2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
53 #include "qemu/range.h"
55 #include "block/coroutines.h"
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
81 QLIST_HEAD_INITIALIZER(bdrv_drivers);
83 static BlockDriverState *bdrv_open_inherit(const char *filename,
84 const char *reference,
85 QDict *options, int flags,
86 BlockDriverState *parent,
87 const BdrvChildClass *child_class,
88 BdrvChildRole child_role,
91 static bool bdrv_recurse_has_child(BlockDriverState *bs,
92 BlockDriverState *child);
94 static void GRAPH_WRLOCK
95 bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs);
97 static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
99 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
100 BlockReopenQueue *queue,
101 Transaction *change_child_tran, Error **errp);
102 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
103 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
105 static bool bdrv_backing_overridden(BlockDriverState *bs);
107 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
108 GHashTable *visited, Transaction *tran,
111 /* If non-zero, use only whitelisted block drivers */
112 static int use_bdrv_whitelist;
115 static int is_windows_drive_prefix(const char *filename)
117 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
118 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
122 int is_windows_drive(const char *filename)
124 if (is_windows_drive_prefix(filename) &&
127 if (strstart(filename, "\\\\.\\", NULL) ||
128 strstart(filename, "//./", NULL))
134 size_t bdrv_opt_mem_align(BlockDriverState *bs)
136 if (!bs || !bs->drv) {
137 /* page size or 4k (hdd sector size) should be on the safe side */
138 return MAX(4096, qemu_real_host_page_size());
142 return bs->bl.opt_mem_alignment;
145 size_t bdrv_min_mem_align(BlockDriverState *bs)
147 if (!bs || !bs->drv) {
148 /* page size or 4k (hdd sector size) should be on the safe side */
149 return MAX(4096, qemu_real_host_page_size());
153 return bs->bl.min_mem_alignment;
156 /* check if the path starts with "<protocol>:" */
157 int path_has_protocol(const char *path)
162 if (is_windows_drive(path) ||
163 is_windows_drive_prefix(path)) {
166 p = path + strcspn(path, ":/\\");
168 p = path + strcspn(path, ":/");
174 int path_is_absolute(const char *path)
177 /* specific case for names like: "\\.\d:" */
178 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
181 return (*path == '/' || *path == '\\');
183 return (*path == '/');
187 /* if filename is absolute, just return its duplicate. Otherwise, build a
188 path to it by considering it is relative to base_path. URL are
190 char *path_combine(const char *base_path, const char *filename)
192 const char *protocol_stripped = NULL;
197 if (path_is_absolute(filename)) {
198 return g_strdup(filename);
201 if (path_has_protocol(base_path)) {
202 protocol_stripped = strchr(base_path, ':');
203 if (protocol_stripped) {
207 p = protocol_stripped ?: base_path;
209 p1 = strrchr(base_path, '/');
213 p2 = strrchr(base_path, '\\');
214 if (!p1 || p2 > p1) {
229 result = g_malloc(len + strlen(filename) + 1);
230 memcpy(result, base_path, len);
231 strcpy(result + len, filename);
237 * Helper function for bdrv_parse_filename() implementations to remove optional
238 * protocol prefixes (especially "file:") from a filename and for putting the
239 * stripped filename into the options QDict if there is such a prefix.
241 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
244 if (strstart(filename, prefix, &filename)) {
245 /* Stripping the explicit protocol prefix may result in a protocol
246 * prefix being (wrongly) detected (if the filename contains a colon) */
247 if (path_has_protocol(filename)) {
248 GString *fat_filename;
250 /* This means there is some colon before the first slash; therefore,
251 * this cannot be an absolute path */
252 assert(!path_is_absolute(filename));
254 /* And we can thus fix the protocol detection issue by prefixing it
256 fat_filename = g_string_new("./");
257 g_string_append(fat_filename, filename);
259 assert(!path_has_protocol(fat_filename->str));
261 qdict_put(options, "filename",
262 qstring_from_gstring(fat_filename));
264 /* If no protocol prefix was detected, we can use the shortened
266 qdict_put_str(options, "filename", filename);
272 /* Returns whether the image file is opened as read-only. Note that this can
273 * return false and writing to the image file is still not possible because the
274 * image is inactivated. */
275 bool bdrv_is_read_only(BlockDriverState *bs)
278 return !(bs->open_flags & BDRV_O_RDWR);
281 static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
282 bool ignore_allow_rdw, Error **errp)
286 /* Do not set read_only if copy_on_read is enabled */
287 if (bs->copy_on_read && read_only) {
288 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
289 bdrv_get_device_or_node_name(bs));
293 /* Do not clear read_only if it is prohibited */
294 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
297 error_setg(errp, "Node '%s' is read only",
298 bdrv_get_device_or_node_name(bs));
306 * Called by a driver that can only provide a read-only image.
308 * Returns 0 if the node is already read-only or it could switch the node to
309 * read-only because BDRV_O_AUTO_RDONLY is set.
311 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
312 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
313 * is not NULL, it is used as the error message for the Error object.
315 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
321 if (!(bs->open_flags & BDRV_O_RDWR)) {
324 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
328 ret = bdrv_can_set_read_only(bs, true, false, NULL);
333 bs->open_flags &= ~BDRV_O_RDWR;
338 error_setg(errp, "%s", errmsg ?: "Image is read-only");
343 * If @backing is empty, this function returns NULL without setting
344 * @errp. In all other cases, NULL will only be returned with @errp
347 * Therefore, a return value of NULL without @errp set means that
348 * there is no backing file; if @errp is set, there is one but its
349 * absolute filename cannot be generated.
351 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
355 if (backing[0] == '\0') {
357 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
358 return g_strdup(backing);
359 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
360 error_setg(errp, "Cannot use relative backing file names for '%s'",
364 return path_combine(backed, backing);
369 * If @filename is empty or NULL, this function returns NULL without
370 * setting @errp. In all other cases, NULL will only be returned with
373 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
374 const char *filename, Error **errp)
376 char *dir, *full_name;
378 if (!filename || filename[0] == '\0') {
380 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
381 return g_strdup(filename);
384 dir = bdrv_dirname(relative_to, errp);
389 full_name = g_strconcat(dir, filename, NULL);
394 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
397 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
400 void bdrv_register(BlockDriver *bdrv)
402 assert(bdrv->format_name);
404 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
407 BlockDriverState *bdrv_new(void)
409 BlockDriverState *bs;
414 bs = g_new0(BlockDriverState, 1);
415 QLIST_INIT(&bs->dirty_bitmaps);
416 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
417 QLIST_INIT(&bs->op_blockers[i]);
419 qemu_mutex_init(&bs->reqs_lock);
420 qemu_mutex_init(&bs->dirty_bitmap_mutex);
422 bs->aio_context = qemu_get_aio_context();
424 qemu_co_queue_init(&bs->flush_queue);
426 qemu_co_mutex_init(&bs->bsc_modify_lock);
427 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
429 for (i = 0; i < bdrv_drain_all_count; i++) {
430 bdrv_drained_begin(bs);
433 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
438 static BlockDriver *bdrv_do_find_format(const char *format_name)
443 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
444 if (!strcmp(drv1->format_name, format_name)) {
452 BlockDriver *bdrv_find_format(const char *format_name)
459 drv1 = bdrv_do_find_format(format_name);
464 /* The driver isn't registered, maybe we need to load a module */
465 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
466 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
467 Error *local_err = NULL;
468 int rv = block_module_load(block_driver_modules[i].library_name,
471 return bdrv_do_find_format(format_name);
473 error_report_err(local_err);
481 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
483 static const char *whitelist_rw[] = {
484 CONFIG_BDRV_RW_WHITELIST
487 static const char *whitelist_ro[] = {
488 CONFIG_BDRV_RO_WHITELIST
493 if (!whitelist_rw[0] && !whitelist_ro[0]) {
494 return 1; /* no whitelist, anything goes */
497 for (p = whitelist_rw; *p; p++) {
498 if (!strcmp(format_name, *p)) {
503 for (p = whitelist_ro; *p; p++) {
504 if (!strcmp(format_name, *p)) {
512 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
515 return bdrv_format_is_whitelisted(drv->format_name, read_only);
518 bool bdrv_uses_whitelist(void)
520 return use_bdrv_whitelist;
523 typedef struct CreateCo {
531 int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
532 QemuOpts *opts, Error **errp)
538 if (!drv->bdrv_co_create_opts) {
539 error_setg(errp, "Driver '%s' does not support image creation",
544 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
545 if (ret < 0 && !*errp) {
546 error_setg_errno(errp, -ret, "Could not create image");
553 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
554 * least the given @minimum_size.
556 * On success, return @blk's actual length.
557 * Otherwise, return -errno.
559 static int64_t coroutine_fn GRAPH_UNLOCKED
560 create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
563 Error *local_err = NULL;
569 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
571 if (ret < 0 && ret != -ENOTSUP) {
572 error_propagate(errp, local_err);
576 size = blk_co_getlength(blk);
578 error_free(local_err);
579 error_setg_errno(errp, -size,
580 "Failed to inquire the new image file's length");
584 if (size < minimum_size) {
585 /* Need to grow the image, but we failed to do that */
586 error_propagate(errp, local_err);
590 error_free(local_err);
597 * Helper function for bdrv_create_file_fallback(): Zero the first
598 * sector to remove any potentially pre-existing image header.
600 static int coroutine_fn
601 create_file_fallback_zero_first_sector(BlockBackend *blk,
602 int64_t current_size,
605 int64_t bytes_to_clear;
610 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
611 if (bytes_to_clear) {
612 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
614 error_setg_errno(errp, -ret,
615 "Failed to clear the new image's first sector");
624 * Simple implementation of bdrv_co_create_opts for protocol drivers
625 * which only support creation via opening a file
626 * (usually existing raw storage device)
628 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
629 const char *filename,
637 PreallocMode prealloc;
638 Error *local_err = NULL;
643 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
644 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
645 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
646 PREALLOC_MODE_OFF, &local_err);
649 error_propagate(errp, local_err);
653 if (prealloc != PREALLOC_MODE_OFF) {
654 error_setg(errp, "Unsupported preallocation mode '%s'",
655 PreallocMode_str(prealloc));
659 options = qdict_new();
660 qdict_put_str(options, "driver", drv->format_name);
662 blk = blk_co_new_open(filename, NULL, options,
663 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
665 error_prepend(errp, "Protocol driver '%s' does not support creating "
666 "new images, so an existing image must be selected as "
667 "the target; however, opening the given target as an "
668 "existing image failed: ",
673 size = create_file_fallback_truncate(blk, size, errp);
679 ret = create_file_fallback_zero_first_sector(blk, size, errp);
690 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
693 QemuOpts *protocol_opts;
700 drv = bdrv_find_protocol(filename, true, errp);
705 if (!drv->create_opts) {
706 error_setg(errp, "Driver '%s' does not support image creation",
712 * 'opts' contains a QemuOptsList with a combination of format and protocol
715 * The format properly removes its options, but the default values remain
716 * in 'opts->list'. So if the protocol has options with the same name
717 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
718 * of the format, since for overlapping options, the format wins.
720 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
721 * only the set options, and then convert it back to QemuOpts, using the
722 * create_opts of the protocol. So the new QemuOpts, will contain only the
725 qdict = qemu_opts_to_qdict(opts, NULL);
726 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
727 if (protocol_opts == NULL) {
732 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
734 qemu_opts_del(protocol_opts);
735 qobject_unref(qdict);
739 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
741 Error *local_err = NULL;
746 assert_bdrv_graph_readable();
749 error_setg(errp, "Block node '%s' is not opened", bs->filename);
753 if (!bs->drv->bdrv_co_delete_file) {
754 error_setg(errp, "Driver '%s' does not support image deletion",
755 bs->drv->format_name);
759 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
761 error_propagate(errp, local_err);
767 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
769 Error *local_err = NULL;
777 ret = bdrv_co_delete_file(bs, &local_err);
779 * ENOTSUP will happen if the block driver doesn't support
780 * the 'bdrv_co_delete_file' interface. This is a predictable
781 * scenario and shouldn't be reported back to the user.
783 if (ret == -ENOTSUP) {
784 error_free(local_err);
785 } else if (ret < 0) {
786 error_report_err(local_err);
791 * Try to get @bs's logical and physical block size.
792 * On success, store them in @bsz struct and return 0.
793 * On failure return -errno.
794 * @bs must not be empty.
796 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
798 BlockDriver *drv = bs->drv;
799 BlockDriverState *filtered = bdrv_filter_bs(bs);
802 if (drv && drv->bdrv_probe_blocksizes) {
803 return drv->bdrv_probe_blocksizes(bs, bsz);
804 } else if (filtered) {
805 return bdrv_probe_blocksizes(filtered, bsz);
812 * Try to get @bs's geometry (cyls, heads, sectors).
813 * On success, store them in @geo struct and return 0.
814 * On failure return -errno.
815 * @bs must not be empty.
817 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
819 BlockDriver *drv = bs->drv;
820 BlockDriverState *filtered = bdrv_filter_bs(bs);
823 if (drv && drv->bdrv_probe_geometry) {
824 return drv->bdrv_probe_geometry(bs, geo);
825 } else if (filtered) {
826 return bdrv_probe_geometry(filtered, geo);
833 * Create a uniquely-named empty temporary file.
834 * Return the actual file name used upon success, otherwise NULL.
835 * This string should be freed with g_free() when not needed any longer.
837 * Note: creating a temporary file for the caller to (re)open is
838 * inherently racy. Use g_file_open_tmp() instead whenever practical.
840 char *create_tmp_file(Error **errp)
844 g_autofree char *filename = NULL;
846 tmpdir = g_get_tmp_dir();
849 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
851 * This function is used to create temporary disk images (like -snapshot),
852 * so the files can become very large. /tmp is often a tmpfs where as
853 * /var/tmp is usually on a disk, so more appropriate for disk images.
855 if (!g_strcmp0(tmpdir, "/tmp")) {
860 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
861 fd = g_mkstemp(filename);
863 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
869 return g_steal_pointer(&filename);
873 * Detect host devices. By convention, /dev/cdrom[N] is always
874 * recognized as a host CDROM.
876 static BlockDriver *find_hdev_driver(const char *filename)
878 int score_max = 0, score;
879 BlockDriver *drv = NULL, *d;
882 QLIST_FOREACH(d, &bdrv_drivers, list) {
883 if (d->bdrv_probe_device) {
884 score = d->bdrv_probe_device(filename);
885 if (score > score_max) {
895 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
900 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
901 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
909 BlockDriver *bdrv_find_protocol(const char *filename,
910 bool allow_protocol_prefix,
920 /* TODO Drivers without bdrv_file_open must be specified explicitly */
923 * XXX(hch): we really should not let host device detection
924 * override an explicit protocol specification, but moving this
925 * later breaks access to device names with colons in them.
926 * Thanks to the brain-dead persistent naming schemes on udev-
927 * based Linux systems those actually are quite common.
929 drv1 = find_hdev_driver(filename);
934 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
938 p = strchr(filename, ':');
941 if (len > sizeof(protocol) - 1)
942 len = sizeof(protocol) - 1;
943 memcpy(protocol, filename, len);
944 protocol[len] = '\0';
946 drv1 = bdrv_do_find_protocol(protocol);
951 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
952 if (block_driver_modules[i].protocol_name &&
953 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
954 int rv = block_module_load(block_driver_modules[i].library_name, errp);
956 drv1 = bdrv_do_find_protocol(protocol);
965 error_setg(errp, "Unknown protocol '%s'", protocol);
971 * Guess image format by probing its contents.
972 * This is not a good idea when your image is raw (CVE-2008-2004), but
973 * we do it anyway for backward compatibility.
975 * @buf contains the image's first @buf_size bytes.
976 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
977 * but can be smaller if the image file is smaller)
978 * @filename is its filename.
980 * For all block drivers, call the bdrv_probe() method to get its
982 * Return the first block driver with the highest probing score.
984 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
985 const char *filename)
987 int score_max = 0, score;
988 BlockDriver *drv = NULL, *d;
991 QLIST_FOREACH(d, &bdrv_drivers, list) {
993 score = d->bdrv_probe(buf, buf_size, filename);
994 if (score > score_max) {
1004 static int find_image_format(BlockBackend *file, const char *filename,
1005 BlockDriver **pdrv, Error **errp)
1008 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1011 GLOBAL_STATE_CODE();
1013 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1014 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1019 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1021 error_setg_errno(errp, -ret, "Could not read image for determining its "
1027 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1029 error_setg(errp, "Could not determine image format: No compatible "
1040 * Set the current 'total_sectors' value
1041 * Return 0 on success, -errno on error.
1043 int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1046 BlockDriver *drv = bs->drv;
1048 assert_bdrv_graph_readable();
1054 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1058 /* query actual device if possible, otherwise just trust the hint */
1059 if (drv->bdrv_co_getlength) {
1060 int64_t length = drv->bdrv_co_getlength(bs);
1064 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1067 bs->total_sectors = hint;
1069 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1077 * Combines a QDict of new block driver @options with any missing options taken
1078 * from @old_options, so that leaving out an option defaults to its old value.
1080 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1083 GLOBAL_STATE_CODE();
1084 if (bs->drv && bs->drv->bdrv_join_options) {
1085 bs->drv->bdrv_join_options(options, old_options);
1087 qdict_join(options, old_options, false);
1091 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1095 Error *local_err = NULL;
1096 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1097 BlockdevDetectZeroesOptions detect_zeroes =
1098 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1099 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1100 GLOBAL_STATE_CODE();
1103 error_propagate(errp, local_err);
1104 return detect_zeroes;
1107 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1108 !(open_flags & BDRV_O_UNMAP))
1110 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1111 "without setting discard operation to unmap");
1114 return detect_zeroes;
1118 * Set open flags for aio engine
1120 * Return 0 on success, -1 if the engine specified is invalid
1122 int bdrv_parse_aio(const char *mode, int *flags)
1124 if (!strcmp(mode, "threads")) {
1125 /* do nothing, default */
1126 } else if (!strcmp(mode, "native")) {
1127 *flags |= BDRV_O_NATIVE_AIO;
1128 #ifdef CONFIG_LINUX_IO_URING
1129 } else if (!strcmp(mode, "io_uring")) {
1130 *flags |= BDRV_O_IO_URING;
1140 * Set open flags for a given discard mode
1142 * Return 0 on success, -1 if the discard mode was invalid.
1144 int bdrv_parse_discard_flags(const char *mode, int *flags)
1146 *flags &= ~BDRV_O_UNMAP;
1148 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1150 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1151 *flags |= BDRV_O_UNMAP;
1160 * Set open flags for a given cache mode
1162 * Return 0 on success, -1 if the cache mode was invalid.
1164 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1166 *flags &= ~BDRV_O_CACHE_MASK;
1168 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1169 *writethrough = false;
1170 *flags |= BDRV_O_NOCACHE;
1171 } else if (!strcmp(mode, "directsync")) {
1172 *writethrough = true;
1173 *flags |= BDRV_O_NOCACHE;
1174 } else if (!strcmp(mode, "writeback")) {
1175 *writethrough = false;
1176 } else if (!strcmp(mode, "unsafe")) {
1177 *writethrough = false;
1178 *flags |= BDRV_O_NO_FLUSH;
1179 } else if (!strcmp(mode, "writethrough")) {
1180 *writethrough = true;
1188 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1190 BlockDriverState *parent = c->opaque;
1191 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1194 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1196 BlockDriverState *bs = child->opaque;
1197 bdrv_do_drained_begin_quiesce(bs, NULL);
1200 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1202 BlockDriverState *bs = child->opaque;
1203 return bdrv_drain_poll(bs, NULL, false);
1206 static void bdrv_child_cb_drained_end(BdrvChild *child)
1208 BlockDriverState *bs = child->opaque;
1209 bdrv_drained_end(bs);
1212 static int bdrv_child_cb_inactivate(BdrvChild *child)
1214 BlockDriverState *bs = child->opaque;
1215 GLOBAL_STATE_CODE();
1216 assert(bs->open_flags & BDRV_O_INACTIVE);
1220 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1221 GHashTable *visited, Transaction *tran,
1224 BlockDriverState *bs = child->opaque;
1225 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1229 * Returns the options and flags that a temporary snapshot should get, based on
1230 * the originally requested flags (the originally requested image will have
1231 * flags like a backing file)
1233 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1234 int parent_flags, QDict *parent_options)
1236 GLOBAL_STATE_CODE();
1237 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1239 /* For temporary files, unconditional cache=unsafe is fine */
1240 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1241 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1243 /* Copy the read-only and discard options from the parent */
1244 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1245 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1247 /* aio=native doesn't work for cache.direct=off, so disable it for the
1248 * temporary snapshot */
1249 *child_flags &= ~BDRV_O_NATIVE_AIO;
1252 static void bdrv_backing_attach(BdrvChild *c)
1254 BlockDriverState *parent = c->opaque;
1255 BlockDriverState *backing_hd = c->bs;
1257 GLOBAL_STATE_CODE();
1258 assert(!parent->backing_blocker);
1259 error_setg(&parent->backing_blocker,
1260 "node is used as backing hd of '%s'",
1261 bdrv_get_device_or_node_name(parent));
1263 bdrv_refresh_filename(backing_hd);
1265 parent->open_flags &= ~BDRV_O_NO_BACKING;
1267 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1268 /* Otherwise we won't be able to commit or stream */
1269 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1270 parent->backing_blocker);
1271 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1272 parent->backing_blocker);
1274 * We do backup in 3 ways:
1276 * The target bs is new opened, and the source is top BDS
1277 * 2. blockdev backup
1278 * Both the source and the target are top BDSes.
1279 * 3. internal backup(used for block replication)
1280 * Both the source and the target are backing file
1282 * In case 1 and 2, neither the source nor the target is the backing file.
1283 * In case 3, we will block the top BDS, so there is only one block job
1284 * for the top BDS and its backing chain.
1286 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1287 parent->backing_blocker);
1288 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1289 parent->backing_blocker);
1292 static void bdrv_backing_detach(BdrvChild *c)
1294 BlockDriverState *parent = c->opaque;
1296 GLOBAL_STATE_CODE();
1297 assert(parent->backing_blocker);
1298 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1299 error_free(parent->backing_blocker);
1300 parent->backing_blocker = NULL;
1303 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1304 const char *filename, Error **errp)
1306 BlockDriverState *parent = c->opaque;
1307 bool read_only = bdrv_is_read_only(parent);
1309 GLOBAL_STATE_CODE();
1312 ret = bdrv_reopen_set_read_only(parent, false, errp);
1318 ret = bdrv_change_backing_file(parent, filename,
1319 base->drv ? base->drv->format_name : "",
1322 error_setg_errno(errp, -ret, "Could not update backing file link");
1326 bdrv_reopen_set_read_only(parent, true, NULL);
1333 * Returns the options and flags that a generic child of a BDS should
1334 * get, based on the given options and flags for the parent BDS.
1336 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1337 int *child_flags, QDict *child_options,
1338 int parent_flags, QDict *parent_options)
1340 int flags = parent_flags;
1341 GLOBAL_STATE_CODE();
1344 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1345 * Generally, the question to answer is: Should this child be
1346 * format-probed by default?
1350 * Pure and non-filtered data children of non-format nodes should
1351 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1352 * set). This only affects a very limited set of drivers (namely
1353 * quorum and blkverify when this comment was written).
1354 * Force-clear BDRV_O_PROTOCOL then.
1356 if (!parent_is_format &&
1357 (role & BDRV_CHILD_DATA) &&
1358 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1360 flags &= ~BDRV_O_PROTOCOL;
1364 * All children of format nodes (except for COW children) and all
1365 * metadata children in general should never be format-probed.
1366 * Force-set BDRV_O_PROTOCOL then.
1368 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1369 (role & BDRV_CHILD_METADATA))
1371 flags |= BDRV_O_PROTOCOL;
1375 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1378 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1379 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1380 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1382 if (role & BDRV_CHILD_COW) {
1383 /* backing files are opened read-only by default */
1384 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1385 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1387 /* Inherit the read-only option from the parent if it's not set */
1388 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1389 qdict_copy_default(child_options, parent_options,
1390 BDRV_OPT_AUTO_READ_ONLY);
1394 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1395 * can default to enable it on lower layers regardless of the
1398 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1400 /* Clear flags that only apply to the top layer */
1401 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1403 if (role & BDRV_CHILD_METADATA) {
1404 flags &= ~BDRV_O_NO_IO;
1406 if (role & BDRV_CHILD_COW) {
1407 flags &= ~BDRV_O_TEMPORARY;
1410 *child_flags = flags;
1413 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1415 BlockDriverState *bs = child->opaque;
1417 assert_bdrv_graph_writable();
1418 QLIST_INSERT_HEAD(&bs->children, child, next);
1419 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1421 * Here we handle filters and block/raw-format.c when it behave like
1422 * filter. They generally have a single PRIMARY child, which is also the
1423 * FILTERED child, and that they may have multiple more children, which
1424 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1425 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1426 * into bs->backing on exceptional cases; and bs->backing will be
1429 assert(!(child->role & BDRV_CHILD_COW));
1430 if (child->role & BDRV_CHILD_PRIMARY) {
1431 assert(child->role & BDRV_CHILD_FILTERED);
1432 assert(!bs->backing);
1435 if (bs->drv->filtered_child_is_backing) {
1436 bs->backing = child;
1441 assert(!(child->role & BDRV_CHILD_FILTERED));
1443 } else if (child->role & BDRV_CHILD_COW) {
1444 assert(bs->drv->supports_backing);
1445 assert(!(child->role & BDRV_CHILD_PRIMARY));
1446 assert(!bs->backing);
1447 bs->backing = child;
1448 bdrv_backing_attach(child);
1449 } else if (child->role & BDRV_CHILD_PRIMARY) {
1455 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1457 BlockDriverState *bs = child->opaque;
1459 if (child->role & BDRV_CHILD_COW) {
1460 bdrv_backing_detach(child);
1463 assert_bdrv_graph_writable();
1464 QLIST_REMOVE(child, next);
1465 if (child == bs->backing) {
1466 assert(child != bs->file);
1468 } else if (child == bs->file) {
1473 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1474 const char *filename, Error **errp)
1476 if (c->role & BDRV_CHILD_COW) {
1477 return bdrv_backing_update_filename(c, base, filename, errp);
1482 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1484 BlockDriverState *bs = c->opaque;
1487 return bdrv_get_aio_context(bs);
1490 const BdrvChildClass child_of_bds = {
1491 .parent_is_bds = true,
1492 .get_parent_desc = bdrv_child_get_parent_desc,
1493 .inherit_options = bdrv_inherited_options,
1494 .drained_begin = bdrv_child_cb_drained_begin,
1495 .drained_poll = bdrv_child_cb_drained_poll,
1496 .drained_end = bdrv_child_cb_drained_end,
1497 .attach = bdrv_child_cb_attach,
1498 .detach = bdrv_child_cb_detach,
1499 .inactivate = bdrv_child_cb_inactivate,
1500 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1501 .update_filename = bdrv_child_cb_update_filename,
1502 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1505 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1508 return c->klass->get_parent_aio_context(c);
1511 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1513 int open_flags = flags;
1514 GLOBAL_STATE_CODE();
1517 * Clear flags that are internal to the block layer before opening the
1520 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1525 static void update_flags_from_options(int *flags, QemuOpts *opts)
1527 GLOBAL_STATE_CODE();
1529 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1531 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1532 *flags |= BDRV_O_NO_FLUSH;
1535 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1536 *flags |= BDRV_O_NOCACHE;
1539 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1540 *flags |= BDRV_O_RDWR;
1543 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1544 *flags |= BDRV_O_AUTO_RDONLY;
1548 static void update_options_from_flags(QDict *options, int flags)
1550 GLOBAL_STATE_CODE();
1551 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1552 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1554 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1555 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1556 flags & BDRV_O_NO_FLUSH);
1558 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1559 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1561 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1562 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1563 flags & BDRV_O_AUTO_RDONLY);
1567 static void bdrv_assign_node_name(BlockDriverState *bs,
1568 const char *node_name,
1571 char *gen_node_name = NULL;
1572 GLOBAL_STATE_CODE();
1575 node_name = gen_node_name = id_generate(ID_BLOCK);
1576 } else if (!id_wellformed(node_name)) {
1578 * Check for empty string or invalid characters, but not if it is
1579 * generated (generated names use characters not available to the user)
1581 error_setg(errp, "Invalid node-name: '%s'", node_name);
1585 /* takes care of avoiding namespaces collisions */
1586 if (blk_by_name(node_name)) {
1587 error_setg(errp, "node-name=%s is conflicting with a device id",
1592 /* takes care of avoiding duplicates node names */
1593 if (bdrv_find_node(node_name)) {
1594 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1598 /* Make sure that the node name isn't truncated */
1599 if (strlen(node_name) >= sizeof(bs->node_name)) {
1600 error_setg(errp, "Node name too long");
1604 /* copy node name into the bs and insert it into the graph list */
1605 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1606 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1608 g_free(gen_node_name);
1612 * The caller must always hold @bs AioContext lock, because this function calls
1613 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1616 static int no_coroutine_fn GRAPH_UNLOCKED
1617 bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1618 QDict *options, int open_flags, Error **errp)
1621 Error *local_err = NULL;
1623 GLOBAL_STATE_CODE();
1625 bdrv_assign_node_name(bs, node_name, &local_err);
1627 error_propagate(errp, local_err);
1632 bs->opaque = g_malloc0(drv->instance_size);
1634 if (drv->bdrv_file_open) {
1635 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1636 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1637 } else if (drv->bdrv_open) {
1638 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1645 error_propagate(errp, local_err);
1646 } else if (bs->filename[0]) {
1647 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1649 error_setg_errno(errp, -ret, "Could not open image");
1654 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1655 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1658 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1659 * drivers that pass read/write requests through to a child the trouble of
1660 * declaring support explicitly.
1662 * Drivers must not propagate this flag accidentally when they initiate I/O
1663 * to a bounce buffer. That case should be rare though.
1665 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1666 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1668 /* Get the context after .bdrv_open, it can change the context */
1669 ctx = bdrv_get_aio_context(bs);
1670 aio_context_acquire(ctx);
1672 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
1674 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1675 aio_context_release(ctx);
1679 bdrv_graph_rdlock_main_loop();
1680 bdrv_refresh_limits(bs, NULL, &local_err);
1681 bdrv_graph_rdunlock_main_loop();
1682 aio_context_release(ctx);
1685 error_propagate(errp, local_err);
1689 assert(bdrv_opt_mem_align(bs) != 0);
1690 assert(bdrv_min_mem_align(bs) != 0);
1691 assert(is_power_of_2(bs->bl.request_alignment));
1693 for (i = 0; i < bs->quiesce_counter; i++) {
1694 if (drv->bdrv_drain_begin) {
1695 drv->bdrv_drain_begin(bs);
1702 if (bs->file != NULL) {
1703 bdrv_unref_child(bs, bs->file);
1712 * Create and open a block node.
1714 * @options is a QDict of options to pass to the block drivers, or NULL for an
1715 * empty set of options. The reference to the QDict belongs to the block layer
1716 * after the call (even on failure), so if the caller intends to reuse the
1717 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1719 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1720 const char *node_name,
1721 QDict *options, int flags,
1724 BlockDriverState *bs;
1727 GLOBAL_STATE_CODE();
1730 bs->open_flags = flags;
1731 bs->options = options ?: qdict_new();
1732 bs->explicit_options = qdict_clone_shallow(bs->options);
1735 update_options_from_flags(bs->options, flags);
1737 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1739 qobject_unref(bs->explicit_options);
1740 bs->explicit_options = NULL;
1741 qobject_unref(bs->options);
1750 /* Create and open a block node. */
1751 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1752 int flags, Error **errp)
1754 GLOBAL_STATE_CODE();
1755 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1758 QemuOptsList bdrv_runtime_opts = {
1759 .name = "bdrv_common",
1760 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1763 .name = "node-name",
1764 .type = QEMU_OPT_STRING,
1765 .help = "Node name of the block device node",
1769 .type = QEMU_OPT_STRING,
1770 .help = "Block driver to use for the node",
1773 .name = BDRV_OPT_CACHE_DIRECT,
1774 .type = QEMU_OPT_BOOL,
1775 .help = "Bypass software writeback cache on the host",
1778 .name = BDRV_OPT_CACHE_NO_FLUSH,
1779 .type = QEMU_OPT_BOOL,
1780 .help = "Ignore flush requests",
1783 .name = BDRV_OPT_READ_ONLY,
1784 .type = QEMU_OPT_BOOL,
1785 .help = "Node is opened in read-only mode",
1788 .name = BDRV_OPT_AUTO_READ_ONLY,
1789 .type = QEMU_OPT_BOOL,
1790 .help = "Node can become read-only if opening read-write fails",
1793 .name = "detect-zeroes",
1794 .type = QEMU_OPT_STRING,
1795 .help = "try to optimize zero writes (off, on, unmap)",
1798 .name = BDRV_OPT_DISCARD,
1799 .type = QEMU_OPT_STRING,
1800 .help = "discard operation (ignore/off, unmap/on)",
1803 .name = BDRV_OPT_FORCE_SHARE,
1804 .type = QEMU_OPT_BOOL,
1805 .help = "always accept other writers (default: off)",
1807 { /* end of list */ }
1811 QemuOptsList bdrv_create_opts_simple = {
1812 .name = "simple-create-opts",
1813 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1816 .name = BLOCK_OPT_SIZE,
1817 .type = QEMU_OPT_SIZE,
1818 .help = "Virtual disk size"
1821 .name = BLOCK_OPT_PREALLOC,
1822 .type = QEMU_OPT_STRING,
1823 .help = "Preallocation mode (allowed values: off)"
1825 { /* end of list */ }
1830 * Common part for opening disk images and files
1832 * Removes all processed options from *options.
1834 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1835 QDict *options, Error **errp)
1837 int ret, open_flags;
1838 const char *filename;
1839 const char *driver_name = NULL;
1840 const char *node_name = NULL;
1841 const char *discard;
1844 Error *local_err = NULL;
1847 assert(bs->file == NULL);
1848 assert(options != NULL && bs->options != options);
1849 GLOBAL_STATE_CODE();
1851 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1852 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1857 update_flags_from_options(&bs->open_flags, opts);
1859 driver_name = qemu_opt_get(opts, "driver");
1860 drv = bdrv_find_format(driver_name);
1861 assert(drv != NULL);
1863 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1865 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1867 BDRV_OPT_FORCE_SHARE
1868 "=on can only be used with read-only images");
1874 bdrv_refresh_filename(blk_bs(file));
1875 filename = blk_bs(file)->filename;
1878 * Caution: while qdict_get_try_str() is fine, getting
1879 * non-string types would require more care. When @options
1880 * come from -blockdev or blockdev_add, its members are typed
1881 * according to the QAPI schema, but when they come from
1882 * -drive, they're all QString.
1884 filename = qdict_get_try_str(options, "filename");
1887 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1888 error_setg(errp, "The '%s' block driver requires a file name",
1894 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1897 ro = bdrv_is_read_only(bs);
1899 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1900 if (!ro && bdrv_is_whitelisted(drv, true)) {
1901 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1907 !ro && bdrv_is_whitelisted(drv, true)
1908 ? "Driver '%s' can only be used for read-only devices"
1909 : "Driver '%s' is not whitelisted",
1915 /* bdrv_new() and bdrv_close() make it so */
1916 assert(qatomic_read(&bs->copy_on_read) == 0);
1918 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1920 bdrv_enable_copy_on_read(bs);
1922 error_setg(errp, "Can't use copy-on-read on read-only device");
1928 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1929 if (discard != NULL) {
1930 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1931 error_setg(errp, "Invalid discard option");
1938 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1940 error_propagate(errp, local_err);
1945 if (filename != NULL) {
1946 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1948 bs->filename[0] = '\0';
1950 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1952 /* Open the image, either directly or using a protocol */
1953 open_flags = bdrv_open_flags(bs, bs->open_flags);
1954 node_name = qemu_opt_get(opts, "node-name");
1956 assert(!drv->bdrv_file_open || file == NULL);
1957 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1962 qemu_opts_del(opts);
1966 qemu_opts_del(opts);
1970 static QDict *parse_json_filename(const char *filename, Error **errp)
1972 QObject *options_obj;
1975 GLOBAL_STATE_CODE();
1977 ret = strstart(filename, "json:", &filename);
1980 options_obj = qobject_from_json(filename, errp);
1982 error_prepend(errp, "Could not parse the JSON options: ");
1986 options = qobject_to(QDict, options_obj);
1988 qobject_unref(options_obj);
1989 error_setg(errp, "Invalid JSON object given");
1993 qdict_flatten(options);
1998 static void parse_json_protocol(QDict *options, const char **pfilename,
2001 QDict *json_options;
2002 Error *local_err = NULL;
2003 GLOBAL_STATE_CODE();
2005 /* Parse json: pseudo-protocol */
2006 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2010 json_options = parse_json_filename(*pfilename, &local_err);
2012 error_propagate(errp, local_err);
2016 /* Options given in the filename have lower priority than options
2017 * specified directly */
2018 qdict_join(options, json_options, false);
2019 qobject_unref(json_options);
2024 * Fills in default options for opening images and converts the legacy
2025 * filename/flags pair to option QDict entries.
2026 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2027 * block driver has been specified explicitly.
2029 static int bdrv_fill_options(QDict **options, const char *filename,
2030 int *flags, Error **errp)
2032 const char *drvname;
2033 bool protocol = *flags & BDRV_O_PROTOCOL;
2034 bool parse_filename = false;
2035 BlockDriver *drv = NULL;
2036 Error *local_err = NULL;
2038 GLOBAL_STATE_CODE();
2041 * Caution: while qdict_get_try_str() is fine, getting non-string
2042 * types would require more care. When @options come from
2043 * -blockdev or blockdev_add, its members are typed according to
2044 * the QAPI schema, but when they come from -drive, they're all
2047 drvname = qdict_get_try_str(*options, "driver");
2049 drv = bdrv_find_format(drvname);
2051 error_setg(errp, "Unknown driver '%s'", drvname);
2054 /* If the user has explicitly specified the driver, this choice should
2055 * override the BDRV_O_PROTOCOL flag */
2056 protocol = drv->bdrv_file_open;
2060 *flags |= BDRV_O_PROTOCOL;
2062 *flags &= ~BDRV_O_PROTOCOL;
2065 /* Translate cache options from flags into options */
2066 update_options_from_flags(*options, *flags);
2068 /* Fetch the file name from the options QDict if necessary */
2069 if (protocol && filename) {
2070 if (!qdict_haskey(*options, "filename")) {
2071 qdict_put_str(*options, "filename", filename);
2072 parse_filename = true;
2074 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2080 /* Find the right block driver */
2081 /* See cautionary note on accessing @options above */
2082 filename = qdict_get_try_str(*options, "filename");
2084 if (!drvname && protocol) {
2086 drv = bdrv_find_protocol(filename, parse_filename, errp);
2091 drvname = drv->format_name;
2092 qdict_put_str(*options, "driver", drvname);
2094 error_setg(errp, "Must specify either driver or file");
2099 assert(drv || !protocol);
2101 /* Driver-specific filename parsing */
2102 if (drv && drv->bdrv_parse_filename && parse_filename) {
2103 drv->bdrv_parse_filename(filename, *options, &local_err);
2105 error_propagate(errp, local_err);
2109 if (!drv->bdrv_needs_filename) {
2110 qdict_del(*options, "filename");
2117 typedef struct BlockReopenQueueEntry {
2119 BDRVReopenState state;
2120 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2121 } BlockReopenQueueEntry;
2124 * Return the flags that @bs will have after the reopens in @q have
2125 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2126 * return the current flags.
2128 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2130 BlockReopenQueueEntry *entry;
2133 QTAILQ_FOREACH(entry, q, entry) {
2134 if (entry->state.bs == bs) {
2135 return entry->state.flags;
2140 return bs->open_flags;
2143 /* Returns whether the image file can be written to after the reopen queue @q
2144 * has been successfully applied, or right now if @q is NULL. */
2145 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2146 BlockReopenQueue *q)
2148 int flags = bdrv_reopen_get_flags(q, bs);
2150 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2154 * Return whether the BDS can be written to. This is not necessarily
2155 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2156 * be written to but do not count as read-only images.
2158 bool bdrv_is_writable(BlockDriverState *bs)
2161 return bdrv_is_writable_after_reopen(bs, NULL);
2164 static char *bdrv_child_user_desc(BdrvChild *c)
2166 GLOBAL_STATE_CODE();
2167 return c->klass->get_parent_desc(c);
2171 * Check that @a allows everything that @b needs. @a and @b must reference same
2174 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2176 const char *child_bs_name;
2177 g_autofree char *a_user = NULL;
2178 g_autofree char *b_user = NULL;
2179 g_autofree char *perms = NULL;
2182 assert(a->bs == b->bs);
2183 GLOBAL_STATE_CODE();
2185 if ((b->perm & a->shared_perm) == b->perm) {
2189 child_bs_name = bdrv_get_node_name(b->bs);
2190 a_user = bdrv_child_user_desc(a);
2191 b_user = bdrv_child_user_desc(b);
2192 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2194 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2195 "both required by %s (uses node '%s' as '%s' child) and "
2196 "unshared by %s (uses node '%s' as '%s' child).",
2197 child_bs_name, perms,
2198 b_user, child_bs_name, b->name,
2199 a_user, child_bs_name, a->name);
2204 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2207 GLOBAL_STATE_CODE();
2210 * During the loop we'll look at each pair twice. That's correct because
2211 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2214 QLIST_FOREACH(a, &bs->parents, next_parent) {
2215 QLIST_FOREACH(b, &bs->parents, next_parent) {
2220 if (!bdrv_a_allow_b(a, b, errp)) {
2229 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2230 BdrvChild *c, BdrvChildRole role,
2231 BlockReopenQueue *reopen_queue,
2232 uint64_t parent_perm, uint64_t parent_shared,
2233 uint64_t *nperm, uint64_t *nshared)
2235 assert(bs->drv && bs->drv->bdrv_child_perm);
2236 GLOBAL_STATE_CODE();
2237 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2238 parent_perm, parent_shared,
2240 /* TODO Take force_share from reopen_queue */
2241 if (child_bs && child_bs->force_share) {
2242 *nshared = BLK_PERM_ALL;
2247 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2248 * nodes that are already in the @list, of course) so that final list is
2249 * topologically sorted. Return the result (GSList @list object is updated, so
2250 * don't use old reference after function call).
2252 * On function start @list must be already topologically sorted and for any node
2253 * in the @list the whole subtree of the node must be in the @list as well. The
2254 * simplest way to satisfy this criteria: use only result of
2255 * bdrv_topological_dfs() or NULL as @list parameter.
2257 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2258 BlockDriverState *bs)
2261 g_autoptr(GHashTable) local_found = NULL;
2263 GLOBAL_STATE_CODE();
2267 found = local_found = g_hash_table_new(NULL, NULL);
2270 if (g_hash_table_contains(found, bs)) {
2273 g_hash_table_add(found, bs);
2275 QLIST_FOREACH(child, &bs->children, next) {
2276 list = bdrv_topological_dfs(list, found, child->bs);
2279 return g_slist_prepend(list, bs);
2282 typedef struct BdrvChildSetPermState {
2285 uint64_t old_shared_perm;
2286 } BdrvChildSetPermState;
2288 static void bdrv_child_set_perm_abort(void *opaque)
2290 BdrvChildSetPermState *s = opaque;
2292 GLOBAL_STATE_CODE();
2294 s->child->perm = s->old_perm;
2295 s->child->shared_perm = s->old_shared_perm;
2298 static TransactionActionDrv bdrv_child_set_pem_drv = {
2299 .abort = bdrv_child_set_perm_abort,
2303 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2304 uint64_t shared, Transaction *tran)
2306 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2307 GLOBAL_STATE_CODE();
2309 *s = (BdrvChildSetPermState) {
2311 .old_perm = c->perm,
2312 .old_shared_perm = c->shared_perm,
2316 c->shared_perm = shared;
2318 tran_add(tran, &bdrv_child_set_pem_drv, s);
2321 static void bdrv_drv_set_perm_commit(void *opaque)
2323 BlockDriverState *bs = opaque;
2324 uint64_t cumulative_perms, cumulative_shared_perms;
2325 GLOBAL_STATE_CODE();
2327 if (bs->drv->bdrv_set_perm) {
2328 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2329 &cumulative_shared_perms);
2330 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2334 static void bdrv_drv_set_perm_abort(void *opaque)
2336 BlockDriverState *bs = opaque;
2337 GLOBAL_STATE_CODE();
2339 if (bs->drv->bdrv_abort_perm_update) {
2340 bs->drv->bdrv_abort_perm_update(bs);
2344 TransactionActionDrv bdrv_drv_set_perm_drv = {
2345 .abort = bdrv_drv_set_perm_abort,
2346 .commit = bdrv_drv_set_perm_commit,
2349 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2350 uint64_t shared_perm, Transaction *tran,
2353 GLOBAL_STATE_CODE();
2358 if (bs->drv->bdrv_check_perm) {
2359 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2366 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2372 typedef struct BdrvReplaceChildState {
2374 BlockDriverState *old_bs;
2375 } BdrvReplaceChildState;
2377 static void bdrv_replace_child_commit(void *opaque)
2379 BdrvReplaceChildState *s = opaque;
2380 GLOBAL_STATE_CODE();
2382 bdrv_unref(s->old_bs);
2385 static void bdrv_replace_child_abort(void *opaque)
2387 BdrvReplaceChildState *s = opaque;
2388 BlockDriverState *new_bs = s->child->bs;
2390 GLOBAL_STATE_CODE();
2391 bdrv_graph_wrlock(s->old_bs);
2393 /* old_bs reference is transparently moved from @s to @s->child */
2394 if (!s->child->bs) {
2396 * The parents were undrained when removing old_bs from the child. New
2397 * requests can't have been made, though, because the child was empty.
2399 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2400 * undraining the parent in the first place. Once this is done, having
2401 * new_bs drained when calling bdrv_replace_child_tran() is not a
2402 * requirement any more.
2404 bdrv_parent_drained_begin_single(s->child);
2405 assert(!bdrv_parent_drained_poll_single(s->child));
2407 assert(s->child->quiesced_parent);
2408 bdrv_replace_child_noperm(s->child, s->old_bs);
2410 bdrv_graph_wrunlock();
2414 static TransactionActionDrv bdrv_replace_child_drv = {
2415 .commit = bdrv_replace_child_commit,
2416 .abort = bdrv_replace_child_abort,
2421 * bdrv_replace_child_tran
2423 * Note: real unref of old_bs is done only on commit.
2425 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2426 * kept drained until the transaction is completed.
2428 * The function doesn't update permissions, caller is responsible for this.
2430 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2433 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2435 assert(child->quiesced_parent);
2436 assert(!new_bs || new_bs->quiesce_counter);
2438 *s = (BdrvReplaceChildState) {
2440 .old_bs = child->bs,
2442 tran_add(tran, &bdrv_replace_child_drv, s);
2448 bdrv_graph_wrlock(new_bs);
2449 bdrv_replace_child_noperm(child, new_bs);
2450 bdrv_graph_wrunlock();
2451 /* old_bs reference is transparently moved from @child to @s */
2455 * Refresh permissions in @bs subtree. The function is intended to be called
2456 * after some graph modification that was done without permission update.
2458 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2459 Transaction *tran, Error **errp)
2461 BlockDriver *drv = bs->drv;
2464 uint64_t cumulative_perms, cumulative_shared_perms;
2465 GLOBAL_STATE_CODE();
2467 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2469 /* Write permissions never work with read-only images */
2470 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2471 !bdrv_is_writable_after_reopen(bs, q))
2473 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2474 error_setg(errp, "Block node is read-only");
2476 error_setg(errp, "Read-only block node '%s' cannot support "
2477 "read-write users", bdrv_get_node_name(bs));
2484 * Unaligned requests will automatically be aligned to bl.request_alignment
2485 * and without RESIZE we can't extend requests to write to space beyond the
2486 * end of the image, so it's required that the image size is aligned.
2488 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2489 !(cumulative_perms & BLK_PERM_RESIZE))
2491 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2492 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2493 "Image size is not a multiple of request "
2499 /* Check this node */
2504 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2510 /* Drivers that never have children can omit .bdrv_child_perm() */
2511 if (!drv->bdrv_child_perm) {
2512 assert(QLIST_EMPTY(&bs->children));
2516 /* Check all children */
2517 QLIST_FOREACH(c, &bs->children, next) {
2518 uint64_t cur_perm, cur_shared;
2520 bdrv_child_perm(bs, c->bs, c, c->role, q,
2521 cumulative_perms, cumulative_shared_perms,
2522 &cur_perm, &cur_shared);
2523 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2530 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2531 * a topologically sorted subgraph.
2533 static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2534 Transaction *tran, Error **errp)
2537 BlockDriverState *bs;
2538 GLOBAL_STATE_CODE();
2540 for ( ; list; list = list->next) {
2543 if (bdrv_parent_perms_conflict(bs, errp)) {
2547 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2557 * @list is any list of nodes. List is completed by all subtrees and
2558 * topologically sorted. It's not a problem if some node occurs in the @list
2561 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2562 Transaction *tran, Error **errp)
2564 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2565 g_autoptr(GSList) refresh_list = NULL;
2567 for ( ; list; list = list->next) {
2568 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2571 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2574 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2575 uint64_t *shared_perm)
2578 uint64_t cumulative_perms = 0;
2579 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2581 GLOBAL_STATE_CODE();
2583 QLIST_FOREACH(c, &bs->parents, next_parent) {
2584 cumulative_perms |= c->perm;
2585 cumulative_shared_perms &= c->shared_perm;
2588 *perm = cumulative_perms;
2589 *shared_perm = cumulative_shared_perms;
2592 char *bdrv_perm_names(uint64_t perm)
2598 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2599 { BLK_PERM_WRITE, "write" },
2600 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2601 { BLK_PERM_RESIZE, "resize" },
2605 GString *result = g_string_sized_new(30);
2606 struct perm_name *p;
2608 for (p = permissions; p->name; p++) {
2609 if (perm & p->perm) {
2610 if (result->len > 0) {
2611 g_string_append(result, ", ");
2613 g_string_append(result, p->name);
2617 return g_string_free(result, FALSE);
2621 /* @tran is allowed to be NULL. In this case no rollback is possible */
2622 static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2626 Transaction *local_tran = NULL;
2627 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2628 GLOBAL_STATE_CODE();
2631 tran = local_tran = tran_new();
2634 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2637 tran_finalize(local_tran, ret);
2643 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2646 Error *local_err = NULL;
2647 Transaction *tran = tran_new();
2650 GLOBAL_STATE_CODE();
2652 bdrv_child_set_perm(c, perm, shared, tran);
2654 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2656 tran_finalize(tran, ret);
2659 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2660 /* tighten permissions */
2661 error_propagate(errp, local_err);
2664 * Our caller may intend to only loosen restrictions and
2665 * does not expect this function to fail. Errors are not
2666 * fatal in such a case, so we can just hide them from our
2669 error_free(local_err);
2677 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2679 uint64_t parent_perms, parent_shared;
2680 uint64_t perms, shared;
2682 GLOBAL_STATE_CODE();
2684 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2685 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2686 parent_perms, parent_shared, &perms, &shared);
2688 return bdrv_child_try_set_perm(c, perms, shared, errp);
2692 * Default implementation for .bdrv_child_perm() for block filters:
2693 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2696 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2698 BlockReopenQueue *reopen_queue,
2699 uint64_t perm, uint64_t shared,
2700 uint64_t *nperm, uint64_t *nshared)
2702 GLOBAL_STATE_CODE();
2703 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2704 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2707 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2709 BlockReopenQueue *reopen_queue,
2710 uint64_t perm, uint64_t shared,
2711 uint64_t *nperm, uint64_t *nshared)
2713 assert(role & BDRV_CHILD_COW);
2714 GLOBAL_STATE_CODE();
2717 * We want consistent read from backing files if the parent needs it.
2718 * No other operations are performed on backing files.
2720 perm &= BLK_PERM_CONSISTENT_READ;
2723 * If the parent can deal with changing data, we're okay with a
2724 * writable and resizable backing file.
2725 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2727 if (shared & BLK_PERM_WRITE) {
2728 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2733 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2735 if (bs->open_flags & BDRV_O_INACTIVE) {
2736 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2743 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2745 BlockReopenQueue *reopen_queue,
2746 uint64_t perm, uint64_t shared,
2747 uint64_t *nperm, uint64_t *nshared)
2751 GLOBAL_STATE_CODE();
2752 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2754 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2757 * Apart from the modifications below, the same permissions are
2758 * forwarded and left alone as for filters
2760 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2761 perm, shared, &perm, &shared);
2763 if (role & BDRV_CHILD_METADATA) {
2764 /* Format drivers may touch metadata even if the guest doesn't write */
2765 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2766 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2770 * bs->file always needs to be consistent because of the
2771 * metadata. We can never allow other users to resize or write
2774 if (!(flags & BDRV_O_NO_IO)) {
2775 perm |= BLK_PERM_CONSISTENT_READ;
2777 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2780 if (role & BDRV_CHILD_DATA) {
2782 * Technically, everything in this block is a subset of the
2783 * BDRV_CHILD_METADATA path taken above, and so this could
2784 * be an "else if" branch. However, that is not obvious, and
2785 * this function is not performance critical, therefore we let
2786 * this be an independent "if".
2790 * We cannot allow other users to resize the file because the
2791 * format driver might have some assumptions about the size
2792 * (e.g. because it is stored in metadata, or because the file
2793 * is split into fixed-size data files).
2795 shared &= ~BLK_PERM_RESIZE;
2798 * WRITE_UNCHANGED often cannot be performed as such on the
2799 * data file. For example, the qcow2 driver may still need to
2800 * write copied clusters on copy-on-read.
2802 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2803 perm |= BLK_PERM_WRITE;
2807 * If the data file is written to, the format driver may
2808 * expect to be able to resize it by writing beyond the EOF.
2810 if (perm & BLK_PERM_WRITE) {
2811 perm |= BLK_PERM_RESIZE;
2815 if (bs->open_flags & BDRV_O_INACTIVE) {
2816 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2823 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2824 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2825 uint64_t perm, uint64_t shared,
2826 uint64_t *nperm, uint64_t *nshared)
2828 GLOBAL_STATE_CODE();
2829 if (role & BDRV_CHILD_FILTERED) {
2830 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2832 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2833 perm, shared, nperm, nshared);
2834 } else if (role & BDRV_CHILD_COW) {
2835 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2836 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2837 perm, shared, nperm, nshared);
2838 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2839 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2840 perm, shared, nperm, nshared);
2842 g_assert_not_reached();
2846 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2848 static const uint64_t permissions[] = {
2849 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2850 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2851 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2852 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2855 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2856 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2858 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2860 return permissions[qapi_perm];
2864 * Replaces the node that a BdrvChild points to without updating permissions.
2866 * If @new_bs is non-NULL, the parent of @child must already be drained through
2867 * @child and the caller must hold the AioContext lock for @new_bs.
2869 static void GRAPH_WRLOCK
2870 bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs)
2872 BlockDriverState *old_bs = child->bs;
2873 int new_bs_quiesce_counter;
2875 assert(!child->frozen);
2878 * If we want to change the BdrvChild to point to a drained node as its new
2879 * child->bs, we need to make sure that its new parent is drained, too. In
2880 * other words, either child->quiesce_parent must already be true or we must
2881 * be able to set it and keep the parent's quiesce_counter consistent with
2882 * that, but without polling or starting new requests (this function
2883 * guarantees that it doesn't poll, and starting new requests would be
2884 * against the invariants of drain sections).
2886 * To keep things simple, we pick the first option (child->quiesce_parent
2887 * must already be true). We also generalise the rule a bit to make it
2888 * easier to verify in callers and more likely to be covered in test cases:
2889 * The parent must be quiesced through this child even if new_bs isn't
2890 * currently drained.
2892 * The only exception is for callers that always pass new_bs == NULL. In
2893 * this case, we obviously never need to consider the case of a drained
2894 * new_bs, so we can keep the callers simpler by allowing them not to drain
2897 assert(!new_bs || child->quiesced_parent);
2898 assert(old_bs != new_bs);
2899 GLOBAL_STATE_CODE();
2901 if (old_bs && new_bs) {
2902 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2906 if (child->klass->detach) {
2907 child->klass->detach(child);
2909 QLIST_REMOVE(child, next_parent);
2915 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2916 if (child->klass->attach) {
2917 child->klass->attach(child);
2922 * If the parent was drained through this BdrvChild previously, but new_bs
2923 * is not drained, allow requests to come in only after the new node has
2926 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2927 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2928 bdrv_parent_drained_end_single(child);
2933 * Free the given @child.
2935 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2936 * unused (i.e. not in a children list).
2938 static void bdrv_child_free(BdrvChild *child)
2941 GLOBAL_STATE_CODE();
2942 assert(!child->next.le_prev); /* not in children list */
2944 g_free(child->name);
2948 typedef struct BdrvAttachChildCommonState {
2950 AioContext *old_parent_ctx;
2951 AioContext *old_child_ctx;
2952 } BdrvAttachChildCommonState;
2954 static void bdrv_attach_child_common_abort(void *opaque)
2956 BdrvAttachChildCommonState *s = opaque;
2957 BlockDriverState *bs = s->child->bs;
2959 GLOBAL_STATE_CODE();
2961 bdrv_graph_wrlock(NULL);
2962 bdrv_replace_child_noperm(s->child, NULL);
2963 bdrv_graph_wrunlock();
2965 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2966 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2969 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2971 GHashTable *visited;
2976 /* No need to visit `child`, because it has been detached already */
2977 visited = g_hash_table_new(NULL, NULL);
2978 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2979 visited, tran, &error_abort);
2980 g_hash_table_destroy(visited);
2982 /* transaction is supposed to always succeed */
2983 assert(ret == true);
2988 bdrv_child_free(s->child);
2991 static TransactionActionDrv bdrv_attach_child_common_drv = {
2992 .abort = bdrv_attach_child_common_abort,
2997 * Common part of attaching bdrv child to bs or to blk or to job
2999 * Function doesn't update permissions, caller is responsible for this.
3001 * Returns new created child.
3003 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3004 * @child_bs can move to a different AioContext in this function. Callers must
3005 * make sure that their AioContext locking is still correct after this.
3007 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
3008 const char *child_name,
3009 const BdrvChildClass *child_class,
3010 BdrvChildRole child_role,
3011 uint64_t perm, uint64_t shared_perm,
3013 Transaction *tran, Error **errp)
3015 BdrvChild *new_child;
3016 AioContext *parent_ctx, *new_child_ctx;
3017 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3019 assert(child_class->get_parent_desc);
3020 GLOBAL_STATE_CODE();
3022 new_child = g_new(BdrvChild, 1);
3023 *new_child = (BdrvChild) {
3025 .name = g_strdup(child_name),
3026 .klass = child_class,
3029 .shared_perm = shared_perm,
3034 * If the AioContexts don't match, first try to move the subtree of
3035 * child_bs into the AioContext of the new parent. If this doesn't work,
3036 * try moving the parent into the AioContext of child_bs instead.
3038 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3039 if (child_ctx != parent_ctx) {
3040 Error *local_err = NULL;
3041 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3044 if (ret < 0 && child_class->change_aio_ctx) {
3045 Transaction *tran = tran_new();
3046 GHashTable *visited = g_hash_table_new(NULL, NULL);
3049 g_hash_table_add(visited, new_child);
3050 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3051 visited, tran, NULL);
3052 if (ret_child == true) {
3053 error_free(local_err);
3056 tran_finalize(tran, ret_child == true ? 0 : -1);
3057 g_hash_table_destroy(visited);
3061 error_propagate(errp, local_err);
3062 bdrv_child_free(new_child);
3067 new_child_ctx = bdrv_get_aio_context(child_bs);
3068 if (new_child_ctx != child_ctx) {
3069 aio_context_release(child_ctx);
3070 aio_context_acquire(new_child_ctx);
3075 * Let every new BdrvChild start with a drained parent. Inserting the child
3076 * in the graph with bdrv_replace_child_noperm() will undrain it if
3077 * @child_bs is not drained.
3079 * The child was only just created and is not yet visible in global state
3080 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3081 * could have sent requests and polling is not necessary.
3083 * Note that this means that the parent isn't fully drained yet, we only
3084 * stop new requests from coming in. This is fine, we don't care about the
3085 * old requests here, they are not for this child. If another place enters a
3086 * drain section for the same parent, but wants it to be fully quiesced, it
3087 * will not run most of the the code in .drained_begin() again (which is not
3088 * a problem, we already did this), but it will still poll until the parent
3089 * is fully quiesced, so it will not be negatively affected either.
3091 bdrv_graph_wrlock(child_bs);
3092 bdrv_parent_drained_begin_single(new_child);
3093 bdrv_replace_child_noperm(new_child, child_bs);
3094 bdrv_graph_wrunlock();
3096 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3097 *s = (BdrvAttachChildCommonState) {
3099 .old_parent_ctx = parent_ctx,
3100 .old_child_ctx = child_ctx,
3102 tran_add(tran, &bdrv_attach_child_common_drv, s);
3104 if (new_child_ctx != child_ctx) {
3105 aio_context_release(new_child_ctx);
3106 aio_context_acquire(child_ctx);
3113 * Function doesn't update permissions, caller is responsible for this.
3115 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3116 * @child_bs can move to a different AioContext in this function. Callers must
3117 * make sure that their AioContext locking is still correct after this.
3119 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3120 BlockDriverState *child_bs,
3121 const char *child_name,
3122 const BdrvChildClass *child_class,
3123 BdrvChildRole child_role,
3127 uint64_t perm, shared_perm;
3129 assert(parent_bs->drv);
3130 GLOBAL_STATE_CODE();
3132 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3133 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3134 child_bs->node_name, child_name, parent_bs->node_name);
3138 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3139 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3140 perm, shared_perm, &perm, &shared_perm);
3142 return bdrv_attach_child_common(child_bs, child_name, child_class,
3143 child_role, perm, shared_perm, parent_bs,
3148 * This function steals the reference to child_bs from the caller.
3149 * That reference is later dropped by bdrv_root_unref_child().
3151 * On failure NULL is returned, errp is set and the reference to
3152 * child_bs is also dropped.
3154 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3155 * (unless @child_bs is already in @ctx).
3157 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3158 const char *child_name,
3159 const BdrvChildClass *child_class,
3160 BdrvChildRole child_role,
3161 uint64_t perm, uint64_t shared_perm,
3162 void *opaque, Error **errp)
3166 Transaction *tran = tran_new();
3168 GLOBAL_STATE_CODE();
3170 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3171 child_role, perm, shared_perm, opaque,
3178 ret = bdrv_refresh_perms(child_bs, tran, errp);
3181 tran_finalize(tran, ret);
3183 bdrv_unref(child_bs);
3185 return ret < 0 ? NULL : child;
3189 * This function transfers the reference to child_bs from the caller
3190 * to parent_bs. That reference is later dropped by parent_bs on
3191 * bdrv_close() or if someone calls bdrv_unref_child().
3193 * On failure NULL is returned, errp is set and the reference to
3194 * child_bs is also dropped.
3196 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3197 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3199 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3200 BlockDriverState *child_bs,
3201 const char *child_name,
3202 const BdrvChildClass *child_class,
3203 BdrvChildRole child_role,
3208 Transaction *tran = tran_new();
3210 GLOBAL_STATE_CODE();
3212 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3213 child_class, child_role, tran, errp);
3219 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3225 tran_finalize(tran, ret);
3227 bdrv_unref(child_bs);
3229 return ret < 0 ? NULL : child;
3232 /* Callers must ensure that child->frozen is false. */
3233 void bdrv_root_unref_child(BdrvChild *child)
3235 BlockDriverState *child_bs = child->bs;
3237 GLOBAL_STATE_CODE();
3238 bdrv_graph_wrlock(NULL);
3239 bdrv_replace_child_noperm(child, NULL);
3240 bdrv_graph_wrunlock();
3241 bdrv_child_free(child);
3245 * Update permissions for old node. We're just taking a parent away, so
3246 * we're loosening restrictions. Errors of permission update are not
3247 * fatal in this case, ignore them.
3249 bdrv_refresh_perms(child_bs, NULL, NULL);
3252 * When the parent requiring a non-default AioContext is removed, the
3253 * node moves back to the main AioContext
3255 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3259 bdrv_unref(child_bs);
3262 typedef struct BdrvSetInheritsFrom {
3263 BlockDriverState *bs;
3264 BlockDriverState *old_inherits_from;
3265 } BdrvSetInheritsFrom;
3267 static void bdrv_set_inherits_from_abort(void *opaque)
3269 BdrvSetInheritsFrom *s = opaque;
3271 s->bs->inherits_from = s->old_inherits_from;
3274 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3275 .abort = bdrv_set_inherits_from_abort,
3279 /* @tran is allowed to be NULL. In this case no rollback is possible */
3280 static void bdrv_set_inherits_from(BlockDriverState *bs,
3281 BlockDriverState *new_inherits_from,
3285 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3287 *s = (BdrvSetInheritsFrom) {
3289 .old_inherits_from = bs->inherits_from,
3292 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3295 bs->inherits_from = new_inherits_from;
3299 * Clear all inherits_from pointers from children and grandchildren of
3300 * @root that point to @root, where necessary.
3301 * @tran is allowed to be NULL. In this case no rollback is possible
3303 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3308 if (child->bs->inherits_from == root) {
3310 * Remove inherits_from only when the last reference between root and
3311 * child->bs goes away.
3313 QLIST_FOREACH(c, &root->children, next) {
3314 if (c != child && c->bs == child->bs) {
3319 bdrv_set_inherits_from(child->bs, NULL, tran);
3323 QLIST_FOREACH(c, &child->bs->children, next) {
3324 bdrv_unset_inherits_from(root, c, tran);
3328 /* Callers must ensure that child->frozen is false. */
3329 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3331 GLOBAL_STATE_CODE();
3332 if (child == NULL) {
3336 bdrv_unset_inherits_from(parent, child, NULL);
3337 bdrv_root_unref_child(child);
3341 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3344 GLOBAL_STATE_CODE();
3345 QLIST_FOREACH(c, &bs->parents, next_parent) {
3346 if (c->klass->change_media) {
3347 c->klass->change_media(c, load);
3352 /* Return true if you can reach parent going through child->inherits_from
3353 * recursively. If parent or child are NULL, return false */
3354 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3355 BlockDriverState *parent)
3357 while (child && child != parent) {
3358 child = child->inherits_from;
3361 return child != NULL;
3365 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3366 * mostly used for COW backing children (role = COW), but also for
3367 * filtered children (role = FILTERED | PRIMARY).
3369 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3371 if (bs->drv && bs->drv->is_filter) {
3372 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3374 return BDRV_CHILD_COW;
3379 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3380 * callers which don't need their own reference any more must call bdrv_unref().
3382 * Function doesn't update permissions, caller is responsible for this.
3384 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3385 * @child_bs can move to a different AioContext in this function. Callers must
3386 * make sure that their AioContext locking is still correct after this.
3388 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3389 BlockDriverState *child_bs,
3391 Transaction *tran, Error **errp)
3393 bool update_inherits_from =
3394 bdrv_inherits_from_recursive(child_bs, parent_bs);
3395 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3398 GLOBAL_STATE_CODE();
3400 if (!parent_bs->drv) {
3402 * Node without drv is an object without a class :/. TODO: finally fix
3403 * qcow2 driver to never clear bs->drv and implement format corruption
3404 * handling in other way.
3406 error_setg(errp, "Node corrupted");
3410 if (child && child->frozen) {
3411 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3412 child->name, parent_bs->node_name, child->bs->node_name);
3416 if (is_backing && !parent_bs->drv->is_filter &&
3417 !parent_bs->drv->supports_backing)
3419 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3420 "files", parent_bs->drv->format_name, parent_bs->node_name);
3424 if (parent_bs->drv->is_filter) {
3425 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3426 } else if (is_backing) {
3427 role = BDRV_CHILD_COW;
3430 * We only can use same role as it is in existing child. We don't have
3431 * infrastructure to determine role of file child in generic way
3434 error_setg(errp, "Cannot set file child to format node without "
3442 bdrv_unset_inherits_from(parent_bs, child, tran);
3443 bdrv_remove_child(child, tran);
3450 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3451 is_backing ? "backing" : "file",
3452 &child_of_bds, role,
3460 * If inherits_from pointed recursively to bs then let's update it to
3461 * point directly to bs (else it will become NULL).
3463 if (update_inherits_from) {
3464 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3468 bdrv_graph_rdlock_main_loop();
3469 bdrv_refresh_limits(parent_bs, tran, NULL);
3470 bdrv_graph_rdunlock_main_loop();
3476 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3477 * @backing_hd can move to a different AioContext in this function. Callers must
3478 * make sure that their AioContext locking is still correct after this.
3480 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3481 BlockDriverState *backing_hd,
3482 Transaction *tran, Error **errp)
3484 GLOBAL_STATE_CODE();
3485 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3488 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3489 BlockDriverState *backing_hd,
3493 Transaction *tran = tran_new();
3495 GLOBAL_STATE_CODE();
3496 assert(bs->quiesce_counter > 0);
3498 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3503 ret = bdrv_refresh_perms(bs, tran, errp);
3505 tran_finalize(tran, ret);
3509 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3513 GLOBAL_STATE_CODE();
3515 bdrv_drained_begin(bs);
3516 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3517 bdrv_drained_end(bs);
3523 * Opens the backing file for a BlockDriverState if not yet open
3525 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3526 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3527 * itself, all options starting with "${bdref_key}." are considered part of the
3530 * The caller must hold the main AioContext lock.
3532 * TODO Can this be unified with bdrv_open_image()?
3534 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3535 const char *bdref_key, Error **errp)
3537 char *backing_filename = NULL;
3538 char *bdref_key_dot;
3539 const char *reference = NULL;
3541 bool implicit_backing = false;
3542 BlockDriverState *backing_hd;
3543 AioContext *backing_hd_ctx;
3545 QDict *tmp_parent_options = NULL;
3546 Error *local_err = NULL;
3548 GLOBAL_STATE_CODE();
3550 if (bs->backing != NULL) {
3554 /* NULL means an empty set of options */
3555 if (parent_options == NULL) {
3556 tmp_parent_options = qdict_new();
3557 parent_options = tmp_parent_options;
3560 bs->open_flags &= ~BDRV_O_NO_BACKING;
3562 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3563 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3564 g_free(bdref_key_dot);
3567 * Caution: while qdict_get_try_str() is fine, getting non-string
3568 * types would require more care. When @parent_options come from
3569 * -blockdev or blockdev_add, its members are typed according to
3570 * the QAPI schema, but when they come from -drive, they're all
3573 reference = qdict_get_try_str(parent_options, bdref_key);
3574 if (reference || qdict_haskey(options, "file.filename")) {
3575 /* keep backing_filename NULL */
3576 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3577 qobject_unref(options);
3580 if (qdict_size(options) == 0) {
3581 /* If the user specifies options that do not modify the
3582 * backing file's behavior, we might still consider it the
3583 * implicit backing file. But it's easier this way, and
3584 * just specifying some of the backing BDS's options is
3585 * only possible with -drive anyway (otherwise the QAPI
3586 * schema forces the user to specify everything). */
3587 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3590 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3593 error_propagate(errp, local_err);
3594 qobject_unref(options);
3599 if (!bs->drv || !bs->drv->supports_backing) {
3601 error_setg(errp, "Driver doesn't support backing files");
3602 qobject_unref(options);
3607 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3608 qdict_put_str(options, "driver", bs->backing_format);
3611 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3612 &child_of_bds, bdrv_backing_role(bs), errp);
3614 bs->open_flags |= BDRV_O_NO_BACKING;
3615 error_prepend(errp, "Could not open backing file: ");
3620 if (implicit_backing) {
3621 bdrv_refresh_filename(backing_hd);
3622 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3623 backing_hd->filename);
3626 /* Hook up the backing file link; drop our reference, bs owns the
3627 * backing_hd reference now */
3628 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3629 aio_context_acquire(backing_hd_ctx);
3630 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3631 bdrv_unref(backing_hd);
3632 aio_context_release(backing_hd_ctx);
3638 qdict_del(parent_options, bdref_key);
3641 g_free(backing_filename);
3642 qobject_unref(tmp_parent_options);
3646 static BlockDriverState *
3647 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3648 BlockDriverState *parent, const BdrvChildClass *child_class,
3649 BdrvChildRole child_role, bool allow_none, Error **errp)
3651 BlockDriverState *bs = NULL;
3652 QDict *image_options;
3653 char *bdref_key_dot;
3654 const char *reference;
3656 assert(child_class != NULL);
3658 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3659 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3660 g_free(bdref_key_dot);
3663 * Caution: while qdict_get_try_str() is fine, getting non-string
3664 * types would require more care. When @options come from
3665 * -blockdev or blockdev_add, its members are typed according to
3666 * the QAPI schema, but when they come from -drive, they're all
3669 reference = qdict_get_try_str(options, bdref_key);
3670 if (!filename && !reference && !qdict_size(image_options)) {
3672 error_setg(errp, "A block device must be specified for \"%s\"",
3675 qobject_unref(image_options);
3679 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3680 parent, child_class, child_role, errp);
3686 qdict_del(options, bdref_key);
3691 * Opens a disk image whose options are given as BlockdevRef in another block
3694 * If allow_none is true, no image will be opened if filename is false and no
3695 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3697 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3698 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3699 * itself, all options starting with "${bdref_key}." are considered part of the
3702 * The BlockdevRef will be removed from the options QDict.
3704 * The caller must hold the lock of the main AioContext and no other AioContext.
3705 * @parent can move to a different AioContext in this function. Callers must
3706 * make sure that their AioContext locking is still correct after this.
3708 BdrvChild *bdrv_open_child(const char *filename,
3709 QDict *options, const char *bdref_key,
3710 BlockDriverState *parent,
3711 const BdrvChildClass *child_class,
3712 BdrvChildRole child_role,
3713 bool allow_none, Error **errp)
3715 BlockDriverState *bs;
3719 GLOBAL_STATE_CODE();
3721 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3722 child_role, allow_none, errp);
3727 ctx = bdrv_get_aio_context(bs);
3728 aio_context_acquire(ctx);
3729 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3731 aio_context_release(ctx);
3737 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3739 * The caller must hold the lock of the main AioContext and no other AioContext.
3740 * @parent can move to a different AioContext in this function. Callers must
3741 * make sure that their AioContext locking is still correct after this.
3743 int bdrv_open_file_child(const char *filename,
3744 QDict *options, const char *bdref_key,
3745 BlockDriverState *parent, Error **errp)
3749 /* commit_top and mirror_top don't use this function */
3750 assert(!parent->drv->filtered_child_is_backing);
3751 role = parent->drv->is_filter ?
3752 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3754 if (!bdrv_open_child(filename, options, bdref_key, parent,
3755 &child_of_bds, role, false, errp))
3764 * TODO Future callers may need to specify parent/child_class in order for
3765 * option inheritance to work. Existing callers use it for the root node.
3767 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3769 BlockDriverState *bs = NULL;
3770 QObject *obj = NULL;
3771 QDict *qdict = NULL;
3772 const char *reference = NULL;
3775 GLOBAL_STATE_CODE();
3777 if (ref->type == QTYPE_QSTRING) {
3778 reference = ref->u.reference;
3780 BlockdevOptions *options = &ref->u.definition;
3781 assert(ref->type == QTYPE_QDICT);
3783 v = qobject_output_visitor_new(&obj);
3784 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3785 visit_complete(v, &obj);
3787 qdict = qobject_to(QDict, obj);
3788 qdict_flatten(qdict);
3790 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3791 * compatibility with other callers) rather than what we want as the
3792 * real defaults. Apply the defaults here instead. */
3793 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3794 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3795 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3796 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3800 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3807 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3809 QDict *snapshot_options,
3812 g_autofree char *tmp_filename = NULL;
3814 QemuOpts *opts = NULL;
3815 BlockDriverState *bs_snapshot = NULL;
3816 AioContext *ctx = bdrv_get_aio_context(bs);
3819 GLOBAL_STATE_CODE();
3821 /* if snapshot, we create a temporary backing file and open it
3822 instead of opening 'filename' directly */
3824 /* Get the required size from the image */
3825 aio_context_acquire(ctx);
3826 total_size = bdrv_getlength(bs);
3827 aio_context_release(ctx);
3829 if (total_size < 0) {
3830 error_setg_errno(errp, -total_size, "Could not get image size");
3834 /* Create the temporary image */
3835 tmp_filename = create_tmp_file(errp);
3836 if (!tmp_filename) {
3840 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3842 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3843 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3844 qemu_opts_del(opts);
3846 error_prepend(errp, "Could not create temporary overlay '%s': ",
3851 /* Prepare options QDict for the temporary file */
3852 qdict_put_str(snapshot_options, "file.driver", "file");
3853 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3854 qdict_put_str(snapshot_options, "driver", "qcow2");
3856 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3857 snapshot_options = NULL;
3862 aio_context_acquire(ctx);
3863 ret = bdrv_append(bs_snapshot, bs, errp);
3864 aio_context_release(ctx);
3872 qobject_unref(snapshot_options);
3877 * Opens a disk image (raw, qcow2, vmdk, ...)
3879 * options is a QDict of options to pass to the block drivers, or NULL for an
3880 * empty set of options. The reference to the QDict belongs to the block layer
3881 * after the call (even on failure), so if the caller intends to reuse the
3882 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3884 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3885 * If it is not NULL, the referenced BDS will be reused.
3887 * The reference parameter may be used to specify an existing block device which
3888 * should be opened. If specified, neither options nor a filename may be given,
3889 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3891 * The caller must always hold the main AioContext lock.
3893 static BlockDriverState * no_coroutine_fn
3894 bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3895 int flags, BlockDriverState *parent,
3896 const BdrvChildClass *child_class, BdrvChildRole child_role,
3900 BlockBackend *file = NULL;
3901 BlockDriverState *bs;
3902 BlockDriver *drv = NULL;
3904 const char *drvname;
3905 const char *backing;
3906 Error *local_err = NULL;
3907 QDict *snapshot_options = NULL;
3908 int snapshot_flags = 0;
3909 AioContext *ctx = qemu_get_aio_context();
3911 assert(!child_class || !flags);
3912 assert(!child_class == !parent);
3913 GLOBAL_STATE_CODE();
3914 assert(!qemu_in_coroutine());
3917 bool options_non_empty = options ? qdict_size(options) : false;
3918 qobject_unref(options);
3920 if (filename || options_non_empty) {
3921 error_setg(errp, "Cannot reference an existing block device with "
3922 "additional options or a new filename");
3926 bs = bdrv_lookup_bs(reference, reference, errp);
3937 /* NULL means an empty set of options */
3938 if (options == NULL) {
3939 options = qdict_new();
3942 /* json: syntax counts as explicit options, as if in the QDict */
3943 parse_json_protocol(options, &filename, &local_err);
3948 bs->explicit_options = qdict_clone_shallow(options);
3951 bool parent_is_format;
3954 parent_is_format = parent->drv->is_format;
3957 * parent->drv is not set yet because this node is opened for
3958 * (potential) format probing. That means that @parent is going
3959 * to be a format node.
3961 parent_is_format = true;
3964 bs->inherits_from = parent;
3965 child_class->inherit_options(child_role, parent_is_format,
3967 parent->open_flags, parent->options);
3970 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3976 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3977 * Caution: getting a boolean member of @options requires care.
3978 * When @options come from -blockdev or blockdev_add, members are
3979 * typed according to the QAPI schema, but when they come from
3980 * -drive, they're all QString.
3982 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3983 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3984 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3986 flags &= ~BDRV_O_RDWR;
3989 if (flags & BDRV_O_SNAPSHOT) {
3990 snapshot_options = qdict_new();
3991 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3993 /* Let bdrv_backing_options() override "read-only" */
3994 qdict_del(options, BDRV_OPT_READ_ONLY);
3995 bdrv_inherited_options(BDRV_CHILD_COW, true,
3996 &flags, options, flags, options);
3999 bs->open_flags = flags;
4000 bs->options = options;
4001 options = qdict_clone_shallow(options);
4003 /* Find the right image format driver */
4004 /* See cautionary note on accessing @options above */
4005 drvname = qdict_get_try_str(options, "driver");
4007 drv = bdrv_find_format(drvname);
4009 error_setg(errp, "Unknown driver: '%s'", drvname);
4014 assert(drvname || !(flags & BDRV_O_PROTOCOL));
4016 /* See cautionary note on accessing @options above */
4017 backing = qdict_get_try_str(options, "backing");
4018 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4019 (backing && *backing == '\0'))
4022 warn_report("Use of \"backing\": \"\" is deprecated; "
4023 "use \"backing\": null instead");
4025 flags |= BDRV_O_NO_BACKING;
4026 qdict_del(bs->explicit_options, "backing");
4027 qdict_del(bs->options, "backing");
4028 qdict_del(options, "backing");
4031 /* Open image file without format layer. This BlockBackend is only used for
4032 * probing, the block drivers will do their own bdrv_open_child() for the
4033 * same BDS, which is why we put the node name back into options. */
4034 if ((flags & BDRV_O_PROTOCOL) == 0) {
4035 BlockDriverState *file_bs;
4037 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
4038 &child_of_bds, BDRV_CHILD_IMAGE,
4043 if (file_bs != NULL) {
4044 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4045 * looking at the header to guess the image format. This works even
4046 * in cases where a guest would not see a consistent state. */
4047 ctx = bdrv_get_aio_context(file_bs);
4048 aio_context_acquire(ctx);
4049 file = blk_new(ctx, 0, BLK_PERM_ALL);
4050 blk_insert_bs(file, file_bs, &local_err);
4051 bdrv_unref(file_bs);
4052 aio_context_release(ctx);
4058 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4062 /* Image format probing */
4065 ret = find_image_format(file, filename, &drv, &local_err);
4070 * This option update would logically belong in bdrv_fill_options(),
4071 * but we first need to open bs->file for the probing to work, while
4072 * opening bs->file already requires the (mostly) final set of options
4073 * so that cache mode etc. can be inherited.
4075 * Adding the driver later is somewhat ugly, but it's not an option
4076 * that would ever be inherited, so it's correct. We just need to make
4077 * sure to update both bs->options (which has the full effective
4078 * options for bs) and options (which has file.* already removed).
4080 qdict_put_str(bs->options, "driver", drv->format_name);
4081 qdict_put_str(options, "driver", drv->format_name);
4083 error_setg(errp, "Must specify either driver or file");
4087 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4088 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4089 /* file must be NULL if a protocol BDS is about to be created
4090 * (the inverse results in an error message from bdrv_open_common()) */
4091 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4093 /* Open the image */
4094 ret = bdrv_open_common(bs, file, options, &local_err);
4099 /* The AioContext could have changed during bdrv_open_common() */
4100 ctx = bdrv_get_aio_context(bs);
4103 aio_context_acquire(ctx);
4105 aio_context_release(ctx);
4109 /* If there is a backing file, use it */
4110 if ((flags & BDRV_O_NO_BACKING) == 0) {
4111 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4113 goto close_and_fail;
4117 /* Remove all children options and references
4118 * from bs->options and bs->explicit_options */
4119 QLIST_FOREACH(child, &bs->children, next) {
4120 char *child_key_dot;
4121 child_key_dot = g_strdup_printf("%s.", child->name);
4122 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4123 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4124 qdict_del(bs->explicit_options, child->name);
4125 qdict_del(bs->options, child->name);
4126 g_free(child_key_dot);
4129 /* Check if any unknown options were used */
4130 if (qdict_size(options) != 0) {
4131 const QDictEntry *entry = qdict_first(options);
4132 if (flags & BDRV_O_PROTOCOL) {
4133 error_setg(errp, "Block protocol '%s' doesn't support the option "
4134 "'%s'", drv->format_name, entry->key);
4137 "Block format '%s' does not support the option '%s'",
4138 drv->format_name, entry->key);
4141 goto close_and_fail;
4144 bdrv_parent_cb_change_media(bs, true);
4146 qobject_unref(options);
4149 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4150 * temporary snapshot afterwards. */
4151 if (snapshot_flags) {
4152 BlockDriverState *snapshot_bs;
4153 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4154 snapshot_options, &local_err);
4155 snapshot_options = NULL;
4157 goto close_and_fail;
4159 /* We are not going to return bs but the overlay on top of it
4160 * (snapshot_bs); thus, we have to drop the strong reference to bs
4161 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4162 * though, because the overlay still has a reference to it. */
4163 aio_context_acquire(ctx);
4165 aio_context_release(ctx);
4172 aio_context_acquire(ctx);
4174 qobject_unref(snapshot_options);
4175 qobject_unref(bs->explicit_options);
4176 qobject_unref(bs->options);
4177 qobject_unref(options);
4179 bs->explicit_options = NULL;
4181 aio_context_release(ctx);
4182 error_propagate(errp, local_err);
4186 aio_context_acquire(ctx);
4188 aio_context_release(ctx);
4189 qobject_unref(snapshot_options);
4190 qobject_unref(options);
4191 error_propagate(errp, local_err);
4195 /* The caller must always hold the main AioContext lock. */
4196 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4197 QDict *options, int flags, Error **errp)
4199 GLOBAL_STATE_CODE();
4201 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4205 /* Return true if the NULL-terminated @list contains @str */
4206 static bool is_str_in_list(const char *str, const char *const *list)
4210 for (i = 0; list[i] != NULL; i++) {
4211 if (!strcmp(str, list[i])) {
4220 * Check that every option set in @bs->options is also set in
4223 * Options listed in the common_options list and in
4224 * @bs->drv->mutable_opts are skipped.
4226 * Return 0 on success, otherwise return -EINVAL and set @errp.
4228 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4229 const QDict *new_opts, Error **errp)
4231 const QDictEntry *e;
4232 /* These options are common to all block drivers and are handled
4233 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4234 const char *const common_options[] = {
4235 "node-name", "discard", "cache.direct", "cache.no-flush",
4236 "read-only", "auto-read-only", "detect-zeroes", NULL
4239 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4240 if (!qdict_haskey(new_opts, e->key) &&
4241 !is_str_in_list(e->key, common_options) &&
4242 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4243 error_setg(errp, "Option '%s' cannot be reset "
4244 "to its default value", e->key);
4253 * Returns true if @child can be reached recursively from @bs
4255 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4256 BlockDriverState *child)
4264 QLIST_FOREACH(c, &bs->children, next) {
4265 if (bdrv_recurse_has_child(c->bs, child)) {
4274 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4275 * reopen of multiple devices.
4277 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4278 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4279 * be created and initialized. This newly created BlockReopenQueue should be
4280 * passed back in for subsequent calls that are intended to be of the same
4283 * bs is the BlockDriverState to add to the reopen queue.
4285 * options contains the changed options for the associated bs
4286 * (the BlockReopenQueue takes ownership)
4288 * flags contains the open flags for the associated bs
4290 * returns a pointer to bs_queue, which is either the newly allocated
4291 * bs_queue, or the existing bs_queue being used.
4293 * bs is drained here and undrained by bdrv_reopen_queue_free().
4295 * To be called with bs->aio_context locked.
4297 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4298 BlockDriverState *bs,
4300 const BdrvChildClass *klass,
4302 bool parent_is_format,
4303 QDict *parent_options,
4309 BlockReopenQueueEntry *bs_entry;
4311 QDict *old_options, *explicit_options, *options_copy;
4315 GLOBAL_STATE_CODE();
4317 bdrv_drained_begin(bs);
4319 if (bs_queue == NULL) {
4320 bs_queue = g_new0(BlockReopenQueue, 1);
4321 QTAILQ_INIT(bs_queue);
4325 options = qdict_new();
4328 /* Check if this BlockDriverState is already in the queue */
4329 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4330 if (bs == bs_entry->state.bs) {
4336 * Precedence of options:
4337 * 1. Explicitly passed in options (highest)
4338 * 2. Retained from explicitly set options of bs
4339 * 3. Inherited from parent node
4340 * 4. Retained from effective options of bs
4343 /* Old explicitly set values (don't overwrite by inherited value) */
4344 if (bs_entry || keep_old_opts) {
4345 old_options = qdict_clone_shallow(bs_entry ?
4346 bs_entry->state.explicit_options :
4347 bs->explicit_options);
4348 bdrv_join_options(bs, options, old_options);
4349 qobject_unref(old_options);
4352 explicit_options = qdict_clone_shallow(options);
4354 /* Inherit from parent node */
4355 if (parent_options) {
4357 klass->inherit_options(role, parent_is_format, &flags, options,
4358 parent_flags, parent_options);
4360 flags = bdrv_get_flags(bs);
4363 if (keep_old_opts) {
4364 /* Old values are used for options that aren't set yet */
4365 old_options = qdict_clone_shallow(bs->options);
4366 bdrv_join_options(bs, options, old_options);
4367 qobject_unref(old_options);
4370 /* We have the final set of options so let's update the flags */
4371 options_copy = qdict_clone_shallow(options);
4372 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4373 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4374 update_flags_from_options(&flags, opts);
4375 qemu_opts_del(opts);
4376 qobject_unref(options_copy);
4378 /* bdrv_open_inherit() sets and clears some additional flags internally */
4379 flags &= ~BDRV_O_PROTOCOL;
4380 if (flags & BDRV_O_RDWR) {
4381 flags |= BDRV_O_ALLOW_RDWR;
4385 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4386 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4388 qobject_unref(bs_entry->state.options);
4389 qobject_unref(bs_entry->state.explicit_options);
4392 bs_entry->state.bs = bs;
4393 bs_entry->state.options = options;
4394 bs_entry->state.explicit_options = explicit_options;
4395 bs_entry->state.flags = flags;
4398 * If keep_old_opts is false then it means that unspecified
4399 * options must be reset to their original value. We don't allow
4400 * resetting 'backing' but we need to know if the option is
4401 * missing in order to decide if we have to return an error.
4403 if (!keep_old_opts) {
4404 bs_entry->state.backing_missing =
4405 !qdict_haskey(options, "backing") &&
4406 !qdict_haskey(options, "backing.driver");
4409 QLIST_FOREACH(child, &bs->children, next) {
4410 QDict *new_child_options = NULL;
4411 bool child_keep_old = keep_old_opts;
4413 /* reopen can only change the options of block devices that were
4414 * implicitly created and inherited options. For other (referenced)
4415 * block devices, a syntax like "backing.foo" results in an error. */
4416 if (child->bs->inherits_from != bs) {
4420 /* Check if the options contain a child reference */
4421 if (qdict_haskey(options, child->name)) {
4422 const char *childref = qdict_get_try_str(options, child->name);
4424 * The current child must not be reopened if the child
4425 * reference is null or points to a different node.
4427 if (g_strcmp0(childref, child->bs->node_name)) {
4431 * If the child reference points to the current child then
4432 * reopen it with its existing set of options (note that
4433 * it can still inherit new options from the parent).
4435 child_keep_old = true;
4437 /* Extract child options ("child-name.*") */
4438 char *child_key_dot = g_strdup_printf("%s.", child->name);
4439 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4440 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4441 g_free(child_key_dot);
4444 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4445 child->klass, child->role, bs->drv->is_format,
4446 options, flags, child_keep_old);
4452 /* To be called with bs->aio_context locked */
4453 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4454 BlockDriverState *bs,
4455 QDict *options, bool keep_old_opts)
4457 GLOBAL_STATE_CODE();
4459 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4460 NULL, 0, keep_old_opts);
4463 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4465 GLOBAL_STATE_CODE();
4467 BlockReopenQueueEntry *bs_entry, *next;
4468 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4469 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4471 aio_context_acquire(ctx);
4472 bdrv_drained_end(bs_entry->state.bs);
4473 aio_context_release(ctx);
4475 qobject_unref(bs_entry->state.explicit_options);
4476 qobject_unref(bs_entry->state.options);
4484 * Reopen multiple BlockDriverStates atomically & transactionally.
4486 * The queue passed in (bs_queue) must have been built up previous
4487 * via bdrv_reopen_queue().
4489 * Reopens all BDS specified in the queue, with the appropriate
4490 * flags. All devices are prepared for reopen, and failure of any
4491 * device will cause all device changes to be abandoned, and intermediate
4494 * If all devices prepare successfully, then the changes are committed
4497 * All affected nodes must be drained between bdrv_reopen_queue() and
4498 * bdrv_reopen_multiple().
4500 * To be called from the main thread, with all other AioContexts unlocked.
4502 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4505 BlockReopenQueueEntry *bs_entry, *next;
4507 Transaction *tran = tran_new();
4508 g_autoptr(GSList) refresh_list = NULL;
4510 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4511 assert(bs_queue != NULL);
4512 GLOBAL_STATE_CODE();
4514 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4515 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4516 aio_context_acquire(ctx);
4517 ret = bdrv_flush(bs_entry->state.bs);
4518 aio_context_release(ctx);
4520 error_setg_errno(errp, -ret, "Error flushing drive");
4525 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4526 assert(bs_entry->state.bs->quiesce_counter > 0);
4527 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4528 aio_context_acquire(ctx);
4529 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4530 aio_context_release(ctx);
4534 bs_entry->prepared = true;
4537 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4538 BDRVReopenState *state = &bs_entry->state;
4540 refresh_list = g_slist_prepend(refresh_list, state->bs);
4541 if (state->old_backing_bs) {
4542 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4544 if (state->old_file_bs) {
4545 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4550 * Note that file-posix driver rely on permission update done during reopen
4551 * (even if no permission changed), because it wants "new" permissions for
4552 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4553 * in raw_reopen_prepare() which is called with "old" permissions.
4555 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4561 * If we reach this point, we have success and just need to apply the
4564 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4565 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4566 * children are usually goes after parents in reopen-queue, so go from last
4569 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4570 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4571 aio_context_acquire(ctx);
4572 bdrv_reopen_commit(&bs_entry->state);
4573 aio_context_release(ctx);
4578 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4579 BlockDriverState *bs = bs_entry->state.bs;
4581 if (bs->drv->bdrv_reopen_commit_post) {
4582 ctx = bdrv_get_aio_context(bs);
4583 aio_context_acquire(ctx);
4584 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4585 aio_context_release(ctx);
4594 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4595 if (bs_entry->prepared) {
4596 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4597 aio_context_acquire(ctx);
4598 bdrv_reopen_abort(&bs_entry->state);
4599 aio_context_release(ctx);
4604 bdrv_reopen_queue_free(bs_queue);
4609 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4612 AioContext *ctx = bdrv_get_aio_context(bs);
4613 BlockReopenQueue *queue;
4616 GLOBAL_STATE_CODE();
4618 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4620 if (ctx != qemu_get_aio_context()) {
4621 aio_context_release(ctx);
4623 ret = bdrv_reopen_multiple(queue, errp);
4625 if (ctx != qemu_get_aio_context()) {
4626 aio_context_acquire(ctx);
4632 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4635 QDict *opts = qdict_new();
4637 GLOBAL_STATE_CODE();
4639 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4641 return bdrv_reopen(bs, opts, true, errp);
4645 * Take a BDRVReopenState and check if the value of 'backing' in the
4646 * reopen_state->options QDict is valid or not.
4648 * If 'backing' is missing from the QDict then return 0.
4650 * If 'backing' contains the node name of the backing file of
4651 * reopen_state->bs then return 0.
4653 * If 'backing' contains a different node name (or is null) then check
4654 * whether the current backing file can be replaced with the new one.
4655 * If that's the case then reopen_state->replace_backing_bs is set to
4656 * true and reopen_state->new_backing_bs contains a pointer to the new
4657 * backing BlockDriverState (or NULL).
4659 * Return 0 on success, otherwise return < 0 and set @errp.
4661 * The caller must hold the AioContext lock of @reopen_state->bs.
4662 * @reopen_state->bs can move to a different AioContext in this function.
4663 * Callers must make sure that their AioContext locking is still correct after
4666 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4667 bool is_backing, Transaction *tran,
4670 BlockDriverState *bs = reopen_state->bs;
4671 BlockDriverState *new_child_bs;
4672 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4674 const char *child_name = is_backing ? "backing" : "file";
4677 AioContext *ctx, *old_ctx;
4680 GLOBAL_STATE_CODE();
4682 value = qdict_get(reopen_state->options, child_name);
4683 if (value == NULL) {
4687 switch (qobject_type(value)) {
4689 assert(is_backing); /* The 'file' option does not allow a null value */
4690 new_child_bs = NULL;
4693 str = qstring_get_str(qobject_to(QString, value));
4694 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4695 if (new_child_bs == NULL) {
4697 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4698 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4699 "cycle", str, child_name, bs->node_name);
4705 * The options QDict has been flattened, so 'backing' and 'file'
4706 * do not allow any other data type here.
4708 g_assert_not_reached();
4711 if (old_child_bs == new_child_bs) {
4716 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4720 if (old_child_bs->implicit) {
4721 error_setg(errp, "Cannot replace implicit %s child of %s",
4722 child_name, bs->node_name);
4727 if (bs->drv->is_filter && !old_child_bs) {
4729 * Filters always have a file or a backing child, so we are trying to
4730 * change wrong child
4732 error_setg(errp, "'%s' is a %s filter node that does not support a "
4733 "%s child", bs->node_name, bs->drv->format_name, child_name);
4738 reopen_state->old_backing_bs = old_child_bs;
4740 reopen_state->old_file_bs = old_child_bs;
4743 old_ctx = bdrv_get_aio_context(bs);
4744 ctx = bdrv_get_aio_context(new_child_bs);
4745 if (old_ctx != ctx) {
4746 aio_context_release(old_ctx);
4747 aio_context_acquire(ctx);
4750 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4753 if (old_ctx != ctx) {
4754 aio_context_release(ctx);
4755 aio_context_acquire(old_ctx);
4762 * Prepares a BlockDriverState for reopen. All changes are staged in the
4763 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4764 * the block driver layer .bdrv_reopen_prepare()
4766 * bs is the BlockDriverState to reopen
4767 * flags are the new open flags
4768 * queue is the reopen queue
4770 * Returns 0 on success, non-zero on error. On error errp will be set
4773 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4774 * It is the responsibility of the caller to then call the abort() or
4775 * commit() for any other BDS that have been left in a prepare() state
4777 * The caller must hold the AioContext lock of @reopen_state->bs.
4779 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4780 BlockReopenQueue *queue,
4781 Transaction *change_child_tran, Error **errp)
4785 Error *local_err = NULL;
4788 QDict *orig_reopen_opts;
4789 char *discard = NULL;
4791 bool drv_prepared = false;
4793 assert(reopen_state != NULL);
4794 assert(reopen_state->bs->drv != NULL);
4795 GLOBAL_STATE_CODE();
4796 drv = reopen_state->bs->drv;
4798 /* This function and each driver's bdrv_reopen_prepare() remove
4799 * entries from reopen_state->options as they are processed, so
4800 * we need to make a copy of the original QDict. */
4801 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4803 /* Process generic block layer options */
4804 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4805 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4810 /* This was already called in bdrv_reopen_queue_child() so the flags
4811 * are up-to-date. This time we simply want to remove the options from
4812 * QemuOpts in order to indicate that they have been processed. */
4813 old_flags = reopen_state->flags;
4814 update_flags_from_options(&reopen_state->flags, opts);
4815 assert(old_flags == reopen_state->flags);
4817 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4818 if (discard != NULL) {
4819 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4820 error_setg(errp, "Invalid discard option");
4826 reopen_state->detect_zeroes =
4827 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4829 error_propagate(errp, local_err);
4834 /* All other options (including node-name and driver) must be unchanged.
4835 * Put them back into the QDict, so that they are checked at the end
4836 * of this function. */
4837 qemu_opts_to_qdict(opts, reopen_state->options);
4839 /* If we are to stay read-only, do not allow permission change
4840 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4841 * not set, or if the BDS still has copy_on_read enabled */
4842 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4843 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4845 error_propagate(errp, local_err);
4849 if (drv->bdrv_reopen_prepare) {
4851 * If a driver-specific option is missing, it means that we
4852 * should reset it to its default value.
4853 * But not all options allow that, so we need to check it first.
4855 ret = bdrv_reset_options_allowed(reopen_state->bs,
4856 reopen_state->options, errp);
4861 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4863 if (local_err != NULL) {
4864 error_propagate(errp, local_err);
4866 bdrv_refresh_filename(reopen_state->bs);
4867 error_setg(errp, "failed while preparing to reopen image '%s'",
4868 reopen_state->bs->filename);
4873 /* It is currently mandatory to have a bdrv_reopen_prepare()
4874 * handler for each supported drv. */
4875 error_setg(errp, "Block format '%s' used by node '%s' "
4876 "does not support reopening files", drv->format_name,
4877 bdrv_get_device_or_node_name(reopen_state->bs));
4882 drv_prepared = true;
4885 * We must provide the 'backing' option if the BDS has a backing
4886 * file or if the image file has a backing file name as part of
4887 * its metadata. Otherwise the 'backing' option can be omitted.
4889 if (drv->supports_backing && reopen_state->backing_missing &&
4890 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4891 error_setg(errp, "backing is missing for '%s'",
4892 reopen_state->bs->node_name);
4898 * Allow changing the 'backing' option. The new value can be
4899 * either a reference to an existing node (using its node name)
4900 * or NULL to simply detach the current backing file.
4902 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4903 change_child_tran, errp);
4907 qdict_del(reopen_state->options, "backing");
4909 /* Allow changing the 'file' option. In this case NULL is not allowed */
4910 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4911 change_child_tran, errp);
4915 qdict_del(reopen_state->options, "file");
4917 /* Options that are not handled are only okay if they are unchanged
4918 * compared to the old state. It is expected that some options are only
4919 * used for the initial open, but not reopen (e.g. filename) */
4920 if (qdict_size(reopen_state->options)) {
4921 const QDictEntry *entry = qdict_first(reopen_state->options);
4924 QObject *new = entry->value;
4925 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4927 /* Allow child references (child_name=node_name) as long as they
4928 * point to the current child (i.e. everything stays the same). */
4929 if (qobject_type(new) == QTYPE_QSTRING) {
4931 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4932 if (!strcmp(child->name, entry->key)) {
4938 if (!strcmp(child->bs->node_name,
4939 qstring_get_str(qobject_to(QString, new)))) {
4940 continue; /* Found child with this name, skip option */
4946 * TODO: When using -drive to specify blockdev options, all values
4947 * will be strings; however, when using -blockdev, blockdev-add or
4948 * filenames using the json:{} pseudo-protocol, they will be
4950 * In contrast, reopening options are (currently) always strings
4951 * (because you can only specify them through qemu-io; all other
4952 * callers do not specify any options).
4953 * Therefore, when using anything other than -drive to create a BDS,
4954 * this cannot detect non-string options as unchanged, because
4955 * qobject_is_equal() always returns false for objects of different
4956 * type. In the future, this should be remedied by correctly typing
4957 * all options. For now, this is not too big of an issue because
4958 * the user can simply omit options which cannot be changed anyway,
4959 * so they will stay unchanged.
4961 if (!qobject_is_equal(new, old)) {
4962 error_setg(errp, "Cannot change the option '%s'", entry->key);
4966 } while ((entry = qdict_next(reopen_state->options, entry)));
4971 /* Restore the original reopen_state->options QDict */
4972 qobject_unref(reopen_state->options);
4973 reopen_state->options = qobject_ref(orig_reopen_opts);
4976 if (ret < 0 && drv_prepared) {
4977 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4978 * call drv->bdrv_reopen_abort() before signaling an error
4979 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4980 * when the respective bdrv_reopen_prepare() has failed) */
4981 if (drv->bdrv_reopen_abort) {
4982 drv->bdrv_reopen_abort(reopen_state);
4985 qemu_opts_del(opts);
4986 qobject_unref(orig_reopen_opts);
4992 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4993 * makes them final by swapping the staging BlockDriverState contents into
4994 * the active BlockDriverState contents.
4996 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4999 BlockDriverState *bs;
5002 assert(reopen_state != NULL);
5003 bs = reopen_state->bs;
5005 assert(drv != NULL);
5006 GLOBAL_STATE_CODE();
5008 /* If there are any driver level actions to take */
5009 if (drv->bdrv_reopen_commit) {
5010 drv->bdrv_reopen_commit(reopen_state);
5013 /* set BDS specific flags now */
5014 qobject_unref(bs->explicit_options);
5015 qobject_unref(bs->options);
5016 qobject_ref(reopen_state->explicit_options);
5017 qobject_ref(reopen_state->options);
5019 bs->explicit_options = reopen_state->explicit_options;
5020 bs->options = reopen_state->options;
5021 bs->open_flags = reopen_state->flags;
5022 bs->detect_zeroes = reopen_state->detect_zeroes;
5024 /* Remove child references from bs->options and bs->explicit_options.
5025 * Child options were already removed in bdrv_reopen_queue_child() */
5026 QLIST_FOREACH(child, &bs->children, next) {
5027 qdict_del(bs->explicit_options, child->name);
5028 qdict_del(bs->options, child->name);
5030 /* backing is probably removed, so it's not handled by previous loop */
5031 qdict_del(bs->explicit_options, "backing");
5032 qdict_del(bs->options, "backing");
5034 bdrv_graph_rdlock_main_loop();
5035 bdrv_refresh_limits(bs, NULL, NULL);
5036 bdrv_graph_rdunlock_main_loop();
5037 bdrv_refresh_total_sectors(bs, bs->total_sectors);
5041 * Abort the reopen, and delete and free the staged changes in
5044 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
5048 assert(reopen_state != NULL);
5049 drv = reopen_state->bs->drv;
5050 assert(drv != NULL);
5051 GLOBAL_STATE_CODE();
5053 if (drv->bdrv_reopen_abort) {
5054 drv->bdrv_reopen_abort(reopen_state);
5059 static void bdrv_close(BlockDriverState *bs)
5061 BdrvAioNotifier *ban, *ban_next;
5062 BdrvChild *child, *next;
5064 GLOBAL_STATE_CODE();
5065 assert(!bs->refcnt);
5067 bdrv_drained_begin(bs); /* complete I/O */
5069 bdrv_drain(bs); /* in case flush left pending I/O */
5072 if (bs->drv->bdrv_close) {
5073 /* Must unfreeze all children, so bdrv_unref_child() works */
5074 bs->drv->bdrv_close(bs);
5079 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
5080 bdrv_unref_child(bs, child);
5083 assert(!bs->backing);
5087 qatomic_set(&bs->copy_on_read, 0);
5088 bs->backing_file[0] = '\0';
5089 bs->backing_format[0] = '\0';
5090 bs->total_sectors = 0;
5091 bs->encrypted = false;
5093 qobject_unref(bs->options);
5094 qobject_unref(bs->explicit_options);
5096 bs->explicit_options = NULL;
5097 qobject_unref(bs->full_open_options);
5098 bs->full_open_options = NULL;
5099 g_free(bs->block_status_cache);
5100 bs->block_status_cache = NULL;
5102 bdrv_release_named_dirty_bitmaps(bs);
5103 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5105 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5108 QLIST_INIT(&bs->aio_notifiers);
5109 bdrv_drained_end(bs);
5112 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5113 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5116 if (bs->quiesce_counter) {
5117 bdrv_drain_all_end_quiesce(bs);
5121 void bdrv_close_all(void)
5123 GLOBAL_STATE_CODE();
5124 assert(job_next(NULL) == NULL);
5126 /* Drop references from requests still in flight, such as canceled block
5127 * jobs whose AIO context has not been polled yet */
5130 blk_remove_all_bs();
5131 blockdev_close_all_bdrv_states();
5133 assert(QTAILQ_EMPTY(&all_bdrv_states));
5136 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
5142 if (c->klass->stay_at_node) {
5146 /* If the child @c belongs to the BDS @to, replacing the current
5147 * c->bs by @to would mean to create a loop.
5149 * Such a case occurs when appending a BDS to a backing chain.
5150 * For instance, imagine the following chain:
5152 * guest device -> node A -> further backing chain...
5154 * Now we create a new BDS B which we want to put on top of this
5155 * chain, so we first attach A as its backing node:
5160 * guest device -> node A -> further backing chain...
5162 * Finally we want to replace A by B. When doing that, we want to
5163 * replace all pointers to A by pointers to B -- except for the
5164 * pointer from B because (1) that would create a loop, and (2)
5165 * that pointer should simply stay intact:
5167 * guest device -> node B
5170 * node A -> further backing chain...
5172 * In general, when replacing a node A (c->bs) by a node B (@to),
5173 * if A is a child of B, that means we cannot replace A by B there
5174 * because that would create a loop. Silently detaching A from B
5175 * is also not really an option. So overall just leaving A in
5176 * place there is the most sensible choice.
5178 * We would also create a loop in any cases where @c is only
5179 * indirectly referenced by @to. Prevent this by returning false
5180 * if @c is found (by breadth-first search) anywhere in the whole
5185 found = g_hash_table_new(NULL, NULL);
5186 g_hash_table_add(found, to);
5187 queue = g_queue_new();
5188 g_queue_push_tail(queue, to);
5190 while (!g_queue_is_empty(queue)) {
5191 BlockDriverState *v = g_queue_pop_head(queue);
5194 QLIST_FOREACH(c2, &v->children, next) {
5200 if (g_hash_table_contains(found, c2->bs)) {
5204 g_queue_push_tail(queue, c2->bs);
5205 g_hash_table_add(found, c2->bs);
5209 g_queue_free(queue);
5210 g_hash_table_destroy(found);
5215 static void bdrv_remove_child_commit(void *opaque)
5217 GLOBAL_STATE_CODE();
5218 bdrv_child_free(opaque);
5221 static TransactionActionDrv bdrv_remove_child_drv = {
5222 .commit = bdrv_remove_child_commit,
5225 /* Function doesn't update permissions, caller is responsible for this. */
5226 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5233 BlockDriverState *bs = child->bs;
5234 bdrv_drained_begin(bs);
5235 bdrv_replace_child_tran(child, NULL, tran);
5236 bdrv_drained_end(bs);
5239 tran_add(tran, &bdrv_remove_child_drv, child);
5242 static void undrain_on_clean_cb(void *opaque)
5244 bdrv_drained_end(opaque);
5247 static TransactionActionDrv undrain_on_clean = {
5248 .clean = undrain_on_clean_cb,
5251 static int bdrv_replace_node_noperm(BlockDriverState *from,
5252 BlockDriverState *to,
5253 bool auto_skip, Transaction *tran,
5256 BdrvChild *c, *next;
5258 GLOBAL_STATE_CODE();
5260 bdrv_drained_begin(from);
5261 bdrv_drained_begin(to);
5262 tran_add(tran, &undrain_on_clean, from);
5263 tran_add(tran, &undrain_on_clean, to);
5265 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5266 assert(c->bs == from);
5267 if (!should_update_child(c, to)) {
5271 error_setg(errp, "Should not change '%s' link to '%s'",
5272 c->name, from->node_name);
5276 error_setg(errp, "Cannot change '%s' link to '%s'",
5277 c->name, from->node_name);
5280 bdrv_replace_child_tran(c, to, tran);
5287 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5288 * if it creates a parent-child relation loop or if parent is block-job.
5290 * With auto_skip=false the error is returned if from has a parent which should
5293 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5294 * case backing link of the cow-parent of @to is removed.
5296 static int bdrv_replace_node_common(BlockDriverState *from,
5297 BlockDriverState *to,
5298 bool auto_skip, bool detach_subchain,
5301 Transaction *tran = tran_new();
5302 g_autoptr(GSList) refresh_list = NULL;
5303 BlockDriverState *to_cow_parent = NULL;
5306 GLOBAL_STATE_CODE();
5308 if (detach_subchain) {
5309 assert(bdrv_chain_contains(from, to));
5311 for (to_cow_parent = from;
5312 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5313 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5319 /* Make sure that @from doesn't go away until we have successfully attached
5320 * all of its parents to @to. */
5323 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5324 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5325 bdrv_drained_begin(from);
5328 * Do the replacement without permission update.
5329 * Replacement may influence the permissions, we should calculate new
5330 * permissions based on new graph. If we fail, we'll roll-back the
5333 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5338 if (detach_subchain) {
5339 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5342 refresh_list = g_slist_prepend(refresh_list, to);
5343 refresh_list = g_slist_prepend(refresh_list, from);
5345 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5353 tran_finalize(tran, ret);
5355 bdrv_drained_end(from);
5361 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5364 GLOBAL_STATE_CODE();
5366 return bdrv_replace_node_common(from, to, true, false, errp);
5369 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5371 GLOBAL_STATE_CODE();
5373 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5378 * Add new bs contents at the top of an image chain while the chain is
5379 * live, while keeping required fields on the top layer.
5381 * This will modify the BlockDriverState fields, and swap contents
5382 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5384 * bs_new must not be attached to a BlockBackend and must not have backing
5387 * This function does not create any image files.
5389 * The caller must hold the AioContext lock for @bs_top.
5391 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5396 Transaction *tran = tran_new();
5397 AioContext *old_context, *new_context = NULL;
5399 GLOBAL_STATE_CODE();
5401 assert(!bs_new->backing);
5403 old_context = bdrv_get_aio_context(bs_top);
5405 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5406 &child_of_bds, bdrv_backing_role(bs_new),
5414 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5415 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5416 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5417 * that assumes the new lock is taken.
5419 new_context = bdrv_get_aio_context(bs_top);
5421 if (old_context != new_context) {
5422 aio_context_release(old_context);
5423 aio_context_acquire(new_context);
5426 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5431 ret = bdrv_refresh_perms(bs_new, tran, errp);
5433 tran_finalize(tran, ret);
5435 bdrv_graph_rdlock_main_loop();
5436 bdrv_refresh_limits(bs_top, NULL, NULL);
5437 bdrv_graph_rdunlock_main_loop();
5439 if (new_context && old_context != new_context) {
5440 aio_context_release(new_context);
5441 aio_context_acquire(old_context);
5447 /* Not for empty child */
5448 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5452 Transaction *tran = tran_new();
5453 g_autoptr(GSList) refresh_list = NULL;
5454 BlockDriverState *old_bs = child->bs;
5456 GLOBAL_STATE_CODE();
5459 bdrv_drained_begin(old_bs);
5460 bdrv_drained_begin(new_bs);
5462 bdrv_replace_child_tran(child, new_bs, tran);
5464 refresh_list = g_slist_prepend(refresh_list, old_bs);
5465 refresh_list = g_slist_prepend(refresh_list, new_bs);
5467 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5469 tran_finalize(tran, ret);
5471 bdrv_drained_end(old_bs);
5472 bdrv_drained_end(new_bs);
5478 static void bdrv_delete(BlockDriverState *bs)
5480 assert(bdrv_op_blocker_is_empty(bs));
5481 assert(!bs->refcnt);
5482 GLOBAL_STATE_CODE();
5484 /* remove from list, if necessary */
5485 if (bs->node_name[0] != '\0') {
5486 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5488 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5492 qemu_mutex_destroy(&bs->reqs_lock);
5499 * Replace @bs by newly created block node.
5501 * @options is a QDict of options to pass to the block drivers, or NULL for an
5502 * empty set of options. The reference to the QDict belongs to the block layer
5503 * after the call (even on failure), so if the caller intends to reuse the
5504 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5506 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5507 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5508 * different AioContext.
5510 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5511 int flags, Error **errp)
5515 AioContext *ctx = bdrv_get_aio_context(bs);
5516 BlockDriverState *new_node_bs = NULL;
5517 const char *drvname, *node_name;
5520 drvname = qdict_get_try_str(options, "driver");
5522 error_setg(errp, "driver is not specified");
5526 drv = bdrv_find_format(drvname);
5528 error_setg(errp, "Unknown driver: '%s'", drvname);
5532 node_name = qdict_get_try_str(options, "node-name");
5534 GLOBAL_STATE_CODE();
5536 aio_context_release(ctx);
5537 aio_context_acquire(qemu_get_aio_context());
5538 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5540 aio_context_release(qemu_get_aio_context());
5541 aio_context_acquire(ctx);
5542 assert(bdrv_get_aio_context(bs) == ctx);
5544 options = NULL; /* bdrv_new_open_driver() eats options */
5546 error_prepend(errp, "Could not create node: ");
5550 bdrv_drained_begin(bs);
5551 ret = bdrv_replace_node(bs, new_node_bs, errp);
5552 bdrv_drained_end(bs);
5555 error_prepend(errp, "Could not replace node: ");
5562 qobject_unref(options);
5563 bdrv_unref(new_node_bs);
5568 * Run consistency checks on an image
5570 * Returns 0 if the check could be completed (it doesn't mean that the image is
5571 * free of errors) or -errno when an internal error occurred. The results of the
5572 * check are stored in res.
5574 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5575 BdrvCheckResult *res, BdrvCheckMode fix)
5578 assert_bdrv_graph_readable();
5579 if (bs->drv == NULL) {
5582 if (bs->drv->bdrv_co_check == NULL) {
5586 memset(res, 0, sizeof(*res));
5587 return bs->drv->bdrv_co_check(bs, res, fix);
5593 * -EINVAL - backing format specified, but no file
5594 * -ENOSPC - can't update the backing file because no space is left in the
5596 * -ENOTSUP - format driver doesn't support changing the backing file
5598 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5599 const char *backing_fmt, bool require)
5601 BlockDriver *drv = bs->drv;
5604 GLOBAL_STATE_CODE();
5610 /* Backing file format doesn't make sense without a backing file */
5611 if (backing_fmt && !backing_file) {
5615 if (require && backing_file && !backing_fmt) {
5619 if (drv->bdrv_change_backing_file != NULL) {
5620 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5626 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5627 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5628 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5629 backing_file ?: "");
5635 * Finds the first non-filter node above bs in the chain between
5636 * active and bs. The returned node is either an immediate parent of
5637 * bs, or there are only filter nodes between the two.
5639 * Returns NULL if bs is not found in active's image chain,
5640 * or if active == bs.
5642 * Returns the bottommost base image if bs == NULL.
5644 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5645 BlockDriverState *bs)
5648 GLOBAL_STATE_CODE();
5650 bs = bdrv_skip_filters(bs);
5651 active = bdrv_skip_filters(active);
5654 BlockDriverState *next = bdrv_backing_chain_next(active);
5664 /* Given a BDS, searches for the base layer. */
5665 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5667 GLOBAL_STATE_CODE();
5669 return bdrv_find_overlay(bs, NULL);
5673 * Return true if at least one of the COW (backing) and filter links
5674 * between @bs and @base is frozen. @errp is set if that's the case.
5675 * @base must be reachable from @bs, or NULL.
5677 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5680 BlockDriverState *i;
5683 GLOBAL_STATE_CODE();
5685 for (i = bs; i != base; i = child_bs(child)) {
5686 child = bdrv_filter_or_cow_child(i);
5688 if (child && child->frozen) {
5689 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5690 child->name, i->node_name, child->bs->node_name);
5699 * Freeze all COW (backing) and filter links between @bs and @base.
5700 * If any of the links is already frozen the operation is aborted and
5701 * none of the links are modified.
5702 * @base must be reachable from @bs, or NULL.
5703 * Returns 0 on success. On failure returns < 0 and sets @errp.
5705 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5708 BlockDriverState *i;
5711 GLOBAL_STATE_CODE();
5713 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5717 for (i = bs; i != base; i = child_bs(child)) {
5718 child = bdrv_filter_or_cow_child(i);
5719 if (child && child->bs->never_freeze) {
5720 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5721 child->name, child->bs->node_name);
5726 for (i = bs; i != base; i = child_bs(child)) {
5727 child = bdrv_filter_or_cow_child(i);
5729 child->frozen = true;
5737 * Unfreeze all COW (backing) and filter links between @bs and @base.
5738 * The caller must ensure that all links are frozen before using this
5740 * @base must be reachable from @bs, or NULL.
5742 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5744 BlockDriverState *i;
5747 GLOBAL_STATE_CODE();
5749 for (i = bs; i != base; i = child_bs(child)) {
5750 child = bdrv_filter_or_cow_child(i);
5752 assert(child->frozen);
5753 child->frozen = false;
5759 * Drops images above 'base' up to and including 'top', and sets the image
5760 * above 'top' to have base as its backing file.
5762 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5763 * information in 'bs' can be properly updated.
5765 * E.g., this will convert the following chain:
5766 * bottom <- base <- intermediate <- top <- active
5770 * bottom <- base <- active
5772 * It is allowed for bottom==base, in which case it converts:
5774 * base <- intermediate <- top <- active
5780 * If backing_file_str is non-NULL, it will be used when modifying top's
5781 * overlay image metadata.
5784 * if active == top, that is considered an error
5787 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5788 const char *backing_file_str)
5790 BlockDriverState *explicit_top = top;
5791 bool update_inherits_from;
5793 Error *local_err = NULL;
5795 g_autoptr(GSList) updated_children = NULL;
5798 GLOBAL_STATE_CODE();
5801 bdrv_drained_begin(base);
5803 if (!top->drv || !base->drv) {
5807 /* Make sure that base is in the backing chain of top */
5808 if (!bdrv_chain_contains(top, base)) {
5812 /* If 'base' recursively inherits from 'top' then we should set
5813 * base->inherits_from to top->inherits_from after 'top' and all
5814 * other intermediate nodes have been dropped.
5815 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5816 * it because no one inherits from it. We use explicit_top for that. */
5817 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5818 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5820 /* success - we can delete the intermediate states, and link top->base */
5821 if (!backing_file_str) {
5822 bdrv_refresh_filename(base);
5823 backing_file_str = base->filename;
5826 QLIST_FOREACH(c, &top->parents, next_parent) {
5827 updated_children = g_slist_prepend(updated_children, c);
5831 * It seems correct to pass detach_subchain=true here, but it triggers
5832 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5833 * another drained section, which modify the graph (for example, removing
5834 * the child, which we keep in updated_children list). So, it's a TODO.
5836 * Note, bug triggered if pass detach_subchain=true here and run
5837 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5840 bdrv_replace_node_common(top, base, false, false, &local_err);
5842 error_report_err(local_err);
5846 for (p = updated_children; p; p = p->next) {
5849 if (c->klass->update_filename) {
5850 ret = c->klass->update_filename(c, base, backing_file_str,
5854 * TODO: Actually, we want to rollback all previous iterations
5855 * of this loop, and (which is almost impossible) previous
5856 * bdrv_replace_node()...
5858 * Note, that c->klass->update_filename may lead to permission
5859 * update, so it's a bad idea to call it inside permission
5860 * update transaction of bdrv_replace_node.
5862 error_report_err(local_err);
5868 if (update_inherits_from) {
5869 base->inherits_from = explicit_top->inherits_from;
5874 bdrv_drained_end(base);
5880 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5881 * sums the size of all data-bearing children. (This excludes backing
5884 static int64_t coroutine_fn GRAPH_RDLOCK
5885 bdrv_sum_allocated_file_size(BlockDriverState *bs)
5888 int64_t child_size, sum = 0;
5890 QLIST_FOREACH(child, &bs->children, next) {
5891 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5892 BDRV_CHILD_FILTERED))
5894 child_size = bdrv_co_get_allocated_file_size(child->bs);
5895 if (child_size < 0) {
5906 * Length of a allocated file in bytes. Sparse files are counted by actual
5907 * allocated space. Return < 0 if error or unknown.
5909 int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
5911 BlockDriver *drv = bs->drv;
5913 assert_bdrv_graph_readable();
5918 if (drv->bdrv_co_get_allocated_file_size) {
5919 return drv->bdrv_co_get_allocated_file_size(bs);
5922 if (drv->bdrv_file_open) {
5924 * Protocol drivers default to -ENOTSUP (most of their data is
5925 * not stored in any of their children (if they even have any),
5926 * so there is no generic way to figure it out).
5929 } else if (drv->is_filter) {
5930 /* Filter drivers default to the size of their filtered child */
5931 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
5933 /* Other drivers default to summing their children's sizes */
5934 return bdrv_sum_allocated_file_size(bs);
5940 * @drv: Format driver
5941 * @opts: Creation options for new image
5942 * @in_bs: Existing image containing data for new image (may be NULL)
5943 * @errp: Error object
5944 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5947 * Calculate file size required to create a new image.
5949 * If @in_bs is given then space for allocated clusters and zero clusters
5950 * from that image are included in the calculation. If @opts contains a
5951 * backing file that is shared by @in_bs then backing clusters may be omitted
5952 * from the calculation.
5954 * If @in_bs is NULL then the calculation includes no allocated clusters
5955 * unless a preallocation option is given in @opts.
5957 * Note that @in_bs may use a different BlockDriver from @drv.
5959 * If an error occurs the @errp pointer is set.
5961 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5962 BlockDriverState *in_bs, Error **errp)
5965 if (!drv->bdrv_measure) {
5966 error_setg(errp, "Block driver '%s' does not support size measurement",
5971 return drv->bdrv_measure(opts, in_bs, errp);
5975 * Return number of sectors on success, -errno on error.
5977 int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
5979 BlockDriver *drv = bs->drv;
5981 assert_bdrv_graph_readable();
5986 if (bs->bl.has_variable_length) {
5987 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
5992 return bs->total_sectors;
5996 * This wrapper is written by hand because this function is in the hot I/O path,
5997 * via blk_get_geometry.
5999 int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
6001 BlockDriver *drv = bs->drv;
6007 if (bs->bl.has_variable_length) {
6008 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6014 return bs->total_sectors;
6018 * Return length in bytes on success, -errno on error.
6019 * The length is always a multiple of BDRV_SECTOR_SIZE.
6021 int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
6025 assert_bdrv_graph_readable();
6027 ret = bdrv_co_nb_sectors(bs);
6031 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6034 return ret * BDRV_SECTOR_SIZE;
6037 bool bdrv_is_sg(BlockDriverState *bs)
6044 * Return whether the given node supports compressed writes.
6046 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6048 BlockDriverState *filtered;
6051 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6055 filtered = bdrv_filter_bs(bs);
6058 * Filters can only forward compressed writes, so we have to
6061 return bdrv_supports_compressed_writes(filtered);
6067 const char *bdrv_get_format_name(BlockDriverState *bs)
6070 return bs->drv ? bs->drv->format_name : NULL;
6073 static int qsort_strcmp(const void *a, const void *b)
6075 return strcmp(*(char *const *)a, *(char *const *)b);
6078 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
6079 void *opaque, bool read_only)
6084 const char **formats = NULL;
6086 GLOBAL_STATE_CODE();
6088 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6089 if (drv->format_name) {
6093 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6097 while (formats && i && !found) {
6098 found = !strcmp(formats[--i], drv->format_name);
6102 formats = g_renew(const char *, formats, count + 1);
6103 formats[count++] = drv->format_name;
6108 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6109 const char *format_name = block_driver_modules[i].format_name;
6115 if (use_bdrv_whitelist &&
6116 !bdrv_format_is_whitelisted(format_name, read_only)) {
6120 while (formats && j && !found) {
6121 found = !strcmp(formats[--j], format_name);
6125 formats = g_renew(const char *, formats, count + 1);
6126 formats[count++] = format_name;
6131 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6133 for (i = 0; i < count; i++) {
6134 it(opaque, formats[i]);
6140 /* This function is to find a node in the bs graph */
6141 BlockDriverState *bdrv_find_node(const char *node_name)
6143 BlockDriverState *bs;
6146 GLOBAL_STATE_CODE();
6148 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6149 if (!strcmp(node_name, bs->node_name)) {
6156 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6157 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6160 BlockDeviceInfoList *list;
6161 BlockDriverState *bs;
6163 GLOBAL_STATE_CODE();
6166 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6167 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6169 qapi_free_BlockDeviceInfoList(list);
6172 QAPI_LIST_PREPEND(list, info);
6178 typedef struct XDbgBlockGraphConstructor {
6179 XDbgBlockGraph *graph;
6180 GHashTable *graph_nodes;
6181 } XDbgBlockGraphConstructor;
6183 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6185 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6187 gr->graph = g_new0(XDbgBlockGraph, 1);
6188 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6193 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6195 XDbgBlockGraph *graph = gr->graph;
6197 g_hash_table_destroy(gr->graph_nodes);
6203 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6205 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6212 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6213 * answer of g_hash_table_lookup.
6215 ret = g_hash_table_size(gr->graph_nodes) + 1;
6216 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6221 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6222 XDbgBlockGraphNodeType type, const char *name)
6224 XDbgBlockGraphNode *n;
6226 n = g_new0(XDbgBlockGraphNode, 1);
6228 n->id = xdbg_graph_node_num(gr, node);
6230 n->name = g_strdup(name);
6232 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6235 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6236 const BdrvChild *child)
6238 BlockPermission qapi_perm;
6239 XDbgBlockGraphEdge *edge;
6240 GLOBAL_STATE_CODE();
6242 edge = g_new0(XDbgBlockGraphEdge, 1);
6244 edge->parent = xdbg_graph_node_num(gr, parent);
6245 edge->child = xdbg_graph_node_num(gr, child->bs);
6246 edge->name = g_strdup(child->name);
6248 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6249 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6251 if (flag & child->perm) {
6252 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6254 if (flag & child->shared_perm) {
6255 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6259 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6263 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6267 BlockDriverState *bs;
6269 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6271 GLOBAL_STATE_CODE();
6273 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6274 char *allocated_name = NULL;
6275 const char *name = blk_name(blk);
6278 name = allocated_name = blk_get_attached_dev_id(blk);
6280 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6282 g_free(allocated_name);
6283 if (blk_root(blk)) {
6284 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6288 WITH_JOB_LOCK_GUARD() {
6289 for (job = block_job_next_locked(NULL); job;
6290 job = block_job_next_locked(job)) {
6293 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6295 for (el = job->nodes; el; el = el->next) {
6296 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6301 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6302 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6304 QLIST_FOREACH(child, &bs->children, next) {
6305 xdbg_graph_add_edge(gr, bs, child);
6309 return xdbg_graph_finalize(gr);
6312 BlockDriverState *bdrv_lookup_bs(const char *device,
6313 const char *node_name,
6317 BlockDriverState *bs;
6319 GLOBAL_STATE_CODE();
6322 blk = blk_by_name(device);
6327 error_setg(errp, "Device '%s' has no medium", device);
6335 bs = bdrv_find_node(node_name);
6342 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6343 device ? device : "",
6344 node_name ? node_name : "");
6348 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6349 * return false. If either argument is NULL, return false. */
6350 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6353 GLOBAL_STATE_CODE();
6355 while (top && top != base) {
6356 top = bdrv_filter_or_cow_bs(top);
6362 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6364 GLOBAL_STATE_CODE();
6366 return QTAILQ_FIRST(&graph_bdrv_states);
6368 return QTAILQ_NEXT(bs, node_list);
6371 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6373 GLOBAL_STATE_CODE();
6375 return QTAILQ_FIRST(&all_bdrv_states);
6377 return QTAILQ_NEXT(bs, bs_list);
6380 const char *bdrv_get_node_name(const BlockDriverState *bs)
6383 return bs->node_name;
6386 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6392 /* If multiple parents have a name, just pick the first one. */
6393 QLIST_FOREACH(c, &bs->parents, next_parent) {
6394 if (c->klass->get_name) {
6395 name = c->klass->get_name(c);
6396 if (name && *name) {
6405 /* TODO check what callers really want: bs->node_name or blk_name() */
6406 const char *bdrv_get_device_name(const BlockDriverState *bs)
6409 return bdrv_get_parent_name(bs) ?: "";
6412 /* This can be used to identify nodes that might not have a device
6413 * name associated. Since node and device names live in the same
6414 * namespace, the result is unambiguous. The exception is if both are
6415 * absent, then this returns an empty (non-null) string. */
6416 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6419 return bdrv_get_parent_name(bs) ?: bs->node_name;
6422 int bdrv_get_flags(BlockDriverState *bs)
6425 return bs->open_flags;
6428 int bdrv_has_zero_init_1(BlockDriverState *bs)
6430 GLOBAL_STATE_CODE();
6434 int bdrv_has_zero_init(BlockDriverState *bs)
6436 BlockDriverState *filtered;
6437 GLOBAL_STATE_CODE();
6443 /* If BS is a copy on write image, it is initialized to
6444 the contents of the base image, which may not be zeroes. */
6445 if (bdrv_cow_child(bs)) {
6448 if (bs->drv->bdrv_has_zero_init) {
6449 return bs->drv->bdrv_has_zero_init(bs);
6452 filtered = bdrv_filter_bs(bs);
6454 return bdrv_has_zero_init(filtered);
6461 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6464 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6468 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6471 void bdrv_get_backing_filename(BlockDriverState *bs,
6472 char *filename, int filename_size)
6475 pstrcpy(filename, filename_size, bs->backing_file);
6478 int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6481 BlockDriver *drv = bs->drv;
6483 assert_bdrv_graph_readable();
6485 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6489 if (!drv->bdrv_co_get_info) {
6490 BlockDriverState *filtered = bdrv_filter_bs(bs);
6492 return bdrv_co_get_info(filtered, bdi);
6496 memset(bdi, 0, sizeof(*bdi));
6497 ret = drv->bdrv_co_get_info(bs, bdi);
6498 if (bdi->subcluster_size == 0) {
6500 * If the driver left this unset, subclusters are not supported.
6501 * Then it is safe to treat each cluster as having only one subcluster.
6503 bdi->subcluster_size = bdi->cluster_size;
6509 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6516 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6519 BlockDriver *drv = bs->drv;
6521 if (drv && drv->bdrv_get_specific_info) {
6522 return drv->bdrv_get_specific_info(bs, errp);
6527 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6529 BlockDriver *drv = bs->drv;
6531 if (!drv || !drv->bdrv_get_specific_stats) {
6534 return drv->bdrv_get_specific_stats(bs);
6537 void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6540 assert_bdrv_graph_readable();
6542 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
6546 bs->drv->bdrv_co_debug_event(bs, event);
6549 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6551 GLOBAL_STATE_CODE();
6552 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6553 bs = bdrv_primary_bs(bs);
6556 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6557 assert(bs->drv->bdrv_debug_remove_breakpoint);
6564 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6567 GLOBAL_STATE_CODE();
6568 bs = bdrv_find_debug_node(bs);
6570 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6576 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6578 GLOBAL_STATE_CODE();
6579 bs = bdrv_find_debug_node(bs);
6581 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6587 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6589 GLOBAL_STATE_CODE();
6590 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6591 bs = bdrv_primary_bs(bs);
6594 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6595 return bs->drv->bdrv_debug_resume(bs, tag);
6601 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6603 GLOBAL_STATE_CODE();
6604 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6605 bs = bdrv_primary_bs(bs);
6608 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6609 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6615 /* backing_file can either be relative, or absolute, or a protocol. If it is
6616 * relative, it must be relative to the chain. So, passing in bs->filename
6617 * from a BDS as backing_file should not be done, as that may be relative to
6618 * the CWD rather than the chain. */
6619 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6620 const char *backing_file)
6622 char *filename_full = NULL;
6623 char *backing_file_full = NULL;
6624 char *filename_tmp = NULL;
6625 int is_protocol = 0;
6626 bool filenames_refreshed = false;
6627 BlockDriverState *curr_bs = NULL;
6628 BlockDriverState *retval = NULL;
6629 BlockDriverState *bs_below;
6631 GLOBAL_STATE_CODE();
6633 if (!bs || !bs->drv || !backing_file) {
6637 filename_full = g_malloc(PATH_MAX);
6638 backing_file_full = g_malloc(PATH_MAX);
6640 is_protocol = path_has_protocol(backing_file);
6643 * Being largely a legacy function, skip any filters here
6644 * (because filters do not have normal filenames, so they cannot
6645 * match anyway; and allowing json:{} filenames is a bit out of
6648 for (curr_bs = bdrv_skip_filters(bs);
6649 bdrv_cow_child(curr_bs) != NULL;
6652 bs_below = bdrv_backing_chain_next(curr_bs);
6654 if (bdrv_backing_overridden(curr_bs)) {
6656 * If the backing file was overridden, we can only compare
6657 * directly against the backing node's filename.
6660 if (!filenames_refreshed) {
6662 * This will automatically refresh all of the
6663 * filenames in the rest of the backing chain, so we
6664 * only need to do this once.
6666 bdrv_refresh_filename(bs_below);
6667 filenames_refreshed = true;
6670 if (strcmp(backing_file, bs_below->filename) == 0) {
6674 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6676 * If either of the filename paths is actually a protocol, then
6677 * compare unmodified paths; otherwise make paths relative.
6679 char *backing_file_full_ret;
6681 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6685 /* Also check against the full backing filename for the image */
6686 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6688 if (backing_file_full_ret) {
6689 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6690 g_free(backing_file_full_ret);
6697 /* If not an absolute filename path, make it relative to the current
6698 * image's filename path */
6699 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6701 /* We are going to compare canonicalized absolute pathnames */
6702 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6703 g_free(filename_tmp);
6706 g_free(filename_tmp);
6708 /* We need to make sure the backing filename we are comparing against
6709 * is relative to the current image filename (or absolute) */
6710 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6711 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6712 g_free(filename_tmp);
6715 g_free(filename_tmp);
6717 if (strcmp(backing_file_full, filename_full) == 0) {
6724 g_free(filename_full);
6725 g_free(backing_file_full);
6729 void bdrv_init(void)
6731 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6732 use_bdrv_whitelist = 1;
6734 module_call_init(MODULE_INIT_BLOCK);
6737 void bdrv_init_with_whitelist(void)
6739 use_bdrv_whitelist = 1;
6743 int bdrv_activate(BlockDriverState *bs, Error **errp)
6745 BdrvChild *child, *parent;
6746 Error *local_err = NULL;
6748 BdrvDirtyBitmap *bm;
6750 GLOBAL_STATE_CODE();
6756 QLIST_FOREACH(child, &bs->children, next) {
6757 bdrv_activate(child->bs, &local_err);
6759 error_propagate(errp, local_err);
6765 * Update permissions, they may differ for inactive nodes.
6767 * Note that the required permissions of inactive images are always a
6768 * subset of the permissions required after activating the image. This
6769 * allows us to just get the permissions upfront without restricting
6770 * bdrv_co_invalidate_cache().
6772 * It also means that in error cases, we don't have to try and revert to
6773 * the old permissions (which is an operation that could fail, too). We can
6774 * just keep the extended permissions for the next time that an activation
6775 * of the image is tried.
6777 if (bs->open_flags & BDRV_O_INACTIVE) {
6778 bs->open_flags &= ~BDRV_O_INACTIVE;
6779 ret = bdrv_refresh_perms(bs, NULL, errp);
6781 bs->open_flags |= BDRV_O_INACTIVE;
6785 ret = bdrv_invalidate_cache(bs, errp);
6787 bs->open_flags |= BDRV_O_INACTIVE;
6791 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6792 bdrv_dirty_bitmap_skip_store(bm, false);
6795 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6797 bs->open_flags |= BDRV_O_INACTIVE;
6798 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6803 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6804 if (parent->klass->activate) {
6805 parent->klass->activate(parent, &local_err);
6807 bs->open_flags |= BDRV_O_INACTIVE;
6808 error_propagate(errp, local_err);
6817 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6819 Error *local_err = NULL;
6822 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6823 assert_bdrv_graph_readable();
6825 if (bs->drv->bdrv_co_invalidate_cache) {
6826 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6828 error_propagate(errp, local_err);
6836 void bdrv_activate_all(Error **errp)
6838 BlockDriverState *bs;
6839 BdrvNextIterator it;
6841 GLOBAL_STATE_CODE();
6843 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6844 AioContext *aio_context = bdrv_get_aio_context(bs);
6847 aio_context_acquire(aio_context);
6848 ret = bdrv_activate(bs, errp);
6849 aio_context_release(aio_context);
6851 bdrv_next_cleanup(&it);
6857 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6860 GLOBAL_STATE_CODE();
6862 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6863 if (parent->klass->parent_is_bds) {
6864 BlockDriverState *parent_bs = parent->opaque;
6865 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6874 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6876 BdrvChild *child, *parent;
6878 uint64_t cumulative_perms, cumulative_shared_perms;
6880 GLOBAL_STATE_CODE();
6886 /* Make sure that we don't inactivate a child before its parent.
6887 * It will be covered by recursion from the yet active parent. */
6888 if (bdrv_has_bds_parent(bs, true)) {
6892 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6894 /* Inactivate this node */
6895 if (bs->drv->bdrv_inactivate) {
6896 ret = bs->drv->bdrv_inactivate(bs);
6902 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6903 if (parent->klass->inactivate) {
6904 ret = parent->klass->inactivate(parent);
6911 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6912 &cumulative_shared_perms);
6913 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6914 /* Our inactive parents still need write access. Inactivation failed. */
6918 bs->open_flags |= BDRV_O_INACTIVE;
6921 * Update permissions, they may differ for inactive nodes.
6922 * We only tried to loosen restrictions, so errors are not fatal, ignore
6925 bdrv_refresh_perms(bs, NULL, NULL);
6927 /* Recursively inactivate children */
6928 QLIST_FOREACH(child, &bs->children, next) {
6929 ret = bdrv_inactivate_recurse(child->bs);
6938 int bdrv_inactivate_all(void)
6940 BlockDriverState *bs = NULL;
6941 BdrvNextIterator it;
6943 GSList *aio_ctxs = NULL, *ctx;
6945 GLOBAL_STATE_CODE();
6947 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6948 AioContext *aio_context = bdrv_get_aio_context(bs);
6950 if (!g_slist_find(aio_ctxs, aio_context)) {
6951 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6952 aio_context_acquire(aio_context);
6956 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6957 /* Nodes with BDS parents are covered by recursion from the last
6958 * parent that gets inactivated. Don't inactivate them a second
6959 * time if that has already happened. */
6960 if (bdrv_has_bds_parent(bs, false)) {
6963 ret = bdrv_inactivate_recurse(bs);
6965 bdrv_next_cleanup(&it);
6971 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6972 AioContext *aio_context = ctx->data;
6973 aio_context_release(aio_context);
6975 g_slist_free(aio_ctxs);
6980 /**************************************************************/
6981 /* removable device support */
6984 * Return TRUE if the media is present
6986 bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
6988 BlockDriver *drv = bs->drv;
6991 assert_bdrv_graph_readable();
6996 if (drv->bdrv_co_is_inserted) {
6997 return drv->bdrv_co_is_inserted(bs);
6999 QLIST_FOREACH(child, &bs->children, next) {
7000 if (!bdrv_co_is_inserted(child->bs)) {
7008 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7010 void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
7012 BlockDriver *drv = bs->drv;
7014 assert_bdrv_graph_readable();
7016 if (drv && drv->bdrv_co_eject) {
7017 drv->bdrv_co_eject(bs, eject_flag);
7022 * Lock or unlock the media (if it is locked, the user won't be able
7023 * to eject it manually).
7025 void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
7027 BlockDriver *drv = bs->drv;
7029 assert_bdrv_graph_readable();
7030 trace_bdrv_lock_medium(bs, locked);
7032 if (drv && drv->bdrv_co_lock_medium) {
7033 drv->bdrv_co_lock_medium(bs, locked);
7037 /* Get a reference to bs */
7038 void bdrv_ref(BlockDriverState *bs)
7040 GLOBAL_STATE_CODE();
7044 /* Release a previously grabbed reference to bs.
7045 * If after releasing, reference count is zero, the BlockDriverState is
7047 void bdrv_unref(BlockDriverState *bs)
7049 GLOBAL_STATE_CODE();
7053 assert(bs->refcnt > 0);
7054 if (--bs->refcnt == 0) {
7060 * Release a BlockDriverState reference while holding the graph write lock.
7062 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7063 * because bdrv_close() both involves polling and taking the graph lock
7064 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7065 * possibly closing @bs until the graph lock is released.
7067 void bdrv_schedule_unref(BlockDriverState *bs)
7072 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7073 (QEMUBHFunc *) bdrv_unref, bs);
7076 struct BdrvOpBlocker {
7078 QLIST_ENTRY(BdrvOpBlocker) list;
7081 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7083 BdrvOpBlocker *blocker;
7084 GLOBAL_STATE_CODE();
7085 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7086 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7087 blocker = QLIST_FIRST(&bs->op_blockers[op]);
7088 error_propagate_prepend(errp, error_copy(blocker->reason),
7089 "Node '%s' is busy: ",
7090 bdrv_get_device_or_node_name(bs));
7096 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7098 BdrvOpBlocker *blocker;
7099 GLOBAL_STATE_CODE();
7100 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7102 blocker = g_new0(BdrvOpBlocker, 1);
7103 blocker->reason = reason;
7104 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7107 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7109 BdrvOpBlocker *blocker, *next;
7110 GLOBAL_STATE_CODE();
7111 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7112 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7113 if (blocker->reason == reason) {
7114 QLIST_REMOVE(blocker, list);
7120 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7123 GLOBAL_STATE_CODE();
7124 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7125 bdrv_op_block(bs, i, reason);
7129 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7132 GLOBAL_STATE_CODE();
7133 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7134 bdrv_op_unblock(bs, i, reason);
7138 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7141 GLOBAL_STATE_CODE();
7142 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7143 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7151 * Must not be called while holding the lock of an AioContext other than the
7154 void bdrv_img_create(const char *filename, const char *fmt,
7155 const char *base_filename, const char *base_fmt,
7156 char *options, uint64_t img_size, int flags, bool quiet,
7159 QemuOptsList *create_opts = NULL;
7160 QemuOpts *opts = NULL;
7161 const char *backing_fmt, *backing_file;
7163 BlockDriver *drv, *proto_drv;
7164 Error *local_err = NULL;
7167 GLOBAL_STATE_CODE();
7169 /* Find driver and parse its options */
7170 drv = bdrv_find_format(fmt);
7172 error_setg(errp, "Unknown file format '%s'", fmt);
7176 proto_drv = bdrv_find_protocol(filename, true, errp);
7181 if (!drv->create_opts) {
7182 error_setg(errp, "Format driver '%s' does not support image creation",
7187 if (!proto_drv->create_opts) {
7188 error_setg(errp, "Protocol driver '%s' does not support image creation",
7189 proto_drv->format_name);
7193 aio_context_acquire(qemu_get_aio_context());
7195 /* Create parameter list */
7196 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7197 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7199 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7201 /* Parse -o options */
7203 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7208 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7209 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7210 } else if (img_size != UINT64_C(-1)) {
7211 error_setg(errp, "The image size must be specified only once");
7215 if (base_filename) {
7216 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7218 error_setg(errp, "Backing file not supported for file format '%s'",
7225 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7226 error_setg(errp, "Backing file format not supported for file "
7227 "format '%s'", fmt);
7232 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7234 if (!strcmp(filename, backing_file)) {
7235 error_setg(errp, "Error: Trying to create an image with the "
7236 "same filename as the backing file");
7239 if (backing_file[0] == '\0') {
7240 error_setg(errp, "Expected backing file name, got empty string");
7245 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7247 /* The size for the image must always be specified, unless we have a backing
7248 * file and we have not been forbidden from opening it. */
7249 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7250 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7251 BlockDriverState *bs;
7254 QDict *backing_options = NULL;
7257 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7262 assert(full_backing);
7265 * No need to do I/O here, which allows us to open encrypted
7266 * backing images without needing the secret
7269 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7270 back_flags |= BDRV_O_NO_IO;
7272 backing_options = qdict_new();
7274 qdict_put_str(backing_options, "driver", backing_fmt);
7276 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7278 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7280 g_free(full_backing);
7282 error_append_hint(&local_err, "Could not open backing image.\n");
7286 error_setg(&local_err,
7287 "Backing file specified without backing format");
7288 error_append_hint(&local_err, "Detected format of %s.\n",
7289 bs->drv->format_name);
7293 /* Opened BS, have no size */
7294 size = bdrv_getlength(bs);
7296 error_setg_errno(errp, -size, "Could not get size of '%s'",
7301 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7305 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7306 } else if (backing_file && !backing_fmt) {
7307 error_setg(&local_err,
7308 "Backing file specified without backing format");
7313 error_setg(errp, "Image creation needs a size parameter");
7318 printf("Formatting '%s', fmt=%s ", filename, fmt);
7319 qemu_opts_print(opts, " ");
7324 ret = bdrv_create(drv, filename, opts, &local_err);
7326 if (ret == -EFBIG) {
7327 /* This is generally a better message than whatever the driver would
7328 * deliver (especially because of the cluster_size_hint), since that
7329 * is most probably not much different from "image too large". */
7330 const char *cluster_size_hint = "";
7331 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7332 cluster_size_hint = " (try using a larger cluster size)";
7334 error_setg(errp, "The image size is too large for file format '%s'"
7335 "%s", fmt, cluster_size_hint);
7336 error_free(local_err);
7341 qemu_opts_del(opts);
7342 qemu_opts_free(create_opts);
7343 error_propagate(errp, local_err);
7344 aio_context_release(qemu_get_aio_context());
7347 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7350 return bs ? bs->aio_context : qemu_get_aio_context();
7353 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7355 Coroutine *self = qemu_coroutine_self();
7356 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7357 AioContext *new_ctx;
7361 * Increase bs->in_flight to ensure that this operation is completed before
7362 * moving the node to a different AioContext. Read new_ctx only afterwards.
7364 bdrv_inc_in_flight(bs);
7366 new_ctx = bdrv_get_aio_context(bs);
7367 aio_co_reschedule_self(new_ctx);
7371 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7374 aio_co_reschedule_self(old_ctx);
7375 bdrv_dec_in_flight(bs);
7378 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7380 AioContext *ctx = bdrv_get_aio_context(bs);
7382 /* In the main thread, bs->aio_context won't change concurrently */
7383 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7386 * We're in coroutine context, so we already hold the lock of the main
7387 * loop AioContext. Don't lock it twice to avoid deadlocks.
7389 assert(qemu_in_coroutine());
7390 if (ctx != qemu_get_aio_context()) {
7391 aio_context_acquire(ctx);
7395 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7397 AioContext *ctx = bdrv_get_aio_context(bs);
7399 assert(qemu_in_coroutine());
7400 if (ctx != qemu_get_aio_context()) {
7401 aio_context_release(ctx);
7405 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7407 GLOBAL_STATE_CODE();
7408 QLIST_REMOVE(ban, list);
7412 static void bdrv_detach_aio_context(BlockDriverState *bs)
7414 BdrvAioNotifier *baf, *baf_tmp;
7416 assert(!bs->walking_aio_notifiers);
7417 GLOBAL_STATE_CODE();
7418 bs->walking_aio_notifiers = true;
7419 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7421 bdrv_do_remove_aio_context_notifier(baf);
7423 baf->detach_aio_context(baf->opaque);
7426 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7427 * remove remaining aio notifiers if we aren't called again.
7429 bs->walking_aio_notifiers = false;
7431 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7432 bs->drv->bdrv_detach_aio_context(bs);
7435 bs->aio_context = NULL;
7438 static void bdrv_attach_aio_context(BlockDriverState *bs,
7439 AioContext *new_context)
7441 BdrvAioNotifier *ban, *ban_tmp;
7442 GLOBAL_STATE_CODE();
7444 bs->aio_context = new_context;
7446 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7447 bs->drv->bdrv_attach_aio_context(bs, new_context);
7450 assert(!bs->walking_aio_notifiers);
7451 bs->walking_aio_notifiers = true;
7452 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7454 bdrv_do_remove_aio_context_notifier(ban);
7456 ban->attached_aio_context(new_context, ban->opaque);
7459 bs->walking_aio_notifiers = false;
7462 typedef struct BdrvStateSetAioContext {
7463 AioContext *new_ctx;
7464 BlockDriverState *bs;
7465 } BdrvStateSetAioContext;
7467 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7468 GHashTable *visited,
7472 GLOBAL_STATE_CODE();
7473 if (g_hash_table_contains(visited, c)) {
7476 g_hash_table_add(visited, c);
7479 * A BdrvChildClass that doesn't handle AioContext changes cannot
7480 * tolerate any AioContext changes
7482 if (!c->klass->change_aio_ctx) {
7483 char *user = bdrv_child_user_desc(c);
7484 error_setg(errp, "Changing iothreads is not supported by %s", user);
7488 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7489 assert(!errp || *errp);
7495 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7496 GHashTable *visited, Transaction *tran,
7499 GLOBAL_STATE_CODE();
7500 if (g_hash_table_contains(visited, c)) {
7503 g_hash_table_add(visited, c);
7504 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7507 static void bdrv_set_aio_context_clean(void *opaque)
7509 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7510 BlockDriverState *bs = (BlockDriverState *) state->bs;
7512 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7513 bdrv_drained_end(bs);
7518 static void bdrv_set_aio_context_commit(void *opaque)
7520 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7521 BlockDriverState *bs = (BlockDriverState *) state->bs;
7522 AioContext *new_context = state->new_ctx;
7523 AioContext *old_context = bdrv_get_aio_context(bs);
7526 * Take the old AioContex when detaching it from bs.
7527 * At this point, new_context lock is already acquired, and we are now
7528 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7529 * does not call AIO_POLL_WHILE().
7531 if (old_context != qemu_get_aio_context()) {
7532 aio_context_acquire(old_context);
7534 bdrv_detach_aio_context(bs);
7535 if (old_context != qemu_get_aio_context()) {
7536 aio_context_release(old_context);
7538 bdrv_attach_aio_context(bs, new_context);
7541 static TransactionActionDrv set_aio_context = {
7542 .commit = bdrv_set_aio_context_commit,
7543 .clean = bdrv_set_aio_context_clean,
7547 * Changes the AioContext used for fd handlers, timers, and BHs by this
7548 * BlockDriverState and all its children and parents.
7550 * Must be called from the main AioContext.
7552 * The caller must own the AioContext lock for the old AioContext of bs, but it
7553 * must not own the AioContext lock for new_context (unless new_context is the
7554 * same as the current context of bs).
7556 * @visited will accumulate all visited BdrvChild objects. The caller is
7557 * responsible for freeing the list afterwards.
7559 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7560 GHashTable *visited, Transaction *tran,
7564 BdrvStateSetAioContext *state;
7566 GLOBAL_STATE_CODE();
7568 if (bdrv_get_aio_context(bs) == ctx) {
7572 QLIST_FOREACH(c, &bs->parents, next_parent) {
7573 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7578 QLIST_FOREACH(c, &bs->children, next) {
7579 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7584 state = g_new(BdrvStateSetAioContext, 1);
7585 *state = (BdrvStateSetAioContext) {
7590 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7591 bdrv_drained_begin(bs);
7593 tran_add(tran, &set_aio_context, state);
7599 * Change bs's and recursively all of its parents' and children's AioContext
7600 * to the given new context, returning an error if that isn't possible.
7602 * If ignore_child is not NULL, that child (and its subgraph) will not
7605 * This function still requires the caller to take the bs current
7606 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7607 * assumes the lock is always held if bs is in another AioContext.
7608 * For the same reason, it temporarily also holds the new AioContext, since
7609 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7610 * Therefore the new AioContext lock must not be taken by the caller.
7612 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7613 BdrvChild *ignore_child, Error **errp)
7616 GHashTable *visited;
7618 AioContext *old_context = bdrv_get_aio_context(bs);
7619 GLOBAL_STATE_CODE();
7622 * Recursion phase: go through all nodes of the graph.
7623 * Take care of checking that all nodes support changing AioContext
7624 * and drain them, building a linear list of callbacks to run if everything
7625 * is successful (the transaction itself).
7628 visited = g_hash_table_new(NULL, NULL);
7630 g_hash_table_add(visited, ignore_child);
7632 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7633 g_hash_table_destroy(visited);
7636 * Linear phase: go through all callbacks collected in the transaction.
7637 * Run all callbacks collected in the recursion to switch all nodes
7638 * AioContext lock (transaction commit), or undo all changes done in the
7639 * recursion (transaction abort).
7643 /* Just run clean() callbacks. No AioContext changed. */
7649 * Release old AioContext, it won't be needed anymore, as all
7650 * bdrv_drained_begin() have been called already.
7652 if (qemu_get_aio_context() != old_context) {
7653 aio_context_release(old_context);
7657 * Acquire new AioContext since bdrv_drained_end() is going to be called
7658 * after we switched all nodes in the new AioContext, and the function
7659 * assumes that the lock of the bs is always taken.
7661 if (qemu_get_aio_context() != ctx) {
7662 aio_context_acquire(ctx);
7667 if (qemu_get_aio_context() != ctx) {
7668 aio_context_release(ctx);
7671 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7672 if (qemu_get_aio_context() != old_context) {
7673 aio_context_acquire(old_context);
7679 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7680 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7681 void (*detach_aio_context)(void *opaque), void *opaque)
7683 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7684 *ban = (BdrvAioNotifier){
7685 .attached_aio_context = attached_aio_context,
7686 .detach_aio_context = detach_aio_context,
7689 GLOBAL_STATE_CODE();
7691 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7694 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7695 void (*attached_aio_context)(AioContext *,
7697 void (*detach_aio_context)(void *),
7700 BdrvAioNotifier *ban, *ban_next;
7701 GLOBAL_STATE_CODE();
7703 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7704 if (ban->attached_aio_context == attached_aio_context &&
7705 ban->detach_aio_context == detach_aio_context &&
7706 ban->opaque == opaque &&
7707 ban->deleted == false)
7709 if (bs->walking_aio_notifiers) {
7710 ban->deleted = true;
7712 bdrv_do_remove_aio_context_notifier(ban);
7721 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7722 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7726 GLOBAL_STATE_CODE();
7728 error_setg(errp, "Node is ejected");
7731 if (!bs->drv->bdrv_amend_options) {
7732 error_setg(errp, "Block driver '%s' does not support option amendment",
7733 bs->drv->format_name);
7736 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7737 cb_opaque, force, errp);
7741 * This function checks whether the given @to_replace is allowed to be
7742 * replaced by a node that always shows the same data as @bs. This is
7743 * used for example to verify whether the mirror job can replace
7744 * @to_replace by the target mirrored from @bs.
7745 * To be replaceable, @bs and @to_replace may either be guaranteed to
7746 * always show the same data (because they are only connected through
7747 * filters), or some driver may allow replacing one of its children
7748 * because it can guarantee that this child's data is not visible at
7749 * all (for example, for dissenting quorum children that have no other
7752 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7753 BlockDriverState *to_replace)
7755 BlockDriverState *filtered;
7757 GLOBAL_STATE_CODE();
7759 if (!bs || !bs->drv) {
7763 if (bs == to_replace) {
7767 /* See what the driver can do */
7768 if (bs->drv->bdrv_recurse_can_replace) {
7769 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7772 /* For filters without an own implementation, we can recurse on our own */
7773 filtered = bdrv_filter_bs(bs);
7775 return bdrv_recurse_can_replace(filtered, to_replace);
7783 * Check whether the given @node_name can be replaced by a node that
7784 * has the same data as @parent_bs. If so, return @node_name's BDS;
7787 * @node_name must be a (recursive) *child of @parent_bs (or this
7788 * function will return NULL).
7790 * The result (whether the node can be replaced or not) is only valid
7791 * for as long as no graph or permission changes occur.
7793 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7794 const char *node_name, Error **errp)
7796 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7797 AioContext *aio_context;
7799 GLOBAL_STATE_CODE();
7801 if (!to_replace_bs) {
7802 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7806 aio_context = bdrv_get_aio_context(to_replace_bs);
7807 aio_context_acquire(aio_context);
7809 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7810 to_replace_bs = NULL;
7814 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7815 * most non filter in order to prevent data corruption.
7816 * Another benefit is that this tests exclude backing files which are
7817 * blocked by the backing blockers.
7819 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7820 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7821 "because it cannot be guaranteed that doing so would not "
7822 "lead to an abrupt change of visible data",
7823 node_name, parent_bs->node_name);
7824 to_replace_bs = NULL;
7829 aio_context_release(aio_context);
7830 return to_replace_bs;
7834 * Iterates through the list of runtime option keys that are said to
7835 * be "strong" for a BDS. An option is called "strong" if it changes
7836 * a BDS's data. For example, the null block driver's "size" and
7837 * "read-zeroes" options are strong, but its "latency-ns" option is
7840 * If a key returned by this function ends with a dot, all options
7841 * starting with that prefix are strong.
7843 static const char *const *strong_options(BlockDriverState *bs,
7844 const char *const *curopt)
7846 static const char *const global_options[] = {
7847 "driver", "filename", NULL
7851 return &global_options[0];
7855 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7856 curopt = bs->drv->strong_runtime_opts;
7859 return (curopt && *curopt) ? curopt : NULL;
7863 * Copies all strong runtime options from bs->options to the given
7864 * QDict. The set of strong option keys is determined by invoking
7867 * Returns true iff any strong option was present in bs->options (and
7868 * thus copied to the target QDict) with the exception of "filename"
7869 * and "driver". The caller is expected to use this value to decide
7870 * whether the existence of strong options prevents the generation of
7873 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7875 bool found_any = false;
7876 const char *const *option_name = NULL;
7882 while ((option_name = strong_options(bs, option_name))) {
7883 bool option_given = false;
7885 assert(strlen(*option_name) > 0);
7886 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7887 QObject *entry = qdict_get(bs->options, *option_name);
7892 qdict_put_obj(d, *option_name, qobject_ref(entry));
7893 option_given = true;
7895 const QDictEntry *entry;
7896 for (entry = qdict_first(bs->options); entry;
7897 entry = qdict_next(bs->options, entry))
7899 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7900 qdict_put_obj(d, qdict_entry_key(entry),
7901 qobject_ref(qdict_entry_value(entry)));
7902 option_given = true;
7907 /* While "driver" and "filename" need to be included in a JSON filename,
7908 * their existence does not prohibit generation of a plain filename. */
7909 if (!found_any && option_given &&
7910 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7916 if (!qdict_haskey(d, "driver")) {
7917 /* Drivers created with bdrv_new_open_driver() may not have a
7918 * @driver option. Add it here. */
7919 qdict_put_str(d, "driver", bs->drv->format_name);
7925 /* Note: This function may return false positives; it may return true
7926 * even if opening the backing file specified by bs's image header
7927 * would result in exactly bs->backing. */
7928 static bool bdrv_backing_overridden(BlockDriverState *bs)
7930 GLOBAL_STATE_CODE();
7932 return strcmp(bs->auto_backing_file,
7933 bs->backing->bs->filename);
7935 /* No backing BDS, so if the image header reports any backing
7936 * file, it must have been suppressed */
7937 return bs->auto_backing_file[0] != '\0';
7941 /* Updates the following BDS fields:
7942 * - exact_filename: A filename which may be used for opening a block device
7943 * which (mostly) equals the given BDS (even without any
7944 * other options; so reading and writing must return the same
7945 * results, but caching etc. may be different)
7946 * - full_open_options: Options which, when given when opening a block device
7947 * (without a filename), result in a BDS (mostly)
7948 * equalling the given one
7949 * - filename: If exact_filename is set, it is copied here. Otherwise,
7950 * full_open_options is converted to a JSON object, prefixed with
7951 * "json:" (for use through the JSON pseudo protocol) and put here.
7953 void bdrv_refresh_filename(BlockDriverState *bs)
7955 BlockDriver *drv = bs->drv;
7957 BlockDriverState *primary_child_bs;
7959 bool backing_overridden;
7960 bool generate_json_filename; /* Whether our default implementation should
7961 fill exact_filename (false) or not (true) */
7963 GLOBAL_STATE_CODE();
7969 /* This BDS's file name may depend on any of its children's file names, so
7970 * refresh those first */
7971 QLIST_FOREACH(child, &bs->children, next) {
7972 bdrv_refresh_filename(child->bs);
7976 /* For implicit nodes, just copy everything from the single child */
7977 child = QLIST_FIRST(&bs->children);
7978 assert(QLIST_NEXT(child, next) == NULL);
7980 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7981 child->bs->exact_filename);
7982 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7984 qobject_unref(bs->full_open_options);
7985 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7990 backing_overridden = bdrv_backing_overridden(bs);
7992 if (bs->open_flags & BDRV_O_NO_IO) {
7993 /* Without I/O, the backing file does not change anything.
7994 * Therefore, in such a case (primarily qemu-img), we can
7995 * pretend the backing file has not been overridden even if
7996 * it technically has been. */
7997 backing_overridden = false;
8000 /* Gather the options QDict */
8002 generate_json_filename = append_strong_runtime_options(opts, bs);
8003 generate_json_filename |= backing_overridden;
8005 if (drv->bdrv_gather_child_options) {
8006 /* Some block drivers may not want to present all of their children's
8007 * options, or name them differently from BdrvChild.name */
8008 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
8010 QLIST_FOREACH(child, &bs->children, next) {
8011 if (child == bs->backing && !backing_overridden) {
8012 /* We can skip the backing BDS if it has not been overridden */
8016 qdict_put(opts, child->name,
8017 qobject_ref(child->bs->full_open_options));
8020 if (backing_overridden && !bs->backing) {
8021 /* Force no backing file */
8022 qdict_put_null(opts, "backing");
8026 qobject_unref(bs->full_open_options);
8027 bs->full_open_options = opts;
8029 primary_child_bs = bdrv_primary_bs(bs);
8031 if (drv->bdrv_refresh_filename) {
8032 /* Obsolete information is of no use here, so drop the old file name
8033 * information before refreshing it */
8034 bs->exact_filename[0] = '\0';
8036 drv->bdrv_refresh_filename(bs);
8037 } else if (primary_child_bs) {
8039 * Try to reconstruct valid information from the underlying
8040 * file -- this only works for format nodes (filter nodes
8041 * cannot be probed and as such must be selected by the user
8042 * either through an options dict, or through a special
8043 * filename which the filter driver must construct in its
8044 * .bdrv_refresh_filename() implementation).
8047 bs->exact_filename[0] = '\0';
8050 * We can use the underlying file's filename if:
8051 * - it has a filename,
8052 * - the current BDS is not a filter,
8053 * - the file is a protocol BDS, and
8054 * - opening that file (as this BDS's format) will automatically create
8055 * the BDS tree we have right now, that is:
8056 * - the user did not significantly change this BDS's behavior with
8057 * some explicit (strong) options
8058 * - no non-file child of this BDS has been overridden by the user
8059 * Both of these conditions are represented by generate_json_filename.
8061 if (primary_child_bs->exact_filename[0] &&
8062 primary_child_bs->drv->bdrv_file_open &&
8063 !drv->is_filter && !generate_json_filename)
8065 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
8069 if (bs->exact_filename[0]) {
8070 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
8072 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
8073 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
8074 json->str) >= sizeof(bs->filename)) {
8075 /* Give user a hint if we truncated things. */
8076 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8078 g_string_free(json, true);
8082 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8084 BlockDriver *drv = bs->drv;
8085 BlockDriverState *child_bs;
8087 GLOBAL_STATE_CODE();
8090 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8094 if (drv->bdrv_dirname) {
8095 return drv->bdrv_dirname(bs, errp);
8098 child_bs = bdrv_primary_bs(bs);
8100 return bdrv_dirname(child_bs, errp);
8103 bdrv_refresh_filename(bs);
8104 if (bs->exact_filename[0] != '\0') {
8105 return path_combine(bs->exact_filename, "");
8108 error_setg(errp, "Cannot generate a base directory for %s nodes",
8114 * Hot add/remove a BDS's child. So the user can take a child offline when
8115 * it is broken and take a new child online
8117 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8120 GLOBAL_STATE_CODE();
8121 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8122 error_setg(errp, "The node %s does not support adding a child",
8123 bdrv_get_device_or_node_name(parent_bs));
8128 * Non-zoned block drivers do not follow zoned storage constraints
8129 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8130 * drivers in a graph.
8132 if (!parent_bs->drv->supports_zoned_children &&
8133 child_bs->bl.zoned == BLK_Z_HM) {
8135 * The host-aware model allows zoned storage constraints and random
8136 * write. Allow mixing host-aware and non-zoned drivers. Using
8137 * host-aware device as a regular device.
8139 error_setg(errp, "Cannot add a %s child to a %s parent",
8140 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8141 parent_bs->drv->supports_zoned_children ?
8142 "support zoned children" : "not support zoned children");
8146 if (!QLIST_EMPTY(&child_bs->parents)) {
8147 error_setg(errp, "The node %s already has a parent",
8148 child_bs->node_name);
8152 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8155 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8159 GLOBAL_STATE_CODE();
8160 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8161 error_setg(errp, "The node %s does not support removing a child",
8162 bdrv_get_device_or_node_name(parent_bs));
8166 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8173 error_setg(errp, "The node %s does not have a child named %s",
8174 bdrv_get_device_or_node_name(parent_bs),
8175 bdrv_get_device_or_node_name(child->bs));
8179 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8182 int bdrv_make_empty(BdrvChild *c, Error **errp)
8184 BlockDriver *drv = c->bs->drv;
8187 GLOBAL_STATE_CODE();
8188 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8190 if (!drv->bdrv_make_empty) {
8191 error_setg(errp, "%s does not support emptying nodes",
8196 ret = drv->bdrv_make_empty(c->bs);
8198 error_setg_errno(errp, -ret, "Failed to empty %s",
8207 * Return the child that @bs acts as an overlay for, and from which data may be
8208 * copied in COW or COR operations. Usually this is the backing file.
8210 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8214 if (!bs || !bs->drv) {
8218 if (bs->drv->is_filter) {
8226 assert(bs->backing->role & BDRV_CHILD_COW);
8231 * If @bs acts as a filter for exactly one of its children, return
8234 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8239 if (!bs || !bs->drv) {
8243 if (!bs->drv->is_filter) {
8247 /* Only one of @backing or @file may be used */
8248 assert(!(bs->backing && bs->file));
8250 c = bs->backing ?: bs->file;
8255 assert(c->role & BDRV_CHILD_FILTERED);
8260 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8261 * whichever is non-NULL.
8263 * Return NULL if both are NULL.
8265 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8267 BdrvChild *cow_child = bdrv_cow_child(bs);
8268 BdrvChild *filter_child = bdrv_filter_child(bs);
8271 /* Filter nodes cannot have COW backing files */
8272 assert(!(cow_child && filter_child));
8274 return cow_child ?: filter_child;
8278 * Return the primary child of this node: For filters, that is the
8279 * filtered child. For other nodes, that is usually the child storing
8281 * (A generally more helpful description is that this is (usually) the
8282 * child that has the same filename as @bs.)
8284 * Drivers do not necessarily have a primary child; for example quorum
8287 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8289 BdrvChild *c, *found = NULL;
8292 QLIST_FOREACH(c, &bs->children, next) {
8293 if (c->role & BDRV_CHILD_PRIMARY) {
8302 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8303 bool stop_on_explicit_filter)
8311 while (!(stop_on_explicit_filter && !bs->implicit)) {
8312 c = bdrv_filter_child(bs);
8315 * A filter that is embedded in a working block graph must
8316 * have a child. Assert this here so this function does
8317 * not return a filter node that is not expected by the
8320 assert(!bs->drv || !bs->drv->is_filter);
8326 * Note that this treats nodes with bs->drv == NULL as not being
8327 * filters (bs->drv == NULL should be replaced by something else
8329 * The advantage of this behavior is that this function will thus
8330 * always return a non-NULL value (given a non-NULL @bs).
8337 * Return the first BDS that has not been added implicitly or that
8338 * does not have a filtered child down the chain starting from @bs
8339 * (including @bs itself).
8341 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8343 GLOBAL_STATE_CODE();
8344 return bdrv_do_skip_filters(bs, true);
8348 * Return the first BDS that does not have a filtered child down the
8349 * chain starting from @bs (including @bs itself).
8351 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8354 return bdrv_do_skip_filters(bs, false);
8358 * For a backing chain, return the first non-filter backing image of
8359 * the first non-filter image.
8361 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8364 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8368 * Check whether [offset, offset + bytes) overlaps with the cached
8369 * block-status data region.
8371 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8372 * which is what bdrv_bsc_is_data()'s interface needs.
8373 * Otherwise, *pnum is not touched.
8375 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8376 int64_t offset, int64_t bytes,
8379 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8383 qatomic_read(&bsc->valid) &&
8384 ranges_overlap(offset, bytes, bsc->data_start,
8385 bsc->data_end - bsc->data_start);
8387 if (overlaps && pnum) {
8388 *pnum = bsc->data_end - offset;
8395 * See block_int.h for this function's documentation.
8397 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8400 RCU_READ_LOCK_GUARD();
8401 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8405 * See block_int.h for this function's documentation.
8407 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8408 int64_t offset, int64_t bytes)
8411 RCU_READ_LOCK_GUARD();
8413 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8414 qatomic_set(&bs->block_status_cache->valid, false);
8419 * See block_int.h for this function's documentation.
8421 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8423 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8424 BdrvBlockStatusCache *old_bsc;
8427 *new_bsc = (BdrvBlockStatusCache) {
8429 .data_start = offset,
8430 .data_end = offset + bytes,
8433 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8435 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8436 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8438 g_free_rcu(old_bsc, rcu);