vhost: ensure vhost_ops are set before calling iotlb callback
[qemu/kevin.git] / block.c
blob694396281be1dd0eb806b558b6558a04733965be
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.
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
44 #include "qemu/id.h"
45 #include "qapi/util.h"
47 #ifdef CONFIG_BSD
48 #include <sys/ioctl.h>
49 #include <sys/queue.h>
50 #ifndef __DragonFly__
51 #include <sys/disk.h>
52 #endif
53 #endif
55 #ifdef _WIN32
56 #include <windows.h>
57 #endif
59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
61 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
64 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
65 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
67 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
68 QLIST_HEAD_INITIALIZER(bdrv_drivers);
70 static BlockDriverState *bdrv_open_inherit(const char *filename,
71 const char *reference,
72 QDict *options, int flags,
73 BlockDriverState *parent,
74 const BdrvChildRole *child_role,
75 Error **errp);
77 /* If non-zero, use only whitelisted block drivers */
78 static int use_bdrv_whitelist;
80 #ifdef _WIN32
81 static int is_windows_drive_prefix(const char *filename)
83 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
84 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
85 filename[1] == ':');
88 int is_windows_drive(const char *filename)
90 if (is_windows_drive_prefix(filename) &&
91 filename[2] == '\0')
92 return 1;
93 if (strstart(filename, "\\\\.\\", NULL) ||
94 strstart(filename, "//./", NULL))
95 return 1;
96 return 0;
98 #endif
100 size_t bdrv_opt_mem_align(BlockDriverState *bs)
102 if (!bs || !bs->drv) {
103 /* page size or 4k (hdd sector size) should be on the safe side */
104 return MAX(4096, getpagesize());
107 return bs->bl.opt_mem_alignment;
110 size_t bdrv_min_mem_align(BlockDriverState *bs)
112 if (!bs || !bs->drv) {
113 /* page size or 4k (hdd sector size) should be on the safe side */
114 return MAX(4096, getpagesize());
117 return bs->bl.min_mem_alignment;
120 /* check if the path starts with "<protocol>:" */
121 int path_has_protocol(const char *path)
123 const char *p;
125 #ifdef _WIN32
126 if (is_windows_drive(path) ||
127 is_windows_drive_prefix(path)) {
128 return 0;
130 p = path + strcspn(path, ":/\\");
131 #else
132 p = path + strcspn(path, ":/");
133 #endif
135 return *p == ':';
138 int path_is_absolute(const char *path)
140 #ifdef _WIN32
141 /* specific case for names like: "\\.\d:" */
142 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
143 return 1;
145 return (*path == '/' || *path == '\\');
146 #else
147 return (*path == '/');
148 #endif
151 /* if filename is absolute, just copy it to dest. Otherwise, build a
152 path to it by considering it is relative to base_path. URL are
153 supported. */
154 void path_combine(char *dest, int dest_size,
155 const char *base_path,
156 const char *filename)
158 const char *p, *p1;
159 int len;
161 if (dest_size <= 0)
162 return;
163 if (path_is_absolute(filename)) {
164 pstrcpy(dest, dest_size, filename);
165 } else {
166 const char *protocol_stripped = NULL;
168 if (path_has_protocol(base_path)) {
169 protocol_stripped = strchr(base_path, ':');
170 if (protocol_stripped) {
171 protocol_stripped++;
174 p = protocol_stripped ?: base_path;
176 p1 = strrchr(base_path, '/');
177 #ifdef _WIN32
179 const char *p2;
180 p2 = strrchr(base_path, '\\');
181 if (!p1 || p2 > p1)
182 p1 = p2;
184 #endif
185 if (p1)
186 p1++;
187 else
188 p1 = base_path;
189 if (p1 > p)
190 p = p1;
191 len = p - base_path;
192 if (len > dest_size - 1)
193 len = dest_size - 1;
194 memcpy(dest, base_path, len);
195 dest[len] = '\0';
196 pstrcat(dest, dest_size, filename);
201 * Helper function for bdrv_parse_filename() implementations to remove optional
202 * protocol prefixes (especially "file:") from a filename and for putting the
203 * stripped filename into the options QDict if there is such a prefix.
205 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
206 QDict *options)
208 if (strstart(filename, prefix, &filename)) {
209 /* Stripping the explicit protocol prefix may result in a protocol
210 * prefix being (wrongly) detected (if the filename contains a colon) */
211 if (path_has_protocol(filename)) {
212 QString *fat_filename;
214 /* This means there is some colon before the first slash; therefore,
215 * this cannot be an absolute path */
216 assert(!path_is_absolute(filename));
218 /* And we can thus fix the protocol detection issue by prefixing it
219 * by "./" */
220 fat_filename = qstring_from_str("./");
221 qstring_append(fat_filename, filename);
223 assert(!path_has_protocol(qstring_get_str(fat_filename)));
225 qdict_put(options, "filename", fat_filename);
226 } else {
227 /* If no protocol prefix was detected, we can use the shortened
228 * filename as-is */
229 qdict_put_str(options, "filename", filename);
235 /* Returns whether the image file is opened as read-only. Note that this can
236 * return false and writing to the image file is still not possible because the
237 * image is inactivated. */
238 bool bdrv_is_read_only(BlockDriverState *bs)
240 return bs->read_only;
243 /* Returns whether the image file can be written to right now */
244 bool bdrv_is_writable(BlockDriverState *bs)
246 return !bdrv_is_read_only(bs) && !(bs->open_flags & BDRV_O_INACTIVE);
249 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, 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 error_setg(errp, "Node '%s' is read only",
261 bdrv_get_device_or_node_name(bs));
262 return -EPERM;
265 return 0;
268 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
270 int ret = 0;
272 ret = bdrv_can_set_read_only(bs, read_only, errp);
273 if (ret < 0) {
274 return ret;
277 bs->read_only = read_only;
278 return 0;
281 void bdrv_get_full_backing_filename_from_filename(const char *backed,
282 const char *backing,
283 char *dest, size_t sz,
284 Error **errp)
286 if (backing[0] == '\0' || path_has_protocol(backing) ||
287 path_is_absolute(backing))
289 pstrcpy(dest, sz, backing);
290 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
291 error_setg(errp, "Cannot use relative backing file names for '%s'",
292 backed);
293 } else {
294 path_combine(dest, sz, backed, backing);
298 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
299 Error **errp)
301 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
303 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
304 dest, sz, errp);
307 void bdrv_register(BlockDriver *bdrv)
309 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
312 BlockDriverState *bdrv_new(void)
314 BlockDriverState *bs;
315 int i;
317 bs = g_new0(BlockDriverState, 1);
318 QLIST_INIT(&bs->dirty_bitmaps);
319 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
320 QLIST_INIT(&bs->op_blockers[i]);
322 notifier_with_return_list_init(&bs->before_write_notifiers);
323 qemu_co_mutex_init(&bs->reqs_lock);
324 qemu_mutex_init(&bs->dirty_bitmap_mutex);
325 bs->refcnt = 1;
326 bs->aio_context = qemu_get_aio_context();
328 qemu_co_queue_init(&bs->flush_queue);
330 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
332 return bs;
335 static BlockDriver *bdrv_do_find_format(const char *format_name)
337 BlockDriver *drv1;
339 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
340 if (!strcmp(drv1->format_name, format_name)) {
341 return drv1;
345 return NULL;
348 BlockDriver *bdrv_find_format(const char *format_name)
350 BlockDriver *drv1;
351 int i;
353 drv1 = bdrv_do_find_format(format_name);
354 if (drv1) {
355 return drv1;
358 /* The driver isn't registered, maybe we need to load a module */
359 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
360 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
361 block_module_load_one(block_driver_modules[i].library_name);
362 break;
366 return bdrv_do_find_format(format_name);
369 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
371 static const char *whitelist_rw[] = {
372 CONFIG_BDRV_RW_WHITELIST
374 static const char *whitelist_ro[] = {
375 CONFIG_BDRV_RO_WHITELIST
377 const char **p;
379 if (!whitelist_rw[0] && !whitelist_ro[0]) {
380 return 1; /* no whitelist, anything goes */
383 for (p = whitelist_rw; *p; p++) {
384 if (!strcmp(drv->format_name, *p)) {
385 return 1;
388 if (read_only) {
389 for (p = whitelist_ro; *p; p++) {
390 if (!strcmp(drv->format_name, *p)) {
391 return 1;
395 return 0;
398 bool bdrv_uses_whitelist(void)
400 return use_bdrv_whitelist;
403 typedef struct CreateCo {
404 BlockDriver *drv;
405 char *filename;
406 QemuOpts *opts;
407 int ret;
408 Error *err;
409 } CreateCo;
411 static void coroutine_fn bdrv_create_co_entry(void *opaque)
413 Error *local_err = NULL;
414 int ret;
416 CreateCo *cco = opaque;
417 assert(cco->drv);
419 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
420 error_propagate(&cco->err, local_err);
421 cco->ret = ret;
424 int bdrv_create(BlockDriver *drv, const char* filename,
425 QemuOpts *opts, Error **errp)
427 int ret;
429 Coroutine *co;
430 CreateCo cco = {
431 .drv = drv,
432 .filename = g_strdup(filename),
433 .opts = opts,
434 .ret = NOT_DONE,
435 .err = NULL,
438 if (!drv->bdrv_create) {
439 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
440 ret = -ENOTSUP;
441 goto out;
444 if (qemu_in_coroutine()) {
445 /* Fast-path if already in coroutine context */
446 bdrv_create_co_entry(&cco);
447 } else {
448 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
449 qemu_coroutine_enter(co);
450 while (cco.ret == NOT_DONE) {
451 aio_poll(qemu_get_aio_context(), true);
455 ret = cco.ret;
456 if (ret < 0) {
457 if (cco.err) {
458 error_propagate(errp, cco.err);
459 } else {
460 error_setg_errno(errp, -ret, "Could not create image");
464 out:
465 g_free(cco.filename);
466 return ret;
469 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
471 BlockDriver *drv;
472 Error *local_err = NULL;
473 int ret;
475 drv = bdrv_find_protocol(filename, true, errp);
476 if (drv == NULL) {
477 return -ENOENT;
480 ret = bdrv_create(drv, filename, opts, &local_err);
481 error_propagate(errp, local_err);
482 return ret;
486 * Try to get @bs's logical and physical block size.
487 * On success, store them in @bsz struct and return 0.
488 * On failure return -errno.
489 * @bs must not be empty.
491 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
493 BlockDriver *drv = bs->drv;
495 if (drv && drv->bdrv_probe_blocksizes) {
496 return drv->bdrv_probe_blocksizes(bs, bsz);
499 return -ENOTSUP;
503 * Try to get @bs's geometry (cyls, heads, sectors).
504 * On success, store them in @geo struct and return 0.
505 * On failure return -errno.
506 * @bs must not be empty.
508 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
510 BlockDriver *drv = bs->drv;
512 if (drv && drv->bdrv_probe_geometry) {
513 return drv->bdrv_probe_geometry(bs, geo);
516 return -ENOTSUP;
520 * Create a uniquely-named empty temporary file.
521 * Return 0 upon success, otherwise a negative errno value.
523 int get_tmp_filename(char *filename, int size)
525 #ifdef _WIN32
526 char temp_dir[MAX_PATH];
527 /* GetTempFileName requires that its output buffer (4th param)
528 have length MAX_PATH or greater. */
529 assert(size >= MAX_PATH);
530 return (GetTempPath(MAX_PATH, temp_dir)
531 && GetTempFileName(temp_dir, "qem", 0, filename)
532 ? 0 : -GetLastError());
533 #else
534 int fd;
535 const char *tmpdir;
536 tmpdir = getenv("TMPDIR");
537 if (!tmpdir) {
538 tmpdir = "/var/tmp";
540 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
541 return -EOVERFLOW;
543 fd = mkstemp(filename);
544 if (fd < 0) {
545 return -errno;
547 if (close(fd) != 0) {
548 unlink(filename);
549 return -errno;
551 return 0;
552 #endif
556 * Detect host devices. By convention, /dev/cdrom[N] is always
557 * recognized as a host CDROM.
559 static BlockDriver *find_hdev_driver(const char *filename)
561 int score_max = 0, score;
562 BlockDriver *drv = NULL, *d;
564 QLIST_FOREACH(d, &bdrv_drivers, list) {
565 if (d->bdrv_probe_device) {
566 score = d->bdrv_probe_device(filename);
567 if (score > score_max) {
568 score_max = score;
569 drv = d;
574 return drv;
577 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
579 BlockDriver *drv1;
581 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
582 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
583 return drv1;
587 return NULL;
590 BlockDriver *bdrv_find_protocol(const char *filename,
591 bool allow_protocol_prefix,
592 Error **errp)
594 BlockDriver *drv1;
595 char protocol[128];
596 int len;
597 const char *p;
598 int i;
600 /* TODO Drivers without bdrv_file_open must be specified explicitly */
603 * XXX(hch): we really should not let host device detection
604 * override an explicit protocol specification, but moving this
605 * later breaks access to device names with colons in them.
606 * Thanks to the brain-dead persistent naming schemes on udev-
607 * based Linux systems those actually are quite common.
609 drv1 = find_hdev_driver(filename);
610 if (drv1) {
611 return drv1;
614 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
615 return &bdrv_file;
618 p = strchr(filename, ':');
619 assert(p != NULL);
620 len = p - filename;
621 if (len > sizeof(protocol) - 1)
622 len = sizeof(protocol) - 1;
623 memcpy(protocol, filename, len);
624 protocol[len] = '\0';
626 drv1 = bdrv_do_find_protocol(protocol);
627 if (drv1) {
628 return drv1;
631 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
632 if (block_driver_modules[i].protocol_name &&
633 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
634 block_module_load_one(block_driver_modules[i].library_name);
635 break;
639 drv1 = bdrv_do_find_protocol(protocol);
640 if (!drv1) {
641 error_setg(errp, "Unknown protocol '%s'", protocol);
643 return drv1;
647 * Guess image format by probing its contents.
648 * This is not a good idea when your image is raw (CVE-2008-2004), but
649 * we do it anyway for backward compatibility.
651 * @buf contains the image's first @buf_size bytes.
652 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
653 * but can be smaller if the image file is smaller)
654 * @filename is its filename.
656 * For all block drivers, call the bdrv_probe() method to get its
657 * probing score.
658 * Return the first block driver with the highest probing score.
660 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
661 const char *filename)
663 int score_max = 0, score;
664 BlockDriver *drv = NULL, *d;
666 QLIST_FOREACH(d, &bdrv_drivers, list) {
667 if (d->bdrv_probe) {
668 score = d->bdrv_probe(buf, buf_size, filename);
669 if (score > score_max) {
670 score_max = score;
671 drv = d;
676 return drv;
679 static int find_image_format(BlockBackend *file, const char *filename,
680 BlockDriver **pdrv, Error **errp)
682 BlockDriver *drv;
683 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
684 int ret = 0;
686 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
687 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
688 *pdrv = &bdrv_raw;
689 return ret;
692 ret = blk_pread(file, 0, buf, sizeof(buf));
693 if (ret < 0) {
694 error_setg_errno(errp, -ret, "Could not read image for determining its "
695 "format");
696 *pdrv = NULL;
697 return ret;
700 drv = bdrv_probe_all(buf, ret, filename);
701 if (!drv) {
702 error_setg(errp, "Could not determine image format: No compatible "
703 "driver found");
704 ret = -ENOENT;
706 *pdrv = drv;
707 return ret;
711 * Set the current 'total_sectors' value
712 * Return 0 on success, -errno on error.
714 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
716 BlockDriver *drv = bs->drv;
718 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
719 if (bdrv_is_sg(bs))
720 return 0;
722 /* query actual device if possible, otherwise just trust the hint */
723 if (drv->bdrv_getlength) {
724 int64_t length = drv->bdrv_getlength(bs);
725 if (length < 0) {
726 return length;
728 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
731 bs->total_sectors = hint;
732 return 0;
736 * Combines a QDict of new block driver @options with any missing options taken
737 * from @old_options, so that leaving out an option defaults to its old value.
739 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
740 QDict *old_options)
742 if (bs->drv && bs->drv->bdrv_join_options) {
743 bs->drv->bdrv_join_options(options, old_options);
744 } else {
745 qdict_join(options, old_options, false);
750 * Set open flags for a given discard mode
752 * Return 0 on success, -1 if the discard mode was invalid.
754 int bdrv_parse_discard_flags(const char *mode, int *flags)
756 *flags &= ~BDRV_O_UNMAP;
758 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
759 /* do nothing */
760 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
761 *flags |= BDRV_O_UNMAP;
762 } else {
763 return -1;
766 return 0;
770 * Set open flags for a given cache mode
772 * Return 0 on success, -1 if the cache mode was invalid.
774 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
776 *flags &= ~BDRV_O_CACHE_MASK;
778 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
779 *writethrough = false;
780 *flags |= BDRV_O_NOCACHE;
781 } else if (!strcmp(mode, "directsync")) {
782 *writethrough = true;
783 *flags |= BDRV_O_NOCACHE;
784 } else if (!strcmp(mode, "writeback")) {
785 *writethrough = false;
786 } else if (!strcmp(mode, "unsafe")) {
787 *writethrough = false;
788 *flags |= BDRV_O_NO_FLUSH;
789 } else if (!strcmp(mode, "writethrough")) {
790 *writethrough = true;
791 } else {
792 return -1;
795 return 0;
798 static char *bdrv_child_get_parent_desc(BdrvChild *c)
800 BlockDriverState *parent = c->opaque;
801 return g_strdup(bdrv_get_device_or_node_name(parent));
804 static void bdrv_child_cb_drained_begin(BdrvChild *child)
806 BlockDriverState *bs = child->opaque;
807 bdrv_drained_begin(bs);
810 static void bdrv_child_cb_drained_end(BdrvChild *child)
812 BlockDriverState *bs = child->opaque;
813 bdrv_drained_end(bs);
816 static int bdrv_child_cb_inactivate(BdrvChild *child)
818 BlockDriverState *bs = child->opaque;
819 assert(bs->open_flags & BDRV_O_INACTIVE);
820 return 0;
824 * Returns the options and flags that a temporary snapshot should get, based on
825 * the originally requested flags (the originally requested image will have
826 * flags like a backing file)
828 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
829 int parent_flags, QDict *parent_options)
831 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
833 /* For temporary files, unconditional cache=unsafe is fine */
834 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
835 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
837 /* Copy the read-only option from the parent */
838 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
840 /* aio=native doesn't work for cache.direct=off, so disable it for the
841 * temporary snapshot */
842 *child_flags &= ~BDRV_O_NATIVE_AIO;
846 * Returns the options and flags that bs->file should get if a protocol driver
847 * is expected, based on the given options and flags for the parent BDS
849 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
850 int parent_flags, QDict *parent_options)
852 int flags = parent_flags;
854 /* Enable protocol handling, disable format probing for bs->file */
855 flags |= BDRV_O_PROTOCOL;
857 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
858 * the parent. */
859 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
860 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
861 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
863 /* Inherit the read-only option from the parent if it's not set */
864 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
866 /* Our block drivers take care to send flushes and respect unmap policy,
867 * so we can default to enable both on lower layers regardless of the
868 * corresponding parent options. */
869 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
871 /* Clear flags that only apply to the top layer */
872 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
873 BDRV_O_NO_IO);
875 *child_flags = flags;
878 const BdrvChildRole child_file = {
879 .get_parent_desc = bdrv_child_get_parent_desc,
880 .inherit_options = bdrv_inherited_options,
881 .drained_begin = bdrv_child_cb_drained_begin,
882 .drained_end = bdrv_child_cb_drained_end,
883 .inactivate = bdrv_child_cb_inactivate,
887 * Returns the options and flags that bs->file should get if the use of formats
888 * (and not only protocols) is permitted for it, based on the given options and
889 * flags for the parent BDS
891 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
892 int parent_flags, QDict *parent_options)
894 child_file.inherit_options(child_flags, child_options,
895 parent_flags, parent_options);
897 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
900 const BdrvChildRole child_format = {
901 .get_parent_desc = bdrv_child_get_parent_desc,
902 .inherit_options = bdrv_inherited_fmt_options,
903 .drained_begin = bdrv_child_cb_drained_begin,
904 .drained_end = bdrv_child_cb_drained_end,
905 .inactivate = bdrv_child_cb_inactivate,
908 static void bdrv_backing_attach(BdrvChild *c)
910 BlockDriverState *parent = c->opaque;
911 BlockDriverState *backing_hd = c->bs;
913 assert(!parent->backing_blocker);
914 error_setg(&parent->backing_blocker,
915 "node is used as backing hd of '%s'",
916 bdrv_get_device_or_node_name(parent));
918 parent->open_flags &= ~BDRV_O_NO_BACKING;
919 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
920 backing_hd->filename);
921 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
922 backing_hd->drv ? backing_hd->drv->format_name : "");
924 bdrv_op_block_all(backing_hd, parent->backing_blocker);
925 /* Otherwise we won't be able to commit or stream */
926 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
927 parent->backing_blocker);
928 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
929 parent->backing_blocker);
931 * We do backup in 3 ways:
932 * 1. drive backup
933 * The target bs is new opened, and the source is top BDS
934 * 2. blockdev backup
935 * Both the source and the target are top BDSes.
936 * 3. internal backup(used for block replication)
937 * Both the source and the target are backing file
939 * In case 1 and 2, neither the source nor the target is the backing file.
940 * In case 3, we will block the top BDS, so there is only one block job
941 * for the top BDS and its backing chain.
943 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
944 parent->backing_blocker);
945 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
946 parent->backing_blocker);
949 static void bdrv_backing_detach(BdrvChild *c)
951 BlockDriverState *parent = c->opaque;
953 assert(parent->backing_blocker);
954 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
955 error_free(parent->backing_blocker);
956 parent->backing_blocker = NULL;
960 * Returns the options and flags that bs->backing should get, based on the
961 * given options and flags for the parent BDS
963 static void bdrv_backing_options(int *child_flags, QDict *child_options,
964 int parent_flags, QDict *parent_options)
966 int flags = parent_flags;
968 /* The cache mode is inherited unmodified for backing files; except WCE,
969 * which is only applied on the top level (BlockBackend) */
970 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
971 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
972 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
974 /* backing files always opened read-only */
975 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
976 flags &= ~BDRV_O_COPY_ON_READ;
978 /* snapshot=on is handled on the top layer */
979 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
981 *child_flags = flags;
984 const BdrvChildRole child_backing = {
985 .get_parent_desc = bdrv_child_get_parent_desc,
986 .attach = bdrv_backing_attach,
987 .detach = bdrv_backing_detach,
988 .inherit_options = bdrv_backing_options,
989 .drained_begin = bdrv_child_cb_drained_begin,
990 .drained_end = bdrv_child_cb_drained_end,
991 .inactivate = bdrv_child_cb_inactivate,
994 static int bdrv_open_flags(BlockDriverState *bs, int flags)
996 int open_flags = flags;
999 * Clear flags that are internal to the block layer before opening the
1000 * image.
1002 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1005 * Snapshots should be writable.
1007 if (flags & BDRV_O_TEMPORARY) {
1008 open_flags |= BDRV_O_RDWR;
1011 return open_flags;
1014 static void update_flags_from_options(int *flags, QemuOpts *opts)
1016 *flags &= ~BDRV_O_CACHE_MASK;
1018 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1019 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1020 *flags |= BDRV_O_NO_FLUSH;
1023 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1024 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1025 *flags |= BDRV_O_NOCACHE;
1028 *flags &= ~BDRV_O_RDWR;
1030 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1031 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1032 *flags |= BDRV_O_RDWR;
1037 static void update_options_from_flags(QDict *options, int flags)
1039 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1040 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1042 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1043 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1044 flags & BDRV_O_NO_FLUSH);
1046 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1047 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1051 static void bdrv_assign_node_name(BlockDriverState *bs,
1052 const char *node_name,
1053 Error **errp)
1055 char *gen_node_name = NULL;
1057 if (!node_name) {
1058 node_name = gen_node_name = id_generate(ID_BLOCK);
1059 } else if (!id_wellformed(node_name)) {
1061 * Check for empty string or invalid characters, but not if it is
1062 * generated (generated names use characters not available to the user)
1064 error_setg(errp, "Invalid node name");
1065 return;
1068 /* takes care of avoiding namespaces collisions */
1069 if (blk_by_name(node_name)) {
1070 error_setg(errp, "node-name=%s is conflicting with a device id",
1071 node_name);
1072 goto out;
1075 /* takes care of avoiding duplicates node names */
1076 if (bdrv_find_node(node_name)) {
1077 error_setg(errp, "Duplicate node name");
1078 goto out;
1081 /* copy node name into the bs and insert it into the graph list */
1082 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1083 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1084 out:
1085 g_free(gen_node_name);
1088 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1089 const char *node_name, QDict *options,
1090 int open_flags, Error **errp)
1092 Error *local_err = NULL;
1093 int ret;
1095 bdrv_assign_node_name(bs, node_name, &local_err);
1096 if (local_err) {
1097 error_propagate(errp, local_err);
1098 return -EINVAL;
1101 bs->drv = drv;
1102 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1103 bs->opaque = g_malloc0(drv->instance_size);
1105 if (drv->bdrv_file_open) {
1106 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1107 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1108 } else if (drv->bdrv_open) {
1109 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1110 } else {
1111 ret = 0;
1114 if (ret < 0) {
1115 if (local_err) {
1116 error_propagate(errp, local_err);
1117 } else if (bs->filename[0]) {
1118 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1119 } else {
1120 error_setg_errno(errp, -ret, "Could not open image");
1122 goto free_and_fail;
1125 ret = refresh_total_sectors(bs, bs->total_sectors);
1126 if (ret < 0) {
1127 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1128 goto free_and_fail;
1131 bdrv_refresh_limits(bs, &local_err);
1132 if (local_err) {
1133 error_propagate(errp, local_err);
1134 ret = -EINVAL;
1135 goto free_and_fail;
1138 assert(bdrv_opt_mem_align(bs) != 0);
1139 assert(bdrv_min_mem_align(bs) != 0);
1140 assert(is_power_of_2(bs->bl.request_alignment));
1142 return 0;
1144 free_and_fail:
1145 /* FIXME Close bs first if already opened*/
1146 g_free(bs->opaque);
1147 bs->opaque = NULL;
1148 bs->drv = NULL;
1149 return ret;
1152 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1153 int flags, Error **errp)
1155 BlockDriverState *bs;
1156 int ret;
1158 bs = bdrv_new();
1159 bs->open_flags = flags;
1160 bs->explicit_options = qdict_new();
1161 bs->options = qdict_new();
1162 bs->opaque = NULL;
1164 update_options_from_flags(bs->options, flags);
1166 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1167 if (ret < 0) {
1168 QDECREF(bs->explicit_options);
1169 QDECREF(bs->options);
1170 bdrv_unref(bs);
1171 return NULL;
1174 return bs;
1177 QemuOptsList bdrv_runtime_opts = {
1178 .name = "bdrv_common",
1179 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1180 .desc = {
1182 .name = "node-name",
1183 .type = QEMU_OPT_STRING,
1184 .help = "Node name of the block device node",
1187 .name = "driver",
1188 .type = QEMU_OPT_STRING,
1189 .help = "Block driver to use for the node",
1192 .name = BDRV_OPT_CACHE_DIRECT,
1193 .type = QEMU_OPT_BOOL,
1194 .help = "Bypass software writeback cache on the host",
1197 .name = BDRV_OPT_CACHE_NO_FLUSH,
1198 .type = QEMU_OPT_BOOL,
1199 .help = "Ignore flush requests",
1202 .name = BDRV_OPT_READ_ONLY,
1203 .type = QEMU_OPT_BOOL,
1204 .help = "Node is opened in read-only mode",
1207 .name = "detect-zeroes",
1208 .type = QEMU_OPT_STRING,
1209 .help = "try to optimize zero writes (off, on, unmap)",
1212 .name = "discard",
1213 .type = QEMU_OPT_STRING,
1214 .help = "discard operation (ignore/off, unmap/on)",
1217 .name = BDRV_OPT_FORCE_SHARE,
1218 .type = QEMU_OPT_BOOL,
1219 .help = "always accept other writers (default: off)",
1221 { /* end of list */ }
1226 * Common part for opening disk images and files
1228 * Removes all processed options from *options.
1230 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1231 QDict *options, Error **errp)
1233 int ret, open_flags;
1234 const char *filename;
1235 const char *driver_name = NULL;
1236 const char *node_name = NULL;
1237 const char *discard;
1238 const char *detect_zeroes;
1239 QemuOpts *opts;
1240 BlockDriver *drv;
1241 Error *local_err = NULL;
1243 assert(bs->file == NULL);
1244 assert(options != NULL && bs->options != options);
1246 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1247 qemu_opts_absorb_qdict(opts, options, &local_err);
1248 if (local_err) {
1249 error_propagate(errp, local_err);
1250 ret = -EINVAL;
1251 goto fail_opts;
1254 update_flags_from_options(&bs->open_flags, opts);
1256 driver_name = qemu_opt_get(opts, "driver");
1257 drv = bdrv_find_format(driver_name);
1258 assert(drv != NULL);
1260 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1262 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1263 error_setg(errp,
1264 BDRV_OPT_FORCE_SHARE
1265 "=on can only be used with read-only images");
1266 ret = -EINVAL;
1267 goto fail_opts;
1270 if (file != NULL) {
1271 filename = blk_bs(file)->filename;
1272 } else {
1274 * Caution: while qdict_get_try_str() is fine, getting
1275 * non-string types would require more care. When @options
1276 * come from -blockdev or blockdev_add, its members are typed
1277 * according to the QAPI schema, but when they come from
1278 * -drive, they're all QString.
1280 filename = qdict_get_try_str(options, "filename");
1283 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1284 error_setg(errp, "The '%s' block driver requires a file name",
1285 drv->format_name);
1286 ret = -EINVAL;
1287 goto fail_opts;
1290 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1291 drv->format_name);
1293 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1295 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1296 error_setg(errp,
1297 !bs->read_only && bdrv_is_whitelisted(drv, true)
1298 ? "Driver '%s' can only be used for read-only devices"
1299 : "Driver '%s' is not whitelisted",
1300 drv->format_name);
1301 ret = -ENOTSUP;
1302 goto fail_opts;
1305 /* bdrv_new() and bdrv_close() make it so */
1306 assert(atomic_read(&bs->copy_on_read) == 0);
1308 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1309 if (!bs->read_only) {
1310 bdrv_enable_copy_on_read(bs);
1311 } else {
1312 error_setg(errp, "Can't use copy-on-read on read-only device");
1313 ret = -EINVAL;
1314 goto fail_opts;
1318 discard = qemu_opt_get(opts, "discard");
1319 if (discard != NULL) {
1320 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1321 error_setg(errp, "Invalid discard option");
1322 ret = -EINVAL;
1323 goto fail_opts;
1327 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1328 if (detect_zeroes) {
1329 BlockdevDetectZeroesOptions value =
1330 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1331 detect_zeroes,
1332 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1333 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1334 &local_err);
1335 if (local_err) {
1336 error_propagate(errp, local_err);
1337 ret = -EINVAL;
1338 goto fail_opts;
1341 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1342 !(bs->open_flags & BDRV_O_UNMAP))
1344 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1345 "without setting discard operation to unmap");
1346 ret = -EINVAL;
1347 goto fail_opts;
1350 bs->detect_zeroes = value;
1353 if (filename != NULL) {
1354 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1355 } else {
1356 bs->filename[0] = '\0';
1358 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1360 /* Open the image, either directly or using a protocol */
1361 open_flags = bdrv_open_flags(bs, bs->open_flags);
1362 node_name = qemu_opt_get(opts, "node-name");
1364 assert(!drv->bdrv_file_open || file == NULL);
1365 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1366 if (ret < 0) {
1367 goto fail_opts;
1370 qemu_opts_del(opts);
1371 return 0;
1373 fail_opts:
1374 qemu_opts_del(opts);
1375 return ret;
1378 static QDict *parse_json_filename(const char *filename, Error **errp)
1380 QObject *options_obj;
1381 QDict *options;
1382 int ret;
1384 ret = strstart(filename, "json:", &filename);
1385 assert(ret);
1387 options_obj = qobject_from_json(filename, errp);
1388 if (!options_obj) {
1389 /* Work around qobject_from_json() lossage TODO fix that */
1390 if (errp && !*errp) {
1391 error_setg(errp, "Could not parse the JSON options");
1392 return NULL;
1394 error_prepend(errp, "Could not parse the JSON options: ");
1395 return NULL;
1398 options = qobject_to_qdict(options_obj);
1399 if (!options) {
1400 qobject_decref(options_obj);
1401 error_setg(errp, "Invalid JSON object given");
1402 return NULL;
1405 qdict_flatten(options);
1407 return options;
1410 static void parse_json_protocol(QDict *options, const char **pfilename,
1411 Error **errp)
1413 QDict *json_options;
1414 Error *local_err = NULL;
1416 /* Parse json: pseudo-protocol */
1417 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1418 return;
1421 json_options = parse_json_filename(*pfilename, &local_err);
1422 if (local_err) {
1423 error_propagate(errp, local_err);
1424 return;
1427 /* Options given in the filename have lower priority than options
1428 * specified directly */
1429 qdict_join(options, json_options, false);
1430 QDECREF(json_options);
1431 *pfilename = NULL;
1435 * Fills in default options for opening images and converts the legacy
1436 * filename/flags pair to option QDict entries.
1437 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1438 * block driver has been specified explicitly.
1440 static int bdrv_fill_options(QDict **options, const char *filename,
1441 int *flags, Error **errp)
1443 const char *drvname;
1444 bool protocol = *flags & BDRV_O_PROTOCOL;
1445 bool parse_filename = false;
1446 BlockDriver *drv = NULL;
1447 Error *local_err = NULL;
1450 * Caution: while qdict_get_try_str() is fine, getting non-string
1451 * types would require more care. When @options come from
1452 * -blockdev or blockdev_add, its members are typed according to
1453 * the QAPI schema, but when they come from -drive, they're all
1454 * QString.
1456 drvname = qdict_get_try_str(*options, "driver");
1457 if (drvname) {
1458 drv = bdrv_find_format(drvname);
1459 if (!drv) {
1460 error_setg(errp, "Unknown driver '%s'", drvname);
1461 return -ENOENT;
1463 /* If the user has explicitly specified the driver, this choice should
1464 * override the BDRV_O_PROTOCOL flag */
1465 protocol = drv->bdrv_file_open;
1468 if (protocol) {
1469 *flags |= BDRV_O_PROTOCOL;
1470 } else {
1471 *flags &= ~BDRV_O_PROTOCOL;
1474 /* Translate cache options from flags into options */
1475 update_options_from_flags(*options, *flags);
1477 /* Fetch the file name from the options QDict if necessary */
1478 if (protocol && filename) {
1479 if (!qdict_haskey(*options, "filename")) {
1480 qdict_put_str(*options, "filename", filename);
1481 parse_filename = true;
1482 } else {
1483 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1484 "the same time");
1485 return -EINVAL;
1489 /* Find the right block driver */
1490 /* See cautionary note on accessing @options above */
1491 filename = qdict_get_try_str(*options, "filename");
1493 if (!drvname && protocol) {
1494 if (filename) {
1495 drv = bdrv_find_protocol(filename, parse_filename, errp);
1496 if (!drv) {
1497 return -EINVAL;
1500 drvname = drv->format_name;
1501 qdict_put_str(*options, "driver", drvname);
1502 } else {
1503 error_setg(errp, "Must specify either driver or file");
1504 return -EINVAL;
1508 assert(drv || !protocol);
1510 /* Driver-specific filename parsing */
1511 if (drv && drv->bdrv_parse_filename && parse_filename) {
1512 drv->bdrv_parse_filename(filename, *options, &local_err);
1513 if (local_err) {
1514 error_propagate(errp, local_err);
1515 return -EINVAL;
1518 if (!drv->bdrv_needs_filename) {
1519 qdict_del(*options, "filename");
1523 return 0;
1526 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1527 GSList *ignore_children, Error **errp);
1528 static void bdrv_child_abort_perm_update(BdrvChild *c);
1529 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1531 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1532 BdrvChild *c,
1533 const BdrvChildRole *role,
1534 uint64_t parent_perm, uint64_t parent_shared,
1535 uint64_t *nperm, uint64_t *nshared)
1537 if (bs->drv && bs->drv->bdrv_child_perm) {
1538 bs->drv->bdrv_child_perm(bs, c, role,
1539 parent_perm, parent_shared,
1540 nperm, nshared);
1542 if (child_bs && child_bs->force_share) {
1543 *nshared = BLK_PERM_ALL;
1548 * Check whether permissions on this node can be changed in a way that
1549 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1550 * permissions of all its parents. This involves checking whether all necessary
1551 * permission changes to child nodes can be performed.
1553 * A call to this function must always be followed by a call to bdrv_set_perm()
1554 * or bdrv_abort_perm_update().
1556 static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1557 uint64_t cumulative_shared_perms,
1558 GSList *ignore_children, Error **errp)
1560 BlockDriver *drv = bs->drv;
1561 BdrvChild *c;
1562 int ret;
1564 /* Write permissions never work with read-only images */
1565 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1566 !bdrv_is_writable(bs))
1568 error_setg(errp, "Block node is read-only");
1569 return -EPERM;
1572 /* Check this node */
1573 if (!drv) {
1574 return 0;
1577 if (drv->bdrv_check_perm) {
1578 return drv->bdrv_check_perm(bs, cumulative_perms,
1579 cumulative_shared_perms, errp);
1582 /* Drivers that never have children can omit .bdrv_child_perm() */
1583 if (!drv->bdrv_child_perm) {
1584 assert(QLIST_EMPTY(&bs->children));
1585 return 0;
1588 /* Check all children */
1589 QLIST_FOREACH(c, &bs->children, next) {
1590 uint64_t cur_perm, cur_shared;
1591 bdrv_child_perm(bs, c->bs, c, c->role,
1592 cumulative_perms, cumulative_shared_perms,
1593 &cur_perm, &cur_shared);
1594 ret = bdrv_child_check_perm(c, cur_perm, cur_shared, ignore_children,
1595 errp);
1596 if (ret < 0) {
1597 return ret;
1601 return 0;
1605 * Notifies drivers that after a previous bdrv_check_perm() call, the
1606 * permission update is not performed and any preparations made for it (e.g.
1607 * taken file locks) need to be undone.
1609 * This function recursively notifies all child nodes.
1611 static void bdrv_abort_perm_update(BlockDriverState *bs)
1613 BlockDriver *drv = bs->drv;
1614 BdrvChild *c;
1616 if (!drv) {
1617 return;
1620 if (drv->bdrv_abort_perm_update) {
1621 drv->bdrv_abort_perm_update(bs);
1624 QLIST_FOREACH(c, &bs->children, next) {
1625 bdrv_child_abort_perm_update(c);
1629 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1630 uint64_t cumulative_shared_perms)
1632 BlockDriver *drv = bs->drv;
1633 BdrvChild *c;
1635 if (!drv) {
1636 return;
1639 /* Update this node */
1640 if (drv->bdrv_set_perm) {
1641 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1644 /* Drivers that never have children can omit .bdrv_child_perm() */
1645 if (!drv->bdrv_child_perm) {
1646 assert(QLIST_EMPTY(&bs->children));
1647 return;
1650 /* Update all children */
1651 QLIST_FOREACH(c, &bs->children, next) {
1652 uint64_t cur_perm, cur_shared;
1653 bdrv_child_perm(bs, c->bs, c, c->role,
1654 cumulative_perms, cumulative_shared_perms,
1655 &cur_perm, &cur_shared);
1656 bdrv_child_set_perm(c, cur_perm, cur_shared);
1660 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1661 uint64_t *shared_perm)
1663 BdrvChild *c;
1664 uint64_t cumulative_perms = 0;
1665 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1667 QLIST_FOREACH(c, &bs->parents, next_parent) {
1668 cumulative_perms |= c->perm;
1669 cumulative_shared_perms &= c->shared_perm;
1672 *perm = cumulative_perms;
1673 *shared_perm = cumulative_shared_perms;
1676 static char *bdrv_child_user_desc(BdrvChild *c)
1678 if (c->role->get_parent_desc) {
1679 return c->role->get_parent_desc(c);
1682 return g_strdup("another user");
1685 char *bdrv_perm_names(uint64_t perm)
1687 struct perm_name {
1688 uint64_t perm;
1689 const char *name;
1690 } permissions[] = {
1691 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1692 { BLK_PERM_WRITE, "write" },
1693 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1694 { BLK_PERM_RESIZE, "resize" },
1695 { BLK_PERM_GRAPH_MOD, "change children" },
1696 { 0, NULL }
1699 char *result = g_strdup("");
1700 struct perm_name *p;
1702 for (p = permissions; p->name; p++) {
1703 if (perm & p->perm) {
1704 char *old = result;
1705 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1706 g_free(old);
1710 return result;
1714 * Checks whether a new reference to @bs can be added if the new user requires
1715 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1716 * set, the BdrvChild objects in this list are ignored in the calculations;
1717 * this allows checking permission updates for an existing reference.
1719 * Needs to be followed by a call to either bdrv_set_perm() or
1720 * bdrv_abort_perm_update(). */
1721 static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm,
1722 uint64_t new_shared_perm,
1723 GSList *ignore_children, Error **errp)
1725 BdrvChild *c;
1726 uint64_t cumulative_perms = new_used_perm;
1727 uint64_t cumulative_shared_perms = new_shared_perm;
1729 /* There is no reason why anyone couldn't tolerate write_unchanged */
1730 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1732 QLIST_FOREACH(c, &bs->parents, next_parent) {
1733 if (g_slist_find(ignore_children, c)) {
1734 continue;
1737 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1738 char *user = bdrv_child_user_desc(c);
1739 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1740 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1741 "allow '%s' on %s",
1742 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1743 g_free(user);
1744 g_free(perm_names);
1745 return -EPERM;
1748 if ((c->perm & new_shared_perm) != c->perm) {
1749 char *user = bdrv_child_user_desc(c);
1750 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1751 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1752 "'%s' on %s",
1753 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1754 g_free(user);
1755 g_free(perm_names);
1756 return -EPERM;
1759 cumulative_perms |= c->perm;
1760 cumulative_shared_perms &= c->shared_perm;
1763 return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms,
1764 ignore_children, errp);
1767 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1768 * bdrv_child_abort_perm_update(). */
1769 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1770 GSList *ignore_children, Error **errp)
1772 int ret;
1774 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1775 ret = bdrv_check_update_perm(c->bs, perm, shared, ignore_children, errp);
1776 g_slist_free(ignore_children);
1778 return ret;
1781 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1783 uint64_t cumulative_perms, cumulative_shared_perms;
1785 c->perm = perm;
1786 c->shared_perm = shared;
1788 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1789 &cumulative_shared_perms);
1790 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1793 static void bdrv_child_abort_perm_update(BdrvChild *c)
1795 bdrv_abort_perm_update(c->bs);
1798 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1799 Error **errp)
1801 int ret;
1803 ret = bdrv_child_check_perm(c, perm, shared, NULL, errp);
1804 if (ret < 0) {
1805 bdrv_child_abort_perm_update(c);
1806 return ret;
1809 bdrv_child_set_perm(c, perm, shared);
1811 return 0;
1814 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1815 | BLK_PERM_WRITE \
1816 | BLK_PERM_WRITE_UNCHANGED \
1817 | BLK_PERM_RESIZE)
1818 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1820 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1821 const BdrvChildRole *role,
1822 uint64_t perm, uint64_t shared,
1823 uint64_t *nperm, uint64_t *nshared)
1825 if (c == NULL) {
1826 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1827 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1828 return;
1831 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1832 (c->perm & DEFAULT_PERM_UNCHANGED);
1833 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1834 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1837 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1838 const BdrvChildRole *role,
1839 uint64_t perm, uint64_t shared,
1840 uint64_t *nperm, uint64_t *nshared)
1842 bool backing = (role == &child_backing);
1843 assert(role == &child_backing || role == &child_file);
1845 if (!backing) {
1846 /* Apart from the modifications below, the same permissions are
1847 * forwarded and left alone as for filters */
1848 bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared);
1850 /* Format drivers may touch metadata even if the guest doesn't write */
1851 if (bdrv_is_writable(bs)) {
1852 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1855 /* bs->file always needs to be consistent because of the metadata. We
1856 * can never allow other users to resize or write to it. */
1857 perm |= BLK_PERM_CONSISTENT_READ;
1858 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1859 } else {
1860 /* We want consistent read from backing files if the parent needs it.
1861 * No other operations are performed on backing files. */
1862 perm &= BLK_PERM_CONSISTENT_READ;
1864 /* If the parent can deal with changing data, we're okay with a
1865 * writable and resizable backing file. */
1866 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1867 if (shared & BLK_PERM_WRITE) {
1868 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1869 } else {
1870 shared = 0;
1873 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1874 BLK_PERM_WRITE_UNCHANGED;
1877 if (bs->open_flags & BDRV_O_INACTIVE) {
1878 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1881 *nperm = perm;
1882 *nshared = shared;
1885 static void bdrv_replace_child_noperm(BdrvChild *child,
1886 BlockDriverState *new_bs)
1888 BlockDriverState *old_bs = child->bs;
1890 if (old_bs && new_bs) {
1891 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1893 if (old_bs) {
1894 if (old_bs->quiesce_counter && child->role->drained_end) {
1895 child->role->drained_end(child);
1897 if (child->role->detach) {
1898 child->role->detach(child);
1900 QLIST_REMOVE(child, next_parent);
1903 child->bs = new_bs;
1905 if (new_bs) {
1906 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1907 if (new_bs->quiesce_counter && child->role->drained_begin) {
1908 child->role->drained_begin(child);
1911 if (child->role->attach) {
1912 child->role->attach(child);
1918 * Updates @child to change its reference to point to @new_bs, including
1919 * checking and applying the necessary permisson updates both to the old node
1920 * and to @new_bs.
1922 * NULL is passed as @new_bs for removing the reference before freeing @child.
1924 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1925 * function uses bdrv_set_perm() to update the permissions according to the new
1926 * reference that @new_bs gets.
1928 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1930 BlockDriverState *old_bs = child->bs;
1931 uint64_t perm, shared_perm;
1933 if (old_bs) {
1934 /* Update permissions for old node. This is guaranteed to succeed
1935 * because we're just taking a parent away, so we're loosening
1936 * restrictions. */
1937 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1938 bdrv_check_perm(old_bs, perm, shared_perm, NULL, &error_abort);
1939 bdrv_set_perm(old_bs, perm, shared_perm);
1942 bdrv_replace_child_noperm(child, new_bs);
1944 if (new_bs) {
1945 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
1946 bdrv_set_perm(new_bs, perm, shared_perm);
1950 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1951 const char *child_name,
1952 const BdrvChildRole *child_role,
1953 uint64_t perm, uint64_t shared_perm,
1954 void *opaque, Error **errp)
1956 BdrvChild *child;
1957 int ret;
1959 ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp);
1960 if (ret < 0) {
1961 bdrv_abort_perm_update(child_bs);
1962 return NULL;
1965 child = g_new(BdrvChild, 1);
1966 *child = (BdrvChild) {
1967 .bs = NULL,
1968 .name = g_strdup(child_name),
1969 .role = child_role,
1970 .perm = perm,
1971 .shared_perm = shared_perm,
1972 .opaque = opaque,
1975 /* This performs the matching bdrv_set_perm() for the above check. */
1976 bdrv_replace_child(child, child_bs);
1978 return child;
1981 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1982 BlockDriverState *child_bs,
1983 const char *child_name,
1984 const BdrvChildRole *child_role,
1985 Error **errp)
1987 BdrvChild *child;
1988 uint64_t perm, shared_perm;
1990 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
1992 assert(parent_bs->drv);
1993 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
1994 bdrv_child_perm(parent_bs, child_bs, NULL, child_role,
1995 perm, shared_perm, &perm, &shared_perm);
1997 child = bdrv_root_attach_child(child_bs, child_name, child_role,
1998 perm, shared_perm, parent_bs, errp);
1999 if (child == NULL) {
2000 return NULL;
2003 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2004 return child;
2007 static void bdrv_detach_child(BdrvChild *child)
2009 if (child->next.le_prev) {
2010 QLIST_REMOVE(child, next);
2011 child->next.le_prev = NULL;
2014 bdrv_replace_child(child, NULL);
2016 g_free(child->name);
2017 g_free(child);
2020 void bdrv_root_unref_child(BdrvChild *child)
2022 BlockDriverState *child_bs;
2024 child_bs = child->bs;
2025 bdrv_detach_child(child);
2026 bdrv_unref(child_bs);
2029 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2031 if (child == NULL) {
2032 return;
2035 if (child->bs->inherits_from == parent) {
2036 BdrvChild *c;
2038 /* Remove inherits_from only when the last reference between parent and
2039 * child->bs goes away. */
2040 QLIST_FOREACH(c, &parent->children, next) {
2041 if (c != child && c->bs == child->bs) {
2042 break;
2045 if (c == NULL) {
2046 child->bs->inherits_from = NULL;
2050 bdrv_root_unref_child(child);
2054 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2056 BdrvChild *c;
2057 QLIST_FOREACH(c, &bs->parents, next_parent) {
2058 if (c->role->change_media) {
2059 c->role->change_media(c, load);
2064 static void bdrv_parent_cb_resize(BlockDriverState *bs)
2066 BdrvChild *c;
2067 QLIST_FOREACH(c, &bs->parents, next_parent) {
2068 if (c->role->resize) {
2069 c->role->resize(c);
2075 * Sets the backing file link of a BDS. A new reference is created; callers
2076 * which don't need their own reference any more must call bdrv_unref().
2078 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2079 Error **errp)
2081 if (backing_hd) {
2082 bdrv_ref(backing_hd);
2085 if (bs->backing) {
2086 bdrv_unref_child(bs, bs->backing);
2089 if (!backing_hd) {
2090 bs->backing = NULL;
2091 goto out;
2094 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2095 errp);
2096 if (!bs->backing) {
2097 bdrv_unref(backing_hd);
2100 bdrv_refresh_filename(bs);
2102 out:
2103 bdrv_refresh_limits(bs, NULL);
2107 * Opens the backing file for a BlockDriverState if not yet open
2109 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2110 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2111 * itself, all options starting with "${bdref_key}." are considered part of the
2112 * BlockdevRef.
2114 * TODO Can this be unified with bdrv_open_image()?
2116 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2117 const char *bdref_key, Error **errp)
2119 char *backing_filename = g_malloc0(PATH_MAX);
2120 char *bdref_key_dot;
2121 const char *reference = NULL;
2122 int ret = 0;
2123 BlockDriverState *backing_hd;
2124 QDict *options;
2125 QDict *tmp_parent_options = NULL;
2126 Error *local_err = NULL;
2128 if (bs->backing != NULL) {
2129 goto free_exit;
2132 /* NULL means an empty set of options */
2133 if (parent_options == NULL) {
2134 tmp_parent_options = qdict_new();
2135 parent_options = tmp_parent_options;
2138 bs->open_flags &= ~BDRV_O_NO_BACKING;
2140 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2141 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2142 g_free(bdref_key_dot);
2145 * Caution: while qdict_get_try_str() is fine, getting non-string
2146 * types would require more care. When @parent_options come from
2147 * -blockdev or blockdev_add, its members are typed according to
2148 * the QAPI schema, but when they come from -drive, they're all
2149 * QString.
2151 reference = qdict_get_try_str(parent_options, bdref_key);
2152 if (reference || qdict_haskey(options, "file.filename")) {
2153 backing_filename[0] = '\0';
2154 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2155 QDECREF(options);
2156 goto free_exit;
2157 } else {
2158 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2159 &local_err);
2160 if (local_err) {
2161 ret = -EINVAL;
2162 error_propagate(errp, local_err);
2163 QDECREF(options);
2164 goto free_exit;
2168 if (!bs->drv || !bs->drv->supports_backing) {
2169 ret = -EINVAL;
2170 error_setg(errp, "Driver doesn't support backing files");
2171 QDECREF(options);
2172 goto free_exit;
2175 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2176 qdict_put_str(options, "driver", bs->backing_format);
2179 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2180 reference, options, 0, bs, &child_backing,
2181 errp);
2182 if (!backing_hd) {
2183 bs->open_flags |= BDRV_O_NO_BACKING;
2184 error_prepend(errp, "Could not open backing file: ");
2185 ret = -EINVAL;
2186 goto free_exit;
2189 /* Hook up the backing file link; drop our reference, bs owns the
2190 * backing_hd reference now */
2191 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2192 bdrv_unref(backing_hd);
2193 if (local_err) {
2194 error_propagate(errp, local_err);
2195 ret = -EINVAL;
2196 goto free_exit;
2199 qdict_del(parent_options, bdref_key);
2201 free_exit:
2202 g_free(backing_filename);
2203 QDECREF(tmp_parent_options);
2204 return ret;
2207 static BlockDriverState *
2208 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2209 BlockDriverState *parent, const BdrvChildRole *child_role,
2210 bool allow_none, Error **errp)
2212 BlockDriverState *bs = NULL;
2213 QDict *image_options;
2214 char *bdref_key_dot;
2215 const char *reference;
2217 assert(child_role != NULL);
2219 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2220 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2221 g_free(bdref_key_dot);
2224 * Caution: while qdict_get_try_str() is fine, getting non-string
2225 * types would require more care. When @options come from
2226 * -blockdev or blockdev_add, its members are typed according to
2227 * the QAPI schema, but when they come from -drive, they're all
2228 * QString.
2230 reference = qdict_get_try_str(options, bdref_key);
2231 if (!filename && !reference && !qdict_size(image_options)) {
2232 if (!allow_none) {
2233 error_setg(errp, "A block device must be specified for \"%s\"",
2234 bdref_key);
2236 QDECREF(image_options);
2237 goto done;
2240 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2241 parent, child_role, errp);
2242 if (!bs) {
2243 goto done;
2246 done:
2247 qdict_del(options, bdref_key);
2248 return bs;
2252 * Opens a disk image whose options are given as BlockdevRef in another block
2253 * device's options.
2255 * If allow_none is true, no image will be opened if filename is false and no
2256 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2258 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2259 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2260 * itself, all options starting with "${bdref_key}." are considered part of the
2261 * BlockdevRef.
2263 * The BlockdevRef will be removed from the options QDict.
2265 BdrvChild *bdrv_open_child(const char *filename,
2266 QDict *options, const char *bdref_key,
2267 BlockDriverState *parent,
2268 const BdrvChildRole *child_role,
2269 bool allow_none, Error **errp)
2271 BdrvChild *c;
2272 BlockDriverState *bs;
2274 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2275 allow_none, errp);
2276 if (bs == NULL) {
2277 return NULL;
2280 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2281 if (!c) {
2282 bdrv_unref(bs);
2283 return NULL;
2286 return c;
2289 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2290 int flags,
2291 QDict *snapshot_options,
2292 Error **errp)
2294 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2295 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2296 int64_t total_size;
2297 QemuOpts *opts = NULL;
2298 BlockDriverState *bs_snapshot = NULL;
2299 Error *local_err = NULL;
2300 int ret;
2302 /* if snapshot, we create a temporary backing file and open it
2303 instead of opening 'filename' directly */
2305 /* Get the required size from the image */
2306 total_size = bdrv_getlength(bs);
2307 if (total_size < 0) {
2308 error_setg_errno(errp, -total_size, "Could not get image size");
2309 goto out;
2312 /* Create the temporary image */
2313 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2314 if (ret < 0) {
2315 error_setg_errno(errp, -ret, "Could not get temporary filename");
2316 goto out;
2319 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2320 &error_abort);
2321 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2322 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2323 qemu_opts_del(opts);
2324 if (ret < 0) {
2325 error_prepend(errp, "Could not create temporary overlay '%s': ",
2326 tmp_filename);
2327 goto out;
2330 /* Prepare options QDict for the temporary file */
2331 qdict_put_str(snapshot_options, "file.driver", "file");
2332 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2333 qdict_put_str(snapshot_options, "driver", "qcow2");
2335 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2336 snapshot_options = NULL;
2337 if (!bs_snapshot) {
2338 goto out;
2341 /* bdrv_append() consumes a strong reference to bs_snapshot
2342 * (i.e. it will call bdrv_unref() on it) even on error, so in
2343 * order to be able to return one, we have to increase
2344 * bs_snapshot's refcount here */
2345 bdrv_ref(bs_snapshot);
2346 bdrv_append(bs_snapshot, bs, &local_err);
2347 if (local_err) {
2348 error_propagate(errp, local_err);
2349 bs_snapshot = NULL;
2350 goto out;
2353 out:
2354 QDECREF(snapshot_options);
2355 g_free(tmp_filename);
2356 return bs_snapshot;
2360 * Opens a disk image (raw, qcow2, vmdk, ...)
2362 * options is a QDict of options to pass to the block drivers, or NULL for an
2363 * empty set of options. The reference to the QDict belongs to the block layer
2364 * after the call (even on failure), so if the caller intends to reuse the
2365 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2367 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2368 * If it is not NULL, the referenced BDS will be reused.
2370 * The reference parameter may be used to specify an existing block device which
2371 * should be opened. If specified, neither options nor a filename may be given,
2372 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2374 static BlockDriverState *bdrv_open_inherit(const char *filename,
2375 const char *reference,
2376 QDict *options, int flags,
2377 BlockDriverState *parent,
2378 const BdrvChildRole *child_role,
2379 Error **errp)
2381 int ret;
2382 BlockBackend *file = NULL;
2383 BlockDriverState *bs;
2384 BlockDriver *drv = NULL;
2385 const char *drvname;
2386 const char *backing;
2387 Error *local_err = NULL;
2388 QDict *snapshot_options = NULL;
2389 int snapshot_flags = 0;
2391 assert(!child_role || !flags);
2392 assert(!child_role == !parent);
2394 if (reference) {
2395 bool options_non_empty = options ? qdict_size(options) : false;
2396 QDECREF(options);
2398 if (filename || options_non_empty) {
2399 error_setg(errp, "Cannot reference an existing block device with "
2400 "additional options or a new filename");
2401 return NULL;
2404 bs = bdrv_lookup_bs(reference, reference, errp);
2405 if (!bs) {
2406 return NULL;
2409 bdrv_ref(bs);
2410 return bs;
2413 bs = bdrv_new();
2415 /* NULL means an empty set of options */
2416 if (options == NULL) {
2417 options = qdict_new();
2420 /* json: syntax counts as explicit options, as if in the QDict */
2421 parse_json_protocol(options, &filename, &local_err);
2422 if (local_err) {
2423 goto fail;
2426 bs->explicit_options = qdict_clone_shallow(options);
2428 if (child_role) {
2429 bs->inherits_from = parent;
2430 child_role->inherit_options(&flags, options,
2431 parent->open_flags, parent->options);
2434 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2435 if (local_err) {
2436 goto fail;
2440 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2441 * Caution: getting a boolean member of @options requires care.
2442 * When @options come from -blockdev or blockdev_add, members are
2443 * typed according to the QAPI schema, but when they come from
2444 * -drive, they're all QString.
2446 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2447 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2448 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2449 } else {
2450 flags &= ~BDRV_O_RDWR;
2453 if (flags & BDRV_O_SNAPSHOT) {
2454 snapshot_options = qdict_new();
2455 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2456 flags, options);
2457 /* Let bdrv_backing_options() override "read-only" */
2458 qdict_del(options, BDRV_OPT_READ_ONLY);
2459 bdrv_backing_options(&flags, options, flags, options);
2462 bs->open_flags = flags;
2463 bs->options = options;
2464 options = qdict_clone_shallow(options);
2466 /* Find the right image format driver */
2467 /* See cautionary note on accessing @options above */
2468 drvname = qdict_get_try_str(options, "driver");
2469 if (drvname) {
2470 drv = bdrv_find_format(drvname);
2471 if (!drv) {
2472 error_setg(errp, "Unknown driver: '%s'", drvname);
2473 goto fail;
2477 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2479 /* See cautionary note on accessing @options above */
2480 backing = qdict_get_try_str(options, "backing");
2481 if (backing && *backing == '\0') {
2482 flags |= BDRV_O_NO_BACKING;
2483 qdict_del(options, "backing");
2486 /* Open image file without format layer. This BlockBackend is only used for
2487 * probing, the block drivers will do their own bdrv_open_child() for the
2488 * same BDS, which is why we put the node name back into options. */
2489 if ((flags & BDRV_O_PROTOCOL) == 0) {
2490 BlockDriverState *file_bs;
2492 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2493 &child_file, true, &local_err);
2494 if (local_err) {
2495 goto fail;
2497 if (file_bs != NULL) {
2498 file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
2499 blk_insert_bs(file, file_bs, &local_err);
2500 bdrv_unref(file_bs);
2501 if (local_err) {
2502 goto fail;
2505 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2509 /* Image format probing */
2510 bs->probed = !drv;
2511 if (!drv && file) {
2512 ret = find_image_format(file, filename, &drv, &local_err);
2513 if (ret < 0) {
2514 goto fail;
2517 * This option update would logically belong in bdrv_fill_options(),
2518 * but we first need to open bs->file for the probing to work, while
2519 * opening bs->file already requires the (mostly) final set of options
2520 * so that cache mode etc. can be inherited.
2522 * Adding the driver later is somewhat ugly, but it's not an option
2523 * that would ever be inherited, so it's correct. We just need to make
2524 * sure to update both bs->options (which has the full effective
2525 * options for bs) and options (which has file.* already removed).
2527 qdict_put_str(bs->options, "driver", drv->format_name);
2528 qdict_put_str(options, "driver", drv->format_name);
2529 } else if (!drv) {
2530 error_setg(errp, "Must specify either driver or file");
2531 goto fail;
2534 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2535 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2536 /* file must be NULL if a protocol BDS is about to be created
2537 * (the inverse results in an error message from bdrv_open_common()) */
2538 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2540 /* Open the image */
2541 ret = bdrv_open_common(bs, file, options, &local_err);
2542 if (ret < 0) {
2543 goto fail;
2546 if (file) {
2547 blk_unref(file);
2548 file = NULL;
2551 /* If there is a backing file, use it */
2552 if ((flags & BDRV_O_NO_BACKING) == 0) {
2553 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2554 if (ret < 0) {
2555 goto close_and_fail;
2559 bdrv_refresh_filename(bs);
2561 /* Check if any unknown options were used */
2562 if (qdict_size(options) != 0) {
2563 const QDictEntry *entry = qdict_first(options);
2564 if (flags & BDRV_O_PROTOCOL) {
2565 error_setg(errp, "Block protocol '%s' doesn't support the option "
2566 "'%s'", drv->format_name, entry->key);
2567 } else {
2568 error_setg(errp,
2569 "Block format '%s' does not support the option '%s'",
2570 drv->format_name, entry->key);
2573 goto close_and_fail;
2576 if (!bdrv_key_required(bs)) {
2577 bdrv_parent_cb_change_media(bs, true);
2578 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
2579 && !runstate_check(RUN_STATE_INMIGRATE)
2580 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
2581 error_setg(errp,
2582 "Guest must be stopped for opening of encrypted image");
2583 goto close_and_fail;
2586 QDECREF(options);
2588 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2589 * temporary snapshot afterwards. */
2590 if (snapshot_flags) {
2591 BlockDriverState *snapshot_bs;
2592 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2593 snapshot_options, &local_err);
2594 snapshot_options = NULL;
2595 if (local_err) {
2596 goto close_and_fail;
2598 /* We are not going to return bs but the overlay on top of it
2599 * (snapshot_bs); thus, we have to drop the strong reference to bs
2600 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2601 * though, because the overlay still has a reference to it. */
2602 bdrv_unref(bs);
2603 bs = snapshot_bs;
2606 return bs;
2608 fail:
2609 blk_unref(file);
2610 if (bs->file != NULL) {
2611 bdrv_unref_child(bs, bs->file);
2613 QDECREF(snapshot_options);
2614 QDECREF(bs->explicit_options);
2615 QDECREF(bs->options);
2616 QDECREF(options);
2617 bs->options = NULL;
2618 bdrv_unref(bs);
2619 error_propagate(errp, local_err);
2620 return NULL;
2622 close_and_fail:
2623 bdrv_unref(bs);
2624 QDECREF(snapshot_options);
2625 QDECREF(options);
2626 error_propagate(errp, local_err);
2627 return NULL;
2630 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2631 QDict *options, int flags, Error **errp)
2633 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2634 NULL, errp);
2637 typedef struct BlockReopenQueueEntry {
2638 bool prepared;
2639 BDRVReopenState state;
2640 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2641 } BlockReopenQueueEntry;
2644 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2645 * reopen of multiple devices.
2647 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2648 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2649 * be created and initialized. This newly created BlockReopenQueue should be
2650 * passed back in for subsequent calls that are intended to be of the same
2651 * atomic 'set'.
2653 * bs is the BlockDriverState to add to the reopen queue.
2655 * options contains the changed options for the associated bs
2656 * (the BlockReopenQueue takes ownership)
2658 * flags contains the open flags for the associated bs
2660 * returns a pointer to bs_queue, which is either the newly allocated
2661 * bs_queue, or the existing bs_queue being used.
2664 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2665 BlockDriverState *bs,
2666 QDict *options,
2667 int flags,
2668 const BdrvChildRole *role,
2669 QDict *parent_options,
2670 int parent_flags)
2672 assert(bs != NULL);
2674 BlockReopenQueueEntry *bs_entry;
2675 BdrvChild *child;
2676 QDict *old_options, *explicit_options;
2678 if (bs_queue == NULL) {
2679 bs_queue = g_new0(BlockReopenQueue, 1);
2680 QSIMPLEQ_INIT(bs_queue);
2683 if (!options) {
2684 options = qdict_new();
2687 /* Check if this BlockDriverState is already in the queue */
2688 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2689 if (bs == bs_entry->state.bs) {
2690 break;
2695 * Precedence of options:
2696 * 1. Explicitly passed in options (highest)
2697 * 2. Set in flags (only for top level)
2698 * 3. Retained from explicitly set options of bs
2699 * 4. Inherited from parent node
2700 * 5. Retained from effective options of bs
2703 if (!parent_options) {
2705 * Any setting represented by flags is always updated. If the
2706 * corresponding QDict option is set, it takes precedence. Otherwise
2707 * the flag is translated into a QDict option. The old setting of bs is
2708 * not considered.
2710 update_options_from_flags(options, flags);
2713 /* Old explicitly set values (don't overwrite by inherited value) */
2714 if (bs_entry) {
2715 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2716 } else {
2717 old_options = qdict_clone_shallow(bs->explicit_options);
2719 bdrv_join_options(bs, options, old_options);
2720 QDECREF(old_options);
2722 explicit_options = qdict_clone_shallow(options);
2724 /* Inherit from parent node */
2725 if (parent_options) {
2726 assert(!flags);
2727 role->inherit_options(&flags, options, parent_flags, parent_options);
2730 /* Old values are used for options that aren't set yet */
2731 old_options = qdict_clone_shallow(bs->options);
2732 bdrv_join_options(bs, options, old_options);
2733 QDECREF(old_options);
2735 /* bdrv_open() masks this flag out */
2736 flags &= ~BDRV_O_PROTOCOL;
2738 QLIST_FOREACH(child, &bs->children, next) {
2739 QDict *new_child_options;
2740 char *child_key_dot;
2742 /* reopen can only change the options of block devices that were
2743 * implicitly created and inherited options. For other (referenced)
2744 * block devices, a syntax like "backing.foo" results in an error. */
2745 if (child->bs->inherits_from != bs) {
2746 continue;
2749 child_key_dot = g_strdup_printf("%s.", child->name);
2750 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2751 g_free(child_key_dot);
2753 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2754 child->role, options, flags);
2757 if (!bs_entry) {
2758 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2759 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2760 } else {
2761 QDECREF(bs_entry->state.options);
2762 QDECREF(bs_entry->state.explicit_options);
2765 bs_entry->state.bs = bs;
2766 bs_entry->state.options = options;
2767 bs_entry->state.explicit_options = explicit_options;
2768 bs_entry->state.flags = flags;
2770 return bs_queue;
2773 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2774 BlockDriverState *bs,
2775 QDict *options, int flags)
2777 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2778 NULL, NULL, 0);
2782 * Reopen multiple BlockDriverStates atomically & transactionally.
2784 * The queue passed in (bs_queue) must have been built up previous
2785 * via bdrv_reopen_queue().
2787 * Reopens all BDS specified in the queue, with the appropriate
2788 * flags. All devices are prepared for reopen, and failure of any
2789 * device will cause all device changes to be abandonded, and intermediate
2790 * data cleaned up.
2792 * If all devices prepare successfully, then the changes are committed
2793 * to all devices.
2796 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2798 int ret = -1;
2799 BlockReopenQueueEntry *bs_entry, *next;
2800 Error *local_err = NULL;
2802 assert(bs_queue != NULL);
2804 aio_context_release(ctx);
2805 bdrv_drain_all_begin();
2806 aio_context_acquire(ctx);
2808 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2809 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2810 error_propagate(errp, local_err);
2811 goto cleanup;
2813 bs_entry->prepared = true;
2816 /* If we reach this point, we have success and just need to apply the
2817 * changes
2819 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2820 bdrv_reopen_commit(&bs_entry->state);
2823 ret = 0;
2825 cleanup:
2826 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2827 if (ret && bs_entry->prepared) {
2828 bdrv_reopen_abort(&bs_entry->state);
2829 } else if (ret) {
2830 QDECREF(bs_entry->state.explicit_options);
2832 QDECREF(bs_entry->state.options);
2833 g_free(bs_entry);
2835 g_free(bs_queue);
2837 bdrv_drain_all_end();
2839 return ret;
2843 /* Reopen a single BlockDriverState with the specified flags. */
2844 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2846 int ret = -1;
2847 Error *local_err = NULL;
2848 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2850 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2851 if (local_err != NULL) {
2852 error_propagate(errp, local_err);
2854 return ret;
2859 * Prepares a BlockDriverState for reopen. All changes are staged in the
2860 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2861 * the block driver layer .bdrv_reopen_prepare()
2863 * bs is the BlockDriverState to reopen
2864 * flags are the new open flags
2865 * queue is the reopen queue
2867 * Returns 0 on success, non-zero on error. On error errp will be set
2868 * as well.
2870 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2871 * It is the responsibility of the caller to then call the abort() or
2872 * commit() for any other BDS that have been left in a prepare() state
2875 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2876 Error **errp)
2878 int ret = -1;
2879 Error *local_err = NULL;
2880 BlockDriver *drv;
2881 QemuOpts *opts;
2882 const char *value;
2883 bool read_only;
2885 assert(reopen_state != NULL);
2886 assert(reopen_state->bs->drv != NULL);
2887 drv = reopen_state->bs->drv;
2889 /* Process generic block layer options */
2890 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2891 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2892 if (local_err) {
2893 error_propagate(errp, local_err);
2894 ret = -EINVAL;
2895 goto error;
2898 update_flags_from_options(&reopen_state->flags, opts);
2900 /* node-name and driver must be unchanged. Put them back into the QDict, so
2901 * that they are checked at the end of this function. */
2902 value = qemu_opt_get(opts, "node-name");
2903 if (value) {
2904 qdict_put_str(reopen_state->options, "node-name", value);
2907 value = qemu_opt_get(opts, "driver");
2908 if (value) {
2909 qdict_put_str(reopen_state->options, "driver", value);
2912 /* If we are to stay read-only, do not allow permission change
2913 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
2914 * not set, or if the BDS still has copy_on_read enabled */
2915 read_only = !(reopen_state->flags & BDRV_O_RDWR);
2916 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, &local_err);
2917 if (local_err) {
2918 error_propagate(errp, local_err);
2919 goto error;
2923 ret = bdrv_flush(reopen_state->bs);
2924 if (ret) {
2925 error_setg_errno(errp, -ret, "Error flushing drive");
2926 goto error;
2929 if (drv->bdrv_reopen_prepare) {
2930 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2931 if (ret) {
2932 if (local_err != NULL) {
2933 error_propagate(errp, local_err);
2934 } else {
2935 error_setg(errp, "failed while preparing to reopen image '%s'",
2936 reopen_state->bs->filename);
2938 goto error;
2940 } else {
2941 /* It is currently mandatory to have a bdrv_reopen_prepare()
2942 * handler for each supported drv. */
2943 error_setg(errp, "Block format '%s' used by node '%s' "
2944 "does not support reopening files", drv->format_name,
2945 bdrv_get_device_or_node_name(reopen_state->bs));
2946 ret = -1;
2947 goto error;
2950 /* Options that are not handled are only okay if they are unchanged
2951 * compared to the old state. It is expected that some options are only
2952 * used for the initial open, but not reopen (e.g. filename) */
2953 if (qdict_size(reopen_state->options)) {
2954 const QDictEntry *entry = qdict_first(reopen_state->options);
2956 do {
2957 QString *new_obj = qobject_to_qstring(entry->value);
2958 const char *new = qstring_get_str(new_obj);
2960 * Caution: while qdict_get_try_str() is fine, getting
2961 * non-string types would require more care. When
2962 * bs->options come from -blockdev or blockdev_add, its
2963 * members are typed according to the QAPI schema, but
2964 * when they come from -drive, they're all QString.
2966 const char *old = qdict_get_try_str(reopen_state->bs->options,
2967 entry->key);
2969 if (!old || strcmp(new, old)) {
2970 error_setg(errp, "Cannot change the option '%s'", entry->key);
2971 ret = -EINVAL;
2972 goto error;
2974 } while ((entry = qdict_next(reopen_state->options, entry)));
2977 ret = 0;
2979 error:
2980 qemu_opts_del(opts);
2981 return ret;
2985 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2986 * makes them final by swapping the staging BlockDriverState contents into
2987 * the active BlockDriverState contents.
2989 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2991 BlockDriver *drv;
2993 assert(reopen_state != NULL);
2994 drv = reopen_state->bs->drv;
2995 assert(drv != NULL);
2997 /* If there are any driver level actions to take */
2998 if (drv->bdrv_reopen_commit) {
2999 drv->bdrv_reopen_commit(reopen_state);
3002 /* set BDS specific flags now */
3003 QDECREF(reopen_state->bs->explicit_options);
3005 reopen_state->bs->explicit_options = reopen_state->explicit_options;
3006 reopen_state->bs->open_flags = reopen_state->flags;
3007 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3009 bdrv_refresh_limits(reopen_state->bs, NULL);
3013 * Abort the reopen, and delete and free the staged changes in
3014 * reopen_state
3016 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3018 BlockDriver *drv;
3020 assert(reopen_state != NULL);
3021 drv = reopen_state->bs->drv;
3022 assert(drv != NULL);
3024 if (drv->bdrv_reopen_abort) {
3025 drv->bdrv_reopen_abort(reopen_state);
3028 QDECREF(reopen_state->explicit_options);
3032 static void bdrv_close(BlockDriverState *bs)
3034 BdrvAioNotifier *ban, *ban_next;
3036 assert(!bs->job);
3037 assert(!bs->refcnt);
3039 bdrv_drained_begin(bs); /* complete I/O */
3040 bdrv_flush(bs);
3041 bdrv_drain(bs); /* in case flush left pending I/O */
3043 bdrv_release_named_dirty_bitmaps(bs);
3044 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3046 if (bs->drv) {
3047 BdrvChild *child, *next;
3049 bs->drv->bdrv_close(bs);
3050 bs->drv = NULL;
3052 bdrv_set_backing_hd(bs, NULL, &error_abort);
3054 if (bs->file != NULL) {
3055 bdrv_unref_child(bs, bs->file);
3056 bs->file = NULL;
3059 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3060 /* TODO Remove bdrv_unref() from drivers' close function and use
3061 * bdrv_unref_child() here */
3062 if (child->bs->inherits_from == bs) {
3063 child->bs->inherits_from = NULL;
3065 bdrv_detach_child(child);
3068 g_free(bs->opaque);
3069 bs->opaque = NULL;
3070 atomic_set(&bs->copy_on_read, 0);
3071 bs->backing_file[0] = '\0';
3072 bs->backing_format[0] = '\0';
3073 bs->total_sectors = 0;
3074 bs->encrypted = false;
3075 bs->valid_key = false;
3076 bs->sg = false;
3077 QDECREF(bs->options);
3078 QDECREF(bs->explicit_options);
3079 bs->options = NULL;
3080 QDECREF(bs->full_open_options);
3081 bs->full_open_options = NULL;
3084 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3085 g_free(ban);
3087 QLIST_INIT(&bs->aio_notifiers);
3088 bdrv_drained_end(bs);
3091 void bdrv_close_all(void)
3093 block_job_cancel_sync_all();
3094 nbd_export_close_all();
3096 /* Drop references from requests still in flight, such as canceled block
3097 * jobs whose AIO context has not been polled yet */
3098 bdrv_drain_all();
3100 blk_remove_all_bs();
3101 blockdev_close_all_bdrv_states();
3103 assert(QTAILQ_EMPTY(&all_bdrv_states));
3106 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3108 BdrvChild *to_c;
3110 if (c->role->stay_at_node) {
3111 return false;
3114 if (c->role == &child_backing) {
3115 /* If @from is a backing file of @to, ignore the child to avoid
3116 * creating a loop. We only want to change the pointer of other
3117 * parents. */
3118 QLIST_FOREACH(to_c, &to->children, next) {
3119 if (to_c == c) {
3120 break;
3123 if (to_c) {
3124 return false;
3128 return true;
3131 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3132 Error **errp)
3134 BdrvChild *c, *next;
3135 GSList *list = NULL, *p;
3136 uint64_t old_perm, old_shared;
3137 uint64_t perm = 0, shared = BLK_PERM_ALL;
3138 int ret;
3140 assert(!atomic_read(&from->in_flight));
3141 assert(!atomic_read(&to->in_flight));
3143 /* Make sure that @from doesn't go away until we have successfully attached
3144 * all of its parents to @to. */
3145 bdrv_ref(from);
3147 /* Put all parents into @list and calculate their cumulative permissions */
3148 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3149 if (!should_update_child(c, to)) {
3150 continue;
3152 list = g_slist_prepend(list, c);
3153 perm |= c->perm;
3154 shared &= c->shared_perm;
3157 /* Check whether the required permissions can be granted on @to, ignoring
3158 * all BdrvChild in @list so that they can't block themselves. */
3159 ret = bdrv_check_update_perm(to, perm, shared, list, errp);
3160 if (ret < 0) {
3161 bdrv_abort_perm_update(to);
3162 goto out;
3165 /* Now actually perform the change. We performed the permission check for
3166 * all elements of @list at once, so set the permissions all at once at the
3167 * very end. */
3168 for (p = list; p != NULL; p = p->next) {
3169 c = p->data;
3171 bdrv_ref(to);
3172 bdrv_replace_child_noperm(c, to);
3173 bdrv_unref(from);
3176 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3177 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3179 out:
3180 g_slist_free(list);
3181 bdrv_unref(from);
3185 * Add new bs contents at the top of an image chain while the chain is
3186 * live, while keeping required fields on the top layer.
3188 * This will modify the BlockDriverState fields, and swap contents
3189 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3191 * bs_new must not be attached to a BlockBackend.
3193 * This function does not create any image files.
3195 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3196 * that's what the callers commonly need. bs_new will be referenced by the old
3197 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3198 * reference of its own, it must call bdrv_ref().
3200 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3201 Error **errp)
3203 Error *local_err = NULL;
3205 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3206 if (local_err) {
3207 error_propagate(errp, local_err);
3208 goto out;
3211 bdrv_replace_node(bs_top, bs_new, &local_err);
3212 if (local_err) {
3213 error_propagate(errp, local_err);
3214 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3215 goto out;
3218 /* bs_new is now referenced by its new parents, we don't need the
3219 * additional reference any more. */
3220 out:
3221 bdrv_unref(bs_new);
3224 static void bdrv_delete(BlockDriverState *bs)
3226 assert(!bs->job);
3227 assert(bdrv_op_blocker_is_empty(bs));
3228 assert(!bs->refcnt);
3230 bdrv_close(bs);
3232 /* remove from list, if necessary */
3233 if (bs->node_name[0] != '\0') {
3234 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3236 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3238 g_free(bs);
3242 * Run consistency checks on an image
3244 * Returns 0 if the check could be completed (it doesn't mean that the image is
3245 * free of errors) or -errno when an internal error occurred. The results of the
3246 * check are stored in res.
3248 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3250 if (bs->drv == NULL) {
3251 return -ENOMEDIUM;
3253 if (bs->drv->bdrv_check == NULL) {
3254 return -ENOTSUP;
3257 memset(res, 0, sizeof(*res));
3258 return bs->drv->bdrv_check(bs, res, fix);
3262 * Return values:
3263 * 0 - success
3264 * -EINVAL - backing format specified, but no file
3265 * -ENOSPC - can't update the backing file because no space is left in the
3266 * image file header
3267 * -ENOTSUP - format driver doesn't support changing the backing file
3269 int bdrv_change_backing_file(BlockDriverState *bs,
3270 const char *backing_file, const char *backing_fmt)
3272 BlockDriver *drv = bs->drv;
3273 int ret;
3275 /* Backing file format doesn't make sense without a backing file */
3276 if (backing_fmt && !backing_file) {
3277 return -EINVAL;
3280 if (drv->bdrv_change_backing_file != NULL) {
3281 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3282 } else {
3283 ret = -ENOTSUP;
3286 if (ret == 0) {
3287 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3288 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3290 return ret;
3294 * Finds the image layer in the chain that has 'bs' as its backing file.
3296 * active is the current topmost image.
3298 * Returns NULL if bs is not found in active's image chain,
3299 * or if active == bs.
3301 * Returns the bottommost base image if bs == NULL.
3303 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3304 BlockDriverState *bs)
3306 while (active && bs != backing_bs(active)) {
3307 active = backing_bs(active);
3310 return active;
3313 /* Given a BDS, searches for the base layer. */
3314 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3316 return bdrv_find_overlay(bs, NULL);
3320 * Drops images above 'base' up to and including 'top', and sets the image
3321 * above 'top' to have base as its backing file.
3323 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3324 * information in 'bs' can be properly updated.
3326 * E.g., this will convert the following chain:
3327 * bottom <- base <- intermediate <- top <- active
3329 * to
3331 * bottom <- base <- active
3333 * It is allowed for bottom==base, in which case it converts:
3335 * base <- intermediate <- top <- active
3337 * to
3339 * base <- active
3341 * If backing_file_str is non-NULL, it will be used when modifying top's
3342 * overlay image metadata.
3344 * Error conditions:
3345 * if active == top, that is considered an error
3348 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
3349 BlockDriverState *base, const char *backing_file_str)
3351 BlockDriverState *new_top_bs = NULL;
3352 Error *local_err = NULL;
3353 int ret = -EIO;
3355 if (!top->drv || !base->drv) {
3356 goto exit;
3359 new_top_bs = bdrv_find_overlay(active, top);
3361 if (new_top_bs == NULL) {
3362 /* we could not find the image above 'top', this is an error */
3363 goto exit;
3366 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3367 * to do, no intermediate images */
3368 if (backing_bs(new_top_bs) == base) {
3369 ret = 0;
3370 goto exit;
3373 /* Make sure that base is in the backing chain of top */
3374 if (!bdrv_chain_contains(top, base)) {
3375 goto exit;
3378 /* success - we can delete the intermediate states, and link top->base */
3379 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3380 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
3381 base->drv ? base->drv->format_name : "");
3382 if (ret) {
3383 goto exit;
3386 bdrv_set_backing_hd(new_top_bs, base, &local_err);
3387 if (local_err) {
3388 ret = -EPERM;
3389 error_report_err(local_err);
3390 goto exit;
3393 ret = 0;
3394 exit:
3395 return ret;
3399 * Truncate file to 'offset' bytes (needed only for file protocols)
3401 int bdrv_truncate(BdrvChild *child, int64_t offset, Error **errp)
3403 BlockDriverState *bs = child->bs;
3404 BlockDriver *drv = bs->drv;
3405 int ret;
3407 assert(child->perm & BLK_PERM_RESIZE);
3409 if (!drv) {
3410 error_setg(errp, "No medium inserted");
3411 return -ENOMEDIUM;
3413 if (!drv->bdrv_truncate) {
3414 error_setg(errp, "Image format driver does not support resize");
3415 return -ENOTSUP;
3417 if (bs->read_only) {
3418 error_setg(errp, "Image is read-only");
3419 return -EACCES;
3422 assert(!(bs->open_flags & BDRV_O_INACTIVE));
3424 ret = drv->bdrv_truncate(bs, offset, errp);
3425 if (ret == 0) {
3426 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3427 bdrv_dirty_bitmap_truncate(bs);
3428 bdrv_parent_cb_resize(bs);
3429 atomic_inc(&bs->write_gen);
3431 return ret;
3435 * Length of a allocated file in bytes. Sparse files are counted by actual
3436 * allocated space. Return < 0 if error or unknown.
3438 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3440 BlockDriver *drv = bs->drv;
3441 if (!drv) {
3442 return -ENOMEDIUM;
3444 if (drv->bdrv_get_allocated_file_size) {
3445 return drv->bdrv_get_allocated_file_size(bs);
3447 if (bs->file) {
3448 return bdrv_get_allocated_file_size(bs->file->bs);
3450 return -ENOTSUP;
3454 * Return number of sectors on success, -errno on error.
3456 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3458 BlockDriver *drv = bs->drv;
3460 if (!drv)
3461 return -ENOMEDIUM;
3463 if (drv->has_variable_length) {
3464 int ret = refresh_total_sectors(bs, bs->total_sectors);
3465 if (ret < 0) {
3466 return ret;
3469 return bs->total_sectors;
3473 * Return length in bytes on success, -errno on error.
3474 * The length is always a multiple of BDRV_SECTOR_SIZE.
3476 int64_t bdrv_getlength(BlockDriverState *bs)
3478 int64_t ret = bdrv_nb_sectors(bs);
3480 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3481 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3484 /* return 0 as number of sectors if no device present or error */
3485 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3487 int64_t nb_sectors = bdrv_nb_sectors(bs);
3489 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3492 bool bdrv_is_sg(BlockDriverState *bs)
3494 return bs->sg;
3497 bool bdrv_is_encrypted(BlockDriverState *bs)
3499 if (bs->backing && bs->backing->bs->encrypted) {
3500 return true;
3502 return bs->encrypted;
3505 bool bdrv_key_required(BlockDriverState *bs)
3507 BdrvChild *backing = bs->backing;
3509 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
3510 return true;
3512 return (bs->encrypted && !bs->valid_key);
3515 int bdrv_set_key(BlockDriverState *bs, const char *key)
3517 int ret;
3518 if (bs->backing && bs->backing->bs->encrypted) {
3519 ret = bdrv_set_key(bs->backing->bs, key);
3520 if (ret < 0)
3521 return ret;
3522 if (!bs->encrypted)
3523 return 0;
3525 if (!bs->encrypted) {
3526 return -EINVAL;
3527 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3528 return -ENOMEDIUM;
3530 ret = bs->drv->bdrv_set_key(bs, key);
3531 if (ret < 0) {
3532 bs->valid_key = false;
3533 } else if (!bs->valid_key) {
3534 /* call the change callback now, we skipped it on open */
3535 bs->valid_key = true;
3536 bdrv_parent_cb_change_media(bs, true);
3538 return ret;
3542 * Provide an encryption key for @bs.
3543 * If @key is non-null:
3544 * If @bs is not encrypted, fail.
3545 * Else if the key is invalid, fail.
3546 * Else set @bs's key to @key, replacing the existing key, if any.
3547 * If @key is null:
3548 * If @bs is encrypted and still lacks a key, fail.
3549 * Else do nothing.
3550 * On failure, store an error object through @errp if non-null.
3552 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
3554 if (key) {
3555 if (!bdrv_is_encrypted(bs)) {
3556 error_setg(errp, "Node '%s' is not encrypted",
3557 bdrv_get_device_or_node_name(bs));
3558 } else if (bdrv_set_key(bs, key) < 0) {
3559 error_setg(errp, QERR_INVALID_PASSWORD);
3561 } else {
3562 if (bdrv_key_required(bs)) {
3563 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
3564 "'%s' (%s) is encrypted",
3565 bdrv_get_device_or_node_name(bs),
3566 bdrv_get_encrypted_filename(bs));
3571 const char *bdrv_get_format_name(BlockDriverState *bs)
3573 return bs->drv ? bs->drv->format_name : NULL;
3576 static int qsort_strcmp(const void *a, const void *b)
3578 return strcmp(*(char *const *)a, *(char *const *)b);
3581 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3582 void *opaque)
3584 BlockDriver *drv;
3585 int count = 0;
3586 int i;
3587 const char **formats = NULL;
3589 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3590 if (drv->format_name) {
3591 bool found = false;
3592 int i = count;
3593 while (formats && i && !found) {
3594 found = !strcmp(formats[--i], drv->format_name);
3597 if (!found) {
3598 formats = g_renew(const char *, formats, count + 1);
3599 formats[count++] = drv->format_name;
3604 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3605 const char *format_name = block_driver_modules[i].format_name;
3607 if (format_name) {
3608 bool found = false;
3609 int j = count;
3611 while (formats && j && !found) {
3612 found = !strcmp(formats[--j], format_name);
3615 if (!found) {
3616 formats = g_renew(const char *, formats, count + 1);
3617 formats[count++] = format_name;
3622 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3624 for (i = 0; i < count; i++) {
3625 it(opaque, formats[i]);
3628 g_free(formats);
3631 /* This function is to find a node in the bs graph */
3632 BlockDriverState *bdrv_find_node(const char *node_name)
3634 BlockDriverState *bs;
3636 assert(node_name);
3638 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3639 if (!strcmp(node_name, bs->node_name)) {
3640 return bs;
3643 return NULL;
3646 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3647 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3649 BlockDeviceInfoList *list, *entry;
3650 BlockDriverState *bs;
3652 list = NULL;
3653 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3654 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3655 if (!info) {
3656 qapi_free_BlockDeviceInfoList(list);
3657 return NULL;
3659 entry = g_malloc0(sizeof(*entry));
3660 entry->value = info;
3661 entry->next = list;
3662 list = entry;
3665 return list;
3668 BlockDriverState *bdrv_lookup_bs(const char *device,
3669 const char *node_name,
3670 Error **errp)
3672 BlockBackend *blk;
3673 BlockDriverState *bs;
3675 if (device) {
3676 blk = blk_by_name(device);
3678 if (blk) {
3679 bs = blk_bs(blk);
3680 if (!bs) {
3681 error_setg(errp, "Device '%s' has no medium", device);
3684 return bs;
3688 if (node_name) {
3689 bs = bdrv_find_node(node_name);
3691 if (bs) {
3692 return bs;
3696 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3697 device ? device : "",
3698 node_name ? node_name : "");
3699 return NULL;
3702 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3703 * return false. If either argument is NULL, return false. */
3704 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3706 while (top && top != base) {
3707 top = backing_bs(top);
3710 return top != NULL;
3713 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3715 if (!bs) {
3716 return QTAILQ_FIRST(&graph_bdrv_states);
3718 return QTAILQ_NEXT(bs, node_list);
3721 const char *bdrv_get_node_name(const BlockDriverState *bs)
3723 return bs->node_name;
3726 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3728 BdrvChild *c;
3729 const char *name;
3731 /* If multiple parents have a name, just pick the first one. */
3732 QLIST_FOREACH(c, &bs->parents, next_parent) {
3733 if (c->role->get_name) {
3734 name = c->role->get_name(c);
3735 if (name && *name) {
3736 return name;
3741 return NULL;
3744 /* TODO check what callers really want: bs->node_name or blk_name() */
3745 const char *bdrv_get_device_name(const BlockDriverState *bs)
3747 return bdrv_get_parent_name(bs) ?: "";
3750 /* This can be used to identify nodes that might not have a device
3751 * name associated. Since node and device names live in the same
3752 * namespace, the result is unambiguous. The exception is if both are
3753 * absent, then this returns an empty (non-null) string. */
3754 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3756 return bdrv_get_parent_name(bs) ?: bs->node_name;
3759 int bdrv_get_flags(BlockDriverState *bs)
3761 return bs->open_flags;
3764 int bdrv_has_zero_init_1(BlockDriverState *bs)
3766 return 1;
3769 int bdrv_has_zero_init(BlockDriverState *bs)
3771 assert(bs->drv);
3773 /* If BS is a copy on write image, it is initialized to
3774 the contents of the base image, which may not be zeroes. */
3775 if (bs->backing) {
3776 return 0;
3778 if (bs->drv->bdrv_has_zero_init) {
3779 return bs->drv->bdrv_has_zero_init(bs);
3782 /* safe default */
3783 return 0;
3786 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3788 BlockDriverInfo bdi;
3790 if (bs->backing) {
3791 return false;
3794 if (bdrv_get_info(bs, &bdi) == 0) {
3795 return bdi.unallocated_blocks_are_zero;
3798 return false;
3801 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3803 BlockDriverInfo bdi;
3805 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3806 return false;
3809 if (bdrv_get_info(bs, &bdi) == 0) {
3810 return bdi.can_write_zeroes_with_unmap;
3813 return false;
3816 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3818 if (bs->backing && bs->backing->bs->encrypted)
3819 return bs->backing_file;
3820 else if (bs->encrypted)
3821 return bs->filename;
3822 else
3823 return NULL;
3826 void bdrv_get_backing_filename(BlockDriverState *bs,
3827 char *filename, int filename_size)
3829 pstrcpy(filename, filename_size, bs->backing_file);
3832 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3834 BlockDriver *drv = bs->drv;
3835 if (!drv)
3836 return -ENOMEDIUM;
3837 if (!drv->bdrv_get_info)
3838 return -ENOTSUP;
3839 memset(bdi, 0, sizeof(*bdi));
3840 return drv->bdrv_get_info(bs, bdi);
3843 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3845 BlockDriver *drv = bs->drv;
3846 if (drv && drv->bdrv_get_specific_info) {
3847 return drv->bdrv_get_specific_info(bs);
3849 return NULL;
3852 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3854 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3855 return;
3858 bs->drv->bdrv_debug_event(bs, event);
3861 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3862 const char *tag)
3864 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3865 bs = bs->file ? bs->file->bs : NULL;
3868 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3869 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3872 return -ENOTSUP;
3875 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3877 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3878 bs = bs->file ? bs->file->bs : NULL;
3881 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3882 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3885 return -ENOTSUP;
3888 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3890 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3891 bs = bs->file ? bs->file->bs : NULL;
3894 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3895 return bs->drv->bdrv_debug_resume(bs, tag);
3898 return -ENOTSUP;
3901 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3903 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3904 bs = bs->file ? bs->file->bs : NULL;
3907 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3908 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3911 return false;
3914 /* backing_file can either be relative, or absolute, or a protocol. If it is
3915 * relative, it must be relative to the chain. So, passing in bs->filename
3916 * from a BDS as backing_file should not be done, as that may be relative to
3917 * the CWD rather than the chain. */
3918 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3919 const char *backing_file)
3921 char *filename_full = NULL;
3922 char *backing_file_full = NULL;
3923 char *filename_tmp = NULL;
3924 int is_protocol = 0;
3925 BlockDriverState *curr_bs = NULL;
3926 BlockDriverState *retval = NULL;
3927 Error *local_error = NULL;
3929 if (!bs || !bs->drv || !backing_file) {
3930 return NULL;
3933 filename_full = g_malloc(PATH_MAX);
3934 backing_file_full = g_malloc(PATH_MAX);
3935 filename_tmp = g_malloc(PATH_MAX);
3937 is_protocol = path_has_protocol(backing_file);
3939 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3941 /* If either of the filename paths is actually a protocol, then
3942 * compare unmodified paths; otherwise make paths relative */
3943 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3944 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3945 retval = curr_bs->backing->bs;
3946 break;
3948 /* Also check against the full backing filename for the image */
3949 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
3950 &local_error);
3951 if (local_error == NULL) {
3952 if (strcmp(backing_file, backing_file_full) == 0) {
3953 retval = curr_bs->backing->bs;
3954 break;
3956 } else {
3957 error_free(local_error);
3958 local_error = NULL;
3960 } else {
3961 /* If not an absolute filename path, make it relative to the current
3962 * image's filename path */
3963 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3964 backing_file);
3966 /* We are going to compare absolute pathnames */
3967 if (!realpath(filename_tmp, filename_full)) {
3968 continue;
3971 /* We need to make sure the backing filename we are comparing against
3972 * is relative to the current image filename (or absolute) */
3973 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3974 curr_bs->backing_file);
3976 if (!realpath(filename_tmp, backing_file_full)) {
3977 continue;
3980 if (strcmp(backing_file_full, filename_full) == 0) {
3981 retval = curr_bs->backing->bs;
3982 break;
3987 g_free(filename_full);
3988 g_free(backing_file_full);
3989 g_free(filename_tmp);
3990 return retval;
3993 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3995 if (!bs->drv) {
3996 return 0;
3999 if (!bs->backing) {
4000 return 0;
4003 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
4006 void bdrv_init(void)
4008 module_call_init(MODULE_INIT_BLOCK);
4011 void bdrv_init_with_whitelist(void)
4013 use_bdrv_whitelist = 1;
4014 bdrv_init();
4017 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4019 BdrvChild *child, *parent;
4020 uint64_t perm, shared_perm;
4021 Error *local_err = NULL;
4022 int ret;
4024 if (!bs->drv) {
4025 return;
4028 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4029 return;
4032 QLIST_FOREACH(child, &bs->children, next) {
4033 bdrv_invalidate_cache(child->bs, &local_err);
4034 if (local_err) {
4035 error_propagate(errp, local_err);
4036 return;
4040 bs->open_flags &= ~BDRV_O_INACTIVE;
4041 if (bs->drv->bdrv_invalidate_cache) {
4042 bs->drv->bdrv_invalidate_cache(bs, &local_err);
4043 if (local_err) {
4044 bs->open_flags |= BDRV_O_INACTIVE;
4045 error_propagate(errp, local_err);
4046 return;
4050 ret = refresh_total_sectors(bs, bs->total_sectors);
4051 if (ret < 0) {
4052 bs->open_flags |= BDRV_O_INACTIVE;
4053 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4054 return;
4057 /* Update permissions, they may differ for inactive nodes */
4058 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4059 ret = bdrv_check_perm(bs, perm, shared_perm, NULL, &local_err);
4060 if (ret < 0) {
4061 bs->open_flags |= BDRV_O_INACTIVE;
4062 error_propagate(errp, local_err);
4063 return;
4065 bdrv_set_perm(bs, perm, shared_perm);
4067 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4068 if (parent->role->activate) {
4069 parent->role->activate(parent, &local_err);
4070 if (local_err) {
4071 error_propagate(errp, local_err);
4072 return;
4078 void bdrv_invalidate_cache_all(Error **errp)
4080 BlockDriverState *bs;
4081 Error *local_err = NULL;
4082 BdrvNextIterator it;
4084 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4085 AioContext *aio_context = bdrv_get_aio_context(bs);
4087 aio_context_acquire(aio_context);
4088 bdrv_invalidate_cache(bs, &local_err);
4089 aio_context_release(aio_context);
4090 if (local_err) {
4091 error_propagate(errp, local_err);
4092 return;
4097 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4098 bool setting_flag)
4100 BdrvChild *child, *parent;
4101 int ret;
4103 if (!setting_flag && bs->drv->bdrv_inactivate) {
4104 ret = bs->drv->bdrv_inactivate(bs);
4105 if (ret < 0) {
4106 return ret;
4110 if (setting_flag) {
4111 uint64_t perm, shared_perm;
4113 bs->open_flags |= BDRV_O_INACTIVE;
4115 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4116 if (parent->role->inactivate) {
4117 ret = parent->role->inactivate(parent);
4118 if (ret < 0) {
4119 bs->open_flags &= ~BDRV_O_INACTIVE;
4120 return ret;
4125 /* Update permissions, they may differ for inactive nodes */
4126 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4127 bdrv_check_perm(bs, perm, shared_perm, NULL, &error_abort);
4128 bdrv_set_perm(bs, perm, shared_perm);
4131 QLIST_FOREACH(child, &bs->children, next) {
4132 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4133 if (ret < 0) {
4134 return ret;
4138 return 0;
4141 int bdrv_inactivate_all(void)
4143 BlockDriverState *bs = NULL;
4144 BdrvNextIterator it;
4145 int ret = 0;
4146 int pass;
4148 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4149 aio_context_acquire(bdrv_get_aio_context(bs));
4152 /* We do two passes of inactivation. The first pass calls to drivers'
4153 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4154 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4155 * is allowed. */
4156 for (pass = 0; pass < 2; pass++) {
4157 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4158 ret = bdrv_inactivate_recurse(bs, pass);
4159 if (ret < 0) {
4160 goto out;
4165 out:
4166 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4167 aio_context_release(bdrv_get_aio_context(bs));
4170 return ret;
4173 /**************************************************************/
4174 /* removable device support */
4177 * Return TRUE if the media is present
4179 bool bdrv_is_inserted(BlockDriverState *bs)
4181 BlockDriver *drv = bs->drv;
4182 BdrvChild *child;
4184 if (!drv) {
4185 return false;
4187 if (drv->bdrv_is_inserted) {
4188 return drv->bdrv_is_inserted(bs);
4190 QLIST_FOREACH(child, &bs->children, next) {
4191 if (!bdrv_is_inserted(child->bs)) {
4192 return false;
4195 return true;
4199 * Return whether the media changed since the last call to this
4200 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4202 int bdrv_media_changed(BlockDriverState *bs)
4204 BlockDriver *drv = bs->drv;
4206 if (drv && drv->bdrv_media_changed) {
4207 return drv->bdrv_media_changed(bs);
4209 return -ENOTSUP;
4213 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4215 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4217 BlockDriver *drv = bs->drv;
4219 if (drv && drv->bdrv_eject) {
4220 drv->bdrv_eject(bs, eject_flag);
4225 * Lock or unlock the media (if it is locked, the user won't be able
4226 * to eject it manually).
4228 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4230 BlockDriver *drv = bs->drv;
4232 trace_bdrv_lock_medium(bs, locked);
4234 if (drv && drv->bdrv_lock_medium) {
4235 drv->bdrv_lock_medium(bs, locked);
4239 /* Get a reference to bs */
4240 void bdrv_ref(BlockDriverState *bs)
4242 bs->refcnt++;
4245 /* Release a previously grabbed reference to bs.
4246 * If after releasing, reference count is zero, the BlockDriverState is
4247 * deleted. */
4248 void bdrv_unref(BlockDriverState *bs)
4250 if (!bs) {
4251 return;
4253 assert(bs->refcnt > 0);
4254 if (--bs->refcnt == 0) {
4255 bdrv_delete(bs);
4259 struct BdrvOpBlocker {
4260 Error *reason;
4261 QLIST_ENTRY(BdrvOpBlocker) list;
4264 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4266 BdrvOpBlocker *blocker;
4267 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4268 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4269 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4270 if (errp) {
4271 *errp = error_copy(blocker->reason);
4272 error_prepend(errp, "Node '%s' is busy: ",
4273 bdrv_get_device_or_node_name(bs));
4275 return true;
4277 return false;
4280 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4282 BdrvOpBlocker *blocker;
4283 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4285 blocker = g_new0(BdrvOpBlocker, 1);
4286 blocker->reason = reason;
4287 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4290 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4292 BdrvOpBlocker *blocker, *next;
4293 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4294 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4295 if (blocker->reason == reason) {
4296 QLIST_REMOVE(blocker, list);
4297 g_free(blocker);
4302 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4304 int i;
4305 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4306 bdrv_op_block(bs, i, reason);
4310 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4312 int i;
4313 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4314 bdrv_op_unblock(bs, i, reason);
4318 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4320 int i;
4322 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4323 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4324 return false;
4327 return true;
4330 void bdrv_img_create(const char *filename, const char *fmt,
4331 const char *base_filename, const char *base_fmt,
4332 char *options, uint64_t img_size, int flags, bool quiet,
4333 Error **errp)
4335 QemuOptsList *create_opts = NULL;
4336 QemuOpts *opts = NULL;
4337 const char *backing_fmt, *backing_file;
4338 int64_t size;
4339 BlockDriver *drv, *proto_drv;
4340 Error *local_err = NULL;
4341 int ret = 0;
4343 /* Find driver and parse its options */
4344 drv = bdrv_find_format(fmt);
4345 if (!drv) {
4346 error_setg(errp, "Unknown file format '%s'", fmt);
4347 return;
4350 proto_drv = bdrv_find_protocol(filename, true, errp);
4351 if (!proto_drv) {
4352 return;
4355 if (!drv->create_opts) {
4356 error_setg(errp, "Format driver '%s' does not support image creation",
4357 drv->format_name);
4358 return;
4361 if (!proto_drv->create_opts) {
4362 error_setg(errp, "Protocol driver '%s' does not support image creation",
4363 proto_drv->format_name);
4364 return;
4367 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4368 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4370 /* Create parameter list with default values */
4371 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4372 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4374 /* Parse -o options */
4375 if (options) {
4376 qemu_opts_do_parse(opts, options, NULL, &local_err);
4377 if (local_err) {
4378 error_report_err(local_err);
4379 local_err = NULL;
4380 error_setg(errp, "Invalid options for file format '%s'", fmt);
4381 goto out;
4385 if (base_filename) {
4386 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4387 if (local_err) {
4388 error_setg(errp, "Backing file not supported for file format '%s'",
4389 fmt);
4390 goto out;
4394 if (base_fmt) {
4395 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4396 if (local_err) {
4397 error_setg(errp, "Backing file format not supported for file "
4398 "format '%s'", fmt);
4399 goto out;
4403 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4404 if (backing_file) {
4405 if (!strcmp(filename, backing_file)) {
4406 error_setg(errp, "Error: Trying to create an image with the "
4407 "same filename as the backing file");
4408 goto out;
4412 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4414 // The size for the image must always be specified, with one exception:
4415 // If we are using a backing file, we can obtain the size from there
4416 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4417 if (size == -1) {
4418 if (backing_file) {
4419 BlockDriverState *bs;
4420 char *full_backing = g_new0(char, PATH_MAX);
4421 int64_t size;
4422 int back_flags;
4423 QDict *backing_options = NULL;
4425 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4426 full_backing, PATH_MAX,
4427 &local_err);
4428 if (local_err) {
4429 g_free(full_backing);
4430 goto out;
4433 /* backing files always opened read-only */
4434 back_flags = flags;
4435 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4437 if (backing_fmt) {
4438 backing_options = qdict_new();
4439 qdict_put_str(backing_options, "driver", backing_fmt);
4442 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4443 &local_err);
4444 g_free(full_backing);
4445 if (!bs) {
4446 goto out;
4448 size = bdrv_getlength(bs);
4449 if (size < 0) {
4450 error_setg_errno(errp, -size, "Could not get size of '%s'",
4451 backing_file);
4452 bdrv_unref(bs);
4453 goto out;
4456 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4458 bdrv_unref(bs);
4459 } else {
4460 error_setg(errp, "Image creation needs a size parameter");
4461 goto out;
4465 if (!quiet) {
4466 printf("Formatting '%s', fmt=%s ", filename, fmt);
4467 qemu_opts_print(opts, " ");
4468 puts("");
4471 ret = bdrv_create(drv, filename, opts, &local_err);
4473 if (ret == -EFBIG) {
4474 /* This is generally a better message than whatever the driver would
4475 * deliver (especially because of the cluster_size_hint), since that
4476 * is most probably not much different from "image too large". */
4477 const char *cluster_size_hint = "";
4478 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4479 cluster_size_hint = " (try using a larger cluster size)";
4481 error_setg(errp, "The image size is too large for file format '%s'"
4482 "%s", fmt, cluster_size_hint);
4483 error_free(local_err);
4484 local_err = NULL;
4487 out:
4488 qemu_opts_del(opts);
4489 qemu_opts_free(create_opts);
4490 error_propagate(errp, local_err);
4493 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4495 return bs->aio_context;
4498 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4500 aio_co_enter(bdrv_get_aio_context(bs), co);
4503 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4505 QLIST_REMOVE(ban, list);
4506 g_free(ban);
4509 void bdrv_detach_aio_context(BlockDriverState *bs)
4511 BdrvAioNotifier *baf, *baf_tmp;
4512 BdrvChild *child;
4514 if (!bs->drv) {
4515 return;
4518 assert(!bs->walking_aio_notifiers);
4519 bs->walking_aio_notifiers = true;
4520 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4521 if (baf->deleted) {
4522 bdrv_do_remove_aio_context_notifier(baf);
4523 } else {
4524 baf->detach_aio_context(baf->opaque);
4527 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4528 * remove remaining aio notifiers if we aren't called again.
4530 bs->walking_aio_notifiers = false;
4532 if (bs->drv->bdrv_detach_aio_context) {
4533 bs->drv->bdrv_detach_aio_context(bs);
4535 QLIST_FOREACH(child, &bs->children, next) {
4536 bdrv_detach_aio_context(child->bs);
4539 bs->aio_context = NULL;
4542 void bdrv_attach_aio_context(BlockDriverState *bs,
4543 AioContext *new_context)
4545 BdrvAioNotifier *ban, *ban_tmp;
4546 BdrvChild *child;
4548 if (!bs->drv) {
4549 return;
4552 bs->aio_context = new_context;
4554 QLIST_FOREACH(child, &bs->children, next) {
4555 bdrv_attach_aio_context(child->bs, new_context);
4557 if (bs->drv->bdrv_attach_aio_context) {
4558 bs->drv->bdrv_attach_aio_context(bs, new_context);
4561 assert(!bs->walking_aio_notifiers);
4562 bs->walking_aio_notifiers = true;
4563 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4564 if (ban->deleted) {
4565 bdrv_do_remove_aio_context_notifier(ban);
4566 } else {
4567 ban->attached_aio_context(new_context, ban->opaque);
4570 bs->walking_aio_notifiers = false;
4573 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4575 AioContext *ctx = bdrv_get_aio_context(bs);
4577 aio_disable_external(ctx);
4578 bdrv_parent_drained_begin(bs);
4579 bdrv_drain(bs); /* ensure there are no in-flight requests */
4581 while (aio_poll(ctx, false)) {
4582 /* wait for all bottom halves to execute */
4585 bdrv_detach_aio_context(bs);
4587 /* This function executes in the old AioContext so acquire the new one in
4588 * case it runs in a different thread.
4590 aio_context_acquire(new_context);
4591 bdrv_attach_aio_context(bs, new_context);
4592 bdrv_parent_drained_end(bs);
4593 aio_enable_external(ctx);
4594 aio_context_release(new_context);
4597 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4598 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4599 void (*detach_aio_context)(void *opaque), void *opaque)
4601 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4602 *ban = (BdrvAioNotifier){
4603 .attached_aio_context = attached_aio_context,
4604 .detach_aio_context = detach_aio_context,
4605 .opaque = opaque
4608 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4611 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4612 void (*attached_aio_context)(AioContext *,
4613 void *),
4614 void (*detach_aio_context)(void *),
4615 void *opaque)
4617 BdrvAioNotifier *ban, *ban_next;
4619 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4620 if (ban->attached_aio_context == attached_aio_context &&
4621 ban->detach_aio_context == detach_aio_context &&
4622 ban->opaque == opaque &&
4623 ban->deleted == false)
4625 if (bs->walking_aio_notifiers) {
4626 ban->deleted = true;
4627 } else {
4628 bdrv_do_remove_aio_context_notifier(ban);
4630 return;
4634 abort();
4637 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4638 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4640 if (!bs->drv->bdrv_amend_options) {
4641 return -ENOTSUP;
4643 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
4646 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4647 * of block filter and by bdrv_is_first_non_filter.
4648 * It is used to test if the given bs is the candidate or recurse more in the
4649 * node graph.
4651 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
4652 BlockDriverState *candidate)
4654 /* return false if basic checks fails */
4655 if (!bs || !bs->drv) {
4656 return false;
4659 /* the code reached a non block filter driver -> check if the bs is
4660 * the same as the candidate. It's the recursion termination condition.
4662 if (!bs->drv->is_filter) {
4663 return bs == candidate;
4665 /* Down this path the driver is a block filter driver */
4667 /* If the block filter recursion method is defined use it to recurse down
4668 * the node graph.
4670 if (bs->drv->bdrv_recurse_is_first_non_filter) {
4671 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
4674 /* the driver is a block filter but don't allow to recurse -> return false
4676 return false;
4679 /* This function checks if the candidate is the first non filter bs down it's
4680 * bs chain. Since we don't have pointers to parents it explore all bs chains
4681 * from the top. Some filters can choose not to pass down the recursion.
4683 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
4685 BlockDriverState *bs;
4686 BdrvNextIterator it;
4688 /* walk down the bs forest recursively */
4689 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4690 bool perm;
4692 /* try to recurse in this top level bs */
4693 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
4695 /* candidate is the first non filter */
4696 if (perm) {
4697 return true;
4701 return false;
4704 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
4705 const char *node_name, Error **errp)
4707 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
4708 AioContext *aio_context;
4710 if (!to_replace_bs) {
4711 error_setg(errp, "Node name '%s' not found", node_name);
4712 return NULL;
4715 aio_context = bdrv_get_aio_context(to_replace_bs);
4716 aio_context_acquire(aio_context);
4718 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
4719 to_replace_bs = NULL;
4720 goto out;
4723 /* We don't want arbitrary node of the BDS chain to be replaced only the top
4724 * most non filter in order to prevent data corruption.
4725 * Another benefit is that this tests exclude backing files which are
4726 * blocked by the backing blockers.
4728 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
4729 error_setg(errp, "Only top most non filter can be replaced");
4730 to_replace_bs = NULL;
4731 goto out;
4734 out:
4735 aio_context_release(aio_context);
4736 return to_replace_bs;
4739 static bool append_open_options(QDict *d, BlockDriverState *bs)
4741 const QDictEntry *entry;
4742 QemuOptDesc *desc;
4743 BdrvChild *child;
4744 bool found_any = false;
4745 const char *p;
4747 for (entry = qdict_first(bs->options); entry;
4748 entry = qdict_next(bs->options, entry))
4750 /* Exclude options for children */
4751 QLIST_FOREACH(child, &bs->children, next) {
4752 if (strstart(qdict_entry_key(entry), child->name, &p)
4753 && (!*p || *p == '.'))
4755 break;
4758 if (child) {
4759 continue;
4762 /* And exclude all non-driver-specific options */
4763 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4764 if (!strcmp(qdict_entry_key(entry), desc->name)) {
4765 break;
4768 if (desc->name) {
4769 continue;
4772 qobject_incref(qdict_entry_value(entry));
4773 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
4774 found_any = true;
4777 return found_any;
4780 /* Updates the following BDS fields:
4781 * - exact_filename: A filename which may be used for opening a block device
4782 * which (mostly) equals the given BDS (even without any
4783 * other options; so reading and writing must return the same
4784 * results, but caching etc. may be different)
4785 * - full_open_options: Options which, when given when opening a block device
4786 * (without a filename), result in a BDS (mostly)
4787 * equalling the given one
4788 * - filename: If exact_filename is set, it is copied here. Otherwise,
4789 * full_open_options is converted to a JSON object, prefixed with
4790 * "json:" (for use through the JSON pseudo protocol) and put here.
4792 void bdrv_refresh_filename(BlockDriverState *bs)
4794 BlockDriver *drv = bs->drv;
4795 QDict *opts;
4797 if (!drv) {
4798 return;
4801 /* This BDS's file name will most probably depend on its file's name, so
4802 * refresh that first */
4803 if (bs->file) {
4804 bdrv_refresh_filename(bs->file->bs);
4807 if (drv->bdrv_refresh_filename) {
4808 /* Obsolete information is of no use here, so drop the old file name
4809 * information before refreshing it */
4810 bs->exact_filename[0] = '\0';
4811 if (bs->full_open_options) {
4812 QDECREF(bs->full_open_options);
4813 bs->full_open_options = NULL;
4816 opts = qdict_new();
4817 append_open_options(opts, bs);
4818 drv->bdrv_refresh_filename(bs, opts);
4819 QDECREF(opts);
4820 } else if (bs->file) {
4821 /* Try to reconstruct valid information from the underlying file */
4822 bool has_open_options;
4824 bs->exact_filename[0] = '\0';
4825 if (bs->full_open_options) {
4826 QDECREF(bs->full_open_options);
4827 bs->full_open_options = NULL;
4830 opts = qdict_new();
4831 has_open_options = append_open_options(opts, bs);
4833 /* If no specific options have been given for this BDS, the filename of
4834 * the underlying file should suffice for this one as well */
4835 if (bs->file->bs->exact_filename[0] && !has_open_options) {
4836 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
4838 /* Reconstructing the full options QDict is simple for most format block
4839 * drivers, as long as the full options are known for the underlying
4840 * file BDS. The full options QDict of that file BDS should somehow
4841 * contain a representation of the filename, therefore the following
4842 * suffices without querying the (exact_)filename of this BDS. */
4843 if (bs->file->bs->full_open_options) {
4844 qdict_put_str(opts, "driver", drv->format_name);
4845 QINCREF(bs->file->bs->full_open_options);
4846 qdict_put(opts, "file", bs->file->bs->full_open_options);
4848 bs->full_open_options = opts;
4849 } else {
4850 QDECREF(opts);
4852 } else if (!bs->full_open_options && qdict_size(bs->options)) {
4853 /* There is no underlying file BDS (at least referenced by BDS.file),
4854 * so the full options QDict should be equal to the options given
4855 * specifically for this block device when it was opened (plus the
4856 * driver specification).
4857 * Because those options don't change, there is no need to update
4858 * full_open_options when it's already set. */
4860 opts = qdict_new();
4861 append_open_options(opts, bs);
4862 qdict_put_str(opts, "driver", drv->format_name);
4864 if (bs->exact_filename[0]) {
4865 /* This may not work for all block protocol drivers (some may
4866 * require this filename to be parsed), but we have to find some
4867 * default solution here, so just include it. If some block driver
4868 * does not support pure options without any filename at all or
4869 * needs some special format of the options QDict, it needs to
4870 * implement the driver-specific bdrv_refresh_filename() function.
4872 qdict_put_str(opts, "filename", bs->exact_filename);
4875 bs->full_open_options = opts;
4878 if (bs->exact_filename[0]) {
4879 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
4880 } else if (bs->full_open_options) {
4881 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
4882 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
4883 qstring_get_str(json));
4884 QDECREF(json);
4889 * Hot add/remove a BDS's child. So the user can take a child offline when
4890 * it is broken and take a new child online
4892 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
4893 Error **errp)
4896 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
4897 error_setg(errp, "The node %s does not support adding a child",
4898 bdrv_get_device_or_node_name(parent_bs));
4899 return;
4902 if (!QLIST_EMPTY(&child_bs->parents)) {
4903 error_setg(errp, "The node %s already has a parent",
4904 child_bs->node_name);
4905 return;
4908 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
4911 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
4913 BdrvChild *tmp;
4915 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
4916 error_setg(errp, "The node %s does not support removing a child",
4917 bdrv_get_device_or_node_name(parent_bs));
4918 return;
4921 QLIST_FOREACH(tmp, &parent_bs->children, next) {
4922 if (tmp == child) {
4923 break;
4927 if (!tmp) {
4928 error_setg(errp, "The node %s does not have a child named %s",
4929 bdrv_get_device_or_node_name(parent_bs),
4930 bdrv_get_device_or_node_name(child->bs));
4931 return;
4934 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);