bios-tables-test: tell people how to update
[qemu.git] / block.c
blob1bdb9c679d77557b244bc7dd3a9a87480f1e1ce6
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/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
50 #include "qemu/id.h"
52 #ifdef CONFIG_BSD
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
55 #ifndef __DragonFly__
56 #include <sys/disk.h>
57 #endif
58 #endif
60 #ifdef _WIN32
61 #include <windows.h>
62 #endif
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
66 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
69 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
72 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
75 static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const BdrvChildRole *child_role,
80 Error **errp);
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist;
85 #ifdef _WIN32
86 static int is_windows_drive_prefix(const char *filename)
88 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
89 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
90 filename[1] == ':');
93 int is_windows_drive(const char *filename)
95 if (is_windows_drive_prefix(filename) &&
96 filename[2] == '\0')
97 return 1;
98 if (strstart(filename, "\\\\.\\", NULL) ||
99 strstart(filename, "//./", NULL))
100 return 1;
101 return 0;
103 #endif
105 size_t bdrv_opt_mem_align(BlockDriverState *bs)
107 if (!bs || !bs->drv) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, qemu_real_host_page_size);
112 return bs->bl.opt_mem_alignment;
115 size_t bdrv_min_mem_align(BlockDriverState *bs)
117 if (!bs || !bs->drv) {
118 /* page size or 4k (hdd sector size) should be on the safe side */
119 return MAX(4096, qemu_real_host_page_size);
122 return bs->bl.min_mem_alignment;
125 /* check if the path starts with "<protocol>:" */
126 int path_has_protocol(const char *path)
128 const char *p;
130 #ifdef _WIN32
131 if (is_windows_drive(path) ||
132 is_windows_drive_prefix(path)) {
133 return 0;
135 p = path + strcspn(path, ":/\\");
136 #else
137 p = path + strcspn(path, ":/");
138 #endif
140 return *p == ':';
143 int path_is_absolute(const char *path)
145 #ifdef _WIN32
146 /* specific case for names like: "\\.\d:" */
147 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
148 return 1;
150 return (*path == '/' || *path == '\\');
151 #else
152 return (*path == '/');
153 #endif
156 /* if filename is absolute, just return its duplicate. Otherwise, build a
157 path to it by considering it is relative to base_path. URL are
158 supported. */
159 char *path_combine(const char *base_path, const char *filename)
161 const char *protocol_stripped = NULL;
162 const char *p, *p1;
163 char *result;
164 int len;
166 if (path_is_absolute(filename)) {
167 return g_strdup(filename);
170 if (path_has_protocol(base_path)) {
171 protocol_stripped = strchr(base_path, ':');
172 if (protocol_stripped) {
173 protocol_stripped++;
176 p = protocol_stripped ?: base_path;
178 p1 = strrchr(base_path, '/');
179 #ifdef _WIN32
181 const char *p2;
182 p2 = strrchr(base_path, '\\');
183 if (!p1 || p2 > p1) {
184 p1 = p2;
187 #endif
188 if (p1) {
189 p1++;
190 } else {
191 p1 = base_path;
193 if (p1 > p) {
194 p = p1;
196 len = p - base_path;
198 result = g_malloc(len + strlen(filename) + 1);
199 memcpy(result, base_path, len);
200 strcpy(result + len, filename);
202 return result;
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
210 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
211 QDict *options)
213 if (strstart(filename, prefix, &filename)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename)) {
217 QString *fat_filename;
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename));
223 /* And we can thus fix the protocol detection issue by prefixing it
224 * by "./" */
225 fat_filename = qstring_from_str("./");
226 qstring_append(fat_filename, filename);
228 assert(!path_has_protocol(qstring_get_str(fat_filename)));
230 qdict_put(options, "filename", fat_filename);
231 } else {
232 /* If no protocol prefix was detected, we can use the shortened
233 * filename as-is */
234 qdict_put_str(options, "filename", filename);
240 /* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
243 bool bdrv_is_read_only(BlockDriverState *bs)
245 return bs->read_only;
248 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
249 bool ignore_allow_rdw, Error **errp)
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs->copy_on_read && read_only) {
253 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs));
255 return -EINVAL;
258 /* Do not clear read_only if it is prohibited */
259 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
260 !ignore_allow_rdw)
262 error_setg(errp, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs));
264 return -EPERM;
267 return 0;
271 * Called by a driver that can only provide a read-only image.
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
280 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
281 Error **errp)
283 int ret = 0;
285 if (!(bs->open_flags & BDRV_O_RDWR)) {
286 return 0;
288 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
289 goto fail;
292 ret = bdrv_can_set_read_only(bs, true, false, NULL);
293 if (ret < 0) {
294 goto fail;
297 bs->read_only = true;
298 bs->open_flags &= ~BDRV_O_RDWR;
300 return 0;
302 fail:
303 error_setg(errp, "%s", errmsg ?: "Image is read-only");
304 return -EACCES;
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
310 * set.
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
316 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
317 const char *backing,
318 Error **errp)
320 if (backing[0] == '\0') {
321 return NULL;
322 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
323 return g_strdup(backing);
324 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
325 error_setg(errp, "Cannot use relative backing file names for '%s'",
326 backed);
327 return NULL;
328 } else {
329 return path_combine(backed, backing);
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
336 * @errp set.
338 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
339 const char *filename, Error **errp)
341 char *dir, *full_name;
343 if (!filename || filename[0] == '\0') {
344 return NULL;
345 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
346 return g_strdup(filename);
349 dir = bdrv_dirname(relative_to, errp);
350 if (!dir) {
351 return NULL;
354 full_name = g_strconcat(dir, filename, NULL);
355 g_free(dir);
356 return full_name;
359 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
361 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
364 void bdrv_register(BlockDriver *bdrv)
366 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
369 BlockDriverState *bdrv_new(void)
371 BlockDriverState *bs;
372 int i;
374 bs = g_new0(BlockDriverState, 1);
375 QLIST_INIT(&bs->dirty_bitmaps);
376 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
377 QLIST_INIT(&bs->op_blockers[i]);
379 notifier_with_return_list_init(&bs->before_write_notifiers);
380 qemu_co_mutex_init(&bs->reqs_lock);
381 qemu_mutex_init(&bs->dirty_bitmap_mutex);
382 bs->refcnt = 1;
383 bs->aio_context = qemu_get_aio_context();
385 qemu_co_queue_init(&bs->flush_queue);
387 for (i = 0; i < bdrv_drain_all_count; i++) {
388 bdrv_drained_begin(bs);
391 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
393 return bs;
396 static BlockDriver *bdrv_do_find_format(const char *format_name)
398 BlockDriver *drv1;
400 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
401 if (!strcmp(drv1->format_name, format_name)) {
402 return drv1;
406 return NULL;
409 BlockDriver *bdrv_find_format(const char *format_name)
411 BlockDriver *drv1;
412 int i;
414 drv1 = bdrv_do_find_format(format_name);
415 if (drv1) {
416 return drv1;
419 /* The driver isn't registered, maybe we need to load a module */
420 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
421 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
422 block_module_load_one(block_driver_modules[i].library_name);
423 break;
427 return bdrv_do_find_format(format_name);
430 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
432 static const char *whitelist_rw[] = {
433 CONFIG_BDRV_RW_WHITELIST
435 static const char *whitelist_ro[] = {
436 CONFIG_BDRV_RO_WHITELIST
438 const char **p;
440 if (!whitelist_rw[0] && !whitelist_ro[0]) {
441 return 1; /* no whitelist, anything goes */
444 for (p = whitelist_rw; *p; p++) {
445 if (!strcmp(format_name, *p)) {
446 return 1;
449 if (read_only) {
450 for (p = whitelist_ro; *p; p++) {
451 if (!strcmp(format_name, *p)) {
452 return 1;
456 return 0;
459 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
461 return bdrv_format_is_whitelisted(drv->format_name, read_only);
464 bool bdrv_uses_whitelist(void)
466 return use_bdrv_whitelist;
469 typedef struct CreateCo {
470 BlockDriver *drv;
471 char *filename;
472 QemuOpts *opts;
473 int ret;
474 Error *err;
475 } CreateCo;
477 static void coroutine_fn bdrv_create_co_entry(void *opaque)
479 Error *local_err = NULL;
480 int ret;
482 CreateCo *cco = opaque;
483 assert(cco->drv);
485 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
486 error_propagate(&cco->err, local_err);
487 cco->ret = ret;
490 int bdrv_create(BlockDriver *drv, const char* filename,
491 QemuOpts *opts, Error **errp)
493 int ret;
495 Coroutine *co;
496 CreateCo cco = {
497 .drv = drv,
498 .filename = g_strdup(filename),
499 .opts = opts,
500 .ret = NOT_DONE,
501 .err = NULL,
504 if (!drv->bdrv_co_create_opts) {
505 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
506 ret = -ENOTSUP;
507 goto out;
510 if (qemu_in_coroutine()) {
511 /* Fast-path if already in coroutine context */
512 bdrv_create_co_entry(&cco);
513 } else {
514 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
515 qemu_coroutine_enter(co);
516 while (cco.ret == NOT_DONE) {
517 aio_poll(qemu_get_aio_context(), true);
521 ret = cco.ret;
522 if (ret < 0) {
523 if (cco.err) {
524 error_propagate(errp, cco.err);
525 } else {
526 error_setg_errno(errp, -ret, "Could not create image");
530 out:
531 g_free(cco.filename);
532 return ret;
536 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
537 * least the given @minimum_size.
539 * On success, return @blk's actual length.
540 * Otherwise, return -errno.
542 static int64_t create_file_fallback_truncate(BlockBackend *blk,
543 int64_t minimum_size, Error **errp)
545 Error *local_err = NULL;
546 int64_t size;
547 int ret;
549 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, &local_err);
550 if (ret < 0 && ret != -ENOTSUP) {
551 error_propagate(errp, local_err);
552 return ret;
555 size = blk_getlength(blk);
556 if (size < 0) {
557 error_free(local_err);
558 error_setg_errno(errp, -size,
559 "Failed to inquire the new image file's length");
560 return size;
563 if (size < minimum_size) {
564 /* Need to grow the image, but we failed to do that */
565 error_propagate(errp, local_err);
566 return -ENOTSUP;
569 error_free(local_err);
570 local_err = NULL;
572 return size;
576 * Helper function for bdrv_create_file_fallback(): Zero the first
577 * sector to remove any potentially pre-existing image header.
579 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
580 int64_t current_size,
581 Error **errp)
583 int64_t bytes_to_clear;
584 int ret;
586 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
587 if (bytes_to_clear) {
588 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
589 if (ret < 0) {
590 error_setg_errno(errp, -ret,
591 "Failed to clear the new image's first sector");
592 return ret;
596 return 0;
599 static int bdrv_create_file_fallback(const char *filename, BlockDriver *drv,
600 QemuOpts *opts, Error **errp)
602 BlockBackend *blk;
603 QDict *options = qdict_new();
604 int64_t size = 0;
605 char *buf = NULL;
606 PreallocMode prealloc;
607 Error *local_err = NULL;
608 int ret;
610 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
611 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
612 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
613 PREALLOC_MODE_OFF, &local_err);
614 g_free(buf);
615 if (local_err) {
616 error_propagate(errp, local_err);
617 return -EINVAL;
620 if (prealloc != PREALLOC_MODE_OFF) {
621 error_setg(errp, "Unsupported preallocation mode '%s'",
622 PreallocMode_str(prealloc));
623 return -ENOTSUP;
626 qdict_put_str(options, "driver", drv->format_name);
628 blk = blk_new_open(filename, NULL, options,
629 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
630 if (!blk) {
631 error_prepend(errp, "Protocol driver '%s' does not support image "
632 "creation, and opening the image failed: ",
633 drv->format_name);
634 return -EINVAL;
637 size = create_file_fallback_truncate(blk, size, errp);
638 if (size < 0) {
639 ret = size;
640 goto out;
643 ret = create_file_fallback_zero_first_sector(blk, size, errp);
644 if (ret < 0) {
645 goto out;
648 ret = 0;
649 out:
650 blk_unref(blk);
651 return ret;
654 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
656 BlockDriver *drv;
658 drv = bdrv_find_protocol(filename, true, errp);
659 if (drv == NULL) {
660 return -ENOENT;
663 if (drv->bdrv_co_create_opts) {
664 return bdrv_create(drv, filename, opts, errp);
665 } else {
666 return bdrv_create_file_fallback(filename, drv, opts, errp);
671 * Try to get @bs's logical and physical block size.
672 * On success, store them in @bsz struct and return 0.
673 * On failure return -errno.
674 * @bs must not be empty.
676 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
678 BlockDriver *drv = bs->drv;
680 if (drv && drv->bdrv_probe_blocksizes) {
681 return drv->bdrv_probe_blocksizes(bs, bsz);
682 } else if (drv && drv->is_filter && bs->file) {
683 return bdrv_probe_blocksizes(bs->file->bs, bsz);
686 return -ENOTSUP;
690 * Try to get @bs's geometry (cyls, heads, sectors).
691 * On success, store them in @geo struct and return 0.
692 * On failure return -errno.
693 * @bs must not be empty.
695 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
697 BlockDriver *drv = bs->drv;
699 if (drv && drv->bdrv_probe_geometry) {
700 return drv->bdrv_probe_geometry(bs, geo);
701 } else if (drv && drv->is_filter && bs->file) {
702 return bdrv_probe_geometry(bs->file->bs, geo);
705 return -ENOTSUP;
709 * Create a uniquely-named empty temporary file.
710 * Return 0 upon success, otherwise a negative errno value.
712 int get_tmp_filename(char *filename, int size)
714 #ifdef _WIN32
715 char temp_dir[MAX_PATH];
716 /* GetTempFileName requires that its output buffer (4th param)
717 have length MAX_PATH or greater. */
718 assert(size >= MAX_PATH);
719 return (GetTempPath(MAX_PATH, temp_dir)
720 && GetTempFileName(temp_dir, "qem", 0, filename)
721 ? 0 : -GetLastError());
722 #else
723 int fd;
724 const char *tmpdir;
725 tmpdir = getenv("TMPDIR");
726 if (!tmpdir) {
727 tmpdir = "/var/tmp";
729 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
730 return -EOVERFLOW;
732 fd = mkstemp(filename);
733 if (fd < 0) {
734 return -errno;
736 if (close(fd) != 0) {
737 unlink(filename);
738 return -errno;
740 return 0;
741 #endif
745 * Detect host devices. By convention, /dev/cdrom[N] is always
746 * recognized as a host CDROM.
748 static BlockDriver *find_hdev_driver(const char *filename)
750 int score_max = 0, score;
751 BlockDriver *drv = NULL, *d;
753 QLIST_FOREACH(d, &bdrv_drivers, list) {
754 if (d->bdrv_probe_device) {
755 score = d->bdrv_probe_device(filename);
756 if (score > score_max) {
757 score_max = score;
758 drv = d;
763 return drv;
766 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
768 BlockDriver *drv1;
770 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
771 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
772 return drv1;
776 return NULL;
779 BlockDriver *bdrv_find_protocol(const char *filename,
780 bool allow_protocol_prefix,
781 Error **errp)
783 BlockDriver *drv1;
784 char protocol[128];
785 int len;
786 const char *p;
787 int i;
789 /* TODO Drivers without bdrv_file_open must be specified explicitly */
792 * XXX(hch): we really should not let host device detection
793 * override an explicit protocol specification, but moving this
794 * later breaks access to device names with colons in them.
795 * Thanks to the brain-dead persistent naming schemes on udev-
796 * based Linux systems those actually are quite common.
798 drv1 = find_hdev_driver(filename);
799 if (drv1) {
800 return drv1;
803 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
804 return &bdrv_file;
807 p = strchr(filename, ':');
808 assert(p != NULL);
809 len = p - filename;
810 if (len > sizeof(protocol) - 1)
811 len = sizeof(protocol) - 1;
812 memcpy(protocol, filename, len);
813 protocol[len] = '\0';
815 drv1 = bdrv_do_find_protocol(protocol);
816 if (drv1) {
817 return drv1;
820 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
821 if (block_driver_modules[i].protocol_name &&
822 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
823 block_module_load_one(block_driver_modules[i].library_name);
824 break;
828 drv1 = bdrv_do_find_protocol(protocol);
829 if (!drv1) {
830 error_setg(errp, "Unknown protocol '%s'", protocol);
832 return drv1;
836 * Guess image format by probing its contents.
837 * This is not a good idea when your image is raw (CVE-2008-2004), but
838 * we do it anyway for backward compatibility.
840 * @buf contains the image's first @buf_size bytes.
841 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
842 * but can be smaller if the image file is smaller)
843 * @filename is its filename.
845 * For all block drivers, call the bdrv_probe() method to get its
846 * probing score.
847 * Return the first block driver with the highest probing score.
849 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
850 const char *filename)
852 int score_max = 0, score;
853 BlockDriver *drv = NULL, *d;
855 QLIST_FOREACH(d, &bdrv_drivers, list) {
856 if (d->bdrv_probe) {
857 score = d->bdrv_probe(buf, buf_size, filename);
858 if (score > score_max) {
859 score_max = score;
860 drv = d;
865 return drv;
868 static int find_image_format(BlockBackend *file, const char *filename,
869 BlockDriver **pdrv, Error **errp)
871 BlockDriver *drv;
872 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
873 int ret = 0;
875 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
876 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
877 *pdrv = &bdrv_raw;
878 return ret;
881 ret = blk_pread(file, 0, buf, sizeof(buf));
882 if (ret < 0) {
883 error_setg_errno(errp, -ret, "Could not read image for determining its "
884 "format");
885 *pdrv = NULL;
886 return ret;
889 drv = bdrv_probe_all(buf, ret, filename);
890 if (!drv) {
891 error_setg(errp, "Could not determine image format: No compatible "
892 "driver found");
893 ret = -ENOENT;
895 *pdrv = drv;
896 return ret;
900 * Set the current 'total_sectors' value
901 * Return 0 on success, -errno on error.
903 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
905 BlockDriver *drv = bs->drv;
907 if (!drv) {
908 return -ENOMEDIUM;
911 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
912 if (bdrv_is_sg(bs))
913 return 0;
915 /* query actual device if possible, otherwise just trust the hint */
916 if (drv->bdrv_getlength) {
917 int64_t length = drv->bdrv_getlength(bs);
918 if (length < 0) {
919 return length;
921 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
924 bs->total_sectors = hint;
925 return 0;
929 * Combines a QDict of new block driver @options with any missing options taken
930 * from @old_options, so that leaving out an option defaults to its old value.
932 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
933 QDict *old_options)
935 if (bs->drv && bs->drv->bdrv_join_options) {
936 bs->drv->bdrv_join_options(options, old_options);
937 } else {
938 qdict_join(options, old_options, false);
942 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
943 int open_flags,
944 Error **errp)
946 Error *local_err = NULL;
947 char *value = qemu_opt_get_del(opts, "detect-zeroes");
948 BlockdevDetectZeroesOptions detect_zeroes =
949 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
950 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
951 g_free(value);
952 if (local_err) {
953 error_propagate(errp, local_err);
954 return detect_zeroes;
957 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
958 !(open_flags & BDRV_O_UNMAP))
960 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
961 "without setting discard operation to unmap");
964 return detect_zeroes;
968 * Set open flags for aio engine
970 * Return 0 on success, -1 if the engine specified is invalid
972 int bdrv_parse_aio(const char *mode, int *flags)
974 if (!strcmp(mode, "threads")) {
975 /* do nothing, default */
976 } else if (!strcmp(mode, "native")) {
977 *flags |= BDRV_O_NATIVE_AIO;
978 #ifdef CONFIG_LINUX_IO_URING
979 } else if (!strcmp(mode, "io_uring")) {
980 *flags |= BDRV_O_IO_URING;
981 #endif
982 } else {
983 return -1;
986 return 0;
990 * Set open flags for a given discard mode
992 * Return 0 on success, -1 if the discard mode was invalid.
994 int bdrv_parse_discard_flags(const char *mode, int *flags)
996 *flags &= ~BDRV_O_UNMAP;
998 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
999 /* do nothing */
1000 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1001 *flags |= BDRV_O_UNMAP;
1002 } else {
1003 return -1;
1006 return 0;
1010 * Set open flags for a given cache mode
1012 * Return 0 on success, -1 if the cache mode was invalid.
1014 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1016 *flags &= ~BDRV_O_CACHE_MASK;
1018 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1019 *writethrough = false;
1020 *flags |= BDRV_O_NOCACHE;
1021 } else if (!strcmp(mode, "directsync")) {
1022 *writethrough = true;
1023 *flags |= BDRV_O_NOCACHE;
1024 } else if (!strcmp(mode, "writeback")) {
1025 *writethrough = false;
1026 } else if (!strcmp(mode, "unsafe")) {
1027 *writethrough = false;
1028 *flags |= BDRV_O_NO_FLUSH;
1029 } else if (!strcmp(mode, "writethrough")) {
1030 *writethrough = true;
1031 } else {
1032 return -1;
1035 return 0;
1038 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1040 BlockDriverState *parent = c->opaque;
1041 return g_strdup(bdrv_get_device_or_node_name(parent));
1044 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1046 BlockDriverState *bs = child->opaque;
1047 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1050 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1052 BlockDriverState *bs = child->opaque;
1053 return bdrv_drain_poll(bs, false, NULL, false);
1056 static void bdrv_child_cb_drained_end(BdrvChild *child,
1057 int *drained_end_counter)
1059 BlockDriverState *bs = child->opaque;
1060 bdrv_drained_end_no_poll(bs, drained_end_counter);
1063 static void bdrv_child_cb_attach(BdrvChild *child)
1065 BlockDriverState *bs = child->opaque;
1066 bdrv_apply_subtree_drain(child, bs);
1069 static void bdrv_child_cb_detach(BdrvChild *child)
1071 BlockDriverState *bs = child->opaque;
1072 bdrv_unapply_subtree_drain(child, bs);
1075 static int bdrv_child_cb_inactivate(BdrvChild *child)
1077 BlockDriverState *bs = child->opaque;
1078 assert(bs->open_flags & BDRV_O_INACTIVE);
1079 return 0;
1082 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1083 GSList **ignore, Error **errp)
1085 BlockDriverState *bs = child->opaque;
1086 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1089 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1090 GSList **ignore)
1092 BlockDriverState *bs = child->opaque;
1093 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1097 * Returns the options and flags that a temporary snapshot should get, based on
1098 * the originally requested flags (the originally requested image will have
1099 * flags like a backing file)
1101 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1102 int parent_flags, QDict *parent_options)
1104 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1106 /* For temporary files, unconditional cache=unsafe is fine */
1107 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1108 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1110 /* Copy the read-only and discard options from the parent */
1111 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1112 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1114 /* aio=native doesn't work for cache.direct=off, so disable it for the
1115 * temporary snapshot */
1116 *child_flags &= ~BDRV_O_NATIVE_AIO;
1120 * Returns the options and flags that bs->file should get if a protocol driver
1121 * is expected, based on the given options and flags for the parent BDS
1123 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
1124 int parent_flags, QDict *parent_options)
1126 int flags = parent_flags;
1128 /* Enable protocol handling, disable format probing for bs->file */
1129 flags |= BDRV_O_PROTOCOL;
1131 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1132 * the parent. */
1133 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1134 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1135 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1137 /* Inherit the read-only option from the parent if it's not set */
1138 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1139 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
1141 /* Our block drivers take care to send flushes and respect unmap policy,
1142 * so we can default to enable both on lower layers regardless of the
1143 * corresponding parent options. */
1144 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1146 /* Clear flags that only apply to the top layer */
1147 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
1148 BDRV_O_NO_IO);
1150 *child_flags = flags;
1153 const BdrvChildRole child_file = {
1154 .parent_is_bds = true,
1155 .get_parent_desc = bdrv_child_get_parent_desc,
1156 .inherit_options = bdrv_inherited_options,
1157 .drained_begin = bdrv_child_cb_drained_begin,
1158 .drained_poll = bdrv_child_cb_drained_poll,
1159 .drained_end = bdrv_child_cb_drained_end,
1160 .attach = bdrv_child_cb_attach,
1161 .detach = bdrv_child_cb_detach,
1162 .inactivate = bdrv_child_cb_inactivate,
1163 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1164 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1168 * Returns the options and flags that bs->file should get if the use of formats
1169 * (and not only protocols) is permitted for it, based on the given options and
1170 * flags for the parent BDS
1172 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1173 int parent_flags, QDict *parent_options)
1175 child_file.inherit_options(child_flags, child_options,
1176 parent_flags, parent_options);
1178 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
1181 const BdrvChildRole child_format = {
1182 .parent_is_bds = true,
1183 .get_parent_desc = bdrv_child_get_parent_desc,
1184 .inherit_options = bdrv_inherited_fmt_options,
1185 .drained_begin = bdrv_child_cb_drained_begin,
1186 .drained_poll = bdrv_child_cb_drained_poll,
1187 .drained_end = bdrv_child_cb_drained_end,
1188 .attach = bdrv_child_cb_attach,
1189 .detach = bdrv_child_cb_detach,
1190 .inactivate = bdrv_child_cb_inactivate,
1191 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1192 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1195 static void bdrv_backing_attach(BdrvChild *c)
1197 BlockDriverState *parent = c->opaque;
1198 BlockDriverState *backing_hd = c->bs;
1200 assert(!parent->backing_blocker);
1201 error_setg(&parent->backing_blocker,
1202 "node is used as backing hd of '%s'",
1203 bdrv_get_device_or_node_name(parent));
1205 bdrv_refresh_filename(backing_hd);
1207 parent->open_flags &= ~BDRV_O_NO_BACKING;
1208 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1209 backing_hd->filename);
1210 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1211 backing_hd->drv ? backing_hd->drv->format_name : "");
1213 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1214 /* Otherwise we won't be able to commit or stream */
1215 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1216 parent->backing_blocker);
1217 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1218 parent->backing_blocker);
1220 * We do backup in 3 ways:
1221 * 1. drive backup
1222 * The target bs is new opened, and the source is top BDS
1223 * 2. blockdev backup
1224 * Both the source and the target are top BDSes.
1225 * 3. internal backup(used for block replication)
1226 * Both the source and the target are backing file
1228 * In case 1 and 2, neither the source nor the target is the backing file.
1229 * In case 3, we will block the top BDS, so there is only one block job
1230 * for the top BDS and its backing chain.
1232 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1233 parent->backing_blocker);
1234 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1235 parent->backing_blocker);
1237 bdrv_child_cb_attach(c);
1240 static void bdrv_backing_detach(BdrvChild *c)
1242 BlockDriverState *parent = c->opaque;
1244 assert(parent->backing_blocker);
1245 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1246 error_free(parent->backing_blocker);
1247 parent->backing_blocker = NULL;
1249 bdrv_child_cb_detach(c);
1253 * Returns the options and flags that bs->backing should get, based on the
1254 * given options and flags for the parent BDS
1256 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1257 int parent_flags, QDict *parent_options)
1259 int flags = parent_flags;
1261 /* The cache mode is inherited unmodified for backing files; except WCE,
1262 * which is only applied on the top level (BlockBackend) */
1263 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1264 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1265 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1267 /* backing files always opened read-only */
1268 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1269 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1270 flags &= ~BDRV_O_COPY_ON_READ;
1272 /* snapshot=on is handled on the top layer */
1273 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1275 *child_flags = flags;
1278 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1279 const char *filename, Error **errp)
1281 BlockDriverState *parent = c->opaque;
1282 bool read_only = bdrv_is_read_only(parent);
1283 int ret;
1285 if (read_only) {
1286 ret = bdrv_reopen_set_read_only(parent, false, errp);
1287 if (ret < 0) {
1288 return ret;
1292 ret = bdrv_change_backing_file(parent, filename,
1293 base->drv ? base->drv->format_name : "");
1294 if (ret < 0) {
1295 error_setg_errno(errp, -ret, "Could not update backing file link");
1298 if (read_only) {
1299 bdrv_reopen_set_read_only(parent, true, NULL);
1302 return ret;
1305 const BdrvChildRole child_backing = {
1306 .parent_is_bds = true,
1307 .get_parent_desc = bdrv_child_get_parent_desc,
1308 .attach = bdrv_backing_attach,
1309 .detach = bdrv_backing_detach,
1310 .inherit_options = bdrv_backing_options,
1311 .drained_begin = bdrv_child_cb_drained_begin,
1312 .drained_poll = bdrv_child_cb_drained_poll,
1313 .drained_end = bdrv_child_cb_drained_end,
1314 .inactivate = bdrv_child_cb_inactivate,
1315 .update_filename = bdrv_backing_update_filename,
1316 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1317 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1320 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1322 int open_flags = flags;
1325 * Clear flags that are internal to the block layer before opening the
1326 * image.
1328 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1330 return open_flags;
1333 static void update_flags_from_options(int *flags, QemuOpts *opts)
1335 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1337 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1338 *flags |= BDRV_O_NO_FLUSH;
1341 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1342 *flags |= BDRV_O_NOCACHE;
1345 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1346 *flags |= BDRV_O_RDWR;
1349 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1350 *flags |= BDRV_O_AUTO_RDONLY;
1354 static void update_options_from_flags(QDict *options, int flags)
1356 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1357 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1359 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1360 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1361 flags & BDRV_O_NO_FLUSH);
1363 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1364 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1366 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1367 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1368 flags & BDRV_O_AUTO_RDONLY);
1372 static void bdrv_assign_node_name(BlockDriverState *bs,
1373 const char *node_name,
1374 Error **errp)
1376 char *gen_node_name = NULL;
1378 if (!node_name) {
1379 node_name = gen_node_name = id_generate(ID_BLOCK);
1380 } else if (!id_wellformed(node_name)) {
1382 * Check for empty string or invalid characters, but not if it is
1383 * generated (generated names use characters not available to the user)
1385 error_setg(errp, "Invalid node name");
1386 return;
1389 /* takes care of avoiding namespaces collisions */
1390 if (blk_by_name(node_name)) {
1391 error_setg(errp, "node-name=%s is conflicting with a device id",
1392 node_name);
1393 goto out;
1396 /* takes care of avoiding duplicates node names */
1397 if (bdrv_find_node(node_name)) {
1398 error_setg(errp, "Duplicate node name");
1399 goto out;
1402 /* Make sure that the node name isn't truncated */
1403 if (strlen(node_name) >= sizeof(bs->node_name)) {
1404 error_setg(errp, "Node name too long");
1405 goto out;
1408 /* copy node name into the bs and insert it into the graph list */
1409 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1410 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1411 out:
1412 g_free(gen_node_name);
1415 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1416 const char *node_name, QDict *options,
1417 int open_flags, Error **errp)
1419 Error *local_err = NULL;
1420 int i, ret;
1422 bdrv_assign_node_name(bs, node_name, &local_err);
1423 if (local_err) {
1424 error_propagate(errp, local_err);
1425 return -EINVAL;
1428 bs->drv = drv;
1429 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1430 bs->opaque = g_malloc0(drv->instance_size);
1432 if (drv->bdrv_file_open) {
1433 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1434 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1435 } else if (drv->bdrv_open) {
1436 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1437 } else {
1438 ret = 0;
1441 if (ret < 0) {
1442 if (local_err) {
1443 error_propagate(errp, local_err);
1444 } else if (bs->filename[0]) {
1445 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1446 } else {
1447 error_setg_errno(errp, -ret, "Could not open image");
1449 goto open_failed;
1452 ret = refresh_total_sectors(bs, bs->total_sectors);
1453 if (ret < 0) {
1454 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1455 return ret;
1458 bdrv_refresh_limits(bs, &local_err);
1459 if (local_err) {
1460 error_propagate(errp, local_err);
1461 return -EINVAL;
1464 assert(bdrv_opt_mem_align(bs) != 0);
1465 assert(bdrv_min_mem_align(bs) != 0);
1466 assert(is_power_of_2(bs->bl.request_alignment));
1468 for (i = 0; i < bs->quiesce_counter; i++) {
1469 if (drv->bdrv_co_drain_begin) {
1470 drv->bdrv_co_drain_begin(bs);
1474 return 0;
1475 open_failed:
1476 bs->drv = NULL;
1477 if (bs->file != NULL) {
1478 bdrv_unref_child(bs, bs->file);
1479 bs->file = NULL;
1481 g_free(bs->opaque);
1482 bs->opaque = NULL;
1483 return ret;
1486 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1487 int flags, Error **errp)
1489 BlockDriverState *bs;
1490 int ret;
1492 bs = bdrv_new();
1493 bs->open_flags = flags;
1494 bs->explicit_options = qdict_new();
1495 bs->options = qdict_new();
1496 bs->opaque = NULL;
1498 update_options_from_flags(bs->options, flags);
1500 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1501 if (ret < 0) {
1502 qobject_unref(bs->explicit_options);
1503 bs->explicit_options = NULL;
1504 qobject_unref(bs->options);
1505 bs->options = NULL;
1506 bdrv_unref(bs);
1507 return NULL;
1510 return bs;
1513 QemuOptsList bdrv_runtime_opts = {
1514 .name = "bdrv_common",
1515 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1516 .desc = {
1518 .name = "node-name",
1519 .type = QEMU_OPT_STRING,
1520 .help = "Node name of the block device node",
1523 .name = "driver",
1524 .type = QEMU_OPT_STRING,
1525 .help = "Block driver to use for the node",
1528 .name = BDRV_OPT_CACHE_DIRECT,
1529 .type = QEMU_OPT_BOOL,
1530 .help = "Bypass software writeback cache on the host",
1533 .name = BDRV_OPT_CACHE_NO_FLUSH,
1534 .type = QEMU_OPT_BOOL,
1535 .help = "Ignore flush requests",
1538 .name = BDRV_OPT_READ_ONLY,
1539 .type = QEMU_OPT_BOOL,
1540 .help = "Node is opened in read-only mode",
1543 .name = BDRV_OPT_AUTO_READ_ONLY,
1544 .type = QEMU_OPT_BOOL,
1545 .help = "Node can become read-only if opening read-write fails",
1548 .name = "detect-zeroes",
1549 .type = QEMU_OPT_STRING,
1550 .help = "try to optimize zero writes (off, on, unmap)",
1553 .name = BDRV_OPT_DISCARD,
1554 .type = QEMU_OPT_STRING,
1555 .help = "discard operation (ignore/off, unmap/on)",
1558 .name = BDRV_OPT_FORCE_SHARE,
1559 .type = QEMU_OPT_BOOL,
1560 .help = "always accept other writers (default: off)",
1562 { /* end of list */ }
1566 static QemuOptsList fallback_create_opts = {
1567 .name = "fallback-create-opts",
1568 .head = QTAILQ_HEAD_INITIALIZER(fallback_create_opts.head),
1569 .desc = {
1571 .name = BLOCK_OPT_SIZE,
1572 .type = QEMU_OPT_SIZE,
1573 .help = "Virtual disk size"
1576 .name = BLOCK_OPT_PREALLOC,
1577 .type = QEMU_OPT_STRING,
1578 .help = "Preallocation mode (allowed values: off)"
1580 { /* end of list */ }
1585 * Common part for opening disk images and files
1587 * Removes all processed options from *options.
1589 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1590 QDict *options, Error **errp)
1592 int ret, open_flags;
1593 const char *filename;
1594 const char *driver_name = NULL;
1595 const char *node_name = NULL;
1596 const char *discard;
1597 QemuOpts *opts;
1598 BlockDriver *drv;
1599 Error *local_err = NULL;
1601 assert(bs->file == NULL);
1602 assert(options != NULL && bs->options != options);
1604 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1605 qemu_opts_absorb_qdict(opts, options, &local_err);
1606 if (local_err) {
1607 error_propagate(errp, local_err);
1608 ret = -EINVAL;
1609 goto fail_opts;
1612 update_flags_from_options(&bs->open_flags, opts);
1614 driver_name = qemu_opt_get(opts, "driver");
1615 drv = bdrv_find_format(driver_name);
1616 assert(drv != NULL);
1618 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1620 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1621 error_setg(errp,
1622 BDRV_OPT_FORCE_SHARE
1623 "=on can only be used with read-only images");
1624 ret = -EINVAL;
1625 goto fail_opts;
1628 if (file != NULL) {
1629 bdrv_refresh_filename(blk_bs(file));
1630 filename = blk_bs(file)->filename;
1631 } else {
1633 * Caution: while qdict_get_try_str() is fine, getting
1634 * non-string types would require more care. When @options
1635 * come from -blockdev or blockdev_add, its members are typed
1636 * according to the QAPI schema, but when they come from
1637 * -drive, they're all QString.
1639 filename = qdict_get_try_str(options, "filename");
1642 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1643 error_setg(errp, "The '%s' block driver requires a file name",
1644 drv->format_name);
1645 ret = -EINVAL;
1646 goto fail_opts;
1649 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1650 drv->format_name);
1652 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1654 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1655 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1656 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1657 } else {
1658 ret = -ENOTSUP;
1660 if (ret < 0) {
1661 error_setg(errp,
1662 !bs->read_only && bdrv_is_whitelisted(drv, true)
1663 ? "Driver '%s' can only be used for read-only devices"
1664 : "Driver '%s' is not whitelisted",
1665 drv->format_name);
1666 goto fail_opts;
1670 /* bdrv_new() and bdrv_close() make it so */
1671 assert(atomic_read(&bs->copy_on_read) == 0);
1673 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1674 if (!bs->read_only) {
1675 bdrv_enable_copy_on_read(bs);
1676 } else {
1677 error_setg(errp, "Can't use copy-on-read on read-only device");
1678 ret = -EINVAL;
1679 goto fail_opts;
1683 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1684 if (discard != NULL) {
1685 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1686 error_setg(errp, "Invalid discard option");
1687 ret = -EINVAL;
1688 goto fail_opts;
1692 bs->detect_zeroes =
1693 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1694 if (local_err) {
1695 error_propagate(errp, local_err);
1696 ret = -EINVAL;
1697 goto fail_opts;
1700 if (filename != NULL) {
1701 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1702 } else {
1703 bs->filename[0] = '\0';
1705 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1707 /* Open the image, either directly or using a protocol */
1708 open_flags = bdrv_open_flags(bs, bs->open_flags);
1709 node_name = qemu_opt_get(opts, "node-name");
1711 assert(!drv->bdrv_file_open || file == NULL);
1712 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1713 if (ret < 0) {
1714 goto fail_opts;
1717 qemu_opts_del(opts);
1718 return 0;
1720 fail_opts:
1721 qemu_opts_del(opts);
1722 return ret;
1725 static QDict *parse_json_filename(const char *filename, Error **errp)
1727 QObject *options_obj;
1728 QDict *options;
1729 int ret;
1731 ret = strstart(filename, "json:", &filename);
1732 assert(ret);
1734 options_obj = qobject_from_json(filename, errp);
1735 if (!options_obj) {
1736 error_prepend(errp, "Could not parse the JSON options: ");
1737 return NULL;
1740 options = qobject_to(QDict, options_obj);
1741 if (!options) {
1742 qobject_unref(options_obj);
1743 error_setg(errp, "Invalid JSON object given");
1744 return NULL;
1747 qdict_flatten(options);
1749 return options;
1752 static void parse_json_protocol(QDict *options, const char **pfilename,
1753 Error **errp)
1755 QDict *json_options;
1756 Error *local_err = NULL;
1758 /* Parse json: pseudo-protocol */
1759 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1760 return;
1763 json_options = parse_json_filename(*pfilename, &local_err);
1764 if (local_err) {
1765 error_propagate(errp, local_err);
1766 return;
1769 /* Options given in the filename have lower priority than options
1770 * specified directly */
1771 qdict_join(options, json_options, false);
1772 qobject_unref(json_options);
1773 *pfilename = NULL;
1777 * Fills in default options for opening images and converts the legacy
1778 * filename/flags pair to option QDict entries.
1779 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1780 * block driver has been specified explicitly.
1782 static int bdrv_fill_options(QDict **options, const char *filename,
1783 int *flags, Error **errp)
1785 const char *drvname;
1786 bool protocol = *flags & BDRV_O_PROTOCOL;
1787 bool parse_filename = false;
1788 BlockDriver *drv = NULL;
1789 Error *local_err = NULL;
1792 * Caution: while qdict_get_try_str() is fine, getting non-string
1793 * types would require more care. When @options come from
1794 * -blockdev or blockdev_add, its members are typed according to
1795 * the QAPI schema, but when they come from -drive, they're all
1796 * QString.
1798 drvname = qdict_get_try_str(*options, "driver");
1799 if (drvname) {
1800 drv = bdrv_find_format(drvname);
1801 if (!drv) {
1802 error_setg(errp, "Unknown driver '%s'", drvname);
1803 return -ENOENT;
1805 /* If the user has explicitly specified the driver, this choice should
1806 * override the BDRV_O_PROTOCOL flag */
1807 protocol = drv->bdrv_file_open;
1810 if (protocol) {
1811 *flags |= BDRV_O_PROTOCOL;
1812 } else {
1813 *flags &= ~BDRV_O_PROTOCOL;
1816 /* Translate cache options from flags into options */
1817 update_options_from_flags(*options, *flags);
1819 /* Fetch the file name from the options QDict if necessary */
1820 if (protocol && filename) {
1821 if (!qdict_haskey(*options, "filename")) {
1822 qdict_put_str(*options, "filename", filename);
1823 parse_filename = true;
1824 } else {
1825 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1826 "the same time");
1827 return -EINVAL;
1831 /* Find the right block driver */
1832 /* See cautionary note on accessing @options above */
1833 filename = qdict_get_try_str(*options, "filename");
1835 if (!drvname && protocol) {
1836 if (filename) {
1837 drv = bdrv_find_protocol(filename, parse_filename, errp);
1838 if (!drv) {
1839 return -EINVAL;
1842 drvname = drv->format_name;
1843 qdict_put_str(*options, "driver", drvname);
1844 } else {
1845 error_setg(errp, "Must specify either driver or file");
1846 return -EINVAL;
1850 assert(drv || !protocol);
1852 /* Driver-specific filename parsing */
1853 if (drv && drv->bdrv_parse_filename && parse_filename) {
1854 drv->bdrv_parse_filename(filename, *options, &local_err);
1855 if (local_err) {
1856 error_propagate(errp, local_err);
1857 return -EINVAL;
1860 if (!drv->bdrv_needs_filename) {
1861 qdict_del(*options, "filename");
1865 return 0;
1868 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1869 uint64_t perm, uint64_t shared,
1870 GSList *ignore_children,
1871 bool *tighten_restrictions, Error **errp);
1872 static void bdrv_child_abort_perm_update(BdrvChild *c);
1873 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1874 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1875 uint64_t *shared_perm);
1877 typedef struct BlockReopenQueueEntry {
1878 bool prepared;
1879 bool perms_checked;
1880 BDRVReopenState state;
1881 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1882 } BlockReopenQueueEntry;
1885 * Return the flags that @bs will have after the reopens in @q have
1886 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1887 * return the current flags.
1889 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1891 BlockReopenQueueEntry *entry;
1893 if (q != NULL) {
1894 QTAILQ_FOREACH(entry, q, entry) {
1895 if (entry->state.bs == bs) {
1896 return entry->state.flags;
1901 return bs->open_flags;
1904 /* Returns whether the image file can be written to after the reopen queue @q
1905 * has been successfully applied, or right now if @q is NULL. */
1906 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1907 BlockReopenQueue *q)
1909 int flags = bdrv_reopen_get_flags(q, bs);
1911 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1915 * Return whether the BDS can be written to. This is not necessarily
1916 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1917 * be written to but do not count as read-only images.
1919 bool bdrv_is_writable(BlockDriverState *bs)
1921 return bdrv_is_writable_after_reopen(bs, NULL);
1924 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1925 BdrvChild *c, const BdrvChildRole *role,
1926 BlockReopenQueue *reopen_queue,
1927 uint64_t parent_perm, uint64_t parent_shared,
1928 uint64_t *nperm, uint64_t *nshared)
1930 assert(bs->drv && bs->drv->bdrv_child_perm);
1931 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1932 parent_perm, parent_shared,
1933 nperm, nshared);
1934 /* TODO Take force_share from reopen_queue */
1935 if (child_bs && child_bs->force_share) {
1936 *nshared = BLK_PERM_ALL;
1941 * Check whether permissions on this node can be changed in a way that
1942 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1943 * permissions of all its parents. This involves checking whether all necessary
1944 * permission changes to child nodes can be performed.
1946 * Will set *tighten_restrictions to true if and only if new permissions have to
1947 * be taken or currently shared permissions are to be unshared. Otherwise,
1948 * errors are not fatal as long as the caller accepts that the restrictions
1949 * remain tighter than they need to be. The caller still has to abort the
1950 * transaction.
1951 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1952 * encounter fatal errors even though no restrictions are to be tightened. For
1953 * example, changing a node from RW to RO will fail if the WRITE permission is
1954 * to be kept.
1956 * A call to this function must always be followed by a call to bdrv_set_perm()
1957 * or bdrv_abort_perm_update().
1959 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1960 uint64_t cumulative_perms,
1961 uint64_t cumulative_shared_perms,
1962 GSList *ignore_children,
1963 bool *tighten_restrictions, Error **errp)
1965 BlockDriver *drv = bs->drv;
1966 BdrvChild *c;
1967 int ret;
1969 assert(!q || !tighten_restrictions);
1971 if (tighten_restrictions) {
1972 uint64_t current_perms, current_shared;
1973 uint64_t added_perms, removed_shared_perms;
1975 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1977 added_perms = cumulative_perms & ~current_perms;
1978 removed_shared_perms = current_shared & ~cumulative_shared_perms;
1980 *tighten_restrictions = added_perms || removed_shared_perms;
1983 /* Write permissions never work with read-only images */
1984 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1985 !bdrv_is_writable_after_reopen(bs, q))
1987 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
1988 error_setg(errp, "Block node is read-only");
1989 } else {
1990 uint64_t current_perms, current_shared;
1991 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1992 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
1993 error_setg(errp, "Cannot make block node read-only, there is "
1994 "a writer on it");
1995 } else {
1996 error_setg(errp, "Cannot make block node read-only and create "
1997 "a writer on it");
2001 return -EPERM;
2004 /* Check this node */
2005 if (!drv) {
2006 return 0;
2009 if (drv->bdrv_check_perm) {
2010 return drv->bdrv_check_perm(bs, cumulative_perms,
2011 cumulative_shared_perms, errp);
2014 /* Drivers that never have children can omit .bdrv_child_perm() */
2015 if (!drv->bdrv_child_perm) {
2016 assert(QLIST_EMPTY(&bs->children));
2017 return 0;
2020 /* Check all children */
2021 QLIST_FOREACH(c, &bs->children, next) {
2022 uint64_t cur_perm, cur_shared;
2023 bool child_tighten_restr;
2025 bdrv_child_perm(bs, c->bs, c, c->role, q,
2026 cumulative_perms, cumulative_shared_perms,
2027 &cur_perm, &cur_shared);
2028 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2029 tighten_restrictions ? &child_tighten_restr
2030 : NULL,
2031 errp);
2032 if (tighten_restrictions) {
2033 *tighten_restrictions |= child_tighten_restr;
2035 if (ret < 0) {
2036 return ret;
2040 return 0;
2044 * Notifies drivers that after a previous bdrv_check_perm() call, the
2045 * permission update is not performed and any preparations made for it (e.g.
2046 * taken file locks) need to be undone.
2048 * This function recursively notifies all child nodes.
2050 static void bdrv_abort_perm_update(BlockDriverState *bs)
2052 BlockDriver *drv = bs->drv;
2053 BdrvChild *c;
2055 if (!drv) {
2056 return;
2059 if (drv->bdrv_abort_perm_update) {
2060 drv->bdrv_abort_perm_update(bs);
2063 QLIST_FOREACH(c, &bs->children, next) {
2064 bdrv_child_abort_perm_update(c);
2068 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2069 uint64_t cumulative_shared_perms)
2071 BlockDriver *drv = bs->drv;
2072 BdrvChild *c;
2074 if (!drv) {
2075 return;
2078 /* Update this node */
2079 if (drv->bdrv_set_perm) {
2080 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2083 /* Drivers that never have children can omit .bdrv_child_perm() */
2084 if (!drv->bdrv_child_perm) {
2085 assert(QLIST_EMPTY(&bs->children));
2086 return;
2089 /* Update all children */
2090 QLIST_FOREACH(c, &bs->children, next) {
2091 uint64_t cur_perm, cur_shared;
2092 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2093 cumulative_perms, cumulative_shared_perms,
2094 &cur_perm, &cur_shared);
2095 bdrv_child_set_perm(c, cur_perm, cur_shared);
2099 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2100 uint64_t *shared_perm)
2102 BdrvChild *c;
2103 uint64_t cumulative_perms = 0;
2104 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2106 QLIST_FOREACH(c, &bs->parents, next_parent) {
2107 cumulative_perms |= c->perm;
2108 cumulative_shared_perms &= c->shared_perm;
2111 *perm = cumulative_perms;
2112 *shared_perm = cumulative_shared_perms;
2115 static char *bdrv_child_user_desc(BdrvChild *c)
2117 if (c->role->get_parent_desc) {
2118 return c->role->get_parent_desc(c);
2121 return g_strdup("another user");
2124 char *bdrv_perm_names(uint64_t perm)
2126 struct perm_name {
2127 uint64_t perm;
2128 const char *name;
2129 } permissions[] = {
2130 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2131 { BLK_PERM_WRITE, "write" },
2132 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2133 { BLK_PERM_RESIZE, "resize" },
2134 { BLK_PERM_GRAPH_MOD, "change children" },
2135 { 0, NULL }
2138 GString *result = g_string_sized_new(30);
2139 struct perm_name *p;
2141 for (p = permissions; p->name; p++) {
2142 if (perm & p->perm) {
2143 if (result->len > 0) {
2144 g_string_append(result, ", ");
2146 g_string_append(result, p->name);
2150 return g_string_free(result, FALSE);
2154 * Checks whether a new reference to @bs can be added if the new user requires
2155 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2156 * set, the BdrvChild objects in this list are ignored in the calculations;
2157 * this allows checking permission updates for an existing reference.
2159 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2161 * Needs to be followed by a call to either bdrv_set_perm() or
2162 * bdrv_abort_perm_update(). */
2163 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2164 uint64_t new_used_perm,
2165 uint64_t new_shared_perm,
2166 GSList *ignore_children,
2167 bool *tighten_restrictions,
2168 Error **errp)
2170 BdrvChild *c;
2171 uint64_t cumulative_perms = new_used_perm;
2172 uint64_t cumulative_shared_perms = new_shared_perm;
2174 assert(!q || !tighten_restrictions);
2176 /* There is no reason why anyone couldn't tolerate write_unchanged */
2177 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2179 QLIST_FOREACH(c, &bs->parents, next_parent) {
2180 if (g_slist_find(ignore_children, c)) {
2181 continue;
2184 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2185 char *user = bdrv_child_user_desc(c);
2186 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2188 if (tighten_restrictions) {
2189 *tighten_restrictions = true;
2192 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2193 "allow '%s' on %s",
2194 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2195 g_free(user);
2196 g_free(perm_names);
2197 return -EPERM;
2200 if ((c->perm & new_shared_perm) != c->perm) {
2201 char *user = bdrv_child_user_desc(c);
2202 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2204 if (tighten_restrictions) {
2205 *tighten_restrictions = true;
2208 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2209 "'%s' on %s",
2210 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2211 g_free(user);
2212 g_free(perm_names);
2213 return -EPERM;
2216 cumulative_perms |= c->perm;
2217 cumulative_shared_perms &= c->shared_perm;
2220 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2221 ignore_children, tighten_restrictions, errp);
2224 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2225 * bdrv_child_abort_perm_update(). */
2226 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2227 uint64_t perm, uint64_t shared,
2228 GSList *ignore_children,
2229 bool *tighten_restrictions, Error **errp)
2231 int ret;
2233 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2234 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2235 tighten_restrictions, errp);
2236 g_slist_free(ignore_children);
2238 if (ret < 0) {
2239 return ret;
2242 if (!c->has_backup_perm) {
2243 c->has_backup_perm = true;
2244 c->backup_perm = c->perm;
2245 c->backup_shared_perm = c->shared_perm;
2248 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2249 * same child twice during check_perm procedure
2252 c->perm = perm;
2253 c->shared_perm = shared;
2255 return 0;
2258 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2260 uint64_t cumulative_perms, cumulative_shared_perms;
2262 c->has_backup_perm = false;
2264 c->perm = perm;
2265 c->shared_perm = shared;
2267 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2268 &cumulative_shared_perms);
2269 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2272 static void bdrv_child_abort_perm_update(BdrvChild *c)
2274 if (c->has_backup_perm) {
2275 c->perm = c->backup_perm;
2276 c->shared_perm = c->backup_shared_perm;
2277 c->has_backup_perm = false;
2280 bdrv_abort_perm_update(c->bs);
2283 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2284 Error **errp)
2286 Error *local_err = NULL;
2287 int ret;
2288 bool tighten_restrictions;
2290 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2291 &tighten_restrictions, &local_err);
2292 if (ret < 0) {
2293 bdrv_child_abort_perm_update(c);
2294 if (tighten_restrictions) {
2295 error_propagate(errp, local_err);
2296 } else {
2298 * Our caller may intend to only loosen restrictions and
2299 * does not expect this function to fail. Errors are not
2300 * fatal in such a case, so we can just hide them from our
2301 * caller.
2303 error_free(local_err);
2304 ret = 0;
2306 return ret;
2309 bdrv_child_set_perm(c, perm, shared);
2311 return 0;
2314 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2316 uint64_t parent_perms, parent_shared;
2317 uint64_t perms, shared;
2319 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2320 bdrv_child_perm(bs, c->bs, c, c->role, NULL, parent_perms, parent_shared,
2321 &perms, &shared);
2323 return bdrv_child_try_set_perm(c, perms, shared, errp);
2326 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2327 const BdrvChildRole *role,
2328 BlockReopenQueue *reopen_queue,
2329 uint64_t perm, uint64_t shared,
2330 uint64_t *nperm, uint64_t *nshared)
2332 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2333 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2336 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2337 const BdrvChildRole *role,
2338 BlockReopenQueue *reopen_queue,
2339 uint64_t perm, uint64_t shared,
2340 uint64_t *nperm, uint64_t *nshared)
2342 bool backing = (role == &child_backing);
2343 assert(role == &child_backing || role == &child_file);
2345 if (!backing) {
2346 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2348 /* Apart from the modifications below, the same permissions are
2349 * forwarded and left alone as for filters */
2350 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2351 &perm, &shared);
2353 /* Format drivers may touch metadata even if the guest doesn't write */
2354 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2355 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2358 /* bs->file always needs to be consistent because of the metadata. We
2359 * can never allow other users to resize or write to it. */
2360 if (!(flags & BDRV_O_NO_IO)) {
2361 perm |= BLK_PERM_CONSISTENT_READ;
2363 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2364 } else {
2365 /* We want consistent read from backing files if the parent needs it.
2366 * No other operations are performed on backing files. */
2367 perm &= BLK_PERM_CONSISTENT_READ;
2369 /* If the parent can deal with changing data, we're okay with a
2370 * writable and resizable backing file. */
2371 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2372 if (shared & BLK_PERM_WRITE) {
2373 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2374 } else {
2375 shared = 0;
2378 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2379 BLK_PERM_WRITE_UNCHANGED;
2382 if (bs->open_flags & BDRV_O_INACTIVE) {
2383 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2386 *nperm = perm;
2387 *nshared = shared;
2390 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2392 static const uint64_t permissions[] = {
2393 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2394 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2395 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2396 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2397 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2400 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2401 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2403 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2405 return permissions[qapi_perm];
2408 static void bdrv_replace_child_noperm(BdrvChild *child,
2409 BlockDriverState *new_bs)
2411 BlockDriverState *old_bs = child->bs;
2412 int new_bs_quiesce_counter;
2413 int drain_saldo;
2415 assert(!child->frozen);
2417 if (old_bs && new_bs) {
2418 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2421 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2422 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2425 * If the new child node is drained but the old one was not, flush
2426 * all outstanding requests to the old child node.
2428 while (drain_saldo > 0 && child->role->drained_begin) {
2429 bdrv_parent_drained_begin_single(child, true);
2430 drain_saldo--;
2433 if (old_bs) {
2434 /* Detach first so that the recursive drain sections coming from @child
2435 * are already gone and we only end the drain sections that came from
2436 * elsewhere. */
2437 if (child->role->detach) {
2438 child->role->detach(child);
2440 QLIST_REMOVE(child, next_parent);
2443 child->bs = new_bs;
2445 if (new_bs) {
2446 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2449 * Detaching the old node may have led to the new node's
2450 * quiesce_counter having been decreased. Not a problem, we
2451 * just need to recognize this here and then invoke
2452 * drained_end appropriately more often.
2454 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2455 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2457 /* Attach only after starting new drained sections, so that recursive
2458 * drain sections coming from @child don't get an extra .drained_begin
2459 * callback. */
2460 if (child->role->attach) {
2461 child->role->attach(child);
2466 * If the old child node was drained but the new one is not, allow
2467 * requests to come in only after the new node has been attached.
2469 while (drain_saldo < 0 && child->role->drained_end) {
2470 bdrv_parent_drained_end_single(child);
2471 drain_saldo++;
2476 * Updates @child to change its reference to point to @new_bs, including
2477 * checking and applying the necessary permisson updates both to the old node
2478 * and to @new_bs.
2480 * NULL is passed as @new_bs for removing the reference before freeing @child.
2482 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2483 * function uses bdrv_set_perm() to update the permissions according to the new
2484 * reference that @new_bs gets.
2486 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2488 BlockDriverState *old_bs = child->bs;
2489 uint64_t perm, shared_perm;
2491 bdrv_replace_child_noperm(child, new_bs);
2494 * Start with the new node's permissions. If @new_bs is a (direct
2495 * or indirect) child of @old_bs, we must complete the permission
2496 * update on @new_bs before we loosen the restrictions on @old_bs.
2497 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2498 * updating the permissions of @new_bs, and thus not purely loosen
2499 * restrictions.
2501 if (new_bs) {
2502 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2503 bdrv_set_perm(new_bs, perm, shared_perm);
2506 if (old_bs) {
2507 /* Update permissions for old node. This is guaranteed to succeed
2508 * because we're just taking a parent away, so we're loosening
2509 * restrictions. */
2510 bool tighten_restrictions;
2511 int ret;
2513 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2514 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2515 &tighten_restrictions, NULL);
2516 assert(tighten_restrictions == false);
2517 if (ret < 0) {
2518 /* We only tried to loosen restrictions, so errors are not fatal */
2519 bdrv_abort_perm_update(old_bs);
2520 } else {
2521 bdrv_set_perm(old_bs, perm, shared_perm);
2524 /* When the parent requiring a non-default AioContext is removed, the
2525 * node moves back to the main AioContext */
2526 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2531 * This function steals the reference to child_bs from the caller.
2532 * That reference is later dropped by bdrv_root_unref_child().
2534 * On failure NULL is returned, errp is set and the reference to
2535 * child_bs is also dropped.
2537 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2538 * (unless @child_bs is already in @ctx).
2540 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2541 const char *child_name,
2542 const BdrvChildRole *child_role,
2543 AioContext *ctx,
2544 uint64_t perm, uint64_t shared_perm,
2545 void *opaque, Error **errp)
2547 BdrvChild *child;
2548 Error *local_err = NULL;
2549 int ret;
2551 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2552 errp);
2553 if (ret < 0) {
2554 bdrv_abort_perm_update(child_bs);
2555 bdrv_unref(child_bs);
2556 return NULL;
2559 child = g_new(BdrvChild, 1);
2560 *child = (BdrvChild) {
2561 .bs = NULL,
2562 .name = g_strdup(child_name),
2563 .role = child_role,
2564 .perm = perm,
2565 .shared_perm = shared_perm,
2566 .opaque = opaque,
2569 /* If the AioContexts don't match, first try to move the subtree of
2570 * child_bs into the AioContext of the new parent. If this doesn't work,
2571 * try moving the parent into the AioContext of child_bs instead. */
2572 if (bdrv_get_aio_context(child_bs) != ctx) {
2573 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2574 if (ret < 0 && child_role->can_set_aio_ctx) {
2575 GSList *ignore = g_slist_prepend(NULL, child);
2576 ctx = bdrv_get_aio_context(child_bs);
2577 if (child_role->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2578 error_free(local_err);
2579 ret = 0;
2580 g_slist_free(ignore);
2581 ignore = g_slist_prepend(NULL, child);
2582 child_role->set_aio_ctx(child, ctx, &ignore);
2584 g_slist_free(ignore);
2586 if (ret < 0) {
2587 error_propagate(errp, local_err);
2588 g_free(child);
2589 bdrv_abort_perm_update(child_bs);
2590 return NULL;
2594 /* This performs the matching bdrv_set_perm() for the above check. */
2595 bdrv_replace_child(child, child_bs);
2597 return child;
2601 * This function transfers the reference to child_bs from the caller
2602 * to parent_bs. That reference is later dropped by parent_bs on
2603 * bdrv_close() or if someone calls bdrv_unref_child().
2605 * On failure NULL is returned, errp is set and the reference to
2606 * child_bs is also dropped.
2608 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2609 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2611 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2612 BlockDriverState *child_bs,
2613 const char *child_name,
2614 const BdrvChildRole *child_role,
2615 Error **errp)
2617 BdrvChild *child;
2618 uint64_t perm, shared_perm;
2620 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2622 assert(parent_bs->drv);
2623 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2624 perm, shared_perm, &perm, &shared_perm);
2626 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2627 bdrv_get_aio_context(parent_bs),
2628 perm, shared_perm, parent_bs, errp);
2629 if (child == NULL) {
2630 return NULL;
2633 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2634 return child;
2637 static void bdrv_detach_child(BdrvChild *child)
2639 QLIST_SAFE_REMOVE(child, next);
2641 bdrv_replace_child(child, NULL);
2643 g_free(child->name);
2644 g_free(child);
2647 void bdrv_root_unref_child(BdrvChild *child)
2649 BlockDriverState *child_bs;
2651 child_bs = child->bs;
2652 bdrv_detach_child(child);
2653 bdrv_unref(child_bs);
2657 * Clear all inherits_from pointers from children and grandchildren of
2658 * @root that point to @root, where necessary.
2660 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2662 BdrvChild *c;
2664 if (child->bs->inherits_from == root) {
2666 * Remove inherits_from only when the last reference between root and
2667 * child->bs goes away.
2669 QLIST_FOREACH(c, &root->children, next) {
2670 if (c != child && c->bs == child->bs) {
2671 break;
2674 if (c == NULL) {
2675 child->bs->inherits_from = NULL;
2679 QLIST_FOREACH(c, &child->bs->children, next) {
2680 bdrv_unset_inherits_from(root, c);
2684 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2686 if (child == NULL) {
2687 return;
2690 bdrv_unset_inherits_from(parent, child);
2691 bdrv_root_unref_child(child);
2695 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2697 BdrvChild *c;
2698 QLIST_FOREACH(c, &bs->parents, next_parent) {
2699 if (c->role->change_media) {
2700 c->role->change_media(c, load);
2705 /* Return true if you can reach parent going through child->inherits_from
2706 * recursively. If parent or child are NULL, return false */
2707 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2708 BlockDriverState *parent)
2710 while (child && child != parent) {
2711 child = child->inherits_from;
2714 return child != NULL;
2718 * Sets the backing file link of a BDS. A new reference is created; callers
2719 * which don't need their own reference any more must call bdrv_unref().
2721 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2722 Error **errp)
2724 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2725 bdrv_inherits_from_recursive(backing_hd, bs);
2727 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2728 return;
2731 if (backing_hd) {
2732 bdrv_ref(backing_hd);
2735 if (bs->backing) {
2736 bdrv_unref_child(bs, bs->backing);
2739 if (!backing_hd) {
2740 bs->backing = NULL;
2741 goto out;
2744 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2745 errp);
2746 /* If backing_hd was already part of bs's backing chain, and
2747 * inherits_from pointed recursively to bs then let's update it to
2748 * point directly to bs (else it will become NULL). */
2749 if (bs->backing && update_inherits_from) {
2750 backing_hd->inherits_from = bs;
2753 out:
2754 bdrv_refresh_limits(bs, NULL);
2758 * Opens the backing file for a BlockDriverState if not yet open
2760 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2761 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2762 * itself, all options starting with "${bdref_key}." are considered part of the
2763 * BlockdevRef.
2765 * TODO Can this be unified with bdrv_open_image()?
2767 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2768 const char *bdref_key, Error **errp)
2770 char *backing_filename = NULL;
2771 char *bdref_key_dot;
2772 const char *reference = NULL;
2773 int ret = 0;
2774 bool implicit_backing = false;
2775 BlockDriverState *backing_hd;
2776 QDict *options;
2777 QDict *tmp_parent_options = NULL;
2778 Error *local_err = NULL;
2780 if (bs->backing != NULL) {
2781 goto free_exit;
2784 /* NULL means an empty set of options */
2785 if (parent_options == NULL) {
2786 tmp_parent_options = qdict_new();
2787 parent_options = tmp_parent_options;
2790 bs->open_flags &= ~BDRV_O_NO_BACKING;
2792 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2793 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2794 g_free(bdref_key_dot);
2797 * Caution: while qdict_get_try_str() is fine, getting non-string
2798 * types would require more care. When @parent_options come from
2799 * -blockdev or blockdev_add, its members are typed according to
2800 * the QAPI schema, but when they come from -drive, they're all
2801 * QString.
2803 reference = qdict_get_try_str(parent_options, bdref_key);
2804 if (reference || qdict_haskey(options, "file.filename")) {
2805 /* keep backing_filename NULL */
2806 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2807 qobject_unref(options);
2808 goto free_exit;
2809 } else {
2810 if (qdict_size(options) == 0) {
2811 /* If the user specifies options that do not modify the
2812 * backing file's behavior, we might still consider it the
2813 * implicit backing file. But it's easier this way, and
2814 * just specifying some of the backing BDS's options is
2815 * only possible with -drive anyway (otherwise the QAPI
2816 * schema forces the user to specify everything). */
2817 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2820 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2821 if (local_err) {
2822 ret = -EINVAL;
2823 error_propagate(errp, local_err);
2824 qobject_unref(options);
2825 goto free_exit;
2829 if (!bs->drv || !bs->drv->supports_backing) {
2830 ret = -EINVAL;
2831 error_setg(errp, "Driver doesn't support backing files");
2832 qobject_unref(options);
2833 goto free_exit;
2836 if (!reference &&
2837 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2838 qdict_put_str(options, "driver", bs->backing_format);
2841 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2842 &child_backing, errp);
2843 if (!backing_hd) {
2844 bs->open_flags |= BDRV_O_NO_BACKING;
2845 error_prepend(errp, "Could not open backing file: ");
2846 ret = -EINVAL;
2847 goto free_exit;
2850 if (implicit_backing) {
2851 bdrv_refresh_filename(backing_hd);
2852 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2853 backing_hd->filename);
2856 /* Hook up the backing file link; drop our reference, bs owns the
2857 * backing_hd reference now */
2858 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2859 bdrv_unref(backing_hd);
2860 if (local_err) {
2861 error_propagate(errp, local_err);
2862 ret = -EINVAL;
2863 goto free_exit;
2866 qdict_del(parent_options, bdref_key);
2868 free_exit:
2869 g_free(backing_filename);
2870 qobject_unref(tmp_parent_options);
2871 return ret;
2874 static BlockDriverState *
2875 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2876 BlockDriverState *parent, const BdrvChildRole *child_role,
2877 bool allow_none, Error **errp)
2879 BlockDriverState *bs = NULL;
2880 QDict *image_options;
2881 char *bdref_key_dot;
2882 const char *reference;
2884 assert(child_role != NULL);
2886 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2887 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2888 g_free(bdref_key_dot);
2891 * Caution: while qdict_get_try_str() is fine, getting non-string
2892 * types would require more care. When @options come from
2893 * -blockdev or blockdev_add, its members are typed according to
2894 * the QAPI schema, but when they come from -drive, they're all
2895 * QString.
2897 reference = qdict_get_try_str(options, bdref_key);
2898 if (!filename && !reference && !qdict_size(image_options)) {
2899 if (!allow_none) {
2900 error_setg(errp, "A block device must be specified for \"%s\"",
2901 bdref_key);
2903 qobject_unref(image_options);
2904 goto done;
2907 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2908 parent, child_role, errp);
2909 if (!bs) {
2910 goto done;
2913 done:
2914 qdict_del(options, bdref_key);
2915 return bs;
2919 * Opens a disk image whose options are given as BlockdevRef in another block
2920 * device's options.
2922 * If allow_none is true, no image will be opened if filename is false and no
2923 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2925 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2926 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2927 * itself, all options starting with "${bdref_key}." are considered part of the
2928 * BlockdevRef.
2930 * The BlockdevRef will be removed from the options QDict.
2932 BdrvChild *bdrv_open_child(const char *filename,
2933 QDict *options, const char *bdref_key,
2934 BlockDriverState *parent,
2935 const BdrvChildRole *child_role,
2936 bool allow_none, Error **errp)
2938 BlockDriverState *bs;
2940 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2941 allow_none, errp);
2942 if (bs == NULL) {
2943 return NULL;
2946 return bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2949 /* TODO Future callers may need to specify parent/child_role in order for
2950 * option inheritance to work. Existing callers use it for the root node. */
2951 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2953 BlockDriverState *bs = NULL;
2954 Error *local_err = NULL;
2955 QObject *obj = NULL;
2956 QDict *qdict = NULL;
2957 const char *reference = NULL;
2958 Visitor *v = NULL;
2960 if (ref->type == QTYPE_QSTRING) {
2961 reference = ref->u.reference;
2962 } else {
2963 BlockdevOptions *options = &ref->u.definition;
2964 assert(ref->type == QTYPE_QDICT);
2966 v = qobject_output_visitor_new(&obj);
2967 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2968 if (local_err) {
2969 error_propagate(errp, local_err);
2970 goto fail;
2972 visit_complete(v, &obj);
2974 qdict = qobject_to(QDict, obj);
2975 qdict_flatten(qdict);
2977 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2978 * compatibility with other callers) rather than what we want as the
2979 * real defaults. Apply the defaults here instead. */
2980 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2981 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2982 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2983 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2987 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2988 obj = NULL;
2990 fail:
2991 qobject_unref(obj);
2992 visit_free(v);
2993 return bs;
2996 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2997 int flags,
2998 QDict *snapshot_options,
2999 Error **errp)
3001 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3002 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3003 int64_t total_size;
3004 QemuOpts *opts = NULL;
3005 BlockDriverState *bs_snapshot = NULL;
3006 Error *local_err = NULL;
3007 int ret;
3009 /* if snapshot, we create a temporary backing file and open it
3010 instead of opening 'filename' directly */
3012 /* Get the required size from the image */
3013 total_size = bdrv_getlength(bs);
3014 if (total_size < 0) {
3015 error_setg_errno(errp, -total_size, "Could not get image size");
3016 goto out;
3019 /* Create the temporary image */
3020 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3021 if (ret < 0) {
3022 error_setg_errno(errp, -ret, "Could not get temporary filename");
3023 goto out;
3026 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3027 &error_abort);
3028 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3029 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3030 qemu_opts_del(opts);
3031 if (ret < 0) {
3032 error_prepend(errp, "Could not create temporary overlay '%s': ",
3033 tmp_filename);
3034 goto out;
3037 /* Prepare options QDict for the temporary file */
3038 qdict_put_str(snapshot_options, "file.driver", "file");
3039 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3040 qdict_put_str(snapshot_options, "driver", "qcow2");
3042 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3043 snapshot_options = NULL;
3044 if (!bs_snapshot) {
3045 goto out;
3048 /* bdrv_append() consumes a strong reference to bs_snapshot
3049 * (i.e. it will call bdrv_unref() on it) even on error, so in
3050 * order to be able to return one, we have to increase
3051 * bs_snapshot's refcount here */
3052 bdrv_ref(bs_snapshot);
3053 bdrv_append(bs_snapshot, bs, &local_err);
3054 if (local_err) {
3055 error_propagate(errp, local_err);
3056 bs_snapshot = NULL;
3057 goto out;
3060 out:
3061 qobject_unref(snapshot_options);
3062 g_free(tmp_filename);
3063 return bs_snapshot;
3067 * Opens a disk image (raw, qcow2, vmdk, ...)
3069 * options is a QDict of options to pass to the block drivers, or NULL for an
3070 * empty set of options. The reference to the QDict belongs to the block layer
3071 * after the call (even on failure), so if the caller intends to reuse the
3072 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3074 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3075 * If it is not NULL, the referenced BDS will be reused.
3077 * The reference parameter may be used to specify an existing block device which
3078 * should be opened. If specified, neither options nor a filename may be given,
3079 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3081 static BlockDriverState *bdrv_open_inherit(const char *filename,
3082 const char *reference,
3083 QDict *options, int flags,
3084 BlockDriverState *parent,
3085 const BdrvChildRole *child_role,
3086 Error **errp)
3088 int ret;
3089 BlockBackend *file = NULL;
3090 BlockDriverState *bs;
3091 BlockDriver *drv = NULL;
3092 BdrvChild *child;
3093 const char *drvname;
3094 const char *backing;
3095 Error *local_err = NULL;
3096 QDict *snapshot_options = NULL;
3097 int snapshot_flags = 0;
3099 assert(!child_role || !flags);
3100 assert(!child_role == !parent);
3102 if (reference) {
3103 bool options_non_empty = options ? qdict_size(options) : false;
3104 qobject_unref(options);
3106 if (filename || options_non_empty) {
3107 error_setg(errp, "Cannot reference an existing block device with "
3108 "additional options or a new filename");
3109 return NULL;
3112 bs = bdrv_lookup_bs(reference, reference, errp);
3113 if (!bs) {
3114 return NULL;
3117 bdrv_ref(bs);
3118 return bs;
3121 bs = bdrv_new();
3123 /* NULL means an empty set of options */
3124 if (options == NULL) {
3125 options = qdict_new();
3128 /* json: syntax counts as explicit options, as if in the QDict */
3129 parse_json_protocol(options, &filename, &local_err);
3130 if (local_err) {
3131 goto fail;
3134 bs->explicit_options = qdict_clone_shallow(options);
3136 if (child_role) {
3137 bs->inherits_from = parent;
3138 child_role->inherit_options(&flags, options,
3139 parent->open_flags, parent->options);
3142 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3143 if (local_err) {
3144 goto fail;
3148 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3149 * Caution: getting a boolean member of @options requires care.
3150 * When @options come from -blockdev or blockdev_add, members are
3151 * typed according to the QAPI schema, but when they come from
3152 * -drive, they're all QString.
3154 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3155 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3156 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3157 } else {
3158 flags &= ~BDRV_O_RDWR;
3161 if (flags & BDRV_O_SNAPSHOT) {
3162 snapshot_options = qdict_new();
3163 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3164 flags, options);
3165 /* Let bdrv_backing_options() override "read-only" */
3166 qdict_del(options, BDRV_OPT_READ_ONLY);
3167 bdrv_backing_options(&flags, options, flags, options);
3170 bs->open_flags = flags;
3171 bs->options = options;
3172 options = qdict_clone_shallow(options);
3174 /* Find the right image format driver */
3175 /* See cautionary note on accessing @options above */
3176 drvname = qdict_get_try_str(options, "driver");
3177 if (drvname) {
3178 drv = bdrv_find_format(drvname);
3179 if (!drv) {
3180 error_setg(errp, "Unknown driver: '%s'", drvname);
3181 goto fail;
3185 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3187 /* See cautionary note on accessing @options above */
3188 backing = qdict_get_try_str(options, "backing");
3189 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3190 (backing && *backing == '\0'))
3192 if (backing) {
3193 warn_report("Use of \"backing\": \"\" is deprecated; "
3194 "use \"backing\": null instead");
3196 flags |= BDRV_O_NO_BACKING;
3197 qdict_del(bs->explicit_options, "backing");
3198 qdict_del(bs->options, "backing");
3199 qdict_del(options, "backing");
3202 /* Open image file without format layer. This BlockBackend is only used for
3203 * probing, the block drivers will do their own bdrv_open_child() for the
3204 * same BDS, which is why we put the node name back into options. */
3205 if ((flags & BDRV_O_PROTOCOL) == 0) {
3206 BlockDriverState *file_bs;
3208 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3209 &child_file, true, &local_err);
3210 if (local_err) {
3211 goto fail;
3213 if (file_bs != NULL) {
3214 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3215 * looking at the header to guess the image format. This works even
3216 * in cases where a guest would not see a consistent state. */
3217 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3218 blk_insert_bs(file, file_bs, &local_err);
3219 bdrv_unref(file_bs);
3220 if (local_err) {
3221 goto fail;
3224 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3228 /* Image format probing */
3229 bs->probed = !drv;
3230 if (!drv && file) {
3231 ret = find_image_format(file, filename, &drv, &local_err);
3232 if (ret < 0) {
3233 goto fail;
3236 * This option update would logically belong in bdrv_fill_options(),
3237 * but we first need to open bs->file for the probing to work, while
3238 * opening bs->file already requires the (mostly) final set of options
3239 * so that cache mode etc. can be inherited.
3241 * Adding the driver later is somewhat ugly, but it's not an option
3242 * that would ever be inherited, so it's correct. We just need to make
3243 * sure to update both bs->options (which has the full effective
3244 * options for bs) and options (which has file.* already removed).
3246 qdict_put_str(bs->options, "driver", drv->format_name);
3247 qdict_put_str(options, "driver", drv->format_name);
3248 } else if (!drv) {
3249 error_setg(errp, "Must specify either driver or file");
3250 goto fail;
3253 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3254 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3255 /* file must be NULL if a protocol BDS is about to be created
3256 * (the inverse results in an error message from bdrv_open_common()) */
3257 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3259 /* Open the image */
3260 ret = bdrv_open_common(bs, file, options, &local_err);
3261 if (ret < 0) {
3262 goto fail;
3265 if (file) {
3266 blk_unref(file);
3267 file = NULL;
3270 /* If there is a backing file, use it */
3271 if ((flags & BDRV_O_NO_BACKING) == 0) {
3272 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3273 if (ret < 0) {
3274 goto close_and_fail;
3278 /* Remove all children options and references
3279 * from bs->options and bs->explicit_options */
3280 QLIST_FOREACH(child, &bs->children, next) {
3281 char *child_key_dot;
3282 child_key_dot = g_strdup_printf("%s.", child->name);
3283 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3284 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3285 qdict_del(bs->explicit_options, child->name);
3286 qdict_del(bs->options, child->name);
3287 g_free(child_key_dot);
3290 /* Check if any unknown options were used */
3291 if (qdict_size(options) != 0) {
3292 const QDictEntry *entry = qdict_first(options);
3293 if (flags & BDRV_O_PROTOCOL) {
3294 error_setg(errp, "Block protocol '%s' doesn't support the option "
3295 "'%s'", drv->format_name, entry->key);
3296 } else {
3297 error_setg(errp,
3298 "Block format '%s' does not support the option '%s'",
3299 drv->format_name, entry->key);
3302 goto close_and_fail;
3305 bdrv_parent_cb_change_media(bs, true);
3307 qobject_unref(options);
3308 options = NULL;
3310 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3311 * temporary snapshot afterwards. */
3312 if (snapshot_flags) {
3313 BlockDriverState *snapshot_bs;
3314 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3315 snapshot_options, &local_err);
3316 snapshot_options = NULL;
3317 if (local_err) {
3318 goto close_and_fail;
3320 /* We are not going to return bs but the overlay on top of it
3321 * (snapshot_bs); thus, we have to drop the strong reference to bs
3322 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3323 * though, because the overlay still has a reference to it. */
3324 bdrv_unref(bs);
3325 bs = snapshot_bs;
3328 return bs;
3330 fail:
3331 blk_unref(file);
3332 qobject_unref(snapshot_options);
3333 qobject_unref(bs->explicit_options);
3334 qobject_unref(bs->options);
3335 qobject_unref(options);
3336 bs->options = NULL;
3337 bs->explicit_options = NULL;
3338 bdrv_unref(bs);
3339 error_propagate(errp, local_err);
3340 return NULL;
3342 close_and_fail:
3343 bdrv_unref(bs);
3344 qobject_unref(snapshot_options);
3345 qobject_unref(options);
3346 error_propagate(errp, local_err);
3347 return NULL;
3350 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3351 QDict *options, int flags, Error **errp)
3353 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3354 NULL, errp);
3357 /* Return true if the NULL-terminated @list contains @str */
3358 static bool is_str_in_list(const char *str, const char *const *list)
3360 if (str && list) {
3361 int i;
3362 for (i = 0; list[i] != NULL; i++) {
3363 if (!strcmp(str, list[i])) {
3364 return true;
3368 return false;
3372 * Check that every option set in @bs->options is also set in
3373 * @new_opts.
3375 * Options listed in the common_options list and in
3376 * @bs->drv->mutable_opts are skipped.
3378 * Return 0 on success, otherwise return -EINVAL and set @errp.
3380 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3381 const QDict *new_opts, Error **errp)
3383 const QDictEntry *e;
3384 /* These options are common to all block drivers and are handled
3385 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3386 const char *const common_options[] = {
3387 "node-name", "discard", "cache.direct", "cache.no-flush",
3388 "read-only", "auto-read-only", "detect-zeroes", NULL
3391 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3392 if (!qdict_haskey(new_opts, e->key) &&
3393 !is_str_in_list(e->key, common_options) &&
3394 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3395 error_setg(errp, "Option '%s' cannot be reset "
3396 "to its default value", e->key);
3397 return -EINVAL;
3401 return 0;
3405 * Returns true if @child can be reached recursively from @bs
3407 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3408 BlockDriverState *child)
3410 BdrvChild *c;
3412 if (bs == child) {
3413 return true;
3416 QLIST_FOREACH(c, &bs->children, next) {
3417 if (bdrv_recurse_has_child(c->bs, child)) {
3418 return true;
3422 return false;
3426 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3427 * reopen of multiple devices.
3429 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3430 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3431 * be created and initialized. This newly created BlockReopenQueue should be
3432 * passed back in for subsequent calls that are intended to be of the same
3433 * atomic 'set'.
3435 * bs is the BlockDriverState to add to the reopen queue.
3437 * options contains the changed options for the associated bs
3438 * (the BlockReopenQueue takes ownership)
3440 * flags contains the open flags for the associated bs
3442 * returns a pointer to bs_queue, which is either the newly allocated
3443 * bs_queue, or the existing bs_queue being used.
3445 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3447 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3448 BlockDriverState *bs,
3449 QDict *options,
3450 const BdrvChildRole *role,
3451 QDict *parent_options,
3452 int parent_flags,
3453 bool keep_old_opts)
3455 assert(bs != NULL);
3457 BlockReopenQueueEntry *bs_entry;
3458 BdrvChild *child;
3459 QDict *old_options, *explicit_options, *options_copy;
3460 int flags;
3461 QemuOpts *opts;
3463 /* Make sure that the caller remembered to use a drained section. This is
3464 * important to avoid graph changes between the recursive queuing here and
3465 * bdrv_reopen_multiple(). */
3466 assert(bs->quiesce_counter > 0);
3468 if (bs_queue == NULL) {
3469 bs_queue = g_new0(BlockReopenQueue, 1);
3470 QTAILQ_INIT(bs_queue);
3473 if (!options) {
3474 options = qdict_new();
3477 /* Check if this BlockDriverState is already in the queue */
3478 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3479 if (bs == bs_entry->state.bs) {
3480 break;
3485 * Precedence of options:
3486 * 1. Explicitly passed in options (highest)
3487 * 2. Retained from explicitly set options of bs
3488 * 3. Inherited from parent node
3489 * 4. Retained from effective options of bs
3492 /* Old explicitly set values (don't overwrite by inherited value) */
3493 if (bs_entry || keep_old_opts) {
3494 old_options = qdict_clone_shallow(bs_entry ?
3495 bs_entry->state.explicit_options :
3496 bs->explicit_options);
3497 bdrv_join_options(bs, options, old_options);
3498 qobject_unref(old_options);
3501 explicit_options = qdict_clone_shallow(options);
3503 /* Inherit from parent node */
3504 if (parent_options) {
3505 flags = 0;
3506 role->inherit_options(&flags, options, parent_flags, parent_options);
3507 } else {
3508 flags = bdrv_get_flags(bs);
3511 if (keep_old_opts) {
3512 /* Old values are used for options that aren't set yet */
3513 old_options = qdict_clone_shallow(bs->options);
3514 bdrv_join_options(bs, options, old_options);
3515 qobject_unref(old_options);
3518 /* We have the final set of options so let's update the flags */
3519 options_copy = qdict_clone_shallow(options);
3520 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3521 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3522 update_flags_from_options(&flags, opts);
3523 qemu_opts_del(opts);
3524 qobject_unref(options_copy);
3526 /* bdrv_open_inherit() sets and clears some additional flags internally */
3527 flags &= ~BDRV_O_PROTOCOL;
3528 if (flags & BDRV_O_RDWR) {
3529 flags |= BDRV_O_ALLOW_RDWR;
3532 if (!bs_entry) {
3533 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3534 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3535 } else {
3536 qobject_unref(bs_entry->state.options);
3537 qobject_unref(bs_entry->state.explicit_options);
3540 bs_entry->state.bs = bs;
3541 bs_entry->state.options = options;
3542 bs_entry->state.explicit_options = explicit_options;
3543 bs_entry->state.flags = flags;
3545 /* This needs to be overwritten in bdrv_reopen_prepare() */
3546 bs_entry->state.perm = UINT64_MAX;
3547 bs_entry->state.shared_perm = 0;
3550 * If keep_old_opts is false then it means that unspecified
3551 * options must be reset to their original value. We don't allow
3552 * resetting 'backing' but we need to know if the option is
3553 * missing in order to decide if we have to return an error.
3555 if (!keep_old_opts) {
3556 bs_entry->state.backing_missing =
3557 !qdict_haskey(options, "backing") &&
3558 !qdict_haskey(options, "backing.driver");
3561 QLIST_FOREACH(child, &bs->children, next) {
3562 QDict *new_child_options = NULL;
3563 bool child_keep_old = keep_old_opts;
3565 /* reopen can only change the options of block devices that were
3566 * implicitly created and inherited options. For other (referenced)
3567 * block devices, a syntax like "backing.foo" results in an error. */
3568 if (child->bs->inherits_from != bs) {
3569 continue;
3572 /* Check if the options contain a child reference */
3573 if (qdict_haskey(options, child->name)) {
3574 const char *childref = qdict_get_try_str(options, child->name);
3576 * The current child must not be reopened if the child
3577 * reference is null or points to a different node.
3579 if (g_strcmp0(childref, child->bs->node_name)) {
3580 continue;
3583 * If the child reference points to the current child then
3584 * reopen it with its existing set of options (note that
3585 * it can still inherit new options from the parent).
3587 child_keep_old = true;
3588 } else {
3589 /* Extract child options ("child-name.*") */
3590 char *child_key_dot = g_strdup_printf("%s.", child->name);
3591 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3592 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3593 g_free(child_key_dot);
3596 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3597 child->role, options, flags, child_keep_old);
3600 return bs_queue;
3603 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3604 BlockDriverState *bs,
3605 QDict *options, bool keep_old_opts)
3607 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0,
3608 keep_old_opts);
3612 * Reopen multiple BlockDriverStates atomically & transactionally.
3614 * The queue passed in (bs_queue) must have been built up previous
3615 * via bdrv_reopen_queue().
3617 * Reopens all BDS specified in the queue, with the appropriate
3618 * flags. All devices are prepared for reopen, and failure of any
3619 * device will cause all device changes to be abandoned, and intermediate
3620 * data cleaned up.
3622 * If all devices prepare successfully, then the changes are committed
3623 * to all devices.
3625 * All affected nodes must be drained between bdrv_reopen_queue() and
3626 * bdrv_reopen_multiple().
3628 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
3630 int ret = -1;
3631 BlockReopenQueueEntry *bs_entry, *next;
3633 assert(bs_queue != NULL);
3635 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3636 assert(bs_entry->state.bs->quiesce_counter > 0);
3637 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
3638 goto cleanup;
3640 bs_entry->prepared = true;
3643 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3644 BDRVReopenState *state = &bs_entry->state;
3645 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
3646 state->shared_perm, NULL, NULL, errp);
3647 if (ret < 0) {
3648 goto cleanup_perm;
3650 /* Check if new_backing_bs would accept the new permissions */
3651 if (state->replace_backing_bs && state->new_backing_bs) {
3652 uint64_t nperm, nshared;
3653 bdrv_child_perm(state->bs, state->new_backing_bs,
3654 NULL, &child_backing, bs_queue,
3655 state->perm, state->shared_perm,
3656 &nperm, &nshared);
3657 ret = bdrv_check_update_perm(state->new_backing_bs, NULL,
3658 nperm, nshared, NULL, NULL, errp);
3659 if (ret < 0) {
3660 goto cleanup_perm;
3663 bs_entry->perms_checked = true;
3667 * If we reach this point, we have success and just need to apply the
3668 * changes.
3670 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3671 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3672 * children are usually goes after parents in reopen-queue, so go from last
3673 * to first element.
3675 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3676 bdrv_reopen_commit(&bs_entry->state);
3679 ret = 0;
3680 cleanup_perm:
3681 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3682 BDRVReopenState *state = &bs_entry->state;
3684 if (!bs_entry->perms_checked) {
3685 continue;
3688 if (ret == 0) {
3689 bdrv_set_perm(state->bs, state->perm, state->shared_perm);
3690 } else {
3691 bdrv_abort_perm_update(state->bs);
3692 if (state->replace_backing_bs && state->new_backing_bs) {
3693 bdrv_abort_perm_update(state->new_backing_bs);
3697 cleanup:
3698 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3699 if (ret) {
3700 if (bs_entry->prepared) {
3701 bdrv_reopen_abort(&bs_entry->state);
3703 qobject_unref(bs_entry->state.explicit_options);
3704 qobject_unref(bs_entry->state.options);
3706 if (bs_entry->state.new_backing_bs) {
3707 bdrv_unref(bs_entry->state.new_backing_bs);
3709 g_free(bs_entry);
3711 g_free(bs_queue);
3713 return ret;
3716 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3717 Error **errp)
3719 int ret;
3720 BlockReopenQueue *queue;
3721 QDict *opts = qdict_new();
3723 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3725 bdrv_subtree_drained_begin(bs);
3726 queue = bdrv_reopen_queue(NULL, bs, opts, true);
3727 ret = bdrv_reopen_multiple(queue, errp);
3728 bdrv_subtree_drained_end(bs);
3730 return ret;
3733 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3734 BdrvChild *c)
3736 BlockReopenQueueEntry *entry;
3738 QTAILQ_FOREACH(entry, q, entry) {
3739 BlockDriverState *bs = entry->state.bs;
3740 BdrvChild *child;
3742 QLIST_FOREACH(child, &bs->children, next) {
3743 if (child == c) {
3744 return entry;
3749 return NULL;
3752 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3753 uint64_t *perm, uint64_t *shared)
3755 BdrvChild *c;
3756 BlockReopenQueueEntry *parent;
3757 uint64_t cumulative_perms = 0;
3758 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3760 QLIST_FOREACH(c, &bs->parents, next_parent) {
3761 parent = find_parent_in_reopen_queue(q, c);
3762 if (!parent) {
3763 cumulative_perms |= c->perm;
3764 cumulative_shared_perms &= c->shared_perm;
3765 } else {
3766 uint64_t nperm, nshared;
3768 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3769 parent->state.perm, parent->state.shared_perm,
3770 &nperm, &nshared);
3772 cumulative_perms |= nperm;
3773 cumulative_shared_perms &= nshared;
3776 *perm = cumulative_perms;
3777 *shared = cumulative_shared_perms;
3781 * Take a BDRVReopenState and check if the value of 'backing' in the
3782 * reopen_state->options QDict is valid or not.
3784 * If 'backing' is missing from the QDict then return 0.
3786 * If 'backing' contains the node name of the backing file of
3787 * reopen_state->bs then return 0.
3789 * If 'backing' contains a different node name (or is null) then check
3790 * whether the current backing file can be replaced with the new one.
3791 * If that's the case then reopen_state->replace_backing_bs is set to
3792 * true and reopen_state->new_backing_bs contains a pointer to the new
3793 * backing BlockDriverState (or NULL).
3795 * Return 0 on success, otherwise return < 0 and set @errp.
3797 static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
3798 Error **errp)
3800 BlockDriverState *bs = reopen_state->bs;
3801 BlockDriverState *overlay_bs, *new_backing_bs;
3802 QObject *value;
3803 const char *str;
3805 value = qdict_get(reopen_state->options, "backing");
3806 if (value == NULL) {
3807 return 0;
3810 switch (qobject_type(value)) {
3811 case QTYPE_QNULL:
3812 new_backing_bs = NULL;
3813 break;
3814 case QTYPE_QSTRING:
3815 str = qobject_get_try_str(value);
3816 new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
3817 if (new_backing_bs == NULL) {
3818 return -EINVAL;
3819 } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
3820 error_setg(errp, "Making '%s' a backing file of '%s' "
3821 "would create a cycle", str, bs->node_name);
3822 return -EINVAL;
3824 break;
3825 default:
3826 /* 'backing' does not allow any other data type */
3827 g_assert_not_reached();
3831 * TODO: before removing the x- prefix from x-blockdev-reopen we
3832 * should move the new backing file into the right AioContext
3833 * instead of returning an error.
3835 if (new_backing_bs) {
3836 if (bdrv_get_aio_context(new_backing_bs) != bdrv_get_aio_context(bs)) {
3837 error_setg(errp, "Cannot use a new backing file "
3838 "with a different AioContext");
3839 return -EINVAL;
3844 * Find the "actual" backing file by skipping all links that point
3845 * to an implicit node, if any (e.g. a commit filter node).
3847 overlay_bs = bs;
3848 while (backing_bs(overlay_bs) && backing_bs(overlay_bs)->implicit) {
3849 overlay_bs = backing_bs(overlay_bs);
3852 /* If we want to replace the backing file we need some extra checks */
3853 if (new_backing_bs != backing_bs(overlay_bs)) {
3854 /* Check for implicit nodes between bs and its backing file */
3855 if (bs != overlay_bs) {
3856 error_setg(errp, "Cannot change backing link if '%s' has "
3857 "an implicit backing file", bs->node_name);
3858 return -EPERM;
3860 /* Check if the backing link that we want to replace is frozen */
3861 if (bdrv_is_backing_chain_frozen(overlay_bs, backing_bs(overlay_bs),
3862 errp)) {
3863 return -EPERM;
3865 reopen_state->replace_backing_bs = true;
3866 if (new_backing_bs) {
3867 bdrv_ref(new_backing_bs);
3868 reopen_state->new_backing_bs = new_backing_bs;
3872 return 0;
3876 * Prepares a BlockDriverState for reopen. All changes are staged in the
3877 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3878 * the block driver layer .bdrv_reopen_prepare()
3880 * bs is the BlockDriverState to reopen
3881 * flags are the new open flags
3882 * queue is the reopen queue
3884 * Returns 0 on success, non-zero on error. On error errp will be set
3885 * as well.
3887 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3888 * It is the responsibility of the caller to then call the abort() or
3889 * commit() for any other BDS that have been left in a prepare() state
3892 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3893 Error **errp)
3895 int ret = -1;
3896 int old_flags;
3897 Error *local_err = NULL;
3898 BlockDriver *drv;
3899 QemuOpts *opts;
3900 QDict *orig_reopen_opts;
3901 char *discard = NULL;
3902 bool read_only;
3903 bool drv_prepared = false;
3905 assert(reopen_state != NULL);
3906 assert(reopen_state->bs->drv != NULL);
3907 drv = reopen_state->bs->drv;
3909 /* This function and each driver's bdrv_reopen_prepare() remove
3910 * entries from reopen_state->options as they are processed, so
3911 * we need to make a copy of the original QDict. */
3912 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3914 /* Process generic block layer options */
3915 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3916 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3917 if (local_err) {
3918 error_propagate(errp, local_err);
3919 ret = -EINVAL;
3920 goto error;
3923 /* This was already called in bdrv_reopen_queue_child() so the flags
3924 * are up-to-date. This time we simply want to remove the options from
3925 * QemuOpts in order to indicate that they have been processed. */
3926 old_flags = reopen_state->flags;
3927 update_flags_from_options(&reopen_state->flags, opts);
3928 assert(old_flags == reopen_state->flags);
3930 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
3931 if (discard != NULL) {
3932 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3933 error_setg(errp, "Invalid discard option");
3934 ret = -EINVAL;
3935 goto error;
3939 reopen_state->detect_zeroes =
3940 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3941 if (local_err) {
3942 error_propagate(errp, local_err);
3943 ret = -EINVAL;
3944 goto error;
3947 /* All other options (including node-name and driver) must be unchanged.
3948 * Put them back into the QDict, so that they are checked at the end
3949 * of this function. */
3950 qemu_opts_to_qdict(opts, reopen_state->options);
3952 /* If we are to stay read-only, do not allow permission change
3953 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3954 * not set, or if the BDS still has copy_on_read enabled */
3955 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3956 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3957 if (local_err) {
3958 error_propagate(errp, local_err);
3959 goto error;
3962 /* Calculate required permissions after reopening */
3963 bdrv_reopen_perm(queue, reopen_state->bs,
3964 &reopen_state->perm, &reopen_state->shared_perm);
3966 ret = bdrv_flush(reopen_state->bs);
3967 if (ret) {
3968 error_setg_errno(errp, -ret, "Error flushing drive");
3969 goto error;
3972 if (drv->bdrv_reopen_prepare) {
3974 * If a driver-specific option is missing, it means that we
3975 * should reset it to its default value.
3976 * But not all options allow that, so we need to check it first.
3978 ret = bdrv_reset_options_allowed(reopen_state->bs,
3979 reopen_state->options, errp);
3980 if (ret) {
3981 goto error;
3984 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3985 if (ret) {
3986 if (local_err != NULL) {
3987 error_propagate(errp, local_err);
3988 } else {
3989 bdrv_refresh_filename(reopen_state->bs);
3990 error_setg(errp, "failed while preparing to reopen image '%s'",
3991 reopen_state->bs->filename);
3993 goto error;
3995 } else {
3996 /* It is currently mandatory to have a bdrv_reopen_prepare()
3997 * handler for each supported drv. */
3998 error_setg(errp, "Block format '%s' used by node '%s' "
3999 "does not support reopening files", drv->format_name,
4000 bdrv_get_device_or_node_name(reopen_state->bs));
4001 ret = -1;
4002 goto error;
4005 drv_prepared = true;
4008 * We must provide the 'backing' option if the BDS has a backing
4009 * file or if the image file has a backing file name as part of
4010 * its metadata. Otherwise the 'backing' option can be omitted.
4012 if (drv->supports_backing && reopen_state->backing_missing &&
4013 (backing_bs(reopen_state->bs) || reopen_state->bs->backing_file[0])) {
4014 error_setg(errp, "backing is missing for '%s'",
4015 reopen_state->bs->node_name);
4016 ret = -EINVAL;
4017 goto error;
4021 * Allow changing the 'backing' option. The new value can be
4022 * either a reference to an existing node (using its node name)
4023 * or NULL to simply detach the current backing file.
4025 ret = bdrv_reopen_parse_backing(reopen_state, errp);
4026 if (ret < 0) {
4027 goto error;
4029 qdict_del(reopen_state->options, "backing");
4031 /* Options that are not handled are only okay if they are unchanged
4032 * compared to the old state. It is expected that some options are only
4033 * used for the initial open, but not reopen (e.g. filename) */
4034 if (qdict_size(reopen_state->options)) {
4035 const QDictEntry *entry = qdict_first(reopen_state->options);
4037 do {
4038 QObject *new = entry->value;
4039 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4041 /* Allow child references (child_name=node_name) as long as they
4042 * point to the current child (i.e. everything stays the same). */
4043 if (qobject_type(new) == QTYPE_QSTRING) {
4044 BdrvChild *child;
4045 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4046 if (!strcmp(child->name, entry->key)) {
4047 break;
4051 if (child) {
4052 const char *str = qobject_get_try_str(new);
4053 if (!strcmp(child->bs->node_name, str)) {
4054 continue; /* Found child with this name, skip option */
4060 * TODO: When using -drive to specify blockdev options, all values
4061 * will be strings; however, when using -blockdev, blockdev-add or
4062 * filenames using the json:{} pseudo-protocol, they will be
4063 * correctly typed.
4064 * In contrast, reopening options are (currently) always strings
4065 * (because you can only specify them through qemu-io; all other
4066 * callers do not specify any options).
4067 * Therefore, when using anything other than -drive to create a BDS,
4068 * this cannot detect non-string options as unchanged, because
4069 * qobject_is_equal() always returns false for objects of different
4070 * type. In the future, this should be remedied by correctly typing
4071 * all options. For now, this is not too big of an issue because
4072 * the user can simply omit options which cannot be changed anyway,
4073 * so they will stay unchanged.
4075 if (!qobject_is_equal(new, old)) {
4076 error_setg(errp, "Cannot change the option '%s'", entry->key);
4077 ret = -EINVAL;
4078 goto error;
4080 } while ((entry = qdict_next(reopen_state->options, entry)));
4083 ret = 0;
4085 /* Restore the original reopen_state->options QDict */
4086 qobject_unref(reopen_state->options);
4087 reopen_state->options = qobject_ref(orig_reopen_opts);
4089 error:
4090 if (ret < 0 && drv_prepared) {
4091 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4092 * call drv->bdrv_reopen_abort() before signaling an error
4093 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4094 * when the respective bdrv_reopen_prepare() has failed) */
4095 if (drv->bdrv_reopen_abort) {
4096 drv->bdrv_reopen_abort(reopen_state);
4099 qemu_opts_del(opts);
4100 qobject_unref(orig_reopen_opts);
4101 g_free(discard);
4102 return ret;
4106 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4107 * makes them final by swapping the staging BlockDriverState contents into
4108 * the active BlockDriverState contents.
4110 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4112 BlockDriver *drv;
4113 BlockDriverState *bs;
4114 BdrvChild *child;
4116 assert(reopen_state != NULL);
4117 bs = reopen_state->bs;
4118 drv = bs->drv;
4119 assert(drv != NULL);
4121 /* If there are any driver level actions to take */
4122 if (drv->bdrv_reopen_commit) {
4123 drv->bdrv_reopen_commit(reopen_state);
4126 /* set BDS specific flags now */
4127 qobject_unref(bs->explicit_options);
4128 qobject_unref(bs->options);
4130 bs->explicit_options = reopen_state->explicit_options;
4131 bs->options = reopen_state->options;
4132 bs->open_flags = reopen_state->flags;
4133 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
4134 bs->detect_zeroes = reopen_state->detect_zeroes;
4136 if (reopen_state->replace_backing_bs) {
4137 qdict_del(bs->explicit_options, "backing");
4138 qdict_del(bs->options, "backing");
4141 /* Remove child references from bs->options and bs->explicit_options.
4142 * Child options were already removed in bdrv_reopen_queue_child() */
4143 QLIST_FOREACH(child, &bs->children, next) {
4144 qdict_del(bs->explicit_options, child->name);
4145 qdict_del(bs->options, child->name);
4149 * Change the backing file if a new one was specified. We do this
4150 * after updating bs->options, so bdrv_refresh_filename() (called
4151 * from bdrv_set_backing_hd()) has the new values.
4153 if (reopen_state->replace_backing_bs) {
4154 BlockDriverState *old_backing_bs = backing_bs(bs);
4155 assert(!old_backing_bs || !old_backing_bs->implicit);
4156 /* Abort the permission update on the backing bs we're detaching */
4157 if (old_backing_bs) {
4158 bdrv_abort_perm_update(old_backing_bs);
4160 bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort);
4163 bdrv_refresh_limits(bs, NULL);
4167 * Abort the reopen, and delete and free the staged changes in
4168 * reopen_state
4170 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4172 BlockDriver *drv;
4174 assert(reopen_state != NULL);
4175 drv = reopen_state->bs->drv;
4176 assert(drv != NULL);
4178 if (drv->bdrv_reopen_abort) {
4179 drv->bdrv_reopen_abort(reopen_state);
4184 static void bdrv_close(BlockDriverState *bs)
4186 BdrvAioNotifier *ban, *ban_next;
4187 BdrvChild *child, *next;
4189 assert(!bs->refcnt);
4191 bdrv_drained_begin(bs); /* complete I/O */
4192 bdrv_flush(bs);
4193 bdrv_drain(bs); /* in case flush left pending I/O */
4195 if (bs->drv) {
4196 if (bs->drv->bdrv_close) {
4197 bs->drv->bdrv_close(bs);
4199 bs->drv = NULL;
4202 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4203 bdrv_unref_child(bs, child);
4206 bs->backing = NULL;
4207 bs->file = NULL;
4208 g_free(bs->opaque);
4209 bs->opaque = NULL;
4210 atomic_set(&bs->copy_on_read, 0);
4211 bs->backing_file[0] = '\0';
4212 bs->backing_format[0] = '\0';
4213 bs->total_sectors = 0;
4214 bs->encrypted = false;
4215 bs->sg = false;
4216 qobject_unref(bs->options);
4217 qobject_unref(bs->explicit_options);
4218 bs->options = NULL;
4219 bs->explicit_options = NULL;
4220 qobject_unref(bs->full_open_options);
4221 bs->full_open_options = NULL;
4223 bdrv_release_named_dirty_bitmaps(bs);
4224 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4226 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4227 g_free(ban);
4229 QLIST_INIT(&bs->aio_notifiers);
4230 bdrv_drained_end(bs);
4233 void bdrv_close_all(void)
4235 assert(job_next(NULL) == NULL);
4236 nbd_export_close_all();
4238 /* Drop references from requests still in flight, such as canceled block
4239 * jobs whose AIO context has not been polled yet */
4240 bdrv_drain_all();
4242 blk_remove_all_bs();
4243 blockdev_close_all_bdrv_states();
4245 assert(QTAILQ_EMPTY(&all_bdrv_states));
4248 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4250 GQueue *queue;
4251 GHashTable *found;
4252 bool ret;
4254 if (c->role->stay_at_node) {
4255 return false;
4258 /* If the child @c belongs to the BDS @to, replacing the current
4259 * c->bs by @to would mean to create a loop.
4261 * Such a case occurs when appending a BDS to a backing chain.
4262 * For instance, imagine the following chain:
4264 * guest device -> node A -> further backing chain...
4266 * Now we create a new BDS B which we want to put on top of this
4267 * chain, so we first attach A as its backing node:
4269 * node B
4272 * guest device -> node A -> further backing chain...
4274 * Finally we want to replace A by B. When doing that, we want to
4275 * replace all pointers to A by pointers to B -- except for the
4276 * pointer from B because (1) that would create a loop, and (2)
4277 * that pointer should simply stay intact:
4279 * guest device -> node B
4282 * node A -> further backing chain...
4284 * In general, when replacing a node A (c->bs) by a node B (@to),
4285 * if A is a child of B, that means we cannot replace A by B there
4286 * because that would create a loop. Silently detaching A from B
4287 * is also not really an option. So overall just leaving A in
4288 * place there is the most sensible choice.
4290 * We would also create a loop in any cases where @c is only
4291 * indirectly referenced by @to. Prevent this by returning false
4292 * if @c is found (by breadth-first search) anywhere in the whole
4293 * subtree of @to.
4296 ret = true;
4297 found = g_hash_table_new(NULL, NULL);
4298 g_hash_table_add(found, to);
4299 queue = g_queue_new();
4300 g_queue_push_tail(queue, to);
4302 while (!g_queue_is_empty(queue)) {
4303 BlockDriverState *v = g_queue_pop_head(queue);
4304 BdrvChild *c2;
4306 QLIST_FOREACH(c2, &v->children, next) {
4307 if (c2 == c) {
4308 ret = false;
4309 break;
4312 if (g_hash_table_contains(found, c2->bs)) {
4313 continue;
4316 g_queue_push_tail(queue, c2->bs);
4317 g_hash_table_add(found, c2->bs);
4321 g_queue_free(queue);
4322 g_hash_table_destroy(found);
4324 return ret;
4327 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
4328 Error **errp)
4330 BdrvChild *c, *next;
4331 GSList *list = NULL, *p;
4332 uint64_t perm = 0, shared = BLK_PERM_ALL;
4333 int ret;
4335 /* Make sure that @from doesn't go away until we have successfully attached
4336 * all of its parents to @to. */
4337 bdrv_ref(from);
4339 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4340 bdrv_drained_begin(from);
4342 /* Put all parents into @list and calculate their cumulative permissions */
4343 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4344 assert(c->bs == from);
4345 if (!should_update_child(c, to)) {
4346 continue;
4348 if (c->frozen) {
4349 error_setg(errp, "Cannot change '%s' link to '%s'",
4350 c->name, from->node_name);
4351 goto out;
4353 list = g_slist_prepend(list, c);
4354 perm |= c->perm;
4355 shared &= c->shared_perm;
4358 /* Check whether the required permissions can be granted on @to, ignoring
4359 * all BdrvChild in @list so that they can't block themselves. */
4360 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, NULL, errp);
4361 if (ret < 0) {
4362 bdrv_abort_perm_update(to);
4363 goto out;
4366 /* Now actually perform the change. We performed the permission check for
4367 * all elements of @list at once, so set the permissions all at once at the
4368 * very end. */
4369 for (p = list; p != NULL; p = p->next) {
4370 c = p->data;
4372 bdrv_ref(to);
4373 bdrv_replace_child_noperm(c, to);
4374 bdrv_unref(from);
4377 bdrv_get_cumulative_perm(to, &perm, &shared);
4378 bdrv_set_perm(to, perm, shared);
4380 out:
4381 g_slist_free(list);
4382 bdrv_drained_end(from);
4383 bdrv_unref(from);
4387 * Add new bs contents at the top of an image chain while the chain is
4388 * live, while keeping required fields on the top layer.
4390 * This will modify the BlockDriverState fields, and swap contents
4391 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4393 * bs_new must not be attached to a BlockBackend.
4395 * This function does not create any image files.
4397 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4398 * that's what the callers commonly need. bs_new will be referenced by the old
4399 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4400 * reference of its own, it must call bdrv_ref().
4402 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
4403 Error **errp)
4405 Error *local_err = NULL;
4407 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
4408 if (local_err) {
4409 error_propagate(errp, local_err);
4410 goto out;
4413 bdrv_replace_node(bs_top, bs_new, &local_err);
4414 if (local_err) {
4415 error_propagate(errp, local_err);
4416 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
4417 goto out;
4420 /* bs_new is now referenced by its new parents, we don't need the
4421 * additional reference any more. */
4422 out:
4423 bdrv_unref(bs_new);
4426 static void bdrv_delete(BlockDriverState *bs)
4428 assert(bdrv_op_blocker_is_empty(bs));
4429 assert(!bs->refcnt);
4431 /* remove from list, if necessary */
4432 if (bs->node_name[0] != '\0') {
4433 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
4435 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
4437 bdrv_close(bs);
4439 g_free(bs);
4443 * Run consistency checks on an image
4445 * Returns 0 if the check could be completed (it doesn't mean that the image is
4446 * free of errors) or -errno when an internal error occurred. The results of the
4447 * check are stored in res.
4449 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
4450 BdrvCheckResult *res, BdrvCheckMode fix)
4452 if (bs->drv == NULL) {
4453 return -ENOMEDIUM;
4455 if (bs->drv->bdrv_co_check == NULL) {
4456 return -ENOTSUP;
4459 memset(res, 0, sizeof(*res));
4460 return bs->drv->bdrv_co_check(bs, res, fix);
4463 typedef struct CheckCo {
4464 BlockDriverState *bs;
4465 BdrvCheckResult *res;
4466 BdrvCheckMode fix;
4467 int ret;
4468 } CheckCo;
4470 static void coroutine_fn bdrv_check_co_entry(void *opaque)
4472 CheckCo *cco = opaque;
4473 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
4474 aio_wait_kick();
4477 int bdrv_check(BlockDriverState *bs,
4478 BdrvCheckResult *res, BdrvCheckMode fix)
4480 Coroutine *co;
4481 CheckCo cco = {
4482 .bs = bs,
4483 .res = res,
4484 .ret = -EINPROGRESS,
4485 .fix = fix,
4488 if (qemu_in_coroutine()) {
4489 /* Fast-path if already in coroutine context */
4490 bdrv_check_co_entry(&cco);
4491 } else {
4492 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
4493 bdrv_coroutine_enter(bs, co);
4494 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
4497 return cco.ret;
4501 * Return values:
4502 * 0 - success
4503 * -EINVAL - backing format specified, but no file
4504 * -ENOSPC - can't update the backing file because no space is left in the
4505 * image file header
4506 * -ENOTSUP - format driver doesn't support changing the backing file
4508 int bdrv_change_backing_file(BlockDriverState *bs,
4509 const char *backing_file, const char *backing_fmt)
4511 BlockDriver *drv = bs->drv;
4512 int ret;
4514 if (!drv) {
4515 return -ENOMEDIUM;
4518 /* Backing file format doesn't make sense without a backing file */
4519 if (backing_fmt && !backing_file) {
4520 return -EINVAL;
4523 if (drv->bdrv_change_backing_file != NULL) {
4524 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
4525 } else {
4526 ret = -ENOTSUP;
4529 if (ret == 0) {
4530 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
4531 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
4532 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
4533 backing_file ?: "");
4535 return ret;
4539 * Finds the image layer in the chain that has 'bs' as its backing file.
4541 * active is the current topmost image.
4543 * Returns NULL if bs is not found in active's image chain,
4544 * or if active == bs.
4546 * Returns the bottommost base image if bs == NULL.
4548 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
4549 BlockDriverState *bs)
4551 while (active && bs != backing_bs(active)) {
4552 active = backing_bs(active);
4555 return active;
4558 /* Given a BDS, searches for the base layer. */
4559 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4561 return bdrv_find_overlay(bs, NULL);
4565 * Return true if at least one of the backing links between @bs and
4566 * @base is frozen. @errp is set if that's the case.
4567 * @base must be reachable from @bs, or NULL.
4569 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
4570 Error **errp)
4572 BlockDriverState *i;
4574 for (i = bs; i != base; i = backing_bs(i)) {
4575 if (i->backing && i->backing->frozen) {
4576 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4577 i->backing->name, i->node_name,
4578 backing_bs(i)->node_name);
4579 return true;
4583 return false;
4587 * Freeze all backing links between @bs and @base.
4588 * If any of the links is already frozen the operation is aborted and
4589 * none of the links are modified.
4590 * @base must be reachable from @bs, or NULL.
4591 * Returns 0 on success. On failure returns < 0 and sets @errp.
4593 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4594 Error **errp)
4596 BlockDriverState *i;
4598 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4599 return -EPERM;
4602 for (i = bs; i != base; i = backing_bs(i)) {
4603 if (i->backing && backing_bs(i)->never_freeze) {
4604 error_setg(errp, "Cannot freeze '%s' link to '%s'",
4605 i->backing->name, backing_bs(i)->node_name);
4606 return -EPERM;
4610 for (i = bs; i != base; i = backing_bs(i)) {
4611 if (i->backing) {
4612 i->backing->frozen = true;
4616 return 0;
4620 * Unfreeze all backing links between @bs and @base. The caller must
4621 * ensure that all links are frozen before using this function.
4622 * @base must be reachable from @bs, or NULL.
4624 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4626 BlockDriverState *i;
4628 for (i = bs; i != base; i = backing_bs(i)) {
4629 if (i->backing) {
4630 assert(i->backing->frozen);
4631 i->backing->frozen = false;
4637 * Drops images above 'base' up to and including 'top', and sets the image
4638 * above 'top' to have base as its backing file.
4640 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4641 * information in 'bs' can be properly updated.
4643 * E.g., this will convert the following chain:
4644 * bottom <- base <- intermediate <- top <- active
4646 * to
4648 * bottom <- base <- active
4650 * It is allowed for bottom==base, in which case it converts:
4652 * base <- intermediate <- top <- active
4654 * to
4656 * base <- active
4658 * If backing_file_str is non-NULL, it will be used when modifying top's
4659 * overlay image metadata.
4661 * Error conditions:
4662 * if active == top, that is considered an error
4665 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4666 const char *backing_file_str)
4668 BlockDriverState *explicit_top = top;
4669 bool update_inherits_from;
4670 BdrvChild *c, *next;
4671 Error *local_err = NULL;
4672 int ret = -EIO;
4674 bdrv_ref(top);
4675 bdrv_subtree_drained_begin(top);
4677 if (!top->drv || !base->drv) {
4678 goto exit;
4681 /* Make sure that base is in the backing chain of top */
4682 if (!bdrv_chain_contains(top, base)) {
4683 goto exit;
4686 /* This function changes all links that point to top and makes
4687 * them point to base. Check that none of them is frozen. */
4688 QLIST_FOREACH(c, &top->parents, next_parent) {
4689 if (c->frozen) {
4690 goto exit;
4694 /* If 'base' recursively inherits from 'top' then we should set
4695 * base->inherits_from to top->inherits_from after 'top' and all
4696 * other intermediate nodes have been dropped.
4697 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4698 * it because no one inherits from it. We use explicit_top for that. */
4699 while (explicit_top && explicit_top->implicit) {
4700 explicit_top = backing_bs(explicit_top);
4702 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4704 /* success - we can delete the intermediate states, and link top->base */
4705 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4706 * we've figured out how they should work. */
4707 if (!backing_file_str) {
4708 bdrv_refresh_filename(base);
4709 backing_file_str = base->filename;
4712 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
4713 /* Check whether we are allowed to switch c from top to base */
4714 GSList *ignore_children = g_slist_prepend(NULL, c);
4715 ret = bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
4716 ignore_children, NULL, &local_err);
4717 g_slist_free(ignore_children);
4718 if (ret < 0) {
4719 error_report_err(local_err);
4720 goto exit;
4723 /* If so, update the backing file path in the image file */
4724 if (c->role->update_filename) {
4725 ret = c->role->update_filename(c, base, backing_file_str,
4726 &local_err);
4727 if (ret < 0) {
4728 bdrv_abort_perm_update(base);
4729 error_report_err(local_err);
4730 goto exit;
4734 /* Do the actual switch in the in-memory graph.
4735 * Completes bdrv_check_update_perm() transaction internally. */
4736 bdrv_ref(base);
4737 bdrv_replace_child(c, base);
4738 bdrv_unref(top);
4741 if (update_inherits_from) {
4742 base->inherits_from = explicit_top->inherits_from;
4745 ret = 0;
4746 exit:
4747 bdrv_subtree_drained_end(top);
4748 bdrv_unref(top);
4749 return ret;
4753 * Length of a allocated file in bytes. Sparse files are counted by actual
4754 * allocated space. Return < 0 if error or unknown.
4756 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
4758 BlockDriver *drv = bs->drv;
4759 if (!drv) {
4760 return -ENOMEDIUM;
4762 if (drv->bdrv_get_allocated_file_size) {
4763 return drv->bdrv_get_allocated_file_size(bs);
4765 if (bs->file) {
4766 return bdrv_get_allocated_file_size(bs->file->bs);
4768 return -ENOTSUP;
4772 * bdrv_measure:
4773 * @drv: Format driver
4774 * @opts: Creation options for new image
4775 * @in_bs: Existing image containing data for new image (may be NULL)
4776 * @errp: Error object
4777 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4778 * or NULL on error
4780 * Calculate file size required to create a new image.
4782 * If @in_bs is given then space for allocated clusters and zero clusters
4783 * from that image are included in the calculation. If @opts contains a
4784 * backing file that is shared by @in_bs then backing clusters may be omitted
4785 * from the calculation.
4787 * If @in_bs is NULL then the calculation includes no allocated clusters
4788 * unless a preallocation option is given in @opts.
4790 * Note that @in_bs may use a different BlockDriver from @drv.
4792 * If an error occurs the @errp pointer is set.
4794 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
4795 BlockDriverState *in_bs, Error **errp)
4797 if (!drv->bdrv_measure) {
4798 error_setg(errp, "Block driver '%s' does not support size measurement",
4799 drv->format_name);
4800 return NULL;
4803 return drv->bdrv_measure(opts, in_bs, errp);
4807 * Return number of sectors on success, -errno on error.
4809 int64_t bdrv_nb_sectors(BlockDriverState *bs)
4811 BlockDriver *drv = bs->drv;
4813 if (!drv)
4814 return -ENOMEDIUM;
4816 if (drv->has_variable_length) {
4817 int ret = refresh_total_sectors(bs, bs->total_sectors);
4818 if (ret < 0) {
4819 return ret;
4822 return bs->total_sectors;
4826 * Return length in bytes on success, -errno on error.
4827 * The length is always a multiple of BDRV_SECTOR_SIZE.
4829 int64_t bdrv_getlength(BlockDriverState *bs)
4831 int64_t ret = bdrv_nb_sectors(bs);
4833 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
4834 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
4837 /* return 0 as number of sectors if no device present or error */
4838 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
4840 int64_t nb_sectors = bdrv_nb_sectors(bs);
4842 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
4845 bool bdrv_is_sg(BlockDriverState *bs)
4847 return bs->sg;
4850 bool bdrv_is_encrypted(BlockDriverState *bs)
4852 if (bs->backing && bs->backing->bs->encrypted) {
4853 return true;
4855 return bs->encrypted;
4858 const char *bdrv_get_format_name(BlockDriverState *bs)
4860 return bs->drv ? bs->drv->format_name : NULL;
4863 static int qsort_strcmp(const void *a, const void *b)
4865 return strcmp(*(char *const *)a, *(char *const *)b);
4868 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
4869 void *opaque, bool read_only)
4871 BlockDriver *drv;
4872 int count = 0;
4873 int i;
4874 const char **formats = NULL;
4876 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4877 if (drv->format_name) {
4878 bool found = false;
4879 int i = count;
4881 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
4882 continue;
4885 while (formats && i && !found) {
4886 found = !strcmp(formats[--i], drv->format_name);
4889 if (!found) {
4890 formats = g_renew(const char *, formats, count + 1);
4891 formats[count++] = drv->format_name;
4896 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4897 const char *format_name = block_driver_modules[i].format_name;
4899 if (format_name) {
4900 bool found = false;
4901 int j = count;
4903 if (use_bdrv_whitelist &&
4904 !bdrv_format_is_whitelisted(format_name, read_only)) {
4905 continue;
4908 while (formats && j && !found) {
4909 found = !strcmp(formats[--j], format_name);
4912 if (!found) {
4913 formats = g_renew(const char *, formats, count + 1);
4914 formats[count++] = format_name;
4919 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
4921 for (i = 0; i < count; i++) {
4922 it(opaque, formats[i]);
4925 g_free(formats);
4928 /* This function is to find a node in the bs graph */
4929 BlockDriverState *bdrv_find_node(const char *node_name)
4931 BlockDriverState *bs;
4933 assert(node_name);
4935 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4936 if (!strcmp(node_name, bs->node_name)) {
4937 return bs;
4940 return NULL;
4943 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4944 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
4945 Error **errp)
4947 BlockDeviceInfoList *list, *entry;
4948 BlockDriverState *bs;
4950 list = NULL;
4951 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4952 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
4953 if (!info) {
4954 qapi_free_BlockDeviceInfoList(list);
4955 return NULL;
4957 entry = g_malloc0(sizeof(*entry));
4958 entry->value = info;
4959 entry->next = list;
4960 list = entry;
4963 return list;
4966 #define QAPI_LIST_ADD(list, element) do { \
4967 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4968 _tmp->value = (element); \
4969 _tmp->next = (list); \
4970 (list) = _tmp; \
4971 } while (0)
4973 typedef struct XDbgBlockGraphConstructor {
4974 XDbgBlockGraph *graph;
4975 GHashTable *graph_nodes;
4976 } XDbgBlockGraphConstructor;
4978 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
4980 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
4982 gr->graph = g_new0(XDbgBlockGraph, 1);
4983 gr->graph_nodes = g_hash_table_new(NULL, NULL);
4985 return gr;
4988 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
4990 XDbgBlockGraph *graph = gr->graph;
4992 g_hash_table_destroy(gr->graph_nodes);
4993 g_free(gr);
4995 return graph;
4998 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5000 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5002 if (ret != 0) {
5003 return ret;
5007 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5008 * answer of g_hash_table_lookup.
5010 ret = g_hash_table_size(gr->graph_nodes) + 1;
5011 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5013 return ret;
5016 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5017 XDbgBlockGraphNodeType type, const char *name)
5019 XDbgBlockGraphNode *n;
5021 n = g_new0(XDbgBlockGraphNode, 1);
5023 n->id = xdbg_graph_node_num(gr, node);
5024 n->type = type;
5025 n->name = g_strdup(name);
5027 QAPI_LIST_ADD(gr->graph->nodes, n);
5030 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5031 const BdrvChild *child)
5033 BlockPermission qapi_perm;
5034 XDbgBlockGraphEdge *edge;
5036 edge = g_new0(XDbgBlockGraphEdge, 1);
5038 edge->parent = xdbg_graph_node_num(gr, parent);
5039 edge->child = xdbg_graph_node_num(gr, child->bs);
5040 edge->name = g_strdup(child->name);
5042 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5043 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5045 if (flag & child->perm) {
5046 QAPI_LIST_ADD(edge->perm, qapi_perm);
5048 if (flag & child->shared_perm) {
5049 QAPI_LIST_ADD(edge->shared_perm, qapi_perm);
5053 QAPI_LIST_ADD(gr->graph->edges, edge);
5057 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5059 BlockBackend *blk;
5060 BlockJob *job;
5061 BlockDriverState *bs;
5062 BdrvChild *child;
5063 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5065 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5066 char *allocated_name = NULL;
5067 const char *name = blk_name(blk);
5069 if (!*name) {
5070 name = allocated_name = blk_get_attached_dev_id(blk);
5072 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5073 name);
5074 g_free(allocated_name);
5075 if (blk_root(blk)) {
5076 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5080 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5081 GSList *el;
5083 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5084 job->job.id);
5085 for (el = job->nodes; el; el = el->next) {
5086 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5090 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5091 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5092 bs->node_name);
5093 QLIST_FOREACH(child, &bs->children, next) {
5094 xdbg_graph_add_edge(gr, bs, child);
5098 return xdbg_graph_finalize(gr);
5101 BlockDriverState *bdrv_lookup_bs(const char *device,
5102 const char *node_name,
5103 Error **errp)
5105 BlockBackend *blk;
5106 BlockDriverState *bs;
5108 if (device) {
5109 blk = blk_by_name(device);
5111 if (blk) {
5112 bs = blk_bs(blk);
5113 if (!bs) {
5114 error_setg(errp, "Device '%s' has no medium", device);
5117 return bs;
5121 if (node_name) {
5122 bs = bdrv_find_node(node_name);
5124 if (bs) {
5125 return bs;
5129 error_setg(errp, "Cannot find device=%s nor node_name=%s",
5130 device ? device : "",
5131 node_name ? node_name : "");
5132 return NULL;
5135 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5136 * return false. If either argument is NULL, return false. */
5137 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
5139 while (top && top != base) {
5140 top = backing_bs(top);
5143 return top != NULL;
5146 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
5148 if (!bs) {
5149 return QTAILQ_FIRST(&graph_bdrv_states);
5151 return QTAILQ_NEXT(bs, node_list);
5154 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5156 if (!bs) {
5157 return QTAILQ_FIRST(&all_bdrv_states);
5159 return QTAILQ_NEXT(bs, bs_list);
5162 const char *bdrv_get_node_name(const BlockDriverState *bs)
5164 return bs->node_name;
5167 const char *bdrv_get_parent_name(const BlockDriverState *bs)
5169 BdrvChild *c;
5170 const char *name;
5172 /* If multiple parents have a name, just pick the first one. */
5173 QLIST_FOREACH(c, &bs->parents, next_parent) {
5174 if (c->role->get_name) {
5175 name = c->role->get_name(c);
5176 if (name && *name) {
5177 return name;
5182 return NULL;
5185 /* TODO check what callers really want: bs->node_name or blk_name() */
5186 const char *bdrv_get_device_name(const BlockDriverState *bs)
5188 return bdrv_get_parent_name(bs) ?: "";
5191 /* This can be used to identify nodes that might not have a device
5192 * name associated. Since node and device names live in the same
5193 * namespace, the result is unambiguous. The exception is if both are
5194 * absent, then this returns an empty (non-null) string. */
5195 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
5197 return bdrv_get_parent_name(bs) ?: bs->node_name;
5200 int bdrv_get_flags(BlockDriverState *bs)
5202 return bs->open_flags;
5205 int bdrv_has_zero_init_1(BlockDriverState *bs)
5207 return 1;
5210 int bdrv_has_zero_init(BlockDriverState *bs)
5212 if (!bs->drv) {
5213 return 0;
5216 /* If BS is a copy on write image, it is initialized to
5217 the contents of the base image, which may not be zeroes. */
5218 if (bs->backing) {
5219 return 0;
5221 if (bs->drv->bdrv_has_zero_init) {
5222 return bs->drv->bdrv_has_zero_init(bs);
5224 if (bs->file && bs->drv->is_filter) {
5225 return bdrv_has_zero_init(bs->file->bs);
5228 /* safe default */
5229 return 0;
5232 int bdrv_has_zero_init_truncate(BlockDriverState *bs)
5234 if (!bs->drv) {
5235 return 0;
5238 if (bs->backing) {
5239 /* Depends on the backing image length, but better safe than sorry */
5240 return 0;
5242 if (bs->drv->bdrv_has_zero_init_truncate) {
5243 return bs->drv->bdrv_has_zero_init_truncate(bs);
5245 if (bs->file && bs->drv->is_filter) {
5246 return bdrv_has_zero_init_truncate(bs->file->bs);
5249 /* safe default */
5250 return 0;
5253 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
5255 BlockDriverInfo bdi;
5257 if (bs->backing) {
5258 return false;
5261 if (bdrv_get_info(bs, &bdi) == 0) {
5262 return bdi.unallocated_blocks_are_zero;
5265 return false;
5268 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
5270 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5271 return false;
5274 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
5277 void bdrv_get_backing_filename(BlockDriverState *bs,
5278 char *filename, int filename_size)
5280 pstrcpy(filename, filename_size, bs->backing_file);
5283 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5285 BlockDriver *drv = bs->drv;
5286 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5287 if (!drv) {
5288 return -ENOMEDIUM;
5290 if (!drv->bdrv_get_info) {
5291 if (bs->file && drv->is_filter) {
5292 return bdrv_get_info(bs->file->bs, bdi);
5294 return -ENOTSUP;
5296 memset(bdi, 0, sizeof(*bdi));
5297 return drv->bdrv_get_info(bs, bdi);
5300 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
5301 Error **errp)
5303 BlockDriver *drv = bs->drv;
5304 if (drv && drv->bdrv_get_specific_info) {
5305 return drv->bdrv_get_specific_info(bs, errp);
5307 return NULL;
5310 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
5312 BlockDriver *drv = bs->drv;
5313 if (!drv || !drv->bdrv_get_specific_stats) {
5314 return NULL;
5316 return drv->bdrv_get_specific_stats(bs);
5319 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
5321 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
5322 return;
5325 bs->drv->bdrv_debug_event(bs, event);
5328 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
5330 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
5331 if (bs->file) {
5332 bs = bs->file->bs;
5333 continue;
5336 if (bs->drv->is_filter && bs->backing) {
5337 bs = bs->backing->bs;
5338 continue;
5341 break;
5344 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
5345 assert(bs->drv->bdrv_debug_remove_breakpoint);
5346 return bs;
5349 return NULL;
5352 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
5353 const char *tag)
5355 bs = bdrv_find_debug_node(bs);
5356 if (bs) {
5357 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
5360 return -ENOTSUP;
5363 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
5365 bs = bdrv_find_debug_node(bs);
5366 if (bs) {
5367 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
5370 return -ENOTSUP;
5373 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
5375 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
5376 bs = bs->file ? bs->file->bs : NULL;
5379 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
5380 return bs->drv->bdrv_debug_resume(bs, tag);
5383 return -ENOTSUP;
5386 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
5388 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
5389 bs = bs->file ? bs->file->bs : NULL;
5392 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
5393 return bs->drv->bdrv_debug_is_suspended(bs, tag);
5396 return false;
5399 /* backing_file can either be relative, or absolute, or a protocol. If it is
5400 * relative, it must be relative to the chain. So, passing in bs->filename
5401 * from a BDS as backing_file should not be done, as that may be relative to
5402 * the CWD rather than the chain. */
5403 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
5404 const char *backing_file)
5406 char *filename_full = NULL;
5407 char *backing_file_full = NULL;
5408 char *filename_tmp = NULL;
5409 int is_protocol = 0;
5410 BlockDriverState *curr_bs = NULL;
5411 BlockDriverState *retval = NULL;
5413 if (!bs || !bs->drv || !backing_file) {
5414 return NULL;
5417 filename_full = g_malloc(PATH_MAX);
5418 backing_file_full = g_malloc(PATH_MAX);
5420 is_protocol = path_has_protocol(backing_file);
5422 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
5424 /* If either of the filename paths is actually a protocol, then
5425 * compare unmodified paths; otherwise make paths relative */
5426 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
5427 char *backing_file_full_ret;
5429 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
5430 retval = curr_bs->backing->bs;
5431 break;
5433 /* Also check against the full backing filename for the image */
5434 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
5435 NULL);
5436 if (backing_file_full_ret) {
5437 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
5438 g_free(backing_file_full_ret);
5439 if (equal) {
5440 retval = curr_bs->backing->bs;
5441 break;
5444 } else {
5445 /* If not an absolute filename path, make it relative to the current
5446 * image's filename path */
5447 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
5448 NULL);
5449 /* We are going to compare canonicalized absolute pathnames */
5450 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
5451 g_free(filename_tmp);
5452 continue;
5454 g_free(filename_tmp);
5456 /* We need to make sure the backing filename we are comparing against
5457 * is relative to the current image filename (or absolute) */
5458 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
5459 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
5460 g_free(filename_tmp);
5461 continue;
5463 g_free(filename_tmp);
5465 if (strcmp(backing_file_full, filename_full) == 0) {
5466 retval = curr_bs->backing->bs;
5467 break;
5472 g_free(filename_full);
5473 g_free(backing_file_full);
5474 return retval;
5477 void bdrv_init(void)
5479 module_call_init(MODULE_INIT_BLOCK);
5482 void bdrv_init_with_whitelist(void)
5484 use_bdrv_whitelist = 1;
5485 bdrv_init();
5488 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
5489 Error **errp)
5491 BdrvChild *child, *parent;
5492 uint64_t perm, shared_perm;
5493 Error *local_err = NULL;
5494 int ret;
5495 BdrvDirtyBitmap *bm;
5497 if (!bs->drv) {
5498 return;
5501 QLIST_FOREACH(child, &bs->children, next) {
5502 bdrv_co_invalidate_cache(child->bs, &local_err);
5503 if (local_err) {
5504 error_propagate(errp, local_err);
5505 return;
5510 * Update permissions, they may differ for inactive nodes.
5512 * Note that the required permissions of inactive images are always a
5513 * subset of the permissions required after activating the image. This
5514 * allows us to just get the permissions upfront without restricting
5515 * drv->bdrv_invalidate_cache().
5517 * It also means that in error cases, we don't have to try and revert to
5518 * the old permissions (which is an operation that could fail, too). We can
5519 * just keep the extended permissions for the next time that an activation
5520 * of the image is tried.
5522 if (bs->open_flags & BDRV_O_INACTIVE) {
5523 bs->open_flags &= ~BDRV_O_INACTIVE;
5524 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5525 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, NULL, &local_err);
5526 if (ret < 0) {
5527 bs->open_flags |= BDRV_O_INACTIVE;
5528 error_propagate(errp, local_err);
5529 return;
5531 bdrv_set_perm(bs, perm, shared_perm);
5533 if (bs->drv->bdrv_co_invalidate_cache) {
5534 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
5535 if (local_err) {
5536 bs->open_flags |= BDRV_O_INACTIVE;
5537 error_propagate(errp, local_err);
5538 return;
5542 FOR_EACH_DIRTY_BITMAP(bs, bm) {
5543 bdrv_dirty_bitmap_skip_store(bm, false);
5546 ret = refresh_total_sectors(bs, bs->total_sectors);
5547 if (ret < 0) {
5548 bs->open_flags |= BDRV_O_INACTIVE;
5549 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5550 return;
5554 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5555 if (parent->role->activate) {
5556 parent->role->activate(parent, &local_err);
5557 if (local_err) {
5558 bs->open_flags |= BDRV_O_INACTIVE;
5559 error_propagate(errp, local_err);
5560 return;
5566 typedef struct InvalidateCacheCo {
5567 BlockDriverState *bs;
5568 Error **errp;
5569 bool done;
5570 } InvalidateCacheCo;
5572 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
5574 InvalidateCacheCo *ico = opaque;
5575 bdrv_co_invalidate_cache(ico->bs, ico->errp);
5576 ico->done = true;
5577 aio_wait_kick();
5580 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
5582 Coroutine *co;
5583 InvalidateCacheCo ico = {
5584 .bs = bs,
5585 .done = false,
5586 .errp = errp
5589 if (qemu_in_coroutine()) {
5590 /* Fast-path if already in coroutine context */
5591 bdrv_invalidate_cache_co_entry(&ico);
5592 } else {
5593 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
5594 bdrv_coroutine_enter(bs, co);
5595 BDRV_POLL_WHILE(bs, !ico.done);
5599 void bdrv_invalidate_cache_all(Error **errp)
5601 BlockDriverState *bs;
5602 Error *local_err = NULL;
5603 BdrvNextIterator it;
5605 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5606 AioContext *aio_context = bdrv_get_aio_context(bs);
5608 aio_context_acquire(aio_context);
5609 bdrv_invalidate_cache(bs, &local_err);
5610 aio_context_release(aio_context);
5611 if (local_err) {
5612 error_propagate(errp, local_err);
5613 bdrv_next_cleanup(&it);
5614 return;
5619 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5621 BdrvChild *parent;
5623 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5624 if (parent->role->parent_is_bds) {
5625 BlockDriverState *parent_bs = parent->opaque;
5626 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5627 return true;
5632 return false;
5635 static int bdrv_inactivate_recurse(BlockDriverState *bs)
5637 BdrvChild *child, *parent;
5638 bool tighten_restrictions;
5639 uint64_t perm, shared_perm;
5640 int ret;
5642 if (!bs->drv) {
5643 return -ENOMEDIUM;
5646 /* Make sure that we don't inactivate a child before its parent.
5647 * It will be covered by recursion from the yet active parent. */
5648 if (bdrv_has_bds_parent(bs, true)) {
5649 return 0;
5652 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5654 /* Inactivate this node */
5655 if (bs->drv->bdrv_inactivate) {
5656 ret = bs->drv->bdrv_inactivate(bs);
5657 if (ret < 0) {
5658 return ret;
5662 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5663 if (parent->role->inactivate) {
5664 ret = parent->role->inactivate(parent);
5665 if (ret < 0) {
5666 return ret;
5671 bs->open_flags |= BDRV_O_INACTIVE;
5673 /* Update permissions, they may differ for inactive nodes */
5674 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5675 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL,
5676 &tighten_restrictions, NULL);
5677 assert(tighten_restrictions == false);
5678 if (ret < 0) {
5679 /* We only tried to loosen restrictions, so errors are not fatal */
5680 bdrv_abort_perm_update(bs);
5681 } else {
5682 bdrv_set_perm(bs, perm, shared_perm);
5686 /* Recursively inactivate children */
5687 QLIST_FOREACH(child, &bs->children, next) {
5688 ret = bdrv_inactivate_recurse(child->bs);
5689 if (ret < 0) {
5690 return ret;
5694 return 0;
5697 int bdrv_inactivate_all(void)
5699 BlockDriverState *bs = NULL;
5700 BdrvNextIterator it;
5701 int ret = 0;
5702 GSList *aio_ctxs = NULL, *ctx;
5704 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5705 AioContext *aio_context = bdrv_get_aio_context(bs);
5707 if (!g_slist_find(aio_ctxs, aio_context)) {
5708 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5709 aio_context_acquire(aio_context);
5713 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5714 /* Nodes with BDS parents are covered by recursion from the last
5715 * parent that gets inactivated. Don't inactivate them a second
5716 * time if that has already happened. */
5717 if (bdrv_has_bds_parent(bs, false)) {
5718 continue;
5720 ret = bdrv_inactivate_recurse(bs);
5721 if (ret < 0) {
5722 bdrv_next_cleanup(&it);
5723 goto out;
5727 out:
5728 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5729 AioContext *aio_context = ctx->data;
5730 aio_context_release(aio_context);
5732 g_slist_free(aio_ctxs);
5734 return ret;
5737 /**************************************************************/
5738 /* removable device support */
5741 * Return TRUE if the media is present
5743 bool bdrv_is_inserted(BlockDriverState *bs)
5745 BlockDriver *drv = bs->drv;
5746 BdrvChild *child;
5748 if (!drv) {
5749 return false;
5751 if (drv->bdrv_is_inserted) {
5752 return drv->bdrv_is_inserted(bs);
5754 QLIST_FOREACH(child, &bs->children, next) {
5755 if (!bdrv_is_inserted(child->bs)) {
5756 return false;
5759 return true;
5763 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5765 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5767 BlockDriver *drv = bs->drv;
5769 if (drv && drv->bdrv_eject) {
5770 drv->bdrv_eject(bs, eject_flag);
5775 * Lock or unlock the media (if it is locked, the user won't be able
5776 * to eject it manually).
5778 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5780 BlockDriver *drv = bs->drv;
5782 trace_bdrv_lock_medium(bs, locked);
5784 if (drv && drv->bdrv_lock_medium) {
5785 drv->bdrv_lock_medium(bs, locked);
5789 /* Get a reference to bs */
5790 void bdrv_ref(BlockDriverState *bs)
5792 bs->refcnt++;
5795 /* Release a previously grabbed reference to bs.
5796 * If after releasing, reference count is zero, the BlockDriverState is
5797 * deleted. */
5798 void bdrv_unref(BlockDriverState *bs)
5800 if (!bs) {
5801 return;
5803 assert(bs->refcnt > 0);
5804 if (--bs->refcnt == 0) {
5805 bdrv_delete(bs);
5809 struct BdrvOpBlocker {
5810 Error *reason;
5811 QLIST_ENTRY(BdrvOpBlocker) list;
5814 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5816 BdrvOpBlocker *blocker;
5817 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5818 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5819 blocker = QLIST_FIRST(&bs->op_blockers[op]);
5820 error_propagate_prepend(errp, error_copy(blocker->reason),
5821 "Node '%s' is busy: ",
5822 bdrv_get_device_or_node_name(bs));
5823 return true;
5825 return false;
5828 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5830 BdrvOpBlocker *blocker;
5831 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5833 blocker = g_new0(BdrvOpBlocker, 1);
5834 blocker->reason = reason;
5835 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5838 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5840 BdrvOpBlocker *blocker, *next;
5841 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5842 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5843 if (blocker->reason == reason) {
5844 QLIST_REMOVE(blocker, list);
5845 g_free(blocker);
5850 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5852 int i;
5853 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5854 bdrv_op_block(bs, i, reason);
5858 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5860 int i;
5861 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5862 bdrv_op_unblock(bs, i, reason);
5866 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5868 int i;
5870 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5871 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5872 return false;
5875 return true;
5878 void bdrv_img_create(const char *filename, const char *fmt,
5879 const char *base_filename, const char *base_fmt,
5880 char *options, uint64_t img_size, int flags, bool quiet,
5881 Error **errp)
5883 QemuOptsList *create_opts = NULL;
5884 QemuOpts *opts = NULL;
5885 const char *backing_fmt, *backing_file;
5886 int64_t size;
5887 BlockDriver *drv, *proto_drv;
5888 Error *local_err = NULL;
5889 int ret = 0;
5891 /* Find driver and parse its options */
5892 drv = bdrv_find_format(fmt);
5893 if (!drv) {
5894 error_setg(errp, "Unknown file format '%s'", fmt);
5895 return;
5898 proto_drv = bdrv_find_protocol(filename, true, errp);
5899 if (!proto_drv) {
5900 return;
5903 if (!drv->create_opts) {
5904 error_setg(errp, "Format driver '%s' does not support image creation",
5905 drv->format_name);
5906 return;
5909 /* Create parameter list */
5910 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5911 if (proto_drv->create_opts) {
5912 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5913 } else {
5914 create_opts = qemu_opts_append(create_opts, &fallback_create_opts);
5917 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5919 /* Parse -o options */
5920 if (options) {
5921 qemu_opts_do_parse(opts, options, NULL, &local_err);
5922 if (local_err) {
5923 goto out;
5927 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
5928 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5929 } else if (img_size != UINT64_C(-1)) {
5930 error_setg(errp, "The image size must be specified only once");
5931 goto out;
5934 if (base_filename) {
5935 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
5936 if (local_err) {
5937 error_setg(errp, "Backing file not supported for file format '%s'",
5938 fmt);
5939 goto out;
5943 if (base_fmt) {
5944 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
5945 if (local_err) {
5946 error_setg(errp, "Backing file format not supported for file "
5947 "format '%s'", fmt);
5948 goto out;
5952 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5953 if (backing_file) {
5954 if (!strcmp(filename, backing_file)) {
5955 error_setg(errp, "Error: Trying to create an image with the "
5956 "same filename as the backing file");
5957 goto out;
5961 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5963 /* The size for the image must always be specified, unless we have a backing
5964 * file and we have not been forbidden from opening it. */
5965 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
5966 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
5967 BlockDriverState *bs;
5968 char *full_backing;
5969 int back_flags;
5970 QDict *backing_options = NULL;
5972 full_backing =
5973 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
5974 &local_err);
5975 if (local_err) {
5976 goto out;
5978 assert(full_backing);
5980 /* backing files always opened read-only */
5981 back_flags = flags;
5982 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5984 backing_options = qdict_new();
5985 if (backing_fmt) {
5986 qdict_put_str(backing_options, "driver", backing_fmt);
5988 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
5990 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
5991 &local_err);
5992 g_free(full_backing);
5993 if (!bs && size != -1) {
5994 /* Couldn't open BS, but we have a size, so it's nonfatal */
5995 warn_reportf_err(local_err,
5996 "Could not verify backing image. "
5997 "This may become an error in future versions.\n");
5998 local_err = NULL;
5999 } else if (!bs) {
6000 /* Couldn't open bs, do not have size */
6001 error_append_hint(&local_err,
6002 "Could not open backing image to determine size.\n");
6003 goto out;
6004 } else {
6005 if (size == -1) {
6006 /* Opened BS, have no size */
6007 size = bdrv_getlength(bs);
6008 if (size < 0) {
6009 error_setg_errno(errp, -size, "Could not get size of '%s'",
6010 backing_file);
6011 bdrv_unref(bs);
6012 goto out;
6014 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
6016 bdrv_unref(bs);
6018 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6020 if (size == -1) {
6021 error_setg(errp, "Image creation needs a size parameter");
6022 goto out;
6025 if (!quiet) {
6026 printf("Formatting '%s', fmt=%s ", filename, fmt);
6027 qemu_opts_print(opts, " ");
6028 puts("");
6031 ret = bdrv_create(drv, filename, opts, &local_err);
6033 if (ret == -EFBIG) {
6034 /* This is generally a better message than whatever the driver would
6035 * deliver (especially because of the cluster_size_hint), since that
6036 * is most probably not much different from "image too large". */
6037 const char *cluster_size_hint = "";
6038 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
6039 cluster_size_hint = " (try using a larger cluster size)";
6041 error_setg(errp, "The image size is too large for file format '%s'"
6042 "%s", fmt, cluster_size_hint);
6043 error_free(local_err);
6044 local_err = NULL;
6047 out:
6048 qemu_opts_del(opts);
6049 qemu_opts_free(create_opts);
6050 error_propagate(errp, local_err);
6053 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6055 return bs ? bs->aio_context : qemu_get_aio_context();
6058 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6060 aio_co_enter(bdrv_get_aio_context(bs), co);
6063 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6065 QLIST_REMOVE(ban, list);
6066 g_free(ban);
6069 static void bdrv_detach_aio_context(BlockDriverState *bs)
6071 BdrvAioNotifier *baf, *baf_tmp;
6073 assert(!bs->walking_aio_notifiers);
6074 bs->walking_aio_notifiers = true;
6075 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6076 if (baf->deleted) {
6077 bdrv_do_remove_aio_context_notifier(baf);
6078 } else {
6079 baf->detach_aio_context(baf->opaque);
6082 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6083 * remove remaining aio notifiers if we aren't called again.
6085 bs->walking_aio_notifiers = false;
6087 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
6088 bs->drv->bdrv_detach_aio_context(bs);
6091 if (bs->quiesce_counter) {
6092 aio_enable_external(bs->aio_context);
6094 bs->aio_context = NULL;
6097 static void bdrv_attach_aio_context(BlockDriverState *bs,
6098 AioContext *new_context)
6100 BdrvAioNotifier *ban, *ban_tmp;
6102 if (bs->quiesce_counter) {
6103 aio_disable_external(new_context);
6106 bs->aio_context = new_context;
6108 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
6109 bs->drv->bdrv_attach_aio_context(bs, new_context);
6112 assert(!bs->walking_aio_notifiers);
6113 bs->walking_aio_notifiers = true;
6114 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6115 if (ban->deleted) {
6116 bdrv_do_remove_aio_context_notifier(ban);
6117 } else {
6118 ban->attached_aio_context(new_context, ban->opaque);
6121 bs->walking_aio_notifiers = false;
6125 * Changes the AioContext used for fd handlers, timers, and BHs by this
6126 * BlockDriverState and all its children and parents.
6128 * Must be called from the main AioContext.
6130 * The caller must own the AioContext lock for the old AioContext of bs, but it
6131 * must not own the AioContext lock for new_context (unless new_context is the
6132 * same as the current context of bs).
6134 * @ignore will accumulate all visited BdrvChild object. The caller is
6135 * responsible for freeing the list afterwards.
6137 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6138 AioContext *new_context, GSList **ignore)
6140 AioContext *old_context = bdrv_get_aio_context(bs);
6141 BdrvChild *child;
6143 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6145 if (old_context == new_context) {
6146 return;
6149 bdrv_drained_begin(bs);
6151 QLIST_FOREACH(child, &bs->children, next) {
6152 if (g_slist_find(*ignore, child)) {
6153 continue;
6155 *ignore = g_slist_prepend(*ignore, child);
6156 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6158 QLIST_FOREACH(child, &bs->parents, next_parent) {
6159 if (g_slist_find(*ignore, child)) {
6160 continue;
6162 assert(child->role->set_aio_ctx);
6163 *ignore = g_slist_prepend(*ignore, child);
6164 child->role->set_aio_ctx(child, new_context, ignore);
6167 bdrv_detach_aio_context(bs);
6169 /* Acquire the new context, if necessary */
6170 if (qemu_get_aio_context() != new_context) {
6171 aio_context_acquire(new_context);
6174 bdrv_attach_aio_context(bs, new_context);
6177 * If this function was recursively called from
6178 * bdrv_set_aio_context_ignore(), there may be nodes in the
6179 * subtree that have not yet been moved to the new AioContext.
6180 * Release the old one so bdrv_drained_end() can poll them.
6182 if (qemu_get_aio_context() != old_context) {
6183 aio_context_release(old_context);
6186 bdrv_drained_end(bs);
6188 if (qemu_get_aio_context() != old_context) {
6189 aio_context_acquire(old_context);
6191 if (qemu_get_aio_context() != new_context) {
6192 aio_context_release(new_context);
6196 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6197 GSList **ignore, Error **errp)
6199 if (g_slist_find(*ignore, c)) {
6200 return true;
6202 *ignore = g_slist_prepend(*ignore, c);
6204 /* A BdrvChildRole that doesn't handle AioContext changes cannot
6205 * tolerate any AioContext changes */
6206 if (!c->role->can_set_aio_ctx) {
6207 char *user = bdrv_child_user_desc(c);
6208 error_setg(errp, "Changing iothreads is not supported by %s", user);
6209 g_free(user);
6210 return false;
6212 if (!c->role->can_set_aio_ctx(c, ctx, ignore, errp)) {
6213 assert(!errp || *errp);
6214 return false;
6216 return true;
6219 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6220 GSList **ignore, Error **errp)
6222 if (g_slist_find(*ignore, c)) {
6223 return true;
6225 *ignore = g_slist_prepend(*ignore, c);
6226 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6229 /* @ignore will accumulate all visited BdrvChild object. The caller is
6230 * responsible for freeing the list afterwards. */
6231 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6232 GSList **ignore, Error **errp)
6234 BdrvChild *c;
6236 if (bdrv_get_aio_context(bs) == ctx) {
6237 return true;
6240 QLIST_FOREACH(c, &bs->parents, next_parent) {
6241 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
6242 return false;
6245 QLIST_FOREACH(c, &bs->children, next) {
6246 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
6247 return false;
6251 return true;
6254 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6255 BdrvChild *ignore_child, Error **errp)
6257 GSList *ignore;
6258 bool ret;
6260 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6261 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
6262 g_slist_free(ignore);
6264 if (!ret) {
6265 return -EPERM;
6268 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6269 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
6270 g_slist_free(ignore);
6272 return 0;
6275 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6276 Error **errp)
6278 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
6281 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
6282 void (*attached_aio_context)(AioContext *new_context, void *opaque),
6283 void (*detach_aio_context)(void *opaque), void *opaque)
6285 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
6286 *ban = (BdrvAioNotifier){
6287 .attached_aio_context = attached_aio_context,
6288 .detach_aio_context = detach_aio_context,
6289 .opaque = opaque
6292 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
6295 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
6296 void (*attached_aio_context)(AioContext *,
6297 void *),
6298 void (*detach_aio_context)(void *),
6299 void *opaque)
6301 BdrvAioNotifier *ban, *ban_next;
6303 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
6304 if (ban->attached_aio_context == attached_aio_context &&
6305 ban->detach_aio_context == detach_aio_context &&
6306 ban->opaque == opaque &&
6307 ban->deleted == false)
6309 if (bs->walking_aio_notifiers) {
6310 ban->deleted = true;
6311 } else {
6312 bdrv_do_remove_aio_context_notifier(ban);
6314 return;
6318 abort();
6321 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
6322 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
6323 Error **errp)
6325 if (!bs->drv) {
6326 error_setg(errp, "Node is ejected");
6327 return -ENOMEDIUM;
6329 if (!bs->drv->bdrv_amend_options) {
6330 error_setg(errp, "Block driver '%s' does not support option amendment",
6331 bs->drv->format_name);
6332 return -ENOTSUP;
6334 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
6338 * This function checks whether the given @to_replace is allowed to be
6339 * replaced by a node that always shows the same data as @bs. This is
6340 * used for example to verify whether the mirror job can replace
6341 * @to_replace by the target mirrored from @bs.
6342 * To be replaceable, @bs and @to_replace may either be guaranteed to
6343 * always show the same data (because they are only connected through
6344 * filters), or some driver may allow replacing one of its children
6345 * because it can guarantee that this child's data is not visible at
6346 * all (for example, for dissenting quorum children that have no other
6347 * parents).
6349 bool bdrv_recurse_can_replace(BlockDriverState *bs,
6350 BlockDriverState *to_replace)
6352 if (!bs || !bs->drv) {
6353 return false;
6356 if (bs == to_replace) {
6357 return true;
6360 /* See what the driver can do */
6361 if (bs->drv->bdrv_recurse_can_replace) {
6362 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
6365 /* For filters without an own implementation, we can recurse on our own */
6366 if (bs->drv->is_filter) {
6367 BdrvChild *child = bs->file ?: bs->backing;
6368 return bdrv_recurse_can_replace(child->bs, to_replace);
6371 /* Safe default */
6372 return false;
6376 * Check whether the given @node_name can be replaced by a node that
6377 * has the same data as @parent_bs. If so, return @node_name's BDS;
6378 * NULL otherwise.
6380 * @node_name must be a (recursive) *child of @parent_bs (or this
6381 * function will return NULL).
6383 * The result (whether the node can be replaced or not) is only valid
6384 * for as long as no graph or permission changes occur.
6386 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
6387 const char *node_name, Error **errp)
6389 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
6390 AioContext *aio_context;
6392 if (!to_replace_bs) {
6393 error_setg(errp, "Node name '%s' not found", node_name);
6394 return NULL;
6397 aio_context = bdrv_get_aio_context(to_replace_bs);
6398 aio_context_acquire(aio_context);
6400 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
6401 to_replace_bs = NULL;
6402 goto out;
6405 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6406 * most non filter in order to prevent data corruption.
6407 * Another benefit is that this tests exclude backing files which are
6408 * blocked by the backing blockers.
6410 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
6411 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
6412 "because it cannot be guaranteed that doing so would not "
6413 "lead to an abrupt change of visible data",
6414 node_name, parent_bs->node_name);
6415 to_replace_bs = NULL;
6416 goto out;
6419 out:
6420 aio_context_release(aio_context);
6421 return to_replace_bs;
6425 * Iterates through the list of runtime option keys that are said to
6426 * be "strong" for a BDS. An option is called "strong" if it changes
6427 * a BDS's data. For example, the null block driver's "size" and
6428 * "read-zeroes" options are strong, but its "latency-ns" option is
6429 * not.
6431 * If a key returned by this function ends with a dot, all options
6432 * starting with that prefix are strong.
6434 static const char *const *strong_options(BlockDriverState *bs,
6435 const char *const *curopt)
6437 static const char *const global_options[] = {
6438 "driver", "filename", NULL
6441 if (!curopt) {
6442 return &global_options[0];
6445 curopt++;
6446 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
6447 curopt = bs->drv->strong_runtime_opts;
6450 return (curopt && *curopt) ? curopt : NULL;
6454 * Copies all strong runtime options from bs->options to the given
6455 * QDict. The set of strong option keys is determined by invoking
6456 * strong_options().
6458 * Returns true iff any strong option was present in bs->options (and
6459 * thus copied to the target QDict) with the exception of "filename"
6460 * and "driver". The caller is expected to use this value to decide
6461 * whether the existence of strong options prevents the generation of
6462 * a plain filename.
6464 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
6466 bool found_any = false;
6467 const char *const *option_name = NULL;
6469 if (!bs->drv) {
6470 return false;
6473 while ((option_name = strong_options(bs, option_name))) {
6474 bool option_given = false;
6476 assert(strlen(*option_name) > 0);
6477 if ((*option_name)[strlen(*option_name) - 1] != '.') {
6478 QObject *entry = qdict_get(bs->options, *option_name);
6479 if (!entry) {
6480 continue;
6483 qdict_put_obj(d, *option_name, qobject_ref(entry));
6484 option_given = true;
6485 } else {
6486 const QDictEntry *entry;
6487 for (entry = qdict_first(bs->options); entry;
6488 entry = qdict_next(bs->options, entry))
6490 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
6491 qdict_put_obj(d, qdict_entry_key(entry),
6492 qobject_ref(qdict_entry_value(entry)));
6493 option_given = true;
6498 /* While "driver" and "filename" need to be included in a JSON filename,
6499 * their existence does not prohibit generation of a plain filename. */
6500 if (!found_any && option_given &&
6501 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
6503 found_any = true;
6507 if (!qdict_haskey(d, "driver")) {
6508 /* Drivers created with bdrv_new_open_driver() may not have a
6509 * @driver option. Add it here. */
6510 qdict_put_str(d, "driver", bs->drv->format_name);
6513 return found_any;
6516 /* Note: This function may return false positives; it may return true
6517 * even if opening the backing file specified by bs's image header
6518 * would result in exactly bs->backing. */
6519 static bool bdrv_backing_overridden(BlockDriverState *bs)
6521 if (bs->backing) {
6522 return strcmp(bs->auto_backing_file,
6523 bs->backing->bs->filename);
6524 } else {
6525 /* No backing BDS, so if the image header reports any backing
6526 * file, it must have been suppressed */
6527 return bs->auto_backing_file[0] != '\0';
6531 /* Updates the following BDS fields:
6532 * - exact_filename: A filename which may be used for opening a block device
6533 * which (mostly) equals the given BDS (even without any
6534 * other options; so reading and writing must return the same
6535 * results, but caching etc. may be different)
6536 * - full_open_options: Options which, when given when opening a block device
6537 * (without a filename), result in a BDS (mostly)
6538 * equalling the given one
6539 * - filename: If exact_filename is set, it is copied here. Otherwise,
6540 * full_open_options is converted to a JSON object, prefixed with
6541 * "json:" (for use through the JSON pseudo protocol) and put here.
6543 void bdrv_refresh_filename(BlockDriverState *bs)
6545 BlockDriver *drv = bs->drv;
6546 BdrvChild *child;
6547 QDict *opts;
6548 bool backing_overridden;
6549 bool generate_json_filename; /* Whether our default implementation should
6550 fill exact_filename (false) or not (true) */
6552 if (!drv) {
6553 return;
6556 /* This BDS's file name may depend on any of its children's file names, so
6557 * refresh those first */
6558 QLIST_FOREACH(child, &bs->children, next) {
6559 bdrv_refresh_filename(child->bs);
6562 if (bs->implicit) {
6563 /* For implicit nodes, just copy everything from the single child */
6564 child = QLIST_FIRST(&bs->children);
6565 assert(QLIST_NEXT(child, next) == NULL);
6567 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
6568 child->bs->exact_filename);
6569 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
6571 qobject_unref(bs->full_open_options);
6572 bs->full_open_options = qobject_ref(child->bs->full_open_options);
6574 return;
6577 backing_overridden = bdrv_backing_overridden(bs);
6579 if (bs->open_flags & BDRV_O_NO_IO) {
6580 /* Without I/O, the backing file does not change anything.
6581 * Therefore, in such a case (primarily qemu-img), we can
6582 * pretend the backing file has not been overridden even if
6583 * it technically has been. */
6584 backing_overridden = false;
6587 /* Gather the options QDict */
6588 opts = qdict_new();
6589 generate_json_filename = append_strong_runtime_options(opts, bs);
6590 generate_json_filename |= backing_overridden;
6592 if (drv->bdrv_gather_child_options) {
6593 /* Some block drivers may not want to present all of their children's
6594 * options, or name them differently from BdrvChild.name */
6595 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
6596 } else {
6597 QLIST_FOREACH(child, &bs->children, next) {
6598 if (child->role == &child_backing && !backing_overridden) {
6599 /* We can skip the backing BDS if it has not been overridden */
6600 continue;
6603 qdict_put(opts, child->name,
6604 qobject_ref(child->bs->full_open_options));
6607 if (backing_overridden && !bs->backing) {
6608 /* Force no backing file */
6609 qdict_put_null(opts, "backing");
6613 qobject_unref(bs->full_open_options);
6614 bs->full_open_options = opts;
6616 if (drv->bdrv_refresh_filename) {
6617 /* Obsolete information is of no use here, so drop the old file name
6618 * information before refreshing it */
6619 bs->exact_filename[0] = '\0';
6621 drv->bdrv_refresh_filename(bs);
6622 } else if (bs->file) {
6623 /* Try to reconstruct valid information from the underlying file */
6625 bs->exact_filename[0] = '\0';
6628 * We can use the underlying file's filename if:
6629 * - it has a filename,
6630 * - the file is a protocol BDS, and
6631 * - opening that file (as this BDS's format) will automatically create
6632 * the BDS tree we have right now, that is:
6633 * - the user did not significantly change this BDS's behavior with
6634 * some explicit (strong) options
6635 * - no non-file child of this BDS has been overridden by the user
6636 * Both of these conditions are represented by generate_json_filename.
6638 if (bs->file->bs->exact_filename[0] &&
6639 bs->file->bs->drv->bdrv_file_open &&
6640 !generate_json_filename)
6642 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
6646 if (bs->exact_filename[0]) {
6647 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
6648 } else {
6649 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
6650 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
6651 qstring_get_str(json));
6652 qobject_unref(json);
6656 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
6658 BlockDriver *drv = bs->drv;
6660 if (!drv) {
6661 error_setg(errp, "Node '%s' is ejected", bs->node_name);
6662 return NULL;
6665 if (drv->bdrv_dirname) {
6666 return drv->bdrv_dirname(bs, errp);
6669 if (bs->file) {
6670 return bdrv_dirname(bs->file->bs, errp);
6673 bdrv_refresh_filename(bs);
6674 if (bs->exact_filename[0] != '\0') {
6675 return path_combine(bs->exact_filename, "");
6678 error_setg(errp, "Cannot generate a base directory for %s nodes",
6679 drv->format_name);
6680 return NULL;
6684 * Hot add/remove a BDS's child. So the user can take a child offline when
6685 * it is broken and take a new child online
6687 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
6688 Error **errp)
6691 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
6692 error_setg(errp, "The node %s does not support adding a child",
6693 bdrv_get_device_or_node_name(parent_bs));
6694 return;
6697 if (!QLIST_EMPTY(&child_bs->parents)) {
6698 error_setg(errp, "The node %s already has a parent",
6699 child_bs->node_name);
6700 return;
6703 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
6706 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
6708 BdrvChild *tmp;
6710 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
6711 error_setg(errp, "The node %s does not support removing a child",
6712 bdrv_get_device_or_node_name(parent_bs));
6713 return;
6716 QLIST_FOREACH(tmp, &parent_bs->children, next) {
6717 if (tmp == child) {
6718 break;
6722 if (!tmp) {
6723 error_setg(errp, "The node %s does not have a child named %s",
6724 bdrv_get_device_or_node_name(parent_bs),
6725 bdrv_get_device_or_node_name(child->bs));
6726 return;
6729 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);