KVM: remove support for kernel-irqchip=off
[qemu.git] / block.c
blob9c2ac757e49586140372b41bf47de6616c99a89e
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 /* Protected by BQL */
71 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
72 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
74 /* Protected by BQL */
75 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
76 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
78 /* Protected by BQL */
79 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
80 QLIST_HEAD_INITIALIZER(bdrv_drivers);
82 static BlockDriverState *bdrv_open_inherit(const char *filename,
83 const char *reference,
84 QDict *options, int flags,
85 BlockDriverState *parent,
86 const BdrvChildClass *child_class,
87 BdrvChildRole child_role,
88 Error **errp);
90 static bool bdrv_recurse_has_child(BlockDriverState *bs,
91 BlockDriverState *child);
93 static void bdrv_replace_child_noperm(BdrvChild *child,
94 BlockDriverState *new_bs);
95 static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
97 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
98 BlockReopenQueue *queue,
99 Transaction *change_child_tran, Error **errp);
100 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
101 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
103 static bool bdrv_backing_overridden(BlockDriverState *bs);
105 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
106 GHashTable *visited, Transaction *tran,
107 Error **errp);
109 /* If non-zero, use only whitelisted block drivers */
110 static int use_bdrv_whitelist;
112 #ifdef _WIN32
113 static int is_windows_drive_prefix(const char *filename)
115 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
116 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
117 filename[1] == ':');
120 int is_windows_drive(const char *filename)
122 if (is_windows_drive_prefix(filename) &&
123 filename[2] == '\0')
124 return 1;
125 if (strstart(filename, "\\\\.\\", NULL) ||
126 strstart(filename, "//./", NULL))
127 return 1;
128 return 0;
130 #endif
132 size_t bdrv_opt_mem_align(BlockDriverState *bs)
134 if (!bs || !bs->drv) {
135 /* page size or 4k (hdd sector size) should be on the safe side */
136 return MAX(4096, qemu_real_host_page_size());
138 IO_CODE();
140 return bs->bl.opt_mem_alignment;
143 size_t bdrv_min_mem_align(BlockDriverState *bs)
145 if (!bs || !bs->drv) {
146 /* page size or 4k (hdd sector size) should be on the safe side */
147 return MAX(4096, qemu_real_host_page_size());
149 IO_CODE();
151 return bs->bl.min_mem_alignment;
154 /* check if the path starts with "<protocol>:" */
155 int path_has_protocol(const char *path)
157 const char *p;
159 #ifdef _WIN32
160 if (is_windows_drive(path) ||
161 is_windows_drive_prefix(path)) {
162 return 0;
164 p = path + strcspn(path, ":/\\");
165 #else
166 p = path + strcspn(path, ":/");
167 #endif
169 return *p == ':';
172 int path_is_absolute(const char *path)
174 #ifdef _WIN32
175 /* specific case for names like: "\\.\d:" */
176 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
177 return 1;
179 return (*path == '/' || *path == '\\');
180 #else
181 return (*path == '/');
182 #endif
185 /* if filename is absolute, just return its duplicate. Otherwise, build a
186 path to it by considering it is relative to base_path. URL are
187 supported. */
188 char *path_combine(const char *base_path, const char *filename)
190 const char *protocol_stripped = NULL;
191 const char *p, *p1;
192 char *result;
193 int len;
195 if (path_is_absolute(filename)) {
196 return g_strdup(filename);
199 if (path_has_protocol(base_path)) {
200 protocol_stripped = strchr(base_path, ':');
201 if (protocol_stripped) {
202 protocol_stripped++;
205 p = protocol_stripped ?: base_path;
207 p1 = strrchr(base_path, '/');
208 #ifdef _WIN32
210 const char *p2;
211 p2 = strrchr(base_path, '\\');
212 if (!p1 || p2 > p1) {
213 p1 = p2;
216 #endif
217 if (p1) {
218 p1++;
219 } else {
220 p1 = base_path;
222 if (p1 > p) {
223 p = p1;
225 len = p - base_path;
227 result = g_malloc(len + strlen(filename) + 1);
228 memcpy(result, base_path, len);
229 strcpy(result + len, filename);
231 return result;
235 * Helper function for bdrv_parse_filename() implementations to remove optional
236 * protocol prefixes (especially "file:") from a filename and for putting the
237 * stripped filename into the options QDict if there is such a prefix.
239 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
240 QDict *options)
242 if (strstart(filename, prefix, &filename)) {
243 /* Stripping the explicit protocol prefix may result in a protocol
244 * prefix being (wrongly) detected (if the filename contains a colon) */
245 if (path_has_protocol(filename)) {
246 GString *fat_filename;
248 /* This means there is some colon before the first slash; therefore,
249 * this cannot be an absolute path */
250 assert(!path_is_absolute(filename));
252 /* And we can thus fix the protocol detection issue by prefixing it
253 * by "./" */
254 fat_filename = g_string_new("./");
255 g_string_append(fat_filename, filename);
257 assert(!path_has_protocol(fat_filename->str));
259 qdict_put(options, "filename",
260 qstring_from_gstring(fat_filename));
261 } else {
262 /* If no protocol prefix was detected, we can use the shortened
263 * filename as-is */
264 qdict_put_str(options, "filename", filename);
270 /* Returns whether the image file is opened as read-only. Note that this can
271 * return false and writing to the image file is still not possible because the
272 * image is inactivated. */
273 bool bdrv_is_read_only(BlockDriverState *bs)
275 IO_CODE();
276 return !(bs->open_flags & BDRV_O_RDWR);
279 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
280 bool ignore_allow_rdw, Error **errp)
282 IO_CODE();
284 /* Do not set read_only if copy_on_read is enabled */
285 if (bs->copy_on_read && read_only) {
286 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
287 bdrv_get_device_or_node_name(bs));
288 return -EINVAL;
291 /* Do not clear read_only if it is prohibited */
292 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
293 !ignore_allow_rdw)
295 error_setg(errp, "Node '%s' is read only",
296 bdrv_get_device_or_node_name(bs));
297 return -EPERM;
300 return 0;
304 * Called by a driver that can only provide a read-only image.
306 * Returns 0 if the node is already read-only or it could switch the node to
307 * read-only because BDRV_O_AUTO_RDONLY is set.
309 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
310 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
311 * is not NULL, it is used as the error message for the Error object.
313 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
314 Error **errp)
316 int ret = 0;
317 IO_CODE();
319 if (!(bs->open_flags & BDRV_O_RDWR)) {
320 return 0;
322 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
323 goto fail;
326 ret = bdrv_can_set_read_only(bs, true, false, NULL);
327 if (ret < 0) {
328 goto fail;
331 bs->open_flags &= ~BDRV_O_RDWR;
333 return 0;
335 fail:
336 error_setg(errp, "%s", errmsg ?: "Image is read-only");
337 return -EACCES;
341 * If @backing is empty, this function returns NULL without setting
342 * @errp. In all other cases, NULL will only be returned with @errp
343 * set.
345 * Therefore, a return value of NULL without @errp set means that
346 * there is no backing file; if @errp is set, there is one but its
347 * absolute filename cannot be generated.
349 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
350 const char *backing,
351 Error **errp)
353 if (backing[0] == '\0') {
354 return NULL;
355 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
356 return g_strdup(backing);
357 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
358 error_setg(errp, "Cannot use relative backing file names for '%s'",
359 backed);
360 return NULL;
361 } else {
362 return path_combine(backed, backing);
367 * If @filename is empty or NULL, this function returns NULL without
368 * setting @errp. In all other cases, NULL will only be returned with
369 * @errp set.
371 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
372 const char *filename, Error **errp)
374 char *dir, *full_name;
376 if (!filename || filename[0] == '\0') {
377 return NULL;
378 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
379 return g_strdup(filename);
382 dir = bdrv_dirname(relative_to, errp);
383 if (!dir) {
384 return NULL;
387 full_name = g_strconcat(dir, filename, NULL);
388 g_free(dir);
389 return full_name;
392 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
394 GLOBAL_STATE_CODE();
395 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
398 void bdrv_register(BlockDriver *bdrv)
400 assert(bdrv->format_name);
401 GLOBAL_STATE_CODE();
402 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
405 BlockDriverState *bdrv_new(void)
407 BlockDriverState *bs;
408 int i;
410 GLOBAL_STATE_CODE();
412 bs = g_new0(BlockDriverState, 1);
413 QLIST_INIT(&bs->dirty_bitmaps);
414 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
415 QLIST_INIT(&bs->op_blockers[i]);
417 qemu_co_mutex_init(&bs->reqs_lock);
418 qemu_mutex_init(&bs->dirty_bitmap_mutex);
419 bs->refcnt = 1;
420 bs->aio_context = qemu_get_aio_context();
422 qemu_co_queue_init(&bs->flush_queue);
424 qemu_co_mutex_init(&bs->bsc_modify_lock);
425 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
427 for (i = 0; i < bdrv_drain_all_count; i++) {
428 bdrv_drained_begin(bs);
431 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
433 return bs;
436 static BlockDriver *bdrv_do_find_format(const char *format_name)
438 BlockDriver *drv1;
439 GLOBAL_STATE_CODE();
441 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
442 if (!strcmp(drv1->format_name, format_name)) {
443 return drv1;
447 return NULL;
450 BlockDriver *bdrv_find_format(const char *format_name)
452 BlockDriver *drv1;
453 int i;
455 GLOBAL_STATE_CODE();
457 drv1 = bdrv_do_find_format(format_name);
458 if (drv1) {
459 return drv1;
462 /* The driver isn't registered, maybe we need to load a module */
463 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
464 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
465 Error *local_err = NULL;
466 int rv = block_module_load(block_driver_modules[i].library_name,
467 &local_err);
468 if (rv > 0) {
469 return bdrv_do_find_format(format_name);
470 } else if (rv < 0) {
471 error_report_err(local_err);
473 break;
476 return NULL;
479 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
481 static const char *whitelist_rw[] = {
482 CONFIG_BDRV_RW_WHITELIST
483 NULL
485 static const char *whitelist_ro[] = {
486 CONFIG_BDRV_RO_WHITELIST
487 NULL
489 const char **p;
491 if (!whitelist_rw[0] && !whitelist_ro[0]) {
492 return 1; /* no whitelist, anything goes */
495 for (p = whitelist_rw; *p; p++) {
496 if (!strcmp(format_name, *p)) {
497 return 1;
500 if (read_only) {
501 for (p = whitelist_ro; *p; p++) {
502 if (!strcmp(format_name, *p)) {
503 return 1;
507 return 0;
510 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
512 GLOBAL_STATE_CODE();
513 return bdrv_format_is_whitelisted(drv->format_name, read_only);
516 bool bdrv_uses_whitelist(void)
518 return use_bdrv_whitelist;
521 typedef struct CreateCo {
522 BlockDriver *drv;
523 char *filename;
524 QemuOpts *opts;
525 int ret;
526 Error *err;
527 } CreateCo;
529 int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
530 QemuOpts *opts, Error **errp)
532 int ret;
533 GLOBAL_STATE_CODE();
534 ERRP_GUARD();
536 if (!drv->bdrv_co_create_opts) {
537 error_setg(errp, "Driver '%s' does not support image creation",
538 drv->format_name);
539 return -ENOTSUP;
542 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
543 if (ret < 0 && !*errp) {
544 error_setg_errno(errp, -ret, "Could not create image");
547 return ret;
551 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
552 * least the given @minimum_size.
554 * On success, return @blk's actual length.
555 * Otherwise, return -errno.
557 static int64_t create_file_fallback_truncate(BlockBackend *blk,
558 int64_t minimum_size, Error **errp)
560 Error *local_err = NULL;
561 int64_t size;
562 int ret;
564 GLOBAL_STATE_CODE();
566 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
567 &local_err);
568 if (ret < 0 && ret != -ENOTSUP) {
569 error_propagate(errp, local_err);
570 return ret;
573 size = blk_getlength(blk);
574 if (size < 0) {
575 error_free(local_err);
576 error_setg_errno(errp, -size,
577 "Failed to inquire the new image file's length");
578 return size;
581 if (size < minimum_size) {
582 /* Need to grow the image, but we failed to do that */
583 error_propagate(errp, local_err);
584 return -ENOTSUP;
587 error_free(local_err);
588 local_err = NULL;
590 return size;
594 * Helper function for bdrv_create_file_fallback(): Zero the first
595 * sector to remove any potentially pre-existing image header.
597 static int coroutine_fn
598 create_file_fallback_zero_first_sector(BlockBackend *blk,
599 int64_t current_size,
600 Error **errp)
602 int64_t bytes_to_clear;
603 int ret;
605 GLOBAL_STATE_CODE();
607 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
608 if (bytes_to_clear) {
609 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
610 if (ret < 0) {
611 error_setg_errno(errp, -ret,
612 "Failed to clear the new image's first sector");
613 return ret;
617 return 0;
621 * Simple implementation of bdrv_co_create_opts for protocol drivers
622 * which only support creation via opening a file
623 * (usually existing raw storage device)
625 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
626 const char *filename,
627 QemuOpts *opts,
628 Error **errp)
630 BlockBackend *blk;
631 QDict *options;
632 int64_t size = 0;
633 char *buf = NULL;
634 PreallocMode prealloc;
635 Error *local_err = NULL;
636 int ret;
638 GLOBAL_STATE_CODE();
640 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
641 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
642 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
643 PREALLOC_MODE_OFF, &local_err);
644 g_free(buf);
645 if (local_err) {
646 error_propagate(errp, local_err);
647 return -EINVAL;
650 if (prealloc != PREALLOC_MODE_OFF) {
651 error_setg(errp, "Unsupported preallocation mode '%s'",
652 PreallocMode_str(prealloc));
653 return -ENOTSUP;
656 options = qdict_new();
657 qdict_put_str(options, "driver", drv->format_name);
659 blk = blk_new_open(filename, NULL, options,
660 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
661 if (!blk) {
662 error_prepend(errp, "Protocol driver '%s' does not support image "
663 "creation, and opening the image failed: ",
664 drv->format_name);
665 return -EINVAL;
668 size = create_file_fallback_truncate(blk, size, errp);
669 if (size < 0) {
670 ret = size;
671 goto out;
674 ret = create_file_fallback_zero_first_sector(blk, size, errp);
675 if (ret < 0) {
676 goto out;
679 ret = 0;
680 out:
681 blk_unref(blk);
682 return ret;
685 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
686 Error **errp)
688 QemuOpts *protocol_opts;
689 BlockDriver *drv;
690 QDict *qdict;
691 int ret;
693 GLOBAL_STATE_CODE();
695 drv = bdrv_find_protocol(filename, true, errp);
696 if (drv == NULL) {
697 return -ENOENT;
700 if (!drv->create_opts) {
701 error_setg(errp, "Driver '%s' does not support image creation",
702 drv->format_name);
703 return -ENOTSUP;
707 * 'opts' contains a QemuOptsList with a combination of format and protocol
708 * default values.
710 * The format properly removes its options, but the default values remain
711 * in 'opts->list'. So if the protocol has options with the same name
712 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
713 * of the format, since for overlapping options, the format wins.
715 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
716 * only the set options, and then convert it back to QemuOpts, using the
717 * create_opts of the protocol. So the new QemuOpts, will contain only the
718 * protocol defaults.
720 qdict = qemu_opts_to_qdict(opts, NULL);
721 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
722 if (protocol_opts == NULL) {
723 ret = -EINVAL;
724 goto out;
727 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
728 out:
729 qemu_opts_del(protocol_opts);
730 qobject_unref(qdict);
731 return ret;
734 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
736 Error *local_err = NULL;
737 int ret;
739 IO_CODE();
740 assert(bs != NULL);
742 if (!bs->drv) {
743 error_setg(errp, "Block node '%s' is not opened", bs->filename);
744 return -ENOMEDIUM;
747 if (!bs->drv->bdrv_co_delete_file) {
748 error_setg(errp, "Driver '%s' does not support image deletion",
749 bs->drv->format_name);
750 return -ENOTSUP;
753 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
754 if (ret < 0) {
755 error_propagate(errp, local_err);
758 return ret;
761 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
763 Error *local_err = NULL;
764 int ret;
765 IO_CODE();
767 if (!bs) {
768 return;
771 ret = bdrv_co_delete_file(bs, &local_err);
773 * ENOTSUP will happen if the block driver doesn't support
774 * the 'bdrv_co_delete_file' interface. This is a predictable
775 * scenario and shouldn't be reported back to the user.
777 if (ret == -ENOTSUP) {
778 error_free(local_err);
779 } else if (ret < 0) {
780 error_report_err(local_err);
785 * Try to get @bs's logical and physical block size.
786 * On success, store them in @bsz struct and return 0.
787 * On failure return -errno.
788 * @bs must not be empty.
790 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
792 BlockDriver *drv = bs->drv;
793 BlockDriverState *filtered = bdrv_filter_bs(bs);
794 GLOBAL_STATE_CODE();
796 if (drv && drv->bdrv_probe_blocksizes) {
797 return drv->bdrv_probe_blocksizes(bs, bsz);
798 } else if (filtered) {
799 return bdrv_probe_blocksizes(filtered, bsz);
802 return -ENOTSUP;
806 * Try to get @bs's geometry (cyls, heads, sectors).
807 * On success, store them in @geo struct and return 0.
808 * On failure return -errno.
809 * @bs must not be empty.
811 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
813 BlockDriver *drv = bs->drv;
814 BlockDriverState *filtered = bdrv_filter_bs(bs);
815 GLOBAL_STATE_CODE();
817 if (drv && drv->bdrv_probe_geometry) {
818 return drv->bdrv_probe_geometry(bs, geo);
819 } else if (filtered) {
820 return bdrv_probe_geometry(filtered, geo);
823 return -ENOTSUP;
827 * Create a uniquely-named empty temporary file.
828 * Return the actual file name used upon success, otherwise NULL.
829 * This string should be freed with g_free() when not needed any longer.
831 * Note: creating a temporary file for the caller to (re)open is
832 * inherently racy. Use g_file_open_tmp() instead whenever practical.
834 char *create_tmp_file(Error **errp)
836 int fd;
837 const char *tmpdir;
838 g_autofree char *filename = NULL;
840 tmpdir = g_get_tmp_dir();
841 #ifndef _WIN32
843 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
845 * This function is used to create temporary disk images (like -snapshot),
846 * so the files can become very large. /tmp is often a tmpfs where as
847 * /var/tmp is usually on a disk, so more appropriate for disk images.
849 if (!g_strcmp0(tmpdir, "/tmp")) {
850 tmpdir = "/var/tmp";
852 #endif
854 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
855 fd = g_mkstemp(filename);
856 if (fd < 0) {
857 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
858 filename);
859 return NULL;
861 close(fd);
863 return g_steal_pointer(&filename);
867 * Detect host devices. By convention, /dev/cdrom[N] is always
868 * recognized as a host CDROM.
870 static BlockDriver *find_hdev_driver(const char *filename)
872 int score_max = 0, score;
873 BlockDriver *drv = NULL, *d;
874 GLOBAL_STATE_CODE();
876 QLIST_FOREACH(d, &bdrv_drivers, list) {
877 if (d->bdrv_probe_device) {
878 score = d->bdrv_probe_device(filename);
879 if (score > score_max) {
880 score_max = score;
881 drv = d;
886 return drv;
889 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
891 BlockDriver *drv1;
892 GLOBAL_STATE_CODE();
894 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
895 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
896 return drv1;
900 return NULL;
903 BlockDriver *bdrv_find_protocol(const char *filename,
904 bool allow_protocol_prefix,
905 Error **errp)
907 BlockDriver *drv1;
908 char protocol[128];
909 int len;
910 const char *p;
911 int i;
913 GLOBAL_STATE_CODE();
914 /* TODO Drivers without bdrv_file_open must be specified explicitly */
917 * XXX(hch): we really should not let host device detection
918 * override an explicit protocol specification, but moving this
919 * later breaks access to device names with colons in them.
920 * Thanks to the brain-dead persistent naming schemes on udev-
921 * based Linux systems those actually are quite common.
923 drv1 = find_hdev_driver(filename);
924 if (drv1) {
925 return drv1;
928 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
929 return &bdrv_file;
932 p = strchr(filename, ':');
933 assert(p != NULL);
934 len = p - filename;
935 if (len > sizeof(protocol) - 1)
936 len = sizeof(protocol) - 1;
937 memcpy(protocol, filename, len);
938 protocol[len] = '\0';
940 drv1 = bdrv_do_find_protocol(protocol);
941 if (drv1) {
942 return drv1;
945 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
946 if (block_driver_modules[i].protocol_name &&
947 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
948 int rv = block_module_load(block_driver_modules[i].library_name, errp);
949 if (rv > 0) {
950 drv1 = bdrv_do_find_protocol(protocol);
951 } else if (rv < 0) {
952 return NULL;
954 break;
958 if (!drv1) {
959 error_setg(errp, "Unknown protocol '%s'", protocol);
961 return drv1;
965 * Guess image format by probing its contents.
966 * This is not a good idea when your image is raw (CVE-2008-2004), but
967 * we do it anyway for backward compatibility.
969 * @buf contains the image's first @buf_size bytes.
970 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
971 * but can be smaller if the image file is smaller)
972 * @filename is its filename.
974 * For all block drivers, call the bdrv_probe() method to get its
975 * probing score.
976 * Return the first block driver with the highest probing score.
978 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
979 const char *filename)
981 int score_max = 0, score;
982 BlockDriver *drv = NULL, *d;
983 IO_CODE();
985 QLIST_FOREACH(d, &bdrv_drivers, list) {
986 if (d->bdrv_probe) {
987 score = d->bdrv_probe(buf, buf_size, filename);
988 if (score > score_max) {
989 score_max = score;
990 drv = d;
995 return drv;
998 static int find_image_format(BlockBackend *file, const char *filename,
999 BlockDriver **pdrv, Error **errp)
1001 BlockDriver *drv;
1002 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1003 int ret = 0;
1005 GLOBAL_STATE_CODE();
1007 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1008 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1009 *pdrv = &bdrv_raw;
1010 return ret;
1013 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1014 if (ret < 0) {
1015 error_setg_errno(errp, -ret, "Could not read image for determining its "
1016 "format");
1017 *pdrv = NULL;
1018 return ret;
1021 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1022 if (!drv) {
1023 error_setg(errp, "Could not determine image format: No compatible "
1024 "driver found");
1025 *pdrv = NULL;
1026 return -ENOENT;
1029 *pdrv = drv;
1030 return 0;
1034 * Set the current 'total_sectors' value
1035 * Return 0 on success, -errno on error.
1037 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
1039 BlockDriver *drv = bs->drv;
1040 IO_CODE();
1042 if (!drv) {
1043 return -ENOMEDIUM;
1046 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1047 if (bdrv_is_sg(bs))
1048 return 0;
1050 /* query actual device if possible, otherwise just trust the hint */
1051 if (drv->bdrv_getlength) {
1052 int64_t length = drv->bdrv_getlength(bs);
1053 if (length < 0) {
1054 return length;
1056 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1059 bs->total_sectors = hint;
1061 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1062 return -EFBIG;
1065 return 0;
1069 * Combines a QDict of new block driver @options with any missing options taken
1070 * from @old_options, so that leaving out an option defaults to its old value.
1072 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1073 QDict *old_options)
1075 GLOBAL_STATE_CODE();
1076 if (bs->drv && bs->drv->bdrv_join_options) {
1077 bs->drv->bdrv_join_options(options, old_options);
1078 } else {
1079 qdict_join(options, old_options, false);
1083 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1084 int open_flags,
1085 Error **errp)
1087 Error *local_err = NULL;
1088 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1089 BlockdevDetectZeroesOptions detect_zeroes =
1090 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1091 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1092 GLOBAL_STATE_CODE();
1093 g_free(value);
1094 if (local_err) {
1095 error_propagate(errp, local_err);
1096 return detect_zeroes;
1099 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1100 !(open_flags & BDRV_O_UNMAP))
1102 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1103 "without setting discard operation to unmap");
1106 return detect_zeroes;
1110 * Set open flags for aio engine
1112 * Return 0 on success, -1 if the engine specified is invalid
1114 int bdrv_parse_aio(const char *mode, int *flags)
1116 if (!strcmp(mode, "threads")) {
1117 /* do nothing, default */
1118 } else if (!strcmp(mode, "native")) {
1119 *flags |= BDRV_O_NATIVE_AIO;
1120 #ifdef CONFIG_LINUX_IO_URING
1121 } else if (!strcmp(mode, "io_uring")) {
1122 *flags |= BDRV_O_IO_URING;
1123 #endif
1124 } else {
1125 return -1;
1128 return 0;
1132 * Set open flags for a given discard mode
1134 * Return 0 on success, -1 if the discard mode was invalid.
1136 int bdrv_parse_discard_flags(const char *mode, int *flags)
1138 *flags &= ~BDRV_O_UNMAP;
1140 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1141 /* do nothing */
1142 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1143 *flags |= BDRV_O_UNMAP;
1144 } else {
1145 return -1;
1148 return 0;
1152 * Set open flags for a given cache mode
1154 * Return 0 on success, -1 if the cache mode was invalid.
1156 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1158 *flags &= ~BDRV_O_CACHE_MASK;
1160 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1161 *writethrough = false;
1162 *flags |= BDRV_O_NOCACHE;
1163 } else if (!strcmp(mode, "directsync")) {
1164 *writethrough = true;
1165 *flags |= BDRV_O_NOCACHE;
1166 } else if (!strcmp(mode, "writeback")) {
1167 *writethrough = false;
1168 } else if (!strcmp(mode, "unsafe")) {
1169 *writethrough = false;
1170 *flags |= BDRV_O_NO_FLUSH;
1171 } else if (!strcmp(mode, "writethrough")) {
1172 *writethrough = true;
1173 } else {
1174 return -1;
1177 return 0;
1180 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1182 BlockDriverState *parent = c->opaque;
1183 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1186 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1188 BlockDriverState *bs = child->opaque;
1189 bdrv_do_drained_begin_quiesce(bs, NULL);
1192 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1194 BlockDriverState *bs = child->opaque;
1195 return bdrv_drain_poll(bs, NULL, false);
1198 static void bdrv_child_cb_drained_end(BdrvChild *child)
1200 BlockDriverState *bs = child->opaque;
1201 bdrv_drained_end(bs);
1204 static int bdrv_child_cb_inactivate(BdrvChild *child)
1206 BlockDriverState *bs = child->opaque;
1207 GLOBAL_STATE_CODE();
1208 assert(bs->open_flags & BDRV_O_INACTIVE);
1209 return 0;
1212 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1213 GHashTable *visited, Transaction *tran,
1214 Error **errp)
1216 BlockDriverState *bs = child->opaque;
1217 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1221 * Returns the options and flags that a temporary snapshot should get, based on
1222 * the originally requested flags (the originally requested image will have
1223 * flags like a backing file)
1225 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1226 int parent_flags, QDict *parent_options)
1228 GLOBAL_STATE_CODE();
1229 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1231 /* For temporary files, unconditional cache=unsafe is fine */
1232 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1233 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1235 /* Copy the read-only and discard options from the parent */
1236 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1237 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1239 /* aio=native doesn't work for cache.direct=off, so disable it for the
1240 * temporary snapshot */
1241 *child_flags &= ~BDRV_O_NATIVE_AIO;
1244 static void bdrv_backing_attach(BdrvChild *c)
1246 BlockDriverState *parent = c->opaque;
1247 BlockDriverState *backing_hd = c->bs;
1249 GLOBAL_STATE_CODE();
1250 assert(!parent->backing_blocker);
1251 error_setg(&parent->backing_blocker,
1252 "node is used as backing hd of '%s'",
1253 bdrv_get_device_or_node_name(parent));
1255 bdrv_refresh_filename(backing_hd);
1257 parent->open_flags &= ~BDRV_O_NO_BACKING;
1259 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1260 /* Otherwise we won't be able to commit or stream */
1261 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1262 parent->backing_blocker);
1263 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1264 parent->backing_blocker);
1266 * We do backup in 3 ways:
1267 * 1. drive backup
1268 * The target bs is new opened, and the source is top BDS
1269 * 2. blockdev backup
1270 * Both the source and the target are top BDSes.
1271 * 3. internal backup(used for block replication)
1272 * Both the source and the target are backing file
1274 * In case 1 and 2, neither the source nor the target is the backing file.
1275 * In case 3, we will block the top BDS, so there is only one block job
1276 * for the top BDS and its backing chain.
1278 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1279 parent->backing_blocker);
1280 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1281 parent->backing_blocker);
1284 static void bdrv_backing_detach(BdrvChild *c)
1286 BlockDriverState *parent = c->opaque;
1288 GLOBAL_STATE_CODE();
1289 assert(parent->backing_blocker);
1290 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1291 error_free(parent->backing_blocker);
1292 parent->backing_blocker = NULL;
1295 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1296 const char *filename, Error **errp)
1298 BlockDriverState *parent = c->opaque;
1299 bool read_only = bdrv_is_read_only(parent);
1300 int ret;
1301 GLOBAL_STATE_CODE();
1303 if (read_only) {
1304 ret = bdrv_reopen_set_read_only(parent, false, errp);
1305 if (ret < 0) {
1306 return ret;
1310 ret = bdrv_change_backing_file(parent, filename,
1311 base->drv ? base->drv->format_name : "",
1312 false);
1313 if (ret < 0) {
1314 error_setg_errno(errp, -ret, "Could not update backing file link");
1317 if (read_only) {
1318 bdrv_reopen_set_read_only(parent, true, NULL);
1321 return ret;
1325 * Returns the options and flags that a generic child of a BDS should
1326 * get, based on the given options and flags for the parent BDS.
1328 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1329 int *child_flags, QDict *child_options,
1330 int parent_flags, QDict *parent_options)
1332 int flags = parent_flags;
1333 GLOBAL_STATE_CODE();
1336 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1337 * Generally, the question to answer is: Should this child be
1338 * format-probed by default?
1342 * Pure and non-filtered data children of non-format nodes should
1343 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1344 * set). This only affects a very limited set of drivers (namely
1345 * quorum and blkverify when this comment was written).
1346 * Force-clear BDRV_O_PROTOCOL then.
1348 if (!parent_is_format &&
1349 (role & BDRV_CHILD_DATA) &&
1350 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1352 flags &= ~BDRV_O_PROTOCOL;
1356 * All children of format nodes (except for COW children) and all
1357 * metadata children in general should never be format-probed.
1358 * Force-set BDRV_O_PROTOCOL then.
1360 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1361 (role & BDRV_CHILD_METADATA))
1363 flags |= BDRV_O_PROTOCOL;
1367 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1368 * the parent.
1370 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1371 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1372 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1374 if (role & BDRV_CHILD_COW) {
1375 /* backing files are opened read-only by default */
1376 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1377 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1378 } else {
1379 /* Inherit the read-only option from the parent if it's not set */
1380 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1381 qdict_copy_default(child_options, parent_options,
1382 BDRV_OPT_AUTO_READ_ONLY);
1386 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1387 * can default to enable it on lower layers regardless of the
1388 * parent option.
1390 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1392 /* Clear flags that only apply to the top layer */
1393 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1395 if (role & BDRV_CHILD_METADATA) {
1396 flags &= ~BDRV_O_NO_IO;
1398 if (role & BDRV_CHILD_COW) {
1399 flags &= ~BDRV_O_TEMPORARY;
1402 *child_flags = flags;
1405 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1407 BlockDriverState *bs = child->opaque;
1409 assert_bdrv_graph_writable();
1410 QLIST_INSERT_HEAD(&bs->children, child, next);
1411 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1413 * Here we handle filters and block/raw-format.c when it behave like
1414 * filter. They generally have a single PRIMARY child, which is also the
1415 * FILTERED child, and that they may have multiple more children, which
1416 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1417 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1418 * into bs->backing on exceptional cases; and bs->backing will be
1419 * nothing else.
1421 assert(!(child->role & BDRV_CHILD_COW));
1422 if (child->role & BDRV_CHILD_PRIMARY) {
1423 assert(child->role & BDRV_CHILD_FILTERED);
1424 assert(!bs->backing);
1425 assert(!bs->file);
1427 if (bs->drv->filtered_child_is_backing) {
1428 bs->backing = child;
1429 } else {
1430 bs->file = child;
1432 } else {
1433 assert(!(child->role & BDRV_CHILD_FILTERED));
1435 } else if (child->role & BDRV_CHILD_COW) {
1436 assert(bs->drv->supports_backing);
1437 assert(!(child->role & BDRV_CHILD_PRIMARY));
1438 assert(!bs->backing);
1439 bs->backing = child;
1440 bdrv_backing_attach(child);
1441 } else if (child->role & BDRV_CHILD_PRIMARY) {
1442 assert(!bs->file);
1443 bs->file = child;
1447 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1449 BlockDriverState *bs = child->opaque;
1451 if (child->role & BDRV_CHILD_COW) {
1452 bdrv_backing_detach(child);
1455 assert_bdrv_graph_writable();
1456 QLIST_REMOVE(child, next);
1457 if (child == bs->backing) {
1458 assert(child != bs->file);
1459 bs->backing = NULL;
1460 } else if (child == bs->file) {
1461 bs->file = NULL;
1465 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1466 const char *filename, Error **errp)
1468 if (c->role & BDRV_CHILD_COW) {
1469 return bdrv_backing_update_filename(c, base, filename, errp);
1471 return 0;
1474 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1476 BlockDriverState *bs = c->opaque;
1477 IO_CODE();
1479 return bdrv_get_aio_context(bs);
1482 const BdrvChildClass child_of_bds = {
1483 .parent_is_bds = true,
1484 .get_parent_desc = bdrv_child_get_parent_desc,
1485 .inherit_options = bdrv_inherited_options,
1486 .drained_begin = bdrv_child_cb_drained_begin,
1487 .drained_poll = bdrv_child_cb_drained_poll,
1488 .drained_end = bdrv_child_cb_drained_end,
1489 .attach = bdrv_child_cb_attach,
1490 .detach = bdrv_child_cb_detach,
1491 .inactivate = bdrv_child_cb_inactivate,
1492 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1493 .update_filename = bdrv_child_cb_update_filename,
1494 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1497 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1499 IO_CODE();
1500 return c->klass->get_parent_aio_context(c);
1503 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1505 int open_flags = flags;
1506 GLOBAL_STATE_CODE();
1509 * Clear flags that are internal to the block layer before opening the
1510 * image.
1512 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1514 return open_flags;
1517 static void update_flags_from_options(int *flags, QemuOpts *opts)
1519 GLOBAL_STATE_CODE();
1521 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1523 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1524 *flags |= BDRV_O_NO_FLUSH;
1527 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1528 *flags |= BDRV_O_NOCACHE;
1531 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1532 *flags |= BDRV_O_RDWR;
1535 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1536 *flags |= BDRV_O_AUTO_RDONLY;
1540 static void update_options_from_flags(QDict *options, int flags)
1542 GLOBAL_STATE_CODE();
1543 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1544 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1546 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1547 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1548 flags & BDRV_O_NO_FLUSH);
1550 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1551 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1553 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1554 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1555 flags & BDRV_O_AUTO_RDONLY);
1559 static void bdrv_assign_node_name(BlockDriverState *bs,
1560 const char *node_name,
1561 Error **errp)
1563 char *gen_node_name = NULL;
1564 GLOBAL_STATE_CODE();
1566 if (!node_name) {
1567 node_name = gen_node_name = id_generate(ID_BLOCK);
1568 } else if (!id_wellformed(node_name)) {
1570 * Check for empty string or invalid characters, but not if it is
1571 * generated (generated names use characters not available to the user)
1573 error_setg(errp, "Invalid node-name: '%s'", node_name);
1574 return;
1577 /* takes care of avoiding namespaces collisions */
1578 if (blk_by_name(node_name)) {
1579 error_setg(errp, "node-name=%s is conflicting with a device id",
1580 node_name);
1581 goto out;
1584 /* takes care of avoiding duplicates node names */
1585 if (bdrv_find_node(node_name)) {
1586 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1587 goto out;
1590 /* Make sure that the node name isn't truncated */
1591 if (strlen(node_name) >= sizeof(bs->node_name)) {
1592 error_setg(errp, "Node name too long");
1593 goto out;
1596 /* copy node name into the bs and insert it into the graph list */
1597 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1598 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1599 out:
1600 g_free(gen_node_name);
1603 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1604 const char *node_name, QDict *options,
1605 int open_flags, Error **errp)
1607 Error *local_err = NULL;
1608 int i, ret;
1609 GLOBAL_STATE_CODE();
1611 bdrv_assign_node_name(bs, node_name, &local_err);
1612 if (local_err) {
1613 error_propagate(errp, local_err);
1614 return -EINVAL;
1617 bs->drv = drv;
1618 bs->opaque = g_malloc0(drv->instance_size);
1620 if (drv->bdrv_file_open) {
1621 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1622 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1623 } else if (drv->bdrv_open) {
1624 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1625 } else {
1626 ret = 0;
1629 if (ret < 0) {
1630 if (local_err) {
1631 error_propagate(errp, local_err);
1632 } else if (bs->filename[0]) {
1633 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1634 } else {
1635 error_setg_errno(errp, -ret, "Could not open image");
1637 goto open_failed;
1640 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1641 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1644 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1645 * drivers that pass read/write requests through to a child the trouble of
1646 * declaring support explicitly.
1648 * Drivers must not propagate this flag accidentally when they initiate I/O
1649 * to a bounce buffer. That case should be rare though.
1651 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1652 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1654 ret = refresh_total_sectors(bs, bs->total_sectors);
1655 if (ret < 0) {
1656 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1657 return ret;
1660 bdrv_refresh_limits(bs, NULL, &local_err);
1661 if (local_err) {
1662 error_propagate(errp, local_err);
1663 return -EINVAL;
1666 assert(bdrv_opt_mem_align(bs) != 0);
1667 assert(bdrv_min_mem_align(bs) != 0);
1668 assert(is_power_of_2(bs->bl.request_alignment));
1670 for (i = 0; i < bs->quiesce_counter; i++) {
1671 if (drv->bdrv_drain_begin) {
1672 drv->bdrv_drain_begin(bs);
1676 return 0;
1677 open_failed:
1678 bs->drv = NULL;
1679 if (bs->file != NULL) {
1680 bdrv_unref_child(bs, bs->file);
1681 assert(!bs->file);
1683 g_free(bs->opaque);
1684 bs->opaque = NULL;
1685 return ret;
1689 * Create and open a block node.
1691 * @options is a QDict of options to pass to the block drivers, or NULL for an
1692 * empty set of options. The reference to the QDict belongs to the block layer
1693 * after the call (even on failure), so if the caller intends to reuse the
1694 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1696 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1697 const char *node_name,
1698 QDict *options, int flags,
1699 Error **errp)
1701 BlockDriverState *bs;
1702 int ret;
1704 GLOBAL_STATE_CODE();
1706 bs = bdrv_new();
1707 bs->open_flags = flags;
1708 bs->options = options ?: qdict_new();
1709 bs->explicit_options = qdict_clone_shallow(bs->options);
1710 bs->opaque = NULL;
1712 update_options_from_flags(bs->options, flags);
1714 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1715 if (ret < 0) {
1716 qobject_unref(bs->explicit_options);
1717 bs->explicit_options = NULL;
1718 qobject_unref(bs->options);
1719 bs->options = NULL;
1720 bdrv_unref(bs);
1721 return NULL;
1724 return bs;
1727 /* Create and open a block node. */
1728 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1729 int flags, Error **errp)
1731 GLOBAL_STATE_CODE();
1732 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1735 QemuOptsList bdrv_runtime_opts = {
1736 .name = "bdrv_common",
1737 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1738 .desc = {
1740 .name = "node-name",
1741 .type = QEMU_OPT_STRING,
1742 .help = "Node name of the block device node",
1745 .name = "driver",
1746 .type = QEMU_OPT_STRING,
1747 .help = "Block driver to use for the node",
1750 .name = BDRV_OPT_CACHE_DIRECT,
1751 .type = QEMU_OPT_BOOL,
1752 .help = "Bypass software writeback cache on the host",
1755 .name = BDRV_OPT_CACHE_NO_FLUSH,
1756 .type = QEMU_OPT_BOOL,
1757 .help = "Ignore flush requests",
1760 .name = BDRV_OPT_READ_ONLY,
1761 .type = QEMU_OPT_BOOL,
1762 .help = "Node is opened in read-only mode",
1765 .name = BDRV_OPT_AUTO_READ_ONLY,
1766 .type = QEMU_OPT_BOOL,
1767 .help = "Node can become read-only if opening read-write fails",
1770 .name = "detect-zeroes",
1771 .type = QEMU_OPT_STRING,
1772 .help = "try to optimize zero writes (off, on, unmap)",
1775 .name = BDRV_OPT_DISCARD,
1776 .type = QEMU_OPT_STRING,
1777 .help = "discard operation (ignore/off, unmap/on)",
1780 .name = BDRV_OPT_FORCE_SHARE,
1781 .type = QEMU_OPT_BOOL,
1782 .help = "always accept other writers (default: off)",
1784 { /* end of list */ }
1788 QemuOptsList bdrv_create_opts_simple = {
1789 .name = "simple-create-opts",
1790 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1791 .desc = {
1793 .name = BLOCK_OPT_SIZE,
1794 .type = QEMU_OPT_SIZE,
1795 .help = "Virtual disk size"
1798 .name = BLOCK_OPT_PREALLOC,
1799 .type = QEMU_OPT_STRING,
1800 .help = "Preallocation mode (allowed values: off)"
1802 { /* end of list */ }
1807 * Common part for opening disk images and files
1809 * Removes all processed options from *options.
1811 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1812 QDict *options, Error **errp)
1814 int ret, open_flags;
1815 const char *filename;
1816 const char *driver_name = NULL;
1817 const char *node_name = NULL;
1818 const char *discard;
1819 QemuOpts *opts;
1820 BlockDriver *drv;
1821 Error *local_err = NULL;
1822 bool ro;
1824 assert(bs->file == NULL);
1825 assert(options != NULL && bs->options != options);
1826 GLOBAL_STATE_CODE();
1828 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1829 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1830 ret = -EINVAL;
1831 goto fail_opts;
1834 update_flags_from_options(&bs->open_flags, opts);
1836 driver_name = qemu_opt_get(opts, "driver");
1837 drv = bdrv_find_format(driver_name);
1838 assert(drv != NULL);
1840 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1842 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1843 error_setg(errp,
1844 BDRV_OPT_FORCE_SHARE
1845 "=on can only be used with read-only images");
1846 ret = -EINVAL;
1847 goto fail_opts;
1850 if (file != NULL) {
1851 bdrv_refresh_filename(blk_bs(file));
1852 filename = blk_bs(file)->filename;
1853 } else {
1855 * Caution: while qdict_get_try_str() is fine, getting
1856 * non-string types would require more care. When @options
1857 * come from -blockdev or blockdev_add, its members are typed
1858 * according to the QAPI schema, but when they come from
1859 * -drive, they're all QString.
1861 filename = qdict_get_try_str(options, "filename");
1864 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1865 error_setg(errp, "The '%s' block driver requires a file name",
1866 drv->format_name);
1867 ret = -EINVAL;
1868 goto fail_opts;
1871 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1872 drv->format_name);
1874 ro = bdrv_is_read_only(bs);
1876 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1877 if (!ro && bdrv_is_whitelisted(drv, true)) {
1878 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1879 } else {
1880 ret = -ENOTSUP;
1882 if (ret < 0) {
1883 error_setg(errp,
1884 !ro && bdrv_is_whitelisted(drv, true)
1885 ? "Driver '%s' can only be used for read-only devices"
1886 : "Driver '%s' is not whitelisted",
1887 drv->format_name);
1888 goto fail_opts;
1892 /* bdrv_new() and bdrv_close() make it so */
1893 assert(qatomic_read(&bs->copy_on_read) == 0);
1895 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1896 if (!ro) {
1897 bdrv_enable_copy_on_read(bs);
1898 } else {
1899 error_setg(errp, "Can't use copy-on-read on read-only device");
1900 ret = -EINVAL;
1901 goto fail_opts;
1905 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1906 if (discard != NULL) {
1907 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1908 error_setg(errp, "Invalid discard option");
1909 ret = -EINVAL;
1910 goto fail_opts;
1914 bs->detect_zeroes =
1915 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1916 if (local_err) {
1917 error_propagate(errp, local_err);
1918 ret = -EINVAL;
1919 goto fail_opts;
1922 if (filename != NULL) {
1923 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1924 } else {
1925 bs->filename[0] = '\0';
1927 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1929 /* Open the image, either directly or using a protocol */
1930 open_flags = bdrv_open_flags(bs, bs->open_flags);
1931 node_name = qemu_opt_get(opts, "node-name");
1933 assert(!drv->bdrv_file_open || file == NULL);
1934 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1935 if (ret < 0) {
1936 goto fail_opts;
1939 qemu_opts_del(opts);
1940 return 0;
1942 fail_opts:
1943 qemu_opts_del(opts);
1944 return ret;
1947 static QDict *parse_json_filename(const char *filename, Error **errp)
1949 QObject *options_obj;
1950 QDict *options;
1951 int ret;
1952 GLOBAL_STATE_CODE();
1954 ret = strstart(filename, "json:", &filename);
1955 assert(ret);
1957 options_obj = qobject_from_json(filename, errp);
1958 if (!options_obj) {
1959 error_prepend(errp, "Could not parse the JSON options: ");
1960 return NULL;
1963 options = qobject_to(QDict, options_obj);
1964 if (!options) {
1965 qobject_unref(options_obj);
1966 error_setg(errp, "Invalid JSON object given");
1967 return NULL;
1970 qdict_flatten(options);
1972 return options;
1975 static void parse_json_protocol(QDict *options, const char **pfilename,
1976 Error **errp)
1978 QDict *json_options;
1979 Error *local_err = NULL;
1980 GLOBAL_STATE_CODE();
1982 /* Parse json: pseudo-protocol */
1983 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1984 return;
1987 json_options = parse_json_filename(*pfilename, &local_err);
1988 if (local_err) {
1989 error_propagate(errp, local_err);
1990 return;
1993 /* Options given in the filename have lower priority than options
1994 * specified directly */
1995 qdict_join(options, json_options, false);
1996 qobject_unref(json_options);
1997 *pfilename = NULL;
2001 * Fills in default options for opening images and converts the legacy
2002 * filename/flags pair to option QDict entries.
2003 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2004 * block driver has been specified explicitly.
2006 static int bdrv_fill_options(QDict **options, const char *filename,
2007 int *flags, Error **errp)
2009 const char *drvname;
2010 bool protocol = *flags & BDRV_O_PROTOCOL;
2011 bool parse_filename = false;
2012 BlockDriver *drv = NULL;
2013 Error *local_err = NULL;
2015 GLOBAL_STATE_CODE();
2018 * Caution: while qdict_get_try_str() is fine, getting non-string
2019 * types would require more care. When @options come from
2020 * -blockdev or blockdev_add, its members are typed according to
2021 * the QAPI schema, but when they come from -drive, they're all
2022 * QString.
2024 drvname = qdict_get_try_str(*options, "driver");
2025 if (drvname) {
2026 drv = bdrv_find_format(drvname);
2027 if (!drv) {
2028 error_setg(errp, "Unknown driver '%s'", drvname);
2029 return -ENOENT;
2031 /* If the user has explicitly specified the driver, this choice should
2032 * override the BDRV_O_PROTOCOL flag */
2033 protocol = drv->bdrv_file_open;
2036 if (protocol) {
2037 *flags |= BDRV_O_PROTOCOL;
2038 } else {
2039 *flags &= ~BDRV_O_PROTOCOL;
2042 /* Translate cache options from flags into options */
2043 update_options_from_flags(*options, *flags);
2045 /* Fetch the file name from the options QDict if necessary */
2046 if (protocol && filename) {
2047 if (!qdict_haskey(*options, "filename")) {
2048 qdict_put_str(*options, "filename", filename);
2049 parse_filename = true;
2050 } else {
2051 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2052 "the same time");
2053 return -EINVAL;
2057 /* Find the right block driver */
2058 /* See cautionary note on accessing @options above */
2059 filename = qdict_get_try_str(*options, "filename");
2061 if (!drvname && protocol) {
2062 if (filename) {
2063 drv = bdrv_find_protocol(filename, parse_filename, errp);
2064 if (!drv) {
2065 return -EINVAL;
2068 drvname = drv->format_name;
2069 qdict_put_str(*options, "driver", drvname);
2070 } else {
2071 error_setg(errp, "Must specify either driver or file");
2072 return -EINVAL;
2076 assert(drv || !protocol);
2078 /* Driver-specific filename parsing */
2079 if (drv && drv->bdrv_parse_filename && parse_filename) {
2080 drv->bdrv_parse_filename(filename, *options, &local_err);
2081 if (local_err) {
2082 error_propagate(errp, local_err);
2083 return -EINVAL;
2086 if (!drv->bdrv_needs_filename) {
2087 qdict_del(*options, "filename");
2091 return 0;
2094 typedef struct BlockReopenQueueEntry {
2095 bool prepared;
2096 bool perms_checked;
2097 BDRVReopenState state;
2098 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2099 } BlockReopenQueueEntry;
2102 * Return the flags that @bs will have after the reopens in @q have
2103 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2104 * return the current flags.
2106 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2108 BlockReopenQueueEntry *entry;
2110 if (q != NULL) {
2111 QTAILQ_FOREACH(entry, q, entry) {
2112 if (entry->state.bs == bs) {
2113 return entry->state.flags;
2118 return bs->open_flags;
2121 /* Returns whether the image file can be written to after the reopen queue @q
2122 * has been successfully applied, or right now if @q is NULL. */
2123 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2124 BlockReopenQueue *q)
2126 int flags = bdrv_reopen_get_flags(q, bs);
2128 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2132 * Return whether the BDS can be written to. This is not necessarily
2133 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2134 * be written to but do not count as read-only images.
2136 bool bdrv_is_writable(BlockDriverState *bs)
2138 IO_CODE();
2139 return bdrv_is_writable_after_reopen(bs, NULL);
2142 static char *bdrv_child_user_desc(BdrvChild *c)
2144 GLOBAL_STATE_CODE();
2145 return c->klass->get_parent_desc(c);
2149 * Check that @a allows everything that @b needs. @a and @b must reference same
2150 * child node.
2152 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2154 const char *child_bs_name;
2155 g_autofree char *a_user = NULL;
2156 g_autofree char *b_user = NULL;
2157 g_autofree char *perms = NULL;
2159 assert(a->bs);
2160 assert(a->bs == b->bs);
2161 GLOBAL_STATE_CODE();
2163 if ((b->perm & a->shared_perm) == b->perm) {
2164 return true;
2167 child_bs_name = bdrv_get_node_name(b->bs);
2168 a_user = bdrv_child_user_desc(a);
2169 b_user = bdrv_child_user_desc(b);
2170 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2172 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2173 "both required by %s (uses node '%s' as '%s' child) and "
2174 "unshared by %s (uses node '%s' as '%s' child).",
2175 child_bs_name, perms,
2176 b_user, child_bs_name, b->name,
2177 a_user, child_bs_name, a->name);
2179 return false;
2182 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2184 BdrvChild *a, *b;
2185 GLOBAL_STATE_CODE();
2188 * During the loop we'll look at each pair twice. That's correct because
2189 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2190 * directions.
2192 QLIST_FOREACH(a, &bs->parents, next_parent) {
2193 QLIST_FOREACH(b, &bs->parents, next_parent) {
2194 if (a == b) {
2195 continue;
2198 if (!bdrv_a_allow_b(a, b, errp)) {
2199 return true;
2204 return false;
2207 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2208 BdrvChild *c, BdrvChildRole role,
2209 BlockReopenQueue *reopen_queue,
2210 uint64_t parent_perm, uint64_t parent_shared,
2211 uint64_t *nperm, uint64_t *nshared)
2213 assert(bs->drv && bs->drv->bdrv_child_perm);
2214 GLOBAL_STATE_CODE();
2215 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2216 parent_perm, parent_shared,
2217 nperm, nshared);
2218 /* TODO Take force_share from reopen_queue */
2219 if (child_bs && child_bs->force_share) {
2220 *nshared = BLK_PERM_ALL;
2225 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2226 * nodes that are already in the @list, of course) so that final list is
2227 * topologically sorted. Return the result (GSList @list object is updated, so
2228 * don't use old reference after function call).
2230 * On function start @list must be already topologically sorted and for any node
2231 * in the @list the whole subtree of the node must be in the @list as well. The
2232 * simplest way to satisfy this criteria: use only result of
2233 * bdrv_topological_dfs() or NULL as @list parameter.
2235 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2236 BlockDriverState *bs)
2238 BdrvChild *child;
2239 g_autoptr(GHashTable) local_found = NULL;
2241 GLOBAL_STATE_CODE();
2243 if (!found) {
2244 assert(!list);
2245 found = local_found = g_hash_table_new(NULL, NULL);
2248 if (g_hash_table_contains(found, bs)) {
2249 return list;
2251 g_hash_table_add(found, bs);
2253 QLIST_FOREACH(child, &bs->children, next) {
2254 list = bdrv_topological_dfs(list, found, child->bs);
2257 return g_slist_prepend(list, bs);
2260 typedef struct BdrvChildSetPermState {
2261 BdrvChild *child;
2262 uint64_t old_perm;
2263 uint64_t old_shared_perm;
2264 } BdrvChildSetPermState;
2266 static void bdrv_child_set_perm_abort(void *opaque)
2268 BdrvChildSetPermState *s = opaque;
2270 GLOBAL_STATE_CODE();
2272 s->child->perm = s->old_perm;
2273 s->child->shared_perm = s->old_shared_perm;
2276 static TransactionActionDrv bdrv_child_set_pem_drv = {
2277 .abort = bdrv_child_set_perm_abort,
2278 .clean = g_free,
2281 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2282 uint64_t shared, Transaction *tran)
2284 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2285 GLOBAL_STATE_CODE();
2287 *s = (BdrvChildSetPermState) {
2288 .child = c,
2289 .old_perm = c->perm,
2290 .old_shared_perm = c->shared_perm,
2293 c->perm = perm;
2294 c->shared_perm = shared;
2296 tran_add(tran, &bdrv_child_set_pem_drv, s);
2299 static void bdrv_drv_set_perm_commit(void *opaque)
2301 BlockDriverState *bs = opaque;
2302 uint64_t cumulative_perms, cumulative_shared_perms;
2303 GLOBAL_STATE_CODE();
2305 if (bs->drv->bdrv_set_perm) {
2306 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2307 &cumulative_shared_perms);
2308 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2312 static void bdrv_drv_set_perm_abort(void *opaque)
2314 BlockDriverState *bs = opaque;
2315 GLOBAL_STATE_CODE();
2317 if (bs->drv->bdrv_abort_perm_update) {
2318 bs->drv->bdrv_abort_perm_update(bs);
2322 TransactionActionDrv bdrv_drv_set_perm_drv = {
2323 .abort = bdrv_drv_set_perm_abort,
2324 .commit = bdrv_drv_set_perm_commit,
2327 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2328 uint64_t shared_perm, Transaction *tran,
2329 Error **errp)
2331 GLOBAL_STATE_CODE();
2332 if (!bs->drv) {
2333 return 0;
2336 if (bs->drv->bdrv_check_perm) {
2337 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2338 if (ret < 0) {
2339 return ret;
2343 if (tran) {
2344 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2347 return 0;
2350 typedef struct BdrvReplaceChildState {
2351 BdrvChild *child;
2352 BlockDriverState *old_bs;
2353 } BdrvReplaceChildState;
2355 static void bdrv_replace_child_commit(void *opaque)
2357 BdrvReplaceChildState *s = opaque;
2358 GLOBAL_STATE_CODE();
2360 bdrv_unref(s->old_bs);
2363 static void bdrv_replace_child_abort(void *opaque)
2365 BdrvReplaceChildState *s = opaque;
2366 BlockDriverState *new_bs = s->child->bs;
2368 GLOBAL_STATE_CODE();
2369 /* old_bs reference is transparently moved from @s to @s->child */
2370 if (!s->child->bs) {
2372 * The parents were undrained when removing old_bs from the child. New
2373 * requests can't have been made, though, because the child was empty.
2375 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2376 * undraining the parent in the first place. Once this is done, having
2377 * new_bs drained when calling bdrv_replace_child_tran() is not a
2378 * requirement any more.
2380 bdrv_parent_drained_begin_single(s->child);
2381 assert(!bdrv_parent_drained_poll_single(s->child));
2383 assert(s->child->quiesced_parent);
2384 bdrv_replace_child_noperm(s->child, s->old_bs);
2385 bdrv_unref(new_bs);
2388 static TransactionActionDrv bdrv_replace_child_drv = {
2389 .commit = bdrv_replace_child_commit,
2390 .abort = bdrv_replace_child_abort,
2391 .clean = g_free,
2395 * bdrv_replace_child_tran
2397 * Note: real unref of old_bs is done only on commit.
2399 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2400 * kept drained until the transaction is completed.
2402 * The function doesn't update permissions, caller is responsible for this.
2404 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2405 Transaction *tran)
2407 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2409 assert(child->quiesced_parent);
2410 assert(!new_bs || new_bs->quiesce_counter);
2412 *s = (BdrvReplaceChildState) {
2413 .child = child,
2414 .old_bs = child->bs,
2416 tran_add(tran, &bdrv_replace_child_drv, s);
2418 if (new_bs) {
2419 bdrv_ref(new_bs);
2421 bdrv_replace_child_noperm(child, new_bs);
2422 /* old_bs reference is transparently moved from @child to @s */
2426 * Refresh permissions in @bs subtree. The function is intended to be called
2427 * after some graph modification that was done without permission update.
2429 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2430 Transaction *tran, Error **errp)
2432 BlockDriver *drv = bs->drv;
2433 BdrvChild *c;
2434 int ret;
2435 uint64_t cumulative_perms, cumulative_shared_perms;
2436 GLOBAL_STATE_CODE();
2438 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2440 /* Write permissions never work with read-only images */
2441 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2442 !bdrv_is_writable_after_reopen(bs, q))
2444 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2445 error_setg(errp, "Block node is read-only");
2446 } else {
2447 error_setg(errp, "Read-only block node '%s' cannot support "
2448 "read-write users", bdrv_get_node_name(bs));
2451 return -EPERM;
2455 * Unaligned requests will automatically be aligned to bl.request_alignment
2456 * and without RESIZE we can't extend requests to write to space beyond the
2457 * end of the image, so it's required that the image size is aligned.
2459 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2460 !(cumulative_perms & BLK_PERM_RESIZE))
2462 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2463 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2464 "Image size is not a multiple of request "
2465 "alignment");
2466 return -EPERM;
2470 /* Check this node */
2471 if (!drv) {
2472 return 0;
2475 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2476 errp);
2477 if (ret < 0) {
2478 return ret;
2481 /* Drivers that never have children can omit .bdrv_child_perm() */
2482 if (!drv->bdrv_child_perm) {
2483 assert(QLIST_EMPTY(&bs->children));
2484 return 0;
2487 /* Check all children */
2488 QLIST_FOREACH(c, &bs->children, next) {
2489 uint64_t cur_perm, cur_shared;
2491 bdrv_child_perm(bs, c->bs, c, c->role, q,
2492 cumulative_perms, cumulative_shared_perms,
2493 &cur_perm, &cur_shared);
2494 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2497 return 0;
2501 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2502 * a topologically sorted subgraph.
2504 static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2505 Transaction *tran, Error **errp)
2507 int ret;
2508 BlockDriverState *bs;
2509 GLOBAL_STATE_CODE();
2511 for ( ; list; list = list->next) {
2512 bs = list->data;
2514 if (bdrv_parent_perms_conflict(bs, errp)) {
2515 return -EINVAL;
2518 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2519 if (ret < 0) {
2520 return ret;
2524 return 0;
2528 * @list is any list of nodes. List is completed by all subtrees and
2529 * topologically sorted. It's not a problem if some node occurs in the @list
2530 * several times.
2532 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2533 Transaction *tran, Error **errp)
2535 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2536 g_autoptr(GSList) refresh_list = NULL;
2538 for ( ; list; list = list->next) {
2539 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2542 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2545 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2546 uint64_t *shared_perm)
2548 BdrvChild *c;
2549 uint64_t cumulative_perms = 0;
2550 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2552 GLOBAL_STATE_CODE();
2554 QLIST_FOREACH(c, &bs->parents, next_parent) {
2555 cumulative_perms |= c->perm;
2556 cumulative_shared_perms &= c->shared_perm;
2559 *perm = cumulative_perms;
2560 *shared_perm = cumulative_shared_perms;
2563 char *bdrv_perm_names(uint64_t perm)
2565 struct perm_name {
2566 uint64_t perm;
2567 const char *name;
2568 } permissions[] = {
2569 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2570 { BLK_PERM_WRITE, "write" },
2571 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2572 { BLK_PERM_RESIZE, "resize" },
2573 { 0, NULL }
2576 GString *result = g_string_sized_new(30);
2577 struct perm_name *p;
2579 for (p = permissions; p->name; p++) {
2580 if (perm & p->perm) {
2581 if (result->len > 0) {
2582 g_string_append(result, ", ");
2584 g_string_append(result, p->name);
2588 return g_string_free(result, FALSE);
2592 /* @tran is allowed to be NULL. In this case no rollback is possible */
2593 static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2594 Error **errp)
2596 int ret;
2597 Transaction *local_tran = NULL;
2598 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2599 GLOBAL_STATE_CODE();
2601 if (!tran) {
2602 tran = local_tran = tran_new();
2605 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2607 if (local_tran) {
2608 tran_finalize(local_tran, ret);
2611 return ret;
2614 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2615 Error **errp)
2617 Error *local_err = NULL;
2618 Transaction *tran = tran_new();
2619 int ret;
2621 GLOBAL_STATE_CODE();
2623 bdrv_child_set_perm(c, perm, shared, tran);
2625 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2627 tran_finalize(tran, ret);
2629 if (ret < 0) {
2630 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2631 /* tighten permissions */
2632 error_propagate(errp, local_err);
2633 } else {
2635 * Our caller may intend to only loosen restrictions and
2636 * does not expect this function to fail. Errors are not
2637 * fatal in such a case, so we can just hide them from our
2638 * caller.
2640 error_free(local_err);
2641 ret = 0;
2645 return ret;
2648 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2650 uint64_t parent_perms, parent_shared;
2651 uint64_t perms, shared;
2653 GLOBAL_STATE_CODE();
2655 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2656 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2657 parent_perms, parent_shared, &perms, &shared);
2659 return bdrv_child_try_set_perm(c, perms, shared, errp);
2663 * Default implementation for .bdrv_child_perm() for block filters:
2664 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2665 * filtered child.
2667 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2668 BdrvChildRole role,
2669 BlockReopenQueue *reopen_queue,
2670 uint64_t perm, uint64_t shared,
2671 uint64_t *nperm, uint64_t *nshared)
2673 GLOBAL_STATE_CODE();
2674 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2675 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2678 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2679 BdrvChildRole role,
2680 BlockReopenQueue *reopen_queue,
2681 uint64_t perm, uint64_t shared,
2682 uint64_t *nperm, uint64_t *nshared)
2684 assert(role & BDRV_CHILD_COW);
2685 GLOBAL_STATE_CODE();
2688 * We want consistent read from backing files if the parent needs it.
2689 * No other operations are performed on backing files.
2691 perm &= BLK_PERM_CONSISTENT_READ;
2694 * If the parent can deal with changing data, we're okay with a
2695 * writable and resizable backing file.
2696 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2698 if (shared & BLK_PERM_WRITE) {
2699 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2700 } else {
2701 shared = 0;
2704 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2706 if (bs->open_flags & BDRV_O_INACTIVE) {
2707 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2710 *nperm = perm;
2711 *nshared = shared;
2714 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2715 BdrvChildRole role,
2716 BlockReopenQueue *reopen_queue,
2717 uint64_t perm, uint64_t shared,
2718 uint64_t *nperm, uint64_t *nshared)
2720 int flags;
2722 GLOBAL_STATE_CODE();
2723 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2725 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2728 * Apart from the modifications below, the same permissions are
2729 * forwarded and left alone as for filters
2731 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2732 perm, shared, &perm, &shared);
2734 if (role & BDRV_CHILD_METADATA) {
2735 /* Format drivers may touch metadata even if the guest doesn't write */
2736 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2737 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2741 * bs->file always needs to be consistent because of the
2742 * metadata. We can never allow other users to resize or write
2743 * to it.
2745 if (!(flags & BDRV_O_NO_IO)) {
2746 perm |= BLK_PERM_CONSISTENT_READ;
2748 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2751 if (role & BDRV_CHILD_DATA) {
2753 * Technically, everything in this block is a subset of the
2754 * BDRV_CHILD_METADATA path taken above, and so this could
2755 * be an "else if" branch. However, that is not obvious, and
2756 * this function is not performance critical, therefore we let
2757 * this be an independent "if".
2761 * We cannot allow other users to resize the file because the
2762 * format driver might have some assumptions about the size
2763 * (e.g. because it is stored in metadata, or because the file
2764 * is split into fixed-size data files).
2766 shared &= ~BLK_PERM_RESIZE;
2769 * WRITE_UNCHANGED often cannot be performed as such on the
2770 * data file. For example, the qcow2 driver may still need to
2771 * write copied clusters on copy-on-read.
2773 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2774 perm |= BLK_PERM_WRITE;
2778 * If the data file is written to, the format driver may
2779 * expect to be able to resize it by writing beyond the EOF.
2781 if (perm & BLK_PERM_WRITE) {
2782 perm |= BLK_PERM_RESIZE;
2786 if (bs->open_flags & BDRV_O_INACTIVE) {
2787 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2790 *nperm = perm;
2791 *nshared = shared;
2794 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2795 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2796 uint64_t perm, uint64_t shared,
2797 uint64_t *nperm, uint64_t *nshared)
2799 GLOBAL_STATE_CODE();
2800 if (role & BDRV_CHILD_FILTERED) {
2801 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2802 BDRV_CHILD_COW)));
2803 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2804 perm, shared, nperm, nshared);
2805 } else if (role & BDRV_CHILD_COW) {
2806 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2807 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2808 perm, shared, nperm, nshared);
2809 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2810 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2811 perm, shared, nperm, nshared);
2812 } else {
2813 g_assert_not_reached();
2817 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2819 static const uint64_t permissions[] = {
2820 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2821 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2822 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2823 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2826 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2827 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2829 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2831 return permissions[qapi_perm];
2835 * Replaces the node that a BdrvChild points to without updating permissions.
2837 * If @new_bs is non-NULL, the parent of @child must already be drained through
2838 * @child.
2840 static void bdrv_replace_child_noperm(BdrvChild *child,
2841 BlockDriverState *new_bs)
2843 BlockDriverState *old_bs = child->bs;
2844 int new_bs_quiesce_counter;
2846 assert(!child->frozen);
2849 * If we want to change the BdrvChild to point to a drained node as its new
2850 * child->bs, we need to make sure that its new parent is drained, too. In
2851 * other words, either child->quiesce_parent must already be true or we must
2852 * be able to set it and keep the parent's quiesce_counter consistent with
2853 * that, but without polling or starting new requests (this function
2854 * guarantees that it doesn't poll, and starting new requests would be
2855 * against the invariants of drain sections).
2857 * To keep things simple, we pick the first option (child->quiesce_parent
2858 * must already be true). We also generalise the rule a bit to make it
2859 * easier to verify in callers and more likely to be covered in test cases:
2860 * The parent must be quiesced through this child even if new_bs isn't
2861 * currently drained.
2863 * The only exception is for callers that always pass new_bs == NULL. In
2864 * this case, we obviously never need to consider the case of a drained
2865 * new_bs, so we can keep the callers simpler by allowing them not to drain
2866 * the parent.
2868 assert(!new_bs || child->quiesced_parent);
2869 assert(old_bs != new_bs);
2870 GLOBAL_STATE_CODE();
2872 if (old_bs && new_bs) {
2873 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2876 /* TODO Pull this up into the callers to avoid polling here */
2877 bdrv_graph_wrlock();
2878 if (old_bs) {
2879 if (child->klass->detach) {
2880 child->klass->detach(child);
2882 QLIST_REMOVE(child, next_parent);
2885 child->bs = new_bs;
2887 if (new_bs) {
2888 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2889 if (child->klass->attach) {
2890 child->klass->attach(child);
2893 bdrv_graph_wrunlock();
2896 * If the parent was drained through this BdrvChild previously, but new_bs
2897 * is not drained, allow requests to come in only after the new node has
2898 * been attached.
2900 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2901 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2902 bdrv_parent_drained_end_single(child);
2907 * Free the given @child.
2909 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2910 * unused (i.e. not in a children list).
2912 static void bdrv_child_free(BdrvChild *child)
2914 assert(!child->bs);
2915 GLOBAL_STATE_CODE();
2916 assert(!child->next.le_prev); /* not in children list */
2918 g_free(child->name);
2919 g_free(child);
2922 typedef struct BdrvAttachChildCommonState {
2923 BdrvChild *child;
2924 AioContext *old_parent_ctx;
2925 AioContext *old_child_ctx;
2926 } BdrvAttachChildCommonState;
2928 static void bdrv_attach_child_common_abort(void *opaque)
2930 BdrvAttachChildCommonState *s = opaque;
2931 BlockDriverState *bs = s->child->bs;
2933 GLOBAL_STATE_CODE();
2934 bdrv_replace_child_noperm(s->child, NULL);
2936 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2937 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2940 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2941 Transaction *tran;
2942 GHashTable *visited;
2943 bool ret;
2945 tran = tran_new();
2947 /* No need to visit `child`, because it has been detached already */
2948 visited = g_hash_table_new(NULL, NULL);
2949 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2950 visited, tran, &error_abort);
2951 g_hash_table_destroy(visited);
2953 /* transaction is supposed to always succeed */
2954 assert(ret == true);
2955 tran_commit(tran);
2958 bdrv_unref(bs);
2959 bdrv_child_free(s->child);
2962 static TransactionActionDrv bdrv_attach_child_common_drv = {
2963 .abort = bdrv_attach_child_common_abort,
2964 .clean = g_free,
2968 * Common part of attaching bdrv child to bs or to blk or to job
2970 * Function doesn't update permissions, caller is responsible for this.
2972 * Returns new created child.
2974 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2975 const char *child_name,
2976 const BdrvChildClass *child_class,
2977 BdrvChildRole child_role,
2978 uint64_t perm, uint64_t shared_perm,
2979 void *opaque,
2980 Transaction *tran, Error **errp)
2982 BdrvChild *new_child;
2983 AioContext *parent_ctx;
2984 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2986 assert(child_class->get_parent_desc);
2987 GLOBAL_STATE_CODE();
2989 new_child = g_new(BdrvChild, 1);
2990 *new_child = (BdrvChild) {
2991 .bs = NULL,
2992 .name = g_strdup(child_name),
2993 .klass = child_class,
2994 .role = child_role,
2995 .perm = perm,
2996 .shared_perm = shared_perm,
2997 .opaque = opaque,
3001 * If the AioContexts don't match, first try to move the subtree of
3002 * child_bs into the AioContext of the new parent. If this doesn't work,
3003 * try moving the parent into the AioContext of child_bs instead.
3005 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3006 if (child_ctx != parent_ctx) {
3007 Error *local_err = NULL;
3008 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3009 &local_err);
3011 if (ret < 0 && child_class->change_aio_ctx) {
3012 Transaction *tran = tran_new();
3013 GHashTable *visited = g_hash_table_new(NULL, NULL);
3014 bool ret_child;
3016 g_hash_table_add(visited, new_child);
3017 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3018 visited, tran, NULL);
3019 if (ret_child == true) {
3020 error_free(local_err);
3021 ret = 0;
3023 tran_finalize(tran, ret_child == true ? 0 : -1);
3024 g_hash_table_destroy(visited);
3027 if (ret < 0) {
3028 error_propagate(errp, local_err);
3029 bdrv_child_free(new_child);
3030 return NULL;
3034 bdrv_ref(child_bs);
3036 * Let every new BdrvChild start with a drained parent. Inserting the child
3037 * in the graph with bdrv_replace_child_noperm() will undrain it if
3038 * @child_bs is not drained.
3040 * The child was only just created and is not yet visible in global state
3041 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3042 * could have sent requests and polling is not necessary.
3044 * Note that this means that the parent isn't fully drained yet, we only
3045 * stop new requests from coming in. This is fine, we don't care about the
3046 * old requests here, they are not for this child. If another place enters a
3047 * drain section for the same parent, but wants it to be fully quiesced, it
3048 * will not run most of the the code in .drained_begin() again (which is not
3049 * a problem, we already did this), but it will still poll until the parent
3050 * is fully quiesced, so it will not be negatively affected either.
3052 bdrv_parent_drained_begin_single(new_child);
3053 bdrv_replace_child_noperm(new_child, child_bs);
3055 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3056 *s = (BdrvAttachChildCommonState) {
3057 .child = new_child,
3058 .old_parent_ctx = parent_ctx,
3059 .old_child_ctx = child_ctx,
3061 tran_add(tran, &bdrv_attach_child_common_drv, s);
3063 return new_child;
3067 * Function doesn't update permissions, caller is responsible for this.
3069 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3070 BlockDriverState *child_bs,
3071 const char *child_name,
3072 const BdrvChildClass *child_class,
3073 BdrvChildRole child_role,
3074 Transaction *tran,
3075 Error **errp)
3077 uint64_t perm, shared_perm;
3079 assert(parent_bs->drv);
3080 GLOBAL_STATE_CODE();
3082 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3083 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3084 child_bs->node_name, child_name, parent_bs->node_name);
3085 return NULL;
3088 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3089 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3090 perm, shared_perm, &perm, &shared_perm);
3092 return bdrv_attach_child_common(child_bs, child_name, child_class,
3093 child_role, perm, shared_perm, parent_bs,
3094 tran, errp);
3098 * This function steals the reference to child_bs from the caller.
3099 * That reference is later dropped by bdrv_root_unref_child().
3101 * On failure NULL is returned, errp is set and the reference to
3102 * child_bs is also dropped.
3104 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3105 * (unless @child_bs is already in @ctx).
3107 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3108 const char *child_name,
3109 const BdrvChildClass *child_class,
3110 BdrvChildRole child_role,
3111 uint64_t perm, uint64_t shared_perm,
3112 void *opaque, Error **errp)
3114 int ret;
3115 BdrvChild *child;
3116 Transaction *tran = tran_new();
3118 GLOBAL_STATE_CODE();
3120 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3121 child_role, perm, shared_perm, opaque,
3122 tran, errp);
3123 if (!child) {
3124 ret = -EINVAL;
3125 goto out;
3128 ret = bdrv_refresh_perms(child_bs, tran, errp);
3130 out:
3131 tran_finalize(tran, ret);
3133 bdrv_unref(child_bs);
3135 return ret < 0 ? NULL : child;
3139 * This function transfers the reference to child_bs from the caller
3140 * to parent_bs. That reference is later dropped by parent_bs on
3141 * bdrv_close() or if someone calls bdrv_unref_child().
3143 * On failure NULL is returned, errp is set and the reference to
3144 * child_bs is also dropped.
3146 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3147 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3149 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3150 BlockDriverState *child_bs,
3151 const char *child_name,
3152 const BdrvChildClass *child_class,
3153 BdrvChildRole child_role,
3154 Error **errp)
3156 int ret;
3157 BdrvChild *child;
3158 Transaction *tran = tran_new();
3160 GLOBAL_STATE_CODE();
3162 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3163 child_class, child_role, tran, errp);
3164 if (!child) {
3165 ret = -EINVAL;
3166 goto out;
3169 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3170 if (ret < 0) {
3171 goto out;
3174 out:
3175 tran_finalize(tran, ret);
3177 bdrv_unref(child_bs);
3179 return ret < 0 ? NULL : child;
3182 /* Callers must ensure that child->frozen is false. */
3183 void bdrv_root_unref_child(BdrvChild *child)
3185 BlockDriverState *child_bs = child->bs;
3187 GLOBAL_STATE_CODE();
3188 bdrv_replace_child_noperm(child, NULL);
3189 bdrv_child_free(child);
3191 if (child_bs) {
3193 * Update permissions for old node. We're just taking a parent away, so
3194 * we're loosening restrictions. Errors of permission update are not
3195 * fatal in this case, ignore them.
3197 bdrv_refresh_perms(child_bs, NULL, NULL);
3200 * When the parent requiring a non-default AioContext is removed, the
3201 * node moves back to the main AioContext
3203 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3204 NULL);
3207 bdrv_unref(child_bs);
3210 typedef struct BdrvSetInheritsFrom {
3211 BlockDriverState *bs;
3212 BlockDriverState *old_inherits_from;
3213 } BdrvSetInheritsFrom;
3215 static void bdrv_set_inherits_from_abort(void *opaque)
3217 BdrvSetInheritsFrom *s = opaque;
3219 s->bs->inherits_from = s->old_inherits_from;
3222 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3223 .abort = bdrv_set_inherits_from_abort,
3224 .clean = g_free,
3227 /* @tran is allowed to be NULL. In this case no rollback is possible */
3228 static void bdrv_set_inherits_from(BlockDriverState *bs,
3229 BlockDriverState *new_inherits_from,
3230 Transaction *tran)
3232 if (tran) {
3233 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3235 *s = (BdrvSetInheritsFrom) {
3236 .bs = bs,
3237 .old_inherits_from = bs->inherits_from,
3240 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3243 bs->inherits_from = new_inherits_from;
3247 * Clear all inherits_from pointers from children and grandchildren of
3248 * @root that point to @root, where necessary.
3249 * @tran is allowed to be NULL. In this case no rollback is possible
3251 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3252 Transaction *tran)
3254 BdrvChild *c;
3256 if (child->bs->inherits_from == root) {
3258 * Remove inherits_from only when the last reference between root and
3259 * child->bs goes away.
3261 QLIST_FOREACH(c, &root->children, next) {
3262 if (c != child && c->bs == child->bs) {
3263 break;
3266 if (c == NULL) {
3267 bdrv_set_inherits_from(child->bs, NULL, tran);
3271 QLIST_FOREACH(c, &child->bs->children, next) {
3272 bdrv_unset_inherits_from(root, c, tran);
3276 /* Callers must ensure that child->frozen is false. */
3277 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3279 GLOBAL_STATE_CODE();
3280 if (child == NULL) {
3281 return;
3284 bdrv_unset_inherits_from(parent, child, NULL);
3285 bdrv_root_unref_child(child);
3289 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3291 BdrvChild *c;
3292 GLOBAL_STATE_CODE();
3293 QLIST_FOREACH(c, &bs->parents, next_parent) {
3294 if (c->klass->change_media) {
3295 c->klass->change_media(c, load);
3300 /* Return true if you can reach parent going through child->inherits_from
3301 * recursively. If parent or child are NULL, return false */
3302 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3303 BlockDriverState *parent)
3305 while (child && child != parent) {
3306 child = child->inherits_from;
3309 return child != NULL;
3313 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3314 * mostly used for COW backing children (role = COW), but also for
3315 * filtered children (role = FILTERED | PRIMARY).
3317 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3319 if (bs->drv && bs->drv->is_filter) {
3320 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3321 } else {
3322 return BDRV_CHILD_COW;
3327 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3328 * callers which don't need their own reference any more must call bdrv_unref().
3330 * Function doesn't update permissions, caller is responsible for this.
3332 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3333 BlockDriverState *child_bs,
3334 bool is_backing,
3335 Transaction *tran, Error **errp)
3337 bool update_inherits_from =
3338 bdrv_inherits_from_recursive(child_bs, parent_bs);
3339 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3340 BdrvChildRole role;
3342 GLOBAL_STATE_CODE();
3344 if (!parent_bs->drv) {
3346 * Node without drv is an object without a class :/. TODO: finally fix
3347 * qcow2 driver to never clear bs->drv and implement format corruption
3348 * handling in other way.
3350 error_setg(errp, "Node corrupted");
3351 return -EINVAL;
3354 if (child && child->frozen) {
3355 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3356 child->name, parent_bs->node_name, child->bs->node_name);
3357 return -EPERM;
3360 if (is_backing && !parent_bs->drv->is_filter &&
3361 !parent_bs->drv->supports_backing)
3363 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3364 "files", parent_bs->drv->format_name, parent_bs->node_name);
3365 return -EINVAL;
3368 if (parent_bs->drv->is_filter) {
3369 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3370 } else if (is_backing) {
3371 role = BDRV_CHILD_COW;
3372 } else {
3374 * We only can use same role as it is in existing child. We don't have
3375 * infrastructure to determine role of file child in generic way
3377 if (!child) {
3378 error_setg(errp, "Cannot set file child to format node without "
3379 "file child");
3380 return -EINVAL;
3382 role = child->role;
3385 if (child) {
3386 bdrv_unset_inherits_from(parent_bs, child, tran);
3387 bdrv_remove_child(child, tran);
3390 if (!child_bs) {
3391 goto out;
3394 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3395 is_backing ? "backing" : "file",
3396 &child_of_bds, role,
3397 tran, errp);
3398 if (!child) {
3399 return -EINVAL;
3404 * If inherits_from pointed recursively to bs then let's update it to
3405 * point directly to bs (else it will become NULL).
3407 if (update_inherits_from) {
3408 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3411 out:
3412 bdrv_refresh_limits(parent_bs, tran, NULL);
3414 return 0;
3417 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3418 BlockDriverState *backing_hd,
3419 Transaction *tran, Error **errp)
3421 GLOBAL_STATE_CODE();
3422 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3425 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3426 BlockDriverState *backing_hd,
3427 Error **errp)
3429 int ret;
3430 Transaction *tran = tran_new();
3432 GLOBAL_STATE_CODE();
3433 assert(bs->quiesce_counter > 0);
3435 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3436 if (ret < 0) {
3437 goto out;
3440 ret = bdrv_refresh_perms(bs, tran, errp);
3441 out:
3442 tran_finalize(tran, ret);
3443 return ret;
3446 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3447 Error **errp)
3449 int ret;
3450 GLOBAL_STATE_CODE();
3452 bdrv_drained_begin(bs);
3453 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3454 bdrv_drained_end(bs);
3456 return ret;
3460 * Opens the backing file for a BlockDriverState if not yet open
3462 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3463 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3464 * itself, all options starting with "${bdref_key}." are considered part of the
3465 * BlockdevRef.
3467 * TODO Can this be unified with bdrv_open_image()?
3469 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3470 const char *bdref_key, Error **errp)
3472 char *backing_filename = NULL;
3473 char *bdref_key_dot;
3474 const char *reference = NULL;
3475 int ret = 0;
3476 bool implicit_backing = false;
3477 BlockDriverState *backing_hd;
3478 QDict *options;
3479 QDict *tmp_parent_options = NULL;
3480 Error *local_err = NULL;
3482 GLOBAL_STATE_CODE();
3484 if (bs->backing != NULL) {
3485 goto free_exit;
3488 /* NULL means an empty set of options */
3489 if (parent_options == NULL) {
3490 tmp_parent_options = qdict_new();
3491 parent_options = tmp_parent_options;
3494 bs->open_flags &= ~BDRV_O_NO_BACKING;
3496 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3497 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3498 g_free(bdref_key_dot);
3501 * Caution: while qdict_get_try_str() is fine, getting non-string
3502 * types would require more care. When @parent_options come from
3503 * -blockdev or blockdev_add, its members are typed according to
3504 * the QAPI schema, but when they come from -drive, they're all
3505 * QString.
3507 reference = qdict_get_try_str(parent_options, bdref_key);
3508 if (reference || qdict_haskey(options, "file.filename")) {
3509 /* keep backing_filename NULL */
3510 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3511 qobject_unref(options);
3512 goto free_exit;
3513 } else {
3514 if (qdict_size(options) == 0) {
3515 /* If the user specifies options that do not modify the
3516 * backing file's behavior, we might still consider it the
3517 * implicit backing file. But it's easier this way, and
3518 * just specifying some of the backing BDS's options is
3519 * only possible with -drive anyway (otherwise the QAPI
3520 * schema forces the user to specify everything). */
3521 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3524 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3525 if (local_err) {
3526 ret = -EINVAL;
3527 error_propagate(errp, local_err);
3528 qobject_unref(options);
3529 goto free_exit;
3533 if (!bs->drv || !bs->drv->supports_backing) {
3534 ret = -EINVAL;
3535 error_setg(errp, "Driver doesn't support backing files");
3536 qobject_unref(options);
3537 goto free_exit;
3540 if (!reference &&
3541 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3542 qdict_put_str(options, "driver", bs->backing_format);
3545 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3546 &child_of_bds, bdrv_backing_role(bs), errp);
3547 if (!backing_hd) {
3548 bs->open_flags |= BDRV_O_NO_BACKING;
3549 error_prepend(errp, "Could not open backing file: ");
3550 ret = -EINVAL;
3551 goto free_exit;
3554 if (implicit_backing) {
3555 bdrv_refresh_filename(backing_hd);
3556 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3557 backing_hd->filename);
3560 /* Hook up the backing file link; drop our reference, bs owns the
3561 * backing_hd reference now */
3562 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3563 bdrv_unref(backing_hd);
3564 if (ret < 0) {
3565 goto free_exit;
3568 qdict_del(parent_options, bdref_key);
3570 free_exit:
3571 g_free(backing_filename);
3572 qobject_unref(tmp_parent_options);
3573 return ret;
3576 static BlockDriverState *
3577 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3578 BlockDriverState *parent, const BdrvChildClass *child_class,
3579 BdrvChildRole child_role, bool allow_none, Error **errp)
3581 BlockDriverState *bs = NULL;
3582 QDict *image_options;
3583 char *bdref_key_dot;
3584 const char *reference;
3586 assert(child_class != NULL);
3588 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3589 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3590 g_free(bdref_key_dot);
3593 * Caution: while qdict_get_try_str() is fine, getting non-string
3594 * types would require more care. When @options come from
3595 * -blockdev or blockdev_add, its members are typed according to
3596 * the QAPI schema, but when they come from -drive, they're all
3597 * QString.
3599 reference = qdict_get_try_str(options, bdref_key);
3600 if (!filename && !reference && !qdict_size(image_options)) {
3601 if (!allow_none) {
3602 error_setg(errp, "A block device must be specified for \"%s\"",
3603 bdref_key);
3605 qobject_unref(image_options);
3606 goto done;
3609 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3610 parent, child_class, child_role, errp);
3611 if (!bs) {
3612 goto done;
3615 done:
3616 qdict_del(options, bdref_key);
3617 return bs;
3621 * Opens a disk image whose options are given as BlockdevRef in another block
3622 * device's options.
3624 * If allow_none is true, no image will be opened if filename is false and no
3625 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3627 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3628 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3629 * itself, all options starting with "${bdref_key}." are considered part of the
3630 * BlockdevRef.
3632 * The BlockdevRef will be removed from the options QDict.
3634 BdrvChild *bdrv_open_child(const char *filename,
3635 QDict *options, const char *bdref_key,
3636 BlockDriverState *parent,
3637 const BdrvChildClass *child_class,
3638 BdrvChildRole child_role,
3639 bool allow_none, Error **errp)
3641 BlockDriverState *bs;
3643 GLOBAL_STATE_CODE();
3645 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3646 child_role, allow_none, errp);
3647 if (bs == NULL) {
3648 return NULL;
3651 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3652 errp);
3656 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3658 int bdrv_open_file_child(const char *filename,
3659 QDict *options, const char *bdref_key,
3660 BlockDriverState *parent, Error **errp)
3662 BdrvChildRole role;
3664 /* commit_top and mirror_top don't use this function */
3665 assert(!parent->drv->filtered_child_is_backing);
3666 role = parent->drv->is_filter ?
3667 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3669 if (!bdrv_open_child(filename, options, bdref_key, parent,
3670 &child_of_bds, role, false, errp))
3672 return -EINVAL;
3675 return 0;
3679 * TODO Future callers may need to specify parent/child_class in order for
3680 * option inheritance to work. Existing callers use it for the root node.
3682 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3684 BlockDriverState *bs = NULL;
3685 QObject *obj = NULL;
3686 QDict *qdict = NULL;
3687 const char *reference = NULL;
3688 Visitor *v = NULL;
3690 GLOBAL_STATE_CODE();
3692 if (ref->type == QTYPE_QSTRING) {
3693 reference = ref->u.reference;
3694 } else {
3695 BlockdevOptions *options = &ref->u.definition;
3696 assert(ref->type == QTYPE_QDICT);
3698 v = qobject_output_visitor_new(&obj);
3699 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3700 visit_complete(v, &obj);
3702 qdict = qobject_to(QDict, obj);
3703 qdict_flatten(qdict);
3705 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3706 * compatibility with other callers) rather than what we want as the
3707 * real defaults. Apply the defaults here instead. */
3708 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3709 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3710 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3711 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3715 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3716 obj = NULL;
3717 qobject_unref(obj);
3718 visit_free(v);
3719 return bs;
3722 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3723 int flags,
3724 QDict *snapshot_options,
3725 Error **errp)
3727 g_autofree char *tmp_filename = NULL;
3728 int64_t total_size;
3729 QemuOpts *opts = NULL;
3730 BlockDriverState *bs_snapshot = NULL;
3731 int ret;
3733 GLOBAL_STATE_CODE();
3735 /* if snapshot, we create a temporary backing file and open it
3736 instead of opening 'filename' directly */
3738 /* Get the required size from the image */
3739 total_size = bdrv_getlength(bs);
3740 if (total_size < 0) {
3741 error_setg_errno(errp, -total_size, "Could not get image size");
3742 goto out;
3745 /* Create the temporary image */
3746 tmp_filename = create_tmp_file(errp);
3747 if (!tmp_filename) {
3748 goto out;
3751 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3752 &error_abort);
3753 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3754 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3755 qemu_opts_del(opts);
3756 if (ret < 0) {
3757 error_prepend(errp, "Could not create temporary overlay '%s': ",
3758 tmp_filename);
3759 goto out;
3762 /* Prepare options QDict for the temporary file */
3763 qdict_put_str(snapshot_options, "file.driver", "file");
3764 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3765 qdict_put_str(snapshot_options, "driver", "qcow2");
3767 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3768 snapshot_options = NULL;
3769 if (!bs_snapshot) {
3770 goto out;
3773 ret = bdrv_append(bs_snapshot, bs, errp);
3774 if (ret < 0) {
3775 bs_snapshot = NULL;
3776 goto out;
3779 out:
3780 qobject_unref(snapshot_options);
3781 return bs_snapshot;
3785 * Opens a disk image (raw, qcow2, vmdk, ...)
3787 * options is a QDict of options to pass to the block drivers, or NULL for an
3788 * empty set of options. The reference to the QDict belongs to the block layer
3789 * after the call (even on failure), so if the caller intends to reuse the
3790 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3792 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3793 * If it is not NULL, the referenced BDS will be reused.
3795 * The reference parameter may be used to specify an existing block device which
3796 * should be opened. If specified, neither options nor a filename may be given,
3797 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3799 static BlockDriverState *bdrv_open_inherit(const char *filename,
3800 const char *reference,
3801 QDict *options, int flags,
3802 BlockDriverState *parent,
3803 const BdrvChildClass *child_class,
3804 BdrvChildRole child_role,
3805 Error **errp)
3807 int ret;
3808 BlockBackend *file = NULL;
3809 BlockDriverState *bs;
3810 BlockDriver *drv = NULL;
3811 BdrvChild *child;
3812 const char *drvname;
3813 const char *backing;
3814 Error *local_err = NULL;
3815 QDict *snapshot_options = NULL;
3816 int snapshot_flags = 0;
3818 assert(!child_class || !flags);
3819 assert(!child_class == !parent);
3820 GLOBAL_STATE_CODE();
3822 if (reference) {
3823 bool options_non_empty = options ? qdict_size(options) : false;
3824 qobject_unref(options);
3826 if (filename || options_non_empty) {
3827 error_setg(errp, "Cannot reference an existing block device with "
3828 "additional options or a new filename");
3829 return NULL;
3832 bs = bdrv_lookup_bs(reference, reference, errp);
3833 if (!bs) {
3834 return NULL;
3837 bdrv_ref(bs);
3838 return bs;
3841 bs = bdrv_new();
3843 /* NULL means an empty set of options */
3844 if (options == NULL) {
3845 options = qdict_new();
3848 /* json: syntax counts as explicit options, as if in the QDict */
3849 parse_json_protocol(options, &filename, &local_err);
3850 if (local_err) {
3851 goto fail;
3854 bs->explicit_options = qdict_clone_shallow(options);
3856 if (child_class) {
3857 bool parent_is_format;
3859 if (parent->drv) {
3860 parent_is_format = parent->drv->is_format;
3861 } else {
3863 * parent->drv is not set yet because this node is opened for
3864 * (potential) format probing. That means that @parent is going
3865 * to be a format node.
3867 parent_is_format = true;
3870 bs->inherits_from = parent;
3871 child_class->inherit_options(child_role, parent_is_format,
3872 &flags, options,
3873 parent->open_flags, parent->options);
3876 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3877 if (ret < 0) {
3878 goto fail;
3882 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3883 * Caution: getting a boolean member of @options requires care.
3884 * When @options come from -blockdev or blockdev_add, members are
3885 * typed according to the QAPI schema, but when they come from
3886 * -drive, they're all QString.
3888 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3889 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3890 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3891 } else {
3892 flags &= ~BDRV_O_RDWR;
3895 if (flags & BDRV_O_SNAPSHOT) {
3896 snapshot_options = qdict_new();
3897 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3898 flags, options);
3899 /* Let bdrv_backing_options() override "read-only" */
3900 qdict_del(options, BDRV_OPT_READ_ONLY);
3901 bdrv_inherited_options(BDRV_CHILD_COW, true,
3902 &flags, options, flags, options);
3905 bs->open_flags = flags;
3906 bs->options = options;
3907 options = qdict_clone_shallow(options);
3909 /* Find the right image format driver */
3910 /* See cautionary note on accessing @options above */
3911 drvname = qdict_get_try_str(options, "driver");
3912 if (drvname) {
3913 drv = bdrv_find_format(drvname);
3914 if (!drv) {
3915 error_setg(errp, "Unknown driver: '%s'", drvname);
3916 goto fail;
3920 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3922 /* See cautionary note on accessing @options above */
3923 backing = qdict_get_try_str(options, "backing");
3924 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3925 (backing && *backing == '\0'))
3927 if (backing) {
3928 warn_report("Use of \"backing\": \"\" is deprecated; "
3929 "use \"backing\": null instead");
3931 flags |= BDRV_O_NO_BACKING;
3932 qdict_del(bs->explicit_options, "backing");
3933 qdict_del(bs->options, "backing");
3934 qdict_del(options, "backing");
3937 /* Open image file without format layer. This BlockBackend is only used for
3938 * probing, the block drivers will do their own bdrv_open_child() for the
3939 * same BDS, which is why we put the node name back into options. */
3940 if ((flags & BDRV_O_PROTOCOL) == 0) {
3941 BlockDriverState *file_bs;
3943 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3944 &child_of_bds, BDRV_CHILD_IMAGE,
3945 true, &local_err);
3946 if (local_err) {
3947 goto fail;
3949 if (file_bs != NULL) {
3950 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3951 * looking at the header to guess the image format. This works even
3952 * in cases where a guest would not see a consistent state. */
3953 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3954 blk_insert_bs(file, file_bs, &local_err);
3955 bdrv_unref(file_bs);
3956 if (local_err) {
3957 goto fail;
3960 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3964 /* Image format probing */
3965 bs->probed = !drv;
3966 if (!drv && file) {
3967 ret = find_image_format(file, filename, &drv, &local_err);
3968 if (ret < 0) {
3969 goto fail;
3972 * This option update would logically belong in bdrv_fill_options(),
3973 * but we first need to open bs->file for the probing to work, while
3974 * opening bs->file already requires the (mostly) final set of options
3975 * so that cache mode etc. can be inherited.
3977 * Adding the driver later is somewhat ugly, but it's not an option
3978 * that would ever be inherited, so it's correct. We just need to make
3979 * sure to update both bs->options (which has the full effective
3980 * options for bs) and options (which has file.* already removed).
3982 qdict_put_str(bs->options, "driver", drv->format_name);
3983 qdict_put_str(options, "driver", drv->format_name);
3984 } else if (!drv) {
3985 error_setg(errp, "Must specify either driver or file");
3986 goto fail;
3989 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3990 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3991 /* file must be NULL if a protocol BDS is about to be created
3992 * (the inverse results in an error message from bdrv_open_common()) */
3993 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3995 /* Open the image */
3996 ret = bdrv_open_common(bs, file, options, &local_err);
3997 if (ret < 0) {
3998 goto fail;
4001 if (file) {
4002 blk_unref(file);
4003 file = NULL;
4006 /* If there is a backing file, use it */
4007 if ((flags & BDRV_O_NO_BACKING) == 0) {
4008 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4009 if (ret < 0) {
4010 goto close_and_fail;
4014 /* Remove all children options and references
4015 * from bs->options and bs->explicit_options */
4016 QLIST_FOREACH(child, &bs->children, next) {
4017 char *child_key_dot;
4018 child_key_dot = g_strdup_printf("%s.", child->name);
4019 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4020 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4021 qdict_del(bs->explicit_options, child->name);
4022 qdict_del(bs->options, child->name);
4023 g_free(child_key_dot);
4026 /* Check if any unknown options were used */
4027 if (qdict_size(options) != 0) {
4028 const QDictEntry *entry = qdict_first(options);
4029 if (flags & BDRV_O_PROTOCOL) {
4030 error_setg(errp, "Block protocol '%s' doesn't support the option "
4031 "'%s'", drv->format_name, entry->key);
4032 } else {
4033 error_setg(errp,
4034 "Block format '%s' does not support the option '%s'",
4035 drv->format_name, entry->key);
4038 goto close_and_fail;
4041 bdrv_parent_cb_change_media(bs, true);
4043 qobject_unref(options);
4044 options = NULL;
4046 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4047 * temporary snapshot afterwards. */
4048 if (snapshot_flags) {
4049 BlockDriverState *snapshot_bs;
4050 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4051 snapshot_options, &local_err);
4052 snapshot_options = NULL;
4053 if (local_err) {
4054 goto close_and_fail;
4056 /* We are not going to return bs but the overlay on top of it
4057 * (snapshot_bs); thus, we have to drop the strong reference to bs
4058 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4059 * though, because the overlay still has a reference to it. */
4060 bdrv_unref(bs);
4061 bs = snapshot_bs;
4064 return bs;
4066 fail:
4067 blk_unref(file);
4068 qobject_unref(snapshot_options);
4069 qobject_unref(bs->explicit_options);
4070 qobject_unref(bs->options);
4071 qobject_unref(options);
4072 bs->options = NULL;
4073 bs->explicit_options = NULL;
4074 bdrv_unref(bs);
4075 error_propagate(errp, local_err);
4076 return NULL;
4078 close_and_fail:
4079 bdrv_unref(bs);
4080 qobject_unref(snapshot_options);
4081 qobject_unref(options);
4082 error_propagate(errp, local_err);
4083 return NULL;
4086 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4087 QDict *options, int flags, Error **errp)
4089 GLOBAL_STATE_CODE();
4091 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4092 NULL, 0, errp);
4095 /* Return true if the NULL-terminated @list contains @str */
4096 static bool is_str_in_list(const char *str, const char *const *list)
4098 if (str && list) {
4099 int i;
4100 for (i = 0; list[i] != NULL; i++) {
4101 if (!strcmp(str, list[i])) {
4102 return true;
4106 return false;
4110 * Check that every option set in @bs->options is also set in
4111 * @new_opts.
4113 * Options listed in the common_options list and in
4114 * @bs->drv->mutable_opts are skipped.
4116 * Return 0 on success, otherwise return -EINVAL and set @errp.
4118 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4119 const QDict *new_opts, Error **errp)
4121 const QDictEntry *e;
4122 /* These options are common to all block drivers and are handled
4123 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4124 const char *const common_options[] = {
4125 "node-name", "discard", "cache.direct", "cache.no-flush",
4126 "read-only", "auto-read-only", "detect-zeroes", NULL
4129 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4130 if (!qdict_haskey(new_opts, e->key) &&
4131 !is_str_in_list(e->key, common_options) &&
4132 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4133 error_setg(errp, "Option '%s' cannot be reset "
4134 "to its default value", e->key);
4135 return -EINVAL;
4139 return 0;
4143 * Returns true if @child can be reached recursively from @bs
4145 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4146 BlockDriverState *child)
4148 BdrvChild *c;
4150 if (bs == child) {
4151 return true;
4154 QLIST_FOREACH(c, &bs->children, next) {
4155 if (bdrv_recurse_has_child(c->bs, child)) {
4156 return true;
4160 return false;
4164 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4165 * reopen of multiple devices.
4167 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4168 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4169 * be created and initialized. This newly created BlockReopenQueue should be
4170 * passed back in for subsequent calls that are intended to be of the same
4171 * atomic 'set'.
4173 * bs is the BlockDriverState to add to the reopen queue.
4175 * options contains the changed options for the associated bs
4176 * (the BlockReopenQueue takes ownership)
4178 * flags contains the open flags for the associated bs
4180 * returns a pointer to bs_queue, which is either the newly allocated
4181 * bs_queue, or the existing bs_queue being used.
4183 * bs is drained here and undrained by bdrv_reopen_queue_free().
4185 * To be called with bs->aio_context locked.
4187 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4188 BlockDriverState *bs,
4189 QDict *options,
4190 const BdrvChildClass *klass,
4191 BdrvChildRole role,
4192 bool parent_is_format,
4193 QDict *parent_options,
4194 int parent_flags,
4195 bool keep_old_opts)
4197 assert(bs != NULL);
4199 BlockReopenQueueEntry *bs_entry;
4200 BdrvChild *child;
4201 QDict *old_options, *explicit_options, *options_copy;
4202 int flags;
4203 QemuOpts *opts;
4205 GLOBAL_STATE_CODE();
4207 bdrv_drained_begin(bs);
4209 if (bs_queue == NULL) {
4210 bs_queue = g_new0(BlockReopenQueue, 1);
4211 QTAILQ_INIT(bs_queue);
4214 if (!options) {
4215 options = qdict_new();
4218 /* Check if this BlockDriverState is already in the queue */
4219 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4220 if (bs == bs_entry->state.bs) {
4221 break;
4226 * Precedence of options:
4227 * 1. Explicitly passed in options (highest)
4228 * 2. Retained from explicitly set options of bs
4229 * 3. Inherited from parent node
4230 * 4. Retained from effective options of bs
4233 /* Old explicitly set values (don't overwrite by inherited value) */
4234 if (bs_entry || keep_old_opts) {
4235 old_options = qdict_clone_shallow(bs_entry ?
4236 bs_entry->state.explicit_options :
4237 bs->explicit_options);
4238 bdrv_join_options(bs, options, old_options);
4239 qobject_unref(old_options);
4242 explicit_options = qdict_clone_shallow(options);
4244 /* Inherit from parent node */
4245 if (parent_options) {
4246 flags = 0;
4247 klass->inherit_options(role, parent_is_format, &flags, options,
4248 parent_flags, parent_options);
4249 } else {
4250 flags = bdrv_get_flags(bs);
4253 if (keep_old_opts) {
4254 /* Old values are used for options that aren't set yet */
4255 old_options = qdict_clone_shallow(bs->options);
4256 bdrv_join_options(bs, options, old_options);
4257 qobject_unref(old_options);
4260 /* We have the final set of options so let's update the flags */
4261 options_copy = qdict_clone_shallow(options);
4262 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4263 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4264 update_flags_from_options(&flags, opts);
4265 qemu_opts_del(opts);
4266 qobject_unref(options_copy);
4268 /* bdrv_open_inherit() sets and clears some additional flags internally */
4269 flags &= ~BDRV_O_PROTOCOL;
4270 if (flags & BDRV_O_RDWR) {
4271 flags |= BDRV_O_ALLOW_RDWR;
4274 if (!bs_entry) {
4275 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4276 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4277 } else {
4278 qobject_unref(bs_entry->state.options);
4279 qobject_unref(bs_entry->state.explicit_options);
4282 bs_entry->state.bs = bs;
4283 bs_entry->state.options = options;
4284 bs_entry->state.explicit_options = explicit_options;
4285 bs_entry->state.flags = flags;
4288 * If keep_old_opts is false then it means that unspecified
4289 * options must be reset to their original value. We don't allow
4290 * resetting 'backing' but we need to know if the option is
4291 * missing in order to decide if we have to return an error.
4293 if (!keep_old_opts) {
4294 bs_entry->state.backing_missing =
4295 !qdict_haskey(options, "backing") &&
4296 !qdict_haskey(options, "backing.driver");
4299 QLIST_FOREACH(child, &bs->children, next) {
4300 QDict *new_child_options = NULL;
4301 bool child_keep_old = keep_old_opts;
4303 /* reopen can only change the options of block devices that were
4304 * implicitly created and inherited options. For other (referenced)
4305 * block devices, a syntax like "backing.foo" results in an error. */
4306 if (child->bs->inherits_from != bs) {
4307 continue;
4310 /* Check if the options contain a child reference */
4311 if (qdict_haskey(options, child->name)) {
4312 const char *childref = qdict_get_try_str(options, child->name);
4314 * The current child must not be reopened if the child
4315 * reference is null or points to a different node.
4317 if (g_strcmp0(childref, child->bs->node_name)) {
4318 continue;
4321 * If the child reference points to the current child then
4322 * reopen it with its existing set of options (note that
4323 * it can still inherit new options from the parent).
4325 child_keep_old = true;
4326 } else {
4327 /* Extract child options ("child-name.*") */
4328 char *child_key_dot = g_strdup_printf("%s.", child->name);
4329 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4330 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4331 g_free(child_key_dot);
4334 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4335 child->klass, child->role, bs->drv->is_format,
4336 options, flags, child_keep_old);
4339 return bs_queue;
4342 /* To be called with bs->aio_context locked */
4343 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4344 BlockDriverState *bs,
4345 QDict *options, bool keep_old_opts)
4347 GLOBAL_STATE_CODE();
4349 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4350 NULL, 0, keep_old_opts);
4353 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4355 GLOBAL_STATE_CODE();
4356 if (bs_queue) {
4357 BlockReopenQueueEntry *bs_entry, *next;
4358 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4359 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4361 aio_context_acquire(ctx);
4362 bdrv_drained_end(bs_entry->state.bs);
4363 aio_context_release(ctx);
4365 qobject_unref(bs_entry->state.explicit_options);
4366 qobject_unref(bs_entry->state.options);
4367 g_free(bs_entry);
4369 g_free(bs_queue);
4374 * Reopen multiple BlockDriverStates atomically & transactionally.
4376 * The queue passed in (bs_queue) must have been built up previous
4377 * via bdrv_reopen_queue().
4379 * Reopens all BDS specified in the queue, with the appropriate
4380 * flags. All devices are prepared for reopen, and failure of any
4381 * device will cause all device changes to be abandoned, and intermediate
4382 * data cleaned up.
4384 * If all devices prepare successfully, then the changes are committed
4385 * to all devices.
4387 * All affected nodes must be drained between bdrv_reopen_queue() and
4388 * bdrv_reopen_multiple().
4390 * To be called from the main thread, with all other AioContexts unlocked.
4392 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4394 int ret = -1;
4395 BlockReopenQueueEntry *bs_entry, *next;
4396 AioContext *ctx;
4397 Transaction *tran = tran_new();
4398 g_autoptr(GSList) refresh_list = NULL;
4400 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4401 assert(bs_queue != NULL);
4402 GLOBAL_STATE_CODE();
4404 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4405 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4406 aio_context_acquire(ctx);
4407 ret = bdrv_flush(bs_entry->state.bs);
4408 aio_context_release(ctx);
4409 if (ret < 0) {
4410 error_setg_errno(errp, -ret, "Error flushing drive");
4411 goto abort;
4415 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4416 assert(bs_entry->state.bs->quiesce_counter > 0);
4417 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4418 aio_context_acquire(ctx);
4419 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4420 aio_context_release(ctx);
4421 if (ret < 0) {
4422 goto abort;
4424 bs_entry->prepared = true;
4427 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4428 BDRVReopenState *state = &bs_entry->state;
4430 refresh_list = g_slist_prepend(refresh_list, state->bs);
4431 if (state->old_backing_bs) {
4432 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4434 if (state->old_file_bs) {
4435 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4440 * Note that file-posix driver rely on permission update done during reopen
4441 * (even if no permission changed), because it wants "new" permissions for
4442 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4443 * in raw_reopen_prepare() which is called with "old" permissions.
4445 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4446 if (ret < 0) {
4447 goto abort;
4451 * If we reach this point, we have success and just need to apply the
4452 * changes.
4454 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4455 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4456 * children are usually goes after parents in reopen-queue, so go from last
4457 * to first element.
4459 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4460 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4461 aio_context_acquire(ctx);
4462 bdrv_reopen_commit(&bs_entry->state);
4463 aio_context_release(ctx);
4466 tran_commit(tran);
4468 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4469 BlockDriverState *bs = bs_entry->state.bs;
4471 if (bs->drv->bdrv_reopen_commit_post) {
4472 ctx = bdrv_get_aio_context(bs);
4473 aio_context_acquire(ctx);
4474 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4475 aio_context_release(ctx);
4479 ret = 0;
4480 goto cleanup;
4482 abort:
4483 tran_abort(tran);
4484 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4485 if (bs_entry->prepared) {
4486 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4487 aio_context_acquire(ctx);
4488 bdrv_reopen_abort(&bs_entry->state);
4489 aio_context_release(ctx);
4493 cleanup:
4494 bdrv_reopen_queue_free(bs_queue);
4496 return ret;
4499 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4500 Error **errp)
4502 AioContext *ctx = bdrv_get_aio_context(bs);
4503 BlockReopenQueue *queue;
4504 int ret;
4506 GLOBAL_STATE_CODE();
4508 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4510 if (ctx != qemu_get_aio_context()) {
4511 aio_context_release(ctx);
4513 ret = bdrv_reopen_multiple(queue, errp);
4515 if (ctx != qemu_get_aio_context()) {
4516 aio_context_acquire(ctx);
4519 return ret;
4522 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4523 Error **errp)
4525 QDict *opts = qdict_new();
4527 GLOBAL_STATE_CODE();
4529 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4531 return bdrv_reopen(bs, opts, true, errp);
4535 * Take a BDRVReopenState and check if the value of 'backing' in the
4536 * reopen_state->options QDict is valid or not.
4538 * If 'backing' is missing from the QDict then return 0.
4540 * If 'backing' contains the node name of the backing file of
4541 * reopen_state->bs then return 0.
4543 * If 'backing' contains a different node name (or is null) then check
4544 * whether the current backing file can be replaced with the new one.
4545 * If that's the case then reopen_state->replace_backing_bs is set to
4546 * true and reopen_state->new_backing_bs contains a pointer to the new
4547 * backing BlockDriverState (or NULL).
4549 * Return 0 on success, otherwise return < 0 and set @errp.
4551 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4552 bool is_backing, Transaction *tran,
4553 Error **errp)
4555 BlockDriverState *bs = reopen_state->bs;
4556 BlockDriverState *new_child_bs;
4557 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4558 child_bs(bs->file);
4559 const char *child_name = is_backing ? "backing" : "file";
4560 QObject *value;
4561 const char *str;
4563 GLOBAL_STATE_CODE();
4565 value = qdict_get(reopen_state->options, child_name);
4566 if (value == NULL) {
4567 return 0;
4570 switch (qobject_type(value)) {
4571 case QTYPE_QNULL:
4572 assert(is_backing); /* The 'file' option does not allow a null value */
4573 new_child_bs = NULL;
4574 break;
4575 case QTYPE_QSTRING:
4576 str = qstring_get_str(qobject_to(QString, value));
4577 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4578 if (new_child_bs == NULL) {
4579 return -EINVAL;
4580 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4581 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4582 "cycle", str, child_name, bs->node_name);
4583 return -EINVAL;
4585 break;
4586 default:
4588 * The options QDict has been flattened, so 'backing' and 'file'
4589 * do not allow any other data type here.
4591 g_assert_not_reached();
4594 if (old_child_bs == new_child_bs) {
4595 return 0;
4598 if (old_child_bs) {
4599 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4600 return 0;
4603 if (old_child_bs->implicit) {
4604 error_setg(errp, "Cannot replace implicit %s child of %s",
4605 child_name, bs->node_name);
4606 return -EPERM;
4610 if (bs->drv->is_filter && !old_child_bs) {
4612 * Filters always have a file or a backing child, so we are trying to
4613 * change wrong child
4615 error_setg(errp, "'%s' is a %s filter node that does not support a "
4616 "%s child", bs->node_name, bs->drv->format_name, child_name);
4617 return -EINVAL;
4620 if (is_backing) {
4621 reopen_state->old_backing_bs = old_child_bs;
4622 } else {
4623 reopen_state->old_file_bs = old_child_bs;
4626 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4627 tran, errp);
4631 * Prepares a BlockDriverState for reopen. All changes are staged in the
4632 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4633 * the block driver layer .bdrv_reopen_prepare()
4635 * bs is the BlockDriverState to reopen
4636 * flags are the new open flags
4637 * queue is the reopen queue
4639 * Returns 0 on success, non-zero on error. On error errp will be set
4640 * as well.
4642 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4643 * It is the responsibility of the caller to then call the abort() or
4644 * commit() for any other BDS that have been left in a prepare() state
4647 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4648 BlockReopenQueue *queue,
4649 Transaction *change_child_tran, Error **errp)
4651 int ret = -1;
4652 int old_flags;
4653 Error *local_err = NULL;
4654 BlockDriver *drv;
4655 QemuOpts *opts;
4656 QDict *orig_reopen_opts;
4657 char *discard = NULL;
4658 bool read_only;
4659 bool drv_prepared = false;
4661 assert(reopen_state != NULL);
4662 assert(reopen_state->bs->drv != NULL);
4663 GLOBAL_STATE_CODE();
4664 drv = reopen_state->bs->drv;
4666 /* This function and each driver's bdrv_reopen_prepare() remove
4667 * entries from reopen_state->options as they are processed, so
4668 * we need to make a copy of the original QDict. */
4669 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4671 /* Process generic block layer options */
4672 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4673 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4674 ret = -EINVAL;
4675 goto error;
4678 /* This was already called in bdrv_reopen_queue_child() so the flags
4679 * are up-to-date. This time we simply want to remove the options from
4680 * QemuOpts in order to indicate that they have been processed. */
4681 old_flags = reopen_state->flags;
4682 update_flags_from_options(&reopen_state->flags, opts);
4683 assert(old_flags == reopen_state->flags);
4685 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4686 if (discard != NULL) {
4687 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4688 error_setg(errp, "Invalid discard option");
4689 ret = -EINVAL;
4690 goto error;
4694 reopen_state->detect_zeroes =
4695 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4696 if (local_err) {
4697 error_propagate(errp, local_err);
4698 ret = -EINVAL;
4699 goto error;
4702 /* All other options (including node-name and driver) must be unchanged.
4703 * Put them back into the QDict, so that they are checked at the end
4704 * of this function. */
4705 qemu_opts_to_qdict(opts, reopen_state->options);
4707 /* If we are to stay read-only, do not allow permission change
4708 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4709 * not set, or if the BDS still has copy_on_read enabled */
4710 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4711 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4712 if (local_err) {
4713 error_propagate(errp, local_err);
4714 goto error;
4717 if (drv->bdrv_reopen_prepare) {
4719 * If a driver-specific option is missing, it means that we
4720 * should reset it to its default value.
4721 * But not all options allow that, so we need to check it first.
4723 ret = bdrv_reset_options_allowed(reopen_state->bs,
4724 reopen_state->options, errp);
4725 if (ret) {
4726 goto error;
4729 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4730 if (ret) {
4731 if (local_err != NULL) {
4732 error_propagate(errp, local_err);
4733 } else {
4734 bdrv_refresh_filename(reopen_state->bs);
4735 error_setg(errp, "failed while preparing to reopen image '%s'",
4736 reopen_state->bs->filename);
4738 goto error;
4740 } else {
4741 /* It is currently mandatory to have a bdrv_reopen_prepare()
4742 * handler for each supported drv. */
4743 error_setg(errp, "Block format '%s' used by node '%s' "
4744 "does not support reopening files", drv->format_name,
4745 bdrv_get_device_or_node_name(reopen_state->bs));
4746 ret = -1;
4747 goto error;
4750 drv_prepared = true;
4753 * We must provide the 'backing' option if the BDS has a backing
4754 * file or if the image file has a backing file name as part of
4755 * its metadata. Otherwise the 'backing' option can be omitted.
4757 if (drv->supports_backing && reopen_state->backing_missing &&
4758 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4759 error_setg(errp, "backing is missing for '%s'",
4760 reopen_state->bs->node_name);
4761 ret = -EINVAL;
4762 goto error;
4766 * Allow changing the 'backing' option. The new value can be
4767 * either a reference to an existing node (using its node name)
4768 * or NULL to simply detach the current backing file.
4770 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4771 change_child_tran, errp);
4772 if (ret < 0) {
4773 goto error;
4775 qdict_del(reopen_state->options, "backing");
4777 /* Allow changing the 'file' option. In this case NULL is not allowed */
4778 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4779 change_child_tran, errp);
4780 if (ret < 0) {
4781 goto error;
4783 qdict_del(reopen_state->options, "file");
4785 /* Options that are not handled are only okay if they are unchanged
4786 * compared to the old state. It is expected that some options are only
4787 * used for the initial open, but not reopen (e.g. filename) */
4788 if (qdict_size(reopen_state->options)) {
4789 const QDictEntry *entry = qdict_first(reopen_state->options);
4791 do {
4792 QObject *new = entry->value;
4793 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4795 /* Allow child references (child_name=node_name) as long as they
4796 * point to the current child (i.e. everything stays the same). */
4797 if (qobject_type(new) == QTYPE_QSTRING) {
4798 BdrvChild *child;
4799 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4800 if (!strcmp(child->name, entry->key)) {
4801 break;
4805 if (child) {
4806 if (!strcmp(child->bs->node_name,
4807 qstring_get_str(qobject_to(QString, new)))) {
4808 continue; /* Found child with this name, skip option */
4814 * TODO: When using -drive to specify blockdev options, all values
4815 * will be strings; however, when using -blockdev, blockdev-add or
4816 * filenames using the json:{} pseudo-protocol, they will be
4817 * correctly typed.
4818 * In contrast, reopening options are (currently) always strings
4819 * (because you can only specify them through qemu-io; all other
4820 * callers do not specify any options).
4821 * Therefore, when using anything other than -drive to create a BDS,
4822 * this cannot detect non-string options as unchanged, because
4823 * qobject_is_equal() always returns false for objects of different
4824 * type. In the future, this should be remedied by correctly typing
4825 * all options. For now, this is not too big of an issue because
4826 * the user can simply omit options which cannot be changed anyway,
4827 * so they will stay unchanged.
4829 if (!qobject_is_equal(new, old)) {
4830 error_setg(errp, "Cannot change the option '%s'", entry->key);
4831 ret = -EINVAL;
4832 goto error;
4834 } while ((entry = qdict_next(reopen_state->options, entry)));
4837 ret = 0;
4839 /* Restore the original reopen_state->options QDict */
4840 qobject_unref(reopen_state->options);
4841 reopen_state->options = qobject_ref(orig_reopen_opts);
4843 error:
4844 if (ret < 0 && drv_prepared) {
4845 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4846 * call drv->bdrv_reopen_abort() before signaling an error
4847 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4848 * when the respective bdrv_reopen_prepare() has failed) */
4849 if (drv->bdrv_reopen_abort) {
4850 drv->bdrv_reopen_abort(reopen_state);
4853 qemu_opts_del(opts);
4854 qobject_unref(orig_reopen_opts);
4855 g_free(discard);
4856 return ret;
4860 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4861 * makes them final by swapping the staging BlockDriverState contents into
4862 * the active BlockDriverState contents.
4864 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4866 BlockDriver *drv;
4867 BlockDriverState *bs;
4868 BdrvChild *child;
4870 assert(reopen_state != NULL);
4871 bs = reopen_state->bs;
4872 drv = bs->drv;
4873 assert(drv != NULL);
4874 GLOBAL_STATE_CODE();
4876 /* If there are any driver level actions to take */
4877 if (drv->bdrv_reopen_commit) {
4878 drv->bdrv_reopen_commit(reopen_state);
4881 /* set BDS specific flags now */
4882 qobject_unref(bs->explicit_options);
4883 qobject_unref(bs->options);
4884 qobject_ref(reopen_state->explicit_options);
4885 qobject_ref(reopen_state->options);
4887 bs->explicit_options = reopen_state->explicit_options;
4888 bs->options = reopen_state->options;
4889 bs->open_flags = reopen_state->flags;
4890 bs->detect_zeroes = reopen_state->detect_zeroes;
4892 /* Remove child references from bs->options and bs->explicit_options.
4893 * Child options were already removed in bdrv_reopen_queue_child() */
4894 QLIST_FOREACH(child, &bs->children, next) {
4895 qdict_del(bs->explicit_options, child->name);
4896 qdict_del(bs->options, child->name);
4898 /* backing is probably removed, so it's not handled by previous loop */
4899 qdict_del(bs->explicit_options, "backing");
4900 qdict_del(bs->options, "backing");
4902 bdrv_refresh_limits(bs, NULL, NULL);
4906 * Abort the reopen, and delete and free the staged changes in
4907 * reopen_state
4909 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4911 BlockDriver *drv;
4913 assert(reopen_state != NULL);
4914 drv = reopen_state->bs->drv;
4915 assert(drv != NULL);
4916 GLOBAL_STATE_CODE();
4918 if (drv->bdrv_reopen_abort) {
4919 drv->bdrv_reopen_abort(reopen_state);
4924 static void bdrv_close(BlockDriverState *bs)
4926 BdrvAioNotifier *ban, *ban_next;
4927 BdrvChild *child, *next;
4929 GLOBAL_STATE_CODE();
4930 assert(!bs->refcnt);
4932 bdrv_drained_begin(bs); /* complete I/O */
4933 bdrv_flush(bs);
4934 bdrv_drain(bs); /* in case flush left pending I/O */
4936 if (bs->drv) {
4937 if (bs->drv->bdrv_close) {
4938 /* Must unfreeze all children, so bdrv_unref_child() works */
4939 bs->drv->bdrv_close(bs);
4941 bs->drv = NULL;
4944 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4945 bdrv_unref_child(bs, child);
4948 assert(!bs->backing);
4949 assert(!bs->file);
4950 g_free(bs->opaque);
4951 bs->opaque = NULL;
4952 qatomic_set(&bs->copy_on_read, 0);
4953 bs->backing_file[0] = '\0';
4954 bs->backing_format[0] = '\0';
4955 bs->total_sectors = 0;
4956 bs->encrypted = false;
4957 bs->sg = false;
4958 qobject_unref(bs->options);
4959 qobject_unref(bs->explicit_options);
4960 bs->options = NULL;
4961 bs->explicit_options = NULL;
4962 qobject_unref(bs->full_open_options);
4963 bs->full_open_options = NULL;
4964 g_free(bs->block_status_cache);
4965 bs->block_status_cache = NULL;
4967 bdrv_release_named_dirty_bitmaps(bs);
4968 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4970 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4971 g_free(ban);
4973 QLIST_INIT(&bs->aio_notifiers);
4974 bdrv_drained_end(bs);
4977 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4978 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4979 * gets called.
4981 if (bs->quiesce_counter) {
4982 bdrv_drain_all_end_quiesce(bs);
4986 void bdrv_close_all(void)
4988 GLOBAL_STATE_CODE();
4989 assert(job_next(NULL) == NULL);
4991 /* Drop references from requests still in flight, such as canceled block
4992 * jobs whose AIO context has not been polled yet */
4993 bdrv_drain_all();
4995 blk_remove_all_bs();
4996 blockdev_close_all_bdrv_states();
4998 assert(QTAILQ_EMPTY(&all_bdrv_states));
5001 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
5003 GQueue *queue;
5004 GHashTable *found;
5005 bool ret;
5007 if (c->klass->stay_at_node) {
5008 return false;
5011 /* If the child @c belongs to the BDS @to, replacing the current
5012 * c->bs by @to would mean to create a loop.
5014 * Such a case occurs when appending a BDS to a backing chain.
5015 * For instance, imagine the following chain:
5017 * guest device -> node A -> further backing chain...
5019 * Now we create a new BDS B which we want to put on top of this
5020 * chain, so we first attach A as its backing node:
5022 * node B
5025 * guest device -> node A -> further backing chain...
5027 * Finally we want to replace A by B. When doing that, we want to
5028 * replace all pointers to A by pointers to B -- except for the
5029 * pointer from B because (1) that would create a loop, and (2)
5030 * that pointer should simply stay intact:
5032 * guest device -> node B
5035 * node A -> further backing chain...
5037 * In general, when replacing a node A (c->bs) by a node B (@to),
5038 * if A is a child of B, that means we cannot replace A by B there
5039 * because that would create a loop. Silently detaching A from B
5040 * is also not really an option. So overall just leaving A in
5041 * place there is the most sensible choice.
5043 * We would also create a loop in any cases where @c is only
5044 * indirectly referenced by @to. Prevent this by returning false
5045 * if @c is found (by breadth-first search) anywhere in the whole
5046 * subtree of @to.
5049 ret = true;
5050 found = g_hash_table_new(NULL, NULL);
5051 g_hash_table_add(found, to);
5052 queue = g_queue_new();
5053 g_queue_push_tail(queue, to);
5055 while (!g_queue_is_empty(queue)) {
5056 BlockDriverState *v = g_queue_pop_head(queue);
5057 BdrvChild *c2;
5059 QLIST_FOREACH(c2, &v->children, next) {
5060 if (c2 == c) {
5061 ret = false;
5062 break;
5065 if (g_hash_table_contains(found, c2->bs)) {
5066 continue;
5069 g_queue_push_tail(queue, c2->bs);
5070 g_hash_table_add(found, c2->bs);
5074 g_queue_free(queue);
5075 g_hash_table_destroy(found);
5077 return ret;
5080 static void bdrv_remove_child_commit(void *opaque)
5082 GLOBAL_STATE_CODE();
5083 bdrv_child_free(opaque);
5086 static TransactionActionDrv bdrv_remove_child_drv = {
5087 .commit = bdrv_remove_child_commit,
5090 /* Function doesn't update permissions, caller is responsible for this. */
5091 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5093 if (!child) {
5094 return;
5097 if (child->bs) {
5098 BlockDriverState *bs = child->bs;
5099 bdrv_drained_begin(bs);
5100 bdrv_replace_child_tran(child, NULL, tran);
5101 bdrv_drained_end(bs);
5104 tran_add(tran, &bdrv_remove_child_drv, child);
5107 static void undrain_on_clean_cb(void *opaque)
5109 bdrv_drained_end(opaque);
5112 static TransactionActionDrv undrain_on_clean = {
5113 .clean = undrain_on_clean_cb,
5116 static int bdrv_replace_node_noperm(BlockDriverState *from,
5117 BlockDriverState *to,
5118 bool auto_skip, Transaction *tran,
5119 Error **errp)
5121 BdrvChild *c, *next;
5123 GLOBAL_STATE_CODE();
5125 bdrv_drained_begin(from);
5126 bdrv_drained_begin(to);
5127 tran_add(tran, &undrain_on_clean, from);
5128 tran_add(tran, &undrain_on_clean, to);
5130 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5131 assert(c->bs == from);
5132 if (!should_update_child(c, to)) {
5133 if (auto_skip) {
5134 continue;
5136 error_setg(errp, "Should not change '%s' link to '%s'",
5137 c->name, from->node_name);
5138 return -EINVAL;
5140 if (c->frozen) {
5141 error_setg(errp, "Cannot change '%s' link to '%s'",
5142 c->name, from->node_name);
5143 return -EPERM;
5145 bdrv_replace_child_tran(c, to, tran);
5148 return 0;
5152 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5153 * if it creates a parent-child relation loop or if parent is block-job.
5155 * With auto_skip=false the error is returned if from has a parent which should
5156 * not be updated.
5158 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5159 * case backing link of the cow-parent of @to is removed.
5161 static int bdrv_replace_node_common(BlockDriverState *from,
5162 BlockDriverState *to,
5163 bool auto_skip, bool detach_subchain,
5164 Error **errp)
5166 Transaction *tran = tran_new();
5167 g_autoptr(GSList) refresh_list = NULL;
5168 BlockDriverState *to_cow_parent = NULL;
5169 int ret;
5171 GLOBAL_STATE_CODE();
5173 if (detach_subchain) {
5174 assert(bdrv_chain_contains(from, to));
5175 assert(from != to);
5176 for (to_cow_parent = from;
5177 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5178 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5184 /* Make sure that @from doesn't go away until we have successfully attached
5185 * all of its parents to @to. */
5186 bdrv_ref(from);
5188 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5189 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5190 bdrv_drained_begin(from);
5193 * Do the replacement without permission update.
5194 * Replacement may influence the permissions, we should calculate new
5195 * permissions based on new graph. If we fail, we'll roll-back the
5196 * replacement.
5198 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5199 if (ret < 0) {
5200 goto out;
5203 if (detach_subchain) {
5204 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5207 refresh_list = g_slist_prepend(refresh_list, to);
5208 refresh_list = g_slist_prepend(refresh_list, from);
5210 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5211 if (ret < 0) {
5212 goto out;
5215 ret = 0;
5217 out:
5218 tran_finalize(tran, ret);
5220 bdrv_drained_end(from);
5221 bdrv_unref(from);
5223 return ret;
5226 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5227 Error **errp)
5229 GLOBAL_STATE_CODE();
5231 return bdrv_replace_node_common(from, to, true, false, errp);
5234 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5236 GLOBAL_STATE_CODE();
5238 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5239 errp);
5243 * Add new bs contents at the top of an image chain while the chain is
5244 * live, while keeping required fields on the top layer.
5246 * This will modify the BlockDriverState fields, and swap contents
5247 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5249 * bs_new must not be attached to a BlockBackend and must not have backing
5250 * child.
5252 * This function does not create any image files.
5254 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5255 Error **errp)
5257 int ret;
5258 BdrvChild *child;
5259 Transaction *tran = tran_new();
5261 GLOBAL_STATE_CODE();
5263 assert(!bs_new->backing);
5265 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5266 &child_of_bds, bdrv_backing_role(bs_new),
5267 tran, errp);
5268 if (!child) {
5269 ret = -EINVAL;
5270 goto out;
5273 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5274 if (ret < 0) {
5275 goto out;
5278 ret = bdrv_refresh_perms(bs_new, tran, errp);
5279 out:
5280 tran_finalize(tran, ret);
5282 bdrv_refresh_limits(bs_top, NULL, NULL);
5284 return ret;
5287 /* Not for empty child */
5288 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5289 Error **errp)
5291 int ret;
5292 Transaction *tran = tran_new();
5293 g_autoptr(GSList) refresh_list = NULL;
5294 BlockDriverState *old_bs = child->bs;
5296 GLOBAL_STATE_CODE();
5298 bdrv_ref(old_bs);
5299 bdrv_drained_begin(old_bs);
5300 bdrv_drained_begin(new_bs);
5302 bdrv_replace_child_tran(child, new_bs, tran);
5304 refresh_list = g_slist_prepend(refresh_list, old_bs);
5305 refresh_list = g_slist_prepend(refresh_list, new_bs);
5307 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5309 tran_finalize(tran, ret);
5311 bdrv_drained_end(old_bs);
5312 bdrv_drained_end(new_bs);
5313 bdrv_unref(old_bs);
5315 return ret;
5318 static void bdrv_delete(BlockDriverState *bs)
5320 assert(bdrv_op_blocker_is_empty(bs));
5321 assert(!bs->refcnt);
5322 GLOBAL_STATE_CODE();
5324 /* remove from list, if necessary */
5325 if (bs->node_name[0] != '\0') {
5326 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5328 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5330 bdrv_close(bs);
5332 g_free(bs);
5337 * Replace @bs by newly created block node.
5339 * @options is a QDict of options to pass to the block drivers, or NULL for an
5340 * empty set of options. The reference to the QDict belongs to the block layer
5341 * after the call (even on failure), so if the caller intends to reuse the
5342 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5344 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5345 int flags, Error **errp)
5347 ERRP_GUARD();
5348 int ret;
5349 BlockDriverState *new_node_bs = NULL;
5350 const char *drvname, *node_name;
5351 BlockDriver *drv;
5353 drvname = qdict_get_try_str(options, "driver");
5354 if (!drvname) {
5355 error_setg(errp, "driver is not specified");
5356 goto fail;
5359 drv = bdrv_find_format(drvname);
5360 if (!drv) {
5361 error_setg(errp, "Unknown driver: '%s'", drvname);
5362 goto fail;
5365 node_name = qdict_get_try_str(options, "node-name");
5367 GLOBAL_STATE_CODE();
5369 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5370 errp);
5371 options = NULL; /* bdrv_new_open_driver() eats options */
5372 if (!new_node_bs) {
5373 error_prepend(errp, "Could not create node: ");
5374 goto fail;
5377 bdrv_drained_begin(bs);
5378 ret = bdrv_replace_node(bs, new_node_bs, errp);
5379 bdrv_drained_end(bs);
5381 if (ret < 0) {
5382 error_prepend(errp, "Could not replace node: ");
5383 goto fail;
5386 return new_node_bs;
5388 fail:
5389 qobject_unref(options);
5390 bdrv_unref(new_node_bs);
5391 return NULL;
5395 * Run consistency checks on an image
5397 * Returns 0 if the check could be completed (it doesn't mean that the image is
5398 * free of errors) or -errno when an internal error occurred. The results of the
5399 * check are stored in res.
5401 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5402 BdrvCheckResult *res, BdrvCheckMode fix)
5404 IO_CODE();
5405 assert_bdrv_graph_readable();
5406 if (bs->drv == NULL) {
5407 return -ENOMEDIUM;
5409 if (bs->drv->bdrv_co_check == NULL) {
5410 return -ENOTSUP;
5413 memset(res, 0, sizeof(*res));
5414 return bs->drv->bdrv_co_check(bs, res, fix);
5418 * Return values:
5419 * 0 - success
5420 * -EINVAL - backing format specified, but no file
5421 * -ENOSPC - can't update the backing file because no space is left in the
5422 * image file header
5423 * -ENOTSUP - format driver doesn't support changing the backing file
5425 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5426 const char *backing_fmt, bool require)
5428 BlockDriver *drv = bs->drv;
5429 int ret;
5431 GLOBAL_STATE_CODE();
5433 if (!drv) {
5434 return -ENOMEDIUM;
5437 /* Backing file format doesn't make sense without a backing file */
5438 if (backing_fmt && !backing_file) {
5439 return -EINVAL;
5442 if (require && backing_file && !backing_fmt) {
5443 return -EINVAL;
5446 if (drv->bdrv_change_backing_file != NULL) {
5447 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5448 } else {
5449 ret = -ENOTSUP;
5452 if (ret == 0) {
5453 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5454 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5455 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5456 backing_file ?: "");
5458 return ret;
5462 * Finds the first non-filter node above bs in the chain between
5463 * active and bs. The returned node is either an immediate parent of
5464 * bs, or there are only filter nodes between the two.
5466 * Returns NULL if bs is not found in active's image chain,
5467 * or if active == bs.
5469 * Returns the bottommost base image if bs == NULL.
5471 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5472 BlockDriverState *bs)
5475 GLOBAL_STATE_CODE();
5477 bs = bdrv_skip_filters(bs);
5478 active = bdrv_skip_filters(active);
5480 while (active) {
5481 BlockDriverState *next = bdrv_backing_chain_next(active);
5482 if (bs == next) {
5483 return active;
5485 active = next;
5488 return NULL;
5491 /* Given a BDS, searches for the base layer. */
5492 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5494 GLOBAL_STATE_CODE();
5496 return bdrv_find_overlay(bs, NULL);
5500 * Return true if at least one of the COW (backing) and filter links
5501 * between @bs and @base is frozen. @errp is set if that's the case.
5502 * @base must be reachable from @bs, or NULL.
5504 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5505 Error **errp)
5507 BlockDriverState *i;
5508 BdrvChild *child;
5510 GLOBAL_STATE_CODE();
5512 for (i = bs; i != base; i = child_bs(child)) {
5513 child = bdrv_filter_or_cow_child(i);
5515 if (child && child->frozen) {
5516 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5517 child->name, i->node_name, child->bs->node_name);
5518 return true;
5522 return false;
5526 * Freeze all COW (backing) and filter links between @bs and @base.
5527 * If any of the links is already frozen the operation is aborted and
5528 * none of the links are modified.
5529 * @base must be reachable from @bs, or NULL.
5530 * Returns 0 on success. On failure returns < 0 and sets @errp.
5532 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5533 Error **errp)
5535 BlockDriverState *i;
5536 BdrvChild *child;
5538 GLOBAL_STATE_CODE();
5540 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5541 return -EPERM;
5544 for (i = bs; i != base; i = child_bs(child)) {
5545 child = bdrv_filter_or_cow_child(i);
5546 if (child && child->bs->never_freeze) {
5547 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5548 child->name, child->bs->node_name);
5549 return -EPERM;
5553 for (i = bs; i != base; i = child_bs(child)) {
5554 child = bdrv_filter_or_cow_child(i);
5555 if (child) {
5556 child->frozen = true;
5560 return 0;
5564 * Unfreeze all COW (backing) and filter links between @bs and @base.
5565 * The caller must ensure that all links are frozen before using this
5566 * function.
5567 * @base must be reachable from @bs, or NULL.
5569 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5571 BlockDriverState *i;
5572 BdrvChild *child;
5574 GLOBAL_STATE_CODE();
5576 for (i = bs; i != base; i = child_bs(child)) {
5577 child = bdrv_filter_or_cow_child(i);
5578 if (child) {
5579 assert(child->frozen);
5580 child->frozen = false;
5586 * Drops images above 'base' up to and including 'top', and sets the image
5587 * above 'top' to have base as its backing file.
5589 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5590 * information in 'bs' can be properly updated.
5592 * E.g., this will convert the following chain:
5593 * bottom <- base <- intermediate <- top <- active
5595 * to
5597 * bottom <- base <- active
5599 * It is allowed for bottom==base, in which case it converts:
5601 * base <- intermediate <- top <- active
5603 * to
5605 * base <- active
5607 * If backing_file_str is non-NULL, it will be used when modifying top's
5608 * overlay image metadata.
5610 * Error conditions:
5611 * if active == top, that is considered an error
5614 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5615 const char *backing_file_str)
5617 BlockDriverState *explicit_top = top;
5618 bool update_inherits_from;
5619 BdrvChild *c;
5620 Error *local_err = NULL;
5621 int ret = -EIO;
5622 g_autoptr(GSList) updated_children = NULL;
5623 GSList *p;
5625 GLOBAL_STATE_CODE();
5627 bdrv_ref(top);
5628 bdrv_drained_begin(base);
5630 if (!top->drv || !base->drv) {
5631 goto exit;
5634 /* Make sure that base is in the backing chain of top */
5635 if (!bdrv_chain_contains(top, base)) {
5636 goto exit;
5639 /* If 'base' recursively inherits from 'top' then we should set
5640 * base->inherits_from to top->inherits_from after 'top' and all
5641 * other intermediate nodes have been dropped.
5642 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5643 * it because no one inherits from it. We use explicit_top for that. */
5644 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5645 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5647 /* success - we can delete the intermediate states, and link top->base */
5648 if (!backing_file_str) {
5649 bdrv_refresh_filename(base);
5650 backing_file_str = base->filename;
5653 QLIST_FOREACH(c, &top->parents, next_parent) {
5654 updated_children = g_slist_prepend(updated_children, c);
5658 * It seems correct to pass detach_subchain=true here, but it triggers
5659 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5660 * another drained section, which modify the graph (for example, removing
5661 * the child, which we keep in updated_children list). So, it's a TODO.
5663 * Note, bug triggered if pass detach_subchain=true here and run
5664 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5665 * That's a FIXME.
5667 bdrv_replace_node_common(top, base, false, false, &local_err);
5668 if (local_err) {
5669 error_report_err(local_err);
5670 goto exit;
5673 for (p = updated_children; p; p = p->next) {
5674 c = p->data;
5676 if (c->klass->update_filename) {
5677 ret = c->klass->update_filename(c, base, backing_file_str,
5678 &local_err);
5679 if (ret < 0) {
5681 * TODO: Actually, we want to rollback all previous iterations
5682 * of this loop, and (which is almost impossible) previous
5683 * bdrv_replace_node()...
5685 * Note, that c->klass->update_filename may lead to permission
5686 * update, so it's a bad idea to call it inside permission
5687 * update transaction of bdrv_replace_node.
5689 error_report_err(local_err);
5690 goto exit;
5695 if (update_inherits_from) {
5696 base->inherits_from = explicit_top->inherits_from;
5699 ret = 0;
5700 exit:
5701 bdrv_drained_end(base);
5702 bdrv_unref(top);
5703 return ret;
5707 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5708 * sums the size of all data-bearing children. (This excludes backing
5709 * children.)
5711 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5713 BdrvChild *child;
5714 int64_t child_size, sum = 0;
5716 QLIST_FOREACH(child, &bs->children, next) {
5717 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5718 BDRV_CHILD_FILTERED))
5720 child_size = bdrv_get_allocated_file_size(child->bs);
5721 if (child_size < 0) {
5722 return child_size;
5724 sum += child_size;
5728 return sum;
5732 * Length of a allocated file in bytes. Sparse files are counted by actual
5733 * allocated space. Return < 0 if error or unknown.
5735 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5737 BlockDriver *drv = bs->drv;
5738 IO_CODE();
5740 if (!drv) {
5741 return -ENOMEDIUM;
5743 if (drv->bdrv_get_allocated_file_size) {
5744 return drv->bdrv_get_allocated_file_size(bs);
5747 if (drv->bdrv_file_open) {
5749 * Protocol drivers default to -ENOTSUP (most of their data is
5750 * not stored in any of their children (if they even have any),
5751 * so there is no generic way to figure it out).
5753 return -ENOTSUP;
5754 } else if (drv->is_filter) {
5755 /* Filter drivers default to the size of their filtered child */
5756 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5757 } else {
5758 /* Other drivers default to summing their children's sizes */
5759 return bdrv_sum_allocated_file_size(bs);
5764 * bdrv_measure:
5765 * @drv: Format driver
5766 * @opts: Creation options for new image
5767 * @in_bs: Existing image containing data for new image (may be NULL)
5768 * @errp: Error object
5769 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5770 * or NULL on error
5772 * Calculate file size required to create a new image.
5774 * If @in_bs is given then space for allocated clusters and zero clusters
5775 * from that image are included in the calculation. If @opts contains a
5776 * backing file that is shared by @in_bs then backing clusters may be omitted
5777 * from the calculation.
5779 * If @in_bs is NULL then the calculation includes no allocated clusters
5780 * unless a preallocation option is given in @opts.
5782 * Note that @in_bs may use a different BlockDriver from @drv.
5784 * If an error occurs the @errp pointer is set.
5786 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5787 BlockDriverState *in_bs, Error **errp)
5789 IO_CODE();
5790 if (!drv->bdrv_measure) {
5791 error_setg(errp, "Block driver '%s' does not support size measurement",
5792 drv->format_name);
5793 return NULL;
5796 return drv->bdrv_measure(opts, in_bs, errp);
5800 * Return number of sectors on success, -errno on error.
5802 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5804 BlockDriver *drv = bs->drv;
5805 IO_CODE();
5807 if (!drv)
5808 return -ENOMEDIUM;
5810 if (drv->has_variable_length) {
5811 int ret = refresh_total_sectors(bs, bs->total_sectors);
5812 if (ret < 0) {
5813 return ret;
5816 return bs->total_sectors;
5820 * Return length in bytes on success, -errno on error.
5821 * The length is always a multiple of BDRV_SECTOR_SIZE.
5823 int64_t bdrv_getlength(BlockDriverState *bs)
5825 int64_t ret = bdrv_nb_sectors(bs);
5826 IO_CODE();
5828 if (ret < 0) {
5829 return ret;
5831 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5832 return -EFBIG;
5834 return ret * BDRV_SECTOR_SIZE;
5837 /* return 0 as number of sectors if no device present or error */
5838 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5840 int64_t nb_sectors = bdrv_nb_sectors(bs);
5841 IO_CODE();
5843 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5846 bool bdrv_is_sg(BlockDriverState *bs)
5848 IO_CODE();
5849 return bs->sg;
5853 * Return whether the given node supports compressed writes.
5855 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5857 BlockDriverState *filtered;
5858 IO_CODE();
5860 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5861 return false;
5864 filtered = bdrv_filter_bs(bs);
5865 if (filtered) {
5867 * Filters can only forward compressed writes, so we have to
5868 * check the child.
5870 return bdrv_supports_compressed_writes(filtered);
5873 return true;
5876 const char *bdrv_get_format_name(BlockDriverState *bs)
5878 IO_CODE();
5879 return bs->drv ? bs->drv->format_name : NULL;
5882 static int qsort_strcmp(const void *a, const void *b)
5884 return strcmp(*(char *const *)a, *(char *const *)b);
5887 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5888 void *opaque, bool read_only)
5890 BlockDriver *drv;
5891 int count = 0;
5892 int i;
5893 const char **formats = NULL;
5895 GLOBAL_STATE_CODE();
5897 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5898 if (drv->format_name) {
5899 bool found = false;
5900 int i = count;
5902 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5903 continue;
5906 while (formats && i && !found) {
5907 found = !strcmp(formats[--i], drv->format_name);
5910 if (!found) {
5911 formats = g_renew(const char *, formats, count + 1);
5912 formats[count++] = drv->format_name;
5917 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5918 const char *format_name = block_driver_modules[i].format_name;
5920 if (format_name) {
5921 bool found = false;
5922 int j = count;
5924 if (use_bdrv_whitelist &&
5925 !bdrv_format_is_whitelisted(format_name, read_only)) {
5926 continue;
5929 while (formats && j && !found) {
5930 found = !strcmp(formats[--j], format_name);
5933 if (!found) {
5934 formats = g_renew(const char *, formats, count + 1);
5935 formats[count++] = format_name;
5940 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5942 for (i = 0; i < count; i++) {
5943 it(opaque, formats[i]);
5946 g_free(formats);
5949 /* This function is to find a node in the bs graph */
5950 BlockDriverState *bdrv_find_node(const char *node_name)
5952 BlockDriverState *bs;
5954 assert(node_name);
5955 GLOBAL_STATE_CODE();
5957 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5958 if (!strcmp(node_name, bs->node_name)) {
5959 return bs;
5962 return NULL;
5965 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5966 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5967 Error **errp)
5969 BlockDeviceInfoList *list;
5970 BlockDriverState *bs;
5972 GLOBAL_STATE_CODE();
5974 list = NULL;
5975 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5976 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5977 if (!info) {
5978 qapi_free_BlockDeviceInfoList(list);
5979 return NULL;
5981 QAPI_LIST_PREPEND(list, info);
5984 return list;
5987 typedef struct XDbgBlockGraphConstructor {
5988 XDbgBlockGraph *graph;
5989 GHashTable *graph_nodes;
5990 } XDbgBlockGraphConstructor;
5992 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5994 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5996 gr->graph = g_new0(XDbgBlockGraph, 1);
5997 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5999 return gr;
6002 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6004 XDbgBlockGraph *graph = gr->graph;
6006 g_hash_table_destroy(gr->graph_nodes);
6007 g_free(gr);
6009 return graph;
6012 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6014 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6016 if (ret != 0) {
6017 return ret;
6021 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6022 * answer of g_hash_table_lookup.
6024 ret = g_hash_table_size(gr->graph_nodes) + 1;
6025 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6027 return ret;
6030 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6031 XDbgBlockGraphNodeType type, const char *name)
6033 XDbgBlockGraphNode *n;
6035 n = g_new0(XDbgBlockGraphNode, 1);
6037 n->id = xdbg_graph_node_num(gr, node);
6038 n->type = type;
6039 n->name = g_strdup(name);
6041 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6044 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6045 const BdrvChild *child)
6047 BlockPermission qapi_perm;
6048 XDbgBlockGraphEdge *edge;
6049 GLOBAL_STATE_CODE();
6051 edge = g_new0(XDbgBlockGraphEdge, 1);
6053 edge->parent = xdbg_graph_node_num(gr, parent);
6054 edge->child = xdbg_graph_node_num(gr, child->bs);
6055 edge->name = g_strdup(child->name);
6057 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6058 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6060 if (flag & child->perm) {
6061 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6063 if (flag & child->shared_perm) {
6064 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6068 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6072 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6074 BlockBackend *blk;
6075 BlockJob *job;
6076 BlockDriverState *bs;
6077 BdrvChild *child;
6078 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6080 GLOBAL_STATE_CODE();
6082 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6083 char *allocated_name = NULL;
6084 const char *name = blk_name(blk);
6086 if (!*name) {
6087 name = allocated_name = blk_get_attached_dev_id(blk);
6089 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6090 name);
6091 g_free(allocated_name);
6092 if (blk_root(blk)) {
6093 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6097 WITH_JOB_LOCK_GUARD() {
6098 for (job = block_job_next_locked(NULL); job;
6099 job = block_job_next_locked(job)) {
6100 GSList *el;
6102 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6103 job->job.id);
6104 for (el = job->nodes; el; el = el->next) {
6105 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6110 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6111 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6112 bs->node_name);
6113 QLIST_FOREACH(child, &bs->children, next) {
6114 xdbg_graph_add_edge(gr, bs, child);
6118 return xdbg_graph_finalize(gr);
6121 BlockDriverState *bdrv_lookup_bs(const char *device,
6122 const char *node_name,
6123 Error **errp)
6125 BlockBackend *blk;
6126 BlockDriverState *bs;
6128 GLOBAL_STATE_CODE();
6130 if (device) {
6131 blk = blk_by_name(device);
6133 if (blk) {
6134 bs = blk_bs(blk);
6135 if (!bs) {
6136 error_setg(errp, "Device '%s' has no medium", device);
6139 return bs;
6143 if (node_name) {
6144 bs = bdrv_find_node(node_name);
6146 if (bs) {
6147 return bs;
6151 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6152 device ? device : "",
6153 node_name ? node_name : "");
6154 return NULL;
6157 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6158 * return false. If either argument is NULL, return false. */
6159 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6162 GLOBAL_STATE_CODE();
6164 while (top && top != base) {
6165 top = bdrv_filter_or_cow_bs(top);
6168 return top != NULL;
6171 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6173 GLOBAL_STATE_CODE();
6174 if (!bs) {
6175 return QTAILQ_FIRST(&graph_bdrv_states);
6177 return QTAILQ_NEXT(bs, node_list);
6180 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6182 GLOBAL_STATE_CODE();
6183 if (!bs) {
6184 return QTAILQ_FIRST(&all_bdrv_states);
6186 return QTAILQ_NEXT(bs, bs_list);
6189 const char *bdrv_get_node_name(const BlockDriverState *bs)
6191 IO_CODE();
6192 return bs->node_name;
6195 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6197 BdrvChild *c;
6198 const char *name;
6199 IO_CODE();
6201 /* If multiple parents have a name, just pick the first one. */
6202 QLIST_FOREACH(c, &bs->parents, next_parent) {
6203 if (c->klass->get_name) {
6204 name = c->klass->get_name(c);
6205 if (name && *name) {
6206 return name;
6211 return NULL;
6214 /* TODO check what callers really want: bs->node_name or blk_name() */
6215 const char *bdrv_get_device_name(const BlockDriverState *bs)
6217 IO_CODE();
6218 return bdrv_get_parent_name(bs) ?: "";
6221 /* This can be used to identify nodes that might not have a device
6222 * name associated. Since node and device names live in the same
6223 * namespace, the result is unambiguous. The exception is if both are
6224 * absent, then this returns an empty (non-null) string. */
6225 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6227 IO_CODE();
6228 return bdrv_get_parent_name(bs) ?: bs->node_name;
6231 int bdrv_get_flags(BlockDriverState *bs)
6233 IO_CODE();
6234 return bs->open_flags;
6237 int bdrv_has_zero_init_1(BlockDriverState *bs)
6239 GLOBAL_STATE_CODE();
6240 return 1;
6243 int bdrv_has_zero_init(BlockDriverState *bs)
6245 BlockDriverState *filtered;
6246 GLOBAL_STATE_CODE();
6248 if (!bs->drv) {
6249 return 0;
6252 /* If BS is a copy on write image, it is initialized to
6253 the contents of the base image, which may not be zeroes. */
6254 if (bdrv_cow_child(bs)) {
6255 return 0;
6257 if (bs->drv->bdrv_has_zero_init) {
6258 return bs->drv->bdrv_has_zero_init(bs);
6261 filtered = bdrv_filter_bs(bs);
6262 if (filtered) {
6263 return bdrv_has_zero_init(filtered);
6266 /* safe default */
6267 return 0;
6270 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6272 IO_CODE();
6273 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6274 return false;
6277 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6280 void bdrv_get_backing_filename(BlockDriverState *bs,
6281 char *filename, int filename_size)
6283 IO_CODE();
6284 pstrcpy(filename, filename_size, bs->backing_file);
6287 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6289 int ret;
6290 BlockDriver *drv = bs->drv;
6291 IO_CODE();
6292 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6293 if (!drv) {
6294 return -ENOMEDIUM;
6296 if (!drv->bdrv_get_info) {
6297 BlockDriverState *filtered = bdrv_filter_bs(bs);
6298 if (filtered) {
6299 return bdrv_get_info(filtered, bdi);
6301 return -ENOTSUP;
6303 memset(bdi, 0, sizeof(*bdi));
6304 ret = drv->bdrv_get_info(bs, bdi);
6305 if (ret < 0) {
6306 return ret;
6309 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6310 return -EINVAL;
6313 return 0;
6316 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6317 Error **errp)
6319 BlockDriver *drv = bs->drv;
6320 IO_CODE();
6321 if (drv && drv->bdrv_get_specific_info) {
6322 return drv->bdrv_get_specific_info(bs, errp);
6324 return NULL;
6327 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6329 BlockDriver *drv = bs->drv;
6330 IO_CODE();
6331 if (!drv || !drv->bdrv_get_specific_stats) {
6332 return NULL;
6334 return drv->bdrv_get_specific_stats(bs);
6337 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6339 IO_CODE();
6340 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6341 return;
6344 bs->drv->bdrv_debug_event(bs, event);
6347 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6349 GLOBAL_STATE_CODE();
6350 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6351 bs = bdrv_primary_bs(bs);
6354 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6355 assert(bs->drv->bdrv_debug_remove_breakpoint);
6356 return bs;
6359 return NULL;
6362 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6363 const char *tag)
6365 GLOBAL_STATE_CODE();
6366 bs = bdrv_find_debug_node(bs);
6367 if (bs) {
6368 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6371 return -ENOTSUP;
6374 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6376 GLOBAL_STATE_CODE();
6377 bs = bdrv_find_debug_node(bs);
6378 if (bs) {
6379 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6382 return -ENOTSUP;
6385 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6387 GLOBAL_STATE_CODE();
6388 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6389 bs = bdrv_primary_bs(bs);
6392 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6393 return bs->drv->bdrv_debug_resume(bs, tag);
6396 return -ENOTSUP;
6399 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6401 GLOBAL_STATE_CODE();
6402 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6403 bs = bdrv_primary_bs(bs);
6406 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6407 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6410 return false;
6413 /* backing_file can either be relative, or absolute, or a protocol. If it is
6414 * relative, it must be relative to the chain. So, passing in bs->filename
6415 * from a BDS as backing_file should not be done, as that may be relative to
6416 * the CWD rather than the chain. */
6417 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6418 const char *backing_file)
6420 char *filename_full = NULL;
6421 char *backing_file_full = NULL;
6422 char *filename_tmp = NULL;
6423 int is_protocol = 0;
6424 bool filenames_refreshed = false;
6425 BlockDriverState *curr_bs = NULL;
6426 BlockDriverState *retval = NULL;
6427 BlockDriverState *bs_below;
6429 GLOBAL_STATE_CODE();
6431 if (!bs || !bs->drv || !backing_file) {
6432 return NULL;
6435 filename_full = g_malloc(PATH_MAX);
6436 backing_file_full = g_malloc(PATH_MAX);
6438 is_protocol = path_has_protocol(backing_file);
6441 * Being largely a legacy function, skip any filters here
6442 * (because filters do not have normal filenames, so they cannot
6443 * match anyway; and allowing json:{} filenames is a bit out of
6444 * scope).
6446 for (curr_bs = bdrv_skip_filters(bs);
6447 bdrv_cow_child(curr_bs) != NULL;
6448 curr_bs = bs_below)
6450 bs_below = bdrv_backing_chain_next(curr_bs);
6452 if (bdrv_backing_overridden(curr_bs)) {
6454 * If the backing file was overridden, we can only compare
6455 * directly against the backing node's filename.
6458 if (!filenames_refreshed) {
6460 * This will automatically refresh all of the
6461 * filenames in the rest of the backing chain, so we
6462 * only need to do this once.
6464 bdrv_refresh_filename(bs_below);
6465 filenames_refreshed = true;
6468 if (strcmp(backing_file, bs_below->filename) == 0) {
6469 retval = bs_below;
6470 break;
6472 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6474 * If either of the filename paths is actually a protocol, then
6475 * compare unmodified paths; otherwise make paths relative.
6477 char *backing_file_full_ret;
6479 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6480 retval = bs_below;
6481 break;
6483 /* Also check against the full backing filename for the image */
6484 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6485 NULL);
6486 if (backing_file_full_ret) {
6487 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6488 g_free(backing_file_full_ret);
6489 if (equal) {
6490 retval = bs_below;
6491 break;
6494 } else {
6495 /* If not an absolute filename path, make it relative to the current
6496 * image's filename path */
6497 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6498 NULL);
6499 /* We are going to compare canonicalized absolute pathnames */
6500 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6501 g_free(filename_tmp);
6502 continue;
6504 g_free(filename_tmp);
6506 /* We need to make sure the backing filename we are comparing against
6507 * is relative to the current image filename (or absolute) */
6508 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6509 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6510 g_free(filename_tmp);
6511 continue;
6513 g_free(filename_tmp);
6515 if (strcmp(backing_file_full, filename_full) == 0) {
6516 retval = bs_below;
6517 break;
6522 g_free(filename_full);
6523 g_free(backing_file_full);
6524 return retval;
6527 void bdrv_init(void)
6529 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6530 use_bdrv_whitelist = 1;
6531 #endif
6532 module_call_init(MODULE_INIT_BLOCK);
6535 void bdrv_init_with_whitelist(void)
6537 use_bdrv_whitelist = 1;
6538 bdrv_init();
6541 int bdrv_activate(BlockDriverState *bs, Error **errp)
6543 BdrvChild *child, *parent;
6544 Error *local_err = NULL;
6545 int ret;
6546 BdrvDirtyBitmap *bm;
6548 GLOBAL_STATE_CODE();
6550 if (!bs->drv) {
6551 return -ENOMEDIUM;
6554 QLIST_FOREACH(child, &bs->children, next) {
6555 bdrv_activate(child->bs, &local_err);
6556 if (local_err) {
6557 error_propagate(errp, local_err);
6558 return -EINVAL;
6563 * Update permissions, they may differ for inactive nodes.
6565 * Note that the required permissions of inactive images are always a
6566 * subset of the permissions required after activating the image. This
6567 * allows us to just get the permissions upfront without restricting
6568 * bdrv_co_invalidate_cache().
6570 * It also means that in error cases, we don't have to try and revert to
6571 * the old permissions (which is an operation that could fail, too). We can
6572 * just keep the extended permissions for the next time that an activation
6573 * of the image is tried.
6575 if (bs->open_flags & BDRV_O_INACTIVE) {
6576 bs->open_flags &= ~BDRV_O_INACTIVE;
6577 ret = bdrv_refresh_perms(bs, NULL, errp);
6578 if (ret < 0) {
6579 bs->open_flags |= BDRV_O_INACTIVE;
6580 return ret;
6583 ret = bdrv_invalidate_cache(bs, errp);
6584 if (ret < 0) {
6585 bs->open_flags |= BDRV_O_INACTIVE;
6586 return ret;
6589 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6590 bdrv_dirty_bitmap_skip_store(bm, false);
6593 ret = refresh_total_sectors(bs, bs->total_sectors);
6594 if (ret < 0) {
6595 bs->open_flags |= BDRV_O_INACTIVE;
6596 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6597 return ret;
6601 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6602 if (parent->klass->activate) {
6603 parent->klass->activate(parent, &local_err);
6604 if (local_err) {
6605 bs->open_flags |= BDRV_O_INACTIVE;
6606 error_propagate(errp, local_err);
6607 return -EINVAL;
6612 return 0;
6615 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6617 Error *local_err = NULL;
6618 IO_CODE();
6620 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6621 assert_bdrv_graph_readable();
6623 if (bs->drv->bdrv_co_invalidate_cache) {
6624 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6625 if (local_err) {
6626 error_propagate(errp, local_err);
6627 return -EINVAL;
6631 return 0;
6634 void bdrv_activate_all(Error **errp)
6636 BlockDriverState *bs;
6637 BdrvNextIterator it;
6639 GLOBAL_STATE_CODE();
6641 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6642 AioContext *aio_context = bdrv_get_aio_context(bs);
6643 int ret;
6645 aio_context_acquire(aio_context);
6646 ret = bdrv_activate(bs, errp);
6647 aio_context_release(aio_context);
6648 if (ret < 0) {
6649 bdrv_next_cleanup(&it);
6650 return;
6655 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6657 BdrvChild *parent;
6658 GLOBAL_STATE_CODE();
6660 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6661 if (parent->klass->parent_is_bds) {
6662 BlockDriverState *parent_bs = parent->opaque;
6663 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6664 return true;
6669 return false;
6672 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6674 BdrvChild *child, *parent;
6675 int ret;
6676 uint64_t cumulative_perms, cumulative_shared_perms;
6678 GLOBAL_STATE_CODE();
6680 if (!bs->drv) {
6681 return -ENOMEDIUM;
6684 /* Make sure that we don't inactivate a child before its parent.
6685 * It will be covered by recursion from the yet active parent. */
6686 if (bdrv_has_bds_parent(bs, true)) {
6687 return 0;
6690 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6692 /* Inactivate this node */
6693 if (bs->drv->bdrv_inactivate) {
6694 ret = bs->drv->bdrv_inactivate(bs);
6695 if (ret < 0) {
6696 return ret;
6700 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6701 if (parent->klass->inactivate) {
6702 ret = parent->klass->inactivate(parent);
6703 if (ret < 0) {
6704 return ret;
6709 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6710 &cumulative_shared_perms);
6711 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6712 /* Our inactive parents still need write access. Inactivation failed. */
6713 return -EPERM;
6716 bs->open_flags |= BDRV_O_INACTIVE;
6719 * Update permissions, they may differ for inactive nodes.
6720 * We only tried to loosen restrictions, so errors are not fatal, ignore
6721 * them.
6723 bdrv_refresh_perms(bs, NULL, NULL);
6725 /* Recursively inactivate children */
6726 QLIST_FOREACH(child, &bs->children, next) {
6727 ret = bdrv_inactivate_recurse(child->bs);
6728 if (ret < 0) {
6729 return ret;
6733 return 0;
6736 int bdrv_inactivate_all(void)
6738 BlockDriverState *bs = NULL;
6739 BdrvNextIterator it;
6740 int ret = 0;
6741 GSList *aio_ctxs = NULL, *ctx;
6743 GLOBAL_STATE_CODE();
6745 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6746 AioContext *aio_context = bdrv_get_aio_context(bs);
6748 if (!g_slist_find(aio_ctxs, aio_context)) {
6749 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6750 aio_context_acquire(aio_context);
6754 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6755 /* Nodes with BDS parents are covered by recursion from the last
6756 * parent that gets inactivated. Don't inactivate them a second
6757 * time if that has already happened. */
6758 if (bdrv_has_bds_parent(bs, false)) {
6759 continue;
6761 ret = bdrv_inactivate_recurse(bs);
6762 if (ret < 0) {
6763 bdrv_next_cleanup(&it);
6764 goto out;
6768 out:
6769 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6770 AioContext *aio_context = ctx->data;
6771 aio_context_release(aio_context);
6773 g_slist_free(aio_ctxs);
6775 return ret;
6778 /**************************************************************/
6779 /* removable device support */
6782 * Return TRUE if the media is present
6784 bool bdrv_is_inserted(BlockDriverState *bs)
6786 BlockDriver *drv = bs->drv;
6787 BdrvChild *child;
6788 IO_CODE();
6790 if (!drv) {
6791 return false;
6793 if (drv->bdrv_is_inserted) {
6794 return drv->bdrv_is_inserted(bs);
6796 QLIST_FOREACH(child, &bs->children, next) {
6797 if (!bdrv_is_inserted(child->bs)) {
6798 return false;
6801 return true;
6805 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6807 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6809 BlockDriver *drv = bs->drv;
6810 IO_CODE();
6812 if (drv && drv->bdrv_eject) {
6813 drv->bdrv_eject(bs, eject_flag);
6818 * Lock or unlock the media (if it is locked, the user won't be able
6819 * to eject it manually).
6821 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6823 BlockDriver *drv = bs->drv;
6824 IO_CODE();
6825 trace_bdrv_lock_medium(bs, locked);
6827 if (drv && drv->bdrv_lock_medium) {
6828 drv->bdrv_lock_medium(bs, locked);
6832 /* Get a reference to bs */
6833 void bdrv_ref(BlockDriverState *bs)
6835 GLOBAL_STATE_CODE();
6836 bs->refcnt++;
6839 /* Release a previously grabbed reference to bs.
6840 * If after releasing, reference count is zero, the BlockDriverState is
6841 * deleted. */
6842 void bdrv_unref(BlockDriverState *bs)
6844 GLOBAL_STATE_CODE();
6845 if (!bs) {
6846 return;
6848 assert(bs->refcnt > 0);
6849 if (--bs->refcnt == 0) {
6850 bdrv_delete(bs);
6854 struct BdrvOpBlocker {
6855 Error *reason;
6856 QLIST_ENTRY(BdrvOpBlocker) list;
6859 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6861 BdrvOpBlocker *blocker;
6862 GLOBAL_STATE_CODE();
6863 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6864 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6865 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6866 error_propagate_prepend(errp, error_copy(blocker->reason),
6867 "Node '%s' is busy: ",
6868 bdrv_get_device_or_node_name(bs));
6869 return true;
6871 return false;
6874 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6876 BdrvOpBlocker *blocker;
6877 GLOBAL_STATE_CODE();
6878 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6880 blocker = g_new0(BdrvOpBlocker, 1);
6881 blocker->reason = reason;
6882 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6885 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6887 BdrvOpBlocker *blocker, *next;
6888 GLOBAL_STATE_CODE();
6889 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6890 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6891 if (blocker->reason == reason) {
6892 QLIST_REMOVE(blocker, list);
6893 g_free(blocker);
6898 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6900 int i;
6901 GLOBAL_STATE_CODE();
6902 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6903 bdrv_op_block(bs, i, reason);
6907 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6909 int i;
6910 GLOBAL_STATE_CODE();
6911 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6912 bdrv_op_unblock(bs, i, reason);
6916 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6918 int i;
6919 GLOBAL_STATE_CODE();
6920 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6921 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6922 return false;
6925 return true;
6929 * Must not be called while holding the lock of an AioContext other than the
6930 * current one.
6932 void bdrv_img_create(const char *filename, const char *fmt,
6933 const char *base_filename, const char *base_fmt,
6934 char *options, uint64_t img_size, int flags, bool quiet,
6935 Error **errp)
6937 QemuOptsList *create_opts = NULL;
6938 QemuOpts *opts = NULL;
6939 const char *backing_fmt, *backing_file;
6940 int64_t size;
6941 BlockDriver *drv, *proto_drv;
6942 Error *local_err = NULL;
6943 int ret = 0;
6945 GLOBAL_STATE_CODE();
6947 /* Find driver and parse its options */
6948 drv = bdrv_find_format(fmt);
6949 if (!drv) {
6950 error_setg(errp, "Unknown file format '%s'", fmt);
6951 return;
6954 proto_drv = bdrv_find_protocol(filename, true, errp);
6955 if (!proto_drv) {
6956 return;
6959 if (!drv->create_opts) {
6960 error_setg(errp, "Format driver '%s' does not support image creation",
6961 drv->format_name);
6962 return;
6965 if (!proto_drv->create_opts) {
6966 error_setg(errp, "Protocol driver '%s' does not support image creation",
6967 proto_drv->format_name);
6968 return;
6971 /* Create parameter list */
6972 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6973 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6975 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6977 /* Parse -o options */
6978 if (options) {
6979 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6980 goto out;
6984 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6985 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6986 } else if (img_size != UINT64_C(-1)) {
6987 error_setg(errp, "The image size must be specified only once");
6988 goto out;
6991 if (base_filename) {
6992 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6993 NULL)) {
6994 error_setg(errp, "Backing file not supported for file format '%s'",
6995 fmt);
6996 goto out;
7000 if (base_fmt) {
7001 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7002 error_setg(errp, "Backing file format not supported for file "
7003 "format '%s'", fmt);
7004 goto out;
7008 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7009 if (backing_file) {
7010 if (!strcmp(filename, backing_file)) {
7011 error_setg(errp, "Error: Trying to create an image with the "
7012 "same filename as the backing file");
7013 goto out;
7015 if (backing_file[0] == '\0') {
7016 error_setg(errp, "Expected backing file name, got empty string");
7017 goto out;
7021 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7023 /* The size for the image must always be specified, unless we have a backing
7024 * file and we have not been forbidden from opening it. */
7025 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7026 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7027 BlockDriverState *bs;
7028 char *full_backing;
7029 int back_flags;
7030 QDict *backing_options = NULL;
7032 full_backing =
7033 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7034 &local_err);
7035 if (local_err) {
7036 goto out;
7038 assert(full_backing);
7041 * No need to do I/O here, which allows us to open encrypted
7042 * backing images without needing the secret
7044 back_flags = flags;
7045 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7046 back_flags |= BDRV_O_NO_IO;
7048 backing_options = qdict_new();
7049 if (backing_fmt) {
7050 qdict_put_str(backing_options, "driver", backing_fmt);
7052 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7054 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7055 &local_err);
7056 g_free(full_backing);
7057 if (!bs) {
7058 error_append_hint(&local_err, "Could not open backing image.\n");
7059 goto out;
7060 } else {
7061 if (!backing_fmt) {
7062 error_setg(&local_err,
7063 "Backing file specified without backing format");
7064 error_append_hint(&local_err, "Detected format of %s.",
7065 bs->drv->format_name);
7066 goto out;
7068 if (size == -1) {
7069 /* Opened BS, have no size */
7070 size = bdrv_getlength(bs);
7071 if (size < 0) {
7072 error_setg_errno(errp, -size, "Could not get size of '%s'",
7073 backing_file);
7074 bdrv_unref(bs);
7075 goto out;
7077 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7079 bdrv_unref(bs);
7081 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7082 } else if (backing_file && !backing_fmt) {
7083 error_setg(&local_err,
7084 "Backing file specified without backing format");
7085 goto out;
7088 if (size == -1) {
7089 error_setg(errp, "Image creation needs a size parameter");
7090 goto out;
7093 if (!quiet) {
7094 printf("Formatting '%s', fmt=%s ", filename, fmt);
7095 qemu_opts_print(opts, " ");
7096 puts("");
7097 fflush(stdout);
7100 ret = bdrv_create(drv, filename, opts, &local_err);
7102 if (ret == -EFBIG) {
7103 /* This is generally a better message than whatever the driver would
7104 * deliver (especially because of the cluster_size_hint), since that
7105 * is most probably not much different from "image too large". */
7106 const char *cluster_size_hint = "";
7107 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7108 cluster_size_hint = " (try using a larger cluster size)";
7110 error_setg(errp, "The image size is too large for file format '%s'"
7111 "%s", fmt, cluster_size_hint);
7112 error_free(local_err);
7113 local_err = NULL;
7116 out:
7117 qemu_opts_del(opts);
7118 qemu_opts_free(create_opts);
7119 error_propagate(errp, local_err);
7122 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7124 IO_CODE();
7125 return bs ? bs->aio_context : qemu_get_aio_context();
7128 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7130 Coroutine *self = qemu_coroutine_self();
7131 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7132 AioContext *new_ctx;
7133 IO_CODE();
7136 * Increase bs->in_flight to ensure that this operation is completed before
7137 * moving the node to a different AioContext. Read new_ctx only afterwards.
7139 bdrv_inc_in_flight(bs);
7141 new_ctx = bdrv_get_aio_context(bs);
7142 aio_co_reschedule_self(new_ctx);
7143 return old_ctx;
7146 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7148 IO_CODE();
7149 aio_co_reschedule_self(old_ctx);
7150 bdrv_dec_in_flight(bs);
7153 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7155 AioContext *ctx = bdrv_get_aio_context(bs);
7157 /* In the main thread, bs->aio_context won't change concurrently */
7158 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7161 * We're in coroutine context, so we already hold the lock of the main
7162 * loop AioContext. Don't lock it twice to avoid deadlocks.
7164 assert(qemu_in_coroutine());
7165 if (ctx != qemu_get_aio_context()) {
7166 aio_context_acquire(ctx);
7170 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7172 AioContext *ctx = bdrv_get_aio_context(bs);
7174 assert(qemu_in_coroutine());
7175 if (ctx != qemu_get_aio_context()) {
7176 aio_context_release(ctx);
7180 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
7182 IO_CODE();
7183 aio_co_enter(bdrv_get_aio_context(bs), co);
7186 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7188 GLOBAL_STATE_CODE();
7189 QLIST_REMOVE(ban, list);
7190 g_free(ban);
7193 static void bdrv_detach_aio_context(BlockDriverState *bs)
7195 BdrvAioNotifier *baf, *baf_tmp;
7197 assert(!bs->walking_aio_notifiers);
7198 GLOBAL_STATE_CODE();
7199 bs->walking_aio_notifiers = true;
7200 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7201 if (baf->deleted) {
7202 bdrv_do_remove_aio_context_notifier(baf);
7203 } else {
7204 baf->detach_aio_context(baf->opaque);
7207 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7208 * remove remaining aio notifiers if we aren't called again.
7210 bs->walking_aio_notifiers = false;
7212 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7213 bs->drv->bdrv_detach_aio_context(bs);
7216 if (bs->quiesce_counter) {
7217 aio_enable_external(bs->aio_context);
7219 bs->aio_context = NULL;
7222 static void bdrv_attach_aio_context(BlockDriverState *bs,
7223 AioContext *new_context)
7225 BdrvAioNotifier *ban, *ban_tmp;
7226 GLOBAL_STATE_CODE();
7228 if (bs->quiesce_counter) {
7229 aio_disable_external(new_context);
7232 bs->aio_context = new_context;
7234 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7235 bs->drv->bdrv_attach_aio_context(bs, new_context);
7238 assert(!bs->walking_aio_notifiers);
7239 bs->walking_aio_notifiers = true;
7240 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7241 if (ban->deleted) {
7242 bdrv_do_remove_aio_context_notifier(ban);
7243 } else {
7244 ban->attached_aio_context(new_context, ban->opaque);
7247 bs->walking_aio_notifiers = false;
7250 typedef struct BdrvStateSetAioContext {
7251 AioContext *new_ctx;
7252 BlockDriverState *bs;
7253 } BdrvStateSetAioContext;
7255 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7256 GHashTable *visited,
7257 Transaction *tran,
7258 Error **errp)
7260 GLOBAL_STATE_CODE();
7261 if (g_hash_table_contains(visited, c)) {
7262 return true;
7264 g_hash_table_add(visited, c);
7267 * A BdrvChildClass that doesn't handle AioContext changes cannot
7268 * tolerate any AioContext changes
7270 if (!c->klass->change_aio_ctx) {
7271 char *user = bdrv_child_user_desc(c);
7272 error_setg(errp, "Changing iothreads is not supported by %s", user);
7273 g_free(user);
7274 return false;
7276 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7277 assert(!errp || *errp);
7278 return false;
7280 return true;
7283 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7284 GHashTable *visited, Transaction *tran,
7285 Error **errp)
7287 GLOBAL_STATE_CODE();
7288 if (g_hash_table_contains(visited, c)) {
7289 return true;
7291 g_hash_table_add(visited, c);
7292 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7295 static void bdrv_set_aio_context_clean(void *opaque)
7297 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7298 BlockDriverState *bs = (BlockDriverState *) state->bs;
7300 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7301 bdrv_drained_end(bs);
7303 g_free(state);
7306 static void bdrv_set_aio_context_commit(void *opaque)
7308 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7309 BlockDriverState *bs = (BlockDriverState *) state->bs;
7310 AioContext *new_context = state->new_ctx;
7311 AioContext *old_context = bdrv_get_aio_context(bs);
7314 * Take the old AioContex when detaching it from bs.
7315 * At this point, new_context lock is already acquired, and we are now
7316 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7317 * does not call AIO_POLL_WHILE().
7319 if (old_context != qemu_get_aio_context()) {
7320 aio_context_acquire(old_context);
7322 bdrv_detach_aio_context(bs);
7323 if (old_context != qemu_get_aio_context()) {
7324 aio_context_release(old_context);
7326 bdrv_attach_aio_context(bs, new_context);
7329 static TransactionActionDrv set_aio_context = {
7330 .commit = bdrv_set_aio_context_commit,
7331 .clean = bdrv_set_aio_context_clean,
7335 * Changes the AioContext used for fd handlers, timers, and BHs by this
7336 * BlockDriverState and all its children and parents.
7338 * Must be called from the main AioContext.
7340 * The caller must own the AioContext lock for the old AioContext of bs, but it
7341 * must not own the AioContext lock for new_context (unless new_context is the
7342 * same as the current context of bs).
7344 * @visited will accumulate all visited BdrvChild objects. The caller is
7345 * responsible for freeing the list afterwards.
7347 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7348 GHashTable *visited, Transaction *tran,
7349 Error **errp)
7351 BdrvChild *c;
7352 BdrvStateSetAioContext *state;
7354 GLOBAL_STATE_CODE();
7356 if (bdrv_get_aio_context(bs) == ctx) {
7357 return true;
7360 QLIST_FOREACH(c, &bs->parents, next_parent) {
7361 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7362 return false;
7366 QLIST_FOREACH(c, &bs->children, next) {
7367 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7368 return false;
7372 state = g_new(BdrvStateSetAioContext, 1);
7373 *state = (BdrvStateSetAioContext) {
7374 .new_ctx = ctx,
7375 .bs = bs,
7378 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7379 bdrv_drained_begin(bs);
7381 tran_add(tran, &set_aio_context, state);
7383 return true;
7387 * Change bs's and recursively all of its parents' and children's AioContext
7388 * to the given new context, returning an error if that isn't possible.
7390 * If ignore_child is not NULL, that child (and its subgraph) will not
7391 * be touched.
7393 * This function still requires the caller to take the bs current
7394 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7395 * assumes the lock is always held if bs is in another AioContext.
7396 * For the same reason, it temporarily also holds the new AioContext, since
7397 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7398 * Therefore the new AioContext lock must not be taken by the caller.
7400 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7401 BdrvChild *ignore_child, Error **errp)
7403 Transaction *tran;
7404 GHashTable *visited;
7405 int ret;
7406 AioContext *old_context = bdrv_get_aio_context(bs);
7407 GLOBAL_STATE_CODE();
7410 * Recursion phase: go through all nodes of the graph.
7411 * Take care of checking that all nodes support changing AioContext
7412 * and drain them, builing a linear list of callbacks to run if everything
7413 * is successful (the transaction itself).
7415 tran = tran_new();
7416 visited = g_hash_table_new(NULL, NULL);
7417 if (ignore_child) {
7418 g_hash_table_add(visited, ignore_child);
7420 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7421 g_hash_table_destroy(visited);
7424 * Linear phase: go through all callbacks collected in the transaction.
7425 * Run all callbacks collected in the recursion to switch all nodes
7426 * AioContext lock (transaction commit), or undo all changes done in the
7427 * recursion (transaction abort).
7430 if (!ret) {
7431 /* Just run clean() callbacks. No AioContext changed. */
7432 tran_abort(tran);
7433 return -EPERM;
7437 * Release old AioContext, it won't be needed anymore, as all
7438 * bdrv_drained_begin() have been called already.
7440 if (qemu_get_aio_context() != old_context) {
7441 aio_context_release(old_context);
7445 * Acquire new AioContext since bdrv_drained_end() is going to be called
7446 * after we switched all nodes in the new AioContext, and the function
7447 * assumes that the lock of the bs is always taken.
7449 if (qemu_get_aio_context() != ctx) {
7450 aio_context_acquire(ctx);
7453 tran_commit(tran);
7455 if (qemu_get_aio_context() != ctx) {
7456 aio_context_release(ctx);
7459 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7460 if (qemu_get_aio_context() != old_context) {
7461 aio_context_acquire(old_context);
7464 return 0;
7467 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7468 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7469 void (*detach_aio_context)(void *opaque), void *opaque)
7471 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7472 *ban = (BdrvAioNotifier){
7473 .attached_aio_context = attached_aio_context,
7474 .detach_aio_context = detach_aio_context,
7475 .opaque = opaque
7477 GLOBAL_STATE_CODE();
7479 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7482 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7483 void (*attached_aio_context)(AioContext *,
7484 void *),
7485 void (*detach_aio_context)(void *),
7486 void *opaque)
7488 BdrvAioNotifier *ban, *ban_next;
7489 GLOBAL_STATE_CODE();
7491 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7492 if (ban->attached_aio_context == attached_aio_context &&
7493 ban->detach_aio_context == detach_aio_context &&
7494 ban->opaque == opaque &&
7495 ban->deleted == false)
7497 if (bs->walking_aio_notifiers) {
7498 ban->deleted = true;
7499 } else {
7500 bdrv_do_remove_aio_context_notifier(ban);
7502 return;
7506 abort();
7509 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7510 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7511 bool force,
7512 Error **errp)
7514 GLOBAL_STATE_CODE();
7515 if (!bs->drv) {
7516 error_setg(errp, "Node is ejected");
7517 return -ENOMEDIUM;
7519 if (!bs->drv->bdrv_amend_options) {
7520 error_setg(errp, "Block driver '%s' does not support option amendment",
7521 bs->drv->format_name);
7522 return -ENOTSUP;
7524 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7525 cb_opaque, force, errp);
7529 * This function checks whether the given @to_replace is allowed to be
7530 * replaced by a node that always shows the same data as @bs. This is
7531 * used for example to verify whether the mirror job can replace
7532 * @to_replace by the target mirrored from @bs.
7533 * To be replaceable, @bs and @to_replace may either be guaranteed to
7534 * always show the same data (because they are only connected through
7535 * filters), or some driver may allow replacing one of its children
7536 * because it can guarantee that this child's data is not visible at
7537 * all (for example, for dissenting quorum children that have no other
7538 * parents).
7540 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7541 BlockDriverState *to_replace)
7543 BlockDriverState *filtered;
7545 GLOBAL_STATE_CODE();
7547 if (!bs || !bs->drv) {
7548 return false;
7551 if (bs == to_replace) {
7552 return true;
7555 /* See what the driver can do */
7556 if (bs->drv->bdrv_recurse_can_replace) {
7557 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7560 /* For filters without an own implementation, we can recurse on our own */
7561 filtered = bdrv_filter_bs(bs);
7562 if (filtered) {
7563 return bdrv_recurse_can_replace(filtered, to_replace);
7566 /* Safe default */
7567 return false;
7571 * Check whether the given @node_name can be replaced by a node that
7572 * has the same data as @parent_bs. If so, return @node_name's BDS;
7573 * NULL otherwise.
7575 * @node_name must be a (recursive) *child of @parent_bs (or this
7576 * function will return NULL).
7578 * The result (whether the node can be replaced or not) is only valid
7579 * for as long as no graph or permission changes occur.
7581 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7582 const char *node_name, Error **errp)
7584 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7585 AioContext *aio_context;
7587 GLOBAL_STATE_CODE();
7589 if (!to_replace_bs) {
7590 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7591 return NULL;
7594 aio_context = bdrv_get_aio_context(to_replace_bs);
7595 aio_context_acquire(aio_context);
7597 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7598 to_replace_bs = NULL;
7599 goto out;
7602 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7603 * most non filter in order to prevent data corruption.
7604 * Another benefit is that this tests exclude backing files which are
7605 * blocked by the backing blockers.
7607 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7608 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7609 "because it cannot be guaranteed that doing so would not "
7610 "lead to an abrupt change of visible data",
7611 node_name, parent_bs->node_name);
7612 to_replace_bs = NULL;
7613 goto out;
7616 out:
7617 aio_context_release(aio_context);
7618 return to_replace_bs;
7622 * Iterates through the list of runtime option keys that are said to
7623 * be "strong" for a BDS. An option is called "strong" if it changes
7624 * a BDS's data. For example, the null block driver's "size" and
7625 * "read-zeroes" options are strong, but its "latency-ns" option is
7626 * not.
7628 * If a key returned by this function ends with a dot, all options
7629 * starting with that prefix are strong.
7631 static const char *const *strong_options(BlockDriverState *bs,
7632 const char *const *curopt)
7634 static const char *const global_options[] = {
7635 "driver", "filename", NULL
7638 if (!curopt) {
7639 return &global_options[0];
7642 curopt++;
7643 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7644 curopt = bs->drv->strong_runtime_opts;
7647 return (curopt && *curopt) ? curopt : NULL;
7651 * Copies all strong runtime options from bs->options to the given
7652 * QDict. The set of strong option keys is determined by invoking
7653 * strong_options().
7655 * Returns true iff any strong option was present in bs->options (and
7656 * thus copied to the target QDict) with the exception of "filename"
7657 * and "driver". The caller is expected to use this value to decide
7658 * whether the existence of strong options prevents the generation of
7659 * a plain filename.
7661 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7663 bool found_any = false;
7664 const char *const *option_name = NULL;
7666 if (!bs->drv) {
7667 return false;
7670 while ((option_name = strong_options(bs, option_name))) {
7671 bool option_given = false;
7673 assert(strlen(*option_name) > 0);
7674 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7675 QObject *entry = qdict_get(bs->options, *option_name);
7676 if (!entry) {
7677 continue;
7680 qdict_put_obj(d, *option_name, qobject_ref(entry));
7681 option_given = true;
7682 } else {
7683 const QDictEntry *entry;
7684 for (entry = qdict_first(bs->options); entry;
7685 entry = qdict_next(bs->options, entry))
7687 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7688 qdict_put_obj(d, qdict_entry_key(entry),
7689 qobject_ref(qdict_entry_value(entry)));
7690 option_given = true;
7695 /* While "driver" and "filename" need to be included in a JSON filename,
7696 * their existence does not prohibit generation of a plain filename. */
7697 if (!found_any && option_given &&
7698 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7700 found_any = true;
7704 if (!qdict_haskey(d, "driver")) {
7705 /* Drivers created with bdrv_new_open_driver() may not have a
7706 * @driver option. Add it here. */
7707 qdict_put_str(d, "driver", bs->drv->format_name);
7710 return found_any;
7713 /* Note: This function may return false positives; it may return true
7714 * even if opening the backing file specified by bs's image header
7715 * would result in exactly bs->backing. */
7716 static bool bdrv_backing_overridden(BlockDriverState *bs)
7718 GLOBAL_STATE_CODE();
7719 if (bs->backing) {
7720 return strcmp(bs->auto_backing_file,
7721 bs->backing->bs->filename);
7722 } else {
7723 /* No backing BDS, so if the image header reports any backing
7724 * file, it must have been suppressed */
7725 return bs->auto_backing_file[0] != '\0';
7729 /* Updates the following BDS fields:
7730 * - exact_filename: A filename which may be used for opening a block device
7731 * which (mostly) equals the given BDS (even without any
7732 * other options; so reading and writing must return the same
7733 * results, but caching etc. may be different)
7734 * - full_open_options: Options which, when given when opening a block device
7735 * (without a filename), result in a BDS (mostly)
7736 * equalling the given one
7737 * - filename: If exact_filename is set, it is copied here. Otherwise,
7738 * full_open_options is converted to a JSON object, prefixed with
7739 * "json:" (for use through the JSON pseudo protocol) and put here.
7741 void bdrv_refresh_filename(BlockDriverState *bs)
7743 BlockDriver *drv = bs->drv;
7744 BdrvChild *child;
7745 BlockDriverState *primary_child_bs;
7746 QDict *opts;
7747 bool backing_overridden;
7748 bool generate_json_filename; /* Whether our default implementation should
7749 fill exact_filename (false) or not (true) */
7751 GLOBAL_STATE_CODE();
7753 if (!drv) {
7754 return;
7757 /* This BDS's file name may depend on any of its children's file names, so
7758 * refresh those first */
7759 QLIST_FOREACH(child, &bs->children, next) {
7760 bdrv_refresh_filename(child->bs);
7763 if (bs->implicit) {
7764 /* For implicit nodes, just copy everything from the single child */
7765 child = QLIST_FIRST(&bs->children);
7766 assert(QLIST_NEXT(child, next) == NULL);
7768 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7769 child->bs->exact_filename);
7770 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7772 qobject_unref(bs->full_open_options);
7773 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7775 return;
7778 backing_overridden = bdrv_backing_overridden(bs);
7780 if (bs->open_flags & BDRV_O_NO_IO) {
7781 /* Without I/O, the backing file does not change anything.
7782 * Therefore, in such a case (primarily qemu-img), we can
7783 * pretend the backing file has not been overridden even if
7784 * it technically has been. */
7785 backing_overridden = false;
7788 /* Gather the options QDict */
7789 opts = qdict_new();
7790 generate_json_filename = append_strong_runtime_options(opts, bs);
7791 generate_json_filename |= backing_overridden;
7793 if (drv->bdrv_gather_child_options) {
7794 /* Some block drivers may not want to present all of their children's
7795 * options, or name them differently from BdrvChild.name */
7796 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7797 } else {
7798 QLIST_FOREACH(child, &bs->children, next) {
7799 if (child == bs->backing && !backing_overridden) {
7800 /* We can skip the backing BDS if it has not been overridden */
7801 continue;
7804 qdict_put(opts, child->name,
7805 qobject_ref(child->bs->full_open_options));
7808 if (backing_overridden && !bs->backing) {
7809 /* Force no backing file */
7810 qdict_put_null(opts, "backing");
7814 qobject_unref(bs->full_open_options);
7815 bs->full_open_options = opts;
7817 primary_child_bs = bdrv_primary_bs(bs);
7819 if (drv->bdrv_refresh_filename) {
7820 /* Obsolete information is of no use here, so drop the old file name
7821 * information before refreshing it */
7822 bs->exact_filename[0] = '\0';
7824 drv->bdrv_refresh_filename(bs);
7825 } else if (primary_child_bs) {
7827 * Try to reconstruct valid information from the underlying
7828 * file -- this only works for format nodes (filter nodes
7829 * cannot be probed and as such must be selected by the user
7830 * either through an options dict, or through a special
7831 * filename which the filter driver must construct in its
7832 * .bdrv_refresh_filename() implementation).
7835 bs->exact_filename[0] = '\0';
7838 * We can use the underlying file's filename if:
7839 * - it has a filename,
7840 * - the current BDS is not a filter,
7841 * - the file is a protocol BDS, and
7842 * - opening that file (as this BDS's format) will automatically create
7843 * the BDS tree we have right now, that is:
7844 * - the user did not significantly change this BDS's behavior with
7845 * some explicit (strong) options
7846 * - no non-file child of this BDS has been overridden by the user
7847 * Both of these conditions are represented by generate_json_filename.
7849 if (primary_child_bs->exact_filename[0] &&
7850 primary_child_bs->drv->bdrv_file_open &&
7851 !drv->is_filter && !generate_json_filename)
7853 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7857 if (bs->exact_filename[0]) {
7858 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7859 } else {
7860 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7861 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7862 json->str) >= sizeof(bs->filename)) {
7863 /* Give user a hint if we truncated things. */
7864 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7866 g_string_free(json, true);
7870 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7872 BlockDriver *drv = bs->drv;
7873 BlockDriverState *child_bs;
7875 GLOBAL_STATE_CODE();
7877 if (!drv) {
7878 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7879 return NULL;
7882 if (drv->bdrv_dirname) {
7883 return drv->bdrv_dirname(bs, errp);
7886 child_bs = bdrv_primary_bs(bs);
7887 if (child_bs) {
7888 return bdrv_dirname(child_bs, errp);
7891 bdrv_refresh_filename(bs);
7892 if (bs->exact_filename[0] != '\0') {
7893 return path_combine(bs->exact_filename, "");
7896 error_setg(errp, "Cannot generate a base directory for %s nodes",
7897 drv->format_name);
7898 return NULL;
7902 * Hot add/remove a BDS's child. So the user can take a child offline when
7903 * it is broken and take a new child online
7905 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7906 Error **errp)
7908 GLOBAL_STATE_CODE();
7909 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7910 error_setg(errp, "The node %s does not support adding a child",
7911 bdrv_get_device_or_node_name(parent_bs));
7912 return;
7915 if (!QLIST_EMPTY(&child_bs->parents)) {
7916 error_setg(errp, "The node %s already has a parent",
7917 child_bs->node_name);
7918 return;
7921 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7924 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7926 BdrvChild *tmp;
7928 GLOBAL_STATE_CODE();
7929 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7930 error_setg(errp, "The node %s does not support removing a child",
7931 bdrv_get_device_or_node_name(parent_bs));
7932 return;
7935 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7936 if (tmp == child) {
7937 break;
7941 if (!tmp) {
7942 error_setg(errp, "The node %s does not have a child named %s",
7943 bdrv_get_device_or_node_name(parent_bs),
7944 bdrv_get_device_or_node_name(child->bs));
7945 return;
7948 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7951 int bdrv_make_empty(BdrvChild *c, Error **errp)
7953 BlockDriver *drv = c->bs->drv;
7954 int ret;
7956 GLOBAL_STATE_CODE();
7957 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7959 if (!drv->bdrv_make_empty) {
7960 error_setg(errp, "%s does not support emptying nodes",
7961 drv->format_name);
7962 return -ENOTSUP;
7965 ret = drv->bdrv_make_empty(c->bs);
7966 if (ret < 0) {
7967 error_setg_errno(errp, -ret, "Failed to empty %s",
7968 c->bs->filename);
7969 return ret;
7972 return 0;
7976 * Return the child that @bs acts as an overlay for, and from which data may be
7977 * copied in COW or COR operations. Usually this is the backing file.
7979 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7981 IO_CODE();
7983 if (!bs || !bs->drv) {
7984 return NULL;
7987 if (bs->drv->is_filter) {
7988 return NULL;
7991 if (!bs->backing) {
7992 return NULL;
7995 assert(bs->backing->role & BDRV_CHILD_COW);
7996 return bs->backing;
8000 * If @bs acts as a filter for exactly one of its children, return
8001 * that child.
8003 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8005 BdrvChild *c;
8006 IO_CODE();
8008 if (!bs || !bs->drv) {
8009 return NULL;
8012 if (!bs->drv->is_filter) {
8013 return NULL;
8016 /* Only one of @backing or @file may be used */
8017 assert(!(bs->backing && bs->file));
8019 c = bs->backing ?: bs->file;
8020 if (!c) {
8021 return NULL;
8024 assert(c->role & BDRV_CHILD_FILTERED);
8025 return c;
8029 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8030 * whichever is non-NULL.
8032 * Return NULL if both are NULL.
8034 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8036 BdrvChild *cow_child = bdrv_cow_child(bs);
8037 BdrvChild *filter_child = bdrv_filter_child(bs);
8038 IO_CODE();
8040 /* Filter nodes cannot have COW backing files */
8041 assert(!(cow_child && filter_child));
8043 return cow_child ?: filter_child;
8047 * Return the primary child of this node: For filters, that is the
8048 * filtered child. For other nodes, that is usually the child storing
8049 * metadata.
8050 * (A generally more helpful description is that this is (usually) the
8051 * child that has the same filename as @bs.)
8053 * Drivers do not necessarily have a primary child; for example quorum
8054 * does not.
8056 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8058 BdrvChild *c, *found = NULL;
8059 IO_CODE();
8061 QLIST_FOREACH(c, &bs->children, next) {
8062 if (c->role & BDRV_CHILD_PRIMARY) {
8063 assert(!found);
8064 found = c;
8068 return found;
8071 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8072 bool stop_on_explicit_filter)
8074 BdrvChild *c;
8076 if (!bs) {
8077 return NULL;
8080 while (!(stop_on_explicit_filter && !bs->implicit)) {
8081 c = bdrv_filter_child(bs);
8082 if (!c) {
8084 * A filter that is embedded in a working block graph must
8085 * have a child. Assert this here so this function does
8086 * not return a filter node that is not expected by the
8087 * caller.
8089 assert(!bs->drv || !bs->drv->is_filter);
8090 break;
8092 bs = c->bs;
8095 * Note that this treats nodes with bs->drv == NULL as not being
8096 * filters (bs->drv == NULL should be replaced by something else
8097 * anyway).
8098 * The advantage of this behavior is that this function will thus
8099 * always return a non-NULL value (given a non-NULL @bs).
8102 return bs;
8106 * Return the first BDS that has not been added implicitly or that
8107 * does not have a filtered child down the chain starting from @bs
8108 * (including @bs itself).
8110 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8112 GLOBAL_STATE_CODE();
8113 return bdrv_do_skip_filters(bs, true);
8117 * Return the first BDS that does not have a filtered child down the
8118 * chain starting from @bs (including @bs itself).
8120 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8122 IO_CODE();
8123 return bdrv_do_skip_filters(bs, false);
8127 * For a backing chain, return the first non-filter backing image of
8128 * the first non-filter image.
8130 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8132 IO_CODE();
8133 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8137 * Check whether [offset, offset + bytes) overlaps with the cached
8138 * block-status data region.
8140 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8141 * which is what bdrv_bsc_is_data()'s interface needs.
8142 * Otherwise, *pnum is not touched.
8144 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8145 int64_t offset, int64_t bytes,
8146 int64_t *pnum)
8148 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8149 bool overlaps;
8151 overlaps =
8152 qatomic_read(&bsc->valid) &&
8153 ranges_overlap(offset, bytes, bsc->data_start,
8154 bsc->data_end - bsc->data_start);
8156 if (overlaps && pnum) {
8157 *pnum = bsc->data_end - offset;
8160 return overlaps;
8164 * See block_int.h for this function's documentation.
8166 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8168 IO_CODE();
8169 RCU_READ_LOCK_GUARD();
8170 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8174 * See block_int.h for this function's documentation.
8176 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8177 int64_t offset, int64_t bytes)
8179 IO_CODE();
8180 RCU_READ_LOCK_GUARD();
8182 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8183 qatomic_set(&bs->block_status_cache->valid, false);
8188 * See block_int.h for this function's documentation.
8190 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8192 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8193 BdrvBlockStatusCache *old_bsc;
8194 IO_CODE();
8196 *new_bsc = (BdrvBlockStatusCache) {
8197 .valid = true,
8198 .data_start = offset,
8199 .data_end = offset + bytes,
8202 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8204 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8205 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8206 if (old_bsc) {
8207 g_free_rcu(old_bsc, rcu);