909 const QDictEntry *entry = qdict_first(reopen_state->options);
4912 QObject *new = entry->value;
4913 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4915 /* Allow child references (child_name=node_name) as long as they
4916 * point to the current child (i.e. everything stays the same). */
4917 if (qobject_type(new) == QTYPE_QSTRING) {
4919 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4920 if (!strcmp(child->name, entry->key)) {
4926 if (!strcmp(child->bs->node_name,
4927 qstring_get_str(qobject_to(QString, new)))) {
4928 continue; /* Found child with this name, skip option */
4934 * TODO: When using -drive to specify blockdev options, all values
4935 * will be strings; however, when using -blockdev, blockdev-add or
4936 * filenames using the json:{} pseudo-protocol, they will be
4938 * In contrast, reopening options are (currently) always strings
4939 * (because you can only specify them through qemu-io; all other
4940 * callers do not specify any options).
4941 * Therefore, when using anything other than -drive to create a BDS,
4942 * this cannot detect non-string options as unchanged, because
4943 * qobject_is_equal() always returns false for objects of different
4944 * type. In the future, this should be remedied by correctly typing
4945 * all options. For now, this is not too big of an issue because
4946 * the user can simply omit options which cannot be changed anyway,
4947 * so they will stay unchanged.
4949 if (!qobject_is_equal(new, old)) {
4950 error_setg(errp, "Cannot change the option '%s'", entry->key);
4954 } while ((entry = qdict_next(reopen_state->options, entry)));
4959 /* Restore the original reopen_state->options QDict */
4960 qobject_unref(reopen_state->options);
4961 reopen_state->options = qobject_ref(orig_reopen_opts);
4964 if (ret < 0 && drv_prepared) {
4965 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4966 * call drv->bdrv_reopen_abort() before signaling an error
4967 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4968 * when the respective bdrv_reopen_prepare() has failed) */
4969 if (drv->bdrv_reopen_abort) {
4970 drv->bdrv_reopen_abort(reopen_state);
4973 qemu_opts_del(opts);
4974 qobject_unref(orig_reopen_opts);
4980 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4981 * makes them final by swapping the staging BlockDriverState contents into
4982 * the active BlockDriverState contents.
4984 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4987 BlockDriverState *bs;
4990 assert(reopen_state != NULL);
4991 bs = reopen_state->bs;
4993 assert(drv != NULL);
4994 GLOBAL_STATE_CODE();
4996 /* If there are any driver level actions to take */
4997 if (drv->bdrv_reopen_commit) {
4998 drv->bdrv_reopen_commit(reopen_state);
5001 /* set BDS specific flags now */
5002 qobject_unref(bs->explicit_options);
5003 qobject_unref(bs->options);
5004 qobject_ref(reopen_state->explicit_options);
5005 qobject_ref(reopen_state->options);
5007 bs->explicit_options = reopen_state->explicit_options;
5008 bs->options = reopen_state->options;
5009 bs->open_flags = reopen_state->flags;
5010 bs->detect_zeroes = reopen_state->detect_zeroes;
5012 /* Remove child references from bs->options and bs->explicit_options.
5013 * Child options were already removed in bdrv_reopen_queue_child() */
5014 QLIST_FOREACH(child, &bs->children, next) {
5015 qdict_del(bs->explicit_options, child->name);
5016 qdict_del(bs->options, child->name);
5018 /* backing is probably removed, so it's not handled by previous loop */
5019 qdict_del(bs->explicit_options, "backing");
5020 qdict_del(bs->options, "backing");
5022 bdrv_graph_rdlock_main_loop();
5023 bdrv_refresh_limits(bs, NULL, NULL);
5024 bdrv_graph_rdunlock_main_loop();
5025 bdrv_refresh_total_sectors(bs, bs->total_sectors);
5029 * Abort the reopen, and delete and free the staged changes in
5032 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
5036 assert(reopen_state != NULL);
5037 drv = reopen_state->bs->drv;
5038 assert(drv != NULL);
5039 GLOBAL_STATE_CODE();
5041 if (drv->bdrv_reopen_abort) {
5042 drv->bdrv_reopen_abort(reopen_state);
5047 static void bdrv_close(BlockDriverState *bs)
5049 BdrvAioNotifier *ban, *ban_next;
5050 BdrvChild *child, *next;
5052 GLOBAL_STATE_CODE();
5053 assert(!bs->refcnt);
5055 bdrv_drained_begin(bs); /* complete I/O */
5057 bdrv_drain(bs); /* in case flush left pending I/O */
5060 if (bs->drv->bdrv_close) {
5061 /* Must unfreeze all children, so bdrv_unref_child() works */
5062 bs->drv->bdrv_close(bs);
5067 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
5068 bdrv_unref_child(bs, child);
5071 assert(!bs->backing);
5075 qatomic_set(&bs->copy_on_read, 0);
5076 bs->backing_file[0] = '\0';
5077 bs->backing_format[0] = '\0';
5078 bs->total_sectors = 0;
5079 bs->encrypted = false;
5081 qobject_unref(bs->options);
5082 qobject_unref(bs->explicit_options);
5084 bs->explicit_options = NULL;
5085 qobject_unref(bs->full_open_options);
5086 bs->full_open_options = NULL;
5087 g_free(bs->block_status_cache);
5088 bs->block_status_cache = NULL;
5090 bdrv_release_named_dirty_bitmaps(bs);
5091 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5093 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5096 QLIST_INIT(&bs->aio_notifiers);
5097 bdrv_drained_end(bs);
5100 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5101 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5104 if (bs->quiesce_counter) {
5105 bdrv_drain_all_end_quiesce(bs);
5109 void bdrv_close_all(void)
5111 GLOBAL_STATE_CODE();
5112 assert(job_next(NULL) == NULL);
5114 /* Drop references from requests still in flight, such as canceled block
5115 * jobs whose AIO context has not been polled yet */
5118 blk_remove_all_bs();
5119 blockdev_close_all_bdrv_states();
5121 assert(QTAILQ_EMPTY(&all_bdrv_states));
5124 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
5130 if (c->klass->stay_at_node) {
5134 /* If the child @c belongs to the BDS @to, replacing the current
5135 * c->bs by @to would mean to create a loop.
5137 * Such a case occurs when appending a BDS to a backing chain.
5138 * For instance, imagine the following chain:
5140 * guest device -> node A -> further backing chain...
5142 * Now we create a new BDS B which we want to put on top of this
5143 * chain, so we first attach A as its backing node:
5148 * guest device -> node A -> further backing chain...
5150 * Finally we want to replace A by B. When doing that, we want to
5151 * replace all pointers to A by pointers to B -- except for the
5152 * pointer from B because (1) that would create a loop, and (2)
5153 * that pointer should simply stay intact:
5155 * guest device -> node B
5158 * node A -> further backing chain...
5160 * In general, when replacing a node A (c->bs) by a node B (@to),
5161 * if A is a child of B, that means we cannot replace A by B there
5162 * because that would create a loop. Silently detaching A from B
5163 * is also not really an option. So overall just leaving A in
5164 * place there is the most sensible choice.
5166 * We would also create a loop in any cases where @c is only
5167 * indirectly referenced by @to. Prevent this by returning false
5168 * if @c is found (by breadth-first search) anywhere in the whole
5173 found = g_hash_table_new(NULL, NULL);
5174 g_hash_table_add(found, to);
5175 queue = g_queue_new();
5176 g_queue_push_tail(queue, to);
5178 while (!g_queue_is_empty(queue)) {
5179 BlockDriverState *v = g_queue_pop_head(queue);
5182 QLIST_FOREACH(c2, &v->children, next) {
5188 if (g_hash_table_contains(found, c2->bs)) {
5192 g_queue_push_tail(queue, c2->bs);
5193 g_hash_table_add(found, c2->bs);
5197 g_queue_free(queue);
5198 g_hash_table_destroy(found);
5203 static void bdrv_remove_child_commit(void *opaque)
5205 GLOBAL_STATE_CODE();
5206 bdrv_child_free(opaque);
5209 static TransactionActionDrv bdrv_remove_child_drv = {
5210 .commit = bdrv_remove_child_commit,
5213 /* Function doesn't update permissions, caller is responsible for this. */
5214 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5221 BlockDriverState *bs = child->bs;
5222 bdrv_drained_begin(bs);
5223 bdrv_replace_child_tran(child, NULL, tran);
5224 bdrv_drained_end(bs);
5227 tran_add(tran, &bdrv_remove_child_drv, child);
5230 static void undrain_on_clean_cb(void *opaque)
5232 bdrv_drained_end(opaque);
5235 static TransactionActionDrv undrain_on_clean = {
5236 .clean = undrain_on_clean_cb,
5239 static int bdrv_replace_node_noperm(BlockDriverState *from,
5240 BlockDriverState *to,
5241 bool auto_skip, Transaction *tran,
5244 BdrvChild *c, *next;
5246 GLOBAL_STATE_CODE();
5248 bdrv_drained_begin(from);
5249 bdrv_drained_begin(to);
5250 tran_add(tran, &undrain_on_clean, from);
5251 tran_add(tran, &undrain_on_clean, to);
5253 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5254 assert(c->bs == from);
5255 if (!should_update_child(c, to)) {
5259 error_setg(errp, "Should not change '%s' link to '%s'",
5260 c->name, from->node_name);
5264 error_setg(errp, "Cannot change '%s' link to '%s'",
5265 c->name, from->node_name);
5268 bdrv_replace_child_tran(c, to, tran);
5275 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5276 * if it creates a parent-child relation loop or if parent is block-job.
5278 * With auto_skip=false the error is returned if from has a parent which should
5281 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5282 * case backing link of the cow-parent of @to is removed.
5284 static int bdrv_replace_node_common(BlockDriverState *from,
5285 BlockDriverState *to,
5286 bool auto_skip, bool detach_subchain,
5289 Transaction *tran = tran_new();
5290 g_autoptr(GSList) refresh_list = NULL;
5291 BlockDriverState *to_cow_parent = NULL;
5294 GLOBAL_STATE_CODE();
5296 if (detach_subchain) {
5297 assert(bdrv_chain_contains(from, to));
5299 for (to_cow_parent = from;
5300 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5301 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5307 /* Make sure that @from doesn't go away until we have successfully attached
5308 * all of its parents to @to. */
5311 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5312 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5313 bdrv_drained_begin(from);
5316 * Do the replacement without permission update.
5317 * Replacement may influence the permissions, we should calculate new
5318 * permissions based on new graph. If we fail, we'll roll-back the
5321 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5326 if (detach_subchain) {
5327 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5330 refresh_list = g_slist_prepend(refresh_list, to);
5331 refresh_list = g_slist_prepend(refresh_list, from);
5333 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5341 tran_finalize(tran, ret);
5343 bdrv_drained_end(from);
5349 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5352 GLOBAL_STATE_CODE();
5354 return bdrv_replace_node_common(from, to, true, false, errp);
5357 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5359 GLOBAL_STATE_CODE();
5361 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5366 * Add new bs contents at the top of an image chain while the chain is
5367 * live, while keeping required fields on the top layer.
5369 * This will modify the BlockDriverState fields, and swap contents
5370 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5372 * bs_new must not be attached to a BlockBackend and must not have backing
5375 * This function does not create any image files.
5377 * The caller must hold the AioContext lock for @bs_top.
5379 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5384 Transaction *tran = tran_new();
5385 AioContext *old_context, *new_context = NULL;
5387 GLOBAL_STATE_CODE();
5389 assert(!bs_new->backing);
5391 old_context = bdrv_get_aio_context(bs_top);
5393 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5394 &child_of_bds, bdrv_backing_role(bs_new),
5402 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5403 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5404 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5405 * that assumes the new lock is taken.
5407 new_context = bdrv_get_aio_context(bs_top);
5409 if (old_context != new_context) {
5410 aio_context_release(old_context);
5411 aio_context_acquire(new_context);
5414 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5419 ret = bdrv_refresh_perms(bs_new, tran, errp);
5421 tran_finalize(tran, ret);
5423 bdrv_graph_rdlock_main_loop();
5424 bdrv_refresh_limits(bs_top, NULL, NULL);
5425 bdrv_graph_rdunlock_main_loop();
5427 if (new_context && old_context != new_context) {
5428 aio_context_release(new_context);
5429 aio_context_acquire(old_context);
5435 /* Not for empty child */
5436 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5440 Transaction *tran = tran_new();
5441 g_autoptr(GSList) refresh_list = NULL;
5442 BlockDriverState *old_bs = child->bs;
5444 GLOBAL_STATE_CODE();
5447 bdrv_drained_begin(old_bs);
5448 bdrv_drained_begin(new_bs);
5450 bdrv_replace_child_tran(child, new_bs, tran);
5452 refresh_list = g_slist_prepend(refresh_list, old_bs);
5453 refresh_list = g_slist_prepend(refresh_list, new_bs);
5455 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5457 tran_finalize(tran, ret);
5459 bdrv_drained_end(old_bs);
5460 bdrv_drained_end(new_bs);
5466 static void bdrv_delete(BlockDriverState *bs)
5468 assert(bdrv_op_blocker_is_empty(bs));
5469 assert(!bs->refcnt);
5470 GLOBAL_STATE_CODE();
5472 /* remove from list, if necessary */
5473 if (bs->node_name[0] != '\0') {
5474 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5476 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5480 qemu_mutex_destroy(&bs->reqs_lock);
5487 * Replace @bs by newly created block node.
5489 * @options is a QDict of options to pass to the block drivers, or NULL for an
5490 * empty set of options. The reference to the QDict belongs to the block layer
5491 * after the call (even on failure), so if the caller intends to reuse the
5492 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5494 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5495 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5496 * different AioContext.
5498 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5499 int flags, Error **errp)
5503 AioContext *ctx = bdrv_get_aio_context(bs);
5504 BlockDriverState *new_node_bs = NULL;
5505 const char *drvname, *node_name;
5508 drvname = qdict_get_try_str(options, "driver");
5510 error_setg(errp, "driver is not specified");
5514 drv = bdrv_find_format(drvname);
5516 error_setg(errp, "Unknown driver: '%s'", drvname);
5520 node_name = qdict_get_try_str(options, "node-name");
5522 GLOBAL_STATE_CODE();
5524 aio_context_release(ctx);
5525 aio_context_acquire(qemu_get_aio_context());
5526 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5528 aio_context_release(qemu_get_aio_context());
5529 aio_context_acquire(ctx);
5530 assert(bdrv_get_aio_context(bs) == ctx);
5532 options = NULL; /* bdrv_new_open_driver() eats options */
5534 error_prepend(errp, "Could not create node: ");
5538 bdrv_drained_begin(bs);
5539 ret = bdrv_replace_node(bs, new_node_bs, errp);
5540 bdrv_drained_end(bs);
5543 error_prepend(errp, "Could not replace node: ");
5550 qobject_unref(options);
5551 bdrv_unref(new_node_bs);
5556 * Run consistency checks on an image
5558 * Returns 0 if the check could be completed (it doesn't mean that the image is
5559 * free of errors) or -errno when an internal error occurred. The results of the
5560 * check are stored in res.
5562 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5563 BdrvCheckResult *res, BdrvCheckMode fix)
5566 assert_bdrv_graph_readable();
5567 if (bs->drv == NULL) {
5570 if (bs->drv->bdrv_co_check == NULL) {
5574 memset(res, 0, sizeof(*res));
5575 return bs->drv->bdrv_co_check(bs, res, fix);
5581 * -EINVAL - backing format specified, but no file
5582 * -ENOSPC - can't update the backing file because no space is left in the
5584 * -ENOTSUP - format driver doesn't support changing the backing file
5586 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5587 const char *backing_fmt, bool require)
5589 BlockDriver *drv = bs->drv;
5592 GLOBAL_STATE_CODE();
5598 /* Backing file format doesn't make sense without a backing file */
5599 if (backing_fmt && !backing_file) {
5603 if (require && backing_file && !backing_fmt) {
5607 if (drv->bdrv_change_backing_file != NULL) {
5608 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5614 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5615 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5616 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5617 backing_file ?: "");
5623 * Finds the first non-filter node above bs in the chain between
5624 * active and bs. The returned node is either an immediate parent of
5625 * bs, or there are only filter nodes between the two.
5627 * Returns NULL if bs is not found in active's image chain,
5628 * or if active == bs.
5630 * Returns the bottommost base image if bs == NULL.
5632 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5633 BlockDriverState *bs)
5636 GLOBAL_STATE_CODE();
5638 bs = bdrv_skip_filters(bs);
5639 active = bdrv_skip_filters(active);
5642 BlockDriverState *next = bdrv_backing_chain_next(active);
5652 /* Given a BDS, searches for the base layer. */
5653 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5655 GLOBAL_STATE_CODE();
5657 return bdrv_find_overlay(bs, NULL);
5661 * Return true if at least one of the COW (backing) and filter links
5662 * between @bs and @base is frozen. @errp is set if that's the case.
5663 * @base must be reachable from @bs, or NULL.
5665 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5668 BlockDriverState *i;
5671 GLOBAL_STATE_CODE();
5673 for (i = bs; i != base; i = child_bs(child)) {
5674 child = bdrv_filter_or_cow_child(i);
5676 if (child && child->frozen) {
5677 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5678 child->name, i->node_name, child->bs->node_name);
5687 * Freeze all COW (backing) and filter links between @bs and @base.
5688 * If any of the links is already frozen the operation is aborted and
5689 * none of the links are modified.
5690 * @base must be reachable from @bs, or NULL.
5691 * Returns 0 on success. On failure returns < 0 and sets @errp.
5693 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5696 BlockDriverState *i;
5699 GLOBAL_STATE_CODE();
5701 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5705 for (i = bs; i != base; i = child_bs(child)) {
5706 child = bdrv_filter_or_cow_child(i);
5707 if (child && child->bs->never_freeze) {
5708 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5709 child->name, child->bs->node_name);
5714 for (i = bs; i != base; i = child_bs(child)) {
5715 child = bdrv_filter_or_cow_child(i);
5717 child->frozen = true;
5725 * Unfreeze all COW (backing) and filter links between @bs and @base.
5726 * The caller must ensure that all links are frozen before using this
5728 * @base must be reachable from @bs, or NULL.
5730 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5732 BlockDriverState *i;
5735 GLOBAL_STATE_CODE();
5737 for (i = bs; i != base; i = child_bs(child)) {
5738 child = bdrv_filter_or_cow_child(i);
5740 assert(child->frozen);
5741 child->frozen = false;
5747 * Drops images above 'base' up to and including 'top', and sets the image
5748 * above 'top' to have base as its backing file.
5750 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5751 * information in 'bs' can be properly updated.
5753 * E.g., this will convert the following chain:
5754 * bottom <- base <- intermediate <- top <- active
5758 * bottom <- base <- active
5760 * It is allowed for bottom==base, in which case it converts:
5762 * base <- intermediate <- top <- active
5768 * If backing_file_str is non-NULL, it will be used when modifying top's
5769 * overlay image metadata.
5772 * if active == top, that is considered an error
5775 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5776 const char *backing_file_str)
5778 BlockDriverState *explicit_top = top;
5779 bool update_inherits_from;
5781 Error *local_err = NULL;
5783 g_autoptr(GSList) updated_children = NULL;
5786 GLOBAL_STATE_CODE();
5789 bdrv_drained_begin(base);
5791 if (!top->drv || !base->drv) {
5795 /* Make sure that base is in the backing chain of top */
5796 if (!bdrv_chain_contains(top, base)) {
5800 /* If 'base' recursively inherits from 'top' then we should set
5801 * base->inherits_from to top->inherits_from after 'top' and all
5802 * other intermediate nodes have been dropped.
5803 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5804 * it because no one inherits from it. We use explicit_top for that. */
5805 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5806 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5808 /* success - we can delete the intermediate states, and link top->base */
5809 if (!backing_file_str) {
5810 bdrv_refresh_filename(base);
5811 backing_file_str = base->filename;
5814 QLIST_FOREACH(c, &top->parents, next_parent) {
5815 updated_children = g_slist_prepend(updated_children, c);
5819 * It seems correct to pass detach_subchain=true here, but it triggers
5820 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5821 * another drained section, which modify the graph (for example, removing
5822 * the child, which we keep in updated_children list). So, it's a TODO.
5824 * Note, bug triggered if pass detach_subchain=true here and run
5825 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5828 bdrv_replace_node_common(top, base, false, false, &local_err);
5830 error_report_err(local_err);
5834 for (p = updated_children; p; p = p->next) {
5837 if (c->klass->update_filename) {
5838 ret = c->klass->update_filename(c, base, backing_file_str,
5842 * TODO: Actually, we want to rollback all previous iterations
5843 * of this loop, and (which is almost impossible) previous
5844 * bdrv_replace_node()...
5846 * Note, that c->klass->update_filename may lead to permission
5847 * update, so it's a bad idea to call it inside permission
5848 * update transaction of bdrv_replace_node.
5850 error_report_err(local_err);
5856 if (update_inherits_from) {
5857 base->inherits_from = explicit_top->inherits_from;
5862 bdrv_drained_end(base);
5868 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5869 * sums the size of all data-bearing children. (This excludes backing
5872 static int64_t coroutine_fn GRAPH_RDLOCK
5873 bdrv_sum_allocated_file_size(BlockDriverState *bs)
5876 int64_t child_size, sum = 0;
5878 QLIST_FOREACH(child, &bs->children, next) {
5879 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5880 BDRV_CHILD_FILTERED))
5882 child_size = bdrv_co_get_allocated_file_size(child->bs);
5883 if (child_size < 0) {
5894 * Length of a allocated file in bytes. Sparse files are counted by actual
5895 * allocated space. Return < 0 if error or unknown.
5897 int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
5899 BlockDriver *drv = bs->drv;
5901 assert_bdrv_graph_readable();
5906 if (drv->bdrv_co_get_allocated_file_size) {
5907 return drv->bdrv_co_get_allocated_file_size(bs);
5910 if (drv->bdrv_file_open) {
5912 * Protocol drivers default to -ENOTSUP (most of their data is
5913 * not stored in any of their children (if they even have any),
5914 * so there is no generic way to figure it out).
5917 } else if (drv->is_filter) {
5918 /* Filter drivers default to the size of their filtered child */
5919 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
5921 /* Other drivers default to summing their children's sizes */
5922 return bdrv_sum_allocated_file_size(bs);
5928 * @drv: Format driver
5929 * @opts: Creation options for new image
5930 * @in_bs: Existing image containing data for new image (may be NULL)
5931 * @errp: Error object
5932 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5935 * Calculate file size required to create a new image.
5937 * If @in_bs is given then space for allocated clusters and zero clusters
5938 * from that image are included in the calculation. If @opts contains a
5939 * backing file that is shared by @in_bs then backing clusters may be omitted
5940 * from the calculation.
5942 * If @in_bs is NULL then the calculation includes no allocated clusters
5943 * unless a preallocation option is given in @opts.
5945 * Note that @in_bs may use a different BlockDriver from @drv.
5947 * If an error occurs the @errp pointer is set.
5949 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5950 BlockDriverState *in_bs, Error **errp)
5953 if (!drv->bdrv_measure) {
5954 error_setg(errp, "Block driver '%s' does not support size measurement",
5959 return drv->bdrv_measure(opts, in_bs, errp);
5963 * Return number of sectors on success, -errno on error.
5965 int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
5967 BlockDriver *drv = bs->drv;
5969 assert_bdrv_graph_readable();
5974 if (bs->bl.has_variable_length) {
5975 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
5980 return bs->total_sectors;
5984 * This wrapper is written by hand because this function is in the hot I/O path,
5985 * via blk_get_geometry.
5987 int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
5989 BlockDriver *drv = bs->drv;
5995 if (bs->bl.has_variable_length) {
5996 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6002 return bs->total_sectors;
6006 * Return length in bytes on success, -errno on error.
6007 * The length is always a multiple of BDRV_SECTOR_SIZE.
6009 int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
6013 assert_bdrv_graph_readable();
6015 ret = bdrv_co_nb_sectors(bs);
6019 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6022 return ret * BDRV_SECTOR_SIZE;
6025 bool bdrv_is_sg(BlockDriverState *bs)
6032 * Return whether the given node supports compressed writes.
6034 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6036 BlockDriverState *filtered;
6039 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6043 filtered = bdrv_filter_bs(bs);
6046 * Filters can only forward compressed writes, so we have to
6049 return bdrv_supports_compressed_writes(filtered);
6055 const char *bdrv_get_format_name(BlockDriverState *bs)
6058 return bs->drv ? bs->drv->format_name : NULL;
6061 static int qsort_strcmp(const void *a, const void *b)
6063 return strcmp(*(char *const *)a, *(char *const *)b);
6066 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
6067 void *opaque, bool read_only)
6072 const char **formats = NULL;
6074 GLOBAL_STATE_CODE();
6076 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6077 if (drv->format_name) {
6081 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6085 while (formats && i && !found) {
6086 found = !strcmp(formats[--i], drv->format_name);
6090 formats = g_renew(const char *, formats, count + 1);
6091 formats[count++] = drv->format_name;
6096 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6097 const char *format_name = block_driver_modules[i].format_name;
6103 if (use_bdrv_whitelist &&
6104 !bdrv_format_is_whitelisted(format_name, read_only)) {
6108 while (formats && j && !found) {
6109 found = !strcmp(formats[--j], format_name);
6113 formats = g_renew(const char *, formats, count + 1);
6114 formats[count++] = format_name;
6119 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6121 for (i = 0; i < count; i++) {
6122 it(opaque, formats[i]);
6128 /* This function is to find a node in the bs graph */
6129 BlockDriverState *bdrv_find_node(const char *node_name)
6131 BlockDriverState *bs;
6134 GLOBAL_STATE_CODE();
6136 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6137 if (!strcmp(node_name, bs->node_name)) {
6144 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6145 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6148 BlockDeviceInfoList *list;
6149 BlockDriverState *bs;
6151 GLOBAL_STATE_CODE();
6154 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6155 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6157 qapi_free_BlockDeviceInfoList(list);
6160 QAPI_LIST_PREPEND(list, info);
6166 typedef struct XDbgBlockGraphConstructor {
6167 XDbgBlockGraph *graph;
6168 GHashTable *graph_nodes;
6169 } XDbgBlockGraphConstructor;
6171 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6173 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6175 gr->graph = g_new0(XDbgBlockGraph, 1);
6176 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6181 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6183 XDbgBlockGraph *graph = gr->graph;
6185 g_hash_table_destroy(gr->graph_nodes);
6191 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6193 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6200 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6201 * answer of g_hash_table_lookup.
6203 ret = g_hash_table_size(gr->graph_nodes) + 1;
6204 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6209 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6210 XDbgBlockGraphNodeType type, const char *name)
6212 XDbgBlockGraphNode *n;
6214 n = g_new0(XDbgBlockGraphNode, 1);
6216 n->id = xdbg_graph_node_num(gr, node);
6218 n->name = g_strdup(name);
6220 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6223 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6224 const BdrvChild *child)
6226 BlockPermission qapi_perm;
6227 XDbgBlockGraphEdge *edge;
6228 GLOBAL_STATE_CODE();
6230 edge = g_new0(XDbgBlockGraphEdge, 1);
6232 edge->parent = xdbg_graph_node_num(gr, parent);
6233 edge->child = xdbg_graph_node_num(gr, child->bs);
6234 edge->name = g_strdup(child->name);
6236 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6237 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6239 if (flag & child->perm) {
6240 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6242 if (flag & child->shared_perm) {
6243 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6247 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6251 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6255 BlockDriverState *bs;
6257 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6259 GLOBAL_STATE_CODE();
6261 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6262 char *allocated_name = NULL;
6263 const char *name = blk_name(blk);
6266 name = allocated_name = blk_get_attached_dev_id(blk);
6268 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6270 g_free(allocated_name);
6271 if (blk_root(blk)) {
6272 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6276 WITH_JOB_LOCK_GUARD() {
6277 for (job = block_job_next_locked(NULL); job;
6278 job = block_job_next_locked(job)) {
6281 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6283 for (el = job->nodes; el; el = el->next) {
6284 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6289 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6290 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6292 QLIST_FOREACH(child, &bs->children, next) {
6293 xdbg_graph_add_edge(gr, bs, child);
6297 return xdbg_graph_finalize(gr);
6300 BlockDriverState *bdrv_lookup_bs(const char *device,
6301 const char *node_name,
6305 BlockDriverState *bs;
6307 GLOBAL_STATE_CODE();
6310 blk = blk_by_name(device);
6315 error_setg(errp, "Device '%s' has no medium", device);
6323 bs = bdrv_find_node(node_name);
6330 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6331 device ? device : "",
6332 node_name ? node_name : "");
6336 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6337 * return false. If either argument is NULL, return false. */
6338 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6341 GLOBAL_STATE_CODE();
6343 while (top && top != base) {
6344 top = bdrv_filter_or_cow_bs(top);
6350 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6352 GLOBAL_STATE_CODE();
6354 return QTAILQ_FIRST(&graph_bdrv_states);
6356 return QTAILQ_NEXT(bs, node_list);
6359 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6361 GLOBAL_STATE_CODE();
6363 return QTAILQ_FIRST(&all_bdrv_states);
6365 return QTAILQ_NEXT(bs, bs_list);
6368 const char *bdrv_get_node_name(const BlockDriverState *bs)
6371 return bs->node_name;
6374 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6380 /* If multiple parents have a name, just pick the first one. */
6381 QLIST_FOREACH(c, &bs->parents, next_parent) {
6382 if (c->klass->get_name) {
6383 name = c->klass->get_name(c);
6384 if (name && *name) {
6393 /* TODO check what callers really want: bs->node_name or blk_name() */
6394 const char *bdrv_get_device_name(const BlockDriverState *bs)
6397 return bdrv_get_parent_name(bs) ?: "";
6400 /* This can be used to identify nodes that might not have a device
6401 * name associated. Since node and device names live in the same
6402 * namespace, the result is unambiguous. The exception is if both are
6403 * absent, then this returns an empty (non-null) string. */
6404 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6407 return bdrv_get_parent_name(bs) ?: bs->node_name;
6410 int bdrv_get_flags(BlockDriverState *bs)
6413 return bs->open_flags;
6416 int bdrv_has_zero_init_1(BlockDriverState *bs)
6418 GLOBAL_STATE_CODE();
6422 int bdrv_has_zero_init(BlockDriverState *bs)
6424 BlockDriverState *filtered;
6425 GLOBAL_STATE_CODE();
6431 /* If BS is a copy on write image, it is initialized to
6432 the contents of the base image, which may not be zeroes. */
6433 if (bdrv_cow_child(bs)) {
6436 if (bs->drv->bdrv_has_zero_init) {
6437 return bs->drv->bdrv_has_zero_init(bs);
6440 filtered = bdrv_filter_bs(bs);
6442 return bdrv_has_zero_init(filtered);
6449 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6452 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6456 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6459 void bdrv_get_backing_filename(BlockDriverState *bs,
6460 char *filename, int filename_size)
6463 pstrcpy(filename, filename_size, bs->backing_file);
6466 int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6469 BlockDriver *drv = bs->drv;
6471 assert_bdrv_graph_readable();
6473 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6477 if (!drv->bdrv_co_get_info) {
6478 BlockDriverState *filtered = bdrv_filter_bs(bs);
6480 return bdrv_co_get_info(filtered, bdi);
6484 memset(bdi, 0, sizeof(*bdi));
6485 ret = drv->bdrv_co_get_info(bs, bdi);
6486 if (bdi->subcluster_size == 0) {
6488 * If the driver left this unset, subclusters are not supported.
6489 * Then it is safe to treat each cluster as having only one subcluster.
6491 bdi->subcluster_size = bdi->cluster_size;
6497 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6504 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6507 BlockDriver *drv = bs->drv;
6509 if (drv && drv->bdrv_get_specific_info) {
6510 return drv->bdrv_get_specific_info(bs, errp);
6515 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6517 BlockDriver *drv = bs->drv;
6519 if (!drv || !drv->bdrv_get_specific_stats) {
6522 return drv->bdrv_get_specific_stats(bs);
6525 void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6528 assert_bdrv_graph_readable();
6530 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
6534 bs->drv->bdrv_co_debug_event(bs, event);
6537 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6539 GLOBAL_STATE_CODE();
6540 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6541 bs = bdrv_primary_bs(bs);
6544 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6545 assert(bs->drv->bdrv_debug_remove_breakpoint);
6552 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6555 GLOBAL_STATE_CODE();
6556 bs = bdrv_find_debug_node(bs);
6558 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6564 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6566 GLOBAL_STATE_CODE();
6567 bs = bdrv_find_debug_node(bs);
6569 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6575 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6577 GLOBAL_STATE_CODE();
6578 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6579 bs = bdrv_primary_bs(bs);
6582 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6583 return bs->drv->bdrv_debug_resume(bs, tag);
6589 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6591 GLOBAL_STATE_CODE();
6592 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6593 bs = bdrv_primary_bs(bs);
6596 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6597 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6603 /* backing_file can either be relative, or absolute, or a protocol. If it is
6604 * relative, it must be relative to the chain. So, passing in bs->filename
6605 * from a BDS as backing_file should not be done, as that may be relative to
6606 * the CWD rather than the chain. */
6607 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6608 const char *backing_file)
6610 char *filename_full = NULL;
6611 char *backing_file_full = NULL;
6612 char *filename_tmp = NULL;
6613 int is_protocol = 0;
6614 bool filenames_refreshed = false;
6615 BlockDriverState *curr_bs = NULL;
6616 BlockDriverState *retval = NULL;
6617 BlockDriverState *bs_below;
6619 GLOBAL_STATE_CODE();
6621 if (!bs || !bs->drv || !backing_file) {
6625 filename_full = g_malloc(PATH_MAX);
6626 backing_file_full = g_malloc(PATH_MAX);
6628 is_protocol = path_has_protocol(backing_file);
6631 * Being largely a legacy function, skip any filters here
6632 * (because filters do not have normal filenames, so they cannot
6633 * match anyway; and allowing json:{} filenames is a bit out of
6636 for (curr_bs = bdrv_skip_filters(bs);
6637 bdrv_cow_child(curr_bs) != NULL;
6640 bs_below = bdrv_backing_chain_next(curr_bs);
6642 if (bdrv_backing_overridden(curr_bs)) {
6644 * If the backing file was overridden, we can only compare
6645 * directly against the backing node's filename.
6648 if (!filenames_refreshed) {
6650 * This will automatically refresh all of the
6651 * filenames in the rest of the backing chain, so we
6652 * only need to do this once.
6654 bdrv_refresh_filename(bs_below);
6655 filenames_refreshed = true;
6658 if (strcmp(backing_file, bs_below->filename) == 0) {
6662 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6664 * If either of the filename paths is actually a protocol, then
6665 * compare unmodified paths; otherwise make paths relative.
6667 char *backing_file_full_ret;
6669 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6673 /* Also check against the full backing filename for the image */
6674 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6676 if (backing_file_full_ret) {
6677 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6678 g_free(backing_file_full_ret);
6685 /* If not an absolute filename path, make it relative to the current
6686 * image's filename path */
6687 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6689 /* We are going to compare canonicalized absolute pathnames */
6690 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6691 g_free(filename_tmp);
6694 g_free(filename_tmp);
6696 /* We need to make sure the backing filename we are comparing against
6697 * is relative to the current image filename (or absolute) */
6698 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6699 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6700 g_free(filename_tmp);
6703 g_free(filename_tmp);
6705 if (strcmp(backing_file_full, filename_full) == 0) {
6712 g_free(filename_full);
6713 g_free(backing_file_full);
6717 void bdrv_init(void)
6719 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6720 use_bdrv_whitelist = 1;
6722 module_call_init(MODULE_INIT_BLOCK);
6725 void bdrv_init_with_whitelist(void)
6727 use_bdrv_whitelist = 1;
6731 int bdrv_activate(BlockDriverState *bs, Error **errp)
6733 BdrvChild *child, *parent;
6734 Error *local_err = NULL;
6736 BdrvDirtyBitmap *bm;
6738 GLOBAL_STATE_CODE();
6744 QLIST_FOREACH(child, &bs->children, next) {
6745 bdrv_activate(child->bs, &local_err);
6747 error_propagate(errp, local_err);
6753 * Update permissions, they may differ for inactive nodes.
6755 * Note that the required permissions of inactive images are always a
6756 * subset of the permissions required after activating the image. This
6757 * allows us to just get the permissions upfront without restricting
6758 * bdrv_co_invalidate_cache().
6760 * It also means that in error cases, we don't have to try and revert to
6761 * the old permissions (which is an operation that could fail, too). We can
6762 * just keep the extended permissions for the next time that an activation
6763 * of the image is tried.
6765 if (bs->open_flags & BDRV_O_INACTIVE) {
6766 bs->open_flags &= ~BDRV_O_INACTIVE;
6767 ret = bdrv_refresh_perms(bs, NULL, errp);
6769 bs->open_flags |= BDRV_O_INACTIVE;
6773 ret = bdrv_invalidate_cache(bs, errp);
6775 bs->open_flags |= BDRV_O_INACTIVE;
6779 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6780 bdrv_dirty_bitmap_skip_store(bm, false);
6783 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6785 bs->open_flags |= BDRV_O_INACTIVE;
6786 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6791 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6792 if (parent->klass->activate) {
6793 parent->klass->activate(parent, &local_err);
6795 bs->open_flags |= BDRV_O_INACTIVE;
6796 error_propagate(errp, local_err);
6805 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6807 Error *local_err = NULL;
6810 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6811 assert_bdrv_graph_readable();
6813 if (bs->drv->bdrv_co_invalidate_cache) {
6814 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6816 error_propagate(errp, local_err);
6824 void bdrv_activate_all(Error **errp)
6826 BlockDriverState *bs;
6827 BdrvNextIterator it;
6829 GLOBAL_STATE_CODE();
6831 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6832 AioContext *aio_context = bdrv_get_aio_context(bs);
6835 aio_context_acquire(aio_context);
6836 ret = bdrv_activate(bs, errp);
6837 aio_context_release(aio_context);
6839 bdrv_next_cleanup(&it);
6845 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6848 GLOBAL_STATE_CODE();
6850 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6851 if (parent->klass->parent_is_bds) {
6852 BlockDriverState *parent_bs = parent->opaque;
6853 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6862 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6864 BdrvChild *child, *parent;
6866 uint64_t cumulative_perms, cumulative_shared_perms;
6868 GLOBAL_STATE_CODE();
6874 /* Make sure that we don't inactivate a child before its parent.
6875 * It will be covered by recursion from the yet active parent. */
6876 if (bdrv_has_bds_parent(bs, true)) {
6880 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6882 /* Inactivate this node */
6883 if (bs->drv->bdrv_inactivate) {
6884 ret = bs->drv->bdrv_inactivate(bs);
6890 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6891 if (parent->klass->inactivate) {
6892 ret = parent->klass->inactivate(parent);
6899 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6900 &cumulative_shared_perms);
6901 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6902 /* Our inactive parents still need write access. Inactivation failed. */
6906 bs->open_flags |= BDRV_O_INACTIVE;
6909 * Update permissions, they may differ for inactive nodes.
6910 * We only tried to loosen restrictions, so errors are not fatal, ignore
6913 bdrv_refresh_perms(bs, NULL, NULL);
6915 /* Recursively inactivate children */
6916 QLIST_FOREACH(child, &bs->children, next) {
6917 ret = bdrv_inactivate_recurse(child->bs);
6926 int bdrv_inactivate_all(void)
6928 BlockDriverState *bs = NULL;
6929 BdrvNextIterator it;
6931 GSList *aio_ctxs = NULL, *ctx;
6933 GLOBAL_STATE_CODE();
6935 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6936 AioContext *aio_context = bdrv_get_aio_context(bs);
6938 if (!g_slist_find(aio_ctxs, aio_context)) {
6939 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6940 aio_context_acquire(aio_context);
6944 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6945 /* Nodes with BDS parents are covered by recursion from the last
6946 * parent that gets inactivated. Don't inactivate them a second
6947 * time if that has already happened. */
6948 if (bdrv_has_bds_parent(bs, false)) {
6951 ret = bdrv_inactivate_recurse(bs);
6953 bdrv_next_cleanup(&it);
6959 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6960 AioContext *aio_context = ctx->data;
6961 aio_context_release(aio_context);
6963 g_slist_free(aio_ctxs);
6968 /**************************************************************/
6969 /* removable device support */
6972 * Return TRUE if the media is present
6974 bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
6976 BlockDriver *drv = bs->drv;
6979 assert_bdrv_graph_readable();
6984 if (drv->bdrv_co_is_inserted) {
6985 return drv->bdrv_co_is_inserted(bs);
6987 QLIST_FOREACH(child, &bs->children, next) {
6988 if (!bdrv_co_is_inserted(child->bs)) {
6996 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6998 void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
7000 BlockDriver *drv = bs->drv;
7002 assert_bdrv_graph_readable();
7004 if (drv && drv->bdrv_co_eject) {
7005 drv->bdrv_co_eject(bs, eject_flag);
7010 * Lock or unlock the media (if it is locked, the user won't be able
7011 * to eject it manually).
7013 void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
7015 BlockDriver *drv = bs->drv;
7017 assert_bdrv_graph_readable();
7018 trace_bdrv_lock_medium(bs, locked);
7020 if (drv && drv->bdrv_co_lock_medium) {
7021 drv->bdrv_co_lock_medium(bs, locked);
7025 /* Get a reference to bs */
7026 void bdrv_ref(BlockDriverState *bs)
7028 GLOBAL_STATE_CODE();
7032 /* Release a previously grabbed reference to bs.
7033 * If after releasing, reference count is zero, the BlockDriverState is
7035 void bdrv_unref(BlockDriverState *bs)
7037 GLOBAL_STATE_CODE();
7041 assert(bs->refcnt > 0);
7042 if (--bs->refcnt == 0) {
7048 * Release a BlockDriverState reference while holding the graph write lock.
7050 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7051 * because bdrv_close() both involves polling and taking the graph lock
7052 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7053 * possibly closing @bs until the graph lock is released.
7055 void bdrv_schedule_unref(BlockDriverState *bs)
7060 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7061 (QEMUBHFunc *) bdrv_unref, bs);
7064 struct BdrvOpBlocker {
7066 QLIST_ENTRY(BdrvOpBlocker) list;
7069 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7071 BdrvOpBlocker *blocker;
7072 GLOBAL_STATE_CODE();
7073 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7074 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7075 blocker = QLIST_FIRST(&bs->op_blockers[op]);
7076 error_propagate_prepend(errp, error_copy(blocker->reason),
7077 "Node '%s' is busy: ",
7078 bdrv_get_device_or_node_name(bs));
7084 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7086 BdrvOpBlocker *blocker;
7087 GLOBAL_STATE_CODE();
7088 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7090 blocker = g_new0(BdrvOpBlocker, 1);
7091 blocker->reason = reason;
7092 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7095 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7097 BdrvOpBlocker *blocker, *next;
7098 GLOBAL_STATE_CODE();
7099 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7100 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7101 if (blocker->reason == reason) {
7102 QLIST_REMOVE(blocker, list);
7108 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7111 GLOBAL_STATE_CODE();
7112 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7113 bdrv_op_block(bs, i, reason);
7117 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7120 GLOBAL_STATE_CODE();
7121 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7122 bdrv_op_unblock(bs, i, reason);
7126 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7129 GLOBAL_STATE_CODE();
7130 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7131 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7139 * Must not be called while holding the lock of an AioContext other than the
7142 void bdrv_img_create(const char *filename, const char *fmt,
7143 const char *base_filename, const char *base_fmt,
7144 char *options, uint64_t img_size, int flags, bool quiet,
7147 QemuOptsList *create_opts = NULL;
7148 QemuOpts *opts = NULL;
7149 const char *backing_fmt, *backing_file;
7151 BlockDriver *drv, *proto_drv;
7152 Error *local_err = NULL;
7155 GLOBAL_STATE_CODE();
7157 /* Find driver and parse its options */
7158 drv = bdrv_find_format(fmt);
7160 error_setg(errp, "Unknown file format '%s'", fmt);
7164 proto_drv = bdrv_find_protocol(filename, true, errp);
7169 if (!drv->create_opts) {
7170 error_setg(errp, "Format driver '%s' does not support image creation",
7175 if (!proto_drv->create_opts) {
7176 error_setg(errp, "Protocol driver '%s' does not support image creation",
7177 proto_drv->format_name);
7181 aio_context_acquire(qemu_get_aio_context());
7183 /* Create parameter list */
7184 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7185 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7187 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7189 /* Parse -o options */
7191 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7196 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7197 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7198 } else if (img_size != UINT64_C(-1)) {
7199 error_setg(errp, "The image size must be specified only once");
7203 if (base_filename) {
7204 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7206 error_setg(errp, "Backing file not supported for file format '%s'",
7213 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7214 error_setg(errp, "Backing file format not supported for file "
7215 "format '%s'", fmt);
7220 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7222 if (!strcmp(filename, backing_file)) {
7223 error_setg(errp, "Error: Trying to create an image with the "
7224 "same filename as the backing file");
7227 if (backing_file[0] == '\0') {
7228 error_setg(errp, "Expected backing file name, got empty string");
7233 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7235 /* The size for the image must always be specified, unless we have a backing
7236 * file and we have not been forbidden from opening it. */
7237 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7238 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7239 BlockDriverState *bs;
7242 QDict *backing_options = NULL;
7245 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7250 assert(full_backing);
7253 * No need to do I/O here, which allows us to open encrypted
7254 * backing images without needing the secret
7257 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7258 back_flags |= BDRV_O_NO_IO;
7260 backing_options = qdict_new();
7262 qdict_put_str(backing_options, "driver", backing_fmt);
7264 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7266 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7268 g_free(full_backing);
7270 error_append_hint(&local_err, "Could not open backing image.\n");
7274 error_setg(&local_err,
7275 "Backing file specified without backing format");
7276 error_append_hint(&local_err, "Detected format of %s.\n",
7277 bs->drv->format_name);
7281 /* Opened BS, have no size */
7282 size = bdrv_getlength(bs);
7284 error_setg_errno(errp, -size, "Could not get size of '%s'",
7289 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7293 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7294 } else if (backing_file && !backing_fmt) {
7295 error_setg(&local_err,
7296 "Backing file specified without backing format");
7301 error_setg(errp, "Image creation needs a size parameter");
7306 printf("Formatting '%s', fmt=%s ", filename, fmt);
7307 qemu_opts_print(opts, " ");
7312 ret = bdrv_create(drv, filename, opts, &local_err);
7314 if (ret == -EFBIG) {
7315 /* This is generally a better message than whatever the driver would
7316 * deliver (especially because of the cluster_size_hint), since that
7317 * is most probably not much different from "image too large". */
7318 const char *cluster_size_hint = "";
7319 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7320 cluster_size_hint = " (try using a larger cluster size)";
7322 error_setg(errp, "The image size is too large for file format '%s'"
7323 "%s", fmt, cluster_size_hint);
7324 error_free(local_err);
7329 qemu_opts_del(opts);
7330 qemu_opts_free(create_opts);
7331 error_propagate(errp, local_err);
7332 aio_context_release(qemu_get_aio_context());
7335 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7338 return bs ? bs->aio_context : qemu_get_aio_context();
7341 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7343 Coroutine *self = qemu_coroutine_self();
7344 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7345 AioContext *new_ctx;
7349 * Increase bs->in_flight to ensure that this operation is completed before
7350 * moving the node to a different AioContext. Read new_ctx only afterwards.
7352 bdrv_inc_in_flight(bs);
7354 new_ctx = bdrv_get_aio_context(bs);
7355 aio_co_reschedule_self(new_ctx);
7359 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7362 aio_co_reschedule_self(old_ctx);
7363 bdrv_dec_in_flight(bs);
7366 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7368 AioContext *ctx = bdrv_get_aio_context(bs);
7370 /* In the main thread, bs->aio_context won't change concurrently */
7371 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7374 * We're in coroutine context, so we already hold the lock of the main
7375 * loop AioContext. Don't lock it twice to avoid deadlocks.
7377 assert(qemu_in_coroutine());
7378 if (ctx != qemu_get_aio_context()) {
7379 aio_context_acquire(ctx);
7383 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7385 AioContext *ctx = bdrv_get_aio_context(bs);
7387 assert(qemu_in_coroutine());
7388 if (ctx != qemu_get_aio_context()) {
7389 aio_context_release(ctx);
7393 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7395 GLOBAL_STATE_CODE();
7396 QLIST_REMOVE(ban, list);
7400 static void bdrv_detach_aio_context(BlockDriverState *bs)
7402 BdrvAioNotifier *baf, *baf_tmp;
7404 assert(!bs->walking_aio_notifiers);
7405 GLOBAL_STATE_CODE();
7406 bs->walking_aio_notifiers = true;
7407 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7409 bdrv_do_remove_aio_context_notifier(baf);
7411 baf->detach_aio_context(baf->opaque);
7414 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7415 * remove remaining aio notifiers if we aren't called again.
7417 bs->walking_aio_notifiers = false;
7419 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7420 bs->drv->bdrv_detach_aio_context(bs);
7423 bs->aio_context = NULL;
7426 static void bdrv_attach_aio_context(BlockDriverState *bs,
7427 AioContext *new_context)
7429 BdrvAioNotifier *ban, *ban_tmp;
7430 GLOBAL_STATE_CODE();
7432 bs->aio_context = new_context;
7434 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7435 bs->drv->bdrv_attach_aio_context(bs, new_context);
7438 assert(!bs->walking_aio_notifiers);
7439 bs->walking_aio_notifiers = true;
7440 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7442 bdrv_do_remove_aio_context_notifier(ban);
7444 ban->attached_aio_context(new_context, ban->opaque);
7447 bs->walking_aio_notifiers = false;
7450 typedef struct BdrvStateSetAioContext {
7451 AioContext *new_ctx;
7452 BlockDriverState *bs;
7453 } BdrvStateSetAioContext;
7455 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7456 GHashTable *visited,
7460 GLOBAL_STATE_CODE();
7461 if (g_hash_table_contains(visited, c)) {
7464 g_hash_table_add(visited, c);
7467 * A BdrvChildClass that doesn't handle AioContext changes cannot
7468 * tolerate any AioContext changes
7470 if (!c->klass->change_aio_ctx) {
7471 char *user = bdrv_child_user_desc(c);
7472 error_setg(errp, "Changing iothreads is not supported by %s", user);
7476 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7477 assert(!errp || *errp);
7483 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7484 GHashTable *visited, Transaction *tran,
7487 GLOBAL_STATE_CODE();
7488 if (g_hash_table_contains(visited, c)) {
7491 g_hash_table_add(visited, c);
7492 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7495 static void bdrv_set_aio_context_clean(void *opaque)
7497 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7498 BlockDriverState *bs = (BlockDriverState *) state->bs;
7500 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7501 bdrv_drained_end(bs);
7506 static void bdrv_set_aio_context_commit(void *opaque)
7508 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7509 BlockDriverState *bs = (BlockDriverState *) state->bs;
7510 AioContext *new_context = state->new_ctx;
7511 AioContext *old_context = bdrv_get_aio_context(bs);
7514 * Take the old AioContex when detaching it from bs.
7515 * At this point, new_context lock is already acquired, and we are now
7516 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7517 * does not call AIO_POLL_WHILE().
7519 if (old_context != qemu_get_aio_context()) {
7520 aio_context_acquire(old_context);
7522 bdrv_detach_aio_context(bs);
7523 if (old_context != qemu_get_aio_context()) {
7524 aio_context_release(old_context);
7526 bdrv_attach_aio_context(bs, new_context);
7529 static TransactionActionDrv set_aio_context = {
7530 .commit = bdrv_set_aio_context_commit,
7531 .clean = bdrv_set_aio_context_clean,
7535 * Changes the AioContext used for fd handlers, timers, and BHs by this
7536 * BlockDriverState and all its children and parents.
7538 * Must be called from the main AioContext.
7540 * The caller must own the AioContext lock for the old AioContext of bs, but it
7541 * must not own the AioContext lock for new_context (unless new_context is the
7542 * same as the current context of bs).
7544 * @visited will accumulate all visited BdrvChild objects. The caller is
7545 * responsible for freeing the list afterwards.
7547 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7548 GHashTable *visited, Transaction *tran,
7552 BdrvStateSetAioContext *state;
7554 GLOBAL_STATE_CODE();
7556 if (bdrv_get_aio_context(bs) == ctx) {
7560 QLIST_FOREACH(c, &bs->parents, next_parent) {
7561 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7566 QLIST_FOREACH(c, &bs->children, next) {
7567 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7572 state = g_new(BdrvStateSetAioContext, 1);
7573 *state = (BdrvStateSetAioContext) {
7578 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7579 bdrv_drained_begin(bs);
7581 tran_add(tran, &set_aio_context, state);
7587 * Change bs's and recursively all of its parents' and children's AioContext
7588 * to the given new context, returning an error if that isn't possible.
7590 * If ignore_child is not NULL, that child (and its subgraph) will not
7593 * This function still requires the caller to take the bs current
7594 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7595 * assumes the lock is always held if bs is in another AioContext.
7596 * For the same reason, it temporarily also holds the new AioContext, since
7597 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7598 * Therefore the new AioContext lock must not be taken by the caller.
7600 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7601 BdrvChild *ignore_child, Error **errp)
7604 GHashTable *visited;
7606 AioContext *old_context = bdrv_get_aio_context(bs);
7607 GLOBAL_STATE_CODE();
7610 * Recursion phase: go through all nodes of the graph.
7611 * Take care of checking that all nodes support changing AioContext
7612 * and drain them, building a linear list of callbacks to run if everything
7613 * is successful (the transaction itself).
7616 visited = g_hash_table_new(NULL, NULL);
7618 g_hash_table_add(visited, ignore_child);
7620 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7621 g_hash_table_destroy(visited);
7624 * Linear phase: go through all callbacks collected in the transaction.
7625 * Run all callbacks collected in the recursion to switch all nodes
7626 * AioContext lock (transaction commit), or undo all changes done in the
7627 * recursion (transaction abort).
7631 /* Just run clean() callbacks. No AioContext changed. */
7637 * Release old AioContext, it won't be needed anymore, as all
7638 * bdrv_drained_begin() have been called already.
7640 if (qemu_get_aio_context() != old_context) {
7641 aio_context_release(old_context);
7645 * Acquire new AioContext since bdrv_drained_end() is going to be called
7646 * after we switched all nodes in the new AioContext, and the function
7647 * assumes that the lock of the bs is always taken.
7649 if (qemu_get_aio_context() != ctx) {
7650 aio_context_acquire(ctx);
7655 if (qemu_get_aio_context() != ctx) {
7656 aio_context_release(ctx);
7659 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7660 if (qemu_get_aio_context() != old_context) {
7661 aio_context_acquire(old_context);
7667 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7668 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7669 void (*detach_aio_context)(void *opaque), void *opaque)
7671 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7672 *ban = (BdrvAioNotifier){
7673 .attached_aio_context = attached_aio_context,
7674 .detach_aio_context = detach_aio_context,
7677 GLOBAL_STATE_CODE();
7679 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7682 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7683 void (*attached_aio_context)(AioContext *,
7685 void (*detach_aio_context)(void *),
7688 BdrvAioNotifier *ban, *ban_next;
7689 GLOBAL_STATE_CODE();
7691 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7692 if (ban->attached_aio_context == attached_aio_context &&
7693 ban->detach_aio_context == detach_aio_context &&
7694 ban->opaque == opaque &&
7695 ban->deleted == false)
7697 if (bs->walking_aio_notifiers) {
7698 ban->deleted = true;
7700 bdrv_do_remove_aio_context_notifier(ban);
7709 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7710 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7714 GLOBAL_STATE_CODE();
7716 error_setg(errp, "Node is ejected");
7719 if (!bs->drv->bdrv_amend_options) {
7720 error_setg(errp, "Block driver '%s' does not support option amendment",
7721 bs->drv->format_name);
7724 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7725 cb_opaque, force, errp);
7729 * This function checks whether the given @to_replace is allowed to be
7730 * replaced by a node that always shows the same data as @bs. This is
7731 * used for example to verify whether the mirror job can replace
7732 * @to_replace by the target mirrored from @bs.
7733 * To be replaceable, @bs and @to_replace may either be guaranteed to
7734 * always show the same data (because they are only connected through
7735 * filters), or some driver may allow replacing one of its children
7736 * because it can guarantee that this child's data is not visible at
7737 * all (for example, for dissenting quorum children that have no other
7740 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7741 BlockDriverState *to_replace)
7743 BlockDriverState *filtered;
7745 GLOBAL_STATE_CODE();
7747 if (!bs || !bs->drv) {
7751 if (bs == to_replace) {
7755 /* See what the driver can do */
7756 if (bs->drv->bdrv_recurse_can_replace) {
7757 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7760 /* For filters without an own implementation, we can recurse on our own */
7761 filtered = bdrv_filter_bs(bs);
7763 return bdrv_recurse_can_replace(filtered, to_replace);
7771 * Check whether the given @node_name can be replaced by a node that
7772 * has the same data as @parent_bs. If so, return @node_name's BDS;
7775 * @node_name must be a (recursive) *child of @parent_bs (or this
7776 * function will return NULL).
7778 * The result (whether the node can be replaced or not) is only valid
7779 * for as long as no graph or permission changes occur.
7781 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7782 const char *node_name, Error **errp)
7784 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7785 AioContext *aio_context;
7787 GLOBAL_STATE_CODE();
7789 if (!to_replace_bs) {
7790 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7794 aio_context = bdrv_get_aio_context(to_replace_bs);
7795 aio_context_acquire(aio_context);
7797 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7798 to_replace_bs = NULL;
7802 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7803 * most non filter in order to prevent data corruption.
7804 * Another benefit is that this tests exclude backing files which are
7805 * blocked by the backing blockers.
7807 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7808 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7809 "because it cannot be guaranteed that doing so would not "
7810 "lead to an abrupt change of visible data",
7811 node_name, parent_bs->node_name);
7812 to_replace_bs = NULL;
7817 aio_context_release(aio_context);
7818 return to_replace_bs;
7822 * Iterates through the list of runtime option keys that are said to
7823 * be "strong" for a BDS. An option is called "strong" if it changes
7824 * a BDS's data. For example, the null block driver's "size" and
7825 * "read-zeroes" options are strong, but its "latency-ns" option is
7828 * If a key returned by this function ends with a dot, all options
7829 * starting with that prefix are strong.
7831 static const char *const *strong_options(BlockDriverState *bs,
7832 const char *const *curopt)
7834 static const char *const global_options[] = {
7835 "driver", "filename", NULL
7839 return &global_options[0];
7843 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7844 curopt = bs->drv->strong_runtime_opts;
7847 return (curopt && *curopt) ? curopt : NULL;
7851 * Copies all strong runtime options from bs->options to the given
7852 * QDict. The set of strong option keys is determined by invoking
7855 * Returns true iff any strong option was present in bs->options (and
7856 * thus copied to the target QDict) with the exception of "filename"
7857 * and "driver". The caller is expected to use this value to decide
7858 * whether the existence of strong options prevents the generation of
7861 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7863 bool found_any = false;
7864 const char *const *option_name = NULL;
7870 while ((option_name = strong_options(bs, option_name))) {
7871 bool option_given = false;
7873 assert(strlen(*option_name) > 0);
7874 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7875 QObject *entry = qdict_get(bs->options, *option_name);
7880 qdict_put_obj(d, *option_name, qobject_ref(entry));
7881 option_given = true;
7883 const QDictEntry *entry;
7884 for (entry = qdict_first(bs->options); entry;
7885 entry = qdict_next(bs->options, entry))
7887 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7888 qdict_put_obj(d, qdict_entry_key(entry),
7889 qobject_ref(qdict_entry_value(entry)));
7890 option_given = true;
7895 /* While "driver" and "filename" need to be included in a JSON filename,
7896 * their existence does not prohibit generation of a plain filename. */
7897 if (!found_any && option_given &&
7898 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7904 if (!qdict_haskey(d, "driver")) {
7905 /* Drivers created with bdrv_new_open_driver() may not have a
7906 * @driver option. Add it here. */
7907 qdict_put_str(d, "driver", bs->drv->format_name);
7913 /* Note: This function may return false positives; it may return true
7914 * even if opening the backing file specified by bs's image header
7915 * would result in exactly bs->backing. */
7916 static bool bdrv_backing_overridden(BlockDriverState *bs)
7918 GLOBAL_STATE_CODE();
7920 return strcmp(bs->auto_backing_file,
7921 bs->backing->bs->filename);
7923 /* No backing BDS, so if the image header reports any backing
7924 * file, it must have been suppressed */
7925 return bs->auto_backing_file[0] != '\0';
7929 /* Updates the following BDS fields:
7930 * - exact_filename: A filename which may be used for opening a block device
7931 * which (mostly) equals the given BDS (even without any
7932 * other options; so reading and writing must return the same
7933 * results, but caching etc. may be different)
7934 * - full_open_options: Options which, when given when opening a block device
7935 * (without a filename), result in a BDS (mostly)
7936 * equalling the given one
7937 * - filename: If exact_filename is set, it is copied here. Otherwise,
7938 * full_open_options is converted to a JSON object, prefixed with
7939 * "json:" (for use through the JSON pseudo protocol) and put here.
7941 void bdrv_refresh_filename(BlockDriverState *bs)
7943 BlockDriver *drv = bs->drv;
7945 BlockDriverState *primary_child_bs;
7947 bool backing_overridden;
7948 bool generate_json_filename; /* Whether our default implementation should
7949 fill exact_filename (false) or not (true) */
7951 GLOBAL_STATE_CODE();
7957 /* This BDS's file name may depend on any of its children's file names, so
7958 * refresh those first */
7959 QLIST_FOREACH(child, &bs->children, next) {
7960 bdrv_refresh_filename(child->bs);
7964 /* For implicit nodes, just copy everything from the single child */
7965 child = QLIST_FIRST(&bs->children);
7966 assert(QLIST_NEXT(child, next) == NULL);
7968 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7969 child->bs->exact_filename);
7970 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7972 qobject_unref(bs->full_open_options);
7973 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7978 backing_overridden = bdrv_backing_overridden(bs);
7980 if (bs->open_flags & BDRV_O_NO_IO) {
7981 /* Without I/O, the backing file does not change anything.
7982 * Therefore, in such a case (primarily qemu-img), we can
7983 * pretend the backing file has not been overridden even if
7984 * it technically has been. */
7985 backing_overridden = false;
7988 /* Gather the options QDict */
7990 generate_json_filename = append_strong_runtime_options(opts, bs);
7991 generate_json_filename |= backing_overridden;
7993 if (drv->bdrv_gather_child_options) {
7994 /* Some block drivers may not want to present all of their children's
7995 * options, or name them differently from BdrvChild.name */
7996 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7998 QLIST_FOREACH(child, &bs->children, next) {
7999 if (child == bs->backing && !backing_overridden) {
8000 /* We can skip the backing BDS if it has not been overridden */
8004 qdict_put(opts, child->name,
8005 qobject_ref(child->bs->full_open_options));
8008 if (backing_overridden && !bs->backing) {
8009 /* Force no backing file */
8010 qdict_put_null(opts, "backing");
8014 qobject_unref(bs->full_open_options);
8015 bs->full_open_options = opts;
8017 primary_child_bs = bdrv_primary_bs(bs);
8019 if (drv->bdrv_refresh_filename) {
8020 /* Obsolete information is of no use here, so drop the old file name
8021 * information before refreshing it */
8022 bs->exact_filename[0] = '\0';
8024 drv->bdrv_refresh_filename(bs);
8025 } else if (primary_child_bs) {
8027 * Try to reconstruct valid information from the underlying
8028 * file -- this only works for format nodes (filter nodes
8029 * cannot be probed and as such must be selected by the user
8030 * either through an options dict, or through a special
8031 * filename which the filter driver must construct in its
8032 * .bdrv_refresh_filename() implementation).
8035 bs->exact_filename[0] = '\0';
8038 * We can use the underlying file's filename if:
8039 * - it has a filename,
8040 * - the current BDS is not a filter,
8041 * - the file is a protocol BDS, and
8042 * - opening that file (as this BDS's format) will automatically create
8043 * the BDS tree we have right now, that is:
8044 * - the user did not significantly change this BDS's behavior with
8045 * some explicit (strong) options
8046 * - no non-file child of this BDS has been overridden by the user
8047 * Both of these conditions are represented by generate_json_filename.
8049 if (primary_child_bs->exact_filename[0] &&
8050 primary_child_bs->drv->bdrv_file_open &&
8051 !drv->is_filter && !generate_json_filename)
8053 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
8057 if (bs->exact_filename[0]) {
8058 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
8060 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
8061 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
8062 json->str) >= sizeof(bs->filename)) {
8063 /* Give user a hint if we truncated things. */
8064 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8066 g_string_free(json, true);
8070 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8072 BlockDriver *drv = bs->drv;
8073 BlockDriverState *child_bs;
8075 GLOBAL_STATE_CODE();
8078 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8082 if (drv->bdrv_dirname) {
8083 return drv->bdrv_dirname(bs, errp);
8086 child_bs = bdrv_primary_bs(bs);
8088 return bdrv_dirname(child_bs, errp);
8091 bdrv_refresh_filename(bs);
8092 if (bs->exact_filename[0] != '\0') {
8093 return path_combine(bs->exact_filename, "");
8096 error_setg(errp, "Cannot generate a base directory for %s nodes",
8102 * Hot add/remove a BDS's child. So the user can take a child offline when
8103 * it is broken and take a new child online
8105 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8108 GLOBAL_STATE_CODE();
8109 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8110 error_setg(errp, "The node %s does not support adding a child",
8111 bdrv_get_device_or_node_name(parent_bs));
8116 * Non-zoned block drivers do not follow zoned storage constraints
8117 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8118 * drivers in a graph.
8120 if (!parent_bs->drv->supports_zoned_children &&
8121 child_bs->bl.zoned == BLK_Z_HM) {
8123 * The host-aware model allows zoned storage constraints and random
8124 * write. Allow mixing host-aware and non-zoned drivers. Using
8125 * host-aware device as a regular device.
8127 error_setg(errp, "Cannot add a %s child to a %s parent",
8128 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8129 parent_bs->drv->supports_zoned_children ?
8130 "support zoned children" : "not support zoned children");
8134 if (!QLIST_EMPTY(&child_bs->parents)) {
8135 error_setg(errp, "The node %s already has a parent",
8136 child_bs->node_name);
8140 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8143 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8147 GLOBAL_STATE_CODE();
8148 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8149 error_setg(errp, "The node %s does not support removing a child",
8150 bdrv_get_device_or_node_name(parent_bs));
8154 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8161 error_setg(errp, "The node %s does not have a child named %s",
8162 bdrv_get_device_or_node_name(parent_bs),
8163 bdrv_get_device_or_node_name(child->bs));
8167 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8170 int bdrv_make_empty(BdrvChild *c, Error **errp)
8172 BlockDriver *drv = c->bs->drv;
8175 GLOBAL_STATE_CODE();
8176 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8178 if (!drv->bdrv_make_empty) {
8179 error_setg(errp, "%s does not support emptying nodes",
8184 ret = drv->bdrv_make_empty(c->bs);
8186 error_setg_errno(errp, -ret, "Failed to empty %s",
8195 * Return the child that @bs acts as an overlay for, and from which data may be
8196 * copied in COW or COR operations. Usually this is the backing file.
8198 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8202 if (!bs || !bs->drv) {
8206 if (bs->drv->is_filter) {
8214 assert(bs->backing->role & BDRV_CHILD_COW);
8219 * If @bs acts as a filter for exactly one of its children, return
8222 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8227 if (!bs || !bs->drv) {
8231 if (!bs->drv->is_filter) {
8235 /* Only one of @backing or @file may be used */
8236 assert(!(bs->backing && bs->file));
8238 c = bs->backing ?: bs->file;
8243 assert(c->role & BDRV_CHILD_FILTERED);
8248 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8249 * whichever is non-NULL.
8251 * Return NULL if both are NULL.
8253 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8255 BdrvChild *cow_child = bdrv_cow_child(bs);
8256 BdrvChild *filter_child = bdrv_filter_child(bs);
8259 /* Filter nodes cannot have COW backing files */
8260 assert(!(cow_child && filter_child));
8262 return cow_child ?: filter_child;
8266 * Return the primary child of this node: For filters, that is the
8267 * filtered child. For other nodes, that is usually the child storing
8269 * (A generally more helpful description is that this is (usually) the
8270 * child that has the same filename as @bs.)
8272 * Drivers do not necessarily have a primary child; for example quorum
8275 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8277 BdrvChild *c, *found = NULL;
8280 QLIST_FOREACH(c, &bs->children, next) {
8281 if (c->role & BDRV_CHILD_PRIMARY) {
8290 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8291 bool stop_on_explicit_filter)
8299 while (!(stop_on_explicit_filter && !bs->implicit)) {
8300 c = bdrv_filter_child(bs);
8303 * A filter that is embedded in a working block graph must
8304 * have a child. Assert this here so this function does
8305 * not return a filter node that is not expected by the
8308 assert(!bs->drv || !bs->drv->is_filter);
8314 * Note that this treats nodes with bs->drv == NULL as not being
8315 * filters (bs->drv == NULL should be replaced by something else
8317 * The advantage of this behavior is that this function will thus
8318 * always return a non-NULL value (given a non-NULL @bs).
8325 * Return the first BDS that has not been added implicitly or that
8326 * does not have a filtered child down the chain starting from @bs
8327 * (including @bs itself).
8329 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8331 GLOBAL_STATE_CODE();
8332 return bdrv_do_skip_filters(bs, true);
8336 * Return the first BDS that does not have a filtered child down the
8337 * chain starting from @bs (including @bs itself).
8339 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8342 return bdrv_do_skip_filters(bs, false);
8346 * For a backing chain, return the first non-filter backing image of
8347 * the first non-filter image.
8349 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8352 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8356 * Check whether [offset, offset + bytes) overlaps with the cached
8357 * block-status data region.
8359 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8360 * which is what bdrv_bsc_is_data()'s interface needs.
8361 * Otherwise, *pnum is not touched.
8363 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8364 int64_t offset, int64_t bytes,
8367 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8371 qatomic_read(&bsc->valid) &&
8372 ranges_overlap(offset, bytes, bsc->data_start,
8373 bsc->data_end - bsc->data_start);
8375 if (overlaps && pnum) {
8376 *pnum = bsc->data_end - offset;
8383 * See block_int.h for this function's documentation.
8385 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8388 RCU_READ_LOCK_GUARD();
8389 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8393 * See block_int.h for this function's documentation.
8395 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8396 int64_t offset, int64_t bytes)
8399 RCU_READ_LOCK_GUARD();
8401 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8402 qatomic_set(&bs->block_status_cache->valid, false);
8407 * See block_int.h for this function's documentation.
8409 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8411 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8412 BdrvBlockStatusCache *old_bsc;
8415 *new_bsc = (BdrvBlockStatusCache) {
8417 .data_start = offset,
8418 .data_end = offset + bytes,
8421 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8423 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8424 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8426 g_free_rcu(old_bsc, rcu);