Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging
[qemu/ar7.git] / block.c
blob718e4cae8b3a530b25fce832db3a5b4b1f05845f
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_child_free(BdrvChild *child);
94 static void bdrv_replace_child_noperm(BdrvChild **child,
95 BlockDriverState *new_bs,
96 bool free_empty_child);
97 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
98 BdrvChild *child,
99 Transaction *tran);
100 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
101 Transaction *tran);
103 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
104 BlockReopenQueue *queue,
105 Transaction *change_child_tran, Error **errp);
106 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
107 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
109 static bool bdrv_backing_overridden(BlockDriverState *bs);
111 /* If non-zero, use only whitelisted block drivers */
112 static int use_bdrv_whitelist;
114 #ifdef _WIN32
115 static int is_windows_drive_prefix(const char *filename)
117 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
118 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
119 filename[1] == ':');
122 int is_windows_drive(const char *filename)
124 if (is_windows_drive_prefix(filename) &&
125 filename[2] == '\0')
126 return 1;
127 if (strstart(filename, "\\\\.\\", NULL) ||
128 strstart(filename, "//./", NULL))
129 return 1;
130 return 0;
132 #endif
134 size_t bdrv_opt_mem_align(BlockDriverState *bs)
136 if (!bs || !bs->drv) {
137 /* page size or 4k (hdd sector size) should be on the safe side */
138 return MAX(4096, qemu_real_host_page_size);
140 IO_CODE();
142 return bs->bl.opt_mem_alignment;
145 size_t bdrv_min_mem_align(BlockDriverState *bs)
147 if (!bs || !bs->drv) {
148 /* page size or 4k (hdd sector size) should be on the safe side */
149 return MAX(4096, qemu_real_host_page_size);
151 IO_CODE();
153 return bs->bl.min_mem_alignment;
156 /* check if the path starts with "<protocol>:" */
157 int path_has_protocol(const char *path)
159 const char *p;
161 #ifdef _WIN32
162 if (is_windows_drive(path) ||
163 is_windows_drive_prefix(path)) {
164 return 0;
166 p = path + strcspn(path, ":/\\");
167 #else
168 p = path + strcspn(path, ":/");
169 #endif
171 return *p == ':';
174 int path_is_absolute(const char *path)
176 #ifdef _WIN32
177 /* specific case for names like: "\\.\d:" */
178 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
179 return 1;
181 return (*path == '/' || *path == '\\');
182 #else
183 return (*path == '/');
184 #endif
187 /* if filename is absolute, just return its duplicate. Otherwise, build a
188 path to it by considering it is relative to base_path. URL are
189 supported. */
190 char *path_combine(const char *base_path, const char *filename)
192 const char *protocol_stripped = NULL;
193 const char *p, *p1;
194 char *result;
195 int len;
197 if (path_is_absolute(filename)) {
198 return g_strdup(filename);
201 if (path_has_protocol(base_path)) {
202 protocol_stripped = strchr(base_path, ':');
203 if (protocol_stripped) {
204 protocol_stripped++;
207 p = protocol_stripped ?: base_path;
209 p1 = strrchr(base_path, '/');
210 #ifdef _WIN32
212 const char *p2;
213 p2 = strrchr(base_path, '\\');
214 if (!p1 || p2 > p1) {
215 p1 = p2;
218 #endif
219 if (p1) {
220 p1++;
221 } else {
222 p1 = base_path;
224 if (p1 > p) {
225 p = p1;
227 len = p - base_path;
229 result = g_malloc(len + strlen(filename) + 1);
230 memcpy(result, base_path, len);
231 strcpy(result + len, filename);
233 return result;
237 * Helper function for bdrv_parse_filename() implementations to remove optional
238 * protocol prefixes (especially "file:") from a filename and for putting the
239 * stripped filename into the options QDict if there is such a prefix.
241 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
242 QDict *options)
244 if (strstart(filename, prefix, &filename)) {
245 /* Stripping the explicit protocol prefix may result in a protocol
246 * prefix being (wrongly) detected (if the filename contains a colon) */
247 if (path_has_protocol(filename)) {
248 GString *fat_filename;
250 /* This means there is some colon before the first slash; therefore,
251 * this cannot be an absolute path */
252 assert(!path_is_absolute(filename));
254 /* And we can thus fix the protocol detection issue by prefixing it
255 * by "./" */
256 fat_filename = g_string_new("./");
257 g_string_append(fat_filename, filename);
259 assert(!path_has_protocol(fat_filename->str));
261 qdict_put(options, "filename",
262 qstring_from_gstring(fat_filename));
263 } else {
264 /* If no protocol prefix was detected, we can use the shortened
265 * filename as-is */
266 qdict_put_str(options, "filename", filename);
272 /* Returns whether the image file is opened as read-only. Note that this can
273 * return false and writing to the image file is still not possible because the
274 * image is inactivated. */
275 bool bdrv_is_read_only(BlockDriverState *bs)
277 IO_CODE();
278 return !(bs->open_flags & BDRV_O_RDWR);
281 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
282 bool ignore_allow_rdw, Error **errp)
284 IO_CODE();
286 /* Do not set read_only if copy_on_read is enabled */
287 if (bs->copy_on_read && read_only) {
288 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
289 bdrv_get_device_or_node_name(bs));
290 return -EINVAL;
293 /* Do not clear read_only if it is prohibited */
294 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
295 !ignore_allow_rdw)
297 error_setg(errp, "Node '%s' is read only",
298 bdrv_get_device_or_node_name(bs));
299 return -EPERM;
302 return 0;
306 * Called by a driver that can only provide a read-only image.
308 * Returns 0 if the node is already read-only or it could switch the node to
309 * read-only because BDRV_O_AUTO_RDONLY is set.
311 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
312 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
313 * is not NULL, it is used as the error message for the Error object.
315 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
316 Error **errp)
318 int ret = 0;
319 IO_CODE();
321 if (!(bs->open_flags & BDRV_O_RDWR)) {
322 return 0;
324 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
325 goto fail;
328 ret = bdrv_can_set_read_only(bs, true, false, NULL);
329 if (ret < 0) {
330 goto fail;
333 bs->open_flags &= ~BDRV_O_RDWR;
335 return 0;
337 fail:
338 error_setg(errp, "%s", errmsg ?: "Image is read-only");
339 return -EACCES;
343 * If @backing is empty, this function returns NULL without setting
344 * @errp. In all other cases, NULL will only be returned with @errp
345 * set.
347 * Therefore, a return value of NULL without @errp set means that
348 * there is no backing file; if @errp is set, there is one but its
349 * absolute filename cannot be generated.
351 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
352 const char *backing,
353 Error **errp)
355 if (backing[0] == '\0') {
356 return NULL;
357 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
358 return g_strdup(backing);
359 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
360 error_setg(errp, "Cannot use relative backing file names for '%s'",
361 backed);
362 return NULL;
363 } else {
364 return path_combine(backed, backing);
369 * If @filename is empty or NULL, this function returns NULL without
370 * setting @errp. In all other cases, NULL will only be returned with
371 * @errp set.
373 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
374 const char *filename, Error **errp)
376 char *dir, *full_name;
378 if (!filename || filename[0] == '\0') {
379 return NULL;
380 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
381 return g_strdup(filename);
384 dir = bdrv_dirname(relative_to, errp);
385 if (!dir) {
386 return NULL;
389 full_name = g_strconcat(dir, filename, NULL);
390 g_free(dir);
391 return full_name;
394 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
396 GLOBAL_STATE_CODE();
397 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
400 void bdrv_register(BlockDriver *bdrv)
402 assert(bdrv->format_name);
403 GLOBAL_STATE_CODE();
404 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
407 BlockDriverState *bdrv_new(void)
409 BlockDriverState *bs;
410 int i;
412 GLOBAL_STATE_CODE();
414 bs = g_new0(BlockDriverState, 1);
415 QLIST_INIT(&bs->dirty_bitmaps);
416 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
417 QLIST_INIT(&bs->op_blockers[i]);
419 qemu_co_mutex_init(&bs->reqs_lock);
420 qemu_mutex_init(&bs->dirty_bitmap_mutex);
421 bs->refcnt = 1;
422 bs->aio_context = qemu_get_aio_context();
424 qemu_co_queue_init(&bs->flush_queue);
426 qemu_co_mutex_init(&bs->bsc_modify_lock);
427 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
429 for (i = 0; i < bdrv_drain_all_count; i++) {
430 bdrv_drained_begin(bs);
433 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
435 return bs;
438 static BlockDriver *bdrv_do_find_format(const char *format_name)
440 BlockDriver *drv1;
441 GLOBAL_STATE_CODE();
443 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
444 if (!strcmp(drv1->format_name, format_name)) {
445 return drv1;
449 return NULL;
452 BlockDriver *bdrv_find_format(const char *format_name)
454 BlockDriver *drv1;
455 int i;
457 GLOBAL_STATE_CODE();
459 drv1 = bdrv_do_find_format(format_name);
460 if (drv1) {
461 return drv1;
464 /* The driver isn't registered, maybe we need to load a module */
465 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
466 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
467 block_module_load_one(block_driver_modules[i].library_name);
468 break;
472 return bdrv_do_find_format(format_name);
475 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
477 static const char *whitelist_rw[] = {
478 CONFIG_BDRV_RW_WHITELIST
479 NULL
481 static const char *whitelist_ro[] = {
482 CONFIG_BDRV_RO_WHITELIST
483 NULL
485 const char **p;
487 if (!whitelist_rw[0] && !whitelist_ro[0]) {
488 return 1; /* no whitelist, anything goes */
491 for (p = whitelist_rw; *p; p++) {
492 if (!strcmp(format_name, *p)) {
493 return 1;
496 if (read_only) {
497 for (p = whitelist_ro; *p; p++) {
498 if (!strcmp(format_name, *p)) {
499 return 1;
503 return 0;
506 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
508 GLOBAL_STATE_CODE();
509 return bdrv_format_is_whitelisted(drv->format_name, read_only);
512 bool bdrv_uses_whitelist(void)
514 return use_bdrv_whitelist;
517 typedef struct CreateCo {
518 BlockDriver *drv;
519 char *filename;
520 QemuOpts *opts;
521 int ret;
522 Error *err;
523 } CreateCo;
525 static void coroutine_fn bdrv_create_co_entry(void *opaque)
527 Error *local_err = NULL;
528 int ret;
530 CreateCo *cco = opaque;
531 assert(cco->drv);
532 GLOBAL_STATE_CODE();
534 ret = cco->drv->bdrv_co_create_opts(cco->drv,
535 cco->filename, cco->opts, &local_err);
536 error_propagate(&cco->err, local_err);
537 cco->ret = ret;
540 int bdrv_create(BlockDriver *drv, const char* filename,
541 QemuOpts *opts, Error **errp)
543 int ret;
545 GLOBAL_STATE_CODE();
547 Coroutine *co;
548 CreateCo cco = {
549 .drv = drv,
550 .filename = g_strdup(filename),
551 .opts = opts,
552 .ret = NOT_DONE,
553 .err = NULL,
556 if (!drv->bdrv_co_create_opts) {
557 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
558 ret = -ENOTSUP;
559 goto out;
562 if (qemu_in_coroutine()) {
563 /* Fast-path if already in coroutine context */
564 bdrv_create_co_entry(&cco);
565 } else {
566 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
567 qemu_coroutine_enter(co);
568 while (cco.ret == NOT_DONE) {
569 aio_poll(qemu_get_aio_context(), true);
573 ret = cco.ret;
574 if (ret < 0) {
575 if (cco.err) {
576 error_propagate(errp, cco.err);
577 } else {
578 error_setg_errno(errp, -ret, "Could not create image");
582 out:
583 g_free(cco.filename);
584 return ret;
588 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
589 * least the given @minimum_size.
591 * On success, return @blk's actual length.
592 * Otherwise, return -errno.
594 static int64_t create_file_fallback_truncate(BlockBackend *blk,
595 int64_t minimum_size, Error **errp)
597 Error *local_err = NULL;
598 int64_t size;
599 int ret;
601 GLOBAL_STATE_CODE();
603 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
604 &local_err);
605 if (ret < 0 && ret != -ENOTSUP) {
606 error_propagate(errp, local_err);
607 return ret;
610 size = blk_getlength(blk);
611 if (size < 0) {
612 error_free(local_err);
613 error_setg_errno(errp, -size,
614 "Failed to inquire the new image file's length");
615 return size;
618 if (size < minimum_size) {
619 /* Need to grow the image, but we failed to do that */
620 error_propagate(errp, local_err);
621 return -ENOTSUP;
624 error_free(local_err);
625 local_err = NULL;
627 return size;
631 * Helper function for bdrv_create_file_fallback(): Zero the first
632 * sector to remove any potentially pre-existing image header.
634 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
635 int64_t current_size,
636 Error **errp)
638 int64_t bytes_to_clear;
639 int ret;
641 GLOBAL_STATE_CODE();
643 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
644 if (bytes_to_clear) {
645 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
646 if (ret < 0) {
647 error_setg_errno(errp, -ret,
648 "Failed to clear the new image's first sector");
649 return ret;
653 return 0;
657 * Simple implementation of bdrv_co_create_opts for protocol drivers
658 * which only support creation via opening a file
659 * (usually existing raw storage device)
661 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
662 const char *filename,
663 QemuOpts *opts,
664 Error **errp)
666 BlockBackend *blk;
667 QDict *options;
668 int64_t size = 0;
669 char *buf = NULL;
670 PreallocMode prealloc;
671 Error *local_err = NULL;
672 int ret;
674 GLOBAL_STATE_CODE();
676 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
677 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
678 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
679 PREALLOC_MODE_OFF, &local_err);
680 g_free(buf);
681 if (local_err) {
682 error_propagate(errp, local_err);
683 return -EINVAL;
686 if (prealloc != PREALLOC_MODE_OFF) {
687 error_setg(errp, "Unsupported preallocation mode '%s'",
688 PreallocMode_str(prealloc));
689 return -ENOTSUP;
692 options = qdict_new();
693 qdict_put_str(options, "driver", drv->format_name);
695 blk = blk_new_open(filename, NULL, options,
696 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
697 if (!blk) {
698 error_prepend(errp, "Protocol driver '%s' does not support image "
699 "creation, and opening the image failed: ",
700 drv->format_name);
701 return -EINVAL;
704 size = create_file_fallback_truncate(blk, size, errp);
705 if (size < 0) {
706 ret = size;
707 goto out;
710 ret = create_file_fallback_zero_first_sector(blk, size, errp);
711 if (ret < 0) {
712 goto out;
715 ret = 0;
716 out:
717 blk_unref(blk);
718 return ret;
721 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
723 QemuOpts *protocol_opts;
724 BlockDriver *drv;
725 QDict *qdict;
726 int ret;
728 GLOBAL_STATE_CODE();
730 drv = bdrv_find_protocol(filename, true, errp);
731 if (drv == NULL) {
732 return -ENOENT;
735 if (!drv->create_opts) {
736 error_setg(errp, "Driver '%s' does not support image creation",
737 drv->format_name);
738 return -ENOTSUP;
742 * 'opts' contains a QemuOptsList with a combination of format and protocol
743 * default values.
745 * The format properly removes its options, but the default values remain
746 * in 'opts->list'. So if the protocol has options with the same name
747 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
748 * of the format, since for overlapping options, the format wins.
750 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
751 * only the set options, and then convert it back to QemuOpts, using the
752 * create_opts of the protocol. So the new QemuOpts, will contain only the
753 * protocol defaults.
755 qdict = qemu_opts_to_qdict(opts, NULL);
756 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
757 if (protocol_opts == NULL) {
758 ret = -EINVAL;
759 goto out;
762 ret = bdrv_create(drv, filename, protocol_opts, errp);
763 out:
764 qemu_opts_del(protocol_opts);
765 qobject_unref(qdict);
766 return ret;
769 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
771 Error *local_err = NULL;
772 int ret;
774 IO_CODE();
775 assert(bs != NULL);
777 if (!bs->drv) {
778 error_setg(errp, "Block node '%s' is not opened", bs->filename);
779 return -ENOMEDIUM;
782 if (!bs->drv->bdrv_co_delete_file) {
783 error_setg(errp, "Driver '%s' does not support image deletion",
784 bs->drv->format_name);
785 return -ENOTSUP;
788 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
789 if (ret < 0) {
790 error_propagate(errp, local_err);
793 return ret;
796 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
798 Error *local_err = NULL;
799 int ret;
800 IO_CODE();
802 if (!bs) {
803 return;
806 ret = bdrv_co_delete_file(bs, &local_err);
808 * ENOTSUP will happen if the block driver doesn't support
809 * the 'bdrv_co_delete_file' interface. This is a predictable
810 * scenario and shouldn't be reported back to the user.
812 if (ret == -ENOTSUP) {
813 error_free(local_err);
814 } else if (ret < 0) {
815 error_report_err(local_err);
820 * Try to get @bs's logical and physical block size.
821 * On success, store them in @bsz struct and return 0.
822 * On failure return -errno.
823 * @bs must not be empty.
825 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
827 BlockDriver *drv = bs->drv;
828 BlockDriverState *filtered = bdrv_filter_bs(bs);
829 GLOBAL_STATE_CODE();
831 if (drv && drv->bdrv_probe_blocksizes) {
832 return drv->bdrv_probe_blocksizes(bs, bsz);
833 } else if (filtered) {
834 return bdrv_probe_blocksizes(filtered, bsz);
837 return -ENOTSUP;
841 * Try to get @bs's geometry (cyls, heads, sectors).
842 * On success, store them in @geo struct and return 0.
843 * On failure return -errno.
844 * @bs must not be empty.
846 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
848 BlockDriver *drv = bs->drv;
849 BlockDriverState *filtered = bdrv_filter_bs(bs);
850 GLOBAL_STATE_CODE();
852 if (drv && drv->bdrv_probe_geometry) {
853 return drv->bdrv_probe_geometry(bs, geo);
854 } else if (filtered) {
855 return bdrv_probe_geometry(filtered, geo);
858 return -ENOTSUP;
862 * Create a uniquely-named empty temporary file.
863 * Return 0 upon success, otherwise a negative errno value.
865 int get_tmp_filename(char *filename, int size)
867 #ifdef _WIN32
868 char temp_dir[MAX_PATH];
869 /* GetTempFileName requires that its output buffer (4th param)
870 have length MAX_PATH or greater. */
871 assert(size >= MAX_PATH);
872 return (GetTempPath(MAX_PATH, temp_dir)
873 && GetTempFileName(temp_dir, "qem", 0, filename)
874 ? 0 : -GetLastError());
875 #else
876 int fd;
877 const char *tmpdir;
878 tmpdir = getenv("TMPDIR");
879 if (!tmpdir) {
880 tmpdir = "/var/tmp";
882 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
883 return -EOVERFLOW;
885 fd = mkstemp(filename);
886 if (fd < 0) {
887 return -errno;
889 if (close(fd) != 0) {
890 unlink(filename);
891 return -errno;
893 return 0;
894 #endif
898 * Detect host devices. By convention, /dev/cdrom[N] is always
899 * recognized as a host CDROM.
901 static BlockDriver *find_hdev_driver(const char *filename)
903 int score_max = 0, score;
904 BlockDriver *drv = NULL, *d;
905 GLOBAL_STATE_CODE();
907 QLIST_FOREACH(d, &bdrv_drivers, list) {
908 if (d->bdrv_probe_device) {
909 score = d->bdrv_probe_device(filename);
910 if (score > score_max) {
911 score_max = score;
912 drv = d;
917 return drv;
920 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
922 BlockDriver *drv1;
923 GLOBAL_STATE_CODE();
925 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
926 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
927 return drv1;
931 return NULL;
934 BlockDriver *bdrv_find_protocol(const char *filename,
935 bool allow_protocol_prefix,
936 Error **errp)
938 BlockDriver *drv1;
939 char protocol[128];
940 int len;
941 const char *p;
942 int i;
944 GLOBAL_STATE_CODE();
945 /* TODO Drivers without bdrv_file_open must be specified explicitly */
948 * XXX(hch): we really should not let host device detection
949 * override an explicit protocol specification, but moving this
950 * later breaks access to device names with colons in them.
951 * Thanks to the brain-dead persistent naming schemes on udev-
952 * based Linux systems those actually are quite common.
954 drv1 = find_hdev_driver(filename);
955 if (drv1) {
956 return drv1;
959 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
960 return &bdrv_file;
963 p = strchr(filename, ':');
964 assert(p != NULL);
965 len = p - filename;
966 if (len > sizeof(protocol) - 1)
967 len = sizeof(protocol) - 1;
968 memcpy(protocol, filename, len);
969 protocol[len] = '\0';
971 drv1 = bdrv_do_find_protocol(protocol);
972 if (drv1) {
973 return drv1;
976 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
977 if (block_driver_modules[i].protocol_name &&
978 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
979 block_module_load_one(block_driver_modules[i].library_name);
980 break;
984 drv1 = bdrv_do_find_protocol(protocol);
985 if (!drv1) {
986 error_setg(errp, "Unknown protocol '%s'", protocol);
988 return drv1;
992 * Guess image format by probing its contents.
993 * This is not a good idea when your image is raw (CVE-2008-2004), but
994 * we do it anyway for backward compatibility.
996 * @buf contains the image's first @buf_size bytes.
997 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
998 * but can be smaller if the image file is smaller)
999 * @filename is its filename.
1001 * For all block drivers, call the bdrv_probe() method to get its
1002 * probing score.
1003 * Return the first block driver with the highest probing score.
1005 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
1006 const char *filename)
1008 int score_max = 0, score;
1009 BlockDriver *drv = NULL, *d;
1010 IO_CODE();
1012 QLIST_FOREACH(d, &bdrv_drivers, list) {
1013 if (d->bdrv_probe) {
1014 score = d->bdrv_probe(buf, buf_size, filename);
1015 if (score > score_max) {
1016 score_max = score;
1017 drv = d;
1022 return drv;
1025 static int find_image_format(BlockBackend *file, const char *filename,
1026 BlockDriver **pdrv, Error **errp)
1028 BlockDriver *drv;
1029 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1030 int ret = 0;
1032 GLOBAL_STATE_CODE();
1034 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1035 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1036 *pdrv = &bdrv_raw;
1037 return ret;
1040 ret = blk_pread(file, 0, buf, sizeof(buf));
1041 if (ret < 0) {
1042 error_setg_errno(errp, -ret, "Could not read image for determining its "
1043 "format");
1044 *pdrv = NULL;
1045 return ret;
1048 drv = bdrv_probe_all(buf, ret, filename);
1049 if (!drv) {
1050 error_setg(errp, "Could not determine image format: No compatible "
1051 "driver found");
1052 ret = -ENOENT;
1054 *pdrv = drv;
1055 return ret;
1059 * Set the current 'total_sectors' value
1060 * Return 0 on success, -errno on error.
1062 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
1064 BlockDriver *drv = bs->drv;
1065 IO_CODE();
1067 if (!drv) {
1068 return -ENOMEDIUM;
1071 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1072 if (bdrv_is_sg(bs))
1073 return 0;
1075 /* query actual device if possible, otherwise just trust the hint */
1076 if (drv->bdrv_getlength) {
1077 int64_t length = drv->bdrv_getlength(bs);
1078 if (length < 0) {
1079 return length;
1081 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1084 bs->total_sectors = hint;
1086 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1087 return -EFBIG;
1090 return 0;
1094 * Combines a QDict of new block driver @options with any missing options taken
1095 * from @old_options, so that leaving out an option defaults to its old value.
1097 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1098 QDict *old_options)
1100 GLOBAL_STATE_CODE();
1101 if (bs->drv && bs->drv->bdrv_join_options) {
1102 bs->drv->bdrv_join_options(options, old_options);
1103 } else {
1104 qdict_join(options, old_options, false);
1108 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1109 int open_flags,
1110 Error **errp)
1112 Error *local_err = NULL;
1113 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1114 BlockdevDetectZeroesOptions detect_zeroes =
1115 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1116 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1117 GLOBAL_STATE_CODE();
1118 g_free(value);
1119 if (local_err) {
1120 error_propagate(errp, local_err);
1121 return detect_zeroes;
1124 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1125 !(open_flags & BDRV_O_UNMAP))
1127 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1128 "without setting discard operation to unmap");
1131 return detect_zeroes;
1135 * Set open flags for aio engine
1137 * Return 0 on success, -1 if the engine specified is invalid
1139 int bdrv_parse_aio(const char *mode, int *flags)
1141 if (!strcmp(mode, "threads")) {
1142 /* do nothing, default */
1143 } else if (!strcmp(mode, "native")) {
1144 *flags |= BDRV_O_NATIVE_AIO;
1145 #ifdef CONFIG_LINUX_IO_URING
1146 } else if (!strcmp(mode, "io_uring")) {
1147 *flags |= BDRV_O_IO_URING;
1148 #endif
1149 } else {
1150 return -1;
1153 return 0;
1157 * Set open flags for a given discard mode
1159 * Return 0 on success, -1 if the discard mode was invalid.
1161 int bdrv_parse_discard_flags(const char *mode, int *flags)
1163 *flags &= ~BDRV_O_UNMAP;
1165 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1166 /* do nothing */
1167 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1168 *flags |= BDRV_O_UNMAP;
1169 } else {
1170 return -1;
1173 return 0;
1177 * Set open flags for a given cache mode
1179 * Return 0 on success, -1 if the cache mode was invalid.
1181 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1183 *flags &= ~BDRV_O_CACHE_MASK;
1185 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1186 *writethrough = false;
1187 *flags |= BDRV_O_NOCACHE;
1188 } else if (!strcmp(mode, "directsync")) {
1189 *writethrough = true;
1190 *flags |= BDRV_O_NOCACHE;
1191 } else if (!strcmp(mode, "writeback")) {
1192 *writethrough = false;
1193 } else if (!strcmp(mode, "unsafe")) {
1194 *writethrough = false;
1195 *flags |= BDRV_O_NO_FLUSH;
1196 } else if (!strcmp(mode, "writethrough")) {
1197 *writethrough = true;
1198 } else {
1199 return -1;
1202 return 0;
1205 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1207 BlockDriverState *parent = c->opaque;
1208 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1211 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1213 BlockDriverState *bs = child->opaque;
1214 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1217 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1219 BlockDriverState *bs = child->opaque;
1220 return bdrv_drain_poll(bs, false, NULL, false);
1223 static void bdrv_child_cb_drained_end(BdrvChild *child,
1224 int *drained_end_counter)
1226 BlockDriverState *bs = child->opaque;
1227 bdrv_drained_end_no_poll(bs, drained_end_counter);
1230 static int bdrv_child_cb_inactivate(BdrvChild *child)
1232 BlockDriverState *bs = child->opaque;
1233 GLOBAL_STATE_CODE();
1234 assert(bs->open_flags & BDRV_O_INACTIVE);
1235 return 0;
1238 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1239 GSList **ignore, Error **errp)
1241 BlockDriverState *bs = child->opaque;
1242 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1245 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1246 GSList **ignore)
1248 BlockDriverState *bs = child->opaque;
1249 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1253 * Returns the options and flags that a temporary snapshot should get, based on
1254 * the originally requested flags (the originally requested image will have
1255 * flags like a backing file)
1257 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1258 int parent_flags, QDict *parent_options)
1260 GLOBAL_STATE_CODE();
1261 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1263 /* For temporary files, unconditional cache=unsafe is fine */
1264 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1265 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1267 /* Copy the read-only and discard options from the parent */
1268 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1269 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1271 /* aio=native doesn't work for cache.direct=off, so disable it for the
1272 * temporary snapshot */
1273 *child_flags &= ~BDRV_O_NATIVE_AIO;
1276 static void bdrv_backing_attach(BdrvChild *c)
1278 BlockDriverState *parent = c->opaque;
1279 BlockDriverState *backing_hd = c->bs;
1281 GLOBAL_STATE_CODE();
1282 assert(!parent->backing_blocker);
1283 error_setg(&parent->backing_blocker,
1284 "node is used as backing hd of '%s'",
1285 bdrv_get_device_or_node_name(parent));
1287 bdrv_refresh_filename(backing_hd);
1289 parent->open_flags &= ~BDRV_O_NO_BACKING;
1291 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1292 /* Otherwise we won't be able to commit or stream */
1293 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1294 parent->backing_blocker);
1295 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1296 parent->backing_blocker);
1298 * We do backup in 3 ways:
1299 * 1. drive backup
1300 * The target bs is new opened, and the source is top BDS
1301 * 2. blockdev backup
1302 * Both the source and the target are top BDSes.
1303 * 3. internal backup(used for block replication)
1304 * Both the source and the target are backing file
1306 * In case 1 and 2, neither the source nor the target is the backing file.
1307 * In case 3, we will block the top BDS, so there is only one block job
1308 * for the top BDS and its backing chain.
1310 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1311 parent->backing_blocker);
1312 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1313 parent->backing_blocker);
1316 static void bdrv_backing_detach(BdrvChild *c)
1318 BlockDriverState *parent = c->opaque;
1320 GLOBAL_STATE_CODE();
1321 assert(parent->backing_blocker);
1322 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1323 error_free(parent->backing_blocker);
1324 parent->backing_blocker = NULL;
1327 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1328 const char *filename, Error **errp)
1330 BlockDriverState *parent = c->opaque;
1331 bool read_only = bdrv_is_read_only(parent);
1332 int ret;
1333 GLOBAL_STATE_CODE();
1335 if (read_only) {
1336 ret = bdrv_reopen_set_read_only(parent, false, errp);
1337 if (ret < 0) {
1338 return ret;
1342 ret = bdrv_change_backing_file(parent, filename,
1343 base->drv ? base->drv->format_name : "",
1344 false);
1345 if (ret < 0) {
1346 error_setg_errno(errp, -ret, "Could not update backing file link");
1349 if (read_only) {
1350 bdrv_reopen_set_read_only(parent, true, NULL);
1353 return ret;
1357 * Returns the options and flags that a generic child of a BDS should
1358 * get, based on the given options and flags for the parent BDS.
1360 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1361 int *child_flags, QDict *child_options,
1362 int parent_flags, QDict *parent_options)
1364 int flags = parent_flags;
1365 GLOBAL_STATE_CODE();
1368 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1369 * Generally, the question to answer is: Should this child be
1370 * format-probed by default?
1374 * Pure and non-filtered data children of non-format nodes should
1375 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1376 * set). This only affects a very limited set of drivers (namely
1377 * quorum and blkverify when this comment was written).
1378 * Force-clear BDRV_O_PROTOCOL then.
1380 if (!parent_is_format &&
1381 (role & BDRV_CHILD_DATA) &&
1382 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1384 flags &= ~BDRV_O_PROTOCOL;
1388 * All children of format nodes (except for COW children) and all
1389 * metadata children in general should never be format-probed.
1390 * Force-set BDRV_O_PROTOCOL then.
1392 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1393 (role & BDRV_CHILD_METADATA))
1395 flags |= BDRV_O_PROTOCOL;
1399 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1400 * the parent.
1402 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1403 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1404 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1406 if (role & BDRV_CHILD_COW) {
1407 /* backing files are opened read-only by default */
1408 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1409 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1410 } else {
1411 /* Inherit the read-only option from the parent if it's not set */
1412 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1413 qdict_copy_default(child_options, parent_options,
1414 BDRV_OPT_AUTO_READ_ONLY);
1418 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1419 * can default to enable it on lower layers regardless of the
1420 * parent option.
1422 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1424 /* Clear flags that only apply to the top layer */
1425 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1427 if (role & BDRV_CHILD_METADATA) {
1428 flags &= ~BDRV_O_NO_IO;
1430 if (role & BDRV_CHILD_COW) {
1431 flags &= ~BDRV_O_TEMPORARY;
1434 *child_flags = flags;
1437 static void bdrv_child_cb_attach(BdrvChild *child)
1439 BlockDriverState *bs = child->opaque;
1441 assert_bdrv_graph_writable(bs);
1442 QLIST_INSERT_HEAD(&bs->children, child, next);
1444 if (child->role & BDRV_CHILD_COW) {
1445 bdrv_backing_attach(child);
1448 bdrv_apply_subtree_drain(child, bs);
1451 static void bdrv_child_cb_detach(BdrvChild *child)
1453 BlockDriverState *bs = child->opaque;
1455 if (child->role & BDRV_CHILD_COW) {
1456 bdrv_backing_detach(child);
1459 bdrv_unapply_subtree_drain(child, bs);
1461 assert_bdrv_graph_writable(bs);
1462 QLIST_REMOVE(child, next);
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 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1493 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1494 .update_filename = bdrv_child_cb_update_filename,
1495 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1498 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1500 GLOBAL_STATE_CODE();
1501 return c->klass->get_parent_aio_context(c);
1504 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1506 int open_flags = flags;
1507 GLOBAL_STATE_CODE();
1510 * Clear flags that are internal to the block layer before opening the
1511 * image.
1513 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1515 return open_flags;
1518 static void update_flags_from_options(int *flags, QemuOpts *opts)
1520 GLOBAL_STATE_CODE();
1522 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1524 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1525 *flags |= BDRV_O_NO_FLUSH;
1528 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1529 *flags |= BDRV_O_NOCACHE;
1532 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1533 *flags |= BDRV_O_RDWR;
1536 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1537 *flags |= BDRV_O_AUTO_RDONLY;
1541 static void update_options_from_flags(QDict *options, int flags)
1543 GLOBAL_STATE_CODE();
1544 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1545 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1547 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1548 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1549 flags & BDRV_O_NO_FLUSH);
1551 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1552 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1554 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1555 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1556 flags & BDRV_O_AUTO_RDONLY);
1560 static void bdrv_assign_node_name(BlockDriverState *bs,
1561 const char *node_name,
1562 Error **errp)
1564 char *gen_node_name = NULL;
1565 GLOBAL_STATE_CODE();
1567 if (!node_name) {
1568 node_name = gen_node_name = id_generate(ID_BLOCK);
1569 } else if (!id_wellformed(node_name)) {
1571 * Check for empty string or invalid characters, but not if it is
1572 * generated (generated names use characters not available to the user)
1574 error_setg(errp, "Invalid node-name: '%s'", node_name);
1575 return;
1578 /* takes care of avoiding namespaces collisions */
1579 if (blk_by_name(node_name)) {
1580 error_setg(errp, "node-name=%s is conflicting with a device id",
1581 node_name);
1582 goto out;
1585 /* takes care of avoiding duplicates node names */
1586 if (bdrv_find_node(node_name)) {
1587 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1588 goto out;
1591 /* Make sure that the node name isn't truncated */
1592 if (strlen(node_name) >= sizeof(bs->node_name)) {
1593 error_setg(errp, "Node name too long");
1594 goto out;
1597 /* copy node name into the bs and insert it into the graph list */
1598 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1599 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1600 out:
1601 g_free(gen_node_name);
1604 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1605 const char *node_name, QDict *options,
1606 int open_flags, Error **errp)
1608 Error *local_err = NULL;
1609 int i, ret;
1610 GLOBAL_STATE_CODE();
1612 bdrv_assign_node_name(bs, node_name, &local_err);
1613 if (local_err) {
1614 error_propagate(errp, local_err);
1615 return -EINVAL;
1618 bs->drv = drv;
1619 bs->opaque = g_malloc0(drv->instance_size);
1621 if (drv->bdrv_file_open) {
1622 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1623 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1624 } else if (drv->bdrv_open) {
1625 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1626 } else {
1627 ret = 0;
1630 if (ret < 0) {
1631 if (local_err) {
1632 error_propagate(errp, local_err);
1633 } else if (bs->filename[0]) {
1634 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1635 } else {
1636 error_setg_errno(errp, -ret, "Could not open image");
1638 goto open_failed;
1641 ret = refresh_total_sectors(bs, bs->total_sectors);
1642 if (ret < 0) {
1643 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1644 return ret;
1647 bdrv_refresh_limits(bs, NULL, &local_err);
1648 if (local_err) {
1649 error_propagate(errp, local_err);
1650 return -EINVAL;
1653 assert(bdrv_opt_mem_align(bs) != 0);
1654 assert(bdrv_min_mem_align(bs) != 0);
1655 assert(is_power_of_2(bs->bl.request_alignment));
1657 for (i = 0; i < bs->quiesce_counter; i++) {
1658 if (drv->bdrv_co_drain_begin) {
1659 drv->bdrv_co_drain_begin(bs);
1663 return 0;
1664 open_failed:
1665 bs->drv = NULL;
1666 if (bs->file != NULL) {
1667 bdrv_unref_child(bs, bs->file);
1668 bs->file = NULL;
1670 g_free(bs->opaque);
1671 bs->opaque = NULL;
1672 return ret;
1676 * Create and open a block node.
1678 * @options is a QDict of options to pass to the block drivers, or NULL for an
1679 * empty set of options. The reference to the QDict belongs to the block layer
1680 * after the call (even on failure), so if the caller intends to reuse the
1681 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1683 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1684 const char *node_name,
1685 QDict *options, int flags,
1686 Error **errp)
1688 BlockDriverState *bs;
1689 int ret;
1691 GLOBAL_STATE_CODE();
1693 bs = bdrv_new();
1694 bs->open_flags = flags;
1695 bs->options = options ?: qdict_new();
1696 bs->explicit_options = qdict_clone_shallow(bs->options);
1697 bs->opaque = NULL;
1699 update_options_from_flags(bs->options, flags);
1701 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1702 if (ret < 0) {
1703 qobject_unref(bs->explicit_options);
1704 bs->explicit_options = NULL;
1705 qobject_unref(bs->options);
1706 bs->options = NULL;
1707 bdrv_unref(bs);
1708 return NULL;
1711 return bs;
1714 /* Create and open a block node. */
1715 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1716 int flags, Error **errp)
1718 GLOBAL_STATE_CODE();
1719 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1722 QemuOptsList bdrv_runtime_opts = {
1723 .name = "bdrv_common",
1724 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1725 .desc = {
1727 .name = "node-name",
1728 .type = QEMU_OPT_STRING,
1729 .help = "Node name of the block device node",
1732 .name = "driver",
1733 .type = QEMU_OPT_STRING,
1734 .help = "Block driver to use for the node",
1737 .name = BDRV_OPT_CACHE_DIRECT,
1738 .type = QEMU_OPT_BOOL,
1739 .help = "Bypass software writeback cache on the host",
1742 .name = BDRV_OPT_CACHE_NO_FLUSH,
1743 .type = QEMU_OPT_BOOL,
1744 .help = "Ignore flush requests",
1747 .name = BDRV_OPT_READ_ONLY,
1748 .type = QEMU_OPT_BOOL,
1749 .help = "Node is opened in read-only mode",
1752 .name = BDRV_OPT_AUTO_READ_ONLY,
1753 .type = QEMU_OPT_BOOL,
1754 .help = "Node can become read-only if opening read-write fails",
1757 .name = "detect-zeroes",
1758 .type = QEMU_OPT_STRING,
1759 .help = "try to optimize zero writes (off, on, unmap)",
1762 .name = BDRV_OPT_DISCARD,
1763 .type = QEMU_OPT_STRING,
1764 .help = "discard operation (ignore/off, unmap/on)",
1767 .name = BDRV_OPT_FORCE_SHARE,
1768 .type = QEMU_OPT_BOOL,
1769 .help = "always accept other writers (default: off)",
1771 { /* end of list */ }
1775 QemuOptsList bdrv_create_opts_simple = {
1776 .name = "simple-create-opts",
1777 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1778 .desc = {
1780 .name = BLOCK_OPT_SIZE,
1781 .type = QEMU_OPT_SIZE,
1782 .help = "Virtual disk size"
1785 .name = BLOCK_OPT_PREALLOC,
1786 .type = QEMU_OPT_STRING,
1787 .help = "Preallocation mode (allowed values: off)"
1789 { /* end of list */ }
1794 * Common part for opening disk images and files
1796 * Removes all processed options from *options.
1798 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1799 QDict *options, Error **errp)
1801 int ret, open_flags;
1802 const char *filename;
1803 const char *driver_name = NULL;
1804 const char *node_name = NULL;
1805 const char *discard;
1806 QemuOpts *opts;
1807 BlockDriver *drv;
1808 Error *local_err = NULL;
1809 bool ro;
1811 assert(bs->file == NULL);
1812 assert(options != NULL && bs->options != options);
1813 GLOBAL_STATE_CODE();
1815 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1816 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1817 ret = -EINVAL;
1818 goto fail_opts;
1821 update_flags_from_options(&bs->open_flags, opts);
1823 driver_name = qemu_opt_get(opts, "driver");
1824 drv = bdrv_find_format(driver_name);
1825 assert(drv != NULL);
1827 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1829 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1830 error_setg(errp,
1831 BDRV_OPT_FORCE_SHARE
1832 "=on can only be used with read-only images");
1833 ret = -EINVAL;
1834 goto fail_opts;
1837 if (file != NULL) {
1838 bdrv_refresh_filename(blk_bs(file));
1839 filename = blk_bs(file)->filename;
1840 } else {
1842 * Caution: while qdict_get_try_str() is fine, getting
1843 * non-string types would require more care. When @options
1844 * come from -blockdev or blockdev_add, its members are typed
1845 * according to the QAPI schema, but when they come from
1846 * -drive, they're all QString.
1848 filename = qdict_get_try_str(options, "filename");
1851 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1852 error_setg(errp, "The '%s' block driver requires a file name",
1853 drv->format_name);
1854 ret = -EINVAL;
1855 goto fail_opts;
1858 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1859 drv->format_name);
1861 ro = bdrv_is_read_only(bs);
1863 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1864 if (!ro && bdrv_is_whitelisted(drv, true)) {
1865 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1866 } else {
1867 ret = -ENOTSUP;
1869 if (ret < 0) {
1870 error_setg(errp,
1871 !ro && bdrv_is_whitelisted(drv, true)
1872 ? "Driver '%s' can only be used for read-only devices"
1873 : "Driver '%s' is not whitelisted",
1874 drv->format_name);
1875 goto fail_opts;
1879 /* bdrv_new() and bdrv_close() make it so */
1880 assert(qatomic_read(&bs->copy_on_read) == 0);
1882 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1883 if (!ro) {
1884 bdrv_enable_copy_on_read(bs);
1885 } else {
1886 error_setg(errp, "Can't use copy-on-read on read-only device");
1887 ret = -EINVAL;
1888 goto fail_opts;
1892 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1893 if (discard != NULL) {
1894 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1895 error_setg(errp, "Invalid discard option");
1896 ret = -EINVAL;
1897 goto fail_opts;
1901 bs->detect_zeroes =
1902 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1903 if (local_err) {
1904 error_propagate(errp, local_err);
1905 ret = -EINVAL;
1906 goto fail_opts;
1909 if (filename != NULL) {
1910 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1911 } else {
1912 bs->filename[0] = '\0';
1914 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1916 /* Open the image, either directly or using a protocol */
1917 open_flags = bdrv_open_flags(bs, bs->open_flags);
1918 node_name = qemu_opt_get(opts, "node-name");
1920 assert(!drv->bdrv_file_open || file == NULL);
1921 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1922 if (ret < 0) {
1923 goto fail_opts;
1926 qemu_opts_del(opts);
1927 return 0;
1929 fail_opts:
1930 qemu_opts_del(opts);
1931 return ret;
1934 static QDict *parse_json_filename(const char *filename, Error **errp)
1936 QObject *options_obj;
1937 QDict *options;
1938 int ret;
1939 GLOBAL_STATE_CODE();
1941 ret = strstart(filename, "json:", &filename);
1942 assert(ret);
1944 options_obj = qobject_from_json(filename, errp);
1945 if (!options_obj) {
1946 error_prepend(errp, "Could not parse the JSON options: ");
1947 return NULL;
1950 options = qobject_to(QDict, options_obj);
1951 if (!options) {
1952 qobject_unref(options_obj);
1953 error_setg(errp, "Invalid JSON object given");
1954 return NULL;
1957 qdict_flatten(options);
1959 return options;
1962 static void parse_json_protocol(QDict *options, const char **pfilename,
1963 Error **errp)
1965 QDict *json_options;
1966 Error *local_err = NULL;
1967 GLOBAL_STATE_CODE();
1969 /* Parse json: pseudo-protocol */
1970 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1971 return;
1974 json_options = parse_json_filename(*pfilename, &local_err);
1975 if (local_err) {
1976 error_propagate(errp, local_err);
1977 return;
1980 /* Options given in the filename have lower priority than options
1981 * specified directly */
1982 qdict_join(options, json_options, false);
1983 qobject_unref(json_options);
1984 *pfilename = NULL;
1988 * Fills in default options for opening images and converts the legacy
1989 * filename/flags pair to option QDict entries.
1990 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1991 * block driver has been specified explicitly.
1993 static int bdrv_fill_options(QDict **options, const char *filename,
1994 int *flags, Error **errp)
1996 const char *drvname;
1997 bool protocol = *flags & BDRV_O_PROTOCOL;
1998 bool parse_filename = false;
1999 BlockDriver *drv = NULL;
2000 Error *local_err = NULL;
2002 GLOBAL_STATE_CODE();
2005 * Caution: while qdict_get_try_str() is fine, getting non-string
2006 * types would require more care. When @options come from
2007 * -blockdev or blockdev_add, its members are typed according to
2008 * the QAPI schema, but when they come from -drive, they're all
2009 * QString.
2011 drvname = qdict_get_try_str(*options, "driver");
2012 if (drvname) {
2013 drv = bdrv_find_format(drvname);
2014 if (!drv) {
2015 error_setg(errp, "Unknown driver '%s'", drvname);
2016 return -ENOENT;
2018 /* If the user has explicitly specified the driver, this choice should
2019 * override the BDRV_O_PROTOCOL flag */
2020 protocol = drv->bdrv_file_open;
2023 if (protocol) {
2024 *flags |= BDRV_O_PROTOCOL;
2025 } else {
2026 *flags &= ~BDRV_O_PROTOCOL;
2029 /* Translate cache options from flags into options */
2030 update_options_from_flags(*options, *flags);
2032 /* Fetch the file name from the options QDict if necessary */
2033 if (protocol && filename) {
2034 if (!qdict_haskey(*options, "filename")) {
2035 qdict_put_str(*options, "filename", filename);
2036 parse_filename = true;
2037 } else {
2038 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2039 "the same time");
2040 return -EINVAL;
2044 /* Find the right block driver */
2045 /* See cautionary note on accessing @options above */
2046 filename = qdict_get_try_str(*options, "filename");
2048 if (!drvname && protocol) {
2049 if (filename) {
2050 drv = bdrv_find_protocol(filename, parse_filename, errp);
2051 if (!drv) {
2052 return -EINVAL;
2055 drvname = drv->format_name;
2056 qdict_put_str(*options, "driver", drvname);
2057 } else {
2058 error_setg(errp, "Must specify either driver or file");
2059 return -EINVAL;
2063 assert(drv || !protocol);
2065 /* Driver-specific filename parsing */
2066 if (drv && drv->bdrv_parse_filename && parse_filename) {
2067 drv->bdrv_parse_filename(filename, *options, &local_err);
2068 if (local_err) {
2069 error_propagate(errp, local_err);
2070 return -EINVAL;
2073 if (!drv->bdrv_needs_filename) {
2074 qdict_del(*options, "filename");
2078 return 0;
2081 typedef struct BlockReopenQueueEntry {
2082 bool prepared;
2083 bool perms_checked;
2084 BDRVReopenState state;
2085 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2086 } BlockReopenQueueEntry;
2089 * Return the flags that @bs will have after the reopens in @q have
2090 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2091 * return the current flags.
2093 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2095 BlockReopenQueueEntry *entry;
2097 if (q != NULL) {
2098 QTAILQ_FOREACH(entry, q, entry) {
2099 if (entry->state.bs == bs) {
2100 return entry->state.flags;
2105 return bs->open_flags;
2108 /* Returns whether the image file can be written to after the reopen queue @q
2109 * has been successfully applied, or right now if @q is NULL. */
2110 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2111 BlockReopenQueue *q)
2113 int flags = bdrv_reopen_get_flags(q, bs);
2115 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2119 * Return whether the BDS can be written to. This is not necessarily
2120 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2121 * be written to but do not count as read-only images.
2123 bool bdrv_is_writable(BlockDriverState *bs)
2125 IO_CODE();
2126 return bdrv_is_writable_after_reopen(bs, NULL);
2129 static char *bdrv_child_user_desc(BdrvChild *c)
2131 GLOBAL_STATE_CODE();
2132 return c->klass->get_parent_desc(c);
2136 * Check that @a allows everything that @b needs. @a and @b must reference same
2137 * child node.
2139 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2141 const char *child_bs_name;
2142 g_autofree char *a_user = NULL;
2143 g_autofree char *b_user = NULL;
2144 g_autofree char *perms = NULL;
2146 assert(a->bs);
2147 assert(a->bs == b->bs);
2148 GLOBAL_STATE_CODE();
2150 if ((b->perm & a->shared_perm) == b->perm) {
2151 return true;
2154 child_bs_name = bdrv_get_node_name(b->bs);
2155 a_user = bdrv_child_user_desc(a);
2156 b_user = bdrv_child_user_desc(b);
2157 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2159 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2160 "both required by %s (uses node '%s' as '%s' child) and "
2161 "unshared by %s (uses node '%s' as '%s' child).",
2162 child_bs_name, perms,
2163 b_user, child_bs_name, b->name,
2164 a_user, child_bs_name, a->name);
2166 return false;
2169 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2171 BdrvChild *a, *b;
2172 GLOBAL_STATE_CODE();
2175 * During the loop we'll look at each pair twice. That's correct because
2176 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2177 * directions.
2179 QLIST_FOREACH(a, &bs->parents, next_parent) {
2180 QLIST_FOREACH(b, &bs->parents, next_parent) {
2181 if (a == b) {
2182 continue;
2185 if (!bdrv_a_allow_b(a, b, errp)) {
2186 return true;
2191 return false;
2194 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2195 BdrvChild *c, BdrvChildRole role,
2196 BlockReopenQueue *reopen_queue,
2197 uint64_t parent_perm, uint64_t parent_shared,
2198 uint64_t *nperm, uint64_t *nshared)
2200 assert(bs->drv && bs->drv->bdrv_child_perm);
2201 GLOBAL_STATE_CODE();
2202 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2203 parent_perm, parent_shared,
2204 nperm, nshared);
2205 /* TODO Take force_share from reopen_queue */
2206 if (child_bs && child_bs->force_share) {
2207 *nshared = BLK_PERM_ALL;
2212 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2213 * nodes that are already in the @list, of course) so that final list is
2214 * topologically sorted. Return the result (GSList @list object is updated, so
2215 * don't use old reference after function call).
2217 * On function start @list must be already topologically sorted and for any node
2218 * in the @list the whole subtree of the node must be in the @list as well. The
2219 * simplest way to satisfy this criteria: use only result of
2220 * bdrv_topological_dfs() or NULL as @list parameter.
2222 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2223 BlockDriverState *bs)
2225 BdrvChild *child;
2226 g_autoptr(GHashTable) local_found = NULL;
2228 GLOBAL_STATE_CODE();
2230 if (!found) {
2231 assert(!list);
2232 found = local_found = g_hash_table_new(NULL, NULL);
2235 if (g_hash_table_contains(found, bs)) {
2236 return list;
2238 g_hash_table_add(found, bs);
2240 QLIST_FOREACH(child, &bs->children, next) {
2241 list = bdrv_topological_dfs(list, found, child->bs);
2244 return g_slist_prepend(list, bs);
2247 typedef struct BdrvChildSetPermState {
2248 BdrvChild *child;
2249 uint64_t old_perm;
2250 uint64_t old_shared_perm;
2251 } BdrvChildSetPermState;
2253 static void bdrv_child_set_perm_abort(void *opaque)
2255 BdrvChildSetPermState *s = opaque;
2257 GLOBAL_STATE_CODE();
2259 s->child->perm = s->old_perm;
2260 s->child->shared_perm = s->old_shared_perm;
2263 static TransactionActionDrv bdrv_child_set_pem_drv = {
2264 .abort = bdrv_child_set_perm_abort,
2265 .clean = g_free,
2268 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2269 uint64_t shared, Transaction *tran)
2271 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2272 GLOBAL_STATE_CODE();
2274 *s = (BdrvChildSetPermState) {
2275 .child = c,
2276 .old_perm = c->perm,
2277 .old_shared_perm = c->shared_perm,
2280 c->perm = perm;
2281 c->shared_perm = shared;
2283 tran_add(tran, &bdrv_child_set_pem_drv, s);
2286 static void bdrv_drv_set_perm_commit(void *opaque)
2288 BlockDriverState *bs = opaque;
2289 uint64_t cumulative_perms, cumulative_shared_perms;
2290 GLOBAL_STATE_CODE();
2292 if (bs->drv->bdrv_set_perm) {
2293 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2294 &cumulative_shared_perms);
2295 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2299 static void bdrv_drv_set_perm_abort(void *opaque)
2301 BlockDriverState *bs = opaque;
2302 GLOBAL_STATE_CODE();
2304 if (bs->drv->bdrv_abort_perm_update) {
2305 bs->drv->bdrv_abort_perm_update(bs);
2309 TransactionActionDrv bdrv_drv_set_perm_drv = {
2310 .abort = bdrv_drv_set_perm_abort,
2311 .commit = bdrv_drv_set_perm_commit,
2314 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2315 uint64_t shared_perm, Transaction *tran,
2316 Error **errp)
2318 GLOBAL_STATE_CODE();
2319 if (!bs->drv) {
2320 return 0;
2323 if (bs->drv->bdrv_check_perm) {
2324 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2325 if (ret < 0) {
2326 return ret;
2330 if (tran) {
2331 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2334 return 0;
2337 typedef struct BdrvReplaceChildState {
2338 BdrvChild *child;
2339 BdrvChild **childp;
2340 BlockDriverState *old_bs;
2341 bool free_empty_child;
2342 } BdrvReplaceChildState;
2344 static void bdrv_replace_child_commit(void *opaque)
2346 BdrvReplaceChildState *s = opaque;
2347 GLOBAL_STATE_CODE();
2349 if (s->free_empty_child && !s->child->bs) {
2350 bdrv_child_free(s->child);
2352 bdrv_unref(s->old_bs);
2355 static void bdrv_replace_child_abort(void *opaque)
2357 BdrvReplaceChildState *s = opaque;
2358 BlockDriverState *new_bs = s->child->bs;
2360 GLOBAL_STATE_CODE();
2362 * old_bs reference is transparently moved from @s to s->child.
2364 * Pass &s->child here instead of s->childp, because:
2365 * (1) s->old_bs must be non-NULL, so bdrv_replace_child_noperm() will not
2366 * modify the BdrvChild * pointer we indirectly pass to it, i.e. it
2367 * will not modify s->child. From that perspective, it does not matter
2368 * whether we pass s->childp or &s->child.
2369 * (2) If new_bs is not NULL, s->childp will be NULL. We then cannot use
2370 * it here.
2371 * (3) If new_bs is NULL, *s->childp will have been NULLed by
2372 * bdrv_replace_child_tran()'s bdrv_replace_child_noperm() call, and we
2373 * must not pass a NULL *s->childp here.
2375 * So whether new_bs was NULL or not, we cannot pass s->childp here; and in
2376 * any case, there is no reason to pass it anyway.
2378 bdrv_replace_child_noperm(&s->child, s->old_bs, true);
2380 * The child was pre-existing, so s->old_bs must be non-NULL, and
2381 * s->child thus must not have been freed
2383 assert(s->child != NULL);
2384 if (!new_bs) {
2385 /* As described above, *s->childp was cleared, so restore it */
2386 assert(s->childp != NULL);
2387 *s->childp = s->child;
2389 bdrv_unref(new_bs);
2392 static TransactionActionDrv bdrv_replace_child_drv = {
2393 .commit = bdrv_replace_child_commit,
2394 .abort = bdrv_replace_child_abort,
2395 .clean = g_free,
2399 * bdrv_replace_child_tran
2401 * Note: real unref of old_bs is done only on commit.
2403 * The function doesn't update permissions, caller is responsible for this.
2405 * (*childp)->bs must not be NULL.
2407 * Note that if new_bs == NULL, @childp is stored in a state object attached
2408 * to @tran, so that the old child can be reinstated in the abort handler.
2409 * Therefore, if @new_bs can be NULL, @childp must stay valid until the
2410 * transaction is committed or aborted.
2412 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2413 * freed (on commit). @free_empty_child should only be false if the
2414 * caller will free the BDrvChild themselves (which may be important
2415 * if this is in turn called in another transactional context).
2417 static void bdrv_replace_child_tran(BdrvChild **childp,
2418 BlockDriverState *new_bs,
2419 Transaction *tran,
2420 bool free_empty_child)
2422 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2423 *s = (BdrvReplaceChildState) {
2424 .child = *childp,
2425 .childp = new_bs == NULL ? childp : NULL,
2426 .old_bs = (*childp)->bs,
2427 .free_empty_child = free_empty_child,
2429 tran_add(tran, &bdrv_replace_child_drv, s);
2431 /* The abort handler relies on this */
2432 assert(s->old_bs != NULL);
2434 if (new_bs) {
2435 bdrv_ref(new_bs);
2438 * Pass free_empty_child=false, we will free the child (if
2439 * necessary) in bdrv_replace_child_commit() (if our
2440 * @free_empty_child parameter was true).
2442 bdrv_replace_child_noperm(childp, new_bs, false);
2443 /* old_bs reference is transparently moved from *childp to @s */
2447 * Refresh permissions in @bs subtree. The function is intended to be called
2448 * after some graph modification that was done without permission update.
2450 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2451 Transaction *tran, Error **errp)
2453 BlockDriver *drv = bs->drv;
2454 BdrvChild *c;
2455 int ret;
2456 uint64_t cumulative_perms, cumulative_shared_perms;
2457 GLOBAL_STATE_CODE();
2459 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2461 /* Write permissions never work with read-only images */
2462 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2463 !bdrv_is_writable_after_reopen(bs, q))
2465 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2466 error_setg(errp, "Block node is read-only");
2467 } else {
2468 error_setg(errp, "Read-only block node '%s' cannot support "
2469 "read-write users", bdrv_get_node_name(bs));
2472 return -EPERM;
2476 * Unaligned requests will automatically be aligned to bl.request_alignment
2477 * and without RESIZE we can't extend requests to write to space beyond the
2478 * end of the image, so it's required that the image size is aligned.
2480 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2481 !(cumulative_perms & BLK_PERM_RESIZE))
2483 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2484 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2485 "Image size is not a multiple of request "
2486 "alignment");
2487 return -EPERM;
2491 /* Check this node */
2492 if (!drv) {
2493 return 0;
2496 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2497 errp);
2498 if (ret < 0) {
2499 return ret;
2502 /* Drivers that never have children can omit .bdrv_child_perm() */
2503 if (!drv->bdrv_child_perm) {
2504 assert(QLIST_EMPTY(&bs->children));
2505 return 0;
2508 /* Check all children */
2509 QLIST_FOREACH(c, &bs->children, next) {
2510 uint64_t cur_perm, cur_shared;
2512 bdrv_child_perm(bs, c->bs, c, c->role, q,
2513 cumulative_perms, cumulative_shared_perms,
2514 &cur_perm, &cur_shared);
2515 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2518 return 0;
2521 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2522 Transaction *tran, Error **errp)
2524 int ret;
2525 BlockDriverState *bs;
2526 GLOBAL_STATE_CODE();
2528 for ( ; list; list = list->next) {
2529 bs = list->data;
2531 if (bdrv_parent_perms_conflict(bs, errp)) {
2532 return -EINVAL;
2535 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2536 if (ret < 0) {
2537 return ret;
2541 return 0;
2544 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2545 uint64_t *shared_perm)
2547 BdrvChild *c;
2548 uint64_t cumulative_perms = 0;
2549 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2551 GLOBAL_STATE_CODE();
2553 QLIST_FOREACH(c, &bs->parents, next_parent) {
2554 cumulative_perms |= c->perm;
2555 cumulative_shared_perms &= c->shared_perm;
2558 *perm = cumulative_perms;
2559 *shared_perm = cumulative_shared_perms;
2562 char *bdrv_perm_names(uint64_t perm)
2564 struct perm_name {
2565 uint64_t perm;
2566 const char *name;
2567 } permissions[] = {
2568 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2569 { BLK_PERM_WRITE, "write" },
2570 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2571 { BLK_PERM_RESIZE, "resize" },
2572 { 0, NULL }
2575 GString *result = g_string_sized_new(30);
2576 struct perm_name *p;
2578 for (p = permissions; p->name; p++) {
2579 if (perm & p->perm) {
2580 if (result->len > 0) {
2581 g_string_append(result, ", ");
2583 g_string_append(result, p->name);
2587 return g_string_free(result, FALSE);
2591 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2593 int ret;
2594 Transaction *tran = tran_new();
2595 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2596 GLOBAL_STATE_CODE();
2598 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2599 tran_finalize(tran, ret);
2601 return ret;
2604 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2605 Error **errp)
2607 Error *local_err = NULL;
2608 Transaction *tran = tran_new();
2609 int ret;
2611 GLOBAL_STATE_CODE();
2613 bdrv_child_set_perm(c, perm, shared, tran);
2615 ret = bdrv_refresh_perms(c->bs, &local_err);
2617 tran_finalize(tran, ret);
2619 if (ret < 0) {
2620 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2621 /* tighten permissions */
2622 error_propagate(errp, local_err);
2623 } else {
2625 * Our caller may intend to only loosen restrictions and
2626 * does not expect this function to fail. Errors are not
2627 * fatal in such a case, so we can just hide them from our
2628 * caller.
2630 error_free(local_err);
2631 ret = 0;
2635 return ret;
2638 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2640 uint64_t parent_perms, parent_shared;
2641 uint64_t perms, shared;
2643 GLOBAL_STATE_CODE();
2645 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2646 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2647 parent_perms, parent_shared, &perms, &shared);
2649 return bdrv_child_try_set_perm(c, perms, shared, errp);
2653 * Default implementation for .bdrv_child_perm() for block filters:
2654 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2655 * filtered child.
2657 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2658 BdrvChildRole role,
2659 BlockReopenQueue *reopen_queue,
2660 uint64_t perm, uint64_t shared,
2661 uint64_t *nperm, uint64_t *nshared)
2663 GLOBAL_STATE_CODE();
2664 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2665 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2668 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2669 BdrvChildRole role,
2670 BlockReopenQueue *reopen_queue,
2671 uint64_t perm, uint64_t shared,
2672 uint64_t *nperm, uint64_t *nshared)
2674 assert(role & BDRV_CHILD_COW);
2675 GLOBAL_STATE_CODE();
2678 * We want consistent read from backing files if the parent needs it.
2679 * No other operations are performed on backing files.
2681 perm &= BLK_PERM_CONSISTENT_READ;
2684 * If the parent can deal with changing data, we're okay with a
2685 * writable and resizable backing file.
2686 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2688 if (shared & BLK_PERM_WRITE) {
2689 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2690 } else {
2691 shared = 0;
2694 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2696 if (bs->open_flags & BDRV_O_INACTIVE) {
2697 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2700 *nperm = perm;
2701 *nshared = shared;
2704 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2705 BdrvChildRole role,
2706 BlockReopenQueue *reopen_queue,
2707 uint64_t perm, uint64_t shared,
2708 uint64_t *nperm, uint64_t *nshared)
2710 int flags;
2712 GLOBAL_STATE_CODE();
2713 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2715 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2718 * Apart from the modifications below, the same permissions are
2719 * forwarded and left alone as for filters
2721 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2722 perm, shared, &perm, &shared);
2724 if (role & BDRV_CHILD_METADATA) {
2725 /* Format drivers may touch metadata even if the guest doesn't write */
2726 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2727 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2731 * bs->file always needs to be consistent because of the
2732 * metadata. We can never allow other users to resize or write
2733 * to it.
2735 if (!(flags & BDRV_O_NO_IO)) {
2736 perm |= BLK_PERM_CONSISTENT_READ;
2738 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2741 if (role & BDRV_CHILD_DATA) {
2743 * Technically, everything in this block is a subset of the
2744 * BDRV_CHILD_METADATA path taken above, and so this could
2745 * be an "else if" branch. However, that is not obvious, and
2746 * this function is not performance critical, therefore we let
2747 * this be an independent "if".
2751 * We cannot allow other users to resize the file because the
2752 * format driver might have some assumptions about the size
2753 * (e.g. because it is stored in metadata, or because the file
2754 * is split into fixed-size data files).
2756 shared &= ~BLK_PERM_RESIZE;
2759 * WRITE_UNCHANGED often cannot be performed as such on the
2760 * data file. For example, the qcow2 driver may still need to
2761 * write copied clusters on copy-on-read.
2763 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2764 perm |= BLK_PERM_WRITE;
2768 * If the data file is written to, the format driver may
2769 * expect to be able to resize it by writing beyond the EOF.
2771 if (perm & BLK_PERM_WRITE) {
2772 perm |= BLK_PERM_RESIZE;
2776 if (bs->open_flags & BDRV_O_INACTIVE) {
2777 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2780 *nperm = perm;
2781 *nshared = shared;
2784 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2785 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2786 uint64_t perm, uint64_t shared,
2787 uint64_t *nperm, uint64_t *nshared)
2789 GLOBAL_STATE_CODE();
2790 if (role & BDRV_CHILD_FILTERED) {
2791 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2792 BDRV_CHILD_COW)));
2793 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2794 perm, shared, nperm, nshared);
2795 } else if (role & BDRV_CHILD_COW) {
2796 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2797 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2798 perm, shared, nperm, nshared);
2799 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2800 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2801 perm, shared, nperm, nshared);
2802 } else {
2803 g_assert_not_reached();
2807 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2809 static const uint64_t permissions[] = {
2810 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2811 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2812 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2813 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2816 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2817 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2819 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2821 return permissions[qapi_perm];
2825 * Replace (*childp)->bs by @new_bs.
2827 * If @new_bs is NULL, *childp will be set to NULL, too: BDS parents
2828 * generally cannot handle a BdrvChild with .bs == NULL, so clearing
2829 * BdrvChild.bs should generally immediately be followed by the
2830 * BdrvChild pointer being cleared as well.
2832 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2833 * freed. @free_empty_child should only be false if the caller will
2834 * free the BdrvChild themselves (this may be important in a
2835 * transactional context, where it may only be freed on commit).
2837 static void bdrv_replace_child_noperm(BdrvChild **childp,
2838 BlockDriverState *new_bs,
2839 bool free_empty_child)
2841 BdrvChild *child = *childp;
2842 BlockDriverState *old_bs = child->bs;
2843 int new_bs_quiesce_counter;
2844 int drain_saldo;
2846 assert(!child->frozen);
2847 assert(old_bs != new_bs);
2848 GLOBAL_STATE_CODE();
2850 if (old_bs && new_bs) {
2851 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2854 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2855 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2858 * If the new child node is drained but the old one was not, flush
2859 * all outstanding requests to the old child node.
2861 while (drain_saldo > 0 && child->klass->drained_begin) {
2862 bdrv_parent_drained_begin_single(child, true);
2863 drain_saldo--;
2866 if (old_bs) {
2867 /* Detach first so that the recursive drain sections coming from @child
2868 * are already gone and we only end the drain sections that came from
2869 * elsewhere. */
2870 if (child->klass->detach) {
2871 child->klass->detach(child);
2873 assert_bdrv_graph_writable(old_bs);
2874 QLIST_REMOVE(child, next_parent);
2877 child->bs = new_bs;
2878 if (!new_bs) {
2879 *childp = NULL;
2882 if (new_bs) {
2883 assert_bdrv_graph_writable(new_bs);
2884 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2887 * Detaching the old node may have led to the new node's
2888 * quiesce_counter having been decreased. Not a problem, we
2889 * just need to recognize this here and then invoke
2890 * drained_end appropriately more often.
2892 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2893 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2895 /* Attach only after starting new drained sections, so that recursive
2896 * drain sections coming from @child don't get an extra .drained_begin
2897 * callback. */
2898 if (child->klass->attach) {
2899 child->klass->attach(child);
2904 * If the old child node was drained but the new one is not, allow
2905 * requests to come in only after the new node has been attached.
2907 while (drain_saldo < 0 && child->klass->drained_end) {
2908 bdrv_parent_drained_end_single(child);
2909 drain_saldo++;
2912 if (free_empty_child && !child->bs) {
2913 bdrv_child_free(child);
2918 * Free the given @child.
2920 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2921 * unused (i.e. not in a children list).
2923 static void bdrv_child_free(BdrvChild *child)
2925 assert(!child->bs);
2926 GLOBAL_STATE_CODE();
2927 assert(!child->next.le_prev); /* not in children list */
2929 g_free(child->name);
2930 g_free(child);
2933 typedef struct BdrvAttachChildCommonState {
2934 BdrvChild **child;
2935 AioContext *old_parent_ctx;
2936 AioContext *old_child_ctx;
2937 } BdrvAttachChildCommonState;
2939 static void bdrv_attach_child_common_abort(void *opaque)
2941 BdrvAttachChildCommonState *s = opaque;
2942 BdrvChild *child = *s->child;
2943 BlockDriverState *bs = child->bs;
2945 GLOBAL_STATE_CODE();
2947 * Pass free_empty_child=false, because we still need the child
2948 * for the AioContext operations on the parent below; those
2949 * BdrvChildClass methods all work on a BdrvChild object, so we
2950 * need to keep it as an empty shell (after this function, it will
2951 * not be attached to any parent, and it will not have a .bs).
2953 bdrv_replace_child_noperm(s->child, NULL, false);
2955 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2956 bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort);
2959 if (bdrv_child_get_parent_aio_context(child) != s->old_parent_ctx) {
2960 GSList *ignore;
2962 /* No need to ignore `child`, because it has been detached already */
2963 ignore = NULL;
2964 child->klass->can_set_aio_ctx(child, s->old_parent_ctx, &ignore,
2965 &error_abort);
2966 g_slist_free(ignore);
2968 ignore = NULL;
2969 child->klass->set_aio_ctx(child, s->old_parent_ctx, &ignore);
2970 g_slist_free(ignore);
2973 bdrv_unref(bs);
2974 bdrv_child_free(child);
2977 static TransactionActionDrv bdrv_attach_child_common_drv = {
2978 .abort = bdrv_attach_child_common_abort,
2979 .clean = g_free,
2983 * Common part of attaching bdrv child to bs or to blk or to job
2985 * Resulting new child is returned through @child.
2986 * At start *@child must be NULL.
2987 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2988 * NULL on abort(). So referenced variable must live at least until transaction
2989 * end.
2991 * Function doesn't update permissions, caller is responsible for this.
2993 static int bdrv_attach_child_common(BlockDriverState *child_bs,
2994 const char *child_name,
2995 const BdrvChildClass *child_class,
2996 BdrvChildRole child_role,
2997 uint64_t perm, uint64_t shared_perm,
2998 void *opaque, BdrvChild **child,
2999 Transaction *tran, Error **errp)
3001 BdrvChild *new_child;
3002 AioContext *parent_ctx;
3003 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3005 assert(child);
3006 assert(*child == NULL);
3007 assert(child_class->get_parent_desc);
3008 GLOBAL_STATE_CODE();
3010 new_child = g_new(BdrvChild, 1);
3011 *new_child = (BdrvChild) {
3012 .bs = NULL,
3013 .name = g_strdup(child_name),
3014 .klass = child_class,
3015 .role = child_role,
3016 .perm = perm,
3017 .shared_perm = shared_perm,
3018 .opaque = opaque,
3022 * If the AioContexts don't match, first try to move the subtree of
3023 * child_bs into the AioContext of the new parent. If this doesn't work,
3024 * try moving the parent into the AioContext of child_bs instead.
3026 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3027 if (child_ctx != parent_ctx) {
3028 Error *local_err = NULL;
3029 int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err);
3031 if (ret < 0 && child_class->can_set_aio_ctx) {
3032 GSList *ignore = g_slist_prepend(NULL, new_child);
3033 if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore,
3034 NULL))
3036 error_free(local_err);
3037 ret = 0;
3038 g_slist_free(ignore);
3039 ignore = g_slist_prepend(NULL, new_child);
3040 child_class->set_aio_ctx(new_child, child_ctx, &ignore);
3042 g_slist_free(ignore);
3045 if (ret < 0) {
3046 error_propagate(errp, local_err);
3047 bdrv_child_free(new_child);
3048 return ret;
3052 bdrv_ref(child_bs);
3053 bdrv_replace_child_noperm(&new_child, child_bs, true);
3054 /* child_bs was non-NULL, so new_child must not have been freed */
3055 assert(new_child != NULL);
3057 *child = new_child;
3059 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3060 *s = (BdrvAttachChildCommonState) {
3061 .child = child,
3062 .old_parent_ctx = parent_ctx,
3063 .old_child_ctx = child_ctx,
3065 tran_add(tran, &bdrv_attach_child_common_drv, s);
3067 return 0;
3071 * Variable referenced by @child must live at least until transaction end.
3072 * (see bdrv_attach_child_common() doc for details)
3074 * Function doesn't update permissions, caller is responsible for this.
3076 static int bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3077 BlockDriverState *child_bs,
3078 const char *child_name,
3079 const BdrvChildClass *child_class,
3080 BdrvChildRole child_role,
3081 BdrvChild **child,
3082 Transaction *tran,
3083 Error **errp)
3085 int ret;
3086 uint64_t perm, shared_perm;
3088 assert(parent_bs->drv);
3089 GLOBAL_STATE_CODE();
3091 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3092 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3093 child_bs->node_name, child_name, parent_bs->node_name);
3094 return -EINVAL;
3097 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3098 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3099 perm, shared_perm, &perm, &shared_perm);
3101 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
3102 child_role, perm, shared_perm, parent_bs,
3103 child, tran, errp);
3104 if (ret < 0) {
3105 return ret;
3108 return 0;
3111 static void bdrv_detach_child(BdrvChild **childp)
3113 BlockDriverState *old_bs = (*childp)->bs;
3115 GLOBAL_STATE_CODE();
3116 bdrv_replace_child_noperm(childp, NULL, true);
3118 if (old_bs) {
3120 * Update permissions for old node. We're just taking a parent away, so
3121 * we're loosening restrictions. Errors of permission update are not
3122 * fatal in this case, ignore them.
3124 bdrv_refresh_perms(old_bs, NULL);
3127 * When the parent requiring a non-default AioContext is removed, the
3128 * node moves back to the main AioContext
3130 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
3135 * This function steals the reference to child_bs from the caller.
3136 * That reference is later dropped by bdrv_root_unref_child().
3138 * On failure NULL is returned, errp is set and the reference to
3139 * child_bs is also dropped.
3141 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3142 * (unless @child_bs is already in @ctx).
3144 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3145 const char *child_name,
3146 const BdrvChildClass *child_class,
3147 BdrvChildRole child_role,
3148 uint64_t perm, uint64_t shared_perm,
3149 void *opaque, Error **errp)
3151 int ret;
3152 BdrvChild *child = NULL;
3153 Transaction *tran = tran_new();
3155 GLOBAL_STATE_CODE();
3157 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
3158 child_role, perm, shared_perm, opaque,
3159 &child, tran, errp);
3160 if (ret < 0) {
3161 goto out;
3164 ret = bdrv_refresh_perms(child_bs, errp);
3166 out:
3167 tran_finalize(tran, ret);
3168 /* child is unset on failure by bdrv_attach_child_common_abort() */
3169 assert((ret < 0) == !child);
3171 bdrv_unref(child_bs);
3172 return child;
3176 * This function transfers the reference to child_bs from the caller
3177 * to parent_bs. That reference is later dropped by parent_bs on
3178 * bdrv_close() or if someone calls bdrv_unref_child().
3180 * On failure NULL is returned, errp is set and the reference to
3181 * child_bs is also dropped.
3183 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3184 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3186 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3187 BlockDriverState *child_bs,
3188 const char *child_name,
3189 const BdrvChildClass *child_class,
3190 BdrvChildRole child_role,
3191 Error **errp)
3193 int ret;
3194 BdrvChild *child = NULL;
3195 Transaction *tran = tran_new();
3197 GLOBAL_STATE_CODE();
3199 ret = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, child_class,
3200 child_role, &child, tran, errp);
3201 if (ret < 0) {
3202 goto out;
3205 ret = bdrv_refresh_perms(parent_bs, errp);
3206 if (ret < 0) {
3207 goto out;
3210 out:
3211 tran_finalize(tran, ret);
3212 /* child is unset on failure by bdrv_attach_child_common_abort() */
3213 assert((ret < 0) == !child);
3215 bdrv_unref(child_bs);
3217 return child;
3220 /* Callers must ensure that child->frozen is false. */
3221 void bdrv_root_unref_child(BdrvChild *child)
3223 BlockDriverState *child_bs;
3225 GLOBAL_STATE_CODE();
3227 child_bs = child->bs;
3228 bdrv_detach_child(&child);
3229 bdrv_unref(child_bs);
3232 typedef struct BdrvSetInheritsFrom {
3233 BlockDriverState *bs;
3234 BlockDriverState *old_inherits_from;
3235 } BdrvSetInheritsFrom;
3237 static void bdrv_set_inherits_from_abort(void *opaque)
3239 BdrvSetInheritsFrom *s = opaque;
3241 s->bs->inherits_from = s->old_inherits_from;
3244 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3245 .abort = bdrv_set_inherits_from_abort,
3246 .clean = g_free,
3249 /* @tran is allowed to be NULL. In this case no rollback is possible */
3250 static void bdrv_set_inherits_from(BlockDriverState *bs,
3251 BlockDriverState *new_inherits_from,
3252 Transaction *tran)
3254 if (tran) {
3255 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3257 *s = (BdrvSetInheritsFrom) {
3258 .bs = bs,
3259 .old_inherits_from = bs->inherits_from,
3262 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3265 bs->inherits_from = new_inherits_from;
3269 * Clear all inherits_from pointers from children and grandchildren of
3270 * @root that point to @root, where necessary.
3271 * @tran is allowed to be NULL. In this case no rollback is possible
3273 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3274 Transaction *tran)
3276 BdrvChild *c;
3278 if (child->bs->inherits_from == root) {
3280 * Remove inherits_from only when the last reference between root and
3281 * child->bs goes away.
3283 QLIST_FOREACH(c, &root->children, next) {
3284 if (c != child && c->bs == child->bs) {
3285 break;
3288 if (c == NULL) {
3289 bdrv_set_inherits_from(child->bs, NULL, tran);
3293 QLIST_FOREACH(c, &child->bs->children, next) {
3294 bdrv_unset_inherits_from(root, c, tran);
3298 /* Callers must ensure that child->frozen is false. */
3299 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3301 GLOBAL_STATE_CODE();
3302 if (child == NULL) {
3303 return;
3306 bdrv_unset_inherits_from(parent, child, NULL);
3307 bdrv_root_unref_child(child);
3311 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3313 BdrvChild *c;
3314 GLOBAL_STATE_CODE();
3315 QLIST_FOREACH(c, &bs->parents, next_parent) {
3316 if (c->klass->change_media) {
3317 c->klass->change_media(c, load);
3322 /* Return true if you can reach parent going through child->inherits_from
3323 * recursively. If parent or child are NULL, return false */
3324 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3325 BlockDriverState *parent)
3327 while (child && child != parent) {
3328 child = child->inherits_from;
3331 return child != NULL;
3335 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3336 * mostly used for COW backing children (role = COW), but also for
3337 * filtered children (role = FILTERED | PRIMARY).
3339 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3341 if (bs->drv && bs->drv->is_filter) {
3342 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3343 } else {
3344 return BDRV_CHILD_COW;
3349 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3350 * callers which don't need their own reference any more must call bdrv_unref().
3352 * Function doesn't update permissions, caller is responsible for this.
3354 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3355 BlockDriverState *child_bs,
3356 bool is_backing,
3357 Transaction *tran, Error **errp)
3359 int ret = 0;
3360 bool update_inherits_from =
3361 bdrv_inherits_from_recursive(child_bs, parent_bs);
3362 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3363 BdrvChildRole role;
3365 GLOBAL_STATE_CODE();
3367 if (!parent_bs->drv) {
3369 * Node without drv is an object without a class :/. TODO: finally fix
3370 * qcow2 driver to never clear bs->drv and implement format corruption
3371 * handling in other way.
3373 error_setg(errp, "Node corrupted");
3374 return -EINVAL;
3377 if (child && child->frozen) {
3378 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3379 child->name, parent_bs->node_name, child->bs->node_name);
3380 return -EPERM;
3383 if (is_backing && !parent_bs->drv->is_filter &&
3384 !parent_bs->drv->supports_backing)
3386 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3387 "files", parent_bs->drv->format_name, parent_bs->node_name);
3388 return -EINVAL;
3391 if (parent_bs->drv->is_filter) {
3392 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3393 } else if (is_backing) {
3394 role = BDRV_CHILD_COW;
3395 } else {
3397 * We only can use same role as it is in existing child. We don't have
3398 * infrastructure to determine role of file child in generic way
3400 if (!child) {
3401 error_setg(errp, "Cannot set file child to format node without "
3402 "file child");
3403 return -EINVAL;
3405 role = child->role;
3408 if (child) {
3409 bdrv_unset_inherits_from(parent_bs, child, tran);
3410 bdrv_remove_file_or_backing_child(parent_bs, child, tran);
3413 if (!child_bs) {
3414 goto out;
3417 ret = bdrv_attach_child_noperm(parent_bs, child_bs,
3418 is_backing ? "backing" : "file",
3419 &child_of_bds, role,
3420 is_backing ? &parent_bs->backing :
3421 &parent_bs->file,
3422 tran, errp);
3423 if (ret < 0) {
3424 return ret;
3429 * If inherits_from pointed recursively to bs then let's update it to
3430 * point directly to bs (else it will become NULL).
3432 if (update_inherits_from) {
3433 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3436 out:
3437 bdrv_refresh_limits(parent_bs, tran, NULL);
3439 return 0;
3442 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3443 BlockDriverState *backing_hd,
3444 Transaction *tran, Error **errp)
3446 GLOBAL_STATE_CODE();
3447 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3450 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3451 Error **errp)
3453 int ret;
3454 Transaction *tran = tran_new();
3456 GLOBAL_STATE_CODE();
3457 bdrv_drained_begin(bs);
3459 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3460 if (ret < 0) {
3461 goto out;
3464 ret = bdrv_refresh_perms(bs, errp);
3465 out:
3466 tran_finalize(tran, ret);
3468 bdrv_drained_end(bs);
3470 return ret;
3474 * Opens the backing file for a BlockDriverState if not yet open
3476 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3477 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3478 * itself, all options starting with "${bdref_key}." are considered part of the
3479 * BlockdevRef.
3481 * TODO Can this be unified with bdrv_open_image()?
3483 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3484 const char *bdref_key, Error **errp)
3486 char *backing_filename = NULL;
3487 char *bdref_key_dot;
3488 const char *reference = NULL;
3489 int ret = 0;
3490 bool implicit_backing = false;
3491 BlockDriverState *backing_hd;
3492 QDict *options;
3493 QDict *tmp_parent_options = NULL;
3494 Error *local_err = NULL;
3496 GLOBAL_STATE_CODE();
3498 if (bs->backing != NULL) {
3499 goto free_exit;
3502 /* NULL means an empty set of options */
3503 if (parent_options == NULL) {
3504 tmp_parent_options = qdict_new();
3505 parent_options = tmp_parent_options;
3508 bs->open_flags &= ~BDRV_O_NO_BACKING;
3510 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3511 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3512 g_free(bdref_key_dot);
3515 * Caution: while qdict_get_try_str() is fine, getting non-string
3516 * types would require more care. When @parent_options come from
3517 * -blockdev or blockdev_add, its members are typed according to
3518 * the QAPI schema, but when they come from -drive, they're all
3519 * QString.
3521 reference = qdict_get_try_str(parent_options, bdref_key);
3522 if (reference || qdict_haskey(options, "file.filename")) {
3523 /* keep backing_filename NULL */
3524 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3525 qobject_unref(options);
3526 goto free_exit;
3527 } else {
3528 if (qdict_size(options) == 0) {
3529 /* If the user specifies options that do not modify the
3530 * backing file's behavior, we might still consider it the
3531 * implicit backing file. But it's easier this way, and
3532 * just specifying some of the backing BDS's options is
3533 * only possible with -drive anyway (otherwise the QAPI
3534 * schema forces the user to specify everything). */
3535 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3538 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3539 if (local_err) {
3540 ret = -EINVAL;
3541 error_propagate(errp, local_err);
3542 qobject_unref(options);
3543 goto free_exit;
3547 if (!bs->drv || !bs->drv->supports_backing) {
3548 ret = -EINVAL;
3549 error_setg(errp, "Driver doesn't support backing files");
3550 qobject_unref(options);
3551 goto free_exit;
3554 if (!reference &&
3555 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3556 qdict_put_str(options, "driver", bs->backing_format);
3559 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3560 &child_of_bds, bdrv_backing_role(bs), errp);
3561 if (!backing_hd) {
3562 bs->open_flags |= BDRV_O_NO_BACKING;
3563 error_prepend(errp, "Could not open backing file: ");
3564 ret = -EINVAL;
3565 goto free_exit;
3568 if (implicit_backing) {
3569 bdrv_refresh_filename(backing_hd);
3570 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3571 backing_hd->filename);
3574 /* Hook up the backing file link; drop our reference, bs owns the
3575 * backing_hd reference now */
3576 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3577 bdrv_unref(backing_hd);
3578 if (ret < 0) {
3579 goto free_exit;
3582 qdict_del(parent_options, bdref_key);
3584 free_exit:
3585 g_free(backing_filename);
3586 qobject_unref(tmp_parent_options);
3587 return ret;
3590 static BlockDriverState *
3591 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3592 BlockDriverState *parent, const BdrvChildClass *child_class,
3593 BdrvChildRole child_role, bool allow_none, Error **errp)
3595 BlockDriverState *bs = NULL;
3596 QDict *image_options;
3597 char *bdref_key_dot;
3598 const char *reference;
3600 assert(child_class != NULL);
3602 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3603 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3604 g_free(bdref_key_dot);
3607 * Caution: while qdict_get_try_str() is fine, getting non-string
3608 * types would require more care. When @options come from
3609 * -blockdev or blockdev_add, its members are typed according to
3610 * the QAPI schema, but when they come from -drive, they're all
3611 * QString.
3613 reference = qdict_get_try_str(options, bdref_key);
3614 if (!filename && !reference && !qdict_size(image_options)) {
3615 if (!allow_none) {
3616 error_setg(errp, "A block device must be specified for \"%s\"",
3617 bdref_key);
3619 qobject_unref(image_options);
3620 goto done;
3623 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3624 parent, child_class, child_role, errp);
3625 if (!bs) {
3626 goto done;
3629 done:
3630 qdict_del(options, bdref_key);
3631 return bs;
3635 * Opens a disk image whose options are given as BlockdevRef in another block
3636 * device's options.
3638 * If allow_none is true, no image will be opened if filename is false and no
3639 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3641 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3642 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3643 * itself, all options starting with "${bdref_key}." are considered part of the
3644 * BlockdevRef.
3646 * The BlockdevRef will be removed from the options QDict.
3648 BdrvChild *bdrv_open_child(const char *filename,
3649 QDict *options, const char *bdref_key,
3650 BlockDriverState *parent,
3651 const BdrvChildClass *child_class,
3652 BdrvChildRole child_role,
3653 bool allow_none, Error **errp)
3655 BlockDriverState *bs;
3657 GLOBAL_STATE_CODE();
3659 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3660 child_role, allow_none, errp);
3661 if (bs == NULL) {
3662 return NULL;
3665 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3666 errp);
3670 * TODO Future callers may need to specify parent/child_class in order for
3671 * option inheritance to work. Existing callers use it for the root node.
3673 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3675 BlockDriverState *bs = NULL;
3676 QObject *obj = NULL;
3677 QDict *qdict = NULL;
3678 const char *reference = NULL;
3679 Visitor *v = NULL;
3681 GLOBAL_STATE_CODE();
3683 if (ref->type == QTYPE_QSTRING) {
3684 reference = ref->u.reference;
3685 } else {
3686 BlockdevOptions *options = &ref->u.definition;
3687 assert(ref->type == QTYPE_QDICT);
3689 v = qobject_output_visitor_new(&obj);
3690 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3691 visit_complete(v, &obj);
3693 qdict = qobject_to(QDict, obj);
3694 qdict_flatten(qdict);
3696 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3697 * compatibility with other callers) rather than what we want as the
3698 * real defaults. Apply the defaults here instead. */
3699 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3700 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3701 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3702 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3706 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3707 obj = NULL;
3708 qobject_unref(obj);
3709 visit_free(v);
3710 return bs;
3713 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3714 int flags,
3715 QDict *snapshot_options,
3716 Error **errp)
3718 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3719 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3720 int64_t total_size;
3721 QemuOpts *opts = NULL;
3722 BlockDriverState *bs_snapshot = NULL;
3723 int ret;
3725 GLOBAL_STATE_CODE();
3727 /* if snapshot, we create a temporary backing file and open it
3728 instead of opening 'filename' directly */
3730 /* Get the required size from the image */
3731 total_size = bdrv_getlength(bs);
3732 if (total_size < 0) {
3733 error_setg_errno(errp, -total_size, "Could not get image size");
3734 goto out;
3737 /* Create the temporary image */
3738 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3739 if (ret < 0) {
3740 error_setg_errno(errp, -ret, "Could not get temporary filename");
3741 goto out;
3744 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3745 &error_abort);
3746 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3747 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3748 qemu_opts_del(opts);
3749 if (ret < 0) {
3750 error_prepend(errp, "Could not create temporary overlay '%s': ",
3751 tmp_filename);
3752 goto out;
3755 /* Prepare options QDict for the temporary file */
3756 qdict_put_str(snapshot_options, "file.driver", "file");
3757 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3758 qdict_put_str(snapshot_options, "driver", "qcow2");
3760 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3761 snapshot_options = NULL;
3762 if (!bs_snapshot) {
3763 goto out;
3766 ret = bdrv_append(bs_snapshot, bs, errp);
3767 if (ret < 0) {
3768 bs_snapshot = NULL;
3769 goto out;
3772 out:
3773 qobject_unref(snapshot_options);
3774 g_free(tmp_filename);
3775 return bs_snapshot;
3779 * Opens a disk image (raw, qcow2, vmdk, ...)
3781 * options is a QDict of options to pass to the block drivers, or NULL for an
3782 * empty set of options. The reference to the QDict belongs to the block layer
3783 * after the call (even on failure), so if the caller intends to reuse the
3784 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3786 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3787 * If it is not NULL, the referenced BDS will be reused.
3789 * The reference parameter may be used to specify an existing block device which
3790 * should be opened. If specified, neither options nor a filename may be given,
3791 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3793 static BlockDriverState *bdrv_open_inherit(const char *filename,
3794 const char *reference,
3795 QDict *options, int flags,
3796 BlockDriverState *parent,
3797 const BdrvChildClass *child_class,
3798 BdrvChildRole child_role,
3799 Error **errp)
3801 int ret;
3802 BlockBackend *file = NULL;
3803 BlockDriverState *bs;
3804 BlockDriver *drv = NULL;
3805 BdrvChild *child;
3806 const char *drvname;
3807 const char *backing;
3808 Error *local_err = NULL;
3809 QDict *snapshot_options = NULL;
3810 int snapshot_flags = 0;
3812 assert(!child_class || !flags);
3813 assert(!child_class == !parent);
3814 GLOBAL_STATE_CODE();
3816 if (reference) {
3817 bool options_non_empty = options ? qdict_size(options) : false;
3818 qobject_unref(options);
3820 if (filename || options_non_empty) {
3821 error_setg(errp, "Cannot reference an existing block device with "
3822 "additional options or a new filename");
3823 return NULL;
3826 bs = bdrv_lookup_bs(reference, reference, errp);
3827 if (!bs) {
3828 return NULL;
3831 bdrv_ref(bs);
3832 return bs;
3835 bs = bdrv_new();
3837 /* NULL means an empty set of options */
3838 if (options == NULL) {
3839 options = qdict_new();
3842 /* json: syntax counts as explicit options, as if in the QDict */
3843 parse_json_protocol(options, &filename, &local_err);
3844 if (local_err) {
3845 goto fail;
3848 bs->explicit_options = qdict_clone_shallow(options);
3850 if (child_class) {
3851 bool parent_is_format;
3853 if (parent->drv) {
3854 parent_is_format = parent->drv->is_format;
3855 } else {
3857 * parent->drv is not set yet because this node is opened for
3858 * (potential) format probing. That means that @parent is going
3859 * to be a format node.
3861 parent_is_format = true;
3864 bs->inherits_from = parent;
3865 child_class->inherit_options(child_role, parent_is_format,
3866 &flags, options,
3867 parent->open_flags, parent->options);
3870 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3871 if (ret < 0) {
3872 goto fail;
3876 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3877 * Caution: getting a boolean member of @options requires care.
3878 * When @options come from -blockdev or blockdev_add, members are
3879 * typed according to the QAPI schema, but when they come from
3880 * -drive, they're all QString.
3882 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3883 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3884 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3885 } else {
3886 flags &= ~BDRV_O_RDWR;
3889 if (flags & BDRV_O_SNAPSHOT) {
3890 snapshot_options = qdict_new();
3891 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3892 flags, options);
3893 /* Let bdrv_backing_options() override "read-only" */
3894 qdict_del(options, BDRV_OPT_READ_ONLY);
3895 bdrv_inherited_options(BDRV_CHILD_COW, true,
3896 &flags, options, flags, options);
3899 bs->open_flags = flags;
3900 bs->options = options;
3901 options = qdict_clone_shallow(options);
3903 /* Find the right image format driver */
3904 /* See cautionary note on accessing @options above */
3905 drvname = qdict_get_try_str(options, "driver");
3906 if (drvname) {
3907 drv = bdrv_find_format(drvname);
3908 if (!drv) {
3909 error_setg(errp, "Unknown driver: '%s'", drvname);
3910 goto fail;
3914 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3916 /* See cautionary note on accessing @options above */
3917 backing = qdict_get_try_str(options, "backing");
3918 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3919 (backing && *backing == '\0'))
3921 if (backing) {
3922 warn_report("Use of \"backing\": \"\" is deprecated; "
3923 "use \"backing\": null instead");
3925 flags |= BDRV_O_NO_BACKING;
3926 qdict_del(bs->explicit_options, "backing");
3927 qdict_del(bs->options, "backing");
3928 qdict_del(options, "backing");
3931 /* Open image file without format layer. This BlockBackend is only used for
3932 * probing, the block drivers will do their own bdrv_open_child() for the
3933 * same BDS, which is why we put the node name back into options. */
3934 if ((flags & BDRV_O_PROTOCOL) == 0) {
3935 BlockDriverState *file_bs;
3937 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3938 &child_of_bds, BDRV_CHILD_IMAGE,
3939 true, &local_err);
3940 if (local_err) {
3941 goto fail;
3943 if (file_bs != NULL) {
3944 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3945 * looking at the header to guess the image format. This works even
3946 * in cases where a guest would not see a consistent state. */
3947 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3948 blk_insert_bs(file, file_bs, &local_err);
3949 bdrv_unref(file_bs);
3950 if (local_err) {
3951 goto fail;
3954 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3958 /* Image format probing */
3959 bs->probed = !drv;
3960 if (!drv && file) {
3961 ret = find_image_format(file, filename, &drv, &local_err);
3962 if (ret < 0) {
3963 goto fail;
3966 * This option update would logically belong in bdrv_fill_options(),
3967 * but we first need to open bs->file for the probing to work, while
3968 * opening bs->file already requires the (mostly) final set of options
3969 * so that cache mode etc. can be inherited.
3971 * Adding the driver later is somewhat ugly, but it's not an option
3972 * that would ever be inherited, so it's correct. We just need to make
3973 * sure to update both bs->options (which has the full effective
3974 * options for bs) and options (which has file.* already removed).
3976 qdict_put_str(bs->options, "driver", drv->format_name);
3977 qdict_put_str(options, "driver", drv->format_name);
3978 } else if (!drv) {
3979 error_setg(errp, "Must specify either driver or file");
3980 goto fail;
3983 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3984 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3985 /* file must be NULL if a protocol BDS is about to be created
3986 * (the inverse results in an error message from bdrv_open_common()) */
3987 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3989 /* Open the image */
3990 ret = bdrv_open_common(bs, file, options, &local_err);
3991 if (ret < 0) {
3992 goto fail;
3995 if (file) {
3996 blk_unref(file);
3997 file = NULL;
4000 /* If there is a backing file, use it */
4001 if ((flags & BDRV_O_NO_BACKING) == 0) {
4002 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4003 if (ret < 0) {
4004 goto close_and_fail;
4008 /* Remove all children options and references
4009 * from bs->options and bs->explicit_options */
4010 QLIST_FOREACH(child, &bs->children, next) {
4011 char *child_key_dot;
4012 child_key_dot = g_strdup_printf("%s.", child->name);
4013 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4014 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4015 qdict_del(bs->explicit_options, child->name);
4016 qdict_del(bs->options, child->name);
4017 g_free(child_key_dot);
4020 /* Check if any unknown options were used */
4021 if (qdict_size(options) != 0) {
4022 const QDictEntry *entry = qdict_first(options);
4023 if (flags & BDRV_O_PROTOCOL) {
4024 error_setg(errp, "Block protocol '%s' doesn't support the option "
4025 "'%s'", drv->format_name, entry->key);
4026 } else {
4027 error_setg(errp,
4028 "Block format '%s' does not support the option '%s'",
4029 drv->format_name, entry->key);
4032 goto close_and_fail;
4035 bdrv_parent_cb_change_media(bs, true);
4037 qobject_unref(options);
4038 options = NULL;
4040 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4041 * temporary snapshot afterwards. */
4042 if (snapshot_flags) {
4043 BlockDriverState *snapshot_bs;
4044 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4045 snapshot_options, &local_err);
4046 snapshot_options = NULL;
4047 if (local_err) {
4048 goto close_and_fail;
4050 /* We are not going to return bs but the overlay on top of it
4051 * (snapshot_bs); thus, we have to drop the strong reference to bs
4052 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4053 * though, because the overlay still has a reference to it. */
4054 bdrv_unref(bs);
4055 bs = snapshot_bs;
4058 return bs;
4060 fail:
4061 blk_unref(file);
4062 qobject_unref(snapshot_options);
4063 qobject_unref(bs->explicit_options);
4064 qobject_unref(bs->options);
4065 qobject_unref(options);
4066 bs->options = NULL;
4067 bs->explicit_options = NULL;
4068 bdrv_unref(bs);
4069 error_propagate(errp, local_err);
4070 return NULL;
4072 close_and_fail:
4073 bdrv_unref(bs);
4074 qobject_unref(snapshot_options);
4075 qobject_unref(options);
4076 error_propagate(errp, local_err);
4077 return NULL;
4080 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4081 QDict *options, int flags, Error **errp)
4083 GLOBAL_STATE_CODE();
4085 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4086 NULL, 0, errp);
4089 /* Return true if the NULL-terminated @list contains @str */
4090 static bool is_str_in_list(const char *str, const char *const *list)
4092 if (str && list) {
4093 int i;
4094 for (i = 0; list[i] != NULL; i++) {
4095 if (!strcmp(str, list[i])) {
4096 return true;
4100 return false;
4104 * Check that every option set in @bs->options is also set in
4105 * @new_opts.
4107 * Options listed in the common_options list and in
4108 * @bs->drv->mutable_opts are skipped.
4110 * Return 0 on success, otherwise return -EINVAL and set @errp.
4112 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4113 const QDict *new_opts, Error **errp)
4115 const QDictEntry *e;
4116 /* These options are common to all block drivers and are handled
4117 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4118 const char *const common_options[] = {
4119 "node-name", "discard", "cache.direct", "cache.no-flush",
4120 "read-only", "auto-read-only", "detect-zeroes", NULL
4123 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4124 if (!qdict_haskey(new_opts, e->key) &&
4125 !is_str_in_list(e->key, common_options) &&
4126 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4127 error_setg(errp, "Option '%s' cannot be reset "
4128 "to its default value", e->key);
4129 return -EINVAL;
4133 return 0;
4137 * Returns true if @child can be reached recursively from @bs
4139 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4140 BlockDriverState *child)
4142 BdrvChild *c;
4144 if (bs == child) {
4145 return true;
4148 QLIST_FOREACH(c, &bs->children, next) {
4149 if (bdrv_recurse_has_child(c->bs, child)) {
4150 return true;
4154 return false;
4158 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4159 * reopen of multiple devices.
4161 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4162 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4163 * be created and initialized. This newly created BlockReopenQueue should be
4164 * passed back in for subsequent calls that are intended to be of the same
4165 * atomic 'set'.
4167 * bs is the BlockDriverState to add to the reopen queue.
4169 * options contains the changed options for the associated bs
4170 * (the BlockReopenQueue takes ownership)
4172 * flags contains the open flags for the associated bs
4174 * returns a pointer to bs_queue, which is either the newly allocated
4175 * bs_queue, or the existing bs_queue being used.
4177 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4179 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4180 BlockDriverState *bs,
4181 QDict *options,
4182 const BdrvChildClass *klass,
4183 BdrvChildRole role,
4184 bool parent_is_format,
4185 QDict *parent_options,
4186 int parent_flags,
4187 bool keep_old_opts)
4189 assert(bs != NULL);
4191 BlockReopenQueueEntry *bs_entry;
4192 BdrvChild *child;
4193 QDict *old_options, *explicit_options, *options_copy;
4194 int flags;
4195 QemuOpts *opts;
4197 /* Make sure that the caller remembered to use a drained section. This is
4198 * important to avoid graph changes between the recursive queuing here and
4199 * bdrv_reopen_multiple(). */
4200 assert(bs->quiesce_counter > 0);
4201 GLOBAL_STATE_CODE();
4203 if (bs_queue == NULL) {
4204 bs_queue = g_new0(BlockReopenQueue, 1);
4205 QTAILQ_INIT(bs_queue);
4208 if (!options) {
4209 options = qdict_new();
4212 /* Check if this BlockDriverState is already in the queue */
4213 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4214 if (bs == bs_entry->state.bs) {
4215 break;
4220 * Precedence of options:
4221 * 1. Explicitly passed in options (highest)
4222 * 2. Retained from explicitly set options of bs
4223 * 3. Inherited from parent node
4224 * 4. Retained from effective options of bs
4227 /* Old explicitly set values (don't overwrite by inherited value) */
4228 if (bs_entry || keep_old_opts) {
4229 old_options = qdict_clone_shallow(bs_entry ?
4230 bs_entry->state.explicit_options :
4231 bs->explicit_options);
4232 bdrv_join_options(bs, options, old_options);
4233 qobject_unref(old_options);
4236 explicit_options = qdict_clone_shallow(options);
4238 /* Inherit from parent node */
4239 if (parent_options) {
4240 flags = 0;
4241 klass->inherit_options(role, parent_is_format, &flags, options,
4242 parent_flags, parent_options);
4243 } else {
4244 flags = bdrv_get_flags(bs);
4247 if (keep_old_opts) {
4248 /* Old values are used for options that aren't set yet */
4249 old_options = qdict_clone_shallow(bs->options);
4250 bdrv_join_options(bs, options, old_options);
4251 qobject_unref(old_options);
4254 /* We have the final set of options so let's update the flags */
4255 options_copy = qdict_clone_shallow(options);
4256 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4257 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4258 update_flags_from_options(&flags, opts);
4259 qemu_opts_del(opts);
4260 qobject_unref(options_copy);
4262 /* bdrv_open_inherit() sets and clears some additional flags internally */
4263 flags &= ~BDRV_O_PROTOCOL;
4264 if (flags & BDRV_O_RDWR) {
4265 flags |= BDRV_O_ALLOW_RDWR;
4268 if (!bs_entry) {
4269 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4270 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4271 } else {
4272 qobject_unref(bs_entry->state.options);
4273 qobject_unref(bs_entry->state.explicit_options);
4276 bs_entry->state.bs = bs;
4277 bs_entry->state.options = options;
4278 bs_entry->state.explicit_options = explicit_options;
4279 bs_entry->state.flags = flags;
4282 * If keep_old_opts is false then it means that unspecified
4283 * options must be reset to their original value. We don't allow
4284 * resetting 'backing' but we need to know if the option is
4285 * missing in order to decide if we have to return an error.
4287 if (!keep_old_opts) {
4288 bs_entry->state.backing_missing =
4289 !qdict_haskey(options, "backing") &&
4290 !qdict_haskey(options, "backing.driver");
4293 QLIST_FOREACH(child, &bs->children, next) {
4294 QDict *new_child_options = NULL;
4295 bool child_keep_old = keep_old_opts;
4297 /* reopen can only change the options of block devices that were
4298 * implicitly created and inherited options. For other (referenced)
4299 * block devices, a syntax like "backing.foo" results in an error. */
4300 if (child->bs->inherits_from != bs) {
4301 continue;
4304 /* Check if the options contain a child reference */
4305 if (qdict_haskey(options, child->name)) {
4306 const char *childref = qdict_get_try_str(options, child->name);
4308 * The current child must not be reopened if the child
4309 * reference is null or points to a different node.
4311 if (g_strcmp0(childref, child->bs->node_name)) {
4312 continue;
4315 * If the child reference points to the current child then
4316 * reopen it with its existing set of options (note that
4317 * it can still inherit new options from the parent).
4319 child_keep_old = true;
4320 } else {
4321 /* Extract child options ("child-name.*") */
4322 char *child_key_dot = g_strdup_printf("%s.", child->name);
4323 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4324 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4325 g_free(child_key_dot);
4328 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4329 child->klass, child->role, bs->drv->is_format,
4330 options, flags, child_keep_old);
4333 return bs_queue;
4336 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4337 BlockDriverState *bs,
4338 QDict *options, bool keep_old_opts)
4340 GLOBAL_STATE_CODE();
4342 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4343 NULL, 0, keep_old_opts);
4346 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4348 GLOBAL_STATE_CODE();
4349 if (bs_queue) {
4350 BlockReopenQueueEntry *bs_entry, *next;
4351 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4352 qobject_unref(bs_entry->state.explicit_options);
4353 qobject_unref(bs_entry->state.options);
4354 g_free(bs_entry);
4356 g_free(bs_queue);
4361 * Reopen multiple BlockDriverStates atomically & transactionally.
4363 * The queue passed in (bs_queue) must have been built up previous
4364 * via bdrv_reopen_queue().
4366 * Reopens all BDS specified in the queue, with the appropriate
4367 * flags. All devices are prepared for reopen, and failure of any
4368 * device will cause all device changes to be abandoned, and intermediate
4369 * data cleaned up.
4371 * If all devices prepare successfully, then the changes are committed
4372 * to all devices.
4374 * All affected nodes must be drained between bdrv_reopen_queue() and
4375 * bdrv_reopen_multiple().
4377 * To be called from the main thread, with all other AioContexts unlocked.
4379 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4381 int ret = -1;
4382 BlockReopenQueueEntry *bs_entry, *next;
4383 AioContext *ctx;
4384 Transaction *tran = tran_new();
4385 g_autoptr(GHashTable) found = NULL;
4386 g_autoptr(GSList) refresh_list = NULL;
4388 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4389 assert(bs_queue != NULL);
4390 GLOBAL_STATE_CODE();
4392 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4393 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4394 aio_context_acquire(ctx);
4395 ret = bdrv_flush(bs_entry->state.bs);
4396 aio_context_release(ctx);
4397 if (ret < 0) {
4398 error_setg_errno(errp, -ret, "Error flushing drive");
4399 goto abort;
4403 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4404 assert(bs_entry->state.bs->quiesce_counter > 0);
4405 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4406 aio_context_acquire(ctx);
4407 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4408 aio_context_release(ctx);
4409 if (ret < 0) {
4410 goto abort;
4412 bs_entry->prepared = true;
4415 found = g_hash_table_new(NULL, NULL);
4416 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4417 BDRVReopenState *state = &bs_entry->state;
4419 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4420 if (state->old_backing_bs) {
4421 refresh_list = bdrv_topological_dfs(refresh_list, found,
4422 state->old_backing_bs);
4424 if (state->old_file_bs) {
4425 refresh_list = bdrv_topological_dfs(refresh_list, found,
4426 state->old_file_bs);
4431 * Note that file-posix driver rely on permission update done during reopen
4432 * (even if no permission changed), because it wants "new" permissions for
4433 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4434 * in raw_reopen_prepare() which is called with "old" permissions.
4436 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4437 if (ret < 0) {
4438 goto abort;
4442 * If we reach this point, we have success and just need to apply the
4443 * changes.
4445 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4446 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4447 * children are usually goes after parents in reopen-queue, so go from last
4448 * to first element.
4450 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4451 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4452 aio_context_acquire(ctx);
4453 bdrv_reopen_commit(&bs_entry->state);
4454 aio_context_release(ctx);
4457 tran_commit(tran);
4459 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4460 BlockDriverState *bs = bs_entry->state.bs;
4462 if (bs->drv->bdrv_reopen_commit_post) {
4463 ctx = bdrv_get_aio_context(bs);
4464 aio_context_acquire(ctx);
4465 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4466 aio_context_release(ctx);
4470 ret = 0;
4471 goto cleanup;
4473 abort:
4474 tran_abort(tran);
4475 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4476 if (bs_entry->prepared) {
4477 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4478 aio_context_acquire(ctx);
4479 bdrv_reopen_abort(&bs_entry->state);
4480 aio_context_release(ctx);
4484 cleanup:
4485 bdrv_reopen_queue_free(bs_queue);
4487 return ret;
4490 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4491 Error **errp)
4493 AioContext *ctx = bdrv_get_aio_context(bs);
4494 BlockReopenQueue *queue;
4495 int ret;
4497 GLOBAL_STATE_CODE();
4499 bdrv_subtree_drained_begin(bs);
4500 if (ctx != qemu_get_aio_context()) {
4501 aio_context_release(ctx);
4504 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4505 ret = bdrv_reopen_multiple(queue, errp);
4507 if (ctx != qemu_get_aio_context()) {
4508 aio_context_acquire(ctx);
4510 bdrv_subtree_drained_end(bs);
4512 return ret;
4515 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4516 Error **errp)
4518 QDict *opts = qdict_new();
4520 GLOBAL_STATE_CODE();
4522 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4524 return bdrv_reopen(bs, opts, true, errp);
4528 * Take a BDRVReopenState and check if the value of 'backing' in the
4529 * reopen_state->options QDict is valid or not.
4531 * If 'backing' is missing from the QDict then return 0.
4533 * If 'backing' contains the node name of the backing file of
4534 * reopen_state->bs then return 0.
4536 * If 'backing' contains a different node name (or is null) then check
4537 * whether the current backing file can be replaced with the new one.
4538 * If that's the case then reopen_state->replace_backing_bs is set to
4539 * true and reopen_state->new_backing_bs contains a pointer to the new
4540 * backing BlockDriverState (or NULL).
4542 * Return 0 on success, otherwise return < 0 and set @errp.
4544 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4545 bool is_backing, Transaction *tran,
4546 Error **errp)
4548 BlockDriverState *bs = reopen_state->bs;
4549 BlockDriverState *new_child_bs;
4550 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4551 child_bs(bs->file);
4552 const char *child_name = is_backing ? "backing" : "file";
4553 QObject *value;
4554 const char *str;
4556 GLOBAL_STATE_CODE();
4558 value = qdict_get(reopen_state->options, child_name);
4559 if (value == NULL) {
4560 return 0;
4563 switch (qobject_type(value)) {
4564 case QTYPE_QNULL:
4565 assert(is_backing); /* The 'file' option does not allow a null value */
4566 new_child_bs = NULL;
4567 break;
4568 case QTYPE_QSTRING:
4569 str = qstring_get_str(qobject_to(QString, value));
4570 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4571 if (new_child_bs == NULL) {
4572 return -EINVAL;
4573 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4574 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4575 "cycle", str, child_name, bs->node_name);
4576 return -EINVAL;
4578 break;
4579 default:
4581 * The options QDict has been flattened, so 'backing' and 'file'
4582 * do not allow any other data type here.
4584 g_assert_not_reached();
4587 if (old_child_bs == new_child_bs) {
4588 return 0;
4591 if (old_child_bs) {
4592 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4593 return 0;
4596 if (old_child_bs->implicit) {
4597 error_setg(errp, "Cannot replace implicit %s child of %s",
4598 child_name, bs->node_name);
4599 return -EPERM;
4603 if (bs->drv->is_filter && !old_child_bs) {
4605 * Filters always have a file or a backing child, so we are trying to
4606 * change wrong child
4608 error_setg(errp, "'%s' is a %s filter node that does not support a "
4609 "%s child", bs->node_name, bs->drv->format_name, child_name);
4610 return -EINVAL;
4613 if (is_backing) {
4614 reopen_state->old_backing_bs = old_child_bs;
4615 } else {
4616 reopen_state->old_file_bs = old_child_bs;
4619 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4620 tran, errp);
4624 * Prepares a BlockDriverState for reopen. All changes are staged in the
4625 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4626 * the block driver layer .bdrv_reopen_prepare()
4628 * bs is the BlockDriverState to reopen
4629 * flags are the new open flags
4630 * queue is the reopen queue
4632 * Returns 0 on success, non-zero on error. On error errp will be set
4633 * as well.
4635 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4636 * It is the responsibility of the caller to then call the abort() or
4637 * commit() for any other BDS that have been left in a prepare() state
4640 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4641 BlockReopenQueue *queue,
4642 Transaction *change_child_tran, Error **errp)
4644 int ret = -1;
4645 int old_flags;
4646 Error *local_err = NULL;
4647 BlockDriver *drv;
4648 QemuOpts *opts;
4649 QDict *orig_reopen_opts;
4650 char *discard = NULL;
4651 bool read_only;
4652 bool drv_prepared = false;
4654 assert(reopen_state != NULL);
4655 assert(reopen_state->bs->drv != NULL);
4656 GLOBAL_STATE_CODE();
4657 drv = reopen_state->bs->drv;
4659 /* This function and each driver's bdrv_reopen_prepare() remove
4660 * entries from reopen_state->options as they are processed, so
4661 * we need to make a copy of the original QDict. */
4662 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4664 /* Process generic block layer options */
4665 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4666 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4667 ret = -EINVAL;
4668 goto error;
4671 /* This was already called in bdrv_reopen_queue_child() so the flags
4672 * are up-to-date. This time we simply want to remove the options from
4673 * QemuOpts in order to indicate that they have been processed. */
4674 old_flags = reopen_state->flags;
4675 update_flags_from_options(&reopen_state->flags, opts);
4676 assert(old_flags == reopen_state->flags);
4678 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4679 if (discard != NULL) {
4680 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4681 error_setg(errp, "Invalid discard option");
4682 ret = -EINVAL;
4683 goto error;
4687 reopen_state->detect_zeroes =
4688 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4689 if (local_err) {
4690 error_propagate(errp, local_err);
4691 ret = -EINVAL;
4692 goto error;
4695 /* All other options (including node-name and driver) must be unchanged.
4696 * Put them back into the QDict, so that they are checked at the end
4697 * of this function. */
4698 qemu_opts_to_qdict(opts, reopen_state->options);
4700 /* If we are to stay read-only, do not allow permission change
4701 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4702 * not set, or if the BDS still has copy_on_read enabled */
4703 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4704 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4705 if (local_err) {
4706 error_propagate(errp, local_err);
4707 goto error;
4710 if (drv->bdrv_reopen_prepare) {
4712 * If a driver-specific option is missing, it means that we
4713 * should reset it to its default value.
4714 * But not all options allow that, so we need to check it first.
4716 ret = bdrv_reset_options_allowed(reopen_state->bs,
4717 reopen_state->options, errp);
4718 if (ret) {
4719 goto error;
4722 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4723 if (ret) {
4724 if (local_err != NULL) {
4725 error_propagate(errp, local_err);
4726 } else {
4727 bdrv_refresh_filename(reopen_state->bs);
4728 error_setg(errp, "failed while preparing to reopen image '%s'",
4729 reopen_state->bs->filename);
4731 goto error;
4733 } else {
4734 /* It is currently mandatory to have a bdrv_reopen_prepare()
4735 * handler for each supported drv. */
4736 error_setg(errp, "Block format '%s' used by node '%s' "
4737 "does not support reopening files", drv->format_name,
4738 bdrv_get_device_or_node_name(reopen_state->bs));
4739 ret = -1;
4740 goto error;
4743 drv_prepared = true;
4746 * We must provide the 'backing' option if the BDS has a backing
4747 * file or if the image file has a backing file name as part of
4748 * its metadata. Otherwise the 'backing' option can be omitted.
4750 if (drv->supports_backing && reopen_state->backing_missing &&
4751 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4752 error_setg(errp, "backing is missing for '%s'",
4753 reopen_state->bs->node_name);
4754 ret = -EINVAL;
4755 goto error;
4759 * Allow changing the 'backing' option. The new value can be
4760 * either a reference to an existing node (using its node name)
4761 * or NULL to simply detach the current backing file.
4763 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4764 change_child_tran, errp);
4765 if (ret < 0) {
4766 goto error;
4768 qdict_del(reopen_state->options, "backing");
4770 /* Allow changing the 'file' option. In this case NULL is not allowed */
4771 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4772 change_child_tran, errp);
4773 if (ret < 0) {
4774 goto error;
4776 qdict_del(reopen_state->options, "file");
4778 /* Options that are not handled are only okay if they are unchanged
4779 * compared to the old state. It is expected that some options are only
4780 * used for the initial open, but not reopen (e.g. filename) */
4781 if (qdict_size(reopen_state->options)) {
4782 const QDictEntry *entry = qdict_first(reopen_state->options);
4784 do {
4785 QObject *new = entry->value;
4786 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4788 /* Allow child references (child_name=node_name) as long as they
4789 * point to the current child (i.e. everything stays the same). */
4790 if (qobject_type(new) == QTYPE_QSTRING) {
4791 BdrvChild *child;
4792 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4793 if (!strcmp(child->name, entry->key)) {
4794 break;
4798 if (child) {
4799 if (!strcmp(child->bs->node_name,
4800 qstring_get_str(qobject_to(QString, new)))) {
4801 continue; /* Found child with this name, skip option */
4807 * TODO: When using -drive to specify blockdev options, all values
4808 * will be strings; however, when using -blockdev, blockdev-add or
4809 * filenames using the json:{} pseudo-protocol, they will be
4810 * correctly typed.
4811 * In contrast, reopening options are (currently) always strings
4812 * (because you can only specify them through qemu-io; all other
4813 * callers do not specify any options).
4814 * Therefore, when using anything other than -drive to create a BDS,
4815 * this cannot detect non-string options as unchanged, because
4816 * qobject_is_equal() always returns false for objects of different
4817 * type. In the future, this should be remedied by correctly typing
4818 * all options. For now, this is not too big of an issue because
4819 * the user can simply omit options which cannot be changed anyway,
4820 * so they will stay unchanged.
4822 if (!qobject_is_equal(new, old)) {
4823 error_setg(errp, "Cannot change the option '%s'", entry->key);
4824 ret = -EINVAL;
4825 goto error;
4827 } while ((entry = qdict_next(reopen_state->options, entry)));
4830 ret = 0;
4832 /* Restore the original reopen_state->options QDict */
4833 qobject_unref(reopen_state->options);
4834 reopen_state->options = qobject_ref(orig_reopen_opts);
4836 error:
4837 if (ret < 0 && drv_prepared) {
4838 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4839 * call drv->bdrv_reopen_abort() before signaling an error
4840 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4841 * when the respective bdrv_reopen_prepare() has failed) */
4842 if (drv->bdrv_reopen_abort) {
4843 drv->bdrv_reopen_abort(reopen_state);
4846 qemu_opts_del(opts);
4847 qobject_unref(orig_reopen_opts);
4848 g_free(discard);
4849 return ret;
4853 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4854 * makes them final by swapping the staging BlockDriverState contents into
4855 * the active BlockDriverState contents.
4857 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4859 BlockDriver *drv;
4860 BlockDriverState *bs;
4861 BdrvChild *child;
4863 assert(reopen_state != NULL);
4864 bs = reopen_state->bs;
4865 drv = bs->drv;
4866 assert(drv != NULL);
4867 GLOBAL_STATE_CODE();
4869 /* If there are any driver level actions to take */
4870 if (drv->bdrv_reopen_commit) {
4871 drv->bdrv_reopen_commit(reopen_state);
4874 /* set BDS specific flags now */
4875 qobject_unref(bs->explicit_options);
4876 qobject_unref(bs->options);
4877 qobject_ref(reopen_state->explicit_options);
4878 qobject_ref(reopen_state->options);
4880 bs->explicit_options = reopen_state->explicit_options;
4881 bs->options = reopen_state->options;
4882 bs->open_flags = reopen_state->flags;
4883 bs->detect_zeroes = reopen_state->detect_zeroes;
4885 /* Remove child references from bs->options and bs->explicit_options.
4886 * Child options were already removed in bdrv_reopen_queue_child() */
4887 QLIST_FOREACH(child, &bs->children, next) {
4888 qdict_del(bs->explicit_options, child->name);
4889 qdict_del(bs->options, child->name);
4891 /* backing is probably removed, so it's not handled by previous loop */
4892 qdict_del(bs->explicit_options, "backing");
4893 qdict_del(bs->options, "backing");
4895 bdrv_refresh_limits(bs, NULL, NULL);
4899 * Abort the reopen, and delete and free the staged changes in
4900 * reopen_state
4902 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4904 BlockDriver *drv;
4906 assert(reopen_state != NULL);
4907 drv = reopen_state->bs->drv;
4908 assert(drv != NULL);
4909 GLOBAL_STATE_CODE();
4911 if (drv->bdrv_reopen_abort) {
4912 drv->bdrv_reopen_abort(reopen_state);
4917 static void bdrv_close(BlockDriverState *bs)
4919 BdrvAioNotifier *ban, *ban_next;
4920 BdrvChild *child, *next;
4922 GLOBAL_STATE_CODE();
4923 assert(!bs->refcnt);
4925 bdrv_drained_begin(bs); /* complete I/O */
4926 bdrv_flush(bs);
4927 bdrv_drain(bs); /* in case flush left pending I/O */
4929 if (bs->drv) {
4930 if (bs->drv->bdrv_close) {
4931 /* Must unfreeze all children, so bdrv_unref_child() works */
4932 bs->drv->bdrv_close(bs);
4934 bs->drv = NULL;
4937 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4938 bdrv_unref_child(bs, child);
4941 bs->backing = NULL;
4942 bs->file = NULL;
4943 g_free(bs->opaque);
4944 bs->opaque = NULL;
4945 qatomic_set(&bs->copy_on_read, 0);
4946 bs->backing_file[0] = '\0';
4947 bs->backing_format[0] = '\0';
4948 bs->total_sectors = 0;
4949 bs->encrypted = false;
4950 bs->sg = false;
4951 qobject_unref(bs->options);
4952 qobject_unref(bs->explicit_options);
4953 bs->options = NULL;
4954 bs->explicit_options = NULL;
4955 qobject_unref(bs->full_open_options);
4956 bs->full_open_options = NULL;
4957 g_free(bs->block_status_cache);
4958 bs->block_status_cache = NULL;
4960 bdrv_release_named_dirty_bitmaps(bs);
4961 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4963 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4964 g_free(ban);
4966 QLIST_INIT(&bs->aio_notifiers);
4967 bdrv_drained_end(bs);
4970 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4971 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4972 * gets called.
4974 if (bs->quiesce_counter) {
4975 bdrv_drain_all_end_quiesce(bs);
4979 void bdrv_close_all(void)
4981 assert(job_next(NULL) == NULL);
4982 GLOBAL_STATE_CODE();
4984 /* Drop references from requests still in flight, such as canceled block
4985 * jobs whose AIO context has not been polled yet */
4986 bdrv_drain_all();
4988 blk_remove_all_bs();
4989 blockdev_close_all_bdrv_states();
4991 assert(QTAILQ_EMPTY(&all_bdrv_states));
4994 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4996 GQueue *queue;
4997 GHashTable *found;
4998 bool ret;
5000 if (c->klass->stay_at_node) {
5001 return false;
5004 /* If the child @c belongs to the BDS @to, replacing the current
5005 * c->bs by @to would mean to create a loop.
5007 * Such a case occurs when appending a BDS to a backing chain.
5008 * For instance, imagine the following chain:
5010 * guest device -> node A -> further backing chain...
5012 * Now we create a new BDS B which we want to put on top of this
5013 * chain, so we first attach A as its backing node:
5015 * node B
5018 * guest device -> node A -> further backing chain...
5020 * Finally we want to replace A by B. When doing that, we want to
5021 * replace all pointers to A by pointers to B -- except for the
5022 * pointer from B because (1) that would create a loop, and (2)
5023 * that pointer should simply stay intact:
5025 * guest device -> node B
5028 * node A -> further backing chain...
5030 * In general, when replacing a node A (c->bs) by a node B (@to),
5031 * if A is a child of B, that means we cannot replace A by B there
5032 * because that would create a loop. Silently detaching A from B
5033 * is also not really an option. So overall just leaving A in
5034 * place there is the most sensible choice.
5036 * We would also create a loop in any cases where @c is only
5037 * indirectly referenced by @to. Prevent this by returning false
5038 * if @c is found (by breadth-first search) anywhere in the whole
5039 * subtree of @to.
5042 ret = true;
5043 found = g_hash_table_new(NULL, NULL);
5044 g_hash_table_add(found, to);
5045 queue = g_queue_new();
5046 g_queue_push_tail(queue, to);
5048 while (!g_queue_is_empty(queue)) {
5049 BlockDriverState *v = g_queue_pop_head(queue);
5050 BdrvChild *c2;
5052 QLIST_FOREACH(c2, &v->children, next) {
5053 if (c2 == c) {
5054 ret = false;
5055 break;
5058 if (g_hash_table_contains(found, c2->bs)) {
5059 continue;
5062 g_queue_push_tail(queue, c2->bs);
5063 g_hash_table_add(found, c2->bs);
5067 g_queue_free(queue);
5068 g_hash_table_destroy(found);
5070 return ret;
5073 typedef struct BdrvRemoveFilterOrCowChild {
5074 BdrvChild *child;
5075 BlockDriverState *bs;
5076 bool is_backing;
5077 } BdrvRemoveFilterOrCowChild;
5079 static void bdrv_remove_filter_or_cow_child_abort(void *opaque)
5081 BdrvRemoveFilterOrCowChild *s = opaque;
5082 BlockDriverState *parent_bs = s->child->opaque;
5084 if (s->is_backing) {
5085 parent_bs->backing = s->child;
5086 } else {
5087 parent_bs->file = s->child;
5091 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
5092 * because that function is transactionable and it registered own completion
5093 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
5094 * called automatically.
5098 static void bdrv_remove_filter_or_cow_child_commit(void *opaque)
5100 BdrvRemoveFilterOrCowChild *s = opaque;
5101 GLOBAL_STATE_CODE();
5102 bdrv_child_free(s->child);
5105 static void bdrv_remove_filter_or_cow_child_clean(void *opaque)
5107 BdrvRemoveFilterOrCowChild *s = opaque;
5109 /* Drop the bs reference after the transaction is done */
5110 bdrv_unref(s->bs);
5111 g_free(s);
5114 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = {
5115 .abort = bdrv_remove_filter_or_cow_child_abort,
5116 .commit = bdrv_remove_filter_or_cow_child_commit,
5117 .clean = bdrv_remove_filter_or_cow_child_clean,
5121 * A function to remove backing or file child of @bs.
5122 * Function doesn't update permissions, caller is responsible for this.
5124 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
5125 BdrvChild *child,
5126 Transaction *tran)
5128 BdrvChild **childp;
5129 BdrvRemoveFilterOrCowChild *s;
5131 if (!child) {
5132 return;
5136 * Keep a reference to @bs so @childp will stay valid throughout the
5137 * transaction (required by bdrv_replace_child_tran())
5139 bdrv_ref(bs);
5140 if (child == bs->backing) {
5141 childp = &bs->backing;
5142 } else if (child == bs->file) {
5143 childp = &bs->file;
5144 } else {
5145 g_assert_not_reached();
5148 if (child->bs) {
5150 * Pass free_empty_child=false, we will free the child in
5151 * bdrv_remove_filter_or_cow_child_commit()
5153 bdrv_replace_child_tran(childp, NULL, tran, false);
5156 s = g_new(BdrvRemoveFilterOrCowChild, 1);
5157 *s = (BdrvRemoveFilterOrCowChild) {
5158 .child = child,
5159 .bs = bs,
5160 .is_backing = (childp == &bs->backing),
5162 tran_add(tran, &bdrv_remove_filter_or_cow_child_drv, s);
5166 * A function to remove backing-chain child of @bs if exists: cow child for
5167 * format nodes (always .backing) and filter child for filters (may be .file or
5168 * .backing)
5170 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
5171 Transaction *tran)
5173 bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran);
5176 static int bdrv_replace_node_noperm(BlockDriverState *from,
5177 BlockDriverState *to,
5178 bool auto_skip, Transaction *tran,
5179 Error **errp)
5181 BdrvChild *c, *next;
5183 assert(to != NULL);
5184 GLOBAL_STATE_CODE();
5186 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5187 assert(c->bs == from);
5188 if (!should_update_child(c, to)) {
5189 if (auto_skip) {
5190 continue;
5192 error_setg(errp, "Should not change '%s' link to '%s'",
5193 c->name, from->node_name);
5194 return -EINVAL;
5196 if (c->frozen) {
5197 error_setg(errp, "Cannot change '%s' link to '%s'",
5198 c->name, from->node_name);
5199 return -EPERM;
5203 * Passing a pointer to the local variable @c is fine here, because
5204 * @to is not NULL, and so &c will not be attached to the transaction.
5206 bdrv_replace_child_tran(&c, to, tran, true);
5209 return 0;
5213 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5214 * if it creates a parent-child relation loop or if parent is block-job.
5216 * With auto_skip=false the error is returned if from has a parent which should
5217 * not be updated.
5219 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5220 * case backing link of the cow-parent of @to is removed.
5222 * @to must not be NULL.
5224 static int bdrv_replace_node_common(BlockDriverState *from,
5225 BlockDriverState *to,
5226 bool auto_skip, bool detach_subchain,
5227 Error **errp)
5229 Transaction *tran = tran_new();
5230 g_autoptr(GHashTable) found = NULL;
5231 g_autoptr(GSList) refresh_list = NULL;
5232 BlockDriverState *to_cow_parent = NULL;
5233 int ret;
5235 GLOBAL_STATE_CODE();
5236 assert(to != NULL);
5238 if (detach_subchain) {
5239 assert(bdrv_chain_contains(from, to));
5240 assert(from != to);
5241 for (to_cow_parent = from;
5242 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5243 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5249 /* Make sure that @from doesn't go away until we have successfully attached
5250 * all of its parents to @to. */
5251 bdrv_ref(from);
5253 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5254 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5255 bdrv_drained_begin(from);
5258 * Do the replacement without permission update.
5259 * Replacement may influence the permissions, we should calculate new
5260 * permissions based on new graph. If we fail, we'll roll-back the
5261 * replacement.
5263 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5264 if (ret < 0) {
5265 goto out;
5268 if (detach_subchain) {
5269 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
5272 found = g_hash_table_new(NULL, NULL);
5274 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
5275 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
5277 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5278 if (ret < 0) {
5279 goto out;
5282 ret = 0;
5284 out:
5285 tran_finalize(tran, ret);
5287 bdrv_drained_end(from);
5288 bdrv_unref(from);
5290 return ret;
5294 * Replace node @from by @to (where neither may be NULL).
5296 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5297 Error **errp)
5299 GLOBAL_STATE_CODE();
5301 return bdrv_replace_node_common(from, to, true, false, errp);
5304 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5306 GLOBAL_STATE_CODE();
5308 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5309 errp);
5313 * Add new bs contents at the top of an image chain while the chain is
5314 * live, while keeping required fields on the top layer.
5316 * This will modify the BlockDriverState fields, and swap contents
5317 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5319 * bs_new must not be attached to a BlockBackend and must not have backing
5320 * child.
5322 * This function does not create any image files.
5324 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5325 Error **errp)
5327 int ret;
5328 Transaction *tran = tran_new();
5330 GLOBAL_STATE_CODE();
5332 assert(!bs_new->backing);
5334 ret = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5335 &child_of_bds, bdrv_backing_role(bs_new),
5336 &bs_new->backing, tran, errp);
5337 if (ret < 0) {
5338 goto out;
5341 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5342 if (ret < 0) {
5343 goto out;
5346 ret = bdrv_refresh_perms(bs_new, errp);
5347 out:
5348 tran_finalize(tran, ret);
5350 bdrv_refresh_limits(bs_top, NULL, NULL);
5352 return ret;
5355 /* Not for empty child */
5356 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5357 Error **errp)
5359 int ret;
5360 Transaction *tran = tran_new();
5361 g_autoptr(GHashTable) found = NULL;
5362 g_autoptr(GSList) refresh_list = NULL;
5363 BlockDriverState *old_bs = child->bs;
5365 GLOBAL_STATE_CODE();
5367 bdrv_ref(old_bs);
5368 bdrv_drained_begin(old_bs);
5369 bdrv_drained_begin(new_bs);
5371 bdrv_replace_child_tran(&child, new_bs, tran, true);
5372 /* @new_bs must have been non-NULL, so @child must not have been freed */
5373 assert(child != NULL);
5375 found = g_hash_table_new(NULL, NULL);
5376 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5377 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5379 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5381 tran_finalize(tran, ret);
5383 bdrv_drained_end(old_bs);
5384 bdrv_drained_end(new_bs);
5385 bdrv_unref(old_bs);
5387 return ret;
5390 static void bdrv_delete(BlockDriverState *bs)
5392 assert(bdrv_op_blocker_is_empty(bs));
5393 assert(!bs->refcnt);
5394 GLOBAL_STATE_CODE();
5396 /* remove from list, if necessary */
5397 if (bs->node_name[0] != '\0') {
5398 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5400 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5402 bdrv_close(bs);
5404 g_free(bs);
5409 * Replace @bs by newly created block node.
5411 * @options is a QDict of options to pass to the block drivers, or NULL for an
5412 * empty set of options. The reference to the QDict belongs to the block layer
5413 * after the call (even on failure), so if the caller intends to reuse the
5414 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5416 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5417 int flags, Error **errp)
5419 ERRP_GUARD();
5420 int ret;
5421 BlockDriverState *new_node_bs = NULL;
5422 const char *drvname, *node_name;
5423 BlockDriver *drv;
5425 drvname = qdict_get_try_str(options, "driver");
5426 if (!drvname) {
5427 error_setg(errp, "driver is not specified");
5428 goto fail;
5431 drv = bdrv_find_format(drvname);
5432 if (!drv) {
5433 error_setg(errp, "Unknown driver: '%s'", drvname);
5434 goto fail;
5437 node_name = qdict_get_try_str(options, "node-name");
5439 GLOBAL_STATE_CODE();
5441 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5442 errp);
5443 options = NULL; /* bdrv_new_open_driver() eats options */
5444 if (!new_node_bs) {
5445 error_prepend(errp, "Could not create node: ");
5446 goto fail;
5449 bdrv_drained_begin(bs);
5450 ret = bdrv_replace_node(bs, new_node_bs, errp);
5451 bdrv_drained_end(bs);
5453 if (ret < 0) {
5454 error_prepend(errp, "Could not replace node: ");
5455 goto fail;
5458 return new_node_bs;
5460 fail:
5461 qobject_unref(options);
5462 bdrv_unref(new_node_bs);
5463 return NULL;
5467 * Run consistency checks on an image
5469 * Returns 0 if the check could be completed (it doesn't mean that the image is
5470 * free of errors) or -errno when an internal error occurred. The results of the
5471 * check are stored in res.
5473 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5474 BdrvCheckResult *res, BdrvCheckMode fix)
5476 IO_CODE();
5477 if (bs->drv == NULL) {
5478 return -ENOMEDIUM;
5480 if (bs->drv->bdrv_co_check == NULL) {
5481 return -ENOTSUP;
5484 memset(res, 0, sizeof(*res));
5485 return bs->drv->bdrv_co_check(bs, res, fix);
5489 * Return values:
5490 * 0 - success
5491 * -EINVAL - backing format specified, but no file
5492 * -ENOSPC - can't update the backing file because no space is left in the
5493 * image file header
5494 * -ENOTSUP - format driver doesn't support changing the backing file
5496 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5497 const char *backing_fmt, bool require)
5499 BlockDriver *drv = bs->drv;
5500 int ret;
5502 GLOBAL_STATE_CODE();
5504 if (!drv) {
5505 return -ENOMEDIUM;
5508 /* Backing file format doesn't make sense without a backing file */
5509 if (backing_fmt && !backing_file) {
5510 return -EINVAL;
5513 if (require && backing_file && !backing_fmt) {
5514 return -EINVAL;
5517 if (drv->bdrv_change_backing_file != NULL) {
5518 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5519 } else {
5520 ret = -ENOTSUP;
5523 if (ret == 0) {
5524 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5525 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5526 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5527 backing_file ?: "");
5529 return ret;
5533 * Finds the first non-filter node above bs in the chain between
5534 * active and bs. The returned node is either an immediate parent of
5535 * bs, or there are only filter nodes between the two.
5537 * Returns NULL if bs is not found in active's image chain,
5538 * or if active == bs.
5540 * Returns the bottommost base image if bs == NULL.
5542 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5543 BlockDriverState *bs)
5546 GLOBAL_STATE_CODE();
5548 bs = bdrv_skip_filters(bs);
5549 active = bdrv_skip_filters(active);
5551 while (active) {
5552 BlockDriverState *next = bdrv_backing_chain_next(active);
5553 if (bs == next) {
5554 return active;
5556 active = next;
5559 return NULL;
5562 /* Given a BDS, searches for the base layer. */
5563 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5565 GLOBAL_STATE_CODE();
5567 return bdrv_find_overlay(bs, NULL);
5571 * Return true if at least one of the COW (backing) and filter links
5572 * between @bs and @base is frozen. @errp is set if that's the case.
5573 * @base must be reachable from @bs, or NULL.
5575 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5576 Error **errp)
5578 BlockDriverState *i;
5579 BdrvChild *child;
5581 GLOBAL_STATE_CODE();
5583 for (i = bs; i != base; i = child_bs(child)) {
5584 child = bdrv_filter_or_cow_child(i);
5586 if (child && child->frozen) {
5587 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5588 child->name, i->node_name, child->bs->node_name);
5589 return true;
5593 return false;
5597 * Freeze all COW (backing) and filter links between @bs and @base.
5598 * If any of the links is already frozen the operation is aborted and
5599 * none of the links are modified.
5600 * @base must be reachable from @bs, or NULL.
5601 * Returns 0 on success. On failure returns < 0 and sets @errp.
5603 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5604 Error **errp)
5606 BlockDriverState *i;
5607 BdrvChild *child;
5609 GLOBAL_STATE_CODE();
5611 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5612 return -EPERM;
5615 for (i = bs; i != base; i = child_bs(child)) {
5616 child = bdrv_filter_or_cow_child(i);
5617 if (child && child->bs->never_freeze) {
5618 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5619 child->name, child->bs->node_name);
5620 return -EPERM;
5624 for (i = bs; i != base; i = child_bs(child)) {
5625 child = bdrv_filter_or_cow_child(i);
5626 if (child) {
5627 child->frozen = true;
5631 return 0;
5635 * Unfreeze all COW (backing) and filter links between @bs and @base.
5636 * The caller must ensure that all links are frozen before using this
5637 * function.
5638 * @base must be reachable from @bs, or NULL.
5640 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5642 BlockDriverState *i;
5643 BdrvChild *child;
5645 GLOBAL_STATE_CODE();
5647 for (i = bs; i != base; i = child_bs(child)) {
5648 child = bdrv_filter_or_cow_child(i);
5649 if (child) {
5650 assert(child->frozen);
5651 child->frozen = false;
5657 * Drops images above 'base' up to and including 'top', and sets the image
5658 * above 'top' to have base as its backing file.
5660 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5661 * information in 'bs' can be properly updated.
5663 * E.g., this will convert the following chain:
5664 * bottom <- base <- intermediate <- top <- active
5666 * to
5668 * bottom <- base <- active
5670 * It is allowed for bottom==base, in which case it converts:
5672 * base <- intermediate <- top <- active
5674 * to
5676 * base <- active
5678 * If backing_file_str is non-NULL, it will be used when modifying top's
5679 * overlay image metadata.
5681 * Error conditions:
5682 * if active == top, that is considered an error
5685 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5686 const char *backing_file_str)
5688 BlockDriverState *explicit_top = top;
5689 bool update_inherits_from;
5690 BdrvChild *c;
5691 Error *local_err = NULL;
5692 int ret = -EIO;
5693 g_autoptr(GSList) updated_children = NULL;
5694 GSList *p;
5696 GLOBAL_STATE_CODE();
5698 bdrv_ref(top);
5699 bdrv_subtree_drained_begin(top);
5701 if (!top->drv || !base->drv) {
5702 goto exit;
5705 /* Make sure that base is in the backing chain of top */
5706 if (!bdrv_chain_contains(top, base)) {
5707 goto exit;
5710 /* If 'base' recursively inherits from 'top' then we should set
5711 * base->inherits_from to top->inherits_from after 'top' and all
5712 * other intermediate nodes have been dropped.
5713 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5714 * it because no one inherits from it. We use explicit_top for that. */
5715 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5716 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5718 /* success - we can delete the intermediate states, and link top->base */
5719 if (!backing_file_str) {
5720 bdrv_refresh_filename(base);
5721 backing_file_str = base->filename;
5724 QLIST_FOREACH(c, &top->parents, next_parent) {
5725 updated_children = g_slist_prepend(updated_children, c);
5729 * It seems correct to pass detach_subchain=true here, but it triggers
5730 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5731 * another drained section, which modify the graph (for example, removing
5732 * the child, which we keep in updated_children list). So, it's a TODO.
5734 * Note, bug triggered if pass detach_subchain=true here and run
5735 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5736 * That's a FIXME.
5738 bdrv_replace_node_common(top, base, false, false, &local_err);
5739 if (local_err) {
5740 error_report_err(local_err);
5741 goto exit;
5744 for (p = updated_children; p; p = p->next) {
5745 c = p->data;
5747 if (c->klass->update_filename) {
5748 ret = c->klass->update_filename(c, base, backing_file_str,
5749 &local_err);
5750 if (ret < 0) {
5752 * TODO: Actually, we want to rollback all previous iterations
5753 * of this loop, and (which is almost impossible) previous
5754 * bdrv_replace_node()...
5756 * Note, that c->klass->update_filename may lead to permission
5757 * update, so it's a bad idea to call it inside permission
5758 * update transaction of bdrv_replace_node.
5760 error_report_err(local_err);
5761 goto exit;
5766 if (update_inherits_from) {
5767 base->inherits_from = explicit_top->inherits_from;
5770 ret = 0;
5771 exit:
5772 bdrv_subtree_drained_end(top);
5773 bdrv_unref(top);
5774 return ret;
5778 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5779 * sums the size of all data-bearing children. (This excludes backing
5780 * children.)
5782 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5784 BdrvChild *child;
5785 int64_t child_size, sum = 0;
5787 QLIST_FOREACH(child, &bs->children, next) {
5788 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5789 BDRV_CHILD_FILTERED))
5791 child_size = bdrv_get_allocated_file_size(child->bs);
5792 if (child_size < 0) {
5793 return child_size;
5795 sum += child_size;
5799 return sum;
5803 * Length of a allocated file in bytes. Sparse files are counted by actual
5804 * allocated space. Return < 0 if error or unknown.
5806 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5808 BlockDriver *drv = bs->drv;
5809 IO_CODE();
5811 if (!drv) {
5812 return -ENOMEDIUM;
5814 if (drv->bdrv_get_allocated_file_size) {
5815 return drv->bdrv_get_allocated_file_size(bs);
5818 if (drv->bdrv_file_open) {
5820 * Protocol drivers default to -ENOTSUP (most of their data is
5821 * not stored in any of their children (if they even have any),
5822 * so there is no generic way to figure it out).
5824 return -ENOTSUP;
5825 } else if (drv->is_filter) {
5826 /* Filter drivers default to the size of their filtered child */
5827 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5828 } else {
5829 /* Other drivers default to summing their children's sizes */
5830 return bdrv_sum_allocated_file_size(bs);
5835 * bdrv_measure:
5836 * @drv: Format driver
5837 * @opts: Creation options for new image
5838 * @in_bs: Existing image containing data for new image (may be NULL)
5839 * @errp: Error object
5840 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5841 * or NULL on error
5843 * Calculate file size required to create a new image.
5845 * If @in_bs is given then space for allocated clusters and zero clusters
5846 * from that image are included in the calculation. If @opts contains a
5847 * backing file that is shared by @in_bs then backing clusters may be omitted
5848 * from the calculation.
5850 * If @in_bs is NULL then the calculation includes no allocated clusters
5851 * unless a preallocation option is given in @opts.
5853 * Note that @in_bs may use a different BlockDriver from @drv.
5855 * If an error occurs the @errp pointer is set.
5857 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5858 BlockDriverState *in_bs, Error **errp)
5860 IO_CODE();
5861 if (!drv->bdrv_measure) {
5862 error_setg(errp, "Block driver '%s' does not support size measurement",
5863 drv->format_name);
5864 return NULL;
5867 return drv->bdrv_measure(opts, in_bs, errp);
5871 * Return number of sectors on success, -errno on error.
5873 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5875 BlockDriver *drv = bs->drv;
5876 IO_CODE();
5878 if (!drv)
5879 return -ENOMEDIUM;
5881 if (drv->has_variable_length) {
5882 int ret = refresh_total_sectors(bs, bs->total_sectors);
5883 if (ret < 0) {
5884 return ret;
5887 return bs->total_sectors;
5891 * Return length in bytes on success, -errno on error.
5892 * The length is always a multiple of BDRV_SECTOR_SIZE.
5894 int64_t bdrv_getlength(BlockDriverState *bs)
5896 int64_t ret = bdrv_nb_sectors(bs);
5897 IO_CODE();
5899 if (ret < 0) {
5900 return ret;
5902 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5903 return -EFBIG;
5905 return ret * BDRV_SECTOR_SIZE;
5908 /* return 0 as number of sectors if no device present or error */
5909 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5911 int64_t nb_sectors = bdrv_nb_sectors(bs);
5912 IO_CODE();
5914 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5917 bool bdrv_is_sg(BlockDriverState *bs)
5919 IO_CODE();
5920 return bs->sg;
5924 * Return whether the given node supports compressed writes.
5926 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5928 BlockDriverState *filtered;
5929 IO_CODE();
5931 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5932 return false;
5935 filtered = bdrv_filter_bs(bs);
5936 if (filtered) {
5938 * Filters can only forward compressed writes, so we have to
5939 * check the child.
5941 return bdrv_supports_compressed_writes(filtered);
5944 return true;
5947 const char *bdrv_get_format_name(BlockDriverState *bs)
5949 IO_CODE();
5950 return bs->drv ? bs->drv->format_name : NULL;
5953 static int qsort_strcmp(const void *a, const void *b)
5955 return strcmp(*(char *const *)a, *(char *const *)b);
5958 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5959 void *opaque, bool read_only)
5961 BlockDriver *drv;
5962 int count = 0;
5963 int i;
5964 const char **formats = NULL;
5966 GLOBAL_STATE_CODE();
5968 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5969 if (drv->format_name) {
5970 bool found = false;
5971 int i = count;
5973 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5974 continue;
5977 while (formats && i && !found) {
5978 found = !strcmp(formats[--i], drv->format_name);
5981 if (!found) {
5982 formats = g_renew(const char *, formats, count + 1);
5983 formats[count++] = drv->format_name;
5988 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5989 const char *format_name = block_driver_modules[i].format_name;
5991 if (format_name) {
5992 bool found = false;
5993 int j = count;
5995 if (use_bdrv_whitelist &&
5996 !bdrv_format_is_whitelisted(format_name, read_only)) {
5997 continue;
6000 while (formats && j && !found) {
6001 found = !strcmp(formats[--j], format_name);
6004 if (!found) {
6005 formats = g_renew(const char *, formats, count + 1);
6006 formats[count++] = format_name;
6011 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6013 for (i = 0; i < count; i++) {
6014 it(opaque, formats[i]);
6017 g_free(formats);
6020 /* This function is to find a node in the bs graph */
6021 BlockDriverState *bdrv_find_node(const char *node_name)
6023 BlockDriverState *bs;
6025 assert(node_name);
6026 GLOBAL_STATE_CODE();
6028 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6029 if (!strcmp(node_name, bs->node_name)) {
6030 return bs;
6033 return NULL;
6036 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6037 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6038 Error **errp)
6040 BlockDeviceInfoList *list;
6041 BlockDriverState *bs;
6043 GLOBAL_STATE_CODE();
6045 list = NULL;
6046 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6047 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6048 if (!info) {
6049 qapi_free_BlockDeviceInfoList(list);
6050 return NULL;
6052 QAPI_LIST_PREPEND(list, info);
6055 return list;
6058 typedef struct XDbgBlockGraphConstructor {
6059 XDbgBlockGraph *graph;
6060 GHashTable *graph_nodes;
6061 } XDbgBlockGraphConstructor;
6063 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6065 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6067 gr->graph = g_new0(XDbgBlockGraph, 1);
6068 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6070 return gr;
6073 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6075 XDbgBlockGraph *graph = gr->graph;
6077 g_hash_table_destroy(gr->graph_nodes);
6078 g_free(gr);
6080 return graph;
6083 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6085 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6087 if (ret != 0) {
6088 return ret;
6092 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6093 * answer of g_hash_table_lookup.
6095 ret = g_hash_table_size(gr->graph_nodes) + 1;
6096 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6098 return ret;
6101 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6102 XDbgBlockGraphNodeType type, const char *name)
6104 XDbgBlockGraphNode *n;
6106 n = g_new0(XDbgBlockGraphNode, 1);
6108 n->id = xdbg_graph_node_num(gr, node);
6109 n->type = type;
6110 n->name = g_strdup(name);
6112 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6115 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6116 const BdrvChild *child)
6118 BlockPermission qapi_perm;
6119 XDbgBlockGraphEdge *edge;
6120 GLOBAL_STATE_CODE();
6122 edge = g_new0(XDbgBlockGraphEdge, 1);
6124 edge->parent = xdbg_graph_node_num(gr, parent);
6125 edge->child = xdbg_graph_node_num(gr, child->bs);
6126 edge->name = g_strdup(child->name);
6128 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6129 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6131 if (flag & child->perm) {
6132 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6134 if (flag & child->shared_perm) {
6135 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6139 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6143 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6145 BlockBackend *blk;
6146 BlockJob *job;
6147 BlockDriverState *bs;
6148 BdrvChild *child;
6149 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6151 GLOBAL_STATE_CODE();
6153 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6154 char *allocated_name = NULL;
6155 const char *name = blk_name(blk);
6157 if (!*name) {
6158 name = allocated_name = blk_get_attached_dev_id(blk);
6160 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6161 name);
6162 g_free(allocated_name);
6163 if (blk_root(blk)) {
6164 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6168 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
6169 GSList *el;
6171 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6172 job->job.id);
6173 for (el = job->nodes; el; el = el->next) {
6174 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6178 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6179 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6180 bs->node_name);
6181 QLIST_FOREACH(child, &bs->children, next) {
6182 xdbg_graph_add_edge(gr, bs, child);
6186 return xdbg_graph_finalize(gr);
6189 BlockDriverState *bdrv_lookup_bs(const char *device,
6190 const char *node_name,
6191 Error **errp)
6193 BlockBackend *blk;
6194 BlockDriverState *bs;
6196 GLOBAL_STATE_CODE();
6198 if (device) {
6199 blk = blk_by_name(device);
6201 if (blk) {
6202 bs = blk_bs(blk);
6203 if (!bs) {
6204 error_setg(errp, "Device '%s' has no medium", device);
6207 return bs;
6211 if (node_name) {
6212 bs = bdrv_find_node(node_name);
6214 if (bs) {
6215 return bs;
6219 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6220 device ? device : "",
6221 node_name ? node_name : "");
6222 return NULL;
6225 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6226 * return false. If either argument is NULL, return false. */
6227 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6230 GLOBAL_STATE_CODE();
6232 while (top && top != base) {
6233 top = bdrv_filter_or_cow_bs(top);
6236 return top != NULL;
6239 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6241 GLOBAL_STATE_CODE();
6242 if (!bs) {
6243 return QTAILQ_FIRST(&graph_bdrv_states);
6245 return QTAILQ_NEXT(bs, node_list);
6248 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6250 GLOBAL_STATE_CODE();
6251 if (!bs) {
6252 return QTAILQ_FIRST(&all_bdrv_states);
6254 return QTAILQ_NEXT(bs, bs_list);
6257 const char *bdrv_get_node_name(const BlockDriverState *bs)
6259 IO_CODE();
6260 return bs->node_name;
6263 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6265 BdrvChild *c;
6266 const char *name;
6267 IO_CODE();
6269 /* If multiple parents have a name, just pick the first one. */
6270 QLIST_FOREACH(c, &bs->parents, next_parent) {
6271 if (c->klass->get_name) {
6272 name = c->klass->get_name(c);
6273 if (name && *name) {
6274 return name;
6279 return NULL;
6282 /* TODO check what callers really want: bs->node_name or blk_name() */
6283 const char *bdrv_get_device_name(const BlockDriverState *bs)
6285 IO_CODE();
6286 return bdrv_get_parent_name(bs) ?: "";
6289 /* This can be used to identify nodes that might not have a device
6290 * name associated. Since node and device names live in the same
6291 * namespace, the result is unambiguous. The exception is if both are
6292 * absent, then this returns an empty (non-null) string. */
6293 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6295 IO_CODE();
6296 return bdrv_get_parent_name(bs) ?: bs->node_name;
6299 int bdrv_get_flags(BlockDriverState *bs)
6301 GLOBAL_STATE_CODE();
6302 return bs->open_flags;
6305 int bdrv_has_zero_init_1(BlockDriverState *bs)
6307 GLOBAL_STATE_CODE();
6308 return 1;
6311 int bdrv_has_zero_init(BlockDriverState *bs)
6313 BlockDriverState *filtered;
6314 GLOBAL_STATE_CODE();
6316 if (!bs->drv) {
6317 return 0;
6320 /* If BS is a copy on write image, it is initialized to
6321 the contents of the base image, which may not be zeroes. */
6322 if (bdrv_cow_child(bs)) {
6323 return 0;
6325 if (bs->drv->bdrv_has_zero_init) {
6326 return bs->drv->bdrv_has_zero_init(bs);
6329 filtered = bdrv_filter_bs(bs);
6330 if (filtered) {
6331 return bdrv_has_zero_init(filtered);
6334 /* safe default */
6335 return 0;
6338 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6340 IO_CODE();
6341 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6342 return false;
6345 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6348 void bdrv_get_backing_filename(BlockDriverState *bs,
6349 char *filename, int filename_size)
6351 IO_CODE();
6352 pstrcpy(filename, filename_size, bs->backing_file);
6355 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6357 int ret;
6358 BlockDriver *drv = bs->drv;
6359 IO_CODE();
6360 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6361 if (!drv) {
6362 return -ENOMEDIUM;
6364 if (!drv->bdrv_get_info) {
6365 BlockDriverState *filtered = bdrv_filter_bs(bs);
6366 if (filtered) {
6367 return bdrv_get_info(filtered, bdi);
6369 return -ENOTSUP;
6371 memset(bdi, 0, sizeof(*bdi));
6372 ret = drv->bdrv_get_info(bs, bdi);
6373 if (ret < 0) {
6374 return ret;
6377 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6378 return -EINVAL;
6381 return 0;
6384 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6385 Error **errp)
6387 BlockDriver *drv = bs->drv;
6388 IO_CODE();
6389 if (drv && drv->bdrv_get_specific_info) {
6390 return drv->bdrv_get_specific_info(bs, errp);
6392 return NULL;
6395 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6397 BlockDriver *drv = bs->drv;
6398 IO_CODE();
6399 if (!drv || !drv->bdrv_get_specific_stats) {
6400 return NULL;
6402 return drv->bdrv_get_specific_stats(bs);
6405 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6407 IO_CODE();
6408 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6409 return;
6412 bs->drv->bdrv_debug_event(bs, event);
6415 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6417 GLOBAL_STATE_CODE();
6418 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6419 bs = bdrv_primary_bs(bs);
6422 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6423 assert(bs->drv->bdrv_debug_remove_breakpoint);
6424 return bs;
6427 return NULL;
6430 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6431 const char *tag)
6433 GLOBAL_STATE_CODE();
6434 bs = bdrv_find_debug_node(bs);
6435 if (bs) {
6436 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6439 return -ENOTSUP;
6442 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6444 GLOBAL_STATE_CODE();
6445 bs = bdrv_find_debug_node(bs);
6446 if (bs) {
6447 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6450 return -ENOTSUP;
6453 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6455 GLOBAL_STATE_CODE();
6456 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6457 bs = bdrv_primary_bs(bs);
6460 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6461 return bs->drv->bdrv_debug_resume(bs, tag);
6464 return -ENOTSUP;
6467 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6469 GLOBAL_STATE_CODE();
6470 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6471 bs = bdrv_primary_bs(bs);
6474 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6475 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6478 return false;
6481 /* backing_file can either be relative, or absolute, or a protocol. If it is
6482 * relative, it must be relative to the chain. So, passing in bs->filename
6483 * from a BDS as backing_file should not be done, as that may be relative to
6484 * the CWD rather than the chain. */
6485 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6486 const char *backing_file)
6488 char *filename_full = NULL;
6489 char *backing_file_full = NULL;
6490 char *filename_tmp = NULL;
6491 int is_protocol = 0;
6492 bool filenames_refreshed = false;
6493 BlockDriverState *curr_bs = NULL;
6494 BlockDriverState *retval = NULL;
6495 BlockDriverState *bs_below;
6497 GLOBAL_STATE_CODE();
6499 if (!bs || !bs->drv || !backing_file) {
6500 return NULL;
6503 filename_full = g_malloc(PATH_MAX);
6504 backing_file_full = g_malloc(PATH_MAX);
6506 is_protocol = path_has_protocol(backing_file);
6509 * Being largely a legacy function, skip any filters here
6510 * (because filters do not have normal filenames, so they cannot
6511 * match anyway; and allowing json:{} filenames is a bit out of
6512 * scope).
6514 for (curr_bs = bdrv_skip_filters(bs);
6515 bdrv_cow_child(curr_bs) != NULL;
6516 curr_bs = bs_below)
6518 bs_below = bdrv_backing_chain_next(curr_bs);
6520 if (bdrv_backing_overridden(curr_bs)) {
6522 * If the backing file was overridden, we can only compare
6523 * directly against the backing node's filename.
6526 if (!filenames_refreshed) {
6528 * This will automatically refresh all of the
6529 * filenames in the rest of the backing chain, so we
6530 * only need to do this once.
6532 bdrv_refresh_filename(bs_below);
6533 filenames_refreshed = true;
6536 if (strcmp(backing_file, bs_below->filename) == 0) {
6537 retval = bs_below;
6538 break;
6540 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6542 * If either of the filename paths is actually a protocol, then
6543 * compare unmodified paths; otherwise make paths relative.
6545 char *backing_file_full_ret;
6547 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6548 retval = bs_below;
6549 break;
6551 /* Also check against the full backing filename for the image */
6552 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6553 NULL);
6554 if (backing_file_full_ret) {
6555 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6556 g_free(backing_file_full_ret);
6557 if (equal) {
6558 retval = bs_below;
6559 break;
6562 } else {
6563 /* If not an absolute filename path, make it relative to the current
6564 * image's filename path */
6565 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6566 NULL);
6567 /* We are going to compare canonicalized absolute pathnames */
6568 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6569 g_free(filename_tmp);
6570 continue;
6572 g_free(filename_tmp);
6574 /* We need to make sure the backing filename we are comparing against
6575 * is relative to the current image filename (or absolute) */
6576 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6577 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6578 g_free(filename_tmp);
6579 continue;
6581 g_free(filename_tmp);
6583 if (strcmp(backing_file_full, filename_full) == 0) {
6584 retval = bs_below;
6585 break;
6590 g_free(filename_full);
6591 g_free(backing_file_full);
6592 return retval;
6595 void bdrv_init(void)
6597 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6598 use_bdrv_whitelist = 1;
6599 #endif
6600 module_call_init(MODULE_INIT_BLOCK);
6603 void bdrv_init_with_whitelist(void)
6605 use_bdrv_whitelist = 1;
6606 bdrv_init();
6609 int bdrv_activate(BlockDriverState *bs, Error **errp)
6611 BdrvChild *child, *parent;
6612 Error *local_err = NULL;
6613 int ret;
6614 BdrvDirtyBitmap *bm;
6616 GLOBAL_STATE_CODE();
6618 if (!bs->drv) {
6619 return -ENOMEDIUM;
6622 QLIST_FOREACH(child, &bs->children, next) {
6623 bdrv_activate(child->bs, &local_err);
6624 if (local_err) {
6625 error_propagate(errp, local_err);
6626 return -EINVAL;
6631 * Update permissions, they may differ for inactive nodes.
6633 * Note that the required permissions of inactive images are always a
6634 * subset of the permissions required after activating the image. This
6635 * allows us to just get the permissions upfront without restricting
6636 * bdrv_co_invalidate_cache().
6638 * It also means that in error cases, we don't have to try and revert to
6639 * the old permissions (which is an operation that could fail, too). We can
6640 * just keep the extended permissions for the next time that an activation
6641 * of the image is tried.
6643 if (bs->open_flags & BDRV_O_INACTIVE) {
6644 bs->open_flags &= ~BDRV_O_INACTIVE;
6645 ret = bdrv_refresh_perms(bs, errp);
6646 if (ret < 0) {
6647 bs->open_flags |= BDRV_O_INACTIVE;
6648 return ret;
6651 ret = bdrv_invalidate_cache(bs, errp);
6652 if (ret < 0) {
6653 bs->open_flags |= BDRV_O_INACTIVE;
6654 return ret;
6657 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6658 bdrv_dirty_bitmap_skip_store(bm, false);
6661 ret = refresh_total_sectors(bs, bs->total_sectors);
6662 if (ret < 0) {
6663 bs->open_flags |= BDRV_O_INACTIVE;
6664 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6665 return ret;
6669 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6670 if (parent->klass->activate) {
6671 parent->klass->activate(parent, &local_err);
6672 if (local_err) {
6673 bs->open_flags |= BDRV_O_INACTIVE;
6674 error_propagate(errp, local_err);
6675 return -EINVAL;
6680 return 0;
6683 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6685 Error *local_err = NULL;
6686 IO_CODE();
6688 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6690 if (bs->drv->bdrv_co_invalidate_cache) {
6691 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6692 if (local_err) {
6693 error_propagate(errp, local_err);
6694 return -EINVAL;
6698 return 0;
6701 void bdrv_activate_all(Error **errp)
6703 BlockDriverState *bs;
6704 BdrvNextIterator it;
6706 GLOBAL_STATE_CODE();
6708 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6709 AioContext *aio_context = bdrv_get_aio_context(bs);
6710 int ret;
6712 aio_context_acquire(aio_context);
6713 ret = bdrv_activate(bs, errp);
6714 aio_context_release(aio_context);
6715 if (ret < 0) {
6716 bdrv_next_cleanup(&it);
6717 return;
6722 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6724 BdrvChild *parent;
6725 GLOBAL_STATE_CODE();
6727 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6728 if (parent->klass->parent_is_bds) {
6729 BlockDriverState *parent_bs = parent->opaque;
6730 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6731 return true;
6736 return false;
6739 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6741 BdrvChild *child, *parent;
6742 int ret;
6743 uint64_t cumulative_perms, cumulative_shared_perms;
6745 GLOBAL_STATE_CODE();
6747 if (!bs->drv) {
6748 return -ENOMEDIUM;
6751 /* Make sure that we don't inactivate a child before its parent.
6752 * It will be covered by recursion from the yet active parent. */
6753 if (bdrv_has_bds_parent(bs, true)) {
6754 return 0;
6757 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6759 /* Inactivate this node */
6760 if (bs->drv->bdrv_inactivate) {
6761 ret = bs->drv->bdrv_inactivate(bs);
6762 if (ret < 0) {
6763 return ret;
6767 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6768 if (parent->klass->inactivate) {
6769 ret = parent->klass->inactivate(parent);
6770 if (ret < 0) {
6771 return ret;
6776 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6777 &cumulative_shared_perms);
6778 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6779 /* Our inactive parents still need write access. Inactivation failed. */
6780 return -EPERM;
6783 bs->open_flags |= BDRV_O_INACTIVE;
6786 * Update permissions, they may differ for inactive nodes.
6787 * We only tried to loosen restrictions, so errors are not fatal, ignore
6788 * them.
6790 bdrv_refresh_perms(bs, NULL);
6792 /* Recursively inactivate children */
6793 QLIST_FOREACH(child, &bs->children, next) {
6794 ret = bdrv_inactivate_recurse(child->bs);
6795 if (ret < 0) {
6796 return ret;
6800 return 0;
6803 int bdrv_inactivate_all(void)
6805 BlockDriverState *bs = NULL;
6806 BdrvNextIterator it;
6807 int ret = 0;
6808 GSList *aio_ctxs = NULL, *ctx;
6810 GLOBAL_STATE_CODE();
6812 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6813 AioContext *aio_context = bdrv_get_aio_context(bs);
6815 if (!g_slist_find(aio_ctxs, aio_context)) {
6816 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6817 aio_context_acquire(aio_context);
6821 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6822 /* Nodes with BDS parents are covered by recursion from the last
6823 * parent that gets inactivated. Don't inactivate them a second
6824 * time if that has already happened. */
6825 if (bdrv_has_bds_parent(bs, false)) {
6826 continue;
6828 ret = bdrv_inactivate_recurse(bs);
6829 if (ret < 0) {
6830 bdrv_next_cleanup(&it);
6831 goto out;
6835 out:
6836 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6837 AioContext *aio_context = ctx->data;
6838 aio_context_release(aio_context);
6840 g_slist_free(aio_ctxs);
6842 return ret;
6845 /**************************************************************/
6846 /* removable device support */
6849 * Return TRUE if the media is present
6851 bool bdrv_is_inserted(BlockDriverState *bs)
6853 BlockDriver *drv = bs->drv;
6854 BdrvChild *child;
6855 IO_CODE();
6857 if (!drv) {
6858 return false;
6860 if (drv->bdrv_is_inserted) {
6861 return drv->bdrv_is_inserted(bs);
6863 QLIST_FOREACH(child, &bs->children, next) {
6864 if (!bdrv_is_inserted(child->bs)) {
6865 return false;
6868 return true;
6872 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6874 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6876 BlockDriver *drv = bs->drv;
6877 IO_CODE();
6879 if (drv && drv->bdrv_eject) {
6880 drv->bdrv_eject(bs, eject_flag);
6885 * Lock or unlock the media (if it is locked, the user won't be able
6886 * to eject it manually).
6888 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6890 BlockDriver *drv = bs->drv;
6891 IO_CODE();
6892 trace_bdrv_lock_medium(bs, locked);
6894 if (drv && drv->bdrv_lock_medium) {
6895 drv->bdrv_lock_medium(bs, locked);
6899 /* Get a reference to bs */
6900 void bdrv_ref(BlockDriverState *bs)
6902 GLOBAL_STATE_CODE();
6903 bs->refcnt++;
6906 /* Release a previously grabbed reference to bs.
6907 * If after releasing, reference count is zero, the BlockDriverState is
6908 * deleted. */
6909 void bdrv_unref(BlockDriverState *bs)
6911 GLOBAL_STATE_CODE();
6912 if (!bs) {
6913 return;
6915 assert(bs->refcnt > 0);
6916 if (--bs->refcnt == 0) {
6917 bdrv_delete(bs);
6921 struct BdrvOpBlocker {
6922 Error *reason;
6923 QLIST_ENTRY(BdrvOpBlocker) list;
6926 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6928 BdrvOpBlocker *blocker;
6929 GLOBAL_STATE_CODE();
6930 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6931 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6932 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6933 error_propagate_prepend(errp, error_copy(blocker->reason),
6934 "Node '%s' is busy: ",
6935 bdrv_get_device_or_node_name(bs));
6936 return true;
6938 return false;
6941 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6943 BdrvOpBlocker *blocker;
6944 GLOBAL_STATE_CODE();
6945 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6947 blocker = g_new0(BdrvOpBlocker, 1);
6948 blocker->reason = reason;
6949 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6952 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6954 BdrvOpBlocker *blocker, *next;
6955 GLOBAL_STATE_CODE();
6956 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6957 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6958 if (blocker->reason == reason) {
6959 QLIST_REMOVE(blocker, list);
6960 g_free(blocker);
6965 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6967 int i;
6968 GLOBAL_STATE_CODE();
6969 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6970 bdrv_op_block(bs, i, reason);
6974 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6976 int i;
6977 GLOBAL_STATE_CODE();
6978 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6979 bdrv_op_unblock(bs, i, reason);
6983 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6985 int i;
6986 GLOBAL_STATE_CODE();
6987 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6988 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6989 return false;
6992 return true;
6995 void bdrv_img_create(const char *filename, const char *fmt,
6996 const char *base_filename, const char *base_fmt,
6997 char *options, uint64_t img_size, int flags, bool quiet,
6998 Error **errp)
7000 QemuOptsList *create_opts = NULL;
7001 QemuOpts *opts = NULL;
7002 const char *backing_fmt, *backing_file;
7003 int64_t size;
7004 BlockDriver *drv, *proto_drv;
7005 Error *local_err = NULL;
7006 int ret = 0;
7008 GLOBAL_STATE_CODE();
7010 /* Find driver and parse its options */
7011 drv = bdrv_find_format(fmt);
7012 if (!drv) {
7013 error_setg(errp, "Unknown file format '%s'", fmt);
7014 return;
7017 proto_drv = bdrv_find_protocol(filename, true, errp);
7018 if (!proto_drv) {
7019 return;
7022 if (!drv->create_opts) {
7023 error_setg(errp, "Format driver '%s' does not support image creation",
7024 drv->format_name);
7025 return;
7028 if (!proto_drv->create_opts) {
7029 error_setg(errp, "Protocol driver '%s' does not support image creation",
7030 proto_drv->format_name);
7031 return;
7034 /* Create parameter list */
7035 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7036 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7038 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7040 /* Parse -o options */
7041 if (options) {
7042 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7043 goto out;
7047 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7048 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7049 } else if (img_size != UINT64_C(-1)) {
7050 error_setg(errp, "The image size must be specified only once");
7051 goto out;
7054 if (base_filename) {
7055 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7056 NULL)) {
7057 error_setg(errp, "Backing file not supported for file format '%s'",
7058 fmt);
7059 goto out;
7063 if (base_fmt) {
7064 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7065 error_setg(errp, "Backing file format not supported for file "
7066 "format '%s'", fmt);
7067 goto out;
7071 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7072 if (backing_file) {
7073 if (!strcmp(filename, backing_file)) {
7074 error_setg(errp, "Error: Trying to create an image with the "
7075 "same filename as the backing file");
7076 goto out;
7078 if (backing_file[0] == '\0') {
7079 error_setg(errp, "Expected backing file name, got empty string");
7080 goto out;
7084 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7086 /* The size for the image must always be specified, unless we have a backing
7087 * file and we have not been forbidden from opening it. */
7088 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7089 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7090 BlockDriverState *bs;
7091 char *full_backing;
7092 int back_flags;
7093 QDict *backing_options = NULL;
7095 full_backing =
7096 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7097 &local_err);
7098 if (local_err) {
7099 goto out;
7101 assert(full_backing);
7104 * No need to do I/O here, which allows us to open encrypted
7105 * backing images without needing the secret
7107 back_flags = flags;
7108 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7109 back_flags |= BDRV_O_NO_IO;
7111 backing_options = qdict_new();
7112 if (backing_fmt) {
7113 qdict_put_str(backing_options, "driver", backing_fmt);
7115 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7117 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7118 &local_err);
7119 g_free(full_backing);
7120 if (!bs) {
7121 error_append_hint(&local_err, "Could not open backing image.\n");
7122 goto out;
7123 } else {
7124 if (!backing_fmt) {
7125 error_setg(&local_err,
7126 "Backing file specified without backing format");
7127 error_append_hint(&local_err, "Detected format of %s.",
7128 bs->drv->format_name);
7129 goto out;
7131 if (size == -1) {
7132 /* Opened BS, have no size */
7133 size = bdrv_getlength(bs);
7134 if (size < 0) {
7135 error_setg_errno(errp, -size, "Could not get size of '%s'",
7136 backing_file);
7137 bdrv_unref(bs);
7138 goto out;
7140 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7142 bdrv_unref(bs);
7144 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7145 } else if (backing_file && !backing_fmt) {
7146 error_setg(&local_err,
7147 "Backing file specified without backing format");
7148 goto out;
7151 if (size == -1) {
7152 error_setg(errp, "Image creation needs a size parameter");
7153 goto out;
7156 if (!quiet) {
7157 printf("Formatting '%s', fmt=%s ", filename, fmt);
7158 qemu_opts_print(opts, " ");
7159 puts("");
7160 fflush(stdout);
7163 ret = bdrv_create(drv, filename, opts, &local_err);
7165 if (ret == -EFBIG) {
7166 /* This is generally a better message than whatever the driver would
7167 * deliver (especially because of the cluster_size_hint), since that
7168 * is most probably not much different from "image too large". */
7169 const char *cluster_size_hint = "";
7170 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7171 cluster_size_hint = " (try using a larger cluster size)";
7173 error_setg(errp, "The image size is too large for file format '%s'"
7174 "%s", fmt, cluster_size_hint);
7175 error_free(local_err);
7176 local_err = NULL;
7179 out:
7180 qemu_opts_del(opts);
7181 qemu_opts_free(create_opts);
7182 error_propagate(errp, local_err);
7185 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7187 IO_CODE();
7188 return bs ? bs->aio_context : qemu_get_aio_context();
7191 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7193 Coroutine *self = qemu_coroutine_self();
7194 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7195 AioContext *new_ctx;
7196 IO_CODE();
7199 * Increase bs->in_flight to ensure that this operation is completed before
7200 * moving the node to a different AioContext. Read new_ctx only afterwards.
7202 bdrv_inc_in_flight(bs);
7204 new_ctx = bdrv_get_aio_context(bs);
7205 aio_co_reschedule_self(new_ctx);
7206 return old_ctx;
7209 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7211 IO_CODE();
7212 aio_co_reschedule_self(old_ctx);
7213 bdrv_dec_in_flight(bs);
7216 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7218 AioContext *ctx = bdrv_get_aio_context(bs);
7220 /* In the main thread, bs->aio_context won't change concurrently */
7221 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7224 * We're in coroutine context, so we already hold the lock of the main
7225 * loop AioContext. Don't lock it twice to avoid deadlocks.
7227 assert(qemu_in_coroutine());
7228 if (ctx != qemu_get_aio_context()) {
7229 aio_context_acquire(ctx);
7233 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7235 AioContext *ctx = bdrv_get_aio_context(bs);
7237 assert(qemu_in_coroutine());
7238 if (ctx != qemu_get_aio_context()) {
7239 aio_context_release(ctx);
7243 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
7245 IO_CODE();
7246 aio_co_enter(bdrv_get_aio_context(bs), co);
7249 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7251 GLOBAL_STATE_CODE();
7252 QLIST_REMOVE(ban, list);
7253 g_free(ban);
7256 static void bdrv_detach_aio_context(BlockDriverState *bs)
7258 BdrvAioNotifier *baf, *baf_tmp;
7260 assert(!bs->walking_aio_notifiers);
7261 GLOBAL_STATE_CODE();
7262 bs->walking_aio_notifiers = true;
7263 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7264 if (baf->deleted) {
7265 bdrv_do_remove_aio_context_notifier(baf);
7266 } else {
7267 baf->detach_aio_context(baf->opaque);
7270 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7271 * remove remaining aio notifiers if we aren't called again.
7273 bs->walking_aio_notifiers = false;
7275 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7276 bs->drv->bdrv_detach_aio_context(bs);
7279 if (bs->quiesce_counter) {
7280 aio_enable_external(bs->aio_context);
7282 bs->aio_context = NULL;
7285 static void bdrv_attach_aio_context(BlockDriverState *bs,
7286 AioContext *new_context)
7288 BdrvAioNotifier *ban, *ban_tmp;
7289 GLOBAL_STATE_CODE();
7291 if (bs->quiesce_counter) {
7292 aio_disable_external(new_context);
7295 bs->aio_context = new_context;
7297 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7298 bs->drv->bdrv_attach_aio_context(bs, new_context);
7301 assert(!bs->walking_aio_notifiers);
7302 bs->walking_aio_notifiers = true;
7303 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7304 if (ban->deleted) {
7305 bdrv_do_remove_aio_context_notifier(ban);
7306 } else {
7307 ban->attached_aio_context(new_context, ban->opaque);
7310 bs->walking_aio_notifiers = false;
7314 * Changes the AioContext used for fd handlers, timers, and BHs by this
7315 * BlockDriverState and all its children and parents.
7317 * Must be called from the main AioContext.
7319 * The caller must own the AioContext lock for the old AioContext of bs, but it
7320 * must not own the AioContext lock for new_context (unless new_context is the
7321 * same as the current context of bs).
7323 * @ignore will accumulate all visited BdrvChild object. The caller is
7324 * responsible for freeing the list afterwards.
7326 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
7327 AioContext *new_context, GSList **ignore)
7329 AioContext *old_context = bdrv_get_aio_context(bs);
7330 GSList *children_to_process = NULL;
7331 GSList *parents_to_process = NULL;
7332 GSList *entry;
7333 BdrvChild *child, *parent;
7335 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7336 GLOBAL_STATE_CODE();
7338 if (old_context == new_context) {
7339 return;
7342 bdrv_drained_begin(bs);
7344 QLIST_FOREACH(child, &bs->children, next) {
7345 if (g_slist_find(*ignore, child)) {
7346 continue;
7348 *ignore = g_slist_prepend(*ignore, child);
7349 children_to_process = g_slist_prepend(children_to_process, child);
7352 QLIST_FOREACH(parent, &bs->parents, next_parent) {
7353 if (g_slist_find(*ignore, parent)) {
7354 continue;
7356 *ignore = g_slist_prepend(*ignore, parent);
7357 parents_to_process = g_slist_prepend(parents_to_process, parent);
7360 for (entry = children_to_process;
7361 entry != NULL;
7362 entry = g_slist_next(entry)) {
7363 child = entry->data;
7364 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
7366 g_slist_free(children_to_process);
7368 for (entry = parents_to_process;
7369 entry != NULL;
7370 entry = g_slist_next(entry)) {
7371 parent = entry->data;
7372 assert(parent->klass->set_aio_ctx);
7373 parent->klass->set_aio_ctx(parent, new_context, ignore);
7375 g_slist_free(parents_to_process);
7377 bdrv_detach_aio_context(bs);
7379 /* Acquire the new context, if necessary */
7380 if (qemu_get_aio_context() != new_context) {
7381 aio_context_acquire(new_context);
7384 bdrv_attach_aio_context(bs, new_context);
7387 * If this function was recursively called from
7388 * bdrv_set_aio_context_ignore(), there may be nodes in the
7389 * subtree that have not yet been moved to the new AioContext.
7390 * Release the old one so bdrv_drained_end() can poll them.
7392 if (qemu_get_aio_context() != old_context) {
7393 aio_context_release(old_context);
7396 bdrv_drained_end(bs);
7398 if (qemu_get_aio_context() != old_context) {
7399 aio_context_acquire(old_context);
7401 if (qemu_get_aio_context() != new_context) {
7402 aio_context_release(new_context);
7406 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7407 GSList **ignore, Error **errp)
7409 GLOBAL_STATE_CODE();
7410 if (g_slist_find(*ignore, c)) {
7411 return true;
7413 *ignore = g_slist_prepend(*ignore, c);
7416 * A BdrvChildClass that doesn't handle AioContext changes cannot
7417 * tolerate any AioContext changes
7419 if (!c->klass->can_set_aio_ctx) {
7420 char *user = bdrv_child_user_desc(c);
7421 error_setg(errp, "Changing iothreads is not supported by %s", user);
7422 g_free(user);
7423 return false;
7425 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
7426 assert(!errp || *errp);
7427 return false;
7429 return true;
7432 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7433 GSList **ignore, Error **errp)
7435 GLOBAL_STATE_CODE();
7436 if (g_slist_find(*ignore, c)) {
7437 return true;
7439 *ignore = g_slist_prepend(*ignore, c);
7440 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
7443 /* @ignore will accumulate all visited BdrvChild object. The caller is
7444 * responsible for freeing the list afterwards. */
7445 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7446 GSList **ignore, Error **errp)
7448 BdrvChild *c;
7450 if (bdrv_get_aio_context(bs) == ctx) {
7451 return true;
7454 GLOBAL_STATE_CODE();
7456 QLIST_FOREACH(c, &bs->parents, next_parent) {
7457 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
7458 return false;
7461 QLIST_FOREACH(c, &bs->children, next) {
7462 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
7463 return false;
7467 return true;
7470 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7471 BdrvChild *ignore_child, Error **errp)
7473 GSList *ignore;
7474 bool ret;
7476 GLOBAL_STATE_CODE();
7478 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7479 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
7480 g_slist_free(ignore);
7482 if (!ret) {
7483 return -EPERM;
7486 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7487 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
7488 g_slist_free(ignore);
7490 return 0;
7493 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7494 Error **errp)
7496 GLOBAL_STATE_CODE();
7497 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
7500 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7501 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7502 void (*detach_aio_context)(void *opaque), void *opaque)
7504 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7505 *ban = (BdrvAioNotifier){
7506 .attached_aio_context = attached_aio_context,
7507 .detach_aio_context = detach_aio_context,
7508 .opaque = opaque
7510 GLOBAL_STATE_CODE();
7512 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7515 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7516 void (*attached_aio_context)(AioContext *,
7517 void *),
7518 void (*detach_aio_context)(void *),
7519 void *opaque)
7521 BdrvAioNotifier *ban, *ban_next;
7522 GLOBAL_STATE_CODE();
7524 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7525 if (ban->attached_aio_context == attached_aio_context &&
7526 ban->detach_aio_context == detach_aio_context &&
7527 ban->opaque == opaque &&
7528 ban->deleted == false)
7530 if (bs->walking_aio_notifiers) {
7531 ban->deleted = true;
7532 } else {
7533 bdrv_do_remove_aio_context_notifier(ban);
7535 return;
7539 abort();
7542 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7543 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7544 bool force,
7545 Error **errp)
7547 GLOBAL_STATE_CODE();
7548 if (!bs->drv) {
7549 error_setg(errp, "Node is ejected");
7550 return -ENOMEDIUM;
7552 if (!bs->drv->bdrv_amend_options) {
7553 error_setg(errp, "Block driver '%s' does not support option amendment",
7554 bs->drv->format_name);
7555 return -ENOTSUP;
7557 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7558 cb_opaque, force, errp);
7562 * This function checks whether the given @to_replace is allowed to be
7563 * replaced by a node that always shows the same data as @bs. This is
7564 * used for example to verify whether the mirror job can replace
7565 * @to_replace by the target mirrored from @bs.
7566 * To be replaceable, @bs and @to_replace may either be guaranteed to
7567 * always show the same data (because they are only connected through
7568 * filters), or some driver may allow replacing one of its children
7569 * because it can guarantee that this child's data is not visible at
7570 * all (for example, for dissenting quorum children that have no other
7571 * parents).
7573 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7574 BlockDriverState *to_replace)
7576 BlockDriverState *filtered;
7578 GLOBAL_STATE_CODE();
7580 if (!bs || !bs->drv) {
7581 return false;
7584 if (bs == to_replace) {
7585 return true;
7588 /* See what the driver can do */
7589 if (bs->drv->bdrv_recurse_can_replace) {
7590 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7593 /* For filters without an own implementation, we can recurse on our own */
7594 filtered = bdrv_filter_bs(bs);
7595 if (filtered) {
7596 return bdrv_recurse_can_replace(filtered, to_replace);
7599 /* Safe default */
7600 return false;
7604 * Check whether the given @node_name can be replaced by a node that
7605 * has the same data as @parent_bs. If so, return @node_name's BDS;
7606 * NULL otherwise.
7608 * @node_name must be a (recursive) *child of @parent_bs (or this
7609 * function will return NULL).
7611 * The result (whether the node can be replaced or not) is only valid
7612 * for as long as no graph or permission changes occur.
7614 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7615 const char *node_name, Error **errp)
7617 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7618 AioContext *aio_context;
7620 GLOBAL_STATE_CODE();
7622 if (!to_replace_bs) {
7623 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7624 return NULL;
7627 aio_context = bdrv_get_aio_context(to_replace_bs);
7628 aio_context_acquire(aio_context);
7630 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7631 to_replace_bs = NULL;
7632 goto out;
7635 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7636 * most non filter in order to prevent data corruption.
7637 * Another benefit is that this tests exclude backing files which are
7638 * blocked by the backing blockers.
7640 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7641 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7642 "because it cannot be guaranteed that doing so would not "
7643 "lead to an abrupt change of visible data",
7644 node_name, parent_bs->node_name);
7645 to_replace_bs = NULL;
7646 goto out;
7649 out:
7650 aio_context_release(aio_context);
7651 return to_replace_bs;
7655 * Iterates through the list of runtime option keys that are said to
7656 * be "strong" for a BDS. An option is called "strong" if it changes
7657 * a BDS's data. For example, the null block driver's "size" and
7658 * "read-zeroes" options are strong, but its "latency-ns" option is
7659 * not.
7661 * If a key returned by this function ends with a dot, all options
7662 * starting with that prefix are strong.
7664 static const char *const *strong_options(BlockDriverState *bs,
7665 const char *const *curopt)
7667 static const char *const global_options[] = {
7668 "driver", "filename", NULL
7671 if (!curopt) {
7672 return &global_options[0];
7675 curopt++;
7676 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7677 curopt = bs->drv->strong_runtime_opts;
7680 return (curopt && *curopt) ? curopt : NULL;
7684 * Copies all strong runtime options from bs->options to the given
7685 * QDict. The set of strong option keys is determined by invoking
7686 * strong_options().
7688 * Returns true iff any strong option was present in bs->options (and
7689 * thus copied to the target QDict) with the exception of "filename"
7690 * and "driver". The caller is expected to use this value to decide
7691 * whether the existence of strong options prevents the generation of
7692 * a plain filename.
7694 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7696 bool found_any = false;
7697 const char *const *option_name = NULL;
7699 if (!bs->drv) {
7700 return false;
7703 while ((option_name = strong_options(bs, option_name))) {
7704 bool option_given = false;
7706 assert(strlen(*option_name) > 0);
7707 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7708 QObject *entry = qdict_get(bs->options, *option_name);
7709 if (!entry) {
7710 continue;
7713 qdict_put_obj(d, *option_name, qobject_ref(entry));
7714 option_given = true;
7715 } else {
7716 const QDictEntry *entry;
7717 for (entry = qdict_first(bs->options); entry;
7718 entry = qdict_next(bs->options, entry))
7720 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7721 qdict_put_obj(d, qdict_entry_key(entry),
7722 qobject_ref(qdict_entry_value(entry)));
7723 option_given = true;
7728 /* While "driver" and "filename" need to be included in a JSON filename,
7729 * their existence does not prohibit generation of a plain filename. */
7730 if (!found_any && option_given &&
7731 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7733 found_any = true;
7737 if (!qdict_haskey(d, "driver")) {
7738 /* Drivers created with bdrv_new_open_driver() may not have a
7739 * @driver option. Add it here. */
7740 qdict_put_str(d, "driver", bs->drv->format_name);
7743 return found_any;
7746 /* Note: This function may return false positives; it may return true
7747 * even if opening the backing file specified by bs's image header
7748 * would result in exactly bs->backing. */
7749 static bool bdrv_backing_overridden(BlockDriverState *bs)
7751 GLOBAL_STATE_CODE();
7752 if (bs->backing) {
7753 return strcmp(bs->auto_backing_file,
7754 bs->backing->bs->filename);
7755 } else {
7756 /* No backing BDS, so if the image header reports any backing
7757 * file, it must have been suppressed */
7758 return bs->auto_backing_file[0] != '\0';
7762 /* Updates the following BDS fields:
7763 * - exact_filename: A filename which may be used for opening a block device
7764 * which (mostly) equals the given BDS (even without any
7765 * other options; so reading and writing must return the same
7766 * results, but caching etc. may be different)
7767 * - full_open_options: Options which, when given when opening a block device
7768 * (without a filename), result in a BDS (mostly)
7769 * equalling the given one
7770 * - filename: If exact_filename is set, it is copied here. Otherwise,
7771 * full_open_options is converted to a JSON object, prefixed with
7772 * "json:" (for use through the JSON pseudo protocol) and put here.
7774 void bdrv_refresh_filename(BlockDriverState *bs)
7776 BlockDriver *drv = bs->drv;
7777 BdrvChild *child;
7778 BlockDriverState *primary_child_bs;
7779 QDict *opts;
7780 bool backing_overridden;
7781 bool generate_json_filename; /* Whether our default implementation should
7782 fill exact_filename (false) or not (true) */
7784 GLOBAL_STATE_CODE();
7786 if (!drv) {
7787 return;
7790 /* This BDS's file name may depend on any of its children's file names, so
7791 * refresh those first */
7792 QLIST_FOREACH(child, &bs->children, next) {
7793 bdrv_refresh_filename(child->bs);
7796 if (bs->implicit) {
7797 /* For implicit nodes, just copy everything from the single child */
7798 child = QLIST_FIRST(&bs->children);
7799 assert(QLIST_NEXT(child, next) == NULL);
7801 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7802 child->bs->exact_filename);
7803 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7805 qobject_unref(bs->full_open_options);
7806 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7808 return;
7811 backing_overridden = bdrv_backing_overridden(bs);
7813 if (bs->open_flags & BDRV_O_NO_IO) {
7814 /* Without I/O, the backing file does not change anything.
7815 * Therefore, in such a case (primarily qemu-img), we can
7816 * pretend the backing file has not been overridden even if
7817 * it technically has been. */
7818 backing_overridden = false;
7821 /* Gather the options QDict */
7822 opts = qdict_new();
7823 generate_json_filename = append_strong_runtime_options(opts, bs);
7824 generate_json_filename |= backing_overridden;
7826 if (drv->bdrv_gather_child_options) {
7827 /* Some block drivers may not want to present all of their children's
7828 * options, or name them differently from BdrvChild.name */
7829 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7830 } else {
7831 QLIST_FOREACH(child, &bs->children, next) {
7832 if (child == bs->backing && !backing_overridden) {
7833 /* We can skip the backing BDS if it has not been overridden */
7834 continue;
7837 qdict_put(opts, child->name,
7838 qobject_ref(child->bs->full_open_options));
7841 if (backing_overridden && !bs->backing) {
7842 /* Force no backing file */
7843 qdict_put_null(opts, "backing");
7847 qobject_unref(bs->full_open_options);
7848 bs->full_open_options = opts;
7850 primary_child_bs = bdrv_primary_bs(bs);
7852 if (drv->bdrv_refresh_filename) {
7853 /* Obsolete information is of no use here, so drop the old file name
7854 * information before refreshing it */
7855 bs->exact_filename[0] = '\0';
7857 drv->bdrv_refresh_filename(bs);
7858 } else if (primary_child_bs) {
7860 * Try to reconstruct valid information from the underlying
7861 * file -- this only works for format nodes (filter nodes
7862 * cannot be probed and as such must be selected by the user
7863 * either through an options dict, or through a special
7864 * filename which the filter driver must construct in its
7865 * .bdrv_refresh_filename() implementation).
7868 bs->exact_filename[0] = '\0';
7871 * We can use the underlying file's filename if:
7872 * - it has a filename,
7873 * - the current BDS is not a filter,
7874 * - the file is a protocol BDS, and
7875 * - opening that file (as this BDS's format) will automatically create
7876 * the BDS tree we have right now, that is:
7877 * - the user did not significantly change this BDS's behavior with
7878 * some explicit (strong) options
7879 * - no non-file child of this BDS has been overridden by the user
7880 * Both of these conditions are represented by generate_json_filename.
7882 if (primary_child_bs->exact_filename[0] &&
7883 primary_child_bs->drv->bdrv_file_open &&
7884 !drv->is_filter && !generate_json_filename)
7886 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7890 if (bs->exact_filename[0]) {
7891 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7892 } else {
7893 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7894 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7895 json->str) >= sizeof(bs->filename)) {
7896 /* Give user a hint if we truncated things. */
7897 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7899 g_string_free(json, true);
7903 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7905 BlockDriver *drv = bs->drv;
7906 BlockDriverState *child_bs;
7908 GLOBAL_STATE_CODE();
7910 if (!drv) {
7911 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7912 return NULL;
7915 if (drv->bdrv_dirname) {
7916 return drv->bdrv_dirname(bs, errp);
7919 child_bs = bdrv_primary_bs(bs);
7920 if (child_bs) {
7921 return bdrv_dirname(child_bs, errp);
7924 bdrv_refresh_filename(bs);
7925 if (bs->exact_filename[0] != '\0') {
7926 return path_combine(bs->exact_filename, "");
7929 error_setg(errp, "Cannot generate a base directory for %s nodes",
7930 drv->format_name);
7931 return NULL;
7935 * Hot add/remove a BDS's child. So the user can take a child offline when
7936 * it is broken and take a new child online
7938 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7939 Error **errp)
7941 GLOBAL_STATE_CODE();
7942 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7943 error_setg(errp, "The node %s does not support adding a child",
7944 bdrv_get_device_or_node_name(parent_bs));
7945 return;
7948 if (!QLIST_EMPTY(&child_bs->parents)) {
7949 error_setg(errp, "The node %s already has a parent",
7950 child_bs->node_name);
7951 return;
7954 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7957 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7959 BdrvChild *tmp;
7961 GLOBAL_STATE_CODE();
7962 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7963 error_setg(errp, "The node %s does not support removing a child",
7964 bdrv_get_device_or_node_name(parent_bs));
7965 return;
7968 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7969 if (tmp == child) {
7970 break;
7974 if (!tmp) {
7975 error_setg(errp, "The node %s does not have a child named %s",
7976 bdrv_get_device_or_node_name(parent_bs),
7977 bdrv_get_device_or_node_name(child->bs));
7978 return;
7981 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7984 int bdrv_make_empty(BdrvChild *c, Error **errp)
7986 BlockDriver *drv = c->bs->drv;
7987 int ret;
7989 GLOBAL_STATE_CODE();
7990 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7992 if (!drv->bdrv_make_empty) {
7993 error_setg(errp, "%s does not support emptying nodes",
7994 drv->format_name);
7995 return -ENOTSUP;
7998 ret = drv->bdrv_make_empty(c->bs);
7999 if (ret < 0) {
8000 error_setg_errno(errp, -ret, "Failed to empty %s",
8001 c->bs->filename);
8002 return ret;
8005 return 0;
8009 * Return the child that @bs acts as an overlay for, and from which data may be
8010 * copied in COW or COR operations. Usually this is the backing file.
8012 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8014 IO_CODE();
8016 if (!bs || !bs->drv) {
8017 return NULL;
8020 if (bs->drv->is_filter) {
8021 return NULL;
8024 if (!bs->backing) {
8025 return NULL;
8028 assert(bs->backing->role & BDRV_CHILD_COW);
8029 return bs->backing;
8033 * If @bs acts as a filter for exactly one of its children, return
8034 * that child.
8036 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8038 BdrvChild *c;
8039 IO_CODE();
8041 if (!bs || !bs->drv) {
8042 return NULL;
8045 if (!bs->drv->is_filter) {
8046 return NULL;
8049 /* Only one of @backing or @file may be used */
8050 assert(!(bs->backing && bs->file));
8052 c = bs->backing ?: bs->file;
8053 if (!c) {
8054 return NULL;
8057 assert(c->role & BDRV_CHILD_FILTERED);
8058 return c;
8062 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8063 * whichever is non-NULL.
8065 * Return NULL if both are NULL.
8067 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8069 BdrvChild *cow_child = bdrv_cow_child(bs);
8070 BdrvChild *filter_child = bdrv_filter_child(bs);
8071 IO_CODE();
8073 /* Filter nodes cannot have COW backing files */
8074 assert(!(cow_child && filter_child));
8076 return cow_child ?: filter_child;
8080 * Return the primary child of this node: For filters, that is the
8081 * filtered child. For other nodes, that is usually the child storing
8082 * metadata.
8083 * (A generally more helpful description is that this is (usually) the
8084 * child that has the same filename as @bs.)
8086 * Drivers do not necessarily have a primary child; for example quorum
8087 * does not.
8089 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8091 BdrvChild *c, *found = NULL;
8092 IO_CODE();
8094 QLIST_FOREACH(c, &bs->children, next) {
8095 if (c->role & BDRV_CHILD_PRIMARY) {
8096 assert(!found);
8097 found = c;
8101 return found;
8104 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8105 bool stop_on_explicit_filter)
8107 BdrvChild *c;
8109 if (!bs) {
8110 return NULL;
8113 while (!(stop_on_explicit_filter && !bs->implicit)) {
8114 c = bdrv_filter_child(bs);
8115 if (!c) {
8117 * A filter that is embedded in a working block graph must
8118 * have a child. Assert this here so this function does
8119 * not return a filter node that is not expected by the
8120 * caller.
8122 assert(!bs->drv || !bs->drv->is_filter);
8123 break;
8125 bs = c->bs;
8128 * Note that this treats nodes with bs->drv == NULL as not being
8129 * filters (bs->drv == NULL should be replaced by something else
8130 * anyway).
8131 * The advantage of this behavior is that this function will thus
8132 * always return a non-NULL value (given a non-NULL @bs).
8135 return bs;
8139 * Return the first BDS that has not been added implicitly or that
8140 * does not have a filtered child down the chain starting from @bs
8141 * (including @bs itself).
8143 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8145 GLOBAL_STATE_CODE();
8146 return bdrv_do_skip_filters(bs, true);
8150 * Return the first BDS that does not have a filtered child down the
8151 * chain starting from @bs (including @bs itself).
8153 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8155 IO_CODE();
8156 return bdrv_do_skip_filters(bs, false);
8160 * For a backing chain, return the first non-filter backing image of
8161 * the first non-filter image.
8163 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8165 IO_CODE();
8166 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8170 * Check whether [offset, offset + bytes) overlaps with the cached
8171 * block-status data region.
8173 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8174 * which is what bdrv_bsc_is_data()'s interface needs.
8175 * Otherwise, *pnum is not touched.
8177 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8178 int64_t offset, int64_t bytes,
8179 int64_t *pnum)
8181 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8182 bool overlaps;
8184 overlaps =
8185 qatomic_read(&bsc->valid) &&
8186 ranges_overlap(offset, bytes, bsc->data_start,
8187 bsc->data_end - bsc->data_start);
8189 if (overlaps && pnum) {
8190 *pnum = bsc->data_end - offset;
8193 return overlaps;
8197 * See block_int.h for this function's documentation.
8199 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8201 IO_CODE();
8202 RCU_READ_LOCK_GUARD();
8203 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8207 * See block_int.h for this function's documentation.
8209 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8210 int64_t offset, int64_t bytes)
8212 IO_CODE();
8213 RCU_READ_LOCK_GUARD();
8215 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8216 qatomic_set(&bs->block_status_cache->valid, false);
8221 * See block_int.h for this function's documentation.
8223 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8225 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8226 BdrvBlockStatusCache *old_bsc;
8227 IO_CODE();
8229 *new_bsc = (BdrvBlockStatusCache) {
8230 .valid = true,
8231 .data_start = offset,
8232 .data_end = offset + bytes,
8235 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8237 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8238 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8239 if (old_bsc) {
8240 g_free_rcu(old_bsc, rcu);