Merge remote-tracking branch 'remotes/vivier2/tags/linux-user-for-6.2-pull-request...
[qemu.git] / block.c
blob5ce08a79fda7184d64e2edc05656072f7692a269
1 /*
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
23 * THE SOFTWARE.
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"
51 #include "qemu/id.h"
52 #include "qemu/range.h"
53 #include "qemu/rcu.h"
54 #include "block/coroutines.h"
56 #ifdef CONFIG_BSD
57 #include <sys/ioctl.h>
58 #include <sys/queue.h>
59 #if defined(HAVE_SYS_DISK_H)
60 #include <sys/disk.h>
61 #endif
62 #endif
64 #ifdef _WIN32
65 #include <windows.h>
66 #endif
68 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
71 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
73 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
74 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
76 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
77 QLIST_HEAD_INITIALIZER(bdrv_drivers);
79 static BlockDriverState *bdrv_open_inherit(const char *filename,
80 const char *reference,
81 QDict *options, int flags,
82 BlockDriverState *parent,
83 const BdrvChildClass *child_class,
84 BdrvChildRole child_role,
85 Error **errp);
87 static void bdrv_replace_child_noperm(BdrvChild *child,
88 BlockDriverState *new_bs);
89 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
90 BdrvChild *child,
91 Transaction *tran);
92 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
93 Transaction *tran);
95 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
96 BlockReopenQueue *queue,
97 Transaction *change_child_tran, Error **errp);
98 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
99 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
101 /* If non-zero, use only whitelisted block drivers */
102 static int use_bdrv_whitelist;
104 #ifdef _WIN32
105 static int is_windows_drive_prefix(const char *filename)
107 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
108 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
109 filename[1] == ':');
112 int is_windows_drive(const char *filename)
114 if (is_windows_drive_prefix(filename) &&
115 filename[2] == '\0')
116 return 1;
117 if (strstart(filename, "\\\\.\\", NULL) ||
118 strstart(filename, "//./", NULL))
119 return 1;
120 return 0;
122 #endif
124 size_t bdrv_opt_mem_align(BlockDriverState *bs)
126 if (!bs || !bs->drv) {
127 /* page size or 4k (hdd sector size) should be on the safe side */
128 return MAX(4096, qemu_real_host_page_size);
131 return bs->bl.opt_mem_alignment;
134 size_t bdrv_min_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);
141 return bs->bl.min_mem_alignment;
144 /* check if the path starts with "<protocol>:" */
145 int path_has_protocol(const char *path)
147 const char *p;
149 #ifdef _WIN32
150 if (is_windows_drive(path) ||
151 is_windows_drive_prefix(path)) {
152 return 0;
154 p = path + strcspn(path, ":/\\");
155 #else
156 p = path + strcspn(path, ":/");
157 #endif
159 return *p == ':';
162 int path_is_absolute(const char *path)
164 #ifdef _WIN32
165 /* specific case for names like: "\\.\d:" */
166 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
167 return 1;
169 return (*path == '/' || *path == '\\');
170 #else
171 return (*path == '/');
172 #endif
175 /* if filename is absolute, just return its duplicate. Otherwise, build a
176 path to it by considering it is relative to base_path. URL are
177 supported. */
178 char *path_combine(const char *base_path, const char *filename)
180 const char *protocol_stripped = NULL;
181 const char *p, *p1;
182 char *result;
183 int len;
185 if (path_is_absolute(filename)) {
186 return g_strdup(filename);
189 if (path_has_protocol(base_path)) {
190 protocol_stripped = strchr(base_path, ':');
191 if (protocol_stripped) {
192 protocol_stripped++;
195 p = protocol_stripped ?: base_path;
197 p1 = strrchr(base_path, '/');
198 #ifdef _WIN32
200 const char *p2;
201 p2 = strrchr(base_path, '\\');
202 if (!p1 || p2 > p1) {
203 p1 = p2;
206 #endif
207 if (p1) {
208 p1++;
209 } else {
210 p1 = base_path;
212 if (p1 > p) {
213 p = p1;
215 len = p - base_path;
217 result = g_malloc(len + strlen(filename) + 1);
218 memcpy(result, base_path, len);
219 strcpy(result + len, filename);
221 return result;
225 * Helper function for bdrv_parse_filename() implementations to remove optional
226 * protocol prefixes (especially "file:") from a filename and for putting the
227 * stripped filename into the options QDict if there is such a prefix.
229 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
230 QDict *options)
232 if (strstart(filename, prefix, &filename)) {
233 /* Stripping the explicit protocol prefix may result in a protocol
234 * prefix being (wrongly) detected (if the filename contains a colon) */
235 if (path_has_protocol(filename)) {
236 GString *fat_filename;
238 /* This means there is some colon before the first slash; therefore,
239 * this cannot be an absolute path */
240 assert(!path_is_absolute(filename));
242 /* And we can thus fix the protocol detection issue by prefixing it
243 * by "./" */
244 fat_filename = g_string_new("./");
245 g_string_append(fat_filename, filename);
247 assert(!path_has_protocol(fat_filename->str));
249 qdict_put(options, "filename",
250 qstring_from_gstring(fat_filename));
251 } else {
252 /* If no protocol prefix was detected, we can use the shortened
253 * filename as-is */
254 qdict_put_str(options, "filename", filename);
260 /* Returns whether the image file is opened as read-only. Note that this can
261 * return false and writing to the image file is still not possible because the
262 * image is inactivated. */
263 bool bdrv_is_read_only(BlockDriverState *bs)
265 return !(bs->open_flags & BDRV_O_RDWR);
268 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
269 bool ignore_allow_rdw, Error **errp)
271 /* Do not set read_only if copy_on_read is enabled */
272 if (bs->copy_on_read && read_only) {
273 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
274 bdrv_get_device_or_node_name(bs));
275 return -EINVAL;
278 /* Do not clear read_only if it is prohibited */
279 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
280 !ignore_allow_rdw)
282 error_setg(errp, "Node '%s' is read only",
283 bdrv_get_device_or_node_name(bs));
284 return -EPERM;
287 return 0;
291 * Called by a driver that can only provide a read-only image.
293 * Returns 0 if the node is already read-only or it could switch the node to
294 * read-only because BDRV_O_AUTO_RDONLY is set.
296 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
297 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
298 * is not NULL, it is used as the error message for the Error object.
300 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
301 Error **errp)
303 int ret = 0;
305 if (!(bs->open_flags & BDRV_O_RDWR)) {
306 return 0;
308 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
309 goto fail;
312 ret = bdrv_can_set_read_only(bs, true, false, NULL);
313 if (ret < 0) {
314 goto fail;
317 bs->open_flags &= ~BDRV_O_RDWR;
319 return 0;
321 fail:
322 error_setg(errp, "%s", errmsg ?: "Image is read-only");
323 return -EACCES;
327 * If @backing is empty, this function returns NULL without setting
328 * @errp. In all other cases, NULL will only be returned with @errp
329 * set.
331 * Therefore, a return value of NULL without @errp set means that
332 * there is no backing file; if @errp is set, there is one but its
333 * absolute filename cannot be generated.
335 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
336 const char *backing,
337 Error **errp)
339 if (backing[0] == '\0') {
340 return NULL;
341 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
342 return g_strdup(backing);
343 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
344 error_setg(errp, "Cannot use relative backing file names for '%s'",
345 backed);
346 return NULL;
347 } else {
348 return path_combine(backed, backing);
353 * If @filename is empty or NULL, this function returns NULL without
354 * setting @errp. In all other cases, NULL will only be returned with
355 * @errp set.
357 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
358 const char *filename, Error **errp)
360 char *dir, *full_name;
362 if (!filename || filename[0] == '\0') {
363 return NULL;
364 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
365 return g_strdup(filename);
368 dir = bdrv_dirname(relative_to, errp);
369 if (!dir) {
370 return NULL;
373 full_name = g_strconcat(dir, filename, NULL);
374 g_free(dir);
375 return full_name;
378 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
380 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
383 void bdrv_register(BlockDriver *bdrv)
385 assert(bdrv->format_name);
386 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
389 BlockDriverState *bdrv_new(void)
391 BlockDriverState *bs;
392 int i;
394 bs = g_new0(BlockDriverState, 1);
395 QLIST_INIT(&bs->dirty_bitmaps);
396 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
397 QLIST_INIT(&bs->op_blockers[i]);
399 qemu_co_mutex_init(&bs->reqs_lock);
400 qemu_mutex_init(&bs->dirty_bitmap_mutex);
401 bs->refcnt = 1;
402 bs->aio_context = qemu_get_aio_context();
404 qemu_co_queue_init(&bs->flush_queue);
406 qemu_co_mutex_init(&bs->bsc_modify_lock);
407 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
409 for (i = 0; i < bdrv_drain_all_count; i++) {
410 bdrv_drained_begin(bs);
413 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
415 return bs;
418 static BlockDriver *bdrv_do_find_format(const char *format_name)
420 BlockDriver *drv1;
422 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
423 if (!strcmp(drv1->format_name, format_name)) {
424 return drv1;
428 return NULL;
431 BlockDriver *bdrv_find_format(const char *format_name)
433 BlockDriver *drv1;
434 int i;
436 drv1 = bdrv_do_find_format(format_name);
437 if (drv1) {
438 return drv1;
441 /* The driver isn't registered, maybe we need to load a module */
442 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
443 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
444 block_module_load_one(block_driver_modules[i].library_name);
445 break;
449 return bdrv_do_find_format(format_name);
452 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
454 static const char *whitelist_rw[] = {
455 CONFIG_BDRV_RW_WHITELIST
456 NULL
458 static const char *whitelist_ro[] = {
459 CONFIG_BDRV_RO_WHITELIST
460 NULL
462 const char **p;
464 if (!whitelist_rw[0] && !whitelist_ro[0]) {
465 return 1; /* no whitelist, anything goes */
468 for (p = whitelist_rw; *p; p++) {
469 if (!strcmp(format_name, *p)) {
470 return 1;
473 if (read_only) {
474 for (p = whitelist_ro; *p; p++) {
475 if (!strcmp(format_name, *p)) {
476 return 1;
480 return 0;
483 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
485 return bdrv_format_is_whitelisted(drv->format_name, read_only);
488 bool bdrv_uses_whitelist(void)
490 return use_bdrv_whitelist;
493 typedef struct CreateCo {
494 BlockDriver *drv;
495 char *filename;
496 QemuOpts *opts;
497 int ret;
498 Error *err;
499 } CreateCo;
501 static void coroutine_fn bdrv_create_co_entry(void *opaque)
503 Error *local_err = NULL;
504 int ret;
506 CreateCo *cco = opaque;
507 assert(cco->drv);
509 ret = cco->drv->bdrv_co_create_opts(cco->drv,
510 cco->filename, cco->opts, &local_err);
511 error_propagate(&cco->err, local_err);
512 cco->ret = ret;
515 int bdrv_create(BlockDriver *drv, const char* filename,
516 QemuOpts *opts, Error **errp)
518 int ret;
520 Coroutine *co;
521 CreateCo cco = {
522 .drv = drv,
523 .filename = g_strdup(filename),
524 .opts = opts,
525 .ret = NOT_DONE,
526 .err = NULL,
529 if (!drv->bdrv_co_create_opts) {
530 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
531 ret = -ENOTSUP;
532 goto out;
535 if (qemu_in_coroutine()) {
536 /* Fast-path if already in coroutine context */
537 bdrv_create_co_entry(&cco);
538 } else {
539 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
540 qemu_coroutine_enter(co);
541 while (cco.ret == NOT_DONE) {
542 aio_poll(qemu_get_aio_context(), true);
546 ret = cco.ret;
547 if (ret < 0) {
548 if (cco.err) {
549 error_propagate(errp, cco.err);
550 } else {
551 error_setg_errno(errp, -ret, "Could not create image");
555 out:
556 g_free(cco.filename);
557 return ret;
561 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
562 * least the given @minimum_size.
564 * On success, return @blk's actual length.
565 * Otherwise, return -errno.
567 static int64_t create_file_fallback_truncate(BlockBackend *blk,
568 int64_t minimum_size, Error **errp)
570 Error *local_err = NULL;
571 int64_t size;
572 int ret;
574 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
575 &local_err);
576 if (ret < 0 && ret != -ENOTSUP) {
577 error_propagate(errp, local_err);
578 return ret;
581 size = blk_getlength(blk);
582 if (size < 0) {
583 error_free(local_err);
584 error_setg_errno(errp, -size,
585 "Failed to inquire the new image file's length");
586 return size;
589 if (size < minimum_size) {
590 /* Need to grow the image, but we failed to do that */
591 error_propagate(errp, local_err);
592 return -ENOTSUP;
595 error_free(local_err);
596 local_err = NULL;
598 return size;
602 * Helper function for bdrv_create_file_fallback(): Zero the first
603 * sector to remove any potentially pre-existing image header.
605 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
606 int64_t current_size,
607 Error **errp)
609 int64_t bytes_to_clear;
610 int ret;
612 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
613 if (bytes_to_clear) {
614 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
615 if (ret < 0) {
616 error_setg_errno(errp, -ret,
617 "Failed to clear the new image's first sector");
618 return ret;
622 return 0;
626 * Simple implementation of bdrv_co_create_opts for protocol drivers
627 * which only support creation via opening a file
628 * (usually existing raw storage device)
630 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
631 const char *filename,
632 QemuOpts *opts,
633 Error **errp)
635 BlockBackend *blk;
636 QDict *options;
637 int64_t size = 0;
638 char *buf = NULL;
639 PreallocMode prealloc;
640 Error *local_err = NULL;
641 int ret;
643 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
644 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
645 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
646 PREALLOC_MODE_OFF, &local_err);
647 g_free(buf);
648 if (local_err) {
649 error_propagate(errp, local_err);
650 return -EINVAL;
653 if (prealloc != PREALLOC_MODE_OFF) {
654 error_setg(errp, "Unsupported preallocation mode '%s'",
655 PreallocMode_str(prealloc));
656 return -ENOTSUP;
659 options = qdict_new();
660 qdict_put_str(options, "driver", drv->format_name);
662 blk = blk_new_open(filename, NULL, options,
663 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
664 if (!blk) {
665 error_prepend(errp, "Protocol driver '%s' does not support image "
666 "creation, and opening the image failed: ",
667 drv->format_name);
668 return -EINVAL;
671 size = create_file_fallback_truncate(blk, size, errp);
672 if (size < 0) {
673 ret = size;
674 goto out;
677 ret = create_file_fallback_zero_first_sector(blk, size, errp);
678 if (ret < 0) {
679 goto out;
682 ret = 0;
683 out:
684 blk_unref(blk);
685 return ret;
688 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
690 QemuOpts *protocol_opts;
691 BlockDriver *drv;
692 QDict *qdict;
693 int ret;
695 drv = bdrv_find_protocol(filename, true, errp);
696 if (drv == NULL) {
697 return -ENOENT;
700 if (!drv->create_opts) {
701 error_setg(errp, "Driver '%s' does not support image creation",
702 drv->format_name);
703 return -ENOTSUP;
707 * 'opts' contains a QemuOptsList with a combination of format and protocol
708 * default values.
710 * The format properly removes its options, but the default values remain
711 * in 'opts->list'. So if the protocol has options with the same name
712 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
713 * of the format, since for overlapping options, the format wins.
715 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
716 * only the set options, and then convert it back to QemuOpts, using the
717 * create_opts of the protocol. So the new QemuOpts, will contain only the
718 * protocol defaults.
720 qdict = qemu_opts_to_qdict(opts, NULL);
721 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
722 if (protocol_opts == NULL) {
723 ret = -EINVAL;
724 goto out;
727 ret = bdrv_create(drv, filename, protocol_opts, errp);
728 out:
729 qemu_opts_del(protocol_opts);
730 qobject_unref(qdict);
731 return ret;
734 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
736 Error *local_err = NULL;
737 int ret;
739 assert(bs != NULL);
741 if (!bs->drv) {
742 error_setg(errp, "Block node '%s' is not opened", bs->filename);
743 return -ENOMEDIUM;
746 if (!bs->drv->bdrv_co_delete_file) {
747 error_setg(errp, "Driver '%s' does not support image deletion",
748 bs->drv->format_name);
749 return -ENOTSUP;
752 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
753 if (ret < 0) {
754 error_propagate(errp, local_err);
757 return ret;
760 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
762 Error *local_err = NULL;
763 int ret;
765 if (!bs) {
766 return;
769 ret = bdrv_co_delete_file(bs, &local_err);
771 * ENOTSUP will happen if the block driver doesn't support
772 * the 'bdrv_co_delete_file' interface. This is a predictable
773 * scenario and shouldn't be reported back to the user.
775 if (ret == -ENOTSUP) {
776 error_free(local_err);
777 } else if (ret < 0) {
778 error_report_err(local_err);
783 * Try to get @bs's logical and physical block size.
784 * On success, store them in @bsz struct and return 0.
785 * On failure return -errno.
786 * @bs must not be empty.
788 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
790 BlockDriver *drv = bs->drv;
791 BlockDriverState *filtered = bdrv_filter_bs(bs);
793 if (drv && drv->bdrv_probe_blocksizes) {
794 return drv->bdrv_probe_blocksizes(bs, bsz);
795 } else if (filtered) {
796 return bdrv_probe_blocksizes(filtered, bsz);
799 return -ENOTSUP;
803 * Try to get @bs's geometry (cyls, heads, sectors).
804 * On success, store them in @geo struct and return 0.
805 * On failure return -errno.
806 * @bs must not be empty.
808 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
810 BlockDriver *drv = bs->drv;
811 BlockDriverState *filtered = bdrv_filter_bs(bs);
813 if (drv && drv->bdrv_probe_geometry) {
814 return drv->bdrv_probe_geometry(bs, geo);
815 } else if (filtered) {
816 return bdrv_probe_geometry(filtered, geo);
819 return -ENOTSUP;
823 * Create a uniquely-named empty temporary file.
824 * Return 0 upon success, otherwise a negative errno value.
826 int get_tmp_filename(char *filename, int size)
828 #ifdef _WIN32
829 char temp_dir[MAX_PATH];
830 /* GetTempFileName requires that its output buffer (4th param)
831 have length MAX_PATH or greater. */
832 assert(size >= MAX_PATH);
833 return (GetTempPath(MAX_PATH, temp_dir)
834 && GetTempFileName(temp_dir, "qem", 0, filename)
835 ? 0 : -GetLastError());
836 #else
837 int fd;
838 const char *tmpdir;
839 tmpdir = getenv("TMPDIR");
840 if (!tmpdir) {
841 tmpdir = "/var/tmp";
843 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
844 return -EOVERFLOW;
846 fd = mkstemp(filename);
847 if (fd < 0) {
848 return -errno;
850 if (close(fd) != 0) {
851 unlink(filename);
852 return -errno;
854 return 0;
855 #endif
859 * Detect host devices. By convention, /dev/cdrom[N] is always
860 * recognized as a host CDROM.
862 static BlockDriver *find_hdev_driver(const char *filename)
864 int score_max = 0, score;
865 BlockDriver *drv = NULL, *d;
867 QLIST_FOREACH(d, &bdrv_drivers, list) {
868 if (d->bdrv_probe_device) {
869 score = d->bdrv_probe_device(filename);
870 if (score > score_max) {
871 score_max = score;
872 drv = d;
877 return drv;
880 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
882 BlockDriver *drv1;
884 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
885 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
886 return drv1;
890 return NULL;
893 BlockDriver *bdrv_find_protocol(const char *filename,
894 bool allow_protocol_prefix,
895 Error **errp)
897 BlockDriver *drv1;
898 char protocol[128];
899 int len;
900 const char *p;
901 int i;
903 /* TODO Drivers without bdrv_file_open must be specified explicitly */
906 * XXX(hch): we really should not let host device detection
907 * override an explicit protocol specification, but moving this
908 * later breaks access to device names with colons in them.
909 * Thanks to the brain-dead persistent naming schemes on udev-
910 * based Linux systems those actually are quite common.
912 drv1 = find_hdev_driver(filename);
913 if (drv1) {
914 return drv1;
917 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
918 return &bdrv_file;
921 p = strchr(filename, ':');
922 assert(p != NULL);
923 len = p - filename;
924 if (len > sizeof(protocol) - 1)
925 len = sizeof(protocol) - 1;
926 memcpy(protocol, filename, len);
927 protocol[len] = '\0';
929 drv1 = bdrv_do_find_protocol(protocol);
930 if (drv1) {
931 return drv1;
934 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
935 if (block_driver_modules[i].protocol_name &&
936 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
937 block_module_load_one(block_driver_modules[i].library_name);
938 break;
942 drv1 = bdrv_do_find_protocol(protocol);
943 if (!drv1) {
944 error_setg(errp, "Unknown protocol '%s'", protocol);
946 return drv1;
950 * Guess image format by probing its contents.
951 * This is not a good idea when your image is raw (CVE-2008-2004), but
952 * we do it anyway for backward compatibility.
954 * @buf contains the image's first @buf_size bytes.
955 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
956 * but can be smaller if the image file is smaller)
957 * @filename is its filename.
959 * For all block drivers, call the bdrv_probe() method to get its
960 * probing score.
961 * Return the first block driver with the highest probing score.
963 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
964 const char *filename)
966 int score_max = 0, score;
967 BlockDriver *drv = NULL, *d;
969 QLIST_FOREACH(d, &bdrv_drivers, list) {
970 if (d->bdrv_probe) {
971 score = d->bdrv_probe(buf, buf_size, filename);
972 if (score > score_max) {
973 score_max = score;
974 drv = d;
979 return drv;
982 static int find_image_format(BlockBackend *file, const char *filename,
983 BlockDriver **pdrv, Error **errp)
985 BlockDriver *drv;
986 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
987 int ret = 0;
989 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
990 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
991 *pdrv = &bdrv_raw;
992 return ret;
995 ret = blk_pread(file, 0, buf, sizeof(buf));
996 if (ret < 0) {
997 error_setg_errno(errp, -ret, "Could not read image for determining its "
998 "format");
999 *pdrv = NULL;
1000 return ret;
1003 drv = bdrv_probe_all(buf, ret, filename);
1004 if (!drv) {
1005 error_setg(errp, "Could not determine image format: No compatible "
1006 "driver found");
1007 ret = -ENOENT;
1009 *pdrv = drv;
1010 return ret;
1014 * Set the current 'total_sectors' value
1015 * Return 0 on success, -errno on error.
1017 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
1019 BlockDriver *drv = bs->drv;
1021 if (!drv) {
1022 return -ENOMEDIUM;
1025 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1026 if (bdrv_is_sg(bs))
1027 return 0;
1029 /* query actual device if possible, otherwise just trust the hint */
1030 if (drv->bdrv_getlength) {
1031 int64_t length = drv->bdrv_getlength(bs);
1032 if (length < 0) {
1033 return length;
1035 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1038 bs->total_sectors = hint;
1040 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1041 return -EFBIG;
1044 return 0;
1048 * Combines a QDict of new block driver @options with any missing options taken
1049 * from @old_options, so that leaving out an option defaults to its old value.
1051 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1052 QDict *old_options)
1054 if (bs->drv && bs->drv->bdrv_join_options) {
1055 bs->drv->bdrv_join_options(options, old_options);
1056 } else {
1057 qdict_join(options, old_options, false);
1061 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1062 int open_flags,
1063 Error **errp)
1065 Error *local_err = NULL;
1066 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1067 BlockdevDetectZeroesOptions detect_zeroes =
1068 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1069 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1070 g_free(value);
1071 if (local_err) {
1072 error_propagate(errp, local_err);
1073 return detect_zeroes;
1076 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1077 !(open_flags & BDRV_O_UNMAP))
1079 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1080 "without setting discard operation to unmap");
1083 return detect_zeroes;
1087 * Set open flags for aio engine
1089 * Return 0 on success, -1 if the engine specified is invalid
1091 int bdrv_parse_aio(const char *mode, int *flags)
1093 if (!strcmp(mode, "threads")) {
1094 /* do nothing, default */
1095 } else if (!strcmp(mode, "native")) {
1096 *flags |= BDRV_O_NATIVE_AIO;
1097 #ifdef CONFIG_LINUX_IO_URING
1098 } else if (!strcmp(mode, "io_uring")) {
1099 *flags |= BDRV_O_IO_URING;
1100 #endif
1101 } else {
1102 return -1;
1105 return 0;
1109 * Set open flags for a given discard mode
1111 * Return 0 on success, -1 if the discard mode was invalid.
1113 int bdrv_parse_discard_flags(const char *mode, int *flags)
1115 *flags &= ~BDRV_O_UNMAP;
1117 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1118 /* do nothing */
1119 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1120 *flags |= BDRV_O_UNMAP;
1121 } else {
1122 return -1;
1125 return 0;
1129 * Set open flags for a given cache mode
1131 * Return 0 on success, -1 if the cache mode was invalid.
1133 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1135 *flags &= ~BDRV_O_CACHE_MASK;
1137 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1138 *writethrough = false;
1139 *flags |= BDRV_O_NOCACHE;
1140 } else if (!strcmp(mode, "directsync")) {
1141 *writethrough = true;
1142 *flags |= BDRV_O_NOCACHE;
1143 } else if (!strcmp(mode, "writeback")) {
1144 *writethrough = false;
1145 } else if (!strcmp(mode, "unsafe")) {
1146 *writethrough = false;
1147 *flags |= BDRV_O_NO_FLUSH;
1148 } else if (!strcmp(mode, "writethrough")) {
1149 *writethrough = true;
1150 } else {
1151 return -1;
1154 return 0;
1157 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1159 BlockDriverState *parent = c->opaque;
1160 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1163 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1165 BlockDriverState *bs = child->opaque;
1166 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1169 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1171 BlockDriverState *bs = child->opaque;
1172 return bdrv_drain_poll(bs, false, NULL, false);
1175 static void bdrv_child_cb_drained_end(BdrvChild *child,
1176 int *drained_end_counter)
1178 BlockDriverState *bs = child->opaque;
1179 bdrv_drained_end_no_poll(bs, drained_end_counter);
1182 static int bdrv_child_cb_inactivate(BdrvChild *child)
1184 BlockDriverState *bs = child->opaque;
1185 assert(bs->open_flags & BDRV_O_INACTIVE);
1186 return 0;
1189 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1190 GSList **ignore, Error **errp)
1192 BlockDriverState *bs = child->opaque;
1193 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1196 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1197 GSList **ignore)
1199 BlockDriverState *bs = child->opaque;
1200 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1204 * Returns the options and flags that a temporary snapshot should get, based on
1205 * the originally requested flags (the originally requested image will have
1206 * flags like a backing file)
1208 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1209 int parent_flags, QDict *parent_options)
1211 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1213 /* For temporary files, unconditional cache=unsafe is fine */
1214 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1215 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1217 /* Copy the read-only and discard options from the parent */
1218 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1219 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1221 /* aio=native doesn't work for cache.direct=off, so disable it for the
1222 * temporary snapshot */
1223 *child_flags &= ~BDRV_O_NATIVE_AIO;
1226 static void bdrv_backing_attach(BdrvChild *c)
1228 BlockDriverState *parent = c->opaque;
1229 BlockDriverState *backing_hd = c->bs;
1231 assert(!parent->backing_blocker);
1232 error_setg(&parent->backing_blocker,
1233 "node is used as backing hd of '%s'",
1234 bdrv_get_device_or_node_name(parent));
1236 bdrv_refresh_filename(backing_hd);
1238 parent->open_flags &= ~BDRV_O_NO_BACKING;
1240 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1241 /* Otherwise we won't be able to commit or stream */
1242 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1243 parent->backing_blocker);
1244 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1245 parent->backing_blocker);
1247 * We do backup in 3 ways:
1248 * 1. drive backup
1249 * The target bs is new opened, and the source is top BDS
1250 * 2. blockdev backup
1251 * Both the source and the target are top BDSes.
1252 * 3. internal backup(used for block replication)
1253 * Both the source and the target are backing file
1255 * In case 1 and 2, neither the source nor the target is the backing file.
1256 * In case 3, we will block the top BDS, so there is only one block job
1257 * for the top BDS and its backing chain.
1259 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1260 parent->backing_blocker);
1261 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1262 parent->backing_blocker);
1265 static void bdrv_backing_detach(BdrvChild *c)
1267 BlockDriverState *parent = c->opaque;
1269 assert(parent->backing_blocker);
1270 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1271 error_free(parent->backing_blocker);
1272 parent->backing_blocker = NULL;
1275 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1276 const char *filename, Error **errp)
1278 BlockDriverState *parent = c->opaque;
1279 bool read_only = bdrv_is_read_only(parent);
1280 int ret;
1282 if (read_only) {
1283 ret = bdrv_reopen_set_read_only(parent, false, errp);
1284 if (ret < 0) {
1285 return ret;
1289 ret = bdrv_change_backing_file(parent, filename,
1290 base->drv ? base->drv->format_name : "",
1291 false);
1292 if (ret < 0) {
1293 error_setg_errno(errp, -ret, "Could not update backing file link");
1296 if (read_only) {
1297 bdrv_reopen_set_read_only(parent, true, NULL);
1300 return ret;
1304 * Returns the options and flags that a generic child of a BDS should
1305 * get, based on the given options and flags for the parent BDS.
1307 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1308 int *child_flags, QDict *child_options,
1309 int parent_flags, QDict *parent_options)
1311 int flags = parent_flags;
1314 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1315 * Generally, the question to answer is: Should this child be
1316 * format-probed by default?
1320 * Pure and non-filtered data children of non-format nodes should
1321 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1322 * set). This only affects a very limited set of drivers (namely
1323 * quorum and blkverify when this comment was written).
1324 * Force-clear BDRV_O_PROTOCOL then.
1326 if (!parent_is_format &&
1327 (role & BDRV_CHILD_DATA) &&
1328 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1330 flags &= ~BDRV_O_PROTOCOL;
1334 * All children of format nodes (except for COW children) and all
1335 * metadata children in general should never be format-probed.
1336 * Force-set BDRV_O_PROTOCOL then.
1338 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1339 (role & BDRV_CHILD_METADATA))
1341 flags |= BDRV_O_PROTOCOL;
1345 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1346 * the parent.
1348 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1349 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1350 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1352 if (role & BDRV_CHILD_COW) {
1353 /* backing files are opened read-only by default */
1354 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1355 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1356 } else {
1357 /* Inherit the read-only option from the parent if it's not set */
1358 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1359 qdict_copy_default(child_options, parent_options,
1360 BDRV_OPT_AUTO_READ_ONLY);
1364 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1365 * can default to enable it on lower layers regardless of the
1366 * parent option.
1368 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1370 /* Clear flags that only apply to the top layer */
1371 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1373 if (role & BDRV_CHILD_METADATA) {
1374 flags &= ~BDRV_O_NO_IO;
1376 if (role & BDRV_CHILD_COW) {
1377 flags &= ~BDRV_O_TEMPORARY;
1380 *child_flags = flags;
1383 static void bdrv_child_cb_attach(BdrvChild *child)
1385 BlockDriverState *bs = child->opaque;
1387 if (child->role & BDRV_CHILD_COW) {
1388 bdrv_backing_attach(child);
1391 bdrv_apply_subtree_drain(child, bs);
1394 static void bdrv_child_cb_detach(BdrvChild *child)
1396 BlockDriverState *bs = child->opaque;
1398 if (child->role & BDRV_CHILD_COW) {
1399 bdrv_backing_detach(child);
1402 bdrv_unapply_subtree_drain(child, bs);
1405 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1406 const char *filename, Error **errp)
1408 if (c->role & BDRV_CHILD_COW) {
1409 return bdrv_backing_update_filename(c, base, filename, errp);
1411 return 0;
1414 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1416 BlockDriverState *bs = c->opaque;
1418 return bdrv_get_aio_context(bs);
1421 const BdrvChildClass child_of_bds = {
1422 .parent_is_bds = true,
1423 .get_parent_desc = bdrv_child_get_parent_desc,
1424 .inherit_options = bdrv_inherited_options,
1425 .drained_begin = bdrv_child_cb_drained_begin,
1426 .drained_poll = bdrv_child_cb_drained_poll,
1427 .drained_end = bdrv_child_cb_drained_end,
1428 .attach = bdrv_child_cb_attach,
1429 .detach = bdrv_child_cb_detach,
1430 .inactivate = bdrv_child_cb_inactivate,
1431 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1432 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1433 .update_filename = bdrv_child_cb_update_filename,
1434 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1437 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1439 return c->klass->get_parent_aio_context(c);
1442 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1444 int open_flags = flags;
1447 * Clear flags that are internal to the block layer before opening the
1448 * image.
1450 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1452 return open_flags;
1455 static void update_flags_from_options(int *flags, QemuOpts *opts)
1457 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1459 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1460 *flags |= BDRV_O_NO_FLUSH;
1463 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1464 *flags |= BDRV_O_NOCACHE;
1467 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1468 *flags |= BDRV_O_RDWR;
1471 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1472 *flags |= BDRV_O_AUTO_RDONLY;
1476 static void update_options_from_flags(QDict *options, int flags)
1478 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1479 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1481 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1482 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1483 flags & BDRV_O_NO_FLUSH);
1485 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1486 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1488 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1489 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1490 flags & BDRV_O_AUTO_RDONLY);
1494 static void bdrv_assign_node_name(BlockDriverState *bs,
1495 const char *node_name,
1496 Error **errp)
1498 char *gen_node_name = NULL;
1500 if (!node_name) {
1501 node_name = gen_node_name = id_generate(ID_BLOCK);
1502 } else if (!id_wellformed(node_name)) {
1504 * Check for empty string or invalid characters, but not if it is
1505 * generated (generated names use characters not available to the user)
1507 error_setg(errp, "Invalid node-name: '%s'", node_name);
1508 return;
1511 /* takes care of avoiding namespaces collisions */
1512 if (blk_by_name(node_name)) {
1513 error_setg(errp, "node-name=%s is conflicting with a device id",
1514 node_name);
1515 goto out;
1518 /* takes care of avoiding duplicates node names */
1519 if (bdrv_find_node(node_name)) {
1520 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1521 goto out;
1524 /* Make sure that the node name isn't truncated */
1525 if (strlen(node_name) >= sizeof(bs->node_name)) {
1526 error_setg(errp, "Node name too long");
1527 goto out;
1530 /* copy node name into the bs and insert it into the graph list */
1531 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1532 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1533 out:
1534 g_free(gen_node_name);
1537 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1538 const char *node_name, QDict *options,
1539 int open_flags, Error **errp)
1541 Error *local_err = NULL;
1542 int i, ret;
1544 bdrv_assign_node_name(bs, node_name, &local_err);
1545 if (local_err) {
1546 error_propagate(errp, local_err);
1547 return -EINVAL;
1550 bs->drv = drv;
1551 bs->opaque = g_malloc0(drv->instance_size);
1553 if (drv->bdrv_file_open) {
1554 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1555 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1556 } else if (drv->bdrv_open) {
1557 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1558 } else {
1559 ret = 0;
1562 if (ret < 0) {
1563 if (local_err) {
1564 error_propagate(errp, local_err);
1565 } else if (bs->filename[0]) {
1566 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1567 } else {
1568 error_setg_errno(errp, -ret, "Could not open image");
1570 goto open_failed;
1573 ret = refresh_total_sectors(bs, bs->total_sectors);
1574 if (ret < 0) {
1575 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1576 return ret;
1579 bdrv_refresh_limits(bs, NULL, &local_err);
1580 if (local_err) {
1581 error_propagate(errp, local_err);
1582 return -EINVAL;
1585 assert(bdrv_opt_mem_align(bs) != 0);
1586 assert(bdrv_min_mem_align(bs) != 0);
1587 assert(is_power_of_2(bs->bl.request_alignment));
1589 for (i = 0; i < bs->quiesce_counter; i++) {
1590 if (drv->bdrv_co_drain_begin) {
1591 drv->bdrv_co_drain_begin(bs);
1595 return 0;
1596 open_failed:
1597 bs->drv = NULL;
1598 if (bs->file != NULL) {
1599 bdrv_unref_child(bs, bs->file);
1600 bs->file = NULL;
1602 g_free(bs->opaque);
1603 bs->opaque = NULL;
1604 return ret;
1607 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1608 int flags, Error **errp)
1610 BlockDriverState *bs;
1611 int ret;
1613 bs = bdrv_new();
1614 bs->open_flags = flags;
1615 bs->explicit_options = qdict_new();
1616 bs->options = qdict_new();
1617 bs->opaque = NULL;
1619 update_options_from_flags(bs->options, flags);
1621 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1622 if (ret < 0) {
1623 qobject_unref(bs->explicit_options);
1624 bs->explicit_options = NULL;
1625 qobject_unref(bs->options);
1626 bs->options = NULL;
1627 bdrv_unref(bs);
1628 return NULL;
1631 return bs;
1634 QemuOptsList bdrv_runtime_opts = {
1635 .name = "bdrv_common",
1636 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1637 .desc = {
1639 .name = "node-name",
1640 .type = QEMU_OPT_STRING,
1641 .help = "Node name of the block device node",
1644 .name = "driver",
1645 .type = QEMU_OPT_STRING,
1646 .help = "Block driver to use for the node",
1649 .name = BDRV_OPT_CACHE_DIRECT,
1650 .type = QEMU_OPT_BOOL,
1651 .help = "Bypass software writeback cache on the host",
1654 .name = BDRV_OPT_CACHE_NO_FLUSH,
1655 .type = QEMU_OPT_BOOL,
1656 .help = "Ignore flush requests",
1659 .name = BDRV_OPT_READ_ONLY,
1660 .type = QEMU_OPT_BOOL,
1661 .help = "Node is opened in read-only mode",
1664 .name = BDRV_OPT_AUTO_READ_ONLY,
1665 .type = QEMU_OPT_BOOL,
1666 .help = "Node can become read-only if opening read-write fails",
1669 .name = "detect-zeroes",
1670 .type = QEMU_OPT_STRING,
1671 .help = "try to optimize zero writes (off, on, unmap)",
1674 .name = BDRV_OPT_DISCARD,
1675 .type = QEMU_OPT_STRING,
1676 .help = "discard operation (ignore/off, unmap/on)",
1679 .name = BDRV_OPT_FORCE_SHARE,
1680 .type = QEMU_OPT_BOOL,
1681 .help = "always accept other writers (default: off)",
1683 { /* end of list */ }
1687 QemuOptsList bdrv_create_opts_simple = {
1688 .name = "simple-create-opts",
1689 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1690 .desc = {
1692 .name = BLOCK_OPT_SIZE,
1693 .type = QEMU_OPT_SIZE,
1694 .help = "Virtual disk size"
1697 .name = BLOCK_OPT_PREALLOC,
1698 .type = QEMU_OPT_STRING,
1699 .help = "Preallocation mode (allowed values: off)"
1701 { /* end of list */ }
1706 * Common part for opening disk images and files
1708 * Removes all processed options from *options.
1710 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1711 QDict *options, Error **errp)
1713 int ret, open_flags;
1714 const char *filename;
1715 const char *driver_name = NULL;
1716 const char *node_name = NULL;
1717 const char *discard;
1718 QemuOpts *opts;
1719 BlockDriver *drv;
1720 Error *local_err = NULL;
1721 bool ro;
1723 assert(bs->file == NULL);
1724 assert(options != NULL && bs->options != options);
1726 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1727 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1728 ret = -EINVAL;
1729 goto fail_opts;
1732 update_flags_from_options(&bs->open_flags, opts);
1734 driver_name = qemu_opt_get(opts, "driver");
1735 drv = bdrv_find_format(driver_name);
1736 assert(drv != NULL);
1738 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1740 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1741 error_setg(errp,
1742 BDRV_OPT_FORCE_SHARE
1743 "=on can only be used with read-only images");
1744 ret = -EINVAL;
1745 goto fail_opts;
1748 if (file != NULL) {
1749 bdrv_refresh_filename(blk_bs(file));
1750 filename = blk_bs(file)->filename;
1751 } else {
1753 * Caution: while qdict_get_try_str() is fine, getting
1754 * non-string types would require more care. When @options
1755 * come from -blockdev or blockdev_add, its members are typed
1756 * according to the QAPI schema, but when they come from
1757 * -drive, they're all QString.
1759 filename = qdict_get_try_str(options, "filename");
1762 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1763 error_setg(errp, "The '%s' block driver requires a file name",
1764 drv->format_name);
1765 ret = -EINVAL;
1766 goto fail_opts;
1769 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1770 drv->format_name);
1772 ro = bdrv_is_read_only(bs);
1774 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1775 if (!ro && bdrv_is_whitelisted(drv, true)) {
1776 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1777 } else {
1778 ret = -ENOTSUP;
1780 if (ret < 0) {
1781 error_setg(errp,
1782 !ro && bdrv_is_whitelisted(drv, true)
1783 ? "Driver '%s' can only be used for read-only devices"
1784 : "Driver '%s' is not whitelisted",
1785 drv->format_name);
1786 goto fail_opts;
1790 /* bdrv_new() and bdrv_close() make it so */
1791 assert(qatomic_read(&bs->copy_on_read) == 0);
1793 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1794 if (!ro) {
1795 bdrv_enable_copy_on_read(bs);
1796 } else {
1797 error_setg(errp, "Can't use copy-on-read on read-only device");
1798 ret = -EINVAL;
1799 goto fail_opts;
1803 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1804 if (discard != NULL) {
1805 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1806 error_setg(errp, "Invalid discard option");
1807 ret = -EINVAL;
1808 goto fail_opts;
1812 bs->detect_zeroes =
1813 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1814 if (local_err) {
1815 error_propagate(errp, local_err);
1816 ret = -EINVAL;
1817 goto fail_opts;
1820 if (filename != NULL) {
1821 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1822 } else {
1823 bs->filename[0] = '\0';
1825 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1827 /* Open the image, either directly or using a protocol */
1828 open_flags = bdrv_open_flags(bs, bs->open_flags);
1829 node_name = qemu_opt_get(opts, "node-name");
1831 assert(!drv->bdrv_file_open || file == NULL);
1832 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1833 if (ret < 0) {
1834 goto fail_opts;
1837 qemu_opts_del(opts);
1838 return 0;
1840 fail_opts:
1841 qemu_opts_del(opts);
1842 return ret;
1845 static QDict *parse_json_filename(const char *filename, Error **errp)
1847 QObject *options_obj;
1848 QDict *options;
1849 int ret;
1851 ret = strstart(filename, "json:", &filename);
1852 assert(ret);
1854 options_obj = qobject_from_json(filename, errp);
1855 if (!options_obj) {
1856 error_prepend(errp, "Could not parse the JSON options: ");
1857 return NULL;
1860 options = qobject_to(QDict, options_obj);
1861 if (!options) {
1862 qobject_unref(options_obj);
1863 error_setg(errp, "Invalid JSON object given");
1864 return NULL;
1867 qdict_flatten(options);
1869 return options;
1872 static void parse_json_protocol(QDict *options, const char **pfilename,
1873 Error **errp)
1875 QDict *json_options;
1876 Error *local_err = NULL;
1878 /* Parse json: pseudo-protocol */
1879 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1880 return;
1883 json_options = parse_json_filename(*pfilename, &local_err);
1884 if (local_err) {
1885 error_propagate(errp, local_err);
1886 return;
1889 /* Options given in the filename have lower priority than options
1890 * specified directly */
1891 qdict_join(options, json_options, false);
1892 qobject_unref(json_options);
1893 *pfilename = NULL;
1897 * Fills in default options for opening images and converts the legacy
1898 * filename/flags pair to option QDict entries.
1899 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1900 * block driver has been specified explicitly.
1902 static int bdrv_fill_options(QDict **options, const char *filename,
1903 int *flags, Error **errp)
1905 const char *drvname;
1906 bool protocol = *flags & BDRV_O_PROTOCOL;
1907 bool parse_filename = false;
1908 BlockDriver *drv = NULL;
1909 Error *local_err = NULL;
1912 * Caution: while qdict_get_try_str() is fine, getting non-string
1913 * types would require more care. When @options come from
1914 * -blockdev or blockdev_add, its members are typed according to
1915 * the QAPI schema, but when they come from -drive, they're all
1916 * QString.
1918 drvname = qdict_get_try_str(*options, "driver");
1919 if (drvname) {
1920 drv = bdrv_find_format(drvname);
1921 if (!drv) {
1922 error_setg(errp, "Unknown driver '%s'", drvname);
1923 return -ENOENT;
1925 /* If the user has explicitly specified the driver, this choice should
1926 * override the BDRV_O_PROTOCOL flag */
1927 protocol = drv->bdrv_file_open;
1930 if (protocol) {
1931 *flags |= BDRV_O_PROTOCOL;
1932 } else {
1933 *flags &= ~BDRV_O_PROTOCOL;
1936 /* Translate cache options from flags into options */
1937 update_options_from_flags(*options, *flags);
1939 /* Fetch the file name from the options QDict if necessary */
1940 if (protocol && filename) {
1941 if (!qdict_haskey(*options, "filename")) {
1942 qdict_put_str(*options, "filename", filename);
1943 parse_filename = true;
1944 } else {
1945 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1946 "the same time");
1947 return -EINVAL;
1951 /* Find the right block driver */
1952 /* See cautionary note on accessing @options above */
1953 filename = qdict_get_try_str(*options, "filename");
1955 if (!drvname && protocol) {
1956 if (filename) {
1957 drv = bdrv_find_protocol(filename, parse_filename, errp);
1958 if (!drv) {
1959 return -EINVAL;
1962 drvname = drv->format_name;
1963 qdict_put_str(*options, "driver", drvname);
1964 } else {
1965 error_setg(errp, "Must specify either driver or file");
1966 return -EINVAL;
1970 assert(drv || !protocol);
1972 /* Driver-specific filename parsing */
1973 if (drv && drv->bdrv_parse_filename && parse_filename) {
1974 drv->bdrv_parse_filename(filename, *options, &local_err);
1975 if (local_err) {
1976 error_propagate(errp, local_err);
1977 return -EINVAL;
1980 if (!drv->bdrv_needs_filename) {
1981 qdict_del(*options, "filename");
1985 return 0;
1988 typedef struct BlockReopenQueueEntry {
1989 bool prepared;
1990 bool perms_checked;
1991 BDRVReopenState state;
1992 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1993 } BlockReopenQueueEntry;
1996 * Return the flags that @bs will have after the reopens in @q have
1997 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1998 * return the current flags.
2000 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2002 BlockReopenQueueEntry *entry;
2004 if (q != NULL) {
2005 QTAILQ_FOREACH(entry, q, entry) {
2006 if (entry->state.bs == bs) {
2007 return entry->state.flags;
2012 return bs->open_flags;
2015 /* Returns whether the image file can be written to after the reopen queue @q
2016 * has been successfully applied, or right now if @q is NULL. */
2017 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2018 BlockReopenQueue *q)
2020 int flags = bdrv_reopen_get_flags(q, bs);
2022 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2026 * Return whether the BDS can be written to. This is not necessarily
2027 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2028 * be written to but do not count as read-only images.
2030 bool bdrv_is_writable(BlockDriverState *bs)
2032 return bdrv_is_writable_after_reopen(bs, NULL);
2035 static char *bdrv_child_user_desc(BdrvChild *c)
2037 return c->klass->get_parent_desc(c);
2041 * Check that @a allows everything that @b needs. @a and @b must reference same
2042 * child node.
2044 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2046 const char *child_bs_name;
2047 g_autofree char *a_user = NULL;
2048 g_autofree char *b_user = NULL;
2049 g_autofree char *perms = NULL;
2051 assert(a->bs);
2052 assert(a->bs == b->bs);
2054 if ((b->perm & a->shared_perm) == b->perm) {
2055 return true;
2058 child_bs_name = bdrv_get_node_name(b->bs);
2059 a_user = bdrv_child_user_desc(a);
2060 b_user = bdrv_child_user_desc(b);
2061 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2063 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2064 "both required by %s (uses node '%s' as '%s' child) and "
2065 "unshared by %s (uses node '%s' as '%s' child).",
2066 child_bs_name, perms,
2067 b_user, child_bs_name, b->name,
2068 a_user, child_bs_name, a->name);
2070 return false;
2073 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2075 BdrvChild *a, *b;
2078 * During the loop we'll look at each pair twice. That's correct because
2079 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2080 * directions.
2082 QLIST_FOREACH(a, &bs->parents, next_parent) {
2083 QLIST_FOREACH(b, &bs->parents, next_parent) {
2084 if (a == b) {
2085 continue;
2088 if (!bdrv_a_allow_b(a, b, errp)) {
2089 return true;
2094 return false;
2097 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2098 BdrvChild *c, BdrvChildRole role,
2099 BlockReopenQueue *reopen_queue,
2100 uint64_t parent_perm, uint64_t parent_shared,
2101 uint64_t *nperm, uint64_t *nshared)
2103 assert(bs->drv && bs->drv->bdrv_child_perm);
2104 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2105 parent_perm, parent_shared,
2106 nperm, nshared);
2107 /* TODO Take force_share from reopen_queue */
2108 if (child_bs && child_bs->force_share) {
2109 *nshared = BLK_PERM_ALL;
2114 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2115 * nodes that are already in the @list, of course) so that final list is
2116 * topologically sorted. Return the result (GSList @list object is updated, so
2117 * don't use old reference after function call).
2119 * On function start @list must be already topologically sorted and for any node
2120 * in the @list the whole subtree of the node must be in the @list as well. The
2121 * simplest way to satisfy this criteria: use only result of
2122 * bdrv_topological_dfs() or NULL as @list parameter.
2124 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2125 BlockDriverState *bs)
2127 BdrvChild *child;
2128 g_autoptr(GHashTable) local_found = NULL;
2130 if (!found) {
2131 assert(!list);
2132 found = local_found = g_hash_table_new(NULL, NULL);
2135 if (g_hash_table_contains(found, bs)) {
2136 return list;
2138 g_hash_table_add(found, bs);
2140 QLIST_FOREACH(child, &bs->children, next) {
2141 list = bdrv_topological_dfs(list, found, child->bs);
2144 return g_slist_prepend(list, bs);
2147 typedef struct BdrvChildSetPermState {
2148 BdrvChild *child;
2149 uint64_t old_perm;
2150 uint64_t old_shared_perm;
2151 } BdrvChildSetPermState;
2153 static void bdrv_child_set_perm_abort(void *opaque)
2155 BdrvChildSetPermState *s = opaque;
2157 s->child->perm = s->old_perm;
2158 s->child->shared_perm = s->old_shared_perm;
2161 static TransactionActionDrv bdrv_child_set_pem_drv = {
2162 .abort = bdrv_child_set_perm_abort,
2163 .clean = g_free,
2166 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2167 uint64_t shared, Transaction *tran)
2169 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2171 *s = (BdrvChildSetPermState) {
2172 .child = c,
2173 .old_perm = c->perm,
2174 .old_shared_perm = c->shared_perm,
2177 c->perm = perm;
2178 c->shared_perm = shared;
2180 tran_add(tran, &bdrv_child_set_pem_drv, s);
2183 static void bdrv_drv_set_perm_commit(void *opaque)
2185 BlockDriverState *bs = opaque;
2186 uint64_t cumulative_perms, cumulative_shared_perms;
2188 if (bs->drv->bdrv_set_perm) {
2189 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2190 &cumulative_shared_perms);
2191 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2195 static void bdrv_drv_set_perm_abort(void *opaque)
2197 BlockDriverState *bs = opaque;
2199 if (bs->drv->bdrv_abort_perm_update) {
2200 bs->drv->bdrv_abort_perm_update(bs);
2204 TransactionActionDrv bdrv_drv_set_perm_drv = {
2205 .abort = bdrv_drv_set_perm_abort,
2206 .commit = bdrv_drv_set_perm_commit,
2209 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2210 uint64_t shared_perm, Transaction *tran,
2211 Error **errp)
2213 if (!bs->drv) {
2214 return 0;
2217 if (bs->drv->bdrv_check_perm) {
2218 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2219 if (ret < 0) {
2220 return ret;
2224 if (tran) {
2225 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2228 return 0;
2231 typedef struct BdrvReplaceChildState {
2232 BdrvChild *child;
2233 BlockDriverState *old_bs;
2234 } BdrvReplaceChildState;
2236 static void bdrv_replace_child_commit(void *opaque)
2238 BdrvReplaceChildState *s = opaque;
2240 bdrv_unref(s->old_bs);
2243 static void bdrv_replace_child_abort(void *opaque)
2245 BdrvReplaceChildState *s = opaque;
2246 BlockDriverState *new_bs = s->child->bs;
2248 /* old_bs reference is transparently moved from @s to @s->child */
2249 bdrv_replace_child_noperm(s->child, s->old_bs);
2250 bdrv_unref(new_bs);
2253 static TransactionActionDrv bdrv_replace_child_drv = {
2254 .commit = bdrv_replace_child_commit,
2255 .abort = bdrv_replace_child_abort,
2256 .clean = g_free,
2260 * bdrv_replace_child_tran
2262 * Note: real unref of old_bs is done only on commit.
2264 * The function doesn't update permissions, caller is responsible for this.
2266 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2267 Transaction *tran)
2269 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2270 *s = (BdrvReplaceChildState) {
2271 .child = child,
2272 .old_bs = child->bs,
2274 tran_add(tran, &bdrv_replace_child_drv, s);
2276 if (new_bs) {
2277 bdrv_ref(new_bs);
2279 bdrv_replace_child_noperm(child, new_bs);
2280 /* old_bs reference is transparently moved from @child to @s */
2284 * Refresh permissions in @bs subtree. The function is intended to be called
2285 * after some graph modification that was done without permission update.
2287 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2288 Transaction *tran, Error **errp)
2290 BlockDriver *drv = bs->drv;
2291 BdrvChild *c;
2292 int ret;
2293 uint64_t cumulative_perms, cumulative_shared_perms;
2295 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2297 /* Write permissions never work with read-only images */
2298 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2299 !bdrv_is_writable_after_reopen(bs, q))
2301 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2302 error_setg(errp, "Block node is read-only");
2303 } else {
2304 error_setg(errp, "Read-only block node '%s' cannot support "
2305 "read-write users", bdrv_get_node_name(bs));
2308 return -EPERM;
2312 * Unaligned requests will automatically be aligned to bl.request_alignment
2313 * and without RESIZE we can't extend requests to write to space beyond the
2314 * end of the image, so it's required that the image size is aligned.
2316 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2317 !(cumulative_perms & BLK_PERM_RESIZE))
2319 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2320 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2321 "Image size is not a multiple of request "
2322 "alignment");
2323 return -EPERM;
2327 /* Check this node */
2328 if (!drv) {
2329 return 0;
2332 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2333 errp);
2334 if (ret < 0) {
2335 return ret;
2338 /* Drivers that never have children can omit .bdrv_child_perm() */
2339 if (!drv->bdrv_child_perm) {
2340 assert(QLIST_EMPTY(&bs->children));
2341 return 0;
2344 /* Check all children */
2345 QLIST_FOREACH(c, &bs->children, next) {
2346 uint64_t cur_perm, cur_shared;
2348 bdrv_child_perm(bs, c->bs, c, c->role, q,
2349 cumulative_perms, cumulative_shared_perms,
2350 &cur_perm, &cur_shared);
2351 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2354 return 0;
2357 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2358 Transaction *tran, Error **errp)
2360 int ret;
2361 BlockDriverState *bs;
2363 for ( ; list; list = list->next) {
2364 bs = list->data;
2366 if (bdrv_parent_perms_conflict(bs, errp)) {
2367 return -EINVAL;
2370 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2371 if (ret < 0) {
2372 return ret;
2376 return 0;
2379 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2380 uint64_t *shared_perm)
2382 BdrvChild *c;
2383 uint64_t cumulative_perms = 0;
2384 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2386 QLIST_FOREACH(c, &bs->parents, next_parent) {
2387 cumulative_perms |= c->perm;
2388 cumulative_shared_perms &= c->shared_perm;
2391 *perm = cumulative_perms;
2392 *shared_perm = cumulative_shared_perms;
2395 char *bdrv_perm_names(uint64_t perm)
2397 struct perm_name {
2398 uint64_t perm;
2399 const char *name;
2400 } permissions[] = {
2401 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2402 { BLK_PERM_WRITE, "write" },
2403 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2404 { BLK_PERM_RESIZE, "resize" },
2405 { BLK_PERM_GRAPH_MOD, "change children" },
2406 { 0, NULL }
2409 GString *result = g_string_sized_new(30);
2410 struct perm_name *p;
2412 for (p = permissions; p->name; p++) {
2413 if (perm & p->perm) {
2414 if (result->len > 0) {
2415 g_string_append(result, ", ");
2417 g_string_append(result, p->name);
2421 return g_string_free(result, FALSE);
2425 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2427 int ret;
2428 Transaction *tran = tran_new();
2429 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2431 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2432 tran_finalize(tran, ret);
2434 return ret;
2437 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2438 Error **errp)
2440 Error *local_err = NULL;
2441 Transaction *tran = tran_new();
2442 int ret;
2444 bdrv_child_set_perm(c, perm, shared, tran);
2446 ret = bdrv_refresh_perms(c->bs, &local_err);
2448 tran_finalize(tran, ret);
2450 if (ret < 0) {
2451 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2452 /* tighten permissions */
2453 error_propagate(errp, local_err);
2454 } else {
2456 * Our caller may intend to only loosen restrictions and
2457 * does not expect this function to fail. Errors are not
2458 * fatal in such a case, so we can just hide them from our
2459 * caller.
2461 error_free(local_err);
2462 ret = 0;
2466 return ret;
2469 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2471 uint64_t parent_perms, parent_shared;
2472 uint64_t perms, shared;
2474 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2475 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2476 parent_perms, parent_shared, &perms, &shared);
2478 return bdrv_child_try_set_perm(c, perms, shared, errp);
2482 * Default implementation for .bdrv_child_perm() for block filters:
2483 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2484 * filtered child.
2486 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2487 BdrvChildRole role,
2488 BlockReopenQueue *reopen_queue,
2489 uint64_t perm, uint64_t shared,
2490 uint64_t *nperm, uint64_t *nshared)
2492 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2493 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2496 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2497 BdrvChildRole role,
2498 BlockReopenQueue *reopen_queue,
2499 uint64_t perm, uint64_t shared,
2500 uint64_t *nperm, uint64_t *nshared)
2502 assert(role & BDRV_CHILD_COW);
2505 * We want consistent read from backing files if the parent needs it.
2506 * No other operations are performed on backing files.
2508 perm &= BLK_PERM_CONSISTENT_READ;
2511 * If the parent can deal with changing data, we're okay with a
2512 * writable and resizable backing file.
2513 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2515 if (shared & BLK_PERM_WRITE) {
2516 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2517 } else {
2518 shared = 0;
2521 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2522 BLK_PERM_WRITE_UNCHANGED;
2524 if (bs->open_flags & BDRV_O_INACTIVE) {
2525 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2528 *nperm = perm;
2529 *nshared = shared;
2532 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2533 BdrvChildRole role,
2534 BlockReopenQueue *reopen_queue,
2535 uint64_t perm, uint64_t shared,
2536 uint64_t *nperm, uint64_t *nshared)
2538 int flags;
2540 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2542 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2545 * Apart from the modifications below, the same permissions are
2546 * forwarded and left alone as for filters
2548 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2549 perm, shared, &perm, &shared);
2551 if (role & BDRV_CHILD_METADATA) {
2552 /* Format drivers may touch metadata even if the guest doesn't write */
2553 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2554 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2558 * bs->file always needs to be consistent because of the
2559 * metadata. We can never allow other users to resize or write
2560 * to it.
2562 if (!(flags & BDRV_O_NO_IO)) {
2563 perm |= BLK_PERM_CONSISTENT_READ;
2565 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2568 if (role & BDRV_CHILD_DATA) {
2570 * Technically, everything in this block is a subset of the
2571 * BDRV_CHILD_METADATA path taken above, and so this could
2572 * be an "else if" branch. However, that is not obvious, and
2573 * this function is not performance critical, therefore we let
2574 * this be an independent "if".
2578 * We cannot allow other users to resize the file because the
2579 * format driver might have some assumptions about the size
2580 * (e.g. because it is stored in metadata, or because the file
2581 * is split into fixed-size data files).
2583 shared &= ~BLK_PERM_RESIZE;
2586 * WRITE_UNCHANGED often cannot be performed as such on the
2587 * data file. For example, the qcow2 driver may still need to
2588 * write copied clusters on copy-on-read.
2590 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2591 perm |= BLK_PERM_WRITE;
2595 * If the data file is written to, the format driver may
2596 * expect to be able to resize it by writing beyond the EOF.
2598 if (perm & BLK_PERM_WRITE) {
2599 perm |= BLK_PERM_RESIZE;
2603 if (bs->open_flags & BDRV_O_INACTIVE) {
2604 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2607 *nperm = perm;
2608 *nshared = shared;
2611 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2612 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2613 uint64_t perm, uint64_t shared,
2614 uint64_t *nperm, uint64_t *nshared)
2616 if (role & BDRV_CHILD_FILTERED) {
2617 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2618 BDRV_CHILD_COW)));
2619 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2620 perm, shared, nperm, nshared);
2621 } else if (role & BDRV_CHILD_COW) {
2622 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2623 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2624 perm, shared, nperm, nshared);
2625 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2626 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2627 perm, shared, nperm, nshared);
2628 } else {
2629 g_assert_not_reached();
2633 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2635 static const uint64_t permissions[] = {
2636 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2637 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2638 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2639 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2640 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2643 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2644 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2646 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2648 return permissions[qapi_perm];
2651 static void bdrv_replace_child_noperm(BdrvChild *child,
2652 BlockDriverState *new_bs)
2654 BlockDriverState *old_bs = child->bs;
2655 int new_bs_quiesce_counter;
2656 int drain_saldo;
2658 assert(!child->frozen);
2660 if (old_bs && new_bs) {
2661 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2664 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2665 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2668 * If the new child node is drained but the old one was not, flush
2669 * all outstanding requests to the old child node.
2671 while (drain_saldo > 0 && child->klass->drained_begin) {
2672 bdrv_parent_drained_begin_single(child, true);
2673 drain_saldo--;
2676 if (old_bs) {
2677 /* Detach first so that the recursive drain sections coming from @child
2678 * are already gone and we only end the drain sections that came from
2679 * elsewhere. */
2680 if (child->klass->detach) {
2681 child->klass->detach(child);
2683 QLIST_REMOVE(child, next_parent);
2686 child->bs = new_bs;
2688 if (new_bs) {
2689 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2692 * Detaching the old node may have led to the new node's
2693 * quiesce_counter having been decreased. Not a problem, we
2694 * just need to recognize this here and then invoke
2695 * drained_end appropriately more often.
2697 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2698 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2700 /* Attach only after starting new drained sections, so that recursive
2701 * drain sections coming from @child don't get an extra .drained_begin
2702 * callback. */
2703 if (child->klass->attach) {
2704 child->klass->attach(child);
2709 * If the old child node was drained but the new one is not, allow
2710 * requests to come in only after the new node has been attached.
2712 while (drain_saldo < 0 && child->klass->drained_end) {
2713 bdrv_parent_drained_end_single(child);
2714 drain_saldo++;
2718 static void bdrv_child_free(void *opaque)
2720 BdrvChild *c = opaque;
2722 g_free(c->name);
2723 g_free(c);
2726 static void bdrv_remove_empty_child(BdrvChild *child)
2728 assert(!child->bs);
2729 QLIST_SAFE_REMOVE(child, next);
2730 bdrv_child_free(child);
2733 typedef struct BdrvAttachChildCommonState {
2734 BdrvChild **child;
2735 AioContext *old_parent_ctx;
2736 AioContext *old_child_ctx;
2737 } BdrvAttachChildCommonState;
2739 static void bdrv_attach_child_common_abort(void *opaque)
2741 BdrvAttachChildCommonState *s = opaque;
2742 BdrvChild *child = *s->child;
2743 BlockDriverState *bs = child->bs;
2745 bdrv_replace_child_noperm(child, NULL);
2747 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2748 bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort);
2751 if (bdrv_child_get_parent_aio_context(child) != s->old_parent_ctx) {
2752 GSList *ignore = g_slist_prepend(NULL, child);
2754 child->klass->can_set_aio_ctx(child, s->old_parent_ctx, &ignore,
2755 &error_abort);
2756 g_slist_free(ignore);
2757 ignore = g_slist_prepend(NULL, child);
2758 child->klass->set_aio_ctx(child, s->old_parent_ctx, &ignore);
2760 g_slist_free(ignore);
2763 bdrv_unref(bs);
2764 bdrv_remove_empty_child(child);
2765 *s->child = NULL;
2768 static TransactionActionDrv bdrv_attach_child_common_drv = {
2769 .abort = bdrv_attach_child_common_abort,
2770 .clean = g_free,
2774 * Common part of attaching bdrv child to bs or to blk or to job
2776 * Resulting new child is returned through @child.
2777 * At start *@child must be NULL.
2778 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2779 * NULL on abort(). So referenced variable must live at least until transaction
2780 * end.
2782 * Function doesn't update permissions, caller is responsible for this.
2784 static int bdrv_attach_child_common(BlockDriverState *child_bs,
2785 const char *child_name,
2786 const BdrvChildClass *child_class,
2787 BdrvChildRole child_role,
2788 uint64_t perm, uint64_t shared_perm,
2789 void *opaque, BdrvChild **child,
2790 Transaction *tran, Error **errp)
2792 BdrvChild *new_child;
2793 AioContext *parent_ctx;
2794 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2796 assert(child);
2797 assert(*child == NULL);
2798 assert(child_class->get_parent_desc);
2800 new_child = g_new(BdrvChild, 1);
2801 *new_child = (BdrvChild) {
2802 .bs = NULL,
2803 .name = g_strdup(child_name),
2804 .klass = child_class,
2805 .role = child_role,
2806 .perm = perm,
2807 .shared_perm = shared_perm,
2808 .opaque = opaque,
2812 * If the AioContexts don't match, first try to move the subtree of
2813 * child_bs into the AioContext of the new parent. If this doesn't work,
2814 * try moving the parent into the AioContext of child_bs instead.
2816 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
2817 if (child_ctx != parent_ctx) {
2818 Error *local_err = NULL;
2819 int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err);
2821 if (ret < 0 && child_class->can_set_aio_ctx) {
2822 GSList *ignore = g_slist_prepend(NULL, new_child);
2823 if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore,
2824 NULL))
2826 error_free(local_err);
2827 ret = 0;
2828 g_slist_free(ignore);
2829 ignore = g_slist_prepend(NULL, new_child);
2830 child_class->set_aio_ctx(new_child, child_ctx, &ignore);
2832 g_slist_free(ignore);
2835 if (ret < 0) {
2836 error_propagate(errp, local_err);
2837 bdrv_remove_empty_child(new_child);
2838 return ret;
2842 bdrv_ref(child_bs);
2843 bdrv_replace_child_noperm(new_child, child_bs);
2845 *child = new_child;
2847 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
2848 *s = (BdrvAttachChildCommonState) {
2849 .child = child,
2850 .old_parent_ctx = parent_ctx,
2851 .old_child_ctx = child_ctx,
2853 tran_add(tran, &bdrv_attach_child_common_drv, s);
2855 return 0;
2859 * Variable referenced by @child must live at least until transaction end.
2860 * (see bdrv_attach_child_common() doc for details)
2862 * Function doesn't update permissions, caller is responsible for this.
2864 static int bdrv_attach_child_noperm(BlockDriverState *parent_bs,
2865 BlockDriverState *child_bs,
2866 const char *child_name,
2867 const BdrvChildClass *child_class,
2868 BdrvChildRole child_role,
2869 BdrvChild **child,
2870 Transaction *tran,
2871 Error **errp)
2873 int ret;
2874 uint64_t perm, shared_perm;
2876 assert(parent_bs->drv);
2878 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2879 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2880 perm, shared_perm, &perm, &shared_perm);
2882 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
2883 child_role, perm, shared_perm, parent_bs,
2884 child, tran, errp);
2885 if (ret < 0) {
2886 return ret;
2889 QLIST_INSERT_HEAD(&parent_bs->children, *child, next);
2891 * child is removed in bdrv_attach_child_common_abort(), so don't care to
2892 * abort this change separately.
2895 return 0;
2898 static void bdrv_detach_child(BdrvChild *child)
2900 BlockDriverState *old_bs = child->bs;
2902 bdrv_replace_child_noperm(child, NULL);
2903 bdrv_remove_empty_child(child);
2905 if (old_bs) {
2907 * Update permissions for old node. We're just taking a parent away, so
2908 * we're loosening restrictions. Errors of permission update are not
2909 * fatal in this case, ignore them.
2911 bdrv_refresh_perms(old_bs, NULL);
2914 * When the parent requiring a non-default AioContext is removed, the
2915 * node moves back to the main AioContext
2917 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2922 * This function steals the reference to child_bs from the caller.
2923 * That reference is later dropped by bdrv_root_unref_child().
2925 * On failure NULL is returned, errp is set and the reference to
2926 * child_bs is also dropped.
2928 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2929 * (unless @child_bs is already in @ctx).
2931 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2932 const char *child_name,
2933 const BdrvChildClass *child_class,
2934 BdrvChildRole child_role,
2935 uint64_t perm, uint64_t shared_perm,
2936 void *opaque, Error **errp)
2938 int ret;
2939 BdrvChild *child = NULL;
2940 Transaction *tran = tran_new();
2942 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
2943 child_role, perm, shared_perm, opaque,
2944 &child, tran, errp);
2945 if (ret < 0) {
2946 goto out;
2949 ret = bdrv_refresh_perms(child_bs, errp);
2951 out:
2952 tran_finalize(tran, ret);
2953 /* child is unset on failure by bdrv_attach_child_common_abort() */
2954 assert((ret < 0) == !child);
2956 bdrv_unref(child_bs);
2957 return child;
2961 * This function transfers the reference to child_bs from the caller
2962 * to parent_bs. That reference is later dropped by parent_bs on
2963 * bdrv_close() or if someone calls bdrv_unref_child().
2965 * On failure NULL is returned, errp is set and the reference to
2966 * child_bs is also dropped.
2968 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2969 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2971 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2972 BlockDriverState *child_bs,
2973 const char *child_name,
2974 const BdrvChildClass *child_class,
2975 BdrvChildRole child_role,
2976 Error **errp)
2978 int ret;
2979 BdrvChild *child = NULL;
2980 Transaction *tran = tran_new();
2982 ret = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, child_class,
2983 child_role, &child, tran, errp);
2984 if (ret < 0) {
2985 goto out;
2988 ret = bdrv_refresh_perms(parent_bs, errp);
2989 if (ret < 0) {
2990 goto out;
2993 out:
2994 tran_finalize(tran, ret);
2995 /* child is unset on failure by bdrv_attach_child_common_abort() */
2996 assert((ret < 0) == !child);
2998 bdrv_unref(child_bs);
3000 return child;
3003 /* Callers must ensure that child->frozen is false. */
3004 void bdrv_root_unref_child(BdrvChild *child)
3006 BlockDriverState *child_bs;
3008 child_bs = child->bs;
3009 bdrv_detach_child(child);
3010 bdrv_unref(child_bs);
3013 typedef struct BdrvSetInheritsFrom {
3014 BlockDriverState *bs;
3015 BlockDriverState *old_inherits_from;
3016 } BdrvSetInheritsFrom;
3018 static void bdrv_set_inherits_from_abort(void *opaque)
3020 BdrvSetInheritsFrom *s = opaque;
3022 s->bs->inherits_from = s->old_inherits_from;
3025 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3026 .abort = bdrv_set_inherits_from_abort,
3027 .clean = g_free,
3030 /* @tran is allowed to be NULL. In this case no rollback is possible */
3031 static void bdrv_set_inherits_from(BlockDriverState *bs,
3032 BlockDriverState *new_inherits_from,
3033 Transaction *tran)
3035 if (tran) {
3036 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3038 *s = (BdrvSetInheritsFrom) {
3039 .bs = bs,
3040 .old_inherits_from = bs->inherits_from,
3043 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3046 bs->inherits_from = new_inherits_from;
3050 * Clear all inherits_from pointers from children and grandchildren of
3051 * @root that point to @root, where necessary.
3052 * @tran is allowed to be NULL. In this case no rollback is possible
3054 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3055 Transaction *tran)
3057 BdrvChild *c;
3059 if (child->bs->inherits_from == root) {
3061 * Remove inherits_from only when the last reference between root and
3062 * child->bs goes away.
3064 QLIST_FOREACH(c, &root->children, next) {
3065 if (c != child && c->bs == child->bs) {
3066 break;
3069 if (c == NULL) {
3070 bdrv_set_inherits_from(child->bs, NULL, tran);
3074 QLIST_FOREACH(c, &child->bs->children, next) {
3075 bdrv_unset_inherits_from(root, c, tran);
3079 /* Callers must ensure that child->frozen is false. */
3080 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3082 if (child == NULL) {
3083 return;
3086 bdrv_unset_inherits_from(parent, child, NULL);
3087 bdrv_root_unref_child(child);
3091 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3093 BdrvChild *c;
3094 QLIST_FOREACH(c, &bs->parents, next_parent) {
3095 if (c->klass->change_media) {
3096 c->klass->change_media(c, load);
3101 /* Return true if you can reach parent going through child->inherits_from
3102 * recursively. If parent or child are NULL, return false */
3103 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3104 BlockDriverState *parent)
3106 while (child && child != parent) {
3107 child = child->inherits_from;
3110 return child != NULL;
3114 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3115 * mostly used for COW backing children (role = COW), but also for
3116 * filtered children (role = FILTERED | PRIMARY).
3118 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3120 if (bs->drv && bs->drv->is_filter) {
3121 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3122 } else {
3123 return BDRV_CHILD_COW;
3128 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3129 * callers which don't need their own reference any more must call bdrv_unref().
3131 * Function doesn't update permissions, caller is responsible for this.
3133 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3134 BlockDriverState *child_bs,
3135 bool is_backing,
3136 Transaction *tran, Error **errp)
3138 int ret = 0;
3139 bool update_inherits_from =
3140 bdrv_inherits_from_recursive(child_bs, parent_bs);
3141 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3142 BdrvChildRole role;
3144 if (!parent_bs->drv) {
3146 * Node without drv is an object without a class :/. TODO: finally fix
3147 * qcow2 driver to never clear bs->drv and implement format corruption
3148 * handling in other way.
3150 error_setg(errp, "Node corrupted");
3151 return -EINVAL;
3154 if (child && child->frozen) {
3155 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3156 child->name, parent_bs->node_name, child->bs->node_name);
3157 return -EPERM;
3160 if (is_backing && !parent_bs->drv->is_filter &&
3161 !parent_bs->drv->supports_backing)
3163 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3164 "files", parent_bs->drv->format_name, parent_bs->node_name);
3165 return -EINVAL;
3168 if (parent_bs->drv->is_filter) {
3169 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3170 } else if (is_backing) {
3171 role = BDRV_CHILD_COW;
3172 } else {
3174 * We only can use same role as it is in existing child. We don't have
3175 * infrastructure to determine role of file child in generic way
3177 if (!child) {
3178 error_setg(errp, "Cannot set file child to format node without "
3179 "file child");
3180 return -EINVAL;
3182 role = child->role;
3185 if (child) {
3186 bdrv_unset_inherits_from(parent_bs, child, tran);
3187 bdrv_remove_file_or_backing_child(parent_bs, child, tran);
3190 if (!child_bs) {
3191 goto out;
3194 ret = bdrv_attach_child_noperm(parent_bs, child_bs,
3195 is_backing ? "backing" : "file",
3196 &child_of_bds, role,
3197 is_backing ? &parent_bs->backing :
3198 &parent_bs->file,
3199 tran, errp);
3200 if (ret < 0) {
3201 return ret;
3206 * If inherits_from pointed recursively to bs then let's update it to
3207 * point directly to bs (else it will become NULL).
3209 if (update_inherits_from) {
3210 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3213 out:
3214 bdrv_refresh_limits(parent_bs, tran, NULL);
3216 return 0;
3219 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3220 BlockDriverState *backing_hd,
3221 Transaction *tran, Error **errp)
3223 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3226 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3227 Error **errp)
3229 int ret;
3230 Transaction *tran = tran_new();
3232 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3233 if (ret < 0) {
3234 goto out;
3237 ret = bdrv_refresh_perms(bs, errp);
3238 out:
3239 tran_finalize(tran, ret);
3241 return ret;
3245 * Opens the backing file for a BlockDriverState if not yet open
3247 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3248 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3249 * itself, all options starting with "${bdref_key}." are considered part of the
3250 * BlockdevRef.
3252 * TODO Can this be unified with bdrv_open_image()?
3254 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3255 const char *bdref_key, Error **errp)
3257 char *backing_filename = NULL;
3258 char *bdref_key_dot;
3259 const char *reference = NULL;
3260 int ret = 0;
3261 bool implicit_backing = false;
3262 BlockDriverState *backing_hd;
3263 QDict *options;
3264 QDict *tmp_parent_options = NULL;
3265 Error *local_err = NULL;
3267 if (bs->backing != NULL) {
3268 goto free_exit;
3271 /* NULL means an empty set of options */
3272 if (parent_options == NULL) {
3273 tmp_parent_options = qdict_new();
3274 parent_options = tmp_parent_options;
3277 bs->open_flags &= ~BDRV_O_NO_BACKING;
3279 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3280 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3281 g_free(bdref_key_dot);
3284 * Caution: while qdict_get_try_str() is fine, getting non-string
3285 * types would require more care. When @parent_options come from
3286 * -blockdev or blockdev_add, its members are typed according to
3287 * the QAPI schema, but when they come from -drive, they're all
3288 * QString.
3290 reference = qdict_get_try_str(parent_options, bdref_key);
3291 if (reference || qdict_haskey(options, "file.filename")) {
3292 /* keep backing_filename NULL */
3293 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3294 qobject_unref(options);
3295 goto free_exit;
3296 } else {
3297 if (qdict_size(options) == 0) {
3298 /* If the user specifies options that do not modify the
3299 * backing file's behavior, we might still consider it the
3300 * implicit backing file. But it's easier this way, and
3301 * just specifying some of the backing BDS's options is
3302 * only possible with -drive anyway (otherwise the QAPI
3303 * schema forces the user to specify everything). */
3304 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3307 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3308 if (local_err) {
3309 ret = -EINVAL;
3310 error_propagate(errp, local_err);
3311 qobject_unref(options);
3312 goto free_exit;
3316 if (!bs->drv || !bs->drv->supports_backing) {
3317 ret = -EINVAL;
3318 error_setg(errp, "Driver doesn't support backing files");
3319 qobject_unref(options);
3320 goto free_exit;
3323 if (!reference &&
3324 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3325 qdict_put_str(options, "driver", bs->backing_format);
3328 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3329 &child_of_bds, bdrv_backing_role(bs), errp);
3330 if (!backing_hd) {
3331 bs->open_flags |= BDRV_O_NO_BACKING;
3332 error_prepend(errp, "Could not open backing file: ");
3333 ret = -EINVAL;
3334 goto free_exit;
3337 if (implicit_backing) {
3338 bdrv_refresh_filename(backing_hd);
3339 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3340 backing_hd->filename);
3343 /* Hook up the backing file link; drop our reference, bs owns the
3344 * backing_hd reference now */
3345 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3346 bdrv_unref(backing_hd);
3347 if (ret < 0) {
3348 goto free_exit;
3351 qdict_del(parent_options, bdref_key);
3353 free_exit:
3354 g_free(backing_filename);
3355 qobject_unref(tmp_parent_options);
3356 return ret;
3359 static BlockDriverState *
3360 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3361 BlockDriverState *parent, const BdrvChildClass *child_class,
3362 BdrvChildRole child_role, bool allow_none, Error **errp)
3364 BlockDriverState *bs = NULL;
3365 QDict *image_options;
3366 char *bdref_key_dot;
3367 const char *reference;
3369 assert(child_class != NULL);
3371 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3372 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3373 g_free(bdref_key_dot);
3376 * Caution: while qdict_get_try_str() is fine, getting non-string
3377 * types would require more care. When @options come from
3378 * -blockdev or blockdev_add, its members are typed according to
3379 * the QAPI schema, but when they come from -drive, they're all
3380 * QString.
3382 reference = qdict_get_try_str(options, bdref_key);
3383 if (!filename && !reference && !qdict_size(image_options)) {
3384 if (!allow_none) {
3385 error_setg(errp, "A block device must be specified for \"%s\"",
3386 bdref_key);
3388 qobject_unref(image_options);
3389 goto done;
3392 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3393 parent, child_class, child_role, errp);
3394 if (!bs) {
3395 goto done;
3398 done:
3399 qdict_del(options, bdref_key);
3400 return bs;
3404 * Opens a disk image whose options are given as BlockdevRef in another block
3405 * device's options.
3407 * If allow_none is true, no image will be opened if filename is false and no
3408 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3410 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3411 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3412 * itself, all options starting with "${bdref_key}." are considered part of the
3413 * BlockdevRef.
3415 * The BlockdevRef will be removed from the options QDict.
3417 BdrvChild *bdrv_open_child(const char *filename,
3418 QDict *options, const char *bdref_key,
3419 BlockDriverState *parent,
3420 const BdrvChildClass *child_class,
3421 BdrvChildRole child_role,
3422 bool allow_none, Error **errp)
3424 BlockDriverState *bs;
3426 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3427 child_role, allow_none, errp);
3428 if (bs == NULL) {
3429 return NULL;
3432 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3433 errp);
3437 * TODO Future callers may need to specify parent/child_class in order for
3438 * option inheritance to work. Existing callers use it for the root node.
3440 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3442 BlockDriverState *bs = NULL;
3443 QObject *obj = NULL;
3444 QDict *qdict = NULL;
3445 const char *reference = NULL;
3446 Visitor *v = NULL;
3448 if (ref->type == QTYPE_QSTRING) {
3449 reference = ref->u.reference;
3450 } else {
3451 BlockdevOptions *options = &ref->u.definition;
3452 assert(ref->type == QTYPE_QDICT);
3454 v = qobject_output_visitor_new(&obj);
3455 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3456 visit_complete(v, &obj);
3458 qdict = qobject_to(QDict, obj);
3459 qdict_flatten(qdict);
3461 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3462 * compatibility with other callers) rather than what we want as the
3463 * real defaults. Apply the defaults here instead. */
3464 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3465 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3466 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3467 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3471 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3472 obj = NULL;
3473 qobject_unref(obj);
3474 visit_free(v);
3475 return bs;
3478 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3479 int flags,
3480 QDict *snapshot_options,
3481 Error **errp)
3483 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3484 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3485 int64_t total_size;
3486 QemuOpts *opts = NULL;
3487 BlockDriverState *bs_snapshot = NULL;
3488 int ret;
3490 /* if snapshot, we create a temporary backing file and open it
3491 instead of opening 'filename' directly */
3493 /* Get the required size from the image */
3494 total_size = bdrv_getlength(bs);
3495 if (total_size < 0) {
3496 error_setg_errno(errp, -total_size, "Could not get image size");
3497 goto out;
3500 /* Create the temporary image */
3501 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3502 if (ret < 0) {
3503 error_setg_errno(errp, -ret, "Could not get temporary filename");
3504 goto out;
3507 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3508 &error_abort);
3509 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3510 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3511 qemu_opts_del(opts);
3512 if (ret < 0) {
3513 error_prepend(errp, "Could not create temporary overlay '%s': ",
3514 tmp_filename);
3515 goto out;
3518 /* Prepare options QDict for the temporary file */
3519 qdict_put_str(snapshot_options, "file.driver", "file");
3520 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3521 qdict_put_str(snapshot_options, "driver", "qcow2");
3523 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3524 snapshot_options = NULL;
3525 if (!bs_snapshot) {
3526 goto out;
3529 ret = bdrv_append(bs_snapshot, bs, errp);
3530 if (ret < 0) {
3531 bs_snapshot = NULL;
3532 goto out;
3535 out:
3536 qobject_unref(snapshot_options);
3537 g_free(tmp_filename);
3538 return bs_snapshot;
3542 * Opens a disk image (raw, qcow2, vmdk, ...)
3544 * options is a QDict of options to pass to the block drivers, or NULL for an
3545 * empty set of options. The reference to the QDict belongs to the block layer
3546 * after the call (even on failure), so if the caller intends to reuse the
3547 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3549 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3550 * If it is not NULL, the referenced BDS will be reused.
3552 * The reference parameter may be used to specify an existing block device which
3553 * should be opened. If specified, neither options nor a filename may be given,
3554 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3556 static BlockDriverState *bdrv_open_inherit(const char *filename,
3557 const char *reference,
3558 QDict *options, int flags,
3559 BlockDriverState *parent,
3560 const BdrvChildClass *child_class,
3561 BdrvChildRole child_role,
3562 Error **errp)
3564 int ret;
3565 BlockBackend *file = NULL;
3566 BlockDriverState *bs;
3567 BlockDriver *drv = NULL;
3568 BdrvChild *child;
3569 const char *drvname;
3570 const char *backing;
3571 Error *local_err = NULL;
3572 QDict *snapshot_options = NULL;
3573 int snapshot_flags = 0;
3575 assert(!child_class || !flags);
3576 assert(!child_class == !parent);
3578 if (reference) {
3579 bool options_non_empty = options ? qdict_size(options) : false;
3580 qobject_unref(options);
3582 if (filename || options_non_empty) {
3583 error_setg(errp, "Cannot reference an existing block device with "
3584 "additional options or a new filename");
3585 return NULL;
3588 bs = bdrv_lookup_bs(reference, reference, errp);
3589 if (!bs) {
3590 return NULL;
3593 bdrv_ref(bs);
3594 return bs;
3597 bs = bdrv_new();
3599 /* NULL means an empty set of options */
3600 if (options == NULL) {
3601 options = qdict_new();
3604 /* json: syntax counts as explicit options, as if in the QDict */
3605 parse_json_protocol(options, &filename, &local_err);
3606 if (local_err) {
3607 goto fail;
3610 bs->explicit_options = qdict_clone_shallow(options);
3612 if (child_class) {
3613 bool parent_is_format;
3615 if (parent->drv) {
3616 parent_is_format = parent->drv->is_format;
3617 } else {
3619 * parent->drv is not set yet because this node is opened for
3620 * (potential) format probing. That means that @parent is going
3621 * to be a format node.
3623 parent_is_format = true;
3626 bs->inherits_from = parent;
3627 child_class->inherit_options(child_role, parent_is_format,
3628 &flags, options,
3629 parent->open_flags, parent->options);
3632 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3633 if (ret < 0) {
3634 goto fail;
3638 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3639 * Caution: getting a boolean member of @options requires care.
3640 * When @options come from -blockdev or blockdev_add, members are
3641 * typed according to the QAPI schema, but when they come from
3642 * -drive, they're all QString.
3644 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3645 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3646 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3647 } else {
3648 flags &= ~BDRV_O_RDWR;
3651 if (flags & BDRV_O_SNAPSHOT) {
3652 snapshot_options = qdict_new();
3653 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3654 flags, options);
3655 /* Let bdrv_backing_options() override "read-only" */
3656 qdict_del(options, BDRV_OPT_READ_ONLY);
3657 bdrv_inherited_options(BDRV_CHILD_COW, true,
3658 &flags, options, flags, options);
3661 bs->open_flags = flags;
3662 bs->options = options;
3663 options = qdict_clone_shallow(options);
3665 /* Find the right image format driver */
3666 /* See cautionary note on accessing @options above */
3667 drvname = qdict_get_try_str(options, "driver");
3668 if (drvname) {
3669 drv = bdrv_find_format(drvname);
3670 if (!drv) {
3671 error_setg(errp, "Unknown driver: '%s'", drvname);
3672 goto fail;
3676 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3678 /* See cautionary note on accessing @options above */
3679 backing = qdict_get_try_str(options, "backing");
3680 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3681 (backing && *backing == '\0'))
3683 if (backing) {
3684 warn_report("Use of \"backing\": \"\" is deprecated; "
3685 "use \"backing\": null instead");
3687 flags |= BDRV_O_NO_BACKING;
3688 qdict_del(bs->explicit_options, "backing");
3689 qdict_del(bs->options, "backing");
3690 qdict_del(options, "backing");
3693 /* Open image file without format layer. This BlockBackend is only used for
3694 * probing, the block drivers will do their own bdrv_open_child() for the
3695 * same BDS, which is why we put the node name back into options. */
3696 if ((flags & BDRV_O_PROTOCOL) == 0) {
3697 BlockDriverState *file_bs;
3699 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3700 &child_of_bds, BDRV_CHILD_IMAGE,
3701 true, &local_err);
3702 if (local_err) {
3703 goto fail;
3705 if (file_bs != NULL) {
3706 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3707 * looking at the header to guess the image format. This works even
3708 * in cases where a guest would not see a consistent state. */
3709 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3710 blk_insert_bs(file, file_bs, &local_err);
3711 bdrv_unref(file_bs);
3712 if (local_err) {
3713 goto fail;
3716 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3720 /* Image format probing */
3721 bs->probed = !drv;
3722 if (!drv && file) {
3723 ret = find_image_format(file, filename, &drv, &local_err);
3724 if (ret < 0) {
3725 goto fail;
3728 * This option update would logically belong in bdrv_fill_options(),
3729 * but we first need to open bs->file for the probing to work, while
3730 * opening bs->file already requires the (mostly) final set of options
3731 * so that cache mode etc. can be inherited.
3733 * Adding the driver later is somewhat ugly, but it's not an option
3734 * that would ever be inherited, so it's correct. We just need to make
3735 * sure to update both bs->options (which has the full effective
3736 * options for bs) and options (which has file.* already removed).
3738 qdict_put_str(bs->options, "driver", drv->format_name);
3739 qdict_put_str(options, "driver", drv->format_name);
3740 } else if (!drv) {
3741 error_setg(errp, "Must specify either driver or file");
3742 goto fail;
3745 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3746 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3747 /* file must be NULL if a protocol BDS is about to be created
3748 * (the inverse results in an error message from bdrv_open_common()) */
3749 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3751 /* Open the image */
3752 ret = bdrv_open_common(bs, file, options, &local_err);
3753 if (ret < 0) {
3754 goto fail;
3757 if (file) {
3758 blk_unref(file);
3759 file = NULL;
3762 /* If there is a backing file, use it */
3763 if ((flags & BDRV_O_NO_BACKING) == 0) {
3764 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3765 if (ret < 0) {
3766 goto close_and_fail;
3770 /* Remove all children options and references
3771 * from bs->options and bs->explicit_options */
3772 QLIST_FOREACH(child, &bs->children, next) {
3773 char *child_key_dot;
3774 child_key_dot = g_strdup_printf("%s.", child->name);
3775 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3776 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3777 qdict_del(bs->explicit_options, child->name);
3778 qdict_del(bs->options, child->name);
3779 g_free(child_key_dot);
3782 /* Check if any unknown options were used */
3783 if (qdict_size(options) != 0) {
3784 const QDictEntry *entry = qdict_first(options);
3785 if (flags & BDRV_O_PROTOCOL) {
3786 error_setg(errp, "Block protocol '%s' doesn't support the option "
3787 "'%s'", drv->format_name, entry->key);
3788 } else {
3789 error_setg(errp,
3790 "Block format '%s' does not support the option '%s'",
3791 drv->format_name, entry->key);
3794 goto close_and_fail;
3797 bdrv_parent_cb_change_media(bs, true);
3799 qobject_unref(options);
3800 options = NULL;
3802 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3803 * temporary snapshot afterwards. */
3804 if (snapshot_flags) {
3805 BlockDriverState *snapshot_bs;
3806 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3807 snapshot_options, &local_err);
3808 snapshot_options = NULL;
3809 if (local_err) {
3810 goto close_and_fail;
3812 /* We are not going to return bs but the overlay on top of it
3813 * (snapshot_bs); thus, we have to drop the strong reference to bs
3814 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3815 * though, because the overlay still has a reference to it. */
3816 bdrv_unref(bs);
3817 bs = snapshot_bs;
3820 return bs;
3822 fail:
3823 blk_unref(file);
3824 qobject_unref(snapshot_options);
3825 qobject_unref(bs->explicit_options);
3826 qobject_unref(bs->options);
3827 qobject_unref(options);
3828 bs->options = NULL;
3829 bs->explicit_options = NULL;
3830 bdrv_unref(bs);
3831 error_propagate(errp, local_err);
3832 return NULL;
3834 close_and_fail:
3835 bdrv_unref(bs);
3836 qobject_unref(snapshot_options);
3837 qobject_unref(options);
3838 error_propagate(errp, local_err);
3839 return NULL;
3842 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3843 QDict *options, int flags, Error **errp)
3845 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3846 NULL, 0, errp);
3849 /* Return true if the NULL-terminated @list contains @str */
3850 static bool is_str_in_list(const char *str, const char *const *list)
3852 if (str && list) {
3853 int i;
3854 for (i = 0; list[i] != NULL; i++) {
3855 if (!strcmp(str, list[i])) {
3856 return true;
3860 return false;
3864 * Check that every option set in @bs->options is also set in
3865 * @new_opts.
3867 * Options listed in the common_options list and in
3868 * @bs->drv->mutable_opts are skipped.
3870 * Return 0 on success, otherwise return -EINVAL and set @errp.
3872 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3873 const QDict *new_opts, Error **errp)
3875 const QDictEntry *e;
3876 /* These options are common to all block drivers and are handled
3877 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3878 const char *const common_options[] = {
3879 "node-name", "discard", "cache.direct", "cache.no-flush",
3880 "read-only", "auto-read-only", "detect-zeroes", NULL
3883 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3884 if (!qdict_haskey(new_opts, e->key) &&
3885 !is_str_in_list(e->key, common_options) &&
3886 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3887 error_setg(errp, "Option '%s' cannot be reset "
3888 "to its default value", e->key);
3889 return -EINVAL;
3893 return 0;
3897 * Returns true if @child can be reached recursively from @bs
3899 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3900 BlockDriverState *child)
3902 BdrvChild *c;
3904 if (bs == child) {
3905 return true;
3908 QLIST_FOREACH(c, &bs->children, next) {
3909 if (bdrv_recurse_has_child(c->bs, child)) {
3910 return true;
3914 return false;
3918 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3919 * reopen of multiple devices.
3921 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3922 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3923 * be created and initialized. This newly created BlockReopenQueue should be
3924 * passed back in for subsequent calls that are intended to be of the same
3925 * atomic 'set'.
3927 * bs is the BlockDriverState to add to the reopen queue.
3929 * options contains the changed options for the associated bs
3930 * (the BlockReopenQueue takes ownership)
3932 * flags contains the open flags for the associated bs
3934 * returns a pointer to bs_queue, which is either the newly allocated
3935 * bs_queue, or the existing bs_queue being used.
3937 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3939 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3940 BlockDriverState *bs,
3941 QDict *options,
3942 const BdrvChildClass *klass,
3943 BdrvChildRole role,
3944 bool parent_is_format,
3945 QDict *parent_options,
3946 int parent_flags,
3947 bool keep_old_opts)
3949 assert(bs != NULL);
3951 BlockReopenQueueEntry *bs_entry;
3952 BdrvChild *child;
3953 QDict *old_options, *explicit_options, *options_copy;
3954 int flags;
3955 QemuOpts *opts;
3957 /* Make sure that the caller remembered to use a drained section. This is
3958 * important to avoid graph changes between the recursive queuing here and
3959 * bdrv_reopen_multiple(). */
3960 assert(bs->quiesce_counter > 0);
3962 if (bs_queue == NULL) {
3963 bs_queue = g_new0(BlockReopenQueue, 1);
3964 QTAILQ_INIT(bs_queue);
3967 if (!options) {
3968 options = qdict_new();
3971 /* Check if this BlockDriverState is already in the queue */
3972 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3973 if (bs == bs_entry->state.bs) {
3974 break;
3979 * Precedence of options:
3980 * 1. Explicitly passed in options (highest)
3981 * 2. Retained from explicitly set options of bs
3982 * 3. Inherited from parent node
3983 * 4. Retained from effective options of bs
3986 /* Old explicitly set values (don't overwrite by inherited value) */
3987 if (bs_entry || keep_old_opts) {
3988 old_options = qdict_clone_shallow(bs_entry ?
3989 bs_entry->state.explicit_options :
3990 bs->explicit_options);
3991 bdrv_join_options(bs, options, old_options);
3992 qobject_unref(old_options);
3995 explicit_options = qdict_clone_shallow(options);
3997 /* Inherit from parent node */
3998 if (parent_options) {
3999 flags = 0;
4000 klass->inherit_options(role, parent_is_format, &flags, options,
4001 parent_flags, parent_options);
4002 } else {
4003 flags = bdrv_get_flags(bs);
4006 if (keep_old_opts) {
4007 /* Old values are used for options that aren't set yet */
4008 old_options = qdict_clone_shallow(bs->options);
4009 bdrv_join_options(bs, options, old_options);
4010 qobject_unref(old_options);
4013 /* We have the final set of options so let's update the flags */
4014 options_copy = qdict_clone_shallow(options);
4015 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4016 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4017 update_flags_from_options(&flags, opts);
4018 qemu_opts_del(opts);
4019 qobject_unref(options_copy);
4021 /* bdrv_open_inherit() sets and clears some additional flags internally */
4022 flags &= ~BDRV_O_PROTOCOL;
4023 if (flags & BDRV_O_RDWR) {
4024 flags |= BDRV_O_ALLOW_RDWR;
4027 if (!bs_entry) {
4028 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4029 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4030 } else {
4031 qobject_unref(bs_entry->state.options);
4032 qobject_unref(bs_entry->state.explicit_options);
4035 bs_entry->state.bs = bs;
4036 bs_entry->state.options = options;
4037 bs_entry->state.explicit_options = explicit_options;
4038 bs_entry->state.flags = flags;
4041 * If keep_old_opts is false then it means that unspecified
4042 * options must be reset to their original value. We don't allow
4043 * resetting 'backing' but we need to know if the option is
4044 * missing in order to decide if we have to return an error.
4046 if (!keep_old_opts) {
4047 bs_entry->state.backing_missing =
4048 !qdict_haskey(options, "backing") &&
4049 !qdict_haskey(options, "backing.driver");
4052 QLIST_FOREACH(child, &bs->children, next) {
4053 QDict *new_child_options = NULL;
4054 bool child_keep_old = keep_old_opts;
4056 /* reopen can only change the options of block devices that were
4057 * implicitly created and inherited options. For other (referenced)
4058 * block devices, a syntax like "backing.foo" results in an error. */
4059 if (child->bs->inherits_from != bs) {
4060 continue;
4063 /* Check if the options contain a child reference */
4064 if (qdict_haskey(options, child->name)) {
4065 const char *childref = qdict_get_try_str(options, child->name);
4067 * The current child must not be reopened if the child
4068 * reference is null or points to a different node.
4070 if (g_strcmp0(childref, child->bs->node_name)) {
4071 continue;
4074 * If the child reference points to the current child then
4075 * reopen it with its existing set of options (note that
4076 * it can still inherit new options from the parent).
4078 child_keep_old = true;
4079 } else {
4080 /* Extract child options ("child-name.*") */
4081 char *child_key_dot = g_strdup_printf("%s.", child->name);
4082 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4083 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4084 g_free(child_key_dot);
4087 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4088 child->klass, child->role, bs->drv->is_format,
4089 options, flags, child_keep_old);
4092 return bs_queue;
4095 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4096 BlockDriverState *bs,
4097 QDict *options, bool keep_old_opts)
4099 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4100 NULL, 0, keep_old_opts);
4103 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4105 if (bs_queue) {
4106 BlockReopenQueueEntry *bs_entry, *next;
4107 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4108 qobject_unref(bs_entry->state.explicit_options);
4109 qobject_unref(bs_entry->state.options);
4110 g_free(bs_entry);
4112 g_free(bs_queue);
4117 * Reopen multiple BlockDriverStates atomically & transactionally.
4119 * The queue passed in (bs_queue) must have been built up previous
4120 * via bdrv_reopen_queue().
4122 * Reopens all BDS specified in the queue, with the appropriate
4123 * flags. All devices are prepared for reopen, and failure of any
4124 * device will cause all device changes to be abandoned, and intermediate
4125 * data cleaned up.
4127 * If all devices prepare successfully, then the changes are committed
4128 * to all devices.
4130 * All affected nodes must be drained between bdrv_reopen_queue() and
4131 * bdrv_reopen_multiple().
4133 * To be called from the main thread, with all other AioContexts unlocked.
4135 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4137 int ret = -1;
4138 BlockReopenQueueEntry *bs_entry, *next;
4139 AioContext *ctx;
4140 Transaction *tran = tran_new();
4141 g_autoptr(GHashTable) found = NULL;
4142 g_autoptr(GSList) refresh_list = NULL;
4144 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4145 assert(bs_queue != NULL);
4147 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4148 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4149 aio_context_acquire(ctx);
4150 ret = bdrv_flush(bs_entry->state.bs);
4151 aio_context_release(ctx);
4152 if (ret < 0) {
4153 error_setg_errno(errp, -ret, "Error flushing drive");
4154 goto abort;
4158 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4159 assert(bs_entry->state.bs->quiesce_counter > 0);
4160 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4161 aio_context_acquire(ctx);
4162 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4163 aio_context_release(ctx);
4164 if (ret < 0) {
4165 goto abort;
4167 bs_entry->prepared = true;
4170 found = g_hash_table_new(NULL, NULL);
4171 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4172 BDRVReopenState *state = &bs_entry->state;
4174 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4175 if (state->old_backing_bs) {
4176 refresh_list = bdrv_topological_dfs(refresh_list, found,
4177 state->old_backing_bs);
4179 if (state->old_file_bs) {
4180 refresh_list = bdrv_topological_dfs(refresh_list, found,
4181 state->old_file_bs);
4186 * Note that file-posix driver rely on permission update done during reopen
4187 * (even if no permission changed), because it wants "new" permissions for
4188 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4189 * in raw_reopen_prepare() which is called with "old" permissions.
4191 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4192 if (ret < 0) {
4193 goto abort;
4197 * If we reach this point, we have success and just need to apply the
4198 * changes.
4200 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4201 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4202 * children are usually goes after parents in reopen-queue, so go from last
4203 * to first element.
4205 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4206 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4207 aio_context_acquire(ctx);
4208 bdrv_reopen_commit(&bs_entry->state);
4209 aio_context_release(ctx);
4212 tran_commit(tran);
4214 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4215 BlockDriverState *bs = bs_entry->state.bs;
4217 if (bs->drv->bdrv_reopen_commit_post) {
4218 ctx = bdrv_get_aio_context(bs);
4219 aio_context_acquire(ctx);
4220 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4221 aio_context_release(ctx);
4225 ret = 0;
4226 goto cleanup;
4228 abort:
4229 tran_abort(tran);
4230 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4231 if (bs_entry->prepared) {
4232 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4233 aio_context_acquire(ctx);
4234 bdrv_reopen_abort(&bs_entry->state);
4235 aio_context_release(ctx);
4239 cleanup:
4240 bdrv_reopen_queue_free(bs_queue);
4242 return ret;
4245 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4246 Error **errp)
4248 AioContext *ctx = bdrv_get_aio_context(bs);
4249 BlockReopenQueue *queue;
4250 int ret;
4252 bdrv_subtree_drained_begin(bs);
4253 if (ctx != qemu_get_aio_context()) {
4254 aio_context_release(ctx);
4257 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4258 ret = bdrv_reopen_multiple(queue, errp);
4260 if (ctx != qemu_get_aio_context()) {
4261 aio_context_acquire(ctx);
4263 bdrv_subtree_drained_end(bs);
4265 return ret;
4268 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4269 Error **errp)
4271 QDict *opts = qdict_new();
4273 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4275 return bdrv_reopen(bs, opts, true, errp);
4279 * Take a BDRVReopenState and check if the value of 'backing' in the
4280 * reopen_state->options QDict is valid or not.
4282 * If 'backing' is missing from the QDict then return 0.
4284 * If 'backing' contains the node name of the backing file of
4285 * reopen_state->bs then return 0.
4287 * If 'backing' contains a different node name (or is null) then check
4288 * whether the current backing file can be replaced with the new one.
4289 * If that's the case then reopen_state->replace_backing_bs is set to
4290 * true and reopen_state->new_backing_bs contains a pointer to the new
4291 * backing BlockDriverState (or NULL).
4293 * Return 0 on success, otherwise return < 0 and set @errp.
4295 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4296 bool is_backing, Transaction *tran,
4297 Error **errp)
4299 BlockDriverState *bs = reopen_state->bs;
4300 BlockDriverState *new_child_bs;
4301 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4302 child_bs(bs->file);
4303 const char *child_name = is_backing ? "backing" : "file";
4304 QObject *value;
4305 const char *str;
4307 value = qdict_get(reopen_state->options, child_name);
4308 if (value == NULL) {
4309 return 0;
4312 switch (qobject_type(value)) {
4313 case QTYPE_QNULL:
4314 assert(is_backing); /* The 'file' option does not allow a null value */
4315 new_child_bs = NULL;
4316 break;
4317 case QTYPE_QSTRING:
4318 str = qstring_get_str(qobject_to(QString, value));
4319 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4320 if (new_child_bs == NULL) {
4321 return -EINVAL;
4322 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4323 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4324 "cycle", str, child_name, bs->node_name);
4325 return -EINVAL;
4327 break;
4328 default:
4330 * The options QDict has been flattened, so 'backing' and 'file'
4331 * do not allow any other data type here.
4333 g_assert_not_reached();
4336 if (old_child_bs == new_child_bs) {
4337 return 0;
4340 if (old_child_bs) {
4341 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4342 return 0;
4345 if (old_child_bs->implicit) {
4346 error_setg(errp, "Cannot replace implicit %s child of %s",
4347 child_name, bs->node_name);
4348 return -EPERM;
4352 if (bs->drv->is_filter && !old_child_bs) {
4354 * Filters always have a file or a backing child, so we are trying to
4355 * change wrong child
4357 error_setg(errp, "'%s' is a %s filter node that does not support a "
4358 "%s child", bs->node_name, bs->drv->format_name, child_name);
4359 return -EINVAL;
4362 if (is_backing) {
4363 reopen_state->old_backing_bs = old_child_bs;
4364 } else {
4365 reopen_state->old_file_bs = old_child_bs;
4368 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4369 tran, errp);
4373 * Prepares a BlockDriverState for reopen. All changes are staged in the
4374 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4375 * the block driver layer .bdrv_reopen_prepare()
4377 * bs is the BlockDriverState to reopen
4378 * flags are the new open flags
4379 * queue is the reopen queue
4381 * Returns 0 on success, non-zero on error. On error errp will be set
4382 * as well.
4384 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4385 * It is the responsibility of the caller to then call the abort() or
4386 * commit() for any other BDS that have been left in a prepare() state
4389 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4390 BlockReopenQueue *queue,
4391 Transaction *change_child_tran, Error **errp)
4393 int ret = -1;
4394 int old_flags;
4395 Error *local_err = NULL;
4396 BlockDriver *drv;
4397 QemuOpts *opts;
4398 QDict *orig_reopen_opts;
4399 char *discard = NULL;
4400 bool read_only;
4401 bool drv_prepared = false;
4403 assert(reopen_state != NULL);
4404 assert(reopen_state->bs->drv != NULL);
4405 drv = reopen_state->bs->drv;
4407 /* This function and each driver's bdrv_reopen_prepare() remove
4408 * entries from reopen_state->options as they are processed, so
4409 * we need to make a copy of the original QDict. */
4410 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4412 /* Process generic block layer options */
4413 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4414 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4415 ret = -EINVAL;
4416 goto error;
4419 /* This was already called in bdrv_reopen_queue_child() so the flags
4420 * are up-to-date. This time we simply want to remove the options from
4421 * QemuOpts in order to indicate that they have been processed. */
4422 old_flags = reopen_state->flags;
4423 update_flags_from_options(&reopen_state->flags, opts);
4424 assert(old_flags == reopen_state->flags);
4426 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4427 if (discard != NULL) {
4428 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4429 error_setg(errp, "Invalid discard option");
4430 ret = -EINVAL;
4431 goto error;
4435 reopen_state->detect_zeroes =
4436 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4437 if (local_err) {
4438 error_propagate(errp, local_err);
4439 ret = -EINVAL;
4440 goto error;
4443 /* All other options (including node-name and driver) must be unchanged.
4444 * Put them back into the QDict, so that they are checked at the end
4445 * of this function. */
4446 qemu_opts_to_qdict(opts, reopen_state->options);
4448 /* If we are to stay read-only, do not allow permission change
4449 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4450 * not set, or if the BDS still has copy_on_read enabled */
4451 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4452 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4453 if (local_err) {
4454 error_propagate(errp, local_err);
4455 goto error;
4458 if (drv->bdrv_reopen_prepare) {
4460 * If a driver-specific option is missing, it means that we
4461 * should reset it to its default value.
4462 * But not all options allow that, so we need to check it first.
4464 ret = bdrv_reset_options_allowed(reopen_state->bs,
4465 reopen_state->options, errp);
4466 if (ret) {
4467 goto error;
4470 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4471 if (ret) {
4472 if (local_err != NULL) {
4473 error_propagate(errp, local_err);
4474 } else {
4475 bdrv_refresh_filename(reopen_state->bs);
4476 error_setg(errp, "failed while preparing to reopen image '%s'",
4477 reopen_state->bs->filename);
4479 goto error;
4481 } else {
4482 /* It is currently mandatory to have a bdrv_reopen_prepare()
4483 * handler for each supported drv. */
4484 error_setg(errp, "Block format '%s' used by node '%s' "
4485 "does not support reopening files", drv->format_name,
4486 bdrv_get_device_or_node_name(reopen_state->bs));
4487 ret = -1;
4488 goto error;
4491 drv_prepared = true;
4494 * We must provide the 'backing' option if the BDS has a backing
4495 * file or if the image file has a backing file name as part of
4496 * its metadata. Otherwise the 'backing' option can be omitted.
4498 if (drv->supports_backing && reopen_state->backing_missing &&
4499 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4500 error_setg(errp, "backing is missing for '%s'",
4501 reopen_state->bs->node_name);
4502 ret = -EINVAL;
4503 goto error;
4507 * Allow changing the 'backing' option. The new value can be
4508 * either a reference to an existing node (using its node name)
4509 * or NULL to simply detach the current backing file.
4511 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4512 change_child_tran, errp);
4513 if (ret < 0) {
4514 goto error;
4516 qdict_del(reopen_state->options, "backing");
4518 /* Allow changing the 'file' option. In this case NULL is not allowed */
4519 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4520 change_child_tran, errp);
4521 if (ret < 0) {
4522 goto error;
4524 qdict_del(reopen_state->options, "file");
4526 /* Options that are not handled are only okay if they are unchanged
4527 * compared to the old state. It is expected that some options are only
4528 * used for the initial open, but not reopen (e.g. filename) */
4529 if (qdict_size(reopen_state->options)) {
4530 const QDictEntry *entry = qdict_first(reopen_state->options);
4532 do {
4533 QObject *new = entry->value;
4534 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4536 /* Allow child references (child_name=node_name) as long as they
4537 * point to the current child (i.e. everything stays the same). */
4538 if (qobject_type(new) == QTYPE_QSTRING) {
4539 BdrvChild *child;
4540 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4541 if (!strcmp(child->name, entry->key)) {
4542 break;
4546 if (child) {
4547 if (!strcmp(child->bs->node_name,
4548 qstring_get_str(qobject_to(QString, new)))) {
4549 continue; /* Found child with this name, skip option */
4555 * TODO: When using -drive to specify blockdev options, all values
4556 * will be strings; however, when using -blockdev, blockdev-add or
4557 * filenames using the json:{} pseudo-protocol, they will be
4558 * correctly typed.
4559 * In contrast, reopening options are (currently) always strings
4560 * (because you can only specify them through qemu-io; all other
4561 * callers do not specify any options).
4562 * Therefore, when using anything other than -drive to create a BDS,
4563 * this cannot detect non-string options as unchanged, because
4564 * qobject_is_equal() always returns false for objects of different
4565 * type. In the future, this should be remedied by correctly typing
4566 * all options. For now, this is not too big of an issue because
4567 * the user can simply omit options which cannot be changed anyway,
4568 * so they will stay unchanged.
4570 if (!qobject_is_equal(new, old)) {
4571 error_setg(errp, "Cannot change the option '%s'", entry->key);
4572 ret = -EINVAL;
4573 goto error;
4575 } while ((entry = qdict_next(reopen_state->options, entry)));
4578 ret = 0;
4580 /* Restore the original reopen_state->options QDict */
4581 qobject_unref(reopen_state->options);
4582 reopen_state->options = qobject_ref(orig_reopen_opts);
4584 error:
4585 if (ret < 0 && drv_prepared) {
4586 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4587 * call drv->bdrv_reopen_abort() before signaling an error
4588 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4589 * when the respective bdrv_reopen_prepare() has failed) */
4590 if (drv->bdrv_reopen_abort) {
4591 drv->bdrv_reopen_abort(reopen_state);
4594 qemu_opts_del(opts);
4595 qobject_unref(orig_reopen_opts);
4596 g_free(discard);
4597 return ret;
4601 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4602 * makes them final by swapping the staging BlockDriverState contents into
4603 * the active BlockDriverState contents.
4605 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4607 BlockDriver *drv;
4608 BlockDriverState *bs;
4609 BdrvChild *child;
4611 assert(reopen_state != NULL);
4612 bs = reopen_state->bs;
4613 drv = bs->drv;
4614 assert(drv != NULL);
4616 /* If there are any driver level actions to take */
4617 if (drv->bdrv_reopen_commit) {
4618 drv->bdrv_reopen_commit(reopen_state);
4621 /* set BDS specific flags now */
4622 qobject_unref(bs->explicit_options);
4623 qobject_unref(bs->options);
4624 qobject_ref(reopen_state->explicit_options);
4625 qobject_ref(reopen_state->options);
4627 bs->explicit_options = reopen_state->explicit_options;
4628 bs->options = reopen_state->options;
4629 bs->open_flags = reopen_state->flags;
4630 bs->detect_zeroes = reopen_state->detect_zeroes;
4632 /* Remove child references from bs->options and bs->explicit_options.
4633 * Child options were already removed in bdrv_reopen_queue_child() */
4634 QLIST_FOREACH(child, &bs->children, next) {
4635 qdict_del(bs->explicit_options, child->name);
4636 qdict_del(bs->options, child->name);
4638 /* backing is probably removed, so it's not handled by previous loop */
4639 qdict_del(bs->explicit_options, "backing");
4640 qdict_del(bs->options, "backing");
4642 bdrv_refresh_limits(bs, NULL, NULL);
4646 * Abort the reopen, and delete and free the staged changes in
4647 * reopen_state
4649 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4651 BlockDriver *drv;
4653 assert(reopen_state != NULL);
4654 drv = reopen_state->bs->drv;
4655 assert(drv != NULL);
4657 if (drv->bdrv_reopen_abort) {
4658 drv->bdrv_reopen_abort(reopen_state);
4663 static void bdrv_close(BlockDriverState *bs)
4665 BdrvAioNotifier *ban, *ban_next;
4666 BdrvChild *child, *next;
4668 assert(!bs->refcnt);
4670 bdrv_drained_begin(bs); /* complete I/O */
4671 bdrv_flush(bs);
4672 bdrv_drain(bs); /* in case flush left pending I/O */
4674 if (bs->drv) {
4675 if (bs->drv->bdrv_close) {
4676 /* Must unfreeze all children, so bdrv_unref_child() works */
4677 bs->drv->bdrv_close(bs);
4679 bs->drv = NULL;
4682 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4683 bdrv_unref_child(bs, child);
4686 bs->backing = NULL;
4687 bs->file = NULL;
4688 g_free(bs->opaque);
4689 bs->opaque = NULL;
4690 qatomic_set(&bs->copy_on_read, 0);
4691 bs->backing_file[0] = '\0';
4692 bs->backing_format[0] = '\0';
4693 bs->total_sectors = 0;
4694 bs->encrypted = false;
4695 bs->sg = false;
4696 qobject_unref(bs->options);
4697 qobject_unref(bs->explicit_options);
4698 bs->options = NULL;
4699 bs->explicit_options = NULL;
4700 qobject_unref(bs->full_open_options);
4701 bs->full_open_options = NULL;
4702 g_free(bs->block_status_cache);
4703 bs->block_status_cache = NULL;
4705 bdrv_release_named_dirty_bitmaps(bs);
4706 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4708 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4709 g_free(ban);
4711 QLIST_INIT(&bs->aio_notifiers);
4712 bdrv_drained_end(bs);
4715 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4716 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4717 * gets called.
4719 if (bs->quiesce_counter) {
4720 bdrv_drain_all_end_quiesce(bs);
4724 void bdrv_close_all(void)
4726 assert(job_next(NULL) == NULL);
4728 /* Drop references from requests still in flight, such as canceled block
4729 * jobs whose AIO context has not been polled yet */
4730 bdrv_drain_all();
4732 blk_remove_all_bs();
4733 blockdev_close_all_bdrv_states();
4735 assert(QTAILQ_EMPTY(&all_bdrv_states));
4738 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4740 GQueue *queue;
4741 GHashTable *found;
4742 bool ret;
4744 if (c->klass->stay_at_node) {
4745 return false;
4748 /* If the child @c belongs to the BDS @to, replacing the current
4749 * c->bs by @to would mean to create a loop.
4751 * Such a case occurs when appending a BDS to a backing chain.
4752 * For instance, imagine the following chain:
4754 * guest device -> node A -> further backing chain...
4756 * Now we create a new BDS B which we want to put on top of this
4757 * chain, so we first attach A as its backing node:
4759 * node B
4762 * guest device -> node A -> further backing chain...
4764 * Finally we want to replace A by B. When doing that, we want to
4765 * replace all pointers to A by pointers to B -- except for the
4766 * pointer from B because (1) that would create a loop, and (2)
4767 * that pointer should simply stay intact:
4769 * guest device -> node B
4772 * node A -> further backing chain...
4774 * In general, when replacing a node A (c->bs) by a node B (@to),
4775 * if A is a child of B, that means we cannot replace A by B there
4776 * because that would create a loop. Silently detaching A from B
4777 * is also not really an option. So overall just leaving A in
4778 * place there is the most sensible choice.
4780 * We would also create a loop in any cases where @c is only
4781 * indirectly referenced by @to. Prevent this by returning false
4782 * if @c is found (by breadth-first search) anywhere in the whole
4783 * subtree of @to.
4786 ret = true;
4787 found = g_hash_table_new(NULL, NULL);
4788 g_hash_table_add(found, to);
4789 queue = g_queue_new();
4790 g_queue_push_tail(queue, to);
4792 while (!g_queue_is_empty(queue)) {
4793 BlockDriverState *v = g_queue_pop_head(queue);
4794 BdrvChild *c2;
4796 QLIST_FOREACH(c2, &v->children, next) {
4797 if (c2 == c) {
4798 ret = false;
4799 break;
4802 if (g_hash_table_contains(found, c2->bs)) {
4803 continue;
4806 g_queue_push_tail(queue, c2->bs);
4807 g_hash_table_add(found, c2->bs);
4811 g_queue_free(queue);
4812 g_hash_table_destroy(found);
4814 return ret;
4817 typedef struct BdrvRemoveFilterOrCowChild {
4818 BdrvChild *child;
4819 bool is_backing;
4820 } BdrvRemoveFilterOrCowChild;
4822 static void bdrv_remove_filter_or_cow_child_abort(void *opaque)
4824 BdrvRemoveFilterOrCowChild *s = opaque;
4825 BlockDriverState *parent_bs = s->child->opaque;
4827 QLIST_INSERT_HEAD(&parent_bs->children, s->child, next);
4828 if (s->is_backing) {
4829 parent_bs->backing = s->child;
4830 } else {
4831 parent_bs->file = s->child;
4835 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
4836 * because that function is transactionable and it registered own completion
4837 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
4838 * called automatically.
4842 static void bdrv_remove_filter_or_cow_child_commit(void *opaque)
4844 BdrvRemoveFilterOrCowChild *s = opaque;
4846 bdrv_child_free(s->child);
4849 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = {
4850 .abort = bdrv_remove_filter_or_cow_child_abort,
4851 .commit = bdrv_remove_filter_or_cow_child_commit,
4852 .clean = g_free,
4856 * A function to remove backing or file child of @bs.
4857 * Function doesn't update permissions, caller is responsible for this.
4859 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
4860 BdrvChild *child,
4861 Transaction *tran)
4863 BdrvRemoveFilterOrCowChild *s;
4865 assert(child == bs->backing || child == bs->file);
4867 if (!child) {
4868 return;
4871 if (child->bs) {
4872 bdrv_replace_child_tran(child, NULL, tran);
4875 s = g_new(BdrvRemoveFilterOrCowChild, 1);
4876 *s = (BdrvRemoveFilterOrCowChild) {
4877 .child = child,
4878 .is_backing = (child == bs->backing),
4880 tran_add(tran, &bdrv_remove_filter_or_cow_child_drv, s);
4882 QLIST_SAFE_REMOVE(child, next);
4883 if (s->is_backing) {
4884 bs->backing = NULL;
4885 } else {
4886 bs->file = NULL;
4891 * A function to remove backing-chain child of @bs if exists: cow child for
4892 * format nodes (always .backing) and filter child for filters (may be .file or
4893 * .backing)
4895 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
4896 Transaction *tran)
4898 bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran);
4901 static int bdrv_replace_node_noperm(BlockDriverState *from,
4902 BlockDriverState *to,
4903 bool auto_skip, Transaction *tran,
4904 Error **errp)
4906 BdrvChild *c, *next;
4908 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4909 assert(c->bs == from);
4910 if (!should_update_child(c, to)) {
4911 if (auto_skip) {
4912 continue;
4914 error_setg(errp, "Should not change '%s' link to '%s'",
4915 c->name, from->node_name);
4916 return -EINVAL;
4918 if (c->frozen) {
4919 error_setg(errp, "Cannot change '%s' link to '%s'",
4920 c->name, from->node_name);
4921 return -EPERM;
4923 bdrv_replace_child_tran(c, to, tran);
4926 return 0;
4930 * With auto_skip=true bdrv_replace_node_common skips updating from parents
4931 * if it creates a parent-child relation loop or if parent is block-job.
4933 * With auto_skip=false the error is returned if from has a parent which should
4934 * not be updated.
4936 * With @detach_subchain=true @to must be in a backing chain of @from. In this
4937 * case backing link of the cow-parent of @to is removed.
4939 static int bdrv_replace_node_common(BlockDriverState *from,
4940 BlockDriverState *to,
4941 bool auto_skip, bool detach_subchain,
4942 Error **errp)
4944 Transaction *tran = tran_new();
4945 g_autoptr(GHashTable) found = NULL;
4946 g_autoptr(GSList) refresh_list = NULL;
4947 BlockDriverState *to_cow_parent = NULL;
4948 int ret;
4950 if (detach_subchain) {
4951 assert(bdrv_chain_contains(from, to));
4952 assert(from != to);
4953 for (to_cow_parent = from;
4954 bdrv_filter_or_cow_bs(to_cow_parent) != to;
4955 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
4961 /* Make sure that @from doesn't go away until we have successfully attached
4962 * all of its parents to @to. */
4963 bdrv_ref(from);
4965 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4966 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
4967 bdrv_drained_begin(from);
4970 * Do the replacement without permission update.
4971 * Replacement may influence the permissions, we should calculate new
4972 * permissions based on new graph. If we fail, we'll roll-back the
4973 * replacement.
4975 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
4976 if (ret < 0) {
4977 goto out;
4980 if (detach_subchain) {
4981 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
4984 found = g_hash_table_new(NULL, NULL);
4986 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
4987 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
4989 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
4990 if (ret < 0) {
4991 goto out;
4994 ret = 0;
4996 out:
4997 tran_finalize(tran, ret);
4999 bdrv_drained_end(from);
5000 bdrv_unref(from);
5002 return ret;
5005 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5006 Error **errp)
5008 return bdrv_replace_node_common(from, to, true, false, errp);
5011 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5013 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5014 errp);
5018 * Add new bs contents at the top of an image chain while the chain is
5019 * live, while keeping required fields on the top layer.
5021 * This will modify the BlockDriverState fields, and swap contents
5022 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5024 * bs_new must not be attached to a BlockBackend and must not have backing
5025 * child.
5027 * This function does not create any image files.
5029 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5030 Error **errp)
5032 int ret;
5033 Transaction *tran = tran_new();
5035 assert(!bs_new->backing);
5037 ret = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5038 &child_of_bds, bdrv_backing_role(bs_new),
5039 &bs_new->backing, tran, errp);
5040 if (ret < 0) {
5041 goto out;
5044 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5045 if (ret < 0) {
5046 goto out;
5049 ret = bdrv_refresh_perms(bs_new, errp);
5050 out:
5051 tran_finalize(tran, ret);
5053 bdrv_refresh_limits(bs_top, NULL, NULL);
5055 return ret;
5058 /* Not for empty child */
5059 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5060 Error **errp)
5062 int ret;
5063 Transaction *tran = tran_new();
5064 g_autoptr(GHashTable) found = NULL;
5065 g_autoptr(GSList) refresh_list = NULL;
5066 BlockDriverState *old_bs = child->bs;
5068 bdrv_ref(old_bs);
5069 bdrv_drained_begin(old_bs);
5070 bdrv_drained_begin(new_bs);
5072 bdrv_replace_child_tran(child, new_bs, tran);
5074 found = g_hash_table_new(NULL, NULL);
5075 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5076 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5078 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5080 tran_finalize(tran, ret);
5082 bdrv_drained_end(old_bs);
5083 bdrv_drained_end(new_bs);
5084 bdrv_unref(old_bs);
5086 return ret;
5089 static void bdrv_delete(BlockDriverState *bs)
5091 assert(bdrv_op_blocker_is_empty(bs));
5092 assert(!bs->refcnt);
5094 /* remove from list, if necessary */
5095 if (bs->node_name[0] != '\0') {
5096 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5098 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5100 bdrv_close(bs);
5102 g_free(bs);
5105 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *node_options,
5106 int flags, Error **errp)
5108 BlockDriverState *new_node_bs;
5109 Error *local_err = NULL;
5111 new_node_bs = bdrv_open(NULL, NULL, node_options, flags, errp);
5112 if (new_node_bs == NULL) {
5113 error_prepend(errp, "Could not create node: ");
5114 return NULL;
5117 bdrv_drained_begin(bs);
5118 bdrv_replace_node(bs, new_node_bs, &local_err);
5119 bdrv_drained_end(bs);
5121 if (local_err) {
5122 bdrv_unref(new_node_bs);
5123 error_propagate(errp, local_err);
5124 return NULL;
5127 return new_node_bs;
5131 * Run consistency checks on an image
5133 * Returns 0 if the check could be completed (it doesn't mean that the image is
5134 * free of errors) or -errno when an internal error occurred. The results of the
5135 * check are stored in res.
5137 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5138 BdrvCheckResult *res, BdrvCheckMode fix)
5140 if (bs->drv == NULL) {
5141 return -ENOMEDIUM;
5143 if (bs->drv->bdrv_co_check == NULL) {
5144 return -ENOTSUP;
5147 memset(res, 0, sizeof(*res));
5148 return bs->drv->bdrv_co_check(bs, res, fix);
5152 * Return values:
5153 * 0 - success
5154 * -EINVAL - backing format specified, but no file
5155 * -ENOSPC - can't update the backing file because no space is left in the
5156 * image file header
5157 * -ENOTSUP - format driver doesn't support changing the backing file
5159 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5160 const char *backing_fmt, bool require)
5162 BlockDriver *drv = bs->drv;
5163 int ret;
5165 if (!drv) {
5166 return -ENOMEDIUM;
5169 /* Backing file format doesn't make sense without a backing file */
5170 if (backing_fmt && !backing_file) {
5171 return -EINVAL;
5174 if (require && backing_file && !backing_fmt) {
5175 return -EINVAL;
5178 if (drv->bdrv_change_backing_file != NULL) {
5179 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5180 } else {
5181 ret = -ENOTSUP;
5184 if (ret == 0) {
5185 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5186 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5187 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5188 backing_file ?: "");
5190 return ret;
5194 * Finds the first non-filter node above bs in the chain between
5195 * active and bs. The returned node is either an immediate parent of
5196 * bs, or there are only filter nodes between the two.
5198 * Returns NULL if bs is not found in active's image chain,
5199 * or if active == bs.
5201 * Returns the bottommost base image if bs == NULL.
5203 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5204 BlockDriverState *bs)
5206 bs = bdrv_skip_filters(bs);
5207 active = bdrv_skip_filters(active);
5209 while (active) {
5210 BlockDriverState *next = bdrv_backing_chain_next(active);
5211 if (bs == next) {
5212 return active;
5214 active = next;
5217 return NULL;
5220 /* Given a BDS, searches for the base layer. */
5221 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5223 return bdrv_find_overlay(bs, NULL);
5227 * Return true if at least one of the COW (backing) and filter links
5228 * between @bs and @base is frozen. @errp is set if that's the case.
5229 * @base must be reachable from @bs, or NULL.
5231 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5232 Error **errp)
5234 BlockDriverState *i;
5235 BdrvChild *child;
5237 for (i = bs; i != base; i = child_bs(child)) {
5238 child = bdrv_filter_or_cow_child(i);
5240 if (child && child->frozen) {
5241 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5242 child->name, i->node_name, child->bs->node_name);
5243 return true;
5247 return false;
5251 * Freeze all COW (backing) and filter links between @bs and @base.
5252 * If any of the links is already frozen the operation is aborted and
5253 * none of the links are modified.
5254 * @base must be reachable from @bs, or NULL.
5255 * Returns 0 on success. On failure returns < 0 and sets @errp.
5257 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5258 Error **errp)
5260 BlockDriverState *i;
5261 BdrvChild *child;
5263 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5264 return -EPERM;
5267 for (i = bs; i != base; i = child_bs(child)) {
5268 child = bdrv_filter_or_cow_child(i);
5269 if (child && child->bs->never_freeze) {
5270 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5271 child->name, child->bs->node_name);
5272 return -EPERM;
5276 for (i = bs; i != base; i = child_bs(child)) {
5277 child = bdrv_filter_or_cow_child(i);
5278 if (child) {
5279 child->frozen = true;
5283 return 0;
5287 * Unfreeze all COW (backing) and filter links between @bs and @base.
5288 * The caller must ensure that all links are frozen before using this
5289 * function.
5290 * @base must be reachable from @bs, or NULL.
5292 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5294 BlockDriverState *i;
5295 BdrvChild *child;
5297 for (i = bs; i != base; i = child_bs(child)) {
5298 child = bdrv_filter_or_cow_child(i);
5299 if (child) {
5300 assert(child->frozen);
5301 child->frozen = false;
5307 * Drops images above 'base' up to and including 'top', and sets the image
5308 * above 'top' to have base as its backing file.
5310 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5311 * information in 'bs' can be properly updated.
5313 * E.g., this will convert the following chain:
5314 * bottom <- base <- intermediate <- top <- active
5316 * to
5318 * bottom <- base <- active
5320 * It is allowed for bottom==base, in which case it converts:
5322 * base <- intermediate <- top <- active
5324 * to
5326 * base <- active
5328 * If backing_file_str is non-NULL, it will be used when modifying top's
5329 * overlay image metadata.
5331 * Error conditions:
5332 * if active == top, that is considered an error
5335 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5336 const char *backing_file_str)
5338 BlockDriverState *explicit_top = top;
5339 bool update_inherits_from;
5340 BdrvChild *c;
5341 Error *local_err = NULL;
5342 int ret = -EIO;
5343 g_autoptr(GSList) updated_children = NULL;
5344 GSList *p;
5346 bdrv_ref(top);
5347 bdrv_subtree_drained_begin(top);
5349 if (!top->drv || !base->drv) {
5350 goto exit;
5353 /* Make sure that base is in the backing chain of top */
5354 if (!bdrv_chain_contains(top, base)) {
5355 goto exit;
5358 /* If 'base' recursively inherits from 'top' then we should set
5359 * base->inherits_from to top->inherits_from after 'top' and all
5360 * other intermediate nodes have been dropped.
5361 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5362 * it because no one inherits from it. We use explicit_top for that. */
5363 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5364 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5366 /* success - we can delete the intermediate states, and link top->base */
5367 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
5368 * we've figured out how they should work. */
5369 if (!backing_file_str) {
5370 bdrv_refresh_filename(base);
5371 backing_file_str = base->filename;
5374 QLIST_FOREACH(c, &top->parents, next_parent) {
5375 updated_children = g_slist_prepend(updated_children, c);
5379 * It seems correct to pass detach_subchain=true here, but it triggers
5380 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5381 * another drained section, which modify the graph (for example, removing
5382 * the child, which we keep in updated_children list). So, it's a TODO.
5384 * Note, bug triggered if pass detach_subchain=true here and run
5385 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5386 * That's a FIXME.
5388 bdrv_replace_node_common(top, base, false, false, &local_err);
5389 if (local_err) {
5390 error_report_err(local_err);
5391 goto exit;
5394 for (p = updated_children; p; p = p->next) {
5395 c = p->data;
5397 if (c->klass->update_filename) {
5398 ret = c->klass->update_filename(c, base, backing_file_str,
5399 &local_err);
5400 if (ret < 0) {
5402 * TODO: Actually, we want to rollback all previous iterations
5403 * of this loop, and (which is almost impossible) previous
5404 * bdrv_replace_node()...
5406 * Note, that c->klass->update_filename may lead to permission
5407 * update, so it's a bad idea to call it inside permission
5408 * update transaction of bdrv_replace_node.
5410 error_report_err(local_err);
5411 goto exit;
5416 if (update_inherits_from) {
5417 base->inherits_from = explicit_top->inherits_from;
5420 ret = 0;
5421 exit:
5422 bdrv_subtree_drained_end(top);
5423 bdrv_unref(top);
5424 return ret;
5428 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5429 * sums the size of all data-bearing children. (This excludes backing
5430 * children.)
5432 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5434 BdrvChild *child;
5435 int64_t child_size, sum = 0;
5437 QLIST_FOREACH(child, &bs->children, next) {
5438 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5439 BDRV_CHILD_FILTERED))
5441 child_size = bdrv_get_allocated_file_size(child->bs);
5442 if (child_size < 0) {
5443 return child_size;
5445 sum += child_size;
5449 return sum;
5453 * Length of a allocated file in bytes. Sparse files are counted by actual
5454 * allocated space. Return < 0 if error or unknown.
5456 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5458 BlockDriver *drv = bs->drv;
5459 if (!drv) {
5460 return -ENOMEDIUM;
5462 if (drv->bdrv_get_allocated_file_size) {
5463 return drv->bdrv_get_allocated_file_size(bs);
5466 if (drv->bdrv_file_open) {
5468 * Protocol drivers default to -ENOTSUP (most of their data is
5469 * not stored in any of their children (if they even have any),
5470 * so there is no generic way to figure it out).
5472 return -ENOTSUP;
5473 } else if (drv->is_filter) {
5474 /* Filter drivers default to the size of their filtered child */
5475 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5476 } else {
5477 /* Other drivers default to summing their children's sizes */
5478 return bdrv_sum_allocated_file_size(bs);
5483 * bdrv_measure:
5484 * @drv: Format driver
5485 * @opts: Creation options for new image
5486 * @in_bs: Existing image containing data for new image (may be NULL)
5487 * @errp: Error object
5488 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5489 * or NULL on error
5491 * Calculate file size required to create a new image.
5493 * If @in_bs is given then space for allocated clusters and zero clusters
5494 * from that image are included in the calculation. If @opts contains a
5495 * backing file that is shared by @in_bs then backing clusters may be omitted
5496 * from the calculation.
5498 * If @in_bs is NULL then the calculation includes no allocated clusters
5499 * unless a preallocation option is given in @opts.
5501 * Note that @in_bs may use a different BlockDriver from @drv.
5503 * If an error occurs the @errp pointer is set.
5505 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5506 BlockDriverState *in_bs, Error **errp)
5508 if (!drv->bdrv_measure) {
5509 error_setg(errp, "Block driver '%s' does not support size measurement",
5510 drv->format_name);
5511 return NULL;
5514 return drv->bdrv_measure(opts, in_bs, errp);
5518 * Return number of sectors on success, -errno on error.
5520 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5522 BlockDriver *drv = bs->drv;
5524 if (!drv)
5525 return -ENOMEDIUM;
5527 if (drv->has_variable_length) {
5528 int ret = refresh_total_sectors(bs, bs->total_sectors);
5529 if (ret < 0) {
5530 return ret;
5533 return bs->total_sectors;
5537 * Return length in bytes on success, -errno on error.
5538 * The length is always a multiple of BDRV_SECTOR_SIZE.
5540 int64_t bdrv_getlength(BlockDriverState *bs)
5542 int64_t ret = bdrv_nb_sectors(bs);
5544 if (ret < 0) {
5545 return ret;
5547 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5548 return -EFBIG;
5550 return ret * BDRV_SECTOR_SIZE;
5553 /* return 0 as number of sectors if no device present or error */
5554 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5556 int64_t nb_sectors = bdrv_nb_sectors(bs);
5558 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5561 bool bdrv_is_sg(BlockDriverState *bs)
5563 return bs->sg;
5567 * Return whether the given node supports compressed writes.
5569 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5571 BlockDriverState *filtered;
5573 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5574 return false;
5577 filtered = bdrv_filter_bs(bs);
5578 if (filtered) {
5580 * Filters can only forward compressed writes, so we have to
5581 * check the child.
5583 return bdrv_supports_compressed_writes(filtered);
5586 return true;
5589 const char *bdrv_get_format_name(BlockDriverState *bs)
5591 return bs->drv ? bs->drv->format_name : NULL;
5594 static int qsort_strcmp(const void *a, const void *b)
5596 return strcmp(*(char *const *)a, *(char *const *)b);
5599 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5600 void *opaque, bool read_only)
5602 BlockDriver *drv;
5603 int count = 0;
5604 int i;
5605 const char **formats = NULL;
5607 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5608 if (drv->format_name) {
5609 bool found = false;
5610 int i = count;
5612 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5613 continue;
5616 while (formats && i && !found) {
5617 found = !strcmp(formats[--i], drv->format_name);
5620 if (!found) {
5621 formats = g_renew(const char *, formats, count + 1);
5622 formats[count++] = drv->format_name;
5627 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5628 const char *format_name = block_driver_modules[i].format_name;
5630 if (format_name) {
5631 bool found = false;
5632 int j = count;
5634 if (use_bdrv_whitelist &&
5635 !bdrv_format_is_whitelisted(format_name, read_only)) {
5636 continue;
5639 while (formats && j && !found) {
5640 found = !strcmp(formats[--j], format_name);
5643 if (!found) {
5644 formats = g_renew(const char *, formats, count + 1);
5645 formats[count++] = format_name;
5650 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5652 for (i = 0; i < count; i++) {
5653 it(opaque, formats[i]);
5656 g_free(formats);
5659 /* This function is to find a node in the bs graph */
5660 BlockDriverState *bdrv_find_node(const char *node_name)
5662 BlockDriverState *bs;
5664 assert(node_name);
5666 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5667 if (!strcmp(node_name, bs->node_name)) {
5668 return bs;
5671 return NULL;
5674 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5675 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5676 Error **errp)
5678 BlockDeviceInfoList *list;
5679 BlockDriverState *bs;
5681 list = NULL;
5682 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5683 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5684 if (!info) {
5685 qapi_free_BlockDeviceInfoList(list);
5686 return NULL;
5688 QAPI_LIST_PREPEND(list, info);
5691 return list;
5694 typedef struct XDbgBlockGraphConstructor {
5695 XDbgBlockGraph *graph;
5696 GHashTable *graph_nodes;
5697 } XDbgBlockGraphConstructor;
5699 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5701 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5703 gr->graph = g_new0(XDbgBlockGraph, 1);
5704 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5706 return gr;
5709 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5711 XDbgBlockGraph *graph = gr->graph;
5713 g_hash_table_destroy(gr->graph_nodes);
5714 g_free(gr);
5716 return graph;
5719 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5721 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5723 if (ret != 0) {
5724 return ret;
5728 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5729 * answer of g_hash_table_lookup.
5731 ret = g_hash_table_size(gr->graph_nodes) + 1;
5732 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5734 return ret;
5737 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5738 XDbgBlockGraphNodeType type, const char *name)
5740 XDbgBlockGraphNode *n;
5742 n = g_new0(XDbgBlockGraphNode, 1);
5744 n->id = xdbg_graph_node_num(gr, node);
5745 n->type = type;
5746 n->name = g_strdup(name);
5748 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5751 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5752 const BdrvChild *child)
5754 BlockPermission qapi_perm;
5755 XDbgBlockGraphEdge *edge;
5757 edge = g_new0(XDbgBlockGraphEdge, 1);
5759 edge->parent = xdbg_graph_node_num(gr, parent);
5760 edge->child = xdbg_graph_node_num(gr, child->bs);
5761 edge->name = g_strdup(child->name);
5763 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5764 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5766 if (flag & child->perm) {
5767 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5769 if (flag & child->shared_perm) {
5770 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5774 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5778 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5780 BlockBackend *blk;
5781 BlockJob *job;
5782 BlockDriverState *bs;
5783 BdrvChild *child;
5784 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5786 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5787 char *allocated_name = NULL;
5788 const char *name = blk_name(blk);
5790 if (!*name) {
5791 name = allocated_name = blk_get_attached_dev_id(blk);
5793 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5794 name);
5795 g_free(allocated_name);
5796 if (blk_root(blk)) {
5797 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5801 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5802 GSList *el;
5804 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5805 job->job.id);
5806 for (el = job->nodes; el; el = el->next) {
5807 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5811 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5812 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5813 bs->node_name);
5814 QLIST_FOREACH(child, &bs->children, next) {
5815 xdbg_graph_add_edge(gr, bs, child);
5819 return xdbg_graph_finalize(gr);
5822 BlockDriverState *bdrv_lookup_bs(const char *device,
5823 const char *node_name,
5824 Error **errp)
5826 BlockBackend *blk;
5827 BlockDriverState *bs;
5829 if (device) {
5830 blk = blk_by_name(device);
5832 if (blk) {
5833 bs = blk_bs(blk);
5834 if (!bs) {
5835 error_setg(errp, "Device '%s' has no medium", device);
5838 return bs;
5842 if (node_name) {
5843 bs = bdrv_find_node(node_name);
5845 if (bs) {
5846 return bs;
5850 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
5851 device ? device : "",
5852 node_name ? node_name : "");
5853 return NULL;
5856 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5857 * return false. If either argument is NULL, return false. */
5858 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
5860 while (top && top != base) {
5861 top = bdrv_filter_or_cow_bs(top);
5864 return top != NULL;
5867 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
5869 if (!bs) {
5870 return QTAILQ_FIRST(&graph_bdrv_states);
5872 return QTAILQ_NEXT(bs, node_list);
5875 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5877 if (!bs) {
5878 return QTAILQ_FIRST(&all_bdrv_states);
5880 return QTAILQ_NEXT(bs, bs_list);
5883 const char *bdrv_get_node_name(const BlockDriverState *bs)
5885 return bs->node_name;
5888 const char *bdrv_get_parent_name(const BlockDriverState *bs)
5890 BdrvChild *c;
5891 const char *name;
5893 /* If multiple parents have a name, just pick the first one. */
5894 QLIST_FOREACH(c, &bs->parents, next_parent) {
5895 if (c->klass->get_name) {
5896 name = c->klass->get_name(c);
5897 if (name && *name) {
5898 return name;
5903 return NULL;
5906 /* TODO check what callers really want: bs->node_name or blk_name() */
5907 const char *bdrv_get_device_name(const BlockDriverState *bs)
5909 return bdrv_get_parent_name(bs) ?: "";
5912 /* This can be used to identify nodes that might not have a device
5913 * name associated. Since node and device names live in the same
5914 * namespace, the result is unambiguous. The exception is if both are
5915 * absent, then this returns an empty (non-null) string. */
5916 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
5918 return bdrv_get_parent_name(bs) ?: bs->node_name;
5921 int bdrv_get_flags(BlockDriverState *bs)
5923 return bs->open_flags;
5926 int bdrv_has_zero_init_1(BlockDriverState *bs)
5928 return 1;
5931 int bdrv_has_zero_init(BlockDriverState *bs)
5933 BlockDriverState *filtered;
5935 if (!bs->drv) {
5936 return 0;
5939 /* If BS is a copy on write image, it is initialized to
5940 the contents of the base image, which may not be zeroes. */
5941 if (bdrv_cow_child(bs)) {
5942 return 0;
5944 if (bs->drv->bdrv_has_zero_init) {
5945 return bs->drv->bdrv_has_zero_init(bs);
5948 filtered = bdrv_filter_bs(bs);
5949 if (filtered) {
5950 return bdrv_has_zero_init(filtered);
5953 /* safe default */
5954 return 0;
5957 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
5959 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5960 return false;
5963 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
5966 void bdrv_get_backing_filename(BlockDriverState *bs,
5967 char *filename, int filename_size)
5969 pstrcpy(filename, filename_size, bs->backing_file);
5972 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5974 int ret;
5975 BlockDriver *drv = bs->drv;
5976 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5977 if (!drv) {
5978 return -ENOMEDIUM;
5980 if (!drv->bdrv_get_info) {
5981 BlockDriverState *filtered = bdrv_filter_bs(bs);
5982 if (filtered) {
5983 return bdrv_get_info(filtered, bdi);
5985 return -ENOTSUP;
5987 memset(bdi, 0, sizeof(*bdi));
5988 ret = drv->bdrv_get_info(bs, bdi);
5989 if (ret < 0) {
5990 return ret;
5993 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
5994 return -EINVAL;
5997 return 0;
6000 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6001 Error **errp)
6003 BlockDriver *drv = bs->drv;
6004 if (drv && drv->bdrv_get_specific_info) {
6005 return drv->bdrv_get_specific_info(bs, errp);
6007 return NULL;
6010 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6012 BlockDriver *drv = bs->drv;
6013 if (!drv || !drv->bdrv_get_specific_stats) {
6014 return NULL;
6016 return drv->bdrv_get_specific_stats(bs);
6019 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6021 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6022 return;
6025 bs->drv->bdrv_debug_event(bs, event);
6028 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6030 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6031 bs = bdrv_primary_bs(bs);
6034 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6035 assert(bs->drv->bdrv_debug_remove_breakpoint);
6036 return bs;
6039 return NULL;
6042 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6043 const char *tag)
6045 bs = bdrv_find_debug_node(bs);
6046 if (bs) {
6047 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6050 return -ENOTSUP;
6053 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6055 bs = bdrv_find_debug_node(bs);
6056 if (bs) {
6057 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6060 return -ENOTSUP;
6063 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6065 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6066 bs = bdrv_primary_bs(bs);
6069 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6070 return bs->drv->bdrv_debug_resume(bs, tag);
6073 return -ENOTSUP;
6076 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6078 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6079 bs = bdrv_primary_bs(bs);
6082 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6083 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6086 return false;
6089 /* backing_file can either be relative, or absolute, or a protocol. If it is
6090 * relative, it must be relative to the chain. So, passing in bs->filename
6091 * from a BDS as backing_file should not be done, as that may be relative to
6092 * the CWD rather than the chain. */
6093 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6094 const char *backing_file)
6096 char *filename_full = NULL;
6097 char *backing_file_full = NULL;
6098 char *filename_tmp = NULL;
6099 int is_protocol = 0;
6100 bool filenames_refreshed = false;
6101 BlockDriverState *curr_bs = NULL;
6102 BlockDriverState *retval = NULL;
6103 BlockDriverState *bs_below;
6105 if (!bs || !bs->drv || !backing_file) {
6106 return NULL;
6109 filename_full = g_malloc(PATH_MAX);
6110 backing_file_full = g_malloc(PATH_MAX);
6112 is_protocol = path_has_protocol(backing_file);
6115 * Being largely a legacy function, skip any filters here
6116 * (because filters do not have normal filenames, so they cannot
6117 * match anyway; and allowing json:{} filenames is a bit out of
6118 * scope).
6120 for (curr_bs = bdrv_skip_filters(bs);
6121 bdrv_cow_child(curr_bs) != NULL;
6122 curr_bs = bs_below)
6124 bs_below = bdrv_backing_chain_next(curr_bs);
6126 if (bdrv_backing_overridden(curr_bs)) {
6128 * If the backing file was overridden, we can only compare
6129 * directly against the backing node's filename.
6132 if (!filenames_refreshed) {
6134 * This will automatically refresh all of the
6135 * filenames in the rest of the backing chain, so we
6136 * only need to do this once.
6138 bdrv_refresh_filename(bs_below);
6139 filenames_refreshed = true;
6142 if (strcmp(backing_file, bs_below->filename) == 0) {
6143 retval = bs_below;
6144 break;
6146 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6148 * If either of the filename paths is actually a protocol, then
6149 * compare unmodified paths; otherwise make paths relative.
6151 char *backing_file_full_ret;
6153 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6154 retval = bs_below;
6155 break;
6157 /* Also check against the full backing filename for the image */
6158 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6159 NULL);
6160 if (backing_file_full_ret) {
6161 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6162 g_free(backing_file_full_ret);
6163 if (equal) {
6164 retval = bs_below;
6165 break;
6168 } else {
6169 /* If not an absolute filename path, make it relative to the current
6170 * image's filename path */
6171 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6172 NULL);
6173 /* We are going to compare canonicalized absolute pathnames */
6174 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6175 g_free(filename_tmp);
6176 continue;
6178 g_free(filename_tmp);
6180 /* We need to make sure the backing filename we are comparing against
6181 * is relative to the current image filename (or absolute) */
6182 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6183 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6184 g_free(filename_tmp);
6185 continue;
6187 g_free(filename_tmp);
6189 if (strcmp(backing_file_full, filename_full) == 0) {
6190 retval = bs_below;
6191 break;
6196 g_free(filename_full);
6197 g_free(backing_file_full);
6198 return retval;
6201 void bdrv_init(void)
6203 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6204 use_bdrv_whitelist = 1;
6205 #endif
6206 module_call_init(MODULE_INIT_BLOCK);
6209 void bdrv_init_with_whitelist(void)
6211 use_bdrv_whitelist = 1;
6212 bdrv_init();
6215 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6217 BdrvChild *child, *parent;
6218 Error *local_err = NULL;
6219 int ret;
6220 BdrvDirtyBitmap *bm;
6222 if (!bs->drv) {
6223 return -ENOMEDIUM;
6226 QLIST_FOREACH(child, &bs->children, next) {
6227 bdrv_co_invalidate_cache(child->bs, &local_err);
6228 if (local_err) {
6229 error_propagate(errp, local_err);
6230 return -EINVAL;
6235 * Update permissions, they may differ for inactive nodes.
6237 * Note that the required permissions of inactive images are always a
6238 * subset of the permissions required after activating the image. This
6239 * allows us to just get the permissions upfront without restricting
6240 * drv->bdrv_invalidate_cache().
6242 * It also means that in error cases, we don't have to try and revert to
6243 * the old permissions (which is an operation that could fail, too). We can
6244 * just keep the extended permissions for the next time that an activation
6245 * of the image is tried.
6247 if (bs->open_flags & BDRV_O_INACTIVE) {
6248 bs->open_flags &= ~BDRV_O_INACTIVE;
6249 ret = bdrv_refresh_perms(bs, errp);
6250 if (ret < 0) {
6251 bs->open_flags |= BDRV_O_INACTIVE;
6252 return ret;
6255 if (bs->drv->bdrv_co_invalidate_cache) {
6256 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6257 if (local_err) {
6258 bs->open_flags |= BDRV_O_INACTIVE;
6259 error_propagate(errp, local_err);
6260 return -EINVAL;
6264 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6265 bdrv_dirty_bitmap_skip_store(bm, false);
6268 ret = refresh_total_sectors(bs, bs->total_sectors);
6269 if (ret < 0) {
6270 bs->open_flags |= BDRV_O_INACTIVE;
6271 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6272 return ret;
6276 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6277 if (parent->klass->activate) {
6278 parent->klass->activate(parent, &local_err);
6279 if (local_err) {
6280 bs->open_flags |= BDRV_O_INACTIVE;
6281 error_propagate(errp, local_err);
6282 return -EINVAL;
6287 return 0;
6290 void bdrv_invalidate_cache_all(Error **errp)
6292 BlockDriverState *bs;
6293 BdrvNextIterator it;
6295 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6296 AioContext *aio_context = bdrv_get_aio_context(bs);
6297 int ret;
6299 aio_context_acquire(aio_context);
6300 ret = bdrv_invalidate_cache(bs, errp);
6301 aio_context_release(aio_context);
6302 if (ret < 0) {
6303 bdrv_next_cleanup(&it);
6304 return;
6309 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6311 BdrvChild *parent;
6313 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6314 if (parent->klass->parent_is_bds) {
6315 BlockDriverState *parent_bs = parent->opaque;
6316 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6317 return true;
6322 return false;
6325 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6327 BdrvChild *child, *parent;
6328 int ret;
6329 uint64_t cumulative_perms, cumulative_shared_perms;
6331 if (!bs->drv) {
6332 return -ENOMEDIUM;
6335 /* Make sure that we don't inactivate a child before its parent.
6336 * It will be covered by recursion from the yet active parent. */
6337 if (bdrv_has_bds_parent(bs, true)) {
6338 return 0;
6341 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6343 /* Inactivate this node */
6344 if (bs->drv->bdrv_inactivate) {
6345 ret = bs->drv->bdrv_inactivate(bs);
6346 if (ret < 0) {
6347 return ret;
6351 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6352 if (parent->klass->inactivate) {
6353 ret = parent->klass->inactivate(parent);
6354 if (ret < 0) {
6355 return ret;
6360 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6361 &cumulative_shared_perms);
6362 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6363 /* Our inactive parents still need write access. Inactivation failed. */
6364 return -EPERM;
6367 bs->open_flags |= BDRV_O_INACTIVE;
6370 * Update permissions, they may differ for inactive nodes.
6371 * We only tried to loosen restrictions, so errors are not fatal, ignore
6372 * them.
6374 bdrv_refresh_perms(bs, NULL);
6376 /* Recursively inactivate children */
6377 QLIST_FOREACH(child, &bs->children, next) {
6378 ret = bdrv_inactivate_recurse(child->bs);
6379 if (ret < 0) {
6380 return ret;
6384 return 0;
6387 int bdrv_inactivate_all(void)
6389 BlockDriverState *bs = NULL;
6390 BdrvNextIterator it;
6391 int ret = 0;
6392 GSList *aio_ctxs = NULL, *ctx;
6394 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6395 AioContext *aio_context = bdrv_get_aio_context(bs);
6397 if (!g_slist_find(aio_ctxs, aio_context)) {
6398 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6399 aio_context_acquire(aio_context);
6403 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6404 /* Nodes with BDS parents are covered by recursion from the last
6405 * parent that gets inactivated. Don't inactivate them a second
6406 * time if that has already happened. */
6407 if (bdrv_has_bds_parent(bs, false)) {
6408 continue;
6410 ret = bdrv_inactivate_recurse(bs);
6411 if (ret < 0) {
6412 bdrv_next_cleanup(&it);
6413 goto out;
6417 out:
6418 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6419 AioContext *aio_context = ctx->data;
6420 aio_context_release(aio_context);
6422 g_slist_free(aio_ctxs);
6424 return ret;
6427 /**************************************************************/
6428 /* removable device support */
6431 * Return TRUE if the media is present
6433 bool bdrv_is_inserted(BlockDriverState *bs)
6435 BlockDriver *drv = bs->drv;
6436 BdrvChild *child;
6438 if (!drv) {
6439 return false;
6441 if (drv->bdrv_is_inserted) {
6442 return drv->bdrv_is_inserted(bs);
6444 QLIST_FOREACH(child, &bs->children, next) {
6445 if (!bdrv_is_inserted(child->bs)) {
6446 return false;
6449 return true;
6453 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6455 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6457 BlockDriver *drv = bs->drv;
6459 if (drv && drv->bdrv_eject) {
6460 drv->bdrv_eject(bs, eject_flag);
6465 * Lock or unlock the media (if it is locked, the user won't be able
6466 * to eject it manually).
6468 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6470 BlockDriver *drv = bs->drv;
6472 trace_bdrv_lock_medium(bs, locked);
6474 if (drv && drv->bdrv_lock_medium) {
6475 drv->bdrv_lock_medium(bs, locked);
6479 /* Get a reference to bs */
6480 void bdrv_ref(BlockDriverState *bs)
6482 bs->refcnt++;
6485 /* Release a previously grabbed reference to bs.
6486 * If after releasing, reference count is zero, the BlockDriverState is
6487 * deleted. */
6488 void bdrv_unref(BlockDriverState *bs)
6490 if (!bs) {
6491 return;
6493 assert(bs->refcnt > 0);
6494 if (--bs->refcnt == 0) {
6495 bdrv_delete(bs);
6499 struct BdrvOpBlocker {
6500 Error *reason;
6501 QLIST_ENTRY(BdrvOpBlocker) list;
6504 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6506 BdrvOpBlocker *blocker;
6507 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6508 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6509 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6510 error_propagate_prepend(errp, error_copy(blocker->reason),
6511 "Node '%s' is busy: ",
6512 bdrv_get_device_or_node_name(bs));
6513 return true;
6515 return false;
6518 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6520 BdrvOpBlocker *blocker;
6521 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6523 blocker = g_new0(BdrvOpBlocker, 1);
6524 blocker->reason = reason;
6525 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6528 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6530 BdrvOpBlocker *blocker, *next;
6531 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6532 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6533 if (blocker->reason == reason) {
6534 QLIST_REMOVE(blocker, list);
6535 g_free(blocker);
6540 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6542 int i;
6543 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6544 bdrv_op_block(bs, i, reason);
6548 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6550 int i;
6551 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6552 bdrv_op_unblock(bs, i, reason);
6556 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6558 int i;
6560 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6561 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6562 return false;
6565 return true;
6568 void bdrv_img_create(const char *filename, const char *fmt,
6569 const char *base_filename, const char *base_fmt,
6570 char *options, uint64_t img_size, int flags, bool quiet,
6571 Error **errp)
6573 QemuOptsList *create_opts = NULL;
6574 QemuOpts *opts = NULL;
6575 const char *backing_fmt, *backing_file;
6576 int64_t size;
6577 BlockDriver *drv, *proto_drv;
6578 Error *local_err = NULL;
6579 int ret = 0;
6581 /* Find driver and parse its options */
6582 drv = bdrv_find_format(fmt);
6583 if (!drv) {
6584 error_setg(errp, "Unknown file format '%s'", fmt);
6585 return;
6588 proto_drv = bdrv_find_protocol(filename, true, errp);
6589 if (!proto_drv) {
6590 return;
6593 if (!drv->create_opts) {
6594 error_setg(errp, "Format driver '%s' does not support image creation",
6595 drv->format_name);
6596 return;
6599 if (!proto_drv->create_opts) {
6600 error_setg(errp, "Protocol driver '%s' does not support image creation",
6601 proto_drv->format_name);
6602 return;
6605 /* Create parameter list */
6606 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6607 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6609 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6611 /* Parse -o options */
6612 if (options) {
6613 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6614 goto out;
6618 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6619 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6620 } else if (img_size != UINT64_C(-1)) {
6621 error_setg(errp, "The image size must be specified only once");
6622 goto out;
6625 if (base_filename) {
6626 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6627 NULL)) {
6628 error_setg(errp, "Backing file not supported for file format '%s'",
6629 fmt);
6630 goto out;
6634 if (base_fmt) {
6635 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
6636 error_setg(errp, "Backing file format not supported for file "
6637 "format '%s'", fmt);
6638 goto out;
6642 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6643 if (backing_file) {
6644 if (!strcmp(filename, backing_file)) {
6645 error_setg(errp, "Error: Trying to create an image with the "
6646 "same filename as the backing file");
6647 goto out;
6649 if (backing_file[0] == '\0') {
6650 error_setg(errp, "Expected backing file name, got empty string");
6651 goto out;
6655 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
6657 /* The size for the image must always be specified, unless we have a backing
6658 * file and we have not been forbidden from opening it. */
6659 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6660 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6661 BlockDriverState *bs;
6662 char *full_backing;
6663 int back_flags;
6664 QDict *backing_options = NULL;
6666 full_backing =
6667 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6668 &local_err);
6669 if (local_err) {
6670 goto out;
6672 assert(full_backing);
6675 * No need to do I/O here, which allows us to open encrypted
6676 * backing images without needing the secret
6678 back_flags = flags;
6679 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
6680 back_flags |= BDRV_O_NO_IO;
6682 backing_options = qdict_new();
6683 if (backing_fmt) {
6684 qdict_put_str(backing_options, "driver", backing_fmt);
6686 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
6688 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6689 &local_err);
6690 g_free(full_backing);
6691 if (!bs) {
6692 error_append_hint(&local_err, "Could not open backing image.\n");
6693 goto out;
6694 } else {
6695 if (!backing_fmt) {
6696 error_setg(&local_err,
6697 "Backing file specified without backing format");
6698 error_append_hint(&local_err, "Detected format of %s.",
6699 bs->drv->format_name);
6700 goto out;
6702 if (size == -1) {
6703 /* Opened BS, have no size */
6704 size = bdrv_getlength(bs);
6705 if (size < 0) {
6706 error_setg_errno(errp, -size, "Could not get size of '%s'",
6707 backing_file);
6708 bdrv_unref(bs);
6709 goto out;
6711 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
6713 bdrv_unref(bs);
6715 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6716 } else if (backing_file && !backing_fmt) {
6717 error_setg(&local_err,
6718 "Backing file specified without backing format");
6719 goto out;
6722 if (size == -1) {
6723 error_setg(errp, "Image creation needs a size parameter");
6724 goto out;
6727 if (!quiet) {
6728 printf("Formatting '%s', fmt=%s ", filename, fmt);
6729 qemu_opts_print(opts, " ");
6730 puts("");
6731 fflush(stdout);
6734 ret = bdrv_create(drv, filename, opts, &local_err);
6736 if (ret == -EFBIG) {
6737 /* This is generally a better message than whatever the driver would
6738 * deliver (especially because of the cluster_size_hint), since that
6739 * is most probably not much different from "image too large". */
6740 const char *cluster_size_hint = "";
6741 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
6742 cluster_size_hint = " (try using a larger cluster size)";
6744 error_setg(errp, "The image size is too large for file format '%s'"
6745 "%s", fmt, cluster_size_hint);
6746 error_free(local_err);
6747 local_err = NULL;
6750 out:
6751 qemu_opts_del(opts);
6752 qemu_opts_free(create_opts);
6753 error_propagate(errp, local_err);
6756 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6758 return bs ? bs->aio_context : qemu_get_aio_context();
6761 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
6763 Coroutine *self = qemu_coroutine_self();
6764 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
6765 AioContext *new_ctx;
6768 * Increase bs->in_flight to ensure that this operation is completed before
6769 * moving the node to a different AioContext. Read new_ctx only afterwards.
6771 bdrv_inc_in_flight(bs);
6773 new_ctx = bdrv_get_aio_context(bs);
6774 aio_co_reschedule_self(new_ctx);
6775 return old_ctx;
6778 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
6780 aio_co_reschedule_self(old_ctx);
6781 bdrv_dec_in_flight(bs);
6784 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
6786 AioContext *ctx = bdrv_get_aio_context(bs);
6788 /* In the main thread, bs->aio_context won't change concurrently */
6789 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6792 * We're in coroutine context, so we already hold the lock of the main
6793 * loop AioContext. Don't lock it twice to avoid deadlocks.
6795 assert(qemu_in_coroutine());
6796 if (ctx != qemu_get_aio_context()) {
6797 aio_context_acquire(ctx);
6801 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
6803 AioContext *ctx = bdrv_get_aio_context(bs);
6805 assert(qemu_in_coroutine());
6806 if (ctx != qemu_get_aio_context()) {
6807 aio_context_release(ctx);
6811 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6813 aio_co_enter(bdrv_get_aio_context(bs), co);
6816 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6818 QLIST_REMOVE(ban, list);
6819 g_free(ban);
6822 static void bdrv_detach_aio_context(BlockDriverState *bs)
6824 BdrvAioNotifier *baf, *baf_tmp;
6826 assert(!bs->walking_aio_notifiers);
6827 bs->walking_aio_notifiers = true;
6828 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6829 if (baf->deleted) {
6830 bdrv_do_remove_aio_context_notifier(baf);
6831 } else {
6832 baf->detach_aio_context(baf->opaque);
6835 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6836 * remove remaining aio notifiers if we aren't called again.
6838 bs->walking_aio_notifiers = false;
6840 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
6841 bs->drv->bdrv_detach_aio_context(bs);
6844 if (bs->quiesce_counter) {
6845 aio_enable_external(bs->aio_context);
6847 bs->aio_context = NULL;
6850 static void bdrv_attach_aio_context(BlockDriverState *bs,
6851 AioContext *new_context)
6853 BdrvAioNotifier *ban, *ban_tmp;
6855 if (bs->quiesce_counter) {
6856 aio_disable_external(new_context);
6859 bs->aio_context = new_context;
6861 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
6862 bs->drv->bdrv_attach_aio_context(bs, new_context);
6865 assert(!bs->walking_aio_notifiers);
6866 bs->walking_aio_notifiers = true;
6867 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6868 if (ban->deleted) {
6869 bdrv_do_remove_aio_context_notifier(ban);
6870 } else {
6871 ban->attached_aio_context(new_context, ban->opaque);
6874 bs->walking_aio_notifiers = false;
6878 * Changes the AioContext used for fd handlers, timers, and BHs by this
6879 * BlockDriverState and all its children and parents.
6881 * Must be called from the main AioContext.
6883 * The caller must own the AioContext lock for the old AioContext of bs, but it
6884 * must not own the AioContext lock for new_context (unless new_context is the
6885 * same as the current context of bs).
6887 * @ignore will accumulate all visited BdrvChild object. The caller is
6888 * responsible for freeing the list afterwards.
6890 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6891 AioContext *new_context, GSList **ignore)
6893 AioContext *old_context = bdrv_get_aio_context(bs);
6894 GSList *children_to_process = NULL;
6895 GSList *parents_to_process = NULL;
6896 GSList *entry;
6897 BdrvChild *child, *parent;
6899 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6901 if (old_context == new_context) {
6902 return;
6905 bdrv_drained_begin(bs);
6907 QLIST_FOREACH(child, &bs->children, next) {
6908 if (g_slist_find(*ignore, child)) {
6909 continue;
6911 *ignore = g_slist_prepend(*ignore, child);
6912 children_to_process = g_slist_prepend(children_to_process, child);
6915 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6916 if (g_slist_find(*ignore, parent)) {
6917 continue;
6919 *ignore = g_slist_prepend(*ignore, parent);
6920 parents_to_process = g_slist_prepend(parents_to_process, parent);
6923 for (entry = children_to_process;
6924 entry != NULL;
6925 entry = g_slist_next(entry)) {
6926 child = entry->data;
6927 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6929 g_slist_free(children_to_process);
6931 for (entry = parents_to_process;
6932 entry != NULL;
6933 entry = g_slist_next(entry)) {
6934 parent = entry->data;
6935 assert(parent->klass->set_aio_ctx);
6936 parent->klass->set_aio_ctx(parent, new_context, ignore);
6938 g_slist_free(parents_to_process);
6940 bdrv_detach_aio_context(bs);
6942 /* Acquire the new context, if necessary */
6943 if (qemu_get_aio_context() != new_context) {
6944 aio_context_acquire(new_context);
6947 bdrv_attach_aio_context(bs, new_context);
6950 * If this function was recursively called from
6951 * bdrv_set_aio_context_ignore(), there may be nodes in the
6952 * subtree that have not yet been moved to the new AioContext.
6953 * Release the old one so bdrv_drained_end() can poll them.
6955 if (qemu_get_aio_context() != old_context) {
6956 aio_context_release(old_context);
6959 bdrv_drained_end(bs);
6961 if (qemu_get_aio_context() != old_context) {
6962 aio_context_acquire(old_context);
6964 if (qemu_get_aio_context() != new_context) {
6965 aio_context_release(new_context);
6969 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6970 GSList **ignore, Error **errp)
6972 if (g_slist_find(*ignore, c)) {
6973 return true;
6975 *ignore = g_slist_prepend(*ignore, c);
6978 * A BdrvChildClass that doesn't handle AioContext changes cannot
6979 * tolerate any AioContext changes
6981 if (!c->klass->can_set_aio_ctx) {
6982 char *user = bdrv_child_user_desc(c);
6983 error_setg(errp, "Changing iothreads is not supported by %s", user);
6984 g_free(user);
6985 return false;
6987 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
6988 assert(!errp || *errp);
6989 return false;
6991 return true;
6994 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6995 GSList **ignore, Error **errp)
6997 if (g_slist_find(*ignore, c)) {
6998 return true;
7000 *ignore = g_slist_prepend(*ignore, c);
7001 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
7004 /* @ignore will accumulate all visited BdrvChild object. The caller is
7005 * responsible for freeing the list afterwards. */
7006 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7007 GSList **ignore, Error **errp)
7009 BdrvChild *c;
7011 if (bdrv_get_aio_context(bs) == ctx) {
7012 return true;
7015 QLIST_FOREACH(c, &bs->parents, next_parent) {
7016 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
7017 return false;
7020 QLIST_FOREACH(c, &bs->children, next) {
7021 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
7022 return false;
7026 return true;
7029 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7030 BdrvChild *ignore_child, Error **errp)
7032 GSList *ignore;
7033 bool ret;
7035 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7036 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
7037 g_slist_free(ignore);
7039 if (!ret) {
7040 return -EPERM;
7043 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7044 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
7045 g_slist_free(ignore);
7047 return 0;
7050 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7051 Error **errp)
7053 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
7056 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7057 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7058 void (*detach_aio_context)(void *opaque), void *opaque)
7060 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7061 *ban = (BdrvAioNotifier){
7062 .attached_aio_context = attached_aio_context,
7063 .detach_aio_context = detach_aio_context,
7064 .opaque = opaque
7067 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7070 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7071 void (*attached_aio_context)(AioContext *,
7072 void *),
7073 void (*detach_aio_context)(void *),
7074 void *opaque)
7076 BdrvAioNotifier *ban, *ban_next;
7078 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7079 if (ban->attached_aio_context == attached_aio_context &&
7080 ban->detach_aio_context == detach_aio_context &&
7081 ban->opaque == opaque &&
7082 ban->deleted == false)
7084 if (bs->walking_aio_notifiers) {
7085 ban->deleted = true;
7086 } else {
7087 bdrv_do_remove_aio_context_notifier(ban);
7089 return;
7093 abort();
7096 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7097 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7098 bool force,
7099 Error **errp)
7101 if (!bs->drv) {
7102 error_setg(errp, "Node is ejected");
7103 return -ENOMEDIUM;
7105 if (!bs->drv->bdrv_amend_options) {
7106 error_setg(errp, "Block driver '%s' does not support option amendment",
7107 bs->drv->format_name);
7108 return -ENOTSUP;
7110 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7111 cb_opaque, force, errp);
7115 * This function checks whether the given @to_replace is allowed to be
7116 * replaced by a node that always shows the same data as @bs. This is
7117 * used for example to verify whether the mirror job can replace
7118 * @to_replace by the target mirrored from @bs.
7119 * To be replaceable, @bs and @to_replace may either be guaranteed to
7120 * always show the same data (because they are only connected through
7121 * filters), or some driver may allow replacing one of its children
7122 * because it can guarantee that this child's data is not visible at
7123 * all (for example, for dissenting quorum children that have no other
7124 * parents).
7126 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7127 BlockDriverState *to_replace)
7129 BlockDriverState *filtered;
7131 if (!bs || !bs->drv) {
7132 return false;
7135 if (bs == to_replace) {
7136 return true;
7139 /* See what the driver can do */
7140 if (bs->drv->bdrv_recurse_can_replace) {
7141 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7144 /* For filters without an own implementation, we can recurse on our own */
7145 filtered = bdrv_filter_bs(bs);
7146 if (filtered) {
7147 return bdrv_recurse_can_replace(filtered, to_replace);
7150 /* Safe default */
7151 return false;
7155 * Check whether the given @node_name can be replaced by a node that
7156 * has the same data as @parent_bs. If so, return @node_name's BDS;
7157 * NULL otherwise.
7159 * @node_name must be a (recursive) *child of @parent_bs (or this
7160 * function will return NULL).
7162 * The result (whether the node can be replaced or not) is only valid
7163 * for as long as no graph or permission changes occur.
7165 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7166 const char *node_name, Error **errp)
7168 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7169 AioContext *aio_context;
7171 if (!to_replace_bs) {
7172 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7173 return NULL;
7176 aio_context = bdrv_get_aio_context(to_replace_bs);
7177 aio_context_acquire(aio_context);
7179 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7180 to_replace_bs = NULL;
7181 goto out;
7184 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7185 * most non filter in order to prevent data corruption.
7186 * Another benefit is that this tests exclude backing files which are
7187 * blocked by the backing blockers.
7189 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7190 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7191 "because it cannot be guaranteed that doing so would not "
7192 "lead to an abrupt change of visible data",
7193 node_name, parent_bs->node_name);
7194 to_replace_bs = NULL;
7195 goto out;
7198 out:
7199 aio_context_release(aio_context);
7200 return to_replace_bs;
7204 * Iterates through the list of runtime option keys that are said to
7205 * be "strong" for a BDS. An option is called "strong" if it changes
7206 * a BDS's data. For example, the null block driver's "size" and
7207 * "read-zeroes" options are strong, but its "latency-ns" option is
7208 * not.
7210 * If a key returned by this function ends with a dot, all options
7211 * starting with that prefix are strong.
7213 static const char *const *strong_options(BlockDriverState *bs,
7214 const char *const *curopt)
7216 static const char *const global_options[] = {
7217 "driver", "filename", NULL
7220 if (!curopt) {
7221 return &global_options[0];
7224 curopt++;
7225 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7226 curopt = bs->drv->strong_runtime_opts;
7229 return (curopt && *curopt) ? curopt : NULL;
7233 * Copies all strong runtime options from bs->options to the given
7234 * QDict. The set of strong option keys is determined by invoking
7235 * strong_options().
7237 * Returns true iff any strong option was present in bs->options (and
7238 * thus copied to the target QDict) with the exception of "filename"
7239 * and "driver". The caller is expected to use this value to decide
7240 * whether the existence of strong options prevents the generation of
7241 * a plain filename.
7243 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7245 bool found_any = false;
7246 const char *const *option_name = NULL;
7248 if (!bs->drv) {
7249 return false;
7252 while ((option_name = strong_options(bs, option_name))) {
7253 bool option_given = false;
7255 assert(strlen(*option_name) > 0);
7256 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7257 QObject *entry = qdict_get(bs->options, *option_name);
7258 if (!entry) {
7259 continue;
7262 qdict_put_obj(d, *option_name, qobject_ref(entry));
7263 option_given = true;
7264 } else {
7265 const QDictEntry *entry;
7266 for (entry = qdict_first(bs->options); entry;
7267 entry = qdict_next(bs->options, entry))
7269 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7270 qdict_put_obj(d, qdict_entry_key(entry),
7271 qobject_ref(qdict_entry_value(entry)));
7272 option_given = true;
7277 /* While "driver" and "filename" need to be included in a JSON filename,
7278 * their existence does not prohibit generation of a plain filename. */
7279 if (!found_any && option_given &&
7280 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7282 found_any = true;
7286 if (!qdict_haskey(d, "driver")) {
7287 /* Drivers created with bdrv_new_open_driver() may not have a
7288 * @driver option. Add it here. */
7289 qdict_put_str(d, "driver", bs->drv->format_name);
7292 return found_any;
7295 /* Note: This function may return false positives; it may return true
7296 * even if opening the backing file specified by bs's image header
7297 * would result in exactly bs->backing. */
7298 bool bdrv_backing_overridden(BlockDriverState *bs)
7300 if (bs->backing) {
7301 return strcmp(bs->auto_backing_file,
7302 bs->backing->bs->filename);
7303 } else {
7304 /* No backing BDS, so if the image header reports any backing
7305 * file, it must have been suppressed */
7306 return bs->auto_backing_file[0] != '\0';
7310 /* Updates the following BDS fields:
7311 * - exact_filename: A filename which may be used for opening a block device
7312 * which (mostly) equals the given BDS (even without any
7313 * other options; so reading and writing must return the same
7314 * results, but caching etc. may be different)
7315 * - full_open_options: Options which, when given when opening a block device
7316 * (without a filename), result in a BDS (mostly)
7317 * equalling the given one
7318 * - filename: If exact_filename is set, it is copied here. Otherwise,
7319 * full_open_options is converted to a JSON object, prefixed with
7320 * "json:" (for use through the JSON pseudo protocol) and put here.
7322 void bdrv_refresh_filename(BlockDriverState *bs)
7324 BlockDriver *drv = bs->drv;
7325 BdrvChild *child;
7326 BlockDriverState *primary_child_bs;
7327 QDict *opts;
7328 bool backing_overridden;
7329 bool generate_json_filename; /* Whether our default implementation should
7330 fill exact_filename (false) or not (true) */
7332 if (!drv) {
7333 return;
7336 /* This BDS's file name may depend on any of its children's file names, so
7337 * refresh those first */
7338 QLIST_FOREACH(child, &bs->children, next) {
7339 bdrv_refresh_filename(child->bs);
7342 if (bs->implicit) {
7343 /* For implicit nodes, just copy everything from the single child */
7344 child = QLIST_FIRST(&bs->children);
7345 assert(QLIST_NEXT(child, next) == NULL);
7347 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7348 child->bs->exact_filename);
7349 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7351 qobject_unref(bs->full_open_options);
7352 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7354 return;
7357 backing_overridden = bdrv_backing_overridden(bs);
7359 if (bs->open_flags & BDRV_O_NO_IO) {
7360 /* Without I/O, the backing file does not change anything.
7361 * Therefore, in such a case (primarily qemu-img), we can
7362 * pretend the backing file has not been overridden even if
7363 * it technically has been. */
7364 backing_overridden = false;
7367 /* Gather the options QDict */
7368 opts = qdict_new();
7369 generate_json_filename = append_strong_runtime_options(opts, bs);
7370 generate_json_filename |= backing_overridden;
7372 if (drv->bdrv_gather_child_options) {
7373 /* Some block drivers may not want to present all of their children's
7374 * options, or name them differently from BdrvChild.name */
7375 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7376 } else {
7377 QLIST_FOREACH(child, &bs->children, next) {
7378 if (child == bs->backing && !backing_overridden) {
7379 /* We can skip the backing BDS if it has not been overridden */
7380 continue;
7383 qdict_put(opts, child->name,
7384 qobject_ref(child->bs->full_open_options));
7387 if (backing_overridden && !bs->backing) {
7388 /* Force no backing file */
7389 qdict_put_null(opts, "backing");
7393 qobject_unref(bs->full_open_options);
7394 bs->full_open_options = opts;
7396 primary_child_bs = bdrv_primary_bs(bs);
7398 if (drv->bdrv_refresh_filename) {
7399 /* Obsolete information is of no use here, so drop the old file name
7400 * information before refreshing it */
7401 bs->exact_filename[0] = '\0';
7403 drv->bdrv_refresh_filename(bs);
7404 } else if (primary_child_bs) {
7406 * Try to reconstruct valid information from the underlying
7407 * file -- this only works for format nodes (filter nodes
7408 * cannot be probed and as such must be selected by the user
7409 * either through an options dict, or through a special
7410 * filename which the filter driver must construct in its
7411 * .bdrv_refresh_filename() implementation).
7414 bs->exact_filename[0] = '\0';
7417 * We can use the underlying file's filename if:
7418 * - it has a filename,
7419 * - the current BDS is not a filter,
7420 * - the file is a protocol BDS, and
7421 * - opening that file (as this BDS's format) will automatically create
7422 * the BDS tree we have right now, that is:
7423 * - the user did not significantly change this BDS's behavior with
7424 * some explicit (strong) options
7425 * - no non-file child of this BDS has been overridden by the user
7426 * Both of these conditions are represented by generate_json_filename.
7428 if (primary_child_bs->exact_filename[0] &&
7429 primary_child_bs->drv->bdrv_file_open &&
7430 !drv->is_filter && !generate_json_filename)
7432 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7436 if (bs->exact_filename[0]) {
7437 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7438 } else {
7439 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7440 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7441 json->str) >= sizeof(bs->filename)) {
7442 /* Give user a hint if we truncated things. */
7443 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7445 g_string_free(json, true);
7449 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7451 BlockDriver *drv = bs->drv;
7452 BlockDriverState *child_bs;
7454 if (!drv) {
7455 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7456 return NULL;
7459 if (drv->bdrv_dirname) {
7460 return drv->bdrv_dirname(bs, errp);
7463 child_bs = bdrv_primary_bs(bs);
7464 if (child_bs) {
7465 return bdrv_dirname(child_bs, errp);
7468 bdrv_refresh_filename(bs);
7469 if (bs->exact_filename[0] != '\0') {
7470 return path_combine(bs->exact_filename, "");
7473 error_setg(errp, "Cannot generate a base directory for %s nodes",
7474 drv->format_name);
7475 return NULL;
7479 * Hot add/remove a BDS's child. So the user can take a child offline when
7480 * it is broken and take a new child online
7482 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7483 Error **errp)
7486 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7487 error_setg(errp, "The node %s does not support adding a child",
7488 bdrv_get_device_or_node_name(parent_bs));
7489 return;
7492 if (!QLIST_EMPTY(&child_bs->parents)) {
7493 error_setg(errp, "The node %s already has a parent",
7494 child_bs->node_name);
7495 return;
7498 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7501 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7503 BdrvChild *tmp;
7505 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7506 error_setg(errp, "The node %s does not support removing a child",
7507 bdrv_get_device_or_node_name(parent_bs));
7508 return;
7511 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7512 if (tmp == child) {
7513 break;
7517 if (!tmp) {
7518 error_setg(errp, "The node %s does not have a child named %s",
7519 bdrv_get_device_or_node_name(parent_bs),
7520 bdrv_get_device_or_node_name(child->bs));
7521 return;
7524 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7527 int bdrv_make_empty(BdrvChild *c, Error **errp)
7529 BlockDriver *drv = c->bs->drv;
7530 int ret;
7532 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7534 if (!drv->bdrv_make_empty) {
7535 error_setg(errp, "%s does not support emptying nodes",
7536 drv->format_name);
7537 return -ENOTSUP;
7540 ret = drv->bdrv_make_empty(c->bs);
7541 if (ret < 0) {
7542 error_setg_errno(errp, -ret, "Failed to empty %s",
7543 c->bs->filename);
7544 return ret;
7547 return 0;
7551 * Return the child that @bs acts as an overlay for, and from which data may be
7552 * copied in COW or COR operations. Usually this is the backing file.
7554 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7556 if (!bs || !bs->drv) {
7557 return NULL;
7560 if (bs->drv->is_filter) {
7561 return NULL;
7564 if (!bs->backing) {
7565 return NULL;
7568 assert(bs->backing->role & BDRV_CHILD_COW);
7569 return bs->backing;
7573 * If @bs acts as a filter for exactly one of its children, return
7574 * that child.
7576 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7578 BdrvChild *c;
7580 if (!bs || !bs->drv) {
7581 return NULL;
7584 if (!bs->drv->is_filter) {
7585 return NULL;
7588 /* Only one of @backing or @file may be used */
7589 assert(!(bs->backing && bs->file));
7591 c = bs->backing ?: bs->file;
7592 if (!c) {
7593 return NULL;
7596 assert(c->role & BDRV_CHILD_FILTERED);
7597 return c;
7601 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7602 * whichever is non-NULL.
7604 * Return NULL if both are NULL.
7606 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7608 BdrvChild *cow_child = bdrv_cow_child(bs);
7609 BdrvChild *filter_child = bdrv_filter_child(bs);
7611 /* Filter nodes cannot have COW backing files */
7612 assert(!(cow_child && filter_child));
7614 return cow_child ?: filter_child;
7618 * Return the primary child of this node: For filters, that is the
7619 * filtered child. For other nodes, that is usually the child storing
7620 * metadata.
7621 * (A generally more helpful description is that this is (usually) the
7622 * child that has the same filename as @bs.)
7624 * Drivers do not necessarily have a primary child; for example quorum
7625 * does not.
7627 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
7629 BdrvChild *c, *found = NULL;
7631 QLIST_FOREACH(c, &bs->children, next) {
7632 if (c->role & BDRV_CHILD_PRIMARY) {
7633 assert(!found);
7634 found = c;
7638 return found;
7641 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
7642 bool stop_on_explicit_filter)
7644 BdrvChild *c;
7646 if (!bs) {
7647 return NULL;
7650 while (!(stop_on_explicit_filter && !bs->implicit)) {
7651 c = bdrv_filter_child(bs);
7652 if (!c) {
7654 * A filter that is embedded in a working block graph must
7655 * have a child. Assert this here so this function does
7656 * not return a filter node that is not expected by the
7657 * caller.
7659 assert(!bs->drv || !bs->drv->is_filter);
7660 break;
7662 bs = c->bs;
7665 * Note that this treats nodes with bs->drv == NULL as not being
7666 * filters (bs->drv == NULL should be replaced by something else
7667 * anyway).
7668 * The advantage of this behavior is that this function will thus
7669 * always return a non-NULL value (given a non-NULL @bs).
7672 return bs;
7676 * Return the first BDS that has not been added implicitly or that
7677 * does not have a filtered child down the chain starting from @bs
7678 * (including @bs itself).
7680 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
7682 return bdrv_do_skip_filters(bs, true);
7686 * Return the first BDS that does not have a filtered child down the
7687 * chain starting from @bs (including @bs itself).
7689 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
7691 return bdrv_do_skip_filters(bs, false);
7695 * For a backing chain, return the first non-filter backing image of
7696 * the first non-filter image.
7698 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
7700 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
7704 * Check whether [offset, offset + bytes) overlaps with the cached
7705 * block-status data region.
7707 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
7708 * which is what bdrv_bsc_is_data()'s interface needs.
7709 * Otherwise, *pnum is not touched.
7711 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
7712 int64_t offset, int64_t bytes,
7713 int64_t *pnum)
7715 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
7716 bool overlaps;
7718 overlaps =
7719 qatomic_read(&bsc->valid) &&
7720 ranges_overlap(offset, bytes, bsc->data_start,
7721 bsc->data_end - bsc->data_start);
7723 if (overlaps && pnum) {
7724 *pnum = bsc->data_end - offset;
7727 return overlaps;
7731 * See block_int.h for this function's documentation.
7733 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
7735 RCU_READ_LOCK_GUARD();
7737 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
7741 * See block_int.h for this function's documentation.
7743 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
7744 int64_t offset, int64_t bytes)
7746 RCU_READ_LOCK_GUARD();
7748 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
7749 qatomic_set(&bs->block_status_cache->valid, false);
7754 * See block_int.h for this function's documentation.
7756 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
7758 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
7759 BdrvBlockStatusCache *old_bsc;
7761 *new_bsc = (BdrvBlockStatusCache) {
7762 .valid = true,
7763 .data_start = offset,
7764 .data_end = offset + bytes,
7767 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
7769 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
7770 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
7771 if (old_bsc) {
7772 g_free_rcu(old_bsc, rcu);