configure: Compile with -Wimplicit-fallthrough=2
[qemu/ar7.git] / block.c
blob8f177504d4966ff882bc6c710de2a7e635938f6f
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/fuse.h"
30 #include "block/nbd.h"
31 #include "block/qdict.h"
32 #include "qemu/error-report.h"
33 #include "block/module_block.h"
34 #include "qemu/main-loop.h"
35 #include "qemu/module.h"
36 #include "qapi/error.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qjson.h"
39 #include "qapi/qmp/qnull.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qapi/qobject-output-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "sysemu/block-backend.h"
44 #include "sysemu/sysemu.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 "block/coroutines.h"
54 #ifdef CONFIG_BSD
55 #include <sys/ioctl.h>
56 #include <sys/queue.h>
57 #ifndef __DragonFly__
58 #include <sys/disk.h>
59 #endif
60 #endif
62 #ifdef _WIN32
63 #include <windows.h>
64 #endif
66 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
68 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
71 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
72 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
74 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
75 QLIST_HEAD_INITIALIZER(bdrv_drivers);
77 static BlockDriverState *bdrv_open_inherit(const char *filename,
78 const char *reference,
79 QDict *options, int flags,
80 BlockDriverState *parent,
81 const BdrvChildClass *child_class,
82 BdrvChildRole child_role,
83 Error **errp);
85 /* If non-zero, use only whitelisted block drivers */
86 static int use_bdrv_whitelist;
88 #ifdef _WIN32
89 static int is_windows_drive_prefix(const char *filename)
91 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
92 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
93 filename[1] == ':');
96 int is_windows_drive(const char *filename)
98 if (is_windows_drive_prefix(filename) &&
99 filename[2] == '\0')
100 return 1;
101 if (strstart(filename, "\\\\.\\", NULL) ||
102 strstart(filename, "//./", NULL))
103 return 1;
104 return 0;
106 #endif
108 size_t bdrv_opt_mem_align(BlockDriverState *bs)
110 if (!bs || !bs->drv) {
111 /* page size or 4k (hdd sector size) should be on the safe side */
112 return MAX(4096, qemu_real_host_page_size);
115 return bs->bl.opt_mem_alignment;
118 size_t bdrv_min_mem_align(BlockDriverState *bs)
120 if (!bs || !bs->drv) {
121 /* page size or 4k (hdd sector size) should be on the safe side */
122 return MAX(4096, qemu_real_host_page_size);
125 return bs->bl.min_mem_alignment;
128 /* check if the path starts with "<protocol>:" */
129 int path_has_protocol(const char *path)
131 const char *p;
133 #ifdef _WIN32
134 if (is_windows_drive(path) ||
135 is_windows_drive_prefix(path)) {
136 return 0;
138 p = path + strcspn(path, ":/\\");
139 #else
140 p = path + strcspn(path, ":/");
141 #endif
143 return *p == ':';
146 int path_is_absolute(const char *path)
148 #ifdef _WIN32
149 /* specific case for names like: "\\.\d:" */
150 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
151 return 1;
153 return (*path == '/' || *path == '\\');
154 #else
155 return (*path == '/');
156 #endif
159 /* if filename is absolute, just return its duplicate. Otherwise, build a
160 path to it by considering it is relative to base_path. URL are
161 supported. */
162 char *path_combine(const char *base_path, const char *filename)
164 const char *protocol_stripped = NULL;
165 const char *p, *p1;
166 char *result;
167 int len;
169 if (path_is_absolute(filename)) {
170 return g_strdup(filename);
173 if (path_has_protocol(base_path)) {
174 protocol_stripped = strchr(base_path, ':');
175 if (protocol_stripped) {
176 protocol_stripped++;
179 p = protocol_stripped ?: base_path;
181 p1 = strrchr(base_path, '/');
182 #ifdef _WIN32
184 const char *p2;
185 p2 = strrchr(base_path, '\\');
186 if (!p1 || p2 > p1) {
187 p1 = p2;
190 #endif
191 if (p1) {
192 p1++;
193 } else {
194 p1 = base_path;
196 if (p1 > p) {
197 p = p1;
199 len = p - base_path;
201 result = g_malloc(len + strlen(filename) + 1);
202 memcpy(result, base_path, len);
203 strcpy(result + len, filename);
205 return result;
209 * Helper function for bdrv_parse_filename() implementations to remove optional
210 * protocol prefixes (especially "file:") from a filename and for putting the
211 * stripped filename into the options QDict if there is such a prefix.
213 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
214 QDict *options)
216 if (strstart(filename, prefix, &filename)) {
217 /* Stripping the explicit protocol prefix may result in a protocol
218 * prefix being (wrongly) detected (if the filename contains a colon) */
219 if (path_has_protocol(filename)) {
220 QString *fat_filename;
222 /* This means there is some colon before the first slash; therefore,
223 * this cannot be an absolute path */
224 assert(!path_is_absolute(filename));
226 /* And we can thus fix the protocol detection issue by prefixing it
227 * by "./" */
228 fat_filename = qstring_from_str("./");
229 qstring_append(fat_filename, filename);
231 assert(!path_has_protocol(qstring_get_str(fat_filename)));
233 qdict_put(options, "filename", fat_filename);
234 } else {
235 /* If no protocol prefix was detected, we can use the shortened
236 * filename as-is */
237 qdict_put_str(options, "filename", filename);
243 /* Returns whether the image file is opened as read-only. Note that this can
244 * return false and writing to the image file is still not possible because the
245 * image is inactivated. */
246 bool bdrv_is_read_only(BlockDriverState *bs)
248 return bs->read_only;
251 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
252 bool ignore_allow_rdw, Error **errp)
254 /* Do not set read_only if copy_on_read is enabled */
255 if (bs->copy_on_read && read_only) {
256 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
257 bdrv_get_device_or_node_name(bs));
258 return -EINVAL;
261 /* Do not clear read_only if it is prohibited */
262 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
263 !ignore_allow_rdw)
265 error_setg(errp, "Node '%s' is read only",
266 bdrv_get_device_or_node_name(bs));
267 return -EPERM;
270 return 0;
274 * Called by a driver that can only provide a read-only image.
276 * Returns 0 if the node is already read-only or it could switch the node to
277 * read-only because BDRV_O_AUTO_RDONLY is set.
279 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
280 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
281 * is not NULL, it is used as the error message for the Error object.
283 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
284 Error **errp)
286 int ret = 0;
288 if (!(bs->open_flags & BDRV_O_RDWR)) {
289 return 0;
291 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
292 goto fail;
295 ret = bdrv_can_set_read_only(bs, true, false, NULL);
296 if (ret < 0) {
297 goto fail;
300 bs->read_only = true;
301 bs->open_flags &= ~BDRV_O_RDWR;
303 return 0;
305 fail:
306 error_setg(errp, "%s", errmsg ?: "Image is read-only");
307 return -EACCES;
311 * If @backing is empty, this function returns NULL without setting
312 * @errp. In all other cases, NULL will only be returned with @errp
313 * set.
315 * Therefore, a return value of NULL without @errp set means that
316 * there is no backing file; if @errp is set, there is one but its
317 * absolute filename cannot be generated.
319 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
320 const char *backing,
321 Error **errp)
323 if (backing[0] == '\0') {
324 return NULL;
325 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
326 return g_strdup(backing);
327 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
328 error_setg(errp, "Cannot use relative backing file names for '%s'",
329 backed);
330 return NULL;
331 } else {
332 return path_combine(backed, backing);
337 * If @filename is empty or NULL, this function returns NULL without
338 * setting @errp. In all other cases, NULL will only be returned with
339 * @errp set.
341 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
342 const char *filename, Error **errp)
344 char *dir, *full_name;
346 if (!filename || filename[0] == '\0') {
347 return NULL;
348 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
349 return g_strdup(filename);
352 dir = bdrv_dirname(relative_to, errp);
353 if (!dir) {
354 return NULL;
357 full_name = g_strconcat(dir, filename, NULL);
358 g_free(dir);
359 return full_name;
362 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
364 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
367 void bdrv_register(BlockDriver *bdrv)
369 assert(bdrv->format_name);
370 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
373 BlockDriverState *bdrv_new(void)
375 BlockDriverState *bs;
376 int i;
378 bs = g_new0(BlockDriverState, 1);
379 QLIST_INIT(&bs->dirty_bitmaps);
380 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
381 QLIST_INIT(&bs->op_blockers[i]);
383 notifier_with_return_list_init(&bs->before_write_notifiers);
384 qemu_co_mutex_init(&bs->reqs_lock);
385 qemu_mutex_init(&bs->dirty_bitmap_mutex);
386 bs->refcnt = 1;
387 bs->aio_context = qemu_get_aio_context();
389 qemu_co_queue_init(&bs->flush_queue);
391 for (i = 0; i < bdrv_drain_all_count; i++) {
392 bdrv_drained_begin(bs);
395 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
397 return bs;
400 static BlockDriver *bdrv_do_find_format(const char *format_name)
402 BlockDriver *drv1;
404 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
405 if (!strcmp(drv1->format_name, format_name)) {
406 return drv1;
410 return NULL;
413 BlockDriver *bdrv_find_format(const char *format_name)
415 BlockDriver *drv1;
416 int i;
418 drv1 = bdrv_do_find_format(format_name);
419 if (drv1) {
420 return drv1;
423 /* The driver isn't registered, maybe we need to load a module */
424 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
425 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
426 block_module_load_one(block_driver_modules[i].library_name);
427 break;
431 return bdrv_do_find_format(format_name);
434 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
436 static const char *whitelist_rw[] = {
437 CONFIG_BDRV_RW_WHITELIST
438 NULL
440 static const char *whitelist_ro[] = {
441 CONFIG_BDRV_RO_WHITELIST
442 NULL
444 const char **p;
446 if (!whitelist_rw[0] && !whitelist_ro[0]) {
447 return 1; /* no whitelist, anything goes */
450 for (p = whitelist_rw; *p; p++) {
451 if (!strcmp(format_name, *p)) {
452 return 1;
455 if (read_only) {
456 for (p = whitelist_ro; *p; p++) {
457 if (!strcmp(format_name, *p)) {
458 return 1;
462 return 0;
465 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
467 return bdrv_format_is_whitelisted(drv->format_name, read_only);
470 bool bdrv_uses_whitelist(void)
472 return use_bdrv_whitelist;
475 typedef struct CreateCo {
476 BlockDriver *drv;
477 char *filename;
478 QemuOpts *opts;
479 int ret;
480 Error *err;
481 } CreateCo;
483 static void coroutine_fn bdrv_create_co_entry(void *opaque)
485 Error *local_err = NULL;
486 int ret;
488 CreateCo *cco = opaque;
489 assert(cco->drv);
491 ret = cco->drv->bdrv_co_create_opts(cco->drv,
492 cco->filename, cco->opts, &local_err);
493 error_propagate(&cco->err, local_err);
494 cco->ret = ret;
497 int bdrv_create(BlockDriver *drv, const char* filename,
498 QemuOpts *opts, Error **errp)
500 int ret;
502 Coroutine *co;
503 CreateCo cco = {
504 .drv = drv,
505 .filename = g_strdup(filename),
506 .opts = opts,
507 .ret = NOT_DONE,
508 .err = NULL,
511 if (!drv->bdrv_co_create_opts) {
512 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
513 ret = -ENOTSUP;
514 goto out;
517 if (qemu_in_coroutine()) {
518 /* Fast-path if already in coroutine context */
519 bdrv_create_co_entry(&cco);
520 } else {
521 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
522 qemu_coroutine_enter(co);
523 while (cco.ret == NOT_DONE) {
524 aio_poll(qemu_get_aio_context(), true);
528 ret = cco.ret;
529 if (ret < 0) {
530 if (cco.err) {
531 error_propagate(errp, cco.err);
532 } else {
533 error_setg_errno(errp, -ret, "Could not create image");
537 out:
538 g_free(cco.filename);
539 return ret;
543 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
544 * least the given @minimum_size.
546 * On success, return @blk's actual length.
547 * Otherwise, return -errno.
549 static int64_t create_file_fallback_truncate(BlockBackend *blk,
550 int64_t minimum_size, Error **errp)
552 Error *local_err = NULL;
553 int64_t size;
554 int ret;
556 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
557 &local_err);
558 if (ret < 0 && ret != -ENOTSUP) {
559 error_propagate(errp, local_err);
560 return ret;
563 size = blk_getlength(blk);
564 if (size < 0) {
565 error_free(local_err);
566 error_setg_errno(errp, -size,
567 "Failed to inquire the new image file's length");
568 return size;
571 if (size < minimum_size) {
572 /* Need to grow the image, but we failed to do that */
573 error_propagate(errp, local_err);
574 return -ENOTSUP;
577 error_free(local_err);
578 local_err = NULL;
580 return size;
584 * Helper function for bdrv_create_file_fallback(): Zero the first
585 * sector to remove any potentially pre-existing image header.
587 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
588 int64_t current_size,
589 Error **errp)
591 int64_t bytes_to_clear;
592 int ret;
594 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
595 if (bytes_to_clear) {
596 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
597 if (ret < 0) {
598 error_setg_errno(errp, -ret,
599 "Failed to clear the new image's first sector");
600 return ret;
604 return 0;
608 * Simple implementation of bdrv_co_create_opts for protocol drivers
609 * which only support creation via opening a file
610 * (usually existing raw storage device)
612 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
613 const char *filename,
614 QemuOpts *opts,
615 Error **errp)
617 BlockBackend *blk;
618 QDict *options;
619 int64_t size = 0;
620 char *buf = NULL;
621 PreallocMode prealloc;
622 Error *local_err = NULL;
623 int ret;
625 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
626 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
627 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
628 PREALLOC_MODE_OFF, &local_err);
629 g_free(buf);
630 if (local_err) {
631 error_propagate(errp, local_err);
632 return -EINVAL;
635 if (prealloc != PREALLOC_MODE_OFF) {
636 error_setg(errp, "Unsupported preallocation mode '%s'",
637 PreallocMode_str(prealloc));
638 return -ENOTSUP;
641 options = qdict_new();
642 qdict_put_str(options, "driver", drv->format_name);
644 blk = blk_new_open(filename, NULL, options,
645 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
646 if (!blk) {
647 error_prepend(errp, "Protocol driver '%s' does not support image "
648 "creation, and opening the image failed: ",
649 drv->format_name);
650 return -EINVAL;
653 size = create_file_fallback_truncate(blk, size, errp);
654 if (size < 0) {
655 ret = size;
656 goto out;
659 ret = create_file_fallback_zero_first_sector(blk, size, errp);
660 if (ret < 0) {
661 goto out;
664 ret = 0;
665 out:
666 blk_unref(blk);
667 return ret;
670 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
672 BlockDriver *drv;
674 drv = bdrv_find_protocol(filename, true, errp);
675 if (drv == NULL) {
676 return -ENOENT;
679 return bdrv_create(drv, filename, opts, errp);
682 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
684 Error *local_err = NULL;
685 int ret;
687 assert(bs != NULL);
689 if (!bs->drv) {
690 error_setg(errp, "Block node '%s' is not opened", bs->filename);
691 return -ENOMEDIUM;
694 if (!bs->drv->bdrv_co_delete_file) {
695 error_setg(errp, "Driver '%s' does not support image deletion",
696 bs->drv->format_name);
697 return -ENOTSUP;
700 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
701 if (ret < 0) {
702 error_propagate(errp, local_err);
705 return ret;
709 * Try to get @bs's logical and physical block size.
710 * On success, store them in @bsz struct and return 0.
711 * On failure return -errno.
712 * @bs must not be empty.
714 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
716 BlockDriver *drv = bs->drv;
717 BlockDriverState *filtered = bdrv_filter_bs(bs);
719 if (drv && drv->bdrv_probe_blocksizes) {
720 return drv->bdrv_probe_blocksizes(bs, bsz);
721 } else if (filtered) {
722 return bdrv_probe_blocksizes(filtered, bsz);
725 return -ENOTSUP;
729 * Try to get @bs's geometry (cyls, heads, sectors).
730 * On success, store them in @geo struct and return 0.
731 * On failure return -errno.
732 * @bs must not be empty.
734 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
736 BlockDriver *drv = bs->drv;
737 BlockDriverState *filtered = bdrv_filter_bs(bs);
739 if (drv && drv->bdrv_probe_geometry) {
740 return drv->bdrv_probe_geometry(bs, geo);
741 } else if (filtered) {
742 return bdrv_probe_geometry(filtered, geo);
745 return -ENOTSUP;
749 * Create a uniquely-named empty temporary file.
750 * Return 0 upon success, otherwise a negative errno value.
752 int get_tmp_filename(char *filename, int size)
754 #ifdef _WIN32
755 char temp_dir[MAX_PATH];
756 /* GetTempFileName requires that its output buffer (4th param)
757 have length MAX_PATH or greater. */
758 assert(size >= MAX_PATH);
759 return (GetTempPath(MAX_PATH, temp_dir)
760 && GetTempFileName(temp_dir, "qem", 0, filename)
761 ? 0 : -GetLastError());
762 #else
763 int fd;
764 const char *tmpdir;
765 tmpdir = getenv("TMPDIR");
766 if (!tmpdir) {
767 tmpdir = "/var/tmp";
769 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
770 return -EOVERFLOW;
772 fd = mkstemp(filename);
773 if (fd < 0) {
774 return -errno;
776 if (close(fd) != 0) {
777 unlink(filename);
778 return -errno;
780 return 0;
781 #endif
785 * Detect host devices. By convention, /dev/cdrom[N] is always
786 * recognized as a host CDROM.
788 static BlockDriver *find_hdev_driver(const char *filename)
790 int score_max = 0, score;
791 BlockDriver *drv = NULL, *d;
793 QLIST_FOREACH(d, &bdrv_drivers, list) {
794 if (d->bdrv_probe_device) {
795 score = d->bdrv_probe_device(filename);
796 if (score > score_max) {
797 score_max = score;
798 drv = d;
803 return drv;
806 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
808 BlockDriver *drv1;
810 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
811 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
812 return drv1;
816 return NULL;
819 BlockDriver *bdrv_find_protocol(const char *filename,
820 bool allow_protocol_prefix,
821 Error **errp)
823 BlockDriver *drv1;
824 char protocol[128];
825 int len;
826 const char *p;
827 int i;
829 /* TODO Drivers without bdrv_file_open must be specified explicitly */
832 * XXX(hch): we really should not let host device detection
833 * override an explicit protocol specification, but moving this
834 * later breaks access to device names with colons in them.
835 * Thanks to the brain-dead persistent naming schemes on udev-
836 * based Linux systems those actually are quite common.
838 drv1 = find_hdev_driver(filename);
839 if (drv1) {
840 return drv1;
843 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
844 return &bdrv_file;
847 p = strchr(filename, ':');
848 assert(p != NULL);
849 len = p - filename;
850 if (len > sizeof(protocol) - 1)
851 len = sizeof(protocol) - 1;
852 memcpy(protocol, filename, len);
853 protocol[len] = '\0';
855 drv1 = bdrv_do_find_protocol(protocol);
856 if (drv1) {
857 return drv1;
860 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
861 if (block_driver_modules[i].protocol_name &&
862 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
863 block_module_load_one(block_driver_modules[i].library_name);
864 break;
868 drv1 = bdrv_do_find_protocol(protocol);
869 if (!drv1) {
870 error_setg(errp, "Unknown protocol '%s'", protocol);
872 return drv1;
876 * Guess image format by probing its contents.
877 * This is not a good idea when your image is raw (CVE-2008-2004), but
878 * we do it anyway for backward compatibility.
880 * @buf contains the image's first @buf_size bytes.
881 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
882 * but can be smaller if the image file is smaller)
883 * @filename is its filename.
885 * For all block drivers, call the bdrv_probe() method to get its
886 * probing score.
887 * Return the first block driver with the highest probing score.
889 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
890 const char *filename)
892 int score_max = 0, score;
893 BlockDriver *drv = NULL, *d;
895 QLIST_FOREACH(d, &bdrv_drivers, list) {
896 if (d->bdrv_probe) {
897 score = d->bdrv_probe(buf, buf_size, filename);
898 if (score > score_max) {
899 score_max = score;
900 drv = d;
905 return drv;
908 static int find_image_format(BlockBackend *file, const char *filename,
909 BlockDriver **pdrv, Error **errp)
911 BlockDriver *drv;
912 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
913 int ret = 0;
915 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
916 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
917 *pdrv = &bdrv_raw;
918 return ret;
921 ret = blk_pread(file, 0, buf, sizeof(buf));
922 if (ret < 0) {
923 error_setg_errno(errp, -ret, "Could not read image for determining its "
924 "format");
925 *pdrv = NULL;
926 return ret;
929 drv = bdrv_probe_all(buf, ret, filename);
930 if (!drv) {
931 error_setg(errp, "Could not determine image format: No compatible "
932 "driver found");
933 ret = -ENOENT;
935 *pdrv = drv;
936 return ret;
940 * Set the current 'total_sectors' value
941 * Return 0 on success, -errno on error.
943 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
945 BlockDriver *drv = bs->drv;
947 if (!drv) {
948 return -ENOMEDIUM;
951 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
952 if (bdrv_is_sg(bs))
953 return 0;
955 /* query actual device if possible, otherwise just trust the hint */
956 if (drv->bdrv_getlength) {
957 int64_t length = drv->bdrv_getlength(bs);
958 if (length < 0) {
959 return length;
961 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
964 bs->total_sectors = hint;
966 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
967 return -EFBIG;
970 return 0;
974 * Combines a QDict of new block driver @options with any missing options taken
975 * from @old_options, so that leaving out an option defaults to its old value.
977 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
978 QDict *old_options)
980 if (bs->drv && bs->drv->bdrv_join_options) {
981 bs->drv->bdrv_join_options(options, old_options);
982 } else {
983 qdict_join(options, old_options, false);
987 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
988 int open_flags,
989 Error **errp)
991 Error *local_err = NULL;
992 char *value = qemu_opt_get_del(opts, "detect-zeroes");
993 BlockdevDetectZeroesOptions detect_zeroes =
994 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
995 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
996 g_free(value);
997 if (local_err) {
998 error_propagate(errp, local_err);
999 return detect_zeroes;
1002 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1003 !(open_flags & BDRV_O_UNMAP))
1005 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1006 "without setting discard operation to unmap");
1009 return detect_zeroes;
1013 * Set open flags for aio engine
1015 * Return 0 on success, -1 if the engine specified is invalid
1017 int bdrv_parse_aio(const char *mode, int *flags)
1019 if (!strcmp(mode, "threads")) {
1020 /* do nothing, default */
1021 } else if (!strcmp(mode, "native")) {
1022 *flags |= BDRV_O_NATIVE_AIO;
1023 #ifdef CONFIG_LINUX_IO_URING
1024 } else if (!strcmp(mode, "io_uring")) {
1025 *flags |= BDRV_O_IO_URING;
1026 #endif
1027 } else {
1028 return -1;
1031 return 0;
1035 * Set open flags for a given discard mode
1037 * Return 0 on success, -1 if the discard mode was invalid.
1039 int bdrv_parse_discard_flags(const char *mode, int *flags)
1041 *flags &= ~BDRV_O_UNMAP;
1043 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1044 /* do nothing */
1045 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1046 *flags |= BDRV_O_UNMAP;
1047 } else {
1048 return -1;
1051 return 0;
1055 * Set open flags for a given cache mode
1057 * Return 0 on success, -1 if the cache mode was invalid.
1059 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1061 *flags &= ~BDRV_O_CACHE_MASK;
1063 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1064 *writethrough = false;
1065 *flags |= BDRV_O_NOCACHE;
1066 } else if (!strcmp(mode, "directsync")) {
1067 *writethrough = true;
1068 *flags |= BDRV_O_NOCACHE;
1069 } else if (!strcmp(mode, "writeback")) {
1070 *writethrough = false;
1071 } else if (!strcmp(mode, "unsafe")) {
1072 *writethrough = false;
1073 *flags |= BDRV_O_NO_FLUSH;
1074 } else if (!strcmp(mode, "writethrough")) {
1075 *writethrough = true;
1076 } else {
1077 return -1;
1080 return 0;
1083 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1085 BlockDriverState *parent = c->opaque;
1086 return g_strdup(bdrv_get_device_or_node_name(parent));
1089 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1091 BlockDriverState *bs = child->opaque;
1092 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1095 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1097 BlockDriverState *bs = child->opaque;
1098 return bdrv_drain_poll(bs, false, NULL, false);
1101 static void bdrv_child_cb_drained_end(BdrvChild *child,
1102 int *drained_end_counter)
1104 BlockDriverState *bs = child->opaque;
1105 bdrv_drained_end_no_poll(bs, drained_end_counter);
1108 static int bdrv_child_cb_inactivate(BdrvChild *child)
1110 BlockDriverState *bs = child->opaque;
1111 assert(bs->open_flags & BDRV_O_INACTIVE);
1112 return 0;
1115 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1116 GSList **ignore, Error **errp)
1118 BlockDriverState *bs = child->opaque;
1119 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1122 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1123 GSList **ignore)
1125 BlockDriverState *bs = child->opaque;
1126 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1130 * Returns the options and flags that a temporary snapshot should get, based on
1131 * the originally requested flags (the originally requested image will have
1132 * flags like a backing file)
1134 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1135 int parent_flags, QDict *parent_options)
1137 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1139 /* For temporary files, unconditional cache=unsafe is fine */
1140 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1141 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1143 /* Copy the read-only and discard options from the parent */
1144 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1145 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1147 /* aio=native doesn't work for cache.direct=off, so disable it for the
1148 * temporary snapshot */
1149 *child_flags &= ~BDRV_O_NATIVE_AIO;
1152 static void bdrv_backing_attach(BdrvChild *c)
1154 BlockDriverState *parent = c->opaque;
1155 BlockDriverState *backing_hd = c->bs;
1157 assert(!parent->backing_blocker);
1158 error_setg(&parent->backing_blocker,
1159 "node is used as backing hd of '%s'",
1160 bdrv_get_device_or_node_name(parent));
1162 bdrv_refresh_filename(backing_hd);
1164 parent->open_flags &= ~BDRV_O_NO_BACKING;
1166 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1167 /* Otherwise we won't be able to commit or stream */
1168 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1169 parent->backing_blocker);
1170 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1171 parent->backing_blocker);
1173 * We do backup in 3 ways:
1174 * 1. drive backup
1175 * The target bs is new opened, and the source is top BDS
1176 * 2. blockdev backup
1177 * Both the source and the target are top BDSes.
1178 * 3. internal backup(used for block replication)
1179 * Both the source and the target are backing file
1181 * In case 1 and 2, neither the source nor the target is the backing file.
1182 * In case 3, we will block the top BDS, so there is only one block job
1183 * for the top BDS and its backing chain.
1185 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1186 parent->backing_blocker);
1187 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1188 parent->backing_blocker);
1191 static void bdrv_backing_detach(BdrvChild *c)
1193 BlockDriverState *parent = c->opaque;
1195 assert(parent->backing_blocker);
1196 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1197 error_free(parent->backing_blocker);
1198 parent->backing_blocker = NULL;
1201 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1202 const char *filename, Error **errp)
1204 BlockDriverState *parent = c->opaque;
1205 bool read_only = bdrv_is_read_only(parent);
1206 int ret;
1208 if (read_only) {
1209 ret = bdrv_reopen_set_read_only(parent, false, errp);
1210 if (ret < 0) {
1211 return ret;
1215 ret = bdrv_change_backing_file(parent, filename,
1216 base->drv ? base->drv->format_name : "",
1217 false);
1218 if (ret < 0) {
1219 error_setg_errno(errp, -ret, "Could not update backing file link");
1222 if (read_only) {
1223 bdrv_reopen_set_read_only(parent, true, NULL);
1226 return ret;
1230 * Returns the options and flags that a generic child of a BDS should
1231 * get, based on the given options and flags for the parent BDS.
1233 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1234 int *child_flags, QDict *child_options,
1235 int parent_flags, QDict *parent_options)
1237 int flags = parent_flags;
1240 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1241 * Generally, the question to answer is: Should this child be
1242 * format-probed by default?
1246 * Pure and non-filtered data children of non-format nodes should
1247 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1248 * set). This only affects a very limited set of drivers (namely
1249 * quorum and blkverify when this comment was written).
1250 * Force-clear BDRV_O_PROTOCOL then.
1252 if (!parent_is_format &&
1253 (role & BDRV_CHILD_DATA) &&
1254 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1256 flags &= ~BDRV_O_PROTOCOL;
1260 * All children of format nodes (except for COW children) and all
1261 * metadata children in general should never be format-probed.
1262 * Force-set BDRV_O_PROTOCOL then.
1264 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1265 (role & BDRV_CHILD_METADATA))
1267 flags |= BDRV_O_PROTOCOL;
1271 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1272 * the parent.
1274 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1275 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1276 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1278 if (role & BDRV_CHILD_COW) {
1279 /* backing files are opened read-only by default */
1280 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1281 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1282 } else {
1283 /* Inherit the read-only option from the parent if it's not set */
1284 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1285 qdict_copy_default(child_options, parent_options,
1286 BDRV_OPT_AUTO_READ_ONLY);
1290 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1291 * can default to enable it on lower layers regardless of the
1292 * parent option.
1294 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1296 /* Clear flags that only apply to the top layer */
1297 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1299 if (role & BDRV_CHILD_METADATA) {
1300 flags &= ~BDRV_O_NO_IO;
1302 if (role & BDRV_CHILD_COW) {
1303 flags &= ~BDRV_O_TEMPORARY;
1306 *child_flags = flags;
1309 static void bdrv_child_cb_attach(BdrvChild *child)
1311 BlockDriverState *bs = child->opaque;
1313 if (child->role & BDRV_CHILD_COW) {
1314 bdrv_backing_attach(child);
1317 bdrv_apply_subtree_drain(child, bs);
1320 static void bdrv_child_cb_detach(BdrvChild *child)
1322 BlockDriverState *bs = child->opaque;
1324 if (child->role & BDRV_CHILD_COW) {
1325 bdrv_backing_detach(child);
1328 bdrv_unapply_subtree_drain(child, bs);
1331 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1332 const char *filename, Error **errp)
1334 if (c->role & BDRV_CHILD_COW) {
1335 return bdrv_backing_update_filename(c, base, filename, errp);
1337 return 0;
1340 const BdrvChildClass child_of_bds = {
1341 .parent_is_bds = true,
1342 .get_parent_desc = bdrv_child_get_parent_desc,
1343 .inherit_options = bdrv_inherited_options,
1344 .drained_begin = bdrv_child_cb_drained_begin,
1345 .drained_poll = bdrv_child_cb_drained_poll,
1346 .drained_end = bdrv_child_cb_drained_end,
1347 .attach = bdrv_child_cb_attach,
1348 .detach = bdrv_child_cb_detach,
1349 .inactivate = bdrv_child_cb_inactivate,
1350 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1351 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1352 .update_filename = bdrv_child_cb_update_filename,
1355 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1357 int open_flags = flags;
1360 * Clear flags that are internal to the block layer before opening the
1361 * image.
1363 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1365 return open_flags;
1368 static void update_flags_from_options(int *flags, QemuOpts *opts)
1370 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1372 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1373 *flags |= BDRV_O_NO_FLUSH;
1376 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1377 *flags |= BDRV_O_NOCACHE;
1380 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1381 *flags |= BDRV_O_RDWR;
1384 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1385 *flags |= BDRV_O_AUTO_RDONLY;
1389 static void update_options_from_flags(QDict *options, int flags)
1391 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1392 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1394 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1395 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1396 flags & BDRV_O_NO_FLUSH);
1398 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1399 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1401 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1402 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1403 flags & BDRV_O_AUTO_RDONLY);
1407 static void bdrv_assign_node_name(BlockDriverState *bs,
1408 const char *node_name,
1409 Error **errp)
1411 char *gen_node_name = NULL;
1413 if (!node_name) {
1414 node_name = gen_node_name = id_generate(ID_BLOCK);
1415 } else if (!id_wellformed(node_name)) {
1417 * Check for empty string or invalid characters, but not if it is
1418 * generated (generated names use characters not available to the user)
1420 error_setg(errp, "Invalid node name");
1421 return;
1424 /* takes care of avoiding namespaces collisions */
1425 if (blk_by_name(node_name)) {
1426 error_setg(errp, "node-name=%s is conflicting with a device id",
1427 node_name);
1428 goto out;
1431 /* takes care of avoiding duplicates node names */
1432 if (bdrv_find_node(node_name)) {
1433 error_setg(errp, "Duplicate node name");
1434 goto out;
1437 /* Make sure that the node name isn't truncated */
1438 if (strlen(node_name) >= sizeof(bs->node_name)) {
1439 error_setg(errp, "Node name too long");
1440 goto out;
1443 /* copy node name into the bs and insert it into the graph list */
1444 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1445 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1446 out:
1447 g_free(gen_node_name);
1450 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1451 const char *node_name, QDict *options,
1452 int open_flags, Error **errp)
1454 Error *local_err = NULL;
1455 int i, ret;
1457 bdrv_assign_node_name(bs, node_name, &local_err);
1458 if (local_err) {
1459 error_propagate(errp, local_err);
1460 return -EINVAL;
1463 bs->drv = drv;
1464 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1465 bs->opaque = g_malloc0(drv->instance_size);
1467 if (drv->bdrv_file_open) {
1468 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1469 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1470 } else if (drv->bdrv_open) {
1471 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1472 } else {
1473 ret = 0;
1476 if (ret < 0) {
1477 if (local_err) {
1478 error_propagate(errp, local_err);
1479 } else if (bs->filename[0]) {
1480 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1481 } else {
1482 error_setg_errno(errp, -ret, "Could not open image");
1484 goto open_failed;
1487 ret = refresh_total_sectors(bs, bs->total_sectors);
1488 if (ret < 0) {
1489 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1490 return ret;
1493 bdrv_refresh_limits(bs, &local_err);
1494 if (local_err) {
1495 error_propagate(errp, local_err);
1496 return -EINVAL;
1499 assert(bdrv_opt_mem_align(bs) != 0);
1500 assert(bdrv_min_mem_align(bs) != 0);
1501 assert(is_power_of_2(bs->bl.request_alignment));
1503 for (i = 0; i < bs->quiesce_counter; i++) {
1504 if (drv->bdrv_co_drain_begin) {
1505 drv->bdrv_co_drain_begin(bs);
1509 return 0;
1510 open_failed:
1511 bs->drv = NULL;
1512 if (bs->file != NULL) {
1513 bdrv_unref_child(bs, bs->file);
1514 bs->file = NULL;
1516 g_free(bs->opaque);
1517 bs->opaque = NULL;
1518 return ret;
1521 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1522 int flags, Error **errp)
1524 BlockDriverState *bs;
1525 int ret;
1527 bs = bdrv_new();
1528 bs->open_flags = flags;
1529 bs->explicit_options = qdict_new();
1530 bs->options = qdict_new();
1531 bs->opaque = NULL;
1533 update_options_from_flags(bs->options, flags);
1535 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1536 if (ret < 0) {
1537 qobject_unref(bs->explicit_options);
1538 bs->explicit_options = NULL;
1539 qobject_unref(bs->options);
1540 bs->options = NULL;
1541 bdrv_unref(bs);
1542 return NULL;
1545 return bs;
1548 QemuOptsList bdrv_runtime_opts = {
1549 .name = "bdrv_common",
1550 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1551 .desc = {
1553 .name = "node-name",
1554 .type = QEMU_OPT_STRING,
1555 .help = "Node name of the block device node",
1558 .name = "driver",
1559 .type = QEMU_OPT_STRING,
1560 .help = "Block driver to use for the node",
1563 .name = BDRV_OPT_CACHE_DIRECT,
1564 .type = QEMU_OPT_BOOL,
1565 .help = "Bypass software writeback cache on the host",
1568 .name = BDRV_OPT_CACHE_NO_FLUSH,
1569 .type = QEMU_OPT_BOOL,
1570 .help = "Ignore flush requests",
1573 .name = BDRV_OPT_READ_ONLY,
1574 .type = QEMU_OPT_BOOL,
1575 .help = "Node is opened in read-only mode",
1578 .name = BDRV_OPT_AUTO_READ_ONLY,
1579 .type = QEMU_OPT_BOOL,
1580 .help = "Node can become read-only if opening read-write fails",
1583 .name = "detect-zeroes",
1584 .type = QEMU_OPT_STRING,
1585 .help = "try to optimize zero writes (off, on, unmap)",
1588 .name = BDRV_OPT_DISCARD,
1589 .type = QEMU_OPT_STRING,
1590 .help = "discard operation (ignore/off, unmap/on)",
1593 .name = BDRV_OPT_FORCE_SHARE,
1594 .type = QEMU_OPT_BOOL,
1595 .help = "always accept other writers (default: off)",
1597 { /* end of list */ }
1601 QemuOptsList bdrv_create_opts_simple = {
1602 .name = "simple-create-opts",
1603 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1604 .desc = {
1606 .name = BLOCK_OPT_SIZE,
1607 .type = QEMU_OPT_SIZE,
1608 .help = "Virtual disk size"
1611 .name = BLOCK_OPT_PREALLOC,
1612 .type = QEMU_OPT_STRING,
1613 .help = "Preallocation mode (allowed values: off)"
1615 { /* end of list */ }
1620 * Common part for opening disk images and files
1622 * Removes all processed options from *options.
1624 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1625 QDict *options, Error **errp)
1627 int ret, open_flags;
1628 const char *filename;
1629 const char *driver_name = NULL;
1630 const char *node_name = NULL;
1631 const char *discard;
1632 QemuOpts *opts;
1633 BlockDriver *drv;
1634 Error *local_err = NULL;
1636 assert(bs->file == NULL);
1637 assert(options != NULL && bs->options != options);
1639 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1640 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1641 ret = -EINVAL;
1642 goto fail_opts;
1645 update_flags_from_options(&bs->open_flags, opts);
1647 driver_name = qemu_opt_get(opts, "driver");
1648 drv = bdrv_find_format(driver_name);
1649 assert(drv != NULL);
1651 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1653 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1654 error_setg(errp,
1655 BDRV_OPT_FORCE_SHARE
1656 "=on can only be used with read-only images");
1657 ret = -EINVAL;
1658 goto fail_opts;
1661 if (file != NULL) {
1662 bdrv_refresh_filename(blk_bs(file));
1663 filename = blk_bs(file)->filename;
1664 } else {
1666 * Caution: while qdict_get_try_str() is fine, getting
1667 * non-string types would require more care. When @options
1668 * come from -blockdev or blockdev_add, its members are typed
1669 * according to the QAPI schema, but when they come from
1670 * -drive, they're all QString.
1672 filename = qdict_get_try_str(options, "filename");
1675 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1676 error_setg(errp, "The '%s' block driver requires a file name",
1677 drv->format_name);
1678 ret = -EINVAL;
1679 goto fail_opts;
1682 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1683 drv->format_name);
1685 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1687 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1688 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1689 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1690 } else {
1691 ret = -ENOTSUP;
1693 if (ret < 0) {
1694 error_setg(errp,
1695 !bs->read_only && bdrv_is_whitelisted(drv, true)
1696 ? "Driver '%s' can only be used for read-only devices"
1697 : "Driver '%s' is not whitelisted",
1698 drv->format_name);
1699 goto fail_opts;
1703 /* bdrv_new() and bdrv_close() make it so */
1704 assert(qatomic_read(&bs->copy_on_read) == 0);
1706 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1707 if (!bs->read_only) {
1708 bdrv_enable_copy_on_read(bs);
1709 } else {
1710 error_setg(errp, "Can't use copy-on-read on read-only device");
1711 ret = -EINVAL;
1712 goto fail_opts;
1716 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1717 if (discard != NULL) {
1718 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1719 error_setg(errp, "Invalid discard option");
1720 ret = -EINVAL;
1721 goto fail_opts;
1725 bs->detect_zeroes =
1726 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1727 if (local_err) {
1728 error_propagate(errp, local_err);
1729 ret = -EINVAL;
1730 goto fail_opts;
1733 if (filename != NULL) {
1734 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1735 } else {
1736 bs->filename[0] = '\0';
1738 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1740 /* Open the image, either directly or using a protocol */
1741 open_flags = bdrv_open_flags(bs, bs->open_flags);
1742 node_name = qemu_opt_get(opts, "node-name");
1744 assert(!drv->bdrv_file_open || file == NULL);
1745 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1746 if (ret < 0) {
1747 goto fail_opts;
1750 qemu_opts_del(opts);
1751 return 0;
1753 fail_opts:
1754 qemu_opts_del(opts);
1755 return ret;
1758 static QDict *parse_json_filename(const char *filename, Error **errp)
1760 QObject *options_obj;
1761 QDict *options;
1762 int ret;
1764 ret = strstart(filename, "json:", &filename);
1765 assert(ret);
1767 options_obj = qobject_from_json(filename, errp);
1768 if (!options_obj) {
1769 error_prepend(errp, "Could not parse the JSON options: ");
1770 return NULL;
1773 options = qobject_to(QDict, options_obj);
1774 if (!options) {
1775 qobject_unref(options_obj);
1776 error_setg(errp, "Invalid JSON object given");
1777 return NULL;
1780 qdict_flatten(options);
1782 return options;
1785 static void parse_json_protocol(QDict *options, const char **pfilename,
1786 Error **errp)
1788 QDict *json_options;
1789 Error *local_err = NULL;
1791 /* Parse json: pseudo-protocol */
1792 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1793 return;
1796 json_options = parse_json_filename(*pfilename, &local_err);
1797 if (local_err) {
1798 error_propagate(errp, local_err);
1799 return;
1802 /* Options given in the filename have lower priority than options
1803 * specified directly */
1804 qdict_join(options, json_options, false);
1805 qobject_unref(json_options);
1806 *pfilename = NULL;
1810 * Fills in default options for opening images and converts the legacy
1811 * filename/flags pair to option QDict entries.
1812 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1813 * block driver has been specified explicitly.
1815 static int bdrv_fill_options(QDict **options, const char *filename,
1816 int *flags, Error **errp)
1818 const char *drvname;
1819 bool protocol = *flags & BDRV_O_PROTOCOL;
1820 bool parse_filename = false;
1821 BlockDriver *drv = NULL;
1822 Error *local_err = NULL;
1825 * Caution: while qdict_get_try_str() is fine, getting non-string
1826 * types would require more care. When @options come from
1827 * -blockdev or blockdev_add, its members are typed according to
1828 * the QAPI schema, but when they come from -drive, they're all
1829 * QString.
1831 drvname = qdict_get_try_str(*options, "driver");
1832 if (drvname) {
1833 drv = bdrv_find_format(drvname);
1834 if (!drv) {
1835 error_setg(errp, "Unknown driver '%s'", drvname);
1836 return -ENOENT;
1838 /* If the user has explicitly specified the driver, this choice should
1839 * override the BDRV_O_PROTOCOL flag */
1840 protocol = drv->bdrv_file_open;
1843 if (protocol) {
1844 *flags |= BDRV_O_PROTOCOL;
1845 } else {
1846 *flags &= ~BDRV_O_PROTOCOL;
1849 /* Translate cache options from flags into options */
1850 update_options_from_flags(*options, *flags);
1852 /* Fetch the file name from the options QDict if necessary */
1853 if (protocol && filename) {
1854 if (!qdict_haskey(*options, "filename")) {
1855 qdict_put_str(*options, "filename", filename);
1856 parse_filename = true;
1857 } else {
1858 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1859 "the same time");
1860 return -EINVAL;
1864 /* Find the right block driver */
1865 /* See cautionary note on accessing @options above */
1866 filename = qdict_get_try_str(*options, "filename");
1868 if (!drvname && protocol) {
1869 if (filename) {
1870 drv = bdrv_find_protocol(filename, parse_filename, errp);
1871 if (!drv) {
1872 return -EINVAL;
1875 drvname = drv->format_name;
1876 qdict_put_str(*options, "driver", drvname);
1877 } else {
1878 error_setg(errp, "Must specify either driver or file");
1879 return -EINVAL;
1883 assert(drv || !protocol);
1885 /* Driver-specific filename parsing */
1886 if (drv && drv->bdrv_parse_filename && parse_filename) {
1887 drv->bdrv_parse_filename(filename, *options, &local_err);
1888 if (local_err) {
1889 error_propagate(errp, local_err);
1890 return -EINVAL;
1893 if (!drv->bdrv_needs_filename) {
1894 qdict_del(*options, "filename");
1898 return 0;
1901 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1902 uint64_t perm, uint64_t shared,
1903 GSList *ignore_children,
1904 bool *tighten_restrictions, Error **errp);
1905 static void bdrv_child_abort_perm_update(BdrvChild *c);
1906 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1908 typedef struct BlockReopenQueueEntry {
1909 bool prepared;
1910 bool perms_checked;
1911 BDRVReopenState state;
1912 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1913 } BlockReopenQueueEntry;
1916 * Return the flags that @bs will have after the reopens in @q have
1917 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1918 * return the current flags.
1920 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1922 BlockReopenQueueEntry *entry;
1924 if (q != NULL) {
1925 QTAILQ_FOREACH(entry, q, entry) {
1926 if (entry->state.bs == bs) {
1927 return entry->state.flags;
1932 return bs->open_flags;
1935 /* Returns whether the image file can be written to after the reopen queue @q
1936 * has been successfully applied, or right now if @q is NULL. */
1937 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1938 BlockReopenQueue *q)
1940 int flags = bdrv_reopen_get_flags(q, bs);
1942 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1946 * Return whether the BDS can be written to. This is not necessarily
1947 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1948 * be written to but do not count as read-only images.
1950 bool bdrv_is_writable(BlockDriverState *bs)
1952 return bdrv_is_writable_after_reopen(bs, NULL);
1955 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1956 BdrvChild *c, BdrvChildRole role,
1957 BlockReopenQueue *reopen_queue,
1958 uint64_t parent_perm, uint64_t parent_shared,
1959 uint64_t *nperm, uint64_t *nshared)
1961 assert(bs->drv && bs->drv->bdrv_child_perm);
1962 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1963 parent_perm, parent_shared,
1964 nperm, nshared);
1965 /* TODO Take force_share from reopen_queue */
1966 if (child_bs && child_bs->force_share) {
1967 *nshared = BLK_PERM_ALL;
1972 * Check whether permissions on this node can be changed in a way that
1973 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1974 * permissions of all its parents. This involves checking whether all necessary
1975 * permission changes to child nodes can be performed.
1977 * Will set *tighten_restrictions to true if and only if new permissions have to
1978 * be taken or currently shared permissions are to be unshared. Otherwise,
1979 * errors are not fatal as long as the caller accepts that the restrictions
1980 * remain tighter than they need to be. The caller still has to abort the
1981 * transaction.
1982 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1983 * encounter fatal errors even though no restrictions are to be tightened. For
1984 * example, changing a node from RW to RO will fail if the WRITE permission is
1985 * to be kept.
1987 * A call to this function must always be followed by a call to bdrv_set_perm()
1988 * or bdrv_abort_perm_update().
1990 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1991 uint64_t cumulative_perms,
1992 uint64_t cumulative_shared_perms,
1993 GSList *ignore_children,
1994 bool *tighten_restrictions, Error **errp)
1996 BlockDriver *drv = bs->drv;
1997 BdrvChild *c;
1998 int ret;
2000 assert(!q || !tighten_restrictions);
2002 if (tighten_restrictions) {
2003 uint64_t current_perms, current_shared;
2004 uint64_t added_perms, removed_shared_perms;
2006 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2008 added_perms = cumulative_perms & ~current_perms;
2009 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2011 *tighten_restrictions = added_perms || removed_shared_perms;
2014 /* Write permissions never work with read-only images */
2015 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2016 !bdrv_is_writable_after_reopen(bs, q))
2018 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2019 error_setg(errp, "Block node is read-only");
2020 } else {
2021 uint64_t current_perms, current_shared;
2022 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2023 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2024 error_setg(errp, "Cannot make block node read-only, there is "
2025 "a writer on it");
2026 } else {
2027 error_setg(errp, "Cannot make block node read-only and create "
2028 "a writer on it");
2032 return -EPERM;
2036 * Unaligned requests will automatically be aligned to bl.request_alignment
2037 * and without RESIZE we can't extend requests to write to space beyond the
2038 * end of the image, so it's required that the image size is aligned.
2040 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2041 !(cumulative_perms & BLK_PERM_RESIZE))
2043 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2044 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2045 "Image size is not a multiple of request "
2046 "alignment");
2047 return -EPERM;
2051 /* Check this node */
2052 if (!drv) {
2053 return 0;
2056 if (drv->bdrv_check_perm) {
2057 return drv->bdrv_check_perm(bs, cumulative_perms,
2058 cumulative_shared_perms, errp);
2061 /* Drivers that never have children can omit .bdrv_child_perm() */
2062 if (!drv->bdrv_child_perm) {
2063 assert(QLIST_EMPTY(&bs->children));
2064 return 0;
2067 /* Check all children */
2068 QLIST_FOREACH(c, &bs->children, next) {
2069 uint64_t cur_perm, cur_shared;
2070 bool child_tighten_restr;
2072 bdrv_child_perm(bs, c->bs, c, c->role, q,
2073 cumulative_perms, cumulative_shared_perms,
2074 &cur_perm, &cur_shared);
2075 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2076 tighten_restrictions ? &child_tighten_restr
2077 : NULL,
2078 errp);
2079 if (tighten_restrictions) {
2080 *tighten_restrictions |= child_tighten_restr;
2082 if (ret < 0) {
2083 return ret;
2087 return 0;
2091 * Notifies drivers that after a previous bdrv_check_perm() call, the
2092 * permission update is not performed and any preparations made for it (e.g.
2093 * taken file locks) need to be undone.
2095 * This function recursively notifies all child nodes.
2097 static void bdrv_abort_perm_update(BlockDriverState *bs)
2099 BlockDriver *drv = bs->drv;
2100 BdrvChild *c;
2102 if (!drv) {
2103 return;
2106 if (drv->bdrv_abort_perm_update) {
2107 drv->bdrv_abort_perm_update(bs);
2110 QLIST_FOREACH(c, &bs->children, next) {
2111 bdrv_child_abort_perm_update(c);
2115 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2116 uint64_t cumulative_shared_perms)
2118 BlockDriver *drv = bs->drv;
2119 BdrvChild *c;
2121 if (!drv) {
2122 return;
2125 /* Update this node */
2126 if (drv->bdrv_set_perm) {
2127 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2130 /* Drivers that never have children can omit .bdrv_child_perm() */
2131 if (!drv->bdrv_child_perm) {
2132 assert(QLIST_EMPTY(&bs->children));
2133 return;
2136 /* Update all children */
2137 QLIST_FOREACH(c, &bs->children, next) {
2138 uint64_t cur_perm, cur_shared;
2139 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2140 cumulative_perms, cumulative_shared_perms,
2141 &cur_perm, &cur_shared);
2142 bdrv_child_set_perm(c, cur_perm, cur_shared);
2146 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2147 uint64_t *shared_perm)
2149 BdrvChild *c;
2150 uint64_t cumulative_perms = 0;
2151 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2153 QLIST_FOREACH(c, &bs->parents, next_parent) {
2154 cumulative_perms |= c->perm;
2155 cumulative_shared_perms &= c->shared_perm;
2158 *perm = cumulative_perms;
2159 *shared_perm = cumulative_shared_perms;
2162 static char *bdrv_child_user_desc(BdrvChild *c)
2164 if (c->klass->get_parent_desc) {
2165 return c->klass->get_parent_desc(c);
2168 return g_strdup("another user");
2171 char *bdrv_perm_names(uint64_t perm)
2173 struct perm_name {
2174 uint64_t perm;
2175 const char *name;
2176 } permissions[] = {
2177 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2178 { BLK_PERM_WRITE, "write" },
2179 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2180 { BLK_PERM_RESIZE, "resize" },
2181 { BLK_PERM_GRAPH_MOD, "change children" },
2182 { 0, NULL }
2185 GString *result = g_string_sized_new(30);
2186 struct perm_name *p;
2188 for (p = permissions; p->name; p++) {
2189 if (perm & p->perm) {
2190 if (result->len > 0) {
2191 g_string_append(result, ", ");
2193 g_string_append(result, p->name);
2197 return g_string_free(result, FALSE);
2201 * Checks whether a new reference to @bs can be added if the new user requires
2202 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2203 * set, the BdrvChild objects in this list are ignored in the calculations;
2204 * this allows checking permission updates for an existing reference.
2206 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2208 * Needs to be followed by a call to either bdrv_set_perm() or
2209 * bdrv_abort_perm_update(). */
2210 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2211 uint64_t new_used_perm,
2212 uint64_t new_shared_perm,
2213 GSList *ignore_children,
2214 bool *tighten_restrictions,
2215 Error **errp)
2217 BdrvChild *c;
2218 uint64_t cumulative_perms = new_used_perm;
2219 uint64_t cumulative_shared_perms = new_shared_perm;
2221 assert(!q || !tighten_restrictions);
2223 /* There is no reason why anyone couldn't tolerate write_unchanged */
2224 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2226 QLIST_FOREACH(c, &bs->parents, next_parent) {
2227 if (g_slist_find(ignore_children, c)) {
2228 continue;
2231 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2232 char *user = bdrv_child_user_desc(c);
2233 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2235 if (tighten_restrictions) {
2236 *tighten_restrictions = true;
2239 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2240 "allow '%s' on %s",
2241 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2242 g_free(user);
2243 g_free(perm_names);
2244 return -EPERM;
2247 if ((c->perm & new_shared_perm) != c->perm) {
2248 char *user = bdrv_child_user_desc(c);
2249 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2251 if (tighten_restrictions) {
2252 *tighten_restrictions = true;
2255 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2256 "'%s' on %s",
2257 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2258 g_free(user);
2259 g_free(perm_names);
2260 return -EPERM;
2263 cumulative_perms |= c->perm;
2264 cumulative_shared_perms &= c->shared_perm;
2267 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2268 ignore_children, tighten_restrictions, errp);
2271 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2272 * bdrv_child_abort_perm_update(). */
2273 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2274 uint64_t perm, uint64_t shared,
2275 GSList *ignore_children,
2276 bool *tighten_restrictions, Error **errp)
2278 int ret;
2280 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2281 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2282 tighten_restrictions, errp);
2283 g_slist_free(ignore_children);
2285 if (ret < 0) {
2286 return ret;
2289 if (!c->has_backup_perm) {
2290 c->has_backup_perm = true;
2291 c->backup_perm = c->perm;
2292 c->backup_shared_perm = c->shared_perm;
2295 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2296 * same child twice during check_perm procedure
2299 c->perm = perm;
2300 c->shared_perm = shared;
2302 return 0;
2305 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2307 uint64_t cumulative_perms, cumulative_shared_perms;
2309 c->has_backup_perm = false;
2311 c->perm = perm;
2312 c->shared_perm = shared;
2314 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2315 &cumulative_shared_perms);
2316 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2319 static void bdrv_child_abort_perm_update(BdrvChild *c)
2321 if (c->has_backup_perm) {
2322 c->perm = c->backup_perm;
2323 c->shared_perm = c->backup_shared_perm;
2324 c->has_backup_perm = false;
2327 bdrv_abort_perm_update(c->bs);
2330 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2331 Error **errp)
2333 Error *local_err = NULL;
2334 int ret;
2335 bool tighten_restrictions;
2337 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2338 &tighten_restrictions, &local_err);
2339 if (ret < 0) {
2340 bdrv_child_abort_perm_update(c);
2341 if (tighten_restrictions) {
2342 error_propagate(errp, local_err);
2343 } else {
2345 * Our caller may intend to only loosen restrictions and
2346 * does not expect this function to fail. Errors are not
2347 * fatal in such a case, so we can just hide them from our
2348 * caller.
2350 error_free(local_err);
2351 ret = 0;
2353 return ret;
2356 bdrv_child_set_perm(c, perm, shared);
2358 return 0;
2361 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2363 uint64_t parent_perms, parent_shared;
2364 uint64_t perms, shared;
2366 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2367 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2368 parent_perms, parent_shared, &perms, &shared);
2370 return bdrv_child_try_set_perm(c, perms, shared, errp);
2374 * Default implementation for .bdrv_child_perm() for block filters:
2375 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2376 * filtered child.
2378 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2379 BdrvChildRole role,
2380 BlockReopenQueue *reopen_queue,
2381 uint64_t perm, uint64_t shared,
2382 uint64_t *nperm, uint64_t *nshared)
2384 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2385 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2388 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2389 BdrvChildRole role,
2390 BlockReopenQueue *reopen_queue,
2391 uint64_t perm, uint64_t shared,
2392 uint64_t *nperm, uint64_t *nshared)
2394 assert(role & BDRV_CHILD_COW);
2397 * We want consistent read from backing files if the parent needs it.
2398 * No other operations are performed on backing files.
2400 perm &= BLK_PERM_CONSISTENT_READ;
2403 * If the parent can deal with changing data, we're okay with a
2404 * writable and resizable backing file.
2405 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2407 if (shared & BLK_PERM_WRITE) {
2408 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2409 } else {
2410 shared = 0;
2413 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2414 BLK_PERM_WRITE_UNCHANGED;
2416 if (bs->open_flags & BDRV_O_INACTIVE) {
2417 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2420 *nperm = perm;
2421 *nshared = shared;
2424 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2425 BdrvChildRole role,
2426 BlockReopenQueue *reopen_queue,
2427 uint64_t perm, uint64_t shared,
2428 uint64_t *nperm, uint64_t *nshared)
2430 int flags;
2432 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2434 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2437 * Apart from the modifications below, the same permissions are
2438 * forwarded and left alone as for filters
2440 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2441 perm, shared, &perm, &shared);
2443 if (role & BDRV_CHILD_METADATA) {
2444 /* Format drivers may touch metadata even if the guest doesn't write */
2445 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2446 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2450 * bs->file always needs to be consistent because of the
2451 * metadata. We can never allow other users to resize or write
2452 * to it.
2454 if (!(flags & BDRV_O_NO_IO)) {
2455 perm |= BLK_PERM_CONSISTENT_READ;
2457 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2460 if (role & BDRV_CHILD_DATA) {
2462 * Technically, everything in this block is a subset of the
2463 * BDRV_CHILD_METADATA path taken above, and so this could
2464 * be an "else if" branch. However, that is not obvious, and
2465 * this function is not performance critical, therefore we let
2466 * this be an independent "if".
2470 * We cannot allow other users to resize the file because the
2471 * format driver might have some assumptions about the size
2472 * (e.g. because it is stored in metadata, or because the file
2473 * is split into fixed-size data files).
2475 shared &= ~BLK_PERM_RESIZE;
2478 * WRITE_UNCHANGED often cannot be performed as such on the
2479 * data file. For example, the qcow2 driver may still need to
2480 * write copied clusters on copy-on-read.
2482 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2483 perm |= BLK_PERM_WRITE;
2487 * If the data file is written to, the format driver may
2488 * expect to be able to resize it by writing beyond the EOF.
2490 if (perm & BLK_PERM_WRITE) {
2491 perm |= BLK_PERM_RESIZE;
2495 if (bs->open_flags & BDRV_O_INACTIVE) {
2496 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2499 *nperm = perm;
2500 *nshared = shared;
2503 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2504 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2505 uint64_t perm, uint64_t shared,
2506 uint64_t *nperm, uint64_t *nshared)
2508 if (role & BDRV_CHILD_FILTERED) {
2509 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2510 BDRV_CHILD_COW)));
2511 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2512 perm, shared, nperm, nshared);
2513 } else if (role & BDRV_CHILD_COW) {
2514 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2515 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2516 perm, shared, nperm, nshared);
2517 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2518 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2519 perm, shared, nperm, nshared);
2520 } else {
2521 g_assert_not_reached();
2525 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2527 static const uint64_t permissions[] = {
2528 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2529 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2530 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2531 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2532 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2535 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2536 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2538 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2540 return permissions[qapi_perm];
2543 static void bdrv_replace_child_noperm(BdrvChild *child,
2544 BlockDriverState *new_bs)
2546 BlockDriverState *old_bs = child->bs;
2547 int new_bs_quiesce_counter;
2548 int drain_saldo;
2550 assert(!child->frozen);
2552 if (old_bs && new_bs) {
2553 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2556 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2557 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2560 * If the new child node is drained but the old one was not, flush
2561 * all outstanding requests to the old child node.
2563 while (drain_saldo > 0 && child->klass->drained_begin) {
2564 bdrv_parent_drained_begin_single(child, true);
2565 drain_saldo--;
2568 if (old_bs) {
2569 /* Detach first so that the recursive drain sections coming from @child
2570 * are already gone and we only end the drain sections that came from
2571 * elsewhere. */
2572 if (child->klass->detach) {
2573 child->klass->detach(child);
2575 QLIST_REMOVE(child, next_parent);
2578 child->bs = new_bs;
2580 if (new_bs) {
2581 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2584 * Detaching the old node may have led to the new node's
2585 * quiesce_counter having been decreased. Not a problem, we
2586 * just need to recognize this here and then invoke
2587 * drained_end appropriately more often.
2589 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2590 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2592 /* Attach only after starting new drained sections, so that recursive
2593 * drain sections coming from @child don't get an extra .drained_begin
2594 * callback. */
2595 if (child->klass->attach) {
2596 child->klass->attach(child);
2601 * If the old child node was drained but the new one is not, allow
2602 * requests to come in only after the new node has been attached.
2604 while (drain_saldo < 0 && child->klass->drained_end) {
2605 bdrv_parent_drained_end_single(child);
2606 drain_saldo++;
2611 * Updates @child to change its reference to point to @new_bs, including
2612 * checking and applying the necessary permission updates both to the old node
2613 * and to @new_bs.
2615 * NULL is passed as @new_bs for removing the reference before freeing @child.
2617 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2618 * function uses bdrv_set_perm() to update the permissions according to the new
2619 * reference that @new_bs gets.
2621 * Callers must ensure that child->frozen is false.
2623 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2625 BlockDriverState *old_bs = child->bs;
2626 uint64_t perm, shared_perm;
2628 /* Asserts that child->frozen == false */
2629 bdrv_replace_child_noperm(child, new_bs);
2632 * Start with the new node's permissions. If @new_bs is a (direct
2633 * or indirect) child of @old_bs, we must complete the permission
2634 * update on @new_bs before we loosen the restrictions on @old_bs.
2635 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2636 * updating the permissions of @new_bs, and thus not purely loosen
2637 * restrictions.
2639 if (new_bs) {
2640 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2641 bdrv_set_perm(new_bs, perm, shared_perm);
2644 if (old_bs) {
2645 /* Update permissions for old node. This is guaranteed to succeed
2646 * because we're just taking a parent away, so we're loosening
2647 * restrictions. */
2648 bool tighten_restrictions;
2649 int ret;
2651 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2652 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2653 &tighten_restrictions, NULL);
2654 assert(tighten_restrictions == false);
2655 if (ret < 0) {
2656 /* We only tried to loosen restrictions, so errors are not fatal */
2657 bdrv_abort_perm_update(old_bs);
2658 } else {
2659 bdrv_set_perm(old_bs, perm, shared_perm);
2662 /* When the parent requiring a non-default AioContext is removed, the
2663 * node moves back to the main AioContext */
2664 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2669 * This function steals the reference to child_bs from the caller.
2670 * That reference is later dropped by bdrv_root_unref_child().
2672 * On failure NULL is returned, errp is set and the reference to
2673 * child_bs is also dropped.
2675 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2676 * (unless @child_bs is already in @ctx).
2678 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2679 const char *child_name,
2680 const BdrvChildClass *child_class,
2681 BdrvChildRole child_role,
2682 AioContext *ctx,
2683 uint64_t perm, uint64_t shared_perm,
2684 void *opaque, Error **errp)
2686 BdrvChild *child;
2687 Error *local_err = NULL;
2688 int ret;
2690 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2691 errp);
2692 if (ret < 0) {
2693 bdrv_abort_perm_update(child_bs);
2694 bdrv_unref(child_bs);
2695 return NULL;
2698 child = g_new(BdrvChild, 1);
2699 *child = (BdrvChild) {
2700 .bs = NULL,
2701 .name = g_strdup(child_name),
2702 .klass = child_class,
2703 .role = child_role,
2704 .perm = perm,
2705 .shared_perm = shared_perm,
2706 .opaque = opaque,
2709 /* If the AioContexts don't match, first try to move the subtree of
2710 * child_bs into the AioContext of the new parent. If this doesn't work,
2711 * try moving the parent into the AioContext of child_bs instead. */
2712 if (bdrv_get_aio_context(child_bs) != ctx) {
2713 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2714 if (ret < 0 && child_class->can_set_aio_ctx) {
2715 GSList *ignore = g_slist_prepend(NULL, child);
2716 ctx = bdrv_get_aio_context(child_bs);
2717 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2718 error_free(local_err);
2719 ret = 0;
2720 g_slist_free(ignore);
2721 ignore = g_slist_prepend(NULL, child);
2722 child_class->set_aio_ctx(child, ctx, &ignore);
2724 g_slist_free(ignore);
2726 if (ret < 0) {
2727 error_propagate(errp, local_err);
2728 g_free(child);
2729 bdrv_abort_perm_update(child_bs);
2730 bdrv_unref(child_bs);
2731 return NULL;
2735 /* This performs the matching bdrv_set_perm() for the above check. */
2736 bdrv_replace_child(child, child_bs);
2738 return child;
2742 * This function transfers the reference to child_bs from the caller
2743 * to parent_bs. That reference is later dropped by parent_bs on
2744 * bdrv_close() or if someone calls bdrv_unref_child().
2746 * On failure NULL is returned, errp is set and the reference to
2747 * child_bs is also dropped.
2749 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2750 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2752 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2753 BlockDriverState *child_bs,
2754 const char *child_name,
2755 const BdrvChildClass *child_class,
2756 BdrvChildRole child_role,
2757 Error **errp)
2759 BdrvChild *child;
2760 uint64_t perm, shared_perm;
2762 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2764 assert(parent_bs->drv);
2765 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2766 perm, shared_perm, &perm, &shared_perm);
2768 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2769 child_role, bdrv_get_aio_context(parent_bs),
2770 perm, shared_perm, parent_bs, errp);
2771 if (child == NULL) {
2772 return NULL;
2775 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2776 return child;
2779 static void bdrv_detach_child(BdrvChild *child)
2781 QLIST_SAFE_REMOVE(child, next);
2783 bdrv_replace_child(child, NULL);
2785 g_free(child->name);
2786 g_free(child);
2789 /* Callers must ensure that child->frozen is false. */
2790 void bdrv_root_unref_child(BdrvChild *child)
2792 BlockDriverState *child_bs;
2794 child_bs = child->bs;
2795 bdrv_detach_child(child);
2796 bdrv_unref(child_bs);
2800 * Clear all inherits_from pointers from children and grandchildren of
2801 * @root that point to @root, where necessary.
2803 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2805 BdrvChild *c;
2807 if (child->bs->inherits_from == root) {
2809 * Remove inherits_from only when the last reference between root and
2810 * child->bs goes away.
2812 QLIST_FOREACH(c, &root->children, next) {
2813 if (c != child && c->bs == child->bs) {
2814 break;
2817 if (c == NULL) {
2818 child->bs->inherits_from = NULL;
2822 QLIST_FOREACH(c, &child->bs->children, next) {
2823 bdrv_unset_inherits_from(root, c);
2827 /* Callers must ensure that child->frozen is false. */
2828 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2830 if (child == NULL) {
2831 return;
2834 bdrv_unset_inherits_from(parent, child);
2835 bdrv_root_unref_child(child);
2839 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2841 BdrvChild *c;
2842 QLIST_FOREACH(c, &bs->parents, next_parent) {
2843 if (c->klass->change_media) {
2844 c->klass->change_media(c, load);
2849 /* Return true if you can reach parent going through child->inherits_from
2850 * recursively. If parent or child are NULL, return false */
2851 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2852 BlockDriverState *parent)
2854 while (child && child != parent) {
2855 child = child->inherits_from;
2858 return child != NULL;
2862 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2863 * mostly used for COW backing children (role = COW), but also for
2864 * filtered children (role = FILTERED | PRIMARY).
2866 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2868 if (bs->drv && bs->drv->is_filter) {
2869 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2870 } else {
2871 return BDRV_CHILD_COW;
2876 * Sets the bs->backing link of a BDS. A new reference is created; callers
2877 * which don't need their own reference any more must call bdrv_unref().
2879 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2880 Error **errp)
2882 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2883 bdrv_inherits_from_recursive(backing_hd, bs);
2885 if (bdrv_is_backing_chain_frozen(bs, child_bs(bs->backing), errp)) {
2886 return;
2889 if (backing_hd) {
2890 bdrv_ref(backing_hd);
2893 if (bs->backing) {
2894 /* Cannot be frozen, we checked that above */
2895 bdrv_unref_child(bs, bs->backing);
2896 bs->backing = NULL;
2899 if (!backing_hd) {
2900 goto out;
2903 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2904 bdrv_backing_role(bs), errp);
2905 /* If backing_hd was already part of bs's backing chain, and
2906 * inherits_from pointed recursively to bs then let's update it to
2907 * point directly to bs (else it will become NULL). */
2908 if (bs->backing && update_inherits_from) {
2909 backing_hd->inherits_from = bs;
2912 out:
2913 bdrv_refresh_limits(bs, NULL);
2917 * Opens the backing file for a BlockDriverState if not yet open
2919 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2920 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2921 * itself, all options starting with "${bdref_key}." are considered part of the
2922 * BlockdevRef.
2924 * TODO Can this be unified with bdrv_open_image()?
2926 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2927 const char *bdref_key, Error **errp)
2929 char *backing_filename = NULL;
2930 char *bdref_key_dot;
2931 const char *reference = NULL;
2932 int ret = 0;
2933 bool implicit_backing = false;
2934 BlockDriverState *backing_hd;
2935 QDict *options;
2936 QDict *tmp_parent_options = NULL;
2937 Error *local_err = NULL;
2939 if (bs->backing != NULL) {
2940 goto free_exit;
2943 /* NULL means an empty set of options */
2944 if (parent_options == NULL) {
2945 tmp_parent_options = qdict_new();
2946 parent_options = tmp_parent_options;
2949 bs->open_flags &= ~BDRV_O_NO_BACKING;
2951 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2952 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2953 g_free(bdref_key_dot);
2956 * Caution: while qdict_get_try_str() is fine, getting non-string
2957 * types would require more care. When @parent_options come from
2958 * -blockdev or blockdev_add, its members are typed according to
2959 * the QAPI schema, but when they come from -drive, they're all
2960 * QString.
2962 reference = qdict_get_try_str(parent_options, bdref_key);
2963 if (reference || qdict_haskey(options, "file.filename")) {
2964 /* keep backing_filename NULL */
2965 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2966 qobject_unref(options);
2967 goto free_exit;
2968 } else {
2969 if (qdict_size(options) == 0) {
2970 /* If the user specifies options that do not modify the
2971 * backing file's behavior, we might still consider it the
2972 * implicit backing file. But it's easier this way, and
2973 * just specifying some of the backing BDS's options is
2974 * only possible with -drive anyway (otherwise the QAPI
2975 * schema forces the user to specify everything). */
2976 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2979 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2980 if (local_err) {
2981 ret = -EINVAL;
2982 error_propagate(errp, local_err);
2983 qobject_unref(options);
2984 goto free_exit;
2988 if (!bs->drv || !bs->drv->supports_backing) {
2989 ret = -EINVAL;
2990 error_setg(errp, "Driver doesn't support backing files");
2991 qobject_unref(options);
2992 goto free_exit;
2995 if (!reference &&
2996 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2997 qdict_put_str(options, "driver", bs->backing_format);
3000 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3001 &child_of_bds, bdrv_backing_role(bs), errp);
3002 if (!backing_hd) {
3003 bs->open_flags |= BDRV_O_NO_BACKING;
3004 error_prepend(errp, "Could not open backing file: ");
3005 ret = -EINVAL;
3006 goto free_exit;
3009 if (implicit_backing) {
3010 bdrv_refresh_filename(backing_hd);
3011 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3012 backing_hd->filename);
3015 /* Hook up the backing file link; drop our reference, bs owns the
3016 * backing_hd reference now */
3017 bdrv_set_backing_hd(bs, backing_hd, &local_err);
3018 bdrv_unref(backing_hd);
3019 if (local_err) {
3020 error_propagate(errp, local_err);
3021 ret = -EINVAL;
3022 goto free_exit;
3025 qdict_del(parent_options, bdref_key);
3027 free_exit:
3028 g_free(backing_filename);
3029 qobject_unref(tmp_parent_options);
3030 return ret;
3033 static BlockDriverState *
3034 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3035 BlockDriverState *parent, const BdrvChildClass *child_class,
3036 BdrvChildRole child_role, bool allow_none, Error **errp)
3038 BlockDriverState *bs = NULL;
3039 QDict *image_options;
3040 char *bdref_key_dot;
3041 const char *reference;
3043 assert(child_class != NULL);
3045 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3046 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3047 g_free(bdref_key_dot);
3050 * Caution: while qdict_get_try_str() is fine, getting non-string
3051 * types would require more care. When @options come from
3052 * -blockdev or blockdev_add, its members are typed according to
3053 * the QAPI schema, but when they come from -drive, they're all
3054 * QString.
3056 reference = qdict_get_try_str(options, bdref_key);
3057 if (!filename && !reference && !qdict_size(image_options)) {
3058 if (!allow_none) {
3059 error_setg(errp, "A block device must be specified for \"%s\"",
3060 bdref_key);
3062 qobject_unref(image_options);
3063 goto done;
3066 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3067 parent, child_class, child_role, errp);
3068 if (!bs) {
3069 goto done;
3072 done:
3073 qdict_del(options, bdref_key);
3074 return bs;
3078 * Opens a disk image whose options are given as BlockdevRef in another block
3079 * device's options.
3081 * If allow_none is true, no image will be opened if filename is false and no
3082 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3084 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3085 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3086 * itself, all options starting with "${bdref_key}." are considered part of the
3087 * BlockdevRef.
3089 * The BlockdevRef will be removed from the options QDict.
3091 BdrvChild *bdrv_open_child(const char *filename,
3092 QDict *options, const char *bdref_key,
3093 BlockDriverState *parent,
3094 const BdrvChildClass *child_class,
3095 BdrvChildRole child_role,
3096 bool allow_none, Error **errp)
3098 BlockDriverState *bs;
3100 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3101 child_role, allow_none, errp);
3102 if (bs == NULL) {
3103 return NULL;
3106 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3107 errp);
3111 * TODO Future callers may need to specify parent/child_class in order for
3112 * option inheritance to work. Existing callers use it for the root node.
3114 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3116 BlockDriverState *bs = NULL;
3117 QObject *obj = NULL;
3118 QDict *qdict = NULL;
3119 const char *reference = NULL;
3120 Visitor *v = NULL;
3122 if (ref->type == QTYPE_QSTRING) {
3123 reference = ref->u.reference;
3124 } else {
3125 BlockdevOptions *options = &ref->u.definition;
3126 assert(ref->type == QTYPE_QDICT);
3128 v = qobject_output_visitor_new(&obj);
3129 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3130 visit_complete(v, &obj);
3132 qdict = qobject_to(QDict, obj);
3133 qdict_flatten(qdict);
3135 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3136 * compatibility with other callers) rather than what we want as the
3137 * real defaults. Apply the defaults here instead. */
3138 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3139 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3140 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3141 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3145 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3146 obj = NULL;
3147 qobject_unref(obj);
3148 visit_free(v);
3149 return bs;
3152 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3153 int flags,
3154 QDict *snapshot_options,
3155 Error **errp)
3157 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3158 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3159 int64_t total_size;
3160 QemuOpts *opts = NULL;
3161 BlockDriverState *bs_snapshot = NULL;
3162 Error *local_err = NULL;
3163 int ret;
3165 /* if snapshot, we create a temporary backing file and open it
3166 instead of opening 'filename' directly */
3168 /* Get the required size from the image */
3169 total_size = bdrv_getlength(bs);
3170 if (total_size < 0) {
3171 error_setg_errno(errp, -total_size, "Could not get image size");
3172 goto out;
3175 /* Create the temporary image */
3176 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3177 if (ret < 0) {
3178 error_setg_errno(errp, -ret, "Could not get temporary filename");
3179 goto out;
3182 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3183 &error_abort);
3184 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3185 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3186 qemu_opts_del(opts);
3187 if (ret < 0) {
3188 error_prepend(errp, "Could not create temporary overlay '%s': ",
3189 tmp_filename);
3190 goto out;
3193 /* Prepare options QDict for the temporary file */
3194 qdict_put_str(snapshot_options, "file.driver", "file");
3195 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3196 qdict_put_str(snapshot_options, "driver", "qcow2");
3198 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3199 snapshot_options = NULL;
3200 if (!bs_snapshot) {
3201 goto out;
3204 /* bdrv_append() consumes a strong reference to bs_snapshot
3205 * (i.e. it will call bdrv_unref() on it) even on error, so in
3206 * order to be able to return one, we have to increase
3207 * bs_snapshot's refcount here */
3208 bdrv_ref(bs_snapshot);
3209 bdrv_append(bs_snapshot, bs, &local_err);
3210 if (local_err) {
3211 error_propagate(errp, local_err);
3212 bs_snapshot = NULL;
3213 goto out;
3216 out:
3217 qobject_unref(snapshot_options);
3218 g_free(tmp_filename);
3219 return bs_snapshot;
3223 * Opens a disk image (raw, qcow2, vmdk, ...)
3225 * options is a QDict of options to pass to the block drivers, or NULL for an
3226 * empty set of options. The reference to the QDict belongs to the block layer
3227 * after the call (even on failure), so if the caller intends to reuse the
3228 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3230 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3231 * If it is not NULL, the referenced BDS will be reused.
3233 * The reference parameter may be used to specify an existing block device which
3234 * should be opened. If specified, neither options nor a filename may be given,
3235 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3237 static BlockDriverState *bdrv_open_inherit(const char *filename,
3238 const char *reference,
3239 QDict *options, int flags,
3240 BlockDriverState *parent,
3241 const BdrvChildClass *child_class,
3242 BdrvChildRole child_role,
3243 Error **errp)
3245 int ret;
3246 BlockBackend *file = NULL;
3247 BlockDriverState *bs;
3248 BlockDriver *drv = NULL;
3249 BdrvChild *child;
3250 const char *drvname;
3251 const char *backing;
3252 Error *local_err = NULL;
3253 QDict *snapshot_options = NULL;
3254 int snapshot_flags = 0;
3256 assert(!child_class || !flags);
3257 assert(!child_class == !parent);
3259 if (reference) {
3260 bool options_non_empty = options ? qdict_size(options) : false;
3261 qobject_unref(options);
3263 if (filename || options_non_empty) {
3264 error_setg(errp, "Cannot reference an existing block device with "
3265 "additional options or a new filename");
3266 return NULL;
3269 bs = bdrv_lookup_bs(reference, reference, errp);
3270 if (!bs) {
3271 return NULL;
3274 bdrv_ref(bs);
3275 return bs;
3278 bs = bdrv_new();
3280 /* NULL means an empty set of options */
3281 if (options == NULL) {
3282 options = qdict_new();
3285 /* json: syntax counts as explicit options, as if in the QDict */
3286 parse_json_protocol(options, &filename, &local_err);
3287 if (local_err) {
3288 goto fail;
3291 bs->explicit_options = qdict_clone_shallow(options);
3293 if (child_class) {
3294 bool parent_is_format;
3296 if (parent->drv) {
3297 parent_is_format = parent->drv->is_format;
3298 } else {
3300 * parent->drv is not set yet because this node is opened for
3301 * (potential) format probing. That means that @parent is going
3302 * to be a format node.
3304 parent_is_format = true;
3307 bs->inherits_from = parent;
3308 child_class->inherit_options(child_role, parent_is_format,
3309 &flags, options,
3310 parent->open_flags, parent->options);
3313 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3314 if (ret < 0) {
3315 goto fail;
3319 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3320 * Caution: getting a boolean member of @options requires care.
3321 * When @options come from -blockdev or blockdev_add, members are
3322 * typed according to the QAPI schema, but when they come from
3323 * -drive, they're all QString.
3325 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3326 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3327 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3328 } else {
3329 flags &= ~BDRV_O_RDWR;
3332 if (flags & BDRV_O_SNAPSHOT) {
3333 snapshot_options = qdict_new();
3334 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3335 flags, options);
3336 /* Let bdrv_backing_options() override "read-only" */
3337 qdict_del(options, BDRV_OPT_READ_ONLY);
3338 bdrv_inherited_options(BDRV_CHILD_COW, true,
3339 &flags, options, flags, options);
3342 bs->open_flags = flags;
3343 bs->options = options;
3344 options = qdict_clone_shallow(options);
3346 /* Find the right image format driver */
3347 /* See cautionary note on accessing @options above */
3348 drvname = qdict_get_try_str(options, "driver");
3349 if (drvname) {
3350 drv = bdrv_find_format(drvname);
3351 if (!drv) {
3352 error_setg(errp, "Unknown driver: '%s'", drvname);
3353 goto fail;
3357 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3359 /* See cautionary note on accessing @options above */
3360 backing = qdict_get_try_str(options, "backing");
3361 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3362 (backing && *backing == '\0'))
3364 if (backing) {
3365 warn_report("Use of \"backing\": \"\" is deprecated; "
3366 "use \"backing\": null instead");
3368 flags |= BDRV_O_NO_BACKING;
3369 qdict_del(bs->explicit_options, "backing");
3370 qdict_del(bs->options, "backing");
3371 qdict_del(options, "backing");
3374 /* Open image file without format layer. This BlockBackend is only used for
3375 * probing, the block drivers will do their own bdrv_open_child() for the
3376 * same BDS, which is why we put the node name back into options. */
3377 if ((flags & BDRV_O_PROTOCOL) == 0) {
3378 BlockDriverState *file_bs;
3380 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3381 &child_of_bds, BDRV_CHILD_IMAGE,
3382 true, &local_err);
3383 if (local_err) {
3384 goto fail;
3386 if (file_bs != NULL) {
3387 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3388 * looking at the header to guess the image format. This works even
3389 * in cases where a guest would not see a consistent state. */
3390 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3391 blk_insert_bs(file, file_bs, &local_err);
3392 bdrv_unref(file_bs);
3393 if (local_err) {
3394 goto fail;
3397 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3401 /* Image format probing */
3402 bs->probed = !drv;
3403 if (!drv && file) {
3404 ret = find_image_format(file, filename, &drv, &local_err);
3405 if (ret < 0) {
3406 goto fail;
3409 * This option update would logically belong in bdrv_fill_options(),
3410 * but we first need to open bs->file for the probing to work, while
3411 * opening bs->file already requires the (mostly) final set of options
3412 * so that cache mode etc. can be inherited.
3414 * Adding the driver later is somewhat ugly, but it's not an option
3415 * that would ever be inherited, so it's correct. We just need to make
3416 * sure to update both bs->options (which has the full effective
3417 * options for bs) and options (which has file.* already removed).
3419 qdict_put_str(bs->options, "driver", drv->format_name);
3420 qdict_put_str(options, "driver", drv->format_name);
3421 } else if (!drv) {
3422 error_setg(errp, "Must specify either driver or file");
3423 goto fail;
3426 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3427 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3428 /* file must be NULL if a protocol BDS is about to be created
3429 * (the inverse results in an error message from bdrv_open_common()) */
3430 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3432 /* Open the image */
3433 ret = bdrv_open_common(bs, file, options, &local_err);
3434 if (ret < 0) {
3435 goto fail;
3438 if (file) {
3439 blk_unref(file);
3440 file = NULL;
3443 /* If there is a backing file, use it */
3444 if ((flags & BDRV_O_NO_BACKING) == 0) {
3445 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3446 if (ret < 0) {
3447 goto close_and_fail;
3451 /* Remove all children options and references
3452 * from bs->options and bs->explicit_options */
3453 QLIST_FOREACH(child, &bs->children, next) {
3454 char *child_key_dot;
3455 child_key_dot = g_strdup_printf("%s.", child->name);
3456 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3457 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3458 qdict_del(bs->explicit_options, child->name);
3459 qdict_del(bs->options, child->name);
3460 g_free(child_key_dot);
3463 /* Check if any unknown options were used */
3464 if (qdict_size(options) != 0) {
3465 const QDictEntry *entry = qdict_first(options);
3466 if (flags & BDRV_O_PROTOCOL) {
3467 error_setg(errp, "Block protocol '%s' doesn't support the option "
3468 "'%s'", drv->format_name, entry->key);
3469 } else {
3470 error_setg(errp,
3471 "Block format '%s' does not support the option '%s'",
3472 drv->format_name, entry->key);
3475 goto close_and_fail;
3478 bdrv_parent_cb_change_media(bs, true);
3480 qobject_unref(options);
3481 options = NULL;
3483 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3484 * temporary snapshot afterwards. */
3485 if (snapshot_flags) {
3486 BlockDriverState *snapshot_bs;
3487 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3488 snapshot_options, &local_err);
3489 snapshot_options = NULL;
3490 if (local_err) {
3491 goto close_and_fail;
3493 /* We are not going to return bs but the overlay on top of it
3494 * (snapshot_bs); thus, we have to drop the strong reference to bs
3495 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3496 * though, because the overlay still has a reference to it. */
3497 bdrv_unref(bs);
3498 bs = snapshot_bs;
3501 return bs;
3503 fail:
3504 blk_unref(file);
3505 qobject_unref(snapshot_options);
3506 qobject_unref(bs->explicit_options);
3507 qobject_unref(bs->options);
3508 qobject_unref(options);
3509 bs->options = NULL;
3510 bs->explicit_options = NULL;
3511 bdrv_unref(bs);
3512 error_propagate(errp, local_err);
3513 return NULL;
3515 close_and_fail:
3516 bdrv_unref(bs);
3517 qobject_unref(snapshot_options);
3518 qobject_unref(options);
3519 error_propagate(errp, local_err);
3520 return NULL;
3523 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3524 QDict *options, int flags, Error **errp)
3526 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3527 NULL, 0, errp);
3530 /* Return true if the NULL-terminated @list contains @str */
3531 static bool is_str_in_list(const char *str, const char *const *list)
3533 if (str && list) {
3534 int i;
3535 for (i = 0; list[i] != NULL; i++) {
3536 if (!strcmp(str, list[i])) {
3537 return true;
3541 return false;
3545 * Check that every option set in @bs->options is also set in
3546 * @new_opts.
3548 * Options listed in the common_options list and in
3549 * @bs->drv->mutable_opts are skipped.
3551 * Return 0 on success, otherwise return -EINVAL and set @errp.
3553 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3554 const QDict *new_opts, Error **errp)
3556 const QDictEntry *e;
3557 /* These options are common to all block drivers and are handled
3558 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3559 const char *const common_options[] = {
3560 "node-name", "discard", "cache.direct", "cache.no-flush",
3561 "read-only", "auto-read-only", "detect-zeroes", NULL
3564 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3565 if (!qdict_haskey(new_opts, e->key) &&
3566 !is_str_in_list(e->key, common_options) &&
3567 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3568 error_setg(errp, "Option '%s' cannot be reset "
3569 "to its default value", e->key);
3570 return -EINVAL;
3574 return 0;
3578 * Returns true if @child can be reached recursively from @bs
3580 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3581 BlockDriverState *child)
3583 BdrvChild *c;
3585 if (bs == child) {
3586 return true;
3589 QLIST_FOREACH(c, &bs->children, next) {
3590 if (bdrv_recurse_has_child(c->bs, child)) {
3591 return true;
3595 return false;
3599 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3600 * reopen of multiple devices.
3602 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3603 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3604 * be created and initialized. This newly created BlockReopenQueue should be
3605 * passed back in for subsequent calls that are intended to be of the same
3606 * atomic 'set'.
3608 * bs is the BlockDriverState to add to the reopen queue.
3610 * options contains the changed options for the associated bs
3611 * (the BlockReopenQueue takes ownership)
3613 * flags contains the open flags for the associated bs
3615 * returns a pointer to bs_queue, which is either the newly allocated
3616 * bs_queue, or the existing bs_queue being used.
3618 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3620 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3621 BlockDriverState *bs,
3622 QDict *options,
3623 const BdrvChildClass *klass,
3624 BdrvChildRole role,
3625 bool parent_is_format,
3626 QDict *parent_options,
3627 int parent_flags,
3628 bool keep_old_opts)
3630 assert(bs != NULL);
3632 BlockReopenQueueEntry *bs_entry;
3633 BdrvChild *child;
3634 QDict *old_options, *explicit_options, *options_copy;
3635 int flags;
3636 QemuOpts *opts;
3638 /* Make sure that the caller remembered to use a drained section. This is
3639 * important to avoid graph changes between the recursive queuing here and
3640 * bdrv_reopen_multiple(). */
3641 assert(bs->quiesce_counter > 0);
3643 if (bs_queue == NULL) {
3644 bs_queue = g_new0(BlockReopenQueue, 1);
3645 QTAILQ_INIT(bs_queue);
3648 if (!options) {
3649 options = qdict_new();
3652 /* Check if this BlockDriverState is already in the queue */
3653 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3654 if (bs == bs_entry->state.bs) {
3655 break;
3660 * Precedence of options:
3661 * 1. Explicitly passed in options (highest)
3662 * 2. Retained from explicitly set options of bs
3663 * 3. Inherited from parent node
3664 * 4. Retained from effective options of bs
3667 /* Old explicitly set values (don't overwrite by inherited value) */
3668 if (bs_entry || keep_old_opts) {
3669 old_options = qdict_clone_shallow(bs_entry ?
3670 bs_entry->state.explicit_options :
3671 bs->explicit_options);
3672 bdrv_join_options(bs, options, old_options);
3673 qobject_unref(old_options);
3676 explicit_options = qdict_clone_shallow(options);
3678 /* Inherit from parent node */
3679 if (parent_options) {
3680 flags = 0;
3681 klass->inherit_options(role, parent_is_format, &flags, options,
3682 parent_flags, parent_options);
3683 } else {
3684 flags = bdrv_get_flags(bs);
3687 if (keep_old_opts) {
3688 /* Old values are used for options that aren't set yet */
3689 old_options = qdict_clone_shallow(bs->options);
3690 bdrv_join_options(bs, options, old_options);
3691 qobject_unref(old_options);
3694 /* We have the final set of options so let's update the flags */
3695 options_copy = qdict_clone_shallow(options);
3696 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3697 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3698 update_flags_from_options(&flags, opts);
3699 qemu_opts_del(opts);
3700 qobject_unref(options_copy);
3702 /* bdrv_open_inherit() sets and clears some additional flags internally */
3703 flags &= ~BDRV_O_PROTOCOL;
3704 if (flags & BDRV_O_RDWR) {
3705 flags |= BDRV_O_ALLOW_RDWR;
3708 if (!bs_entry) {
3709 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3710 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3711 } else {
3712 qobject_unref(bs_entry->state.options);
3713 qobject_unref(bs_entry->state.explicit_options);
3716 bs_entry->state.bs = bs;
3717 bs_entry->state.options = options;
3718 bs_entry->state.explicit_options = explicit_options;
3719 bs_entry->state.flags = flags;
3721 /* This needs to be overwritten in bdrv_reopen_prepare() */
3722 bs_entry->state.perm = UINT64_MAX;
3723 bs_entry->state.shared_perm = 0;
3726 * If keep_old_opts is false then it means that unspecified
3727 * options must be reset to their original value. We don't allow
3728 * resetting 'backing' but we need to know if the option is
3729 * missing in order to decide if we have to return an error.
3731 if (!keep_old_opts) {
3732 bs_entry->state.backing_missing =
3733 !qdict_haskey(options, "backing") &&
3734 !qdict_haskey(options, "backing.driver");
3737 QLIST_FOREACH(child, &bs->children, next) {
3738 QDict *new_child_options = NULL;
3739 bool child_keep_old = keep_old_opts;
3741 /* reopen can only change the options of block devices that were
3742 * implicitly created and inherited options. For other (referenced)
3743 * block devices, a syntax like "backing.foo" results in an error. */
3744 if (child->bs->inherits_from != bs) {
3745 continue;
3748 /* Check if the options contain a child reference */
3749 if (qdict_haskey(options, child->name)) {
3750 const char *childref = qdict_get_try_str(options, child->name);
3752 * The current child must not be reopened if the child
3753 * reference is null or points to a different node.
3755 if (g_strcmp0(childref, child->bs->node_name)) {
3756 continue;
3759 * If the child reference points to the current child then
3760 * reopen it with its existing set of options (note that
3761 * it can still inherit new options from the parent).
3763 child_keep_old = true;
3764 } else {
3765 /* Extract child options ("child-name.*") */
3766 char *child_key_dot = g_strdup_printf("%s.", child->name);
3767 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3768 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3769 g_free(child_key_dot);
3772 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3773 child->klass, child->role, bs->drv->is_format,
3774 options, flags, child_keep_old);
3777 return bs_queue;
3780 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3781 BlockDriverState *bs,
3782 QDict *options, bool keep_old_opts)
3784 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
3785 NULL, 0, keep_old_opts);
3789 * Reopen multiple BlockDriverStates atomically & transactionally.
3791 * The queue passed in (bs_queue) must have been built up previous
3792 * via bdrv_reopen_queue().
3794 * Reopens all BDS specified in the queue, with the appropriate
3795 * flags. All devices are prepared for reopen, and failure of any
3796 * device will cause all device changes to be abandoned, and intermediate
3797 * data cleaned up.
3799 * If all devices prepare successfully, then the changes are committed
3800 * to all devices.
3802 * All affected nodes must be drained between bdrv_reopen_queue() and
3803 * bdrv_reopen_multiple().
3805 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
3807 int ret = -1;
3808 BlockReopenQueueEntry *bs_entry, *next;
3810 assert(bs_queue != NULL);
3812 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3813 assert(bs_entry->state.bs->quiesce_counter > 0);
3814 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
3815 goto cleanup;
3817 bs_entry->prepared = true;
3820 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3821 BDRVReopenState *state = &bs_entry->state;
3822 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
3823 state->shared_perm, NULL, NULL, errp);
3824 if (ret < 0) {
3825 goto cleanup_perm;
3827 /* Check if new_backing_bs would accept the new permissions */
3828 if (state->replace_backing_bs && state->new_backing_bs) {
3829 uint64_t nperm, nshared;
3830 bdrv_child_perm(state->bs, state->new_backing_bs,
3831 NULL, bdrv_backing_role(state->bs),
3832 bs_queue, state->perm, state->shared_perm,
3833 &nperm, &nshared);
3834 ret = bdrv_check_update_perm(state->new_backing_bs, NULL,
3835 nperm, nshared, NULL, NULL, errp);
3836 if (ret < 0) {
3837 goto cleanup_perm;
3840 bs_entry->perms_checked = true;
3844 * If we reach this point, we have success and just need to apply the
3845 * changes.
3847 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3848 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3849 * children are usually goes after parents in reopen-queue, so go from last
3850 * to first element.
3852 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3853 bdrv_reopen_commit(&bs_entry->state);
3856 ret = 0;
3857 cleanup_perm:
3858 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3859 BDRVReopenState *state = &bs_entry->state;
3861 if (!bs_entry->perms_checked) {
3862 continue;
3865 if (ret == 0) {
3866 bdrv_set_perm(state->bs, state->perm, state->shared_perm);
3867 } else {
3868 bdrv_abort_perm_update(state->bs);
3869 if (state->replace_backing_bs && state->new_backing_bs) {
3870 bdrv_abort_perm_update(state->new_backing_bs);
3875 if (ret == 0) {
3876 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3877 BlockDriverState *bs = bs_entry->state.bs;
3879 if (bs->drv->bdrv_reopen_commit_post)
3880 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
3883 cleanup:
3884 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3885 if (ret) {
3886 if (bs_entry->prepared) {
3887 bdrv_reopen_abort(&bs_entry->state);
3889 qobject_unref(bs_entry->state.explicit_options);
3890 qobject_unref(bs_entry->state.options);
3892 if (bs_entry->state.new_backing_bs) {
3893 bdrv_unref(bs_entry->state.new_backing_bs);
3895 g_free(bs_entry);
3897 g_free(bs_queue);
3899 return ret;
3902 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3903 Error **errp)
3905 int ret;
3906 BlockReopenQueue *queue;
3907 QDict *opts = qdict_new();
3909 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3911 bdrv_subtree_drained_begin(bs);
3912 queue = bdrv_reopen_queue(NULL, bs, opts, true);
3913 ret = bdrv_reopen_multiple(queue, errp);
3914 bdrv_subtree_drained_end(bs);
3916 return ret;
3919 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3920 BdrvChild *c)
3922 BlockReopenQueueEntry *entry;
3924 QTAILQ_FOREACH(entry, q, entry) {
3925 BlockDriverState *bs = entry->state.bs;
3926 BdrvChild *child;
3928 QLIST_FOREACH(child, &bs->children, next) {
3929 if (child == c) {
3930 return entry;
3935 return NULL;
3938 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3939 uint64_t *perm, uint64_t *shared)
3941 BdrvChild *c;
3942 BlockReopenQueueEntry *parent;
3943 uint64_t cumulative_perms = 0;
3944 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3946 QLIST_FOREACH(c, &bs->parents, next_parent) {
3947 parent = find_parent_in_reopen_queue(q, c);
3948 if (!parent) {
3949 cumulative_perms |= c->perm;
3950 cumulative_shared_perms &= c->shared_perm;
3951 } else {
3952 uint64_t nperm, nshared;
3954 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3955 parent->state.perm, parent->state.shared_perm,
3956 &nperm, &nshared);
3958 cumulative_perms |= nperm;
3959 cumulative_shared_perms &= nshared;
3962 *perm = cumulative_perms;
3963 *shared = cumulative_shared_perms;
3966 static bool bdrv_reopen_can_attach(BlockDriverState *parent,
3967 BdrvChild *child,
3968 BlockDriverState *new_child,
3969 Error **errp)
3971 AioContext *parent_ctx = bdrv_get_aio_context(parent);
3972 AioContext *child_ctx = bdrv_get_aio_context(new_child);
3973 GSList *ignore;
3974 bool ret;
3976 ignore = g_slist_prepend(NULL, child);
3977 ret = bdrv_can_set_aio_context(new_child, parent_ctx, &ignore, NULL);
3978 g_slist_free(ignore);
3979 if (ret) {
3980 return ret;
3983 ignore = g_slist_prepend(NULL, child);
3984 ret = bdrv_can_set_aio_context(parent, child_ctx, &ignore, errp);
3985 g_slist_free(ignore);
3986 return ret;
3990 * Take a BDRVReopenState and check if the value of 'backing' in the
3991 * reopen_state->options QDict is valid or not.
3993 * If 'backing' is missing from the QDict then return 0.
3995 * If 'backing' contains the node name of the backing file of
3996 * reopen_state->bs then return 0.
3998 * If 'backing' contains a different node name (or is null) then check
3999 * whether the current backing file can be replaced with the new one.
4000 * If that's the case then reopen_state->replace_backing_bs is set to
4001 * true and reopen_state->new_backing_bs contains a pointer to the new
4002 * backing BlockDriverState (or NULL).
4004 * Return 0 on success, otherwise return < 0 and set @errp.
4006 static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
4007 Error **errp)
4009 BlockDriverState *bs = reopen_state->bs;
4010 BlockDriverState *overlay_bs, *below_bs, *new_backing_bs;
4011 QObject *value;
4012 const char *str;
4014 value = qdict_get(reopen_state->options, "backing");
4015 if (value == NULL) {
4016 return 0;
4019 switch (qobject_type(value)) {
4020 case QTYPE_QNULL:
4021 new_backing_bs = NULL;
4022 break;
4023 case QTYPE_QSTRING:
4024 str = qobject_get_try_str(value);
4025 new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
4026 if (new_backing_bs == NULL) {
4027 return -EINVAL;
4028 } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
4029 error_setg(errp, "Making '%s' a backing file of '%s' "
4030 "would create a cycle", str, bs->node_name);
4031 return -EINVAL;
4033 break;
4034 default:
4035 /* 'backing' does not allow any other data type */
4036 g_assert_not_reached();
4040 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4041 * bdrv_reopen_commit() won't fail.
4043 if (new_backing_bs) {
4044 if (!bdrv_reopen_can_attach(bs, bs->backing, new_backing_bs, errp)) {
4045 return -EINVAL;
4050 * Ensure that @bs can really handle backing files, because we are
4051 * about to give it one (or swap the existing one)
4053 if (bs->drv->is_filter) {
4054 /* Filters always have a file or a backing child */
4055 if (!bs->backing) {
4056 error_setg(errp, "'%s' is a %s filter node that does not support a "
4057 "backing child", bs->node_name, bs->drv->format_name);
4058 return -EINVAL;
4060 } else if (!bs->drv->supports_backing) {
4061 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
4062 "files", bs->drv->format_name, bs->node_name);
4063 return -EINVAL;
4067 * Find the "actual" backing file by skipping all links that point
4068 * to an implicit node, if any (e.g. a commit filter node).
4069 * We cannot use any of the bdrv_skip_*() functions here because
4070 * those return the first explicit node, while we are looking for
4071 * its overlay here.
4073 overlay_bs = bs;
4074 for (below_bs = bdrv_filter_or_cow_bs(overlay_bs);
4075 below_bs && below_bs->implicit;
4076 below_bs = bdrv_filter_or_cow_bs(overlay_bs))
4078 overlay_bs = below_bs;
4081 /* If we want to replace the backing file we need some extra checks */
4082 if (new_backing_bs != bdrv_filter_or_cow_bs(overlay_bs)) {
4083 /* Check for implicit nodes between bs and its backing file */
4084 if (bs != overlay_bs) {
4085 error_setg(errp, "Cannot change backing link if '%s' has "
4086 "an implicit backing file", bs->node_name);
4087 return -EPERM;
4090 * Check if the backing link that we want to replace is frozen.
4091 * Note that
4092 * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
4093 * because we know that overlay_bs == bs, and that @bs
4094 * either is a filter that uses ->backing or a COW format BDS
4095 * with bs->drv->supports_backing == true.
4097 if (bdrv_is_backing_chain_frozen(overlay_bs,
4098 child_bs(overlay_bs->backing), errp))
4100 return -EPERM;
4102 reopen_state->replace_backing_bs = true;
4103 if (new_backing_bs) {
4104 bdrv_ref(new_backing_bs);
4105 reopen_state->new_backing_bs = new_backing_bs;
4109 return 0;
4113 * Prepares a BlockDriverState for reopen. All changes are staged in the
4114 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4115 * the block driver layer .bdrv_reopen_prepare()
4117 * bs is the BlockDriverState to reopen
4118 * flags are the new open flags
4119 * queue is the reopen queue
4121 * Returns 0 on success, non-zero on error. On error errp will be set
4122 * as well.
4124 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4125 * It is the responsibility of the caller to then call the abort() or
4126 * commit() for any other BDS that have been left in a prepare() state
4129 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
4130 Error **errp)
4132 int ret = -1;
4133 int old_flags;
4134 Error *local_err = NULL;
4135 BlockDriver *drv;
4136 QemuOpts *opts;
4137 QDict *orig_reopen_opts;
4138 char *discard = NULL;
4139 bool read_only;
4140 bool drv_prepared = false;
4142 assert(reopen_state != NULL);
4143 assert(reopen_state->bs->drv != NULL);
4144 drv = reopen_state->bs->drv;
4146 /* This function and each driver's bdrv_reopen_prepare() remove
4147 * entries from reopen_state->options as they are processed, so
4148 * we need to make a copy of the original QDict. */
4149 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4151 /* Process generic block layer options */
4152 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4153 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4154 ret = -EINVAL;
4155 goto error;
4158 /* This was already called in bdrv_reopen_queue_child() so the flags
4159 * are up-to-date. This time we simply want to remove the options from
4160 * QemuOpts in order to indicate that they have been processed. */
4161 old_flags = reopen_state->flags;
4162 update_flags_from_options(&reopen_state->flags, opts);
4163 assert(old_flags == reopen_state->flags);
4165 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4166 if (discard != NULL) {
4167 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4168 error_setg(errp, "Invalid discard option");
4169 ret = -EINVAL;
4170 goto error;
4174 reopen_state->detect_zeroes =
4175 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4176 if (local_err) {
4177 error_propagate(errp, local_err);
4178 ret = -EINVAL;
4179 goto error;
4182 /* All other options (including node-name and driver) must be unchanged.
4183 * Put them back into the QDict, so that they are checked at the end
4184 * of this function. */
4185 qemu_opts_to_qdict(opts, reopen_state->options);
4187 /* If we are to stay read-only, do not allow permission change
4188 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4189 * not set, or if the BDS still has copy_on_read enabled */
4190 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4191 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4192 if (local_err) {
4193 error_propagate(errp, local_err);
4194 goto error;
4197 /* Calculate required permissions after reopening */
4198 bdrv_reopen_perm(queue, reopen_state->bs,
4199 &reopen_state->perm, &reopen_state->shared_perm);
4201 ret = bdrv_flush(reopen_state->bs);
4202 if (ret) {
4203 error_setg_errno(errp, -ret, "Error flushing drive");
4204 goto error;
4207 if (drv->bdrv_reopen_prepare) {
4209 * If a driver-specific option is missing, it means that we
4210 * should reset it to its default value.
4211 * But not all options allow that, so we need to check it first.
4213 ret = bdrv_reset_options_allowed(reopen_state->bs,
4214 reopen_state->options, errp);
4215 if (ret) {
4216 goto error;
4219 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4220 if (ret) {
4221 if (local_err != NULL) {
4222 error_propagate(errp, local_err);
4223 } else {
4224 bdrv_refresh_filename(reopen_state->bs);
4225 error_setg(errp, "failed while preparing to reopen image '%s'",
4226 reopen_state->bs->filename);
4228 goto error;
4230 } else {
4231 /* It is currently mandatory to have a bdrv_reopen_prepare()
4232 * handler for each supported drv. */
4233 error_setg(errp, "Block format '%s' used by node '%s' "
4234 "does not support reopening files", drv->format_name,
4235 bdrv_get_device_or_node_name(reopen_state->bs));
4236 ret = -1;
4237 goto error;
4240 drv_prepared = true;
4243 * We must provide the 'backing' option if the BDS has a backing
4244 * file or if the image file has a backing file name as part of
4245 * its metadata. Otherwise the 'backing' option can be omitted.
4247 if (drv->supports_backing && reopen_state->backing_missing &&
4248 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4249 error_setg(errp, "backing is missing for '%s'",
4250 reopen_state->bs->node_name);
4251 ret = -EINVAL;
4252 goto error;
4256 * Allow changing the 'backing' option. The new value can be
4257 * either a reference to an existing node (using its node name)
4258 * or NULL to simply detach the current backing file.
4260 ret = bdrv_reopen_parse_backing(reopen_state, errp);
4261 if (ret < 0) {
4262 goto error;
4264 qdict_del(reopen_state->options, "backing");
4266 /* Options that are not handled are only okay if they are unchanged
4267 * compared to the old state. It is expected that some options are only
4268 * used for the initial open, but not reopen (e.g. filename) */
4269 if (qdict_size(reopen_state->options)) {
4270 const QDictEntry *entry = qdict_first(reopen_state->options);
4272 do {
4273 QObject *new = entry->value;
4274 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4276 /* Allow child references (child_name=node_name) as long as they
4277 * point to the current child (i.e. everything stays the same). */
4278 if (qobject_type(new) == QTYPE_QSTRING) {
4279 BdrvChild *child;
4280 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4281 if (!strcmp(child->name, entry->key)) {
4282 break;
4286 if (child) {
4287 const char *str = qobject_get_try_str(new);
4288 if (!strcmp(child->bs->node_name, str)) {
4289 continue; /* Found child with this name, skip option */
4295 * TODO: When using -drive to specify blockdev options, all values
4296 * will be strings; however, when using -blockdev, blockdev-add or
4297 * filenames using the json:{} pseudo-protocol, they will be
4298 * correctly typed.
4299 * In contrast, reopening options are (currently) always strings
4300 * (because you can only specify them through qemu-io; all other
4301 * callers do not specify any options).
4302 * Therefore, when using anything other than -drive to create a BDS,
4303 * this cannot detect non-string options as unchanged, because
4304 * qobject_is_equal() always returns false for objects of different
4305 * type. In the future, this should be remedied by correctly typing
4306 * all options. For now, this is not too big of an issue because
4307 * the user can simply omit options which cannot be changed anyway,
4308 * so they will stay unchanged.
4310 if (!qobject_is_equal(new, old)) {
4311 error_setg(errp, "Cannot change the option '%s'", entry->key);
4312 ret = -EINVAL;
4313 goto error;
4315 } while ((entry = qdict_next(reopen_state->options, entry)));
4318 ret = 0;
4320 /* Restore the original reopen_state->options QDict */
4321 qobject_unref(reopen_state->options);
4322 reopen_state->options = qobject_ref(orig_reopen_opts);
4324 error:
4325 if (ret < 0 && drv_prepared) {
4326 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4327 * call drv->bdrv_reopen_abort() before signaling an error
4328 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4329 * when the respective bdrv_reopen_prepare() has failed) */
4330 if (drv->bdrv_reopen_abort) {
4331 drv->bdrv_reopen_abort(reopen_state);
4334 qemu_opts_del(opts);
4335 qobject_unref(orig_reopen_opts);
4336 g_free(discard);
4337 return ret;
4341 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4342 * makes them final by swapping the staging BlockDriverState contents into
4343 * the active BlockDriverState contents.
4345 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4347 BlockDriver *drv;
4348 BlockDriverState *bs;
4349 BdrvChild *child;
4351 assert(reopen_state != NULL);
4352 bs = reopen_state->bs;
4353 drv = bs->drv;
4354 assert(drv != NULL);
4356 /* If there are any driver level actions to take */
4357 if (drv->bdrv_reopen_commit) {
4358 drv->bdrv_reopen_commit(reopen_state);
4361 /* set BDS specific flags now */
4362 qobject_unref(bs->explicit_options);
4363 qobject_unref(bs->options);
4365 bs->explicit_options = reopen_state->explicit_options;
4366 bs->options = reopen_state->options;
4367 bs->open_flags = reopen_state->flags;
4368 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
4369 bs->detect_zeroes = reopen_state->detect_zeroes;
4371 if (reopen_state->replace_backing_bs) {
4372 qdict_del(bs->explicit_options, "backing");
4373 qdict_del(bs->options, "backing");
4376 /* Remove child references from bs->options and bs->explicit_options.
4377 * Child options were already removed in bdrv_reopen_queue_child() */
4378 QLIST_FOREACH(child, &bs->children, next) {
4379 qdict_del(bs->explicit_options, child->name);
4380 qdict_del(bs->options, child->name);
4384 * Change the backing file if a new one was specified. We do this
4385 * after updating bs->options, so bdrv_refresh_filename() (called
4386 * from bdrv_set_backing_hd()) has the new values.
4388 if (reopen_state->replace_backing_bs) {
4389 BlockDriverState *old_backing_bs = child_bs(bs->backing);
4390 assert(!old_backing_bs || !old_backing_bs->implicit);
4391 /* Abort the permission update on the backing bs we're detaching */
4392 if (old_backing_bs) {
4393 bdrv_abort_perm_update(old_backing_bs);
4395 bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort);
4398 bdrv_refresh_limits(bs, NULL);
4402 * Abort the reopen, and delete and free the staged changes in
4403 * reopen_state
4405 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4407 BlockDriver *drv;
4409 assert(reopen_state != NULL);
4410 drv = reopen_state->bs->drv;
4411 assert(drv != NULL);
4413 if (drv->bdrv_reopen_abort) {
4414 drv->bdrv_reopen_abort(reopen_state);
4419 static void bdrv_close(BlockDriverState *bs)
4421 BdrvAioNotifier *ban, *ban_next;
4422 BdrvChild *child, *next;
4424 assert(!bs->refcnt);
4426 bdrv_drained_begin(bs); /* complete I/O */
4427 bdrv_flush(bs);
4428 bdrv_drain(bs); /* in case flush left pending I/O */
4430 if (bs->drv) {
4431 if (bs->drv->bdrv_close) {
4432 /* Must unfreeze all children, so bdrv_unref_child() works */
4433 bs->drv->bdrv_close(bs);
4435 bs->drv = NULL;
4438 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4439 bdrv_unref_child(bs, child);
4442 bs->backing = NULL;
4443 bs->file = NULL;
4444 g_free(bs->opaque);
4445 bs->opaque = NULL;
4446 qatomic_set(&bs->copy_on_read, 0);
4447 bs->backing_file[0] = '\0';
4448 bs->backing_format[0] = '\0';
4449 bs->total_sectors = 0;
4450 bs->encrypted = false;
4451 bs->sg = false;
4452 qobject_unref(bs->options);
4453 qobject_unref(bs->explicit_options);
4454 bs->options = NULL;
4455 bs->explicit_options = NULL;
4456 qobject_unref(bs->full_open_options);
4457 bs->full_open_options = NULL;
4459 bdrv_release_named_dirty_bitmaps(bs);
4460 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4462 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4463 g_free(ban);
4465 QLIST_INIT(&bs->aio_notifiers);
4466 bdrv_drained_end(bs);
4469 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4470 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4471 * gets called.
4473 if (bs->quiesce_counter) {
4474 bdrv_drain_all_end_quiesce(bs);
4478 void bdrv_close_all(void)
4480 assert(job_next(NULL) == NULL);
4481 blk_exp_close_all();
4483 /* Drop references from requests still in flight, such as canceled block
4484 * jobs whose AIO context has not been polled yet */
4485 bdrv_drain_all();
4487 blk_remove_all_bs();
4488 blockdev_close_all_bdrv_states();
4490 assert(QTAILQ_EMPTY(&all_bdrv_states));
4493 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4495 GQueue *queue;
4496 GHashTable *found;
4497 bool ret;
4499 if (c->klass->stay_at_node) {
4500 return false;
4503 /* If the child @c belongs to the BDS @to, replacing the current
4504 * c->bs by @to would mean to create a loop.
4506 * Such a case occurs when appending a BDS to a backing chain.
4507 * For instance, imagine the following chain:
4509 * guest device -> node A -> further backing chain...
4511 * Now we create a new BDS B which we want to put on top of this
4512 * chain, so we first attach A as its backing node:
4514 * node B
4517 * guest device -> node A -> further backing chain...
4519 * Finally we want to replace A by B. When doing that, we want to
4520 * replace all pointers to A by pointers to B -- except for the
4521 * pointer from B because (1) that would create a loop, and (2)
4522 * that pointer should simply stay intact:
4524 * guest device -> node B
4527 * node A -> further backing chain...
4529 * In general, when replacing a node A (c->bs) by a node B (@to),
4530 * if A is a child of B, that means we cannot replace A by B there
4531 * because that would create a loop. Silently detaching A from B
4532 * is also not really an option. So overall just leaving A in
4533 * place there is the most sensible choice.
4535 * We would also create a loop in any cases where @c is only
4536 * indirectly referenced by @to. Prevent this by returning false
4537 * if @c is found (by breadth-first search) anywhere in the whole
4538 * subtree of @to.
4541 ret = true;
4542 found = g_hash_table_new(NULL, NULL);
4543 g_hash_table_add(found, to);
4544 queue = g_queue_new();
4545 g_queue_push_tail(queue, to);
4547 while (!g_queue_is_empty(queue)) {
4548 BlockDriverState *v = g_queue_pop_head(queue);
4549 BdrvChild *c2;
4551 QLIST_FOREACH(c2, &v->children, next) {
4552 if (c2 == c) {
4553 ret = false;
4554 break;
4557 if (g_hash_table_contains(found, c2->bs)) {
4558 continue;
4561 g_queue_push_tail(queue, c2->bs);
4562 g_hash_table_add(found, c2->bs);
4566 g_queue_free(queue);
4567 g_hash_table_destroy(found);
4569 return ret;
4573 * With auto_skip=true bdrv_replace_node_common skips updating from parents
4574 * if it creates a parent-child relation loop or if parent is block-job.
4576 * With auto_skip=false the error is returned if from has a parent which should
4577 * not be updated.
4579 static void bdrv_replace_node_common(BlockDriverState *from,
4580 BlockDriverState *to,
4581 bool auto_skip, Error **errp)
4583 BdrvChild *c, *next;
4584 GSList *list = NULL, *p;
4585 uint64_t perm = 0, shared = BLK_PERM_ALL;
4586 int ret;
4588 /* Make sure that @from doesn't go away until we have successfully attached
4589 * all of its parents to @to. */
4590 bdrv_ref(from);
4592 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4593 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
4594 bdrv_drained_begin(from);
4596 /* Put all parents into @list and calculate their cumulative permissions */
4597 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4598 assert(c->bs == from);
4599 if (!should_update_child(c, to)) {
4600 if (auto_skip) {
4601 continue;
4603 error_setg(errp, "Should not change '%s' link to '%s'",
4604 c->name, from->node_name);
4605 goto out;
4607 if (c->frozen) {
4608 error_setg(errp, "Cannot change '%s' link to '%s'",
4609 c->name, from->node_name);
4610 goto out;
4612 list = g_slist_prepend(list, c);
4613 perm |= c->perm;
4614 shared &= c->shared_perm;
4617 /* Check whether the required permissions can be granted on @to, ignoring
4618 * all BdrvChild in @list so that they can't block themselves. */
4619 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, NULL, errp);
4620 if (ret < 0) {
4621 bdrv_abort_perm_update(to);
4622 goto out;
4625 /* Now actually perform the change. We performed the permission check for
4626 * all elements of @list at once, so set the permissions all at once at the
4627 * very end. */
4628 for (p = list; p != NULL; p = p->next) {
4629 c = p->data;
4631 bdrv_ref(to);
4632 bdrv_replace_child_noperm(c, to);
4633 bdrv_unref(from);
4636 bdrv_get_cumulative_perm(to, &perm, &shared);
4637 bdrv_set_perm(to, perm, shared);
4639 out:
4640 g_slist_free(list);
4641 bdrv_drained_end(from);
4642 bdrv_unref(from);
4645 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
4646 Error **errp)
4648 return bdrv_replace_node_common(from, to, true, errp);
4652 * Add new bs contents at the top of an image chain while the chain is
4653 * live, while keeping required fields on the top layer.
4655 * This will modify the BlockDriverState fields, and swap contents
4656 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4658 * bs_new must not be attached to a BlockBackend.
4660 * This function does not create any image files.
4662 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4663 * that's what the callers commonly need. bs_new will be referenced by the old
4664 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4665 * reference of its own, it must call bdrv_ref().
4667 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
4668 Error **errp)
4670 Error *local_err = NULL;
4672 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
4673 if (local_err) {
4674 error_propagate(errp, local_err);
4675 goto out;
4678 bdrv_replace_node(bs_top, bs_new, &local_err);
4679 if (local_err) {
4680 error_propagate(errp, local_err);
4681 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
4682 goto out;
4685 /* bs_new is now referenced by its new parents, we don't need the
4686 * additional reference any more. */
4687 out:
4688 bdrv_unref(bs_new);
4691 static void bdrv_delete(BlockDriverState *bs)
4693 assert(bdrv_op_blocker_is_empty(bs));
4694 assert(!bs->refcnt);
4696 /* remove from list, if necessary */
4697 if (bs->node_name[0] != '\0') {
4698 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
4700 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
4702 bdrv_close(bs);
4704 g_free(bs);
4708 * Run consistency checks on an image
4710 * Returns 0 if the check could be completed (it doesn't mean that the image is
4711 * free of errors) or -errno when an internal error occurred. The results of the
4712 * check are stored in res.
4714 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
4715 BdrvCheckResult *res, BdrvCheckMode fix)
4717 if (bs->drv == NULL) {
4718 return -ENOMEDIUM;
4720 if (bs->drv->bdrv_co_check == NULL) {
4721 return -ENOTSUP;
4724 memset(res, 0, sizeof(*res));
4725 return bs->drv->bdrv_co_check(bs, res, fix);
4729 * Return values:
4730 * 0 - success
4731 * -EINVAL - backing format specified, but no file
4732 * -ENOSPC - can't update the backing file because no space is left in the
4733 * image file header
4734 * -ENOTSUP - format driver doesn't support changing the backing file
4736 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
4737 const char *backing_fmt, bool warn)
4739 BlockDriver *drv = bs->drv;
4740 int ret;
4742 if (!drv) {
4743 return -ENOMEDIUM;
4746 /* Backing file format doesn't make sense without a backing file */
4747 if (backing_fmt && !backing_file) {
4748 return -EINVAL;
4751 if (warn && backing_file && !backing_fmt) {
4752 warn_report("Deprecated use of backing file without explicit "
4753 "backing format, use of this image requires "
4754 "potentially unsafe format probing");
4757 if (drv->bdrv_change_backing_file != NULL) {
4758 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
4759 } else {
4760 ret = -ENOTSUP;
4763 if (ret == 0) {
4764 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
4765 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
4766 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
4767 backing_file ?: "");
4769 return ret;
4773 * Finds the first non-filter node above bs in the chain between
4774 * active and bs. The returned node is either an immediate parent of
4775 * bs, or there are only filter nodes between the two.
4777 * Returns NULL if bs is not found in active's image chain,
4778 * or if active == bs.
4780 * Returns the bottommost base image if bs == NULL.
4782 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
4783 BlockDriverState *bs)
4785 bs = bdrv_skip_filters(bs);
4786 active = bdrv_skip_filters(active);
4788 while (active) {
4789 BlockDriverState *next = bdrv_backing_chain_next(active);
4790 if (bs == next) {
4791 return active;
4793 active = next;
4796 return NULL;
4799 /* Given a BDS, searches for the base layer. */
4800 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4802 return bdrv_find_overlay(bs, NULL);
4806 * Return true if at least one of the COW (backing) and filter links
4807 * between @bs and @base is frozen. @errp is set if that's the case.
4808 * @base must be reachable from @bs, or NULL.
4810 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
4811 Error **errp)
4813 BlockDriverState *i;
4814 BdrvChild *child;
4816 for (i = bs; i != base; i = child_bs(child)) {
4817 child = bdrv_filter_or_cow_child(i);
4819 if (child && child->frozen) {
4820 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4821 child->name, i->node_name, child->bs->node_name);
4822 return true;
4826 return false;
4830 * Freeze all COW (backing) and filter links between @bs and @base.
4831 * If any of the links is already frozen the operation is aborted and
4832 * none of the links are modified.
4833 * @base must be reachable from @bs, or NULL.
4834 * Returns 0 on success. On failure returns < 0 and sets @errp.
4836 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4837 Error **errp)
4839 BlockDriverState *i;
4840 BdrvChild *child;
4842 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4843 return -EPERM;
4846 for (i = bs; i != base; i = child_bs(child)) {
4847 child = bdrv_filter_or_cow_child(i);
4848 if (child && child->bs->never_freeze) {
4849 error_setg(errp, "Cannot freeze '%s' link to '%s'",
4850 child->name, child->bs->node_name);
4851 return -EPERM;
4855 for (i = bs; i != base; i = child_bs(child)) {
4856 child = bdrv_filter_or_cow_child(i);
4857 if (child) {
4858 child->frozen = true;
4862 return 0;
4866 * Unfreeze all COW (backing) and filter links between @bs and @base.
4867 * The caller must ensure that all links are frozen before using this
4868 * function.
4869 * @base must be reachable from @bs, or NULL.
4871 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4873 BlockDriverState *i;
4874 BdrvChild *child;
4876 for (i = bs; i != base; i = child_bs(child)) {
4877 child = bdrv_filter_or_cow_child(i);
4878 if (child) {
4879 assert(child->frozen);
4880 child->frozen = false;
4886 * Drops images above 'base' up to and including 'top', and sets the image
4887 * above 'top' to have base as its backing file.
4889 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4890 * information in 'bs' can be properly updated.
4892 * E.g., this will convert the following chain:
4893 * bottom <- base <- intermediate <- top <- active
4895 * to
4897 * bottom <- base <- active
4899 * It is allowed for bottom==base, in which case it converts:
4901 * base <- intermediate <- top <- active
4903 * to
4905 * base <- active
4907 * If backing_file_str is non-NULL, it will be used when modifying top's
4908 * overlay image metadata.
4910 * Error conditions:
4911 * if active == top, that is considered an error
4914 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4915 const char *backing_file_str)
4917 BlockDriverState *explicit_top = top;
4918 bool update_inherits_from;
4919 BdrvChild *c;
4920 Error *local_err = NULL;
4921 int ret = -EIO;
4922 g_autoptr(GSList) updated_children = NULL;
4923 GSList *p;
4925 bdrv_ref(top);
4926 bdrv_subtree_drained_begin(top);
4928 if (!top->drv || !base->drv) {
4929 goto exit;
4932 /* Make sure that base is in the backing chain of top */
4933 if (!bdrv_chain_contains(top, base)) {
4934 goto exit;
4937 /* If 'base' recursively inherits from 'top' then we should set
4938 * base->inherits_from to top->inherits_from after 'top' and all
4939 * other intermediate nodes have been dropped.
4940 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4941 * it because no one inherits from it. We use explicit_top for that. */
4942 explicit_top = bdrv_skip_implicit_filters(explicit_top);
4943 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4945 /* success - we can delete the intermediate states, and link top->base */
4946 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4947 * we've figured out how they should work. */
4948 if (!backing_file_str) {
4949 bdrv_refresh_filename(base);
4950 backing_file_str = base->filename;
4953 QLIST_FOREACH(c, &top->parents, next_parent) {
4954 updated_children = g_slist_prepend(updated_children, c);
4957 bdrv_replace_node_common(top, base, false, &local_err);
4958 if (local_err) {
4959 error_report_err(local_err);
4960 goto exit;
4963 for (p = updated_children; p; p = p->next) {
4964 c = p->data;
4966 if (c->klass->update_filename) {
4967 ret = c->klass->update_filename(c, base, backing_file_str,
4968 &local_err);
4969 if (ret < 0) {
4971 * TODO: Actually, we want to rollback all previous iterations
4972 * of this loop, and (which is almost impossible) previous
4973 * bdrv_replace_node()...
4975 * Note, that c->klass->update_filename may lead to permission
4976 * update, so it's a bad idea to call it inside permission
4977 * update transaction of bdrv_replace_node.
4979 error_report_err(local_err);
4980 goto exit;
4985 if (update_inherits_from) {
4986 base->inherits_from = explicit_top->inherits_from;
4989 ret = 0;
4990 exit:
4991 bdrv_subtree_drained_end(top);
4992 bdrv_unref(top);
4993 return ret;
4997 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
4998 * sums the size of all data-bearing children. (This excludes backing
4999 * children.)
5001 static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5003 BdrvChild *child;
5004 int64_t child_size, sum = 0;
5006 QLIST_FOREACH(child, &bs->children, next) {
5007 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5008 BDRV_CHILD_FILTERED))
5010 child_size = bdrv_get_allocated_file_size(child->bs);
5011 if (child_size < 0) {
5012 return child_size;
5014 sum += child_size;
5018 return sum;
5022 * Length of a allocated file in bytes. Sparse files are counted by actual
5023 * allocated space. Return < 0 if error or unknown.
5025 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
5027 BlockDriver *drv = bs->drv;
5028 if (!drv) {
5029 return -ENOMEDIUM;
5031 if (drv->bdrv_get_allocated_file_size) {
5032 return drv->bdrv_get_allocated_file_size(bs);
5035 if (drv->bdrv_file_open) {
5037 * Protocol drivers default to -ENOTSUP (most of their data is
5038 * not stored in any of their children (if they even have any),
5039 * so there is no generic way to figure it out).
5041 return -ENOTSUP;
5042 } else if (drv->is_filter) {
5043 /* Filter drivers default to the size of their filtered child */
5044 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5045 } else {
5046 /* Other drivers default to summing their children's sizes */
5047 return bdrv_sum_allocated_file_size(bs);
5052 * bdrv_measure:
5053 * @drv: Format driver
5054 * @opts: Creation options for new image
5055 * @in_bs: Existing image containing data for new image (may be NULL)
5056 * @errp: Error object
5057 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5058 * or NULL on error
5060 * Calculate file size required to create a new image.
5062 * If @in_bs is given then space for allocated clusters and zero clusters
5063 * from that image are included in the calculation. If @opts contains a
5064 * backing file that is shared by @in_bs then backing clusters may be omitted
5065 * from the calculation.
5067 * If @in_bs is NULL then the calculation includes no allocated clusters
5068 * unless a preallocation option is given in @opts.
5070 * Note that @in_bs may use a different BlockDriver from @drv.
5072 * If an error occurs the @errp pointer is set.
5074 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5075 BlockDriverState *in_bs, Error **errp)
5077 if (!drv->bdrv_measure) {
5078 error_setg(errp, "Block driver '%s' does not support size measurement",
5079 drv->format_name);
5080 return NULL;
5083 return drv->bdrv_measure(opts, in_bs, errp);
5087 * Return number of sectors on success, -errno on error.
5089 int64_t bdrv_nb_sectors(BlockDriverState *bs)
5091 BlockDriver *drv = bs->drv;
5093 if (!drv)
5094 return -ENOMEDIUM;
5096 if (drv->has_variable_length) {
5097 int ret = refresh_total_sectors(bs, bs->total_sectors);
5098 if (ret < 0) {
5099 return ret;
5102 return bs->total_sectors;
5106 * Return length in bytes on success, -errno on error.
5107 * The length is always a multiple of BDRV_SECTOR_SIZE.
5109 int64_t bdrv_getlength(BlockDriverState *bs)
5111 int64_t ret = bdrv_nb_sectors(bs);
5113 if (ret < 0) {
5114 return ret;
5116 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5117 return -EFBIG;
5119 return ret * BDRV_SECTOR_SIZE;
5122 /* return 0 as number of sectors if no device present or error */
5123 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
5125 int64_t nb_sectors = bdrv_nb_sectors(bs);
5127 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
5130 bool bdrv_is_sg(BlockDriverState *bs)
5132 return bs->sg;
5136 * Return whether the given node supports compressed writes.
5138 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5140 BlockDriverState *filtered;
5142 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5143 return false;
5146 filtered = bdrv_filter_bs(bs);
5147 if (filtered) {
5149 * Filters can only forward compressed writes, so we have to
5150 * check the child.
5152 return bdrv_supports_compressed_writes(filtered);
5155 return true;
5158 const char *bdrv_get_format_name(BlockDriverState *bs)
5160 return bs->drv ? bs->drv->format_name : NULL;
5163 static int qsort_strcmp(const void *a, const void *b)
5165 return strcmp(*(char *const *)a, *(char *const *)b);
5168 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
5169 void *opaque, bool read_only)
5171 BlockDriver *drv;
5172 int count = 0;
5173 int i;
5174 const char **formats = NULL;
5176 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5177 if (drv->format_name) {
5178 bool found = false;
5179 int i = count;
5181 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5182 continue;
5185 while (formats && i && !found) {
5186 found = !strcmp(formats[--i], drv->format_name);
5189 if (!found) {
5190 formats = g_renew(const char *, formats, count + 1);
5191 formats[count++] = drv->format_name;
5196 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5197 const char *format_name = block_driver_modules[i].format_name;
5199 if (format_name) {
5200 bool found = false;
5201 int j = count;
5203 if (use_bdrv_whitelist &&
5204 !bdrv_format_is_whitelisted(format_name, read_only)) {
5205 continue;
5208 while (formats && j && !found) {
5209 found = !strcmp(formats[--j], format_name);
5212 if (!found) {
5213 formats = g_renew(const char *, formats, count + 1);
5214 formats[count++] = format_name;
5219 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
5221 for (i = 0; i < count; i++) {
5222 it(opaque, formats[i]);
5225 g_free(formats);
5228 /* This function is to find a node in the bs graph */
5229 BlockDriverState *bdrv_find_node(const char *node_name)
5231 BlockDriverState *bs;
5233 assert(node_name);
5235 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5236 if (!strcmp(node_name, bs->node_name)) {
5237 return bs;
5240 return NULL;
5243 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5244 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5245 Error **errp)
5247 BlockDeviceInfoList *list;
5248 BlockDriverState *bs;
5250 list = NULL;
5251 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5252 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
5253 if (!info) {
5254 qapi_free_BlockDeviceInfoList(list);
5255 return NULL;
5257 QAPI_LIST_PREPEND(list, info);
5260 return list;
5263 typedef struct XDbgBlockGraphConstructor {
5264 XDbgBlockGraph *graph;
5265 GHashTable *graph_nodes;
5266 } XDbgBlockGraphConstructor;
5268 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5270 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5272 gr->graph = g_new0(XDbgBlockGraph, 1);
5273 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5275 return gr;
5278 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5280 XDbgBlockGraph *graph = gr->graph;
5282 g_hash_table_destroy(gr->graph_nodes);
5283 g_free(gr);
5285 return graph;
5288 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5290 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5292 if (ret != 0) {
5293 return ret;
5297 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5298 * answer of g_hash_table_lookup.
5300 ret = g_hash_table_size(gr->graph_nodes) + 1;
5301 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5303 return ret;
5306 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5307 XDbgBlockGraphNodeType type, const char *name)
5309 XDbgBlockGraphNode *n;
5311 n = g_new0(XDbgBlockGraphNode, 1);
5313 n->id = xdbg_graph_node_num(gr, node);
5314 n->type = type;
5315 n->name = g_strdup(name);
5317 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5320 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5321 const BdrvChild *child)
5323 BlockPermission qapi_perm;
5324 XDbgBlockGraphEdge *edge;
5326 edge = g_new0(XDbgBlockGraphEdge, 1);
5328 edge->parent = xdbg_graph_node_num(gr, parent);
5329 edge->child = xdbg_graph_node_num(gr, child->bs);
5330 edge->name = g_strdup(child->name);
5332 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5333 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5335 if (flag & child->perm) {
5336 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5338 if (flag & child->shared_perm) {
5339 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5343 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5347 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5349 BlockBackend *blk;
5350 BlockJob *job;
5351 BlockDriverState *bs;
5352 BdrvChild *child;
5353 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5355 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5356 char *allocated_name = NULL;
5357 const char *name = blk_name(blk);
5359 if (!*name) {
5360 name = allocated_name = blk_get_attached_dev_id(blk);
5362 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5363 name);
5364 g_free(allocated_name);
5365 if (blk_root(blk)) {
5366 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5370 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5371 GSList *el;
5373 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5374 job->job.id);
5375 for (el = job->nodes; el; el = el->next) {
5376 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5380 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5381 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5382 bs->node_name);
5383 QLIST_FOREACH(child, &bs->children, next) {
5384 xdbg_graph_add_edge(gr, bs, child);
5388 return xdbg_graph_finalize(gr);
5391 BlockDriverState *bdrv_lookup_bs(const char *device,
5392 const char *node_name,
5393 Error **errp)
5395 BlockBackend *blk;
5396 BlockDriverState *bs;
5398 if (device) {
5399 blk = blk_by_name(device);
5401 if (blk) {
5402 bs = blk_bs(blk);
5403 if (!bs) {
5404 error_setg(errp, "Device '%s' has no medium", device);
5407 return bs;
5411 if (node_name) {
5412 bs = bdrv_find_node(node_name);
5414 if (bs) {
5415 return bs;
5419 error_setg(errp, "Cannot find device=%s nor node_name=%s",
5420 device ? device : "",
5421 node_name ? node_name : "");
5422 return NULL;
5425 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5426 * return false. If either argument is NULL, return false. */
5427 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
5429 while (top && top != base) {
5430 top = bdrv_filter_or_cow_bs(top);
5433 return top != NULL;
5436 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
5438 if (!bs) {
5439 return QTAILQ_FIRST(&graph_bdrv_states);
5441 return QTAILQ_NEXT(bs, node_list);
5444 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5446 if (!bs) {
5447 return QTAILQ_FIRST(&all_bdrv_states);
5449 return QTAILQ_NEXT(bs, bs_list);
5452 const char *bdrv_get_node_name(const BlockDriverState *bs)
5454 return bs->node_name;
5457 const char *bdrv_get_parent_name(const BlockDriverState *bs)
5459 BdrvChild *c;
5460 const char *name;
5462 /* If multiple parents have a name, just pick the first one. */
5463 QLIST_FOREACH(c, &bs->parents, next_parent) {
5464 if (c->klass->get_name) {
5465 name = c->klass->get_name(c);
5466 if (name && *name) {
5467 return name;
5472 return NULL;
5475 /* TODO check what callers really want: bs->node_name or blk_name() */
5476 const char *bdrv_get_device_name(const BlockDriverState *bs)
5478 return bdrv_get_parent_name(bs) ?: "";
5481 /* This can be used to identify nodes that might not have a device
5482 * name associated. Since node and device names live in the same
5483 * namespace, the result is unambiguous. The exception is if both are
5484 * absent, then this returns an empty (non-null) string. */
5485 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
5487 return bdrv_get_parent_name(bs) ?: bs->node_name;
5490 int bdrv_get_flags(BlockDriverState *bs)
5492 return bs->open_flags;
5495 int bdrv_has_zero_init_1(BlockDriverState *bs)
5497 return 1;
5500 int bdrv_has_zero_init(BlockDriverState *bs)
5502 BlockDriverState *filtered;
5504 if (!bs->drv) {
5505 return 0;
5508 /* If BS is a copy on write image, it is initialized to
5509 the contents of the base image, which may not be zeroes. */
5510 if (bdrv_cow_child(bs)) {
5511 return 0;
5513 if (bs->drv->bdrv_has_zero_init) {
5514 return bs->drv->bdrv_has_zero_init(bs);
5517 filtered = bdrv_filter_bs(bs);
5518 if (filtered) {
5519 return bdrv_has_zero_init(filtered);
5522 /* safe default */
5523 return 0;
5526 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
5528 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5529 return false;
5532 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
5535 void bdrv_get_backing_filename(BlockDriverState *bs,
5536 char *filename, int filename_size)
5538 pstrcpy(filename, filename_size, bs->backing_file);
5541 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5543 int ret;
5544 BlockDriver *drv = bs->drv;
5545 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5546 if (!drv) {
5547 return -ENOMEDIUM;
5549 if (!drv->bdrv_get_info) {
5550 BlockDriverState *filtered = bdrv_filter_bs(bs);
5551 if (filtered) {
5552 return bdrv_get_info(filtered, bdi);
5554 return -ENOTSUP;
5556 memset(bdi, 0, sizeof(*bdi));
5557 ret = drv->bdrv_get_info(bs, bdi);
5558 if (ret < 0) {
5559 return ret;
5562 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
5563 return -EINVAL;
5566 return 0;
5569 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
5570 Error **errp)
5572 BlockDriver *drv = bs->drv;
5573 if (drv && drv->bdrv_get_specific_info) {
5574 return drv->bdrv_get_specific_info(bs, errp);
5576 return NULL;
5579 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
5581 BlockDriver *drv = bs->drv;
5582 if (!drv || !drv->bdrv_get_specific_stats) {
5583 return NULL;
5585 return drv->bdrv_get_specific_stats(bs);
5588 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
5590 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
5591 return;
5594 bs->drv->bdrv_debug_event(bs, event);
5597 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
5599 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
5600 bs = bdrv_primary_bs(bs);
5603 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
5604 assert(bs->drv->bdrv_debug_remove_breakpoint);
5605 return bs;
5608 return NULL;
5611 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
5612 const char *tag)
5614 bs = bdrv_find_debug_node(bs);
5615 if (bs) {
5616 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
5619 return -ENOTSUP;
5622 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
5624 bs = bdrv_find_debug_node(bs);
5625 if (bs) {
5626 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
5629 return -ENOTSUP;
5632 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
5634 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
5635 bs = bdrv_primary_bs(bs);
5638 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
5639 return bs->drv->bdrv_debug_resume(bs, tag);
5642 return -ENOTSUP;
5645 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
5647 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
5648 bs = bdrv_primary_bs(bs);
5651 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
5652 return bs->drv->bdrv_debug_is_suspended(bs, tag);
5655 return false;
5658 /* backing_file can either be relative, or absolute, or a protocol. If it is
5659 * relative, it must be relative to the chain. So, passing in bs->filename
5660 * from a BDS as backing_file should not be done, as that may be relative to
5661 * the CWD rather than the chain. */
5662 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
5663 const char *backing_file)
5665 char *filename_full = NULL;
5666 char *backing_file_full = NULL;
5667 char *filename_tmp = NULL;
5668 int is_protocol = 0;
5669 bool filenames_refreshed = false;
5670 BlockDriverState *curr_bs = NULL;
5671 BlockDriverState *retval = NULL;
5672 BlockDriverState *bs_below;
5674 if (!bs || !bs->drv || !backing_file) {
5675 return NULL;
5678 filename_full = g_malloc(PATH_MAX);
5679 backing_file_full = g_malloc(PATH_MAX);
5681 is_protocol = path_has_protocol(backing_file);
5684 * Being largely a legacy function, skip any filters here
5685 * (because filters do not have normal filenames, so they cannot
5686 * match anyway; and allowing json:{} filenames is a bit out of
5687 * scope).
5689 for (curr_bs = bdrv_skip_filters(bs);
5690 bdrv_cow_child(curr_bs) != NULL;
5691 curr_bs = bs_below)
5693 bs_below = bdrv_backing_chain_next(curr_bs);
5695 if (bdrv_backing_overridden(curr_bs)) {
5697 * If the backing file was overridden, we can only compare
5698 * directly against the backing node's filename.
5701 if (!filenames_refreshed) {
5703 * This will automatically refresh all of the
5704 * filenames in the rest of the backing chain, so we
5705 * only need to do this once.
5707 bdrv_refresh_filename(bs_below);
5708 filenames_refreshed = true;
5711 if (strcmp(backing_file, bs_below->filename) == 0) {
5712 retval = bs_below;
5713 break;
5715 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
5717 * If either of the filename paths is actually a protocol, then
5718 * compare unmodified paths; otherwise make paths relative.
5720 char *backing_file_full_ret;
5722 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
5723 retval = bs_below;
5724 break;
5726 /* Also check against the full backing filename for the image */
5727 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
5728 NULL);
5729 if (backing_file_full_ret) {
5730 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
5731 g_free(backing_file_full_ret);
5732 if (equal) {
5733 retval = bs_below;
5734 break;
5737 } else {
5738 /* If not an absolute filename path, make it relative to the current
5739 * image's filename path */
5740 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
5741 NULL);
5742 /* We are going to compare canonicalized absolute pathnames */
5743 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
5744 g_free(filename_tmp);
5745 continue;
5747 g_free(filename_tmp);
5749 /* We need to make sure the backing filename we are comparing against
5750 * is relative to the current image filename (or absolute) */
5751 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
5752 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
5753 g_free(filename_tmp);
5754 continue;
5756 g_free(filename_tmp);
5758 if (strcmp(backing_file_full, filename_full) == 0) {
5759 retval = bs_below;
5760 break;
5765 g_free(filename_full);
5766 g_free(backing_file_full);
5767 return retval;
5770 void bdrv_init(void)
5772 module_call_init(MODULE_INIT_BLOCK);
5775 void bdrv_init_with_whitelist(void)
5777 use_bdrv_whitelist = 1;
5778 bdrv_init();
5781 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
5783 BdrvChild *child, *parent;
5784 uint64_t perm, shared_perm;
5785 Error *local_err = NULL;
5786 int ret;
5787 BdrvDirtyBitmap *bm;
5789 if (!bs->drv) {
5790 return -ENOMEDIUM;
5793 QLIST_FOREACH(child, &bs->children, next) {
5794 bdrv_co_invalidate_cache(child->bs, &local_err);
5795 if (local_err) {
5796 error_propagate(errp, local_err);
5797 return -EINVAL;
5802 * Update permissions, they may differ for inactive nodes.
5804 * Note that the required permissions of inactive images are always a
5805 * subset of the permissions required after activating the image. This
5806 * allows us to just get the permissions upfront without restricting
5807 * drv->bdrv_invalidate_cache().
5809 * It also means that in error cases, we don't have to try and revert to
5810 * the old permissions (which is an operation that could fail, too). We can
5811 * just keep the extended permissions for the next time that an activation
5812 * of the image is tried.
5814 if (bs->open_flags & BDRV_O_INACTIVE) {
5815 bs->open_flags &= ~BDRV_O_INACTIVE;
5816 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5817 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, NULL, errp);
5818 if (ret < 0) {
5819 bdrv_abort_perm_update(bs);
5820 bs->open_flags |= BDRV_O_INACTIVE;
5821 return ret;
5823 bdrv_set_perm(bs, perm, shared_perm);
5825 if (bs->drv->bdrv_co_invalidate_cache) {
5826 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
5827 if (local_err) {
5828 bs->open_flags |= BDRV_O_INACTIVE;
5829 error_propagate(errp, local_err);
5830 return -EINVAL;
5834 FOR_EACH_DIRTY_BITMAP(bs, bm) {
5835 bdrv_dirty_bitmap_skip_store(bm, false);
5838 ret = refresh_total_sectors(bs, bs->total_sectors);
5839 if (ret < 0) {
5840 bs->open_flags |= BDRV_O_INACTIVE;
5841 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5842 return ret;
5846 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5847 if (parent->klass->activate) {
5848 parent->klass->activate(parent, &local_err);
5849 if (local_err) {
5850 bs->open_flags |= BDRV_O_INACTIVE;
5851 error_propagate(errp, local_err);
5852 return -EINVAL;
5857 return 0;
5860 void bdrv_invalidate_cache_all(Error **errp)
5862 BlockDriverState *bs;
5863 BdrvNextIterator it;
5865 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5866 AioContext *aio_context = bdrv_get_aio_context(bs);
5867 int ret;
5869 aio_context_acquire(aio_context);
5870 ret = bdrv_invalidate_cache(bs, errp);
5871 aio_context_release(aio_context);
5872 if (ret < 0) {
5873 bdrv_next_cleanup(&it);
5874 return;
5879 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5881 BdrvChild *parent;
5883 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5884 if (parent->klass->parent_is_bds) {
5885 BlockDriverState *parent_bs = parent->opaque;
5886 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5887 return true;
5892 return false;
5895 static int bdrv_inactivate_recurse(BlockDriverState *bs)
5897 BdrvChild *child, *parent;
5898 bool tighten_restrictions;
5899 uint64_t perm, shared_perm;
5900 int ret;
5902 if (!bs->drv) {
5903 return -ENOMEDIUM;
5906 /* Make sure that we don't inactivate a child before its parent.
5907 * It will be covered by recursion from the yet active parent. */
5908 if (bdrv_has_bds_parent(bs, true)) {
5909 return 0;
5912 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5914 /* Inactivate this node */
5915 if (bs->drv->bdrv_inactivate) {
5916 ret = bs->drv->bdrv_inactivate(bs);
5917 if (ret < 0) {
5918 return ret;
5922 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5923 if (parent->klass->inactivate) {
5924 ret = parent->klass->inactivate(parent);
5925 if (ret < 0) {
5926 return ret;
5931 bs->open_flags |= BDRV_O_INACTIVE;
5933 /* Update permissions, they may differ for inactive nodes */
5934 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5935 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL,
5936 &tighten_restrictions, NULL);
5937 assert(tighten_restrictions == false);
5938 if (ret < 0) {
5939 /* We only tried to loosen restrictions, so errors are not fatal */
5940 bdrv_abort_perm_update(bs);
5941 } else {
5942 bdrv_set_perm(bs, perm, shared_perm);
5946 /* Recursively inactivate children */
5947 QLIST_FOREACH(child, &bs->children, next) {
5948 ret = bdrv_inactivate_recurse(child->bs);
5949 if (ret < 0) {
5950 return ret;
5954 return 0;
5957 int bdrv_inactivate_all(void)
5959 BlockDriverState *bs = NULL;
5960 BdrvNextIterator it;
5961 int ret = 0;
5962 GSList *aio_ctxs = NULL, *ctx;
5964 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5965 AioContext *aio_context = bdrv_get_aio_context(bs);
5967 if (!g_slist_find(aio_ctxs, aio_context)) {
5968 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5969 aio_context_acquire(aio_context);
5973 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5974 /* Nodes with BDS parents are covered by recursion from the last
5975 * parent that gets inactivated. Don't inactivate them a second
5976 * time if that has already happened. */
5977 if (bdrv_has_bds_parent(bs, false)) {
5978 continue;
5980 ret = bdrv_inactivate_recurse(bs);
5981 if (ret < 0) {
5982 bdrv_next_cleanup(&it);
5983 goto out;
5987 out:
5988 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5989 AioContext *aio_context = ctx->data;
5990 aio_context_release(aio_context);
5992 g_slist_free(aio_ctxs);
5994 return ret;
5997 /**************************************************************/
5998 /* removable device support */
6001 * Return TRUE if the media is present
6003 bool bdrv_is_inserted(BlockDriverState *bs)
6005 BlockDriver *drv = bs->drv;
6006 BdrvChild *child;
6008 if (!drv) {
6009 return false;
6011 if (drv->bdrv_is_inserted) {
6012 return drv->bdrv_is_inserted(bs);
6014 QLIST_FOREACH(child, &bs->children, next) {
6015 if (!bdrv_is_inserted(child->bs)) {
6016 return false;
6019 return true;
6023 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6025 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
6027 BlockDriver *drv = bs->drv;
6029 if (drv && drv->bdrv_eject) {
6030 drv->bdrv_eject(bs, eject_flag);
6035 * Lock or unlock the media (if it is locked, the user won't be able
6036 * to eject it manually).
6038 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
6040 BlockDriver *drv = bs->drv;
6042 trace_bdrv_lock_medium(bs, locked);
6044 if (drv && drv->bdrv_lock_medium) {
6045 drv->bdrv_lock_medium(bs, locked);
6049 /* Get a reference to bs */
6050 void bdrv_ref(BlockDriverState *bs)
6052 bs->refcnt++;
6055 /* Release a previously grabbed reference to bs.
6056 * If after releasing, reference count is zero, the BlockDriverState is
6057 * deleted. */
6058 void bdrv_unref(BlockDriverState *bs)
6060 if (!bs) {
6061 return;
6063 assert(bs->refcnt > 0);
6064 if (--bs->refcnt == 0) {
6065 bdrv_delete(bs);
6069 struct BdrvOpBlocker {
6070 Error *reason;
6071 QLIST_ENTRY(BdrvOpBlocker) list;
6074 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6076 BdrvOpBlocker *blocker;
6077 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6078 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6079 blocker = QLIST_FIRST(&bs->op_blockers[op]);
6080 error_propagate_prepend(errp, error_copy(blocker->reason),
6081 "Node '%s' is busy: ",
6082 bdrv_get_device_or_node_name(bs));
6083 return true;
6085 return false;
6088 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6090 BdrvOpBlocker *blocker;
6091 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6093 blocker = g_new0(BdrvOpBlocker, 1);
6094 blocker->reason = reason;
6095 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6098 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6100 BdrvOpBlocker *blocker, *next;
6101 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6102 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6103 if (blocker->reason == reason) {
6104 QLIST_REMOVE(blocker, list);
6105 g_free(blocker);
6110 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6112 int i;
6113 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6114 bdrv_op_block(bs, i, reason);
6118 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6120 int i;
6121 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6122 bdrv_op_unblock(bs, i, reason);
6126 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6128 int i;
6130 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6131 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6132 return false;
6135 return true;
6138 void bdrv_img_create(const char *filename, const char *fmt,
6139 const char *base_filename, const char *base_fmt,
6140 char *options, uint64_t img_size, int flags, bool quiet,
6141 Error **errp)
6143 QemuOptsList *create_opts = NULL;
6144 QemuOpts *opts = NULL;
6145 const char *backing_fmt, *backing_file;
6146 int64_t size;
6147 BlockDriver *drv, *proto_drv;
6148 Error *local_err = NULL;
6149 int ret = 0;
6151 /* Find driver and parse its options */
6152 drv = bdrv_find_format(fmt);
6153 if (!drv) {
6154 error_setg(errp, "Unknown file format '%s'", fmt);
6155 return;
6158 proto_drv = bdrv_find_protocol(filename, true, errp);
6159 if (!proto_drv) {
6160 return;
6163 if (!drv->create_opts) {
6164 error_setg(errp, "Format driver '%s' does not support image creation",
6165 drv->format_name);
6166 return;
6169 if (!proto_drv->create_opts) {
6170 error_setg(errp, "Protocol driver '%s' does not support image creation",
6171 proto_drv->format_name);
6172 return;
6175 /* Create parameter list */
6176 create_opts = qemu_opts_append(create_opts, drv->create_opts);
6177 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
6179 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
6181 /* Parse -o options */
6182 if (options) {
6183 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
6184 goto out;
6188 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6189 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6190 } else if (img_size != UINT64_C(-1)) {
6191 error_setg(errp, "The image size must be specified only once");
6192 goto out;
6195 if (base_filename) {
6196 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
6197 NULL)) {
6198 error_setg(errp, "Backing file not supported for file format '%s'",
6199 fmt);
6200 goto out;
6204 if (base_fmt) {
6205 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
6206 error_setg(errp, "Backing file format not supported for file "
6207 "format '%s'", fmt);
6208 goto out;
6212 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6213 if (backing_file) {
6214 if (!strcmp(filename, backing_file)) {
6215 error_setg(errp, "Error: Trying to create an image with the "
6216 "same filename as the backing file");
6217 goto out;
6219 if (backing_file[0] == '\0') {
6220 error_setg(errp, "Expected backing file name, got empty string");
6221 goto out;
6225 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
6227 /* The size for the image must always be specified, unless we have a backing
6228 * file and we have not been forbidden from opening it. */
6229 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6230 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6231 BlockDriverState *bs;
6232 char *full_backing;
6233 int back_flags;
6234 QDict *backing_options = NULL;
6236 full_backing =
6237 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6238 &local_err);
6239 if (local_err) {
6240 goto out;
6242 assert(full_backing);
6244 /* backing files always opened read-only */
6245 back_flags = flags;
6246 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
6248 backing_options = qdict_new();
6249 if (backing_fmt) {
6250 qdict_put_str(backing_options, "driver", backing_fmt);
6252 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
6254 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6255 &local_err);
6256 g_free(full_backing);
6257 if (!bs) {
6258 error_append_hint(&local_err, "Could not open backing image.\n");
6259 goto out;
6260 } else {
6261 if (!backing_fmt) {
6262 warn_report("Deprecated use of backing file without explicit "
6263 "backing format (detected format of %s)",
6264 bs->drv->format_name);
6265 if (bs->drv != &bdrv_raw) {
6267 * A probe of raw deserves the most attention:
6268 * leaving the backing format out of the image
6269 * will ensure bs->probed is set (ensuring we
6270 * don't accidentally commit into the backing
6271 * file), and allow more spots to warn the users
6272 * to fix their toolchain when opening this image
6273 * later. For other images, we can safely record
6274 * the format that we probed.
6276 backing_fmt = bs->drv->format_name;
6277 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, backing_fmt,
6278 NULL);
6281 if (size == -1) {
6282 /* Opened BS, have no size */
6283 size = bdrv_getlength(bs);
6284 if (size < 0) {
6285 error_setg_errno(errp, -size, "Could not get size of '%s'",
6286 backing_file);
6287 bdrv_unref(bs);
6288 goto out;
6290 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
6292 bdrv_unref(bs);
6294 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6295 } else if (backing_file && !backing_fmt) {
6296 warn_report("Deprecated use of unopened backing file without "
6297 "explicit backing format, use of this image requires "
6298 "potentially unsafe format probing");
6301 if (size == -1) {
6302 error_setg(errp, "Image creation needs a size parameter");
6303 goto out;
6306 if (!quiet) {
6307 printf("Formatting '%s', fmt=%s ", filename, fmt);
6308 qemu_opts_print(opts, " ");
6309 puts("");
6310 fflush(stdout);
6313 ret = bdrv_create(drv, filename, opts, &local_err);
6315 if (ret == -EFBIG) {
6316 /* This is generally a better message than whatever the driver would
6317 * deliver (especially because of the cluster_size_hint), since that
6318 * is most probably not much different from "image too large". */
6319 const char *cluster_size_hint = "";
6320 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
6321 cluster_size_hint = " (try using a larger cluster size)";
6323 error_setg(errp, "The image size is too large for file format '%s'"
6324 "%s", fmt, cluster_size_hint);
6325 error_free(local_err);
6326 local_err = NULL;
6329 out:
6330 qemu_opts_del(opts);
6331 qemu_opts_free(create_opts);
6332 error_propagate(errp, local_err);
6335 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6337 return bs ? bs->aio_context : qemu_get_aio_context();
6340 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
6342 Coroutine *self = qemu_coroutine_self();
6343 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
6344 AioContext *new_ctx;
6347 * Increase bs->in_flight to ensure that this operation is completed before
6348 * moving the node to a different AioContext. Read new_ctx only afterwards.
6350 bdrv_inc_in_flight(bs);
6352 new_ctx = bdrv_get_aio_context(bs);
6353 aio_co_reschedule_self(new_ctx);
6354 return old_ctx;
6357 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
6359 aio_co_reschedule_self(old_ctx);
6360 bdrv_dec_in_flight(bs);
6363 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
6365 AioContext *ctx = bdrv_get_aio_context(bs);
6367 /* In the main thread, bs->aio_context won't change concurrently */
6368 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6371 * We're in coroutine context, so we already hold the lock of the main
6372 * loop AioContext. Don't lock it twice to avoid deadlocks.
6374 assert(qemu_in_coroutine());
6375 if (ctx != qemu_get_aio_context()) {
6376 aio_context_acquire(ctx);
6380 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
6382 AioContext *ctx = bdrv_get_aio_context(bs);
6384 assert(qemu_in_coroutine());
6385 if (ctx != qemu_get_aio_context()) {
6386 aio_context_release(ctx);
6390 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6392 aio_co_enter(bdrv_get_aio_context(bs), co);
6395 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6397 QLIST_REMOVE(ban, list);
6398 g_free(ban);
6401 static void bdrv_detach_aio_context(BlockDriverState *bs)
6403 BdrvAioNotifier *baf, *baf_tmp;
6405 assert(!bs->walking_aio_notifiers);
6406 bs->walking_aio_notifiers = true;
6407 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6408 if (baf->deleted) {
6409 bdrv_do_remove_aio_context_notifier(baf);
6410 } else {
6411 baf->detach_aio_context(baf->opaque);
6414 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6415 * remove remaining aio notifiers if we aren't called again.
6417 bs->walking_aio_notifiers = false;
6419 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
6420 bs->drv->bdrv_detach_aio_context(bs);
6423 if (bs->quiesce_counter) {
6424 aio_enable_external(bs->aio_context);
6426 bs->aio_context = NULL;
6429 static void bdrv_attach_aio_context(BlockDriverState *bs,
6430 AioContext *new_context)
6432 BdrvAioNotifier *ban, *ban_tmp;
6434 if (bs->quiesce_counter) {
6435 aio_disable_external(new_context);
6438 bs->aio_context = new_context;
6440 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
6441 bs->drv->bdrv_attach_aio_context(bs, new_context);
6444 assert(!bs->walking_aio_notifiers);
6445 bs->walking_aio_notifiers = true;
6446 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6447 if (ban->deleted) {
6448 bdrv_do_remove_aio_context_notifier(ban);
6449 } else {
6450 ban->attached_aio_context(new_context, ban->opaque);
6453 bs->walking_aio_notifiers = false;
6457 * Changes the AioContext used for fd handlers, timers, and BHs by this
6458 * BlockDriverState and all its children and parents.
6460 * Must be called from the main AioContext.
6462 * The caller must own the AioContext lock for the old AioContext of bs, but it
6463 * must not own the AioContext lock for new_context (unless new_context is the
6464 * same as the current context of bs).
6466 * @ignore will accumulate all visited BdrvChild object. The caller is
6467 * responsible for freeing the list afterwards.
6469 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6470 AioContext *new_context, GSList **ignore)
6472 AioContext *old_context = bdrv_get_aio_context(bs);
6473 BdrvChild *child;
6475 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6477 if (old_context == new_context) {
6478 return;
6481 bdrv_drained_begin(bs);
6483 QLIST_FOREACH(child, &bs->children, next) {
6484 if (g_slist_find(*ignore, child)) {
6485 continue;
6487 *ignore = g_slist_prepend(*ignore, child);
6488 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6490 QLIST_FOREACH(child, &bs->parents, next_parent) {
6491 if (g_slist_find(*ignore, child)) {
6492 continue;
6494 assert(child->klass->set_aio_ctx);
6495 *ignore = g_slist_prepend(*ignore, child);
6496 child->klass->set_aio_ctx(child, new_context, ignore);
6499 bdrv_detach_aio_context(bs);
6501 /* Acquire the new context, if necessary */
6502 if (qemu_get_aio_context() != new_context) {
6503 aio_context_acquire(new_context);
6506 bdrv_attach_aio_context(bs, new_context);
6509 * If this function was recursively called from
6510 * bdrv_set_aio_context_ignore(), there may be nodes in the
6511 * subtree that have not yet been moved to the new AioContext.
6512 * Release the old one so bdrv_drained_end() can poll them.
6514 if (qemu_get_aio_context() != old_context) {
6515 aio_context_release(old_context);
6518 bdrv_drained_end(bs);
6520 if (qemu_get_aio_context() != old_context) {
6521 aio_context_acquire(old_context);
6523 if (qemu_get_aio_context() != new_context) {
6524 aio_context_release(new_context);
6528 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6529 GSList **ignore, Error **errp)
6531 if (g_slist_find(*ignore, c)) {
6532 return true;
6534 *ignore = g_slist_prepend(*ignore, c);
6537 * A BdrvChildClass that doesn't handle AioContext changes cannot
6538 * tolerate any AioContext changes
6540 if (!c->klass->can_set_aio_ctx) {
6541 char *user = bdrv_child_user_desc(c);
6542 error_setg(errp, "Changing iothreads is not supported by %s", user);
6543 g_free(user);
6544 return false;
6546 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
6547 assert(!errp || *errp);
6548 return false;
6550 return true;
6553 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6554 GSList **ignore, Error **errp)
6556 if (g_slist_find(*ignore, c)) {
6557 return true;
6559 *ignore = g_slist_prepend(*ignore, c);
6560 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6563 /* @ignore will accumulate all visited BdrvChild object. The caller is
6564 * responsible for freeing the list afterwards. */
6565 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6566 GSList **ignore, Error **errp)
6568 BdrvChild *c;
6570 if (bdrv_get_aio_context(bs) == ctx) {
6571 return true;
6574 QLIST_FOREACH(c, &bs->parents, next_parent) {
6575 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
6576 return false;
6579 QLIST_FOREACH(c, &bs->children, next) {
6580 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
6581 return false;
6585 return true;
6588 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6589 BdrvChild *ignore_child, Error **errp)
6591 GSList *ignore;
6592 bool ret;
6594 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6595 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
6596 g_slist_free(ignore);
6598 if (!ret) {
6599 return -EPERM;
6602 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6603 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
6604 g_slist_free(ignore);
6606 return 0;
6609 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6610 Error **errp)
6612 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
6615 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
6616 void (*attached_aio_context)(AioContext *new_context, void *opaque),
6617 void (*detach_aio_context)(void *opaque), void *opaque)
6619 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
6620 *ban = (BdrvAioNotifier){
6621 .attached_aio_context = attached_aio_context,
6622 .detach_aio_context = detach_aio_context,
6623 .opaque = opaque
6626 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
6629 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
6630 void (*attached_aio_context)(AioContext *,
6631 void *),
6632 void (*detach_aio_context)(void *),
6633 void *opaque)
6635 BdrvAioNotifier *ban, *ban_next;
6637 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
6638 if (ban->attached_aio_context == attached_aio_context &&
6639 ban->detach_aio_context == detach_aio_context &&
6640 ban->opaque == opaque &&
6641 ban->deleted == false)
6643 if (bs->walking_aio_notifiers) {
6644 ban->deleted = true;
6645 } else {
6646 bdrv_do_remove_aio_context_notifier(ban);
6648 return;
6652 abort();
6655 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
6656 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
6657 bool force,
6658 Error **errp)
6660 if (!bs->drv) {
6661 error_setg(errp, "Node is ejected");
6662 return -ENOMEDIUM;
6664 if (!bs->drv->bdrv_amend_options) {
6665 error_setg(errp, "Block driver '%s' does not support option amendment",
6666 bs->drv->format_name);
6667 return -ENOTSUP;
6669 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
6670 cb_opaque, force, errp);
6674 * This function checks whether the given @to_replace is allowed to be
6675 * replaced by a node that always shows the same data as @bs. This is
6676 * used for example to verify whether the mirror job can replace
6677 * @to_replace by the target mirrored from @bs.
6678 * To be replaceable, @bs and @to_replace may either be guaranteed to
6679 * always show the same data (because they are only connected through
6680 * filters), or some driver may allow replacing one of its children
6681 * because it can guarantee that this child's data is not visible at
6682 * all (for example, for dissenting quorum children that have no other
6683 * parents).
6685 bool bdrv_recurse_can_replace(BlockDriverState *bs,
6686 BlockDriverState *to_replace)
6688 BlockDriverState *filtered;
6690 if (!bs || !bs->drv) {
6691 return false;
6694 if (bs == to_replace) {
6695 return true;
6698 /* See what the driver can do */
6699 if (bs->drv->bdrv_recurse_can_replace) {
6700 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
6703 /* For filters without an own implementation, we can recurse on our own */
6704 filtered = bdrv_filter_bs(bs);
6705 if (filtered) {
6706 return bdrv_recurse_can_replace(filtered, to_replace);
6709 /* Safe default */
6710 return false;
6714 * Check whether the given @node_name can be replaced by a node that
6715 * has the same data as @parent_bs. If so, return @node_name's BDS;
6716 * NULL otherwise.
6718 * @node_name must be a (recursive) *child of @parent_bs (or this
6719 * function will return NULL).
6721 * The result (whether the node can be replaced or not) is only valid
6722 * for as long as no graph or permission changes occur.
6724 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
6725 const char *node_name, Error **errp)
6727 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
6728 AioContext *aio_context;
6730 if (!to_replace_bs) {
6731 error_setg(errp, "Node name '%s' not found", node_name);
6732 return NULL;
6735 aio_context = bdrv_get_aio_context(to_replace_bs);
6736 aio_context_acquire(aio_context);
6738 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
6739 to_replace_bs = NULL;
6740 goto out;
6743 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6744 * most non filter in order to prevent data corruption.
6745 * Another benefit is that this tests exclude backing files which are
6746 * blocked by the backing blockers.
6748 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
6749 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
6750 "because it cannot be guaranteed that doing so would not "
6751 "lead to an abrupt change of visible data",
6752 node_name, parent_bs->node_name);
6753 to_replace_bs = NULL;
6754 goto out;
6757 out:
6758 aio_context_release(aio_context);
6759 return to_replace_bs;
6763 * Iterates through the list of runtime option keys that are said to
6764 * be "strong" for a BDS. An option is called "strong" if it changes
6765 * a BDS's data. For example, the null block driver's "size" and
6766 * "read-zeroes" options are strong, but its "latency-ns" option is
6767 * not.
6769 * If a key returned by this function ends with a dot, all options
6770 * starting with that prefix are strong.
6772 static const char *const *strong_options(BlockDriverState *bs,
6773 const char *const *curopt)
6775 static const char *const global_options[] = {
6776 "driver", "filename", NULL
6779 if (!curopt) {
6780 return &global_options[0];
6783 curopt++;
6784 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
6785 curopt = bs->drv->strong_runtime_opts;
6788 return (curopt && *curopt) ? curopt : NULL;
6792 * Copies all strong runtime options from bs->options to the given
6793 * QDict. The set of strong option keys is determined by invoking
6794 * strong_options().
6796 * Returns true iff any strong option was present in bs->options (and
6797 * thus copied to the target QDict) with the exception of "filename"
6798 * and "driver". The caller is expected to use this value to decide
6799 * whether the existence of strong options prevents the generation of
6800 * a plain filename.
6802 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
6804 bool found_any = false;
6805 const char *const *option_name = NULL;
6807 if (!bs->drv) {
6808 return false;
6811 while ((option_name = strong_options(bs, option_name))) {
6812 bool option_given = false;
6814 assert(strlen(*option_name) > 0);
6815 if ((*option_name)[strlen(*option_name) - 1] != '.') {
6816 QObject *entry = qdict_get(bs->options, *option_name);
6817 if (!entry) {
6818 continue;
6821 qdict_put_obj(d, *option_name, qobject_ref(entry));
6822 option_given = true;
6823 } else {
6824 const QDictEntry *entry;
6825 for (entry = qdict_first(bs->options); entry;
6826 entry = qdict_next(bs->options, entry))
6828 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
6829 qdict_put_obj(d, qdict_entry_key(entry),
6830 qobject_ref(qdict_entry_value(entry)));
6831 option_given = true;
6836 /* While "driver" and "filename" need to be included in a JSON filename,
6837 * their existence does not prohibit generation of a plain filename. */
6838 if (!found_any && option_given &&
6839 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
6841 found_any = true;
6845 if (!qdict_haskey(d, "driver")) {
6846 /* Drivers created with bdrv_new_open_driver() may not have a
6847 * @driver option. Add it here. */
6848 qdict_put_str(d, "driver", bs->drv->format_name);
6851 return found_any;
6854 /* Note: This function may return false positives; it may return true
6855 * even if opening the backing file specified by bs's image header
6856 * would result in exactly bs->backing. */
6857 bool bdrv_backing_overridden(BlockDriverState *bs)
6859 if (bs->backing) {
6860 return strcmp(bs->auto_backing_file,
6861 bs->backing->bs->filename);
6862 } else {
6863 /* No backing BDS, so if the image header reports any backing
6864 * file, it must have been suppressed */
6865 return bs->auto_backing_file[0] != '\0';
6869 /* Updates the following BDS fields:
6870 * - exact_filename: A filename which may be used for opening a block device
6871 * which (mostly) equals the given BDS (even without any
6872 * other options; so reading and writing must return the same
6873 * results, but caching etc. may be different)
6874 * - full_open_options: Options which, when given when opening a block device
6875 * (without a filename), result in a BDS (mostly)
6876 * equalling the given one
6877 * - filename: If exact_filename is set, it is copied here. Otherwise,
6878 * full_open_options is converted to a JSON object, prefixed with
6879 * "json:" (for use through the JSON pseudo protocol) and put here.
6881 void bdrv_refresh_filename(BlockDriverState *bs)
6883 BlockDriver *drv = bs->drv;
6884 BdrvChild *child;
6885 BlockDriverState *primary_child_bs;
6886 QDict *opts;
6887 bool backing_overridden;
6888 bool generate_json_filename; /* Whether our default implementation should
6889 fill exact_filename (false) or not (true) */
6891 if (!drv) {
6892 return;
6895 /* This BDS's file name may depend on any of its children's file names, so
6896 * refresh those first */
6897 QLIST_FOREACH(child, &bs->children, next) {
6898 bdrv_refresh_filename(child->bs);
6901 if (bs->implicit) {
6902 /* For implicit nodes, just copy everything from the single child */
6903 child = QLIST_FIRST(&bs->children);
6904 assert(QLIST_NEXT(child, next) == NULL);
6906 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
6907 child->bs->exact_filename);
6908 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
6910 qobject_unref(bs->full_open_options);
6911 bs->full_open_options = qobject_ref(child->bs->full_open_options);
6913 return;
6916 backing_overridden = bdrv_backing_overridden(bs);
6918 if (bs->open_flags & BDRV_O_NO_IO) {
6919 /* Without I/O, the backing file does not change anything.
6920 * Therefore, in such a case (primarily qemu-img), we can
6921 * pretend the backing file has not been overridden even if
6922 * it technically has been. */
6923 backing_overridden = false;
6926 /* Gather the options QDict */
6927 opts = qdict_new();
6928 generate_json_filename = append_strong_runtime_options(opts, bs);
6929 generate_json_filename |= backing_overridden;
6931 if (drv->bdrv_gather_child_options) {
6932 /* Some block drivers may not want to present all of their children's
6933 * options, or name them differently from BdrvChild.name */
6934 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
6935 } else {
6936 QLIST_FOREACH(child, &bs->children, next) {
6937 if (child == bs->backing && !backing_overridden) {
6938 /* We can skip the backing BDS if it has not been overridden */
6939 continue;
6942 qdict_put(opts, child->name,
6943 qobject_ref(child->bs->full_open_options));
6946 if (backing_overridden && !bs->backing) {
6947 /* Force no backing file */
6948 qdict_put_null(opts, "backing");
6952 qobject_unref(bs->full_open_options);
6953 bs->full_open_options = opts;
6955 primary_child_bs = bdrv_primary_bs(bs);
6957 if (drv->bdrv_refresh_filename) {
6958 /* Obsolete information is of no use here, so drop the old file name
6959 * information before refreshing it */
6960 bs->exact_filename[0] = '\0';
6962 drv->bdrv_refresh_filename(bs);
6963 } else if (primary_child_bs) {
6965 * Try to reconstruct valid information from the underlying
6966 * file -- this only works for format nodes (filter nodes
6967 * cannot be probed and as such must be selected by the user
6968 * either through an options dict, or through a special
6969 * filename which the filter driver must construct in its
6970 * .bdrv_refresh_filename() implementation).
6973 bs->exact_filename[0] = '\0';
6976 * We can use the underlying file's filename if:
6977 * - it has a filename,
6978 * - the current BDS is not a filter,
6979 * - the file is a protocol BDS, and
6980 * - opening that file (as this BDS's format) will automatically create
6981 * the BDS tree we have right now, that is:
6982 * - the user did not significantly change this BDS's behavior with
6983 * some explicit (strong) options
6984 * - no non-file child of this BDS has been overridden by the user
6985 * Both of these conditions are represented by generate_json_filename.
6987 if (primary_child_bs->exact_filename[0] &&
6988 primary_child_bs->drv->bdrv_file_open &&
6989 !drv->is_filter && !generate_json_filename)
6991 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
6995 if (bs->exact_filename[0]) {
6996 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
6997 } else {
6998 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
6999 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
7000 qstring_get_str(json)) >= sizeof(bs->filename)) {
7001 /* Give user a hint if we truncated things. */
7002 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7004 qobject_unref(json);
7008 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7010 BlockDriver *drv = bs->drv;
7011 BlockDriverState *child_bs;
7013 if (!drv) {
7014 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7015 return NULL;
7018 if (drv->bdrv_dirname) {
7019 return drv->bdrv_dirname(bs, errp);
7022 child_bs = bdrv_primary_bs(bs);
7023 if (child_bs) {
7024 return bdrv_dirname(child_bs, errp);
7027 bdrv_refresh_filename(bs);
7028 if (bs->exact_filename[0] != '\0') {
7029 return path_combine(bs->exact_filename, "");
7032 error_setg(errp, "Cannot generate a base directory for %s nodes",
7033 drv->format_name);
7034 return NULL;
7038 * Hot add/remove a BDS's child. So the user can take a child offline when
7039 * it is broken and take a new child online
7041 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7042 Error **errp)
7045 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7046 error_setg(errp, "The node %s does not support adding a child",
7047 bdrv_get_device_or_node_name(parent_bs));
7048 return;
7051 if (!QLIST_EMPTY(&child_bs->parents)) {
7052 error_setg(errp, "The node %s already has a parent",
7053 child_bs->node_name);
7054 return;
7057 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7060 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7062 BdrvChild *tmp;
7064 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7065 error_setg(errp, "The node %s does not support removing a child",
7066 bdrv_get_device_or_node_name(parent_bs));
7067 return;
7070 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7071 if (tmp == child) {
7072 break;
7076 if (!tmp) {
7077 error_setg(errp, "The node %s does not have a child named %s",
7078 bdrv_get_device_or_node_name(parent_bs),
7079 bdrv_get_device_or_node_name(child->bs));
7080 return;
7083 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7086 int bdrv_make_empty(BdrvChild *c, Error **errp)
7088 BlockDriver *drv = c->bs->drv;
7089 int ret;
7091 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7093 if (!drv->bdrv_make_empty) {
7094 error_setg(errp, "%s does not support emptying nodes",
7095 drv->format_name);
7096 return -ENOTSUP;
7099 ret = drv->bdrv_make_empty(c->bs);
7100 if (ret < 0) {
7101 error_setg_errno(errp, -ret, "Failed to empty %s",
7102 c->bs->filename);
7103 return ret;
7106 return 0;
7110 * Return the child that @bs acts as an overlay for, and from which data may be
7111 * copied in COW or COR operations. Usually this is the backing file.
7113 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7115 if (!bs || !bs->drv) {
7116 return NULL;
7119 if (bs->drv->is_filter) {
7120 return NULL;
7123 if (!bs->backing) {
7124 return NULL;
7127 assert(bs->backing->role & BDRV_CHILD_COW);
7128 return bs->backing;
7132 * If @bs acts as a filter for exactly one of its children, return
7133 * that child.
7135 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7137 BdrvChild *c;
7139 if (!bs || !bs->drv) {
7140 return NULL;
7143 if (!bs->drv->is_filter) {
7144 return NULL;
7147 /* Only one of @backing or @file may be used */
7148 assert(!(bs->backing && bs->file));
7150 c = bs->backing ?: bs->file;
7151 if (!c) {
7152 return NULL;
7155 assert(c->role & BDRV_CHILD_FILTERED);
7156 return c;
7160 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7161 * whichever is non-NULL.
7163 * Return NULL if both are NULL.
7165 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7167 BdrvChild *cow_child = bdrv_cow_child(bs);
7168 BdrvChild *filter_child = bdrv_filter_child(bs);
7170 /* Filter nodes cannot have COW backing files */
7171 assert(!(cow_child && filter_child));
7173 return cow_child ?: filter_child;
7177 * Return the primary child of this node: For filters, that is the
7178 * filtered child. For other nodes, that is usually the child storing
7179 * metadata.
7180 * (A generally more helpful description is that this is (usually) the
7181 * child that has the same filename as @bs.)
7183 * Drivers do not necessarily have a primary child; for example quorum
7184 * does not.
7186 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
7188 BdrvChild *c, *found = NULL;
7190 QLIST_FOREACH(c, &bs->children, next) {
7191 if (c->role & BDRV_CHILD_PRIMARY) {
7192 assert(!found);
7193 found = c;
7197 return found;
7200 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
7201 bool stop_on_explicit_filter)
7203 BdrvChild *c;
7205 if (!bs) {
7206 return NULL;
7209 while (!(stop_on_explicit_filter && !bs->implicit)) {
7210 c = bdrv_filter_child(bs);
7211 if (!c) {
7213 * A filter that is embedded in a working block graph must
7214 * have a child. Assert this here so this function does
7215 * not return a filter node that is not expected by the
7216 * caller.
7218 assert(!bs->drv || !bs->drv->is_filter);
7219 break;
7221 bs = c->bs;
7224 * Note that this treats nodes with bs->drv == NULL as not being
7225 * filters (bs->drv == NULL should be replaced by something else
7226 * anyway).
7227 * The advantage of this behavior is that this function will thus
7228 * always return a non-NULL value (given a non-NULL @bs).
7231 return bs;
7235 * Return the first BDS that has not been added implicitly or that
7236 * does not have a filtered child down the chain starting from @bs
7237 * (including @bs itself).
7239 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
7241 return bdrv_do_skip_filters(bs, true);
7245 * Return the first BDS that does not have a filtered child down the
7246 * chain starting from @bs (including @bs itself).
7248 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
7250 return bdrv_do_skip_filters(bs, false);
7254 * For a backing chain, return the first non-filter backing image of
7255 * the first non-filter image.
7257 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
7259 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));