qboot: update to latest submodule
[qemu.git] / block.c
blobbc85f46eed690321657cf1de5d27cbe56765c6b5
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, sizeof(buf), buf, 0);
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, sizeof(buf), filename);
1049 if (!drv) {
1050 error_setg(errp, "Could not determine image format: No compatible "
1051 "driver found");
1052 *pdrv = NULL;
1053 return -ENOENT;
1056 *pdrv = drv;
1057 return 0;
1061 * Set the current 'total_sectors' value
1062 * Return 0 on success, -errno on error.
1064 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
1066 BlockDriver *drv = bs->drv;
1067 IO_CODE();
1069 if (!drv) {
1070 return -ENOMEDIUM;
1073 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1074 if (bdrv_is_sg(bs))
1075 return 0;
1077 /* query actual device if possible, otherwise just trust the hint */
1078 if (drv->bdrv_getlength) {
1079 int64_t length = drv->bdrv_getlength(bs);
1080 if (length < 0) {
1081 return length;
1083 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1086 bs->total_sectors = hint;
1088 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1089 return -EFBIG;
1092 return 0;
1096 * Combines a QDict of new block driver @options with any missing options taken
1097 * from @old_options, so that leaving out an option defaults to its old value.
1099 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1100 QDict *old_options)
1102 GLOBAL_STATE_CODE();
1103 if (bs->drv && bs->drv->bdrv_join_options) {
1104 bs->drv->bdrv_join_options(options, old_options);
1105 } else {
1106 qdict_join(options, old_options, false);
1110 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1111 int open_flags,
1112 Error **errp)
1114 Error *local_err = NULL;
1115 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1116 BlockdevDetectZeroesOptions detect_zeroes =
1117 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1118 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1119 GLOBAL_STATE_CODE();
1120 g_free(value);
1121 if (local_err) {
1122 error_propagate(errp, local_err);
1123 return detect_zeroes;
1126 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1127 !(open_flags & BDRV_O_UNMAP))
1129 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1130 "without setting discard operation to unmap");
1133 return detect_zeroes;
1137 * Set open flags for aio engine
1139 * Return 0 on success, -1 if the engine specified is invalid
1141 int bdrv_parse_aio(const char *mode, int *flags)
1143 if (!strcmp(mode, "threads")) {
1144 /* do nothing, default */
1145 } else if (!strcmp(mode, "native")) {
1146 *flags |= BDRV_O_NATIVE_AIO;
1147 #ifdef CONFIG_LINUX_IO_URING
1148 } else if (!strcmp(mode, "io_uring")) {
1149 *flags |= BDRV_O_IO_URING;
1150 #endif
1151 } else {
1152 return -1;
1155 return 0;
1159 * Set open flags for a given discard mode
1161 * Return 0 on success, -1 if the discard mode was invalid.
1163 int bdrv_parse_discard_flags(const char *mode, int *flags)
1165 *flags &= ~BDRV_O_UNMAP;
1167 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1168 /* do nothing */
1169 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1170 *flags |= BDRV_O_UNMAP;
1171 } else {
1172 return -1;
1175 return 0;
1179 * Set open flags for a given cache mode
1181 * Return 0 on success, -1 if the cache mode was invalid.
1183 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1185 *flags &= ~BDRV_O_CACHE_MASK;
1187 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1188 *writethrough = false;
1189 *flags |= BDRV_O_NOCACHE;
1190 } else if (!strcmp(mode, "directsync")) {
1191 *writethrough = true;
1192 *flags |= BDRV_O_NOCACHE;
1193 } else if (!strcmp(mode, "writeback")) {
1194 *writethrough = false;
1195 } else if (!strcmp(mode, "unsafe")) {
1196 *writethrough = false;
1197 *flags |= BDRV_O_NO_FLUSH;
1198 } else if (!strcmp(mode, "writethrough")) {
1199 *writethrough = true;
1200 } else {
1201 return -1;
1204 return 0;
1207 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1209 BlockDriverState *parent = c->opaque;
1210 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1213 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1215 BlockDriverState *bs = child->opaque;
1216 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1219 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1221 BlockDriverState *bs = child->opaque;
1222 return bdrv_drain_poll(bs, false, NULL, false);
1225 static void bdrv_child_cb_drained_end(BdrvChild *child,
1226 int *drained_end_counter)
1228 BlockDriverState *bs = child->opaque;
1229 bdrv_drained_end_no_poll(bs, drained_end_counter);
1232 static int bdrv_child_cb_inactivate(BdrvChild *child)
1234 BlockDriverState *bs = child->opaque;
1235 GLOBAL_STATE_CODE();
1236 assert(bs->open_flags & BDRV_O_INACTIVE);
1237 return 0;
1240 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1241 GSList **ignore, Error **errp)
1243 BlockDriverState *bs = child->opaque;
1244 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1247 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1248 GSList **ignore)
1250 BlockDriverState *bs = child->opaque;
1251 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1255 * Returns the options and flags that a temporary snapshot should get, based on
1256 * the originally requested flags (the originally requested image will have
1257 * flags like a backing file)
1259 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1260 int parent_flags, QDict *parent_options)
1262 GLOBAL_STATE_CODE();
1263 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1265 /* For temporary files, unconditional cache=unsafe is fine */
1266 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1267 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1269 /* Copy the read-only and discard options from the parent */
1270 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1271 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1273 /* aio=native doesn't work for cache.direct=off, so disable it for the
1274 * temporary snapshot */
1275 *child_flags &= ~BDRV_O_NATIVE_AIO;
1278 static void bdrv_backing_attach(BdrvChild *c)
1280 BlockDriverState *parent = c->opaque;
1281 BlockDriverState *backing_hd = c->bs;
1283 GLOBAL_STATE_CODE();
1284 assert(!parent->backing_blocker);
1285 error_setg(&parent->backing_blocker,
1286 "node is used as backing hd of '%s'",
1287 bdrv_get_device_or_node_name(parent));
1289 bdrv_refresh_filename(backing_hd);
1291 parent->open_flags &= ~BDRV_O_NO_BACKING;
1293 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1294 /* Otherwise we won't be able to commit or stream */
1295 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1296 parent->backing_blocker);
1297 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1298 parent->backing_blocker);
1300 * We do backup in 3 ways:
1301 * 1. drive backup
1302 * The target bs is new opened, and the source is top BDS
1303 * 2. blockdev backup
1304 * Both the source and the target are top BDSes.
1305 * 3. internal backup(used for block replication)
1306 * Both the source and the target are backing file
1308 * In case 1 and 2, neither the source nor the target is the backing file.
1309 * In case 3, we will block the top BDS, so there is only one block job
1310 * for the top BDS and its backing chain.
1312 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1313 parent->backing_blocker);
1314 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1315 parent->backing_blocker);
1318 static void bdrv_backing_detach(BdrvChild *c)
1320 BlockDriverState *parent = c->opaque;
1322 GLOBAL_STATE_CODE();
1323 assert(parent->backing_blocker);
1324 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1325 error_free(parent->backing_blocker);
1326 parent->backing_blocker = NULL;
1329 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1330 const char *filename, Error **errp)
1332 BlockDriverState *parent = c->opaque;
1333 bool read_only = bdrv_is_read_only(parent);
1334 int ret;
1335 GLOBAL_STATE_CODE();
1337 if (read_only) {
1338 ret = bdrv_reopen_set_read_only(parent, false, errp);
1339 if (ret < 0) {
1340 return ret;
1344 ret = bdrv_change_backing_file(parent, filename,
1345 base->drv ? base->drv->format_name : "",
1346 false);
1347 if (ret < 0) {
1348 error_setg_errno(errp, -ret, "Could not update backing file link");
1351 if (read_only) {
1352 bdrv_reopen_set_read_only(parent, true, NULL);
1355 return ret;
1359 * Returns the options and flags that a generic child of a BDS should
1360 * get, based on the given options and flags for the parent BDS.
1362 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1363 int *child_flags, QDict *child_options,
1364 int parent_flags, QDict *parent_options)
1366 int flags = parent_flags;
1367 GLOBAL_STATE_CODE();
1370 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1371 * Generally, the question to answer is: Should this child be
1372 * format-probed by default?
1376 * Pure and non-filtered data children of non-format nodes should
1377 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1378 * set). This only affects a very limited set of drivers (namely
1379 * quorum and blkverify when this comment was written).
1380 * Force-clear BDRV_O_PROTOCOL then.
1382 if (!parent_is_format &&
1383 (role & BDRV_CHILD_DATA) &&
1384 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1386 flags &= ~BDRV_O_PROTOCOL;
1390 * All children of format nodes (except for COW children) and all
1391 * metadata children in general should never be format-probed.
1392 * Force-set BDRV_O_PROTOCOL then.
1394 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1395 (role & BDRV_CHILD_METADATA))
1397 flags |= BDRV_O_PROTOCOL;
1401 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1402 * the parent.
1404 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1405 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1406 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1408 if (role & BDRV_CHILD_COW) {
1409 /* backing files are opened read-only by default */
1410 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1411 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1412 } else {
1413 /* Inherit the read-only option from the parent if it's not set */
1414 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1415 qdict_copy_default(child_options, parent_options,
1416 BDRV_OPT_AUTO_READ_ONLY);
1420 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1421 * can default to enable it on lower layers regardless of the
1422 * parent option.
1424 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1426 /* Clear flags that only apply to the top layer */
1427 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1429 if (role & BDRV_CHILD_METADATA) {
1430 flags &= ~BDRV_O_NO_IO;
1432 if (role & BDRV_CHILD_COW) {
1433 flags &= ~BDRV_O_TEMPORARY;
1436 *child_flags = flags;
1439 static void bdrv_child_cb_attach(BdrvChild *child)
1441 BlockDriverState *bs = child->opaque;
1443 assert_bdrv_graph_writable(bs);
1444 QLIST_INSERT_HEAD(&bs->children, child, next);
1446 if (child->role & BDRV_CHILD_COW) {
1447 bdrv_backing_attach(child);
1450 bdrv_apply_subtree_drain(child, bs);
1453 static void bdrv_child_cb_detach(BdrvChild *child)
1455 BlockDriverState *bs = child->opaque;
1457 if (child->role & BDRV_CHILD_COW) {
1458 bdrv_backing_detach(child);
1461 bdrv_unapply_subtree_drain(child, bs);
1463 assert_bdrv_graph_writable(bs);
1464 QLIST_REMOVE(child, next);
1467 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1468 const char *filename, Error **errp)
1470 if (c->role & BDRV_CHILD_COW) {
1471 return bdrv_backing_update_filename(c, base, filename, errp);
1473 return 0;
1476 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1478 BlockDriverState *bs = c->opaque;
1479 IO_CODE();
1481 return bdrv_get_aio_context(bs);
1484 const BdrvChildClass child_of_bds = {
1485 .parent_is_bds = true,
1486 .get_parent_desc = bdrv_child_get_parent_desc,
1487 .inherit_options = bdrv_inherited_options,
1488 .drained_begin = bdrv_child_cb_drained_begin,
1489 .drained_poll = bdrv_child_cb_drained_poll,
1490 .drained_end = bdrv_child_cb_drained_end,
1491 .attach = bdrv_child_cb_attach,
1492 .detach = bdrv_child_cb_detach,
1493 .inactivate = bdrv_child_cb_inactivate,
1494 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1495 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1496 .update_filename = bdrv_child_cb_update_filename,
1497 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1500 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1502 GLOBAL_STATE_CODE();
1503 return c->klass->get_parent_aio_context(c);
1506 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1508 int open_flags = flags;
1509 GLOBAL_STATE_CODE();
1512 * Clear flags that are internal to the block layer before opening the
1513 * image.
1515 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1517 return open_flags;
1520 static void update_flags_from_options(int *flags, QemuOpts *opts)
1522 GLOBAL_STATE_CODE();
1524 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1526 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1527 *flags |= BDRV_O_NO_FLUSH;
1530 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1531 *flags |= BDRV_O_NOCACHE;
1534 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1535 *flags |= BDRV_O_RDWR;
1538 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1539 *flags |= BDRV_O_AUTO_RDONLY;
1543 static void update_options_from_flags(QDict *options, int flags)
1545 GLOBAL_STATE_CODE();
1546 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1547 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1549 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1550 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1551 flags & BDRV_O_NO_FLUSH);
1553 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1554 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1556 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1557 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1558 flags & BDRV_O_AUTO_RDONLY);
1562 static void bdrv_assign_node_name(BlockDriverState *bs,
1563 const char *node_name,
1564 Error **errp)
1566 char *gen_node_name = NULL;
1567 GLOBAL_STATE_CODE();
1569 if (!node_name) {
1570 node_name = gen_node_name = id_generate(ID_BLOCK);
1571 } else if (!id_wellformed(node_name)) {
1573 * Check for empty string or invalid characters, but not if it is
1574 * generated (generated names use characters not available to the user)
1576 error_setg(errp, "Invalid node-name: '%s'", node_name);
1577 return;
1580 /* takes care of avoiding namespaces collisions */
1581 if (blk_by_name(node_name)) {
1582 error_setg(errp, "node-name=%s is conflicting with a device id",
1583 node_name);
1584 goto out;
1587 /* takes care of avoiding duplicates node names */
1588 if (bdrv_find_node(node_name)) {
1589 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1590 goto out;
1593 /* Make sure that the node name isn't truncated */
1594 if (strlen(node_name) >= sizeof(bs->node_name)) {
1595 error_setg(errp, "Node name too long");
1596 goto out;
1599 /* copy node name into the bs and insert it into the graph list */
1600 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1601 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1602 out:
1603 g_free(gen_node_name);
1606 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1607 const char *node_name, QDict *options,
1608 int open_flags, Error **errp)
1610 Error *local_err = NULL;
1611 int i, ret;
1612 GLOBAL_STATE_CODE();
1614 bdrv_assign_node_name(bs, node_name, &local_err);
1615 if (local_err) {
1616 error_propagate(errp, local_err);
1617 return -EINVAL;
1620 bs->drv = drv;
1621 bs->opaque = g_malloc0(drv->instance_size);
1623 if (drv->bdrv_file_open) {
1624 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1625 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1626 } else if (drv->bdrv_open) {
1627 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1628 } else {
1629 ret = 0;
1632 if (ret < 0) {
1633 if (local_err) {
1634 error_propagate(errp, local_err);
1635 } else if (bs->filename[0]) {
1636 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1637 } else {
1638 error_setg_errno(errp, -ret, "Could not open image");
1640 goto open_failed;
1643 ret = refresh_total_sectors(bs, bs->total_sectors);
1644 if (ret < 0) {
1645 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1646 return ret;
1649 bdrv_refresh_limits(bs, NULL, &local_err);
1650 if (local_err) {
1651 error_propagate(errp, local_err);
1652 return -EINVAL;
1655 assert(bdrv_opt_mem_align(bs) != 0);
1656 assert(bdrv_min_mem_align(bs) != 0);
1657 assert(is_power_of_2(bs->bl.request_alignment));
1659 for (i = 0; i < bs->quiesce_counter; i++) {
1660 if (drv->bdrv_co_drain_begin) {
1661 drv->bdrv_co_drain_begin(bs);
1665 return 0;
1666 open_failed:
1667 bs->drv = NULL;
1668 if (bs->file != NULL) {
1669 bdrv_unref_child(bs, bs->file);
1670 bs->file = NULL;
1672 g_free(bs->opaque);
1673 bs->opaque = NULL;
1674 return ret;
1678 * Create and open a block node.
1680 * @options is a QDict of options to pass to the block drivers, or NULL for an
1681 * empty set of options. The reference to the QDict belongs to the block layer
1682 * after the call (even on failure), so if the caller intends to reuse the
1683 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1685 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1686 const char *node_name,
1687 QDict *options, int flags,
1688 Error **errp)
1690 BlockDriverState *bs;
1691 int ret;
1693 GLOBAL_STATE_CODE();
1695 bs = bdrv_new();
1696 bs->open_flags = flags;
1697 bs->options = options ?: qdict_new();
1698 bs->explicit_options = qdict_clone_shallow(bs->options);
1699 bs->opaque = NULL;
1701 update_options_from_flags(bs->options, flags);
1703 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1704 if (ret < 0) {
1705 qobject_unref(bs->explicit_options);
1706 bs->explicit_options = NULL;
1707 qobject_unref(bs->options);
1708 bs->options = NULL;
1709 bdrv_unref(bs);
1710 return NULL;
1713 return bs;
1716 /* Create and open a block node. */
1717 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1718 int flags, Error **errp)
1720 GLOBAL_STATE_CODE();
1721 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1724 QemuOptsList bdrv_runtime_opts = {
1725 .name = "bdrv_common",
1726 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1727 .desc = {
1729 .name = "node-name",
1730 .type = QEMU_OPT_STRING,
1731 .help = "Node name of the block device node",
1734 .name = "driver",
1735 .type = QEMU_OPT_STRING,
1736 .help = "Block driver to use for the node",
1739 .name = BDRV_OPT_CACHE_DIRECT,
1740 .type = QEMU_OPT_BOOL,
1741 .help = "Bypass software writeback cache on the host",
1744 .name = BDRV_OPT_CACHE_NO_FLUSH,
1745 .type = QEMU_OPT_BOOL,
1746 .help = "Ignore flush requests",
1749 .name = BDRV_OPT_READ_ONLY,
1750 .type = QEMU_OPT_BOOL,
1751 .help = "Node is opened in read-only mode",
1754 .name = BDRV_OPT_AUTO_READ_ONLY,
1755 .type = QEMU_OPT_BOOL,
1756 .help = "Node can become read-only if opening read-write fails",
1759 .name = "detect-zeroes",
1760 .type = QEMU_OPT_STRING,
1761 .help = "try to optimize zero writes (off, on, unmap)",
1764 .name = BDRV_OPT_DISCARD,
1765 .type = QEMU_OPT_STRING,
1766 .help = "discard operation (ignore/off, unmap/on)",
1769 .name = BDRV_OPT_FORCE_SHARE,
1770 .type = QEMU_OPT_BOOL,
1771 .help = "always accept other writers (default: off)",
1773 { /* end of list */ }
1777 QemuOptsList bdrv_create_opts_simple = {
1778 .name = "simple-create-opts",
1779 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1780 .desc = {
1782 .name = BLOCK_OPT_SIZE,
1783 .type = QEMU_OPT_SIZE,
1784 .help = "Virtual disk size"
1787 .name = BLOCK_OPT_PREALLOC,
1788 .type = QEMU_OPT_STRING,
1789 .help = "Preallocation mode (allowed values: off)"
1791 { /* end of list */ }
1796 * Common part for opening disk images and files
1798 * Removes all processed options from *options.
1800 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1801 QDict *options, Error **errp)
1803 int ret, open_flags;
1804 const char *filename;
1805 const char *driver_name = NULL;
1806 const char *node_name = NULL;
1807 const char *discard;
1808 QemuOpts *opts;
1809 BlockDriver *drv;
1810 Error *local_err = NULL;
1811 bool ro;
1813 assert(bs->file == NULL);
1814 assert(options != NULL && bs->options != options);
1815 GLOBAL_STATE_CODE();
1817 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1818 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1819 ret = -EINVAL;
1820 goto fail_opts;
1823 update_flags_from_options(&bs->open_flags, opts);
1825 driver_name = qemu_opt_get(opts, "driver");
1826 drv = bdrv_find_format(driver_name);
1827 assert(drv != NULL);
1829 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1831 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1832 error_setg(errp,
1833 BDRV_OPT_FORCE_SHARE
1834 "=on can only be used with read-only images");
1835 ret = -EINVAL;
1836 goto fail_opts;
1839 if (file != NULL) {
1840 bdrv_refresh_filename(blk_bs(file));
1841 filename = blk_bs(file)->filename;
1842 } else {
1844 * Caution: while qdict_get_try_str() is fine, getting
1845 * non-string types would require more care. When @options
1846 * come from -blockdev or blockdev_add, its members are typed
1847 * according to the QAPI schema, but when they come from
1848 * -drive, they're all QString.
1850 filename = qdict_get_try_str(options, "filename");
1853 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1854 error_setg(errp, "The '%s' block driver requires a file name",
1855 drv->format_name);
1856 ret = -EINVAL;
1857 goto fail_opts;
1860 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1861 drv->format_name);
1863 ro = bdrv_is_read_only(bs);
1865 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1866 if (!ro && bdrv_is_whitelisted(drv, true)) {
1867 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1868 } else {
1869 ret = -ENOTSUP;
1871 if (ret < 0) {
1872 error_setg(errp,
1873 !ro && bdrv_is_whitelisted(drv, true)
1874 ? "Driver '%s' can only be used for read-only devices"
1875 : "Driver '%s' is not whitelisted",
1876 drv->format_name);
1877 goto fail_opts;
1881 /* bdrv_new() and bdrv_close() make it so */
1882 assert(qatomic_read(&bs->copy_on_read) == 0);
1884 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1885 if (!ro) {
1886 bdrv_enable_copy_on_read(bs);
1887 } else {
1888 error_setg(errp, "Can't use copy-on-read on read-only device");
1889 ret = -EINVAL;
1890 goto fail_opts;
1894 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1895 if (discard != NULL) {
1896 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1897 error_setg(errp, "Invalid discard option");
1898 ret = -EINVAL;
1899 goto fail_opts;
1903 bs->detect_zeroes =
1904 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1905 if (local_err) {
1906 error_propagate(errp, local_err);
1907 ret = -EINVAL;
1908 goto fail_opts;
1911 if (filename != NULL) {
1912 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1913 } else {
1914 bs->filename[0] = '\0';
1916 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1918 /* Open the image, either directly or using a protocol */
1919 open_flags = bdrv_open_flags(bs, bs->open_flags);
1920 node_name = qemu_opt_get(opts, "node-name");
1922 assert(!drv->bdrv_file_open || file == NULL);
1923 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1924 if (ret < 0) {
1925 goto fail_opts;
1928 qemu_opts_del(opts);
1929 return 0;
1931 fail_opts:
1932 qemu_opts_del(opts);
1933 return ret;
1936 static QDict *parse_json_filename(const char *filename, Error **errp)
1938 QObject *options_obj;
1939 QDict *options;
1940 int ret;
1941 GLOBAL_STATE_CODE();
1943 ret = strstart(filename, "json:", &filename);
1944 assert(ret);
1946 options_obj = qobject_from_json(filename, errp);
1947 if (!options_obj) {
1948 error_prepend(errp, "Could not parse the JSON options: ");
1949 return NULL;
1952 options = qobject_to(QDict, options_obj);
1953 if (!options) {
1954 qobject_unref(options_obj);
1955 error_setg(errp, "Invalid JSON object given");
1956 return NULL;
1959 qdict_flatten(options);
1961 return options;
1964 static void parse_json_protocol(QDict *options, const char **pfilename,
1965 Error **errp)
1967 QDict *json_options;
1968 Error *local_err = NULL;
1969 GLOBAL_STATE_CODE();
1971 /* Parse json: pseudo-protocol */
1972 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1973 return;
1976 json_options = parse_json_filename(*pfilename, &local_err);
1977 if (local_err) {
1978 error_propagate(errp, local_err);
1979 return;
1982 /* Options given in the filename have lower priority than options
1983 * specified directly */
1984 qdict_join(options, json_options, false);
1985 qobject_unref(json_options);
1986 *pfilename = NULL;
1990 * Fills in default options for opening images and converts the legacy
1991 * filename/flags pair to option QDict entries.
1992 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1993 * block driver has been specified explicitly.
1995 static int bdrv_fill_options(QDict **options, const char *filename,
1996 int *flags, Error **errp)
1998 const char *drvname;
1999 bool protocol = *flags & BDRV_O_PROTOCOL;
2000 bool parse_filename = false;
2001 BlockDriver *drv = NULL;
2002 Error *local_err = NULL;
2004 GLOBAL_STATE_CODE();
2007 * Caution: while qdict_get_try_str() is fine, getting non-string
2008 * types would require more care. When @options come from
2009 * -blockdev or blockdev_add, its members are typed according to
2010 * the QAPI schema, but when they come from -drive, they're all
2011 * QString.
2013 drvname = qdict_get_try_str(*options, "driver");
2014 if (drvname) {
2015 drv = bdrv_find_format(drvname);
2016 if (!drv) {
2017 error_setg(errp, "Unknown driver '%s'", drvname);
2018 return -ENOENT;
2020 /* If the user has explicitly specified the driver, this choice should
2021 * override the BDRV_O_PROTOCOL flag */
2022 protocol = drv->bdrv_file_open;
2025 if (protocol) {
2026 *flags |= BDRV_O_PROTOCOL;
2027 } else {
2028 *flags &= ~BDRV_O_PROTOCOL;
2031 /* Translate cache options from flags into options */
2032 update_options_from_flags(*options, *flags);
2034 /* Fetch the file name from the options QDict if necessary */
2035 if (protocol && filename) {
2036 if (!qdict_haskey(*options, "filename")) {
2037 qdict_put_str(*options, "filename", filename);
2038 parse_filename = true;
2039 } else {
2040 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2041 "the same time");
2042 return -EINVAL;
2046 /* Find the right block driver */
2047 /* See cautionary note on accessing @options above */
2048 filename = qdict_get_try_str(*options, "filename");
2050 if (!drvname && protocol) {
2051 if (filename) {
2052 drv = bdrv_find_protocol(filename, parse_filename, errp);
2053 if (!drv) {
2054 return -EINVAL;
2057 drvname = drv->format_name;
2058 qdict_put_str(*options, "driver", drvname);
2059 } else {
2060 error_setg(errp, "Must specify either driver or file");
2061 return -EINVAL;
2065 assert(drv || !protocol);
2067 /* Driver-specific filename parsing */
2068 if (drv && drv->bdrv_parse_filename && parse_filename) {
2069 drv->bdrv_parse_filename(filename, *options, &local_err);
2070 if (local_err) {
2071 error_propagate(errp, local_err);
2072 return -EINVAL;
2075 if (!drv->bdrv_needs_filename) {
2076 qdict_del(*options, "filename");
2080 return 0;
2083 typedef struct BlockReopenQueueEntry {
2084 bool prepared;
2085 bool perms_checked;
2086 BDRVReopenState state;
2087 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2088 } BlockReopenQueueEntry;
2091 * Return the flags that @bs will have after the reopens in @q have
2092 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2093 * return the current flags.
2095 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2097 BlockReopenQueueEntry *entry;
2099 if (q != NULL) {
2100 QTAILQ_FOREACH(entry, q, entry) {
2101 if (entry->state.bs == bs) {
2102 return entry->state.flags;
2107 return bs->open_flags;
2110 /* Returns whether the image file can be written to after the reopen queue @q
2111 * has been successfully applied, or right now if @q is NULL. */
2112 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2113 BlockReopenQueue *q)
2115 int flags = bdrv_reopen_get_flags(q, bs);
2117 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2121 * Return whether the BDS can be written to. This is not necessarily
2122 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2123 * be written to but do not count as read-only images.
2125 bool bdrv_is_writable(BlockDriverState *bs)
2127 IO_CODE();
2128 return bdrv_is_writable_after_reopen(bs, NULL);
2131 static char *bdrv_child_user_desc(BdrvChild *c)
2133 GLOBAL_STATE_CODE();
2134 return c->klass->get_parent_desc(c);
2138 * Check that @a allows everything that @b needs. @a and @b must reference same
2139 * child node.
2141 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2143 const char *child_bs_name;
2144 g_autofree char *a_user = NULL;
2145 g_autofree char *b_user = NULL;
2146 g_autofree char *perms = NULL;
2148 assert(a->bs);
2149 assert(a->bs == b->bs);
2150 GLOBAL_STATE_CODE();
2152 if ((b->perm & a->shared_perm) == b->perm) {
2153 return true;
2156 child_bs_name = bdrv_get_node_name(b->bs);
2157 a_user = bdrv_child_user_desc(a);
2158 b_user = bdrv_child_user_desc(b);
2159 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2161 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2162 "both required by %s (uses node '%s' as '%s' child) and "
2163 "unshared by %s (uses node '%s' as '%s' child).",
2164 child_bs_name, perms,
2165 b_user, child_bs_name, b->name,
2166 a_user, child_bs_name, a->name);
2168 return false;
2171 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2173 BdrvChild *a, *b;
2174 GLOBAL_STATE_CODE();
2177 * During the loop we'll look at each pair twice. That's correct because
2178 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2179 * directions.
2181 QLIST_FOREACH(a, &bs->parents, next_parent) {
2182 QLIST_FOREACH(b, &bs->parents, next_parent) {
2183 if (a == b) {
2184 continue;
2187 if (!bdrv_a_allow_b(a, b, errp)) {
2188 return true;
2193 return false;
2196 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2197 BdrvChild *c, BdrvChildRole role,
2198 BlockReopenQueue *reopen_queue,
2199 uint64_t parent_perm, uint64_t parent_shared,
2200 uint64_t *nperm, uint64_t *nshared)
2202 assert(bs->drv && bs->drv->bdrv_child_perm);
2203 GLOBAL_STATE_CODE();
2204 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2205 parent_perm, parent_shared,
2206 nperm, nshared);
2207 /* TODO Take force_share from reopen_queue */
2208 if (child_bs && child_bs->force_share) {
2209 *nshared = BLK_PERM_ALL;
2214 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2215 * nodes that are already in the @list, of course) so that final list is
2216 * topologically sorted. Return the result (GSList @list object is updated, so
2217 * don't use old reference after function call).
2219 * On function start @list must be already topologically sorted and for any node
2220 * in the @list the whole subtree of the node must be in the @list as well. The
2221 * simplest way to satisfy this criteria: use only result of
2222 * bdrv_topological_dfs() or NULL as @list parameter.
2224 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2225 BlockDriverState *bs)
2227 BdrvChild *child;
2228 g_autoptr(GHashTable) local_found = NULL;
2230 GLOBAL_STATE_CODE();
2232 if (!found) {
2233 assert(!list);
2234 found = local_found = g_hash_table_new(NULL, NULL);
2237 if (g_hash_table_contains(found, bs)) {
2238 return list;
2240 g_hash_table_add(found, bs);
2242 QLIST_FOREACH(child, &bs->children, next) {
2243 list = bdrv_topological_dfs(list, found, child->bs);
2246 return g_slist_prepend(list, bs);
2249 typedef struct BdrvChildSetPermState {
2250 BdrvChild *child;
2251 uint64_t old_perm;
2252 uint64_t old_shared_perm;
2253 } BdrvChildSetPermState;
2255 static void bdrv_child_set_perm_abort(void *opaque)
2257 BdrvChildSetPermState *s = opaque;
2259 GLOBAL_STATE_CODE();
2261 s->child->perm = s->old_perm;
2262 s->child->shared_perm = s->old_shared_perm;
2265 static TransactionActionDrv bdrv_child_set_pem_drv = {
2266 .abort = bdrv_child_set_perm_abort,
2267 .clean = g_free,
2270 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2271 uint64_t shared, Transaction *tran)
2273 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2274 GLOBAL_STATE_CODE();
2276 *s = (BdrvChildSetPermState) {
2277 .child = c,
2278 .old_perm = c->perm,
2279 .old_shared_perm = c->shared_perm,
2282 c->perm = perm;
2283 c->shared_perm = shared;
2285 tran_add(tran, &bdrv_child_set_pem_drv, s);
2288 static void bdrv_drv_set_perm_commit(void *opaque)
2290 BlockDriverState *bs = opaque;
2291 uint64_t cumulative_perms, cumulative_shared_perms;
2292 GLOBAL_STATE_CODE();
2294 if (bs->drv->bdrv_set_perm) {
2295 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2296 &cumulative_shared_perms);
2297 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2301 static void bdrv_drv_set_perm_abort(void *opaque)
2303 BlockDriverState *bs = opaque;
2304 GLOBAL_STATE_CODE();
2306 if (bs->drv->bdrv_abort_perm_update) {
2307 bs->drv->bdrv_abort_perm_update(bs);
2311 TransactionActionDrv bdrv_drv_set_perm_drv = {
2312 .abort = bdrv_drv_set_perm_abort,
2313 .commit = bdrv_drv_set_perm_commit,
2316 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2317 uint64_t shared_perm, Transaction *tran,
2318 Error **errp)
2320 GLOBAL_STATE_CODE();
2321 if (!bs->drv) {
2322 return 0;
2325 if (bs->drv->bdrv_check_perm) {
2326 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2327 if (ret < 0) {
2328 return ret;
2332 if (tran) {
2333 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2336 return 0;
2339 typedef struct BdrvReplaceChildState {
2340 BdrvChild *child;
2341 BdrvChild **childp;
2342 BlockDriverState *old_bs;
2343 bool free_empty_child;
2344 } BdrvReplaceChildState;
2346 static void bdrv_replace_child_commit(void *opaque)
2348 BdrvReplaceChildState *s = opaque;
2349 GLOBAL_STATE_CODE();
2351 if (s->free_empty_child && !s->child->bs) {
2352 bdrv_child_free(s->child);
2354 bdrv_unref(s->old_bs);
2357 static void bdrv_replace_child_abort(void *opaque)
2359 BdrvReplaceChildState *s = opaque;
2360 BlockDriverState *new_bs = s->child->bs;
2362 GLOBAL_STATE_CODE();
2364 * old_bs reference is transparently moved from @s to s->child.
2366 * Pass &s->child here instead of s->childp, because:
2367 * (1) s->old_bs must be non-NULL, so bdrv_replace_child_noperm() will not
2368 * modify the BdrvChild * pointer we indirectly pass to it, i.e. it
2369 * will not modify s->child. From that perspective, it does not matter
2370 * whether we pass s->childp or &s->child.
2371 * (2) If new_bs is not NULL, s->childp will be NULL. We then cannot use
2372 * it here.
2373 * (3) If new_bs is NULL, *s->childp will have been NULLed by
2374 * bdrv_replace_child_tran()'s bdrv_replace_child_noperm() call, and we
2375 * must not pass a NULL *s->childp here.
2377 * So whether new_bs was NULL or not, we cannot pass s->childp here; and in
2378 * any case, there is no reason to pass it anyway.
2380 bdrv_replace_child_noperm(&s->child, s->old_bs, true);
2382 * The child was pre-existing, so s->old_bs must be non-NULL, and
2383 * s->child thus must not have been freed
2385 assert(s->child != NULL);
2386 if (!new_bs) {
2387 /* As described above, *s->childp was cleared, so restore it */
2388 assert(s->childp != NULL);
2389 *s->childp = s->child;
2391 bdrv_unref(new_bs);
2394 static TransactionActionDrv bdrv_replace_child_drv = {
2395 .commit = bdrv_replace_child_commit,
2396 .abort = bdrv_replace_child_abort,
2397 .clean = g_free,
2401 * bdrv_replace_child_tran
2403 * Note: real unref of old_bs is done only on commit.
2405 * The function doesn't update permissions, caller is responsible for this.
2407 * (*childp)->bs must not be NULL.
2409 * Note that if new_bs == NULL, @childp is stored in a state object attached
2410 * to @tran, so that the old child can be reinstated in the abort handler.
2411 * Therefore, if @new_bs can be NULL, @childp must stay valid until the
2412 * transaction is committed or aborted.
2414 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2415 * freed (on commit). @free_empty_child should only be false if the
2416 * caller will free the BDrvChild themselves (which may be important
2417 * if this is in turn called in another transactional context).
2419 static void bdrv_replace_child_tran(BdrvChild **childp,
2420 BlockDriverState *new_bs,
2421 Transaction *tran,
2422 bool free_empty_child)
2424 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2425 *s = (BdrvReplaceChildState) {
2426 .child = *childp,
2427 .childp = new_bs == NULL ? childp : NULL,
2428 .old_bs = (*childp)->bs,
2429 .free_empty_child = free_empty_child,
2431 tran_add(tran, &bdrv_replace_child_drv, s);
2433 /* The abort handler relies on this */
2434 assert(s->old_bs != NULL);
2436 if (new_bs) {
2437 bdrv_ref(new_bs);
2440 * Pass free_empty_child=false, we will free the child (if
2441 * necessary) in bdrv_replace_child_commit() (if our
2442 * @free_empty_child parameter was true).
2444 bdrv_replace_child_noperm(childp, new_bs, false);
2445 /* old_bs reference is transparently moved from *childp to @s */
2449 * Refresh permissions in @bs subtree. The function is intended to be called
2450 * after some graph modification that was done without permission update.
2452 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2453 Transaction *tran, Error **errp)
2455 BlockDriver *drv = bs->drv;
2456 BdrvChild *c;
2457 int ret;
2458 uint64_t cumulative_perms, cumulative_shared_perms;
2459 GLOBAL_STATE_CODE();
2461 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2463 /* Write permissions never work with read-only images */
2464 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2465 !bdrv_is_writable_after_reopen(bs, q))
2467 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2468 error_setg(errp, "Block node is read-only");
2469 } else {
2470 error_setg(errp, "Read-only block node '%s' cannot support "
2471 "read-write users", bdrv_get_node_name(bs));
2474 return -EPERM;
2478 * Unaligned requests will automatically be aligned to bl.request_alignment
2479 * and without RESIZE we can't extend requests to write to space beyond the
2480 * end of the image, so it's required that the image size is aligned.
2482 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2483 !(cumulative_perms & BLK_PERM_RESIZE))
2485 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2486 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2487 "Image size is not a multiple of request "
2488 "alignment");
2489 return -EPERM;
2493 /* Check this node */
2494 if (!drv) {
2495 return 0;
2498 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2499 errp);
2500 if (ret < 0) {
2501 return ret;
2504 /* Drivers that never have children can omit .bdrv_child_perm() */
2505 if (!drv->bdrv_child_perm) {
2506 assert(QLIST_EMPTY(&bs->children));
2507 return 0;
2510 /* Check all children */
2511 QLIST_FOREACH(c, &bs->children, next) {
2512 uint64_t cur_perm, cur_shared;
2514 bdrv_child_perm(bs, c->bs, c, c->role, q,
2515 cumulative_perms, cumulative_shared_perms,
2516 &cur_perm, &cur_shared);
2517 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2520 return 0;
2523 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2524 Transaction *tran, Error **errp)
2526 int ret;
2527 BlockDriverState *bs;
2528 GLOBAL_STATE_CODE();
2530 for ( ; list; list = list->next) {
2531 bs = list->data;
2533 if (bdrv_parent_perms_conflict(bs, errp)) {
2534 return -EINVAL;
2537 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2538 if (ret < 0) {
2539 return ret;
2543 return 0;
2546 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2547 uint64_t *shared_perm)
2549 BdrvChild *c;
2550 uint64_t cumulative_perms = 0;
2551 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2553 GLOBAL_STATE_CODE();
2555 QLIST_FOREACH(c, &bs->parents, next_parent) {
2556 cumulative_perms |= c->perm;
2557 cumulative_shared_perms &= c->shared_perm;
2560 *perm = cumulative_perms;
2561 *shared_perm = cumulative_shared_perms;
2564 char *bdrv_perm_names(uint64_t perm)
2566 struct perm_name {
2567 uint64_t perm;
2568 const char *name;
2569 } permissions[] = {
2570 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2571 { BLK_PERM_WRITE, "write" },
2572 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2573 { BLK_PERM_RESIZE, "resize" },
2574 { 0, NULL }
2577 GString *result = g_string_sized_new(30);
2578 struct perm_name *p;
2580 for (p = permissions; p->name; p++) {
2581 if (perm & p->perm) {
2582 if (result->len > 0) {
2583 g_string_append(result, ", ");
2585 g_string_append(result, p->name);
2589 return g_string_free(result, FALSE);
2593 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2595 int ret;
2596 Transaction *tran = tran_new();
2597 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2598 GLOBAL_STATE_CODE();
2600 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2601 tran_finalize(tran, ret);
2603 return ret;
2606 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2607 Error **errp)
2609 Error *local_err = NULL;
2610 Transaction *tran = tran_new();
2611 int ret;
2613 GLOBAL_STATE_CODE();
2615 bdrv_child_set_perm(c, perm, shared, tran);
2617 ret = bdrv_refresh_perms(c->bs, &local_err);
2619 tran_finalize(tran, ret);
2621 if (ret < 0) {
2622 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2623 /* tighten permissions */
2624 error_propagate(errp, local_err);
2625 } else {
2627 * Our caller may intend to only loosen restrictions and
2628 * does not expect this function to fail. Errors are not
2629 * fatal in such a case, so we can just hide them from our
2630 * caller.
2632 error_free(local_err);
2633 ret = 0;
2637 return ret;
2640 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2642 uint64_t parent_perms, parent_shared;
2643 uint64_t perms, shared;
2645 GLOBAL_STATE_CODE();
2647 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2648 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2649 parent_perms, parent_shared, &perms, &shared);
2651 return bdrv_child_try_set_perm(c, perms, shared, errp);
2655 * Default implementation for .bdrv_child_perm() for block filters:
2656 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2657 * filtered child.
2659 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2660 BdrvChildRole role,
2661 BlockReopenQueue *reopen_queue,
2662 uint64_t perm, uint64_t shared,
2663 uint64_t *nperm, uint64_t *nshared)
2665 GLOBAL_STATE_CODE();
2666 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2667 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2670 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2671 BdrvChildRole role,
2672 BlockReopenQueue *reopen_queue,
2673 uint64_t perm, uint64_t shared,
2674 uint64_t *nperm, uint64_t *nshared)
2676 assert(role & BDRV_CHILD_COW);
2677 GLOBAL_STATE_CODE();
2680 * We want consistent read from backing files if the parent needs it.
2681 * No other operations are performed on backing files.
2683 perm &= BLK_PERM_CONSISTENT_READ;
2686 * If the parent can deal with changing data, we're okay with a
2687 * writable and resizable backing file.
2688 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2690 if (shared & BLK_PERM_WRITE) {
2691 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2692 } else {
2693 shared = 0;
2696 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2698 if (bs->open_flags & BDRV_O_INACTIVE) {
2699 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2702 *nperm = perm;
2703 *nshared = shared;
2706 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2707 BdrvChildRole role,
2708 BlockReopenQueue *reopen_queue,
2709 uint64_t perm, uint64_t shared,
2710 uint64_t *nperm, uint64_t *nshared)
2712 int flags;
2714 GLOBAL_STATE_CODE();
2715 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2717 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2720 * Apart from the modifications below, the same permissions are
2721 * forwarded and left alone as for filters
2723 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2724 perm, shared, &perm, &shared);
2726 if (role & BDRV_CHILD_METADATA) {
2727 /* Format drivers may touch metadata even if the guest doesn't write */
2728 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2729 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2733 * bs->file always needs to be consistent because of the
2734 * metadata. We can never allow other users to resize or write
2735 * to it.
2737 if (!(flags & BDRV_O_NO_IO)) {
2738 perm |= BLK_PERM_CONSISTENT_READ;
2740 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2743 if (role & BDRV_CHILD_DATA) {
2745 * Technically, everything in this block is a subset of the
2746 * BDRV_CHILD_METADATA path taken above, and so this could
2747 * be an "else if" branch. However, that is not obvious, and
2748 * this function is not performance critical, therefore we let
2749 * this be an independent "if".
2753 * We cannot allow other users to resize the file because the
2754 * format driver might have some assumptions about the size
2755 * (e.g. because it is stored in metadata, or because the file
2756 * is split into fixed-size data files).
2758 shared &= ~BLK_PERM_RESIZE;
2761 * WRITE_UNCHANGED often cannot be performed as such on the
2762 * data file. For example, the qcow2 driver may still need to
2763 * write copied clusters on copy-on-read.
2765 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2766 perm |= BLK_PERM_WRITE;
2770 * If the data file is written to, the format driver may
2771 * expect to be able to resize it by writing beyond the EOF.
2773 if (perm & BLK_PERM_WRITE) {
2774 perm |= BLK_PERM_RESIZE;
2778 if (bs->open_flags & BDRV_O_INACTIVE) {
2779 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2782 *nperm = perm;
2783 *nshared = shared;
2786 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2787 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2788 uint64_t perm, uint64_t shared,
2789 uint64_t *nperm, uint64_t *nshared)
2791 GLOBAL_STATE_CODE();
2792 if (role & BDRV_CHILD_FILTERED) {
2793 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2794 BDRV_CHILD_COW)));
2795 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2796 perm, shared, nperm, nshared);
2797 } else if (role & BDRV_CHILD_COW) {
2798 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2799 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2800 perm, shared, nperm, nshared);
2801 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2802 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2803 perm, shared, nperm, nshared);
2804 } else {
2805 g_assert_not_reached();
2809 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2811 static const uint64_t permissions[] = {
2812 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2813 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2814 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2815 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2818 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2819 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2821 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2823 return permissions[qapi_perm];
2827 * Replace (*childp)->bs by @new_bs.
2829 * If @new_bs is NULL, *childp will be set to NULL, too: BDS parents
2830 * generally cannot handle a BdrvChild with .bs == NULL, so clearing
2831 * BdrvChild.bs should generally immediately be followed by the
2832 * BdrvChild pointer being cleared as well.
2834 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2835 * freed. @free_empty_child should only be false if the caller will
2836 * free the BdrvChild themselves (this may be important in a
2837 * transactional context, where it may only be freed on commit).
2839 static void bdrv_replace_child_noperm(BdrvChild **childp,
2840 BlockDriverState *new_bs,
2841 bool free_empty_child)
2843 BdrvChild *child = *childp;
2844 BlockDriverState *old_bs = child->bs;
2845 int new_bs_quiesce_counter;
2846 int drain_saldo;
2848 assert(!child->frozen);
2849 assert(old_bs != new_bs);
2850 GLOBAL_STATE_CODE();
2852 if (old_bs && new_bs) {
2853 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2856 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2857 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2860 * If the new child node is drained but the old one was not, flush
2861 * all outstanding requests to the old child node.
2863 while (drain_saldo > 0 && child->klass->drained_begin) {
2864 bdrv_parent_drained_begin_single(child, true);
2865 drain_saldo--;
2868 if (old_bs) {
2869 /* Detach first so that the recursive drain sections coming from @child
2870 * are already gone and we only end the drain sections that came from
2871 * elsewhere. */
2872 if (child->klass->detach) {
2873 child->klass->detach(child);
2875 assert_bdrv_graph_writable(old_bs);
2876 QLIST_REMOVE(child, next_parent);
2879 child->bs = new_bs;
2880 if (!new_bs) {
2881 *childp = NULL;
2884 if (new_bs) {
2885 assert_bdrv_graph_writable(new_bs);
2886 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2889 * Detaching the old node may have led to the new node's
2890 * quiesce_counter having been decreased. Not a problem, we
2891 * just need to recognize this here and then invoke
2892 * drained_end appropriately more often.
2894 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2895 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2897 /* Attach only after starting new drained sections, so that recursive
2898 * drain sections coming from @child don't get an extra .drained_begin
2899 * callback. */
2900 if (child->klass->attach) {
2901 child->klass->attach(child);
2906 * If the old child node was drained but the new one is not, allow
2907 * requests to come in only after the new node has been attached.
2909 while (drain_saldo < 0 && child->klass->drained_end) {
2910 bdrv_parent_drained_end_single(child);
2911 drain_saldo++;
2914 if (free_empty_child && !child->bs) {
2915 bdrv_child_free(child);
2920 * Free the given @child.
2922 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2923 * unused (i.e. not in a children list).
2925 static void bdrv_child_free(BdrvChild *child)
2927 assert(!child->bs);
2928 GLOBAL_STATE_CODE();
2929 assert(!child->next.le_prev); /* not in children list */
2931 g_free(child->name);
2932 g_free(child);
2935 typedef struct BdrvAttachChildCommonState {
2936 BdrvChild **child;
2937 AioContext *old_parent_ctx;
2938 AioContext *old_child_ctx;
2939 } BdrvAttachChildCommonState;
2941 static void bdrv_attach_child_common_abort(void *opaque)
2943 BdrvAttachChildCommonState *s = opaque;
2944 BdrvChild *child = *s->child;
2945 BlockDriverState *bs = child->bs;
2947 GLOBAL_STATE_CODE();
2949 * Pass free_empty_child=false, because we still need the child
2950 * for the AioContext operations on the parent below; those
2951 * BdrvChildClass methods all work on a BdrvChild object, so we
2952 * need to keep it as an empty shell (after this function, it will
2953 * not be attached to any parent, and it will not have a .bs).
2955 bdrv_replace_child_noperm(s->child, NULL, false);
2957 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2958 bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort);
2961 if (bdrv_child_get_parent_aio_context(child) != s->old_parent_ctx) {
2962 GSList *ignore;
2964 /* No need to ignore `child`, because it has been detached already */
2965 ignore = NULL;
2966 child->klass->can_set_aio_ctx(child, s->old_parent_ctx, &ignore,
2967 &error_abort);
2968 g_slist_free(ignore);
2970 ignore = NULL;
2971 child->klass->set_aio_ctx(child, s->old_parent_ctx, &ignore);
2972 g_slist_free(ignore);
2975 bdrv_unref(bs);
2976 bdrv_child_free(child);
2979 static TransactionActionDrv bdrv_attach_child_common_drv = {
2980 .abort = bdrv_attach_child_common_abort,
2981 .clean = g_free,
2985 * Common part of attaching bdrv child to bs or to blk or to job
2987 * Resulting new child is returned through @child.
2988 * At start *@child must be NULL.
2989 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2990 * NULL on abort(). So referenced variable must live at least until transaction
2991 * end.
2993 * Function doesn't update permissions, caller is responsible for this.
2995 static int bdrv_attach_child_common(BlockDriverState *child_bs,
2996 const char *child_name,
2997 const BdrvChildClass *child_class,
2998 BdrvChildRole child_role,
2999 uint64_t perm, uint64_t shared_perm,
3000 void *opaque, BdrvChild **child,
3001 Transaction *tran, Error **errp)
3003 BdrvChild *new_child;
3004 AioContext *parent_ctx;
3005 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3007 assert(child);
3008 assert(*child == NULL);
3009 assert(child_class->get_parent_desc);
3010 GLOBAL_STATE_CODE();
3012 new_child = g_new(BdrvChild, 1);
3013 *new_child = (BdrvChild) {
3014 .bs = NULL,
3015 .name = g_strdup(child_name),
3016 .klass = child_class,
3017 .role = child_role,
3018 .perm = perm,
3019 .shared_perm = shared_perm,
3020 .opaque = opaque,
3024 * If the AioContexts don't match, first try to move the subtree of
3025 * child_bs into the AioContext of the new parent. If this doesn't work,
3026 * try moving the parent into the AioContext of child_bs instead.
3028 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3029 if (child_ctx != parent_ctx) {
3030 Error *local_err = NULL;
3031 int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err);
3033 if (ret < 0 && child_class->can_set_aio_ctx) {
3034 GSList *ignore = g_slist_prepend(NULL, new_child);
3035 if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore,
3036 NULL))
3038 error_free(local_err);
3039 ret = 0;
3040 g_slist_free(ignore);
3041 ignore = g_slist_prepend(NULL, new_child);
3042 child_class->set_aio_ctx(new_child, child_ctx, &ignore);
3044 g_slist_free(ignore);
3047 if (ret < 0) {
3048 error_propagate(errp, local_err);
3049 bdrv_child_free(new_child);
3050 return ret;
3054 bdrv_ref(child_bs);
3055 bdrv_replace_child_noperm(&new_child, child_bs, true);
3056 /* child_bs was non-NULL, so new_child must not have been freed */
3057 assert(new_child != NULL);
3059 *child = new_child;
3061 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3062 *s = (BdrvAttachChildCommonState) {
3063 .child = child,
3064 .old_parent_ctx = parent_ctx,
3065 .old_child_ctx = child_ctx,
3067 tran_add(tran, &bdrv_attach_child_common_drv, s);
3069 return 0;
3073 * Variable referenced by @child must live at least until transaction end.
3074 * (see bdrv_attach_child_common() doc for details)
3076 * Function doesn't update permissions, caller is responsible for this.
3078 static int bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3079 BlockDriverState *child_bs,
3080 const char *child_name,
3081 const BdrvChildClass *child_class,
3082 BdrvChildRole child_role,
3083 BdrvChild **child,
3084 Transaction *tran,
3085 Error **errp)
3087 int ret;
3088 uint64_t perm, shared_perm;
3090 assert(parent_bs->drv);
3091 GLOBAL_STATE_CODE();
3093 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3094 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3095 child_bs->node_name, child_name, parent_bs->node_name);
3096 return -EINVAL;
3099 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3100 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3101 perm, shared_perm, &perm, &shared_perm);
3103 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
3104 child_role, perm, shared_perm, parent_bs,
3105 child, tran, errp);
3106 if (ret < 0) {
3107 return ret;
3110 return 0;
3113 static void bdrv_detach_child(BdrvChild **childp)
3115 BlockDriverState *old_bs = (*childp)->bs;
3117 GLOBAL_STATE_CODE();
3118 bdrv_replace_child_noperm(childp, NULL, true);
3120 if (old_bs) {
3122 * Update permissions for old node. We're just taking a parent away, so
3123 * we're loosening restrictions. Errors of permission update are not
3124 * fatal in this case, ignore them.
3126 bdrv_refresh_perms(old_bs, NULL);
3129 * When the parent requiring a non-default AioContext is removed, the
3130 * node moves back to the main AioContext
3132 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
3137 * This function steals the reference to child_bs from the caller.
3138 * That reference is later dropped by bdrv_root_unref_child().
3140 * On failure NULL is returned, errp is set and the reference to
3141 * child_bs is also dropped.
3143 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3144 * (unless @child_bs is already in @ctx).
3146 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3147 const char *child_name,
3148 const BdrvChildClass *child_class,
3149 BdrvChildRole child_role,
3150 uint64_t perm, uint64_t shared_perm,
3151 void *opaque, Error **errp)
3153 int ret;
3154 BdrvChild *child = NULL;
3155 Transaction *tran = tran_new();
3157 GLOBAL_STATE_CODE();
3159 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
3160 child_role, perm, shared_perm, opaque,
3161 &child, tran, errp);
3162 if (ret < 0) {
3163 goto out;
3166 ret = bdrv_refresh_perms(child_bs, errp);
3168 out:
3169 tran_finalize(tran, ret);
3170 /* child is unset on failure by bdrv_attach_child_common_abort() */
3171 assert((ret < 0) == !child);
3173 bdrv_unref(child_bs);
3174 return child;
3178 * This function transfers the reference to child_bs from the caller
3179 * to parent_bs. That reference is later dropped by parent_bs on
3180 * bdrv_close() or if someone calls bdrv_unref_child().
3182 * On failure NULL is returned, errp is set and the reference to
3183 * child_bs is also dropped.
3185 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3186 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3188 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3189 BlockDriverState *child_bs,
3190 const char *child_name,
3191 const BdrvChildClass *child_class,
3192 BdrvChildRole child_role,
3193 Error **errp)
3195 int ret;
3196 BdrvChild *child = NULL;
3197 Transaction *tran = tran_new();
3199 GLOBAL_STATE_CODE();
3201 ret = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, child_class,
3202 child_role, &child, tran, errp);
3203 if (ret < 0) {
3204 goto out;
3207 ret = bdrv_refresh_perms(parent_bs, errp);
3208 if (ret < 0) {
3209 goto out;
3212 out:
3213 tran_finalize(tran, ret);
3214 /* child is unset on failure by bdrv_attach_child_common_abort() */
3215 assert((ret < 0) == !child);
3217 bdrv_unref(child_bs);
3219 return child;
3222 /* Callers must ensure that child->frozen is false. */
3223 void bdrv_root_unref_child(BdrvChild *child)
3225 BlockDriverState *child_bs;
3227 GLOBAL_STATE_CODE();
3229 child_bs = child->bs;
3230 bdrv_detach_child(&child);
3231 bdrv_unref(child_bs);
3234 typedef struct BdrvSetInheritsFrom {
3235 BlockDriverState *bs;
3236 BlockDriverState *old_inherits_from;
3237 } BdrvSetInheritsFrom;
3239 static void bdrv_set_inherits_from_abort(void *opaque)
3241 BdrvSetInheritsFrom *s = opaque;
3243 s->bs->inherits_from = s->old_inherits_from;
3246 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3247 .abort = bdrv_set_inherits_from_abort,
3248 .clean = g_free,
3251 /* @tran is allowed to be NULL. In this case no rollback is possible */
3252 static void bdrv_set_inherits_from(BlockDriverState *bs,
3253 BlockDriverState *new_inherits_from,
3254 Transaction *tran)
3256 if (tran) {
3257 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3259 *s = (BdrvSetInheritsFrom) {
3260 .bs = bs,
3261 .old_inherits_from = bs->inherits_from,
3264 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3267 bs->inherits_from = new_inherits_from;
3271 * Clear all inherits_from pointers from children and grandchildren of
3272 * @root that point to @root, where necessary.
3273 * @tran is allowed to be NULL. In this case no rollback is possible
3275 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3276 Transaction *tran)
3278 BdrvChild *c;
3280 if (child->bs->inherits_from == root) {
3282 * Remove inherits_from only when the last reference between root and
3283 * child->bs goes away.
3285 QLIST_FOREACH(c, &root->children, next) {
3286 if (c != child && c->bs == child->bs) {
3287 break;
3290 if (c == NULL) {
3291 bdrv_set_inherits_from(child->bs, NULL, tran);
3295 QLIST_FOREACH(c, &child->bs->children, next) {
3296 bdrv_unset_inherits_from(root, c, tran);
3300 /* Callers must ensure that child->frozen is false. */
3301 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3303 GLOBAL_STATE_CODE();
3304 if (child == NULL) {
3305 return;
3308 bdrv_unset_inherits_from(parent, child, NULL);
3309 bdrv_root_unref_child(child);
3313 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3315 BdrvChild *c;
3316 GLOBAL_STATE_CODE();
3317 QLIST_FOREACH(c, &bs->parents, next_parent) {
3318 if (c->klass->change_media) {
3319 c->klass->change_media(c, load);
3324 /* Return true if you can reach parent going through child->inherits_from
3325 * recursively. If parent or child are NULL, return false */
3326 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3327 BlockDriverState *parent)
3329 while (child && child != parent) {
3330 child = child->inherits_from;
3333 return child != NULL;
3337 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3338 * mostly used for COW backing children (role = COW), but also for
3339 * filtered children (role = FILTERED | PRIMARY).
3341 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3343 if (bs->drv && bs->drv->is_filter) {
3344 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3345 } else {
3346 return BDRV_CHILD_COW;
3351 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3352 * callers which don't need their own reference any more must call bdrv_unref().
3354 * Function doesn't update permissions, caller is responsible for this.
3356 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3357 BlockDriverState *child_bs,
3358 bool is_backing,
3359 Transaction *tran, Error **errp)
3361 int ret = 0;
3362 bool update_inherits_from =
3363 bdrv_inherits_from_recursive(child_bs, parent_bs);
3364 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3365 BdrvChildRole role;
3367 GLOBAL_STATE_CODE();
3369 if (!parent_bs->drv) {
3371 * Node without drv is an object without a class :/. TODO: finally fix
3372 * qcow2 driver to never clear bs->drv and implement format corruption
3373 * handling in other way.
3375 error_setg(errp, "Node corrupted");
3376 return -EINVAL;
3379 if (child && child->frozen) {
3380 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3381 child->name, parent_bs->node_name, child->bs->node_name);
3382 return -EPERM;
3385 if (is_backing && !parent_bs->drv->is_filter &&
3386 !parent_bs->drv->supports_backing)
3388 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3389 "files", parent_bs->drv->format_name, parent_bs->node_name);
3390 return -EINVAL;
3393 if (parent_bs->drv->is_filter) {
3394 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3395 } else if (is_backing) {
3396 role = BDRV_CHILD_COW;
3397 } else {
3399 * We only can use same role as it is in existing child. We don't have
3400 * infrastructure to determine role of file child in generic way
3402 if (!child) {
3403 error_setg(errp, "Cannot set file child to format node without "
3404 "file child");
3405 return -EINVAL;
3407 role = child->role;
3410 if (child) {
3411 bdrv_unset_inherits_from(parent_bs, child, tran);
3412 bdrv_remove_file_or_backing_child(parent_bs, child, tran);
3415 if (!child_bs) {
3416 goto out;
3419 ret = bdrv_attach_child_noperm(parent_bs, child_bs,
3420 is_backing ? "backing" : "file",
3421 &child_of_bds, role,
3422 is_backing ? &parent_bs->backing :
3423 &parent_bs->file,
3424 tran, errp);
3425 if (ret < 0) {
3426 return ret;
3431 * If inherits_from pointed recursively to bs then let's update it to
3432 * point directly to bs (else it will become NULL).
3434 if (update_inherits_from) {
3435 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3438 out:
3439 bdrv_refresh_limits(parent_bs, tran, NULL);
3441 return 0;
3444 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3445 BlockDriverState *backing_hd,
3446 Transaction *tran, Error **errp)
3448 GLOBAL_STATE_CODE();
3449 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3452 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3453 Error **errp)
3455 int ret;
3456 Transaction *tran = tran_new();
3458 GLOBAL_STATE_CODE();
3459 bdrv_drained_begin(bs);
3461 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3462 if (ret < 0) {
3463 goto out;
3466 ret = bdrv_refresh_perms(bs, errp);
3467 out:
3468 tran_finalize(tran, ret);
3470 bdrv_drained_end(bs);
3472 return ret;
3476 * Opens the backing file for a BlockDriverState if not yet open
3478 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3479 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3480 * itself, all options starting with "${bdref_key}." are considered part of the
3481 * BlockdevRef.
3483 * TODO Can this be unified with bdrv_open_image()?
3485 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3486 const char *bdref_key, Error **errp)
3488 char *backing_filename = NULL;
3489 char *bdref_key_dot;
3490 const char *reference = NULL;
3491 int ret = 0;
3492 bool implicit_backing = false;
3493 BlockDriverState *backing_hd;
3494 QDict *options;
3495 QDict *tmp_parent_options = NULL;
3496 Error *local_err = NULL;
3498 GLOBAL_STATE_CODE();
3500 if (bs->backing != NULL) {
3501 goto free_exit;
3504 /* NULL means an empty set of options */
3505 if (parent_options == NULL) {
3506 tmp_parent_options = qdict_new();
3507 parent_options = tmp_parent_options;
3510 bs->open_flags &= ~BDRV_O_NO_BACKING;
3512 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3513 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3514 g_free(bdref_key_dot);
3517 * Caution: while qdict_get_try_str() is fine, getting non-string
3518 * types would require more care. When @parent_options come from
3519 * -blockdev or blockdev_add, its members are typed according to
3520 * the QAPI schema, but when they come from -drive, they're all
3521 * QString.
3523 reference = qdict_get_try_str(parent_options, bdref_key);
3524 if (reference || qdict_haskey(options, "file.filename")) {
3525 /* keep backing_filename NULL */
3526 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3527 qobject_unref(options);
3528 goto free_exit;
3529 } else {
3530 if (qdict_size(options) == 0) {
3531 /* If the user specifies options that do not modify the
3532 * backing file's behavior, we might still consider it the
3533 * implicit backing file. But it's easier this way, and
3534 * just specifying some of the backing BDS's options is
3535 * only possible with -drive anyway (otherwise the QAPI
3536 * schema forces the user to specify everything). */
3537 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3540 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3541 if (local_err) {
3542 ret = -EINVAL;
3543 error_propagate(errp, local_err);
3544 qobject_unref(options);
3545 goto free_exit;
3549 if (!bs->drv || !bs->drv->supports_backing) {
3550 ret = -EINVAL;
3551 error_setg(errp, "Driver doesn't support backing files");
3552 qobject_unref(options);
3553 goto free_exit;
3556 if (!reference &&
3557 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3558 qdict_put_str(options, "driver", bs->backing_format);
3561 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3562 &child_of_bds, bdrv_backing_role(bs), errp);
3563 if (!backing_hd) {
3564 bs->open_flags |= BDRV_O_NO_BACKING;
3565 error_prepend(errp, "Could not open backing file: ");
3566 ret = -EINVAL;
3567 goto free_exit;
3570 if (implicit_backing) {
3571 bdrv_refresh_filename(backing_hd);
3572 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3573 backing_hd->filename);
3576 /* Hook up the backing file link; drop our reference, bs owns the
3577 * backing_hd reference now */
3578 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3579 bdrv_unref(backing_hd);
3580 if (ret < 0) {
3581 goto free_exit;
3584 qdict_del(parent_options, bdref_key);
3586 free_exit:
3587 g_free(backing_filename);
3588 qobject_unref(tmp_parent_options);
3589 return ret;
3592 static BlockDriverState *
3593 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3594 BlockDriverState *parent, const BdrvChildClass *child_class,
3595 BdrvChildRole child_role, bool allow_none, Error **errp)
3597 BlockDriverState *bs = NULL;
3598 QDict *image_options;
3599 char *bdref_key_dot;
3600 const char *reference;
3602 assert(child_class != NULL);
3604 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3605 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3606 g_free(bdref_key_dot);
3609 * Caution: while qdict_get_try_str() is fine, getting non-string
3610 * types would require more care. When @options come from
3611 * -blockdev or blockdev_add, its members are typed according to
3612 * the QAPI schema, but when they come from -drive, they're all
3613 * QString.
3615 reference = qdict_get_try_str(options, bdref_key);
3616 if (!filename && !reference && !qdict_size(image_options)) {
3617 if (!allow_none) {
3618 error_setg(errp, "A block device must be specified for \"%s\"",
3619 bdref_key);
3621 qobject_unref(image_options);
3622 goto done;
3625 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3626 parent, child_class, child_role, errp);
3627 if (!bs) {
3628 goto done;
3631 done:
3632 qdict_del(options, bdref_key);
3633 return bs;
3637 * Opens a disk image whose options are given as BlockdevRef in another block
3638 * device's options.
3640 * If allow_none is true, no image will be opened if filename is false and no
3641 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3643 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3644 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3645 * itself, all options starting with "${bdref_key}." are considered part of the
3646 * BlockdevRef.
3648 * The BlockdevRef will be removed from the options QDict.
3650 BdrvChild *bdrv_open_child(const char *filename,
3651 QDict *options, const char *bdref_key,
3652 BlockDriverState *parent,
3653 const BdrvChildClass *child_class,
3654 BdrvChildRole child_role,
3655 bool allow_none, Error **errp)
3657 BlockDriverState *bs;
3659 GLOBAL_STATE_CODE();
3661 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3662 child_role, allow_none, errp);
3663 if (bs == NULL) {
3664 return NULL;
3667 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3668 errp);
3672 * TODO Future callers may need to specify parent/child_class in order for
3673 * option inheritance to work. Existing callers use it for the root node.
3675 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3677 BlockDriverState *bs = NULL;
3678 QObject *obj = NULL;
3679 QDict *qdict = NULL;
3680 const char *reference = NULL;
3681 Visitor *v = NULL;
3683 GLOBAL_STATE_CODE();
3685 if (ref->type == QTYPE_QSTRING) {
3686 reference = ref->u.reference;
3687 } else {
3688 BlockdevOptions *options = &ref->u.definition;
3689 assert(ref->type == QTYPE_QDICT);
3691 v = qobject_output_visitor_new(&obj);
3692 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3693 visit_complete(v, &obj);
3695 qdict = qobject_to(QDict, obj);
3696 qdict_flatten(qdict);
3698 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3699 * compatibility with other callers) rather than what we want as the
3700 * real defaults. Apply the defaults here instead. */
3701 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3702 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3703 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3704 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3708 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3709 obj = NULL;
3710 qobject_unref(obj);
3711 visit_free(v);
3712 return bs;
3715 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3716 int flags,
3717 QDict *snapshot_options,
3718 Error **errp)
3720 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3721 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3722 int64_t total_size;
3723 QemuOpts *opts = NULL;
3724 BlockDriverState *bs_snapshot = NULL;
3725 int ret;
3727 GLOBAL_STATE_CODE();
3729 /* if snapshot, we create a temporary backing file and open it
3730 instead of opening 'filename' directly */
3732 /* Get the required size from the image */
3733 total_size = bdrv_getlength(bs);
3734 if (total_size < 0) {
3735 error_setg_errno(errp, -total_size, "Could not get image size");
3736 goto out;
3739 /* Create the temporary image */
3740 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3741 if (ret < 0) {
3742 error_setg_errno(errp, -ret, "Could not get temporary filename");
3743 goto out;
3746 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3747 &error_abort);
3748 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3749 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3750 qemu_opts_del(opts);
3751 if (ret < 0) {
3752 error_prepend(errp, "Could not create temporary overlay '%s': ",
3753 tmp_filename);
3754 goto out;
3757 /* Prepare options QDict for the temporary file */
3758 qdict_put_str(snapshot_options, "file.driver", "file");
3759 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3760 qdict_put_str(snapshot_options, "driver", "qcow2");
3762 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3763 snapshot_options = NULL;
3764 if (!bs_snapshot) {
3765 goto out;
3768 ret = bdrv_append(bs_snapshot, bs, errp);
3769 if (ret < 0) {
3770 bs_snapshot = NULL;
3771 goto out;
3774 out:
3775 qobject_unref(snapshot_options);
3776 g_free(tmp_filename);
3777 return bs_snapshot;
3781 * Opens a disk image (raw, qcow2, vmdk, ...)
3783 * options is a QDict of options to pass to the block drivers, or NULL for an
3784 * empty set of options. The reference to the QDict belongs to the block layer
3785 * after the call (even on failure), so if the caller intends to reuse the
3786 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3788 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3789 * If it is not NULL, the referenced BDS will be reused.
3791 * The reference parameter may be used to specify an existing block device which
3792 * should be opened. If specified, neither options nor a filename may be given,
3793 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3795 static BlockDriverState *bdrv_open_inherit(const char *filename,
3796 const char *reference,
3797 QDict *options, int flags,
3798 BlockDriverState *parent,
3799 const BdrvChildClass *child_class,
3800 BdrvChildRole child_role,
3801 Error **errp)
3803 int ret;
3804 BlockBackend *file = NULL;
3805 BlockDriverState *bs;
3806 BlockDriver *drv = NULL;
3807 BdrvChild *child;
3808 const char *drvname;
3809 const char *backing;
3810 Error *local_err = NULL;
3811 QDict *snapshot_options = NULL;
3812 int snapshot_flags = 0;
3814 assert(!child_class || !flags);
3815 assert(!child_class == !parent);
3816 GLOBAL_STATE_CODE();
3818 if (reference) {
3819 bool options_non_empty = options ? qdict_size(options) : false;
3820 qobject_unref(options);
3822 if (filename || options_non_empty) {
3823 error_setg(errp, "Cannot reference an existing block device with "
3824 "additional options or a new filename");
3825 return NULL;
3828 bs = bdrv_lookup_bs(reference, reference, errp);
3829 if (!bs) {
3830 return NULL;
3833 bdrv_ref(bs);
3834 return bs;
3837 bs = bdrv_new();
3839 /* NULL means an empty set of options */
3840 if (options == NULL) {
3841 options = qdict_new();
3844 /* json: syntax counts as explicit options, as if in the QDict */
3845 parse_json_protocol(options, &filename, &local_err);
3846 if (local_err) {
3847 goto fail;
3850 bs->explicit_options = qdict_clone_shallow(options);
3852 if (child_class) {
3853 bool parent_is_format;
3855 if (parent->drv) {
3856 parent_is_format = parent->drv->is_format;
3857 } else {
3859 * parent->drv is not set yet because this node is opened for
3860 * (potential) format probing. That means that @parent is going
3861 * to be a format node.
3863 parent_is_format = true;
3866 bs->inherits_from = parent;
3867 child_class->inherit_options(child_role, parent_is_format,
3868 &flags, options,
3869 parent->open_flags, parent->options);
3872 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3873 if (ret < 0) {
3874 goto fail;
3878 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3879 * Caution: getting a boolean member of @options requires care.
3880 * When @options come from -blockdev or blockdev_add, members are
3881 * typed according to the QAPI schema, but when they come from
3882 * -drive, they're all QString.
3884 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3885 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3886 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3887 } else {
3888 flags &= ~BDRV_O_RDWR;
3891 if (flags & BDRV_O_SNAPSHOT) {
3892 snapshot_options = qdict_new();
3893 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3894 flags, options);
3895 /* Let bdrv_backing_options() override "read-only" */
3896 qdict_del(options, BDRV_OPT_READ_ONLY);
3897 bdrv_inherited_options(BDRV_CHILD_COW, true,
3898 &flags, options, flags, options);
3901 bs->open_flags = flags;
3902 bs->options = options;
3903 options = qdict_clone_shallow(options);
3905 /* Find the right image format driver */
3906 /* See cautionary note on accessing @options above */
3907 drvname = qdict_get_try_str(options, "driver");
3908 if (drvname) {
3909 drv = bdrv_find_format(drvname);
3910 if (!drv) {
3911 error_setg(errp, "Unknown driver: '%s'", drvname);
3912 goto fail;
3916 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3918 /* See cautionary note on accessing @options above */
3919 backing = qdict_get_try_str(options, "backing");
3920 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3921 (backing && *backing == '\0'))
3923 if (backing) {
3924 warn_report("Use of \"backing\": \"\" is deprecated; "
3925 "use \"backing\": null instead");
3927 flags |= BDRV_O_NO_BACKING;
3928 qdict_del(bs->explicit_options, "backing");
3929 qdict_del(bs->options, "backing");
3930 qdict_del(options, "backing");
3933 /* Open image file without format layer. This BlockBackend is only used for
3934 * probing, the block drivers will do their own bdrv_open_child() for the
3935 * same BDS, which is why we put the node name back into options. */
3936 if ((flags & BDRV_O_PROTOCOL) == 0) {
3937 BlockDriverState *file_bs;
3939 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3940 &child_of_bds, BDRV_CHILD_IMAGE,
3941 true, &local_err);
3942 if (local_err) {
3943 goto fail;
3945 if (file_bs != NULL) {
3946 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3947 * looking at the header to guess the image format. This works even
3948 * in cases where a guest would not see a consistent state. */
3949 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3950 blk_insert_bs(file, file_bs, &local_err);
3951 bdrv_unref(file_bs);
3952 if (local_err) {
3953 goto fail;
3956 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3960 /* Image format probing */
3961 bs->probed = !drv;
3962 if (!drv && file) {
3963 ret = find_image_format(file, filename, &drv, &local_err);
3964 if (ret < 0) {
3965 goto fail;
3968 * This option update would logically belong in bdrv_fill_options(),
3969 * but we first need to open bs->file for the probing to work, while
3970 * opening bs->file already requires the (mostly) final set of options
3971 * so that cache mode etc. can be inherited.
3973 * Adding the driver later is somewhat ugly, but it's not an option
3974 * that would ever be inherited, so it's correct. We just need to make
3975 * sure to update both bs->options (which has the full effective
3976 * options for bs) and options (which has file.* already removed).
3978 qdict_put_str(bs->options, "driver", drv->format_name);
3979 qdict_put_str(options, "driver", drv->format_name);
3980 } else if (!drv) {
3981 error_setg(errp, "Must specify either driver or file");
3982 goto fail;
3985 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3986 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3987 /* file must be NULL if a protocol BDS is about to be created
3988 * (the inverse results in an error message from bdrv_open_common()) */
3989 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3991 /* Open the image */
3992 ret = bdrv_open_common(bs, file, options, &local_err);
3993 if (ret < 0) {
3994 goto fail;
3997 if (file) {
3998 blk_unref(file);
3999 file = NULL;
4002 /* If there is a backing file, use it */
4003 if ((flags & BDRV_O_NO_BACKING) == 0) {
4004 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4005 if (ret < 0) {
4006 goto close_and_fail;
4010 /* Remove all children options and references
4011 * from bs->options and bs->explicit_options */
4012 QLIST_FOREACH(child, &bs->children, next) {
4013 char *child_key_dot;
4014 child_key_dot = g_strdup_printf("%s.", child->name);
4015 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4016 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4017 qdict_del(bs->explicit_options, child->name);
4018 qdict_del(bs->options, child->name);
4019 g_free(child_key_dot);
4022 /* Check if any unknown options were used */
4023 if (qdict_size(options) != 0) {
4024 const QDictEntry *entry = qdict_first(options);
4025 if (flags & BDRV_O_PROTOCOL) {
4026 error_setg(errp, "Block protocol '%s' doesn't support the option "
4027 "'%s'", drv->format_name, entry->key);
4028 } else {
4029 error_setg(errp,
4030 "Block format '%s' does not support the option '%s'",
4031 drv->format_name, entry->key);
4034 goto close_and_fail;
4037 bdrv_parent_cb_change_media(bs, true);
4039 qobject_unref(options);
4040 options = NULL;
4042 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4043 * temporary snapshot afterwards. */
4044 if (snapshot_flags) {
4045 BlockDriverState *snapshot_bs;
4046 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4047 snapshot_options, &local_err);
4048 snapshot_options = NULL;
4049 if (local_err) {
4050 goto close_and_fail;
4052 /* We are not going to return bs but the overlay on top of it
4053 * (snapshot_bs); thus, we have to drop the strong reference to bs
4054 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4055 * though, because the overlay still has a reference to it. */
4056 bdrv_unref(bs);
4057 bs = snapshot_bs;
4060 return bs;
4062 fail:
4063 blk_unref(file);
4064 qobject_unref(snapshot_options);
4065 qobject_unref(bs->explicit_options);
4066 qobject_unref(bs->options);
4067 qobject_unref(options);
4068 bs->options = NULL;
4069 bs->explicit_options = NULL;
4070 bdrv_unref(bs);
4071 error_propagate(errp, local_err);
4072 return NULL;
4074 close_and_fail:
4075 bdrv_unref(bs);
4076 qobject_unref(snapshot_options);
4077 qobject_unref(options);
4078 error_propagate(errp, local_err);
4079 return NULL;
4082 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4083 QDict *options, int flags, Error **errp)
4085 GLOBAL_STATE_CODE();
4087 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4088 NULL, 0, errp);
4091 /* Return true if the NULL-terminated @list contains @str */
4092 static bool is_str_in_list(const char *str, const char *const *list)
4094 if (str && list) {
4095 int i;
4096 for (i = 0; list[i] != NULL; i++) {
4097 if (!strcmp(str, list[i])) {
4098 return true;
4102 return false;
4106 * Check that every option set in @bs->options is also set in
4107 * @new_opts.
4109 * Options listed in the common_options list and in
4110 * @bs->drv->mutable_opts are skipped.
4112 * Return 0 on success, otherwise return -EINVAL and set @errp.
4114 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4115 const QDict *new_opts, Error **errp)
4117 const QDictEntry *e;
4118 /* These options are common to all block drivers and are handled
4119 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4120 const char *const common_options[] = {
4121 "node-name", "discard", "cache.direct", "cache.no-flush",
4122 "read-only", "auto-read-only", "detect-zeroes", NULL
4125 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4126 if (!qdict_haskey(new_opts, e->key) &&
4127 !is_str_in_list(e->key, common_options) &&
4128 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4129 error_setg(errp, "Option '%s' cannot be reset "
4130 "to its default value", e->key);
4131 return -EINVAL;
4135 return 0;
4139 * Returns true if @child can be reached recursively from @bs
4141 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4142 BlockDriverState *child)
4144 BdrvChild *c;
4146 if (bs == child) {
4147 return true;
4150 QLIST_FOREACH(c, &bs->children, next) {
4151 if (bdrv_recurse_has_child(c->bs, child)) {
4152 return true;
4156 return false;
4160 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4161 * reopen of multiple devices.
4163 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4164 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4165 * be created and initialized. This newly created BlockReopenQueue should be
4166 * passed back in for subsequent calls that are intended to be of the same
4167 * atomic 'set'.
4169 * bs is the BlockDriverState to add to the reopen queue.
4171 * options contains the changed options for the associated bs
4172 * (the BlockReopenQueue takes ownership)
4174 * flags contains the open flags for the associated bs
4176 * returns a pointer to bs_queue, which is either the newly allocated
4177 * bs_queue, or the existing bs_queue being used.
4179 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4181 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4182 BlockDriverState *bs,
4183 QDict *options,
4184 const BdrvChildClass *klass,
4185 BdrvChildRole role,
4186 bool parent_is_format,
4187 QDict *parent_options,
4188 int parent_flags,
4189 bool keep_old_opts)
4191 assert(bs != NULL);
4193 BlockReopenQueueEntry *bs_entry;
4194 BdrvChild *child;
4195 QDict *old_options, *explicit_options, *options_copy;
4196 int flags;
4197 QemuOpts *opts;
4199 /* Make sure that the caller remembered to use a drained section. This is
4200 * important to avoid graph changes between the recursive queuing here and
4201 * bdrv_reopen_multiple(). */
4202 assert(bs->quiesce_counter > 0);
4203 GLOBAL_STATE_CODE();
4205 if (bs_queue == NULL) {
4206 bs_queue = g_new0(BlockReopenQueue, 1);
4207 QTAILQ_INIT(bs_queue);
4210 if (!options) {
4211 options = qdict_new();
4214 /* Check if this BlockDriverState is already in the queue */
4215 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4216 if (bs == bs_entry->state.bs) {
4217 break;
4222 * Precedence of options:
4223 * 1. Explicitly passed in options (highest)
4224 * 2. Retained from explicitly set options of bs
4225 * 3. Inherited from parent node
4226 * 4. Retained from effective options of bs
4229 /* Old explicitly set values (don't overwrite by inherited value) */
4230 if (bs_entry || keep_old_opts) {
4231 old_options = qdict_clone_shallow(bs_entry ?
4232 bs_entry->state.explicit_options :
4233 bs->explicit_options);
4234 bdrv_join_options(bs, options, old_options);
4235 qobject_unref(old_options);
4238 explicit_options = qdict_clone_shallow(options);
4240 /* Inherit from parent node */
4241 if (parent_options) {
4242 flags = 0;
4243 klass->inherit_options(role, parent_is_format, &flags, options,
4244 parent_flags, parent_options);
4245 } else {
4246 flags = bdrv_get_flags(bs);
4249 if (keep_old_opts) {
4250 /* Old values are used for options that aren't set yet */
4251 old_options = qdict_clone_shallow(bs->options);
4252 bdrv_join_options(bs, options, old_options);
4253 qobject_unref(old_options);
4256 /* We have the final set of options so let's update the flags */
4257 options_copy = qdict_clone_shallow(options);
4258 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4259 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4260 update_flags_from_options(&flags, opts);
4261 qemu_opts_del(opts);
4262 qobject_unref(options_copy);
4264 /* bdrv_open_inherit() sets and clears some additional flags internally */
4265 flags &= ~BDRV_O_PROTOCOL;
4266 if (flags & BDRV_O_RDWR) {
4267 flags |= BDRV_O_ALLOW_RDWR;
4270 if (!bs_entry) {
4271 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4272 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4273 } else {
4274 qobject_unref(bs_entry->state.options);
4275 qobject_unref(bs_entry->state.explicit_options);
4278 bs_entry->state.bs = bs;
4279 bs_entry->state.options = options;
4280 bs_entry->state.explicit_options = explicit_options;
4281 bs_entry->state.flags = flags;
4284 * If keep_old_opts is false then it means that unspecified
4285 * options must be reset to their original value. We don't allow
4286 * resetting 'backing' but we need to know if the option is
4287 * missing in order to decide if we have to return an error.
4289 if (!keep_old_opts) {
4290 bs_entry->state.backing_missing =
4291 !qdict_haskey(options, "backing") &&
4292 !qdict_haskey(options, "backing.driver");
4295 QLIST_FOREACH(child, &bs->children, next) {
4296 QDict *new_child_options = NULL;
4297 bool child_keep_old = keep_old_opts;
4299 /* reopen can only change the options of block devices that were
4300 * implicitly created and inherited options. For other (referenced)
4301 * block devices, a syntax like "backing.foo" results in an error. */
4302 if (child->bs->inherits_from != bs) {
4303 continue;
4306 /* Check if the options contain a child reference */
4307 if (qdict_haskey(options, child->name)) {
4308 const char *childref = qdict_get_try_str(options, child->name);
4310 * The current child must not be reopened if the child
4311 * reference is null or points to a different node.
4313 if (g_strcmp0(childref, child->bs->node_name)) {
4314 continue;
4317 * If the child reference points to the current child then
4318 * reopen it with its existing set of options (note that
4319 * it can still inherit new options from the parent).
4321 child_keep_old = true;
4322 } else {
4323 /* Extract child options ("child-name.*") */
4324 char *child_key_dot = g_strdup_printf("%s.", child->name);
4325 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4326 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4327 g_free(child_key_dot);
4330 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4331 child->klass, child->role, bs->drv->is_format,
4332 options, flags, child_keep_old);
4335 return bs_queue;
4338 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4339 BlockDriverState *bs,
4340 QDict *options, bool keep_old_opts)
4342 GLOBAL_STATE_CODE();
4344 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4345 NULL, 0, keep_old_opts);
4348 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4350 GLOBAL_STATE_CODE();
4351 if (bs_queue) {
4352 BlockReopenQueueEntry *bs_entry, *next;
4353 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4354 qobject_unref(bs_entry->state.explicit_options);
4355 qobject_unref(bs_entry->state.options);
4356 g_free(bs_entry);
4358 g_free(bs_queue);
4363 * Reopen multiple BlockDriverStates atomically & transactionally.
4365 * The queue passed in (bs_queue) must have been built up previous
4366 * via bdrv_reopen_queue().
4368 * Reopens all BDS specified in the queue, with the appropriate
4369 * flags. All devices are prepared for reopen, and failure of any
4370 * device will cause all device changes to be abandoned, and intermediate
4371 * data cleaned up.
4373 * If all devices prepare successfully, then the changes are committed
4374 * to all devices.
4376 * All affected nodes must be drained between bdrv_reopen_queue() and
4377 * bdrv_reopen_multiple().
4379 * To be called from the main thread, with all other AioContexts unlocked.
4381 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4383 int ret = -1;
4384 BlockReopenQueueEntry *bs_entry, *next;
4385 AioContext *ctx;
4386 Transaction *tran = tran_new();
4387 g_autoptr(GHashTable) found = NULL;
4388 g_autoptr(GSList) refresh_list = NULL;
4390 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4391 assert(bs_queue != NULL);
4392 GLOBAL_STATE_CODE();
4394 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4395 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4396 aio_context_acquire(ctx);
4397 ret = bdrv_flush(bs_entry->state.bs);
4398 aio_context_release(ctx);
4399 if (ret < 0) {
4400 error_setg_errno(errp, -ret, "Error flushing drive");
4401 goto abort;
4405 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4406 assert(bs_entry->state.bs->quiesce_counter > 0);
4407 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4408 aio_context_acquire(ctx);
4409 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4410 aio_context_release(ctx);
4411 if (ret < 0) {
4412 goto abort;
4414 bs_entry->prepared = true;
4417 found = g_hash_table_new(NULL, NULL);
4418 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4419 BDRVReopenState *state = &bs_entry->state;
4421 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4422 if (state->old_backing_bs) {
4423 refresh_list = bdrv_topological_dfs(refresh_list, found,
4424 state->old_backing_bs);
4426 if (state->old_file_bs) {
4427 refresh_list = bdrv_topological_dfs(refresh_list, found,
4428 state->old_file_bs);
4433 * Note that file-posix driver rely on permission update done during reopen
4434 * (even if no permission changed), because it wants "new" permissions for
4435 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4436 * in raw_reopen_prepare() which is called with "old" permissions.
4438 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4439 if (ret < 0) {
4440 goto abort;
4444 * If we reach this point, we have success and just need to apply the
4445 * changes.
4447 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4448 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4449 * children are usually goes after parents in reopen-queue, so go from last
4450 * to first element.
4452 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4453 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4454 aio_context_acquire(ctx);
4455 bdrv_reopen_commit(&bs_entry->state);
4456 aio_context_release(ctx);
4459 tran_commit(tran);
4461 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4462 BlockDriverState *bs = bs_entry->state.bs;
4464 if (bs->drv->bdrv_reopen_commit_post) {
4465 ctx = bdrv_get_aio_context(bs);
4466 aio_context_acquire(ctx);
4467 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4468 aio_context_release(ctx);
4472 ret = 0;
4473 goto cleanup;
4475 abort:
4476 tran_abort(tran);
4477 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4478 if (bs_entry->prepared) {
4479 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4480 aio_context_acquire(ctx);
4481 bdrv_reopen_abort(&bs_entry->state);
4482 aio_context_release(ctx);
4486 cleanup:
4487 bdrv_reopen_queue_free(bs_queue);
4489 return ret;
4492 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4493 Error **errp)
4495 AioContext *ctx = bdrv_get_aio_context(bs);
4496 BlockReopenQueue *queue;
4497 int ret;
4499 GLOBAL_STATE_CODE();
4501 bdrv_subtree_drained_begin(bs);
4502 if (ctx != qemu_get_aio_context()) {
4503 aio_context_release(ctx);
4506 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4507 ret = bdrv_reopen_multiple(queue, errp);
4509 if (ctx != qemu_get_aio_context()) {
4510 aio_context_acquire(ctx);
4512 bdrv_subtree_drained_end(bs);
4514 return ret;
4517 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4518 Error **errp)
4520 QDict *opts = qdict_new();
4522 GLOBAL_STATE_CODE();
4524 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4526 return bdrv_reopen(bs, opts, true, errp);
4530 * Take a BDRVReopenState and check if the value of 'backing' in the
4531 * reopen_state->options QDict is valid or not.
4533 * If 'backing' is missing from the QDict then return 0.
4535 * If 'backing' contains the node name of the backing file of
4536 * reopen_state->bs then return 0.
4538 * If 'backing' contains a different node name (or is null) then check
4539 * whether the current backing file can be replaced with the new one.
4540 * If that's the case then reopen_state->replace_backing_bs is set to
4541 * true and reopen_state->new_backing_bs contains a pointer to the new
4542 * backing BlockDriverState (or NULL).
4544 * Return 0 on success, otherwise return < 0 and set @errp.
4546 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4547 bool is_backing, Transaction *tran,
4548 Error **errp)
4550 BlockDriverState *bs = reopen_state->bs;
4551 BlockDriverState *new_child_bs;
4552 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4553 child_bs(bs->file);
4554 const char *child_name = is_backing ? "backing" : "file";
4555 QObject *value;
4556 const char *str;
4558 GLOBAL_STATE_CODE();
4560 value = qdict_get(reopen_state->options, child_name);
4561 if (value == NULL) {
4562 return 0;
4565 switch (qobject_type(value)) {
4566 case QTYPE_QNULL:
4567 assert(is_backing); /* The 'file' option does not allow a null value */
4568 new_child_bs = NULL;
4569 break;
4570 case QTYPE_QSTRING:
4571 str = qstring_get_str(qobject_to(QString, value));
4572 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4573 if (new_child_bs == NULL) {
4574 return -EINVAL;
4575 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4576 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4577 "cycle", str, child_name, bs->node_name);
4578 return -EINVAL;
4580 break;
4581 default:
4583 * The options QDict has been flattened, so 'backing' and 'file'
4584 * do not allow any other data type here.
4586 g_assert_not_reached();
4589 if (old_child_bs == new_child_bs) {
4590 return 0;
4593 if (old_child_bs) {
4594 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4595 return 0;
4598 if (old_child_bs->implicit) {
4599 error_setg(errp, "Cannot replace implicit %s child of %s",
4600 child_name, bs->node_name);
4601 return -EPERM;
4605 if (bs->drv->is_filter && !old_child_bs) {
4607 * Filters always have a file or a backing child, so we are trying to
4608 * change wrong child
4610 error_setg(errp, "'%s' is a %s filter node that does not support a "
4611 "%s child", bs->node_name, bs->drv->format_name, child_name);
4612 return -EINVAL;
4615 if (is_backing) {
4616 reopen_state->old_backing_bs = old_child_bs;
4617 } else {
4618 reopen_state->old_file_bs = old_child_bs;
4621 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4622 tran, errp);
4626 * Prepares a BlockDriverState for reopen. All changes are staged in the
4627 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4628 * the block driver layer .bdrv_reopen_prepare()
4630 * bs is the BlockDriverState to reopen
4631 * flags are the new open flags
4632 * queue is the reopen queue
4634 * Returns 0 on success, non-zero on error. On error errp will be set
4635 * as well.
4637 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4638 * It is the responsibility of the caller to then call the abort() or
4639 * commit() for any other BDS that have been left in a prepare() state
4642 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4643 BlockReopenQueue *queue,
4644 Transaction *change_child_tran, Error **errp)
4646 int ret = -1;
4647 int old_flags;
4648 Error *local_err = NULL;
4649 BlockDriver *drv;
4650 QemuOpts *opts;
4651 QDict *orig_reopen_opts;
4652 char *discard = NULL;
4653 bool read_only;
4654 bool drv_prepared = false;
4656 assert(reopen_state != NULL);
4657 assert(reopen_state->bs->drv != NULL);
4658 GLOBAL_STATE_CODE();
4659 drv = reopen_state->bs->drv;
4661 /* This function and each driver's bdrv_reopen_prepare() remove
4662 * entries from reopen_state->options as they are processed, so
4663 * we need to make a copy of the original QDict. */
4664 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4666 /* Process generic block layer options */
4667 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4668 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4669 ret = -EINVAL;
4670 goto error;
4673 /* This was already called in bdrv_reopen_queue_child() so the flags
4674 * are up-to-date. This time we simply want to remove the options from
4675 * QemuOpts in order to indicate that they have been processed. */
4676 old_flags = reopen_state->flags;
4677 update_flags_from_options(&reopen_state->flags, opts);
4678 assert(old_flags == reopen_state->flags);
4680 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4681 if (discard != NULL) {
4682 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4683 error_setg(errp, "Invalid discard option");
4684 ret = -EINVAL;
4685 goto error;
4689 reopen_state->detect_zeroes =
4690 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4691 if (local_err) {
4692 error_propagate(errp, local_err);
4693 ret = -EINVAL;
4694 goto error;
4697 /* All other options (including node-name and driver) must be unchanged.
4698 * Put them back into the QDict, so that they are checked at the end
4699 * of this function. */
4700 qemu_opts_to_qdict(opts, reopen_state->options);
4702 /* If we are to stay read-only, do not allow permission change
4703 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4704 * not set, or if the BDS still has copy_on_read enabled */
4705 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4706 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4707 if (local_err) {
4708 error_propagate(errp, local_err);
4709 goto error;
4712 if (drv->bdrv_reopen_prepare) {
4714 * If a driver-specific option is missing, it means that we
4715 * should reset it to its default value.
4716 * But not all options allow that, so we need to check it first.
4718 ret = bdrv_reset_options_allowed(reopen_state->bs,
4719 reopen_state->options, errp);
4720 if (ret) {
4721 goto error;
4724 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4725 if (ret) {
4726 if (local_err != NULL) {
4727 error_propagate(errp, local_err);
4728 } else {
4729 bdrv_refresh_filename(reopen_state->bs);
4730 error_setg(errp, "failed while preparing to reopen image '%s'",
4731 reopen_state->bs->filename);
4733 goto error;
4735 } else {
4736 /* It is currently mandatory to have a bdrv_reopen_prepare()
4737 * handler for each supported drv. */
4738 error_setg(errp, "Block format '%s' used by node '%s' "
4739 "does not support reopening files", drv->format_name,
4740 bdrv_get_device_or_node_name(reopen_state->bs));
4741 ret = -1;
4742 goto error;
4745 drv_prepared = true;
4748 * We must provide the 'backing' option if the BDS has a backing
4749 * file or if the image file has a backing file name as part of
4750 * its metadata. Otherwise the 'backing' option can be omitted.
4752 if (drv->supports_backing && reopen_state->backing_missing &&
4753 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4754 error_setg(errp, "backing is missing for '%s'",
4755 reopen_state->bs->node_name);
4756 ret = -EINVAL;
4757 goto error;
4761 * Allow changing the 'backing' option. The new value can be
4762 * either a reference to an existing node (using its node name)
4763 * or NULL to simply detach the current backing file.
4765 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4766 change_child_tran, errp);
4767 if (ret < 0) {
4768 goto error;
4770 qdict_del(reopen_state->options, "backing");
4772 /* Allow changing the 'file' option. In this case NULL is not allowed */
4773 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4774 change_child_tran, errp);
4775 if (ret < 0) {
4776 goto error;
4778 qdict_del(reopen_state->options, "file");
4780 /* Options that are not handled are only okay if they are unchanged
4781 * compared to the old state. It is expected that some options are only
4782 * used for the initial open, but not reopen (e.g. filename) */
4783 if (qdict_size(reopen_state->options)) {
4784 const QDictEntry *entry = qdict_first(reopen_state->options);
4786 do {
4787 QObject *new = entry->value;
4788 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4790 /* Allow child references (child_name=node_name) as long as they
4791 * point to the current child (i.e. everything stays the same). */
4792 if (qobject_type(new) == QTYPE_QSTRING) {
4793 BdrvChild *child;
4794 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4795 if (!strcmp(child->name, entry->key)) {
4796 break;
4800 if (child) {
4801 if (!strcmp(child->bs->node_name,
4802 qstring_get_str(qobject_to(QString, new)))) {
4803 continue; /* Found child with this name, skip option */
4809 * TODO: When using -drive to specify blockdev options, all values
4810 * will be strings; however, when using -blockdev, blockdev-add or
4811 * filenames using the json:{} pseudo-protocol, they will be
4812 * correctly typed.
4813 * In contrast, reopening options are (currently) always strings
4814 * (because you can only specify them through qemu-io; all other
4815 * callers do not specify any options).
4816 * Therefore, when using anything other than -drive to create a BDS,
4817 * this cannot detect non-string options as unchanged, because
4818 * qobject_is_equal() always returns false for objects of different
4819 * type. In the future, this should be remedied by correctly typing
4820 * all options. For now, this is not too big of an issue because
4821 * the user can simply omit options which cannot be changed anyway,
4822 * so they will stay unchanged.
4824 if (!qobject_is_equal(new, old)) {
4825 error_setg(errp, "Cannot change the option '%s'", entry->key);
4826 ret = -EINVAL;
4827 goto error;
4829 } while ((entry = qdict_next(reopen_state->options, entry)));
4832 ret = 0;
4834 /* Restore the original reopen_state->options QDict */
4835 qobject_unref(reopen_state->options);
4836 reopen_state->options = qobject_ref(orig_reopen_opts);
4838 error:
4839 if (ret < 0 && drv_prepared) {
4840 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4841 * call drv->bdrv_reopen_abort() before signaling an error
4842 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4843 * when the respective bdrv_reopen_prepare() has failed) */
4844 if (drv->bdrv_reopen_abort) {
4845 drv->bdrv_reopen_abort(reopen_state);
4848 qemu_opts_del(opts);
4849 qobject_unref(orig_reopen_opts);
4850 g_free(discard);
4851 return ret;
4855 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4856 * makes them final by swapping the staging BlockDriverState contents into
4857 * the active BlockDriverState contents.
4859 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4861 BlockDriver *drv;
4862 BlockDriverState *bs;
4863 BdrvChild *child;
4865 assert(reopen_state != NULL);
4866 bs = reopen_state->bs;
4867 drv = bs->drv;
4868 assert(drv != NULL);
4869 GLOBAL_STATE_CODE();
4871 /* If there are any driver level actions to take */
4872 if (drv->bdrv_reopen_commit) {
4873 drv->bdrv_reopen_commit(reopen_state);
4876 /* set BDS specific flags now */
4877 qobject_unref(bs->explicit_options);
4878 qobject_unref(bs->options);
4879 qobject_ref(reopen_state->explicit_options);
4880 qobject_ref(reopen_state->options);
4882 bs->explicit_options = reopen_state->explicit_options;
4883 bs->options = reopen_state->options;
4884 bs->open_flags = reopen_state->flags;
4885 bs->detect_zeroes = reopen_state->detect_zeroes;
4887 /* Remove child references from bs->options and bs->explicit_options.
4888 * Child options were already removed in bdrv_reopen_queue_child() */
4889 QLIST_FOREACH(child, &bs->children, next) {
4890 qdict_del(bs->explicit_options, child->name);
4891 qdict_del(bs->options, child->name);
4893 /* backing is probably removed, so it's not handled by previous loop */
4894 qdict_del(bs->explicit_options, "backing");
4895 qdict_del(bs->options, "backing");
4897 bdrv_refresh_limits(bs, NULL, NULL);
4901 * Abort the reopen, and delete and free the staged changes in
4902 * reopen_state
4904 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4906 BlockDriver *drv;
4908 assert(reopen_state != NULL);
4909 drv = reopen_state->bs->drv;
4910 assert(drv != NULL);
4911 GLOBAL_STATE_CODE();
4913 if (drv->bdrv_reopen_abort) {
4914 drv->bdrv_reopen_abort(reopen_state);
4919 static void bdrv_close(BlockDriverState *bs)
4921 BdrvAioNotifier *ban, *ban_next;
4922 BdrvChild *child, *next;
4924 GLOBAL_STATE_CODE();
4925 assert(!bs->refcnt);
4927 bdrv_drained_begin(bs); /* complete I/O */
4928 bdrv_flush(bs);
4929 bdrv_drain(bs); /* in case flush left pending I/O */
4931 if (bs->drv) {
4932 if (bs->drv->bdrv_close) {
4933 /* Must unfreeze all children, so bdrv_unref_child() works */
4934 bs->drv->bdrv_close(bs);
4936 bs->drv = NULL;
4939 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4940 bdrv_unref_child(bs, child);
4943 bs->backing = NULL;
4944 bs->file = NULL;
4945 g_free(bs->opaque);
4946 bs->opaque = NULL;
4947 qatomic_set(&bs->copy_on_read, 0);
4948 bs->backing_file[0] = '\0';
4949 bs->backing_format[0] = '\0';
4950 bs->total_sectors = 0;
4951 bs->encrypted = false;
4952 bs->sg = false;
4953 qobject_unref(bs->options);
4954 qobject_unref(bs->explicit_options);
4955 bs->options = NULL;
4956 bs->explicit_options = NULL;
4957 qobject_unref(bs->full_open_options);
4958 bs->full_open_options = NULL;
4959 g_free(bs->block_status_cache);
4960 bs->block_status_cache = NULL;
4962 bdrv_release_named_dirty_bitmaps(bs);
4963 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4965 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4966 g_free(ban);
4968 QLIST_INIT(&bs->aio_notifiers);
4969 bdrv_drained_end(bs);
4972 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4973 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4974 * gets called.
4976 if (bs->quiesce_counter) {
4977 bdrv_drain_all_end_quiesce(bs);
4981 void bdrv_close_all(void)
4983 assert(job_next(NULL) == NULL);
4984 GLOBAL_STATE_CODE();
4986 /* Drop references from requests still in flight, such as canceled block
4987 * jobs whose AIO context has not been polled yet */
4988 bdrv_drain_all();
4990 blk_remove_all_bs();
4991 blockdev_close_all_bdrv_states();
4993 assert(QTAILQ_EMPTY(&all_bdrv_states));
4996 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4998 GQueue *queue;
4999 GHashTable *found;
5000 bool ret;
5002 if (c->klass->stay_at_node) {
5003 return false;
5006 /* If the child @c belongs to the BDS @to, replacing the current
5007 * c->bs by @to would mean to create a loop.
5009 * Such a case occurs when appending a BDS to a backing chain.
5010 * For instance, imagine the following chain:
5012 * guest device -> node A -> further backing chain...
5014 * Now we create a new BDS B which we want to put on top of this
5015 * chain, so we first attach A as its backing node:
5017 * node B
5020 * guest device -> node A -> further backing chain...
5022 * Finally we want to replace A by B. When doing that, we want to
5023 * replace all pointers to A by pointers to B -- except for the
5024 * pointer from B because (1) that would create a loop, and (2)
5025 * that pointer should simply stay intact:
5027 * guest device -> node B
5030 * node A -> further backing chain...
5032 * In general, when replacing a node A (c->bs) by a node B (@to),
5033 * if A is a child of B, that means we cannot replace A by B there
5034 * because that would create a loop. Silently detaching A from B
5035 * is also not really an option. So overall just leaving A in
5036 * place there is the most sensible choice.
5038 * We would also create a loop in any cases where @c is only
5039 * indirectly referenced by @to. Prevent this by returning false
5040 * if @c is found (by breadth-first search) anywhere in the whole
5041 * subtree of @to.
5044 ret = true;
5045 found = g_hash_table_new(NULL, NULL);
5046 g_hash_table_add(found, to);
5047 queue = g_queue_new();
5048 g_queue_push_tail(queue, to);
5050 while (!g_queue_is_empty(queue)) {
5051 BlockDriverState *v = g_queue_pop_head(queue);
5052 BdrvChild *c2;
5054 QLIST_FOREACH(c2, &v->children, next) {
5055 if (c2 == c) {
5056 ret = false;
5057 break;
5060 if (g_hash_table_contains(found, c2->bs)) {
5061 continue;
5064 g_queue_push_tail(queue, c2->bs);
5065 g_hash_table_add(found, c2->bs);
5069 g_queue_free(queue);
5070 g_hash_table_destroy(found);
5072 return ret;
5075 typedef struct BdrvRemoveFilterOrCowChild {
5076 BdrvChild *child;
5077 BlockDriverState *bs;
5078 bool is_backing;
5079 } BdrvRemoveFilterOrCowChild;
5081 static void bdrv_remove_filter_or_cow_child_abort(void *opaque)
5083 BdrvRemoveFilterOrCowChild *s = opaque;
5084 BlockDriverState *parent_bs = s->child->opaque;
5086 if (s->is_backing) {
5087 parent_bs->backing = s->child;
5088 } else {
5089 parent_bs->file = s->child;
5093 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
5094 * because that function is transactionable and it registered own completion
5095 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
5096 * called automatically.
5100 static void bdrv_remove_filter_or_cow_child_commit(void *opaque)
5102 BdrvRemoveFilterOrCowChild *s = opaque;
5103 GLOBAL_STATE_CODE();
5104 bdrv_child_free(s->child);
5107 static void bdrv_remove_filter_or_cow_child_clean(void *opaque)
5109 BdrvRemoveFilterOrCowChild *s = opaque;
5111 /* Drop the bs reference after the transaction is done */
5112 bdrv_unref(s->bs);
5113 g_free(s);
5116 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = {
5117 .abort = bdrv_remove_filter_or_cow_child_abort,
5118 .commit = bdrv_remove_filter_or_cow_child_commit,
5119 .clean = bdrv_remove_filter_or_cow_child_clean,
5123 * A function to remove backing or file child of @bs.
5124 * Function doesn't update permissions, caller is responsible for this.
5126 static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
5127 BdrvChild *child,
5128 Transaction *tran)
5130 BdrvChild **childp;
5131 BdrvRemoveFilterOrCowChild *s;
5133 if (!child) {
5134 return;
5138 * Keep a reference to @bs so @childp will stay valid throughout the
5139 * transaction (required by bdrv_replace_child_tran())
5141 bdrv_ref(bs);
5142 if (child == bs->backing) {
5143 childp = &bs->backing;
5144 } else if (child == bs->file) {
5145 childp = &bs->file;
5146 } else {
5147 g_assert_not_reached();
5150 if (child->bs) {
5152 * Pass free_empty_child=false, we will free the child in
5153 * bdrv_remove_filter_or_cow_child_commit()
5155 bdrv_replace_child_tran(childp, NULL, tran, false);
5158 s = g_new(BdrvRemoveFilterOrCowChild, 1);
5159 *s = (BdrvRemoveFilterOrCowChild) {
5160 .child = child,
5161 .bs = bs,
5162 .is_backing = (childp == &bs->backing),
5164 tran_add(tran, &bdrv_remove_filter_or_cow_child_drv, s);
5168 * A function to remove backing-chain child of @bs if exists: cow child for
5169 * format nodes (always .backing) and filter child for filters (may be .file or
5170 * .backing)
5172 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
5173 Transaction *tran)
5175 bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran);
5178 static int bdrv_replace_node_noperm(BlockDriverState *from,
5179 BlockDriverState *to,
5180 bool auto_skip, Transaction *tran,
5181 Error **errp)
5183 BdrvChild *c, *next;
5185 assert(to != NULL);
5186 GLOBAL_STATE_CODE();
5188 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5189 assert(c->bs == from);
5190 if (!should_update_child(c, to)) {
5191 if (auto_skip) {
5192 continue;
5194 error_setg(errp, "Should not change '%s' link to '%s'",
5195 c->name, from->node_name);
5196 return -EINVAL;
5198 if (c->frozen) {
5199 error_setg(errp, "Cannot change '%s' link to '%s'",
5200 c->name, from->node_name);
5201 return -EPERM;
5205 * Passing a pointer to the local variable @c is fine here, because
5206 * @to is not NULL, and so &c will not be attached to the transaction.
5208 bdrv_replace_child_tran(&c, to, tran, true);
5211 return 0;
5215 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5216 * if it creates a parent-child relation loop or if parent is block-job.
5218 * With auto_skip=false the error is returned if from has a parent which should
5219 * not be updated.
5221 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5222 * case backing link of the cow-parent of @to is removed.
5224 * @to must not be NULL.
5226 static int bdrv_replace_node_common(BlockDriverState *from,
5227 BlockDriverState *to,
5228 bool auto_skip, bool detach_subchain,
5229 Error **errp)
5231 Transaction *tran = tran_new();
5232 g_autoptr(GHashTable) found = NULL;
5233 g_autoptr(GSList) refresh_list = NULL;
5234 BlockDriverState *to_cow_parent = NULL;
5235 int ret;
5237 GLOBAL_STATE_CODE();
5238 assert(to != NULL);
5240 if (detach_subchain) {
5241 assert(bdrv_chain_contains(from, to));
5242 assert(from != to);
5243 for (to_cow_parent = from;
5244 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5245 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5251 /* Make sure that @from doesn't go away until we have successfully attached
5252 * all of its parents to @to. */
5253 bdrv_ref(from);
5255 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5256 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5257 bdrv_drained_begin(from);
5260 * Do the replacement without permission update.
5261 * Replacement may influence the permissions, we should calculate new
5262 * permissions based on new graph. If we fail, we'll roll-back the
5263 * replacement.
5265 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5266 if (ret < 0) {
5267 goto out;
5270 if (detach_subchain) {
5271 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
5274 found = g_hash_table_new(NULL, NULL);
5276 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
5277 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
5279 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5280 if (ret < 0) {
5281 goto out;
5284 ret = 0;
5286 out:
5287 tran_finalize(tran, ret);
5289 bdrv_drained_end(from);
5290 bdrv_unref(from);
5292 return ret;
5296 * Replace node @from by @to (where neither may be NULL).
5298 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5299 Error **errp)
5301 GLOBAL_STATE_CODE();
5303 return bdrv_replace_node_common(from, to, true, false, errp);
5306 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5308 GLOBAL_STATE_CODE();
5310 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5311 errp);
5315 * Add new bs contents at the top of an image chain while the chain is
5316 * live, while keeping required fields on the top layer.
5318 * This will modify the BlockDriverState fields, and swap contents
5319 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5321 * bs_new must not be attached to a BlockBackend and must not have backing
5322 * child.
5324 * This function does not create any image files.
5326 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5327 Error **errp)
5329 int ret;
5330 Transaction *tran = tran_new();
5332 GLOBAL_STATE_CODE();
5334 assert(!bs_new->backing);
5336 ret = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5337 &child_of_bds, bdrv_backing_role(bs_new),
5338 &bs_new->backing, tran, errp);
5339 if (ret < 0) {
5340 goto out;
5343 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5344 if (ret < 0) {
5345 goto out;
5348 ret = bdrv_refresh_perms(bs_new, errp);
5349 out:
5350 tran_finalize(tran, ret);
5352 bdrv_refresh_limits(bs_top, NULL, NULL);
5354 return ret;
5357 /* Not for empty child */
5358 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5359 Error **errp)
5361 int ret;
5362 Transaction *tran = tran_new();
5363 g_autoptr(GHashTable) found = NULL;
5364 g_autoptr(GSList) refresh_list = NULL;
5365 BlockDriverState *old_bs = child->bs;
5367 GLOBAL_STATE_CODE();
5369 bdrv_ref(old_bs);
5370 bdrv_drained_begin(old_bs);
5371 bdrv_drained_begin(new_bs);
5373 bdrv_replace_child_tran(&child, new_bs, tran, true);
5374 /* @new_bs must have been non-NULL, so @child must not have been freed */
5375 assert(child != NULL);
5377 found = g_hash_table_new(NULL, NULL);
5378 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5379 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5381 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5383 tran_finalize(tran, ret);
5385 bdrv_drained_end(old_bs);
5386 bdrv_drained_end(new_bs);
5387 bdrv_unref(old_bs);
5389 return ret;
5392 static void bdrv_delete(BlockDriverState *bs)
5394 assert(bdrv_op_blocker_is_empty(bs));
5395 assert(!bs->refcnt);
5396 GLOBAL_STATE_CODE();
5398 /* remove from list, if necessary */
5399 if (bs->node_name[0] != '\0') {
5400 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5402 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5404 bdrv_close(bs);
5406 g_free(bs);
5411 * Replace @bs by newly created block node.
5413 * @options is a QDict of options to pass to the block drivers, or NULL for an
5414 * empty set of options. The reference to the QDict belongs to the block layer
5415 * after the call (even on failure), so if the caller intends to reuse the
5416 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5418 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5419 int flags, Error **errp)
5421 ERRP_GUARD();
5422 int ret;
5423 BlockDriverState *new_node_bs = NULL;
5424 const char *drvname, *node_name;
5425 BlockDriver *drv;
5427 drvname = qdict_get_try_str(options, "driver");
5428 if (!drvname) {
5429 error_setg(errp, "driver is not specified");
5430 goto fail;
5433 drv = bdrv_find_format(drvname);
5434 if (!drv) {
5435 error_setg(errp, "Unknown driver: '%s'", drvname);
5436 goto fail;
5439 node_name = qdict_get_try_str(options, "node-name");
5441 GLOBAL_STATE_CODE();
5443 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5444 errp);
5445 options = NULL; /* bdrv_new_open_driver() eats options */
5446 if (!new_node_bs) {
5447 error_prepend(errp, "Could not create node: ");
5448 goto fail;
5451 bdrv_drained_begin(bs);
5452 ret = bdrv_replace_node(bs, new_node_bs, errp);
5453 bdrv_drained_end(bs);
5455 if (ret < 0) {
5456 error_prepend(errp, "Could not replace node: ");
5457 goto fail;
5460 return new_node_bs;
5462 fail:
5463 qobject_unref(options);
5464 bdrv_unref(new_node_bs);
5465 return NULL;
5469 * Run consistency checks on an image
5471 * Returns 0 if the check could be completed (it doesn't mean that the image is
5472 * free of errors) or -errno when an internal error occurred. The results of the
5473 * check are stored in res.
5475 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5476 BdrvCheckResult *res, BdrvCheckMode fix)
5478 IO_CODE();
5479 if (bs->drv == NULL) {
5480 return -ENOMEDIUM;
5482 if (bs->drv->bdrv_co_check == NULL) {
5483 return -ENOTSUP;
5486 memset(res, 0, sizeof(*res));
5487 return bs->drv->bdrv_co_check(bs, res, fix);
5491 * Return values:
5492 * 0 - success
5493 * -EINVAL - backing format specified, but no file
5494 * -ENOSPC - can't update the backing file because no space is left in the
5495 * image file header
5496 * -ENOTSUP - format driver doesn't support changing the backing file
5498 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5499 const char *backing_fmt, bool require)
5501 BlockDriver *drv = bs->drv;
5502 int ret;
5504 GLOBAL_STATE_CODE();
5506 if (!drv) {
5507 return -ENOMEDIUM;
5510 /* Backing file format doesn't make sense without a backing file */
5511 if (backing_fmt && !backing_file) {
5512 return -EINVAL;
5515 if (require && backing_file && !backing_fmt) {
5516 return -EINVAL;
5519 if (drv->bdrv_change_backing_file != NULL) {
5520 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5521 } else {
5522 ret = -ENOTSUP;
5525 if (ret == 0) {
5526 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5527 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5528 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5529 backing_file ?: "");
5531 return ret;
5535 * Finds the first non-filter node above bs in the chain between
5536 * active and bs. The returned node is either an immediate parent of
5537 * bs, or there are only filter nodes between the two.
5539 * Returns NULL if bs is not found in active's image chain,
5540 * or if active == bs.
5542 * Returns the bottommost base image if bs == NULL.
5544 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5545 BlockDriverState *bs)
5548 GLOBAL_STATE_CODE();
5550 bs = bdrv_skip_filters(bs);
5551 active = bdrv_skip_filters(active);
5553 while (active) {
5554 BlockDriverState *next = bdrv_backing_chain_next(active);
5555 if (bs == next) {
5556 return active;
5558 active = next;
5561 return NULL;
5564 /* Given a BDS, searches for the base layer. */
5565 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5567 GLOBAL_STATE_CODE();
5569 return bdrv_find_overlay(bs, NULL);
5573 * Return true if at least one of the COW (backing) and filter links
5574 * between @bs and @base is frozen. @errp is set if that's the case.
5575 * @base must be reachable from @bs, or NULL.
5577 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5578 Error **errp)
5580 BlockDriverState *i;
5581 BdrvChild *child;
5583 GLOBAL_STATE_CODE();
5585 for (i = bs; i != base; i = child_bs(child)) {
5586 child = bdrv_filter_or_cow_child(i);
5588 if (child && child->frozen) {
5589 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5590 child->name, i->node_name, child->bs->node_name);
5591 return true;
5595 return false;
5599 * Freeze all COW (backing) and filter links between @bs and @base.
5600 * If any of the links is already frozen the operation is aborted and
5601 * none of the links are modified.
5602 * @base must be reachable from @bs, or NULL.
5603 * Returns 0 on success. On failure returns < 0 and sets @errp.
5605 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5606 Error **errp)
5608 BlockDriverState *i;
5609 BdrvChild *child;
5611 GLOBAL_STATE_CODE();
5613 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5614 return -EPERM;
5617 for (i = bs; i != base; i = child_bs(child)) {
5618 child = bdrv_filter_or_cow_child(i);
5619 if (child && child->bs->never_freeze) {
5620 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5621 child->name, child->bs->node_name);
5622 return -EPERM;
5626 for (i = bs; i != base; i = child_bs(child)) {
5627 child = bdrv_filter_or_cow_child(i);
5628 if (child) {
5629 child->frozen = true;
5633 return 0;
5637 * Unfreeze all COW (backing) and filter links between @bs and @base.
5638 * The caller must ensure that all links are frozen before using this
5639 * function.
5640 * @base must be reachable from @bs, or NULL.
5642 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5644 BlockDriverState *i;
5645 BdrvChild *child;
5647 GLOBAL_STATE_CODE();
5649 for (i = bs; i != base; i = child_bs(child)) {
5650 child = bdrv_filter_or_cow_child(i);
5651 if (child) {
5652 assert(child->frozen);
5653 child->frozen = false;
5659 * Drops images above 'base' up to and including 'top', and sets the image
5660 * above 'top' to have base as its backing file.
5662 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5663 * information in 'bs' can be properly updated.
5665 * E.g., this will convert the following chain:
5666 * bottom <- base <- intermediate <- top <- active
5668 * to
5670 * bottom <- base <- active
5672 * It is allowed for bottom==base, in which case it converts:
5674 * base <- intermediate <- top <- active
5676 * to
5678 * base <- active
5680 * If backing_file_str is non-NULL, it will be used when modifying top's
5681 * overlay image metadata.
5683 * Error conditions:
5684 * if active == top, that is considered an error
5687 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5688 const char *backing_file_str)
5690 BlockDriverState *explicit_top = top;
5691 bool update_inherits_from;
5692 BdrvChild *c;
5693 Error *local_err = NULL;
5694 int ret = -EIO;
5695 g_autoptr(GSList) updated_children = NULL;
5696 GSList *p;
5698 GLOBAL_STATE_CODE();
5700 bdrv_ref(top);
5701 bdrv_subtree_drained_begin(top);
5703 if (!top->drv || !base->drv) {
5704 goto exit;
5707 /* Make sure that base is in the backing chain of top */
5708 if (!bdrv_chain_contains(top, base)) {
5709 goto exit;
5712 /* If 'base' recursively inherits from 'top' then we should set
5713 * base->inherits_from to top->inherits_from after 'top' and all
5714 * other intermediate nodes have been dropped.
5715 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5716 * it because no one inherits from it. We use explicit_top for that. */
5717 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5718 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5720 /* success - we can delete the intermediate states, and link top->base */
5721 if (!backing_file_str) {
5722 bdrv_refresh_filename(base);
5723 backing_file_str = base->filename;
5726 QLIST_FOREACH(c, &top->parents, next_parent) {
5727 updated_children = g_slist_prepend(updated_children, c);
5731 * It seems correct to pass detach_subchain=true here, but it triggers
5732 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5733 * another drained section, which modify the graph (for example, removing
5734 * the child, which we keep in updated_children list). So, it's a TODO.
5736 * Note, bug triggered if pass detach_subchain=true here and run
5737 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5738 * That's a FIXME.
5740 bdrv_replace_node_common(top, base, false, false, &local_err);
5741 if (local_err) {
5742 error_report_err(local_err);
5743 goto exit;
5746 for (p = updated_children; p; p = p->next) {
5747 c = p->data;
5749 if (c->klass->update_filename) {
5750 ret = c->klass->update_filename(c, base, backing_file_str,
5751 &local_err);
5752 if (ret < 0) {
5754 * TODO: Actually, we want to rollback all previous iterations
5755 * of this loop, and (which is almost impossible) previous
5756 * bdrv_replace_node()...
5758 * Note, that c->klass->update_filename may lead to permission
5759 * update, so it's a bad idea to call it inside permission
5760 * update transaction of bdrv_replace_node.
5762 error_report_err(local_err);
5763 goto exit;
5768 if (update_inherits_from) {
5769 base->inherits_from = explicit_top->inherits_from;
5772 ret = 0;
5773 exit:
5774 bdrv_subtree_drained_end(top);
5775 bdrv_unref(top);
5776 return ret;
5780 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5781 * sums the size of all data-bearing children. (This excludes backing
5782 * children.)
5784 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5786 BdrvChild *child;
5787 int64_t child_size, sum = 0;
5789 QLIST_FOREACH(child, &bs->children, next) {
5790 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5791 BDRV_CHILD_FILTERED))
5793 child_size = bdrv_get_allocated_file_size(child->bs);
5794 if (child_size < 0) {
5795 return child_size;
5797 sum += child_size;
5801 return sum;
5805 * Length of a allocated file in bytes. Sparse files are counted by actual
5806 * allocated space. Return < 0 if error or unknown.
5808 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5810 BlockDriver *drv = bs->drv;
5811 IO_CODE();
5813 if (!drv) {
5814 return -ENOMEDIUM;
5816 if (drv->bdrv_get_allocated_file_size) {
5817 return drv->bdrv_get_allocated_file_size(bs);
5820 if (drv->bdrv_file_open) {
5822 * Protocol drivers default to -ENOTSUP (most of their data is
5823 * not stored in any of their children (if they even have any),
5824 * so there is no generic way to figure it out).
5826 return -ENOTSUP;
5827 } else if (drv->is_filter) {
5828 /* Filter drivers default to the size of their filtered child */
5829 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5830 } else {
5831 /* Other drivers default to summing their children's sizes */
5832 return bdrv_sum_allocated_file_size(bs);
5837 * bdrv_measure:
5838 * @drv: Format driver
5839 * @opts: Creation options for new image
5840 * @in_bs: Existing image containing data for new image (may be NULL)
5841 * @errp: Error object
5842 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5843 * or NULL on error
5845 * Calculate file size required to create a new image.
5847 * If @in_bs is given then space for allocated clusters and zero clusters
5848 * from that image are included in the calculation. If @opts contains a
5849 * backing file that is shared by @in_bs then backing clusters may be omitted
5850 * from the calculation.
5852 * If @in_bs is NULL then the calculation includes no allocated clusters
5853 * unless a preallocation option is given in @opts.
5855 * Note that @in_bs may use a different BlockDriver from @drv.
5857 * If an error occurs the @errp pointer is set.
5859 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5860 BlockDriverState *in_bs, Error **errp)
5862 IO_CODE();
5863 if (!drv->bdrv_measure) {
5864 error_setg(errp, "Block driver '%s' does not support size measurement",
5865 drv->format_name);
5866 return NULL;
5869 return drv->bdrv_measure(opts, in_bs, errp);
5873 * Return number of sectors on success, -errno on error.
5875 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5877 BlockDriver *drv = bs->drv;
5878 IO_CODE();
5880 if (!drv)
5881 return -ENOMEDIUM;
5883 if (drv->has_variable_length) {
5884 int ret = refresh_total_sectors(bs, bs->total_sectors);
5885 if (ret < 0) {
5886 return ret;
5889 return bs->total_sectors;
5893 * Return length in bytes on success, -errno on error.
5894 * The length is always a multiple of BDRV_SECTOR_SIZE.
5896 int64_t bdrv_getlength(BlockDriverState *bs)
5898 int64_t ret = bdrv_nb_sectors(bs);
5899 IO_CODE();
5901 if (ret < 0) {
5902 return ret;
5904 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5905 return -EFBIG;
5907 return ret * BDRV_SECTOR_SIZE;
5910 /* return 0 as number of sectors if no device present or error */
5911 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5913 int64_t nb_sectors = bdrv_nb_sectors(bs);
5914 IO_CODE();
5916 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5919 bool bdrv_is_sg(BlockDriverState *bs)
5921 IO_CODE();
5922 return bs->sg;
5926 * Return whether the given node supports compressed writes.
5928 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5930 BlockDriverState *filtered;
5931 IO_CODE();
5933 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5934 return false;
5937 filtered = bdrv_filter_bs(bs);
5938 if (filtered) {
5940 * Filters can only forward compressed writes, so we have to
5941 * check the child.
5943 return bdrv_supports_compressed_writes(filtered);
5946 return true;
5949 const char *bdrv_get_format_name(BlockDriverState *bs)
5951 IO_CODE();
5952 return bs->drv ? bs->drv->format_name : NULL;
5955 static int qsort_strcmp(const void *a, const void *b)
5957 return strcmp(*(char *const *)a, *(char *const *)b);
5960 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5961 void *opaque, bool read_only)
5963 BlockDriver *drv;
5964 int count = 0;
5965 int i;
5966 const char **formats = NULL;
5968 GLOBAL_STATE_CODE();
5970 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5971 if (drv->format_name) {
5972 bool found = false;
5973 int i = count;
5975 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5976 continue;
5979 while (formats && i && !found) {
5980 found = !strcmp(formats[--i], drv->format_name);
5983 if (!found) {
5984 formats = g_renew(const char *, formats, count + 1);
5985 formats[count++] = drv->format_name;
5990 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5991 const char *format_name = block_driver_modules[i].format_name;
5993 if (format_name) {
5994 bool found = false;
5995 int j = count;
5997 if (use_bdrv_whitelist &&
5998 !bdrv_format_is_whitelisted(format_name, read_only)) {
5999 continue;
6002 while (formats && j && !found) {
6003 found = !strcmp(formats[--j], format_name);
6006 if (!found) {
6007 formats = g_renew(const char *, formats, count + 1);
6008 formats[count++] = format_name;
6013 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6015 for (i = 0; i < count; i++) {
6016 it(opaque, formats[i]);
6019 g_free(formats);
6022 /* This function is to find a node in the bs graph */
6023 BlockDriverState *bdrv_find_node(const char *node_name)
6025 BlockDriverState *bs;
6027 assert(node_name);
6028 GLOBAL_STATE_CODE();
6030 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6031 if (!strcmp(node_name, bs->node_name)) {
6032 return bs;
6035 return NULL;
6038 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6039 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6040 Error **errp)
6042 BlockDeviceInfoList *list;
6043 BlockDriverState *bs;
6045 GLOBAL_STATE_CODE();
6047 list = NULL;
6048 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6049 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6050 if (!info) {
6051 qapi_free_BlockDeviceInfoList(list);
6052 return NULL;
6054 QAPI_LIST_PREPEND(list, info);
6057 return list;
6060 typedef struct XDbgBlockGraphConstructor {
6061 XDbgBlockGraph *graph;
6062 GHashTable *graph_nodes;
6063 } XDbgBlockGraphConstructor;
6065 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6067 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6069 gr->graph = g_new0(XDbgBlockGraph, 1);
6070 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6072 return gr;
6075 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6077 XDbgBlockGraph *graph = gr->graph;
6079 g_hash_table_destroy(gr->graph_nodes);
6080 g_free(gr);
6082 return graph;
6085 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6087 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6089 if (ret != 0) {
6090 return ret;
6094 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6095 * answer of g_hash_table_lookup.
6097 ret = g_hash_table_size(gr->graph_nodes) + 1;
6098 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6100 return ret;
6103 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6104 XDbgBlockGraphNodeType type, const char *name)
6106 XDbgBlockGraphNode *n;
6108 n = g_new0(XDbgBlockGraphNode, 1);
6110 n->id = xdbg_graph_node_num(gr, node);
6111 n->type = type;
6112 n->name = g_strdup(name);
6114 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6117 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6118 const BdrvChild *child)
6120 BlockPermission qapi_perm;
6121 XDbgBlockGraphEdge *edge;
6122 GLOBAL_STATE_CODE();
6124 edge = g_new0(XDbgBlockGraphEdge, 1);
6126 edge->parent = xdbg_graph_node_num(gr, parent);
6127 edge->child = xdbg_graph_node_num(gr, child->bs);
6128 edge->name = g_strdup(child->name);
6130 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6131 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6133 if (flag & child->perm) {
6134 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6136 if (flag & child->shared_perm) {
6137 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6141 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6145 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6147 BlockBackend *blk;
6148 BlockJob *job;
6149 BlockDriverState *bs;
6150 BdrvChild *child;
6151 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6153 GLOBAL_STATE_CODE();
6155 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6156 char *allocated_name = NULL;
6157 const char *name = blk_name(blk);
6159 if (!*name) {
6160 name = allocated_name = blk_get_attached_dev_id(blk);
6162 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6163 name);
6164 g_free(allocated_name);
6165 if (blk_root(blk)) {
6166 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6170 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
6171 GSList *el;
6173 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6174 job->job.id);
6175 for (el = job->nodes; el; el = el->next) {
6176 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6180 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6181 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6182 bs->node_name);
6183 QLIST_FOREACH(child, &bs->children, next) {
6184 xdbg_graph_add_edge(gr, bs, child);
6188 return xdbg_graph_finalize(gr);
6191 BlockDriverState *bdrv_lookup_bs(const char *device,
6192 const char *node_name,
6193 Error **errp)
6195 BlockBackend *blk;
6196 BlockDriverState *bs;
6198 GLOBAL_STATE_CODE();
6200 if (device) {
6201 blk = blk_by_name(device);
6203 if (blk) {
6204 bs = blk_bs(blk);
6205 if (!bs) {
6206 error_setg(errp, "Device '%s' has no medium", device);
6209 return bs;
6213 if (node_name) {
6214 bs = bdrv_find_node(node_name);
6216 if (bs) {
6217 return bs;
6221 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6222 device ? device : "",
6223 node_name ? node_name : "");
6224 return NULL;
6227 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6228 * return false. If either argument is NULL, return false. */
6229 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6232 GLOBAL_STATE_CODE();
6234 while (top && top != base) {
6235 top = bdrv_filter_or_cow_bs(top);
6238 return top != NULL;
6241 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6243 GLOBAL_STATE_CODE();
6244 if (!bs) {
6245 return QTAILQ_FIRST(&graph_bdrv_states);
6247 return QTAILQ_NEXT(bs, node_list);
6250 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6252 GLOBAL_STATE_CODE();
6253 if (!bs) {
6254 return QTAILQ_FIRST(&all_bdrv_states);
6256 return QTAILQ_NEXT(bs, bs_list);
6259 const char *bdrv_get_node_name(const BlockDriverState *bs)
6261 IO_CODE();
6262 return bs->node_name;
6265 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6267 BdrvChild *c;
6268 const char *name;
6269 IO_CODE();
6271 /* If multiple parents have a name, just pick the first one. */
6272 QLIST_FOREACH(c, &bs->parents, next_parent) {
6273 if (c->klass->get_name) {
6274 name = c->klass->get_name(c);
6275 if (name && *name) {
6276 return name;
6281 return NULL;
6284 /* TODO check what callers really want: bs->node_name or blk_name() */
6285 const char *bdrv_get_device_name(const BlockDriverState *bs)
6287 IO_CODE();
6288 return bdrv_get_parent_name(bs) ?: "";
6291 /* This can be used to identify nodes that might not have a device
6292 * name associated. Since node and device names live in the same
6293 * namespace, the result is unambiguous. The exception is if both are
6294 * absent, then this returns an empty (non-null) string. */
6295 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6297 IO_CODE();
6298 return bdrv_get_parent_name(bs) ?: bs->node_name;
6301 int bdrv_get_flags(BlockDriverState *bs)
6303 IO_CODE();
6304 return bs->open_flags;
6307 int bdrv_has_zero_init_1(BlockDriverState *bs)
6309 GLOBAL_STATE_CODE();
6310 return 1;
6313 int bdrv_has_zero_init(BlockDriverState *bs)
6315 BlockDriverState *filtered;
6316 GLOBAL_STATE_CODE();
6318 if (!bs->drv) {
6319 return 0;
6322 /* If BS is a copy on write image, it is initialized to
6323 the contents of the base image, which may not be zeroes. */
6324 if (bdrv_cow_child(bs)) {
6325 return 0;
6327 if (bs->drv->bdrv_has_zero_init) {
6328 return bs->drv->bdrv_has_zero_init(bs);
6331 filtered = bdrv_filter_bs(bs);
6332 if (filtered) {
6333 return bdrv_has_zero_init(filtered);
6336 /* safe default */
6337 return 0;
6340 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6342 IO_CODE();
6343 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6344 return false;
6347 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6350 void bdrv_get_backing_filename(BlockDriverState *bs,
6351 char *filename, int filename_size)
6353 IO_CODE();
6354 pstrcpy(filename, filename_size, bs->backing_file);
6357 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6359 int ret;
6360 BlockDriver *drv = bs->drv;
6361 IO_CODE();
6362 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6363 if (!drv) {
6364 return -ENOMEDIUM;
6366 if (!drv->bdrv_get_info) {
6367 BlockDriverState *filtered = bdrv_filter_bs(bs);
6368 if (filtered) {
6369 return bdrv_get_info(filtered, bdi);
6371 return -ENOTSUP;
6373 memset(bdi, 0, sizeof(*bdi));
6374 ret = drv->bdrv_get_info(bs, bdi);
6375 if (ret < 0) {
6376 return ret;
6379 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6380 return -EINVAL;
6383 return 0;
6386 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6387 Error **errp)
6389 BlockDriver *drv = bs->drv;
6390 IO_CODE();
6391 if (drv && drv->bdrv_get_specific_info) {
6392 return drv->bdrv_get_specific_info(bs, errp);
6394 return NULL;
6397 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6399 BlockDriver *drv = bs->drv;
6400 IO_CODE();
6401 if (!drv || !drv->bdrv_get_specific_stats) {
6402 return NULL;
6404 return drv->bdrv_get_specific_stats(bs);
6407 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6409 IO_CODE();
6410 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6411 return;
6414 bs->drv->bdrv_debug_event(bs, event);
6417 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6419 GLOBAL_STATE_CODE();
6420 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6421 bs = bdrv_primary_bs(bs);
6424 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6425 assert(bs->drv->bdrv_debug_remove_breakpoint);
6426 return bs;
6429 return NULL;
6432 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6433 const char *tag)
6435 GLOBAL_STATE_CODE();
6436 bs = bdrv_find_debug_node(bs);
6437 if (bs) {
6438 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6441 return -ENOTSUP;
6444 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6446 GLOBAL_STATE_CODE();
6447 bs = bdrv_find_debug_node(bs);
6448 if (bs) {
6449 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6452 return -ENOTSUP;
6455 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6457 GLOBAL_STATE_CODE();
6458 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6459 bs = bdrv_primary_bs(bs);
6462 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6463 return bs->drv->bdrv_debug_resume(bs, tag);
6466 return -ENOTSUP;
6469 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6471 GLOBAL_STATE_CODE();
6472 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6473 bs = bdrv_primary_bs(bs);
6476 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6477 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6480 return false;
6483 /* backing_file can either be relative, or absolute, or a protocol. If it is
6484 * relative, it must be relative to the chain. So, passing in bs->filename
6485 * from a BDS as backing_file should not be done, as that may be relative to
6486 * the CWD rather than the chain. */
6487 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6488 const char *backing_file)
6490 char *filename_full = NULL;
6491 char *backing_file_full = NULL;
6492 char *filename_tmp = NULL;
6493 int is_protocol = 0;
6494 bool filenames_refreshed = false;
6495 BlockDriverState *curr_bs = NULL;
6496 BlockDriverState *retval = NULL;
6497 BlockDriverState *bs_below;
6499 GLOBAL_STATE_CODE();
6501 if (!bs || !bs->drv || !backing_file) {
6502 return NULL;
6505 filename_full = g_malloc(PATH_MAX);
6506 backing_file_full = g_malloc(PATH_MAX);
6508 is_protocol = path_has_protocol(backing_file);
6511 * Being largely a legacy function, skip any filters here
6512 * (because filters do not have normal filenames, so they cannot
6513 * match anyway; and allowing json:{} filenames is a bit out of
6514 * scope).
6516 for (curr_bs = bdrv_skip_filters(bs);
6517 bdrv_cow_child(curr_bs) != NULL;
6518 curr_bs = bs_below)
6520 bs_below = bdrv_backing_chain_next(curr_bs);
6522 if (bdrv_backing_overridden(curr_bs)) {
6524 * If the backing file was overridden, we can only compare
6525 * directly against the backing node's filename.
6528 if (!filenames_refreshed) {
6530 * This will automatically refresh all of the
6531 * filenames in the rest of the backing chain, so we
6532 * only need to do this once.
6534 bdrv_refresh_filename(bs_below);
6535 filenames_refreshed = true;
6538 if (strcmp(backing_file, bs_below->filename) == 0) {
6539 retval = bs_below;
6540 break;
6542 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6544 * If either of the filename paths is actually a protocol, then
6545 * compare unmodified paths; otherwise make paths relative.
6547 char *backing_file_full_ret;
6549 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6550 retval = bs_below;
6551 break;
6553 /* Also check against the full backing filename for the image */
6554 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6555 NULL);
6556 if (backing_file_full_ret) {
6557 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6558 g_free(backing_file_full_ret);
6559 if (equal) {
6560 retval = bs_below;
6561 break;
6564 } else {
6565 /* If not an absolute filename path, make it relative to the current
6566 * image's filename path */
6567 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6568 NULL);
6569 /* We are going to compare canonicalized absolute pathnames */
6570 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6571 g_free(filename_tmp);
6572 continue;
6574 g_free(filename_tmp);
6576 /* We need to make sure the backing filename we are comparing against
6577 * is relative to the current image filename (or absolute) */
6578 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6579 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6580 g_free(filename_tmp);
6581 continue;
6583 g_free(filename_tmp);
6585 if (strcmp(backing_file_full, filename_full) == 0) {
6586 retval = bs_below;
6587 break;
6592 g_free(filename_full);
6593 g_free(backing_file_full);
6594 return retval;
6597 void bdrv_init(void)
6599 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6600 use_bdrv_whitelist = 1;
6601 #endif
6602 module_call_init(MODULE_INIT_BLOCK);
6605 void bdrv_init_with_whitelist(void)
6607 use_bdrv_whitelist = 1;
6608 bdrv_init();
6611 int bdrv_activate(BlockDriverState *bs, Error **errp)
6613 BdrvChild *child, *parent;
6614 Error *local_err = NULL;
6615 int ret;
6616 BdrvDirtyBitmap *bm;
6618 GLOBAL_STATE_CODE();
6620 if (!bs->drv) {
6621 return -ENOMEDIUM;
6624 QLIST_FOREACH(child, &bs->children, next) {
6625 bdrv_activate(child->bs, &local_err);
6626 if (local_err) {
6627 error_propagate(errp, local_err);
6628 return -EINVAL;
6633 * Update permissions, they may differ for inactive nodes.
6635 * Note that the required permissions of inactive images are always a
6636 * subset of the permissions required after activating the image. This
6637 * allows us to just get the permissions upfront without restricting
6638 * bdrv_co_invalidate_cache().
6640 * It also means that in error cases, we don't have to try and revert to
6641 * the old permissions (which is an operation that could fail, too). We can
6642 * just keep the extended permissions for the next time that an activation
6643 * of the image is tried.
6645 if (bs->open_flags & BDRV_O_INACTIVE) {
6646 bs->open_flags &= ~BDRV_O_INACTIVE;
6647 ret = bdrv_refresh_perms(bs, errp);
6648 if (ret < 0) {
6649 bs->open_flags |= BDRV_O_INACTIVE;
6650 return ret;
6653 ret = bdrv_invalidate_cache(bs, errp);
6654 if (ret < 0) {
6655 bs->open_flags |= BDRV_O_INACTIVE;
6656 return ret;
6659 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6660 bdrv_dirty_bitmap_skip_store(bm, false);
6663 ret = refresh_total_sectors(bs, bs->total_sectors);
6664 if (ret < 0) {
6665 bs->open_flags |= BDRV_O_INACTIVE;
6666 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6667 return ret;
6671 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6672 if (parent->klass->activate) {
6673 parent->klass->activate(parent, &local_err);
6674 if (local_err) {
6675 bs->open_flags |= BDRV_O_INACTIVE;
6676 error_propagate(errp, local_err);
6677 return -EINVAL;
6682 return 0;
6685 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6687 Error *local_err = NULL;
6688 IO_CODE();
6690 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6692 if (bs->drv->bdrv_co_invalidate_cache) {
6693 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6694 if (local_err) {
6695 error_propagate(errp, local_err);
6696 return -EINVAL;
6700 return 0;
6703 void bdrv_activate_all(Error **errp)
6705 BlockDriverState *bs;
6706 BdrvNextIterator it;
6708 GLOBAL_STATE_CODE();
6710 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6711 AioContext *aio_context = bdrv_get_aio_context(bs);
6712 int ret;
6714 aio_context_acquire(aio_context);
6715 ret = bdrv_activate(bs, errp);
6716 aio_context_release(aio_context);
6717 if (ret < 0) {
6718 bdrv_next_cleanup(&it);
6719 return;
6724 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6726 BdrvChild *parent;
6727 GLOBAL_STATE_CODE();
6729 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6730 if (parent->klass->parent_is_bds) {
6731 BlockDriverState *parent_bs = parent->opaque;
6732 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6733 return true;
6738 return false;
6741 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6743 BdrvChild *child, *parent;
6744 int ret;
6745 uint64_t cumulative_perms, cumulative_shared_perms;
6747 GLOBAL_STATE_CODE();
6749 if (!bs->drv) {
6750 return -ENOMEDIUM;
6753 /* Make sure that we don't inactivate a child before its parent.
6754 * It will be covered by recursion from the yet active parent. */
6755 if (bdrv_has_bds_parent(bs, true)) {
6756 return 0;
6759 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6761 /* Inactivate this node */
6762 if (bs->drv->bdrv_inactivate) {
6763 ret = bs->drv->bdrv_inactivate(bs);
6764 if (ret < 0) {
6765 return ret;
6769 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6770 if (parent->klass->inactivate) {
6771 ret = parent->klass->inactivate(parent);
6772 if (ret < 0) {
6773 return ret;
6778 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6779 &cumulative_shared_perms);
6780 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6781 /* Our inactive parents still need write access. Inactivation failed. */
6782 return -EPERM;
6785 bs->open_flags |= BDRV_O_INACTIVE;
6788 * Update permissions, they may differ for inactive nodes.
6789 * We only tried to loosen restrictions, so errors are not fatal, ignore
6790 * them.
6792 bdrv_refresh_perms(bs, NULL);
6794 /* Recursively inactivate children */
6795 QLIST_FOREACH(child, &bs->children, next) {
6796 ret = bdrv_inactivate_recurse(child->bs);
6797 if (ret < 0) {
6798 return ret;
6802 return 0;
6805 int bdrv_inactivate_all(void)
6807 BlockDriverState *bs = NULL;
6808 BdrvNextIterator it;
6809 int ret = 0;
6810 GSList *aio_ctxs = NULL, *ctx;
6812 GLOBAL_STATE_CODE();
6814 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6815 AioContext *aio_context = bdrv_get_aio_context(bs);
6817 if (!g_slist_find(aio_ctxs, aio_context)) {
6818 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6819 aio_context_acquire(aio_context);
6823 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6824 /* Nodes with BDS parents are covered by recursion from the last
6825 * parent that gets inactivated. Don't inactivate them a second
6826 * time if that has already happened. */
6827 if (bdrv_has_bds_parent(bs, false)) {
6828 continue;
6830 ret = bdrv_inactivate_recurse(bs);
6831 if (ret < 0) {
6832 bdrv_next_cleanup(&it);
6833 goto out;
6837 out:
6838 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6839 AioContext *aio_context = ctx->data;
6840 aio_context_release(aio_context);
6842 g_slist_free(aio_ctxs);
6844 return ret;
6847 /**************************************************************/
6848 /* removable device support */
6851 * Return TRUE if the media is present
6853 bool bdrv_is_inserted(BlockDriverState *bs)
6855 BlockDriver *drv = bs->drv;
6856 BdrvChild *child;
6857 IO_CODE();
6859 if (!drv) {
6860 return false;
6862 if (drv->bdrv_is_inserted) {
6863 return drv->bdrv_is_inserted(bs);
6865 QLIST_FOREACH(child, &bs->children, next) {
6866 if (!bdrv_is_inserted(child->bs)) {
6867 return false;
6870 return true;
6874 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6876 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6878 BlockDriver *drv = bs->drv;
6879 IO_CODE();
6881 if (drv && drv->bdrv_eject) {
6882 drv->bdrv_eject(bs, eject_flag);
6887 * Lock or unlock the media (if it is locked, the user won't be able
6888 * to eject it manually).
6890 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6892 BlockDriver *drv = bs->drv;
6893 IO_CODE();
6894 trace_bdrv_lock_medium(bs, locked);
6896 if (drv && drv->bdrv_lock_medium) {
6897 drv->bdrv_lock_medium(bs, locked);
6901 /* Get a reference to bs */
6902 void bdrv_ref(BlockDriverState *bs)
6904 GLOBAL_STATE_CODE();
6905 bs->refcnt++;
6908 /* Release a previously grabbed reference to bs.
6909 * If after releasing, reference count is zero, the BlockDriverState is
6910 * deleted. */
6911 void bdrv_unref(BlockDriverState *bs)
6913 GLOBAL_STATE_CODE();
6914 if (!bs) {
6915 return;
6917 assert(bs->refcnt > 0);
6918 if (--bs->refcnt == 0) {
6919 bdrv_delete(bs);
6923 struct BdrvOpBlocker {
6924 Error *reason;
6925 QLIST_ENTRY(BdrvOpBlocker) list;
6928 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6930 BdrvOpBlocker *blocker;
6931 GLOBAL_STATE_CODE();
6932 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6933 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6934 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6935 error_propagate_prepend(errp, error_copy(blocker->reason),
6936 "Node '%s' is busy: ",
6937 bdrv_get_device_or_node_name(bs));
6938 return true;
6940 return false;
6943 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6945 BdrvOpBlocker *blocker;
6946 GLOBAL_STATE_CODE();
6947 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6949 blocker = g_new0(BdrvOpBlocker, 1);
6950 blocker->reason = reason;
6951 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6954 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6956 BdrvOpBlocker *blocker, *next;
6957 GLOBAL_STATE_CODE();
6958 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6959 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6960 if (blocker->reason == reason) {
6961 QLIST_REMOVE(blocker, list);
6962 g_free(blocker);
6967 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6969 int i;
6970 GLOBAL_STATE_CODE();
6971 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6972 bdrv_op_block(bs, i, reason);
6976 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6978 int i;
6979 GLOBAL_STATE_CODE();
6980 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6981 bdrv_op_unblock(bs, i, reason);
6985 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6987 int i;
6988 GLOBAL_STATE_CODE();
6989 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6990 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6991 return false;
6994 return true;
6997 void bdrv_img_create(const char *filename, const char *fmt,
6998 const char *base_filename, const char *base_fmt,
6999 char *options, uint64_t img_size, int flags, bool quiet,
7000 Error **errp)
7002 QemuOptsList *create_opts = NULL;
7003 QemuOpts *opts = NULL;
7004 const char *backing_fmt, *backing_file;
7005 int64_t size;
7006 BlockDriver *drv, *proto_drv;
7007 Error *local_err = NULL;
7008 int ret = 0;
7010 GLOBAL_STATE_CODE();
7012 /* Find driver and parse its options */
7013 drv = bdrv_find_format(fmt);
7014 if (!drv) {
7015 error_setg(errp, "Unknown file format '%s'", fmt);
7016 return;
7019 proto_drv = bdrv_find_protocol(filename, true, errp);
7020 if (!proto_drv) {
7021 return;
7024 if (!drv->create_opts) {
7025 error_setg(errp, "Format driver '%s' does not support image creation",
7026 drv->format_name);
7027 return;
7030 if (!proto_drv->create_opts) {
7031 error_setg(errp, "Protocol driver '%s' does not support image creation",
7032 proto_drv->format_name);
7033 return;
7036 /* Create parameter list */
7037 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7038 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7040 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7042 /* Parse -o options */
7043 if (options) {
7044 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7045 goto out;
7049 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7050 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7051 } else if (img_size != UINT64_C(-1)) {
7052 error_setg(errp, "The image size must be specified only once");
7053 goto out;
7056 if (base_filename) {
7057 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7058 NULL)) {
7059 error_setg(errp, "Backing file not supported for file format '%s'",
7060 fmt);
7061 goto out;
7065 if (base_fmt) {
7066 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7067 error_setg(errp, "Backing file format not supported for file "
7068 "format '%s'", fmt);
7069 goto out;
7073 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7074 if (backing_file) {
7075 if (!strcmp(filename, backing_file)) {
7076 error_setg(errp, "Error: Trying to create an image with the "
7077 "same filename as the backing file");
7078 goto out;
7080 if (backing_file[0] == '\0') {
7081 error_setg(errp, "Expected backing file name, got empty string");
7082 goto out;
7086 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7088 /* The size for the image must always be specified, unless we have a backing
7089 * file and we have not been forbidden from opening it. */
7090 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7091 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7092 BlockDriverState *bs;
7093 char *full_backing;
7094 int back_flags;
7095 QDict *backing_options = NULL;
7097 full_backing =
7098 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7099 &local_err);
7100 if (local_err) {
7101 goto out;
7103 assert(full_backing);
7106 * No need to do I/O here, which allows us to open encrypted
7107 * backing images without needing the secret
7109 back_flags = flags;
7110 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7111 back_flags |= BDRV_O_NO_IO;
7113 backing_options = qdict_new();
7114 if (backing_fmt) {
7115 qdict_put_str(backing_options, "driver", backing_fmt);
7117 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7119 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7120 &local_err);
7121 g_free(full_backing);
7122 if (!bs) {
7123 error_append_hint(&local_err, "Could not open backing image.\n");
7124 goto out;
7125 } else {
7126 if (!backing_fmt) {
7127 error_setg(&local_err,
7128 "Backing file specified without backing format");
7129 error_append_hint(&local_err, "Detected format of %s.",
7130 bs->drv->format_name);
7131 goto out;
7133 if (size == -1) {
7134 /* Opened BS, have no size */
7135 size = bdrv_getlength(bs);
7136 if (size < 0) {
7137 error_setg_errno(errp, -size, "Could not get size of '%s'",
7138 backing_file);
7139 bdrv_unref(bs);
7140 goto out;
7142 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7144 bdrv_unref(bs);
7146 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7147 } else if (backing_file && !backing_fmt) {
7148 error_setg(&local_err,
7149 "Backing file specified without backing format");
7150 goto out;
7153 if (size == -1) {
7154 error_setg(errp, "Image creation needs a size parameter");
7155 goto out;
7158 if (!quiet) {
7159 printf("Formatting '%s', fmt=%s ", filename, fmt);
7160 qemu_opts_print(opts, " ");
7161 puts("");
7162 fflush(stdout);
7165 ret = bdrv_create(drv, filename, opts, &local_err);
7167 if (ret == -EFBIG) {
7168 /* This is generally a better message than whatever the driver would
7169 * deliver (especially because of the cluster_size_hint), since that
7170 * is most probably not much different from "image too large". */
7171 const char *cluster_size_hint = "";
7172 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7173 cluster_size_hint = " (try using a larger cluster size)";
7175 error_setg(errp, "The image size is too large for file format '%s'"
7176 "%s", fmt, cluster_size_hint);
7177 error_free(local_err);
7178 local_err = NULL;
7181 out:
7182 qemu_opts_del(opts);
7183 qemu_opts_free(create_opts);
7184 error_propagate(errp, local_err);
7187 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7189 IO_CODE();
7190 return bs ? bs->aio_context : qemu_get_aio_context();
7193 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7195 Coroutine *self = qemu_coroutine_self();
7196 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7197 AioContext *new_ctx;
7198 IO_CODE();
7201 * Increase bs->in_flight to ensure that this operation is completed before
7202 * moving the node to a different AioContext. Read new_ctx only afterwards.
7204 bdrv_inc_in_flight(bs);
7206 new_ctx = bdrv_get_aio_context(bs);
7207 aio_co_reschedule_self(new_ctx);
7208 return old_ctx;
7211 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7213 IO_CODE();
7214 aio_co_reschedule_self(old_ctx);
7215 bdrv_dec_in_flight(bs);
7218 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7220 AioContext *ctx = bdrv_get_aio_context(bs);
7222 /* In the main thread, bs->aio_context won't change concurrently */
7223 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7226 * We're in coroutine context, so we already hold the lock of the main
7227 * loop AioContext. Don't lock it twice to avoid deadlocks.
7229 assert(qemu_in_coroutine());
7230 if (ctx != qemu_get_aio_context()) {
7231 aio_context_acquire(ctx);
7235 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7237 AioContext *ctx = bdrv_get_aio_context(bs);
7239 assert(qemu_in_coroutine());
7240 if (ctx != qemu_get_aio_context()) {
7241 aio_context_release(ctx);
7245 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
7247 IO_CODE();
7248 aio_co_enter(bdrv_get_aio_context(bs), co);
7251 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7253 GLOBAL_STATE_CODE();
7254 QLIST_REMOVE(ban, list);
7255 g_free(ban);
7258 static void bdrv_detach_aio_context(BlockDriverState *bs)
7260 BdrvAioNotifier *baf, *baf_tmp;
7262 assert(!bs->walking_aio_notifiers);
7263 GLOBAL_STATE_CODE();
7264 bs->walking_aio_notifiers = true;
7265 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7266 if (baf->deleted) {
7267 bdrv_do_remove_aio_context_notifier(baf);
7268 } else {
7269 baf->detach_aio_context(baf->opaque);
7272 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7273 * remove remaining aio notifiers if we aren't called again.
7275 bs->walking_aio_notifiers = false;
7277 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7278 bs->drv->bdrv_detach_aio_context(bs);
7281 if (bs->quiesce_counter) {
7282 aio_enable_external(bs->aio_context);
7284 bs->aio_context = NULL;
7287 static void bdrv_attach_aio_context(BlockDriverState *bs,
7288 AioContext *new_context)
7290 BdrvAioNotifier *ban, *ban_tmp;
7291 GLOBAL_STATE_CODE();
7293 if (bs->quiesce_counter) {
7294 aio_disable_external(new_context);
7297 bs->aio_context = new_context;
7299 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7300 bs->drv->bdrv_attach_aio_context(bs, new_context);
7303 assert(!bs->walking_aio_notifiers);
7304 bs->walking_aio_notifiers = true;
7305 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7306 if (ban->deleted) {
7307 bdrv_do_remove_aio_context_notifier(ban);
7308 } else {
7309 ban->attached_aio_context(new_context, ban->opaque);
7312 bs->walking_aio_notifiers = false;
7316 * Changes the AioContext used for fd handlers, timers, and BHs by this
7317 * BlockDriverState and all its children and parents.
7319 * Must be called from the main AioContext.
7321 * The caller must own the AioContext lock for the old AioContext of bs, but it
7322 * must not own the AioContext lock for new_context (unless new_context is the
7323 * same as the current context of bs).
7325 * @ignore will accumulate all visited BdrvChild object. The caller is
7326 * responsible for freeing the list afterwards.
7328 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
7329 AioContext *new_context, GSList **ignore)
7331 AioContext *old_context = bdrv_get_aio_context(bs);
7332 GSList *children_to_process = NULL;
7333 GSList *parents_to_process = NULL;
7334 GSList *entry;
7335 BdrvChild *child, *parent;
7337 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7338 GLOBAL_STATE_CODE();
7340 if (old_context == new_context) {
7341 return;
7344 bdrv_drained_begin(bs);
7346 QLIST_FOREACH(child, &bs->children, next) {
7347 if (g_slist_find(*ignore, child)) {
7348 continue;
7350 *ignore = g_slist_prepend(*ignore, child);
7351 children_to_process = g_slist_prepend(children_to_process, child);
7354 QLIST_FOREACH(parent, &bs->parents, next_parent) {
7355 if (g_slist_find(*ignore, parent)) {
7356 continue;
7358 *ignore = g_slist_prepend(*ignore, parent);
7359 parents_to_process = g_slist_prepend(parents_to_process, parent);
7362 for (entry = children_to_process;
7363 entry != NULL;
7364 entry = g_slist_next(entry)) {
7365 child = entry->data;
7366 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
7368 g_slist_free(children_to_process);
7370 for (entry = parents_to_process;
7371 entry != NULL;
7372 entry = g_slist_next(entry)) {
7373 parent = entry->data;
7374 assert(parent->klass->set_aio_ctx);
7375 parent->klass->set_aio_ctx(parent, new_context, ignore);
7377 g_slist_free(parents_to_process);
7379 bdrv_detach_aio_context(bs);
7381 /* Acquire the new context, if necessary */
7382 if (qemu_get_aio_context() != new_context) {
7383 aio_context_acquire(new_context);
7386 bdrv_attach_aio_context(bs, new_context);
7389 * If this function was recursively called from
7390 * bdrv_set_aio_context_ignore(), there may be nodes in the
7391 * subtree that have not yet been moved to the new AioContext.
7392 * Release the old one so bdrv_drained_end() can poll them.
7394 if (qemu_get_aio_context() != old_context) {
7395 aio_context_release(old_context);
7398 bdrv_drained_end(bs);
7400 if (qemu_get_aio_context() != old_context) {
7401 aio_context_acquire(old_context);
7403 if (qemu_get_aio_context() != new_context) {
7404 aio_context_release(new_context);
7408 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7409 GSList **ignore, Error **errp)
7411 GLOBAL_STATE_CODE();
7412 if (g_slist_find(*ignore, c)) {
7413 return true;
7415 *ignore = g_slist_prepend(*ignore, c);
7418 * A BdrvChildClass that doesn't handle AioContext changes cannot
7419 * tolerate any AioContext changes
7421 if (!c->klass->can_set_aio_ctx) {
7422 char *user = bdrv_child_user_desc(c);
7423 error_setg(errp, "Changing iothreads is not supported by %s", user);
7424 g_free(user);
7425 return false;
7427 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
7428 assert(!errp || *errp);
7429 return false;
7431 return true;
7434 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7435 GSList **ignore, Error **errp)
7437 GLOBAL_STATE_CODE();
7438 if (g_slist_find(*ignore, c)) {
7439 return true;
7441 *ignore = g_slist_prepend(*ignore, c);
7442 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
7445 /* @ignore will accumulate all visited BdrvChild object. The caller is
7446 * responsible for freeing the list afterwards. */
7447 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7448 GSList **ignore, Error **errp)
7450 BdrvChild *c;
7452 if (bdrv_get_aio_context(bs) == ctx) {
7453 return true;
7456 GLOBAL_STATE_CODE();
7458 QLIST_FOREACH(c, &bs->parents, next_parent) {
7459 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
7460 return false;
7463 QLIST_FOREACH(c, &bs->children, next) {
7464 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
7465 return false;
7469 return true;
7472 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7473 BdrvChild *ignore_child, Error **errp)
7475 GSList *ignore;
7476 bool ret;
7478 GLOBAL_STATE_CODE();
7480 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7481 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
7482 g_slist_free(ignore);
7484 if (!ret) {
7485 return -EPERM;
7488 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7489 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
7490 g_slist_free(ignore);
7492 return 0;
7495 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7496 Error **errp)
7498 GLOBAL_STATE_CODE();
7499 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
7502 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7503 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7504 void (*detach_aio_context)(void *opaque), void *opaque)
7506 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7507 *ban = (BdrvAioNotifier){
7508 .attached_aio_context = attached_aio_context,
7509 .detach_aio_context = detach_aio_context,
7510 .opaque = opaque
7512 GLOBAL_STATE_CODE();
7514 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7517 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7518 void (*attached_aio_context)(AioContext *,
7519 void *),
7520 void (*detach_aio_context)(void *),
7521 void *opaque)
7523 BdrvAioNotifier *ban, *ban_next;
7524 GLOBAL_STATE_CODE();
7526 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7527 if (ban->attached_aio_context == attached_aio_context &&
7528 ban->detach_aio_context == detach_aio_context &&
7529 ban->opaque == opaque &&
7530 ban->deleted == false)
7532 if (bs->walking_aio_notifiers) {
7533 ban->deleted = true;
7534 } else {
7535 bdrv_do_remove_aio_context_notifier(ban);
7537 return;
7541 abort();
7544 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7545 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7546 bool force,
7547 Error **errp)
7549 GLOBAL_STATE_CODE();
7550 if (!bs->drv) {
7551 error_setg(errp, "Node is ejected");
7552 return -ENOMEDIUM;
7554 if (!bs->drv->bdrv_amend_options) {
7555 error_setg(errp, "Block driver '%s' does not support option amendment",
7556 bs->drv->format_name);
7557 return -ENOTSUP;
7559 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7560 cb_opaque, force, errp);
7564 * This function checks whether the given @to_replace is allowed to be
7565 * replaced by a node that always shows the same data as @bs. This is
7566 * used for example to verify whether the mirror job can replace
7567 * @to_replace by the target mirrored from @bs.
7568 * To be replaceable, @bs and @to_replace may either be guaranteed to
7569 * always show the same data (because they are only connected through
7570 * filters), or some driver may allow replacing one of its children
7571 * because it can guarantee that this child's data is not visible at
7572 * all (for example, for dissenting quorum children that have no other
7573 * parents).
7575 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7576 BlockDriverState *to_replace)
7578 BlockDriverState *filtered;
7580 GLOBAL_STATE_CODE();
7582 if (!bs || !bs->drv) {
7583 return false;
7586 if (bs == to_replace) {
7587 return true;
7590 /* See what the driver can do */
7591 if (bs->drv->bdrv_recurse_can_replace) {
7592 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7595 /* For filters without an own implementation, we can recurse on our own */
7596 filtered = bdrv_filter_bs(bs);
7597 if (filtered) {
7598 return bdrv_recurse_can_replace(filtered, to_replace);
7601 /* Safe default */
7602 return false;
7606 * Check whether the given @node_name can be replaced by a node that
7607 * has the same data as @parent_bs. If so, return @node_name's BDS;
7608 * NULL otherwise.
7610 * @node_name must be a (recursive) *child of @parent_bs (or this
7611 * function will return NULL).
7613 * The result (whether the node can be replaced or not) is only valid
7614 * for as long as no graph or permission changes occur.
7616 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7617 const char *node_name, Error **errp)
7619 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7620 AioContext *aio_context;
7622 GLOBAL_STATE_CODE();
7624 if (!to_replace_bs) {
7625 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7626 return NULL;
7629 aio_context = bdrv_get_aio_context(to_replace_bs);
7630 aio_context_acquire(aio_context);
7632 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7633 to_replace_bs = NULL;
7634 goto out;
7637 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7638 * most non filter in order to prevent data corruption.
7639 * Another benefit is that this tests exclude backing files which are
7640 * blocked by the backing blockers.
7642 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7643 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7644 "because it cannot be guaranteed that doing so would not "
7645 "lead to an abrupt change of visible data",
7646 node_name, parent_bs->node_name);
7647 to_replace_bs = NULL;
7648 goto out;
7651 out:
7652 aio_context_release(aio_context);
7653 return to_replace_bs;
7657 * Iterates through the list of runtime option keys that are said to
7658 * be "strong" for a BDS. An option is called "strong" if it changes
7659 * a BDS's data. For example, the null block driver's "size" and
7660 * "read-zeroes" options are strong, but its "latency-ns" option is
7661 * not.
7663 * If a key returned by this function ends with a dot, all options
7664 * starting with that prefix are strong.
7666 static const char *const *strong_options(BlockDriverState *bs,
7667 const char *const *curopt)
7669 static const char *const global_options[] = {
7670 "driver", "filename", NULL
7673 if (!curopt) {
7674 return &global_options[0];
7677 curopt++;
7678 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7679 curopt = bs->drv->strong_runtime_opts;
7682 return (curopt && *curopt) ? curopt : NULL;
7686 * Copies all strong runtime options from bs->options to the given
7687 * QDict. The set of strong option keys is determined by invoking
7688 * strong_options().
7690 * Returns true iff any strong option was present in bs->options (and
7691 * thus copied to the target QDict) with the exception of "filename"
7692 * and "driver". The caller is expected to use this value to decide
7693 * whether the existence of strong options prevents the generation of
7694 * a plain filename.
7696 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7698 bool found_any = false;
7699 const char *const *option_name = NULL;
7701 if (!bs->drv) {
7702 return false;
7705 while ((option_name = strong_options(bs, option_name))) {
7706 bool option_given = false;
7708 assert(strlen(*option_name) > 0);
7709 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7710 QObject *entry = qdict_get(bs->options, *option_name);
7711 if (!entry) {
7712 continue;
7715 qdict_put_obj(d, *option_name, qobject_ref(entry));
7716 option_given = true;
7717 } else {
7718 const QDictEntry *entry;
7719 for (entry = qdict_first(bs->options); entry;
7720 entry = qdict_next(bs->options, entry))
7722 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7723 qdict_put_obj(d, qdict_entry_key(entry),
7724 qobject_ref(qdict_entry_value(entry)));
7725 option_given = true;
7730 /* While "driver" and "filename" need to be included in a JSON filename,
7731 * their existence does not prohibit generation of a plain filename. */
7732 if (!found_any && option_given &&
7733 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7735 found_any = true;
7739 if (!qdict_haskey(d, "driver")) {
7740 /* Drivers created with bdrv_new_open_driver() may not have a
7741 * @driver option. Add it here. */
7742 qdict_put_str(d, "driver", bs->drv->format_name);
7745 return found_any;
7748 /* Note: This function may return false positives; it may return true
7749 * even if opening the backing file specified by bs's image header
7750 * would result in exactly bs->backing. */
7751 static bool bdrv_backing_overridden(BlockDriverState *bs)
7753 GLOBAL_STATE_CODE();
7754 if (bs->backing) {
7755 return strcmp(bs->auto_backing_file,
7756 bs->backing->bs->filename);
7757 } else {
7758 /* No backing BDS, so if the image header reports any backing
7759 * file, it must have been suppressed */
7760 return bs->auto_backing_file[0] != '\0';
7764 /* Updates the following BDS fields:
7765 * - exact_filename: A filename which may be used for opening a block device
7766 * which (mostly) equals the given BDS (even without any
7767 * other options; so reading and writing must return the same
7768 * results, but caching etc. may be different)
7769 * - full_open_options: Options which, when given when opening a block device
7770 * (without a filename), result in a BDS (mostly)
7771 * equalling the given one
7772 * - filename: If exact_filename is set, it is copied here. Otherwise,
7773 * full_open_options is converted to a JSON object, prefixed with
7774 * "json:" (for use through the JSON pseudo protocol) and put here.
7776 void bdrv_refresh_filename(BlockDriverState *bs)
7778 BlockDriver *drv = bs->drv;
7779 BdrvChild *child;
7780 BlockDriverState *primary_child_bs;
7781 QDict *opts;
7782 bool backing_overridden;
7783 bool generate_json_filename; /* Whether our default implementation should
7784 fill exact_filename (false) or not (true) */
7786 GLOBAL_STATE_CODE();
7788 if (!drv) {
7789 return;
7792 /* This BDS's file name may depend on any of its children's file names, so
7793 * refresh those first */
7794 QLIST_FOREACH(child, &bs->children, next) {
7795 bdrv_refresh_filename(child->bs);
7798 if (bs->implicit) {
7799 /* For implicit nodes, just copy everything from the single child */
7800 child = QLIST_FIRST(&bs->children);
7801 assert(QLIST_NEXT(child, next) == NULL);
7803 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7804 child->bs->exact_filename);
7805 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7807 qobject_unref(bs->full_open_options);
7808 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7810 return;
7813 backing_overridden = bdrv_backing_overridden(bs);
7815 if (bs->open_flags & BDRV_O_NO_IO) {
7816 /* Without I/O, the backing file does not change anything.
7817 * Therefore, in such a case (primarily qemu-img), we can
7818 * pretend the backing file has not been overridden even if
7819 * it technically has been. */
7820 backing_overridden = false;
7823 /* Gather the options QDict */
7824 opts = qdict_new();
7825 generate_json_filename = append_strong_runtime_options(opts, bs);
7826 generate_json_filename |= backing_overridden;
7828 if (drv->bdrv_gather_child_options) {
7829 /* Some block drivers may not want to present all of their children's
7830 * options, or name them differently from BdrvChild.name */
7831 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7832 } else {
7833 QLIST_FOREACH(child, &bs->children, next) {
7834 if (child == bs->backing && !backing_overridden) {
7835 /* We can skip the backing BDS if it has not been overridden */
7836 continue;
7839 qdict_put(opts, child->name,
7840 qobject_ref(child->bs->full_open_options));
7843 if (backing_overridden && !bs->backing) {
7844 /* Force no backing file */
7845 qdict_put_null(opts, "backing");
7849 qobject_unref(bs->full_open_options);
7850 bs->full_open_options = opts;
7852 primary_child_bs = bdrv_primary_bs(bs);
7854 if (drv->bdrv_refresh_filename) {
7855 /* Obsolete information is of no use here, so drop the old file name
7856 * information before refreshing it */
7857 bs->exact_filename[0] = '\0';
7859 drv->bdrv_refresh_filename(bs);
7860 } else if (primary_child_bs) {
7862 * Try to reconstruct valid information from the underlying
7863 * file -- this only works for format nodes (filter nodes
7864 * cannot be probed and as such must be selected by the user
7865 * either through an options dict, or through a special
7866 * filename which the filter driver must construct in its
7867 * .bdrv_refresh_filename() implementation).
7870 bs->exact_filename[0] = '\0';
7873 * We can use the underlying file's filename if:
7874 * - it has a filename,
7875 * - the current BDS is not a filter,
7876 * - the file is a protocol BDS, and
7877 * - opening that file (as this BDS's format) will automatically create
7878 * the BDS tree we have right now, that is:
7879 * - the user did not significantly change this BDS's behavior with
7880 * some explicit (strong) options
7881 * - no non-file child of this BDS has been overridden by the user
7882 * Both of these conditions are represented by generate_json_filename.
7884 if (primary_child_bs->exact_filename[0] &&
7885 primary_child_bs->drv->bdrv_file_open &&
7886 !drv->is_filter && !generate_json_filename)
7888 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
7892 if (bs->exact_filename[0]) {
7893 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
7894 } else {
7895 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
7896 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7897 json->str) >= sizeof(bs->filename)) {
7898 /* Give user a hint if we truncated things. */
7899 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7901 g_string_free(json, true);
7905 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7907 BlockDriver *drv = bs->drv;
7908 BlockDriverState *child_bs;
7910 GLOBAL_STATE_CODE();
7912 if (!drv) {
7913 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7914 return NULL;
7917 if (drv->bdrv_dirname) {
7918 return drv->bdrv_dirname(bs, errp);
7921 child_bs = bdrv_primary_bs(bs);
7922 if (child_bs) {
7923 return bdrv_dirname(child_bs, errp);
7926 bdrv_refresh_filename(bs);
7927 if (bs->exact_filename[0] != '\0') {
7928 return path_combine(bs->exact_filename, "");
7931 error_setg(errp, "Cannot generate a base directory for %s nodes",
7932 drv->format_name);
7933 return NULL;
7937 * Hot add/remove a BDS's child. So the user can take a child offline when
7938 * it is broken and take a new child online
7940 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7941 Error **errp)
7943 GLOBAL_STATE_CODE();
7944 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7945 error_setg(errp, "The node %s does not support adding a child",
7946 bdrv_get_device_or_node_name(parent_bs));
7947 return;
7950 if (!QLIST_EMPTY(&child_bs->parents)) {
7951 error_setg(errp, "The node %s already has a parent",
7952 child_bs->node_name);
7953 return;
7956 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7959 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7961 BdrvChild *tmp;
7963 GLOBAL_STATE_CODE();
7964 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7965 error_setg(errp, "The node %s does not support removing a child",
7966 bdrv_get_device_or_node_name(parent_bs));
7967 return;
7970 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7971 if (tmp == child) {
7972 break;
7976 if (!tmp) {
7977 error_setg(errp, "The node %s does not have a child named %s",
7978 bdrv_get_device_or_node_name(parent_bs),
7979 bdrv_get_device_or_node_name(child->bs));
7980 return;
7983 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7986 int bdrv_make_empty(BdrvChild *c, Error **errp)
7988 BlockDriver *drv = c->bs->drv;
7989 int ret;
7991 GLOBAL_STATE_CODE();
7992 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7994 if (!drv->bdrv_make_empty) {
7995 error_setg(errp, "%s does not support emptying nodes",
7996 drv->format_name);
7997 return -ENOTSUP;
8000 ret = drv->bdrv_make_empty(c->bs);
8001 if (ret < 0) {
8002 error_setg_errno(errp, -ret, "Failed to empty %s",
8003 c->bs->filename);
8004 return ret;
8007 return 0;
8011 * Return the child that @bs acts as an overlay for, and from which data may be
8012 * copied in COW or COR operations. Usually this is the backing file.
8014 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8016 IO_CODE();
8018 if (!bs || !bs->drv) {
8019 return NULL;
8022 if (bs->drv->is_filter) {
8023 return NULL;
8026 if (!bs->backing) {
8027 return NULL;
8030 assert(bs->backing->role & BDRV_CHILD_COW);
8031 return bs->backing;
8035 * If @bs acts as a filter for exactly one of its children, return
8036 * that child.
8038 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8040 BdrvChild *c;
8041 IO_CODE();
8043 if (!bs || !bs->drv) {
8044 return NULL;
8047 if (!bs->drv->is_filter) {
8048 return NULL;
8051 /* Only one of @backing or @file may be used */
8052 assert(!(bs->backing && bs->file));
8054 c = bs->backing ?: bs->file;
8055 if (!c) {
8056 return NULL;
8059 assert(c->role & BDRV_CHILD_FILTERED);
8060 return c;
8064 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8065 * whichever is non-NULL.
8067 * Return NULL if both are NULL.
8069 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8071 BdrvChild *cow_child = bdrv_cow_child(bs);
8072 BdrvChild *filter_child = bdrv_filter_child(bs);
8073 IO_CODE();
8075 /* Filter nodes cannot have COW backing files */
8076 assert(!(cow_child && filter_child));
8078 return cow_child ?: filter_child;
8082 * Return the primary child of this node: For filters, that is the
8083 * filtered child. For other nodes, that is usually the child storing
8084 * metadata.
8085 * (A generally more helpful description is that this is (usually) the
8086 * child that has the same filename as @bs.)
8088 * Drivers do not necessarily have a primary child; for example quorum
8089 * does not.
8091 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8093 BdrvChild *c, *found = NULL;
8094 IO_CODE();
8096 QLIST_FOREACH(c, &bs->children, next) {
8097 if (c->role & BDRV_CHILD_PRIMARY) {
8098 assert(!found);
8099 found = c;
8103 return found;
8106 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8107 bool stop_on_explicit_filter)
8109 BdrvChild *c;
8111 if (!bs) {
8112 return NULL;
8115 while (!(stop_on_explicit_filter && !bs->implicit)) {
8116 c = bdrv_filter_child(bs);
8117 if (!c) {
8119 * A filter that is embedded in a working block graph must
8120 * have a child. Assert this here so this function does
8121 * not return a filter node that is not expected by the
8122 * caller.
8124 assert(!bs->drv || !bs->drv->is_filter);
8125 break;
8127 bs = c->bs;
8130 * Note that this treats nodes with bs->drv == NULL as not being
8131 * filters (bs->drv == NULL should be replaced by something else
8132 * anyway).
8133 * The advantage of this behavior is that this function will thus
8134 * always return a non-NULL value (given a non-NULL @bs).
8137 return bs;
8141 * Return the first BDS that has not been added implicitly or that
8142 * does not have a filtered child down the chain starting from @bs
8143 * (including @bs itself).
8145 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8147 GLOBAL_STATE_CODE();
8148 return bdrv_do_skip_filters(bs, true);
8152 * Return the first BDS that does not have a filtered child down the
8153 * chain starting from @bs (including @bs itself).
8155 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8157 IO_CODE();
8158 return bdrv_do_skip_filters(bs, false);
8162 * For a backing chain, return the first non-filter backing image of
8163 * the first non-filter image.
8165 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8167 IO_CODE();
8168 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8172 * Check whether [offset, offset + bytes) overlaps with the cached
8173 * block-status data region.
8175 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8176 * which is what bdrv_bsc_is_data()'s interface needs.
8177 * Otherwise, *pnum is not touched.
8179 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8180 int64_t offset, int64_t bytes,
8181 int64_t *pnum)
8183 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8184 bool overlaps;
8186 overlaps =
8187 qatomic_read(&bsc->valid) &&
8188 ranges_overlap(offset, bytes, bsc->data_start,
8189 bsc->data_end - bsc->data_start);
8191 if (overlaps && pnum) {
8192 *pnum = bsc->data_end - offset;
8195 return overlaps;
8199 * See block_int.h for this function's documentation.
8201 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8203 IO_CODE();
8204 RCU_READ_LOCK_GUARD();
8205 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8209 * See block_int.h for this function's documentation.
8211 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8212 int64_t offset, int64_t bytes)
8214 IO_CODE();
8215 RCU_READ_LOCK_GUARD();
8217 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8218 qatomic_set(&bs->block_status_cache->valid, false);
8223 * See block_int.h for this function's documentation.
8225 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8227 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8228 BdrvBlockStatusCache *old_bsc;
8229 IO_CODE();
8231 *new_bsc = (BdrvBlockStatusCache) {
8232 .valid = true,
8233 .data_start = offset,
8234 .data_end = offset + bytes,
8237 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8239 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8240 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8241 if (old_bsc) {
8242 g_free_rcu(old_bsc, rcu);