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/fuse.h"
31 #include "block/nbd.h"
32 #include "block/qdict.h"
33 #include "qemu/error-report.h"
34 #include "block/module_block.h"
35 #include "qemu/main-loop.h"
36 #include "qemu/module.h"
37 #include "qapi/error.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qjson.h"
40 #include "qapi/qmp/qnull.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/qobject-output-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "sysemu/block-backend.h"
45 #include "qemu/notify.h"
46 #include "qemu/option.h"
47 #include "qemu/coroutine.h"
48 #include "block/qapi.h"
49 #include "qemu/timer.h"
50 #include "qemu/cutils.h"
52 #include "qemu/range.h"
54 #include "block/coroutines.h"
57 #include <sys/ioctl.h>
58 #include <sys/queue.h>
59 #if defined(HAVE_SYS_DISK_H)
68 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70 /* Protected by BQL */
71 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
72 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
74 /* Protected by BQL */
75 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
76 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
78 /* Protected by BQL */
79 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
80 QLIST_HEAD_INITIALIZER(bdrv_drivers);
82 static BlockDriverState *bdrv_open_inherit(const char *filename,
83 const char *reference,
84 QDict *options, int flags,
85 BlockDriverState *parent,
86 const BdrvChildClass *child_class,
87 BdrvChildRole child_role,
90 static bool bdrv_recurse_has_child(BlockDriverState *bs,
91 BlockDriverState *child);
93 static void bdrv_replace_child_noperm(BdrvChild *child,
94 BlockDriverState *new_bs);
95 static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
96 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
99 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
100 BlockReopenQueue *queue,
101 Transaction *change_child_tran, Error **errp);
102 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
103 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
105 static bool bdrv_backing_overridden(BlockDriverState *bs);
107 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
108 GHashTable *visited, Transaction *tran,
111 /* If non-zero, use only whitelisted block drivers */
112 static int use_bdrv_whitelist;
115 static int is_windows_drive_prefix(const char *filename)
117 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
118 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
122 int is_windows_drive(const char *filename)
124 if (is_windows_drive_prefix(filename) &&
127 if (strstart(filename, "\\\\.\\", NULL) ||
128 strstart(filename, "//./", NULL))
134 size_t bdrv_opt_mem_align(BlockDriverState *bs)
136 if (!bs || !bs->drv) {
137 /* page size or 4k (hdd sector size) should be on the safe side */
138 return MAX(4096, qemu_real_host_page_size());
142 return bs->bl.opt_mem_alignment;
145 size_t bdrv_min_mem_align(BlockDriverState *bs)
147 if (!bs || !bs->drv) {
148 /* page size or 4k (hdd sector size) should be on the safe side */
149 return MAX(4096, qemu_real_host_page_size());
153 return bs->bl.min_mem_alignment;
156 /* check if the path starts with "<protocol>:" */
157 int path_has_protocol(const char *path)
162 if (is_windows_drive(path) ||
163 is_windows_drive_prefix(path)) {
166 p = path + strcspn(path, ":/\\");
168 p = path + strcspn(path, ":/");
174 int path_is_absolute(const char *path)
177 /* specific case for names like: "\\.\d:" */
178 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
181 return (*path == '/' || *path == '\\');
183 return (*path == '/');
187 /* if filename is absolute, just return its duplicate. Otherwise, build a
188 path to it by considering it is relative to base_path. URL are
190 char *path_combine(const char *base_path, const char *filename)
192 const char *protocol_stripped = NULL;
197 if (path_is_absolute(filename)) {
198 return g_strdup(filename);
201 if (path_has_protocol(base_path)) {
202 protocol_stripped = strchr(base_path, ':');
203 if (protocol_stripped) {
207 p = protocol_stripped ?: base_path;
209 p1 = strrchr(base_path, '/');
213 p2 = strrchr(base_path, '\\');
214 if (!p1 || p2 > p1) {
229 result = g_malloc(len + strlen(filename) + 1);
230 memcpy(result, base_path, len);
231 strcpy(result + len, filename);
237 * Helper function for bdrv_parse_filename() implementations to remove optional
238 * protocol prefixes (especially "file:") from a filename and for putting the
239 * stripped filename into the options QDict if there is such a prefix.
241 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
244 if (strstart(filename, prefix, &filename)) {
245 /* Stripping the explicit protocol prefix may result in a protocol
246 * prefix being (wrongly) detected (if the filename contains a colon) */
247 if (path_has_protocol(filename)) {
248 GString *fat_filename;
250 /* This means there is some colon before the first slash; therefore,
251 * this cannot be an absolute path */
252 assert(!path_is_absolute(filename));
254 /* And we can thus fix the protocol detection issue by prefixing it
256 fat_filename = g_string_new("./");
257 g_string_append(fat_filename, filename);
259 assert(!path_has_protocol(fat_filename->str));
261 qdict_put(options, "filename",
262 qstring_from_gstring(fat_filename));
264 /* If no protocol prefix was detected, we can use the shortened
266 qdict_put_str(options, "filename", filename);
272 /* Returns whether the image file is opened as read-only. Note that this can
273 * return false and writing to the image file is still not possible because the
274 * image is inactivated. */
275 bool bdrv_is_read_only(BlockDriverState *bs)
278 return !(bs->open_flags & BDRV_O_RDWR);
281 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
282 bool ignore_allow_rdw, Error **errp)
286 /* Do not set read_only if copy_on_read is enabled */
287 if (bs->copy_on_read && read_only) {
288 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
289 bdrv_get_device_or_node_name(bs));
293 /* Do not clear read_only if it is prohibited */
294 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
297 error_setg(errp, "Node '%s' is read only",
298 bdrv_get_device_or_node_name(bs));
306 * Called by a driver that can only provide a read-only image.
308 * Returns 0 if the node is already read-only or it could switch the node to
309 * read-only because BDRV_O_AUTO_RDONLY is set.
311 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
312 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
313 * is not NULL, it is used as the error message for the Error object.
315 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
321 if (!(bs->open_flags & BDRV_O_RDWR)) {
324 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
328 ret = bdrv_can_set_read_only(bs, true, false, NULL);
333 bs->open_flags &= ~BDRV_O_RDWR;
338 error_setg(errp, "%s", errmsg ?: "Image is read-only");
343 * If @backing is empty, this function returns NULL without setting
344 * @errp. In all other cases, NULL will only be returned with @errp
347 * Therefore, a return value of NULL without @errp set means that
348 * there is no backing file; if @errp is set, there is one but its
349 * absolute filename cannot be generated.
351 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
355 if (backing[0] == '\0') {
357 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
358 return g_strdup(backing);
359 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
360 error_setg(errp, "Cannot use relative backing file names for '%s'",
364 return path_combine(backed, backing);
369 * If @filename is empty or NULL, this function returns NULL without
370 * setting @errp. In all other cases, NULL will only be returned with
373 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
374 const char *filename, Error **errp)
376 char *dir, *full_name;
378 if (!filename || filename[0] == '\0') {
380 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
381 return g_strdup(filename);
384 dir = bdrv_dirname(relative_to, errp);
389 full_name = g_strconcat(dir, filename, NULL);
394 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
397 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
400 void bdrv_register(BlockDriver *bdrv)
402 assert(bdrv->format_name);
404 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
407 BlockDriverState *bdrv_new(void)
409 BlockDriverState *bs;
414 bs = g_new0(BlockDriverState, 1);
415 QLIST_INIT(&bs->dirty_bitmaps);
416 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
417 QLIST_INIT(&bs->op_blockers[i]);
419 qemu_co_mutex_init(&bs->reqs_lock);
420 qemu_mutex_init(&bs->dirty_bitmap_mutex);
422 bs->aio_context = qemu_get_aio_context();
424 qemu_co_queue_init(&bs->flush_queue);
426 qemu_co_mutex_init(&bs->bsc_modify_lock);
427 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
429 for (i = 0; i < bdrv_drain_all_count; i++) {
430 bdrv_drained_begin(bs);
433 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
438 static BlockDriver *bdrv_do_find_format(const char *format_name)
443 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
444 if (!strcmp(drv1->format_name, format_name)) {
452 BlockDriver *bdrv_find_format(const char *format_name)
459 drv1 = bdrv_do_find_format(format_name);
464 /* The driver isn't registered, maybe we need to load a module */
465 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
466 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
467 block_module_load_one(block_driver_modules[i].library_name);
472 return bdrv_do_find_format(format_name);
475 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
477 static const char *whitelist_rw[] = {
478 CONFIG_BDRV_RW_WHITELIST
481 static const char *whitelist_ro[] = {
482 CONFIG_BDRV_RO_WHITELIST
487 if (!whitelist_rw[0] && !whitelist_ro[0]) {
488 return 1; /* no whitelist, anything goes */
491 for (p = whitelist_rw; *p; p++) {
492 if (!strcmp(format_name, *p)) {
497 for (p = whitelist_ro; *p; p++) {
498 if (!strcmp(format_name, *p)) {
506 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
509 return bdrv_format_is_whitelisted(drv->format_name, read_only);
512 bool bdrv_uses_whitelist(void)
514 return use_bdrv_whitelist;
517 typedef struct CreateCo {
525 static void coroutine_fn bdrv_create_co_entry(void *opaque)
527 Error *local_err = NULL;
530 CreateCo *cco = opaque;
534 ret = cco->drv->bdrv_co_create_opts(cco->drv,
535 cco->filename, cco->opts, &local_err);
536 error_propagate(&cco->err, local_err);
540 int bdrv_create(BlockDriver *drv, const char* filename,
541 QemuOpts *opts, Error **errp)
550 .filename = g_strdup(filename),
556 if (!drv->bdrv_co_create_opts) {
557 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
562 if (qemu_in_coroutine()) {
563 /* Fast-path if already in coroutine context */
564 bdrv_create_co_entry(&cco);
566 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
567 qemu_coroutine_enter(co);
568 while (cco.ret == NOT_DONE) {
569 aio_poll(qemu_get_aio_context(), true);
576 error_propagate(errp, cco.err);
578 error_setg_errno(errp, -ret, "Could not create image");
583 g_free(cco.filename);
588 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
589 * least the given @minimum_size.
591 * On success, return @blk's actual length.
592 * Otherwise, return -errno.
594 static int64_t create_file_fallback_truncate(BlockBackend *blk,
595 int64_t minimum_size, Error **errp)
597 Error *local_err = NULL;
603 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
605 if (ret < 0 && ret != -ENOTSUP) {
606 error_propagate(errp, local_err);
610 size = blk_getlength(blk);
612 error_free(local_err);
613 error_setg_errno(errp, -size,
614 "Failed to inquire the new image file's length");
618 if (size < minimum_size) {
619 /* Need to grow the image, but we failed to do that */
620 error_propagate(errp, local_err);
624 error_free(local_err);
631 * Helper function for bdrv_create_file_fallback(): Zero the first
632 * sector to remove any potentially pre-existing image header.
634 static int coroutine_fn
635 create_file_fallback_zero_first_sector(BlockBackend *blk,
636 int64_t current_size,
639 int64_t bytes_to_clear;
644 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
645 if (bytes_to_clear) {
646 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
648 error_setg_errno(errp, -ret,
649 "Failed to clear the new image's first sector");
658 * Simple implementation of bdrv_co_create_opts for protocol drivers
659 * which only support creation via opening a file
660 * (usually existing raw storage device)
662 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
663 const char *filename,
671 PreallocMode prealloc;
672 Error *local_err = NULL;
677 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
678 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
679 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
680 PREALLOC_MODE_OFF, &local_err);
683 error_propagate(errp, local_err);
687 if (prealloc != PREALLOC_MODE_OFF) {
688 error_setg(errp, "Unsupported preallocation mode '%s'",
689 PreallocMode_str(prealloc));
693 options = qdict_new();
694 qdict_put_str(options, "driver", drv->format_name);
696 blk = blk_new_open(filename, NULL, options,
697 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
699 error_prepend(errp, "Protocol driver '%s' does not support image "
700 "creation, and opening the image failed: ",
705 size = create_file_fallback_truncate(blk, size, errp);
711 ret = create_file_fallback_zero_first_sector(blk, size, errp);
722 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
724 QemuOpts *protocol_opts;
731 drv = bdrv_find_protocol(filename, true, errp);
736 if (!drv->create_opts) {
737 error_setg(errp, "Driver '%s' does not support image creation",
743 * 'opts' contains a QemuOptsList with a combination of format and protocol
746 * The format properly removes its options, but the default values remain
747 * in 'opts->list'. So if the protocol has options with the same name
748 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
749 * of the format, since for overlapping options, the format wins.
751 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
752 * only the set options, and then convert it back to QemuOpts, using the
753 * create_opts of the protocol. So the new QemuOpts, will contain only the
756 qdict = qemu_opts_to_qdict(opts, NULL);
757 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
758 if (protocol_opts == NULL) {
763 ret = bdrv_create(drv, filename, protocol_opts, errp);
765 qemu_opts_del(protocol_opts);
766 qobject_unref(qdict);
770 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
772 Error *local_err = NULL;
779 error_setg(errp, "Block node '%s' is not opened", bs->filename);
783 if (!bs->drv->bdrv_co_delete_file) {
784 error_setg(errp, "Driver '%s' does not support image deletion",
785 bs->drv->format_name);
789 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
791 error_propagate(errp, local_err);
797 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
799 Error *local_err = NULL;
807 ret = bdrv_co_delete_file(bs, &local_err);
809 * ENOTSUP will happen if the block driver doesn't support
810 * the 'bdrv_co_delete_file' interface. This is a predictable
811 * scenario and shouldn't be reported back to the user.
813 if (ret == -ENOTSUP) {
814 error_free(local_err);
815 } else if (ret < 0) {
816 error_report_err(local_err);
821 * Try to get @bs's logical and physical block size.
822 * On success, store them in @bsz struct and return 0.
823 * On failure return -errno.
824 * @bs must not be empty.
826 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
828 BlockDriver *drv = bs->drv;
829 BlockDriverState *filtered = bdrv_filter_bs(bs);
832 if (drv && drv->bdrv_probe_blocksizes) {
833 return drv->bdrv_probe_blocksizes(bs, bsz);
834 } else if (filtered) {
835 return bdrv_probe_blocksizes(filtered, bsz);
842 * Try to get @bs's geometry (cyls, heads, sectors).
843 * On success, store them in @geo struct and return 0.
844 * On failure return -errno.
845 * @bs must not be empty.
847 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
849 BlockDriver *drv = bs->drv;
850 BlockDriverState *filtered = bdrv_filter_bs(bs);
853 if (drv && drv->bdrv_probe_geometry) {
854 return drv->bdrv_probe_geometry(bs, geo);
855 } else if (filtered) {
856 return bdrv_probe_geometry(filtered, geo);
863 * Create a uniquely-named empty temporary file.
864 * Return the actual file name used upon success, otherwise NULL.
865 * This string should be freed with g_free() when not needed any longer.
867 * Note: creating a temporary file for the caller to (re)open is
868 * inherently racy. Use g_file_open_tmp() instead whenever practical.
870 char *create_tmp_file(Error **errp)
874 g_autofree char *filename = NULL;
876 tmpdir = g_get_tmp_dir();
879 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
881 * This function is used to create temporary disk images (like -snapshot),
882 * so the files can become very large. /tmp is often a tmpfs where as
883 * /var/tmp is usually on a disk, so more appropriate for disk images.
885 if (!g_strcmp0(tmpdir, "/tmp")) {
890 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
891 fd = g_mkstemp(filename);
893 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
899 return g_steal_pointer(&filename);
903 * Detect host devices. By convention, /dev/cdrom[N] is always
904 * recognized as a host CDROM.
906 static BlockDriver *find_hdev_driver(const char *filename)
908 int score_max = 0, score;
909 BlockDriver *drv = NULL, *d;
912 QLIST_FOREACH(d, &bdrv_drivers, list) {
913 if (d->bdrv_probe_device) {
914 score = d->bdrv_probe_device(filename);
915 if (score > score_max) {
925 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
930 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
931 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
939 BlockDriver *bdrv_find_protocol(const char *filename,
940 bool allow_protocol_prefix,
950 /* TODO Drivers without bdrv_file_open must be specified explicitly */
953 * XXX(hch): we really should not let host device detection
954 * override an explicit protocol specification, but moving this
955 * later breaks access to device names with colons in them.
956 * Thanks to the brain-dead persistent naming schemes on udev-
957 * based Linux systems those actually are quite common.
959 drv1 = find_hdev_driver(filename);
964 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
968 p = strchr(filename, ':');
971 if (len > sizeof(protocol) - 1)
972 len = sizeof(protocol) - 1;
973 memcpy(protocol, filename, len);
974 protocol[len] = '\0';
976 drv1 = bdrv_do_find_protocol(protocol);
981 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
982 if (block_driver_modules[i].protocol_name &&
983 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
984 block_module_load_one(block_driver_modules[i].library_name);
989 drv1 = bdrv_do_find_protocol(protocol);
991 error_setg(errp, "Unknown protocol '%s'", protocol);
997 * Guess image format by probing its contents.
998 * This is not a good idea when your image is raw (CVE-2008-2004), but
999 * we do it anyway for backward compatibility.
1001 * @buf contains the image's first @buf_size bytes.
1002 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
1003 * but can be smaller if the image file is smaller)
1004 * @filename is its filename.
1006 * For all block drivers, call the bdrv_probe() method to get its
1008 * Return the first block driver with the highest probing score.
1010 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
1011 const char *filename)
1013 int score_max = 0, score;
1014 BlockDriver *drv = NULL, *d;
1017 QLIST_FOREACH(d, &bdrv_drivers, list) {
1018 if (d->bdrv_probe) {
1019 score = d->bdrv_probe(buf, buf_size, filename);
1020 if (score > score_max) {
1030 static int find_image_format(BlockBackend *file, const char *filename,
1031 BlockDriver **pdrv, Error **errp)
1034 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1037 GLOBAL_STATE_CODE();
1039 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1040 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1045 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1047 error_setg_errno(errp, -ret, "Could not read image for determining its "
1053 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1055 error_setg(errp, "Could not determine image format: No compatible "
1066 * Set the current 'total_sectors' value
1067 * Return 0 on success, -errno on error.
1069 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
1071 BlockDriver *drv = bs->drv;
1078 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1082 /* query actual device if possible, otherwise just trust the hint */
1083 if (drv->bdrv_getlength) {
1084 int64_t length = drv->bdrv_getlength(bs);
1088 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1091 bs->total_sectors = hint;
1093 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1101 * Combines a QDict of new block driver @options with any missing options taken
1102 * from @old_options, so that leaving out an option defaults to its old value.
1104 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1107 GLOBAL_STATE_CODE();
1108 if (bs->drv && bs->drv->bdrv_join_options) {
1109 bs->drv->bdrv_join_options(options, old_options);
1111 qdict_join(options, old_options, false);
1115 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1119 Error *local_err = NULL;
1120 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1121 BlockdevDetectZeroesOptions detect_zeroes =
1122 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1123 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1124 GLOBAL_STATE_CODE();
1127 error_propagate(errp, local_err);
1128 return detect_zeroes;
1131 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1132 !(open_flags & BDRV_O_UNMAP))
1134 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1135 "without setting discard operation to unmap");
1138 return detect_zeroes;
1142 * Set open flags for aio engine
1144 * Return 0 on success, -1 if the engine specified is invalid
1146 int bdrv_parse_aio(const char *mode, int *flags)
1148 if (!strcmp(mode, "threads")) {
1149 /* do nothing, default */
1150 } else if (!strcmp(mode, "native")) {
1151 *flags |= BDRV_O_NATIVE_AIO;
1152 #ifdef CONFIG_LINUX_IO_URING
1153 } else if (!strcmp(mode, "io_uring")) {
1154 *flags |= BDRV_O_IO_URING;
1164 * Set open flags for a given discard mode
1166 * Return 0 on success, -1 if the discard mode was invalid.
1168 int bdrv_parse_discard_flags(const char *mode, int *flags)
1170 *flags &= ~BDRV_O_UNMAP;
1172 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1174 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1175 *flags |= BDRV_O_UNMAP;
1184 * Set open flags for a given cache mode
1186 * Return 0 on success, -1 if the cache mode was invalid.
1188 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1190 *flags &= ~BDRV_O_CACHE_MASK;
1192 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1193 *writethrough = false;
1194 *flags |= BDRV_O_NOCACHE;
1195 } else if (!strcmp(mode, "directsync")) {
1196 *writethrough = true;
1197 *flags |= BDRV_O_NOCACHE;
1198 } else if (!strcmp(mode, "writeback")) {
1199 *writethrough = false;
1200 } else if (!strcmp(mode, "unsafe")) {
1201 *writethrough = false;
1202 *flags |= BDRV_O_NO_FLUSH;
1203 } else if (!strcmp(mode, "writethrough")) {
1204 *writethrough = true;
1212 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1214 BlockDriverState *parent = c->opaque;
1215 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1218 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1220 BlockDriverState *bs = child->opaque;
1221 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1224 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1226 BlockDriverState *bs = child->opaque;
1227 return bdrv_drain_poll(bs, false, NULL, false);
1230 static void bdrv_child_cb_drained_end(BdrvChild *child,
1231 int *drained_end_counter)
1233 BlockDriverState *bs = child->opaque;
1234 bdrv_drained_end_no_poll(bs, drained_end_counter);
1237 static int bdrv_child_cb_inactivate(BdrvChild *child)
1239 BlockDriverState *bs = child->opaque;
1240 GLOBAL_STATE_CODE();
1241 assert(bs->open_flags & BDRV_O_INACTIVE);
1245 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1246 GHashTable *visited, Transaction *tran,
1249 BlockDriverState *bs = child->opaque;
1250 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1254 * Returns the options and flags that a temporary snapshot should get, based on
1255 * the originally requested flags (the originally requested image will have
1256 * flags like a backing file)
1258 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1259 int parent_flags, QDict *parent_options)
1261 GLOBAL_STATE_CODE();
1262 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1264 /* For temporary files, unconditional cache=unsafe is fine */
1265 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1266 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1268 /* Copy the read-only and discard options from the parent */
1269 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1270 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1272 /* aio=native doesn't work for cache.direct=off, so disable it for the
1273 * temporary snapshot */
1274 *child_flags &= ~BDRV_O_NATIVE_AIO;
1277 static void bdrv_backing_attach(BdrvChild *c)
1279 BlockDriverState *parent = c->opaque;
1280 BlockDriverState *backing_hd = c->bs;
1282 GLOBAL_STATE_CODE();
1283 assert(!parent->backing_blocker);
1284 error_setg(&parent->backing_blocker,
1285 "node is used as backing hd of '%s'",
1286 bdrv_get_device_or_node_name(parent));
1288 bdrv_refresh_filename(backing_hd);
1290 parent->open_flags &= ~BDRV_O_NO_BACKING;
1292 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1293 /* Otherwise we won't be able to commit or stream */
1294 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1295 parent->backing_blocker);
1296 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1297 parent->backing_blocker);
1299 * We do backup in 3 ways:
1301 * The target bs is new opened, and the source is top BDS
1302 * 2. blockdev backup
1303 * Both the source and the target are top BDSes.
1304 * 3. internal backup(used for block replication)
1305 * Both the source and the target are backing file
1307 * In case 1 and 2, neither the source nor the target is the backing file.
1308 * In case 3, we will block the top BDS, so there is only one block job
1309 * for the top BDS and its backing chain.
1311 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1312 parent->backing_blocker);
1313 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1314 parent->backing_blocker);
1317 static void bdrv_backing_detach(BdrvChild *c)
1319 BlockDriverState *parent = c->opaque;
1321 GLOBAL_STATE_CODE();
1322 assert(parent->backing_blocker);
1323 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1324 error_free(parent->backing_blocker);
1325 parent->backing_blocker = NULL;
1328 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1329 const char *filename, Error **errp)
1331 BlockDriverState *parent = c->opaque;
1332 bool read_only = bdrv_is_read_only(parent);
1334 GLOBAL_STATE_CODE();
1337 ret = bdrv_reopen_set_read_only(parent, false, errp);
1343 ret = bdrv_change_backing_file(parent, filename,
1344 base->drv ? base->drv->format_name : "",
1347 error_setg_errno(errp, -ret, "Could not update backing file link");
1351 bdrv_reopen_set_read_only(parent, true, NULL);
1358 * Returns the options and flags that a generic child of a BDS should
1359 * get, based on the given options and flags for the parent BDS.
1361 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1362 int *child_flags, QDict *child_options,
1363 int parent_flags, QDict *parent_options)
1365 int flags = parent_flags;
1366 GLOBAL_STATE_CODE();
1369 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1370 * Generally, the question to answer is: Should this child be
1371 * format-probed by default?
1375 * Pure and non-filtered data children of non-format nodes should
1376 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1377 * set). This only affects a very limited set of drivers (namely
1378 * quorum and blkverify when this comment was written).
1379 * Force-clear BDRV_O_PROTOCOL then.
1381 if (!parent_is_format &&
1382 (role & BDRV_CHILD_DATA) &&
1383 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1385 flags &= ~BDRV_O_PROTOCOL;
1389 * All children of format nodes (except for COW children) and all
1390 * metadata children in general should never be format-probed.
1391 * Force-set BDRV_O_PROTOCOL then.
1393 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1394 (role & BDRV_CHILD_METADATA))
1396 flags |= BDRV_O_PROTOCOL;
1400 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1403 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1404 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1405 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1407 if (role & BDRV_CHILD_COW) {
1408 /* backing files are opened read-only by default */
1409 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1410 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1412 /* Inherit the read-only option from the parent if it's not set */
1413 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1414 qdict_copy_default(child_options, parent_options,
1415 BDRV_OPT_AUTO_READ_ONLY);
1419 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1420 * can default to enable it on lower layers regardless of the
1423 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1425 /* Clear flags that only apply to the top layer */
1426 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1428 if (role & BDRV_CHILD_METADATA) {
1429 flags &= ~BDRV_O_NO_IO;
1431 if (role & BDRV_CHILD_COW) {
1432 flags &= ~BDRV_O_TEMPORARY;
1435 *child_flags = flags;
1438 static void bdrv_child_cb_attach(BdrvChild *child)
1440 BlockDriverState *bs = child->opaque;
1442 assert_bdrv_graph_writable(bs);
1443 QLIST_INSERT_HEAD(&bs->children, child, next);
1444 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1446 * Here we handle filters and block/raw-format.c when it behave like
1447 * filter. They generally have a single PRIMARY child, which is also the
1448 * FILTERED child, and that they may have multiple more children, which
1449 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1450 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1451 * into bs->backing on exceptional cases; and bs->backing will be
1454 assert(!(child->role & BDRV_CHILD_COW));
1455 if (child->role & BDRV_CHILD_PRIMARY) {
1456 assert(child->role & BDRV_CHILD_FILTERED);
1457 assert(!bs->backing);
1460 if (bs->drv->filtered_child_is_backing) {
1461 bs->backing = child;
1466 assert(!(child->role & BDRV_CHILD_FILTERED));
1468 } else if (child->role & BDRV_CHILD_COW) {
1469 assert(bs->drv->supports_backing);
1470 assert(!(child->role & BDRV_CHILD_PRIMARY));
1471 assert(!bs->backing);
1472 bs->backing = child;
1473 bdrv_backing_attach(child);
1474 } else if (child->role & BDRV_CHILD_PRIMARY) {
1479 bdrv_apply_subtree_drain(child, bs);
1482 static void bdrv_child_cb_detach(BdrvChild *child)
1484 BlockDriverState *bs = child->opaque;
1486 if (child->role & BDRV_CHILD_COW) {
1487 bdrv_backing_detach(child);
1490 bdrv_unapply_subtree_drain(child, bs);
1492 assert_bdrv_graph_writable(bs);
1493 QLIST_REMOVE(child, next);
1494 if (child == bs->backing) {
1495 assert(child != bs->file);
1497 } else if (child == bs->file) {
1502 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1503 const char *filename, Error **errp)
1505 if (c->role & BDRV_CHILD_COW) {
1506 return bdrv_backing_update_filename(c, base, filename, errp);
1511 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1513 BlockDriverState *bs = c->opaque;
1516 return bdrv_get_aio_context(bs);
1519 const BdrvChildClass child_of_bds = {
1520 .parent_is_bds = true,
1521 .get_parent_desc = bdrv_child_get_parent_desc,
1522 .inherit_options = bdrv_inherited_options,
1523 .drained_begin = bdrv_child_cb_drained_begin,
1524 .drained_poll = bdrv_child_cb_drained_poll,
1525 .drained_end = bdrv_child_cb_drained_end,
1526 .attach = bdrv_child_cb_attach,
1527 .detach = bdrv_child_cb_detach,
1528 .inactivate = bdrv_child_cb_inactivate,
1529 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1530 .update_filename = bdrv_child_cb_update_filename,
1531 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1534 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1536 GLOBAL_STATE_CODE();
1537 return c->klass->get_parent_aio_context(c);
1540 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1542 int open_flags = flags;
1543 GLOBAL_STATE_CODE();
1546 * Clear flags that are internal to the block layer before opening the
1549 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1554 static void update_flags_from_options(int *flags, QemuOpts *opts)
1556 GLOBAL_STATE_CODE();
1558 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1560 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1561 *flags |= BDRV_O_NO_FLUSH;
1564 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1565 *flags |= BDRV_O_NOCACHE;
1568 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1569 *flags |= BDRV_O_RDWR;
1572 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1573 *flags |= BDRV_O_AUTO_RDONLY;
1577 static void update_options_from_flags(QDict *options, int flags)
1579 GLOBAL_STATE_CODE();
1580 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1581 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1583 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1584 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1585 flags & BDRV_O_NO_FLUSH);
1587 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1588 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1590 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1591 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1592 flags & BDRV_O_AUTO_RDONLY);
1596 static void bdrv_assign_node_name(BlockDriverState *bs,
1597 const char *node_name,
1600 char *gen_node_name = NULL;
1601 GLOBAL_STATE_CODE();
1604 node_name = gen_node_name = id_generate(ID_BLOCK);
1605 } else if (!id_wellformed(node_name)) {
1607 * Check for empty string or invalid characters, but not if it is
1608 * generated (generated names use characters not available to the user)
1610 error_setg(errp, "Invalid node-name: '%s'", node_name);
1614 /* takes care of avoiding namespaces collisions */
1615 if (blk_by_name(node_name)) {
1616 error_setg(errp, "node-name=%s is conflicting with a device id",
1621 /* takes care of avoiding duplicates node names */
1622 if (bdrv_find_node(node_name)) {
1623 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1627 /* Make sure that the node name isn't truncated */
1628 if (strlen(node_name) >= sizeof(bs->node_name)) {
1629 error_setg(errp, "Node name too long");
1633 /* copy node name into the bs and insert it into the graph list */
1634 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1635 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1637 g_free(gen_node_name);
1640 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1641 const char *node_name, QDict *options,
1642 int open_flags, Error **errp)
1644 Error *local_err = NULL;
1646 GLOBAL_STATE_CODE();
1648 bdrv_assign_node_name(bs, node_name, &local_err);
1650 error_propagate(errp, local_err);
1655 bs->opaque = g_malloc0(drv->instance_size);
1657 if (drv->bdrv_file_open) {
1658 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1659 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1660 } else if (drv->bdrv_open) {
1661 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1668 error_propagate(errp, local_err);
1669 } else if (bs->filename[0]) {
1670 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1672 error_setg_errno(errp, -ret, "Could not open image");
1677 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1678 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1681 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1682 * drivers that pass read/write requests through to a child the trouble of
1683 * declaring support explicitly.
1685 * Drivers must not propagate this flag accidentally when they initiate I/O
1686 * to a bounce buffer. That case should be rare though.
1688 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1689 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1691 ret = refresh_total_sectors(bs, bs->total_sectors);
1693 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1697 bdrv_refresh_limits(bs, NULL, &local_err);
1699 error_propagate(errp, local_err);
1703 assert(bdrv_opt_mem_align(bs) != 0);
1704 assert(bdrv_min_mem_align(bs) != 0);
1705 assert(is_power_of_2(bs->bl.request_alignment));
1707 for (i = 0; i < bs->quiesce_counter; i++) {
1708 if (drv->bdrv_co_drain_begin) {
1709 drv->bdrv_co_drain_begin(bs);
1716 if (bs->file != NULL) {
1717 bdrv_unref_child(bs, bs->file);
1726 * Create and open a block node.
1728 * @options is a QDict of options to pass to the block drivers, or NULL for an
1729 * empty set of options. The reference to the QDict belongs to the block layer
1730 * after the call (even on failure), so if the caller intends to reuse the
1731 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1733 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1734 const char *node_name,
1735 QDict *options, int flags,
1738 BlockDriverState *bs;
1741 GLOBAL_STATE_CODE();
1744 bs->open_flags = flags;
1745 bs->options = options ?: qdict_new();
1746 bs->explicit_options = qdict_clone_shallow(bs->options);
1749 update_options_from_flags(bs->options, flags);
1751 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1753 qobject_unref(bs->explicit_options);
1754 bs->explicit_options = NULL;
1755 qobject_unref(bs->options);
1764 /* Create and open a block node. */
1765 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1766 int flags, Error **errp)
1768 GLOBAL_STATE_CODE();
1769 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1772 QemuOptsList bdrv_runtime_opts = {
1773 .name = "bdrv_common",
1774 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1777 .name = "node-name",
1778 .type = QEMU_OPT_STRING,
1779 .help = "Node name of the block device node",
1783 .type = QEMU_OPT_STRING,
1784 .help = "Block driver to use for the node",
1787 .name = BDRV_OPT_CACHE_DIRECT,
1788 .type = QEMU_OPT_BOOL,
1789 .help = "Bypass software writeback cache on the host",
1792 .name = BDRV_OPT_CACHE_NO_FLUSH,
1793 .type = QEMU_OPT_BOOL,
1794 .help = "Ignore flush requests",
1797 .name = BDRV_OPT_READ_ONLY,
1798 .type = QEMU_OPT_BOOL,
1799 .help = "Node is opened in read-only mode",
1802 .name = BDRV_OPT_AUTO_READ_ONLY,
1803 .type = QEMU_OPT_BOOL,
1804 .help = "Node can become read-only if opening read-write fails",
1807 .name = "detect-zeroes",
1808 .type = QEMU_OPT_STRING,
1809 .help = "try to optimize zero writes (off, on, unmap)",
1812 .name = BDRV_OPT_DISCARD,
1813 .type = QEMU_OPT_STRING,
1814 .help = "discard operation (ignore/off, unmap/on)",
1817 .name = BDRV_OPT_FORCE_SHARE,
1818 .type = QEMU_OPT_BOOL,
1819 .help = "always accept other writers (default: off)",
1821 { /* end of list */ }
1825 QemuOptsList bdrv_create_opts_simple = {
1826 .name = "simple-create-opts",
1827 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1830 .name = BLOCK_OPT_SIZE,
1831 .type = QEMU_OPT_SIZE,
1832 .help = "Virtual disk size"
1835 .name = BLOCK_OPT_PREALLOC,
1836 .type = QEMU_OPT_STRING,
1837 .help = "Preallocation mode (allowed values: off)"
1839 { /* end of list */ }
1844 * Common part for opening disk images and files
1846 * Removes all processed options from *options.
1848 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1849 QDict *options, Error **errp)
1851 int ret, open_flags;
1852 const char *filename;
1853 const char *driver_name = NULL;
1854 const char *node_name = NULL;
1855 const char *discard;
1858 Error *local_err = NULL;
1861 assert(bs->file == NULL);
1862 assert(options != NULL && bs->options != options);
1863 GLOBAL_STATE_CODE();
1865 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1866 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1871 update_flags_from_options(&bs->open_flags, opts);
1873 driver_name = qemu_opt_get(opts, "driver");
1874 drv = bdrv_find_format(driver_name);
1875 assert(drv != NULL);
1877 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1879 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1881 BDRV_OPT_FORCE_SHARE
1882 "=on can only be used with read-only images");
1888 bdrv_refresh_filename(blk_bs(file));
1889 filename = blk_bs(file)->filename;
1892 * Caution: while qdict_get_try_str() is fine, getting
1893 * non-string types would require more care. When @options
1894 * come from -blockdev or blockdev_add, its members are typed
1895 * according to the QAPI schema, but when they come from
1896 * -drive, they're all QString.
1898 filename = qdict_get_try_str(options, "filename");
1901 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1902 error_setg(errp, "The '%s' block driver requires a file name",
1908 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1911 ro = bdrv_is_read_only(bs);
1913 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1914 if (!ro && bdrv_is_whitelisted(drv, true)) {
1915 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1921 !ro && bdrv_is_whitelisted(drv, true)
1922 ? "Driver '%s' can only be used for read-only devices"
1923 : "Driver '%s' is not whitelisted",
1929 /* bdrv_new() and bdrv_close() make it so */
1930 assert(qatomic_read(&bs->copy_on_read) == 0);
1932 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1934 bdrv_enable_copy_on_read(bs);
1936 error_setg(errp, "Can't use copy-on-read on read-only device");
1942 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1943 if (discard != NULL) {
1944 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1945 error_setg(errp, "Invalid discard option");
1952 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1954 error_propagate(errp, local_err);
1959 if (filename != NULL) {
1960 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1962 bs->filename[0] = '\0';
1964 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1966 /* Open the image, either directly or using a protocol */
1967 open_flags = bdrv_open_flags(bs, bs->open_flags);
1968 node_name = qemu_opt_get(opts, "node-name");
1970 assert(!drv->bdrv_file_open || file == NULL);
1971 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1976 qemu_opts_del(opts);
1980 qemu_opts_del(opts);
1984 static QDict *parse_json_filename(const char *filename, Error **errp)
1986 QObject *options_obj;
1989 GLOBAL_STATE_CODE();
1991 ret = strstart(filename, "json:", &filename);
1994 options_obj = qobject_from_json(filename, errp);
1996 error_prepend(errp, "Could not parse the JSON options: ");
2000 options = qobject_to(QDict, options_obj);
2002 qobject_unref(options_obj);
2003 error_setg(errp, "Invalid JSON object given");
2007 qdict_flatten(options);
2012 static void parse_json_protocol(QDict *options, const char **pfilename,
2015 QDict *json_options;
2016 Error *local_err = NULL;
2017 GLOBAL_STATE_CODE();
2019 /* Parse json: pseudo-protocol */
2020 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2024 json_options = parse_json_filename(*pfilename, &local_err);
2026 error_propagate(errp, local_err);
2030 /* Options given in the filename have lower priority than options
2031 * specified directly */
2032 qdict_join(options, json_options, false);
2033 qobject_unref(json_options);
2038 * Fills in default options for opening images and converts the legacy
2039 * filename/flags pair to option QDict entries.
2040 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2041 * block driver has been specified explicitly.
2043 static int bdrv_fill_options(QDict **options, const char *filename,
2044 int *flags, Error **errp)
2046 const char *drvname;
2047 bool protocol = *flags & BDRV_O_PROTOCOL;
2048 bool parse_filename = false;
2049 BlockDriver *drv = NULL;
2050 Error *local_err = NULL;
2052 GLOBAL_STATE_CODE();
2055 * Caution: while qdict_get_try_str() is fine, getting non-string
2056 * types would require more care. When @options come from
2057 * -blockdev or blockdev_add, its members are typed according to
2058 * the QAPI schema, but when they come from -drive, they're all
2061 drvname = qdict_get_try_str(*options, "driver");
2063 drv = bdrv_find_format(drvname);
2065 error_setg(errp, "Unknown driver '%s'", drvname);
2068 /* If the user has explicitly specified the driver, this choice should
2069 * override the BDRV_O_PROTOCOL flag */
2070 protocol = drv->bdrv_file_open;
2074 *flags |= BDRV_O_PROTOCOL;
2076 *flags &= ~BDRV_O_PROTOCOL;
2079 /* Translate cache options from flags into options */
2080 update_options_from_flags(*options, *flags);
2082 /* Fetch the file name from the options QDict if necessary */
2083 if (protocol && filename) {
2084 if (!qdict_haskey(*options, "filename")) {
2085 qdict_put_str(*options, "filename", filename);
2086 parse_filename = true;
2088 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2094 /* Find the right block driver */
2095 /* See cautionary note on accessing @options above */
2096 filename = qdict_get_try_str(*options, "filename");
2098 if (!drvname && protocol) {
2100 drv = bdrv_find_protocol(filename, parse_filename, errp);
2105 drvname = drv->format_name;
2106 qdict_put_str(*options, "driver", drvname);
2108 error_setg(errp, "Must specify either driver or file");
2113 assert(drv || !protocol);
2115 /* Driver-specific filename parsing */
2116 if (drv && drv->bdrv_parse_filename && parse_filename) {
2117 drv->bdrv_parse_filename(filename, *options, &local_err);
2119 error_propagate(errp, local_err);
2123 if (!drv->bdrv_needs_filename) {
2124 qdict_del(*options, "filename");
2131 typedef struct BlockReopenQueueEntry {
2134 BDRVReopenState state;
2135 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2136 } BlockReopenQueueEntry;
2139 * Return the flags that @bs will have after the reopens in @q have
2140 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2141 * return the current flags.
2143 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2145 BlockReopenQueueEntry *entry;
2148 QTAILQ_FOREACH(entry, q, entry) {
2149 if (entry->state.bs == bs) {
2150 return entry->state.flags;
2155 return bs->open_flags;
2158 /* Returns whether the image file can be written to after the reopen queue @q
2159 * has been successfully applied, or right now if @q is NULL. */
2160 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2161 BlockReopenQueue *q)
2163 int flags = bdrv_reopen_get_flags(q, bs);
2165 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2169 * Return whether the BDS can be written to. This is not necessarily
2170 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2171 * be written to but do not count as read-only images.
2173 bool bdrv_is_writable(BlockDriverState *bs)
2176 return bdrv_is_writable_after_reopen(bs, NULL);
2179 static char *bdrv_child_user_desc(BdrvChild *c)
2181 GLOBAL_STATE_CODE();
2182 return c->klass->get_parent_desc(c);
2186 * Check that @a allows everything that @b needs. @a and @b must reference same
2189 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2191 const char *child_bs_name;
2192 g_autofree char *a_user = NULL;
2193 g_autofree char *b_user = NULL;
2194 g_autofree char *perms = NULL;
2197 assert(a->bs == b->bs);
2198 GLOBAL_STATE_CODE();
2200 if ((b->perm & a->shared_perm) == b->perm) {
2204 child_bs_name = bdrv_get_node_name(b->bs);
2205 a_user = bdrv_child_user_desc(a);
2206 b_user = bdrv_child_user_desc(b);
2207 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2209 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2210 "both required by %s (uses node '%s' as '%s' child) and "
2211 "unshared by %s (uses node '%s' as '%s' child).",
2212 child_bs_name, perms,
2213 b_user, child_bs_name, b->name,
2214 a_user, child_bs_name, a->name);
2219 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2222 GLOBAL_STATE_CODE();
2225 * During the loop we'll look at each pair twice. That's correct because
2226 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2229 QLIST_FOREACH(a, &bs->parents, next_parent) {
2230 QLIST_FOREACH(b, &bs->parents, next_parent) {
2235 if (!bdrv_a_allow_b(a, b, errp)) {
2244 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2245 BdrvChild *c, BdrvChildRole role,
2246 BlockReopenQueue *reopen_queue,
2247 uint64_t parent_perm, uint64_t parent_shared,
2248 uint64_t *nperm, uint64_t *nshared)
2250 assert(bs->drv && bs->drv->bdrv_child_perm);
2251 GLOBAL_STATE_CODE();
2252 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2253 parent_perm, parent_shared,
2255 /* TODO Take force_share from reopen_queue */
2256 if (child_bs && child_bs->force_share) {
2257 *nshared = BLK_PERM_ALL;
2262 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2263 * nodes that are already in the @list, of course) so that final list is
2264 * topologically sorted. Return the result (GSList @list object is updated, so
2265 * don't use old reference after function call).
2267 * On function start @list must be already topologically sorted and for any node
2268 * in the @list the whole subtree of the node must be in the @list as well. The
2269 * simplest way to satisfy this criteria: use only result of
2270 * bdrv_topological_dfs() or NULL as @list parameter.
2272 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2273 BlockDriverState *bs)
2276 g_autoptr(GHashTable) local_found = NULL;
2278 GLOBAL_STATE_CODE();
2282 found = local_found = g_hash_table_new(NULL, NULL);
2285 if (g_hash_table_contains(found, bs)) {
2288 g_hash_table_add(found, bs);
2290 QLIST_FOREACH(child, &bs->children, next) {
2291 list = bdrv_topological_dfs(list, found, child->bs);
2294 return g_slist_prepend(list, bs);
2297 typedef struct BdrvChildSetPermState {
2300 uint64_t old_shared_perm;
2301 } BdrvChildSetPermState;
2303 static void bdrv_child_set_perm_abort(void *opaque)
2305 BdrvChildSetPermState *s = opaque;
2307 GLOBAL_STATE_CODE();
2309 s->child->perm = s->old_perm;
2310 s->child->shared_perm = s->old_shared_perm;
2313 static TransactionActionDrv bdrv_child_set_pem_drv = {
2314 .abort = bdrv_child_set_perm_abort,
2318 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2319 uint64_t shared, Transaction *tran)
2321 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2322 GLOBAL_STATE_CODE();
2324 *s = (BdrvChildSetPermState) {
2326 .old_perm = c->perm,
2327 .old_shared_perm = c->shared_perm,
2331 c->shared_perm = shared;
2333 tran_add(tran, &bdrv_child_set_pem_drv, s);
2336 static void bdrv_drv_set_perm_commit(void *opaque)
2338 BlockDriverState *bs = opaque;
2339 uint64_t cumulative_perms, cumulative_shared_perms;
2340 GLOBAL_STATE_CODE();
2342 if (bs->drv->bdrv_set_perm) {
2343 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2344 &cumulative_shared_perms);
2345 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2349 static void bdrv_drv_set_perm_abort(void *opaque)
2351 BlockDriverState *bs = opaque;
2352 GLOBAL_STATE_CODE();
2354 if (bs->drv->bdrv_abort_perm_update) {
2355 bs->drv->bdrv_abort_perm_update(bs);
2359 TransactionActionDrv bdrv_drv_set_perm_drv = {
2360 .abort = bdrv_drv_set_perm_abort,
2361 .commit = bdrv_drv_set_perm_commit,
2364 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2365 uint64_t shared_perm, Transaction *tran,
2368 GLOBAL_STATE_CODE();
2373 if (bs->drv->bdrv_check_perm) {
2374 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2381 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2387 typedef struct BdrvReplaceChildState {
2389 BlockDriverState *old_bs;
2390 } BdrvReplaceChildState;
2392 static void bdrv_replace_child_commit(void *opaque)
2394 BdrvReplaceChildState *s = opaque;
2395 GLOBAL_STATE_CODE();
2397 bdrv_unref(s->old_bs);
2400 static void bdrv_replace_child_abort(void *opaque)
2402 BdrvReplaceChildState *s = opaque;
2403 BlockDriverState *new_bs = s->child->bs;
2405 GLOBAL_STATE_CODE();
2406 /* old_bs reference is transparently moved from @s to @s->child */
2407 bdrv_replace_child_noperm(s->child, s->old_bs);
2411 static TransactionActionDrv bdrv_replace_child_drv = {
2412 .commit = bdrv_replace_child_commit,
2413 .abort = bdrv_replace_child_abort,
2418 * bdrv_replace_child_tran
2420 * Note: real unref of old_bs is done only on commit.
2422 * The function doesn't update permissions, caller is responsible for this.
2424 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2427 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2428 *s = (BdrvReplaceChildState) {
2430 .old_bs = child->bs,
2432 tran_add(tran, &bdrv_replace_child_drv, s);
2437 bdrv_replace_child_noperm(child, new_bs);
2438 /* old_bs reference is transparently moved from @child to @s */
2442 * Refresh permissions in @bs subtree. The function is intended to be called
2443 * after some graph modification that was done without permission update.
2445 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2446 Transaction *tran, Error **errp)
2448 BlockDriver *drv = bs->drv;
2451 uint64_t cumulative_perms, cumulative_shared_perms;
2452 GLOBAL_STATE_CODE();
2454 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2456 /* Write permissions never work with read-only images */
2457 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2458 !bdrv_is_writable_after_reopen(bs, q))
2460 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2461 error_setg(errp, "Block node is read-only");
2463 error_setg(errp, "Read-only block node '%s' cannot support "
2464 "read-write users", bdrv_get_node_name(bs));
2471 * Unaligned requests will automatically be aligned to bl.request_alignment
2472 * and without RESIZE we can't extend requests to write to space beyond the
2473 * end of the image, so it's required that the image size is aligned.
2475 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2476 !(cumulative_perms & BLK_PERM_RESIZE))
2478 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2479 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2480 "Image size is not a multiple of request "
2486 /* Check this node */
2491 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2497 /* Drivers that never have children can omit .bdrv_child_perm() */
2498 if (!drv->bdrv_child_perm) {
2499 assert(QLIST_EMPTY(&bs->children));
2503 /* Check all children */
2504 QLIST_FOREACH(c, &bs->children, next) {
2505 uint64_t cur_perm, cur_shared;
2507 bdrv_child_perm(bs, c->bs, c, c->role, q,
2508 cumulative_perms, cumulative_shared_perms,
2509 &cur_perm, &cur_shared);
2510 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2516 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2517 Transaction *tran, Error **errp)
2520 BlockDriverState *bs;
2521 GLOBAL_STATE_CODE();
2523 for ( ; list; list = list->next) {
2526 if (bdrv_parent_perms_conflict(bs, errp)) {
2530 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2539 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2540 uint64_t *shared_perm)
2543 uint64_t cumulative_perms = 0;
2544 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2546 GLOBAL_STATE_CODE();
2548 QLIST_FOREACH(c, &bs->parents, next_parent) {
2549 cumulative_perms |= c->perm;
2550 cumulative_shared_perms &= c->shared_perm;
2553 *perm = cumulative_perms;
2554 *shared_perm = cumulative_shared_perms;
2557 char *bdrv_perm_names(uint64_t perm)
2563 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2564 { BLK_PERM_WRITE, "write" },
2565 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2566 { BLK_PERM_RESIZE, "resize" },
2570 GString *result = g_string_sized_new(30);
2571 struct perm_name *p;
2573 for (p = permissions; p->name; p++) {
2574 if (perm & p->perm) {
2575 if (result->len > 0) {
2576 g_string_append(result, ", ");
2578 g_string_append(result, p->name);
2582 return g_string_free(result, FALSE);
2586 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2589 Transaction *tran = tran_new();
2590 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2591 GLOBAL_STATE_CODE();
2593 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2594 tran_finalize(tran, ret);
2599 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2602 Error *local_err = NULL;
2603 Transaction *tran = tran_new();
2606 GLOBAL_STATE_CODE();
2608 bdrv_child_set_perm(c, perm, shared, tran);
2610 ret = bdrv_refresh_perms(c->bs, &local_err);
2612 tran_finalize(tran, ret);
2615 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2616 /* tighten permissions */
2617 error_propagate(errp, local_err);
2620 * Our caller may intend to only loosen restrictions and
2621 * does not expect this function to fail. Errors are not
2622 * fatal in such a case, so we can just hide them from our
2625 error_free(local_err);
2633 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2635 uint64_t parent_perms, parent_shared;
2636 uint64_t perms, shared;
2638 GLOBAL_STATE_CODE();
2640 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2641 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2642 parent_perms, parent_shared, &perms, &shared);
2644 return bdrv_child_try_set_perm(c, perms, shared, errp);
2648 * Default implementation for .bdrv_child_perm() for block filters:
2649 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2652 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2654 BlockReopenQueue *reopen_queue,
2655 uint64_t perm, uint64_t shared,
2656 uint64_t *nperm, uint64_t *nshared)
2658 GLOBAL_STATE_CODE();
2659 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2660 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2663 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2665 BlockReopenQueue *reopen_queue,
2666 uint64_t perm, uint64_t shared,
2667 uint64_t *nperm, uint64_t *nshared)
2669 assert(role & BDRV_CHILD_COW);
2670 GLOBAL_STATE_CODE();
2673 * We want consistent read from backing files if the parent needs it.
2674 * No other operations are performed on backing files.
2676 perm &= BLK_PERM_CONSISTENT_READ;
2679 * If the parent can deal with changing data, we're okay with a
2680 * writable and resizable backing file.
2681 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2683 if (shared & BLK_PERM_WRITE) {
2684 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2689 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2691 if (bs->open_flags & BDRV_O_INACTIVE) {
2692 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2699 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2701 BlockReopenQueue *reopen_queue,
2702 uint64_t perm, uint64_t shared,
2703 uint64_t *nperm, uint64_t *nshared)
2707 GLOBAL_STATE_CODE();
2708 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2710 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2713 * Apart from the modifications below, the same permissions are
2714 * forwarded and left alone as for filters
2716 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2717 perm, shared, &perm, &shared);
2719 if (role & BDRV_CHILD_METADATA) {
2720 /* Format drivers may touch metadata even if the guest doesn't write */
2721 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2722 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2726 * bs->file always needs to be consistent because of the
2727 * metadata. We can never allow other users to resize or write
2730 if (!(flags & BDRV_O_NO_IO)) {
2731 perm |= BLK_PERM_CONSISTENT_READ;
2733 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2736 if (role & BDRV_CHILD_DATA) {
2738 * Technically, everything in this block is a subset of the
2739 * BDRV_CHILD_METADATA path taken above, and so this could
2740 * be an "else if" branch. However, that is not obvious, and
2741 * this function is not performance critical, therefore we let
2742 * this be an independent "if".
2746 * We cannot allow other users to resize the file because the
2747 * format driver might have some assumptions about the size
2748 * (e.g. because it is stored in metadata, or because the file
2749 * is split into fixed-size data files).
2751 shared &= ~BLK_PERM_RESIZE;
2754 * WRITE_UNCHANGED often cannot be performed as such on the
2755 * data file. For example, the qcow2 driver may still need to
2756 * write copied clusters on copy-on-read.
2758 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2759 perm |= BLK_PERM_WRITE;
2763 * If the data file is written to, the format driver may
2764 * expect to be able to resize it by writing beyond the EOF.
2766 if (perm & BLK_PERM_WRITE) {
2767 perm |= BLK_PERM_RESIZE;
2771 if (bs->open_flags & BDRV_O_INACTIVE) {
2772 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2779 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2780 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2781 uint64_t perm, uint64_t shared,
2782 uint64_t *nperm, uint64_t *nshared)
2784 GLOBAL_STATE_CODE();
2785 if (role & BDRV_CHILD_FILTERED) {
2786 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2788 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2789 perm, shared, nperm, nshared);
2790 } else if (role & BDRV_CHILD_COW) {
2791 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2792 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2793 perm, shared, nperm, nshared);
2794 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2795 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2796 perm, shared, nperm, nshared);
2798 g_assert_not_reached();
2802 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2804 static const uint64_t permissions[] = {
2805 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2806 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2807 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2808 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2811 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2812 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2814 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2816 return permissions[qapi_perm];
2819 static void bdrv_replace_child_noperm(BdrvChild *child,
2820 BlockDriverState *new_bs)
2822 BlockDriverState *old_bs = child->bs;
2823 int new_bs_quiesce_counter;
2826 assert(!child->frozen);
2827 assert(old_bs != new_bs);
2828 GLOBAL_STATE_CODE();
2830 if (old_bs && new_bs) {
2831 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2834 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2835 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2838 * If the new child node is drained but the old one was not, flush
2839 * all outstanding requests to the old child node.
2841 while (drain_saldo > 0 && child->klass->drained_begin) {
2842 bdrv_parent_drained_begin_single(child, true);
2847 /* Detach first so that the recursive drain sections coming from @child
2848 * are already gone and we only end the drain sections that came from
2850 if (child->klass->detach) {
2851 child->klass->detach(child);
2853 assert_bdrv_graph_writable(old_bs);
2854 QLIST_REMOVE(child, next_parent);
2860 assert_bdrv_graph_writable(new_bs);
2861 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2864 * Detaching the old node may have led to the new node's
2865 * quiesce_counter having been decreased. Not a problem, we
2866 * just need to recognize this here and then invoke
2867 * drained_end appropriately more often.
2869 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2870 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2872 /* Attach only after starting new drained sections, so that recursive
2873 * drain sections coming from @child don't get an extra .drained_begin
2875 if (child->klass->attach) {
2876 child->klass->attach(child);
2881 * If the old child node was drained but the new one is not, allow
2882 * requests to come in only after the new node has been attached.
2884 while (drain_saldo < 0 && child->klass->drained_end) {
2885 bdrv_parent_drained_end_single(child);
2891 * Free the given @child.
2893 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2894 * unused (i.e. not in a children list).
2896 static void bdrv_child_free(BdrvChild *child)
2899 GLOBAL_STATE_CODE();
2900 assert(!child->next.le_prev); /* not in children list */
2902 g_free(child->name);
2906 typedef struct BdrvAttachChildCommonState {
2908 AioContext *old_parent_ctx;
2909 AioContext *old_child_ctx;
2910 } BdrvAttachChildCommonState;
2912 static void bdrv_attach_child_common_abort(void *opaque)
2914 BdrvAttachChildCommonState *s = opaque;
2915 BlockDriverState *bs = s->child->bs;
2917 GLOBAL_STATE_CODE();
2918 bdrv_replace_child_noperm(s->child, NULL);
2920 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2921 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2924 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2926 GHashTable *visited;
2931 /* No need to visit `child`, because it has been detached already */
2932 visited = g_hash_table_new(NULL, NULL);
2933 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2934 visited, tran, &error_abort);
2935 g_hash_table_destroy(visited);
2937 /* transaction is supposed to always succeed */
2938 assert(ret == true);
2943 bdrv_child_free(s->child);
2946 static TransactionActionDrv bdrv_attach_child_common_drv = {
2947 .abort = bdrv_attach_child_common_abort,
2952 * Common part of attaching bdrv child to bs or to blk or to job
2954 * Function doesn't update permissions, caller is responsible for this.
2956 * Returns new created child.
2958 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2959 const char *child_name,
2960 const BdrvChildClass *child_class,
2961 BdrvChildRole child_role,
2962 uint64_t perm, uint64_t shared_perm,
2964 Transaction *tran, Error **errp)
2966 BdrvChild *new_child;
2967 AioContext *parent_ctx;
2968 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2970 assert(child_class->get_parent_desc);
2971 GLOBAL_STATE_CODE();
2973 new_child = g_new(BdrvChild, 1);
2974 *new_child = (BdrvChild) {
2976 .name = g_strdup(child_name),
2977 .klass = child_class,
2980 .shared_perm = shared_perm,
2985 * If the AioContexts don't match, first try to move the subtree of
2986 * child_bs into the AioContext of the new parent. If this doesn't work,
2987 * try moving the parent into the AioContext of child_bs instead.
2989 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
2990 if (child_ctx != parent_ctx) {
2991 Error *local_err = NULL;
2992 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
2995 if (ret < 0 && child_class->change_aio_ctx) {
2996 Transaction *tran = tran_new();
2997 GHashTable *visited = g_hash_table_new(NULL, NULL);
3000 g_hash_table_add(visited, new_child);
3001 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3002 visited, tran, NULL);
3003 if (ret_child == true) {
3004 error_free(local_err);
3007 tran_finalize(tran, ret_child == true ? 0 : -1);
3008 g_hash_table_destroy(visited);
3012 error_propagate(errp, local_err);
3013 bdrv_child_free(new_child);
3019 bdrv_replace_child_noperm(new_child, child_bs);
3021 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3022 *s = (BdrvAttachChildCommonState) {
3024 .old_parent_ctx = parent_ctx,
3025 .old_child_ctx = child_ctx,
3027 tran_add(tran, &bdrv_attach_child_common_drv, s);
3033 * Function doesn't update permissions, caller is responsible for this.
3035 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3036 BlockDriverState *child_bs,
3037 const char *child_name,
3038 const BdrvChildClass *child_class,
3039 BdrvChildRole child_role,
3043 uint64_t perm, shared_perm;
3045 assert(parent_bs->drv);
3046 GLOBAL_STATE_CODE();
3048 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3049 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3050 child_bs->node_name, child_name, parent_bs->node_name);
3054 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3055 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3056 perm, shared_perm, &perm, &shared_perm);
3058 return bdrv_attach_child_common(child_bs, child_name, child_class,
3059 child_role, perm, shared_perm, parent_bs,
3063 static void bdrv_detach_child(BdrvChild *child)
3065 BlockDriverState *old_bs = child->bs;
3067 GLOBAL_STATE_CODE();
3068 bdrv_replace_child_noperm(child, NULL);
3069 bdrv_child_free(child);
3073 * Update permissions for old node. We're just taking a parent away, so
3074 * we're loosening restrictions. Errors of permission update are not
3075 * fatal in this case, ignore them.
3077 bdrv_refresh_perms(old_bs, NULL);
3080 * When the parent requiring a non-default AioContext is removed, the
3081 * node moves back to the main AioContext
3083 bdrv_try_change_aio_context(old_bs, qemu_get_aio_context(), NULL, NULL);
3088 * This function steals the reference to child_bs from the caller.
3089 * That reference is later dropped by bdrv_root_unref_child().
3091 * On failure NULL is returned, errp is set and the reference to
3092 * child_bs is also dropped.
3094 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3095 * (unless @child_bs is already in @ctx).
3097 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3098 const char *child_name,
3099 const BdrvChildClass *child_class,
3100 BdrvChildRole child_role,
3101 uint64_t perm, uint64_t shared_perm,
3102 void *opaque, Error **errp)
3106 Transaction *tran = tran_new();
3108 GLOBAL_STATE_CODE();
3110 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3111 child_role, perm, shared_perm, opaque,
3118 ret = bdrv_refresh_perms(child_bs, errp);
3121 tran_finalize(tran, ret);
3123 bdrv_unref(child_bs);
3125 return ret < 0 ? NULL : child;
3129 * This function transfers the reference to child_bs from the caller
3130 * to parent_bs. That reference is later dropped by parent_bs on
3131 * bdrv_close() or if someone calls bdrv_unref_child().
3133 * On failure NULL is returned, errp is set and the reference to
3134 * child_bs is also dropped.
3136 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3137 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3139 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3140 BlockDriverState *child_bs,
3141 const char *child_name,
3142 const BdrvChildClass *child_class,
3143 BdrvChildRole child_role,
3148 Transaction *tran = tran_new();
3150 GLOBAL_STATE_CODE();
3152 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3153 child_class, child_role, tran, errp);
3159 ret = bdrv_refresh_perms(parent_bs, errp);
3165 tran_finalize(tran, ret);
3167 bdrv_unref(child_bs);
3169 return ret < 0 ? NULL : child;
3172 /* Callers must ensure that child->frozen is false. */
3173 void bdrv_root_unref_child(BdrvChild *child)
3175 BlockDriverState *child_bs;
3177 GLOBAL_STATE_CODE();
3179 child_bs = child->bs;
3180 bdrv_detach_child(child);
3181 bdrv_unref(child_bs);
3184 typedef struct BdrvSetInheritsFrom {
3185 BlockDriverState *bs;
3186 BlockDriverState *old_inherits_from;
3187 } BdrvSetInheritsFrom;
3189 static void bdrv_set_inherits_from_abort(void *opaque)
3191 BdrvSetInheritsFrom *s = opaque;
3193 s->bs->inherits_from = s->old_inherits_from;
3196 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3197 .abort = bdrv_set_inherits_from_abort,
3201 /* @tran is allowed to be NULL. In this case no rollback is possible */
3202 static void bdrv_set_inherits_from(BlockDriverState *bs,
3203 BlockDriverState *new_inherits_from,
3207 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3209 *s = (BdrvSetInheritsFrom) {
3211 .old_inherits_from = bs->inherits_from,
3214 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3217 bs->inherits_from = new_inherits_from;
3221 * Clear all inherits_from pointers from children and grandchildren of
3222 * @root that point to @root, where necessary.
3223 * @tran is allowed to be NULL. In this case no rollback is possible
3225 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3230 if (child->bs->inherits_from == root) {
3232 * Remove inherits_from only when the last reference between root and
3233 * child->bs goes away.
3235 QLIST_FOREACH(c, &root->children, next) {
3236 if (c != child && c->bs == child->bs) {
3241 bdrv_set_inherits_from(child->bs, NULL, tran);
3245 QLIST_FOREACH(c, &child->bs->children, next) {
3246 bdrv_unset_inherits_from(root, c, tran);
3250 /* Callers must ensure that child->frozen is false. */
3251 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3253 GLOBAL_STATE_CODE();
3254 if (child == NULL) {
3258 bdrv_unset_inherits_from(parent, child, NULL);
3259 bdrv_root_unref_child(child);
3263 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3266 GLOBAL_STATE_CODE();
3267 QLIST_FOREACH(c, &bs->parents, next_parent) {
3268 if (c->klass->change_media) {
3269 c->klass->change_media(c, load);
3274 /* Return true if you can reach parent going through child->inherits_from
3275 * recursively. If parent or child are NULL, return false */
3276 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3277 BlockDriverState *parent)
3279 while (child && child != parent) {
3280 child = child->inherits_from;
3283 return child != NULL;
3287 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3288 * mostly used for COW backing children (role = COW), but also for
3289 * filtered children (role = FILTERED | PRIMARY).
3291 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3293 if (bs->drv && bs->drv->is_filter) {
3294 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3296 return BDRV_CHILD_COW;
3301 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3302 * callers which don't need their own reference any more must call bdrv_unref().
3304 * Function doesn't update permissions, caller is responsible for this.
3306 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3307 BlockDriverState *child_bs,
3309 Transaction *tran, Error **errp)
3311 bool update_inherits_from =
3312 bdrv_inherits_from_recursive(child_bs, parent_bs);
3313 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3316 GLOBAL_STATE_CODE();
3318 if (!parent_bs->drv) {
3320 * Node without drv is an object without a class :/. TODO: finally fix
3321 * qcow2 driver to never clear bs->drv and implement format corruption
3322 * handling in other way.
3324 error_setg(errp, "Node corrupted");
3328 if (child && child->frozen) {
3329 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3330 child->name, parent_bs->node_name, child->bs->node_name);
3334 if (is_backing && !parent_bs->drv->is_filter &&
3335 !parent_bs->drv->supports_backing)
3337 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3338 "files", parent_bs->drv->format_name, parent_bs->node_name);
3342 if (parent_bs->drv->is_filter) {
3343 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3344 } else if (is_backing) {
3345 role = BDRV_CHILD_COW;
3348 * We only can use same role as it is in existing child. We don't have
3349 * infrastructure to determine role of file child in generic way
3352 error_setg(errp, "Cannot set file child to format node without "
3360 bdrv_unset_inherits_from(parent_bs, child, tran);
3361 bdrv_remove_child(child, tran);
3368 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3369 is_backing ? "backing" : "file",
3370 &child_of_bds, role,
3378 * If inherits_from pointed recursively to bs then let's update it to
3379 * point directly to bs (else it will become NULL).
3381 if (update_inherits_from) {
3382 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3386 bdrv_refresh_limits(parent_bs, tran, NULL);
3391 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3392 BlockDriverState *backing_hd,
3393 Transaction *tran, Error **errp)
3395 GLOBAL_STATE_CODE();
3396 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3399 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3403 Transaction *tran = tran_new();
3405 GLOBAL_STATE_CODE();
3406 bdrv_drained_begin(bs);
3408 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3413 ret = bdrv_refresh_perms(bs, errp);
3415 tran_finalize(tran, ret);
3417 bdrv_drained_end(bs);
3423 * Opens the backing file for a BlockDriverState if not yet open
3425 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3426 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3427 * itself, all options starting with "${bdref_key}." are considered part of the
3430 * TODO Can this be unified with bdrv_open_image()?
3432 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3433 const char *bdref_key, Error **errp)
3435 char *backing_filename = NULL;
3436 char *bdref_key_dot;
3437 const char *reference = NULL;
3439 bool implicit_backing = false;
3440 BlockDriverState *backing_hd;
3442 QDict *tmp_parent_options = NULL;
3443 Error *local_err = NULL;
3445 GLOBAL_STATE_CODE();
3447 if (bs->backing != NULL) {
3451 /* NULL means an empty set of options */
3452 if (parent_options == NULL) {
3453 tmp_parent_options = qdict_new();
3454 parent_options = tmp_parent_options;
3457 bs->open_flags &= ~BDRV_O_NO_BACKING;
3459 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3460 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3461 g_free(bdref_key_dot);
3464 * Caution: while qdict_get_try_str() is fine, getting non-string
3465 * types would require more care. When @parent_options come from
3466 * -blockdev or blockdev_add, its members are typed according to
3467 * the QAPI schema, but when they come from -drive, they're all
3470 reference = qdict_get_try_str(parent_options, bdref_key);
3471 if (reference || qdict_haskey(options, "file.filename")) {
3472 /* keep backing_filename NULL */
3473 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3474 qobject_unref(options);
3477 if (qdict_size(options) == 0) {
3478 /* If the user specifies options that do not modify the
3479 * backing file's behavior, we might still consider it the
3480 * implicit backing file. But it's easier this way, and
3481 * just specifying some of the backing BDS's options is
3482 * only possible with -drive anyway (otherwise the QAPI
3483 * schema forces the user to specify everything). */
3484 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3487 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3490 error_propagate(errp, local_err);
3491 qobject_unref(options);
3496 if (!bs->drv || !bs->drv->supports_backing) {
3498 error_setg(errp, "Driver doesn't support backing files");
3499 qobject_unref(options);
3504 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3505 qdict_put_str(options, "driver", bs->backing_format);
3508 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3509 &child_of_bds, bdrv_backing_role(bs), errp);
3511 bs->open_flags |= BDRV_O_NO_BACKING;
3512 error_prepend(errp, "Could not open backing file: ");
3517 if (implicit_backing) {
3518 bdrv_refresh_filename(backing_hd);
3519 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3520 backing_hd->filename);
3523 /* Hook up the backing file link; drop our reference, bs owns the
3524 * backing_hd reference now */
3525 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3526 bdrv_unref(backing_hd);
3531 qdict_del(parent_options, bdref_key);
3534 g_free(backing_filename);
3535 qobject_unref(tmp_parent_options);
3539 static BlockDriverState *
3540 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3541 BlockDriverState *parent, const BdrvChildClass *child_class,
3542 BdrvChildRole child_role, bool allow_none, Error **errp)
3544 BlockDriverState *bs = NULL;
3545 QDict *image_options;
3546 char *bdref_key_dot;
3547 const char *reference;
3549 assert(child_class != NULL);
3551 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3552 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3553 g_free(bdref_key_dot);
3556 * Caution: while qdict_get_try_str() is fine, getting non-string
3557 * types would require more care. When @options come from
3558 * -blockdev or blockdev_add, its members are typed according to
3559 * the QAPI schema, but when they come from -drive, they're all
3562 reference = qdict_get_try_str(options, bdref_key);
3563 if (!filename && !reference && !qdict_size(image_options)) {
3565 error_setg(errp, "A block device must be specified for \"%s\"",
3568 qobject_unref(image_options);
3572 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3573 parent, child_class, child_role, errp);
3579 qdict_del(options, bdref_key);
3584 * Opens a disk image whose options are given as BlockdevRef in another block
3587 * If allow_none is true, no image will be opened if filename is false and no
3588 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3590 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3591 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3592 * itself, all options starting with "${bdref_key}." are considered part of the
3595 * The BlockdevRef will be removed from the options QDict.
3597 BdrvChild *bdrv_open_child(const char *filename,
3598 QDict *options, const char *bdref_key,
3599 BlockDriverState *parent,
3600 const BdrvChildClass *child_class,
3601 BdrvChildRole child_role,
3602 bool allow_none, Error **errp)
3604 BlockDriverState *bs;
3606 GLOBAL_STATE_CODE();
3608 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3609 child_role, allow_none, errp);
3614 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3619 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3621 int bdrv_open_file_child(const char *filename,
3622 QDict *options, const char *bdref_key,
3623 BlockDriverState *parent, Error **errp)
3627 /* commit_top and mirror_top don't use this function */
3628 assert(!parent->drv->filtered_child_is_backing);
3629 role = parent->drv->is_filter ?
3630 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3632 if (!bdrv_open_child(filename, options, bdref_key, parent,
3633 &child_of_bds, role, false, errp))
3642 * TODO Future callers may need to specify parent/child_class in order for
3643 * option inheritance to work. Existing callers use it for the root node.
3645 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3647 BlockDriverState *bs = NULL;
3648 QObject *obj = NULL;
3649 QDict *qdict = NULL;
3650 const char *reference = NULL;
3653 GLOBAL_STATE_CODE();
3655 if (ref->type == QTYPE_QSTRING) {
3656 reference = ref->u.reference;
3658 BlockdevOptions *options = &ref->u.definition;
3659 assert(ref->type == QTYPE_QDICT);
3661 v = qobject_output_visitor_new(&obj);
3662 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3663 visit_complete(v, &obj);
3665 qdict = qobject_to(QDict, obj);
3666 qdict_flatten(qdict);
3668 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3669 * compatibility with other callers) rather than what we want as the
3670 * real defaults. Apply the defaults here instead. */
3671 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3672 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3673 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3674 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3678 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3685 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3687 QDict *snapshot_options,
3690 g_autofree char *tmp_filename = NULL;
3692 QemuOpts *opts = NULL;
3693 BlockDriverState *bs_snapshot = NULL;
3696 GLOBAL_STATE_CODE();
3698 /* if snapshot, we create a temporary backing file and open it
3699 instead of opening 'filename' directly */
3701 /* Get the required size from the image */
3702 total_size = bdrv_getlength(bs);
3703 if (total_size < 0) {
3704 error_setg_errno(errp, -total_size, "Could not get image size");
3708 /* Create the temporary image */
3709 tmp_filename = create_tmp_file(errp);
3710 if (!tmp_filename) {
3714 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3716 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3717 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3718 qemu_opts_del(opts);
3720 error_prepend(errp, "Could not create temporary overlay '%s': ",
3725 /* Prepare options QDict for the temporary file */
3726 qdict_put_str(snapshot_options, "file.driver", "file");
3727 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3728 qdict_put_str(snapshot_options, "driver", "qcow2");
3730 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3731 snapshot_options = NULL;
3736 ret = bdrv_append(bs_snapshot, bs, errp);
3743 qobject_unref(snapshot_options);
3748 * Opens a disk image (raw, qcow2, vmdk, ...)
3750 * options is a QDict of options to pass to the block drivers, or NULL for an
3751 * empty set of options. The reference to the QDict belongs to the block layer
3752 * after the call (even on failure), so if the caller intends to reuse the
3753 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3755 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3756 * If it is not NULL, the referenced BDS will be reused.
3758 * The reference parameter may be used to specify an existing block device which
3759 * should be opened. If specified, neither options nor a filename may be given,
3760 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3762 static BlockDriverState *bdrv_open_inherit(const char *filename,
3763 const char *reference,
3764 QDict *options, int flags,
3765 BlockDriverState *parent,
3766 const BdrvChildClass *child_class,
3767 BdrvChildRole child_role,
3771 BlockBackend *file = NULL;
3772 BlockDriverState *bs;
3773 BlockDriver *drv = NULL;
3775 const char *drvname;
3776 const char *backing;
3777 Error *local_err = NULL;
3778 QDict *snapshot_options = NULL;
3779 int snapshot_flags = 0;
3781 assert(!child_class || !flags);
3782 assert(!child_class == !parent);
3783 GLOBAL_STATE_CODE();
3786 bool options_non_empty = options ? qdict_size(options) : false;
3787 qobject_unref(options);
3789 if (filename || options_non_empty) {
3790 error_setg(errp, "Cannot reference an existing block device with "
3791 "additional options or a new filename");
3795 bs = bdrv_lookup_bs(reference, reference, errp);
3806 /* NULL means an empty set of options */
3807 if (options == NULL) {
3808 options = qdict_new();
3811 /* json: syntax counts as explicit options, as if in the QDict */
3812 parse_json_protocol(options, &filename, &local_err);
3817 bs->explicit_options = qdict_clone_shallow(options);
3820 bool parent_is_format;
3823 parent_is_format = parent->drv->is_format;
3826 * parent->drv is not set yet because this node is opened for
3827 * (potential) format probing. That means that @parent is going
3828 * to be a format node.
3830 parent_is_format = true;
3833 bs->inherits_from = parent;
3834 child_class->inherit_options(child_role, parent_is_format,
3836 parent->open_flags, parent->options);
3839 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3845 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3846 * Caution: getting a boolean member of @options requires care.
3847 * When @options come from -blockdev or blockdev_add, members are
3848 * typed according to the QAPI schema, but when they come from
3849 * -drive, they're all QString.
3851 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3852 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3853 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3855 flags &= ~BDRV_O_RDWR;
3858 if (flags & BDRV_O_SNAPSHOT) {
3859 snapshot_options = qdict_new();
3860 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3862 /* Let bdrv_backing_options() override "read-only" */
3863 qdict_del(options, BDRV_OPT_READ_ONLY);
3864 bdrv_inherited_options(BDRV_CHILD_COW, true,
3865 &flags, options, flags, options);
3868 bs->open_flags = flags;
3869 bs->options = options;
3870 options = qdict_clone_shallow(options);
3872 /* Find the right image format driver */
3873 /* See cautionary note on accessing @options above */
3874 drvname = qdict_get_try_str(options, "driver");
3876 drv = bdrv_find_format(drvname);
3878 error_setg(errp, "Unknown driver: '%s'", drvname);
3883 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3885 /* See cautionary note on accessing @options above */
3886 backing = qdict_get_try_str(options, "backing");
3887 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3888 (backing && *backing == '\0'))
3891 warn_report("Use of \"backing\": \"\" is deprecated; "
3892 "use \"backing\": null instead");
3894 flags |= BDRV_O_NO_BACKING;
3895 qdict_del(bs->explicit_options, "backing");
3896 qdict_del(bs->options, "backing");
3897 qdict_del(options, "backing");
3900 /* Open image file without format layer. This BlockBackend is only used for
3901 * probing, the block drivers will do their own bdrv_open_child() for the
3902 * same BDS, which is why we put the node name back into options. */
3903 if ((flags & BDRV_O_PROTOCOL) == 0) {
3904 BlockDriverState *file_bs;
3906 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3907 &child_of_bds, BDRV_CHILD_IMAGE,
3912 if (file_bs != NULL) {
3913 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3914 * looking at the header to guess the image format. This works even
3915 * in cases where a guest would not see a consistent state. */
3916 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3917 blk_insert_bs(file, file_bs, &local_err);
3918 bdrv_unref(file_bs);
3923 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3927 /* Image format probing */
3930 ret = find_image_format(file, filename, &drv, &local_err);
3935 * This option update would logically belong in bdrv_fill_options(),
3936 * but we first need to open bs->file for the probing to work, while
3937 * opening bs->file already requires the (mostly) final set of options
3938 * so that cache mode etc. can be inherited.
3940 * Adding the driver later is somewhat ugly, but it's not an option
3941 * that would ever be inherited, so it's correct. We just need to make
3942 * sure to update both bs->options (which has the full effective
3943 * options for bs) and options (which has file.* already removed).
3945 qdict_put_str(bs->options, "driver", drv->format_name);
3946 qdict_put_str(options, "driver", drv->format_name);
3948 error_setg(errp, "Must specify either driver or file");
3952 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3953 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3954 /* file must be NULL if a protocol BDS is about to be created
3955 * (the inverse results in an error message from bdrv_open_common()) */
3956 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3958 /* Open the image */
3959 ret = bdrv_open_common(bs, file, options, &local_err);
3969 /* If there is a backing file, use it */
3970 if ((flags & BDRV_O_NO_BACKING) == 0) {
3971 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3973 goto close_and_fail;
3977 /* Remove all children options and references
3978 * from bs->options and bs->explicit_options */
3979 QLIST_FOREACH(child, &bs->children, next) {
3980 char *child_key_dot;
3981 child_key_dot = g_strdup_printf("%s.", child->name);
3982 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3983 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3984 qdict_del(bs->explicit_options, child->name);
3985 qdict_del(bs->options, child->name);
3986 g_free(child_key_dot);
3989 /* Check if any unknown options were used */
3990 if (qdict_size(options) != 0) {
3991 const QDictEntry *entry = qdict_first(options);
3992 if (flags & BDRV_O_PROTOCOL) {
3993 error_setg(errp, "Block protocol '%s' doesn't support the option "
3994 "'%s'", drv->format_name, entry->key);
3997 "Block format '%s' does not support the option '%s'",
3998 drv->format_name, entry->key);
4001 goto close_and_fail;
4004 bdrv_parent_cb_change_media(bs, true);
4006 qobject_unref(options);
4009 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4010 * temporary snapshot afterwards. */
4011 if (snapshot_flags) {
4012 BlockDriverState *snapshot_bs;
4013 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4014 snapshot_options, &local_err);
4015 snapshot_options = NULL;
4017 goto close_and_fail;
4019 /* We are not going to return bs but the overlay on top of it
4020 * (snapshot_bs); thus, we have to drop the strong reference to bs
4021 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4022 * though, because the overlay still has a reference to it. */
4031 qobject_unref(snapshot_options);
4032 qobject_unref(bs->explicit_options);
4033 qobject_unref(bs->options);
4034 qobject_unref(options);
4036 bs->explicit_options = NULL;
4038 error_propagate(errp, local_err);
4043 qobject_unref(snapshot_options);
4044 qobject_unref(options);
4045 error_propagate(errp, local_err);
4049 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4050 QDict *options, int flags, Error **errp)
4052 GLOBAL_STATE_CODE();
4054 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4058 /* Return true if the NULL-terminated @list contains @str */
4059 static bool is_str_in_list(const char *str, const char *const *list)
4063 for (i = 0; list[i] != NULL; i++) {
4064 if (!strcmp(str, list[i])) {
4073 * Check that every option set in @bs->options is also set in
4076 * Options listed in the common_options list and in
4077 * @bs->drv->mutable_opts are skipped.
4079 * Return 0 on success, otherwise return -EINVAL and set @errp.
4081 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4082 const QDict *new_opts, Error **errp)
4084 const QDictEntry *e;
4085 /* These options are common to all block drivers and are handled
4086 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4087 const char *const common_options[] = {
4088 "node-name", "discard", "cache.direct", "cache.no-flush",
4089 "read-only", "auto-read-only", "detect-zeroes", NULL
4092 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4093 if (!qdict_haskey(new_opts, e->key) &&
4094 !is_str_in_list(e->key, common_options) &&
4095 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4096 error_setg(errp, "Option '%s' cannot be reset "
4097 "to its default value", e->key);
4106 * Returns true if @child can be reached recursively from @bs
4108 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4109 BlockDriverState *child)
4117 QLIST_FOREACH(c, &bs->children, next) {
4118 if (bdrv_recurse_has_child(c->bs, child)) {
4127 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4128 * reopen of multiple devices.
4130 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4131 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4132 * be created and initialized. This newly created BlockReopenQueue should be
4133 * passed back in for subsequent calls that are intended to be of the same
4136 * bs is the BlockDriverState to add to the reopen queue.
4138 * options contains the changed options for the associated bs
4139 * (the BlockReopenQueue takes ownership)
4141 * flags contains the open flags for the associated bs
4143 * returns a pointer to bs_queue, which is either the newly allocated
4144 * bs_queue, or the existing bs_queue being used.
4146 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4148 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4149 BlockDriverState *bs,
4151 const BdrvChildClass *klass,
4153 bool parent_is_format,
4154 QDict *parent_options,
4160 BlockReopenQueueEntry *bs_entry;
4162 QDict *old_options, *explicit_options, *options_copy;
4166 /* Make sure that the caller remembered to use a drained section. This is
4167 * important to avoid graph changes between the recursive queuing here and
4168 * bdrv_reopen_multiple(). */
4169 assert(bs->quiesce_counter > 0);
4170 GLOBAL_STATE_CODE();
4172 if (bs_queue == NULL) {
4173 bs_queue = g_new0(BlockReopenQueue, 1);
4174 QTAILQ_INIT(bs_queue);
4178 options = qdict_new();
4181 /* Check if this BlockDriverState is already in the queue */
4182 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4183 if (bs == bs_entry->state.bs) {
4189 * Precedence of options:
4190 * 1. Explicitly passed in options (highest)
4191 * 2. Retained from explicitly set options of bs
4192 * 3. Inherited from parent node
4193 * 4. Retained from effective options of bs
4196 /* Old explicitly set values (don't overwrite by inherited value) */
4197 if (bs_entry || keep_old_opts) {
4198 old_options = qdict_clone_shallow(bs_entry ?
4199 bs_entry->state.explicit_options :
4200 bs->explicit_options);
4201 bdrv_join_options(bs, options, old_options);
4202 qobject_unref(old_options);
4205 explicit_options = qdict_clone_shallow(options);
4207 /* Inherit from parent node */
4208 if (parent_options) {
4210 klass->inherit_options(role, parent_is_format, &flags, options,
4211 parent_flags, parent_options);
4213 flags = bdrv_get_flags(bs);
4216 if (keep_old_opts) {
4217 /* Old values are used for options that aren't set yet */
4218 old_options = qdict_clone_shallow(bs->options);
4219 bdrv_join_options(bs, options, old_options);
4220 qobject_unref(old_options);
4223 /* We have the final set of options so let's update the flags */
4224 options_copy = qdict_clone_shallow(options);
4225 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4226 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4227 update_flags_from_options(&flags, opts);
4228 qemu_opts_del(opts);
4229 qobject_unref(options_copy);
4231 /* bdrv_open_inherit() sets and clears some additional flags internally */
4232 flags &= ~BDRV_O_PROTOCOL;
4233 if (flags & BDRV_O_RDWR) {
4234 flags |= BDRV_O_ALLOW_RDWR;
4238 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4239 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4241 qobject_unref(bs_entry->state.options);
4242 qobject_unref(bs_entry->state.explicit_options);
4245 bs_entry->state.bs = bs;
4246 bs_entry->state.options = options;
4247 bs_entry->state.explicit_options = explicit_options;
4248 bs_entry->state.flags = flags;
4251 * If keep_old_opts is false then it means that unspecified
4252 * options must be reset to their original value. We don't allow
4253 * resetting 'backing' but we need to know if the option is
4254 * missing in order to decide if we have to return an error.
4256 if (!keep_old_opts) {
4257 bs_entry->state.backing_missing =
4258 !qdict_haskey(options, "backing") &&
4259 !qdict_haskey(options, "backing.driver");
4262 QLIST_FOREACH(child, &bs->children, next) {
4263 QDict *new_child_options = NULL;
4264 bool child_keep_old = keep_old_opts;
4266 /* reopen can only change the options of block devices that were
4267 * implicitly created and inherited options. For other (referenced)
4268 * block devices, a syntax like "backing.foo" results in an error. */
4269 if (child->bs->inherits_from != bs) {
4273 /* Check if the options contain a child reference */
4274 if (qdict_haskey(options, child->name)) {
4275 const char *childref = qdict_get_try_str(options, child->name);
4277 * The current child must not be reopened if the child
4278 * reference is null or points to a different node.
4280 if (g_strcmp0(childref, child->bs->node_name)) {
4284 * If the child reference points to the current child then
4285 * reopen it with its existing set of options (note that
4286 * it can still inherit new options from the parent).
4288 child_keep_old = true;
4290 /* Extract child options ("child-name.*") */
4291 char *child_key_dot = g_strdup_printf("%s.", child->name);
4292 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4293 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4294 g_free(child_key_dot);
4297 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4298 child->klass, child->role, bs->drv->is_format,
4299 options, flags, child_keep_old);
4305 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4306 BlockDriverState *bs,
4307 QDict *options, bool keep_old_opts)
4309 GLOBAL_STATE_CODE();
4311 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4312 NULL, 0, keep_old_opts);
4315 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4317 GLOBAL_STATE_CODE();
4319 BlockReopenQueueEntry *bs_entry, *next;
4320 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4321 qobject_unref(bs_entry->state.explicit_options);
4322 qobject_unref(bs_entry->state.options);
4330 * Reopen multiple BlockDriverStates atomically & transactionally.
4332 * The queue passed in (bs_queue) must have been built up previous
4333 * via bdrv_reopen_queue().
4335 * Reopens all BDS specified in the queue, with the appropriate
4336 * flags. All devices are prepared for reopen, and failure of any
4337 * device will cause all device changes to be abandoned, and intermediate
4340 * If all devices prepare successfully, then the changes are committed
4343 * All affected nodes must be drained between bdrv_reopen_queue() and
4344 * bdrv_reopen_multiple().
4346 * To be called from the main thread, with all other AioContexts unlocked.
4348 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4351 BlockReopenQueueEntry *bs_entry, *next;
4353 Transaction *tran = tran_new();
4354 g_autoptr(GHashTable) found = NULL;
4355 g_autoptr(GSList) refresh_list = NULL;
4357 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4358 assert(bs_queue != NULL);
4359 GLOBAL_STATE_CODE();
4361 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4362 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4363 aio_context_acquire(ctx);
4364 ret = bdrv_flush(bs_entry->state.bs);
4365 aio_context_release(ctx);
4367 error_setg_errno(errp, -ret, "Error flushing drive");
4372 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4373 assert(bs_entry->state.bs->quiesce_counter > 0);
4374 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4375 aio_context_acquire(ctx);
4376 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4377 aio_context_release(ctx);
4381 bs_entry->prepared = true;
4384 found = g_hash_table_new(NULL, NULL);
4385 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4386 BDRVReopenState *state = &bs_entry->state;
4388 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4389 if (state->old_backing_bs) {
4390 refresh_list = bdrv_topological_dfs(refresh_list, found,
4391 state->old_backing_bs);
4393 if (state->old_file_bs) {
4394 refresh_list = bdrv_topological_dfs(refresh_list, found,
4395 state->old_file_bs);
4400 * Note that file-posix driver rely on permission update done during reopen
4401 * (even if no permission changed), because it wants "new" permissions for
4402 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4403 * in raw_reopen_prepare() which is called with "old" permissions.
4405 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4411 * If we reach this point, we have success and just need to apply the
4414 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4415 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4416 * children are usually goes after parents in reopen-queue, so go from last
4419 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4420 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4421 aio_context_acquire(ctx);
4422 bdrv_reopen_commit(&bs_entry->state);
4423 aio_context_release(ctx);
4428 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4429 BlockDriverState *bs = bs_entry->state.bs;
4431 if (bs->drv->bdrv_reopen_commit_post) {
4432 ctx = bdrv_get_aio_context(bs);
4433 aio_context_acquire(ctx);
4434 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4435 aio_context_release(ctx);
4444 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4445 if (bs_entry->prepared) {
4446 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4447 aio_context_acquire(ctx);
4448 bdrv_reopen_abort(&bs_entry->state);
4449 aio_context_release(ctx);
4454 bdrv_reopen_queue_free(bs_queue);
4459 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4462 AioContext *ctx = bdrv_get_aio_context(bs);
4463 BlockReopenQueue *queue;
4466 GLOBAL_STATE_CODE();
4468 bdrv_subtree_drained_begin(bs);
4469 if (ctx != qemu_get_aio_context()) {
4470 aio_context_release(ctx);
4473 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4474 ret = bdrv_reopen_multiple(queue, errp);
4476 if (ctx != qemu_get_aio_context()) {
4477 aio_context_acquire(ctx);
4479 bdrv_subtree_drained_end(bs);
4484 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4487 QDict *opts = qdict_new();
4489 GLOBAL_STATE_CODE();
4491 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4493 return bdrv_reopen(bs, opts, true, errp);
4497 * Take a BDRVReopenState and check if the value of 'backing' in the
4498 * reopen_state->options QDict is valid or not.
4500 * If 'backing' is missing from the QDict then return 0.
4502 * If 'backing' contains the node name of the backing file of
4503 * reopen_state->bs then return 0.
4505 * If 'backing' contains a different node name (or is null) then check
4506 * whether the current backing file can be replaced with the new one.
4507 * If that's the case then reopen_state->replace_backing_bs is set to
4508 * true and reopen_state->new_backing_bs contains a pointer to the new
4509 * backing BlockDriverState (or NULL).
4511 * Return 0 on success, otherwise return < 0 and set @errp.
4513 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4514 bool is_backing, Transaction *tran,
4517 BlockDriverState *bs = reopen_state->bs;
4518 BlockDriverState *new_child_bs;
4519 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4521 const char *child_name = is_backing ? "backing" : "file";
4525 GLOBAL_STATE_CODE();
4527 value = qdict_get(reopen_state->options, child_name);
4528 if (value == NULL) {
4532 switch (qobject_type(value)) {
4534 assert(is_backing); /* The 'file' option does not allow a null value */
4535 new_child_bs = NULL;
4538 str = qstring_get_str(qobject_to(QString, value));
4539 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4540 if (new_child_bs == NULL) {
4542 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4543 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4544 "cycle", str, child_name, bs->node_name);
4550 * The options QDict has been flattened, so 'backing' and 'file'
4551 * do not allow any other data type here.
4553 g_assert_not_reached();
4556 if (old_child_bs == new_child_bs) {
4561 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4565 if (old_child_bs->implicit) {
4566 error_setg(errp, "Cannot replace implicit %s child of %s",
4567 child_name, bs->node_name);
4572 if (bs->drv->is_filter && !old_child_bs) {
4574 * Filters always have a file or a backing child, so we are trying to
4575 * change wrong child
4577 error_setg(errp, "'%s' is a %s filter node that does not support a "
4578 "%s child", bs->node_name, bs->drv->format_name, child_name);
4583 reopen_state->old_backing_bs = old_child_bs;
4585 reopen_state->old_file_bs = old_child_bs;
4588 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4593 * Prepares a BlockDriverState for reopen. All changes are staged in the
4594 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4595 * the block driver layer .bdrv_reopen_prepare()
4597 * bs is the BlockDriverState to reopen
4598 * flags are the new open flags
4599 * queue is the reopen queue
4601 * Returns 0 on success, non-zero on error. On error errp will be set
4604 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4605 * It is the responsibility of the caller to then call the abort() or
4606 * commit() for any other BDS that have been left in a prepare() state
4609 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4610 BlockReopenQueue *queue,
4611 Transaction *change_child_tran, Error **errp)
4615 Error *local_err = NULL;
4618 QDict *orig_reopen_opts;
4619 char *discard = NULL;
4621 bool drv_prepared = false;
4623 assert(reopen_state != NULL);
4624 assert(reopen_state->bs->drv != NULL);
4625 GLOBAL_STATE_CODE();
4626 drv = reopen_state->bs->drv;
4628 /* This function and each driver's bdrv_reopen_prepare() remove
4629 * entries from reopen_state->options as they are processed, so
4630 * we need to make a copy of the original QDict. */
4631 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4633 /* Process generic block layer options */
4634 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4635 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4640 /* This was already called in bdrv_reopen_queue_child() so the flags
4641 * are up-to-date. This time we simply want to remove the options from
4642 * QemuOpts in order to indicate that they have been processed. */
4643 old_flags = reopen_state->flags;
4644 update_flags_from_options(&reopen_state->flags, opts);
4645 assert(old_flags == reopen_state->flags);
4647 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4648 if (discard != NULL) {
4649 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4650 error_setg(errp, "Invalid discard option");
4656 reopen_state->detect_zeroes =
4657 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4659 error_propagate(errp, local_err);
4664 /* All other options (including node-name and driver) must be unchanged.
4665 * Put them back into the QDict, so that they are checked at the end
4666 * of this function. */
4667 qemu_opts_to_qdict(opts, reopen_state->options);
4669 /* If we are to stay read-only, do not allow permission change
4670 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4671 * not set, or if the BDS still has copy_on_read enabled */
4672 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4673 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4675 error_propagate(errp, local_err);
4679 if (drv->bdrv_reopen_prepare) {
4681 * If a driver-specific option is missing, it means that we
4682 * should reset it to its default value.
4683 * But not all options allow that, so we need to check it first.
4685 ret = bdrv_reset_options_allowed(reopen_state->bs,
4686 reopen_state->options, errp);
4691 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4693 if (local_err != NULL) {
4694 error_propagate(errp, local_err);
4696 bdrv_refresh_filename(reopen_state->bs);
4697 error_setg(errp, "failed while preparing to reopen image '%s'",
4698 reopen_state->bs->filename);
4703 /* It is currently mandatory to have a bdrv_reopen_prepare()
4704 * handler for each supported drv. */
4705 error_setg(errp, "Block format '%s' used by node '%s' "
4706 "does not support reopening files", drv->format_name,
4707 bdrv_get_device_or_node_name(reopen_state->bs));
4712 drv_prepared = true;
4715 * We must provide the 'backing' option if the BDS has a backing
4716 * file or if the image file has a backing file name as part of
4717 * its metadata. Otherwise the 'backing' option can be omitted.
4719 if (drv->supports_backing && reopen_state->backing_missing &&
4720 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4721 error_setg(errp, "backing is missing for '%s'",
4722 reopen_state->bs->node_name);
4728 * Allow changing the 'backing' option. The new value can be
4729 * either a reference to an existing node (using its node name)
4730 * or NULL to simply detach the current backing file.
4732 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4733 change_child_tran, errp);
4737 qdict_del(reopen_state->options, "backing");
4739 /* Allow changing the 'file' option. In this case NULL is not allowed */
4740 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4741 change_child_tran, errp);
4745 qdict_del(reopen_state->options, "file");
4747 /* Options that are not handled are only okay if they are unchanged
4748 * compared to the old state. It is expected that some options are only
4749 * used for the initial open, but not reopen (e.g. filename) */
4750 if (qdict_size(reopen_state->options)) {
4751 const QDictEntry *entry = qdict_first(reopen_state->options);
4754 QObject *new = entry->value;
4755 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4757 /* Allow child references (child_name=node_name) as long as they
4758 * point to the current child (i.e. everything stays the same). */
4759 if (qobject_type(new) == QTYPE_QSTRING) {
4761 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4762 if (!strcmp(child->name, entry->key)) {
4768 if (!strcmp(child->bs->node_name,
4769 qstring_get_str(qobject_to(QString, new)))) {
4770 continue; /* Found child with this name, skip option */
4776 * TODO: When using -drive to specify blockdev options, all values
4777 * will be strings; however, when using -blockdev, blockdev-add or
4778 * filenames using the json:{} pseudo-protocol, they will be
4780 * In contrast, reopening options are (currently) always strings
4781 * (because you can only specify them through qemu-io; all other
4782 * callers do not specify any options).
4783 * Therefore, when using anything other than -drive to create a BDS,
4784 * this cannot detect non-string options as unchanged, because
4785 * qobject_is_equal() always returns false for objects of different
4786 * type. In the future, this should be remedied by correctly typing
4787 * all options. For now, this is not too big of an issue because
4788 * the user can simply omit options which cannot be changed anyway,
4789 * so they will stay unchanged.
4791 if (!qobject_is_equal(new, old)) {
4792 error_setg(errp, "Cannot change the option '%s'", entry->key);
4796 } while ((entry = qdict_next(reopen_state->options, entry)));
4801 /* Restore the original reopen_state->options QDict */
4802 qobject_unref(reopen_state->options);
4803 reopen_state->options = qobject_ref(orig_reopen_opts);
4806 if (ret < 0 && drv_prepared) {
4807 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4808 * call drv->bdrv_reopen_abort() before signaling an error
4809 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4810 * when the respective bdrv_reopen_prepare() has failed) */
4811 if (drv->bdrv_reopen_abort) {
4812 drv->bdrv_reopen_abort(reopen_state);
4815 qemu_opts_del(opts);
4816 qobject_unref(orig_reopen_opts);
4822 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4823 * makes them final by swapping the staging BlockDriverState contents into
4824 * the active BlockDriverState contents.
4826 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4829 BlockDriverState *bs;
4832 assert(reopen_state != NULL);
4833 bs = reopen_state->bs;
4835 assert(drv != NULL);
4836 GLOBAL_STATE_CODE();
4838 /* If there are any driver level actions to take */
4839 if (drv->bdrv_reopen_commit) {
4840 drv->bdrv_reopen_commit(reopen_state);
4843 /* set BDS specific flags now */
4844 qobject_unref(bs->explicit_options);
4845 qobject_unref(bs->options);
4846 qobject_ref(reopen_state->explicit_options);
4847 qobject_ref(reopen_state->options);
4849 bs->explicit_options = reopen_state->explicit_options;
4850 bs->options = reopen_state->options;
4851 bs->open_flags = reopen_state->flags;
4852 bs->detect_zeroes = reopen_state->detect_zeroes;
4854 /* Remove child references from bs->options and bs->explicit_options.
4855 * Child options were already removed in bdrv_reopen_queue_child() */
4856 QLIST_FOREACH(child, &bs->children, next) {
4857 qdict_del(bs->explicit_options, child->name);
4858 qdict_del(bs->options, child->name);
4860 /* backing is probably removed, so it's not handled by previous loop */
4861 qdict_del(bs->explicit_options, "backing");
4862 qdict_del(bs->options, "backing");
4864 bdrv_refresh_limits(bs, NULL, NULL);
4868 * Abort the reopen, and delete and free the staged changes in
4871 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4875 assert(reopen_state != NULL);
4876 drv = reopen_state->bs->drv;
4877 assert(drv != NULL);
4878 GLOBAL_STATE_CODE();
4880 if (drv->bdrv_reopen_abort) {
4881 drv->bdrv_reopen_abort(reopen_state);
4886 static void bdrv_close(BlockDriverState *bs)
4888 BdrvAioNotifier *ban, *ban_next;
4889 BdrvChild *child, *next;
4891 GLOBAL_STATE_CODE();
4892 assert(!bs->refcnt);
4894 bdrv_drained_begin(bs); /* complete I/O */
4896 bdrv_drain(bs); /* in case flush left pending I/O */
4899 if (bs->drv->bdrv_close) {
4900 /* Must unfreeze all children, so bdrv_unref_child() works */
4901 bs->drv->bdrv_close(bs);
4906 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4907 bdrv_unref_child(bs, child);
4910 assert(!bs->backing);
4914 qatomic_set(&bs->copy_on_read, 0);
4915 bs->backing_file[0] = '\0';
4916 bs->backing_format[0] = '\0';
4917 bs->total_sectors = 0;
4918 bs->encrypted = false;
4920 qobject_unref(bs->options);
4921 qobject_unref(bs->explicit_options);
4923 bs->explicit_options = NULL;
4924 qobject_unref(bs->full_open_options);
4925 bs->full_open_options = NULL;
4926 g_free(bs->block_status_cache);
4927 bs->block_status_cache = NULL;
4929 bdrv_release_named_dirty_bitmaps(bs);
4930 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4932 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4935 QLIST_INIT(&bs->aio_notifiers);
4936 bdrv_drained_end(bs);
4939 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4940 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4943 if (bs->quiesce_counter) {
4944 bdrv_drain_all_end_quiesce(bs);
4948 void bdrv_close_all(void)
4950 GLOBAL_STATE_CODE();
4951 assert(job_next(NULL) == NULL);
4953 /* Drop references from requests still in flight, such as canceled block
4954 * jobs whose AIO context has not been polled yet */
4957 blk_remove_all_bs();
4958 blockdev_close_all_bdrv_states();
4960 assert(QTAILQ_EMPTY(&all_bdrv_states));
4963 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4969 if (c->klass->stay_at_node) {
4973 /* If the child @c belongs to the BDS @to, replacing the current
4974 * c->bs by @to would mean to create a loop.
4976 * Such a case occurs when appending a BDS to a backing chain.
4977 * For instance, imagine the following chain:
4979 * guest device -> node A -> further backing chain...
4981 * Now we create a new BDS B which we want to put on top of this
4982 * chain, so we first attach A as its backing node:
4987 * guest device -> node A -> further backing chain...
4989 * Finally we want to replace A by B. When doing that, we want to
4990 * replace all pointers to A by pointers to B -- except for the
4991 * pointer from B because (1) that would create a loop, and (2)
4992 * that pointer should simply stay intact:
4994 * guest device -> node B
4997 * node A -> further backing chain...
4999 * In general, when replacing a node A (c->bs) by a node B (@to),
5000 * if A is a child of B, that means we cannot replace A by B there
5001 * because that would create a loop. Silently detaching A from B
5002 * is also not really an option. So overall just leaving A in
5003 * place there is the most sensible choice.
5005 * We would also create a loop in any cases where @c is only
5006 * indirectly referenced by @to. Prevent this by returning false
5007 * if @c is found (by breadth-first search) anywhere in the whole
5012 found = g_hash_table_new(NULL, NULL);
5013 g_hash_table_add(found, to);
5014 queue = g_queue_new();
5015 g_queue_push_tail(queue, to);
5017 while (!g_queue_is_empty(queue)) {
5018 BlockDriverState *v = g_queue_pop_head(queue);
5021 QLIST_FOREACH(c2, &v->children, next) {
5027 if (g_hash_table_contains(found, c2->bs)) {
5031 g_queue_push_tail(queue, c2->bs);
5032 g_hash_table_add(found, c2->bs);
5036 g_queue_free(queue);
5037 g_hash_table_destroy(found);
5042 static void bdrv_remove_child_commit(void *opaque)
5044 GLOBAL_STATE_CODE();
5045 bdrv_child_free(opaque);
5048 static TransactionActionDrv bdrv_remove_child_drv = {
5049 .commit = bdrv_remove_child_commit,
5052 /* Function doesn't update permissions, caller is responsible for this. */
5053 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5060 bdrv_replace_child_tran(child, NULL, tran);
5063 tran_add(tran, &bdrv_remove_child_drv, child);
5067 * A function to remove backing-chain child of @bs if exists: cow child for
5068 * format nodes (always .backing) and filter child for filters (may be .file or
5071 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
5074 bdrv_remove_child(bdrv_filter_or_cow_child(bs), tran);
5077 static int bdrv_replace_node_noperm(BlockDriverState *from,
5078 BlockDriverState *to,
5079 bool auto_skip, Transaction *tran,
5082 BdrvChild *c, *next;
5084 GLOBAL_STATE_CODE();
5086 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5087 assert(c->bs == from);
5088 if (!should_update_child(c, to)) {
5092 error_setg(errp, "Should not change '%s' link to '%s'",
5093 c->name, from->node_name);
5097 error_setg(errp, "Cannot change '%s' link to '%s'",
5098 c->name, from->node_name);
5101 bdrv_replace_child_tran(c, to, tran);
5108 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5109 * if it creates a parent-child relation loop or if parent is block-job.
5111 * With auto_skip=false the error is returned if from has a parent which should
5114 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5115 * case backing link of the cow-parent of @to is removed.
5117 static int bdrv_replace_node_common(BlockDriverState *from,
5118 BlockDriverState *to,
5119 bool auto_skip, bool detach_subchain,
5122 Transaction *tran = tran_new();
5123 g_autoptr(GHashTable) found = NULL;
5124 g_autoptr(GSList) refresh_list = NULL;
5125 BlockDriverState *to_cow_parent = NULL;
5128 GLOBAL_STATE_CODE();
5130 if (detach_subchain) {
5131 assert(bdrv_chain_contains(from, to));
5133 for (to_cow_parent = from;
5134 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5135 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5141 /* Make sure that @from doesn't go away until we have successfully attached
5142 * all of its parents to @to. */
5145 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5146 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5147 bdrv_drained_begin(from);
5150 * Do the replacement without permission update.
5151 * Replacement may influence the permissions, we should calculate new
5152 * permissions based on new graph. If we fail, we'll roll-back the
5155 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5160 if (detach_subchain) {
5161 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
5164 found = g_hash_table_new(NULL, NULL);
5166 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
5167 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
5169 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5177 tran_finalize(tran, ret);
5179 bdrv_drained_end(from);
5185 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5188 GLOBAL_STATE_CODE();
5190 return bdrv_replace_node_common(from, to, true, false, errp);
5193 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5195 GLOBAL_STATE_CODE();
5197 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5202 * Add new bs contents at the top of an image chain while the chain is
5203 * live, while keeping required fields on the top layer.
5205 * This will modify the BlockDriverState fields, and swap contents
5206 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5208 * bs_new must not be attached to a BlockBackend and must not have backing
5211 * This function does not create any image files.
5213 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5218 Transaction *tran = tran_new();
5220 GLOBAL_STATE_CODE();
5222 assert(!bs_new->backing);
5224 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5225 &child_of_bds, bdrv_backing_role(bs_new),
5232 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5237 ret = bdrv_refresh_perms(bs_new, errp);
5239 tran_finalize(tran, ret);
5241 bdrv_refresh_limits(bs_top, NULL, NULL);
5246 /* Not for empty child */
5247 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5251 Transaction *tran = tran_new();
5252 g_autoptr(GHashTable) found = NULL;
5253 g_autoptr(GSList) refresh_list = NULL;
5254 BlockDriverState *old_bs = child->bs;
5256 GLOBAL_STATE_CODE();
5259 bdrv_drained_begin(old_bs);
5260 bdrv_drained_begin(new_bs);
5262 bdrv_replace_child_tran(child, new_bs, tran);
5264 found = g_hash_table_new(NULL, NULL);
5265 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5266 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5268 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5270 tran_finalize(tran, ret);
5272 bdrv_drained_end(old_bs);
5273 bdrv_drained_end(new_bs);
5279 static void bdrv_delete(BlockDriverState *bs)
5281 assert(bdrv_op_blocker_is_empty(bs));
5282 assert(!bs->refcnt);
5283 GLOBAL_STATE_CODE();
5285 /* remove from list, if necessary */
5286 if (bs->node_name[0] != '\0') {
5287 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5289 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5298 * Replace @bs by newly created block node.
5300 * @options is a QDict of options to pass to the block drivers, or NULL for an
5301 * empty set of options. The reference to the QDict belongs to the block layer
5302 * after the call (even on failure), so if the caller intends to reuse the
5303 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5305 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5306 int flags, Error **errp)
5310 BlockDriverState *new_node_bs = NULL;
5311 const char *drvname, *node_name;
5314 drvname = qdict_get_try_str(options, "driver");
5316 error_setg(errp, "driver is not specified");
5320 drv = bdrv_find_format(drvname);
5322 error_setg(errp, "Unknown driver: '%s'", drvname);
5326 node_name = qdict_get_try_str(options, "node-name");
5328 GLOBAL_STATE_CODE();
5330 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5332 options = NULL; /* bdrv_new_open_driver() eats options */
5334 error_prepend(errp, "Could not create node: ");
5338 bdrv_drained_begin(bs);
5339 ret = bdrv_replace_node(bs, new_node_bs, errp);
5340 bdrv_drained_end(bs);
5343 error_prepend(errp, "Could not replace node: ");
5350 qobject_unref(options);
5351 bdrv_unref(new_node_bs);
5356 * Run consistency checks on an image
5358 * Returns 0 if the check could be completed (it doesn't mean that the image is
5359 * free of errors) or -errno when an internal error occurred. The results of the
5360 * check are stored in res.
5362 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5363 BdrvCheckResult *res, BdrvCheckMode fix)
5366 if (bs->drv == NULL) {
5369 if (bs->drv->bdrv_co_check == NULL) {
5373 memset(res, 0, sizeof(*res));
5374 return bs->drv->bdrv_co_check(bs, res, fix);
5380 * -EINVAL - backing format specified, but no file
5381 * -ENOSPC - can't update the backing file because no space is left in the
5383 * -ENOTSUP - format driver doesn't support changing the backing file
5385 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5386 const char *backing_fmt, bool require)
5388 BlockDriver *drv = bs->drv;
5391 GLOBAL_STATE_CODE();
5397 /* Backing file format doesn't make sense without a backing file */
5398 if (backing_fmt && !backing_file) {
5402 if (require && backing_file && !backing_fmt) {
5406 if (drv->bdrv_change_backing_file != NULL) {
5407 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5413 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5414 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5415 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5416 backing_file ?: "");
5422 * Finds the first non-filter node above bs in the chain between
5423 * active and bs. The returned node is either an immediate parent of
5424 * bs, or there are only filter nodes between the two.
5426 * Returns NULL if bs is not found in active's image chain,
5427 * or if active == bs.
5429 * Returns the bottommost base image if bs == NULL.
5431 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5432 BlockDriverState *bs)
5435 GLOBAL_STATE_CODE();
5437 bs = bdrv_skip_filters(bs);
5438 active = bdrv_skip_filters(active);
5441 BlockDriverState *next = bdrv_backing_chain_next(active);
5451 /* Given a BDS, searches for the base layer. */
5452 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5454 GLOBAL_STATE_CODE();
5456 return bdrv_find_overlay(bs, NULL);
5460 * Return true if at least one of the COW (backing) and filter links
5461 * between @bs and @base is frozen. @errp is set if that's the case.
5462 * @base must be reachable from @bs, or NULL.
5464 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5467 BlockDriverState *i;
5470 GLOBAL_STATE_CODE();
5472 for (i = bs; i != base; i = child_bs(child)) {
5473 child = bdrv_filter_or_cow_child(i);
5475 if (child && child->frozen) {
5476 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5477 child->name, i->node_name, child->bs->node_name);
5486 * Freeze all COW (backing) and filter links between @bs and @base.
5487 * If any of the links is already frozen the operation is aborted and
5488 * none of the links are modified.
5489 * @base must be reachable from @bs, or NULL.
5490 * Returns 0 on success. On failure returns < 0 and sets @errp.
5492 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5495 BlockDriverState *i;
5498 GLOBAL_STATE_CODE();
5500 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5504 for (i = bs; i != base; i = child_bs(child)) {
5505 child = bdrv_filter_or_cow_child(i);
5506 if (child && child->bs->never_freeze) {
5507 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5508 child->name, child->bs->node_name);
5513 for (i = bs; i != base; i = child_bs(child)) {
5514 child = bdrv_filter_or_cow_child(i);
5516 child->frozen = true;
5524 * Unfreeze all COW (backing) and filter links between @bs and @base.
5525 * The caller must ensure that all links are frozen before using this
5527 * @base must be reachable from @bs, or NULL.
5529 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5531 BlockDriverState *i;
5534 GLOBAL_STATE_CODE();
5536 for (i = bs; i != base; i = child_bs(child)) {
5537 child = bdrv_filter_or_cow_child(i);
5539 assert(child->frozen);
5540 child->frozen = false;
5546 * Drops images above 'base' up to and including 'top', and sets the image
5547 * above 'top' to have base as its backing file.
5549 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5550 * information in 'bs' can be properly updated.
5552 * E.g., this will convert the following chain:
5553 * bottom <- base <- intermediate <- top <- active
5557 * bottom <- base <- active
5559 * It is allowed for bottom==base, in which case it converts:
5561 * base <- intermediate <- top <- active
5567 * If backing_file_str is non-NULL, it will be used when modifying top's
5568 * overlay image metadata.
5571 * if active == top, that is considered an error
5574 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5575 const char *backing_file_str)
5577 BlockDriverState *explicit_top = top;
5578 bool update_inherits_from;
5580 Error *local_err = NULL;
5582 g_autoptr(GSList) updated_children = NULL;
5585 GLOBAL_STATE_CODE();
5588 bdrv_subtree_drained_begin(top);
5590 if (!top->drv || !base->drv) {
5594 /* Make sure that base is in the backing chain of top */
5595 if (!bdrv_chain_contains(top, base)) {
5599 /* If 'base' recursively inherits from 'top' then we should set
5600 * base->inherits_from to top->inherits_from after 'top' and all
5601 * other intermediate nodes have been dropped.
5602 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5603 * it because no one inherits from it. We use explicit_top for that. */
5604 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5605 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5607 /* success - we can delete the intermediate states, and link top->base */
5608 if (!backing_file_str) {
5609 bdrv_refresh_filename(base);
5610 backing_file_str = base->filename;
5613 QLIST_FOREACH(c, &top->parents, next_parent) {
5614 updated_children = g_slist_prepend(updated_children, c);
5618 * It seems correct to pass detach_subchain=true here, but it triggers
5619 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5620 * another drained section, which modify the graph (for example, removing
5621 * the child, which we keep in updated_children list). So, it's a TODO.
5623 * Note, bug triggered if pass detach_subchain=true here and run
5624 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5627 bdrv_replace_node_common(top, base, false, false, &local_err);
5629 error_report_err(local_err);
5633 for (p = updated_children; p; p = p->next) {
5636 if (c->klass->update_filename) {
5637 ret = c->klass->update_filename(c, base, backing_file_str,
5641 * TODO: Actually, we want to rollback all previous iterations
5642 * of this loop, and (which is almost impossible) previous
5643 * bdrv_replace_node()...
5645 * Note, that c->klass->update_filename may lead to permission
5646 * update, so it's a bad idea to call it inside permission
5647 * update transaction of bdrv_replace_node.
5649 error_report_err(local_err);
5655 if (update_inherits_from) {
5656 base->inherits_from = explicit_top->inherits_from;
5661 bdrv_subtree_drained_end(top);
5667 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5668 * sums the size of all data-bearing children. (This excludes backing
5671 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5674 int64_t child_size, sum = 0;
5676 QLIST_FOREACH(child, &bs->children, next) {
5677 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5678 BDRV_CHILD_FILTERED))
5680 child_size = bdrv_get_allocated_file_size(child->bs);
5681 if (child_size < 0) {
5692 * Length of a allocated file in bytes. Sparse files are counted by actual
5693 * allocated space. Return < 0 if error or unknown.
5695 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5697 BlockDriver *drv = bs->drv;
5703 if (drv->bdrv_get_allocated_file_size) {
5704 return drv->bdrv_get_allocated_file_size(bs);
5707 if (drv->bdrv_file_open) {
5709 * Protocol drivers default to -ENOTSUP (most of their data is
5710 * not stored in any of their children (if they even have any),
5711 * so there is no generic way to figure it out).
5714 } else if (drv->is_filter) {
5715 /* Filter drivers default to the size of their filtered child */
5716 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5718 /* Other drivers default to summing their children's sizes */
5719 return bdrv_sum_allocated_file_size(bs);
5725 * @drv: Format driver
5726 * @opts: Creation options for new image
5727 * @in_bs: Existing image containing data for new image (may be NULL)
5728 * @errp: Error object
5729 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5732 * Calculate file size required to create a new image.
5734 * If @in_bs is given then space for allocated clusters and zero clusters
5735 * from that image are included in the calculation. If @opts contains a
5736 * backing file that is shared by @in_bs then backing clusters may be omitted
5737 * from the calculation.
5739 * If @in_bs is NULL then the calculation includes no allocated clusters
5740 * unless a preallocation option is given in @opts.
5742 * Note that @in_bs may use a different BlockDriver from @drv.
5744 * If an error occurs the @errp pointer is set.
5746 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5747 BlockDriverState *in_bs, Error **errp)
5750 if (!drv->bdrv_measure) {
5751 error_setg(errp, "Block driver '%s' does not support size measurement",
5756 return drv->bdrv_measure(opts, in_bs, errp);
5760 * Return number of sectors on success, -errno on error.
5762 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5764 BlockDriver *drv = bs->drv;
5770 if (drv->has_variable_length) {
5771 int ret = refresh_total_sectors(bs, bs->total_sectors);
5776 return bs->total_sectors;
5780 * Return length in bytes on success, -errno on error.
5781 * The length is always a multiple of BDRV_SECTOR_SIZE.
5783 int64_t bdrv_getlength(BlockDriverState *bs)
5785 int64_t ret = bdrv_nb_sectors(bs);
5791 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5794 return ret * BDRV_SECTOR_SIZE;
5797 /* return 0 as number of sectors if no device present or error */
5798 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5800 int64_t nb_sectors = bdrv_nb_sectors(bs);
5803 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5806 bool bdrv_is_sg(BlockDriverState *bs)
5813 * Return whether the given node supports compressed writes.
5815 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5817 BlockDriverState *filtered;
5820 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5824 filtered = bdrv_filter_bs(bs);
5827 * Filters can only forward compressed writes, so we have to
5830 return bdrv_supports_compressed_writes(filtered);
5836 const char *bdrv_get_format_name(BlockDriverState *bs)
5839 return bs->drv ? bs->drv->format_name : NULL;
5842 static int qsort_strcmp(const void *a, const void *b)
5844 return strcmp(*(char *const *)a, *(char *const *)b);
5847 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5848 void *opaque, bool read_only)
5853 const char **formats = NULL;
5855 GLOBAL_STATE_CODE();
5857 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5858 if (drv->format_name) {
5862 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5866 while (formats && i && !found) {
5867 found = !strcmp(formats[--i], drv->format_name);
5871 formats = g_renew(const char *, formats, count + 1);
5872 formats[count++] = drv->format_name;
5877 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5878 const char *format_name = block_driver_modules[i].format_name;
5884 if (use_bdrv_whitelist &&
5885 !bdrv_format_is_whitelisted(format_name, read_only)) {
5889 while (formats && j && !found) {
5890 found = !strcmp(formats[--j], format_name);
5894 formats = g_renew(const char *, formats, count + 1);
5895 formats[count++] = format_name;
5900 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5902 for (i = 0; i < count; i++) {
5903 it(opaque, formats[i]);
5909 /* This function is to find a node in the bs graph */
5910 BlockDriverState *bdrv_find_node(const char *node_name)
5912 BlockDriverState *bs;
5915 GLOBAL_STATE_CODE();
5917 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5918 if (!strcmp(node_name, bs->node_name)) {
5925 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5926 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5929 BlockDeviceInfoList *list;
5930 BlockDriverState *bs;
5932 GLOBAL_STATE_CODE();
5935 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5936 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5938 qapi_free_BlockDeviceInfoList(list);
5941 QAPI_LIST_PREPEND(list, info);
5947 typedef struct XDbgBlockGraphConstructor {
5948 XDbgBlockGraph *graph;
5949 GHashTable *graph_nodes;
5950 } XDbgBlockGraphConstructor;
5952 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5954 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5956 gr->graph = g_new0(XDbgBlockGraph, 1);
5957 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5962 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5964 XDbgBlockGraph *graph = gr->graph;
5966 g_hash_table_destroy(gr->graph_nodes);
5972 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5974 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5981 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5982 * answer of g_hash_table_lookup.
5984 ret = g_hash_table_size(gr->graph_nodes) + 1;
5985 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5990 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5991 XDbgBlockGraphNodeType type, const char *name)
5993 XDbgBlockGraphNode *n;
5995 n = g_new0(XDbgBlockGraphNode, 1);
5997 n->id = xdbg_graph_node_num(gr, node);
5999 n->name = g_strdup(name);
6001 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6004 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6005 const BdrvChild *child)
6007 BlockPermission qapi_perm;
6008 XDbgBlockGraphEdge *edge;
6009 GLOBAL_STATE_CODE();
6011 edge = g_new0(XDbgBlockGraphEdge, 1);
6013 edge->parent = xdbg_graph_node_num(gr, parent);
6014 edge->child = xdbg_graph_node_num(gr, child->bs);
6015 edge->name = g_strdup(child->name);
6017 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6018 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6020 if (flag & child->perm) {
6021 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6023 if (flag & child->shared_perm) {
6024 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6028 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6032 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6036 BlockDriverState *bs;
6038 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6040 GLOBAL_STATE_CODE();
6042 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6043 char *allocated_name = NULL;
6044 const char *name = blk_name(blk);
6047 name = allocated_name = blk_get_attached_dev_id(blk);
6049 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6051 g_free(allocated_name);
6052 if (blk_root(blk)) {
6053 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6057 WITH_JOB_LOCK_GUARD() {
6058 for (job = block_job_next_locked(NULL); job;
6059 job = block_job_next_locked(job)) {
6062 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6064 for (el = job->nodes; el; el = el->next) {
6065 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6070 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6071 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6073 QLIST_FOREACH(child, &bs->children, next) {
6074 xdbg_graph_add_edge(gr, bs, child);
6078 return xdbg_graph_finalize(gr);
6081 BlockDriverState *bdrv_lookup_bs(const char *device,
6082 const char *node_name,
6086 BlockDriverState *bs;
6088 GLOBAL_STATE_CODE();
6091 blk = blk_by_name(device);
6096 error_setg(errp, "Device '%s' has no medium", device);
6104 bs = bdrv_find_node(node_name);
6111 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6112 device ? device : "",
6113 node_name ? node_name : "");
6117 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6118 * return false. If either argument is NULL, return false. */
6119 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6122 GLOBAL_STATE_CODE();
6124 while (top && top != base) {
6125 top = bdrv_filter_or_cow_bs(top);
6131 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6133 GLOBAL_STATE_CODE();
6135 return QTAILQ_FIRST(&graph_bdrv_states);
6137 return QTAILQ_NEXT(bs, node_list);
6140 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6142 GLOBAL_STATE_CODE();
6144 return QTAILQ_FIRST(&all_bdrv_states);
6146 return QTAILQ_NEXT(bs, bs_list);
6149 const char *bdrv_get_node_name(const BlockDriverState *bs)
6152 return bs->node_name;
6155 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6161 /* If multiple parents have a name, just pick the first one. */
6162 QLIST_FOREACH(c, &bs->parents, next_parent) {
6163 if (c->klass->get_name) {
6164 name = c->klass->get_name(c);
6165 if (name && *name) {
6174 /* TODO check what callers really want: bs->node_name or blk_name() */
6175 const char *bdrv_get_device_name(const BlockDriverState *bs)
6178 return bdrv_get_parent_name(bs) ?: "";
6181 /* This can be used to identify nodes that might not have a device
6182 * name associated. Since node and device names live in the same
6183 * namespace, the result is unambiguous. The exception is if both are
6184 * absent, then this returns an empty (non-null) string. */
6185 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6188 return bdrv_get_parent_name(bs) ?: bs->node_name;
6191 int bdrv_get_flags(BlockDriverState *bs)
6194 return bs->open_flags;
6197 int bdrv_has_zero_init_1(BlockDriverState *bs)
6199 GLOBAL_STATE_CODE();
6203 int bdrv_has_zero_init(BlockDriverState *bs)
6205 BlockDriverState *filtered;
6206 GLOBAL_STATE_CODE();
6212 /* If BS is a copy on write image, it is initialized to
6213 the contents of the base image, which may not be zeroes. */
6214 if (bdrv_cow_child(bs)) {
6217 if (bs->drv->bdrv_has_zero_init) {
6218 return bs->drv->bdrv_has_zero_init(bs);
6221 filtered = bdrv_filter_bs(bs);
6223 return bdrv_has_zero_init(filtered);
6230 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6233 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6237 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6240 void bdrv_get_backing_filename(BlockDriverState *bs,
6241 char *filename, int filename_size)
6244 pstrcpy(filename, filename_size, bs->backing_file);
6247 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6250 BlockDriver *drv = bs->drv;
6252 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6256 if (!drv->bdrv_get_info) {
6257 BlockDriverState *filtered = bdrv_filter_bs(bs);
6259 return bdrv_get_info(filtered, bdi);
6263 memset(bdi, 0, sizeof(*bdi));
6264 ret = drv->bdrv_get_info(bs, bdi);
6269 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6276 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6279 BlockDriver *drv = bs->drv;
6281 if (drv && drv->bdrv_get_specific_info) {
6282 return drv->bdrv_get_specific_info(bs, errp);
6287 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6289 BlockDriver *drv = bs->drv;
6291 if (!drv || !drv->bdrv_get_specific_stats) {
6294 return drv->bdrv_get_specific_stats(bs);
6297 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6300 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6304 bs->drv->bdrv_debug_event(bs, event);
6307 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6309 GLOBAL_STATE_CODE();
6310 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6311 bs = bdrv_primary_bs(bs);
6314 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6315 assert(bs->drv->bdrv_debug_remove_breakpoint);
6322 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6325 GLOBAL_STATE_CODE();
6326 bs = bdrv_find_debug_node(bs);
6328 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6334 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6336 GLOBAL_STATE_CODE();
6337 bs = bdrv_find_debug_node(bs);
6339 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6345 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6347 GLOBAL_STATE_CODE();
6348 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6349 bs = bdrv_primary_bs(bs);
6352 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6353 return bs->drv->bdrv_debug_resume(bs, tag);
6359 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6361 GLOBAL_STATE_CODE();
6362 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6363 bs = bdrv_primary_bs(bs);
6366 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6367 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6373 /* backing_file can either be relative, or absolute, or a protocol. If it is
6374 * relative, it must be relative to the chain. So, passing in bs->filename
6375 * from a BDS as backing_file should not be done, as that may be relative to
6376 * the CWD rather than the chain. */
6377 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6378 const char *backing_file)
6380 char *filename_full = NULL;
6381 char *backing_file_full = NULL;
6382 char *filename_tmp = NULL;
6383 int is_protocol = 0;
6384 bool filenames_refreshed = false;
6385 BlockDriverState *curr_bs = NULL;
6386 BlockDriverState *retval = NULL;
6387 BlockDriverState *bs_below;
6389 GLOBAL_STATE_CODE();
6391 if (!bs || !bs->drv || !backing_file) {
6395 filename_full = g_malloc(PATH_MAX);
6396 backing_file_full = g_malloc(PATH_MAX);
6398 is_protocol = path_has_protocol(backing_file);
6401 * Being largely a legacy function, skip any filters here
6402 * (because filters do not have normal filenames, so they cannot
6403 * match anyway; and allowing json:{} filenames is a bit out of
6406 for (curr_bs = bdrv_skip_filters(bs);
6407 bdrv_cow_child(curr_bs) != NULL;
6410 bs_below = bdrv_backing_chain_next(curr_bs);
6412 if (bdrv_backing_overridden(curr_bs)) {
6414 * If the backing file was overridden, we can only compare
6415 * directly against the backing node's filename.
6418 if (!filenames_refreshed) {
6420 * This will automatically refresh all of the
6421 * filenames in the rest of the backing chain, so we
6422 * only need to do this once.
6424 bdrv_refresh_filename(bs_below);
6425 filenames_refreshed = true;
6428 if (strcmp(backing_file, bs_below->filename) == 0) {
6432 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6434 * If either of the filename paths is actually a protocol, then
6435 * compare unmodified paths; otherwise make paths relative.
6437 char *backing_file_full_ret;
6439 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6443 /* Also check against the full backing filename for the image */
6444 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6446 if (backing_file_full_ret) {
6447 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6448 g_free(backing_file_full_ret);
6455 /* If not an absolute filename path, make it relative to the current
6456 * image's filename path */
6457 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6459 /* We are going to compare canonicalized absolute pathnames */
6460 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6461 g_free(filename_tmp);
6464 g_free(filename_tmp);
6466 /* We need to make sure the backing filename we are comparing against
6467 * is relative to the current image filename (or absolute) */
6468 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6469 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6470 g_free(filename_tmp);
6473 g_free(filename_tmp);
6475 if (strcmp(backing_file_full, filename_full) == 0) {
6482 g_free(filename_full);
6483 g_free(backing_file_full);
6487 void bdrv_init(void)
6489 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6490 use_bdrv_whitelist = 1;
6492 module_call_init(MODULE_INIT_BLOCK);
6495 void bdrv_init_with_whitelist(void)
6497 use_bdrv_whitelist = 1;
6501 int bdrv_activate(BlockDriverState *bs, Error **errp)
6503 BdrvChild *child, *parent;
6504 Error *local_err = NULL;
6506 BdrvDirtyBitmap *bm;
6508 GLOBAL_STATE_CODE();
6514 QLIST_FOREACH(child, &bs->children, next) {
6515 bdrv_activate(child->bs, &local_err);
6517 error_propagate(errp, local_err);
6523 * Update permissions, they may differ for inactive nodes.
6525 * Note that the required permissions of inactive images are always a
6526 * subset of the permissions required after activating the image. This
6527 * allows us to just get the permissions upfront without restricting
6528 * bdrv_co_invalidate_cache().
6530 * It also means that in error cases, we don't have to try and revert to
6531 * the old permissions (which is an operation that could fail, too). We can
6532 * just keep the extended permissions for the next time that an activation
6533 * of the image is tried.
6535 if (bs->open_flags & BDRV_O_INACTIVE) {
6536 bs->open_flags &= ~BDRV_O_INACTIVE;
6537 ret = bdrv_refresh_perms(bs, errp);
6539 bs->open_flags |= BDRV_O_INACTIVE;
6543 ret = bdrv_invalidate_cache(bs, errp);
6545 bs->open_flags |= BDRV_O_INACTIVE;
6549 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6550 bdrv_dirty_bitmap_skip_store(bm, false);
6553 ret = refresh_total_sectors(bs, bs->total_sectors);
6555 bs->open_flags |= BDRV_O_INACTIVE;
6556 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6561 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6562 if (parent->klass->activate) {
6563 parent->klass->activate(parent, &local_err);
6565 bs->open_flags |= BDRV_O_INACTIVE;
6566 error_propagate(errp, local_err);
6575 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6577 Error *local_err = NULL;
6580 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6582 if (bs->drv->bdrv_co_invalidate_cache) {
6583 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6585 error_propagate(errp, local_err);
6593 void bdrv_activate_all(Error **errp)
6595 BlockDriverState *bs;
6596 BdrvNextIterator it;
6598 GLOBAL_STATE_CODE();
6600 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6601 AioContext *aio_context = bdrv_get_aio_context(bs);
6604 aio_context_acquire(aio_context);
6605 ret = bdrv_activate(bs, errp);
6606 aio_context_release(aio_context);
6608 bdrv_next_cleanup(&it);
6614 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6617 GLOBAL_STATE_CODE();
6619 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6620 if (parent->klass->parent_is_bds) {
6621 BlockDriverState *parent_bs = parent->opaque;
6622 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6631 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6633 BdrvChild *child, *parent;
6635 uint64_t cumulative_perms, cumulative_shared_perms;
6637 GLOBAL_STATE_CODE();
6643 /* Make sure that we don't inactivate a child before its parent.
6644 * It will be covered by recursion from the yet active parent. */
6645 if (bdrv_has_bds_parent(bs, true)) {
6649 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6651 /* Inactivate this node */
6652 if (bs->drv->bdrv_inactivate) {
6653 ret = bs->drv->bdrv_inactivate(bs);
6659 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6660 if (parent->klass->inactivate) {
6661 ret = parent->klass->inactivate(parent);
6668 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6669 &cumulative_shared_perms);
6670 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6671 /* Our inactive parents still need write access. Inactivation failed. */
6675 bs->open_flags |= BDRV_O_INACTIVE;
6678 * Update permissions, they may differ for inactive nodes.
6679 * We only tried to loosen restrictions, so errors are not fatal, ignore
6682 bdrv_refresh_perms(bs, NULL);
6684 /* Recursively inactivate children */
6685 QLIST_FOREACH(child, &bs->children, next) {
6686 ret = bdrv_inactivate_recurse(child->bs);
6695 int bdrv_inactivate_all(void)
6697 BlockDriverState *bs = NULL;
6698 BdrvNextIterator it;
6700 GSList *aio_ctxs = NULL, *ctx;
6702 GLOBAL_STATE_CODE();
6704 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6705 AioContext *aio_context = bdrv_get_aio_context(bs);
6707 if (!g_slist_find(aio_ctxs, aio_context)) {
6708 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6709 aio_context_acquire(aio_context);
6713 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6714 /* Nodes with BDS parents are covered by recursion from the last
6715 * parent that gets inactivated. Don't inactivate them a second
6716 * time if that has already happened. */
6717 if (bdrv_has_bds_parent(bs, false)) {
6720 ret = bdrv_inactivate_recurse(bs);
6722 bdrv_next_cleanup(&it);
6728 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6729 AioContext *aio_context = ctx->data;
6730 aio_context_release(aio_context);
6732 g_slist_free(aio_ctxs);
6737 /**************************************************************/
6738 /* removable device support */
6741 * Return TRUE if the media is present
6743 bool bdrv_is_inserted(BlockDriverState *bs)
6745 BlockDriver *drv = bs->drv;
6752 if (drv->bdrv_is_inserted) {
6753 return drv->bdrv_is_inserted(bs);
6755 QLIST_FOREACH(child, &bs->children, next) {
6756 if (!bdrv_is_inserted(child->bs)) {
6764 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6766 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6768 BlockDriver *drv = bs->drv;
6771 if (drv && drv->bdrv_eject) {
6772 drv->bdrv_eject(bs, eject_flag);
6777 * Lock or unlock the media (if it is locked, the user won't be able
6778 * to eject it manually).
6780 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6782 BlockDriver *drv = bs->drv;
6784 trace_bdrv_lock_medium(bs, locked);
6786 if (drv && drv->bdrv_lock_medium) {
6787 drv->bdrv_lock_medium(bs, locked);
6791 /* Get a reference to bs */
6792 void bdrv_ref(BlockDriverState *bs)
6794 GLOBAL_STATE_CODE();
6798 /* Release a previously grabbed reference to bs.
6799 * If after releasing, reference count is zero, the BlockDriverState is
6801 void bdrv_unref(BlockDriverState *bs)
6803 GLOBAL_STATE_CODE();
6807 assert(bs->refcnt > 0);
6808 if (--bs->refcnt == 0) {
6813 struct BdrvOpBlocker {
6815 QLIST_ENTRY(BdrvOpBlocker) list;
6818 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6820 BdrvOpBlocker *blocker;
6821 GLOBAL_STATE_CODE();
6822 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6823 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6824 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6825 error_propagate_prepend(errp, error_copy(blocker->reason),
6826 "Node '%s' is busy: ",
6827 bdrv_get_device_or_node_name(bs));
6833 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6835 BdrvOpBlocker *blocker;
6836 GLOBAL_STATE_CODE();
6837 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6839 blocker = g_new0(BdrvOpBlocker, 1);
6840 blocker->reason = reason;
6841 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6844 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6846 BdrvOpBlocker *blocker, *next;
6847 GLOBAL_STATE_CODE();
6848 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6849 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6850 if (blocker->reason == reason) {
6851 QLIST_REMOVE(blocker, list);
6857 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6860 GLOBAL_STATE_CODE();
6861 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6862 bdrv_op_block(bs, i, reason);
6866 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6869 GLOBAL_STATE_CODE();
6870 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6871 bdrv_op_unblock(bs, i, reason);
6875 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6878 GLOBAL_STATE_CODE();
6879 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6880 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6887 void bdrv_img_create(const char *filename, const char *fmt,
6888 const char *base_filename, const char *base_fmt,
6889 char *options, uint64_t img_size, int flags, bool quiet,
6892 QemuOptsList *create_opts = NULL;
6893 QemuOpts *opts = NULL;
6894 const char *backing_fmt, *backing_file;
6896 BlockDriver *drv, *proto_drv;
6897 Error *local_err = NULL;
6900 GLOBAL_STATE_CODE();
6902 /* Find driver and parse its options */
6903 drv = bdrv_find_format(fmt);
6905 error_setg(errp, "Unknown file format '%s'", fmt);
6909 proto_drv = bdrv_find_protocol(filename, true, errp);
6914 if (!drv->create_opts) {
6915 error_setg(errp, "Format driver '%s' does not support image creation",
6920 if (!proto_drv->create_opts) {
6921 error_setg(errp, "Protocol driver '%s' does not support image creation",
6922 proto_drv->format_name);
6926 /* Create parameter list */
6927 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6928 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6930 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6932 /* Parse -o options */
6934 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6939 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6940 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6941 } else if (img_size != UINT64_C(-1)) {
6942 error_setg(errp, "The image size must be specified only once");
6946 if (base_filename) {
6947 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6949 error_setg(errp, "Backing file not supported for file format '%s'",
6956 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
6957 error_setg(errp, "Backing file format not supported for file "
6958 "format '%s'", fmt);
6963 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6965 if (!strcmp(filename, backing_file)) {
6966 error_setg(errp, "Error: Trying to create an image with the "
6967 "same filename as the backing file");
6970 if (backing_file[0] == '\0') {
6971 error_setg(errp, "Expected backing file name, got empty string");
6976 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
6978 /* The size for the image must always be specified, unless we have a backing
6979 * file and we have not been forbidden from opening it. */
6980 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6981 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6982 BlockDriverState *bs;
6985 QDict *backing_options = NULL;
6988 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6993 assert(full_backing);
6996 * No need to do I/O here, which allows us to open encrypted
6997 * backing images without needing the secret
7000 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7001 back_flags |= BDRV_O_NO_IO;
7003 backing_options = qdict_new();
7005 qdict_put_str(backing_options, "driver", backing_fmt);
7007 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7009 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7011 g_free(full_backing);
7013 error_append_hint(&local_err, "Could not open backing image.\n");
7017 error_setg(&local_err,
7018 "Backing file specified without backing format");
7019 error_append_hint(&local_err, "Detected format of %s.",
7020 bs->drv->format_name);
7024 /* Opened BS, have no size */
7025 size = bdrv_getlength(bs);
7027 error_setg_errno(errp, -size, "Could not get size of '%s'",
7032 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7036 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7037 } else if (backing_file && !backing_fmt) {
7038 error_setg(&local_err,
7039 "Backing file specified without backing format");
7044 error_setg(errp, "Image creation needs a size parameter");
7049 printf("Formatting '%s', fmt=%s ", filename, fmt);
7050 qemu_opts_print(opts, " ");
7055 ret = bdrv_create(drv, filename, opts, &local_err);
7057 if (ret == -EFBIG) {
7058 /* This is generally a better message than whatever the driver would
7059 * deliver (especially because of the cluster_size_hint), since that
7060 * is most probably not much different from "image too large". */
7061 const char *cluster_size_hint = "";
7062 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7063 cluster_size_hint = " (try using a larger cluster size)";
7065 error_setg(errp, "The image size is too large for file format '%s'"
7066 "%s", fmt, cluster_size_hint);
7067 error_free(local_err);
7072 qemu_opts_del(opts);
7073 qemu_opts_free(create_opts);
7074 error_propagate(errp, local_err);
7077 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7080 return bs ? bs->aio_context : qemu_get_aio_context();
7083 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7085 Coroutine *self = qemu_coroutine_self();
7086 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7087 AioContext *new_ctx;
7091 * Increase bs->in_flight to ensure that this operation is completed before
7092 * moving the node to a different AioContext. Read new_ctx only afterwards.
7094 bdrv_inc_in_flight(bs);
7096 new_ctx = bdrv_get_aio_context(bs);
7097 aio_co_reschedule_self(new_ctx);
7101 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7104 aio_co_reschedule_self(old_ctx);
7105 bdrv_dec_in_flight(bs);
7108 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7110 AioContext *ctx = bdrv_get_aio_context(bs);
7112 /* In the main thread, bs->aio_context won't change concurrently */
7113 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7116 * We're in coroutine context, so we already hold the lock of the main
7117 * loop AioContext. Don't lock it twice to avoid deadlocks.
7119 assert(qemu_in_coroutine());
7120 if (ctx != qemu_get_aio_context()) {
7121 aio_context_acquire(ctx);
7125 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7127 AioContext *ctx = bdrv_get_aio_context(bs);
7129 assert(qemu_in_coroutine());
7130 if (ctx != qemu_get_aio_context()) {
7131 aio_context_release(ctx);
7135 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
7138 aio_co_enter(bdrv_get_aio_context(bs), co);
7141 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7143 GLOBAL_STATE_CODE();
7144 QLIST_REMOVE(ban, list);
7148 static void bdrv_detach_aio_context(BlockDriverState *bs)
7150 BdrvAioNotifier *baf, *baf_tmp;
7152 assert(!bs->walking_aio_notifiers);
7153 GLOBAL_STATE_CODE();
7154 bs->walking_aio_notifiers = true;
7155 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7157 bdrv_do_remove_aio_context_notifier(baf);
7159 baf->detach_aio_context(baf->opaque);
7162 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7163 * remove remaining aio notifiers if we aren't called again.
7165 bs->walking_aio_notifiers = false;
7167 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7168 bs->drv->bdrv_detach_aio_context(bs);
7171 if (bs->quiesce_counter) {
7172 aio_enable_external(bs->aio_context);
7174 assert_bdrv_graph_writable(bs);
7175 bs->aio_context = NULL;
7178 static void bdrv_attach_aio_context(BlockDriverState *bs,
7179 AioContext *new_context)
7181 BdrvAioNotifier *ban, *ban_tmp;
7182 GLOBAL_STATE_CODE();
7184 if (bs->quiesce_counter) {
7185 aio_disable_external(new_context);
7188 assert_bdrv_graph_writable(bs);
7189 bs->aio_context = new_context;
7191 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7192 bs->drv->bdrv_attach_aio_context(bs, new_context);
7195 assert(!bs->walking_aio_notifiers);
7196 bs->walking_aio_notifiers = true;
7197 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7199 bdrv_do_remove_aio_context_notifier(ban);
7201 ban->attached_aio_context(new_context, ban->opaque);
7204 bs->walking_aio_notifiers = false;
7207 typedef struct BdrvStateSetAioContext {
7208 AioContext *new_ctx;
7209 BlockDriverState *bs;
7210 } BdrvStateSetAioContext;
7212 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7213 GHashTable *visited,
7217 GLOBAL_STATE_CODE();
7218 if (g_hash_table_contains(visited, c)) {
7221 g_hash_table_add(visited, c);
7224 * A BdrvChildClass that doesn't handle AioContext changes cannot
7225 * tolerate any AioContext changes
7227 if (!c->klass->change_aio_ctx) {
7228 char *user = bdrv_child_user_desc(c);
7229 error_setg(errp, "Changing iothreads is not supported by %s", user);
7233 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7234 assert(!errp || *errp);
7240 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7241 GHashTable *visited, Transaction *tran,
7244 GLOBAL_STATE_CODE();
7245 if (g_hash_table_contains(visited, c)) {
7248 g_hash_table_add(visited, c);
7249 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7252 static void bdrv_set_aio_context_clean(void *opaque)
7254 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7255 BlockDriverState *bs = (BlockDriverState *) state->bs;
7257 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7258 bdrv_drained_end(bs);
7263 static void bdrv_set_aio_context_commit(void *opaque)
7265 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7266 BlockDriverState *bs = (BlockDriverState *) state->bs;
7267 AioContext *new_context = state->new_ctx;
7268 AioContext *old_context = bdrv_get_aio_context(bs);
7269 assert_bdrv_graph_writable(bs);
7272 * Take the old AioContex when detaching it from bs.
7273 * At this point, new_context lock is already acquired, and we are now
7274 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7275 * does not call AIO_POLL_WHILE().
7277 if (old_context != qemu_get_aio_context()) {
7278 aio_context_acquire(old_context);
7280 bdrv_detach_aio_context(bs);
7281 if (old_context != qemu_get_aio_context()) {
7282 aio_context_release(old_context);
7284 bdrv_attach_aio_context(bs, new_context);
7287 static TransactionActionDrv set_aio_context = {
7288 .commit = bdrv_set_aio_context_commit,
7289 .clean = bdrv_set_aio_context_clean,
7293 * Changes the AioContext used for fd handlers, timers, and BHs by this
7294 * BlockDriverState and all its children and parents.
7296 * Must be called from the main AioContext.
7298 * The caller must own the AioContext lock for the old AioContext of bs, but it
7299 * must not own the AioContext lock for new_context (unless new_context is the
7300 * same as the current context of bs).
7302 * @visited will accumulate all visited BdrvChild objects. The caller is
7303 * responsible for freeing the list afterwards.
7305 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7306 GHashTable *visited, Transaction *tran,
7310 BdrvStateSetAioContext *state;
7312 GLOBAL_STATE_CODE();
7314 if (bdrv_get_aio_context(bs) == ctx) {
7318 QLIST_FOREACH(c, &bs->parents, next_parent) {
7319 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7324 QLIST_FOREACH(c, &bs->children, next) {
7325 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7330 state = g_new(BdrvStateSetAioContext, 1);
7331 *state = (BdrvStateSetAioContext) {
7336 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7337 bdrv_drained_begin(bs);
7339 tran_add(tran, &set_aio_context, state);
7345 * Change bs's and recursively all of its parents' and children's AioContext
7346 * to the given new context, returning an error if that isn't possible.
7348 * If ignore_child is not NULL, that child (and its subgraph) will not
7351 * This function still requires the caller to take the bs current
7352 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7353 * assumes the lock is always held if bs is in another AioContext.
7354 * For the same reason, it temporarily also holds the new AioContext, since
7355 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7356 * Therefore the new AioContext lock must not be taken by the caller.
7358 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7359 BdrvChild *ignore_child, Error **errp)
7362 GHashTable *visited;
7364 AioContext *old_context = bdrv_get_aio_context(bs);
7365 GLOBAL_STATE_CODE();
7368 * Recursion phase: go through all nodes of the graph.
7369 * Take care of checking that all nodes support changing AioContext
7370 * and drain them, builing a linear list of callbacks to run if everything
7371 * is successful (the transaction itself).
7374 visited = g_hash_table_new(NULL, NULL);
7376 g_hash_table_add(visited, ignore_child);
7378 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7379 g_hash_table_destroy(visited);
7382 * Linear phase: go through all callbacks collected in the transaction.
7383 * Run all callbacks collected in the recursion to switch all nodes
7384 * AioContext lock (transaction commit), or undo all changes done in the
7385 * recursion (transaction abort).
7389 /* Just run clean() callbacks. No AioContext changed. */
7395 * Release old AioContext, it won't be needed anymore, as all
7396 * bdrv_drained_begin() have been called already.
7398 if (qemu_get_aio_context() != old_context) {
7399 aio_context_release(old_context);
7403 * Acquire new AioContext since bdrv_drained_end() is going to be called
7404 * after we switched all nodes in the new AioContext, and the function
7405 * assumes that the lock of the bs is always taken.
7407 if (qemu_get_aio_context() != ctx) {
7408 aio_context_acquire(ctx);
7413 if (qemu_get_aio_context() != ctx) {
7414 aio_context_release(ctx);
7417 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7418 if (qemu_get_aio_context() != old_context) {
7419 aio_context_acquire(old_context);
7425 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7426 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7427 void (*detach_aio_context)(void *opaque), void *opaque)
7429 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7430 *ban = (BdrvAioNotifier){
7431 .attached_aio_context = attached_aio_context,
7432 .detach_aio_context = detach_aio_context,
7435 GLOBAL_STATE_CODE();
7437 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7440 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7441 void (*attached_aio_context)(AioContext *,
7443 void (*detach_aio_context)(void *),
7446 BdrvAioNotifier *ban, *ban_next;
7447 GLOBAL_STATE_CODE();
7449 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7450 if (ban->attached_aio_context == attached_aio_context &&
7451 ban->detach_aio_context == detach_aio_context &&
7452 ban->opaque == opaque &&
7453 ban->deleted == false)
7455 if (bs->walking_aio_notifiers) {
7456 ban->deleted = true;
7458 bdrv_do_remove_aio_context_notifier(ban);
7467 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7468 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7472 GLOBAL_STATE_CODE();
7474 error_setg(errp, "Node is ejected");
7477 if (!bs->drv->bdrv_amend_options) {
7478 error_setg(errp, "Block driver '%s' does not support option amendment",
7479 bs->drv->format_name);
7482 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7483 cb_opaque, force, errp);
7487 * This function checks whether the given @to_replace is allowed to be
7488 * replaced by a node that always shows the same data as @bs. This is
7489 * used for example to verify whether the mirror job can replace
7490 * @to_replace by the target mirrored from @bs.
7491 * To be replaceable, @bs and @to_replace may either be guaranteed to
7492 * always show the same data (because they are only connected through
7493 * filters), or some driver may allow replacing one of its children
7494 * because it can guarantee that this child's data is not visible at
7495 * all (for example, for dissenting quorum children that have no other
7498 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7499 BlockDriverState *to_replace)
7501 BlockDriverState *filtered;
7503 GLOBAL_STATE_CODE();
7505 if (!bs || !bs->drv) {
7509 if (bs == to_replace) {
7513 /* See what the driver can do */
7514 if (bs->drv->bdrv_recurse_can_replace) {
7515 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7518 /* For filters without an own implementation, we can recurse on our own */
7519 filtered = bdrv_filter_bs(bs);
7521 return bdrv_recurse_can_replace(filtered, to_replace);
7529 * Check whether the given @node_name can be replaced by a node that
7530 * has the same data as @parent_bs. If so, return @node_name's BDS;
7533 * @node_name must be a (recursive) *child of @parent_bs (or this
7534 * function will return NULL).
7536 * The result (whether the node can be replaced or not) is only valid
7537 * for as long as no graph or permission changes occur.
7539 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7540 const char *node_name, Error **errp)
7542 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7543 AioContext *aio_context;
7545 GLOBAL_STATE_CODE();
7547 if (!to_replace_bs) {
7548 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7552 aio_context = bdrv_get_aio_context(to_replace_bs);
7553 aio_context_acquire(aio_context);
7555 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7556 to_replace_bs = NULL;
7560 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7561 * most non filter in order to prevent data corruption.
7562 * Another benefit is that this tests exclude backing files which are
7563 * blocked by the backing blockers.
7565 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7566 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7567 "because it cannot be guaranteed that doing so would not "
7568 "lead to an abrupt change of visible data",
7569 node_name, parent_bs->node_name);
7570 to_replace_bs = NULL;
7575 aio_context_release(aio_context);
7576 return to_replace_bs;
7580 * Iterates through the list of runtime option keys that are said to
7581 * be "strong" for a BDS. An option is called "strong" if it changes
7582 * a BDS's data. For example, the null block driver's "size" and
7583 * "read-zeroes" options are strong, but its "latency-ns" option is
7586 * If a key returned by this function ends with a dot, all options
7587 * starting with that prefix are strong.
7589 static const char *const *strong_options(BlockDriverState *bs,
7590 const char *const *curopt)
7592 static const char *const global_options[] = {
7593 "driver", "filename", NULL
7597 return &global_options[0];
7601 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7602 curopt = bs->drv->strong_runtime_opts;
7605 return (curopt && *curopt) ? curopt : NULL;
7609 * Copies all strong runtime options from bs->options to the given
7610 * QDict. The set of strong option keys is determined by invoking
7613 * Returns true iff any strong option was present in bs->options (and
7614 * thus copied to the target QDict) with the exception of "filename"
7615 * and "driver". The caller is expected to use this value to decide
7616 * whether the existence of strong options prevents the generation of
7619 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7621 bool found_any = false;
7622 const char *const *option_name = NULL;
7628 while ((option_name = strong_options(bs, option_name))) {
7629 bool option_given = false;
7631 assert(strlen(*option_name) > 0);
7632 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7633 QObject *entry = qdict_get(bs->options, *option_name);
7638 qdict_put_obj(d, *option_name, qobject_ref(entry));
7639 option_given = true;
7641 const QDictEntry *entry;
7642 for (entry = qdict_first(bs->options); entry;
7643 entry = qdict_next(bs->options, entry))
7645 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7646 qdict_put_obj(d, qdict_entry_key(entry),
7647 qobject_ref(qdict_entry_value(entry)));
7648 option_given = true;
7653 /* While "driver" and "filename" need to be included in a JSON filename,
7654 * their existence does not prohibit generation of a plain filename. */
7655 if (!found_any && option_given &&
7656 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7662 if (!qdict_haskey(d, "driver")) {
7663 /* Drivers created with bdrv_new_open_driver() may not have a
7664 * @driver option. Add it here. */
7665 qdict_put_str(d, "driver", bs->drv->format_name);
7671 /* Note: This function may return false positives; it may return true
7672 * even if opening the backing file specified by bs's image header
7673 * would result in exactly bs->backing. */
7674 static bool bdrv_backing_overridden(BlockDriverState *bs)
7676 GLOBAL_STATE_CODE();
7678 return strcmp(bs->auto_backing_file,
7679 bs->backing->bs->filename);
7681 /* No backing BDS, so if the image header reports any backing
7682 * file, it must have been suppressed */
7683 return bs->auto_backing_file[0] != '\0';
7687 /* Updates the following BDS fields:
7688 * - exact_filename: A filename which may be used for opening a block device
7689 * which (mostly) equals the given BDS (even without any
7690 * other options; so reading and writing must return the same
7691 * results, but caching etc. may be different)
7692 * - full_open_options: Options which, when given when opening a block device
7693 * (without a filename), result in a BDS (mostly)
7694 * equalling the given one
7695 * - filename: If exact_filename is set, it is copied here. Otherwise,
7696 * full_open_options is converted to a JSON object, prefixed with
7697 * "json:" (for use through the JSON pseudo protocol) and put here.
7699 void bdrv_refresh_filename(BlockDriverState *bs)
7701 BlockDriver *drv = bs->drv;
7703 BlockDriverState *primary_child_bs;
7705 bool backing_overridden;
7706 bool generate_json_filename; /* Whether our default implementation should
7707 fill exact_filename (false) or not (true) */
7709 GLOBAL_STATE_CODE();
7715 /* This BDS's file name may depend on any of its children's file names, so
7716 * refresh those first */
7717 QLIST_FOREACH(child, &bs->children, next) {
7718 bdrv_refresh_filename(child->bs);
7722 /* For implicit nodes, just copy everything from the single child */
7723 child = QLIST_FIRST(&bs->children);
7724 assert(QLIST_NEXT(child, next) == NULL);
7726 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7727 child->bs->exact_filename);
7728 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7730 qobject_unref(bs->full_open_options);
7731 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7736 backing_overridden = bdrv_backing_overridden(bs);
7738 if (bs->open_flags & BDRV_O_NO_IO) {
7739 /* Without I/O, the backing file does not change anything.
7740 * Therefore, in such a case (primarily qemu-img), we can
7741 * pretend the backing file has not been overridden even if
7742 * it technically has been. */
7743 backing_overridden = false;
7746 /* Gather the options QDict */
7748 generate_json_filename = append_strong_runtime_options(opts, bs);
7749 generate_json_filename |= backing_overridden;
7751 if (drv->bdrv_gather_child_options) {
7752 /* Some block drivers may not want to present all of their children's
7753 * options, or name them differently from BdrvChild.name */
7754 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7756 QLIST_FOREACH(child, &bs->children, next) {
7757 if (child == bs->backing && !backing_overridden) {
7758 /* We can skip the backing BDS if it has not been overridden */
7762 qdict_put(opts, child->name,
7763 qobject_ref(child->bs->full_open_options));
7766 if (backing_overridden && !bs->backing) {
7767 /* Force no backing file */
7768 qdict_put_null(opts, "backing");
7772 qobject_unref(bs->full_open_options);
7773 bs->full_open_options = opts;
7775 primary_child_bs = bdrv_primary_bs(bs);
7777 if (drv->bdrv_refresh_filename) {
7778 /* Obsolete information is of no use here, so drop the old file name
7779 * information before refreshing it */
7780 bs->exact_filename[0] = '\0';
7782 drv->bdrv_refresh_filename(bs);
7783 } else if (primary_child_bs) {
7785 * Try to reconstruct valid information from the underlying
7786 * file -- this only works for format nodes (filter nodes
7787 * cannot be probed and as such must be selected by the user
7788 * either through an options dict, or through a special
7789 * filename which the filter driver must construct in its
7790 * .bdrv_refresh_filename() implementation).
7793 bs->exact_filename[0] = '\0';
7796 * We can use the underlying file's filename if:
7797 * - it has a filename,
7798 * - the current BDS is not a filter,
7799 * - the file is a protocol BDS, and
7800 * - opening that file (as this BDS's format) will automatically create
7801 * the BDS tree we have right now, that is:
7802 * - the user did not significantly change this BDS's behavior with
7803 * some explicit (strong) options
7804 * - no non-file child of this BDS has been overridden by the user
7805 * Both of these conditions are represented by generate_json_filename.
7807 if (primary_child_bs->exact_filename[0] &&
7808 primary_child_bs->drv->bdrv_file_open &&
7809 !drv->is_filter && !generate_json_filename)
7811 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7815 if (bs->exact_filename[0]) {
7816 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7818 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7819 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7820 json->str) >= sizeof(bs->filename)) {
7821 /* Give user a hint if we truncated things. */
7822 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7824 g_string_free(json, true);
7828 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7830 BlockDriver *drv = bs->drv;
7831 BlockDriverState *child_bs;
7833 GLOBAL_STATE_CODE();
7836 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7840 if (drv->bdrv_dirname) {
7841 return drv->bdrv_dirname(bs, errp);
7844 child_bs = bdrv_primary_bs(bs);
7846 return bdrv_dirname(child_bs, errp);
7849 bdrv_refresh_filename(bs);
7850 if (bs->exact_filename[0] != '\0') {
7851 return path_combine(bs->exact_filename, "");
7854 error_setg(errp, "Cannot generate a base directory for %s nodes",
7860 * Hot add/remove a BDS's child. So the user can take a child offline when
7861 * it is broken and take a new child online
7863 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7866 GLOBAL_STATE_CODE();
7867 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7868 error_setg(errp, "The node %s does not support adding a child",
7869 bdrv_get_device_or_node_name(parent_bs));
7873 if (!QLIST_EMPTY(&child_bs->parents)) {
7874 error_setg(errp, "The node %s already has a parent",
7875 child_bs->node_name);
7879 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7882 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7886 GLOBAL_STATE_CODE();
7887 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7888 error_setg(errp, "The node %s does not support removing a child",
7889 bdrv_get_device_or_node_name(parent_bs));
7893 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7900 error_setg(errp, "The node %s does not have a child named %s",
7901 bdrv_get_device_or_node_name(parent_bs),
7902 bdrv_get_device_or_node_name(child->bs));
7906 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7909 int bdrv_make_empty(BdrvChild *c, Error **errp)
7911 BlockDriver *drv = c->bs->drv;
7914 GLOBAL_STATE_CODE();
7915 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7917 if (!drv->bdrv_make_empty) {
7918 error_setg(errp, "%s does not support emptying nodes",
7923 ret = drv->bdrv_make_empty(c->bs);
7925 error_setg_errno(errp, -ret, "Failed to empty %s",
7934 * Return the child that @bs acts as an overlay for, and from which data may be
7935 * copied in COW or COR operations. Usually this is the backing file.
7937 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7941 if (!bs || !bs->drv) {
7945 if (bs->drv->is_filter) {
7953 assert(bs->backing->role & BDRV_CHILD_COW);
7958 * If @bs acts as a filter for exactly one of its children, return
7961 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7966 if (!bs || !bs->drv) {
7970 if (!bs->drv->is_filter) {
7974 /* Only one of @backing or @file may be used */
7975 assert(!(bs->backing && bs->file));
7977 c = bs->backing ?: bs->file;
7982 assert(c->role & BDRV_CHILD_FILTERED);
7987 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7988 * whichever is non-NULL.
7990 * Return NULL if both are NULL.
7992 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7994 BdrvChild *cow_child = bdrv_cow_child(bs);
7995 BdrvChild *filter_child = bdrv_filter_child(bs);
7998 /* Filter nodes cannot have COW backing files */
7999 assert(!(cow_child && filter_child));
8001 return cow_child ?: filter_child;
8005 * Return the primary child of this node: For filters, that is the
8006 * filtered child. For other nodes, that is usually the child storing
8008 * (A generally more helpful description is that this is (usually) the
8009 * child that has the same filename as @bs.)
8011 * Drivers do not necessarily have a primary child; for example quorum
8014 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8016 BdrvChild *c, *found = NULL;
8019 QLIST_FOREACH(c, &bs->children, next) {
8020 if (c->role & BDRV_CHILD_PRIMARY) {
8029 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8030 bool stop_on_explicit_filter)
8038 while (!(stop_on_explicit_filter && !bs->implicit)) {
8039 c = bdrv_filter_child(bs);
8042 * A filter that is embedded in a working block graph must
8043 * have a child. Assert this here so this function does
8044 * not return a filter node that is not expected by the
8047 assert(!bs->drv || !bs->drv->is_filter);
8053 * Note that this treats nodes with bs->drv == NULL as not being
8054 * filters (bs->drv == NULL should be replaced by something else
8056 * The advantage of this behavior is that this function will thus
8057 * always return a non-NULL value (given a non-NULL @bs).
8064 * Return the first BDS that has not been added implicitly or that
8065 * does not have a filtered child down the chain starting from @bs
8066 * (including @bs itself).
8068 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8070 GLOBAL_STATE_CODE();
8071 return bdrv_do_skip_filters(bs, true);
8075 * Return the first BDS that does not have a filtered child down the
8076 * chain starting from @bs (including @bs itself).
8078 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8081 return bdrv_do_skip_filters(bs, false);
8085 * For a backing chain, return the first non-filter backing image of
8086 * the first non-filter image.
8088 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8091 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8095 * Check whether [offset, offset + bytes) overlaps with the cached
8096 * block-status data region.
8098 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8099 * which is what bdrv_bsc_is_data()'s interface needs.
8100 * Otherwise, *pnum is not touched.
8102 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8103 int64_t offset, int64_t bytes,
8106 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8110 qatomic_read(&bsc->valid) &&
8111 ranges_overlap(offset, bytes, bsc->data_start,
8112 bsc->data_end - bsc->data_start);
8114 if (overlaps && pnum) {
8115 *pnum = bsc->data_end - offset;
8122 * See block_int.h for this function's documentation.
8124 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8127 RCU_READ_LOCK_GUARD();
8128 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8132 * See block_int.h for this function's documentation.
8134 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8135 int64_t offset, int64_t bytes)
8138 RCU_READ_LOCK_GUARD();
8140 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8141 qatomic_set(&bs->block_status_cache->valid, false);
8146 * See block_int.h for this function's documentation.
8148 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8150 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8151 BdrvBlockStatusCache *old_bsc;
8154 *new_bsc = (BdrvBlockStatusCache) {
8156 .data_start = offset,
8157 .data_end = offset + bytes,
8160 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8162 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8163 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8165 g_free_rcu(old_bsc, rcu);