seccomp: allow action to be customized per syscall
[qemu/armbru.git] / block.c
blobb54d59d1fa2a9d2920569065cfb0565deb317f63
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 bool bdrv_recurse_has_child(BlockDriverState *bs,
88 BlockDriverState *child);
90 static void bdrv_child_free(BdrvChild *child);
91 static void bdrv_replace_child_noperm(BdrvChild **child,
92 BlockDriverState *new_bs,
93 bool free_empty_child);
94 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
95 BdrvChild *child,
96 Transaction *tran);
97 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
98 Transaction *tran);
100 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
101 BlockReopenQueue *queue,
102 Transaction *change_child_tran, Error **errp);
103 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
104 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
106 static bool bdrv_backing_overridden(BlockDriverState *bs);
108 /* If non-zero, use only whitelisted block drivers */
109 static int use_bdrv_whitelist;
111 #ifdef _WIN32
112 static int is_windows_drive_prefix(const char *filename)
114 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
115 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
116 filename[1] == ':');
119 int is_windows_drive(const char *filename)
121 if (is_windows_drive_prefix(filename) &&
122 filename[2] == '\0')
123 return 1;
124 if (strstart(filename, "\\\\.\\", NULL) ||
125 strstart(filename, "//./", NULL))
126 return 1;
127 return 0;
129 #endif
131 size_t bdrv_opt_mem_align(BlockDriverState *bs)
133 if (!bs || !bs->drv) {
134 /* page size or 4k (hdd sector size) should be on the safe side */
135 return MAX(4096, qemu_real_host_page_size);
138 return bs->bl.opt_mem_alignment;
141 size_t bdrv_min_mem_align(BlockDriverState *bs)
143 if (!bs || !bs->drv) {
144 /* page size or 4k (hdd sector size) should be on the safe side */
145 return MAX(4096, qemu_real_host_page_size);
148 return bs->bl.min_mem_alignment;
151 /* check if the path starts with "<protocol>:" */
152 int path_has_protocol(const char *path)
154 const char *p;
156 #ifdef _WIN32
157 if (is_windows_drive(path) ||
158 is_windows_drive_prefix(path)) {
159 return 0;
161 p = path + strcspn(path, ":/\\");
162 #else
163 p = path + strcspn(path, ":/");
164 #endif
166 return *p == ':';
169 int path_is_absolute(const char *path)
171 #ifdef _WIN32
172 /* specific case for names like: "\\.\d:" */
173 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
174 return 1;
176 return (*path == '/' || *path == '\\');
177 #else
178 return (*path == '/');
179 #endif
182 /* if filename is absolute, just return its duplicate. Otherwise, build a
183 path to it by considering it is relative to base_path. URL are
184 supported. */
185 char *path_combine(const char *base_path, const char *filename)
187 const char *protocol_stripped = NULL;
188 const char *p, *p1;
189 char *result;
190 int len;
192 if (path_is_absolute(filename)) {
193 return g_strdup(filename);
196 if (path_has_protocol(base_path)) {
197 protocol_stripped = strchr(base_path, ':');
198 if (protocol_stripped) {
199 protocol_stripped++;
202 p = protocol_stripped ?: base_path;
204 p1 = strrchr(base_path, '/');
205 #ifdef _WIN32
207 const char *p2;
208 p2 = strrchr(base_path, '\\');
209 if (!p1 || p2 > p1) {
210 p1 = p2;
213 #endif
214 if (p1) {
215 p1++;
216 } else {
217 p1 = base_path;
219 if (p1 > p) {
220 p = p1;
222 len = p - base_path;
224 result = g_malloc(len + strlen(filename) + 1);
225 memcpy(result, base_path, len);
226 strcpy(result + len, filename);
228 return result;
232 * Helper function for bdrv_parse_filename() implementations to remove optional
233 * protocol prefixes (especially "file:") from a filename and for putting the
234 * stripped filename into the options QDict if there is such a prefix.
236 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
237 QDict *options)
239 if (strstart(filename, prefix, &filename)) {
240 /* Stripping the explicit protocol prefix may result in a protocol
241 * prefix being (wrongly) detected (if the filename contains a colon) */
242 if (path_has_protocol(filename)) {
243 GString *fat_filename;
245 /* This means there is some colon before the first slash; therefore,
246 * this cannot be an absolute path */
247 assert(!path_is_absolute(filename));
249 /* And we can thus fix the protocol detection issue by prefixing it
250 * by "./" */
251 fat_filename = g_string_new("./");
252 g_string_append(fat_filename, filename);
254 assert(!path_has_protocol(fat_filename->str));
256 qdict_put(options, "filename",
257 qstring_from_gstring(fat_filename));
258 } else {
259 /* If no protocol prefix was detected, we can use the shortened
260 * filename as-is */
261 qdict_put_str(options, "filename", filename);
267 /* Returns whether the image file is opened as read-only. Note that this can
268 * return false and writing to the image file is still not possible because the
269 * image is inactivated. */
270 bool bdrv_is_read_only(BlockDriverState *bs)
272 return !(bs->open_flags & BDRV_O_RDWR);
275 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
276 bool ignore_allow_rdw, Error **errp)
278 /* Do not set read_only if copy_on_read is enabled */
279 if (bs->copy_on_read && read_only) {
280 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
281 bdrv_get_device_or_node_name(bs));
282 return -EINVAL;
285 /* Do not clear read_only if it is prohibited */
286 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
287 !ignore_allow_rdw)
289 error_setg(errp, "Node '%s' is read only",
290 bdrv_get_device_or_node_name(bs));
291 return -EPERM;
294 return 0;
298 * Called by a driver that can only provide a read-only image.
300 * Returns 0 if the node is already read-only or it could switch the node to
301 * read-only because BDRV_O_AUTO_RDONLY is set.
303 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
304 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
305 * is not NULL, it is used as the error message for the Error object.
307 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
308 Error **errp)
310 int ret = 0;
312 if (!(bs->open_flags & BDRV_O_RDWR)) {
313 return 0;
315 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
316 goto fail;
319 ret = bdrv_can_set_read_only(bs, true, false, NULL);
320 if (ret < 0) {
321 goto fail;
324 bs->open_flags &= ~BDRV_O_RDWR;
326 return 0;
328 fail:
329 error_setg(errp, "%s", errmsg ?: "Image is read-only");
330 return -EACCES;
334 * If @backing is empty, this function returns NULL without setting
335 * @errp. In all other cases, NULL will only be returned with @errp
336 * set.
338 * Therefore, a return value of NULL without @errp set means that
339 * there is no backing file; if @errp is set, there is one but its
340 * absolute filename cannot be generated.
342 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
343 const char *backing,
344 Error **errp)
346 if (backing[0] == '\0') {
347 return NULL;
348 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
349 return g_strdup(backing);
350 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
351 error_setg(errp, "Cannot use relative backing file names for '%s'",
352 backed);
353 return NULL;
354 } else {
355 return path_combine(backed, backing);
360 * If @filename is empty or NULL, this function returns NULL without
361 * setting @errp. In all other cases, NULL will only be returned with
362 * @errp set.
364 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
365 const char *filename, Error **errp)
367 char *dir, *full_name;
369 if (!filename || filename[0] == '\0') {
370 return NULL;
371 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
372 return g_strdup(filename);
375 dir = bdrv_dirname(relative_to, errp);
376 if (!dir) {
377 return NULL;
380 full_name = g_strconcat(dir, filename, NULL);
381 g_free(dir);
382 return full_name;
385 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
387 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
390 void bdrv_register(BlockDriver *bdrv)
392 assert(bdrv->format_name);
393 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
396 BlockDriverState *bdrv_new(void)
398 BlockDriverState *bs;
399 int i;
401 bs = g_new0(BlockDriverState, 1);
402 QLIST_INIT(&bs->dirty_bitmaps);
403 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
404 QLIST_INIT(&bs->op_blockers[i]);
406 qemu_co_mutex_init(&bs->reqs_lock);
407 qemu_mutex_init(&bs->dirty_bitmap_mutex);
408 bs->refcnt = 1;
409 bs->aio_context = qemu_get_aio_context();
411 qemu_co_queue_init(&bs->flush_queue);
413 qemu_co_mutex_init(&bs->bsc_modify_lock);
414 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
416 for (i = 0; i < bdrv_drain_all_count; i++) {
417 bdrv_drained_begin(bs);
420 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
422 return bs;
425 static BlockDriver *bdrv_do_find_format(const char *format_name)
427 BlockDriver *drv1;
429 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
430 if (!strcmp(drv1->format_name, format_name)) {
431 return drv1;
435 return NULL;
438 BlockDriver *bdrv_find_format(const char *format_name)
440 BlockDriver *drv1;
441 int i;
443 drv1 = bdrv_do_find_format(format_name);
444 if (drv1) {
445 return drv1;
448 /* The driver isn't registered, maybe we need to load a module */
449 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
450 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
451 block_module_load_one(block_driver_modules[i].library_name);
452 break;
456 return bdrv_do_find_format(format_name);
459 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
461 static const char *whitelist_rw[] = {
462 CONFIG_BDRV_RW_WHITELIST
463 NULL
465 static const char *whitelist_ro[] = {
466 CONFIG_BDRV_RO_WHITELIST
467 NULL
469 const char **p;
471 if (!whitelist_rw[0] && !whitelist_ro[0]) {
472 return 1; /* no whitelist, anything goes */
475 for (p = whitelist_rw; *p; p++) {
476 if (!strcmp(format_name, *p)) {
477 return 1;
480 if (read_only) {
481 for (p = whitelist_ro; *p; p++) {
482 if (!strcmp(format_name, *p)) {
483 return 1;
487 return 0;
490 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
492 return bdrv_format_is_whitelisted(drv->format_name, read_only);
495 bool bdrv_uses_whitelist(void)
497 return use_bdrv_whitelist;
500 typedef struct CreateCo {
501 BlockDriver *drv;
502 char *filename;
503 QemuOpts *opts;
504 int ret;
505 Error *err;
506 } CreateCo;
508 static void coroutine_fn bdrv_create_co_entry(void *opaque)
510 Error *local_err = NULL;
511 int ret;
513 CreateCo *cco = opaque;
514 assert(cco->drv);
516 ret = cco->drv->bdrv_co_create_opts(cco->drv,
517 cco->filename, cco->opts, &local_err);
518 error_propagate(&cco->err, local_err);
519 cco->ret = ret;
522 int bdrv_create(BlockDriver *drv, const char* filename,
523 QemuOpts *opts, Error **errp)
525 int ret;
527 Coroutine *co;
528 CreateCo cco = {
529 .drv = drv,
530 .filename = g_strdup(filename),
531 .opts = opts,
532 .ret = NOT_DONE,
533 .err = NULL,
536 if (!drv->bdrv_co_create_opts) {
537 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
538 ret = -ENOTSUP;
539 goto out;
542 if (qemu_in_coroutine()) {
543 /* Fast-path if already in coroutine context */
544 bdrv_create_co_entry(&cco);
545 } else {
546 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
547 qemu_coroutine_enter(co);
548 while (cco.ret == NOT_DONE) {
549 aio_poll(qemu_get_aio_context(), true);
553 ret = cco.ret;
554 if (ret < 0) {
555 if (cco.err) {
556 error_propagate(errp, cco.err);
557 } else {
558 error_setg_errno(errp, -ret, "Could not create image");
562 out:
563 g_free(cco.filename);
564 return ret;
568 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
569 * least the given @minimum_size.
571 * On success, return @blk's actual length.
572 * Otherwise, return -errno.
574 static int64_t create_file_fallback_truncate(BlockBackend *blk,
575 int64_t minimum_size, Error **errp)
577 Error *local_err = NULL;
578 int64_t size;
579 int ret;
581 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
582 &local_err);
583 if (ret < 0 && ret != -ENOTSUP) {
584 error_propagate(errp, local_err);
585 return ret;
588 size = blk_getlength(blk);
589 if (size < 0) {
590 error_free(local_err);
591 error_setg_errno(errp, -size,
592 "Failed to inquire the new image file's length");
593 return size;
596 if (size < minimum_size) {
597 /* Need to grow the image, but we failed to do that */
598 error_propagate(errp, local_err);
599 return -ENOTSUP;
602 error_free(local_err);
603 local_err = NULL;
605 return size;
609 * Helper function for bdrv_create_file_fallback(): Zero the first
610 * sector to remove any potentially pre-existing image header.
612 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
613 int64_t current_size,
614 Error **errp)
616 int64_t bytes_to_clear;
617 int ret;
619 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
620 if (bytes_to_clear) {
621 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
622 if (ret < 0) {
623 error_setg_errno(errp, -ret,
624 "Failed to clear the new image's first sector");
625 return ret;
629 return 0;
633 * Simple implementation of bdrv_co_create_opts for protocol drivers
634 * which only support creation via opening a file
635 * (usually existing raw storage device)
637 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
638 const char *filename,
639 QemuOpts *opts,
640 Error **errp)
642 BlockBackend *blk;
643 QDict *options;
644 int64_t size = 0;
645 char *buf = NULL;
646 PreallocMode prealloc;
647 Error *local_err = NULL;
648 int ret;
650 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
651 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
652 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
653 PREALLOC_MODE_OFF, &local_err);
654 g_free(buf);
655 if (local_err) {
656 error_propagate(errp, local_err);
657 return -EINVAL;
660 if (prealloc != PREALLOC_MODE_OFF) {
661 error_setg(errp, "Unsupported preallocation mode '%s'",
662 PreallocMode_str(prealloc));
663 return -ENOTSUP;
666 options = qdict_new();
667 qdict_put_str(options, "driver", drv->format_name);
669 blk = blk_new_open(filename, NULL, options,
670 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
671 if (!blk) {
672 error_prepend(errp, "Protocol driver '%s' does not support image "
673 "creation, and opening the image failed: ",
674 drv->format_name);
675 return -EINVAL;
678 size = create_file_fallback_truncate(blk, size, errp);
679 if (size < 0) {
680 ret = size;
681 goto out;
684 ret = create_file_fallback_zero_first_sector(blk, size, errp);
685 if (ret < 0) {
686 goto out;
689 ret = 0;
690 out:
691 blk_unref(blk);
692 return ret;
695 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
697 QemuOpts *protocol_opts;
698 BlockDriver *drv;
699 QDict *qdict;
700 int ret;
702 drv = bdrv_find_protocol(filename, true, errp);
703 if (drv == NULL) {
704 return -ENOENT;
707 if (!drv->create_opts) {
708 error_setg(errp, "Driver '%s' does not support image creation",
709 drv->format_name);
710 return -ENOTSUP;
714 * 'opts' contains a QemuOptsList with a combination of format and protocol
715 * default values.
717 * The format properly removes its options, but the default values remain
718 * in 'opts->list'. So if the protocol has options with the same name
719 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
720 * of the format, since for overlapping options, the format wins.
722 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
723 * only the set options, and then convert it back to QemuOpts, using the
724 * create_opts of the protocol. So the new QemuOpts, will contain only the
725 * protocol defaults.
727 qdict = qemu_opts_to_qdict(opts, NULL);
728 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
729 if (protocol_opts == NULL) {
730 ret = -EINVAL;
731 goto out;
734 ret = bdrv_create(drv, filename, protocol_opts, errp);
735 out:
736 qemu_opts_del(protocol_opts);
737 qobject_unref(qdict);
738 return ret;
741 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
743 Error *local_err = NULL;
744 int ret;
746 assert(bs != NULL);
748 if (!bs->drv) {
749 error_setg(errp, "Block node '%s' is not opened", bs->filename);
750 return -ENOMEDIUM;
753 if (!bs->drv->bdrv_co_delete_file) {
754 error_setg(errp, "Driver '%s' does not support image deletion",
755 bs->drv->format_name);
756 return -ENOTSUP;
759 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
760 if (ret < 0) {
761 error_propagate(errp, local_err);
764 return ret;
767 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
769 Error *local_err = NULL;
770 int ret;
772 if (!bs) {
773 return;
776 ret = bdrv_co_delete_file(bs, &local_err);
778 * ENOTSUP will happen if the block driver doesn't support
779 * the 'bdrv_co_delete_file' interface. This is a predictable
780 * scenario and shouldn't be reported back to the user.
782 if (ret == -ENOTSUP) {
783 error_free(local_err);
784 } else if (ret < 0) {
785 error_report_err(local_err);
790 * Try to get @bs's logical and physical block size.
791 * On success, store them in @bsz struct and return 0.
792 * On failure return -errno.
793 * @bs must not be empty.
795 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
797 BlockDriver *drv = bs->drv;
798 BlockDriverState *filtered = bdrv_filter_bs(bs);
800 if (drv && drv->bdrv_probe_blocksizes) {
801 return drv->bdrv_probe_blocksizes(bs, bsz);
802 } else if (filtered) {
803 return bdrv_probe_blocksizes(filtered, bsz);
806 return -ENOTSUP;
810 * Try to get @bs's geometry (cyls, heads, sectors).
811 * On success, store them in @geo struct and return 0.
812 * On failure return -errno.
813 * @bs must not be empty.
815 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
817 BlockDriver *drv = bs->drv;
818 BlockDriverState *filtered = bdrv_filter_bs(bs);
820 if (drv && drv->bdrv_probe_geometry) {
821 return drv->bdrv_probe_geometry(bs, geo);
822 } else if (filtered) {
823 return bdrv_probe_geometry(filtered, geo);
826 return -ENOTSUP;
830 * Create a uniquely-named empty temporary file.
831 * Return 0 upon success, otherwise a negative errno value.
833 int get_tmp_filename(char *filename, int size)
835 #ifdef _WIN32
836 char temp_dir[MAX_PATH];
837 /* GetTempFileName requires that its output buffer (4th param)
838 have length MAX_PATH or greater. */
839 assert(size >= MAX_PATH);
840 return (GetTempPath(MAX_PATH, temp_dir)
841 && GetTempFileName(temp_dir, "qem", 0, filename)
842 ? 0 : -GetLastError());
843 #else
844 int fd;
845 const char *tmpdir;
846 tmpdir = getenv("TMPDIR");
847 if (!tmpdir) {
848 tmpdir = "/var/tmp";
850 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
851 return -EOVERFLOW;
853 fd = mkstemp(filename);
854 if (fd < 0) {
855 return -errno;
857 if (close(fd) != 0) {
858 unlink(filename);
859 return -errno;
861 return 0;
862 #endif
866 * Detect host devices. By convention, /dev/cdrom[N] is always
867 * recognized as a host CDROM.
869 static BlockDriver *find_hdev_driver(const char *filename)
871 int score_max = 0, score;
872 BlockDriver *drv = NULL, *d;
874 QLIST_FOREACH(d, &bdrv_drivers, list) {
875 if (d->bdrv_probe_device) {
876 score = d->bdrv_probe_device(filename);
877 if (score > score_max) {
878 score_max = score;
879 drv = d;
884 return drv;
887 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
889 BlockDriver *drv1;
891 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
892 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
893 return drv1;
897 return NULL;
900 BlockDriver *bdrv_find_protocol(const char *filename,
901 bool allow_protocol_prefix,
902 Error **errp)
904 BlockDriver *drv1;
905 char protocol[128];
906 int len;
907 const char *p;
908 int i;
910 /* TODO Drivers without bdrv_file_open must be specified explicitly */
913 * XXX(hch): we really should not let host device detection
914 * override an explicit protocol specification, but moving this
915 * later breaks access to device names with colons in them.
916 * Thanks to the brain-dead persistent naming schemes on udev-
917 * based Linux systems those actually are quite common.
919 drv1 = find_hdev_driver(filename);
920 if (drv1) {
921 return drv1;
924 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
925 return &bdrv_file;
928 p = strchr(filename, ':');
929 assert(p != NULL);
930 len = p - filename;
931 if (len > sizeof(protocol) - 1)
932 len = sizeof(protocol) - 1;
933 memcpy(protocol, filename, len);
934 protocol[len] = '\0';
936 drv1 = bdrv_do_find_protocol(protocol);
937 if (drv1) {
938 return drv1;
941 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
942 if (block_driver_modules[i].protocol_name &&
943 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
944 block_module_load_one(block_driver_modules[i].library_name);
945 break;
949 drv1 = bdrv_do_find_protocol(protocol);
950 if (!drv1) {
951 error_setg(errp, "Unknown protocol '%s'", protocol);
953 return drv1;
957 * Guess image format by probing its contents.
958 * This is not a good idea when your image is raw (CVE-2008-2004), but
959 * we do it anyway for backward compatibility.
961 * @buf contains the image's first @buf_size bytes.
962 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
963 * but can be smaller if the image file is smaller)
964 * @filename is its filename.
966 * For all block drivers, call the bdrv_probe() method to get its
967 * probing score.
968 * Return the first block driver with the highest probing score.
970 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
971 const char *filename)
973 int score_max = 0, score;
974 BlockDriver *drv = NULL, *d;
976 QLIST_FOREACH(d, &bdrv_drivers, list) {
977 if (d->bdrv_probe) {
978 score = d->bdrv_probe(buf, buf_size, filename);
979 if (score > score_max) {
980 score_max = score;
981 drv = d;
986 return drv;
989 static int find_image_format(BlockBackend *file, const char *filename,
990 BlockDriver **pdrv, Error **errp)
992 BlockDriver *drv;
993 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
994 int ret = 0;
996 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
997 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
998 *pdrv = &bdrv_raw;
999 return ret;
1002 ret = blk_pread(file, 0, buf, sizeof(buf));
1003 if (ret < 0) {
1004 error_setg_errno(errp, -ret, "Could not read image for determining its "
1005 "format");
1006 *pdrv = NULL;
1007 return ret;
1010 drv = bdrv_probe_all(buf, ret, filename);
1011 if (!drv) {
1012 error_setg(errp, "Could not determine image format: No compatible "
1013 "driver found");
1014 ret = -ENOENT;
1016 *pdrv = drv;
1017 return ret;
1021 * Set the current 'total_sectors' value
1022 * Return 0 on success, -errno on error.
1024 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
1026 BlockDriver *drv = bs->drv;
1028 if (!drv) {
1029 return -ENOMEDIUM;
1032 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1033 if (bdrv_is_sg(bs))
1034 return 0;
1036 /* query actual device if possible, otherwise just trust the hint */
1037 if (drv->bdrv_getlength) {
1038 int64_t length = drv->bdrv_getlength(bs);
1039 if (length < 0) {
1040 return length;
1042 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1045 bs->total_sectors = hint;
1047 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1048 return -EFBIG;
1051 return 0;
1055 * Combines a QDict of new block driver @options with any missing options taken
1056 * from @old_options, so that leaving out an option defaults to its old value.
1058 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1059 QDict *old_options)
1061 if (bs->drv && bs->drv->bdrv_join_options) {
1062 bs->drv->bdrv_join_options(options, old_options);
1063 } else {
1064 qdict_join(options, old_options, false);
1068 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1069 int open_flags,
1070 Error **errp)
1072 Error *local_err = NULL;
1073 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1074 BlockdevDetectZeroesOptions detect_zeroes =
1075 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1076 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1077 g_free(value);
1078 if (local_err) {
1079 error_propagate(errp, local_err);
1080 return detect_zeroes;
1083 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1084 !(open_flags & BDRV_O_UNMAP))
1086 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1087 "without setting discard operation to unmap");
1090 return detect_zeroes;
1094 * Set open flags for aio engine
1096 * Return 0 on success, -1 if the engine specified is invalid
1098 int bdrv_parse_aio(const char *mode, int *flags)
1100 if (!strcmp(mode, "threads")) {
1101 /* do nothing, default */
1102 } else if (!strcmp(mode, "native")) {
1103 *flags |= BDRV_O_NATIVE_AIO;
1104 #ifdef CONFIG_LINUX_IO_URING
1105 } else if (!strcmp(mode, "io_uring")) {
1106 *flags |= BDRV_O_IO_URING;
1107 #endif
1108 } else {
1109 return -1;
1112 return 0;
1116 * Set open flags for a given discard mode
1118 * Return 0 on success, -1 if the discard mode was invalid.
1120 int bdrv_parse_discard_flags(const char *mode, int *flags)
1122 *flags &= ~BDRV_O_UNMAP;
1124 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1125 /* do nothing */
1126 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1127 *flags |= BDRV_O_UNMAP;
1128 } else {
1129 return -1;
1132 return 0;
1136 * Set open flags for a given cache mode
1138 * Return 0 on success, -1 if the cache mode was invalid.
1140 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1142 *flags &= ~BDRV_O_CACHE_MASK;
1144 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1145 *writethrough = false;
1146 *flags |= BDRV_O_NOCACHE;
1147 } else if (!strcmp(mode, "directsync")) {
1148 *writethrough = true;
1149 *flags |= BDRV_O_NOCACHE;
1150 } else if (!strcmp(mode, "writeback")) {
1151 *writethrough = false;
1152 } else if (!strcmp(mode, "unsafe")) {
1153 *writethrough = false;
1154 *flags |= BDRV_O_NO_FLUSH;
1155 } else if (!strcmp(mode, "writethrough")) {
1156 *writethrough = true;
1157 } else {
1158 return -1;
1161 return 0;
1164 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1166 BlockDriverState *parent = c->opaque;
1167 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1170 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1172 BlockDriverState *bs = child->opaque;
1173 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1176 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1178 BlockDriverState *bs = child->opaque;
1179 return bdrv_drain_poll(bs, false, NULL, false);
1182 static void bdrv_child_cb_drained_end(BdrvChild *child,
1183 int *drained_end_counter)
1185 BlockDriverState *bs = child->opaque;
1186 bdrv_drained_end_no_poll(bs, drained_end_counter);
1189 static int bdrv_child_cb_inactivate(BdrvChild *child)
1191 BlockDriverState *bs = child->opaque;
1192 assert(bs->open_flags & BDRV_O_INACTIVE);
1193 return 0;
1196 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1197 GSList **ignore, Error **errp)
1199 BlockDriverState *bs = child->opaque;
1200 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1203 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1204 GSList **ignore)
1206 BlockDriverState *bs = child->opaque;
1207 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1211 * Returns the options and flags that a temporary snapshot should get, based on
1212 * the originally requested flags (the originally requested image will have
1213 * flags like a backing file)
1215 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1216 int parent_flags, QDict *parent_options)
1218 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1220 /* For temporary files, unconditional cache=unsafe is fine */
1221 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1222 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1224 /* Copy the read-only and discard options from the parent */
1225 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1226 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1228 /* aio=native doesn't work for cache.direct=off, so disable it for the
1229 * temporary snapshot */
1230 *child_flags &= ~BDRV_O_NATIVE_AIO;
1233 static void bdrv_backing_attach(BdrvChild *c)
1235 BlockDriverState *parent = c->opaque;
1236 BlockDriverState *backing_hd = c->bs;
1238 assert(!parent->backing_blocker);
1239 error_setg(&parent->backing_blocker,
1240 "node is used as backing hd of '%s'",
1241 bdrv_get_device_or_node_name(parent));
1243 bdrv_refresh_filename(backing_hd);
1245 parent->open_flags &= ~BDRV_O_NO_BACKING;
1247 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1248 /* Otherwise we won't be able to commit or stream */
1249 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1250 parent->backing_blocker);
1251 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1252 parent->backing_blocker);
1254 * We do backup in 3 ways:
1255 * 1. drive backup
1256 * The target bs is new opened, and the source is top BDS
1257 * 2. blockdev backup
1258 * Both the source and the target are top BDSes.
1259 * 3. internal backup(used for block replication)
1260 * Both the source and the target are backing file
1262 * In case 1 and 2, neither the source nor the target is the backing file.
1263 * In case 3, we will block the top BDS, so there is only one block job
1264 * for the top BDS and its backing chain.
1266 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1267 parent->backing_blocker);
1268 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1269 parent->backing_blocker);
1272 static void bdrv_backing_detach(BdrvChild *c)
1274 BlockDriverState *parent = c->opaque;
1276 assert(parent->backing_blocker);
1277 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1278 error_free(parent->backing_blocker);
1279 parent->backing_blocker = NULL;
1282 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1283 const char *filename, Error **errp)
1285 BlockDriverState *parent = c->opaque;
1286 bool read_only = bdrv_is_read_only(parent);
1287 int ret;
1289 if (read_only) {
1290 ret = bdrv_reopen_set_read_only(parent, false, errp);
1291 if (ret < 0) {
1292 return ret;
1296 ret = bdrv_change_backing_file(parent, filename,
1297 base->drv ? base->drv->format_name : "",
1298 false);
1299 if (ret < 0) {
1300 error_setg_errno(errp, -ret, "Could not update backing file link");
1303 if (read_only) {
1304 bdrv_reopen_set_read_only(parent, true, NULL);
1307 return ret;
1311 * Returns the options and flags that a generic child of a BDS should
1312 * get, based on the given options and flags for the parent BDS.
1314 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1315 int *child_flags, QDict *child_options,
1316 int parent_flags, QDict *parent_options)
1318 int flags = parent_flags;
1321 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1322 * Generally, the question to answer is: Should this child be
1323 * format-probed by default?
1327 * Pure and non-filtered data children of non-format nodes should
1328 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1329 * set). This only affects a very limited set of drivers (namely
1330 * quorum and blkverify when this comment was written).
1331 * Force-clear BDRV_O_PROTOCOL then.
1333 if (!parent_is_format &&
1334 (role & BDRV_CHILD_DATA) &&
1335 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1337 flags &= ~BDRV_O_PROTOCOL;
1341 * All children of format nodes (except for COW children) and all
1342 * metadata children in general should never be format-probed.
1343 * Force-set BDRV_O_PROTOCOL then.
1345 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1346 (role & BDRV_CHILD_METADATA))
1348 flags |= BDRV_O_PROTOCOL;
1352 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1353 * the parent.
1355 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1356 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1357 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1359 if (role & BDRV_CHILD_COW) {
1360 /* backing files are opened read-only by default */
1361 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1362 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1363 } else {
1364 /* Inherit the read-only option from the parent if it's not set */
1365 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1366 qdict_copy_default(child_options, parent_options,
1367 BDRV_OPT_AUTO_READ_ONLY);
1371 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1372 * can default to enable it on lower layers regardless of the
1373 * parent option.
1375 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1377 /* Clear flags that only apply to the top layer */
1378 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1380 if (role & BDRV_CHILD_METADATA) {
1381 flags &= ~BDRV_O_NO_IO;
1383 if (role & BDRV_CHILD_COW) {
1384 flags &= ~BDRV_O_TEMPORARY;
1387 *child_flags = flags;
1390 static void bdrv_child_cb_attach(BdrvChild *child)
1392 BlockDriverState *bs = child->opaque;
1394 QLIST_INSERT_HEAD(&bs->children, child, next);
1396 if (child->role & BDRV_CHILD_COW) {
1397 bdrv_backing_attach(child);
1400 bdrv_apply_subtree_drain(child, bs);
1403 static void bdrv_child_cb_detach(BdrvChild *child)
1405 BlockDriverState *bs = child->opaque;
1407 if (child->role & BDRV_CHILD_COW) {
1408 bdrv_backing_detach(child);
1411 bdrv_unapply_subtree_drain(child, bs);
1413 QLIST_REMOVE(child, next);
1416 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1417 const char *filename, Error **errp)
1419 if (c->role & BDRV_CHILD_COW) {
1420 return bdrv_backing_update_filename(c, base, filename, errp);
1422 return 0;
1425 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1427 BlockDriverState *bs = c->opaque;
1429 return bdrv_get_aio_context(bs);
1432 const BdrvChildClass child_of_bds = {
1433 .parent_is_bds = true,
1434 .get_parent_desc = bdrv_child_get_parent_desc,
1435 .inherit_options = bdrv_inherited_options,
1436 .drained_begin = bdrv_child_cb_drained_begin,
1437 .drained_poll = bdrv_child_cb_drained_poll,
1438 .drained_end = bdrv_child_cb_drained_end,
1439 .attach = bdrv_child_cb_attach,
1440 .detach = bdrv_child_cb_detach,
1441 .inactivate = bdrv_child_cb_inactivate,
1442 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1443 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1444 .update_filename = bdrv_child_cb_update_filename,
1445 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1448 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1450 return c->klass->get_parent_aio_context(c);
1453 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1455 int open_flags = flags;
1458 * Clear flags that are internal to the block layer before opening the
1459 * image.
1461 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1463 return open_flags;
1466 static void update_flags_from_options(int *flags, QemuOpts *opts)
1468 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1470 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1471 *flags |= BDRV_O_NO_FLUSH;
1474 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1475 *flags |= BDRV_O_NOCACHE;
1478 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1479 *flags |= BDRV_O_RDWR;
1482 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1483 *flags |= BDRV_O_AUTO_RDONLY;
1487 static void update_options_from_flags(QDict *options, int flags)
1489 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1490 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1492 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1493 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1494 flags & BDRV_O_NO_FLUSH);
1496 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1497 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1499 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1500 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1501 flags & BDRV_O_AUTO_RDONLY);
1505 static void bdrv_assign_node_name(BlockDriverState *bs,
1506 const char *node_name,
1507 Error **errp)
1509 char *gen_node_name = NULL;
1511 if (!node_name) {
1512 node_name = gen_node_name = id_generate(ID_BLOCK);
1513 } else if (!id_wellformed(node_name)) {
1515 * Check for empty string or invalid characters, but not if it is
1516 * generated (generated names use characters not available to the user)
1518 error_setg(errp, "Invalid node-name: '%s'", node_name);
1519 return;
1522 /* takes care of avoiding namespaces collisions */
1523 if (blk_by_name(node_name)) {
1524 error_setg(errp, "node-name=%s is conflicting with a device id",
1525 node_name);
1526 goto out;
1529 /* takes care of avoiding duplicates node names */
1530 if (bdrv_find_node(node_name)) {
1531 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1532 goto out;
1535 /* Make sure that the node name isn't truncated */
1536 if (strlen(node_name) >= sizeof(bs->node_name)) {
1537 error_setg(errp, "Node name too long");
1538 goto out;
1541 /* copy node name into the bs and insert it into the graph list */
1542 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1543 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1544 out:
1545 g_free(gen_node_name);
1548 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1549 const char *node_name, QDict *options,
1550 int open_flags, Error **errp)
1552 Error *local_err = NULL;
1553 int i, ret;
1555 bdrv_assign_node_name(bs, node_name, &local_err);
1556 if (local_err) {
1557 error_propagate(errp, local_err);
1558 return -EINVAL;
1561 bs->drv = drv;
1562 bs->opaque = g_malloc0(drv->instance_size);
1564 if (drv->bdrv_file_open) {
1565 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1566 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1567 } else if (drv->bdrv_open) {
1568 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1569 } else {
1570 ret = 0;
1573 if (ret < 0) {
1574 if (local_err) {
1575 error_propagate(errp, local_err);
1576 } else if (bs->filename[0]) {
1577 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1578 } else {
1579 error_setg_errno(errp, -ret, "Could not open image");
1581 goto open_failed;
1584 ret = refresh_total_sectors(bs, bs->total_sectors);
1585 if (ret < 0) {
1586 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1587 return ret;
1590 bdrv_refresh_limits(bs, NULL, &local_err);
1591 if (local_err) {
1592 error_propagate(errp, local_err);
1593 return -EINVAL;
1596 assert(bdrv_opt_mem_align(bs) != 0);
1597 assert(bdrv_min_mem_align(bs) != 0);
1598 assert(is_power_of_2(bs->bl.request_alignment));
1600 for (i = 0; i < bs->quiesce_counter; i++) {
1601 if (drv->bdrv_co_drain_begin) {
1602 drv->bdrv_co_drain_begin(bs);
1606 return 0;
1607 open_failed:
1608 bs->drv = NULL;
1609 if (bs->file != NULL) {
1610 bdrv_unref_child(bs, bs->file);
1611 bs->file = NULL;
1613 g_free(bs->opaque);
1614 bs->opaque = NULL;
1615 return ret;
1619 * Create and open a block node.
1621 * @options is a QDict of options to pass to the block drivers, or NULL for an
1622 * empty set of options. The reference to the QDict belongs to the block layer
1623 * after the call (even on failure), so if the caller intends to reuse the
1624 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1626 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1627 const char *node_name,
1628 QDict *options, int flags,
1629 Error **errp)
1631 BlockDriverState *bs;
1632 int ret;
1634 bs = bdrv_new();
1635 bs->open_flags = flags;
1636 bs->options = options ?: qdict_new();
1637 bs->explicit_options = qdict_clone_shallow(bs->options);
1638 bs->opaque = NULL;
1640 update_options_from_flags(bs->options, flags);
1642 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1643 if (ret < 0) {
1644 qobject_unref(bs->explicit_options);
1645 bs->explicit_options = NULL;
1646 qobject_unref(bs->options);
1647 bs->options = NULL;
1648 bdrv_unref(bs);
1649 return NULL;
1652 return bs;
1655 /* Create and open a block node. */
1656 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1657 int flags, Error **errp)
1659 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1662 QemuOptsList bdrv_runtime_opts = {
1663 .name = "bdrv_common",
1664 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1665 .desc = {
1667 .name = "node-name",
1668 .type = QEMU_OPT_STRING,
1669 .help = "Node name of the block device node",
1672 .name = "driver",
1673 .type = QEMU_OPT_STRING,
1674 .help = "Block driver to use for the node",
1677 .name = BDRV_OPT_CACHE_DIRECT,
1678 .type = QEMU_OPT_BOOL,
1679 .help = "Bypass software writeback cache on the host",
1682 .name = BDRV_OPT_CACHE_NO_FLUSH,
1683 .type = QEMU_OPT_BOOL,
1684 .help = "Ignore flush requests",
1687 .name = BDRV_OPT_READ_ONLY,
1688 .type = QEMU_OPT_BOOL,
1689 .help = "Node is opened in read-only mode",
1692 .name = BDRV_OPT_AUTO_READ_ONLY,
1693 .type = QEMU_OPT_BOOL,
1694 .help = "Node can become read-only if opening read-write fails",
1697 .name = "detect-zeroes",
1698 .type = QEMU_OPT_STRING,
1699 .help = "try to optimize zero writes (off, on, unmap)",
1702 .name = BDRV_OPT_DISCARD,
1703 .type = QEMU_OPT_STRING,
1704 .help = "discard operation (ignore/off, unmap/on)",
1707 .name = BDRV_OPT_FORCE_SHARE,
1708 .type = QEMU_OPT_BOOL,
1709 .help = "always accept other writers (default: off)",
1711 { /* end of list */ }
1715 QemuOptsList bdrv_create_opts_simple = {
1716 .name = "simple-create-opts",
1717 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1718 .desc = {
1720 .name = BLOCK_OPT_SIZE,
1721 .type = QEMU_OPT_SIZE,
1722 .help = "Virtual disk size"
1725 .name = BLOCK_OPT_PREALLOC,
1726 .type = QEMU_OPT_STRING,
1727 .help = "Preallocation mode (allowed values: off)"
1729 { /* end of list */ }
1734 * Common part for opening disk images and files
1736 * Removes all processed options from *options.
1738 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1739 QDict *options, Error **errp)
1741 int ret, open_flags;
1742 const char *filename;
1743 const char *driver_name = NULL;
1744 const char *node_name = NULL;
1745 const char *discard;
1746 QemuOpts *opts;
1747 BlockDriver *drv;
1748 Error *local_err = NULL;
1749 bool ro;
1751 assert(bs->file == NULL);
1752 assert(options != NULL && bs->options != options);
1754 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1755 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1756 ret = -EINVAL;
1757 goto fail_opts;
1760 update_flags_from_options(&bs->open_flags, opts);
1762 driver_name = qemu_opt_get(opts, "driver");
1763 drv = bdrv_find_format(driver_name);
1764 assert(drv != NULL);
1766 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1768 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1769 error_setg(errp,
1770 BDRV_OPT_FORCE_SHARE
1771 "=on can only be used with read-only images");
1772 ret = -EINVAL;
1773 goto fail_opts;
1776 if (file != NULL) {
1777 bdrv_refresh_filename(blk_bs(file));
1778 filename = blk_bs(file)->filename;
1779 } else {
1781 * Caution: while qdict_get_try_str() is fine, getting
1782 * non-string types would require more care. When @options
1783 * come from -blockdev or blockdev_add, its members are typed
1784 * according to the QAPI schema, but when they come from
1785 * -drive, they're all QString.
1787 filename = qdict_get_try_str(options, "filename");
1790 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1791 error_setg(errp, "The '%s' block driver requires a file name",
1792 drv->format_name);
1793 ret = -EINVAL;
1794 goto fail_opts;
1797 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1798 drv->format_name);
1800 ro = bdrv_is_read_only(bs);
1802 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1803 if (!ro && bdrv_is_whitelisted(drv, true)) {
1804 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1805 } else {
1806 ret = -ENOTSUP;
1808 if (ret < 0) {
1809 error_setg(errp,
1810 !ro && bdrv_is_whitelisted(drv, true)
1811 ? "Driver '%s' can only be used for read-only devices"
1812 : "Driver '%s' is not whitelisted",
1813 drv->format_name);
1814 goto fail_opts;
1818 /* bdrv_new() and bdrv_close() make it so */
1819 assert(qatomic_read(&bs->copy_on_read) == 0);
1821 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1822 if (!ro) {
1823 bdrv_enable_copy_on_read(bs);
1824 } else {
1825 error_setg(errp, "Can't use copy-on-read on read-only device");
1826 ret = -EINVAL;
1827 goto fail_opts;
1831 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1832 if (discard != NULL) {
1833 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1834 error_setg(errp, "Invalid discard option");
1835 ret = -EINVAL;
1836 goto fail_opts;
1840 bs->detect_zeroes =
1841 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1842 if (local_err) {
1843 error_propagate(errp, local_err);
1844 ret = -EINVAL;
1845 goto fail_opts;
1848 if (filename != NULL) {
1849 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1850 } else {
1851 bs->filename[0] = '\0';
1853 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1855 /* Open the image, either directly or using a protocol */
1856 open_flags = bdrv_open_flags(bs, bs->open_flags);
1857 node_name = qemu_opt_get(opts, "node-name");
1859 assert(!drv->bdrv_file_open || file == NULL);
1860 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1861 if (ret < 0) {
1862 goto fail_opts;
1865 qemu_opts_del(opts);
1866 return 0;
1868 fail_opts:
1869 qemu_opts_del(opts);
1870 return ret;
1873 static QDict *parse_json_filename(const char *filename, Error **errp)
1875 QObject *options_obj;
1876 QDict *options;
1877 int ret;
1879 ret = strstart(filename, "json:", &filename);
1880 assert(ret);
1882 options_obj = qobject_from_json(filename, errp);
1883 if (!options_obj) {
1884 error_prepend(errp, "Could not parse the JSON options: ");
1885 return NULL;
1888 options = qobject_to(QDict, options_obj);
1889 if (!options) {
1890 qobject_unref(options_obj);
1891 error_setg(errp, "Invalid JSON object given");
1892 return NULL;
1895 qdict_flatten(options);
1897 return options;
1900 static void parse_json_protocol(QDict *options, const char **pfilename,
1901 Error **errp)
1903 QDict *json_options;
1904 Error *local_err = NULL;
1906 /* Parse json: pseudo-protocol */
1907 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1908 return;
1911 json_options = parse_json_filename(*pfilename, &local_err);
1912 if (local_err) {
1913 error_propagate(errp, local_err);
1914 return;
1917 /* Options given in the filename have lower priority than options
1918 * specified directly */
1919 qdict_join(options, json_options, false);
1920 qobject_unref(json_options);
1921 *pfilename = NULL;
1925 * Fills in default options for opening images and converts the legacy
1926 * filename/flags pair to option QDict entries.
1927 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1928 * block driver has been specified explicitly.
1930 static int bdrv_fill_options(QDict **options, const char *filename,
1931 int *flags, Error **errp)
1933 const char *drvname;
1934 bool protocol = *flags & BDRV_O_PROTOCOL;
1935 bool parse_filename = false;
1936 BlockDriver *drv = NULL;
1937 Error *local_err = NULL;
1940 * Caution: while qdict_get_try_str() is fine, getting non-string
1941 * types would require more care. When @options come from
1942 * -blockdev or blockdev_add, its members are typed according to
1943 * the QAPI schema, but when they come from -drive, they're all
1944 * QString.
1946 drvname = qdict_get_try_str(*options, "driver");
1947 if (drvname) {
1948 drv = bdrv_find_format(drvname);
1949 if (!drv) {
1950 error_setg(errp, "Unknown driver '%s'", drvname);
1951 return -ENOENT;
1953 /* If the user has explicitly specified the driver, this choice should
1954 * override the BDRV_O_PROTOCOL flag */
1955 protocol = drv->bdrv_file_open;
1958 if (protocol) {
1959 *flags |= BDRV_O_PROTOCOL;
1960 } else {
1961 *flags &= ~BDRV_O_PROTOCOL;
1964 /* Translate cache options from flags into options */
1965 update_options_from_flags(*options, *flags);
1967 /* Fetch the file name from the options QDict if necessary */
1968 if (protocol && filename) {
1969 if (!qdict_haskey(*options, "filename")) {
1970 qdict_put_str(*options, "filename", filename);
1971 parse_filename = true;
1972 } else {
1973 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1974 "the same time");
1975 return -EINVAL;
1979 /* Find the right block driver */
1980 /* See cautionary note on accessing @options above */
1981 filename = qdict_get_try_str(*options, "filename");
1983 if (!drvname && protocol) {
1984 if (filename) {
1985 drv = bdrv_find_protocol(filename, parse_filename, errp);
1986 if (!drv) {
1987 return -EINVAL;
1990 drvname = drv->format_name;
1991 qdict_put_str(*options, "driver", drvname);
1992 } else {
1993 error_setg(errp, "Must specify either driver or file");
1994 return -EINVAL;
1998 assert(drv || !protocol);
2000 /* Driver-specific filename parsing */
2001 if (drv && drv->bdrv_parse_filename && parse_filename) {
2002 drv->bdrv_parse_filename(filename, *options, &local_err);
2003 if (local_err) {
2004 error_propagate(errp, local_err);
2005 return -EINVAL;
2008 if (!drv->bdrv_needs_filename) {
2009 qdict_del(*options, "filename");
2013 return 0;
2016 typedef struct BlockReopenQueueEntry {
2017 bool prepared;
2018 bool perms_checked;
2019 BDRVReopenState state;
2020 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2021 } BlockReopenQueueEntry;
2024 * Return the flags that @bs will have after the reopens in @q have
2025 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2026 * return the current flags.
2028 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2030 BlockReopenQueueEntry *entry;
2032 if (q != NULL) {
2033 QTAILQ_FOREACH(entry, q, entry) {
2034 if (entry->state.bs == bs) {
2035 return entry->state.flags;
2040 return bs->open_flags;
2043 /* Returns whether the image file can be written to after the reopen queue @q
2044 * has been successfully applied, or right now if @q is NULL. */
2045 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2046 BlockReopenQueue *q)
2048 int flags = bdrv_reopen_get_flags(q, bs);
2050 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2054 * Return whether the BDS can be written to. This is not necessarily
2055 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2056 * be written to but do not count as read-only images.
2058 bool bdrv_is_writable(BlockDriverState *bs)
2060 return bdrv_is_writable_after_reopen(bs, NULL);
2063 static char *bdrv_child_user_desc(BdrvChild *c)
2065 return c->klass->get_parent_desc(c);
2069 * Check that @a allows everything that @b needs. @a and @b must reference same
2070 * child node.
2072 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2074 const char *child_bs_name;
2075 g_autofree char *a_user = NULL;
2076 g_autofree char *b_user = NULL;
2077 g_autofree char *perms = NULL;
2079 assert(a->bs);
2080 assert(a->bs == b->bs);
2082 if ((b->perm & a->shared_perm) == b->perm) {
2083 return true;
2086 child_bs_name = bdrv_get_node_name(b->bs);
2087 a_user = bdrv_child_user_desc(a);
2088 b_user = bdrv_child_user_desc(b);
2089 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2091 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2092 "both required by %s (uses node '%s' as '%s' child) and "
2093 "unshared by %s (uses node '%s' as '%s' child).",
2094 child_bs_name, perms,
2095 b_user, child_bs_name, b->name,
2096 a_user, child_bs_name, a->name);
2098 return false;
2101 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2103 BdrvChild *a, *b;
2106 * During the loop we'll look at each pair twice. That's correct because
2107 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2108 * directions.
2110 QLIST_FOREACH(a, &bs->parents, next_parent) {
2111 QLIST_FOREACH(b, &bs->parents, next_parent) {
2112 if (a == b) {
2113 continue;
2116 if (!bdrv_a_allow_b(a, b, errp)) {
2117 return true;
2122 return false;
2125 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2126 BdrvChild *c, BdrvChildRole role,
2127 BlockReopenQueue *reopen_queue,
2128 uint64_t parent_perm, uint64_t parent_shared,
2129 uint64_t *nperm, uint64_t *nshared)
2131 assert(bs->drv && bs->drv->bdrv_child_perm);
2132 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2133 parent_perm, parent_shared,
2134 nperm, nshared);
2135 /* TODO Take force_share from reopen_queue */
2136 if (child_bs && child_bs->force_share) {
2137 *nshared = BLK_PERM_ALL;
2142 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2143 * nodes that are already in the @list, of course) so that final list is
2144 * topologically sorted. Return the result (GSList @list object is updated, so
2145 * don't use old reference after function call).
2147 * On function start @list must be already topologically sorted and for any node
2148 * in the @list the whole subtree of the node must be in the @list as well. The
2149 * simplest way to satisfy this criteria: use only result of
2150 * bdrv_topological_dfs() or NULL as @list parameter.
2152 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2153 BlockDriverState *bs)
2155 BdrvChild *child;
2156 g_autoptr(GHashTable) local_found = NULL;
2158 if (!found) {
2159 assert(!list);
2160 found = local_found = g_hash_table_new(NULL, NULL);
2163 if (g_hash_table_contains(found, bs)) {
2164 return list;
2166 g_hash_table_add(found, bs);
2168 QLIST_FOREACH(child, &bs->children, next) {
2169 list = bdrv_topological_dfs(list, found, child->bs);
2172 return g_slist_prepend(list, bs);
2175 typedef struct BdrvChildSetPermState {
2176 BdrvChild *child;
2177 uint64_t old_perm;
2178 uint64_t old_shared_perm;
2179 } BdrvChildSetPermState;
2181 static void bdrv_child_set_perm_abort(void *opaque)
2183 BdrvChildSetPermState *s = opaque;
2185 s->child->perm = s->old_perm;
2186 s->child->shared_perm = s->old_shared_perm;
2189 static TransactionActionDrv bdrv_child_set_pem_drv = {
2190 .abort = bdrv_child_set_perm_abort,
2191 .clean = g_free,
2194 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2195 uint64_t shared, Transaction *tran)
2197 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2199 *s = (BdrvChildSetPermState) {
2200 .child = c,
2201 .old_perm = c->perm,
2202 .old_shared_perm = c->shared_perm,
2205 c->perm = perm;
2206 c->shared_perm = shared;
2208 tran_add(tran, &bdrv_child_set_pem_drv, s);
2211 static void bdrv_drv_set_perm_commit(void *opaque)
2213 BlockDriverState *bs = opaque;
2214 uint64_t cumulative_perms, cumulative_shared_perms;
2216 if (bs->drv->bdrv_set_perm) {
2217 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2218 &cumulative_shared_perms);
2219 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2223 static void bdrv_drv_set_perm_abort(void *opaque)
2225 BlockDriverState *bs = opaque;
2227 if (bs->drv->bdrv_abort_perm_update) {
2228 bs->drv->bdrv_abort_perm_update(bs);
2232 TransactionActionDrv bdrv_drv_set_perm_drv = {
2233 .abort = bdrv_drv_set_perm_abort,
2234 .commit = bdrv_drv_set_perm_commit,
2237 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2238 uint64_t shared_perm, Transaction *tran,
2239 Error **errp)
2241 if (!bs->drv) {
2242 return 0;
2245 if (bs->drv->bdrv_check_perm) {
2246 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2247 if (ret < 0) {
2248 return ret;
2252 if (tran) {
2253 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2256 return 0;
2259 typedef struct BdrvReplaceChildState {
2260 BdrvChild *child;
2261 BdrvChild **childp;
2262 BlockDriverState *old_bs;
2263 bool free_empty_child;
2264 } BdrvReplaceChildState;
2266 static void bdrv_replace_child_commit(void *opaque)
2268 BdrvReplaceChildState *s = opaque;
2270 if (s->free_empty_child && !s->child->bs) {
2271 bdrv_child_free(s->child);
2273 bdrv_unref(s->old_bs);
2276 static void bdrv_replace_child_abort(void *opaque)
2278 BdrvReplaceChildState *s = opaque;
2279 BlockDriverState *new_bs = s->child->bs;
2282 * old_bs reference is transparently moved from @s to s->child.
2284 * Pass &s->child here instead of s->childp, because:
2285 * (1) s->old_bs must be non-NULL, so bdrv_replace_child_noperm() will not
2286 * modify the BdrvChild * pointer we indirectly pass to it, i.e. it
2287 * will not modify s->child. From that perspective, it does not matter
2288 * whether we pass s->childp or &s->child.
2289 * (2) If new_bs is not NULL, s->childp will be NULL. We then cannot use
2290 * it here.
2291 * (3) If new_bs is NULL, *s->childp will have been NULLed by
2292 * bdrv_replace_child_tran()'s bdrv_replace_child_noperm() call, and we
2293 * must not pass a NULL *s->childp here.
2295 * So whether new_bs was NULL or not, we cannot pass s->childp here; and in
2296 * any case, there is no reason to pass it anyway.
2298 bdrv_replace_child_noperm(&s->child, s->old_bs, true);
2300 * The child was pre-existing, so s->old_bs must be non-NULL, and
2301 * s->child thus must not have been freed
2303 assert(s->child != NULL);
2304 if (!new_bs) {
2305 /* As described above, *s->childp was cleared, so restore it */
2306 assert(s->childp != NULL);
2307 *s->childp = s->child;
2309 bdrv_unref(new_bs);
2312 static TransactionActionDrv bdrv_replace_child_drv = {
2313 .commit = bdrv_replace_child_commit,
2314 .abort = bdrv_replace_child_abort,
2315 .clean = g_free,
2319 * bdrv_replace_child_tran
2321 * Note: real unref of old_bs is done only on commit.
2323 * The function doesn't update permissions, caller is responsible for this.
2325 * (*childp)->bs must not be NULL.
2327 * Note that if new_bs == NULL, @childp is stored in a state object attached
2328 * to @tran, so that the old child can be reinstated in the abort handler.
2329 * Therefore, if @new_bs can be NULL, @childp must stay valid until the
2330 * transaction is committed or aborted.
2332 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2333 * freed (on commit). @free_empty_child should only be false if the
2334 * caller will free the BDrvChild themselves (which may be important
2335 * if this is in turn called in another transactional context).
2337 static void bdrv_replace_child_tran(BdrvChild **childp,
2338 BlockDriverState *new_bs,
2339 Transaction *tran,
2340 bool free_empty_child)
2342 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2343 *s = (BdrvReplaceChildState) {
2344 .child = *childp,
2345 .childp = new_bs == NULL ? childp : NULL,
2346 .old_bs = (*childp)->bs,
2347 .free_empty_child = free_empty_child,
2349 tran_add(tran, &bdrv_replace_child_drv, s);
2351 /* The abort handler relies on this */
2352 assert(s->old_bs != NULL);
2354 if (new_bs) {
2355 bdrv_ref(new_bs);
2358 * Pass free_empty_child=false, we will free the child (if
2359 * necessary) in bdrv_replace_child_commit() (if our
2360 * @free_empty_child parameter was true).
2362 bdrv_replace_child_noperm(childp, new_bs, false);
2363 /* old_bs reference is transparently moved from *childp to @s */
2367 * Refresh permissions in @bs subtree. The function is intended to be called
2368 * after some graph modification that was done without permission update.
2370 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2371 Transaction *tran, Error **errp)
2373 BlockDriver *drv = bs->drv;
2374 BdrvChild *c;
2375 int ret;
2376 uint64_t cumulative_perms, cumulative_shared_perms;
2378 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2380 /* Write permissions never work with read-only images */
2381 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2382 !bdrv_is_writable_after_reopen(bs, q))
2384 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2385 error_setg(errp, "Block node is read-only");
2386 } else {
2387 error_setg(errp, "Read-only block node '%s' cannot support "
2388 "read-write users", bdrv_get_node_name(bs));
2391 return -EPERM;
2395 * Unaligned requests will automatically be aligned to bl.request_alignment
2396 * and without RESIZE we can't extend requests to write to space beyond the
2397 * end of the image, so it's required that the image size is aligned.
2399 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2400 !(cumulative_perms & BLK_PERM_RESIZE))
2402 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2403 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2404 "Image size is not a multiple of request "
2405 "alignment");
2406 return -EPERM;
2410 /* Check this node */
2411 if (!drv) {
2412 return 0;
2415 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2416 errp);
2417 if (ret < 0) {
2418 return ret;
2421 /* Drivers that never have children can omit .bdrv_child_perm() */
2422 if (!drv->bdrv_child_perm) {
2423 assert(QLIST_EMPTY(&bs->children));
2424 return 0;
2427 /* Check all children */
2428 QLIST_FOREACH(c, &bs->children, next) {
2429 uint64_t cur_perm, cur_shared;
2431 bdrv_child_perm(bs, c->bs, c, c->role, q,
2432 cumulative_perms, cumulative_shared_perms,
2433 &cur_perm, &cur_shared);
2434 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2437 return 0;
2440 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2441 Transaction *tran, Error **errp)
2443 int ret;
2444 BlockDriverState *bs;
2446 for ( ; list; list = list->next) {
2447 bs = list->data;
2449 if (bdrv_parent_perms_conflict(bs, errp)) {
2450 return -EINVAL;
2453 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2454 if (ret < 0) {
2455 return ret;
2459 return 0;
2462 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2463 uint64_t *shared_perm)
2465 BdrvChild *c;
2466 uint64_t cumulative_perms = 0;
2467 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2469 QLIST_FOREACH(c, &bs->parents, next_parent) {
2470 cumulative_perms |= c->perm;
2471 cumulative_shared_perms &= c->shared_perm;
2474 *perm = cumulative_perms;
2475 *shared_perm = cumulative_shared_perms;
2478 char *bdrv_perm_names(uint64_t perm)
2480 struct perm_name {
2481 uint64_t perm;
2482 const char *name;
2483 } permissions[] = {
2484 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2485 { BLK_PERM_WRITE, "write" },
2486 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2487 { BLK_PERM_RESIZE, "resize" },
2488 { 0, NULL }
2491 GString *result = g_string_sized_new(30);
2492 struct perm_name *p;
2494 for (p = permissions; p->name; p++) {
2495 if (perm & p->perm) {
2496 if (result->len > 0) {
2497 g_string_append(result, ", ");
2499 g_string_append(result, p->name);
2503 return g_string_free(result, FALSE);
2507 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2509 int ret;
2510 Transaction *tran = tran_new();
2511 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2513 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2514 tran_finalize(tran, ret);
2516 return ret;
2519 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2520 Error **errp)
2522 Error *local_err = NULL;
2523 Transaction *tran = tran_new();
2524 int ret;
2526 bdrv_child_set_perm(c, perm, shared, tran);
2528 ret = bdrv_refresh_perms(c->bs, &local_err);
2530 tran_finalize(tran, ret);
2532 if (ret < 0) {
2533 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2534 /* tighten permissions */
2535 error_propagate(errp, local_err);
2536 } else {
2538 * Our caller may intend to only loosen restrictions and
2539 * does not expect this function to fail. Errors are not
2540 * fatal in such a case, so we can just hide them from our
2541 * caller.
2543 error_free(local_err);
2544 ret = 0;
2548 return ret;
2551 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2553 uint64_t parent_perms, parent_shared;
2554 uint64_t perms, shared;
2556 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2557 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2558 parent_perms, parent_shared, &perms, &shared);
2560 return bdrv_child_try_set_perm(c, perms, shared, errp);
2564 * Default implementation for .bdrv_child_perm() for block filters:
2565 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2566 * filtered child.
2568 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2569 BdrvChildRole role,
2570 BlockReopenQueue *reopen_queue,
2571 uint64_t perm, uint64_t shared,
2572 uint64_t *nperm, uint64_t *nshared)
2574 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2575 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2578 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2579 BdrvChildRole role,
2580 BlockReopenQueue *reopen_queue,
2581 uint64_t perm, uint64_t shared,
2582 uint64_t *nperm, uint64_t *nshared)
2584 assert(role & BDRV_CHILD_COW);
2587 * We want consistent read from backing files if the parent needs it.
2588 * No other operations are performed on backing files.
2590 perm &= BLK_PERM_CONSISTENT_READ;
2593 * If the parent can deal with changing data, we're okay with a
2594 * writable and resizable backing file.
2595 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2597 if (shared & BLK_PERM_WRITE) {
2598 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2599 } else {
2600 shared = 0;
2603 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2605 if (bs->open_flags & BDRV_O_INACTIVE) {
2606 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2609 *nperm = perm;
2610 *nshared = shared;
2613 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2614 BdrvChildRole role,
2615 BlockReopenQueue *reopen_queue,
2616 uint64_t perm, uint64_t shared,
2617 uint64_t *nperm, uint64_t *nshared)
2619 int flags;
2621 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2623 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2626 * Apart from the modifications below, the same permissions are
2627 * forwarded and left alone as for filters
2629 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2630 perm, shared, &perm, &shared);
2632 if (role & BDRV_CHILD_METADATA) {
2633 /* Format drivers may touch metadata even if the guest doesn't write */
2634 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2635 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2639 * bs->file always needs to be consistent because of the
2640 * metadata. We can never allow other users to resize or write
2641 * to it.
2643 if (!(flags & BDRV_O_NO_IO)) {
2644 perm |= BLK_PERM_CONSISTENT_READ;
2646 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2649 if (role & BDRV_CHILD_DATA) {
2651 * Technically, everything in this block is a subset of the
2652 * BDRV_CHILD_METADATA path taken above, and so this could
2653 * be an "else if" branch. However, that is not obvious, and
2654 * this function is not performance critical, therefore we let
2655 * this be an independent "if".
2659 * We cannot allow other users to resize the file because the
2660 * format driver might have some assumptions about the size
2661 * (e.g. because it is stored in metadata, or because the file
2662 * is split into fixed-size data files).
2664 shared &= ~BLK_PERM_RESIZE;
2667 * WRITE_UNCHANGED often cannot be performed as such on the
2668 * data file. For example, the qcow2 driver may still need to
2669 * write copied clusters on copy-on-read.
2671 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2672 perm |= BLK_PERM_WRITE;
2676 * If the data file is written to, the format driver may
2677 * expect to be able to resize it by writing beyond the EOF.
2679 if (perm & BLK_PERM_WRITE) {
2680 perm |= BLK_PERM_RESIZE;
2684 if (bs->open_flags & BDRV_O_INACTIVE) {
2685 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2688 *nperm = perm;
2689 *nshared = shared;
2692 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2693 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2694 uint64_t perm, uint64_t shared,
2695 uint64_t *nperm, uint64_t *nshared)
2697 if (role & BDRV_CHILD_FILTERED) {
2698 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2699 BDRV_CHILD_COW)));
2700 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2701 perm, shared, nperm, nshared);
2702 } else if (role & BDRV_CHILD_COW) {
2703 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2704 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2705 perm, shared, nperm, nshared);
2706 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2707 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2708 perm, shared, nperm, nshared);
2709 } else {
2710 g_assert_not_reached();
2714 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2716 static const uint64_t permissions[] = {
2717 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2718 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2719 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2720 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2723 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2724 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2726 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2728 return permissions[qapi_perm];
2732 * Replace (*childp)->bs by @new_bs.
2734 * If @new_bs is NULL, *childp will be set to NULL, too: BDS parents
2735 * generally cannot handle a BdrvChild with .bs == NULL, so clearing
2736 * BdrvChild.bs should generally immediately be followed by the
2737 * BdrvChild pointer being cleared as well.
2739 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2740 * freed. @free_empty_child should only be false if the caller will
2741 * free the BdrvChild themselves (this may be important in a
2742 * transactional context, where it may only be freed on commit).
2744 static void bdrv_replace_child_noperm(BdrvChild **childp,
2745 BlockDriverState *new_bs,
2746 bool free_empty_child)
2748 BdrvChild *child = *childp;
2749 BlockDriverState *old_bs = child->bs;
2750 int new_bs_quiesce_counter;
2751 int drain_saldo;
2753 assert(!child->frozen);
2754 assert(old_bs != new_bs);
2756 if (old_bs && new_bs) {
2757 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2760 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2761 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2764 * If the new child node is drained but the old one was not, flush
2765 * all outstanding requests to the old child node.
2767 while (drain_saldo > 0 && child->klass->drained_begin) {
2768 bdrv_parent_drained_begin_single(child, true);
2769 drain_saldo--;
2772 if (old_bs) {
2773 /* Detach first so that the recursive drain sections coming from @child
2774 * are already gone and we only end the drain sections that came from
2775 * elsewhere. */
2776 if (child->klass->detach) {
2777 child->klass->detach(child);
2779 QLIST_REMOVE(child, next_parent);
2782 child->bs = new_bs;
2783 if (!new_bs) {
2784 *childp = NULL;
2787 if (new_bs) {
2788 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2791 * Detaching the old node may have led to the new node's
2792 * quiesce_counter having been decreased. Not a problem, we
2793 * just need to recognize this here and then invoke
2794 * drained_end appropriately more often.
2796 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2797 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2799 /* Attach only after starting new drained sections, so that recursive
2800 * drain sections coming from @child don't get an extra .drained_begin
2801 * callback. */
2802 if (child->klass->attach) {
2803 child->klass->attach(child);
2808 * If the old child node was drained but the new one is not, allow
2809 * requests to come in only after the new node has been attached.
2811 while (drain_saldo < 0 && child->klass->drained_end) {
2812 bdrv_parent_drained_end_single(child);
2813 drain_saldo++;
2816 if (free_empty_child && !child->bs) {
2817 bdrv_child_free(child);
2822 * Free the given @child.
2824 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2825 * unused (i.e. not in a children list).
2827 static void bdrv_child_free(BdrvChild *child)
2829 assert(!child->bs);
2830 assert(!child->next.le_prev); /* not in children list */
2832 g_free(child->name);
2833 g_free(child);
2836 typedef struct BdrvAttachChildCommonState {
2837 BdrvChild **child;
2838 AioContext *old_parent_ctx;
2839 AioContext *old_child_ctx;
2840 } BdrvAttachChildCommonState;
2842 static void bdrv_attach_child_common_abort(void *opaque)
2844 BdrvAttachChildCommonState *s = opaque;
2845 BdrvChild *child = *s->child;
2846 BlockDriverState *bs = child->bs;
2849 * Pass free_empty_child=false, because we still need the child
2850 * for the AioContext operations on the parent below; those
2851 * BdrvChildClass methods all work on a BdrvChild object, so we
2852 * need to keep it as an empty shell (after this function, it will
2853 * not be attached to any parent, and it will not have a .bs).
2855 bdrv_replace_child_noperm(s->child, NULL, false);
2857 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2858 bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort);
2861 if (bdrv_child_get_parent_aio_context(child) != s->old_parent_ctx) {
2862 GSList *ignore;
2864 /* No need to ignore `child`, because it has been detached already */
2865 ignore = NULL;
2866 child->klass->can_set_aio_ctx(child, s->old_parent_ctx, &ignore,
2867 &error_abort);
2868 g_slist_free(ignore);
2870 ignore = NULL;
2871 child->klass->set_aio_ctx(child, s->old_parent_ctx, &ignore);
2872 g_slist_free(ignore);
2875 bdrv_unref(bs);
2876 bdrv_child_free(child);
2879 static TransactionActionDrv bdrv_attach_child_common_drv = {
2880 .abort = bdrv_attach_child_common_abort,
2881 .clean = g_free,
2885 * Common part of attaching bdrv child to bs or to blk or to job
2887 * Resulting new child is returned through @child.
2888 * At start *@child must be NULL.
2889 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2890 * NULL on abort(). So referenced variable must live at least until transaction
2891 * end.
2893 * Function doesn't update permissions, caller is responsible for this.
2895 static int bdrv_attach_child_common(BlockDriverState *child_bs,
2896 const char *child_name,
2897 const BdrvChildClass *child_class,
2898 BdrvChildRole child_role,
2899 uint64_t perm, uint64_t shared_perm,
2900 void *opaque, BdrvChild **child,
2901 Transaction *tran, Error **errp)
2903 BdrvChild *new_child;
2904 AioContext *parent_ctx;
2905 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2907 assert(child);
2908 assert(*child == NULL);
2909 assert(child_class->get_parent_desc);
2911 new_child = g_new(BdrvChild, 1);
2912 *new_child = (BdrvChild) {
2913 .bs = NULL,
2914 .name = g_strdup(child_name),
2915 .klass = child_class,
2916 .role = child_role,
2917 .perm = perm,
2918 .shared_perm = shared_perm,
2919 .opaque = opaque,
2923 * If the AioContexts don't match, first try to move the subtree of
2924 * child_bs into the AioContext of the new parent. If this doesn't work,
2925 * try moving the parent into the AioContext of child_bs instead.
2927 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
2928 if (child_ctx != parent_ctx) {
2929 Error *local_err = NULL;
2930 int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err);
2932 if (ret < 0 && child_class->can_set_aio_ctx) {
2933 GSList *ignore = g_slist_prepend(NULL, new_child);
2934 if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore,
2935 NULL))
2937 error_free(local_err);
2938 ret = 0;
2939 g_slist_free(ignore);
2940 ignore = g_slist_prepend(NULL, new_child);
2941 child_class->set_aio_ctx(new_child, child_ctx, &ignore);
2943 g_slist_free(ignore);
2946 if (ret < 0) {
2947 error_propagate(errp, local_err);
2948 bdrv_child_free(new_child);
2949 return ret;
2953 bdrv_ref(child_bs);
2954 bdrv_replace_child_noperm(&new_child, child_bs, true);
2955 /* child_bs was non-NULL, so new_child must not have been freed */
2956 assert(new_child != NULL);
2958 *child = new_child;
2960 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
2961 *s = (BdrvAttachChildCommonState) {
2962 .child = child,
2963 .old_parent_ctx = parent_ctx,
2964 .old_child_ctx = child_ctx,
2966 tran_add(tran, &bdrv_attach_child_common_drv, s);
2968 return 0;
2972 * Variable referenced by @child must live at least until transaction end.
2973 * (see bdrv_attach_child_common() doc for details)
2975 * Function doesn't update permissions, caller is responsible for this.
2977 static int bdrv_attach_child_noperm(BlockDriverState *parent_bs,
2978 BlockDriverState *child_bs,
2979 const char *child_name,
2980 const BdrvChildClass *child_class,
2981 BdrvChildRole child_role,
2982 BdrvChild **child,
2983 Transaction *tran,
2984 Error **errp)
2986 int ret;
2987 uint64_t perm, shared_perm;
2989 assert(parent_bs->drv);
2991 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
2992 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
2993 child_bs->node_name, child_name, parent_bs->node_name);
2994 return -EINVAL;
2997 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2998 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2999 perm, shared_perm, &perm, &shared_perm);
3001 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
3002 child_role, perm, shared_perm, parent_bs,
3003 child, tran, errp);
3004 if (ret < 0) {
3005 return ret;
3008 return 0;
3011 static void bdrv_detach_child(BdrvChild **childp)
3013 BlockDriverState *old_bs = (*childp)->bs;
3015 bdrv_replace_child_noperm(childp, NULL, true);
3017 if (old_bs) {
3019 * Update permissions for old node. We're just taking a parent away, so
3020 * we're loosening restrictions. Errors of permission update are not
3021 * fatal in this case, ignore them.
3023 bdrv_refresh_perms(old_bs, NULL);
3026 * When the parent requiring a non-default AioContext is removed, the
3027 * node moves back to the main AioContext
3029 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
3034 * This function steals the reference to child_bs from the caller.
3035 * That reference is later dropped by bdrv_root_unref_child().
3037 * On failure NULL is returned, errp is set and the reference to
3038 * child_bs is also dropped.
3040 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3041 * (unless @child_bs is already in @ctx).
3043 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3044 const char *child_name,
3045 const BdrvChildClass *child_class,
3046 BdrvChildRole child_role,
3047 uint64_t perm, uint64_t shared_perm,
3048 void *opaque, Error **errp)
3050 int ret;
3051 BdrvChild *child = NULL;
3052 Transaction *tran = tran_new();
3054 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
3055 child_role, perm, shared_perm, opaque,
3056 &child, tran, errp);
3057 if (ret < 0) {
3058 goto out;
3061 ret = bdrv_refresh_perms(child_bs, errp);
3063 out:
3064 tran_finalize(tran, ret);
3065 /* child is unset on failure by bdrv_attach_child_common_abort() */
3066 assert((ret < 0) == !child);
3068 bdrv_unref(child_bs);
3069 return child;
3073 * This function transfers the reference to child_bs from the caller
3074 * to parent_bs. That reference is later dropped by parent_bs on
3075 * bdrv_close() or if someone calls bdrv_unref_child().
3077 * On failure NULL is returned, errp is set and the reference to
3078 * child_bs is also dropped.
3080 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3081 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3083 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3084 BlockDriverState *child_bs,
3085 const char *child_name,
3086 const BdrvChildClass *child_class,
3087 BdrvChildRole child_role,
3088 Error **errp)
3090 int ret;
3091 BdrvChild *child = NULL;
3092 Transaction *tran = tran_new();
3094 ret = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, child_class,
3095 child_role, &child, tran, errp);
3096 if (ret < 0) {
3097 goto out;
3100 ret = bdrv_refresh_perms(parent_bs, errp);
3101 if (ret < 0) {
3102 goto out;
3105 out:
3106 tran_finalize(tran, ret);
3107 /* child is unset on failure by bdrv_attach_child_common_abort() */
3108 assert((ret < 0) == !child);
3110 bdrv_unref(child_bs);
3112 return child;
3115 /* Callers must ensure that child->frozen is false. */
3116 void bdrv_root_unref_child(BdrvChild *child)
3118 BlockDriverState *child_bs;
3120 child_bs = child->bs;
3121 bdrv_detach_child(&child);
3122 bdrv_unref(child_bs);
3125 typedef struct BdrvSetInheritsFrom {
3126 BlockDriverState *bs;
3127 BlockDriverState *old_inherits_from;
3128 } BdrvSetInheritsFrom;
3130 static void bdrv_set_inherits_from_abort(void *opaque)
3132 BdrvSetInheritsFrom *s = opaque;
3134 s->bs->inherits_from = s->old_inherits_from;
3137 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3138 .abort = bdrv_set_inherits_from_abort,
3139 .clean = g_free,
3142 /* @tran is allowed to be NULL. In this case no rollback is possible */
3143 static void bdrv_set_inherits_from(BlockDriverState *bs,
3144 BlockDriverState *new_inherits_from,
3145 Transaction *tran)
3147 if (tran) {
3148 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3150 *s = (BdrvSetInheritsFrom) {
3151 .bs = bs,
3152 .old_inherits_from = bs->inherits_from,
3155 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3158 bs->inherits_from = new_inherits_from;
3162 * Clear all inherits_from pointers from children and grandchildren of
3163 * @root that point to @root, where necessary.
3164 * @tran is allowed to be NULL. In this case no rollback is possible
3166 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3167 Transaction *tran)
3169 BdrvChild *c;
3171 if (child->bs->inherits_from == root) {
3173 * Remove inherits_from only when the last reference between root and
3174 * child->bs goes away.
3176 QLIST_FOREACH(c, &root->children, next) {
3177 if (c != child && c->bs == child->bs) {
3178 break;
3181 if (c == NULL) {
3182 bdrv_set_inherits_from(child->bs, NULL, tran);
3186 QLIST_FOREACH(c, &child->bs->children, next) {
3187 bdrv_unset_inherits_from(root, c, tran);
3191 /* Callers must ensure that child->frozen is false. */
3192 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3194 if (child == NULL) {
3195 return;
3198 bdrv_unset_inherits_from(parent, child, NULL);
3199 bdrv_root_unref_child(child);
3203 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3205 BdrvChild *c;
3206 QLIST_FOREACH(c, &bs->parents, next_parent) {
3207 if (c->klass->change_media) {
3208 c->klass->change_media(c, load);
3213 /* Return true if you can reach parent going through child->inherits_from
3214 * recursively. If parent or child are NULL, return false */
3215 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3216 BlockDriverState *parent)
3218 while (child && child != parent) {
3219 child = child->inherits_from;
3222 return child != NULL;
3226 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3227 * mostly used for COW backing children (role = COW), but also for
3228 * filtered children (role = FILTERED | PRIMARY).
3230 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3232 if (bs->drv && bs->drv->is_filter) {
3233 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3234 } else {
3235 return BDRV_CHILD_COW;
3240 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3241 * callers which don't need their own reference any more must call bdrv_unref().
3243 * Function doesn't update permissions, caller is responsible for this.
3245 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3246 BlockDriverState *child_bs,
3247 bool is_backing,
3248 Transaction *tran, Error **errp)
3250 int ret = 0;
3251 bool update_inherits_from =
3252 bdrv_inherits_from_recursive(child_bs, parent_bs);
3253 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3254 BdrvChildRole role;
3256 if (!parent_bs->drv) {
3258 * Node without drv is an object without a class :/. TODO: finally fix
3259 * qcow2 driver to never clear bs->drv and implement format corruption
3260 * handling in other way.
3262 error_setg(errp, "Node corrupted");
3263 return -EINVAL;
3266 if (child && child->frozen) {
3267 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3268 child->name, parent_bs->node_name, child->bs->node_name);
3269 return -EPERM;
3272 if (is_backing && !parent_bs->drv->is_filter &&
3273 !parent_bs->drv->supports_backing)
3275 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3276 "files", parent_bs->drv->format_name, parent_bs->node_name);
3277 return -EINVAL;
3280 if (parent_bs->drv->is_filter) {
3281 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3282 } else if (is_backing) {
3283 role = BDRV_CHILD_COW;
3284 } else {
3286 * We only can use same role as it is in existing child. We don't have
3287 * infrastructure to determine role of file child in generic way
3289 if (!child) {
3290 error_setg(errp, "Cannot set file child to format node without "
3291 "file child");
3292 return -EINVAL;
3294 role = child->role;
3297 if (child) {
3298 bdrv_unset_inherits_from(parent_bs, child, tran);
3299 bdrv_remove_file_or_backing_child(parent_bs, child, tran);
3302 if (!child_bs) {
3303 goto out;
3306 ret = bdrv_attach_child_noperm(parent_bs, child_bs,
3307 is_backing ? "backing" : "file",
3308 &child_of_bds, role,
3309 is_backing ? &parent_bs->backing :
3310 &parent_bs->file,
3311 tran, errp);
3312 if (ret < 0) {
3313 return ret;
3318 * If inherits_from pointed recursively to bs then let's update it to
3319 * point directly to bs (else it will become NULL).
3321 if (update_inherits_from) {
3322 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3325 out:
3326 bdrv_refresh_limits(parent_bs, tran, NULL);
3328 return 0;
3331 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3332 BlockDriverState *backing_hd,
3333 Transaction *tran, Error **errp)
3335 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3338 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3339 Error **errp)
3341 int ret;
3342 Transaction *tran = tran_new();
3344 bdrv_drained_begin(bs);
3346 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3347 if (ret < 0) {
3348 goto out;
3351 ret = bdrv_refresh_perms(bs, errp);
3352 out:
3353 tran_finalize(tran, ret);
3355 bdrv_drained_end(bs);
3357 return ret;
3361 * Opens the backing file for a BlockDriverState if not yet open
3363 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3364 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3365 * itself, all options starting with "${bdref_key}." are considered part of the
3366 * BlockdevRef.
3368 * TODO Can this be unified with bdrv_open_image()?
3370 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3371 const char *bdref_key, Error **errp)
3373 char *backing_filename = NULL;
3374 char *bdref_key_dot;
3375 const char *reference = NULL;
3376 int ret = 0;
3377 bool implicit_backing = false;
3378 BlockDriverState *backing_hd;
3379 QDict *options;
3380 QDict *tmp_parent_options = NULL;
3381 Error *local_err = NULL;
3383 if (bs->backing != NULL) {
3384 goto free_exit;
3387 /* NULL means an empty set of options */
3388 if (parent_options == NULL) {
3389 tmp_parent_options = qdict_new();
3390 parent_options = tmp_parent_options;
3393 bs->open_flags &= ~BDRV_O_NO_BACKING;
3395 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3396 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3397 g_free(bdref_key_dot);
3400 * Caution: while qdict_get_try_str() is fine, getting non-string
3401 * types would require more care. When @parent_options come from
3402 * -blockdev or blockdev_add, its members are typed according to
3403 * the QAPI schema, but when they come from -drive, they're all
3404 * QString.
3406 reference = qdict_get_try_str(parent_options, bdref_key);
3407 if (reference || qdict_haskey(options, "file.filename")) {
3408 /* keep backing_filename NULL */
3409 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3410 qobject_unref(options);
3411 goto free_exit;
3412 } else {
3413 if (qdict_size(options) == 0) {
3414 /* If the user specifies options that do not modify the
3415 * backing file's behavior, we might still consider it the
3416 * implicit backing file. But it's easier this way, and
3417 * just specifying some of the backing BDS's options is
3418 * only possible with -drive anyway (otherwise the QAPI
3419 * schema forces the user to specify everything). */
3420 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3423 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3424 if (local_err) {
3425 ret = -EINVAL;
3426 error_propagate(errp, local_err);
3427 qobject_unref(options);
3428 goto free_exit;
3432 if (!bs->drv || !bs->drv->supports_backing) {
3433 ret = -EINVAL;
3434 error_setg(errp, "Driver doesn't support backing files");
3435 qobject_unref(options);
3436 goto free_exit;
3439 if (!reference &&
3440 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3441 qdict_put_str(options, "driver", bs->backing_format);
3444 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3445 &child_of_bds, bdrv_backing_role(bs), errp);
3446 if (!backing_hd) {
3447 bs->open_flags |= BDRV_O_NO_BACKING;
3448 error_prepend(errp, "Could not open backing file: ");
3449 ret = -EINVAL;
3450 goto free_exit;
3453 if (implicit_backing) {
3454 bdrv_refresh_filename(backing_hd);
3455 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3456 backing_hd->filename);
3459 /* Hook up the backing file link; drop our reference, bs owns the
3460 * backing_hd reference now */
3461 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3462 bdrv_unref(backing_hd);
3463 if (ret < 0) {
3464 goto free_exit;
3467 qdict_del(parent_options, bdref_key);
3469 free_exit:
3470 g_free(backing_filename);
3471 qobject_unref(tmp_parent_options);
3472 return ret;
3475 static BlockDriverState *
3476 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3477 BlockDriverState *parent, const BdrvChildClass *child_class,
3478 BdrvChildRole child_role, bool allow_none, Error **errp)
3480 BlockDriverState *bs = NULL;
3481 QDict *image_options;
3482 char *bdref_key_dot;
3483 const char *reference;
3485 assert(child_class != NULL);
3487 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3488 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3489 g_free(bdref_key_dot);
3492 * Caution: while qdict_get_try_str() is fine, getting non-string
3493 * types would require more care. When @options come from
3494 * -blockdev or blockdev_add, its members are typed according to
3495 * the QAPI schema, but when they come from -drive, they're all
3496 * QString.
3498 reference = qdict_get_try_str(options, bdref_key);
3499 if (!filename && !reference && !qdict_size(image_options)) {
3500 if (!allow_none) {
3501 error_setg(errp, "A block device must be specified for \"%s\"",
3502 bdref_key);
3504 qobject_unref(image_options);
3505 goto done;
3508 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3509 parent, child_class, child_role, errp);
3510 if (!bs) {
3511 goto done;
3514 done:
3515 qdict_del(options, bdref_key);
3516 return bs;
3520 * Opens a disk image whose options are given as BlockdevRef in another block
3521 * device's options.
3523 * If allow_none is true, no image will be opened if filename is false and no
3524 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3526 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3527 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3528 * itself, all options starting with "${bdref_key}." are considered part of the
3529 * BlockdevRef.
3531 * The BlockdevRef will be removed from the options QDict.
3533 BdrvChild *bdrv_open_child(const char *filename,
3534 QDict *options, const char *bdref_key,
3535 BlockDriverState *parent,
3536 const BdrvChildClass *child_class,
3537 BdrvChildRole child_role,
3538 bool allow_none, Error **errp)
3540 BlockDriverState *bs;
3542 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3543 child_role, allow_none, errp);
3544 if (bs == NULL) {
3545 return NULL;
3548 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3549 errp);
3553 * TODO Future callers may need to specify parent/child_class in order for
3554 * option inheritance to work. Existing callers use it for the root node.
3556 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3558 BlockDriverState *bs = NULL;
3559 QObject *obj = NULL;
3560 QDict *qdict = NULL;
3561 const char *reference = NULL;
3562 Visitor *v = NULL;
3564 if (ref->type == QTYPE_QSTRING) {
3565 reference = ref->u.reference;
3566 } else {
3567 BlockdevOptions *options = &ref->u.definition;
3568 assert(ref->type == QTYPE_QDICT);
3570 v = qobject_output_visitor_new(&obj);
3571 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3572 visit_complete(v, &obj);
3574 qdict = qobject_to(QDict, obj);
3575 qdict_flatten(qdict);
3577 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3578 * compatibility with other callers) rather than what we want as the
3579 * real defaults. Apply the defaults here instead. */
3580 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3581 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3582 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3583 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3587 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3588 obj = NULL;
3589 qobject_unref(obj);
3590 visit_free(v);
3591 return bs;
3594 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3595 int flags,
3596 QDict *snapshot_options,
3597 Error **errp)
3599 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3600 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3601 int64_t total_size;
3602 QemuOpts *opts = NULL;
3603 BlockDriverState *bs_snapshot = NULL;
3604 int ret;
3606 /* if snapshot, we create a temporary backing file and open it
3607 instead of opening 'filename' directly */
3609 /* Get the required size from the image */
3610 total_size = bdrv_getlength(bs);
3611 if (total_size < 0) {
3612 error_setg_errno(errp, -total_size, "Could not get image size");
3613 goto out;
3616 /* Create the temporary image */
3617 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3618 if (ret < 0) {
3619 error_setg_errno(errp, -ret, "Could not get temporary filename");
3620 goto out;
3623 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3624 &error_abort);
3625 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3626 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3627 qemu_opts_del(opts);
3628 if (ret < 0) {
3629 error_prepend(errp, "Could not create temporary overlay '%s': ",
3630 tmp_filename);
3631 goto out;
3634 /* Prepare options QDict for the temporary file */
3635 qdict_put_str(snapshot_options, "file.driver", "file");
3636 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3637 qdict_put_str(snapshot_options, "driver", "qcow2");
3639 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3640 snapshot_options = NULL;
3641 if (!bs_snapshot) {
3642 goto out;
3645 ret = bdrv_append(bs_snapshot, bs, errp);
3646 if (ret < 0) {
3647 bs_snapshot = NULL;
3648 goto out;
3651 out:
3652 qobject_unref(snapshot_options);
3653 g_free(tmp_filename);
3654 return bs_snapshot;
3658 * Opens a disk image (raw, qcow2, vmdk, ...)
3660 * options is a QDict of options to pass to the block drivers, or NULL for an
3661 * empty set of options. The reference to the QDict belongs to the block layer
3662 * after the call (even on failure), so if the caller intends to reuse the
3663 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3665 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3666 * If it is not NULL, the referenced BDS will be reused.
3668 * The reference parameter may be used to specify an existing block device which
3669 * should be opened. If specified, neither options nor a filename may be given,
3670 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3672 static BlockDriverState *bdrv_open_inherit(const char *filename,
3673 const char *reference,
3674 QDict *options, int flags,
3675 BlockDriverState *parent,
3676 const BdrvChildClass *child_class,
3677 BdrvChildRole child_role,
3678 Error **errp)
3680 int ret;
3681 BlockBackend *file = NULL;
3682 BlockDriverState *bs;
3683 BlockDriver *drv = NULL;
3684 BdrvChild *child;
3685 const char *drvname;
3686 const char *backing;
3687 Error *local_err = NULL;
3688 QDict *snapshot_options = NULL;
3689 int snapshot_flags = 0;
3691 assert(!child_class || !flags);
3692 assert(!child_class == !parent);
3694 if (reference) {
3695 bool options_non_empty = options ? qdict_size(options) : false;
3696 qobject_unref(options);
3698 if (filename || options_non_empty) {
3699 error_setg(errp, "Cannot reference an existing block device with "
3700 "additional options or a new filename");
3701 return NULL;
3704 bs = bdrv_lookup_bs(reference, reference, errp);
3705 if (!bs) {
3706 return NULL;
3709 bdrv_ref(bs);
3710 return bs;
3713 bs = bdrv_new();
3715 /* NULL means an empty set of options */
3716 if (options == NULL) {
3717 options = qdict_new();
3720 /* json: syntax counts as explicit options, as if in the QDict */
3721 parse_json_protocol(options, &filename, &local_err);
3722 if (local_err) {
3723 goto fail;
3726 bs->explicit_options = qdict_clone_shallow(options);
3728 if (child_class) {
3729 bool parent_is_format;
3731 if (parent->drv) {
3732 parent_is_format = parent->drv->is_format;
3733 } else {
3735 * parent->drv is not set yet because this node is opened for
3736 * (potential) format probing. That means that @parent is going
3737 * to be a format node.
3739 parent_is_format = true;
3742 bs->inherits_from = parent;
3743 child_class->inherit_options(child_role, parent_is_format,
3744 &flags, options,
3745 parent->open_flags, parent->options);
3748 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3749 if (ret < 0) {
3750 goto fail;
3754 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3755 * Caution: getting a boolean member of @options requires care.
3756 * When @options come from -blockdev or blockdev_add, members are
3757 * typed according to the QAPI schema, but when they come from
3758 * -drive, they're all QString.
3760 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3761 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3762 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3763 } else {
3764 flags &= ~BDRV_O_RDWR;
3767 if (flags & BDRV_O_SNAPSHOT) {
3768 snapshot_options = qdict_new();
3769 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3770 flags, options);
3771 /* Let bdrv_backing_options() override "read-only" */
3772 qdict_del(options, BDRV_OPT_READ_ONLY);
3773 bdrv_inherited_options(BDRV_CHILD_COW, true,
3774 &flags, options, flags, options);
3777 bs->open_flags = flags;
3778 bs->options = options;
3779 options = qdict_clone_shallow(options);
3781 /* Find the right image format driver */
3782 /* See cautionary note on accessing @options above */
3783 drvname = qdict_get_try_str(options, "driver");
3784 if (drvname) {
3785 drv = bdrv_find_format(drvname);
3786 if (!drv) {
3787 error_setg(errp, "Unknown driver: '%s'", drvname);
3788 goto fail;
3792 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3794 /* See cautionary note on accessing @options above */
3795 backing = qdict_get_try_str(options, "backing");
3796 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3797 (backing && *backing == '\0'))
3799 if (backing) {
3800 warn_report("Use of \"backing\": \"\" is deprecated; "
3801 "use \"backing\": null instead");
3803 flags |= BDRV_O_NO_BACKING;
3804 qdict_del(bs->explicit_options, "backing");
3805 qdict_del(bs->options, "backing");
3806 qdict_del(options, "backing");
3809 /* Open image file without format layer. This BlockBackend is only used for
3810 * probing, the block drivers will do their own bdrv_open_child() for the
3811 * same BDS, which is why we put the node name back into options. */
3812 if ((flags & BDRV_O_PROTOCOL) == 0) {
3813 BlockDriverState *file_bs;
3815 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3816 &child_of_bds, BDRV_CHILD_IMAGE,
3817 true, &local_err);
3818 if (local_err) {
3819 goto fail;
3821 if (file_bs != NULL) {
3822 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3823 * looking at the header to guess the image format. This works even
3824 * in cases where a guest would not see a consistent state. */
3825 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3826 blk_insert_bs(file, file_bs, &local_err);
3827 bdrv_unref(file_bs);
3828 if (local_err) {
3829 goto fail;
3832 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3836 /* Image format probing */
3837 bs->probed = !drv;
3838 if (!drv && file) {
3839 ret = find_image_format(file, filename, &drv, &local_err);
3840 if (ret < 0) {
3841 goto fail;
3844 * This option update would logically belong in bdrv_fill_options(),
3845 * but we first need to open bs->file for the probing to work, while
3846 * opening bs->file already requires the (mostly) final set of options
3847 * so that cache mode etc. can be inherited.
3849 * Adding the driver later is somewhat ugly, but it's not an option
3850 * that would ever be inherited, so it's correct. We just need to make
3851 * sure to update both bs->options (which has the full effective
3852 * options for bs) and options (which has file.* already removed).
3854 qdict_put_str(bs->options, "driver", drv->format_name);
3855 qdict_put_str(options, "driver", drv->format_name);
3856 } else if (!drv) {
3857 error_setg(errp, "Must specify either driver or file");
3858 goto fail;
3861 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3862 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3863 /* file must be NULL if a protocol BDS is about to be created
3864 * (the inverse results in an error message from bdrv_open_common()) */
3865 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3867 /* Open the image */
3868 ret = bdrv_open_common(bs, file, options, &local_err);
3869 if (ret < 0) {
3870 goto fail;
3873 if (file) {
3874 blk_unref(file);
3875 file = NULL;
3878 /* If there is a backing file, use it */
3879 if ((flags & BDRV_O_NO_BACKING) == 0) {
3880 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3881 if (ret < 0) {
3882 goto close_and_fail;
3886 /* Remove all children options and references
3887 * from bs->options and bs->explicit_options */
3888 QLIST_FOREACH(child, &bs->children, next) {
3889 char *child_key_dot;
3890 child_key_dot = g_strdup_printf("%s.", child->name);
3891 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3892 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3893 qdict_del(bs->explicit_options, child->name);
3894 qdict_del(bs->options, child->name);
3895 g_free(child_key_dot);
3898 /* Check if any unknown options were used */
3899 if (qdict_size(options) != 0) {
3900 const QDictEntry *entry = qdict_first(options);
3901 if (flags & BDRV_O_PROTOCOL) {
3902 error_setg(errp, "Block protocol '%s' doesn't support the option "
3903 "'%s'", drv->format_name, entry->key);
3904 } else {
3905 error_setg(errp,
3906 "Block format '%s' does not support the option '%s'",
3907 drv->format_name, entry->key);
3910 goto close_and_fail;
3913 bdrv_parent_cb_change_media(bs, true);
3915 qobject_unref(options);
3916 options = NULL;
3918 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3919 * temporary snapshot afterwards. */
3920 if (snapshot_flags) {
3921 BlockDriverState *snapshot_bs;
3922 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3923 snapshot_options, &local_err);
3924 snapshot_options = NULL;
3925 if (local_err) {
3926 goto close_and_fail;
3928 /* We are not going to return bs but the overlay on top of it
3929 * (snapshot_bs); thus, we have to drop the strong reference to bs
3930 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3931 * though, because the overlay still has a reference to it. */
3932 bdrv_unref(bs);
3933 bs = snapshot_bs;
3936 return bs;
3938 fail:
3939 blk_unref(file);
3940 qobject_unref(snapshot_options);
3941 qobject_unref(bs->explicit_options);
3942 qobject_unref(bs->options);
3943 qobject_unref(options);
3944 bs->options = NULL;
3945 bs->explicit_options = NULL;
3946 bdrv_unref(bs);
3947 error_propagate(errp, local_err);
3948 return NULL;
3950 close_and_fail:
3951 bdrv_unref(bs);
3952 qobject_unref(snapshot_options);
3953 qobject_unref(options);
3954 error_propagate(errp, local_err);
3955 return NULL;
3958 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3959 QDict *options, int flags, Error **errp)
3961 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3962 NULL, 0, errp);
3965 /* Return true if the NULL-terminated @list contains @str */
3966 static bool is_str_in_list(const char *str, const char *const *list)
3968 if (str && list) {
3969 int i;
3970 for (i = 0; list[i] != NULL; i++) {
3971 if (!strcmp(str, list[i])) {
3972 return true;
3976 return false;
3980 * Check that every option set in @bs->options is also set in
3981 * @new_opts.
3983 * Options listed in the common_options list and in
3984 * @bs->drv->mutable_opts are skipped.
3986 * Return 0 on success, otherwise return -EINVAL and set @errp.
3988 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3989 const QDict *new_opts, Error **errp)
3991 const QDictEntry *e;
3992 /* These options are common to all block drivers and are handled
3993 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3994 const char *const common_options[] = {
3995 "node-name", "discard", "cache.direct", "cache.no-flush",
3996 "read-only", "auto-read-only", "detect-zeroes", NULL
3999 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4000 if (!qdict_haskey(new_opts, e->key) &&
4001 !is_str_in_list(e->key, common_options) &&
4002 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4003 error_setg(errp, "Option '%s' cannot be reset "
4004 "to its default value", e->key);
4005 return -EINVAL;
4009 return 0;
4013 * Returns true if @child can be reached recursively from @bs
4015 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4016 BlockDriverState *child)
4018 BdrvChild *c;
4020 if (bs == child) {
4021 return true;
4024 QLIST_FOREACH(c, &bs->children, next) {
4025 if (bdrv_recurse_has_child(c->bs, child)) {
4026 return true;
4030 return false;
4034 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4035 * reopen of multiple devices.
4037 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4038 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4039 * be created and initialized. This newly created BlockReopenQueue should be
4040 * passed back in for subsequent calls that are intended to be of the same
4041 * atomic 'set'.
4043 * bs is the BlockDriverState to add to the reopen queue.
4045 * options contains the changed options for the associated bs
4046 * (the BlockReopenQueue takes ownership)
4048 * flags contains the open flags for the associated bs
4050 * returns a pointer to bs_queue, which is either the newly allocated
4051 * bs_queue, or the existing bs_queue being used.
4053 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4055 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4056 BlockDriverState *bs,
4057 QDict *options,
4058 const BdrvChildClass *klass,
4059 BdrvChildRole role,
4060 bool parent_is_format,
4061 QDict *parent_options,
4062 int parent_flags,
4063 bool keep_old_opts)
4065 assert(bs != NULL);
4067 BlockReopenQueueEntry *bs_entry;
4068 BdrvChild *child;
4069 QDict *old_options, *explicit_options, *options_copy;
4070 int flags;
4071 QemuOpts *opts;
4073 /* Make sure that the caller remembered to use a drained section. This is
4074 * important to avoid graph changes between the recursive queuing here and
4075 * bdrv_reopen_multiple(). */
4076 assert(bs->quiesce_counter > 0);
4078 if (bs_queue == NULL) {
4079 bs_queue = g_new0(BlockReopenQueue, 1);
4080 QTAILQ_INIT(bs_queue);
4083 if (!options) {
4084 options = qdict_new();
4087 /* Check if this BlockDriverState is already in the queue */
4088 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4089 if (bs == bs_entry->state.bs) {
4090 break;
4095 * Precedence of options:
4096 * 1. Explicitly passed in options (highest)
4097 * 2. Retained from explicitly set options of bs
4098 * 3. Inherited from parent node
4099 * 4. Retained from effective options of bs
4102 /* Old explicitly set values (don't overwrite by inherited value) */
4103 if (bs_entry || keep_old_opts) {
4104 old_options = qdict_clone_shallow(bs_entry ?
4105 bs_entry->state.explicit_options :
4106 bs->explicit_options);
4107 bdrv_join_options(bs, options, old_options);
4108 qobject_unref(old_options);
4111 explicit_options = qdict_clone_shallow(options);
4113 /* Inherit from parent node */
4114 if (parent_options) {
4115 flags = 0;
4116 klass->inherit_options(role, parent_is_format, &flags, options,
4117 parent_flags, parent_options);
4118 } else {
4119 flags = bdrv_get_flags(bs);
4122 if (keep_old_opts) {
4123 /* Old values are used for options that aren't set yet */
4124 old_options = qdict_clone_shallow(bs->options);
4125 bdrv_join_options(bs, options, old_options);
4126 qobject_unref(old_options);
4129 /* We have the final set of options so let's update the flags */
4130 options_copy = qdict_clone_shallow(options);
4131 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4132 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4133 update_flags_from_options(&flags, opts);
4134 qemu_opts_del(opts);
4135 qobject_unref(options_copy);
4137 /* bdrv_open_inherit() sets and clears some additional flags internally */
4138 flags &= ~BDRV_O_PROTOCOL;
4139 if (flags & BDRV_O_RDWR) {
4140 flags |= BDRV_O_ALLOW_RDWR;
4143 if (!bs_entry) {
4144 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4145 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4146 } else {
4147 qobject_unref(bs_entry->state.options);
4148 qobject_unref(bs_entry->state.explicit_options);
4151 bs_entry->state.bs = bs;
4152 bs_entry->state.options = options;
4153 bs_entry->state.explicit_options = explicit_options;
4154 bs_entry->state.flags = flags;
4157 * If keep_old_opts is false then it means that unspecified
4158 * options must be reset to their original value. We don't allow
4159 * resetting 'backing' but we need to know if the option is
4160 * missing in order to decide if we have to return an error.
4162 if (!keep_old_opts) {
4163 bs_entry->state.backing_missing =
4164 !qdict_haskey(options, "backing") &&
4165 !qdict_haskey(options, "backing.driver");
4168 QLIST_FOREACH(child, &bs->children, next) {
4169 QDict *new_child_options = NULL;
4170 bool child_keep_old = keep_old_opts;
4172 /* reopen can only change the options of block devices that were
4173 * implicitly created and inherited options. For other (referenced)
4174 * block devices, a syntax like "backing.foo" results in an error. */
4175 if (child->bs->inherits_from != bs) {
4176 continue;
4179 /* Check if the options contain a child reference */
4180 if (qdict_haskey(options, child->name)) {
4181 const char *childref = qdict_get_try_str(options, child->name);
4183 * The current child must not be reopened if the child
4184 * reference is null or points to a different node.
4186 if (g_strcmp0(childref, child->bs->node_name)) {
4187 continue;
4190 * If the child reference points to the current child then
4191 * reopen it with its existing set of options (note that
4192 * it can still inherit new options from the parent).
4194 child_keep_old = true;
4195 } else {
4196 /* Extract child options ("child-name.*") */
4197 char *child_key_dot = g_strdup_printf("%s.", child->name);
4198 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4199 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4200 g_free(child_key_dot);
4203 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4204 child->klass, child->role, bs->drv->is_format,
4205 options, flags, child_keep_old);
4208 return bs_queue;
4211 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4212 BlockDriverState *bs,
4213 QDict *options, bool keep_old_opts)
4215 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4216 NULL, 0, keep_old_opts);
4219 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4221 if (bs_queue) {
4222 BlockReopenQueueEntry *bs_entry, *next;
4223 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4224 qobject_unref(bs_entry->state.explicit_options);
4225 qobject_unref(bs_entry->state.options);
4226 g_free(bs_entry);
4228 g_free(bs_queue);
4233 * Reopen multiple BlockDriverStates atomically & transactionally.
4235 * The queue passed in (bs_queue) must have been built up previous
4236 * via bdrv_reopen_queue().
4238 * Reopens all BDS specified in the queue, with the appropriate
4239 * flags. All devices are prepared for reopen, and failure of any
4240 * device will cause all device changes to be abandoned, and intermediate
4241 * data cleaned up.
4243 * If all devices prepare successfully, then the changes are committed
4244 * to all devices.
4246 * All affected nodes must be drained between bdrv_reopen_queue() and
4247 * bdrv_reopen_multiple().
4249 * To be called from the main thread, with all other AioContexts unlocked.
4251 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4253 int ret = -1;
4254 BlockReopenQueueEntry *bs_entry, *next;
4255 AioContext *ctx;
4256 Transaction *tran = tran_new();
4257 g_autoptr(GHashTable) found = NULL;
4258 g_autoptr(GSList) refresh_list = NULL;
4260 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4261 assert(bs_queue != NULL);
4263 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4264 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4265 aio_context_acquire(ctx);
4266 ret = bdrv_flush(bs_entry->state.bs);
4267 aio_context_release(ctx);
4268 if (ret < 0) {
4269 error_setg_errno(errp, -ret, "Error flushing drive");
4270 goto abort;
4274 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4275 assert(bs_entry->state.bs->quiesce_counter > 0);
4276 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4277 aio_context_acquire(ctx);
4278 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4279 aio_context_release(ctx);
4280 if (ret < 0) {
4281 goto abort;
4283 bs_entry->prepared = true;
4286 found = g_hash_table_new(NULL, NULL);
4287 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4288 BDRVReopenState *state = &bs_entry->state;
4290 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4291 if (state->old_backing_bs) {
4292 refresh_list = bdrv_topological_dfs(refresh_list, found,
4293 state->old_backing_bs);
4295 if (state->old_file_bs) {
4296 refresh_list = bdrv_topological_dfs(refresh_list, found,
4297 state->old_file_bs);
4302 * Note that file-posix driver rely on permission update done during reopen
4303 * (even if no permission changed), because it wants "new" permissions for
4304 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4305 * in raw_reopen_prepare() which is called with "old" permissions.
4307 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4308 if (ret < 0) {
4309 goto abort;
4313 * If we reach this point, we have success and just need to apply the
4314 * changes.
4316 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4317 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4318 * children are usually goes after parents in reopen-queue, so go from last
4319 * to first element.
4321 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4322 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4323 aio_context_acquire(ctx);
4324 bdrv_reopen_commit(&bs_entry->state);
4325 aio_context_release(ctx);
4328 tran_commit(tran);
4330 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4331 BlockDriverState *bs = bs_entry->state.bs;
4333 if (bs->drv->bdrv_reopen_commit_post) {
4334 ctx = bdrv_get_aio_context(bs);
4335 aio_context_acquire(ctx);
4336 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4337 aio_context_release(ctx);
4341 ret = 0;
4342 goto cleanup;
4344 abort:
4345 tran_abort(tran);
4346 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4347 if (bs_entry->prepared) {
4348 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4349 aio_context_acquire(ctx);
4350 bdrv_reopen_abort(&bs_entry->state);
4351 aio_context_release(ctx);
4355 cleanup:
4356 bdrv_reopen_queue_free(bs_queue);
4358 return ret;
4361 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4362 Error **errp)
4364 AioContext *ctx = bdrv_get_aio_context(bs);
4365 BlockReopenQueue *queue;
4366 int ret;
4368 bdrv_subtree_drained_begin(bs);
4369 if (ctx != qemu_get_aio_context()) {
4370 aio_context_release(ctx);
4373 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4374 ret = bdrv_reopen_multiple(queue, errp);
4376 if (ctx != qemu_get_aio_context()) {
4377 aio_context_acquire(ctx);
4379 bdrv_subtree_drained_end(bs);
4381 return ret;
4384 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4385 Error **errp)
4387 QDict *opts = qdict_new();
4389 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4391 return bdrv_reopen(bs, opts, true, errp);
4395 * Take a BDRVReopenState and check if the value of 'backing' in the
4396 * reopen_state->options QDict is valid or not.
4398 * If 'backing' is missing from the QDict then return 0.
4400 * If 'backing' contains the node name of the backing file of
4401 * reopen_state->bs then return 0.
4403 * If 'backing' contains a different node name (or is null) then check
4404 * whether the current backing file can be replaced with the new one.
4405 * If that's the case then reopen_state->replace_backing_bs is set to
4406 * true and reopen_state->new_backing_bs contains a pointer to the new
4407 * backing BlockDriverState (or NULL).
4409 * Return 0 on success, otherwise return < 0 and set @errp.
4411 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4412 bool is_backing, Transaction *tran,
4413 Error **errp)
4415 BlockDriverState *bs = reopen_state->bs;
4416 BlockDriverState *new_child_bs;
4417 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4418 child_bs(bs->file);
4419 const char *child_name = is_backing ? "backing" : "file";
4420 QObject *value;
4421 const char *str;
4423 value = qdict_get(reopen_state->options, child_name);
4424 if (value == NULL) {
4425 return 0;
4428 switch (qobject_type(value)) {
4429 case QTYPE_QNULL:
4430 assert(is_backing); /* The 'file' option does not allow a null value */
4431 new_child_bs = NULL;
4432 break;
4433 case QTYPE_QSTRING:
4434 str = qstring_get_str(qobject_to(QString, value));
4435 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4436 if (new_child_bs == NULL) {
4437 return -EINVAL;
4438 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4439 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4440 "cycle", str, child_name, bs->node_name);
4441 return -EINVAL;
4443 break;
4444 default:
4446 * The options QDict has been flattened, so 'backing' and 'file'
4447 * do not allow any other data type here.
4449 g_assert_not_reached();
4452 if (old_child_bs == new_child_bs) {
4453 return 0;
4456 if (old_child_bs) {
4457 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4458 return 0;
4461 if (old_child_bs->implicit) {
4462 error_setg(errp, "Cannot replace implicit %s child of %s",
4463 child_name, bs->node_name);
4464 return -EPERM;
4468 if (bs->drv->is_filter && !old_child_bs) {
4470 * Filters always have a file or a backing child, so we are trying to
4471 * change wrong child
4473 error_setg(errp, "'%s' is a %s filter node that does not support a "
4474 "%s child", bs->node_name, bs->drv->format_name, child_name);
4475 return -EINVAL;
4478 if (is_backing) {
4479 reopen_state->old_backing_bs = old_child_bs;
4480 } else {
4481 reopen_state->old_file_bs = old_child_bs;
4484 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4485 tran, errp);
4489 * Prepares a BlockDriverState for reopen. All changes are staged in the
4490 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4491 * the block driver layer .bdrv_reopen_prepare()
4493 * bs is the BlockDriverState to reopen
4494 * flags are the new open flags
4495 * queue is the reopen queue
4497 * Returns 0 on success, non-zero on error. On error errp will be set
4498 * as well.
4500 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4501 * It is the responsibility of the caller to then call the abort() or
4502 * commit() for any other BDS that have been left in a prepare() state
4505 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4506 BlockReopenQueue *queue,
4507 Transaction *change_child_tran, Error **errp)
4509 int ret = -1;
4510 int old_flags;
4511 Error *local_err = NULL;
4512 BlockDriver *drv;
4513 QemuOpts *opts;
4514 QDict *orig_reopen_opts;
4515 char *discard = NULL;
4516 bool read_only;
4517 bool drv_prepared = false;
4519 assert(reopen_state != NULL);
4520 assert(reopen_state->bs->drv != NULL);
4521 drv = reopen_state->bs->drv;
4523 /* This function and each driver's bdrv_reopen_prepare() remove
4524 * entries from reopen_state->options as they are processed, so
4525 * we need to make a copy of the original QDict. */
4526 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4528 /* Process generic block layer options */
4529 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4530 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4531 ret = -EINVAL;
4532 goto error;
4535 /* This was already called in bdrv_reopen_queue_child() so the flags
4536 * are up-to-date. This time we simply want to remove the options from
4537 * QemuOpts in order to indicate that they have been processed. */
4538 old_flags = reopen_state->flags;
4539 update_flags_from_options(&reopen_state->flags, opts);
4540 assert(old_flags == reopen_state->flags);
4542 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4543 if (discard != NULL) {
4544 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4545 error_setg(errp, "Invalid discard option");
4546 ret = -EINVAL;
4547 goto error;
4551 reopen_state->detect_zeroes =
4552 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4553 if (local_err) {
4554 error_propagate(errp, local_err);
4555 ret = -EINVAL;
4556 goto error;
4559 /* All other options (including node-name and driver) must be unchanged.
4560 * Put them back into the QDict, so that they are checked at the end
4561 * of this function. */
4562 qemu_opts_to_qdict(opts, reopen_state->options);
4564 /* If we are to stay read-only, do not allow permission change
4565 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4566 * not set, or if the BDS still has copy_on_read enabled */
4567 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4568 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4569 if (local_err) {
4570 error_propagate(errp, local_err);
4571 goto error;
4574 if (drv->bdrv_reopen_prepare) {
4576 * If a driver-specific option is missing, it means that we
4577 * should reset it to its default value.
4578 * But not all options allow that, so we need to check it first.
4580 ret = bdrv_reset_options_allowed(reopen_state->bs,
4581 reopen_state->options, errp);
4582 if (ret) {
4583 goto error;
4586 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4587 if (ret) {
4588 if (local_err != NULL) {
4589 error_propagate(errp, local_err);
4590 } else {
4591 bdrv_refresh_filename(reopen_state->bs);
4592 error_setg(errp, "failed while preparing to reopen image '%s'",
4593 reopen_state->bs->filename);
4595 goto error;
4597 } else {
4598 /* It is currently mandatory to have a bdrv_reopen_prepare()
4599 * handler for each supported drv. */
4600 error_setg(errp, "Block format '%s' used by node '%s' "
4601 "does not support reopening files", drv->format_name,
4602 bdrv_get_device_or_node_name(reopen_state->bs));
4603 ret = -1;
4604 goto error;
4607 drv_prepared = true;
4610 * We must provide the 'backing' option if the BDS has a backing
4611 * file or if the image file has a backing file name as part of
4612 * its metadata. Otherwise the 'backing' option can be omitted.
4614 if (drv->supports_backing && reopen_state->backing_missing &&
4615 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4616 error_setg(errp, "backing is missing for '%s'",
4617 reopen_state->bs->node_name);
4618 ret = -EINVAL;
4619 goto error;
4623 * Allow changing the 'backing' option. The new value can be
4624 * either a reference to an existing node (using its node name)
4625 * or NULL to simply detach the current backing file.
4627 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4628 change_child_tran, errp);
4629 if (ret < 0) {
4630 goto error;
4632 qdict_del(reopen_state->options, "backing");
4634 /* Allow changing the 'file' option. In this case NULL is not allowed */
4635 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4636 change_child_tran, errp);
4637 if (ret < 0) {
4638 goto error;
4640 qdict_del(reopen_state->options, "file");
4642 /* Options that are not handled are only okay if they are unchanged
4643 * compared to the old state. It is expected that some options are only
4644 * used for the initial open, but not reopen (e.g. filename) */
4645 if (qdict_size(reopen_state->options)) {
4646 const QDictEntry *entry = qdict_first(reopen_state->options);
4648 do {
4649 QObject *new = entry->value;
4650 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4652 /* Allow child references (child_name=node_name) as long as they
4653 * point to the current child (i.e. everything stays the same). */
4654 if (qobject_type(new) == QTYPE_QSTRING) {
4655 BdrvChild *child;
4656 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4657 if (!strcmp(child->name, entry->key)) {
4658 break;
4662 if (child) {
4663 if (!strcmp(child->bs->node_name,
4664 qstring_get_str(qobject_to(QString, new)))) {
4665 continue; /* Found child with this name, skip option */
4671 * TODO: When using -drive to specify blockdev options, all values
4672 * will be strings; however, when using -blockdev, blockdev-add or
4673 * filenames using the json:{} pseudo-protocol, they will be
4674 * correctly typed.
4675 * In contrast, reopening options are (currently) always strings
4676 * (because you can only specify them through qemu-io; all other
4677 * callers do not specify any options).
4678 * Therefore, when using anything other than -drive to create a BDS,
4679 * this cannot detect non-string options as unchanged, because
4680 * qobject_is_equal() always returns false for objects of different
4681 * type. In the future, this should be remedied by correctly typing
4682 * all options. For now, this is not too big of an issue because
4683 * the user can simply omit options which cannot be changed anyway,
4684 * so they will stay unchanged.
4686 if (!qobject_is_equal(new, old)) {
4687 error_setg(errp, "Cannot change the option '%s'", entry->key);
4688 ret = -EINVAL;
4689 goto error;
4691 } while ((entry = qdict_next(reopen_state->options, entry)));
4694 ret = 0;
4696 /* Restore the original reopen_state->options QDict */
4697 qobject_unref(reopen_state->options);
4698 reopen_state->options = qobject_ref(orig_reopen_opts);
4700 error:
4701 if (ret < 0 && drv_prepared) {
4702 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4703 * call drv->bdrv_reopen_abort() before signaling an error
4704 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4705 * when the respective bdrv_reopen_prepare() has failed) */
4706 if (drv->bdrv_reopen_abort) {
4707 drv->bdrv_reopen_abort(reopen_state);
4710 qemu_opts_del(opts);
4711 qobject_unref(orig_reopen_opts);
4712 g_free(discard);
4713 return ret;
4717 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4718 * makes them final by swapping the staging BlockDriverState contents into
4719 * the active BlockDriverState contents.
4721 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4723 BlockDriver *drv;
4724 BlockDriverState *bs;
4725 BdrvChild *child;
4727 assert(reopen_state != NULL);
4728 bs = reopen_state->bs;
4729 drv = bs->drv;
4730 assert(drv != NULL);
4732 /* If there are any driver level actions to take */
4733 if (drv->bdrv_reopen_commit) {
4734 drv->bdrv_reopen_commit(reopen_state);
4737 /* set BDS specific flags now */
4738 qobject_unref(bs->explicit_options);
4739 qobject_unref(bs->options);
4740 qobject_ref(reopen_state->explicit_options);
4741 qobject_ref(reopen_state->options);
4743 bs->explicit_options = reopen_state->explicit_options;
4744 bs->options = reopen_state->options;
4745 bs->open_flags = reopen_state->flags;
4746 bs->detect_zeroes = reopen_state->detect_zeroes;
4748 /* Remove child references from bs->options and bs->explicit_options.
4749 * Child options were already removed in bdrv_reopen_queue_child() */
4750 QLIST_FOREACH(child, &bs->children, next) {
4751 qdict_del(bs->explicit_options, child->name);
4752 qdict_del(bs->options, child->name);
4754 /* backing is probably removed, so it's not handled by previous loop */
4755 qdict_del(bs->explicit_options, "backing");
4756 qdict_del(bs->options, "backing");
4758 bdrv_refresh_limits(bs, NULL, NULL);
4762 * Abort the reopen, and delete and free the staged changes in
4763 * reopen_state
4765 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4767 BlockDriver *drv;
4769 assert(reopen_state != NULL);
4770 drv = reopen_state->bs->drv;
4771 assert(drv != NULL);
4773 if (drv->bdrv_reopen_abort) {
4774 drv->bdrv_reopen_abort(reopen_state);
4779 static void bdrv_close(BlockDriverState *bs)
4781 BdrvAioNotifier *ban, *ban_next;
4782 BdrvChild *child, *next;
4784 assert(!bs->refcnt);
4786 bdrv_drained_begin(bs); /* complete I/O */
4787 bdrv_flush(bs);
4788 bdrv_drain(bs); /* in case flush left pending I/O */
4790 if (bs->drv) {
4791 if (bs->drv->bdrv_close) {
4792 /* Must unfreeze all children, so bdrv_unref_child() works */
4793 bs->drv->bdrv_close(bs);
4795 bs->drv = NULL;
4798 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4799 bdrv_unref_child(bs, child);
4802 bs->backing = NULL;
4803 bs->file = NULL;
4804 g_free(bs->opaque);
4805 bs->opaque = NULL;
4806 qatomic_set(&bs->copy_on_read, 0);
4807 bs->backing_file[0] = '\0';
4808 bs->backing_format[0] = '\0';
4809 bs->total_sectors = 0;
4810 bs->encrypted = false;
4811 bs->sg = false;
4812 qobject_unref(bs->options);
4813 qobject_unref(bs->explicit_options);
4814 bs->options = NULL;
4815 bs->explicit_options = NULL;
4816 qobject_unref(bs->full_open_options);
4817 bs->full_open_options = NULL;
4818 g_free(bs->block_status_cache);
4819 bs->block_status_cache = NULL;
4821 bdrv_release_named_dirty_bitmaps(bs);
4822 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4824 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4825 g_free(ban);
4827 QLIST_INIT(&bs->aio_notifiers);
4828 bdrv_drained_end(bs);
4831 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4832 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4833 * gets called.
4835 if (bs->quiesce_counter) {
4836 bdrv_drain_all_end_quiesce(bs);
4840 void bdrv_close_all(void)
4842 assert(job_next(NULL) == NULL);
4844 /* Drop references from requests still in flight, such as canceled block
4845 * jobs whose AIO context has not been polled yet */
4846 bdrv_drain_all();
4848 blk_remove_all_bs();
4849 blockdev_close_all_bdrv_states();
4851 assert(QTAILQ_EMPTY(&all_bdrv_states));
4854 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4856 GQueue *queue;
4857 GHashTable *found;
4858 bool ret;
4860 if (c->klass->stay_at_node) {
4861 return false;
4864 /* If the child @c belongs to the BDS @to, replacing the current
4865 * c->bs by @to would mean to create a loop.
4867 * Such a case occurs when appending a BDS to a backing chain.
4868 * For instance, imagine the following chain:
4870 * guest device -> node A -> further backing chain...
4872 * Now we create a new BDS B which we want to put on top of this
4873 * chain, so we first attach A as its backing node:
4875 * node B
4878 * guest device -> node A -> further backing chain...
4880 * Finally we want to replace A by B. When doing that, we want to
4881 * replace all pointers to A by pointers to B -- except for the
4882 * pointer from B because (1) that would create a loop, and (2)
4883 * that pointer should simply stay intact:
4885 * guest device -> node B
4888 * node A -> further backing chain...
4890 * In general, when replacing a node A (c->bs) by a node B (@to),
4891 * if A is a child of B, that means we cannot replace A by B there
4892 * because that would create a loop. Silently detaching A from B
4893 * is also not really an option. So overall just leaving A in
4894 * place there is the most sensible choice.
4896 * We would also create a loop in any cases where @c is only
4897 * indirectly referenced by @to. Prevent this by returning false
4898 * if @c is found (by breadth-first search) anywhere in the whole
4899 * subtree of @to.
4902 ret = true;
4903 found = g_hash_table_new(NULL, NULL);
4904 g_hash_table_add(found, to);
4905 queue = g_queue_new();
4906 g_queue_push_tail(queue, to);
4908 while (!g_queue_is_empty(queue)) {
4909 BlockDriverState *v = g_queue_pop_head(queue);
4910 BdrvChild *c2;
4912 QLIST_FOREACH(c2, &v->children, next) {
4913 if (c2 == c) {
4914 ret = false;
4915 break;
4918 if (g_hash_table_contains(found, c2->bs)) {
4919 continue;
4922 g_queue_push_tail(queue, c2->bs);
4923 g_hash_table_add(found, c2->bs);
4927 g_queue_free(queue);
4928 g_hash_table_destroy(found);
4930 return ret;
4933 typedef struct BdrvRemoveFilterOrCowChild {
4934 BdrvChild *child;
4935 BlockDriverState *bs;
4936 bool is_backing;
4937 } BdrvRemoveFilterOrCowChild;
4939 static void bdrv_remove_filter_or_cow_child_abort(void *opaque)
4941 BdrvRemoveFilterOrCowChild *s = opaque;
4942 BlockDriverState *parent_bs = s->child->opaque;
4944 if (s->is_backing) {
4945 parent_bs->backing = s->child;
4946 } else {
4947 parent_bs->file = s->child;
4951 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
4952 * because that function is transactionable and it registered own completion
4953 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
4954 * called automatically.
4958 static void bdrv_remove_filter_or_cow_child_commit(void *opaque)
4960 BdrvRemoveFilterOrCowChild *s = opaque;
4962 bdrv_child_free(s->child);
4965 static void bdrv_remove_filter_or_cow_child_clean(void *opaque)
4967 BdrvRemoveFilterOrCowChild *s = opaque;
4969 /* Drop the bs reference after the transaction is done */
4970 bdrv_unref(s->bs);
4971 g_free(s);
4974 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = {
4975 .abort = bdrv_remove_filter_or_cow_child_abort,
4976 .commit = bdrv_remove_filter_or_cow_child_commit,
4977 .clean = bdrv_remove_filter_or_cow_child_clean,
4981 * A function to remove backing or file child of @bs.
4982 * Function doesn't update permissions, caller is responsible for this.
4984 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
4985 BdrvChild *child,
4986 Transaction *tran)
4988 BdrvChild **childp;
4989 BdrvRemoveFilterOrCowChild *s;
4991 if (!child) {
4992 return;
4996 * Keep a reference to @bs so @childp will stay valid throughout the
4997 * transaction (required by bdrv_replace_child_tran())
4999 bdrv_ref(bs);
5000 if (child == bs->backing) {
5001 childp = &bs->backing;
5002 } else if (child == bs->file) {
5003 childp = &bs->file;
5004 } else {
5005 g_assert_not_reached();
5008 if (child->bs) {
5010 * Pass free_empty_child=false, we will free the child in
5011 * bdrv_remove_filter_or_cow_child_commit()
5013 bdrv_replace_child_tran(childp, NULL, tran, false);
5016 s = g_new(BdrvRemoveFilterOrCowChild, 1);
5017 *s = (BdrvRemoveFilterOrCowChild) {
5018 .child = child,
5019 .bs = bs,
5020 .is_backing = (childp == &bs->backing),
5022 tran_add(tran, &bdrv_remove_filter_or_cow_child_drv, s);
5026 * A function to remove backing-chain child of @bs if exists: cow child for
5027 * format nodes (always .backing) and filter child for filters (may be .file or
5028 * .backing)
5030 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
5031 Transaction *tran)
5033 bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran);
5036 static int bdrv_replace_node_noperm(BlockDriverState *from,
5037 BlockDriverState *to,
5038 bool auto_skip, Transaction *tran,
5039 Error **errp)
5041 BdrvChild *c, *next;
5043 assert(to != NULL);
5045 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5046 assert(c->bs == from);
5047 if (!should_update_child(c, to)) {
5048 if (auto_skip) {
5049 continue;
5051 error_setg(errp, "Should not change '%s' link to '%s'",
5052 c->name, from->node_name);
5053 return -EINVAL;
5055 if (c->frozen) {
5056 error_setg(errp, "Cannot change '%s' link to '%s'",
5057 c->name, from->node_name);
5058 return -EPERM;
5062 * Passing a pointer to the local variable @c is fine here, because
5063 * @to is not NULL, and so &c will not be attached to the transaction.
5065 bdrv_replace_child_tran(&c, to, tran, true);
5068 return 0;
5072 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5073 * if it creates a parent-child relation loop or if parent is block-job.
5075 * With auto_skip=false the error is returned if from has a parent which should
5076 * not be updated.
5078 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5079 * case backing link of the cow-parent of @to is removed.
5081 * @to must not be NULL.
5083 static int bdrv_replace_node_common(BlockDriverState *from,
5084 BlockDriverState *to,
5085 bool auto_skip, bool detach_subchain,
5086 Error **errp)
5088 Transaction *tran = tran_new();
5089 g_autoptr(GHashTable) found = NULL;
5090 g_autoptr(GSList) refresh_list = NULL;
5091 BlockDriverState *to_cow_parent = NULL;
5092 int ret;
5094 assert(to != NULL);
5096 if (detach_subchain) {
5097 assert(bdrv_chain_contains(from, to));
5098 assert(from != to);
5099 for (to_cow_parent = from;
5100 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5101 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5107 /* Make sure that @from doesn't go away until we have successfully attached
5108 * all of its parents to @to. */
5109 bdrv_ref(from);
5111 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5112 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5113 bdrv_drained_begin(from);
5116 * Do the replacement without permission update.
5117 * Replacement may influence the permissions, we should calculate new
5118 * permissions based on new graph. If we fail, we'll roll-back the
5119 * replacement.
5121 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5122 if (ret < 0) {
5123 goto out;
5126 if (detach_subchain) {
5127 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
5130 found = g_hash_table_new(NULL, NULL);
5132 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
5133 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
5135 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5136 if (ret < 0) {
5137 goto out;
5140 ret = 0;
5142 out:
5143 tran_finalize(tran, ret);
5145 bdrv_drained_end(from);
5146 bdrv_unref(from);
5148 return ret;
5152 * Replace node @from by @to (where neither may be NULL).
5154 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5155 Error **errp)
5157 return bdrv_replace_node_common(from, to, true, false, errp);
5160 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5162 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5163 errp);
5167 * Add new bs contents at the top of an image chain while the chain is
5168 * live, while keeping required fields on the top layer.
5170 * This will modify the BlockDriverState fields, and swap contents
5171 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5173 * bs_new must not be attached to a BlockBackend and must not have backing
5174 * child.
5176 * This function does not create any image files.
5178 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5179 Error **errp)
5181 int ret;
5182 Transaction *tran = tran_new();
5184 assert(!bs_new->backing);
5186 ret = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5187 &child_of_bds, bdrv_backing_role(bs_new),
5188 &bs_new->backing, tran, errp);
5189 if (ret < 0) {
5190 goto out;
5193 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5194 if (ret < 0) {
5195 goto out;
5198 ret = bdrv_refresh_perms(bs_new, errp);
5199 out:
5200 tran_finalize(tran, ret);
5202 bdrv_refresh_limits(bs_top, NULL, NULL);
5204 return ret;
5207 /* Not for empty child */
5208 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5209 Error **errp)
5211 int ret;
5212 Transaction *tran = tran_new();
5213 g_autoptr(GHashTable) found = NULL;
5214 g_autoptr(GSList) refresh_list = NULL;
5215 BlockDriverState *old_bs = child->bs;
5217 bdrv_ref(old_bs);
5218 bdrv_drained_begin(old_bs);
5219 bdrv_drained_begin(new_bs);
5221 bdrv_replace_child_tran(&child, new_bs, tran, true);
5222 /* @new_bs must have been non-NULL, so @child must not have been freed */
5223 assert(child != NULL);
5225 found = g_hash_table_new(NULL, NULL);
5226 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5227 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5229 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5231 tran_finalize(tran, ret);
5233 bdrv_drained_end(old_bs);
5234 bdrv_drained_end(new_bs);
5235 bdrv_unref(old_bs);
5237 return ret;
5240 static void bdrv_delete(BlockDriverState *bs)
5242 assert(bdrv_op_blocker_is_empty(bs));
5243 assert(!bs->refcnt);
5245 /* remove from list, if necessary */
5246 if (bs->node_name[0] != '\0') {
5247 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5249 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5251 bdrv_close(bs);
5253 g_free(bs);
5258 * Replace @bs by newly created block node.
5260 * @options is a QDict of options to pass to the block drivers, or NULL for an
5261 * empty set of options. The reference to the QDict belongs to the block layer
5262 * after the call (even on failure), so if the caller intends to reuse the
5263 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5265 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5266 int flags, Error **errp)
5268 ERRP_GUARD();
5269 int ret;
5270 BlockDriverState *new_node_bs = NULL;
5271 const char *drvname, *node_name;
5272 BlockDriver *drv;
5274 drvname = qdict_get_try_str(options, "driver");
5275 if (!drvname) {
5276 error_setg(errp, "driver is not specified");
5277 goto fail;
5280 drv = bdrv_find_format(drvname);
5281 if (!drv) {
5282 error_setg(errp, "Unknown driver: '%s'", drvname);
5283 goto fail;
5286 node_name = qdict_get_try_str(options, "node-name");
5288 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5289 errp);
5290 options = NULL; /* bdrv_new_open_driver() eats options */
5291 if (!new_node_bs) {
5292 error_prepend(errp, "Could not create node: ");
5293 goto fail;
5296 bdrv_drained_begin(bs);
5297 ret = bdrv_replace_node(bs, new_node_bs, errp);
5298 bdrv_drained_end(bs);
5300 if (ret < 0) {
5301 error_prepend(errp, "Could not replace node: ");
5302 goto fail;
5305 return new_node_bs;
5307 fail:
5308 qobject_unref(options);
5309 bdrv_unref(new_node_bs);
5310 return NULL;
5314 * Run consistency checks on an image
5316 * Returns 0 if the check could be completed (it doesn't mean that the image is
5317 * free of errors) or -errno when an internal error occurred. The results of the
5318 * check are stored in res.
5320 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5321 BdrvCheckResult *res, BdrvCheckMode fix)
5323 if (bs->drv == NULL) {
5324 return -ENOMEDIUM;
5326 if (bs->drv->bdrv_co_check == NULL) {
5327 return -ENOTSUP;
5330 memset(res, 0, sizeof(*res));
5331 return bs->drv->bdrv_co_check(bs, res, fix);
5335 * Return values:
5336 * 0 - success
5337 * -EINVAL - backing format specified, but no file
5338 * -ENOSPC - can't update the backing file because no space is left in the
5339 * image file header
5340 * -ENOTSUP - format driver doesn't support changing the backing file
5342 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5343 const char *backing_fmt, bool require)
5345 BlockDriver *drv = bs->drv;
5346 int ret;
5348 if (!drv) {
5349 return -ENOMEDIUM;
5352 /* Backing file format doesn't make sense without a backing file */
5353 if (backing_fmt && !backing_file) {
5354 return -EINVAL;
5357 if (require && backing_file && !backing_fmt) {
5358 return -EINVAL;
5361 if (drv->bdrv_change_backing_file != NULL) {
5362 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5363 } else {
5364 ret = -ENOTSUP;
5367 if (ret == 0) {
5368 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5369 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5370 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5371 backing_file ?: "");
5373 return ret;
5377 * Finds the first non-filter node above bs in the chain between
5378 * active and bs. The returned node is either an immediate parent of
5379 * bs, or there are only filter nodes between the two.
5381 * Returns NULL if bs is not found in active's image chain,
5382 * or if active == bs.
5384 * Returns the bottommost base image if bs == NULL.
5386 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5387 BlockDriverState *bs)
5389 bs = bdrv_skip_filters(bs);
5390 active = bdrv_skip_filters(active);
5392 while (active) {
5393 BlockDriverState *next = bdrv_backing_chain_next(active);
5394 if (bs == next) {
5395 return active;
5397 active = next;
5400 return NULL;
5403 /* Given a BDS, searches for the base layer. */
5404 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5406 return bdrv_find_overlay(bs, NULL);
5410 * Return true if at least one of the COW (backing) and filter links
5411 * between @bs and @base is frozen. @errp is set if that's the case.
5412 * @base must be reachable from @bs, or NULL.
5414 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5415 Error **errp)
5417 BlockDriverState *i;
5418 BdrvChild *child;
5420 for (i = bs; i != base; i = child_bs(child)) {
5421 child = bdrv_filter_or_cow_child(i);
5423 if (child && child->frozen) {
5424 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5425 child->name, i->node_name, child->bs->node_name);
5426 return true;
5430 return false;
5434 * Freeze all COW (backing) and filter links between @bs and @base.
5435 * If any of the links is already frozen the operation is aborted and
5436 * none of the links are modified.
5437 * @base must be reachable from @bs, or NULL.
5438 * Returns 0 on success. On failure returns < 0 and sets @errp.
5440 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5441 Error **errp)
5443 BlockDriverState *i;
5444 BdrvChild *child;
5446 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5447 return -EPERM;
5450 for (i = bs; i != base; i = child_bs(child)) {
5451 child = bdrv_filter_or_cow_child(i);
5452 if (child && child->bs->never_freeze) {
5453 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5454 child->name, child->bs->node_name);
5455 return -EPERM;
5459 for (i = bs; i != base; i = child_bs(child)) {
5460 child = bdrv_filter_or_cow_child(i);
5461 if (child) {
5462 child->frozen = true;
5466 return 0;
5470 * Unfreeze all COW (backing) and filter links between @bs and @base.
5471 * The caller must ensure that all links are frozen before using this
5472 * function.
5473 * @base must be reachable from @bs, or NULL.
5475 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5477 BlockDriverState *i;
5478 BdrvChild *child;
5480 for (i = bs; i != base; i = child_bs(child)) {
5481 child = bdrv_filter_or_cow_child(i);
5482 if (child) {
5483 assert(child->frozen);
5484 child->frozen = false;
5490 * Drops images above 'base' up to and including 'top', and sets the image
5491 * above 'top' to have base as its backing file.
5493 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5494 * information in 'bs' can be properly updated.
5496 * E.g., this will convert the following chain:
5497 * bottom <- base <- intermediate <- top <- active
5499 * to
5501 * bottom <- base <- active
5503 * It is allowed for bottom==base, in which case it converts:
5505 * base <- intermediate <- top <- active
5507 * to
5509 * base <- active
5511 * If backing_file_str is non-NULL, it will be used when modifying top's
5512 * overlay image metadata.
5514 * Error conditions:
5515 * if active == top, that is considered an error
5518 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5519 const char *backing_file_str)
5521 BlockDriverState *explicit_top = top;
5522 bool update_inherits_from;
5523 BdrvChild *c;
5524 Error *local_err = NULL;
5525 int ret = -EIO;
5526 g_autoptr(GSList) updated_children = NULL;
5527 GSList *p;
5529 bdrv_ref(top);
5530 bdrv_subtree_drained_begin(top);
5532 if (!top->drv || !base->drv) {
5533 goto exit;
5536 /* Make sure that base is in the backing chain of top */
5537 if (!bdrv_chain_contains(top, base)) {
5538 goto exit;
5541 /* If 'base' recursively inherits from 'top' then we should set
5542 * base->inherits_from to top->inherits_from after 'top' and all
5543 * other intermediate nodes have been dropped.
5544 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5545 * it because no one inherits from it. We use explicit_top for that. */
5546 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5547 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5549 /* success - we can delete the intermediate states, and link top->base */
5550 if (!backing_file_str) {
5551 bdrv_refresh_filename(base);
5552 backing_file_str = base->filename;
5555 QLIST_FOREACH(c, &top->parents, next_parent) {
5556 updated_children = g_slist_prepend(updated_children, c);
5560 * It seems correct to pass detach_subchain=true here, but it triggers
5561 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5562 * another drained section, which modify the graph (for example, removing
5563 * the child, which we keep in updated_children list). So, it's a TODO.
5565 * Note, bug triggered if pass detach_subchain=true here and run
5566 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5567 * That's a FIXME.
5569 bdrv_replace_node_common(top, base, false, false, &local_err);
5570 if (local_err) {
5571 error_report_err(local_err);
5572 goto exit;
5575 for (p = updated_children; p; p = p->next) {
5576 c = p->data;
5578 if (c->klass->update_filename) {
5579 ret = c->klass->update_filename(c, base, backing_file_str,
5580 &local_err);
5581 if (ret < 0) {
5583 * TODO: Actually, we want to rollback all previous iterations
5584 * of this loop, and (which is almost impossible) previous
5585 * bdrv_replace_node()...
5587 * Note, that c->klass->update_filename may lead to permission
5588 * update, so it's a bad idea to call it inside permission
5589 * update transaction of bdrv_replace_node.
5591 error_report_err(local_err);
5592 goto exit;
5597 if (update_inherits_from) {
5598 base->inherits_from = explicit_top->inherits_from;
5601 ret = 0;
5602 exit:
5603 bdrv_subtree_drained_end(top);
5604 bdrv_unref(top);
5605 return ret;
5609 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5610 * sums the size of all data-bearing children. (This excludes backing
5611 * children.)
5613 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5615 BdrvChild *child;
5616 int64_t child_size, sum = 0;
5618 QLIST_FOREACH(child, &bs->children, next) {
5619 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5620 BDRV_CHILD_FILTERED))
5622 child_size = bdrv_get_allocated_file_size(child->bs);
5623 if (child_size < 0) {
5624 return child_size;
5626 sum += child_size;
5630 return sum;
5634 * Length of a allocated file in bytes. Sparse files are counted by actual
5635 * allocated space. Return < 0 if error or unknown.
5637 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5639 BlockDriver *drv = bs->drv;
5640 if (!drv) {
5641 return -ENOMEDIUM;
5643 if (drv->bdrv_get_allocated_file_size) {
5644 return drv->bdrv_get_allocated_file_size(bs);
5647 if (drv->bdrv_file_open) {
5649 * Protocol drivers default to -ENOTSUP (most of their data is
5650 * not stored in any of their children (if they even have any),
5651 * so there is no generic way to figure it out).
5653 return -ENOTSUP;
5654 } else if (drv->is_filter) {
5655 /* Filter drivers default to the size of their filtered child */
5656 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5657 } else {
5658 /* Other drivers default to summing their children's sizes */
5659 return bdrv_sum_allocated_file_size(bs);
5664 * bdrv_measure:
5665 * @drv: Format driver
5666 * @opts: Creation options for new image
5667 * @in_bs: Existing image containing data for new image (may be NULL)
5668 * @errp: Error object
5669 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5670 * or NULL on error
5672 * Calculate file size required to create a new image.
5674 * If @in_bs is given then space for allocated clusters and zero clusters
5675 * from that image are included in the calculation. If @opts contains a
5676 * backing file that is shared by @in_bs then backing clusters may be omitted
5677 * from the calculation.
5679 * If @in_bs is NULL then the calculation includes no allocated clusters
5680 * unless a preallocation option is given in @opts.
5682 * Note that @in_bs may use a different BlockDriver from @drv.
5684 * If an error occurs the @errp pointer is set.
5686 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5687 BlockDriverState *in_bs, Error **errp)
5689 if (!drv->bdrv_measure) {
5690 error_setg(errp, "Block driver '%s' does not support size measurement",
5691 drv->format_name);
5692 return NULL;
5695 return drv->bdrv_measure(opts, in_bs, errp);
5699 * Return number of sectors on success, -errno on error.
5701 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5703 BlockDriver *drv = bs->drv;
5705 if (!drv)
5706 return -ENOMEDIUM;
5708 if (drv->has_variable_length) {
5709 int ret = refresh_total_sectors(bs, bs->total_sectors);
5710 if (ret < 0) {
5711 return ret;
5714 return bs->total_sectors;
5718 * Return length in bytes on success, -errno on error.
5719 * The length is always a multiple of BDRV_SECTOR_SIZE.
5721 int64_t bdrv_getlength(BlockDriverState *bs)
5723 int64_t ret = bdrv_nb_sectors(bs);
5725 if (ret < 0) {
5726 return ret;
5728 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5729 return -EFBIG;
5731 return ret * BDRV_SECTOR_SIZE;
5734 /* return 0 as number of sectors if no device present or error */
5735 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5737 int64_t nb_sectors = bdrv_nb_sectors(bs);
5739 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5742 bool bdrv_is_sg(BlockDriverState *bs)
5744 return bs->sg;
5748 * Return whether the given node supports compressed writes.
5750 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5752 BlockDriverState *filtered;
5754 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5755 return false;
5758 filtered = bdrv_filter_bs(bs);
5759 if (filtered) {
5761 * Filters can only forward compressed writes, so we have to
5762 * check the child.
5764 return bdrv_supports_compressed_writes(filtered);
5767 return true;
5770 const char *bdrv_get_format_name(BlockDriverState *bs)
5772 return bs->drv ? bs->drv->format_name : NULL;
5775 static int qsort_strcmp(const void *a, const void *b)
5777 return strcmp(*(char *const *)a, *(char *const *)b);
5780 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5781 void *opaque, bool read_only)
5783 BlockDriver *drv;
5784 int count = 0;
5785 int i;
5786 const char **formats = NULL;
5788 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5789 if (drv->format_name) {
5790 bool found = false;
5791 int i = count;
5793 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5794 continue;
5797 while (formats && i && !found) {
5798 found = !strcmp(formats[--i], drv->format_name);
5801 if (!found) {
5802 formats = g_renew(const char *, formats, count + 1);
5803 formats[count++] = drv->format_name;
5808 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5809 const char *format_name = block_driver_modules[i].format_name;
5811 if (format_name) {
5812 bool found = false;
5813 int j = count;
5815 if (use_bdrv_whitelist &&
5816 !bdrv_format_is_whitelisted(format_name, read_only)) {
5817 continue;
5820 while (formats && j && !found) {
5821 found = !strcmp(formats[--j], format_name);
5824 if (!found) {
5825 formats = g_renew(const char *, formats, count + 1);
5826 formats[count++] = format_name;
5831 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5833 for (i = 0; i < count; i++) {
5834 it(opaque, formats[i]);
5837 g_free(formats);
5840 /* This function is to find a node in the bs graph */
5841 BlockDriverState *bdrv_find_node(const char *node_name)
5843 BlockDriverState *bs;
5845 assert(node_name);
5847 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5848 if (!strcmp(node_name, bs->node_name)) {
5849 return bs;
5852 return NULL;
5855 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5856 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5857 Error **errp)
5859 BlockDeviceInfoList *list;
5860 BlockDriverState *bs;
5862 list = NULL;
5863 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5864 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5865 if (!info) {
5866 qapi_free_BlockDeviceInfoList(list);
5867 return NULL;
5869 QAPI_LIST_PREPEND(list, info);
5872 return list;
5875 typedef struct XDbgBlockGraphConstructor {
5876 XDbgBlockGraph *graph;
5877 GHashTable *graph_nodes;
5878 } XDbgBlockGraphConstructor;
5880 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5882 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5884 gr->graph = g_new0(XDbgBlockGraph, 1);
5885 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5887 return gr;
5890 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5892 XDbgBlockGraph *graph = gr->graph;
5894 g_hash_table_destroy(gr->graph_nodes);
5895 g_free(gr);
5897 return graph;
5900 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5902 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5904 if (ret != 0) {
5905 return ret;
5909 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5910 * answer of g_hash_table_lookup.
5912 ret = g_hash_table_size(gr->graph_nodes) + 1;
5913 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5915 return ret;
5918 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5919 XDbgBlockGraphNodeType type, const char *name)
5921 XDbgBlockGraphNode *n;
5923 n = g_new0(XDbgBlockGraphNode, 1);
5925 n->id = xdbg_graph_node_num(gr, node);
5926 n->type = type;
5927 n->name = g_strdup(name);
5929 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5932 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5933 const BdrvChild *child)
5935 BlockPermission qapi_perm;
5936 XDbgBlockGraphEdge *edge;
5938 edge = g_new0(XDbgBlockGraphEdge, 1);
5940 edge->parent = xdbg_graph_node_num(gr, parent);
5941 edge->child = xdbg_graph_node_num(gr, child->bs);
5942 edge->name = g_strdup(child->name);
5944 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5945 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5947 if (flag & child->perm) {
5948 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5950 if (flag & child->shared_perm) {
5951 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5955 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5959 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5961 BlockBackend *blk;
5962 BlockJob *job;
5963 BlockDriverState *bs;
5964 BdrvChild *child;
5965 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5967 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5968 char *allocated_name = NULL;
5969 const char *name = blk_name(blk);
5971 if (!*name) {
5972 name = allocated_name = blk_get_attached_dev_id(blk);
5974 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5975 name);
5976 g_free(allocated_name);
5977 if (blk_root(blk)) {
5978 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5982 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5983 GSList *el;
5985 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5986 job->job.id);
5987 for (el = job->nodes; el; el = el->next) {
5988 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5992 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5993 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5994 bs->node_name);
5995 QLIST_FOREACH(child, &bs->children, next) {
5996 xdbg_graph_add_edge(gr, bs, child);
6000 return xdbg_graph_finalize(gr);
6003 BlockDriverState *bdrv_lookup_bs(const char *device,
6004 const char *node_name,
6005 Error **errp)
6007 BlockBackend *blk;
6008 BlockDriverState *bs;
6010 if (device) {
6011 blk = blk_by_name(device);
6013 if (blk) {
6014 bs = blk_bs(blk);
6015 if (!bs) {
6016 error_setg(errp, "Device '%s' has no medium", device);
6019 return bs;
6023 if (node_name) {
6024 bs = bdrv_find_node(node_name);
6026 if (bs) {
6027 return bs;
6031 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6032 device ? device : "",
6033 node_name ? node_name : "");
6034 return NULL;
6037 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6038 * return false. If either argument is NULL, return false. */
6039 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6041 while (top && top != base) {
6042 top = bdrv_filter_or_cow_bs(top);
6045 return top != NULL;
6048 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6050 if (!bs) {
6051 return QTAILQ_FIRST(&graph_bdrv_states);
6053 return QTAILQ_NEXT(bs, node_list);
6056 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6058 if (!bs) {
6059 return QTAILQ_FIRST(&all_bdrv_states);
6061 return QTAILQ_NEXT(bs, bs_list);
6064 const char *bdrv_get_node_name(const BlockDriverState *bs)
6066 return bs->node_name;
6069 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6071 BdrvChild *c;
6072 const char *name;
6074 /* If multiple parents have a name, just pick the first one. */
6075 QLIST_FOREACH(c, &bs->parents, next_parent) {
6076 if (c->klass->get_name) {
6077 name = c->klass->get_name(c);
6078 if (name && *name) {
6079 return name;
6084 return NULL;
6087 /* TODO check what callers really want: bs->node_name or blk_name() */
6088 const char *bdrv_get_device_name(const BlockDriverState *bs)
6090 return bdrv_get_parent_name(bs) ?: "";
6093 /* This can be used to identify nodes that might not have a device
6094 * name associated. Since node and device names live in the same
6095 * namespace, the result is unambiguous. The exception is if both are
6096 * absent, then this returns an empty (non-null) string. */
6097 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6099 return bdrv_get_parent_name(bs) ?: bs->node_name;
6102 int bdrv_get_flags(BlockDriverState *bs)
6104 return bs->open_flags;
6107 int bdrv_has_zero_init_1(BlockDriverState *bs)
6109 return 1;
6112 int bdrv_has_zero_init(BlockDriverState *bs)
6114 BlockDriverState *filtered;
6116 if (!bs->drv) {
6117 return 0;
6120 /* If BS is a copy on write image, it is initialized to
6121 the contents of the base image, which may not be zeroes. */
6122 if (bdrv_cow_child(bs)) {
6123 return 0;
6125 if (bs->drv->bdrv_has_zero_init) {
6126 return bs->drv->bdrv_has_zero_init(bs);
6129 filtered = bdrv_filter_bs(bs);
6130 if (filtered) {
6131 return bdrv_has_zero_init(filtered);
6134 /* safe default */
6135 return 0;
6138 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6140 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6141 return false;
6144 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6147 void bdrv_get_backing_filename(BlockDriverState *bs,
6148 char *filename, int filename_size)
6150 pstrcpy(filename, filename_size, bs->backing_file);
6153 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6155 int ret;
6156 BlockDriver *drv = bs->drv;
6157 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6158 if (!drv) {
6159 return -ENOMEDIUM;
6161 if (!drv->bdrv_get_info) {
6162 BlockDriverState *filtered = bdrv_filter_bs(bs);
6163 if (filtered) {
6164 return bdrv_get_info(filtered, bdi);
6166 return -ENOTSUP;
6168 memset(bdi, 0, sizeof(*bdi));
6169 ret = drv->bdrv_get_info(bs, bdi);
6170 if (ret < 0) {
6171 return ret;
6174 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6175 return -EINVAL;
6178 return 0;
6181 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6182 Error **errp)
6184 BlockDriver *drv = bs->drv;
6185 if (drv && drv->bdrv_get_specific_info) {
6186 return drv->bdrv_get_specific_info(bs, errp);
6188 return NULL;
6191 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6193 BlockDriver *drv = bs->drv;
6194 if (!drv || !drv->bdrv_get_specific_stats) {
6195 return NULL;
6197 return drv->bdrv_get_specific_stats(bs);
6200 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6202 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6203 return;
6206 bs->drv->bdrv_debug_event(bs, event);
6209 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6211 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6212 bs = bdrv_primary_bs(bs);
6215 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6216 assert(bs->drv->bdrv_debug_remove_breakpoint);
6217 return bs;
6220 return NULL;
6223 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6224 const char *tag)
6226 bs = bdrv_find_debug_node(bs);
6227 if (bs) {
6228 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6231 return -ENOTSUP;
6234 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6236 bs = bdrv_find_debug_node(bs);
6237 if (bs) {
6238 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6241 return -ENOTSUP;
6244 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6246 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6247 bs = bdrv_primary_bs(bs);
6250 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6251 return bs->drv->bdrv_debug_resume(bs, tag);
6254 return -ENOTSUP;
6257 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6259 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6260 bs = bdrv_primary_bs(bs);
6263 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6264 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6267 return false;
6270 /* backing_file can either be relative, or absolute, or a protocol. If it is
6271 * relative, it must be relative to the chain. So, passing in bs->filename
6272 * from a BDS as backing_file should not be done, as that may be relative to
6273 * the CWD rather than the chain. */
6274 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6275 const char *backing_file)
6277 char *filename_full = NULL;
6278 char *backing_file_full = NULL;
6279 char *filename_tmp = NULL;
6280 int is_protocol = 0;
6281 bool filenames_refreshed = false;
6282 BlockDriverState *curr_bs = NULL;
6283 BlockDriverState *retval = NULL;
6284 BlockDriverState *bs_below;
6286 if (!bs || !bs->drv || !backing_file) {
6287 return NULL;
6290 filename_full = g_malloc(PATH_MAX);
6291 backing_file_full = g_malloc(PATH_MAX);
6293 is_protocol = path_has_protocol(backing_file);
6296 * Being largely a legacy function, skip any filters here
6297 * (because filters do not have normal filenames, so they cannot
6298 * match anyway; and allowing json:{} filenames is a bit out of
6299 * scope).
6301 for (curr_bs = bdrv_skip_filters(bs);
6302 bdrv_cow_child(curr_bs) != NULL;
6303 curr_bs = bs_below)
6305 bs_below = bdrv_backing_chain_next(curr_bs);
6307 if (bdrv_backing_overridden(curr_bs)) {
6309 * If the backing file was overridden, we can only compare
6310 * directly against the backing node's filename.
6313 if (!filenames_refreshed) {
6315 * This will automatically refresh all of the
6316 * filenames in the rest of the backing chain, so we
6317 * only need to do this once.
6319 bdrv_refresh_filename(bs_below);
6320 filenames_refreshed = true;
6323 if (strcmp(backing_file, bs_below->filename) == 0) {
6324 retval = bs_below;
6325 break;
6327 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6329 * If either of the filename paths is actually a protocol, then
6330 * compare unmodified paths; otherwise make paths relative.
6332 char *backing_file_full_ret;
6334 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6335 retval = bs_below;
6336 break;
6338 /* Also check against the full backing filename for the image */
6339 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6340 NULL);
6341 if (backing_file_full_ret) {
6342 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6343 g_free(backing_file_full_ret);
6344 if (equal) {
6345 retval = bs_below;
6346 break;
6349 } else {
6350 /* If not an absolute filename path, make it relative to the current
6351 * image's filename path */
6352 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6353 NULL);
6354 /* We are going to compare canonicalized absolute pathnames */
6355 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6356 g_free(filename_tmp);
6357 continue;
6359 g_free(filename_tmp);
6361 /* We need to make sure the backing filename we are comparing against
6362 * is relative to the current image filename (or absolute) */
6363 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6364 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6365 g_free(filename_tmp);
6366 continue;
6368 g_free(filename_tmp);
6370 if (strcmp(backing_file_full, filename_full) == 0) {
6371 retval = bs_below;
6372 break;
6377 g_free(filename_full);
6378 g_free(backing_file_full);
6379 return retval;
6382 void bdrv_init(void)
6384 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6385 use_bdrv_whitelist = 1;
6386 #endif
6387 module_call_init(MODULE_INIT_BLOCK);
6390 void bdrv_init_with_whitelist(void)
6392 use_bdrv_whitelist = 1;
6393 bdrv_init();
6396 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6398 BdrvChild *child, *parent;
6399 Error *local_err = NULL;
6400 int ret;
6401 BdrvDirtyBitmap *bm;
6403 if (!bs->drv) {
6404 return -ENOMEDIUM;
6407 QLIST_FOREACH(child, &bs->children, next) {
6408 bdrv_co_invalidate_cache(child->bs, &local_err);
6409 if (local_err) {
6410 error_propagate(errp, local_err);
6411 return -EINVAL;
6416 * Update permissions, they may differ for inactive nodes.
6418 * Note that the required permissions of inactive images are always a
6419 * subset of the permissions required after activating the image. This
6420 * allows us to just get the permissions upfront without restricting
6421 * drv->bdrv_invalidate_cache().
6423 * It also means that in error cases, we don't have to try and revert to
6424 * the old permissions (which is an operation that could fail, too). We can
6425 * just keep the extended permissions for the next time that an activation
6426 * of the image is tried.
6428 if (bs->open_flags & BDRV_O_INACTIVE) {
6429 bs->open_flags &= ~BDRV_O_INACTIVE;
6430 ret = bdrv_refresh_perms(bs, errp);
6431 if (ret < 0) {
6432 bs->open_flags |= BDRV_O_INACTIVE;
6433 return ret;
6436 if (bs->drv->bdrv_co_invalidate_cache) {
6437 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6438 if (local_err) {
6439 bs->open_flags |= BDRV_O_INACTIVE;
6440 error_propagate(errp, local_err);
6441 return -EINVAL;
6445 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6446 bdrv_dirty_bitmap_skip_store(bm, false);
6449 ret = refresh_total_sectors(bs, bs->total_sectors);
6450 if (ret < 0) {
6451 bs->open_flags |= BDRV_O_INACTIVE;
6452 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6453 return ret;
6457 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6458 if (parent->klass->activate) {
6459 parent->klass->activate(parent, &local_err);
6460 if (local_err) {
6461 bs->open_flags |= BDRV_O_INACTIVE;
6462 error_propagate(errp, local_err);
6463 return -EINVAL;
6468 return 0;
6471 void bdrv_invalidate_cache_all(Error **errp)
6473 BlockDriverState *bs;
6474 BdrvNextIterator it;
6476 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6477 AioContext *aio_context = bdrv_get_aio_context(bs);
6478 int ret;
6480 aio_context_acquire(aio_context);
6481 ret = bdrv_invalidate_cache(bs, errp);
6482 aio_context_release(aio_context);
6483 if (ret < 0) {
6484 bdrv_next_cleanup(&it);
6485 return;
6490 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6492 BdrvChild *parent;
6494 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6495 if (parent->klass->parent_is_bds) {
6496 BlockDriverState *parent_bs = parent->opaque;
6497 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6498 return true;
6503 return false;
6506 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6508 BdrvChild *child, *parent;
6509 int ret;
6510 uint64_t cumulative_perms, cumulative_shared_perms;
6512 if (!bs->drv) {
6513 return -ENOMEDIUM;
6516 /* Make sure that we don't inactivate a child before its parent.
6517 * It will be covered by recursion from the yet active parent. */
6518 if (bdrv_has_bds_parent(bs, true)) {
6519 return 0;
6522 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6524 /* Inactivate this node */
6525 if (bs->drv->bdrv_inactivate) {
6526 ret = bs->drv->bdrv_inactivate(bs);
6527 if (ret < 0) {
6528 return ret;
6532 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6533 if (parent->klass->inactivate) {
6534 ret = parent->klass->inactivate(parent);
6535 if (ret < 0) {
6536 return ret;
6541 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6542 &cumulative_shared_perms);
6543 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6544 /* Our inactive parents still need write access. Inactivation failed. */
6545 return -EPERM;
6548 bs->open_flags |= BDRV_O_INACTIVE;
6551 * Update permissions, they may differ for inactive nodes.
6552 * We only tried to loosen restrictions, so errors are not fatal, ignore
6553 * them.
6555 bdrv_refresh_perms(bs, NULL);
6557 /* Recursively inactivate children */
6558 QLIST_FOREACH(child, &bs->children, next) {
6559 ret = bdrv_inactivate_recurse(child->bs);
6560 if (ret < 0) {
6561 return ret;
6565 return 0;
6568 int bdrv_inactivate_all(void)
6570 BlockDriverState *bs = NULL;
6571 BdrvNextIterator it;
6572 int ret = 0;
6573 GSList *aio_ctxs = NULL, *ctx;
6575 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6576 AioContext *aio_context = bdrv_get_aio_context(bs);
6578 if (!g_slist_find(aio_ctxs, aio_context)) {
6579 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6580 aio_context_acquire(aio_context);
6584 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6585 /* Nodes with BDS parents are covered by recursion from the last
6586 * parent that gets inactivated. Don't inactivate them a second
6587 * time if that has already happened. */
6588 if (bdrv_has_bds_parent(bs, false)) {
6589 continue;
6591 ret = bdrv_inactivate_recurse(bs);
6592 if (ret < 0) {
6593 bdrv_next_cleanup(&it);
6594 goto out;
6598 out:
6599 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6600 AioContext *aio_context = ctx->data;
6601 aio_context_release(aio_context);
6603 g_slist_free(aio_ctxs);
6605 return ret;
6608 /**************************************************************/
6609 /* removable device support */
6612 * Return TRUE if the media is present
6614 bool bdrv_is_inserted(BlockDriverState *bs)
6616 BlockDriver *drv = bs->drv;
6617 BdrvChild *child;
6619 if (!drv) {
6620 return false;
6622 if (drv->bdrv_is_inserted) {
6623 return drv->bdrv_is_inserted(bs);
6625 QLIST_FOREACH(child, &bs->children, next) {
6626 if (!bdrv_is_inserted(child->bs)) {
6627 return false;
6630 return true;
6634 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6636 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6638 BlockDriver *drv = bs->drv;
6640 if (drv && drv->bdrv_eject) {
6641 drv->bdrv_eject(bs, eject_flag);
6646 * Lock or unlock the media (if it is locked, the user won't be able
6647 * to eject it manually).
6649 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6651 BlockDriver *drv = bs->drv;
6653 trace_bdrv_lock_medium(bs, locked);
6655 if (drv && drv->bdrv_lock_medium) {
6656 drv->bdrv_lock_medium(bs, locked);
6660 /* Get a reference to bs */
6661 void bdrv_ref(BlockDriverState *bs)
6663 bs->refcnt++;
6666 /* Release a previously grabbed reference to bs.
6667 * If after releasing, reference count is zero, the BlockDriverState is
6668 * deleted. */
6669 void bdrv_unref(BlockDriverState *bs)
6671 if (!bs) {
6672 return;
6674 assert(bs->refcnt > 0);
6675 if (--bs->refcnt == 0) {
6676 bdrv_delete(bs);
6680 struct BdrvOpBlocker {
6681 Error *reason;
6682 QLIST_ENTRY(BdrvOpBlocker) list;
6685 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6687 BdrvOpBlocker *blocker;
6688 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6689 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6690 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6691 error_propagate_prepend(errp, error_copy(blocker->reason),
6692 "Node '%s' is busy: ",
6693 bdrv_get_device_or_node_name(bs));
6694 return true;
6696 return false;
6699 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6701 BdrvOpBlocker *blocker;
6702 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6704 blocker = g_new0(BdrvOpBlocker, 1);
6705 blocker->reason = reason;
6706 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6709 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6711 BdrvOpBlocker *blocker, *next;
6712 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6713 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6714 if (blocker->reason == reason) {
6715 QLIST_REMOVE(blocker, list);
6716 g_free(blocker);
6721 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6723 int i;
6724 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6725 bdrv_op_block(bs, i, reason);
6729 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6731 int i;
6732 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6733 bdrv_op_unblock(bs, i, reason);
6737 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6739 int i;
6741 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6742 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6743 return false;
6746 return true;
6749 void bdrv_img_create(const char *filename, const char *fmt,
6750 const char *base_filename, const char *base_fmt,
6751 char *options, uint64_t img_size, int flags, bool quiet,
6752 Error **errp)
6754 QemuOptsList *create_opts = NULL;
6755 QemuOpts *opts = NULL;
6756 const char *backing_fmt, *backing_file;
6757 int64_t size;
6758 BlockDriver *drv, *proto_drv;
6759 Error *local_err = NULL;
6760 int ret = 0;
6762 /* Find driver and parse its options */
6763 drv = bdrv_find_format(fmt);
6764 if (!drv) {
6765 error_setg(errp, "Unknown file format '%s'", fmt);
6766 return;
6769 proto_drv = bdrv_find_protocol(filename, true, errp);
6770 if (!proto_drv) {
6771 return;
6774 if (!drv->create_opts) {
6775 error_setg(errp, "Format driver '%s' does not support image creation",
6776 drv->format_name);
6777 return;
6780 if (!proto_drv->create_opts) {
6781 error_setg(errp, "Protocol driver '%s' does not support image creation",
6782 proto_drv->format_name);
6783 return;
6786 /* Create parameter list */
6787 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6788 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6790 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6792 /* Parse -o options */
6793 if (options) {
6794 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6795 goto out;
6799 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6800 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6801 } else if (img_size != UINT64_C(-1)) {
6802 error_setg(errp, "The image size must be specified only once");
6803 goto out;
6806 if (base_filename) {
6807 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6808 NULL)) {
6809 error_setg(errp, "Backing file not supported for file format '%s'",
6810 fmt);
6811 goto out;
6815 if (base_fmt) {
6816 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
6817 error_setg(errp, "Backing file format not supported for file "
6818 "format '%s'", fmt);
6819 goto out;
6823 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6824 if (backing_file) {
6825 if (!strcmp(filename, backing_file)) {
6826 error_setg(errp, "Error: Trying to create an image with the "
6827 "same filename as the backing file");
6828 goto out;
6830 if (backing_file[0] == '\0') {
6831 error_setg(errp, "Expected backing file name, got empty string");
6832 goto out;
6836 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
6838 /* The size for the image must always be specified, unless we have a backing
6839 * file and we have not been forbidden from opening it. */
6840 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6841 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6842 BlockDriverState *bs;
6843 char *full_backing;
6844 int back_flags;
6845 QDict *backing_options = NULL;
6847 full_backing =
6848 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6849 &local_err);
6850 if (local_err) {
6851 goto out;
6853 assert(full_backing);
6856 * No need to do I/O here, which allows us to open encrypted
6857 * backing images without needing the secret
6859 back_flags = flags;
6860 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
6861 back_flags |= BDRV_O_NO_IO;
6863 backing_options = qdict_new();
6864 if (backing_fmt) {
6865 qdict_put_str(backing_options, "driver", backing_fmt);
6867 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
6869 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6870 &local_err);
6871 g_free(full_backing);
6872 if (!bs) {
6873 error_append_hint(&local_err, "Could not open backing image.\n");
6874 goto out;
6875 } else {
6876 if (!backing_fmt) {
6877 error_setg(&local_err,
6878 "Backing file specified without backing format");
6879 error_append_hint(&local_err, "Detected format of %s.",
6880 bs->drv->format_name);
6881 goto out;
6883 if (size == -1) {
6884 /* Opened BS, have no size */
6885 size = bdrv_getlength(bs);
6886 if (size < 0) {
6887 error_setg_errno(errp, -size, "Could not get size of '%s'",
6888 backing_file);
6889 bdrv_unref(bs);
6890 goto out;
6892 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
6894 bdrv_unref(bs);
6896 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6897 } else if (backing_file && !backing_fmt) {
6898 error_setg(&local_err,
6899 "Backing file specified without backing format");
6900 goto out;
6903 if (size == -1) {
6904 error_setg(errp, "Image creation needs a size parameter");
6905 goto out;
6908 if (!quiet) {
6909 printf("Formatting '%s', fmt=%s ", filename, fmt);
6910 qemu_opts_print(opts, " ");
6911 puts("");
6912 fflush(stdout);
6915 ret = bdrv_create(drv, filename, opts, &local_err);
6917 if (ret == -EFBIG) {
6918 /* This is generally a better message than whatever the driver would
6919 * deliver (especially because of the cluster_size_hint), since that
6920 * is most probably not much different from "image too large". */
6921 const char *cluster_size_hint = "";
6922 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
6923 cluster_size_hint = " (try using a larger cluster size)";
6925 error_setg(errp, "The image size is too large for file format '%s'"
6926 "%s", fmt, cluster_size_hint);
6927 error_free(local_err);
6928 local_err = NULL;
6931 out:
6932 qemu_opts_del(opts);
6933 qemu_opts_free(create_opts);
6934 error_propagate(errp, local_err);
6937 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6939 return bs ? bs->aio_context : qemu_get_aio_context();
6942 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
6944 Coroutine *self = qemu_coroutine_self();
6945 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
6946 AioContext *new_ctx;
6949 * Increase bs->in_flight to ensure that this operation is completed before
6950 * moving the node to a different AioContext. Read new_ctx only afterwards.
6952 bdrv_inc_in_flight(bs);
6954 new_ctx = bdrv_get_aio_context(bs);
6955 aio_co_reschedule_self(new_ctx);
6956 return old_ctx;
6959 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
6961 aio_co_reschedule_self(old_ctx);
6962 bdrv_dec_in_flight(bs);
6965 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
6967 AioContext *ctx = bdrv_get_aio_context(bs);
6969 /* In the main thread, bs->aio_context won't change concurrently */
6970 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6973 * We're in coroutine context, so we already hold the lock of the main
6974 * loop AioContext. Don't lock it twice to avoid deadlocks.
6976 assert(qemu_in_coroutine());
6977 if (ctx != qemu_get_aio_context()) {
6978 aio_context_acquire(ctx);
6982 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
6984 AioContext *ctx = bdrv_get_aio_context(bs);
6986 assert(qemu_in_coroutine());
6987 if (ctx != qemu_get_aio_context()) {
6988 aio_context_release(ctx);
6992 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6994 aio_co_enter(bdrv_get_aio_context(bs), co);
6997 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6999 QLIST_REMOVE(ban, list);
7000 g_free(ban);
7003 static void bdrv_detach_aio_context(BlockDriverState *bs)
7005 BdrvAioNotifier *baf, *baf_tmp;
7007 assert(!bs->walking_aio_notifiers);
7008 bs->walking_aio_notifiers = true;
7009 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7010 if (baf->deleted) {
7011 bdrv_do_remove_aio_context_notifier(baf);
7012 } else {
7013 baf->detach_aio_context(baf->opaque);
7016 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7017 * remove remaining aio notifiers if we aren't called again.
7019 bs->walking_aio_notifiers = false;
7021 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7022 bs->drv->bdrv_detach_aio_context(bs);
7025 if (bs->quiesce_counter) {
7026 aio_enable_external(bs->aio_context);
7028 bs->aio_context = NULL;
7031 static void bdrv_attach_aio_context(BlockDriverState *bs,
7032 AioContext *new_context)
7034 BdrvAioNotifier *ban, *ban_tmp;
7036 if (bs->quiesce_counter) {
7037 aio_disable_external(new_context);
7040 bs->aio_context = new_context;
7042 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7043 bs->drv->bdrv_attach_aio_context(bs, new_context);
7046 assert(!bs->walking_aio_notifiers);
7047 bs->walking_aio_notifiers = true;
7048 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7049 if (ban->deleted) {
7050 bdrv_do_remove_aio_context_notifier(ban);
7051 } else {
7052 ban->attached_aio_context(new_context, ban->opaque);
7055 bs->walking_aio_notifiers = false;
7059 * Changes the AioContext used for fd handlers, timers, and BHs by this
7060 * BlockDriverState and all its children and parents.
7062 * Must be called from the main AioContext.
7064 * The caller must own the AioContext lock for the old AioContext of bs, but it
7065 * must not own the AioContext lock for new_context (unless new_context is the
7066 * same as the current context of bs).
7068 * @ignore will accumulate all visited BdrvChild object. The caller is
7069 * responsible for freeing the list afterwards.
7071 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
7072 AioContext *new_context, GSList **ignore)
7074 AioContext *old_context = bdrv_get_aio_context(bs);
7075 GSList *children_to_process = NULL;
7076 GSList *parents_to_process = NULL;
7077 GSList *entry;
7078 BdrvChild *child, *parent;
7080 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7082 if (old_context == new_context) {
7083 return;
7086 bdrv_drained_begin(bs);
7088 QLIST_FOREACH(child, &bs->children, next) {
7089 if (g_slist_find(*ignore, child)) {
7090 continue;
7092 *ignore = g_slist_prepend(*ignore, child);
7093 children_to_process = g_slist_prepend(children_to_process, child);
7096 QLIST_FOREACH(parent, &bs->parents, next_parent) {
7097 if (g_slist_find(*ignore, parent)) {
7098 continue;
7100 *ignore = g_slist_prepend(*ignore, parent);
7101 parents_to_process = g_slist_prepend(parents_to_process, parent);
7104 for (entry = children_to_process;
7105 entry != NULL;
7106 entry = g_slist_next(entry)) {
7107 child = entry->data;
7108 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
7110 g_slist_free(children_to_process);
7112 for (entry = parents_to_process;
7113 entry != NULL;
7114 entry = g_slist_next(entry)) {
7115 parent = entry->data;
7116 assert(parent->klass->set_aio_ctx);
7117 parent->klass->set_aio_ctx(parent, new_context, ignore);
7119 g_slist_free(parents_to_process);
7121 bdrv_detach_aio_context(bs);
7123 /* Acquire the new context, if necessary */
7124 if (qemu_get_aio_context() != new_context) {
7125 aio_context_acquire(new_context);
7128 bdrv_attach_aio_context(bs, new_context);
7131 * If this function was recursively called from
7132 * bdrv_set_aio_context_ignore(), there may be nodes in the
7133 * subtree that have not yet been moved to the new AioContext.
7134 * Release the old one so bdrv_drained_end() can poll them.
7136 if (qemu_get_aio_context() != old_context) {
7137 aio_context_release(old_context);
7140 bdrv_drained_end(bs);
7142 if (qemu_get_aio_context() != old_context) {
7143 aio_context_acquire(old_context);
7145 if (qemu_get_aio_context() != new_context) {
7146 aio_context_release(new_context);
7150 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7151 GSList **ignore, Error **errp)
7153 if (g_slist_find(*ignore, c)) {
7154 return true;
7156 *ignore = g_slist_prepend(*ignore, c);
7159 * A BdrvChildClass that doesn't handle AioContext changes cannot
7160 * tolerate any AioContext changes
7162 if (!c->klass->can_set_aio_ctx) {
7163 char *user = bdrv_child_user_desc(c);
7164 error_setg(errp, "Changing iothreads is not supported by %s", user);
7165 g_free(user);
7166 return false;
7168 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
7169 assert(!errp || *errp);
7170 return false;
7172 return true;
7175 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7176 GSList **ignore, Error **errp)
7178 if (g_slist_find(*ignore, c)) {
7179 return true;
7181 *ignore = g_slist_prepend(*ignore, c);
7182 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
7185 /* @ignore will accumulate all visited BdrvChild object. The caller is
7186 * responsible for freeing the list afterwards. */
7187 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7188 GSList **ignore, Error **errp)
7190 BdrvChild *c;
7192 if (bdrv_get_aio_context(bs) == ctx) {
7193 return true;
7196 QLIST_FOREACH(c, &bs->parents, next_parent) {
7197 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
7198 return false;
7201 QLIST_FOREACH(c, &bs->children, next) {
7202 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
7203 return false;
7207 return true;
7210 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7211 BdrvChild *ignore_child, Error **errp)
7213 GSList *ignore;
7214 bool ret;
7216 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7217 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
7218 g_slist_free(ignore);
7220 if (!ret) {
7221 return -EPERM;
7224 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7225 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
7226 g_slist_free(ignore);
7228 return 0;
7231 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7232 Error **errp)
7234 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
7237 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7238 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7239 void (*detach_aio_context)(void *opaque), void *opaque)
7241 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7242 *ban = (BdrvAioNotifier){
7243 .attached_aio_context = attached_aio_context,
7244 .detach_aio_context = detach_aio_context,
7245 .opaque = opaque
7248 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7251 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7252 void (*attached_aio_context)(AioContext *,
7253 void *),
7254 void (*detach_aio_context)(void *),
7255 void *opaque)
7257 BdrvAioNotifier *ban, *ban_next;
7259 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7260 if (ban->attached_aio_context == attached_aio_context &&
7261 ban->detach_aio_context == detach_aio_context &&
7262 ban->opaque == opaque &&
7263 ban->deleted == false)
7265 if (bs->walking_aio_notifiers) {
7266 ban->deleted = true;
7267 } else {
7268 bdrv_do_remove_aio_context_notifier(ban);
7270 return;
7274 abort();
7277 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7278 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7279 bool force,
7280 Error **errp)
7282 if (!bs->drv) {
7283 error_setg(errp, "Node is ejected");
7284 return -ENOMEDIUM;
7286 if (!bs->drv->bdrv_amend_options) {
7287 error_setg(errp, "Block driver '%s' does not support option amendment",
7288 bs->drv->format_name);
7289 return -ENOTSUP;
7291 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7292 cb_opaque, force, errp);
7296 * This function checks whether the given @to_replace is allowed to be
7297 * replaced by a node that always shows the same data as @bs. This is
7298 * used for example to verify whether the mirror job can replace
7299 * @to_replace by the target mirrored from @bs.
7300 * To be replaceable, @bs and @to_replace may either be guaranteed to
7301 * always show the same data (because they are only connected through
7302 * filters), or some driver may allow replacing one of its children
7303 * because it can guarantee that this child's data is not visible at
7304 * all (for example, for dissenting quorum children that have no other
7305 * parents).
7307 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7308 BlockDriverState *to_replace)
7310 BlockDriverState *filtered;
7312 if (!bs || !bs->drv) {
7313 return false;
7316 if (bs == to_replace) {
7317 return true;
7320 /* See what the driver can do */
7321 if (bs->drv->bdrv_recurse_can_replace) {
7322 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7325 /* For filters without an own implementation, we can recurse on our own */
7326 filtered = bdrv_filter_bs(bs);
7327 if (filtered) {
7328 return bdrv_recurse_can_replace(filtered, to_replace);
7331 /* Safe default */
7332 return false;
7336 * Check whether the given @node_name can be replaced by a node that
7337 * has the same data as @parent_bs. If so, return @node_name's BDS;
7338 * NULL otherwise.
7340 * @node_name must be a (recursive) *child of @parent_bs (or this
7341 * function will return NULL).
7343 * The result (whether the node can be replaced or not) is only valid
7344 * for as long as no graph or permission changes occur.
7346 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7347 const char *node_name, Error **errp)
7349 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7350 AioContext *aio_context;
7352 if (!to_replace_bs) {
7353 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7354 return NULL;
7357 aio_context = bdrv_get_aio_context(to_replace_bs);
7358 aio_context_acquire(aio_context);
7360 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7361 to_replace_bs = NULL;
7362 goto out;
7365 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7366 * most non filter in order to prevent data corruption.
7367 * Another benefit is that this tests exclude backing files which are
7368 * blocked by the backing blockers.
7370 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7371 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7372 "because it cannot be guaranteed that doing so would not "
7373 "lead to an abrupt change of visible data",
7374 node_name, parent_bs->node_name);
7375 to_replace_bs = NULL;
7376 goto out;
7379 out:
7380 aio_context_release(aio_context);
7381 return to_replace_bs;
7385 * Iterates through the list of runtime option keys that are said to
7386 * be "strong" for a BDS. An option is called "strong" if it changes
7387 * a BDS's data. For example, the null block driver's "size" and
7388 * "read-zeroes" options are strong, but its "latency-ns" option is
7389 * not.
7391 * If a key returned by this function ends with a dot, all options
7392 * starting with that prefix are strong.
7394 static const char *const *strong_options(BlockDriverState *bs,
7395 const char *const *curopt)
7397 static const char *const global_options[] = {
7398 "driver", "filename", NULL
7401 if (!curopt) {
7402 return &global_options[0];
7405 curopt++;
7406 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7407 curopt = bs->drv->strong_runtime_opts;
7410 return (curopt && *curopt) ? curopt : NULL;
7414 * Copies all strong runtime options from bs->options to the given
7415 * QDict. The set of strong option keys is determined by invoking
7416 * strong_options().
7418 * Returns true iff any strong option was present in bs->options (and
7419 * thus copied to the target QDict) with the exception of "filename"
7420 * and "driver". The caller is expected to use this value to decide
7421 * whether the existence of strong options prevents the generation of
7422 * a plain filename.
7424 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7426 bool found_any = false;
7427 const char *const *option_name = NULL;
7429 if (!bs->drv) {
7430 return false;
7433 while ((option_name = strong_options(bs, option_name))) {
7434 bool option_given = false;
7436 assert(strlen(*option_name) > 0);
7437 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7438 QObject *entry = qdict_get(bs->options, *option_name);
7439 if (!entry) {
7440 continue;
7443 qdict_put_obj(d, *option_name, qobject_ref(entry));
7444 option_given = true;
7445 } else {
7446 const QDictEntry *entry;
7447 for (entry = qdict_first(bs->options); entry;
7448 entry = qdict_next(bs->options, entry))
7450 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7451 qdict_put_obj(d, qdict_entry_key(entry),
7452 qobject_ref(qdict_entry_value(entry)));
7453 option_given = true;
7458 /* While "driver" and "filename" need to be included in a JSON filename,
7459 * their existence does not prohibit generation of a plain filename. */
7460 if (!found_any && option_given &&
7461 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7463 found_any = true;
7467 if (!qdict_haskey(d, "driver")) {
7468 /* Drivers created with bdrv_new_open_driver() may not have a
7469 * @driver option. Add it here. */
7470 qdict_put_str(d, "driver", bs->drv->format_name);
7473 return found_any;
7476 /* Note: This function may return false positives; it may return true
7477 * even if opening the backing file specified by bs's image header
7478 * would result in exactly bs->backing. */
7479 static bool bdrv_backing_overridden(BlockDriverState *bs)
7481 if (bs->backing) {
7482 return strcmp(bs->auto_backing_file,
7483 bs->backing->bs->filename);
7484 } else {
7485 /* No backing BDS, so if the image header reports any backing
7486 * file, it must have been suppressed */
7487 return bs->auto_backing_file[0] != '\0';
7491 /* Updates the following BDS fields:
7492 * - exact_filename: A filename which may be used for opening a block device
7493 * which (mostly) equals the given BDS (even without any
7494 * other options; so reading and writing must return the same
7495 * results, but caching etc. may be different)
7496 * - full_open_options: Options which, when given when opening a block device
7497 * (without a filename), result in a BDS (mostly)
7498 * equalling the given one
7499 * - filename: If exact_filename is set, it is copied here. Otherwise,
7500 * full_open_options is converted to a JSON object, prefixed with
7501 * "json:" (for use through the JSON pseudo protocol) and put here.
7503 void bdrv_refresh_filename(BlockDriverState *bs)
7505 BlockDriver *drv = bs->drv;
7506 BdrvChild *child;
7507 BlockDriverState *primary_child_bs;
7508 QDict *opts;
7509 bool backing_overridden;
7510 bool generate_json_filename; /* Whether our default implementation should
7511 fill exact_filename (false) or not (true) */
7513 if (!drv) {
7514 return;
7517 /* This BDS's file name may depend on any of its children's file names, so
7518 * refresh those first */
7519 QLIST_FOREACH(child, &bs->children, next) {
7520 bdrv_refresh_filename(child->bs);
7523 if (bs->implicit) {
7524 /* For implicit nodes, just copy everything from the single child */
7525 child = QLIST_FIRST(&bs->children);
7526 assert(QLIST_NEXT(child, next) == NULL);
7528 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7529 child->bs->exact_filename);
7530 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7532 qobject_unref(bs->full_open_options);
7533 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7535 return;
7538 backing_overridden = bdrv_backing_overridden(bs);
7540 if (bs->open_flags & BDRV_O_NO_IO) {
7541 /* Without I/O, the backing file does not change anything.
7542 * Therefore, in such a case (primarily qemu-img), we can
7543 * pretend the backing file has not been overridden even if
7544 * it technically has been. */
7545 backing_overridden = false;
7548 /* Gather the options QDict */
7549 opts = qdict_new();
7550 generate_json_filename = append_strong_runtime_options(opts, bs);
7551 generate_json_filename |= backing_overridden;
7553 if (drv->bdrv_gather_child_options) {
7554 /* Some block drivers may not want to present all of their children's
7555 * options, or name them differently from BdrvChild.name */
7556 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7557 } else {
7558 QLIST_FOREACH(child, &bs->children, next) {
7559 if (child == bs->backing && !backing_overridden) {
7560 /* We can skip the backing BDS if it has not been overridden */
7561 continue;
7564 qdict_put(opts, child->name,
7565 qobject_ref(child->bs->full_open_options));
7568 if (backing_overridden && !bs->backing) {
7569 /* Force no backing file */
7570 qdict_put_null(opts, "backing");
7574 qobject_unref(bs->full_open_options);
7575 bs->full_open_options = opts;
7577 primary_child_bs = bdrv_primary_bs(bs);
7579 if (drv->bdrv_refresh_filename) {
7580 /* Obsolete information is of no use here, so drop the old file name
7581 * information before refreshing it */
7582 bs->exact_filename[0] = '\0';
7584 drv->bdrv_refresh_filename(bs);
7585 } else if (primary_child_bs) {
7587 * Try to reconstruct valid information from the underlying
7588 * file -- this only works for format nodes (filter nodes
7589 * cannot be probed and as such must be selected by the user
7590 * either through an options dict, or through a special
7591 * filename which the filter driver must construct in its
7592 * .bdrv_refresh_filename() implementation).
7595 bs->exact_filename[0] = '\0';
7598 * We can use the underlying file's filename if:
7599 * - it has a filename,
7600 * - the current BDS is not a filter,
7601 * - the file is a protocol BDS, and
7602 * - opening that file (as this BDS's format) will automatically create
7603 * the BDS tree we have right now, that is:
7604 * - the user did not significantly change this BDS's behavior with
7605 * some explicit (strong) options
7606 * - no non-file child of this BDS has been overridden by the user
7607 * Both of these conditions are represented by generate_json_filename.
7609 if (primary_child_bs->exact_filename[0] &&
7610 primary_child_bs->drv->bdrv_file_open &&
7611 !drv->is_filter && !generate_json_filename)
7613 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7617 if (bs->exact_filename[0]) {
7618 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7619 } else {
7620 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7621 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7622 json->str) >= sizeof(bs->filename)) {
7623 /* Give user a hint if we truncated things. */
7624 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7626 g_string_free(json, true);
7630 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7632 BlockDriver *drv = bs->drv;
7633 BlockDriverState *child_bs;
7635 if (!drv) {
7636 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7637 return NULL;
7640 if (drv->bdrv_dirname) {
7641 return drv->bdrv_dirname(bs, errp);
7644 child_bs = bdrv_primary_bs(bs);
7645 if (child_bs) {
7646 return bdrv_dirname(child_bs, errp);
7649 bdrv_refresh_filename(bs);
7650 if (bs->exact_filename[0] != '\0') {
7651 return path_combine(bs->exact_filename, "");
7654 error_setg(errp, "Cannot generate a base directory for %s nodes",
7655 drv->format_name);
7656 return NULL;
7660 * Hot add/remove a BDS's child. So the user can take a child offline when
7661 * it is broken and take a new child online
7663 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7664 Error **errp)
7667 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7668 error_setg(errp, "The node %s does not support adding a child",
7669 bdrv_get_device_or_node_name(parent_bs));
7670 return;
7673 if (!QLIST_EMPTY(&child_bs->parents)) {
7674 error_setg(errp, "The node %s already has a parent",
7675 child_bs->node_name);
7676 return;
7679 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7682 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7684 BdrvChild *tmp;
7686 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7687 error_setg(errp, "The node %s does not support removing a child",
7688 bdrv_get_device_or_node_name(parent_bs));
7689 return;
7692 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7693 if (tmp == child) {
7694 break;
7698 if (!tmp) {
7699 error_setg(errp, "The node %s does not have a child named %s",
7700 bdrv_get_device_or_node_name(parent_bs),
7701 bdrv_get_device_or_node_name(child->bs));
7702 return;
7705 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7708 int bdrv_make_empty(BdrvChild *c, Error **errp)
7710 BlockDriver *drv = c->bs->drv;
7711 int ret;
7713 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7715 if (!drv->bdrv_make_empty) {
7716 error_setg(errp, "%s does not support emptying nodes",
7717 drv->format_name);
7718 return -ENOTSUP;
7721 ret = drv->bdrv_make_empty(c->bs);
7722 if (ret < 0) {
7723 error_setg_errno(errp, -ret, "Failed to empty %s",
7724 c->bs->filename);
7725 return ret;
7728 return 0;
7732 * Return the child that @bs acts as an overlay for, and from which data may be
7733 * copied in COW or COR operations. Usually this is the backing file.
7735 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7737 if (!bs || !bs->drv) {
7738 return NULL;
7741 if (bs->drv->is_filter) {
7742 return NULL;
7745 if (!bs->backing) {
7746 return NULL;
7749 assert(bs->backing->role & BDRV_CHILD_COW);
7750 return bs->backing;
7754 * If @bs acts as a filter for exactly one of its children, return
7755 * that child.
7757 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7759 BdrvChild *c;
7761 if (!bs || !bs->drv) {
7762 return NULL;
7765 if (!bs->drv->is_filter) {
7766 return NULL;
7769 /* Only one of @backing or @file may be used */
7770 assert(!(bs->backing && bs->file));
7772 c = bs->backing ?: bs->file;
7773 if (!c) {
7774 return NULL;
7777 assert(c->role & BDRV_CHILD_FILTERED);
7778 return c;
7782 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7783 * whichever is non-NULL.
7785 * Return NULL if both are NULL.
7787 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7789 BdrvChild *cow_child = bdrv_cow_child(bs);
7790 BdrvChild *filter_child = bdrv_filter_child(bs);
7792 /* Filter nodes cannot have COW backing files */
7793 assert(!(cow_child && filter_child));
7795 return cow_child ?: filter_child;
7799 * Return the primary child of this node: For filters, that is the
7800 * filtered child. For other nodes, that is usually the child storing
7801 * metadata.
7802 * (A generally more helpful description is that this is (usually) the
7803 * child that has the same filename as @bs.)
7805 * Drivers do not necessarily have a primary child; for example quorum
7806 * does not.
7808 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
7810 BdrvChild *c, *found = NULL;
7812 QLIST_FOREACH(c, &bs->children, next) {
7813 if (c->role & BDRV_CHILD_PRIMARY) {
7814 assert(!found);
7815 found = c;
7819 return found;
7822 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
7823 bool stop_on_explicit_filter)
7825 BdrvChild *c;
7827 if (!bs) {
7828 return NULL;
7831 while (!(stop_on_explicit_filter && !bs->implicit)) {
7832 c = bdrv_filter_child(bs);
7833 if (!c) {
7835 * A filter that is embedded in a working block graph must
7836 * have a child. Assert this here so this function does
7837 * not return a filter node that is not expected by the
7838 * caller.
7840 assert(!bs->drv || !bs->drv->is_filter);
7841 break;
7843 bs = c->bs;
7846 * Note that this treats nodes with bs->drv == NULL as not being
7847 * filters (bs->drv == NULL should be replaced by something else
7848 * anyway).
7849 * The advantage of this behavior is that this function will thus
7850 * always return a non-NULL value (given a non-NULL @bs).
7853 return bs;
7857 * Return the first BDS that has not been added implicitly or that
7858 * does not have a filtered child down the chain starting from @bs
7859 * (including @bs itself).
7861 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
7863 return bdrv_do_skip_filters(bs, true);
7867 * Return the first BDS that does not have a filtered child down the
7868 * chain starting from @bs (including @bs itself).
7870 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
7872 return bdrv_do_skip_filters(bs, false);
7876 * For a backing chain, return the first non-filter backing image of
7877 * the first non-filter image.
7879 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
7881 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
7885 * Check whether [offset, offset + bytes) overlaps with the cached
7886 * block-status data region.
7888 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
7889 * which is what bdrv_bsc_is_data()'s interface needs.
7890 * Otherwise, *pnum is not touched.
7892 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
7893 int64_t offset, int64_t bytes,
7894 int64_t *pnum)
7896 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
7897 bool overlaps;
7899 overlaps =
7900 qatomic_read(&bsc->valid) &&
7901 ranges_overlap(offset, bytes, bsc->data_start,
7902 bsc->data_end - bsc->data_start);
7904 if (overlaps && pnum) {
7905 *pnum = bsc->data_end - offset;
7908 return overlaps;
7912 * See block_int.h for this function's documentation.
7914 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
7916 RCU_READ_LOCK_GUARD();
7918 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
7922 * See block_int.h for this function's documentation.
7924 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
7925 int64_t offset, int64_t bytes)
7927 RCU_READ_LOCK_GUARD();
7929 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
7930 qatomic_set(&bs->block_status_cache->valid, false);
7935 * See block_int.h for this function's documentation.
7937 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
7939 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
7940 BdrvBlockStatusCache *old_bsc;
7942 *new_bsc = (BdrvBlockStatusCache) {
7943 .valid = true,
7944 .data_start = offset,
7945 .data_end = offset + bytes,
7948 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
7950 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
7951 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
7952 if (old_bsc) {
7953 g_free_rcu(old_bsc, rcu);