2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
53 #include "qemu/range.h"
55 #include "block/coroutines.h"
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
81 QLIST_HEAD_INITIALIZER(bdrv_drivers);
83 static BlockDriverState *bdrv_open_inherit(const char *filename,
84 const char *reference,
85 QDict *options, int flags,
86 BlockDriverState *parent,
87 const BdrvChildClass *child_class,
88 BdrvChildRole child_role,
91 static bool bdrv_recurse_has_child(BlockDriverState *bs,
92 BlockDriverState *child);
94 static void bdrv_replace_child_noperm(BdrvChild *child,
95 BlockDriverState *new_bs);
96 static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
98 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
99 BlockReopenQueue *queue,
100 Transaction *change_child_tran, Error **errp);
101 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
102 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
104 static bool bdrv_backing_overridden(BlockDriverState *bs);
106 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
107 GHashTable *visited, Transaction *tran,
110 /* If non-zero, use only whitelisted block drivers */
111 static int use_bdrv_whitelist;
114 static int is_windows_drive_prefix(const char *filename)
116 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
117 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
121 int is_windows_drive(const char *filename)
123 if (is_windows_drive_prefix(filename) &&
126 if (strstart(filename, "\\\\.\\", NULL) ||
127 strstart(filename, "//./", NULL))
133 size_t bdrv_opt_mem_align(BlockDriverState *bs)
135 if (!bs || !bs->drv) {
136 /* page size or 4k (hdd sector size) should be on the safe side */
137 return MAX(4096, qemu_real_host_page_size());
141 return bs->bl.opt_mem_alignment;
144 size_t bdrv_min_mem_align(BlockDriverState *bs)
146 if (!bs || !bs->drv) {
147 /* page size or 4k (hdd sector size) should be on the safe side */
148 return MAX(4096, qemu_real_host_page_size());
152 return bs->bl.min_mem_alignment;
155 /* check if the path starts with "<protocol>:" */
156 int path_has_protocol(const char *path)
161 if (is_windows_drive(path) ||
162 is_windows_drive_prefix(path)) {
165 p = path + strcspn(path, ":/\\");
167 p = path + strcspn(path, ":/");
173 int path_is_absolute(const char *path)
176 /* specific case for names like: "\\.\d:" */
177 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
180 return (*path == '/' || *path == '\\');
182 return (*path == '/');
186 /* if filename is absolute, just return its duplicate. Otherwise, build a
187 path to it by considering it is relative to base_path. URL are
189 char *path_combine(const char *base_path, const char *filename)
191 const char *protocol_stripped = NULL;
196 if (path_is_absolute(filename)) {
197 return g_strdup(filename);
200 if (path_has_protocol(base_path)) {
201 protocol_stripped = strchr(base_path, ':');
202 if (protocol_stripped) {
206 p = protocol_stripped ?: base_path;
208 p1 = strrchr(base_path, '/');
212 p2 = strrchr(base_path, '\\');
213 if (!p1 || p2 > p1) {
228 result = g_malloc(len + strlen(filename) + 1);
229 memcpy(result, base_path, len);
230 strcpy(result + len, filename);
236 * Helper function for bdrv_parse_filename() implementations to remove optional
237 * protocol prefixes (especially "file:") from a filename and for putting the
238 * stripped filename into the options QDict if there is such a prefix.
240 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
243 if (strstart(filename, prefix, &filename)) {
244 /* Stripping the explicit protocol prefix may result in a protocol
245 * prefix being (wrongly) detected (if the filename contains a colon) */
246 if (path_has_protocol(filename)) {
247 GString *fat_filename;
249 /* This means there is some colon before the first slash; therefore,
250 * this cannot be an absolute path */
251 assert(!path_is_absolute(filename));
253 /* And we can thus fix the protocol detection issue by prefixing it
255 fat_filename = g_string_new("./");
256 g_string_append(fat_filename, filename);
258 assert(!path_has_protocol(fat_filename->str));
260 qdict_put(options, "filename",
261 qstring_from_gstring(fat_filename));
263 /* If no protocol prefix was detected, we can use the shortened
265 qdict_put_str(options, "filename", filename);
271 /* Returns whether the image file is opened as read-only. Note that this can
272 * return false and writing to the image file is still not possible because the
273 * image is inactivated. */
274 bool bdrv_is_read_only(BlockDriverState *bs)
277 return !(bs->open_flags & BDRV_O_RDWR);
280 static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
281 bool ignore_allow_rdw, Error **errp)
285 /* Do not set read_only if copy_on_read is enabled */
286 if (bs->copy_on_read && read_only) {
287 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
288 bdrv_get_device_or_node_name(bs));
292 /* Do not clear read_only if it is prohibited */
293 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
296 error_setg(errp, "Node '%s' is read only",
297 bdrv_get_device_or_node_name(bs));
305 * Called by a driver that can only provide a read-only image.
307 * Returns 0 if the node is already read-only or it could switch the node to
308 * read-only because BDRV_O_AUTO_RDONLY is set.
310 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
311 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
312 * is not NULL, it is used as the error message for the Error object.
314 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
320 if (!(bs->open_flags & BDRV_O_RDWR)) {
323 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
327 ret = bdrv_can_set_read_only(bs, true, false, NULL);
332 bs->open_flags &= ~BDRV_O_RDWR;
337 error_setg(errp, "%s", errmsg ?: "Image is read-only");
342 * If @backing is empty, this function returns NULL without setting
343 * @errp. In all other cases, NULL will only be returned with @errp
346 * Therefore, a return value of NULL without @errp set means that
347 * there is no backing file; if @errp is set, there is one but its
348 * absolute filename cannot be generated.
350 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
354 if (backing[0] == '\0') {
356 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
357 return g_strdup(backing);
358 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
359 error_setg(errp, "Cannot use relative backing file names for '%s'",
363 return path_combine(backed, backing);
368 * If @filename is empty or NULL, this function returns NULL without
369 * setting @errp. In all other cases, NULL will only be returned with
372 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
373 const char *filename, Error **errp)
375 char *dir, *full_name;
377 if (!filename || filename[0] == '\0') {
379 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
380 return g_strdup(filename);
383 dir = bdrv_dirname(relative_to, errp);
388 full_name = g_strconcat(dir, filename, NULL);
393 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
396 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
399 void bdrv_register(BlockDriver *bdrv)
401 assert(bdrv->format_name);
403 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
406 BlockDriverState *bdrv_new(void)
408 BlockDriverState *bs;
413 bs = g_new0(BlockDriverState, 1);
414 QLIST_INIT(&bs->dirty_bitmaps);
415 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
416 QLIST_INIT(&bs->op_blockers[i]);
418 qemu_co_mutex_init(&bs->reqs_lock);
419 qemu_mutex_init(&bs->dirty_bitmap_mutex);
421 bs->aio_context = qemu_get_aio_context();
423 qemu_co_queue_init(&bs->flush_queue);
425 qemu_co_mutex_init(&bs->bsc_modify_lock);
426 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
428 for (i = 0; i < bdrv_drain_all_count; i++) {
429 bdrv_drained_begin(bs);
432 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
437 static BlockDriver *bdrv_do_find_format(const char *format_name)
442 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
443 if (!strcmp(drv1->format_name, format_name)) {
451 BlockDriver *bdrv_find_format(const char *format_name)
458 drv1 = bdrv_do_find_format(format_name);
463 /* The driver isn't registered, maybe we need to load a module */
464 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
465 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
466 Error *local_err = NULL;
467 int rv = block_module_load(block_driver_modules[i].library_name,
470 return bdrv_do_find_format(format_name);
472 error_report_err(local_err);
480 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
482 static const char *whitelist_rw[] = {
483 CONFIG_BDRV_RW_WHITELIST
486 static const char *whitelist_ro[] = {
487 CONFIG_BDRV_RO_WHITELIST
492 if (!whitelist_rw[0] && !whitelist_ro[0]) {
493 return 1; /* no whitelist, anything goes */
496 for (p = whitelist_rw; *p; p++) {
497 if (!strcmp(format_name, *p)) {
502 for (p = whitelist_ro; *p; p++) {
503 if (!strcmp(format_name, *p)) {
511 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
514 return bdrv_format_is_whitelisted(drv->format_name, read_only);
517 bool bdrv_uses_whitelist(void)
519 return use_bdrv_whitelist;
522 typedef struct CreateCo {
530 int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
531 QemuOpts *opts, Error **errp)
537 if (!drv->bdrv_co_create_opts) {
538 error_setg(errp, "Driver '%s' does not support image creation",
543 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
544 if (ret < 0 && !*errp) {
545 error_setg_errno(errp, -ret, "Could not create image");
552 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
553 * least the given @minimum_size.
555 * On success, return @blk's actual length.
556 * Otherwise, return -errno.
558 static int64_t create_file_fallback_truncate(BlockBackend *blk,
559 int64_t minimum_size, Error **errp)
561 Error *local_err = NULL;
567 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
569 if (ret < 0 && ret != -ENOTSUP) {
570 error_propagate(errp, local_err);
574 size = blk_getlength(blk);
576 error_free(local_err);
577 error_setg_errno(errp, -size,
578 "Failed to inquire the new image file's length");
582 if (size < minimum_size) {
583 /* Need to grow the image, but we failed to do that */
584 error_propagate(errp, local_err);
588 error_free(local_err);
595 * Helper function for bdrv_create_file_fallback(): Zero the first
596 * sector to remove any potentially pre-existing image header.
598 static int coroutine_fn
599 create_file_fallback_zero_first_sector(BlockBackend *blk,
600 int64_t current_size,
603 int64_t bytes_to_clear;
608 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
609 if (bytes_to_clear) {
610 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
612 error_setg_errno(errp, -ret,
613 "Failed to clear the new image's first sector");
622 * Simple implementation of bdrv_co_create_opts for protocol drivers
623 * which only support creation via opening a file
624 * (usually existing raw storage device)
626 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
627 const char *filename,
635 PreallocMode prealloc;
636 Error *local_err = NULL;
641 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
642 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
643 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
644 PREALLOC_MODE_OFF, &local_err);
647 error_propagate(errp, local_err);
651 if (prealloc != PREALLOC_MODE_OFF) {
652 error_setg(errp, "Unsupported preallocation mode '%s'",
653 PreallocMode_str(prealloc));
657 options = qdict_new();
658 qdict_put_str(options, "driver", drv->format_name);
660 blk = blk_co_new_open(filename, NULL, options,
661 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
663 error_prepend(errp, "Protocol driver '%s' does not support image "
664 "creation, and opening the image failed: ",
669 size = create_file_fallback_truncate(blk, size, errp);
675 ret = create_file_fallback_zero_first_sector(blk, size, errp);
686 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
689 QemuOpts *protocol_opts;
696 drv = bdrv_find_protocol(filename, true, errp);
701 if (!drv->create_opts) {
702 error_setg(errp, "Driver '%s' does not support image creation",
708 * 'opts' contains a QemuOptsList with a combination of format and protocol
711 * The format properly removes its options, but the default values remain
712 * in 'opts->list'. So if the protocol has options with the same name
713 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
714 * of the format, since for overlapping options, the format wins.
716 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
717 * only the set options, and then convert it back to QemuOpts, using the
718 * create_opts of the protocol. So the new QemuOpts, will contain only the
721 qdict = qemu_opts_to_qdict(opts, NULL);
722 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
723 if (protocol_opts == NULL) {
728 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
730 qemu_opts_del(protocol_opts);
731 qobject_unref(qdict);
735 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
737 Error *local_err = NULL;
742 assert_bdrv_graph_readable();
745 error_setg(errp, "Block node '%s' is not opened", bs->filename);
749 if (!bs->drv->bdrv_co_delete_file) {
750 error_setg(errp, "Driver '%s' does not support image deletion",
751 bs->drv->format_name);
755 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
757 error_propagate(errp, local_err);
763 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
765 Error *local_err = NULL;
773 ret = bdrv_co_delete_file(bs, &local_err);
775 * ENOTSUP will happen if the block driver doesn't support
776 * the 'bdrv_co_delete_file' interface. This is a predictable
777 * scenario and shouldn't be reported back to the user.
779 if (ret == -ENOTSUP) {
780 error_free(local_err);
781 } else if (ret < 0) {
782 error_report_err(local_err);
787 * Try to get @bs's logical and physical block size.
788 * On success, store them in @bsz struct and return 0.
789 * On failure return -errno.
790 * @bs must not be empty.
792 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
794 BlockDriver *drv = bs->drv;
795 BlockDriverState *filtered = bdrv_filter_bs(bs);
798 if (drv && drv->bdrv_probe_blocksizes) {
799 return drv->bdrv_probe_blocksizes(bs, bsz);
800 } else if (filtered) {
801 return bdrv_probe_blocksizes(filtered, bsz);
808 * Try to get @bs's geometry (cyls, heads, sectors).
809 * On success, store them in @geo struct and return 0.
810 * On failure return -errno.
811 * @bs must not be empty.
813 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
815 BlockDriver *drv = bs->drv;
816 BlockDriverState *filtered = bdrv_filter_bs(bs);
819 if (drv && drv->bdrv_probe_geometry) {
820 return drv->bdrv_probe_geometry(bs, geo);
821 } else if (filtered) {
822 return bdrv_probe_geometry(filtered, geo);
829 * Create a uniquely-named empty temporary file.
830 * Return the actual file name used upon success, otherwise NULL.
831 * This string should be freed with g_free() when not needed any longer.
833 * Note: creating a temporary file for the caller to (re)open is
834 * inherently racy. Use g_file_open_tmp() instead whenever practical.
836 char *create_tmp_file(Error **errp)
840 g_autofree char *filename = NULL;
842 tmpdir = g_get_tmp_dir();
845 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
847 * This function is used to create temporary disk images (like -snapshot),
848 * so the files can become very large. /tmp is often a tmpfs where as
849 * /var/tmp is usually on a disk, so more appropriate for disk images.
851 if (!g_strcmp0(tmpdir, "/tmp")) {
856 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
857 fd = g_mkstemp(filename);
859 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
865 return g_steal_pointer(&filename);
869 * Detect host devices. By convention, /dev/cdrom[N] is always
870 * recognized as a host CDROM.
872 static BlockDriver *find_hdev_driver(const char *filename)
874 int score_max = 0, score;
875 BlockDriver *drv = NULL, *d;
878 QLIST_FOREACH(d, &bdrv_drivers, list) {
879 if (d->bdrv_probe_device) {
880 score = d->bdrv_probe_device(filename);
881 if (score > score_max) {
891 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
896 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
897 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
905 BlockDriver *bdrv_find_protocol(const char *filename,
906 bool allow_protocol_prefix,
916 /* TODO Drivers without bdrv_file_open must be specified explicitly */
919 * XXX(hch): we really should not let host device detection
920 * override an explicit protocol specification, but moving this
921 * later breaks access to device names with colons in them.
922 * Thanks to the brain-dead persistent naming schemes on udev-
923 * based Linux systems those actually are quite common.
925 drv1 = find_hdev_driver(filename);
930 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
934 p = strchr(filename, ':');
937 if (len > sizeof(protocol) - 1)
938 len = sizeof(protocol) - 1;
939 memcpy(protocol, filename, len);
940 protocol[len] = '\0';
942 drv1 = bdrv_do_find_protocol(protocol);
947 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
948 if (block_driver_modules[i].protocol_name &&
949 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
950 int rv = block_module_load(block_driver_modules[i].library_name, errp);
952 drv1 = bdrv_do_find_protocol(protocol);
961 error_setg(errp, "Unknown protocol '%s'", protocol);
967 * Guess image format by probing its contents.
968 * This is not a good idea when your image is raw (CVE-2008-2004), but
969 * we do it anyway for backward compatibility.
971 * @buf contains the image's first @buf_size bytes.
972 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
973 * but can be smaller if the image file is smaller)
974 * @filename is its filename.
976 * For all block drivers, call the bdrv_probe() method to get its
978 * Return the first block driver with the highest probing score.
980 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
981 const char *filename)
983 int score_max = 0, score;
984 BlockDriver *drv = NULL, *d;
987 QLIST_FOREACH(d, &bdrv_drivers, list) {
989 score = d->bdrv_probe(buf, buf_size, filename);
990 if (score > score_max) {
1000 static int find_image_format(BlockBackend *file, const char *filename,
1001 BlockDriver **pdrv, Error **errp)
1004 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1007 GLOBAL_STATE_CODE();
1009 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1010 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1015 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1017 error_setg_errno(errp, -ret, "Could not read image for determining its "
1023 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1025 error_setg(errp, "Could not determine image format: No compatible "
1036 * Set the current 'total_sectors' value
1037 * Return 0 on success, -errno on error.
1039 int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1042 BlockDriver *drv = bs->drv;
1044 assert_bdrv_graph_readable();
1050 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1054 /* query actual device if possible, otherwise just trust the hint */
1055 if (drv->bdrv_co_getlength) {
1056 int64_t length = drv->bdrv_co_getlength(bs);
1060 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1063 bs->total_sectors = hint;
1065 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1073 * Combines a QDict of new block driver @options with any missing options taken
1074 * from @old_options, so that leaving out an option defaults to its old value.
1076 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1079 GLOBAL_STATE_CODE();
1080 if (bs->drv && bs->drv->bdrv_join_options) {
1081 bs->drv->bdrv_join_options(options, old_options);
1083 qdict_join(options, old_options, false);
1087 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1091 Error *local_err = NULL;
1092 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1093 BlockdevDetectZeroesOptions detect_zeroes =
1094 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1095 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1096 GLOBAL_STATE_CODE();
1099 error_propagate(errp, local_err);
1100 return detect_zeroes;
1103 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1104 !(open_flags & BDRV_O_UNMAP))
1106 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1107 "without setting discard operation to unmap");
1110 return detect_zeroes;
1114 * Set open flags for aio engine
1116 * Return 0 on success, -1 if the engine specified is invalid
1118 int bdrv_parse_aio(const char *mode, int *flags)
1120 if (!strcmp(mode, "threads")) {
1121 /* do nothing, default */
1122 } else if (!strcmp(mode, "native")) {
1123 *flags |= BDRV_O_NATIVE_AIO;
1124 #ifdef CONFIG_LINUX_IO_URING
1125 } else if (!strcmp(mode, "io_uring")) {
1126 *flags |= BDRV_O_IO_URING;
1136 * Set open flags for a given discard mode
1138 * Return 0 on success, -1 if the discard mode was invalid.
1140 int bdrv_parse_discard_flags(const char *mode, int *flags)
1142 *flags &= ~BDRV_O_UNMAP;
1144 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1146 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1147 *flags |= BDRV_O_UNMAP;
1156 * Set open flags for a given cache mode
1158 * Return 0 on success, -1 if the cache mode was invalid.
1160 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1162 *flags &= ~BDRV_O_CACHE_MASK;
1164 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1165 *writethrough = false;
1166 *flags |= BDRV_O_NOCACHE;
1167 } else if (!strcmp(mode, "directsync")) {
1168 *writethrough = true;
1169 *flags |= BDRV_O_NOCACHE;
1170 } else if (!strcmp(mode, "writeback")) {
1171 *writethrough = false;
1172 } else if (!strcmp(mode, "unsafe")) {
1173 *writethrough = false;
1174 *flags |= BDRV_O_NO_FLUSH;
1175 } else if (!strcmp(mode, "writethrough")) {
1176 *writethrough = true;
1184 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1186 BlockDriverState *parent = c->opaque;
1187 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1190 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1192 BlockDriverState *bs = child->opaque;
1193 bdrv_do_drained_begin_quiesce(bs, NULL);
1196 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1198 BlockDriverState *bs = child->opaque;
1199 return bdrv_drain_poll(bs, NULL, false);
1202 static void bdrv_child_cb_drained_end(BdrvChild *child)
1204 BlockDriverState *bs = child->opaque;
1205 bdrv_drained_end(bs);
1208 static int bdrv_child_cb_inactivate(BdrvChild *child)
1210 BlockDriverState *bs = child->opaque;
1211 GLOBAL_STATE_CODE();
1212 assert(bs->open_flags & BDRV_O_INACTIVE);
1216 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1217 GHashTable *visited, Transaction *tran,
1220 BlockDriverState *bs = child->opaque;
1221 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1225 * Returns the options and flags that a temporary snapshot should get, based on
1226 * the originally requested flags (the originally requested image will have
1227 * flags like a backing file)
1229 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1230 int parent_flags, QDict *parent_options)
1232 GLOBAL_STATE_CODE();
1233 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1235 /* For temporary files, unconditional cache=unsafe is fine */
1236 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1237 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1239 /* Copy the read-only and discard options from the parent */
1240 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1241 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1243 /* aio=native doesn't work for cache.direct=off, so disable it for the
1244 * temporary snapshot */
1245 *child_flags &= ~BDRV_O_NATIVE_AIO;
1248 static void bdrv_backing_attach(BdrvChild *c)
1250 BlockDriverState *parent = c->opaque;
1251 BlockDriverState *backing_hd = c->bs;
1253 GLOBAL_STATE_CODE();
1254 assert(!parent->backing_blocker);
1255 error_setg(&parent->backing_blocker,
1256 "node is used as backing hd of '%s'",
1257 bdrv_get_device_or_node_name(parent));
1259 bdrv_refresh_filename(backing_hd);
1261 parent->open_flags &= ~BDRV_O_NO_BACKING;
1263 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1264 /* Otherwise we won't be able to commit or stream */
1265 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1266 parent->backing_blocker);
1267 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1268 parent->backing_blocker);
1270 * We do backup in 3 ways:
1272 * The target bs is new opened, and the source is top BDS
1273 * 2. blockdev backup
1274 * Both the source and the target are top BDSes.
1275 * 3. internal backup(used for block replication)
1276 * Both the source and the target are backing file
1278 * In case 1 and 2, neither the source nor the target is the backing file.
1279 * In case 3, we will block the top BDS, so there is only one block job
1280 * for the top BDS and its backing chain.
1282 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1283 parent->backing_blocker);
1284 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1285 parent->backing_blocker);
1288 static void bdrv_backing_detach(BdrvChild *c)
1290 BlockDriverState *parent = c->opaque;
1292 GLOBAL_STATE_CODE();
1293 assert(parent->backing_blocker);
1294 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1295 error_free(parent->backing_blocker);
1296 parent->backing_blocker = NULL;
1299 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1300 const char *filename, Error **errp)
1302 BlockDriverState *parent = c->opaque;
1303 bool read_only = bdrv_is_read_only(parent);
1305 GLOBAL_STATE_CODE();
1308 ret = bdrv_reopen_set_read_only(parent, false, errp);
1314 ret = bdrv_change_backing_file(parent, filename,
1315 base->drv ? base->drv->format_name : "",
1318 error_setg_errno(errp, -ret, "Could not update backing file link");
1322 bdrv_reopen_set_read_only(parent, true, NULL);
1329 * Returns the options and flags that a generic child of a BDS should
1330 * get, based on the given options and flags for the parent BDS.
1332 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1333 int *child_flags, QDict *child_options,
1334 int parent_flags, QDict *parent_options)
1336 int flags = parent_flags;
1337 GLOBAL_STATE_CODE();
1340 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1341 * Generally, the question to answer is: Should this child be
1342 * format-probed by default?
1346 * Pure and non-filtered data children of non-format nodes should
1347 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1348 * set). This only affects a very limited set of drivers (namely
1349 * quorum and blkverify when this comment was written).
1350 * Force-clear BDRV_O_PROTOCOL then.
1352 if (!parent_is_format &&
1353 (role & BDRV_CHILD_DATA) &&
1354 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1356 flags &= ~BDRV_O_PROTOCOL;
1360 * All children of format nodes (except for COW children) and all
1361 * metadata children in general should never be format-probed.
1362 * Force-set BDRV_O_PROTOCOL then.
1364 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1365 (role & BDRV_CHILD_METADATA))
1367 flags |= BDRV_O_PROTOCOL;
1371 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1374 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1375 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1376 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1378 if (role & BDRV_CHILD_COW) {
1379 /* backing files are opened read-only by default */
1380 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1381 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1383 /* Inherit the read-only option from the parent if it's not set */
1384 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1385 qdict_copy_default(child_options, parent_options,
1386 BDRV_OPT_AUTO_READ_ONLY);
1390 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1391 * can default to enable it on lower layers regardless of the
1394 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1396 /* Clear flags that only apply to the top layer */
1397 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1399 if (role & BDRV_CHILD_METADATA) {
1400 flags &= ~BDRV_O_NO_IO;
1402 if (role & BDRV_CHILD_COW) {
1403 flags &= ~BDRV_O_TEMPORARY;
1406 *child_flags = flags;
1409 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1411 BlockDriverState *bs = child->opaque;
1413 assert_bdrv_graph_writable();
1414 QLIST_INSERT_HEAD(&bs->children, child, next);
1415 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1417 * Here we handle filters and block/raw-format.c when it behave like
1418 * filter. They generally have a single PRIMARY child, which is also the
1419 * FILTERED child, and that they may have multiple more children, which
1420 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1421 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1422 * into bs->backing on exceptional cases; and bs->backing will be
1425 assert(!(child->role & BDRV_CHILD_COW));
1426 if (child->role & BDRV_CHILD_PRIMARY) {
1427 assert(child->role & BDRV_CHILD_FILTERED);
1428 assert(!bs->backing);
1431 if (bs->drv->filtered_child_is_backing) {
1432 bs->backing = child;
1437 assert(!(child->role & BDRV_CHILD_FILTERED));
1439 } else if (child->role & BDRV_CHILD_COW) {
1440 assert(bs->drv->supports_backing);
1441 assert(!(child->role & BDRV_CHILD_PRIMARY));
1442 assert(!bs->backing);
1443 bs->backing = child;
1444 bdrv_backing_attach(child);
1445 } else if (child->role & BDRV_CHILD_PRIMARY) {
1451 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1453 BlockDriverState *bs = child->opaque;
1455 if (child->role & BDRV_CHILD_COW) {
1456 bdrv_backing_detach(child);
1459 assert_bdrv_graph_writable();
1460 QLIST_REMOVE(child, next);
1461 if (child == bs->backing) {
1462 assert(child != bs->file);
1464 } else if (child == bs->file) {
1469 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1470 const char *filename, Error **errp)
1472 if (c->role & BDRV_CHILD_COW) {
1473 return bdrv_backing_update_filename(c, base, filename, errp);
1478 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1480 BlockDriverState *bs = c->opaque;
1483 return bdrv_get_aio_context(bs);
1486 const BdrvChildClass child_of_bds = {
1487 .parent_is_bds = true,
1488 .get_parent_desc = bdrv_child_get_parent_desc,
1489 .inherit_options = bdrv_inherited_options,
1490 .drained_begin = bdrv_child_cb_drained_begin,
1491 .drained_poll = bdrv_child_cb_drained_poll,
1492 .drained_end = bdrv_child_cb_drained_end,
1493 .attach = bdrv_child_cb_attach,
1494 .detach = bdrv_child_cb_detach,
1495 .inactivate = bdrv_child_cb_inactivate,
1496 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1497 .update_filename = bdrv_child_cb_update_filename,
1498 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1501 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1504 return c->klass->get_parent_aio_context(c);
1507 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1509 int open_flags = flags;
1510 GLOBAL_STATE_CODE();
1513 * Clear flags that are internal to the block layer before opening the
1516 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1521 static void update_flags_from_options(int *flags, QemuOpts *opts)
1523 GLOBAL_STATE_CODE();
1525 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1527 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1528 *flags |= BDRV_O_NO_FLUSH;
1531 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1532 *flags |= BDRV_O_NOCACHE;
1535 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1536 *flags |= BDRV_O_RDWR;
1539 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1540 *flags |= BDRV_O_AUTO_RDONLY;
1544 static void update_options_from_flags(QDict *options, int flags)
1546 GLOBAL_STATE_CODE();
1547 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1548 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1550 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1551 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1552 flags & BDRV_O_NO_FLUSH);
1554 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1555 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1557 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1558 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1559 flags & BDRV_O_AUTO_RDONLY);
1563 static void bdrv_assign_node_name(BlockDriverState *bs,
1564 const char *node_name,
1567 char *gen_node_name = NULL;
1568 GLOBAL_STATE_CODE();
1571 node_name = gen_node_name = id_generate(ID_BLOCK);
1572 } else if (!id_wellformed(node_name)) {
1574 * Check for empty string or invalid characters, but not if it is
1575 * generated (generated names use characters not available to the user)
1577 error_setg(errp, "Invalid node-name: '%s'", node_name);
1581 /* takes care of avoiding namespaces collisions */
1582 if (blk_by_name(node_name)) {
1583 error_setg(errp, "node-name=%s is conflicting with a device id",
1588 /* takes care of avoiding duplicates node names */
1589 if (bdrv_find_node(node_name)) {
1590 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1594 /* Make sure that the node name isn't truncated */
1595 if (strlen(node_name) >= sizeof(bs->node_name)) {
1596 error_setg(errp, "Node name too long");
1600 /* copy node name into the bs and insert it into the graph list */
1601 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1602 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1604 g_free(gen_node_name);
1608 * The caller must always hold @bs AioContext lock, because this function calls
1609 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1612 static int no_coroutine_fn GRAPH_UNLOCKED
1613 bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1614 QDict *options, int open_flags, Error **errp)
1617 Error *local_err = NULL;
1619 GLOBAL_STATE_CODE();
1621 bdrv_assign_node_name(bs, node_name, &local_err);
1623 error_propagate(errp, local_err);
1628 bs->opaque = g_malloc0(drv->instance_size);
1630 if (drv->bdrv_file_open) {
1631 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1632 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1633 } else if (drv->bdrv_open) {
1634 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1641 error_propagate(errp, local_err);
1642 } else if (bs->filename[0]) {
1643 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1645 error_setg_errno(errp, -ret, "Could not open image");
1650 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1651 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1654 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1655 * drivers that pass read/write requests through to a child the trouble of
1656 * declaring support explicitly.
1658 * Drivers must not propagate this flag accidentally when they initiate I/O
1659 * to a bounce buffer. That case should be rare though.
1661 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1662 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1664 /* Get the context after .bdrv_open, it can change the context */
1665 ctx = bdrv_get_aio_context(bs);
1666 aio_context_acquire(ctx);
1668 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
1670 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1671 aio_context_release(ctx);
1675 bdrv_graph_rdlock_main_loop();
1676 bdrv_refresh_limits(bs, NULL, &local_err);
1677 bdrv_graph_rdunlock_main_loop();
1678 aio_context_release(ctx);
1681 error_propagate(errp, local_err);
1685 assert(bdrv_opt_mem_align(bs) != 0);
1686 assert(bdrv_min_mem_align(bs) != 0);
1687 assert(is_power_of_2(bs->bl.request_alignment));
1689 for (i = 0; i < bs->quiesce_counter; i++) {
1690 if (drv->bdrv_drain_begin) {
1691 drv->bdrv_drain_begin(bs);
1698 if (bs->file != NULL) {
1699 bdrv_unref_child(bs, bs->file);
1708 * Create and open a block node.
1710 * @options is a QDict of options to pass to the block drivers, or NULL for an
1711 * empty set of options. The reference to the QDict belongs to the block layer
1712 * after the call (even on failure), so if the caller intends to reuse the
1713 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1715 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1716 const char *node_name,
1717 QDict *options, int flags,
1720 BlockDriverState *bs;
1723 GLOBAL_STATE_CODE();
1726 bs->open_flags = flags;
1727 bs->options = options ?: qdict_new();
1728 bs->explicit_options = qdict_clone_shallow(bs->options);
1731 update_options_from_flags(bs->options, flags);
1733 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1735 qobject_unref(bs->explicit_options);
1736 bs->explicit_options = NULL;
1737 qobject_unref(bs->options);
1746 /* Create and open a block node. */
1747 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1748 int flags, Error **errp)
1750 GLOBAL_STATE_CODE();
1751 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1754 QemuOptsList bdrv_runtime_opts = {
1755 .name = "bdrv_common",
1756 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1759 .name = "node-name",
1760 .type = QEMU_OPT_STRING,
1761 .help = "Node name of the block device node",
1765 .type = QEMU_OPT_STRING,
1766 .help = "Block driver to use for the node",
1769 .name = BDRV_OPT_CACHE_DIRECT,
1770 .type = QEMU_OPT_BOOL,
1771 .help = "Bypass software writeback cache on the host",
1774 .name = BDRV_OPT_CACHE_NO_FLUSH,
1775 .type = QEMU_OPT_BOOL,
1776 .help = "Ignore flush requests",
1779 .name = BDRV_OPT_READ_ONLY,
1780 .type = QEMU_OPT_BOOL,
1781 .help = "Node is opened in read-only mode",
1784 .name = BDRV_OPT_AUTO_READ_ONLY,
1785 .type = QEMU_OPT_BOOL,
1786 .help = "Node can become read-only if opening read-write fails",
1789 .name = "detect-zeroes",
1790 .type = QEMU_OPT_STRING,
1791 .help = "try to optimize zero writes (off, on, unmap)",
1794 .name = BDRV_OPT_DISCARD,
1795 .type = QEMU_OPT_STRING,
1796 .help = "discard operation (ignore/off, unmap/on)",
1799 .name = BDRV_OPT_FORCE_SHARE,
1800 .type = QEMU_OPT_BOOL,
1801 .help = "always accept other writers (default: off)",
1803 { /* end of list */ }
1807 QemuOptsList bdrv_create_opts_simple = {
1808 .name = "simple-create-opts",
1809 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1812 .name = BLOCK_OPT_SIZE,
1813 .type = QEMU_OPT_SIZE,
1814 .help = "Virtual disk size"
1817 .name = BLOCK_OPT_PREALLOC,
1818 .type = QEMU_OPT_STRING,
1819 .help = "Preallocation mode (allowed values: off)"
1821 { /* end of list */ }
1826 * Common part for opening disk images and files
1828 * Removes all processed options from *options.
1830 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1831 QDict *options, Error **errp)
1833 int ret, open_flags;
1834 const char *filename;
1835 const char *driver_name = NULL;
1836 const char *node_name = NULL;
1837 const char *discard;
1840 Error *local_err = NULL;
1843 assert(bs->file == NULL);
1844 assert(options != NULL && bs->options != options);
1845 GLOBAL_STATE_CODE();
1847 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1848 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1853 update_flags_from_options(&bs->open_flags, opts);
1855 driver_name = qemu_opt_get(opts, "driver");
1856 drv = bdrv_find_format(driver_name);
1857 assert(drv != NULL);
1859 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1861 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1863 BDRV_OPT_FORCE_SHARE
1864 "=on can only be used with read-only images");
1870 bdrv_refresh_filename(blk_bs(file));
1871 filename = blk_bs(file)->filename;
1874 * Caution: while qdict_get_try_str() is fine, getting
1875 * non-string types would require more care. When @options
1876 * come from -blockdev or blockdev_add, its members are typed
1877 * according to the QAPI schema, but when they come from
1878 * -drive, they're all QString.
1880 filename = qdict_get_try_str(options, "filename");
1883 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1884 error_setg(errp, "The '%s' block driver requires a file name",
1890 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1893 ro = bdrv_is_read_only(bs);
1895 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1896 if (!ro && bdrv_is_whitelisted(drv, true)) {
1897 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1903 !ro && bdrv_is_whitelisted(drv, true)
1904 ? "Driver '%s' can only be used for read-only devices"
1905 : "Driver '%s' is not whitelisted",
1911 /* bdrv_new() and bdrv_close() make it so */
1912 assert(qatomic_read(&bs->copy_on_read) == 0);
1914 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1916 bdrv_enable_copy_on_read(bs);
1918 error_setg(errp, "Can't use copy-on-read on read-only device");
1924 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1925 if (discard != NULL) {
1926 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1927 error_setg(errp, "Invalid discard option");
1934 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1936 error_propagate(errp, local_err);
1941 if (filename != NULL) {
1942 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1944 bs->filename[0] = '\0';
1946 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1948 /* Open the image, either directly or using a protocol */
1949 open_flags = bdrv_open_flags(bs, bs->open_flags);
1950 node_name = qemu_opt_get(opts, "node-name");
1952 assert(!drv->bdrv_file_open || file == NULL);
1953 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1958 qemu_opts_del(opts);
1962 qemu_opts_del(opts);
1966 static QDict *parse_json_filename(const char *filename, Error **errp)
1968 QObject *options_obj;
1971 GLOBAL_STATE_CODE();
1973 ret = strstart(filename, "json:", &filename);
1976 options_obj = qobject_from_json(filename, errp);
1978 error_prepend(errp, "Could not parse the JSON options: ");
1982 options = qobject_to(QDict, options_obj);
1984 qobject_unref(options_obj);
1985 error_setg(errp, "Invalid JSON object given");
1989 qdict_flatten(options);
1994 static void parse_json_protocol(QDict *options, const char **pfilename,
1997 QDict *json_options;
1998 Error *local_err = NULL;
1999 GLOBAL_STATE_CODE();
2001 /* Parse json: pseudo-protocol */
2002 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2006 json_options = parse_json_filename(*pfilename, &local_err);
2008 error_propagate(errp, local_err);
2012 /* Options given in the filename have lower priority than options
2013 * specified directly */
2014 qdict_join(options, json_options, false);
2015 qobject_unref(json_options);
2020 * Fills in default options for opening images and converts the legacy
2021 * filename/flags pair to option QDict entries.
2022 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2023 * block driver has been specified explicitly.
2025 static int bdrv_fill_options(QDict **options, const char *filename,
2026 int *flags, Error **errp)
2028 const char *drvname;
2029 bool protocol = *flags & BDRV_O_PROTOCOL;
2030 bool parse_filename = false;
2031 BlockDriver *drv = NULL;
2032 Error *local_err = NULL;
2034 GLOBAL_STATE_CODE();
2037 * Caution: while qdict_get_try_str() is fine, getting non-string
2038 * types would require more care. When @options come from
2039 * -blockdev or blockdev_add, its members are typed according to
2040 * the QAPI schema, but when they come from -drive, they're all
2043 drvname = qdict_get_try_str(*options, "driver");
2045 drv = bdrv_find_format(drvname);
2047 error_setg(errp, "Unknown driver '%s'", drvname);
2050 /* If the user has explicitly specified the driver, this choice should
2051 * override the BDRV_O_PROTOCOL flag */
2052 protocol = drv->bdrv_file_open;
2056 *flags |= BDRV_O_PROTOCOL;
2058 *flags &= ~BDRV_O_PROTOCOL;
2061 /* Translate cache options from flags into options */
2062 update_options_from_flags(*options, *flags);
2064 /* Fetch the file name from the options QDict if necessary */
2065 if (protocol && filename) {
2066 if (!qdict_haskey(*options, "filename")) {
2067 qdict_put_str(*options, "filename", filename);
2068 parse_filename = true;
2070 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2076 /* Find the right block driver */
2077 /* See cautionary note on accessing @options above */
2078 filename = qdict_get_try_str(*options, "filename");
2080 if (!drvname && protocol) {
2082 drv = bdrv_find_protocol(filename, parse_filename, errp);
2087 drvname = drv->format_name;
2088 qdict_put_str(*options, "driver", drvname);
2090 error_setg(errp, "Must specify either driver or file");
2095 assert(drv || !protocol);
2097 /* Driver-specific filename parsing */
2098 if (drv && drv->bdrv_parse_filename && parse_filename) {
2099 drv->bdrv_parse_filename(filename, *options, &local_err);
2101 error_propagate(errp, local_err);
2105 if (!drv->bdrv_needs_filename) {
2106 qdict_del(*options, "filename");
2113 typedef struct BlockReopenQueueEntry {
2116 BDRVReopenState state;
2117 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2118 } BlockReopenQueueEntry;
2121 * Return the flags that @bs will have after the reopens in @q have
2122 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2123 * return the current flags.
2125 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2127 BlockReopenQueueEntry *entry;
2130 QTAILQ_FOREACH(entry, q, entry) {
2131 if (entry->state.bs == bs) {
2132 return entry->state.flags;
2137 return bs->open_flags;
2140 /* Returns whether the image file can be written to after the reopen queue @q
2141 * has been successfully applied, or right now if @q is NULL. */
2142 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2143 BlockReopenQueue *q)
2145 int flags = bdrv_reopen_get_flags(q, bs);
2147 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2151 * Return whether the BDS can be written to. This is not necessarily
2152 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2153 * be written to but do not count as read-only images.
2155 bool bdrv_is_writable(BlockDriverState *bs)
2158 return bdrv_is_writable_after_reopen(bs, NULL);
2161 static char *bdrv_child_user_desc(BdrvChild *c)
2163 GLOBAL_STATE_CODE();
2164 return c->klass->get_parent_desc(c);
2168 * Check that @a allows everything that @b needs. @a and @b must reference same
2171 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2173 const char *child_bs_name;
2174 g_autofree char *a_user = NULL;
2175 g_autofree char *b_user = NULL;
2176 g_autofree char *perms = NULL;
2179 assert(a->bs == b->bs);
2180 GLOBAL_STATE_CODE();
2182 if ((b->perm & a->shared_perm) == b->perm) {
2186 child_bs_name = bdrv_get_node_name(b->bs);
2187 a_user = bdrv_child_user_desc(a);
2188 b_user = bdrv_child_user_desc(b);
2189 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2191 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2192 "both required by %s (uses node '%s' as '%s' child) and "
2193 "unshared by %s (uses node '%s' as '%s' child).",
2194 child_bs_name, perms,
2195 b_user, child_bs_name, b->name,
2196 a_user, child_bs_name, a->name);
2201 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2204 GLOBAL_STATE_CODE();
2207 * During the loop we'll look at each pair twice. That's correct because
2208 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2211 QLIST_FOREACH(a, &bs->parents, next_parent) {
2212 QLIST_FOREACH(b, &bs->parents, next_parent) {
2217 if (!bdrv_a_allow_b(a, b, errp)) {
2226 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2227 BdrvChild *c, BdrvChildRole role,
2228 BlockReopenQueue *reopen_queue,
2229 uint64_t parent_perm, uint64_t parent_shared,
2230 uint64_t *nperm, uint64_t *nshared)
2232 assert(bs->drv && bs->drv->bdrv_child_perm);
2233 GLOBAL_STATE_CODE();
2234 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2235 parent_perm, parent_shared,
2237 /* TODO Take force_share from reopen_queue */
2238 if (child_bs && child_bs->force_share) {
2239 *nshared = BLK_PERM_ALL;
2244 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2245 * nodes that are already in the @list, of course) so that final list is
2246 * topologically sorted. Return the result (GSList @list object is updated, so
2247 * don't use old reference after function call).
2249 * On function start @list must be already topologically sorted and for any node
2250 * in the @list the whole subtree of the node must be in the @list as well. The
2251 * simplest way to satisfy this criteria: use only result of
2252 * bdrv_topological_dfs() or NULL as @list parameter.
2254 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2255 BlockDriverState *bs)
2258 g_autoptr(GHashTable) local_found = NULL;
2260 GLOBAL_STATE_CODE();
2264 found = local_found = g_hash_table_new(NULL, NULL);
2267 if (g_hash_table_contains(found, bs)) {
2270 g_hash_table_add(found, bs);
2272 QLIST_FOREACH(child, &bs->children, next) {
2273 list = bdrv_topological_dfs(list, found, child->bs);
2276 return g_slist_prepend(list, bs);
2279 typedef struct BdrvChildSetPermState {
2282 uint64_t old_shared_perm;
2283 } BdrvChildSetPermState;
2285 static void bdrv_child_set_perm_abort(void *opaque)
2287 BdrvChildSetPermState *s = opaque;
2289 GLOBAL_STATE_CODE();
2291 s->child->perm = s->old_perm;
2292 s->child->shared_perm = s->old_shared_perm;
2295 static TransactionActionDrv bdrv_child_set_pem_drv = {
2296 .abort = bdrv_child_set_perm_abort,
2300 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2301 uint64_t shared, Transaction *tran)
2303 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2304 GLOBAL_STATE_CODE();
2306 *s = (BdrvChildSetPermState) {
2308 .old_perm = c->perm,
2309 .old_shared_perm = c->shared_perm,
2313 c->shared_perm = shared;
2315 tran_add(tran, &bdrv_child_set_pem_drv, s);
2318 static void bdrv_drv_set_perm_commit(void *opaque)
2320 BlockDriverState *bs = opaque;
2321 uint64_t cumulative_perms, cumulative_shared_perms;
2322 GLOBAL_STATE_CODE();
2324 if (bs->drv->bdrv_set_perm) {
2325 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2326 &cumulative_shared_perms);
2327 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2331 static void bdrv_drv_set_perm_abort(void *opaque)
2333 BlockDriverState *bs = opaque;
2334 GLOBAL_STATE_CODE();
2336 if (bs->drv->bdrv_abort_perm_update) {
2337 bs->drv->bdrv_abort_perm_update(bs);
2341 TransactionActionDrv bdrv_drv_set_perm_drv = {
2342 .abort = bdrv_drv_set_perm_abort,
2343 .commit = bdrv_drv_set_perm_commit,
2346 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2347 uint64_t shared_perm, Transaction *tran,
2350 GLOBAL_STATE_CODE();
2355 if (bs->drv->bdrv_check_perm) {
2356 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2363 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2369 typedef struct BdrvReplaceChildState {
2371 BlockDriverState *old_bs;
2372 } BdrvReplaceChildState;
2374 static void bdrv_replace_child_commit(void *opaque)
2376 BdrvReplaceChildState *s = opaque;
2377 GLOBAL_STATE_CODE();
2379 bdrv_unref(s->old_bs);
2382 static void bdrv_replace_child_abort(void *opaque)
2384 BdrvReplaceChildState *s = opaque;
2385 BlockDriverState *new_bs = s->child->bs;
2387 GLOBAL_STATE_CODE();
2388 /* old_bs reference is transparently moved from @s to @s->child */
2389 if (!s->child->bs) {
2391 * The parents were undrained when removing old_bs from the child. New
2392 * requests can't have been made, though, because the child was empty.
2394 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2395 * undraining the parent in the first place. Once this is done, having
2396 * new_bs drained when calling bdrv_replace_child_tran() is not a
2397 * requirement any more.
2399 bdrv_parent_drained_begin_single(s->child);
2400 assert(!bdrv_parent_drained_poll_single(s->child));
2402 assert(s->child->quiesced_parent);
2403 bdrv_replace_child_noperm(s->child, s->old_bs);
2407 static TransactionActionDrv bdrv_replace_child_drv = {
2408 .commit = bdrv_replace_child_commit,
2409 .abort = bdrv_replace_child_abort,
2414 * bdrv_replace_child_tran
2416 * Note: real unref of old_bs is done only on commit.
2418 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2419 * kept drained until the transaction is completed.
2421 * The function doesn't update permissions, caller is responsible for this.
2423 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2426 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2428 assert(child->quiesced_parent);
2429 assert(!new_bs || new_bs->quiesce_counter);
2431 *s = (BdrvReplaceChildState) {
2433 .old_bs = child->bs,
2435 tran_add(tran, &bdrv_replace_child_drv, s);
2440 bdrv_replace_child_noperm(child, new_bs);
2441 /* old_bs reference is transparently moved from @child to @s */
2445 * Refresh permissions in @bs subtree. The function is intended to be called
2446 * after some graph modification that was done without permission update.
2448 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2449 Transaction *tran, Error **errp)
2451 BlockDriver *drv = bs->drv;
2454 uint64_t cumulative_perms, cumulative_shared_perms;
2455 GLOBAL_STATE_CODE();
2457 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2459 /* Write permissions never work with read-only images */
2460 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2461 !bdrv_is_writable_after_reopen(bs, q))
2463 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2464 error_setg(errp, "Block node is read-only");
2466 error_setg(errp, "Read-only block node '%s' cannot support "
2467 "read-write users", bdrv_get_node_name(bs));
2474 * Unaligned requests will automatically be aligned to bl.request_alignment
2475 * and without RESIZE we can't extend requests to write to space beyond the
2476 * end of the image, so it's required that the image size is aligned.
2478 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2479 !(cumulative_perms & BLK_PERM_RESIZE))
2481 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2482 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2483 "Image size is not a multiple of request "
2489 /* Check this node */
2494 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2500 /* Drivers that never have children can omit .bdrv_child_perm() */
2501 if (!drv->bdrv_child_perm) {
2502 assert(QLIST_EMPTY(&bs->children));
2506 /* Check all children */
2507 QLIST_FOREACH(c, &bs->children, next) {
2508 uint64_t cur_perm, cur_shared;
2510 bdrv_child_perm(bs, c->bs, c, c->role, q,
2511 cumulative_perms, cumulative_shared_perms,
2512 &cur_perm, &cur_shared);
2513 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2520 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2521 * a topologically sorted subgraph.
2523 static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2524 Transaction *tran, Error **errp)
2527 BlockDriverState *bs;
2528 GLOBAL_STATE_CODE();
2530 for ( ; list; list = list->next) {
2533 if (bdrv_parent_perms_conflict(bs, errp)) {
2537 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2547 * @list is any list of nodes. List is completed by all subtrees and
2548 * topologically sorted. It's not a problem if some node occurs in the @list
2551 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2552 Transaction *tran, Error **errp)
2554 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2555 g_autoptr(GSList) refresh_list = NULL;
2557 for ( ; list; list = list->next) {
2558 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2561 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2564 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2565 uint64_t *shared_perm)
2568 uint64_t cumulative_perms = 0;
2569 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2571 GLOBAL_STATE_CODE();
2573 QLIST_FOREACH(c, &bs->parents, next_parent) {
2574 cumulative_perms |= c->perm;
2575 cumulative_shared_perms &= c->shared_perm;
2578 *perm = cumulative_perms;
2579 *shared_perm = cumulative_shared_perms;
2582 char *bdrv_perm_names(uint64_t perm)
2588 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2589 { BLK_PERM_WRITE, "write" },
2590 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2591 { BLK_PERM_RESIZE, "resize" },
2595 GString *result = g_string_sized_new(30);
2596 struct perm_name *p;
2598 for (p = permissions; p->name; p++) {
2599 if (perm & p->perm) {
2600 if (result->len > 0) {
2601 g_string_append(result, ", ");
2603 g_string_append(result, p->name);
2607 return g_string_free(result, FALSE);
2611 /* @tran is allowed to be NULL. In this case no rollback is possible */
2612 static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2616 Transaction *local_tran = NULL;
2617 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2618 GLOBAL_STATE_CODE();
2621 tran = local_tran = tran_new();
2624 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2627 tran_finalize(local_tran, ret);
2633 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2636 Error *local_err = NULL;
2637 Transaction *tran = tran_new();
2640 GLOBAL_STATE_CODE();
2642 bdrv_child_set_perm(c, perm, shared, tran);
2644 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2646 tran_finalize(tran, ret);
2649 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2650 /* tighten permissions */
2651 error_propagate(errp, local_err);
2654 * Our caller may intend to only loosen restrictions and
2655 * does not expect this function to fail. Errors are not
2656 * fatal in such a case, so we can just hide them from our
2659 error_free(local_err);
2667 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2669 uint64_t parent_perms, parent_shared;
2670 uint64_t perms, shared;
2672 GLOBAL_STATE_CODE();
2674 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2675 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2676 parent_perms, parent_shared, &perms, &shared);
2678 return bdrv_child_try_set_perm(c, perms, shared, errp);
2682 * Default implementation for .bdrv_child_perm() for block filters:
2683 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2686 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2688 BlockReopenQueue *reopen_queue,
2689 uint64_t perm, uint64_t shared,
2690 uint64_t *nperm, uint64_t *nshared)
2692 GLOBAL_STATE_CODE();
2693 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2694 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2697 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2699 BlockReopenQueue *reopen_queue,
2700 uint64_t perm, uint64_t shared,
2701 uint64_t *nperm, uint64_t *nshared)
2703 assert(role & BDRV_CHILD_COW);
2704 GLOBAL_STATE_CODE();
2707 * We want consistent read from backing files if the parent needs it.
2708 * No other operations are performed on backing files.
2710 perm &= BLK_PERM_CONSISTENT_READ;
2713 * If the parent can deal with changing data, we're okay with a
2714 * writable and resizable backing file.
2715 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2717 if (shared & BLK_PERM_WRITE) {
2718 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2723 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2725 if (bs->open_flags & BDRV_O_INACTIVE) {
2726 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2733 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2735 BlockReopenQueue *reopen_queue,
2736 uint64_t perm, uint64_t shared,
2737 uint64_t *nperm, uint64_t *nshared)
2741 GLOBAL_STATE_CODE();
2742 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2744 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2747 * Apart from the modifications below, the same permissions are
2748 * forwarded and left alone as for filters
2750 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2751 perm, shared, &perm, &shared);
2753 if (role & BDRV_CHILD_METADATA) {
2754 /* Format drivers may touch metadata even if the guest doesn't write */
2755 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2756 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2760 * bs->file always needs to be consistent because of the
2761 * metadata. We can never allow other users to resize or write
2764 if (!(flags & BDRV_O_NO_IO)) {
2765 perm |= BLK_PERM_CONSISTENT_READ;
2767 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2770 if (role & BDRV_CHILD_DATA) {
2772 * Technically, everything in this block is a subset of the
2773 * BDRV_CHILD_METADATA path taken above, and so this could
2774 * be an "else if" branch. However, that is not obvious, and
2775 * this function is not performance critical, therefore we let
2776 * this be an independent "if".
2780 * We cannot allow other users to resize the file because the
2781 * format driver might have some assumptions about the size
2782 * (e.g. because it is stored in metadata, or because the file
2783 * is split into fixed-size data files).
2785 shared &= ~BLK_PERM_RESIZE;
2788 * WRITE_UNCHANGED often cannot be performed as such on the
2789 * data file. For example, the qcow2 driver may still need to
2790 * write copied clusters on copy-on-read.
2792 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2793 perm |= BLK_PERM_WRITE;
2797 * If the data file is written to, the format driver may
2798 * expect to be able to resize it by writing beyond the EOF.
2800 if (perm & BLK_PERM_WRITE) {
2801 perm |= BLK_PERM_RESIZE;
2805 if (bs->open_flags & BDRV_O_INACTIVE) {
2806 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2813 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2814 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2815 uint64_t perm, uint64_t shared,
2816 uint64_t *nperm, uint64_t *nshared)
2818 GLOBAL_STATE_CODE();
2819 if (role & BDRV_CHILD_FILTERED) {
2820 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2822 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2823 perm, shared, nperm, nshared);
2824 } else if (role & BDRV_CHILD_COW) {
2825 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2826 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2827 perm, shared, nperm, nshared);
2828 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2829 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2830 perm, shared, nperm, nshared);
2832 g_assert_not_reached();
2836 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2838 static const uint64_t permissions[] = {
2839 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2840 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2841 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2842 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2845 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2846 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2848 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2850 return permissions[qapi_perm];
2854 * Replaces the node that a BdrvChild points to without updating permissions.
2856 * If @new_bs is non-NULL, the parent of @child must already be drained through
2859 static void bdrv_replace_child_noperm(BdrvChild *child,
2860 BlockDriverState *new_bs)
2862 BlockDriverState *old_bs = child->bs;
2863 int new_bs_quiesce_counter;
2865 assert(!child->frozen);
2868 * If we want to change the BdrvChild to point to a drained node as its new
2869 * child->bs, we need to make sure that its new parent is drained, too. In
2870 * other words, either child->quiesce_parent must already be true or we must
2871 * be able to set it and keep the parent's quiesce_counter consistent with
2872 * that, but without polling or starting new requests (this function
2873 * guarantees that it doesn't poll, and starting new requests would be
2874 * against the invariants of drain sections).
2876 * To keep things simple, we pick the first option (child->quiesce_parent
2877 * must already be true). We also generalise the rule a bit to make it
2878 * easier to verify in callers and more likely to be covered in test cases:
2879 * The parent must be quiesced through this child even if new_bs isn't
2880 * currently drained.
2882 * The only exception is for callers that always pass new_bs == NULL. In
2883 * this case, we obviously never need to consider the case of a drained
2884 * new_bs, so we can keep the callers simpler by allowing them not to drain
2887 assert(!new_bs || child->quiesced_parent);
2888 assert(old_bs != new_bs);
2889 GLOBAL_STATE_CODE();
2891 if (old_bs && new_bs) {
2892 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2895 /* TODO Pull this up into the callers to avoid polling here */
2896 bdrv_graph_wrlock();
2898 if (child->klass->detach) {
2899 child->klass->detach(child);
2901 QLIST_REMOVE(child, next_parent);
2907 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2908 if (child->klass->attach) {
2909 child->klass->attach(child);
2912 bdrv_graph_wrunlock();
2915 * If the parent was drained through this BdrvChild previously, but new_bs
2916 * is not drained, allow requests to come in only after the new node has
2919 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2920 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2921 bdrv_parent_drained_end_single(child);
2926 * Free the given @child.
2928 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2929 * unused (i.e. not in a children list).
2931 static void bdrv_child_free(BdrvChild *child)
2934 GLOBAL_STATE_CODE();
2935 assert(!child->next.le_prev); /* not in children list */
2937 g_free(child->name);
2941 typedef struct BdrvAttachChildCommonState {
2943 AioContext *old_parent_ctx;
2944 AioContext *old_child_ctx;
2945 } BdrvAttachChildCommonState;
2947 static void bdrv_attach_child_common_abort(void *opaque)
2949 BdrvAttachChildCommonState *s = opaque;
2950 BlockDriverState *bs = s->child->bs;
2952 GLOBAL_STATE_CODE();
2953 bdrv_replace_child_noperm(s->child, NULL);
2955 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2956 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2959 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2961 GHashTable *visited;
2966 /* No need to visit `child`, because it has been detached already */
2967 visited = g_hash_table_new(NULL, NULL);
2968 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2969 visited, tran, &error_abort);
2970 g_hash_table_destroy(visited);
2972 /* transaction is supposed to always succeed */
2973 assert(ret == true);
2978 bdrv_child_free(s->child);
2981 static TransactionActionDrv bdrv_attach_child_common_drv = {
2982 .abort = bdrv_attach_child_common_abort,
2987 * Common part of attaching bdrv child to bs or to blk or to job
2989 * Function doesn't update permissions, caller is responsible for this.
2991 * Returns new created child.
2993 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
2994 * @child_bs can move to a different AioContext in this function. Callers must
2995 * make sure that their AioContext locking is still correct after this.
2997 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2998 const char *child_name,
2999 const BdrvChildClass *child_class,
3000 BdrvChildRole child_role,
3001 uint64_t perm, uint64_t shared_perm,
3003 Transaction *tran, Error **errp)
3005 BdrvChild *new_child;
3006 AioContext *parent_ctx, *new_child_ctx;
3007 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3009 assert(child_class->get_parent_desc);
3010 GLOBAL_STATE_CODE();
3012 new_child = g_new(BdrvChild, 1);
3013 *new_child = (BdrvChild) {
3015 .name = g_strdup(child_name),
3016 .klass = child_class,
3019 .shared_perm = shared_perm,
3024 * If the AioContexts don't match, first try to move the subtree of
3025 * child_bs into the AioContext of the new parent. If this doesn't work,
3026 * try moving the parent into the AioContext of child_bs instead.
3028 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3029 if (child_ctx != parent_ctx) {
3030 Error *local_err = NULL;
3031 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3034 if (ret < 0 && child_class->change_aio_ctx) {
3035 Transaction *tran = tran_new();
3036 GHashTable *visited = g_hash_table_new(NULL, NULL);
3039 g_hash_table_add(visited, new_child);
3040 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3041 visited, tran, NULL);
3042 if (ret_child == true) {
3043 error_free(local_err);
3046 tran_finalize(tran, ret_child == true ? 0 : -1);
3047 g_hash_table_destroy(visited);
3051 error_propagate(errp, local_err);
3052 bdrv_child_free(new_child);
3057 new_child_ctx = bdrv_get_aio_context(child_bs);
3058 if (new_child_ctx != child_ctx) {
3059 aio_context_release(child_ctx);
3060 aio_context_acquire(new_child_ctx);
3065 * Let every new BdrvChild start with a drained parent. Inserting the child
3066 * in the graph with bdrv_replace_child_noperm() will undrain it if
3067 * @child_bs is not drained.
3069 * The child was only just created and is not yet visible in global state
3070 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3071 * could have sent requests and polling is not necessary.
3073 * Note that this means that the parent isn't fully drained yet, we only
3074 * stop new requests from coming in. This is fine, we don't care about the
3075 * old requests here, they are not for this child. If another place enters a
3076 * drain section for the same parent, but wants it to be fully quiesced, it
3077 * will not run most of the the code in .drained_begin() again (which is not
3078 * a problem, we already did this), but it will still poll until the parent
3079 * is fully quiesced, so it will not be negatively affected either.
3081 bdrv_parent_drained_begin_single(new_child);
3082 bdrv_replace_child_noperm(new_child, child_bs);
3084 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3085 *s = (BdrvAttachChildCommonState) {
3087 .old_parent_ctx = parent_ctx,
3088 .old_child_ctx = child_ctx,
3090 tran_add(tran, &bdrv_attach_child_common_drv, s);
3092 if (new_child_ctx != child_ctx) {
3093 aio_context_release(new_child_ctx);
3094 aio_context_acquire(child_ctx);
3101 * Function doesn't update permissions, caller is responsible for this.
3103 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3104 * @child_bs can move to a different AioContext in this function. Callers must
3105 * make sure that their AioContext locking is still correct after this.
3107 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3108 BlockDriverState *child_bs,
3109 const char *child_name,
3110 const BdrvChildClass *child_class,
3111 BdrvChildRole child_role,
3115 uint64_t perm, shared_perm;
3117 assert(parent_bs->drv);
3118 GLOBAL_STATE_CODE();
3120 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3121 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3122 child_bs->node_name, child_name, parent_bs->node_name);
3126 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3127 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3128 perm, shared_perm, &perm, &shared_perm);
3130 return bdrv_attach_child_common(child_bs, child_name, child_class,
3131 child_role, perm, shared_perm, parent_bs,
3136 * This function steals the reference to child_bs from the caller.
3137 * That reference is later dropped by bdrv_root_unref_child().
3139 * On failure NULL is returned, errp is set and the reference to
3140 * child_bs is also dropped.
3142 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3143 * (unless @child_bs is already in @ctx).
3145 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3146 const char *child_name,
3147 const BdrvChildClass *child_class,
3148 BdrvChildRole child_role,
3149 uint64_t perm, uint64_t shared_perm,
3150 void *opaque, Error **errp)
3154 Transaction *tran = tran_new();
3156 GLOBAL_STATE_CODE();
3158 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3159 child_role, perm, shared_perm, opaque,
3166 ret = bdrv_refresh_perms(child_bs, tran, errp);
3169 tran_finalize(tran, ret);
3171 bdrv_unref(child_bs);
3173 return ret < 0 ? NULL : child;
3177 * This function transfers the reference to child_bs from the caller
3178 * to parent_bs. That reference is later dropped by parent_bs on
3179 * bdrv_close() or if someone calls bdrv_unref_child().
3181 * On failure NULL is returned, errp is set and the reference to
3182 * child_bs is also dropped.
3184 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3185 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3187 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3188 BlockDriverState *child_bs,
3189 const char *child_name,
3190 const BdrvChildClass *child_class,
3191 BdrvChildRole child_role,
3196 Transaction *tran = tran_new();
3198 GLOBAL_STATE_CODE();
3200 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3201 child_class, child_role, tran, errp);
3207 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3213 tran_finalize(tran, ret);
3215 bdrv_unref(child_bs);
3217 return ret < 0 ? NULL : child;
3220 /* Callers must ensure that child->frozen is false. */
3221 void bdrv_root_unref_child(BdrvChild *child)
3223 BlockDriverState *child_bs = child->bs;
3225 GLOBAL_STATE_CODE();
3226 bdrv_replace_child_noperm(child, NULL);
3227 bdrv_child_free(child);
3231 * Update permissions for old node. We're just taking a parent away, so
3232 * we're loosening restrictions. Errors of permission update are not
3233 * fatal in this case, ignore them.
3235 bdrv_refresh_perms(child_bs, NULL, NULL);
3238 * When the parent requiring a non-default AioContext is removed, the
3239 * node moves back to the main AioContext
3241 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3245 bdrv_unref(child_bs);
3248 typedef struct BdrvSetInheritsFrom {
3249 BlockDriverState *bs;
3250 BlockDriverState *old_inherits_from;
3251 } BdrvSetInheritsFrom;
3253 static void bdrv_set_inherits_from_abort(void *opaque)
3255 BdrvSetInheritsFrom *s = opaque;
3257 s->bs->inherits_from = s->old_inherits_from;
3260 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3261 .abort = bdrv_set_inherits_from_abort,
3265 /* @tran is allowed to be NULL. In this case no rollback is possible */
3266 static void bdrv_set_inherits_from(BlockDriverState *bs,
3267 BlockDriverState *new_inherits_from,
3271 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3273 *s = (BdrvSetInheritsFrom) {
3275 .old_inherits_from = bs->inherits_from,
3278 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3281 bs->inherits_from = new_inherits_from;
3285 * Clear all inherits_from pointers from children and grandchildren of
3286 * @root that point to @root, where necessary.
3287 * @tran is allowed to be NULL. In this case no rollback is possible
3289 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3294 if (child->bs->inherits_from == root) {
3296 * Remove inherits_from only when the last reference between root and
3297 * child->bs goes away.
3299 QLIST_FOREACH(c, &root->children, next) {
3300 if (c != child && c->bs == child->bs) {
3305 bdrv_set_inherits_from(child->bs, NULL, tran);
3309 QLIST_FOREACH(c, &child->bs->children, next) {
3310 bdrv_unset_inherits_from(root, c, tran);
3314 /* Callers must ensure that child->frozen is false. */
3315 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3317 GLOBAL_STATE_CODE();
3318 if (child == NULL) {
3322 bdrv_unset_inherits_from(parent, child, NULL);
3323 bdrv_root_unref_child(child);
3327 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3330 GLOBAL_STATE_CODE();
3331 QLIST_FOREACH(c, &bs->parents, next_parent) {
3332 if (c->klass->change_media) {
3333 c->klass->change_media(c, load);
3338 /* Return true if you can reach parent going through child->inherits_from
3339 * recursively. If parent or child are NULL, return false */
3340 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3341 BlockDriverState *parent)
3343 while (child && child != parent) {
3344 child = child->inherits_from;
3347 return child != NULL;
3351 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3352 * mostly used for COW backing children (role = COW), but also for
3353 * filtered children (role = FILTERED | PRIMARY).
3355 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3357 if (bs->drv && bs->drv->is_filter) {
3358 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3360 return BDRV_CHILD_COW;
3365 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3366 * callers which don't need their own reference any more must call bdrv_unref().
3368 * Function doesn't update permissions, caller is responsible for this.
3370 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3371 * @child_bs can move to a different AioContext in this function. Callers must
3372 * make sure that their AioContext locking is still correct after this.
3374 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3375 BlockDriverState *child_bs,
3377 Transaction *tran, Error **errp)
3379 bool update_inherits_from =
3380 bdrv_inherits_from_recursive(child_bs, parent_bs);
3381 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3384 GLOBAL_STATE_CODE();
3386 if (!parent_bs->drv) {
3388 * Node without drv is an object without a class :/. TODO: finally fix
3389 * qcow2 driver to never clear bs->drv and implement format corruption
3390 * handling in other way.
3392 error_setg(errp, "Node corrupted");
3396 if (child && child->frozen) {
3397 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3398 child->name, parent_bs->node_name, child->bs->node_name);
3402 if (is_backing && !parent_bs->drv->is_filter &&
3403 !parent_bs->drv->supports_backing)
3405 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3406 "files", parent_bs->drv->format_name, parent_bs->node_name);
3410 if (parent_bs->drv->is_filter) {
3411 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3412 } else if (is_backing) {
3413 role = BDRV_CHILD_COW;
3416 * We only can use same role as it is in existing child. We don't have
3417 * infrastructure to determine role of file child in generic way
3420 error_setg(errp, "Cannot set file child to format node without "
3428 bdrv_unset_inherits_from(parent_bs, child, tran);
3429 bdrv_remove_child(child, tran);
3436 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3437 is_backing ? "backing" : "file",
3438 &child_of_bds, role,
3446 * If inherits_from pointed recursively to bs then let's update it to
3447 * point directly to bs (else it will become NULL).
3449 if (update_inherits_from) {
3450 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3454 bdrv_graph_rdlock_main_loop();
3455 bdrv_refresh_limits(parent_bs, tran, NULL);
3456 bdrv_graph_rdunlock_main_loop();
3462 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3463 * @backing_hd can move to a different AioContext in this function. Callers must
3464 * make sure that their AioContext locking is still correct after this.
3466 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3467 BlockDriverState *backing_hd,
3468 Transaction *tran, Error **errp)
3470 GLOBAL_STATE_CODE();
3471 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3474 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3475 BlockDriverState *backing_hd,
3479 Transaction *tran = tran_new();
3481 GLOBAL_STATE_CODE();
3482 assert(bs->quiesce_counter > 0);
3484 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3489 ret = bdrv_refresh_perms(bs, tran, errp);
3491 tran_finalize(tran, ret);
3495 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3499 GLOBAL_STATE_CODE();
3501 bdrv_drained_begin(bs);
3502 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3503 bdrv_drained_end(bs);
3509 * Opens the backing file for a BlockDriverState if not yet open
3511 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3512 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3513 * itself, all options starting with "${bdref_key}." are considered part of the
3516 * The caller must hold the main AioContext lock.
3518 * TODO Can this be unified with bdrv_open_image()?
3520 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3521 const char *bdref_key, Error **errp)
3523 char *backing_filename = NULL;
3524 char *bdref_key_dot;
3525 const char *reference = NULL;
3527 bool implicit_backing = false;
3528 BlockDriverState *backing_hd;
3530 QDict *tmp_parent_options = NULL;
3531 Error *local_err = NULL;
3533 GLOBAL_STATE_CODE();
3535 if (bs->backing != NULL) {
3539 /* NULL means an empty set of options */
3540 if (parent_options == NULL) {
3541 tmp_parent_options = qdict_new();
3542 parent_options = tmp_parent_options;
3545 bs->open_flags &= ~BDRV_O_NO_BACKING;
3547 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3548 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3549 g_free(bdref_key_dot);
3552 * Caution: while qdict_get_try_str() is fine, getting non-string
3553 * types would require more care. When @parent_options come from
3554 * -blockdev or blockdev_add, its members are typed according to
3555 * the QAPI schema, but when they come from -drive, they're all
3558 reference = qdict_get_try_str(parent_options, bdref_key);
3559 if (reference || qdict_haskey(options, "file.filename")) {
3560 /* keep backing_filename NULL */
3561 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3562 qobject_unref(options);
3565 if (qdict_size(options) == 0) {
3566 /* If the user specifies options that do not modify the
3567 * backing file's behavior, we might still consider it the
3568 * implicit backing file. But it's easier this way, and
3569 * just specifying some of the backing BDS's options is
3570 * only possible with -drive anyway (otherwise the QAPI
3571 * schema forces the user to specify everything). */
3572 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3575 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3578 error_propagate(errp, local_err);
3579 qobject_unref(options);
3584 if (!bs->drv || !bs->drv->supports_backing) {
3586 error_setg(errp, "Driver doesn't support backing files");
3587 qobject_unref(options);
3592 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3593 qdict_put_str(options, "driver", bs->backing_format);
3596 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3597 &child_of_bds, bdrv_backing_role(bs), errp);
3599 bs->open_flags |= BDRV_O_NO_BACKING;
3600 error_prepend(errp, "Could not open backing file: ");
3605 if (implicit_backing) {
3606 bdrv_refresh_filename(backing_hd);
3607 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3608 backing_hd->filename);
3611 /* Hook up the backing file link; drop our reference, bs owns the
3612 * backing_hd reference now */
3613 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3614 bdrv_unref(backing_hd);
3619 qdict_del(parent_options, bdref_key);
3622 g_free(backing_filename);
3623 qobject_unref(tmp_parent_options);
3627 static BlockDriverState *
3628 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3629 BlockDriverState *parent, const BdrvChildClass *child_class,
3630 BdrvChildRole child_role, bool allow_none, Error **errp)
3632 BlockDriverState *bs = NULL;
3633 QDict *image_options;
3634 char *bdref_key_dot;
3635 const char *reference;
3637 assert(child_class != NULL);
3639 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3640 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3641 g_free(bdref_key_dot);
3644 * Caution: while qdict_get_try_str() is fine, getting non-string
3645 * types would require more care. When @options come from
3646 * -blockdev or blockdev_add, its members are typed according to
3647 * the QAPI schema, but when they come from -drive, they're all
3650 reference = qdict_get_try_str(options, bdref_key);
3651 if (!filename && !reference && !qdict_size(image_options)) {
3653 error_setg(errp, "A block device must be specified for \"%s\"",
3656 qobject_unref(image_options);
3660 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3661 parent, child_class, child_role, errp);
3667 qdict_del(options, bdref_key);
3672 * Opens a disk image whose options are given as BlockdevRef in another block
3675 * If allow_none is true, no image will be opened if filename is false and no
3676 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3678 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3679 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3680 * itself, all options starting with "${bdref_key}." are considered part of the
3683 * The BlockdevRef will be removed from the options QDict.
3685 * The caller must hold the lock of the main AioContext and no other AioContext.
3686 * @parent can move to a different AioContext in this function. Callers must
3687 * make sure that their AioContext locking is still correct after this.
3689 BdrvChild *bdrv_open_child(const char *filename,
3690 QDict *options, const char *bdref_key,
3691 BlockDriverState *parent,
3692 const BdrvChildClass *child_class,
3693 BdrvChildRole child_role,
3694 bool allow_none, Error **errp)
3696 BlockDriverState *bs;
3700 GLOBAL_STATE_CODE();
3702 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3703 child_role, allow_none, errp);
3708 ctx = bdrv_get_aio_context(bs);
3709 aio_context_acquire(ctx);
3710 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3712 aio_context_release(ctx);
3718 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3720 * The caller must hold the lock of the main AioContext and no other AioContext.
3721 * @parent can move to a different AioContext in this function. Callers must
3722 * make sure that their AioContext locking is still correct after this.
3724 int bdrv_open_file_child(const char *filename,
3725 QDict *options, const char *bdref_key,
3726 BlockDriverState *parent, Error **errp)
3730 /* commit_top and mirror_top don't use this function */
3731 assert(!parent->drv->filtered_child_is_backing);
3732 role = parent->drv->is_filter ?
3733 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3735 if (!bdrv_open_child(filename, options, bdref_key, parent,
3736 &child_of_bds, role, false, errp))
3745 * TODO Future callers may need to specify parent/child_class in order for
3746 * option inheritance to work. Existing callers use it for the root node.
3748 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3750 BlockDriverState *bs = NULL;
3751 QObject *obj = NULL;
3752 QDict *qdict = NULL;
3753 const char *reference = NULL;
3756 GLOBAL_STATE_CODE();
3758 if (ref->type == QTYPE_QSTRING) {
3759 reference = ref->u.reference;
3761 BlockdevOptions *options = &ref->u.definition;
3762 assert(ref->type == QTYPE_QDICT);
3764 v = qobject_output_visitor_new(&obj);
3765 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3766 visit_complete(v, &obj);
3768 qdict = qobject_to(QDict, obj);
3769 qdict_flatten(qdict);
3771 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3772 * compatibility with other callers) rather than what we want as the
3773 * real defaults. Apply the defaults here instead. */
3774 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3775 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3776 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3777 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3781 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3788 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3790 QDict *snapshot_options,
3793 g_autofree char *tmp_filename = NULL;
3795 QemuOpts *opts = NULL;
3796 BlockDriverState *bs_snapshot = NULL;
3797 AioContext *ctx = bdrv_get_aio_context(bs);
3800 GLOBAL_STATE_CODE();
3802 /* if snapshot, we create a temporary backing file and open it
3803 instead of opening 'filename' directly */
3805 /* Get the required size from the image */
3806 aio_context_acquire(ctx);
3807 total_size = bdrv_getlength(bs);
3808 aio_context_release(ctx);
3810 if (total_size < 0) {
3811 error_setg_errno(errp, -total_size, "Could not get image size");
3815 /* Create the temporary image */
3816 tmp_filename = create_tmp_file(errp);
3817 if (!tmp_filename) {
3821 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3823 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3824 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3825 qemu_opts_del(opts);
3827 error_prepend(errp, "Could not create temporary overlay '%s': ",
3832 /* Prepare options QDict for the temporary file */
3833 qdict_put_str(snapshot_options, "file.driver", "file");
3834 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3835 qdict_put_str(snapshot_options, "driver", "qcow2");
3837 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3838 snapshot_options = NULL;
3843 aio_context_acquire(ctx);
3844 ret = bdrv_append(bs_snapshot, bs, errp);
3845 aio_context_release(ctx);
3853 qobject_unref(snapshot_options);
3858 * Opens a disk image (raw, qcow2, vmdk, ...)
3860 * options is a QDict of options to pass to the block drivers, or NULL for an
3861 * empty set of options. The reference to the QDict belongs to the block layer
3862 * after the call (even on failure), so if the caller intends to reuse the
3863 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3865 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3866 * If it is not NULL, the referenced BDS will be reused.
3868 * The reference parameter may be used to specify an existing block device which
3869 * should be opened. If specified, neither options nor a filename may be given,
3870 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3872 * The caller must always hold the main AioContext lock.
3874 static BlockDriverState * no_coroutine_fn
3875 bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3876 int flags, BlockDriverState *parent,
3877 const BdrvChildClass *child_class, BdrvChildRole child_role,
3881 BlockBackend *file = NULL;
3882 BlockDriverState *bs;
3883 BlockDriver *drv = NULL;
3885 const char *drvname;
3886 const char *backing;
3887 Error *local_err = NULL;
3888 QDict *snapshot_options = NULL;
3889 int snapshot_flags = 0;
3890 AioContext *ctx = qemu_get_aio_context();
3892 assert(!child_class || !flags);
3893 assert(!child_class == !parent);
3894 GLOBAL_STATE_CODE();
3895 assert(!qemu_in_coroutine());
3898 bool options_non_empty = options ? qdict_size(options) : false;
3899 qobject_unref(options);
3901 if (filename || options_non_empty) {
3902 error_setg(errp, "Cannot reference an existing block device with "
3903 "additional options or a new filename");
3907 bs = bdrv_lookup_bs(reference, reference, errp);
3918 /* NULL means an empty set of options */
3919 if (options == NULL) {
3920 options = qdict_new();
3923 /* json: syntax counts as explicit options, as if in the QDict */
3924 parse_json_protocol(options, &filename, &local_err);
3929 bs->explicit_options = qdict_clone_shallow(options);
3932 bool parent_is_format;
3935 parent_is_format = parent->drv->is_format;
3938 * parent->drv is not set yet because this node is opened for
3939 * (potential) format probing. That means that @parent is going
3940 * to be a format node.
3942 parent_is_format = true;
3945 bs->inherits_from = parent;
3946 child_class->inherit_options(child_role, parent_is_format,
3948 parent->open_flags, parent->options);
3951 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3957 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3958 * Caution: getting a boolean member of @options requires care.
3959 * When @options come from -blockdev or blockdev_add, members are
3960 * typed according to the QAPI schema, but when they come from
3961 * -drive, they're all QString.
3963 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3964 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3965 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3967 flags &= ~BDRV_O_RDWR;
3970 if (flags & BDRV_O_SNAPSHOT) {
3971 snapshot_options = qdict_new();
3972 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3974 /* Let bdrv_backing_options() override "read-only" */
3975 qdict_del(options, BDRV_OPT_READ_ONLY);
3976 bdrv_inherited_options(BDRV_CHILD_COW, true,
3977 &flags, options, flags, options);
3980 bs->open_flags = flags;
3981 bs->options = options;
3982 options = qdict_clone_shallow(options);
3984 /* Find the right image format driver */
3985 /* See cautionary note on accessing @options above */
3986 drvname = qdict_get_try_str(options, "driver");
3988 drv = bdrv_find_format(drvname);
3990 error_setg(errp, "Unknown driver: '%s'", drvname);
3995 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3997 /* See cautionary note on accessing @options above */
3998 backing = qdict_get_try_str(options, "backing");
3999 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4000 (backing && *backing == '\0'))
4003 warn_report("Use of \"backing\": \"\" is deprecated; "
4004 "use \"backing\": null instead");
4006 flags |= BDRV_O_NO_BACKING;
4007 qdict_del(bs->explicit_options, "backing");
4008 qdict_del(bs->options, "backing");
4009 qdict_del(options, "backing");
4012 /* Open image file without format layer. This BlockBackend is only used for
4013 * probing, the block drivers will do their own bdrv_open_child() for the
4014 * same BDS, which is why we put the node name back into options. */
4015 if ((flags & BDRV_O_PROTOCOL) == 0) {
4016 BlockDriverState *file_bs;
4018 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
4019 &child_of_bds, BDRV_CHILD_IMAGE,
4024 if (file_bs != NULL) {
4025 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4026 * looking at the header to guess the image format. This works even
4027 * in cases where a guest would not see a consistent state. */
4028 ctx = bdrv_get_aio_context(file_bs);
4029 aio_context_acquire(ctx);
4030 file = blk_new(ctx, 0, BLK_PERM_ALL);
4031 blk_insert_bs(file, file_bs, &local_err);
4032 bdrv_unref(file_bs);
4033 aio_context_release(ctx);
4039 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4043 /* Image format probing */
4046 ret = find_image_format(file, filename, &drv, &local_err);
4051 * This option update would logically belong in bdrv_fill_options(),
4052 * but we first need to open bs->file for the probing to work, while
4053 * opening bs->file already requires the (mostly) final set of options
4054 * so that cache mode etc. can be inherited.
4056 * Adding the driver later is somewhat ugly, but it's not an option
4057 * that would ever be inherited, so it's correct. We just need to make
4058 * sure to update both bs->options (which has the full effective
4059 * options for bs) and options (which has file.* already removed).
4061 qdict_put_str(bs->options, "driver", drv->format_name);
4062 qdict_put_str(options, "driver", drv->format_name);
4064 error_setg(errp, "Must specify either driver or file");
4068 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4069 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4070 /* file must be NULL if a protocol BDS is about to be created
4071 * (the inverse results in an error message from bdrv_open_common()) */
4072 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4074 /* Open the image */
4075 ret = bdrv_open_common(bs, file, options, &local_err);
4080 /* The AioContext could have changed during bdrv_open_common() */
4081 ctx = bdrv_get_aio_context(bs);
4084 aio_context_acquire(ctx);
4086 aio_context_release(ctx);
4090 /* If there is a backing file, use it */
4091 if ((flags & BDRV_O_NO_BACKING) == 0) {
4092 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4094 goto close_and_fail;
4098 /* Remove all children options and references
4099 * from bs->options and bs->explicit_options */
4100 QLIST_FOREACH(child, &bs->children, next) {
4101 char *child_key_dot;
4102 child_key_dot = g_strdup_printf("%s.", child->name);
4103 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4104 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4105 qdict_del(bs->explicit_options, child->name);
4106 qdict_del(bs->options, child->name);
4107 g_free(child_key_dot);
4110 /* Check if any unknown options were used */
4111 if (qdict_size(options) != 0) {
4112 const QDictEntry *entry = qdict_first(options);
4113 if (flags & BDRV_O_PROTOCOL) {
4114 error_setg(errp, "Block protocol '%s' doesn't support the option "
4115 "'%s'", drv->format_name, entry->key);
4118 "Block format '%s' does not support the option '%s'",
4119 drv->format_name, entry->key);
4122 goto close_and_fail;
4125 bdrv_parent_cb_change_media(bs, true);
4127 qobject_unref(options);
4130 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4131 * temporary snapshot afterwards. */
4132 if (snapshot_flags) {
4133 BlockDriverState *snapshot_bs;
4134 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4135 snapshot_options, &local_err);
4136 snapshot_options = NULL;
4138 goto close_and_fail;
4140 /* We are not going to return bs but the overlay on top of it
4141 * (snapshot_bs); thus, we have to drop the strong reference to bs
4142 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4143 * though, because the overlay still has a reference to it. */
4144 aio_context_acquire(ctx);
4146 aio_context_release(ctx);
4153 aio_context_acquire(ctx);
4155 qobject_unref(snapshot_options);
4156 qobject_unref(bs->explicit_options);
4157 qobject_unref(bs->options);
4158 qobject_unref(options);
4160 bs->explicit_options = NULL;
4162 aio_context_release(ctx);
4163 error_propagate(errp, local_err);
4167 aio_context_acquire(ctx);
4169 aio_context_release(ctx);
4170 qobject_unref(snapshot_options);
4171 qobject_unref(options);
4172 error_propagate(errp, local_err);
4176 /* The caller must always hold the main AioContext lock. */
4177 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4178 QDict *options, int flags, Error **errp)
4180 GLOBAL_STATE_CODE();
4182 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4186 /* Return true if the NULL-terminated @list contains @str */
4187 static bool is_str_in_list(const char *str, const char *const *list)
4191 for (i = 0; list[i] != NULL; i++) {
4192 if (!strcmp(str, list[i])) {
4201 * Check that every option set in @bs->options is also set in
4204 * Options listed in the common_options list and in
4205 * @bs->drv->mutable_opts are skipped.
4207 * Return 0 on success, otherwise return -EINVAL and set @errp.
4209 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4210 const QDict *new_opts, Error **errp)
4212 const QDictEntry *e;
4213 /* These options are common to all block drivers and are handled
4214 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4215 const char *const common_options[] = {
4216 "node-name", "discard", "cache.direct", "cache.no-flush",
4217 "read-only", "auto-read-only", "detect-zeroes", NULL
4220 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4221 if (!qdict_haskey(new_opts, e->key) &&
4222 !is_str_in_list(e->key, common_options) &&
4223 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4224 error_setg(errp, "Option '%s' cannot be reset "
4225 "to its default value", e->key);
4234 * Returns true if @child can be reached recursively from @bs
4236 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4237 BlockDriverState *child)
4245 QLIST_FOREACH(c, &bs->children, next) {
4246 if (bdrv_recurse_has_child(c->bs, child)) {
4255 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4256 * reopen of multiple devices.
4258 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4259 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4260 * be created and initialized. This newly created BlockReopenQueue should be
4261 * passed back in for subsequent calls that are intended to be of the same
4264 * bs is the BlockDriverState to add to the reopen queue.
4266 * options contains the changed options for the associated bs
4267 * (the BlockReopenQueue takes ownership)
4269 * flags contains the open flags for the associated bs
4271 * returns a pointer to bs_queue, which is either the newly allocated
4272 * bs_queue, or the existing bs_queue being used.
4274 * bs is drained here and undrained by bdrv_reopen_queue_free().
4276 * To be called with bs->aio_context locked.
4278 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4279 BlockDriverState *bs,
4281 const BdrvChildClass *klass,
4283 bool parent_is_format,
4284 QDict *parent_options,
4290 BlockReopenQueueEntry *bs_entry;
4292 QDict *old_options, *explicit_options, *options_copy;
4296 GLOBAL_STATE_CODE();
4298 bdrv_drained_begin(bs);
4300 if (bs_queue == NULL) {
4301 bs_queue = g_new0(BlockReopenQueue, 1);
4302 QTAILQ_INIT(bs_queue);
4306 options = qdict_new();
4309 /* Check if this BlockDriverState is already in the queue */
4310 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4311 if (bs == bs_entry->state.bs) {
4317 * Precedence of options:
4318 * 1. Explicitly passed in options (highest)
4319 * 2. Retained from explicitly set options of bs
4320 * 3. Inherited from parent node
4321 * 4. Retained from effective options of bs
4324 /* Old explicitly set values (don't overwrite by inherited value) */
4325 if (bs_entry || keep_old_opts) {
4326 old_options = qdict_clone_shallow(bs_entry ?
4327 bs_entry->state.explicit_options :
4328 bs->explicit_options);
4329 bdrv_join_options(bs, options, old_options);
4330 qobject_unref(old_options);
4333 explicit_options = qdict_clone_shallow(options);
4335 /* Inherit from parent node */
4336 if (parent_options) {
4338 klass->inherit_options(role, parent_is_format, &flags, options,
4339 parent_flags, parent_options);
4341 flags = bdrv_get_flags(bs);
4344 if (keep_old_opts) {
4345 /* Old values are used for options that aren't set yet */
4346 old_options = qdict_clone_shallow(bs->options);
4347 bdrv_join_options(bs, options, old_options);
4348 qobject_unref(old_options);
4351 /* We have the final set of options so let's update the flags */
4352 options_copy = qdict_clone_shallow(options);
4353 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4354 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4355 update_flags_from_options(&flags, opts);
4356 qemu_opts_del(opts);
4357 qobject_unref(options_copy);
4359 /* bdrv_open_inherit() sets and clears some additional flags internally */
4360 flags &= ~BDRV_O_PROTOCOL;
4361 if (flags & BDRV_O_RDWR) {
4362 flags |= BDRV_O_ALLOW_RDWR;
4366 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4367 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4369 qobject_unref(bs_entry->state.options);
4370 qobject_unref(bs_entry->state.explicit_options);
4373 bs_entry->state.bs = bs;
4374 bs_entry->state.options = options;
4375 bs_entry->state.explicit_options = explicit_options;
4376 bs_entry->state.flags = flags;
4379 * If keep_old_opts is false then it means that unspecified
4380 * options must be reset to their original value. We don't allow
4381 * resetting 'backing' but we need to know if the option is
4382 * missing in order to decide if we have to return an error.
4384 if (!keep_old_opts) {
4385 bs_entry->state.backing_missing =
4386 !qdict_haskey(options, "backing") &&
4387 !qdict_haskey(options, "backing.driver");
4390 QLIST_FOREACH(child, &bs->children, next) {
4391 QDict *new_child_options = NULL;
4392 bool child_keep_old = keep_old_opts;
4394 /* reopen can only change the options of block devices that were
4395 * implicitly created and inherited options. For other (referenced)
4396 * block devices, a syntax like "backing.foo" results in an error. */
4397 if (child->bs->inherits_from != bs) {
4401 /* Check if the options contain a child reference */
4402 if (qdict_haskey(options, child->name)) {
4403 const char *childref = qdict_get_try_str(options, child->name);
4405 * The current child must not be reopened if the child
4406 * reference is null or points to a different node.
4408 if (g_strcmp0(childref, child->bs->node_name)) {
4412 * If the child reference points to the current child then
4413 * reopen it with its existing set of options (note that
4414 * it can still inherit new options from the parent).
4416 child_keep_old = true;
4418 /* Extract child options ("child-name.*") */
4419 char *child_key_dot = g_strdup_printf("%s.", child->name);
4420 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4421 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4422 g_free(child_key_dot);
4425 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4426 child->klass, child->role, bs->drv->is_format,
4427 options, flags, child_keep_old);
4433 /* To be called with bs->aio_context locked */
4434 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4435 BlockDriverState *bs,
4436 QDict *options, bool keep_old_opts)
4438 GLOBAL_STATE_CODE();
4440 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4441 NULL, 0, keep_old_opts);
4444 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4446 GLOBAL_STATE_CODE();
4448 BlockReopenQueueEntry *bs_entry, *next;
4449 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4450 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4452 aio_context_acquire(ctx);
4453 bdrv_drained_end(bs_entry->state.bs);
4454 aio_context_release(ctx);
4456 qobject_unref(bs_entry->state.explicit_options);
4457 qobject_unref(bs_entry->state.options);
4465 * Reopen multiple BlockDriverStates atomically & transactionally.
4467 * The queue passed in (bs_queue) must have been built up previous
4468 * via bdrv_reopen_queue().
4470 * Reopens all BDS specified in the queue, with the appropriate
4471 * flags. All devices are prepared for reopen, and failure of any
4472 * device will cause all device changes to be abandoned, and intermediate
4475 * If all devices prepare successfully, then the changes are committed
4478 * All affected nodes must be drained between bdrv_reopen_queue() and
4479 * bdrv_reopen_multiple().
4481 * To be called from the main thread, with all other AioContexts unlocked.
4483 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4486 BlockReopenQueueEntry *bs_entry, *next;
4488 Transaction *tran = tran_new();
4489 g_autoptr(GSList) refresh_list = NULL;
4491 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4492 assert(bs_queue != NULL);
4493 GLOBAL_STATE_CODE();
4495 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4496 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4497 aio_context_acquire(ctx);
4498 ret = bdrv_flush(bs_entry->state.bs);
4499 aio_context_release(ctx);
4501 error_setg_errno(errp, -ret, "Error flushing drive");
4506 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4507 assert(bs_entry->state.bs->quiesce_counter > 0);
4508 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4509 aio_context_acquire(ctx);
4510 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4511 aio_context_release(ctx);
4515 bs_entry->prepared = true;
4518 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4519 BDRVReopenState *state = &bs_entry->state;
4521 refresh_list = g_slist_prepend(refresh_list, state->bs);
4522 if (state->old_backing_bs) {
4523 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4525 if (state->old_file_bs) {
4526 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4531 * Note that file-posix driver rely on permission update done during reopen
4532 * (even if no permission changed), because it wants "new" permissions for
4533 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4534 * in raw_reopen_prepare() which is called with "old" permissions.
4536 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4542 * If we reach this point, we have success and just need to apply the
4545 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4546 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4547 * children are usually goes after parents in reopen-queue, so go from last
4550 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4551 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4552 aio_context_acquire(ctx);
4553 bdrv_reopen_commit(&bs_entry->state);
4554 aio_context_release(ctx);
4559 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4560 BlockDriverState *bs = bs_entry->state.bs;
4562 if (bs->drv->bdrv_reopen_commit_post) {
4563 ctx = bdrv_get_aio_context(bs);
4564 aio_context_acquire(ctx);
4565 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4566 aio_context_release(ctx);
4575 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4576 if (bs_entry->prepared) {
4577 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4578 aio_context_acquire(ctx);
4579 bdrv_reopen_abort(&bs_entry->state);
4580 aio_context_release(ctx);
4585 bdrv_reopen_queue_free(bs_queue);
4590 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4593 AioContext *ctx = bdrv_get_aio_context(bs);
4594 BlockReopenQueue *queue;
4597 GLOBAL_STATE_CODE();
4599 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4601 if (ctx != qemu_get_aio_context()) {
4602 aio_context_release(ctx);
4604 ret = bdrv_reopen_multiple(queue, errp);
4606 if (ctx != qemu_get_aio_context()) {
4607 aio_context_acquire(ctx);
4613 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4616 QDict *opts = qdict_new();
4618 GLOBAL_STATE_CODE();
4620 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4622 return bdrv_reopen(bs, opts, true, errp);
4626 * Take a BDRVReopenState and check if the value of 'backing' in the
4627 * reopen_state->options QDict is valid or not.
4629 * If 'backing' is missing from the QDict then return 0.
4631 * If 'backing' contains the node name of the backing file of
4632 * reopen_state->bs then return 0.
4634 * If 'backing' contains a different node name (or is null) then check
4635 * whether the current backing file can be replaced with the new one.
4636 * If that's the case then reopen_state->replace_backing_bs is set to
4637 * true and reopen_state->new_backing_bs contains a pointer to the new
4638 * backing BlockDriverState (or NULL).
4640 * Return 0 on success, otherwise return < 0 and set @errp.
4642 * The caller must hold the AioContext lock of @reopen_state->bs.
4643 * @reopen_state->bs can move to a different AioContext in this function.
4644 * Callers must make sure that their AioContext locking is still correct after
4647 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4648 bool is_backing, Transaction *tran,
4651 BlockDriverState *bs = reopen_state->bs;
4652 BlockDriverState *new_child_bs;
4653 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4655 const char *child_name = is_backing ? "backing" : "file";
4658 AioContext *ctx, *old_ctx;
4661 GLOBAL_STATE_CODE();
4663 value = qdict_get(reopen_state->options, child_name);
4664 if (value == NULL) {
4668 switch (qobject_type(value)) {
4670 assert(is_backing); /* The 'file' option does not allow a null value */
4671 new_child_bs = NULL;
4674 str = qstring_get_str(qobject_to(QString, value));
4675 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4676 if (new_child_bs == NULL) {
4678 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4679 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4680 "cycle", str, child_name, bs->node_name);
4686 * The options QDict has been flattened, so 'backing' and 'file'
4687 * do not allow any other data type here.
4689 g_assert_not_reached();
4692 if (old_child_bs == new_child_bs) {
4697 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4701 if (old_child_bs->implicit) {
4702 error_setg(errp, "Cannot replace implicit %s child of %s",
4703 child_name, bs->node_name);
4708 if (bs->drv->is_filter && !old_child_bs) {
4710 * Filters always have a file or a backing child, so we are trying to
4711 * change wrong child
4713 error_setg(errp, "'%s' is a %s filter node that does not support a "
4714 "%s child", bs->node_name, bs->drv->format_name, child_name);
4719 reopen_state->old_backing_bs = old_child_bs;
4721 reopen_state->old_file_bs = old_child_bs;
4724 old_ctx = bdrv_get_aio_context(bs);
4725 ctx = bdrv_get_aio_context(new_child_bs);
4726 if (old_ctx != ctx) {
4727 aio_context_release(old_ctx);
4728 aio_context_acquire(ctx);
4731 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4734 if (old_ctx != ctx) {
4735 aio_context_release(ctx);
4736 aio_context_acquire(old_ctx);
4743 * Prepares a BlockDriverState for reopen. All changes are staged in the
4744 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4745 * the block driver layer .bdrv_reopen_prepare()
4747 * bs is the BlockDriverState to reopen
4748 * flags are the new open flags
4749 * queue is the reopen queue
4751 * Returns 0 on success, non-zero on error. On error errp will be set
4754 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4755 * It is the responsibility of the caller to then call the abort() or
4756 * commit() for any other BDS that have been left in a prepare() state
4758 * The caller must hold the AioContext lock of @reopen_state->bs.
4760 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4761 BlockReopenQueue *queue,
4762 Transaction *change_child_tran, Error **errp)
4766 Error *local_err = NULL;
4769 QDict *orig_reopen_opts;
4770 char *discard = NULL;
4772 bool drv_prepared = false;
4774 assert(reopen_state != NULL);
4775 assert(reopen_state->bs->drv != NULL);
4776 GLOBAL_STATE_CODE();
4777 drv = reopen_state->bs->drv;
4779 /* This function and each driver's bdrv_reopen_prepare() remove
4780 * entries from reopen_state->options as they are processed, so
4781 * we need to make a copy of the original QDict. */
4782 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4784 /* Process generic block layer options */
4785 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4786 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4791 /* This was already called in bdrv_reopen_queue_child() so the flags
4792 * are up-to-date. This time we simply want to remove the options from
4793 * QemuOpts in order to indicate that they have been processed. */
4794 old_flags = reopen_state->flags;
4795 update_flags_from_options(&reopen_state->flags, opts);
4796 assert(old_flags == reopen_state->flags);
4798 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4799 if (discard != NULL) {
4800 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4801 error_setg(errp, "Invalid discard option");
4807 reopen_state->detect_zeroes =
4808 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4810 error_propagate(errp, local_err);
4815 /* All other options (including node-name and driver) must be unchanged.
4816 * Put them back into the QDict, so that they are checked at the end
4817 * of this function. */
4818 qemu_opts_to_qdict(opts, reopen_state->options);
4820 /* If we are to stay read-only, do not allow permission change
4821 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4822 * not set, or if the BDS still has copy_on_read enabled */
4823 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4824 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4826 error_propagate(errp, local_err);
4830 if (drv->bdrv_reopen_prepare) {
4832 * If a driver-specific option is missing, it means that we
4833 * should reset it to its default value.
4834 * But not all options allow that, so we need to check it first.
4836 ret = bdrv_reset_options_allowed(reopen_state->bs,
4837 reopen_state->options, errp);
4842 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4844 if (local_err != NULL) {
4845 error_propagate(errp, local_err);
4847 bdrv_refresh_filename(reopen_state->bs);
4848 error_setg(errp, "failed while preparing to reopen image '%s'",
4849 reopen_state->bs->filename);
4854 /* It is currently mandatory to have a bdrv_reopen_prepare()
4855 * handler for each supported drv. */
4856 error_setg(errp, "Block format '%s' used by node '%s' "
4857 "does not support reopening files", drv->format_name,
4858 bdrv_get_device_or_node_name(reopen_state->bs));
4863 drv_prepared = true;
4866 * We must provide the 'backing' option if the BDS has a backing
4867 * file or if the image file has a backing file name as part of
4868 * its metadata. Otherwise the 'backing' option can be omitted.
4870 if (drv->supports_backing && reopen_state->backing_missing &&
4871 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4872 error_setg(errp, "backing is missing for '%s'",
4873 reopen_state->bs->node_name);
4879 * Allow changing the 'backing' option. The new value can be
4880 * either a reference to an existing node (using its node name)
4881 * or NULL to simply detach the current backing file.
4883 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4884 change_child_tran, errp);
4888 qdict_del(reopen_state->options, "backing");
4890 /* Allow changing the 'file' option. In this case NULL is not allowed */
4891 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4892 change_child_tran, errp);
4896 qdict_del(reopen_state->options, "file");
4898 /* Options that are not handled are only okay if they are unchanged
4899 * compared to the old state. It is expected that some options are only
4900 * used for the initial open, but not reopen (e.g. filename) */
4901 if (qdict_size(reopen_state->options)) {
4902 const QDictEntry *entry = qdict_first(reopen_state->options);
4905 QObject *new = entry->value;
4906 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4908 /* Allow child references (child_name=node_name) as long as they
4909 * point to the current child (i.e. everything stays the same). */
4910 if (qobject_type(new) == QTYPE_QSTRING) {
4912 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4913 if (!strcmp(child->name, entry->key)) {
4919 if (!strcmp(child->bs->node_name,
4920 qstring_get_str(qobject_to(QString, new)))) {
4921 continue; /* Found child with this name, skip option */
4927 * TODO: When using -drive to specify blockdev options, all values
4928 * will be strings; however, when using -blockdev, blockdev-add or
4929 * filenames using the json:{} pseudo-protocol, they will be
4931 * In contrast, reopening options are (currently) always strings
4932 * (because you can only specify them through qemu-io; all other
4933 * callers do not specify any options).
4934 * Therefore, when using anything other than -drive to create a BDS,
4935 * this cannot detect non-string options as unchanged, because
4936 * qobject_is_equal() always returns false for objects of different
4937 * type. In the future, this should be remedied by correctly typing
4938 * all options. For now, this is not too big of an issue because
4939 * the user can simply omit options which cannot be changed anyway,
4940 * so they will stay unchanged.
4942 if (!qobject_is_equal(new, old)) {
4943 error_setg(errp, "Cannot change the option '%s'", entry->key);
4947 } while ((entry = qdict_next(reopen_state->options, entry)));
4952 /* Restore the original reopen_state->options QDict */
4953 qobject_unref(reopen_state->options);
4954 reopen_state->options = qobject_ref(orig_reopen_opts);
4957 if (ret < 0 && drv_prepared) {
4958 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4959 * call drv->bdrv_reopen_abort() before signaling an error
4960 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4961 * when the respective bdrv_reopen_prepare() has failed) */
4962 if (drv->bdrv_reopen_abort) {
4963 drv->bdrv_reopen_abort(reopen_state);
4966 qemu_opts_del(opts);
4967 qobject_unref(orig_reopen_opts);
4973 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4974 * makes them final by swapping the staging BlockDriverState contents into
4975 * the active BlockDriverState contents.
4977 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4980 BlockDriverState *bs;
4983 assert(reopen_state != NULL);
4984 bs = reopen_state->bs;
4986 assert(drv != NULL);
4987 GLOBAL_STATE_CODE();
4989 /* If there are any driver level actions to take */
4990 if (drv->bdrv_reopen_commit) {
4991 drv->bdrv_reopen_commit(reopen_state);
4994 /* set BDS specific flags now */
4995 qobject_unref(bs->explicit_options);
4996 qobject_unref(bs->options);
4997 qobject_ref(reopen_state->explicit_options);
4998 qobject_ref(reopen_state->options);
5000 bs->explicit_options = reopen_state->explicit_options;
5001 bs->options = reopen_state->options;
5002 bs->open_flags = reopen_state->flags;
5003 bs->detect_zeroes = reopen_state->detect_zeroes;
5005 /* Remove child references from bs->options and bs->explicit_options.
5006 * Child options were already removed in bdrv_reopen_queue_child() */
5007 QLIST_FOREACH(child, &bs->children, next) {
5008 qdict_del(bs->explicit_options, child->name);
5009 qdict_del(bs->options, child->name);
5011 /* backing is probably removed, so it's not handled by previous loop */
5012 qdict_del(bs->explicit_options, "backing");
5013 qdict_del(bs->options, "backing");
5015 bdrv_graph_rdlock_main_loop();
5016 bdrv_refresh_limits(bs, NULL, NULL);
5017 bdrv_graph_rdunlock_main_loop();
5018 bdrv_refresh_total_sectors(bs, bs->total_sectors);
5022 * Abort the reopen, and delete and free the staged changes in
5025 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
5029 assert(reopen_state != NULL);
5030 drv = reopen_state->bs->drv;
5031 assert(drv != NULL);
5032 GLOBAL_STATE_CODE();
5034 if (drv->bdrv_reopen_abort) {
5035 drv->bdrv_reopen_abort(reopen_state);
5040 static void bdrv_close(BlockDriverState *bs)
5042 BdrvAioNotifier *ban, *ban_next;
5043 BdrvChild *child, *next;
5045 GLOBAL_STATE_CODE();
5046 assert(!bs->refcnt);
5048 bdrv_drained_begin(bs); /* complete I/O */
5050 bdrv_drain(bs); /* in case flush left pending I/O */
5053 if (bs->drv->bdrv_close) {
5054 /* Must unfreeze all children, so bdrv_unref_child() works */
5055 bs->drv->bdrv_close(bs);
5060 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
5061 bdrv_unref_child(bs, child);
5064 assert(!bs->backing);
5068 qatomic_set(&bs->copy_on_read, 0);
5069 bs->backing_file[0] = '\0';
5070 bs->backing_format[0] = '\0';
5071 bs->total_sectors = 0;
5072 bs->encrypted = false;
5074 qobject_unref(bs->options);
5075 qobject_unref(bs->explicit_options);
5077 bs->explicit_options = NULL;
5078 qobject_unref(bs->full_open_options);
5079 bs->full_open_options = NULL;
5080 g_free(bs->block_status_cache);
5081 bs->block_status_cache = NULL;
5083 bdrv_release_named_dirty_bitmaps(bs);
5084 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5086 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5089 QLIST_INIT(&bs->aio_notifiers);
5090 bdrv_drained_end(bs);
5093 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5094 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5097 if (bs->quiesce_counter) {
5098 bdrv_drain_all_end_quiesce(bs);
5102 void bdrv_close_all(void)
5104 GLOBAL_STATE_CODE();
5105 assert(job_next(NULL) == NULL);
5107 /* Drop references from requests still in flight, such as canceled block
5108 * jobs whose AIO context has not been polled yet */
5111 blk_remove_all_bs();
5112 blockdev_close_all_bdrv_states();
5114 assert(QTAILQ_EMPTY(&all_bdrv_states));
5117 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
5123 if (c->klass->stay_at_node) {
5127 /* If the child @c belongs to the BDS @to, replacing the current
5128 * c->bs by @to would mean to create a loop.
5130 * Such a case occurs when appending a BDS to a backing chain.
5131 * For instance, imagine the following chain:
5133 * guest device -> node A -> further backing chain...
5135 * Now we create a new BDS B which we want to put on top of this
5136 * chain, so we first attach A as its backing node:
5141 * guest device -> node A -> further backing chain...
5143 * Finally we want to replace A by B. When doing that, we want to
5144 * replace all pointers to A by pointers to B -- except for the
5145 * pointer from B because (1) that would create a loop, and (2)
5146 * that pointer should simply stay intact:
5148 * guest device -> node B
5151 * node A -> further backing chain...
5153 * In general, when replacing a node A (c->bs) by a node B (@to),
5154 * if A is a child of B, that means we cannot replace A by B there
5155 * because that would create a loop. Silently detaching A from B
5156 * is also not really an option. So overall just leaving A in
5157 * place there is the most sensible choice.
5159 * We would also create a loop in any cases where @c is only
5160 * indirectly referenced by @to. Prevent this by returning false
5161 * if @c is found (by breadth-first search) anywhere in the whole
5166 found = g_hash_table_new(NULL, NULL);
5167 g_hash_table_add(found, to);
5168 queue = g_queue_new();
5169 g_queue_push_tail(queue, to);
5171 while (!g_queue_is_empty(queue)) {
5172 BlockDriverState *v = g_queue_pop_head(queue);
5175 QLIST_FOREACH(c2, &v->children, next) {
5181 if (g_hash_table_contains(found, c2->bs)) {
5185 g_queue_push_tail(queue, c2->bs);
5186 g_hash_table_add(found, c2->bs);
5190 g_queue_free(queue);
5191 g_hash_table_destroy(found);
5196 static void bdrv_remove_child_commit(void *opaque)
5198 GLOBAL_STATE_CODE();
5199 bdrv_child_free(opaque);
5202 static TransactionActionDrv bdrv_remove_child_drv = {
5203 .commit = bdrv_remove_child_commit,
5206 /* Function doesn't update permissions, caller is responsible for this. */
5207 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5214 BlockDriverState *bs = child->bs;
5215 bdrv_drained_begin(bs);
5216 bdrv_replace_child_tran(child, NULL, tran);
5217 bdrv_drained_end(bs);
5220 tran_add(tran, &bdrv_remove_child_drv, child);
5223 static void undrain_on_clean_cb(void *opaque)
5225 bdrv_drained_end(opaque);
5228 static TransactionActionDrv undrain_on_clean = {
5229 .clean = undrain_on_clean_cb,
5232 static int bdrv_replace_node_noperm(BlockDriverState *from,
5233 BlockDriverState *to,
5234 bool auto_skip, Transaction *tran,
5237 BdrvChild *c, *next;
5239 GLOBAL_STATE_CODE();
5241 bdrv_drained_begin(from);
5242 bdrv_drained_begin(to);
5243 tran_add(tran, &undrain_on_clean, from);
5244 tran_add(tran, &undrain_on_clean, to);
5246 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5247 assert(c->bs == from);
5248 if (!should_update_child(c, to)) {
5252 error_setg(errp, "Should not change '%s' link to '%s'",
5253 c->name, from->node_name);
5257 error_setg(errp, "Cannot change '%s' link to '%s'",
5258 c->name, from->node_name);
5261 bdrv_replace_child_tran(c, to, tran);
5268 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5269 * if it creates a parent-child relation loop or if parent is block-job.
5271 * With auto_skip=false the error is returned if from has a parent which should
5274 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5275 * case backing link of the cow-parent of @to is removed.
5277 static int bdrv_replace_node_common(BlockDriverState *from,
5278 BlockDriverState *to,
5279 bool auto_skip, bool detach_subchain,
5282 Transaction *tran = tran_new();
5283 g_autoptr(GSList) refresh_list = NULL;
5284 BlockDriverState *to_cow_parent = NULL;
5287 GLOBAL_STATE_CODE();
5289 if (detach_subchain) {
5290 assert(bdrv_chain_contains(from, to));
5292 for (to_cow_parent = from;
5293 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5294 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5300 /* Make sure that @from doesn't go away until we have successfully attached
5301 * all of its parents to @to. */
5304 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5305 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5306 bdrv_drained_begin(from);
5309 * Do the replacement without permission update.
5310 * Replacement may influence the permissions, we should calculate new
5311 * permissions based on new graph. If we fail, we'll roll-back the
5314 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5319 if (detach_subchain) {
5320 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5323 refresh_list = g_slist_prepend(refresh_list, to);
5324 refresh_list = g_slist_prepend(refresh_list, from);
5326 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5334 tran_finalize(tran, ret);
5336 bdrv_drained_end(from);
5342 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5345 GLOBAL_STATE_CODE();
5347 return bdrv_replace_node_common(from, to, true, false, errp);
5350 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5352 GLOBAL_STATE_CODE();
5354 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5359 * Add new bs contents at the top of an image chain while the chain is
5360 * live, while keeping required fields on the top layer.
5362 * This will modify the BlockDriverState fields, and swap contents
5363 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5365 * bs_new must not be attached to a BlockBackend and must not have backing
5368 * This function does not create any image files.
5370 * The caller must hold the AioContext lock for @bs_top.
5372 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5377 Transaction *tran = tran_new();
5378 AioContext *old_context, *new_context = NULL;
5380 GLOBAL_STATE_CODE();
5382 assert(!bs_new->backing);
5384 old_context = bdrv_get_aio_context(bs_top);
5386 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5387 &child_of_bds, bdrv_backing_role(bs_new),
5395 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5396 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5397 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5398 * that assumes the new lock is taken.
5400 new_context = bdrv_get_aio_context(bs_top);
5402 if (old_context != new_context) {
5403 aio_context_release(old_context);
5404 aio_context_acquire(new_context);
5407 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5412 ret = bdrv_refresh_perms(bs_new, tran, errp);
5414 tran_finalize(tran, ret);
5416 bdrv_graph_rdlock_main_loop();
5417 bdrv_refresh_limits(bs_top, NULL, NULL);
5418 bdrv_graph_rdunlock_main_loop();
5420 if (new_context && old_context != new_context) {
5421 aio_context_release(new_context);
5422 aio_context_acquire(old_context);
5428 /* Not for empty child */
5429 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5433 Transaction *tran = tran_new();
5434 g_autoptr(GSList) refresh_list = NULL;
5435 BlockDriverState *old_bs = child->bs;
5437 GLOBAL_STATE_CODE();
5440 bdrv_drained_begin(old_bs);
5441 bdrv_drained_begin(new_bs);
5443 bdrv_replace_child_tran(child, new_bs, tran);
5445 refresh_list = g_slist_prepend(refresh_list, old_bs);
5446 refresh_list = g_slist_prepend(refresh_list, new_bs);
5448 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5450 tran_finalize(tran, ret);
5452 bdrv_drained_end(old_bs);
5453 bdrv_drained_end(new_bs);
5459 static void bdrv_delete(BlockDriverState *bs)
5461 assert(bdrv_op_blocker_is_empty(bs));
5462 assert(!bs->refcnt);
5463 GLOBAL_STATE_CODE();
5465 /* remove from list, if necessary */
5466 if (bs->node_name[0] != '\0') {
5467 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5469 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5478 * Replace @bs by newly created block node.
5480 * @options is a QDict of options to pass to the block drivers, or NULL for an
5481 * empty set of options. The reference to the QDict belongs to the block layer
5482 * after the call (even on failure), so if the caller intends to reuse the
5483 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5485 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5486 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5487 * different AioContext.
5489 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5490 int flags, Error **errp)
5494 AioContext *ctx = bdrv_get_aio_context(bs);
5495 BlockDriverState *new_node_bs = NULL;
5496 const char *drvname, *node_name;
5499 drvname = qdict_get_try_str(options, "driver");
5501 error_setg(errp, "driver is not specified");
5505 drv = bdrv_find_format(drvname);
5507 error_setg(errp, "Unknown driver: '%s'", drvname);
5511 node_name = qdict_get_try_str(options, "node-name");
5513 GLOBAL_STATE_CODE();
5515 aio_context_release(ctx);
5516 aio_context_acquire(qemu_get_aio_context());
5517 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5519 aio_context_release(qemu_get_aio_context());
5520 aio_context_acquire(ctx);
5521 assert(bdrv_get_aio_context(bs) == ctx);
5523 options = NULL; /* bdrv_new_open_driver() eats options */
5525 error_prepend(errp, "Could not create node: ");
5529 bdrv_drained_begin(bs);
5530 ret = bdrv_replace_node(bs, new_node_bs, errp);
5531 bdrv_drained_end(bs);
5534 error_prepend(errp, "Could not replace node: ");
5541 qobject_unref(options);
5542 bdrv_unref(new_node_bs);
5547 * Run consistency checks on an image
5549 * Returns 0 if the check could be completed (it doesn't mean that the image is
5550 * free of errors) or -errno when an internal error occurred. The results of the
5551 * check are stored in res.
5553 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5554 BdrvCheckResult *res, BdrvCheckMode fix)
5557 assert_bdrv_graph_readable();
5558 if (bs->drv == NULL) {
5561 if (bs->drv->bdrv_co_check == NULL) {
5565 memset(res, 0, sizeof(*res));
5566 return bs->drv->bdrv_co_check(bs, res, fix);
5572 * -EINVAL - backing format specified, but no file
5573 * -ENOSPC - can't update the backing file because no space is left in the
5575 * -ENOTSUP - format driver doesn't support changing the backing file
5577 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5578 const char *backing_fmt, bool require)
5580 BlockDriver *drv = bs->drv;
5583 GLOBAL_STATE_CODE();
5589 /* Backing file format doesn't make sense without a backing file */
5590 if (backing_fmt && !backing_file) {
5594 if (require && backing_file && !backing_fmt) {
5598 if (drv->bdrv_change_backing_file != NULL) {
5599 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5605 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5606 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5607 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5608 backing_file ?: "");
5614 * Finds the first non-filter node above bs in the chain between
5615 * active and bs. The returned node is either an immediate parent of
5616 * bs, or there are only filter nodes between the two.
5618 * Returns NULL if bs is not found in active's image chain,
5619 * or if active == bs.
5621 * Returns the bottommost base image if bs == NULL.
5623 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5624 BlockDriverState *bs)
5627 GLOBAL_STATE_CODE();
5629 bs = bdrv_skip_filters(bs);
5630 active = bdrv_skip_filters(active);
5633 BlockDriverState *next = bdrv_backing_chain_next(active);
5643 /* Given a BDS, searches for the base layer. */
5644 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5646 GLOBAL_STATE_CODE();
5648 return bdrv_find_overlay(bs, NULL);
5652 * Return true if at least one of the COW (backing) and filter links
5653 * between @bs and @base is frozen. @errp is set if that's the case.
5654 * @base must be reachable from @bs, or NULL.
5656 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5659 BlockDriverState *i;
5662 GLOBAL_STATE_CODE();
5664 for (i = bs; i != base; i = child_bs(child)) {
5665 child = bdrv_filter_or_cow_child(i);
5667 if (child && child->frozen) {
5668 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5669 child->name, i->node_name, child->bs->node_name);
5678 * Freeze all COW (backing) and filter links between @bs and @base.
5679 * If any of the links is already frozen the operation is aborted and
5680 * none of the links are modified.
5681 * @base must be reachable from @bs, or NULL.
5682 * Returns 0 on success. On failure returns < 0 and sets @errp.
5684 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5687 BlockDriverState *i;
5690 GLOBAL_STATE_CODE();
5692 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5696 for (i = bs; i != base; i = child_bs(child)) {
5697 child = bdrv_filter_or_cow_child(i);
5698 if (child && child->bs->never_freeze) {
5699 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5700 child->name, child->bs->node_name);
5705 for (i = bs; i != base; i = child_bs(child)) {
5706 child = bdrv_filter_or_cow_child(i);
5708 child->frozen = true;
5716 * Unfreeze all COW (backing) and filter links between @bs and @base.
5717 * The caller must ensure that all links are frozen before using this
5719 * @base must be reachable from @bs, or NULL.
5721 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5723 BlockDriverState *i;
5726 GLOBAL_STATE_CODE();
5728 for (i = bs; i != base; i = child_bs(child)) {
5729 child = bdrv_filter_or_cow_child(i);
5731 assert(child->frozen);
5732 child->frozen = false;
5738 * Drops images above 'base' up to and including 'top', and sets the image
5739 * above 'top' to have base as its backing file.
5741 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5742 * information in 'bs' can be properly updated.
5744 * E.g., this will convert the following chain:
5745 * bottom <- base <- intermediate <- top <- active
5749 * bottom <- base <- active
5751 * It is allowed for bottom==base, in which case it converts:
5753 * base <- intermediate <- top <- active
5759 * If backing_file_str is non-NULL, it will be used when modifying top's
5760 * overlay image metadata.
5763 * if active == top, that is considered an error
5766 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5767 const char *backing_file_str)
5769 BlockDriverState *explicit_top = top;
5770 bool update_inherits_from;
5772 Error *local_err = NULL;
5774 g_autoptr(GSList) updated_children = NULL;
5777 GLOBAL_STATE_CODE();
5780 bdrv_drained_begin(base);
5782 if (!top->drv || !base->drv) {
5786 /* Make sure that base is in the backing chain of top */
5787 if (!bdrv_chain_contains(top, base)) {
5791 /* If 'base' recursively inherits from 'top' then we should set
5792 * base->inherits_from to top->inherits_from after 'top' and all
5793 * other intermediate nodes have been dropped.
5794 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5795 * it because no one inherits from it. We use explicit_top for that. */
5796 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5797 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5799 /* success - we can delete the intermediate states, and link top->base */
5800 if (!backing_file_str) {
5801 bdrv_refresh_filename(base);
5802 backing_file_str = base->filename;
5805 QLIST_FOREACH(c, &top->parents, next_parent) {
5806 updated_children = g_slist_prepend(updated_children, c);
5810 * It seems correct to pass detach_subchain=true here, but it triggers
5811 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5812 * another drained section, which modify the graph (for example, removing
5813 * the child, which we keep in updated_children list). So, it's a TODO.
5815 * Note, bug triggered if pass detach_subchain=true here and run
5816 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5819 bdrv_replace_node_common(top, base, false, false, &local_err);
5821 error_report_err(local_err);
5825 for (p = updated_children; p; p = p->next) {
5828 if (c->klass->update_filename) {
5829 ret = c->klass->update_filename(c, base, backing_file_str,
5833 * TODO: Actually, we want to rollback all previous iterations
5834 * of this loop, and (which is almost impossible) previous
5835 * bdrv_replace_node()...
5837 * Note, that c->klass->update_filename may lead to permission
5838 * update, so it's a bad idea to call it inside permission
5839 * update transaction of bdrv_replace_node.
5841 error_report_err(local_err);
5847 if (update_inherits_from) {
5848 base->inherits_from = explicit_top->inherits_from;
5853 bdrv_drained_end(base);
5859 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5860 * sums the size of all data-bearing children. (This excludes backing
5863 static int64_t coroutine_fn GRAPH_RDLOCK
5864 bdrv_sum_allocated_file_size(BlockDriverState *bs)
5867 int64_t child_size, sum = 0;
5869 QLIST_FOREACH(child, &bs->children, next) {
5870 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5871 BDRV_CHILD_FILTERED))
5873 child_size = bdrv_co_get_allocated_file_size(child->bs);
5874 if (child_size < 0) {
5885 * Length of a allocated file in bytes. Sparse files are counted by actual
5886 * allocated space. Return < 0 if error or unknown.
5888 int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
5890 BlockDriver *drv = bs->drv;
5892 assert_bdrv_graph_readable();
5897 if (drv->bdrv_co_get_allocated_file_size) {
5898 return drv->bdrv_co_get_allocated_file_size(bs);
5901 if (drv->bdrv_file_open) {
5903 * Protocol drivers default to -ENOTSUP (most of their data is
5904 * not stored in any of their children (if they even have any),
5905 * so there is no generic way to figure it out).
5908 } else if (drv->is_filter) {
5909 /* Filter drivers default to the size of their filtered child */
5910 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
5912 /* Other drivers default to summing their children's sizes */
5913 return bdrv_sum_allocated_file_size(bs);
5919 * @drv: Format driver
5920 * @opts: Creation options for new image
5921 * @in_bs: Existing image containing data for new image (may be NULL)
5922 * @errp: Error object
5923 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5926 * Calculate file size required to create a new image.
5928 * If @in_bs is given then space for allocated clusters and zero clusters
5929 * from that image are included in the calculation. If @opts contains a
5930 * backing file that is shared by @in_bs then backing clusters may be omitted
5931 * from the calculation.
5933 * If @in_bs is NULL then the calculation includes no allocated clusters
5934 * unless a preallocation option is given in @opts.
5936 * Note that @in_bs may use a different BlockDriver from @drv.
5938 * If an error occurs the @errp pointer is set.
5940 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5941 BlockDriverState *in_bs, Error **errp)
5944 if (!drv->bdrv_measure) {
5945 error_setg(errp, "Block driver '%s' does not support size measurement",
5950 return drv->bdrv_measure(opts, in_bs, errp);
5954 * Return number of sectors on success, -errno on error.
5956 int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
5958 BlockDriver *drv = bs->drv;
5960 assert_bdrv_graph_readable();
5965 if (bs->bl.has_variable_length) {
5966 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
5971 return bs->total_sectors;
5975 * This wrapper is written by hand because this function is in the hot I/O path,
5976 * via blk_get_geometry.
5978 int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
5980 BlockDriver *drv = bs->drv;
5986 if (bs->bl.has_variable_length) {
5987 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
5993 return bs->total_sectors;
5997 * Return length in bytes on success, -errno on error.
5998 * The length is always a multiple of BDRV_SECTOR_SIZE.
6000 int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
6004 assert_bdrv_graph_readable();
6006 ret = bdrv_co_nb_sectors(bs);
6010 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6013 return ret * BDRV_SECTOR_SIZE;
6016 bool bdrv_is_sg(BlockDriverState *bs)
6023 * Return whether the given node supports compressed writes.
6025 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6027 BlockDriverState *filtered;
6030 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6034 filtered = bdrv_filter_bs(bs);
6037 * Filters can only forward compressed writes, so we have to
6040 return bdrv_supports_compressed_writes(filtered);
6046 const char *bdrv_get_format_name(BlockDriverState *bs)
6049 return bs->drv ? bs->drv->format_name : NULL;
6052 static int qsort_strcmp(const void *a, const void *b)
6054 return strcmp(*(char *const *)a, *(char *const *)b);
6057 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
6058 void *opaque, bool read_only)
6063 const char **formats = NULL;
6065 GLOBAL_STATE_CODE();
6067 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6068 if (drv->format_name) {
6072 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6076 while (formats && i && !found) {
6077 found = !strcmp(formats[--i], drv->format_name);
6081 formats = g_renew(const char *, formats, count + 1);
6082 formats[count++] = drv->format_name;
6087 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6088 const char *format_name = block_driver_modules[i].format_name;
6094 if (use_bdrv_whitelist &&
6095 !bdrv_format_is_whitelisted(format_name, read_only)) {
6099 while (formats && j && !found) {
6100 found = !strcmp(formats[--j], format_name);
6104 formats = g_renew(const char *, formats, count + 1);
6105 formats[count++] = format_name;
6110 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6112 for (i = 0; i < count; i++) {
6113 it(opaque, formats[i]);
6119 /* This function is to find a node in the bs graph */
6120 BlockDriverState *bdrv_find_node(const char *node_name)
6122 BlockDriverState *bs;
6125 GLOBAL_STATE_CODE();
6127 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6128 if (!strcmp(node_name, bs->node_name)) {
6135 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6136 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6139 BlockDeviceInfoList *list;
6140 BlockDriverState *bs;
6142 GLOBAL_STATE_CODE();
6145 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6146 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6148 qapi_free_BlockDeviceInfoList(list);
6151 QAPI_LIST_PREPEND(list, info);
6157 typedef struct XDbgBlockGraphConstructor {
6158 XDbgBlockGraph *graph;
6159 GHashTable *graph_nodes;
6160 } XDbgBlockGraphConstructor;
6162 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6164 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6166 gr->graph = g_new0(XDbgBlockGraph, 1);
6167 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6172 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6174 XDbgBlockGraph *graph = gr->graph;
6176 g_hash_table_destroy(gr->graph_nodes);
6182 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6184 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6191 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6192 * answer of g_hash_table_lookup.
6194 ret = g_hash_table_size(gr->graph_nodes) + 1;
6195 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6200 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6201 XDbgBlockGraphNodeType type, const char *name)
6203 XDbgBlockGraphNode *n;
6205 n = g_new0(XDbgBlockGraphNode, 1);
6207 n->id = xdbg_graph_node_num(gr, node);
6209 n->name = g_strdup(name);
6211 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6214 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6215 const BdrvChild *child)
6217 BlockPermission qapi_perm;
6218 XDbgBlockGraphEdge *edge;
6219 GLOBAL_STATE_CODE();
6221 edge = g_new0(XDbgBlockGraphEdge, 1);
6223 edge->parent = xdbg_graph_node_num(gr, parent);
6224 edge->child = xdbg_graph_node_num(gr, child->bs);
6225 edge->name = g_strdup(child->name);
6227 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6228 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6230 if (flag & child->perm) {
6231 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6233 if (flag & child->shared_perm) {
6234 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6238 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6242 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6246 BlockDriverState *bs;
6248 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6250 GLOBAL_STATE_CODE();
6252 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6253 char *allocated_name = NULL;
6254 const char *name = blk_name(blk);
6257 name = allocated_name = blk_get_attached_dev_id(blk);
6259 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6261 g_free(allocated_name);
6262 if (blk_root(blk)) {
6263 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6267 WITH_JOB_LOCK_GUARD() {
6268 for (job = block_job_next_locked(NULL); job;
6269 job = block_job_next_locked(job)) {
6272 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6274 for (el = job->nodes; el; el = el->next) {
6275 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6280 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6281 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6283 QLIST_FOREACH(child, &bs->children, next) {
6284 xdbg_graph_add_edge(gr, bs, child);
6288 return xdbg_graph_finalize(gr);
6291 BlockDriverState *bdrv_lookup_bs(const char *device,
6292 const char *node_name,
6296 BlockDriverState *bs;
6298 GLOBAL_STATE_CODE();
6301 blk = blk_by_name(device);
6306 error_setg(errp, "Device '%s' has no medium", device);
6314 bs = bdrv_find_node(node_name);
6321 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6322 device ? device : "",
6323 node_name ? node_name : "");
6327 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6328 * return false. If either argument is NULL, return false. */
6329 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6332 GLOBAL_STATE_CODE();
6334 while (top && top != base) {
6335 top = bdrv_filter_or_cow_bs(top);
6341 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6343 GLOBAL_STATE_CODE();
6345 return QTAILQ_FIRST(&graph_bdrv_states);
6347 return QTAILQ_NEXT(bs, node_list);
6350 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6352 GLOBAL_STATE_CODE();
6354 return QTAILQ_FIRST(&all_bdrv_states);
6356 return QTAILQ_NEXT(bs, bs_list);
6359 const char *bdrv_get_node_name(const BlockDriverState *bs)
6362 return bs->node_name;
6365 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6371 /* If multiple parents have a name, just pick the first one. */
6372 QLIST_FOREACH(c, &bs->parents, next_parent) {
6373 if (c->klass->get_name) {
6374 name = c->klass->get_name(c);
6375 if (name && *name) {
6384 /* TODO check what callers really want: bs->node_name or blk_name() */
6385 const char *bdrv_get_device_name(const BlockDriverState *bs)
6388 return bdrv_get_parent_name(bs) ?: "";
6391 /* This can be used to identify nodes that might not have a device
6392 * name associated. Since node and device names live in the same
6393 * namespace, the result is unambiguous. The exception is if both are
6394 * absent, then this returns an empty (non-null) string. */
6395 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6398 return bdrv_get_parent_name(bs) ?: bs->node_name;
6401 int bdrv_get_flags(BlockDriverState *bs)
6404 return bs->open_flags;
6407 int bdrv_has_zero_init_1(BlockDriverState *bs)
6409 GLOBAL_STATE_CODE();
6413 int bdrv_has_zero_init(BlockDriverState *bs)
6415 BlockDriverState *filtered;
6416 GLOBAL_STATE_CODE();
6422 /* If BS is a copy on write image, it is initialized to
6423 the contents of the base image, which may not be zeroes. */
6424 if (bdrv_cow_child(bs)) {
6427 if (bs->drv->bdrv_has_zero_init) {
6428 return bs->drv->bdrv_has_zero_init(bs);
6431 filtered = bdrv_filter_bs(bs);
6433 return bdrv_has_zero_init(filtered);
6440 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6443 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6447 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6450 void bdrv_get_backing_filename(BlockDriverState *bs,
6451 char *filename, int filename_size)
6454 pstrcpy(filename, filename_size, bs->backing_file);
6457 int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6460 BlockDriver *drv = bs->drv;
6462 assert_bdrv_graph_readable();
6464 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6468 if (!drv->bdrv_co_get_info) {
6469 BlockDriverState *filtered = bdrv_filter_bs(bs);
6471 return bdrv_co_get_info(filtered, bdi);
6475 memset(bdi, 0, sizeof(*bdi));
6476 ret = drv->bdrv_co_get_info(bs, bdi);
6481 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6488 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6491 BlockDriver *drv = bs->drv;
6493 if (drv && drv->bdrv_get_specific_info) {
6494 return drv->bdrv_get_specific_info(bs, errp);
6499 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6501 BlockDriver *drv = bs->drv;
6503 if (!drv || !drv->bdrv_get_specific_stats) {
6506 return drv->bdrv_get_specific_stats(bs);
6509 void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6512 assert_bdrv_graph_readable();
6514 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
6518 bs->drv->bdrv_co_debug_event(bs, event);
6521 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6523 GLOBAL_STATE_CODE();
6524 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6525 bs = bdrv_primary_bs(bs);
6528 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6529 assert(bs->drv->bdrv_debug_remove_breakpoint);
6536 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6539 GLOBAL_STATE_CODE();
6540 bs = bdrv_find_debug_node(bs);
6542 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6548 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6550 GLOBAL_STATE_CODE();
6551 bs = bdrv_find_debug_node(bs);
6553 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6559 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6561 GLOBAL_STATE_CODE();
6562 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6563 bs = bdrv_primary_bs(bs);
6566 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6567 return bs->drv->bdrv_debug_resume(bs, tag);
6573 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6575 GLOBAL_STATE_CODE();
6576 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6577 bs = bdrv_primary_bs(bs);
6580 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6581 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6587 /* backing_file can either be relative, or absolute, or a protocol. If it is
6588 * relative, it must be relative to the chain. So, passing in bs->filename
6589 * from a BDS as backing_file should not be done, as that may be relative to
6590 * the CWD rather than the chain. */
6591 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6592 const char *backing_file)
6594 char *filename_full = NULL;
6595 char *backing_file_full = NULL;
6596 char *filename_tmp = NULL;
6597 int is_protocol = 0;
6598 bool filenames_refreshed = false;
6599 BlockDriverState *curr_bs = NULL;
6600 BlockDriverState *retval = NULL;
6601 BlockDriverState *bs_below;
6603 GLOBAL_STATE_CODE();
6605 if (!bs || !bs->drv || !backing_file) {
6609 filename_full = g_malloc(PATH_MAX);
6610 backing_file_full = g_malloc(PATH_MAX);
6612 is_protocol = path_has_protocol(backing_file);
6615 * Being largely a legacy function, skip any filters here
6616 * (because filters do not have normal filenames, so they cannot
6617 * match anyway; and allowing json:{} filenames is a bit out of
6620 for (curr_bs = bdrv_skip_filters(bs);
6621 bdrv_cow_child(curr_bs) != NULL;
6624 bs_below = bdrv_backing_chain_next(curr_bs);
6626 if (bdrv_backing_overridden(curr_bs)) {
6628 * If the backing file was overridden, we can only compare
6629 * directly against the backing node's filename.
6632 if (!filenames_refreshed) {
6634 * This will automatically refresh all of the
6635 * filenames in the rest of the backing chain, so we
6636 * only need to do this once.
6638 bdrv_refresh_filename(bs_below);
6639 filenames_refreshed = true;
6642 if (strcmp(backing_file, bs_below->filename) == 0) {
6646 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6648 * If either of the filename paths is actually a protocol, then
6649 * compare unmodified paths; otherwise make paths relative.
6651 char *backing_file_full_ret;
6653 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6657 /* Also check against the full backing filename for the image */
6658 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6660 if (backing_file_full_ret) {
6661 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6662 g_free(backing_file_full_ret);
6669 /* If not an absolute filename path, make it relative to the current
6670 * image's filename path */
6671 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6673 /* We are going to compare canonicalized absolute pathnames */
6674 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6675 g_free(filename_tmp);
6678 g_free(filename_tmp);
6680 /* We need to make sure the backing filename we are comparing against
6681 * is relative to the current image filename (or absolute) */
6682 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6683 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6684 g_free(filename_tmp);
6687 g_free(filename_tmp);
6689 if (strcmp(backing_file_full, filename_full) == 0) {
6696 g_free(filename_full);
6697 g_free(backing_file_full);
6701 void bdrv_init(void)
6703 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6704 use_bdrv_whitelist = 1;
6706 module_call_init(MODULE_INIT_BLOCK);
6709 void bdrv_init_with_whitelist(void)
6711 use_bdrv_whitelist = 1;
6715 int bdrv_activate(BlockDriverState *bs, Error **errp)
6717 BdrvChild *child, *parent;
6718 Error *local_err = NULL;
6720 BdrvDirtyBitmap *bm;
6722 GLOBAL_STATE_CODE();
6728 QLIST_FOREACH(child, &bs->children, next) {
6729 bdrv_activate(child->bs, &local_err);
6731 error_propagate(errp, local_err);
6737 * Update permissions, they may differ for inactive nodes.
6739 * Note that the required permissions of inactive images are always a
6740 * subset of the permissions required after activating the image. This
6741 * allows us to just get the permissions upfront without restricting
6742 * bdrv_co_invalidate_cache().
6744 * It also means that in error cases, we don't have to try and revert to
6745 * the old permissions (which is an operation that could fail, too). We can
6746 * just keep the extended permissions for the next time that an activation
6747 * of the image is tried.
6749 if (bs->open_flags & BDRV_O_INACTIVE) {
6750 bs->open_flags &= ~BDRV_O_INACTIVE;
6751 ret = bdrv_refresh_perms(bs, NULL, errp);
6753 bs->open_flags |= BDRV_O_INACTIVE;
6757 ret = bdrv_invalidate_cache(bs, errp);
6759 bs->open_flags |= BDRV_O_INACTIVE;
6763 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6764 bdrv_dirty_bitmap_skip_store(bm, false);
6767 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6769 bs->open_flags |= BDRV_O_INACTIVE;
6770 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6775 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6776 if (parent->klass->activate) {
6777 parent->klass->activate(parent, &local_err);
6779 bs->open_flags |= BDRV_O_INACTIVE;
6780 error_propagate(errp, local_err);
6789 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6791 Error *local_err = NULL;
6794 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6795 assert_bdrv_graph_readable();
6797 if (bs->drv->bdrv_co_invalidate_cache) {
6798 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6800 error_propagate(errp, local_err);
6808 void bdrv_activate_all(Error **errp)
6810 BlockDriverState *bs;
6811 BdrvNextIterator it;
6813 GLOBAL_STATE_CODE();
6815 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6816 AioContext *aio_context = bdrv_get_aio_context(bs);
6819 aio_context_acquire(aio_context);
6820 ret = bdrv_activate(bs, errp);
6821 aio_context_release(aio_context);
6823 bdrv_next_cleanup(&it);
6829 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6832 GLOBAL_STATE_CODE();
6834 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6835 if (parent->klass->parent_is_bds) {
6836 BlockDriverState *parent_bs = parent->opaque;
6837 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6846 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6848 BdrvChild *child, *parent;
6850 uint64_t cumulative_perms, cumulative_shared_perms;
6852 GLOBAL_STATE_CODE();
6858 /* Make sure that we don't inactivate a child before its parent.
6859 * It will be covered by recursion from the yet active parent. */
6860 if (bdrv_has_bds_parent(bs, true)) {
6864 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6866 /* Inactivate this node */
6867 if (bs->drv->bdrv_inactivate) {
6868 ret = bs->drv->bdrv_inactivate(bs);
6874 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6875 if (parent->klass->inactivate) {
6876 ret = parent->klass->inactivate(parent);
6883 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6884 &cumulative_shared_perms);
6885 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6886 /* Our inactive parents still need write access. Inactivation failed. */
6890 bs->open_flags |= BDRV_O_INACTIVE;
6893 * Update permissions, they may differ for inactive nodes.
6894 * We only tried to loosen restrictions, so errors are not fatal, ignore
6897 bdrv_refresh_perms(bs, NULL, NULL);
6899 /* Recursively inactivate children */
6900 QLIST_FOREACH(child, &bs->children, next) {
6901 ret = bdrv_inactivate_recurse(child->bs);
6910 int bdrv_inactivate_all(void)
6912 BlockDriverState *bs = NULL;
6913 BdrvNextIterator it;
6915 GSList *aio_ctxs = NULL, *ctx;
6917 GLOBAL_STATE_CODE();
6919 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6920 AioContext *aio_context = bdrv_get_aio_context(bs);
6922 if (!g_slist_find(aio_ctxs, aio_context)) {
6923 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6924 aio_context_acquire(aio_context);
6928 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6929 /* Nodes with BDS parents are covered by recursion from the last
6930 * parent that gets inactivated. Don't inactivate them a second
6931 * time if that has already happened. */
6932 if (bdrv_has_bds_parent(bs, false)) {
6935 ret = bdrv_inactivate_recurse(bs);
6937 bdrv_next_cleanup(&it);
6943 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6944 AioContext *aio_context = ctx->data;
6945 aio_context_release(aio_context);
6947 g_slist_free(aio_ctxs);
6952 /**************************************************************/
6953 /* removable device support */
6956 * Return TRUE if the media is present
6958 bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
6960 BlockDriver *drv = bs->drv;
6963 assert_bdrv_graph_readable();
6968 if (drv->bdrv_co_is_inserted) {
6969 return drv->bdrv_co_is_inserted(bs);
6971 QLIST_FOREACH(child, &bs->children, next) {
6972 if (!bdrv_co_is_inserted(child->bs)) {
6980 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6982 void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
6984 BlockDriver *drv = bs->drv;
6986 assert_bdrv_graph_readable();
6988 if (drv && drv->bdrv_co_eject) {
6989 drv->bdrv_co_eject(bs, eject_flag);
6994 * Lock or unlock the media (if it is locked, the user won't be able
6995 * to eject it manually).
6997 void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
6999 BlockDriver *drv = bs->drv;
7001 assert_bdrv_graph_readable();
7002 trace_bdrv_lock_medium(bs, locked);
7004 if (drv && drv->bdrv_co_lock_medium) {
7005 drv->bdrv_co_lock_medium(bs, locked);
7009 /* Get a reference to bs */
7010 void bdrv_ref(BlockDriverState *bs)
7012 GLOBAL_STATE_CODE();
7016 /* Release a previously grabbed reference to bs.
7017 * If after releasing, reference count is zero, the BlockDriverState is
7019 void bdrv_unref(BlockDriverState *bs)
7021 GLOBAL_STATE_CODE();
7025 assert(bs->refcnt > 0);
7026 if (--bs->refcnt == 0) {
7031 struct BdrvOpBlocker {
7033 QLIST_ENTRY(BdrvOpBlocker) list;
7036 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7038 BdrvOpBlocker *blocker;
7039 GLOBAL_STATE_CODE();
7040 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7041 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7042 blocker = QLIST_FIRST(&bs->op_blockers[op]);
7043 error_propagate_prepend(errp, error_copy(blocker->reason),
7044 "Node '%s' is busy: ",
7045 bdrv_get_device_or_node_name(bs));
7051 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7053 BdrvOpBlocker *blocker;
7054 GLOBAL_STATE_CODE();
7055 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7057 blocker = g_new0(BdrvOpBlocker, 1);
7058 blocker->reason = reason;
7059 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7062 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7064 BdrvOpBlocker *blocker, *next;
7065 GLOBAL_STATE_CODE();
7066 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7067 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7068 if (blocker->reason == reason) {
7069 QLIST_REMOVE(blocker, list);
7075 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7078 GLOBAL_STATE_CODE();
7079 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7080 bdrv_op_block(bs, i, reason);
7084 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7087 GLOBAL_STATE_CODE();
7088 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7089 bdrv_op_unblock(bs, i, reason);
7093 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7096 GLOBAL_STATE_CODE();
7097 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7098 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7106 * Must not be called while holding the lock of an AioContext other than the
7109 void bdrv_img_create(const char *filename, const char *fmt,
7110 const char *base_filename, const char *base_fmt,
7111 char *options, uint64_t img_size, int flags, bool quiet,
7114 QemuOptsList *create_opts = NULL;
7115 QemuOpts *opts = NULL;
7116 const char *backing_fmt, *backing_file;
7118 BlockDriver *drv, *proto_drv;
7119 Error *local_err = NULL;
7122 GLOBAL_STATE_CODE();
7124 /* Find driver and parse its options */
7125 drv = bdrv_find_format(fmt);
7127 error_setg(errp, "Unknown file format '%s'", fmt);
7131 proto_drv = bdrv_find_protocol(filename, true, errp);
7136 if (!drv->create_opts) {
7137 error_setg(errp, "Format driver '%s' does not support image creation",
7142 if (!proto_drv->create_opts) {
7143 error_setg(errp, "Protocol driver '%s' does not support image creation",
7144 proto_drv->format_name);
7148 aio_context_acquire(qemu_get_aio_context());
7150 /* Create parameter list */
7151 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7152 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7154 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7156 /* Parse -o options */
7158 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7163 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7164 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7165 } else if (img_size != UINT64_C(-1)) {
7166 error_setg(errp, "The image size must be specified only once");
7170 if (base_filename) {
7171 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7173 error_setg(errp, "Backing file not supported for file format '%s'",
7180 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7181 error_setg(errp, "Backing file format not supported for file "
7182 "format '%s'", fmt);
7187 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7189 if (!strcmp(filename, backing_file)) {
7190 error_setg(errp, "Error: Trying to create an image with the "
7191 "same filename as the backing file");
7194 if (backing_file[0] == '\0') {
7195 error_setg(errp, "Expected backing file name, got empty string");
7200 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7202 /* The size for the image must always be specified, unless we have a backing
7203 * file and we have not been forbidden from opening it. */
7204 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7205 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7206 BlockDriverState *bs;
7209 QDict *backing_options = NULL;
7212 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7217 assert(full_backing);
7220 * No need to do I/O here, which allows us to open encrypted
7221 * backing images without needing the secret
7224 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7225 back_flags |= BDRV_O_NO_IO;
7227 backing_options = qdict_new();
7229 qdict_put_str(backing_options, "driver", backing_fmt);
7231 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7233 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7235 g_free(full_backing);
7237 error_append_hint(&local_err, "Could not open backing image.\n");
7241 error_setg(&local_err,
7242 "Backing file specified without backing format");
7243 error_append_hint(&local_err, "Detected format of %s.\n",
7244 bs->drv->format_name);
7248 /* Opened BS, have no size */
7249 size = bdrv_getlength(bs);
7251 error_setg_errno(errp, -size, "Could not get size of '%s'",
7256 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7260 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7261 } else if (backing_file && !backing_fmt) {
7262 error_setg(&local_err,
7263 "Backing file specified without backing format");
7268 error_setg(errp, "Image creation needs a size parameter");
7273 printf("Formatting '%s', fmt=%s ", filename, fmt);
7274 qemu_opts_print(opts, " ");
7279 ret = bdrv_create(drv, filename, opts, &local_err);
7281 if (ret == -EFBIG) {
7282 /* This is generally a better message than whatever the driver would
7283 * deliver (especially because of the cluster_size_hint), since that
7284 * is most probably not much different from "image too large". */
7285 const char *cluster_size_hint = "";
7286 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7287 cluster_size_hint = " (try using a larger cluster size)";
7289 error_setg(errp, "The image size is too large for file format '%s'"
7290 "%s", fmt, cluster_size_hint);
7291 error_free(local_err);
7296 qemu_opts_del(opts);
7297 qemu_opts_free(create_opts);
7298 error_propagate(errp, local_err);
7299 aio_context_release(qemu_get_aio_context());
7302 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7305 return bs ? bs->aio_context : qemu_get_aio_context();
7308 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7310 Coroutine *self = qemu_coroutine_self();
7311 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7312 AioContext *new_ctx;
7316 * Increase bs->in_flight to ensure that this operation is completed before
7317 * moving the node to a different AioContext. Read new_ctx only afterwards.
7319 bdrv_inc_in_flight(bs);
7321 new_ctx = bdrv_get_aio_context(bs);
7322 aio_co_reschedule_self(new_ctx);
7326 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7329 aio_co_reschedule_self(old_ctx);
7330 bdrv_dec_in_flight(bs);
7333 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7335 AioContext *ctx = bdrv_get_aio_context(bs);
7337 /* In the main thread, bs->aio_context won't change concurrently */
7338 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7341 * We're in coroutine context, so we already hold the lock of the main
7342 * loop AioContext. Don't lock it twice to avoid deadlocks.
7344 assert(qemu_in_coroutine());
7345 if (ctx != qemu_get_aio_context()) {
7346 aio_context_acquire(ctx);
7350 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7352 AioContext *ctx = bdrv_get_aio_context(bs);
7354 assert(qemu_in_coroutine());
7355 if (ctx != qemu_get_aio_context()) {
7356 aio_context_release(ctx);
7360 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7362 GLOBAL_STATE_CODE();
7363 QLIST_REMOVE(ban, list);
7367 static void bdrv_detach_aio_context(BlockDriverState *bs)
7369 BdrvAioNotifier *baf, *baf_tmp;
7371 assert(!bs->walking_aio_notifiers);
7372 GLOBAL_STATE_CODE();
7373 bs->walking_aio_notifiers = true;
7374 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7376 bdrv_do_remove_aio_context_notifier(baf);
7378 baf->detach_aio_context(baf->opaque);
7381 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7382 * remove remaining aio notifiers if we aren't called again.
7384 bs->walking_aio_notifiers = false;
7386 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7387 bs->drv->bdrv_detach_aio_context(bs);
7390 bs->aio_context = NULL;
7393 static void bdrv_attach_aio_context(BlockDriverState *bs,
7394 AioContext *new_context)
7396 BdrvAioNotifier *ban, *ban_tmp;
7397 GLOBAL_STATE_CODE();
7399 bs->aio_context = new_context;
7401 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7402 bs->drv->bdrv_attach_aio_context(bs, new_context);
7405 assert(!bs->walking_aio_notifiers);
7406 bs->walking_aio_notifiers = true;
7407 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7409 bdrv_do_remove_aio_context_notifier(ban);
7411 ban->attached_aio_context(new_context, ban->opaque);
7414 bs->walking_aio_notifiers = false;
7417 typedef struct BdrvStateSetAioContext {
7418 AioContext *new_ctx;
7419 BlockDriverState *bs;
7420 } BdrvStateSetAioContext;
7422 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7423 GHashTable *visited,
7427 GLOBAL_STATE_CODE();
7428 if (g_hash_table_contains(visited, c)) {
7431 g_hash_table_add(visited, c);
7434 * A BdrvChildClass that doesn't handle AioContext changes cannot
7435 * tolerate any AioContext changes
7437 if (!c->klass->change_aio_ctx) {
7438 char *user = bdrv_child_user_desc(c);
7439 error_setg(errp, "Changing iothreads is not supported by %s", user);
7443 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7444 assert(!errp || *errp);
7450 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7451 GHashTable *visited, Transaction *tran,
7454 GLOBAL_STATE_CODE();
7455 if (g_hash_table_contains(visited, c)) {
7458 g_hash_table_add(visited, c);
7459 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7462 static void bdrv_set_aio_context_clean(void *opaque)
7464 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7465 BlockDriverState *bs = (BlockDriverState *) state->bs;
7467 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7468 bdrv_drained_end(bs);
7473 static void bdrv_set_aio_context_commit(void *opaque)
7475 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7476 BlockDriverState *bs = (BlockDriverState *) state->bs;
7477 AioContext *new_context = state->new_ctx;
7478 AioContext *old_context = bdrv_get_aio_context(bs);
7481 * Take the old AioContex when detaching it from bs.
7482 * At this point, new_context lock is already acquired, and we are now
7483 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7484 * does not call AIO_POLL_WHILE().
7486 if (old_context != qemu_get_aio_context()) {
7487 aio_context_acquire(old_context);
7489 bdrv_detach_aio_context(bs);
7490 if (old_context != qemu_get_aio_context()) {
7491 aio_context_release(old_context);
7493 bdrv_attach_aio_context(bs, new_context);
7496 static TransactionActionDrv set_aio_context = {
7497 .commit = bdrv_set_aio_context_commit,
7498 .clean = bdrv_set_aio_context_clean,
7502 * Changes the AioContext used for fd handlers, timers, and BHs by this
7503 * BlockDriverState and all its children and parents.
7505 * Must be called from the main AioContext.
7507 * The caller must own the AioContext lock for the old AioContext of bs, but it
7508 * must not own the AioContext lock for new_context (unless new_context is the
7509 * same as the current context of bs).
7511 * @visited will accumulate all visited BdrvChild objects. The caller is
7512 * responsible for freeing the list afterwards.
7514 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7515 GHashTable *visited, Transaction *tran,
7519 BdrvStateSetAioContext *state;
7521 GLOBAL_STATE_CODE();
7523 if (bdrv_get_aio_context(bs) == ctx) {
7527 QLIST_FOREACH(c, &bs->parents, next_parent) {
7528 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7533 QLIST_FOREACH(c, &bs->children, next) {
7534 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7539 state = g_new(BdrvStateSetAioContext, 1);
7540 *state = (BdrvStateSetAioContext) {
7545 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7546 bdrv_drained_begin(bs);
7548 tran_add(tran, &set_aio_context, state);
7554 * Change bs's and recursively all of its parents' and children's AioContext
7555 * to the given new context, returning an error if that isn't possible.
7557 * If ignore_child is not NULL, that child (and its subgraph) will not
7560 * This function still requires the caller to take the bs current
7561 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7562 * assumes the lock is always held if bs is in another AioContext.
7563 * For the same reason, it temporarily also holds the new AioContext, since
7564 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7565 * Therefore the new AioContext lock must not be taken by the caller.
7567 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7568 BdrvChild *ignore_child, Error **errp)
7571 GHashTable *visited;
7573 AioContext *old_context = bdrv_get_aio_context(bs);
7574 GLOBAL_STATE_CODE();
7577 * Recursion phase: go through all nodes of the graph.
7578 * Take care of checking that all nodes support changing AioContext
7579 * and drain them, builing a linear list of callbacks to run if everything
7580 * is successful (the transaction itself).
7583 visited = g_hash_table_new(NULL, NULL);
7585 g_hash_table_add(visited, ignore_child);
7587 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7588 g_hash_table_destroy(visited);
7591 * Linear phase: go through all callbacks collected in the transaction.
7592 * Run all callbacks collected in the recursion to switch all nodes
7593 * AioContext lock (transaction commit), or undo all changes done in the
7594 * recursion (transaction abort).
7598 /* Just run clean() callbacks. No AioContext changed. */
7604 * Release old AioContext, it won't be needed anymore, as all
7605 * bdrv_drained_begin() have been called already.
7607 if (qemu_get_aio_context() != old_context) {
7608 aio_context_release(old_context);
7612 * Acquire new AioContext since bdrv_drained_end() is going to be called
7613 * after we switched all nodes in the new AioContext, and the function
7614 * assumes that the lock of the bs is always taken.
7616 if (qemu_get_aio_context() != ctx) {
7617 aio_context_acquire(ctx);
7622 if (qemu_get_aio_context() != ctx) {
7623 aio_context_release(ctx);
7626 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7627 if (qemu_get_aio_context() != old_context) {
7628 aio_context_acquire(old_context);
7634 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7635 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7636 void (*detach_aio_context)(void *opaque), void *opaque)
7638 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7639 *ban = (BdrvAioNotifier){
7640 .attached_aio_context = attached_aio_context,
7641 .detach_aio_context = detach_aio_context,
7644 GLOBAL_STATE_CODE();
7646 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7649 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7650 void (*attached_aio_context)(AioContext *,
7652 void (*detach_aio_context)(void *),
7655 BdrvAioNotifier *ban, *ban_next;
7656 GLOBAL_STATE_CODE();
7658 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7659 if (ban->attached_aio_context == attached_aio_context &&
7660 ban->detach_aio_context == detach_aio_context &&
7661 ban->opaque == opaque &&
7662 ban->deleted == false)
7664 if (bs->walking_aio_notifiers) {
7665 ban->deleted = true;
7667 bdrv_do_remove_aio_context_notifier(ban);
7676 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7677 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7681 GLOBAL_STATE_CODE();
7683 error_setg(errp, "Node is ejected");
7686 if (!bs->drv->bdrv_amend_options) {
7687 error_setg(errp, "Block driver '%s' does not support option amendment",
7688 bs->drv->format_name);
7691 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7692 cb_opaque, force, errp);
7696 * This function checks whether the given @to_replace is allowed to be
7697 * replaced by a node that always shows the same data as @bs. This is
7698 * used for example to verify whether the mirror job can replace
7699 * @to_replace by the target mirrored from @bs.
7700 * To be replaceable, @bs and @to_replace may either be guaranteed to
7701 * always show the same data (because they are only connected through
7702 * filters), or some driver may allow replacing one of its children
7703 * because it can guarantee that this child's data is not visible at
7704 * all (for example, for dissenting quorum children that have no other
7707 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7708 BlockDriverState *to_replace)
7710 BlockDriverState *filtered;
7712 GLOBAL_STATE_CODE();
7714 if (!bs || !bs->drv) {
7718 if (bs == to_replace) {
7722 /* See what the driver can do */
7723 if (bs->drv->bdrv_recurse_can_replace) {
7724 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7727 /* For filters without an own implementation, we can recurse on our own */
7728 filtered = bdrv_filter_bs(bs);
7730 return bdrv_recurse_can_replace(filtered, to_replace);
7738 * Check whether the given @node_name can be replaced by a node that
7739 * has the same data as @parent_bs. If so, return @node_name's BDS;
7742 * @node_name must be a (recursive) *child of @parent_bs (or this
7743 * function will return NULL).
7745 * The result (whether the node can be replaced or not) is only valid
7746 * for as long as no graph or permission changes occur.
7748 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7749 const char *node_name, Error **errp)
7751 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7752 AioContext *aio_context;
7754 GLOBAL_STATE_CODE();
7756 if (!to_replace_bs) {
7757 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7761 aio_context = bdrv_get_aio_context(to_replace_bs);
7762 aio_context_acquire(aio_context);
7764 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7765 to_replace_bs = NULL;
7769 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7770 * most non filter in order to prevent data corruption.
7771 * Another benefit is that this tests exclude backing files which are
7772 * blocked by the backing blockers.
7774 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7775 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7776 "because it cannot be guaranteed that doing so would not "
7777 "lead to an abrupt change of visible data",
7778 node_name, parent_bs->node_name);
7779 to_replace_bs = NULL;
7784 aio_context_release(aio_context);
7785 return to_replace_bs;
7789 * Iterates through the list of runtime option keys that are said to
7790 * be "strong" for a BDS. An option is called "strong" if it changes
7791 * a BDS's data. For example, the null block driver's "size" and
7792 * "read-zeroes" options are strong, but its "latency-ns" option is
7795 * If a key returned by this function ends with a dot, all options
7796 * starting with that prefix are strong.
7798 static const char *const *strong_options(BlockDriverState *bs,
7799 const char *const *curopt)
7801 static const char *const global_options[] = {
7802 "driver", "filename", NULL
7806 return &global_options[0];
7810 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7811 curopt = bs->drv->strong_runtime_opts;
7814 return (curopt && *curopt) ? curopt : NULL;
7818 * Copies all strong runtime options from bs->options to the given
7819 * QDict. The set of strong option keys is determined by invoking
7822 * Returns true iff any strong option was present in bs->options (and
7823 * thus copied to the target QDict) with the exception of "filename"
7824 * and "driver". The caller is expected to use this value to decide
7825 * whether the existence of strong options prevents the generation of
7828 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7830 bool found_any = false;
7831 const char *const *option_name = NULL;
7837 while ((option_name = strong_options(bs, option_name))) {
7838 bool option_given = false;
7840 assert(strlen(*option_name) > 0);
7841 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7842 QObject *entry = qdict_get(bs->options, *option_name);
7847 qdict_put_obj(d, *option_name, qobject_ref(entry));
7848 option_given = true;
7850 const QDictEntry *entry;
7851 for (entry = qdict_first(bs->options); entry;
7852 entry = qdict_next(bs->options, entry))
7854 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7855 qdict_put_obj(d, qdict_entry_key(entry),
7856 qobject_ref(qdict_entry_value(entry)));
7857 option_given = true;
7862 /* While "driver" and "filename" need to be included in a JSON filename,
7863 * their existence does not prohibit generation of a plain filename. */
7864 if (!found_any && option_given &&
7865 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7871 if (!qdict_haskey(d, "driver")) {
7872 /* Drivers created with bdrv_new_open_driver() may not have a
7873 * @driver option. Add it here. */
7874 qdict_put_str(d, "driver", bs->drv->format_name);
7880 /* Note: This function may return false positives; it may return true
7881 * even if opening the backing file specified by bs's image header
7882 * would result in exactly bs->backing. */
7883 static bool bdrv_backing_overridden(BlockDriverState *bs)
7885 GLOBAL_STATE_CODE();
7887 return strcmp(bs->auto_backing_file,
7888 bs->backing->bs->filename);
7890 /* No backing BDS, so if the image header reports any backing
7891 * file, it must have been suppressed */
7892 return bs->auto_backing_file[0] != '\0';
7896 /* Updates the following BDS fields:
7897 * - exact_filename: A filename which may be used for opening a block device
7898 * which (mostly) equals the given BDS (even without any
7899 * other options; so reading and writing must return the same
7900 * results, but caching etc. may be different)
7901 * - full_open_options: Options which, when given when opening a block device
7902 * (without a filename), result in a BDS (mostly)
7903 * equalling the given one
7904 * - filename: If exact_filename is set, it is copied here. Otherwise,
7905 * full_open_options is converted to a JSON object, prefixed with
7906 * "json:" (for use through the JSON pseudo protocol) and put here.
7908 void bdrv_refresh_filename(BlockDriverState *bs)
7910 BlockDriver *drv = bs->drv;
7912 BlockDriverState *primary_child_bs;
7914 bool backing_overridden;
7915 bool generate_json_filename; /* Whether our default implementation should
7916 fill exact_filename (false) or not (true) */
7918 GLOBAL_STATE_CODE();
7924 /* This BDS's file name may depend on any of its children's file names, so
7925 * refresh those first */
7926 QLIST_FOREACH(child, &bs->children, next) {
7927 bdrv_refresh_filename(child->bs);
7931 /* For implicit nodes, just copy everything from the single child */
7932 child = QLIST_FIRST(&bs->children);
7933 assert(QLIST_NEXT(child, next) == NULL);
7935 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7936 child->bs->exact_filename);
7937 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7939 qobject_unref(bs->full_open_options);
7940 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7945 backing_overridden = bdrv_backing_overridden(bs);
7947 if (bs->open_flags & BDRV_O_NO_IO) {
7948 /* Without I/O, the backing file does not change anything.
7949 * Therefore, in such a case (primarily qemu-img), we can
7950 * pretend the backing file has not been overridden even if
7951 * it technically has been. */
7952 backing_overridden = false;
7955 /* Gather the options QDict */
7957 generate_json_filename = append_strong_runtime_options(opts, bs);
7958 generate_json_filename |= backing_overridden;
7960 if (drv->bdrv_gather_child_options) {
7961 /* Some block drivers may not want to present all of their children's
7962 * options, or name them differently from BdrvChild.name */
7963 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7965 QLIST_FOREACH(child, &bs->children, next) {
7966 if (child == bs->backing && !backing_overridden) {
7967 /* We can skip the backing BDS if it has not been overridden */
7971 qdict_put(opts, child->name,
7972 qobject_ref(child->bs->full_open_options));
7975 if (backing_overridden && !bs->backing) {
7976 /* Force no backing file */
7977 qdict_put_null(opts, "backing");
7981 qobject_unref(bs->full_open_options);
7982 bs->full_open_options = opts;
7984 primary_child_bs = bdrv_primary_bs(bs);
7986 if (drv->bdrv_refresh_filename) {
7987 /* Obsolete information is of no use here, so drop the old file name
7988 * information before refreshing it */
7989 bs->exact_filename[0] = '\0';
7991 drv->bdrv_refresh_filename(bs);
7992 } else if (primary_child_bs) {
7994 * Try to reconstruct valid information from the underlying
7995 * file -- this only works for format nodes (filter nodes
7996 * cannot be probed and as such must be selected by the user
7997 * either through an options dict, or through a special
7998 * filename which the filter driver must construct in its
7999 * .bdrv_refresh_filename() implementation).
8002 bs->exact_filename[0] = '\0';
8005 * We can use the underlying file's filename if:
8006 * - it has a filename,
8007 * - the current BDS is not a filter,
8008 * - the file is a protocol BDS, and
8009 * - opening that file (as this BDS's format) will automatically create
8010 * the BDS tree we have right now, that is:
8011 * - the user did not significantly change this BDS's behavior with
8012 * some explicit (strong) options
8013 * - no non-file child of this BDS has been overridden by the user
8014 * Both of these conditions are represented by generate_json_filename.
8016 if (primary_child_bs->exact_filename[0] &&
8017 primary_child_bs->drv->bdrv_file_open &&
8018 !drv->is_filter && !generate_json_filename)
8020 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
8024 if (bs->exact_filename[0]) {
8025 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
8027 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
8028 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
8029 json->str) >= sizeof(bs->filename)) {
8030 /* Give user a hint if we truncated things. */
8031 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8033 g_string_free(json, true);
8037 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8039 BlockDriver *drv = bs->drv;
8040 BlockDriverState *child_bs;
8042 GLOBAL_STATE_CODE();
8045 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8049 if (drv->bdrv_dirname) {
8050 return drv->bdrv_dirname(bs, errp);
8053 child_bs = bdrv_primary_bs(bs);
8055 return bdrv_dirname(child_bs, errp);
8058 bdrv_refresh_filename(bs);
8059 if (bs->exact_filename[0] != '\0') {
8060 return path_combine(bs->exact_filename, "");
8063 error_setg(errp, "Cannot generate a base directory for %s nodes",
8069 * Hot add/remove a BDS's child. So the user can take a child offline when
8070 * it is broken and take a new child online
8072 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8075 GLOBAL_STATE_CODE();
8076 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8077 error_setg(errp, "The node %s does not support adding a child",
8078 bdrv_get_device_or_node_name(parent_bs));
8083 * Non-zoned block drivers do not follow zoned storage constraints
8084 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8085 * drivers in a graph.
8087 if (!parent_bs->drv->supports_zoned_children &&
8088 child_bs->bl.zoned == BLK_Z_HM) {
8090 * The host-aware model allows zoned storage constraints and random
8091 * write. Allow mixing host-aware and non-zoned drivers. Using
8092 * host-aware device as a regular device.
8094 error_setg(errp, "Cannot add a %s child to a %s parent",
8095 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8096 parent_bs->drv->supports_zoned_children ?
8097 "support zoned children" : "not support zoned children");
8101 if (!QLIST_EMPTY(&child_bs->parents)) {
8102 error_setg(errp, "The node %s already has a parent",
8103 child_bs->node_name);
8107 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8110 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8114 GLOBAL_STATE_CODE();
8115 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8116 error_setg(errp, "The node %s does not support removing a child",
8117 bdrv_get_device_or_node_name(parent_bs));
8121 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8128 error_setg(errp, "The node %s does not have a child named %s",
8129 bdrv_get_device_or_node_name(parent_bs),
8130 bdrv_get_device_or_node_name(child->bs));
8134 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8137 int bdrv_make_empty(BdrvChild *c, Error **errp)
8139 BlockDriver *drv = c->bs->drv;
8142 GLOBAL_STATE_CODE();
8143 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8145 if (!drv->bdrv_make_empty) {
8146 error_setg(errp, "%s does not support emptying nodes",
8151 ret = drv->bdrv_make_empty(c->bs);
8153 error_setg_errno(errp, -ret, "Failed to empty %s",
8162 * Return the child that @bs acts as an overlay for, and from which data may be
8163 * copied in COW or COR operations. Usually this is the backing file.
8165 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8169 if (!bs || !bs->drv) {
8173 if (bs->drv->is_filter) {
8181 assert(bs->backing->role & BDRV_CHILD_COW);
8186 * If @bs acts as a filter for exactly one of its children, return
8189 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8194 if (!bs || !bs->drv) {
8198 if (!bs->drv->is_filter) {
8202 /* Only one of @backing or @file may be used */
8203 assert(!(bs->backing && bs->file));
8205 c = bs->backing ?: bs->file;
8210 assert(c->role & BDRV_CHILD_FILTERED);
8215 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8216 * whichever is non-NULL.
8218 * Return NULL if both are NULL.
8220 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8222 BdrvChild *cow_child = bdrv_cow_child(bs);
8223 BdrvChild *filter_child = bdrv_filter_child(bs);
8226 /* Filter nodes cannot have COW backing files */
8227 assert(!(cow_child && filter_child));
8229 return cow_child ?: filter_child;
8233 * Return the primary child of this node: For filters, that is the
8234 * filtered child. For other nodes, that is usually the child storing
8236 * (A generally more helpful description is that this is (usually) the
8237 * child that has the same filename as @bs.)
8239 * Drivers do not necessarily have a primary child; for example quorum
8242 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8244 BdrvChild *c, *found = NULL;
8247 QLIST_FOREACH(c, &bs->children, next) {
8248 if (c->role & BDRV_CHILD_PRIMARY) {
8257 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8258 bool stop_on_explicit_filter)
8266 while (!(stop_on_explicit_filter && !bs->implicit)) {
8267 c = bdrv_filter_child(bs);
8270 * A filter that is embedded in a working block graph must
8271 * have a child. Assert this here so this function does
8272 * not return a filter node that is not expected by the
8275 assert(!bs->drv || !bs->drv->is_filter);
8281 * Note that this treats nodes with bs->drv == NULL as not being
8282 * filters (bs->drv == NULL should be replaced by something else
8284 * The advantage of this behavior is that this function will thus
8285 * always return a non-NULL value (given a non-NULL @bs).
8292 * Return the first BDS that has not been added implicitly or that
8293 * does not have a filtered child down the chain starting from @bs
8294 * (including @bs itself).
8296 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8298 GLOBAL_STATE_CODE();
8299 return bdrv_do_skip_filters(bs, true);
8303 * Return the first BDS that does not have a filtered child down the
8304 * chain starting from @bs (including @bs itself).
8306 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8309 return bdrv_do_skip_filters(bs, false);
8313 * For a backing chain, return the first non-filter backing image of
8314 * the first non-filter image.
8316 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8319 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8323 * Check whether [offset, offset + bytes) overlaps with the cached
8324 * block-status data region.
8326 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8327 * which is what bdrv_bsc_is_data()'s interface needs.
8328 * Otherwise, *pnum is not touched.
8330 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8331 int64_t offset, int64_t bytes,
8334 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8338 qatomic_read(&bsc->valid) &&
8339 ranges_overlap(offset, bytes, bsc->data_start,
8340 bsc->data_end - bsc->data_start);
8342 if (overlaps && pnum) {
8343 *pnum = bsc->data_end - offset;
8350 * See block_int.h for this function's documentation.
8352 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8355 RCU_READ_LOCK_GUARD();
8356 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8360 * See block_int.h for this function's documentation.
8362 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8363 int64_t offset, int64_t bytes)
8366 RCU_READ_LOCK_GUARD();
8368 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8369 qatomic_set(&bs->block_status_cache->valid, false);
8374 * See block_int.h for this function's documentation.
8376 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8378 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8379 BdrvBlockStatusCache *old_bsc;
8382 *new_bsc = (BdrvBlockStatusCache) {
8384 .data_start = offset,
8385 .data_end = offset + bytes,
8388 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8390 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8391 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8393 g_free_rcu(old_bsc, rcu);