s390x/css: realize css_create_sch
[qemu.git] / block.c
blob50ba2641430aef9c18b9f1070dd59a1ab3891654
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 p = strchr(base_path, ':');
167 if (p)
168 p++;
169 else
170 p = base_path;
171 p1 = strrchr(base_path, '/');
172 #ifdef _WIN32
174 const char *p2;
175 p2 = strrchr(base_path, '\\');
176 if (!p1 || p2 > p1)
177 p1 = p2;
179 #endif
180 if (p1)
181 p1++;
182 else
183 p1 = base_path;
184 if (p1 > p)
185 p = p1;
186 len = p - base_path;
187 if (len > dest_size - 1)
188 len = dest_size - 1;
189 memcpy(dest, base_path, len);
190 dest[len] = '\0';
191 pstrcat(dest, dest_size, filename);
195 /* Returns whether the image file is opened as read-only. Note that this can
196 * return false and writing to the image file is still not possible because the
197 * image is inactivated. */
198 bool bdrv_is_read_only(BlockDriverState *bs)
200 return bs->read_only;
203 /* Returns whether the image file can be written to right now */
204 bool bdrv_is_writable(BlockDriverState *bs)
206 return !bdrv_is_read_only(bs) && !(bs->open_flags & BDRV_O_INACTIVE);
209 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
211 /* Do not set read_only if copy_on_read is enabled */
212 if (bs->copy_on_read && read_only) {
213 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
214 bdrv_get_device_or_node_name(bs));
215 return -EINVAL;
218 /* Do not clear read_only if it is prohibited */
219 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR)) {
220 error_setg(errp, "Node '%s' is read only",
221 bdrv_get_device_or_node_name(bs));
222 return -EPERM;
225 return 0;
228 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
230 int ret = 0;
232 ret = bdrv_can_set_read_only(bs, read_only, errp);
233 if (ret < 0) {
234 return ret;
237 bs->read_only = read_only;
238 return 0;
241 void bdrv_get_full_backing_filename_from_filename(const char *backed,
242 const char *backing,
243 char *dest, size_t sz,
244 Error **errp)
246 if (backing[0] == '\0' || path_has_protocol(backing) ||
247 path_is_absolute(backing))
249 pstrcpy(dest, sz, backing);
250 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
251 error_setg(errp, "Cannot use relative backing file names for '%s'",
252 backed);
253 } else {
254 path_combine(dest, sz, backed, backing);
258 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
259 Error **errp)
261 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
263 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
264 dest, sz, errp);
267 void bdrv_register(BlockDriver *bdrv)
269 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
272 BlockDriverState *bdrv_new(void)
274 BlockDriverState *bs;
275 int i;
277 bs = g_new0(BlockDriverState, 1);
278 QLIST_INIT(&bs->dirty_bitmaps);
279 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
280 QLIST_INIT(&bs->op_blockers[i]);
282 notifier_with_return_list_init(&bs->before_write_notifiers);
283 bs->refcnt = 1;
284 bs->aio_context = qemu_get_aio_context();
286 qemu_co_queue_init(&bs->flush_queue);
288 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
290 return bs;
293 static BlockDriver *bdrv_do_find_format(const char *format_name)
295 BlockDriver *drv1;
297 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
298 if (!strcmp(drv1->format_name, format_name)) {
299 return drv1;
303 return NULL;
306 BlockDriver *bdrv_find_format(const char *format_name)
308 BlockDriver *drv1;
309 int i;
311 drv1 = bdrv_do_find_format(format_name);
312 if (drv1) {
313 return drv1;
316 /* The driver isn't registered, maybe we need to load a module */
317 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
318 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
319 block_module_load_one(block_driver_modules[i].library_name);
320 break;
324 return bdrv_do_find_format(format_name);
327 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
329 static const char *whitelist_rw[] = {
330 CONFIG_BDRV_RW_WHITELIST
332 static const char *whitelist_ro[] = {
333 CONFIG_BDRV_RO_WHITELIST
335 const char **p;
337 if (!whitelist_rw[0] && !whitelist_ro[0]) {
338 return 1; /* no whitelist, anything goes */
341 for (p = whitelist_rw; *p; p++) {
342 if (!strcmp(drv->format_name, *p)) {
343 return 1;
346 if (read_only) {
347 for (p = whitelist_ro; *p; p++) {
348 if (!strcmp(drv->format_name, *p)) {
349 return 1;
353 return 0;
356 bool bdrv_uses_whitelist(void)
358 return use_bdrv_whitelist;
361 typedef struct CreateCo {
362 BlockDriver *drv;
363 char *filename;
364 QemuOpts *opts;
365 int ret;
366 Error *err;
367 } CreateCo;
369 static void coroutine_fn bdrv_create_co_entry(void *opaque)
371 Error *local_err = NULL;
372 int ret;
374 CreateCo *cco = opaque;
375 assert(cco->drv);
377 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
378 error_propagate(&cco->err, local_err);
379 cco->ret = ret;
382 int bdrv_create(BlockDriver *drv, const char* filename,
383 QemuOpts *opts, Error **errp)
385 int ret;
387 Coroutine *co;
388 CreateCo cco = {
389 .drv = drv,
390 .filename = g_strdup(filename),
391 .opts = opts,
392 .ret = NOT_DONE,
393 .err = NULL,
396 if (!drv->bdrv_create) {
397 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
398 ret = -ENOTSUP;
399 goto out;
402 if (qemu_in_coroutine()) {
403 /* Fast-path if already in coroutine context */
404 bdrv_create_co_entry(&cco);
405 } else {
406 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
407 qemu_coroutine_enter(co);
408 while (cco.ret == NOT_DONE) {
409 aio_poll(qemu_get_aio_context(), true);
413 ret = cco.ret;
414 if (ret < 0) {
415 if (cco.err) {
416 error_propagate(errp, cco.err);
417 } else {
418 error_setg_errno(errp, -ret, "Could not create image");
422 out:
423 g_free(cco.filename);
424 return ret;
427 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
429 BlockDriver *drv;
430 Error *local_err = NULL;
431 int ret;
433 drv = bdrv_find_protocol(filename, true, errp);
434 if (drv == NULL) {
435 return -ENOENT;
438 ret = bdrv_create(drv, filename, opts, &local_err);
439 error_propagate(errp, local_err);
440 return ret;
444 * Try to get @bs's logical and physical block size.
445 * On success, store them in @bsz struct and return 0.
446 * On failure return -errno.
447 * @bs must not be empty.
449 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
451 BlockDriver *drv = bs->drv;
453 if (drv && drv->bdrv_probe_blocksizes) {
454 return drv->bdrv_probe_blocksizes(bs, bsz);
457 return -ENOTSUP;
461 * Try to get @bs's geometry (cyls, heads, sectors).
462 * On success, store them in @geo struct and return 0.
463 * On failure return -errno.
464 * @bs must not be empty.
466 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
468 BlockDriver *drv = bs->drv;
470 if (drv && drv->bdrv_probe_geometry) {
471 return drv->bdrv_probe_geometry(bs, geo);
474 return -ENOTSUP;
478 * Create a uniquely-named empty temporary file.
479 * Return 0 upon success, otherwise a negative errno value.
481 int get_tmp_filename(char *filename, int size)
483 #ifdef _WIN32
484 char temp_dir[MAX_PATH];
485 /* GetTempFileName requires that its output buffer (4th param)
486 have length MAX_PATH or greater. */
487 assert(size >= MAX_PATH);
488 return (GetTempPath(MAX_PATH, temp_dir)
489 && GetTempFileName(temp_dir, "qem", 0, filename)
490 ? 0 : -GetLastError());
491 #else
492 int fd;
493 const char *tmpdir;
494 tmpdir = getenv("TMPDIR");
495 if (!tmpdir) {
496 tmpdir = "/var/tmp";
498 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
499 return -EOVERFLOW;
501 fd = mkstemp(filename);
502 if (fd < 0) {
503 return -errno;
505 if (close(fd) != 0) {
506 unlink(filename);
507 return -errno;
509 return 0;
510 #endif
514 * Detect host devices. By convention, /dev/cdrom[N] is always
515 * recognized as a host CDROM.
517 static BlockDriver *find_hdev_driver(const char *filename)
519 int score_max = 0, score;
520 BlockDriver *drv = NULL, *d;
522 QLIST_FOREACH(d, &bdrv_drivers, list) {
523 if (d->bdrv_probe_device) {
524 score = d->bdrv_probe_device(filename);
525 if (score > score_max) {
526 score_max = score;
527 drv = d;
532 return drv;
535 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
537 BlockDriver *drv1;
539 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
540 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
541 return drv1;
545 return NULL;
548 BlockDriver *bdrv_find_protocol(const char *filename,
549 bool allow_protocol_prefix,
550 Error **errp)
552 BlockDriver *drv1;
553 char protocol[128];
554 int len;
555 const char *p;
556 int i;
558 /* TODO Drivers without bdrv_file_open must be specified explicitly */
561 * XXX(hch): we really should not let host device detection
562 * override an explicit protocol specification, but moving this
563 * later breaks access to device names with colons in them.
564 * Thanks to the brain-dead persistent naming schemes on udev-
565 * based Linux systems those actually are quite common.
567 drv1 = find_hdev_driver(filename);
568 if (drv1) {
569 return drv1;
572 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
573 return &bdrv_file;
576 p = strchr(filename, ':');
577 assert(p != NULL);
578 len = p - filename;
579 if (len > sizeof(protocol) - 1)
580 len = sizeof(protocol) - 1;
581 memcpy(protocol, filename, len);
582 protocol[len] = '\0';
584 drv1 = bdrv_do_find_protocol(protocol);
585 if (drv1) {
586 return drv1;
589 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
590 if (block_driver_modules[i].protocol_name &&
591 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
592 block_module_load_one(block_driver_modules[i].library_name);
593 break;
597 drv1 = bdrv_do_find_protocol(protocol);
598 if (!drv1) {
599 error_setg(errp, "Unknown protocol '%s'", protocol);
601 return drv1;
605 * Guess image format by probing its contents.
606 * This is not a good idea when your image is raw (CVE-2008-2004), but
607 * we do it anyway for backward compatibility.
609 * @buf contains the image's first @buf_size bytes.
610 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
611 * but can be smaller if the image file is smaller)
612 * @filename is its filename.
614 * For all block drivers, call the bdrv_probe() method to get its
615 * probing score.
616 * Return the first block driver with the highest probing score.
618 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
619 const char *filename)
621 int score_max = 0, score;
622 BlockDriver *drv = NULL, *d;
624 QLIST_FOREACH(d, &bdrv_drivers, list) {
625 if (d->bdrv_probe) {
626 score = d->bdrv_probe(buf, buf_size, filename);
627 if (score > score_max) {
628 score_max = score;
629 drv = d;
634 return drv;
637 static int find_image_format(BlockBackend *file, const char *filename,
638 BlockDriver **pdrv, Error **errp)
640 BlockDriver *drv;
641 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
642 int ret = 0;
644 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
645 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
646 *pdrv = &bdrv_raw;
647 return ret;
650 ret = blk_pread(file, 0, buf, sizeof(buf));
651 if (ret < 0) {
652 error_setg_errno(errp, -ret, "Could not read image for determining its "
653 "format");
654 *pdrv = NULL;
655 return ret;
658 drv = bdrv_probe_all(buf, ret, filename);
659 if (!drv) {
660 error_setg(errp, "Could not determine image format: No compatible "
661 "driver found");
662 ret = -ENOENT;
664 *pdrv = drv;
665 return ret;
669 * Set the current 'total_sectors' value
670 * Return 0 on success, -errno on error.
672 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
674 BlockDriver *drv = bs->drv;
676 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
677 if (bdrv_is_sg(bs))
678 return 0;
680 /* query actual device if possible, otherwise just trust the hint */
681 if (drv->bdrv_getlength) {
682 int64_t length = drv->bdrv_getlength(bs);
683 if (length < 0) {
684 return length;
686 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
689 bs->total_sectors = hint;
690 return 0;
694 * Combines a QDict of new block driver @options with any missing options taken
695 * from @old_options, so that leaving out an option defaults to its old value.
697 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
698 QDict *old_options)
700 if (bs->drv && bs->drv->bdrv_join_options) {
701 bs->drv->bdrv_join_options(options, old_options);
702 } else {
703 qdict_join(options, old_options, false);
708 * Set open flags for a given discard mode
710 * Return 0 on success, -1 if the discard mode was invalid.
712 int bdrv_parse_discard_flags(const char *mode, int *flags)
714 *flags &= ~BDRV_O_UNMAP;
716 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
717 /* do nothing */
718 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
719 *flags |= BDRV_O_UNMAP;
720 } else {
721 return -1;
724 return 0;
728 * Set open flags for a given cache mode
730 * Return 0 on success, -1 if the cache mode was invalid.
732 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
734 *flags &= ~BDRV_O_CACHE_MASK;
736 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
737 *writethrough = false;
738 *flags |= BDRV_O_NOCACHE;
739 } else if (!strcmp(mode, "directsync")) {
740 *writethrough = true;
741 *flags |= BDRV_O_NOCACHE;
742 } else if (!strcmp(mode, "writeback")) {
743 *writethrough = false;
744 } else if (!strcmp(mode, "unsafe")) {
745 *writethrough = false;
746 *flags |= BDRV_O_NO_FLUSH;
747 } else if (!strcmp(mode, "writethrough")) {
748 *writethrough = true;
749 } else {
750 return -1;
753 return 0;
756 static char *bdrv_child_get_parent_desc(BdrvChild *c)
758 BlockDriverState *parent = c->opaque;
759 return g_strdup(bdrv_get_device_or_node_name(parent));
762 static void bdrv_child_cb_drained_begin(BdrvChild *child)
764 BlockDriverState *bs = child->opaque;
765 bdrv_drained_begin(bs);
768 static void bdrv_child_cb_drained_end(BdrvChild *child)
770 BlockDriverState *bs = child->opaque;
771 bdrv_drained_end(bs);
774 static int bdrv_child_cb_inactivate(BdrvChild *child)
776 BlockDriverState *bs = child->opaque;
777 assert(bs->open_flags & BDRV_O_INACTIVE);
778 return 0;
782 * Returns the options and flags that a temporary snapshot should get, based on
783 * the originally requested flags (the originally requested image will have
784 * flags like a backing file)
786 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
787 int parent_flags, QDict *parent_options)
789 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
791 /* For temporary files, unconditional cache=unsafe is fine */
792 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
793 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
795 /* Copy the read-only option from the parent */
796 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
798 /* aio=native doesn't work for cache.direct=off, so disable it for the
799 * temporary snapshot */
800 *child_flags &= ~BDRV_O_NATIVE_AIO;
804 * Returns the options and flags that bs->file should get if a protocol driver
805 * is expected, based on the given options and flags for the parent BDS
807 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
808 int parent_flags, QDict *parent_options)
810 int flags = parent_flags;
812 /* Enable protocol handling, disable format probing for bs->file */
813 flags |= BDRV_O_PROTOCOL;
815 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
816 * the parent. */
817 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
818 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
819 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
821 /* Inherit the read-only option from the parent if it's not set */
822 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
824 /* Our block drivers take care to send flushes and respect unmap policy,
825 * so we can default to enable both on lower layers regardless of the
826 * corresponding parent options. */
827 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
829 /* Clear flags that only apply to the top layer */
830 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
831 BDRV_O_NO_IO);
833 *child_flags = flags;
836 const BdrvChildRole child_file = {
837 .get_parent_desc = bdrv_child_get_parent_desc,
838 .inherit_options = bdrv_inherited_options,
839 .drained_begin = bdrv_child_cb_drained_begin,
840 .drained_end = bdrv_child_cb_drained_end,
841 .inactivate = bdrv_child_cb_inactivate,
845 * Returns the options and flags that bs->file should get if the use of formats
846 * (and not only protocols) is permitted for it, based on the given options and
847 * flags for the parent BDS
849 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
850 int parent_flags, QDict *parent_options)
852 child_file.inherit_options(child_flags, child_options,
853 parent_flags, parent_options);
855 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
858 const BdrvChildRole child_format = {
859 .get_parent_desc = bdrv_child_get_parent_desc,
860 .inherit_options = bdrv_inherited_fmt_options,
861 .drained_begin = bdrv_child_cb_drained_begin,
862 .drained_end = bdrv_child_cb_drained_end,
863 .inactivate = bdrv_child_cb_inactivate,
866 static void bdrv_backing_attach(BdrvChild *c)
868 BlockDriverState *parent = c->opaque;
869 BlockDriverState *backing_hd = c->bs;
871 assert(!parent->backing_blocker);
872 error_setg(&parent->backing_blocker,
873 "node is used as backing hd of '%s'",
874 bdrv_get_device_or_node_name(parent));
876 parent->open_flags &= ~BDRV_O_NO_BACKING;
877 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
878 backing_hd->filename);
879 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
880 backing_hd->drv ? backing_hd->drv->format_name : "");
882 bdrv_op_block_all(backing_hd, parent->backing_blocker);
883 /* Otherwise we won't be able to commit or stream */
884 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
885 parent->backing_blocker);
886 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
887 parent->backing_blocker);
889 * We do backup in 3 ways:
890 * 1. drive backup
891 * The target bs is new opened, and the source is top BDS
892 * 2. blockdev backup
893 * Both the source and the target are top BDSes.
894 * 3. internal backup(used for block replication)
895 * Both the source and the target are backing file
897 * In case 1 and 2, neither the source nor the target is the backing file.
898 * In case 3, we will block the top BDS, so there is only one block job
899 * for the top BDS and its backing chain.
901 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
902 parent->backing_blocker);
903 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
904 parent->backing_blocker);
907 static void bdrv_backing_detach(BdrvChild *c)
909 BlockDriverState *parent = c->opaque;
911 assert(parent->backing_blocker);
912 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
913 error_free(parent->backing_blocker);
914 parent->backing_blocker = NULL;
918 * Returns the options and flags that bs->backing should get, based on the
919 * given options and flags for the parent BDS
921 static void bdrv_backing_options(int *child_flags, QDict *child_options,
922 int parent_flags, QDict *parent_options)
924 int flags = parent_flags;
926 /* The cache mode is inherited unmodified for backing files; except WCE,
927 * which is only applied on the top level (BlockBackend) */
928 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
929 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
930 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
932 /* backing files always opened read-only */
933 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
934 flags &= ~BDRV_O_COPY_ON_READ;
936 /* snapshot=on is handled on the top layer */
937 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
939 *child_flags = flags;
942 const BdrvChildRole child_backing = {
943 .get_parent_desc = bdrv_child_get_parent_desc,
944 .attach = bdrv_backing_attach,
945 .detach = bdrv_backing_detach,
946 .inherit_options = bdrv_backing_options,
947 .drained_begin = bdrv_child_cb_drained_begin,
948 .drained_end = bdrv_child_cb_drained_end,
949 .inactivate = bdrv_child_cb_inactivate,
952 static int bdrv_open_flags(BlockDriverState *bs, int flags)
954 int open_flags = flags;
957 * Clear flags that are internal to the block layer before opening the
958 * image.
960 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
963 * Snapshots should be writable.
965 if (flags & BDRV_O_TEMPORARY) {
966 open_flags |= BDRV_O_RDWR;
969 return open_flags;
972 static void update_flags_from_options(int *flags, QemuOpts *opts)
974 *flags &= ~BDRV_O_CACHE_MASK;
976 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
977 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
978 *flags |= BDRV_O_NO_FLUSH;
981 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
982 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
983 *flags |= BDRV_O_NOCACHE;
986 *flags &= ~BDRV_O_RDWR;
988 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
989 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
990 *flags |= BDRV_O_RDWR;
995 static void update_options_from_flags(QDict *options, int flags)
997 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
998 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1000 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1001 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1002 flags & BDRV_O_NO_FLUSH);
1004 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1005 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1009 static void bdrv_assign_node_name(BlockDriverState *bs,
1010 const char *node_name,
1011 Error **errp)
1013 char *gen_node_name = NULL;
1015 if (!node_name) {
1016 node_name = gen_node_name = id_generate(ID_BLOCK);
1017 } else if (!id_wellformed(node_name)) {
1019 * Check for empty string or invalid characters, but not if it is
1020 * generated (generated names use characters not available to the user)
1022 error_setg(errp, "Invalid node name");
1023 return;
1026 /* takes care of avoiding namespaces collisions */
1027 if (blk_by_name(node_name)) {
1028 error_setg(errp, "node-name=%s is conflicting with a device id",
1029 node_name);
1030 goto out;
1033 /* takes care of avoiding duplicates node names */
1034 if (bdrv_find_node(node_name)) {
1035 error_setg(errp, "Duplicate node name");
1036 goto out;
1039 /* copy node name into the bs and insert it into the graph list */
1040 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1041 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1042 out:
1043 g_free(gen_node_name);
1046 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1047 const char *node_name, QDict *options,
1048 int open_flags, Error **errp)
1050 Error *local_err = NULL;
1051 int ret;
1053 bdrv_assign_node_name(bs, node_name, &local_err);
1054 if (local_err) {
1055 error_propagate(errp, local_err);
1056 return -EINVAL;
1059 bs->drv = drv;
1060 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1061 bs->opaque = g_malloc0(drv->instance_size);
1063 if (drv->bdrv_file_open) {
1064 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1065 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1066 } else if (drv->bdrv_open) {
1067 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1068 } else {
1069 ret = 0;
1072 if (ret < 0) {
1073 if (local_err) {
1074 error_propagate(errp, local_err);
1075 } else if (bs->filename[0]) {
1076 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1077 } else {
1078 error_setg_errno(errp, -ret, "Could not open image");
1080 goto free_and_fail;
1083 ret = refresh_total_sectors(bs, bs->total_sectors);
1084 if (ret < 0) {
1085 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1086 goto free_and_fail;
1089 bdrv_refresh_limits(bs, &local_err);
1090 if (local_err) {
1091 error_propagate(errp, local_err);
1092 ret = -EINVAL;
1093 goto free_and_fail;
1096 assert(bdrv_opt_mem_align(bs) != 0);
1097 assert(bdrv_min_mem_align(bs) != 0);
1098 assert(is_power_of_2(bs->bl.request_alignment));
1100 return 0;
1102 free_and_fail:
1103 /* FIXME Close bs first if already opened*/
1104 g_free(bs->opaque);
1105 bs->opaque = NULL;
1106 bs->drv = NULL;
1107 return ret;
1110 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1111 int flags, Error **errp)
1113 BlockDriverState *bs;
1114 int ret;
1116 bs = bdrv_new();
1117 bs->open_flags = flags;
1118 bs->explicit_options = qdict_new();
1119 bs->options = qdict_new();
1120 bs->opaque = NULL;
1122 update_options_from_flags(bs->options, flags);
1124 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1125 if (ret < 0) {
1126 QDECREF(bs->explicit_options);
1127 QDECREF(bs->options);
1128 bdrv_unref(bs);
1129 return NULL;
1132 return bs;
1135 QemuOptsList bdrv_runtime_opts = {
1136 .name = "bdrv_common",
1137 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1138 .desc = {
1140 .name = "node-name",
1141 .type = QEMU_OPT_STRING,
1142 .help = "Node name of the block device node",
1145 .name = "driver",
1146 .type = QEMU_OPT_STRING,
1147 .help = "Block driver to use for the node",
1150 .name = BDRV_OPT_CACHE_DIRECT,
1151 .type = QEMU_OPT_BOOL,
1152 .help = "Bypass software writeback cache on the host",
1155 .name = BDRV_OPT_CACHE_NO_FLUSH,
1156 .type = QEMU_OPT_BOOL,
1157 .help = "Ignore flush requests",
1160 .name = BDRV_OPT_READ_ONLY,
1161 .type = QEMU_OPT_BOOL,
1162 .help = "Node is opened in read-only mode",
1165 .name = "detect-zeroes",
1166 .type = QEMU_OPT_STRING,
1167 .help = "try to optimize zero writes (off, on, unmap)",
1170 .name = "discard",
1171 .type = QEMU_OPT_STRING,
1172 .help = "discard operation (ignore/off, unmap/on)",
1175 .name = BDRV_OPT_FORCE_SHARE,
1176 .type = QEMU_OPT_BOOL,
1177 .help = "always accept other writers (default: off)",
1179 { /* end of list */ }
1184 * Common part for opening disk images and files
1186 * Removes all processed options from *options.
1188 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1189 QDict *options, Error **errp)
1191 int ret, open_flags;
1192 const char *filename;
1193 const char *driver_name = NULL;
1194 const char *node_name = NULL;
1195 const char *discard;
1196 const char *detect_zeroes;
1197 QemuOpts *opts;
1198 BlockDriver *drv;
1199 Error *local_err = NULL;
1201 assert(bs->file == NULL);
1202 assert(options != NULL && bs->options != options);
1204 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1205 qemu_opts_absorb_qdict(opts, options, &local_err);
1206 if (local_err) {
1207 error_propagate(errp, local_err);
1208 ret = -EINVAL;
1209 goto fail_opts;
1212 update_flags_from_options(&bs->open_flags, opts);
1214 driver_name = qemu_opt_get(opts, "driver");
1215 drv = bdrv_find_format(driver_name);
1216 assert(drv != NULL);
1218 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1220 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1221 error_setg(errp,
1222 BDRV_OPT_FORCE_SHARE
1223 "=on can only be used with read-only images");
1224 ret = -EINVAL;
1225 goto fail_opts;
1228 if (file != NULL) {
1229 filename = blk_bs(file)->filename;
1230 } else {
1232 * Caution: while qdict_get_try_str() is fine, getting
1233 * non-string types would require more care. When @options
1234 * come from -blockdev or blockdev_add, its members are typed
1235 * according to the QAPI schema, but when they come from
1236 * -drive, they're all QString.
1238 filename = qdict_get_try_str(options, "filename");
1241 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1242 error_setg(errp, "The '%s' block driver requires a file name",
1243 drv->format_name);
1244 ret = -EINVAL;
1245 goto fail_opts;
1248 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1249 drv->format_name);
1251 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1253 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1254 error_setg(errp,
1255 !bs->read_only && bdrv_is_whitelisted(drv, true)
1256 ? "Driver '%s' can only be used for read-only devices"
1257 : "Driver '%s' is not whitelisted",
1258 drv->format_name);
1259 ret = -ENOTSUP;
1260 goto fail_opts;
1263 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
1264 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1265 if (!bs->read_only) {
1266 bdrv_enable_copy_on_read(bs);
1267 } else {
1268 error_setg(errp, "Can't use copy-on-read on read-only device");
1269 ret = -EINVAL;
1270 goto fail_opts;
1274 discard = qemu_opt_get(opts, "discard");
1275 if (discard != NULL) {
1276 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1277 error_setg(errp, "Invalid discard option");
1278 ret = -EINVAL;
1279 goto fail_opts;
1283 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1284 if (detect_zeroes) {
1285 BlockdevDetectZeroesOptions value =
1286 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1287 detect_zeroes,
1288 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1289 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1290 &local_err);
1291 if (local_err) {
1292 error_propagate(errp, local_err);
1293 ret = -EINVAL;
1294 goto fail_opts;
1297 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1298 !(bs->open_flags & BDRV_O_UNMAP))
1300 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1301 "without setting discard operation to unmap");
1302 ret = -EINVAL;
1303 goto fail_opts;
1306 bs->detect_zeroes = value;
1309 if (filename != NULL) {
1310 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1311 } else {
1312 bs->filename[0] = '\0';
1314 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1316 /* Open the image, either directly or using a protocol */
1317 open_flags = bdrv_open_flags(bs, bs->open_flags);
1318 node_name = qemu_opt_get(opts, "node-name");
1320 assert(!drv->bdrv_file_open || file == NULL);
1321 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1322 if (ret < 0) {
1323 goto fail_opts;
1326 qemu_opts_del(opts);
1327 return 0;
1329 fail_opts:
1330 qemu_opts_del(opts);
1331 return ret;
1334 static QDict *parse_json_filename(const char *filename, Error **errp)
1336 QObject *options_obj;
1337 QDict *options;
1338 int ret;
1340 ret = strstart(filename, "json:", &filename);
1341 assert(ret);
1343 options_obj = qobject_from_json(filename, errp);
1344 if (!options_obj) {
1345 /* Work around qobject_from_json() lossage TODO fix that */
1346 if (errp && !*errp) {
1347 error_setg(errp, "Could not parse the JSON options");
1348 return NULL;
1350 error_prepend(errp, "Could not parse the JSON options: ");
1351 return NULL;
1354 options = qobject_to_qdict(options_obj);
1355 if (!options) {
1356 qobject_decref(options_obj);
1357 error_setg(errp, "Invalid JSON object given");
1358 return NULL;
1361 qdict_flatten(options);
1363 return options;
1366 static void parse_json_protocol(QDict *options, const char **pfilename,
1367 Error **errp)
1369 QDict *json_options;
1370 Error *local_err = NULL;
1372 /* Parse json: pseudo-protocol */
1373 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1374 return;
1377 json_options = parse_json_filename(*pfilename, &local_err);
1378 if (local_err) {
1379 error_propagate(errp, local_err);
1380 return;
1383 /* Options given in the filename have lower priority than options
1384 * specified directly */
1385 qdict_join(options, json_options, false);
1386 QDECREF(json_options);
1387 *pfilename = NULL;
1391 * Fills in default options for opening images and converts the legacy
1392 * filename/flags pair to option QDict entries.
1393 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1394 * block driver has been specified explicitly.
1396 static int bdrv_fill_options(QDict **options, const char *filename,
1397 int *flags, Error **errp)
1399 const char *drvname;
1400 bool protocol = *flags & BDRV_O_PROTOCOL;
1401 bool parse_filename = false;
1402 BlockDriver *drv = NULL;
1403 Error *local_err = NULL;
1406 * Caution: while qdict_get_try_str() is fine, getting non-string
1407 * types would require more care. When @options come from
1408 * -blockdev or blockdev_add, its members are typed according to
1409 * the QAPI schema, but when they come from -drive, they're all
1410 * QString.
1412 drvname = qdict_get_try_str(*options, "driver");
1413 if (drvname) {
1414 drv = bdrv_find_format(drvname);
1415 if (!drv) {
1416 error_setg(errp, "Unknown driver '%s'", drvname);
1417 return -ENOENT;
1419 /* If the user has explicitly specified the driver, this choice should
1420 * override the BDRV_O_PROTOCOL flag */
1421 protocol = drv->bdrv_file_open;
1424 if (protocol) {
1425 *flags |= BDRV_O_PROTOCOL;
1426 } else {
1427 *flags &= ~BDRV_O_PROTOCOL;
1430 /* Translate cache options from flags into options */
1431 update_options_from_flags(*options, *flags);
1433 /* Fetch the file name from the options QDict if necessary */
1434 if (protocol && filename) {
1435 if (!qdict_haskey(*options, "filename")) {
1436 qdict_put_str(*options, "filename", filename);
1437 parse_filename = true;
1438 } else {
1439 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1440 "the same time");
1441 return -EINVAL;
1445 /* Find the right block driver */
1446 /* See cautionary note on accessing @options above */
1447 filename = qdict_get_try_str(*options, "filename");
1449 if (!drvname && protocol) {
1450 if (filename) {
1451 drv = bdrv_find_protocol(filename, parse_filename, errp);
1452 if (!drv) {
1453 return -EINVAL;
1456 drvname = drv->format_name;
1457 qdict_put_str(*options, "driver", drvname);
1458 } else {
1459 error_setg(errp, "Must specify either driver or file");
1460 return -EINVAL;
1464 assert(drv || !protocol);
1466 /* Driver-specific filename parsing */
1467 if (drv && drv->bdrv_parse_filename && parse_filename) {
1468 drv->bdrv_parse_filename(filename, *options, &local_err);
1469 if (local_err) {
1470 error_propagate(errp, local_err);
1471 return -EINVAL;
1474 if (!drv->bdrv_needs_filename) {
1475 qdict_del(*options, "filename");
1479 return 0;
1482 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1483 GSList *ignore_children, Error **errp);
1484 static void bdrv_child_abort_perm_update(BdrvChild *c);
1485 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1487 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1488 BdrvChild *c,
1489 const BdrvChildRole *role,
1490 uint64_t parent_perm, uint64_t parent_shared,
1491 uint64_t *nperm, uint64_t *nshared)
1493 if (bs->drv && bs->drv->bdrv_child_perm) {
1494 bs->drv->bdrv_child_perm(bs, c, role,
1495 parent_perm, parent_shared,
1496 nperm, nshared);
1498 if (child_bs && child_bs->force_share) {
1499 *nshared = BLK_PERM_ALL;
1504 * Check whether permissions on this node can be changed in a way that
1505 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1506 * permissions of all its parents. This involves checking whether all necessary
1507 * permission changes to child nodes can be performed.
1509 * A call to this function must always be followed by a call to bdrv_set_perm()
1510 * or bdrv_abort_perm_update().
1512 static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1513 uint64_t cumulative_shared_perms,
1514 GSList *ignore_children, Error **errp)
1516 BlockDriver *drv = bs->drv;
1517 BdrvChild *c;
1518 int ret;
1520 /* Write permissions never work with read-only images */
1521 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1522 !bdrv_is_writable(bs))
1524 error_setg(errp, "Block node is read-only");
1525 return -EPERM;
1528 /* Check this node */
1529 if (!drv) {
1530 return 0;
1533 if (drv->bdrv_check_perm) {
1534 return drv->bdrv_check_perm(bs, cumulative_perms,
1535 cumulative_shared_perms, errp);
1538 /* Drivers that never have children can omit .bdrv_child_perm() */
1539 if (!drv->bdrv_child_perm) {
1540 assert(QLIST_EMPTY(&bs->children));
1541 return 0;
1544 /* Check all children */
1545 QLIST_FOREACH(c, &bs->children, next) {
1546 uint64_t cur_perm, cur_shared;
1547 bdrv_child_perm(bs, c->bs, c, c->role,
1548 cumulative_perms, cumulative_shared_perms,
1549 &cur_perm, &cur_shared);
1550 ret = bdrv_child_check_perm(c, cur_perm, cur_shared, ignore_children,
1551 errp);
1552 if (ret < 0) {
1553 return ret;
1557 return 0;
1561 * Notifies drivers that after a previous bdrv_check_perm() call, the
1562 * permission update is not performed and any preparations made for it (e.g.
1563 * taken file locks) need to be undone.
1565 * This function recursively notifies all child nodes.
1567 static void bdrv_abort_perm_update(BlockDriverState *bs)
1569 BlockDriver *drv = bs->drv;
1570 BdrvChild *c;
1572 if (!drv) {
1573 return;
1576 if (drv->bdrv_abort_perm_update) {
1577 drv->bdrv_abort_perm_update(bs);
1580 QLIST_FOREACH(c, &bs->children, next) {
1581 bdrv_child_abort_perm_update(c);
1585 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1586 uint64_t cumulative_shared_perms)
1588 BlockDriver *drv = bs->drv;
1589 BdrvChild *c;
1591 if (!drv) {
1592 return;
1595 /* Update this node */
1596 if (drv->bdrv_set_perm) {
1597 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1600 /* Drivers that never have children can omit .bdrv_child_perm() */
1601 if (!drv->bdrv_child_perm) {
1602 assert(QLIST_EMPTY(&bs->children));
1603 return;
1606 /* Update all children */
1607 QLIST_FOREACH(c, &bs->children, next) {
1608 uint64_t cur_perm, cur_shared;
1609 bdrv_child_perm(bs, c->bs, c, c->role,
1610 cumulative_perms, cumulative_shared_perms,
1611 &cur_perm, &cur_shared);
1612 bdrv_child_set_perm(c, cur_perm, cur_shared);
1616 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1617 uint64_t *shared_perm)
1619 BdrvChild *c;
1620 uint64_t cumulative_perms = 0;
1621 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1623 QLIST_FOREACH(c, &bs->parents, next_parent) {
1624 cumulative_perms |= c->perm;
1625 cumulative_shared_perms &= c->shared_perm;
1628 *perm = cumulative_perms;
1629 *shared_perm = cumulative_shared_perms;
1632 static char *bdrv_child_user_desc(BdrvChild *c)
1634 if (c->role->get_parent_desc) {
1635 return c->role->get_parent_desc(c);
1638 return g_strdup("another user");
1641 char *bdrv_perm_names(uint64_t perm)
1643 struct perm_name {
1644 uint64_t perm;
1645 const char *name;
1646 } permissions[] = {
1647 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1648 { BLK_PERM_WRITE, "write" },
1649 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1650 { BLK_PERM_RESIZE, "resize" },
1651 { BLK_PERM_GRAPH_MOD, "change children" },
1652 { 0, NULL }
1655 char *result = g_strdup("");
1656 struct perm_name *p;
1658 for (p = permissions; p->name; p++) {
1659 if (perm & p->perm) {
1660 char *old = result;
1661 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1662 g_free(old);
1666 return result;
1670 * Checks whether a new reference to @bs can be added if the new user requires
1671 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1672 * set, the BdrvChild objects in this list are ignored in the calculations;
1673 * this allows checking permission updates for an existing reference.
1675 * Needs to be followed by a call to either bdrv_set_perm() or
1676 * bdrv_abort_perm_update(). */
1677 static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm,
1678 uint64_t new_shared_perm,
1679 GSList *ignore_children, Error **errp)
1681 BdrvChild *c;
1682 uint64_t cumulative_perms = new_used_perm;
1683 uint64_t cumulative_shared_perms = new_shared_perm;
1685 /* There is no reason why anyone couldn't tolerate write_unchanged */
1686 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1688 QLIST_FOREACH(c, &bs->parents, next_parent) {
1689 if (g_slist_find(ignore_children, c)) {
1690 continue;
1693 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1694 char *user = bdrv_child_user_desc(c);
1695 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1696 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1697 "allow '%s' on %s",
1698 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1699 g_free(user);
1700 g_free(perm_names);
1701 return -EPERM;
1704 if ((c->perm & new_shared_perm) != c->perm) {
1705 char *user = bdrv_child_user_desc(c);
1706 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1707 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1708 "'%s' on %s",
1709 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1710 g_free(user);
1711 g_free(perm_names);
1712 return -EPERM;
1715 cumulative_perms |= c->perm;
1716 cumulative_shared_perms &= c->shared_perm;
1719 return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms,
1720 ignore_children, errp);
1723 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1724 * bdrv_child_abort_perm_update(). */
1725 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1726 GSList *ignore_children, Error **errp)
1728 int ret;
1730 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1731 ret = bdrv_check_update_perm(c->bs, perm, shared, ignore_children, errp);
1732 g_slist_free(ignore_children);
1734 return ret;
1737 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1739 uint64_t cumulative_perms, cumulative_shared_perms;
1741 c->perm = perm;
1742 c->shared_perm = shared;
1744 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1745 &cumulative_shared_perms);
1746 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1749 static void bdrv_child_abort_perm_update(BdrvChild *c)
1751 bdrv_abort_perm_update(c->bs);
1754 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1755 Error **errp)
1757 int ret;
1759 ret = bdrv_child_check_perm(c, perm, shared, NULL, errp);
1760 if (ret < 0) {
1761 bdrv_child_abort_perm_update(c);
1762 return ret;
1765 bdrv_child_set_perm(c, perm, shared);
1767 return 0;
1770 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1771 | BLK_PERM_WRITE \
1772 | BLK_PERM_WRITE_UNCHANGED \
1773 | BLK_PERM_RESIZE)
1774 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1776 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1777 const BdrvChildRole *role,
1778 uint64_t perm, uint64_t shared,
1779 uint64_t *nperm, uint64_t *nshared)
1781 if (c == NULL) {
1782 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1783 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1784 return;
1787 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1788 (c->perm & DEFAULT_PERM_UNCHANGED);
1789 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1790 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1793 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1794 const BdrvChildRole *role,
1795 uint64_t perm, uint64_t shared,
1796 uint64_t *nperm, uint64_t *nshared)
1798 bool backing = (role == &child_backing);
1799 assert(role == &child_backing || role == &child_file);
1801 if (!backing) {
1802 /* Apart from the modifications below, the same permissions are
1803 * forwarded and left alone as for filters */
1804 bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared);
1806 /* Format drivers may touch metadata even if the guest doesn't write */
1807 if (bdrv_is_writable(bs)) {
1808 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1811 /* bs->file always needs to be consistent because of the metadata. We
1812 * can never allow other users to resize or write to it. */
1813 perm |= BLK_PERM_CONSISTENT_READ;
1814 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1815 } else {
1816 /* We want consistent read from backing files if the parent needs it.
1817 * No other operations are performed on backing files. */
1818 perm &= BLK_PERM_CONSISTENT_READ;
1820 /* If the parent can deal with changing data, we're okay with a
1821 * writable and resizable backing file. */
1822 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1823 if (shared & BLK_PERM_WRITE) {
1824 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1825 } else {
1826 shared = 0;
1829 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1830 BLK_PERM_WRITE_UNCHANGED;
1833 if (bs->open_flags & BDRV_O_INACTIVE) {
1834 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1837 *nperm = perm;
1838 *nshared = shared;
1841 static void bdrv_replace_child_noperm(BdrvChild *child,
1842 BlockDriverState *new_bs)
1844 BlockDriverState *old_bs = child->bs;
1846 if (old_bs && new_bs) {
1847 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1849 if (old_bs) {
1850 if (old_bs->quiesce_counter && child->role->drained_end) {
1851 child->role->drained_end(child);
1853 if (child->role->detach) {
1854 child->role->detach(child);
1856 QLIST_REMOVE(child, next_parent);
1859 child->bs = new_bs;
1861 if (new_bs) {
1862 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1863 if (new_bs->quiesce_counter && child->role->drained_begin) {
1864 child->role->drained_begin(child);
1867 if (child->role->attach) {
1868 child->role->attach(child);
1874 * Updates @child to change its reference to point to @new_bs, including
1875 * checking and applying the necessary permisson updates both to the old node
1876 * and to @new_bs.
1878 * NULL is passed as @new_bs for removing the reference before freeing @child.
1880 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1881 * function uses bdrv_set_perm() to update the permissions according to the new
1882 * reference that @new_bs gets.
1884 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1886 BlockDriverState *old_bs = child->bs;
1887 uint64_t perm, shared_perm;
1889 if (old_bs) {
1890 /* Update permissions for old node. This is guaranteed to succeed
1891 * because we're just taking a parent away, so we're loosening
1892 * restrictions. */
1893 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1894 bdrv_check_perm(old_bs, perm, shared_perm, NULL, &error_abort);
1895 bdrv_set_perm(old_bs, perm, shared_perm);
1898 bdrv_replace_child_noperm(child, new_bs);
1900 if (new_bs) {
1901 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
1902 bdrv_set_perm(new_bs, perm, shared_perm);
1906 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1907 const char *child_name,
1908 const BdrvChildRole *child_role,
1909 uint64_t perm, uint64_t shared_perm,
1910 void *opaque, Error **errp)
1912 BdrvChild *child;
1913 int ret;
1915 ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp);
1916 if (ret < 0) {
1917 bdrv_abort_perm_update(child_bs);
1918 return NULL;
1921 child = g_new(BdrvChild, 1);
1922 *child = (BdrvChild) {
1923 .bs = NULL,
1924 .name = g_strdup(child_name),
1925 .role = child_role,
1926 .perm = perm,
1927 .shared_perm = shared_perm,
1928 .opaque = opaque,
1931 /* This performs the matching bdrv_set_perm() for the above check. */
1932 bdrv_replace_child(child, child_bs);
1934 return child;
1937 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1938 BlockDriverState *child_bs,
1939 const char *child_name,
1940 const BdrvChildRole *child_role,
1941 Error **errp)
1943 BdrvChild *child;
1944 uint64_t perm, shared_perm;
1946 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
1948 assert(parent_bs->drv);
1949 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
1950 bdrv_child_perm(parent_bs, child_bs, NULL, child_role,
1951 perm, shared_perm, &perm, &shared_perm);
1953 child = bdrv_root_attach_child(child_bs, child_name, child_role,
1954 perm, shared_perm, parent_bs, errp);
1955 if (child == NULL) {
1956 return NULL;
1959 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1960 return child;
1963 static void bdrv_detach_child(BdrvChild *child)
1965 if (child->next.le_prev) {
1966 QLIST_REMOVE(child, next);
1967 child->next.le_prev = NULL;
1970 bdrv_replace_child(child, NULL);
1972 g_free(child->name);
1973 g_free(child);
1976 void bdrv_root_unref_child(BdrvChild *child)
1978 BlockDriverState *child_bs;
1980 child_bs = child->bs;
1981 bdrv_detach_child(child);
1982 bdrv_unref(child_bs);
1985 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1987 if (child == NULL) {
1988 return;
1991 if (child->bs->inherits_from == parent) {
1992 BdrvChild *c;
1994 /* Remove inherits_from only when the last reference between parent and
1995 * child->bs goes away. */
1996 QLIST_FOREACH(c, &parent->children, next) {
1997 if (c != child && c->bs == child->bs) {
1998 break;
2001 if (c == NULL) {
2002 child->bs->inherits_from = NULL;
2006 bdrv_root_unref_child(child);
2010 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2012 BdrvChild *c;
2013 QLIST_FOREACH(c, &bs->parents, next_parent) {
2014 if (c->role->change_media) {
2015 c->role->change_media(c, load);
2020 static void bdrv_parent_cb_resize(BlockDriverState *bs)
2022 BdrvChild *c;
2023 QLIST_FOREACH(c, &bs->parents, next_parent) {
2024 if (c->role->resize) {
2025 c->role->resize(c);
2031 * Sets the backing file link of a BDS. A new reference is created; callers
2032 * which don't need their own reference any more must call bdrv_unref().
2034 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2035 Error **errp)
2037 if (backing_hd) {
2038 bdrv_ref(backing_hd);
2041 if (bs->backing) {
2042 bdrv_unref_child(bs, bs->backing);
2045 if (!backing_hd) {
2046 bs->backing = NULL;
2047 goto out;
2050 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2051 errp);
2052 if (!bs->backing) {
2053 bdrv_unref(backing_hd);
2056 bdrv_refresh_filename(bs);
2058 out:
2059 bdrv_refresh_limits(bs, NULL);
2063 * Opens the backing file for a BlockDriverState if not yet open
2065 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2066 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2067 * itself, all options starting with "${bdref_key}." are considered part of the
2068 * BlockdevRef.
2070 * TODO Can this be unified with bdrv_open_image()?
2072 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2073 const char *bdref_key, Error **errp)
2075 char *backing_filename = g_malloc0(PATH_MAX);
2076 char *bdref_key_dot;
2077 const char *reference = NULL;
2078 int ret = 0;
2079 BlockDriverState *backing_hd;
2080 QDict *options;
2081 QDict *tmp_parent_options = NULL;
2082 Error *local_err = NULL;
2084 if (bs->backing != NULL) {
2085 goto free_exit;
2088 /* NULL means an empty set of options */
2089 if (parent_options == NULL) {
2090 tmp_parent_options = qdict_new();
2091 parent_options = tmp_parent_options;
2094 bs->open_flags &= ~BDRV_O_NO_BACKING;
2096 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2097 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2098 g_free(bdref_key_dot);
2101 * Caution: while qdict_get_try_str() is fine, getting non-string
2102 * types would require more care. When @parent_options come from
2103 * -blockdev or blockdev_add, its members are typed according to
2104 * the QAPI schema, but when they come from -drive, they're all
2105 * QString.
2107 reference = qdict_get_try_str(parent_options, bdref_key);
2108 if (reference || qdict_haskey(options, "file.filename")) {
2109 backing_filename[0] = '\0';
2110 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2111 QDECREF(options);
2112 goto free_exit;
2113 } else {
2114 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2115 &local_err);
2116 if (local_err) {
2117 ret = -EINVAL;
2118 error_propagate(errp, local_err);
2119 QDECREF(options);
2120 goto free_exit;
2124 if (!bs->drv || !bs->drv->supports_backing) {
2125 ret = -EINVAL;
2126 error_setg(errp, "Driver doesn't support backing files");
2127 QDECREF(options);
2128 goto free_exit;
2131 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2132 qdict_put_str(options, "driver", bs->backing_format);
2135 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2136 reference, options, 0, bs, &child_backing,
2137 errp);
2138 if (!backing_hd) {
2139 bs->open_flags |= BDRV_O_NO_BACKING;
2140 error_prepend(errp, "Could not open backing file: ");
2141 ret = -EINVAL;
2142 goto free_exit;
2145 /* Hook up the backing file link; drop our reference, bs owns the
2146 * backing_hd reference now */
2147 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2148 bdrv_unref(backing_hd);
2149 if (local_err) {
2150 error_propagate(errp, local_err);
2151 ret = -EINVAL;
2152 goto free_exit;
2155 qdict_del(parent_options, bdref_key);
2157 free_exit:
2158 g_free(backing_filename);
2159 QDECREF(tmp_parent_options);
2160 return ret;
2163 static BlockDriverState *
2164 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2165 BlockDriverState *parent, const BdrvChildRole *child_role,
2166 bool allow_none, Error **errp)
2168 BlockDriverState *bs = NULL;
2169 QDict *image_options;
2170 char *bdref_key_dot;
2171 const char *reference;
2173 assert(child_role != NULL);
2175 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2176 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2177 g_free(bdref_key_dot);
2180 * Caution: while qdict_get_try_str() is fine, getting non-string
2181 * types would require more care. When @options come from
2182 * -blockdev or blockdev_add, its members are typed according to
2183 * the QAPI schema, but when they come from -drive, they're all
2184 * QString.
2186 reference = qdict_get_try_str(options, bdref_key);
2187 if (!filename && !reference && !qdict_size(image_options)) {
2188 if (!allow_none) {
2189 error_setg(errp, "A block device must be specified for \"%s\"",
2190 bdref_key);
2192 QDECREF(image_options);
2193 goto done;
2196 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2197 parent, child_role, errp);
2198 if (!bs) {
2199 goto done;
2202 done:
2203 qdict_del(options, bdref_key);
2204 return bs;
2208 * Opens a disk image whose options are given as BlockdevRef in another block
2209 * device's options.
2211 * If allow_none is true, no image will be opened if filename is false and no
2212 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2214 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2215 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2216 * itself, all options starting with "${bdref_key}." are considered part of the
2217 * BlockdevRef.
2219 * The BlockdevRef will be removed from the options QDict.
2221 BdrvChild *bdrv_open_child(const char *filename,
2222 QDict *options, const char *bdref_key,
2223 BlockDriverState *parent,
2224 const BdrvChildRole *child_role,
2225 bool allow_none, Error **errp)
2227 BdrvChild *c;
2228 BlockDriverState *bs;
2230 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2231 allow_none, errp);
2232 if (bs == NULL) {
2233 return NULL;
2236 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2237 if (!c) {
2238 bdrv_unref(bs);
2239 return NULL;
2242 return c;
2245 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2246 int flags,
2247 QDict *snapshot_options,
2248 Error **errp)
2250 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2251 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2252 int64_t total_size;
2253 QemuOpts *opts = NULL;
2254 BlockDriverState *bs_snapshot = NULL;
2255 Error *local_err = NULL;
2256 int ret;
2258 /* if snapshot, we create a temporary backing file and open it
2259 instead of opening 'filename' directly */
2261 /* Get the required size from the image */
2262 total_size = bdrv_getlength(bs);
2263 if (total_size < 0) {
2264 error_setg_errno(errp, -total_size, "Could not get image size");
2265 goto out;
2268 /* Create the temporary image */
2269 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2270 if (ret < 0) {
2271 error_setg_errno(errp, -ret, "Could not get temporary filename");
2272 goto out;
2275 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2276 &error_abort);
2277 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2278 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2279 qemu_opts_del(opts);
2280 if (ret < 0) {
2281 error_prepend(errp, "Could not create temporary overlay '%s': ",
2282 tmp_filename);
2283 goto out;
2286 /* Prepare options QDict for the temporary file */
2287 qdict_put_str(snapshot_options, "file.driver", "file");
2288 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2289 qdict_put_str(snapshot_options, "driver", "qcow2");
2291 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2292 snapshot_options = NULL;
2293 if (!bs_snapshot) {
2294 goto out;
2297 /* bdrv_append() consumes a strong reference to bs_snapshot
2298 * (i.e. it will call bdrv_unref() on it) even on error, so in
2299 * order to be able to return one, we have to increase
2300 * bs_snapshot's refcount here */
2301 bdrv_ref(bs_snapshot);
2302 bdrv_append(bs_snapshot, bs, &local_err);
2303 if (local_err) {
2304 error_propagate(errp, local_err);
2305 bs_snapshot = NULL;
2306 goto out;
2309 out:
2310 QDECREF(snapshot_options);
2311 g_free(tmp_filename);
2312 return bs_snapshot;
2316 * Opens a disk image (raw, qcow2, vmdk, ...)
2318 * options is a QDict of options to pass to the block drivers, or NULL for an
2319 * empty set of options. The reference to the QDict belongs to the block layer
2320 * after the call (even on failure), so if the caller intends to reuse the
2321 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2323 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2324 * If it is not NULL, the referenced BDS will be reused.
2326 * The reference parameter may be used to specify an existing block device which
2327 * should be opened. If specified, neither options nor a filename may be given,
2328 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2330 static BlockDriverState *bdrv_open_inherit(const char *filename,
2331 const char *reference,
2332 QDict *options, int flags,
2333 BlockDriverState *parent,
2334 const BdrvChildRole *child_role,
2335 Error **errp)
2337 int ret;
2338 BlockBackend *file = NULL;
2339 BlockDriverState *bs;
2340 BlockDriver *drv = NULL;
2341 const char *drvname;
2342 const char *backing;
2343 Error *local_err = NULL;
2344 QDict *snapshot_options = NULL;
2345 int snapshot_flags = 0;
2347 assert(!child_role || !flags);
2348 assert(!child_role == !parent);
2350 if (reference) {
2351 bool options_non_empty = options ? qdict_size(options) : false;
2352 QDECREF(options);
2354 if (filename || options_non_empty) {
2355 error_setg(errp, "Cannot reference an existing block device with "
2356 "additional options or a new filename");
2357 return NULL;
2360 bs = bdrv_lookup_bs(reference, reference, errp);
2361 if (!bs) {
2362 return NULL;
2365 bdrv_ref(bs);
2366 return bs;
2369 bs = bdrv_new();
2371 /* NULL means an empty set of options */
2372 if (options == NULL) {
2373 options = qdict_new();
2376 /* json: syntax counts as explicit options, as if in the QDict */
2377 parse_json_protocol(options, &filename, &local_err);
2378 if (local_err) {
2379 goto fail;
2382 bs->explicit_options = qdict_clone_shallow(options);
2384 if (child_role) {
2385 bs->inherits_from = parent;
2386 child_role->inherit_options(&flags, options,
2387 parent->open_flags, parent->options);
2390 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2391 if (local_err) {
2392 goto fail;
2396 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2397 * Caution: getting a boolean member of @options requires care.
2398 * When @options come from -blockdev or blockdev_add, members are
2399 * typed according to the QAPI schema, but when they come from
2400 * -drive, they're all QString.
2402 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2403 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2404 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2405 } else {
2406 flags &= ~BDRV_O_RDWR;
2409 if (flags & BDRV_O_SNAPSHOT) {
2410 snapshot_options = qdict_new();
2411 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2412 flags, options);
2413 /* Let bdrv_backing_options() override "read-only" */
2414 qdict_del(options, BDRV_OPT_READ_ONLY);
2415 bdrv_backing_options(&flags, options, flags, options);
2418 bs->open_flags = flags;
2419 bs->options = options;
2420 options = qdict_clone_shallow(options);
2422 /* Find the right image format driver */
2423 /* See cautionary note on accessing @options above */
2424 drvname = qdict_get_try_str(options, "driver");
2425 if (drvname) {
2426 drv = bdrv_find_format(drvname);
2427 if (!drv) {
2428 error_setg(errp, "Unknown driver: '%s'", drvname);
2429 goto fail;
2433 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2435 /* See cautionary note on accessing @options above */
2436 backing = qdict_get_try_str(options, "backing");
2437 if (backing && *backing == '\0') {
2438 flags |= BDRV_O_NO_BACKING;
2439 qdict_del(options, "backing");
2442 /* Open image file without format layer. This BlockBackend is only used for
2443 * probing, the block drivers will do their own bdrv_open_child() for the
2444 * same BDS, which is why we put the node name back into options. */
2445 if ((flags & BDRV_O_PROTOCOL) == 0) {
2446 BlockDriverState *file_bs;
2448 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2449 &child_file, true, &local_err);
2450 if (local_err) {
2451 goto fail;
2453 if (file_bs != NULL) {
2454 file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
2455 blk_insert_bs(file, file_bs, &local_err);
2456 bdrv_unref(file_bs);
2457 if (local_err) {
2458 goto fail;
2461 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2465 /* Image format probing */
2466 bs->probed = !drv;
2467 if (!drv && file) {
2468 ret = find_image_format(file, filename, &drv, &local_err);
2469 if (ret < 0) {
2470 goto fail;
2473 * This option update would logically belong in bdrv_fill_options(),
2474 * but we first need to open bs->file for the probing to work, while
2475 * opening bs->file already requires the (mostly) final set of options
2476 * so that cache mode etc. can be inherited.
2478 * Adding the driver later is somewhat ugly, but it's not an option
2479 * that would ever be inherited, so it's correct. We just need to make
2480 * sure to update both bs->options (which has the full effective
2481 * options for bs) and options (which has file.* already removed).
2483 qdict_put_str(bs->options, "driver", drv->format_name);
2484 qdict_put_str(options, "driver", drv->format_name);
2485 } else if (!drv) {
2486 error_setg(errp, "Must specify either driver or file");
2487 goto fail;
2490 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2491 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2492 /* file must be NULL if a protocol BDS is about to be created
2493 * (the inverse results in an error message from bdrv_open_common()) */
2494 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2496 /* Open the image */
2497 ret = bdrv_open_common(bs, file, options, &local_err);
2498 if (ret < 0) {
2499 goto fail;
2502 if (file) {
2503 blk_unref(file);
2504 file = NULL;
2507 /* If there is a backing file, use it */
2508 if ((flags & BDRV_O_NO_BACKING) == 0) {
2509 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2510 if (ret < 0) {
2511 goto close_and_fail;
2515 bdrv_refresh_filename(bs);
2517 /* Check if any unknown options were used */
2518 if (qdict_size(options) != 0) {
2519 const QDictEntry *entry = qdict_first(options);
2520 if (flags & BDRV_O_PROTOCOL) {
2521 error_setg(errp, "Block protocol '%s' doesn't support the option "
2522 "'%s'", drv->format_name, entry->key);
2523 } else {
2524 error_setg(errp,
2525 "Block format '%s' does not support the option '%s'",
2526 drv->format_name, entry->key);
2529 goto close_and_fail;
2532 if (!bdrv_key_required(bs)) {
2533 bdrv_parent_cb_change_media(bs, true);
2534 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
2535 && !runstate_check(RUN_STATE_INMIGRATE)
2536 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
2537 error_setg(errp,
2538 "Guest must be stopped for opening of encrypted image");
2539 goto close_and_fail;
2542 QDECREF(options);
2544 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2545 * temporary snapshot afterwards. */
2546 if (snapshot_flags) {
2547 BlockDriverState *snapshot_bs;
2548 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2549 snapshot_options, &local_err);
2550 snapshot_options = NULL;
2551 if (local_err) {
2552 goto close_and_fail;
2554 /* We are not going to return bs but the overlay on top of it
2555 * (snapshot_bs); thus, we have to drop the strong reference to bs
2556 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2557 * though, because the overlay still has a reference to it. */
2558 bdrv_unref(bs);
2559 bs = snapshot_bs;
2562 return bs;
2564 fail:
2565 blk_unref(file);
2566 if (bs->file != NULL) {
2567 bdrv_unref_child(bs, bs->file);
2569 QDECREF(snapshot_options);
2570 QDECREF(bs->explicit_options);
2571 QDECREF(bs->options);
2572 QDECREF(options);
2573 bs->options = NULL;
2574 bdrv_unref(bs);
2575 error_propagate(errp, local_err);
2576 return NULL;
2578 close_and_fail:
2579 bdrv_unref(bs);
2580 QDECREF(snapshot_options);
2581 QDECREF(options);
2582 error_propagate(errp, local_err);
2583 return NULL;
2586 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2587 QDict *options, int flags, Error **errp)
2589 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2590 NULL, errp);
2593 typedef struct BlockReopenQueueEntry {
2594 bool prepared;
2595 BDRVReopenState state;
2596 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2597 } BlockReopenQueueEntry;
2600 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2601 * reopen of multiple devices.
2603 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2604 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2605 * be created and initialized. This newly created BlockReopenQueue should be
2606 * passed back in for subsequent calls that are intended to be of the same
2607 * atomic 'set'.
2609 * bs is the BlockDriverState to add to the reopen queue.
2611 * options contains the changed options for the associated bs
2612 * (the BlockReopenQueue takes ownership)
2614 * flags contains the open flags for the associated bs
2616 * returns a pointer to bs_queue, which is either the newly allocated
2617 * bs_queue, or the existing bs_queue being used.
2620 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2621 BlockDriverState *bs,
2622 QDict *options,
2623 int flags,
2624 const BdrvChildRole *role,
2625 QDict *parent_options,
2626 int parent_flags)
2628 assert(bs != NULL);
2630 BlockReopenQueueEntry *bs_entry;
2631 BdrvChild *child;
2632 QDict *old_options, *explicit_options;
2634 if (bs_queue == NULL) {
2635 bs_queue = g_new0(BlockReopenQueue, 1);
2636 QSIMPLEQ_INIT(bs_queue);
2639 if (!options) {
2640 options = qdict_new();
2643 /* Check if this BlockDriverState is already in the queue */
2644 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2645 if (bs == bs_entry->state.bs) {
2646 break;
2651 * Precedence of options:
2652 * 1. Explicitly passed in options (highest)
2653 * 2. Set in flags (only for top level)
2654 * 3. Retained from explicitly set options of bs
2655 * 4. Inherited from parent node
2656 * 5. Retained from effective options of bs
2659 if (!parent_options) {
2661 * Any setting represented by flags is always updated. If the
2662 * corresponding QDict option is set, it takes precedence. Otherwise
2663 * the flag is translated into a QDict option. The old setting of bs is
2664 * not considered.
2666 update_options_from_flags(options, flags);
2669 /* Old explicitly set values (don't overwrite by inherited value) */
2670 if (bs_entry) {
2671 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2672 } else {
2673 old_options = qdict_clone_shallow(bs->explicit_options);
2675 bdrv_join_options(bs, options, old_options);
2676 QDECREF(old_options);
2678 explicit_options = qdict_clone_shallow(options);
2680 /* Inherit from parent node */
2681 if (parent_options) {
2682 assert(!flags);
2683 role->inherit_options(&flags, options, parent_flags, parent_options);
2686 /* Old values are used for options that aren't set yet */
2687 old_options = qdict_clone_shallow(bs->options);
2688 bdrv_join_options(bs, options, old_options);
2689 QDECREF(old_options);
2691 /* bdrv_open() masks this flag out */
2692 flags &= ~BDRV_O_PROTOCOL;
2694 QLIST_FOREACH(child, &bs->children, next) {
2695 QDict *new_child_options;
2696 char *child_key_dot;
2698 /* reopen can only change the options of block devices that were
2699 * implicitly created and inherited options. For other (referenced)
2700 * block devices, a syntax like "backing.foo" results in an error. */
2701 if (child->bs->inherits_from != bs) {
2702 continue;
2705 child_key_dot = g_strdup_printf("%s.", child->name);
2706 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2707 g_free(child_key_dot);
2709 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2710 child->role, options, flags);
2713 if (!bs_entry) {
2714 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2715 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2716 } else {
2717 QDECREF(bs_entry->state.options);
2718 QDECREF(bs_entry->state.explicit_options);
2721 bs_entry->state.bs = bs;
2722 bs_entry->state.options = options;
2723 bs_entry->state.explicit_options = explicit_options;
2724 bs_entry->state.flags = flags;
2726 return bs_queue;
2729 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2730 BlockDriverState *bs,
2731 QDict *options, int flags)
2733 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2734 NULL, NULL, 0);
2738 * Reopen multiple BlockDriverStates atomically & transactionally.
2740 * The queue passed in (bs_queue) must have been built up previous
2741 * via bdrv_reopen_queue().
2743 * Reopens all BDS specified in the queue, with the appropriate
2744 * flags. All devices are prepared for reopen, and failure of any
2745 * device will cause all device changes to be abandonded, and intermediate
2746 * data cleaned up.
2748 * If all devices prepare successfully, then the changes are committed
2749 * to all devices.
2752 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2754 int ret = -1;
2755 BlockReopenQueueEntry *bs_entry, *next;
2756 Error *local_err = NULL;
2758 assert(bs_queue != NULL);
2760 aio_context_release(ctx);
2761 bdrv_drain_all_begin();
2762 aio_context_acquire(ctx);
2764 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2765 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2766 error_propagate(errp, local_err);
2767 goto cleanup;
2769 bs_entry->prepared = true;
2772 /* If we reach this point, we have success and just need to apply the
2773 * changes
2775 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2776 bdrv_reopen_commit(&bs_entry->state);
2779 ret = 0;
2781 cleanup:
2782 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2783 if (ret && bs_entry->prepared) {
2784 bdrv_reopen_abort(&bs_entry->state);
2785 } else if (ret) {
2786 QDECREF(bs_entry->state.explicit_options);
2788 QDECREF(bs_entry->state.options);
2789 g_free(bs_entry);
2791 g_free(bs_queue);
2793 bdrv_drain_all_end();
2795 return ret;
2799 /* Reopen a single BlockDriverState with the specified flags. */
2800 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2802 int ret = -1;
2803 Error *local_err = NULL;
2804 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2806 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2807 if (local_err != NULL) {
2808 error_propagate(errp, local_err);
2810 return ret;
2815 * Prepares a BlockDriverState for reopen. All changes are staged in the
2816 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2817 * the block driver layer .bdrv_reopen_prepare()
2819 * bs is the BlockDriverState to reopen
2820 * flags are the new open flags
2821 * queue is the reopen queue
2823 * Returns 0 on success, non-zero on error. On error errp will be set
2824 * as well.
2826 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2827 * It is the responsibility of the caller to then call the abort() or
2828 * commit() for any other BDS that have been left in a prepare() state
2831 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2832 Error **errp)
2834 int ret = -1;
2835 Error *local_err = NULL;
2836 BlockDriver *drv;
2837 QemuOpts *opts;
2838 const char *value;
2839 bool read_only;
2841 assert(reopen_state != NULL);
2842 assert(reopen_state->bs->drv != NULL);
2843 drv = reopen_state->bs->drv;
2845 /* Process generic block layer options */
2846 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2847 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2848 if (local_err) {
2849 error_propagate(errp, local_err);
2850 ret = -EINVAL;
2851 goto error;
2854 update_flags_from_options(&reopen_state->flags, opts);
2856 /* node-name and driver must be unchanged. Put them back into the QDict, so
2857 * that they are checked at the end of this function. */
2858 value = qemu_opt_get(opts, "node-name");
2859 if (value) {
2860 qdict_put_str(reopen_state->options, "node-name", value);
2863 value = qemu_opt_get(opts, "driver");
2864 if (value) {
2865 qdict_put_str(reopen_state->options, "driver", value);
2868 /* If we are to stay read-only, do not allow permission change
2869 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
2870 * not set, or if the BDS still has copy_on_read enabled */
2871 read_only = !(reopen_state->flags & BDRV_O_RDWR);
2872 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, &local_err);
2873 if (local_err) {
2874 error_propagate(errp, local_err);
2875 goto error;
2879 ret = bdrv_flush(reopen_state->bs);
2880 if (ret) {
2881 error_setg_errno(errp, -ret, "Error flushing drive");
2882 goto error;
2885 if (drv->bdrv_reopen_prepare) {
2886 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2887 if (ret) {
2888 if (local_err != NULL) {
2889 error_propagate(errp, local_err);
2890 } else {
2891 error_setg(errp, "failed while preparing to reopen image '%s'",
2892 reopen_state->bs->filename);
2894 goto error;
2896 } else {
2897 /* It is currently mandatory to have a bdrv_reopen_prepare()
2898 * handler for each supported drv. */
2899 error_setg(errp, "Block format '%s' used by node '%s' "
2900 "does not support reopening files", drv->format_name,
2901 bdrv_get_device_or_node_name(reopen_state->bs));
2902 ret = -1;
2903 goto error;
2906 /* Options that are not handled are only okay if they are unchanged
2907 * compared to the old state. It is expected that some options are only
2908 * used for the initial open, but not reopen (e.g. filename) */
2909 if (qdict_size(reopen_state->options)) {
2910 const QDictEntry *entry = qdict_first(reopen_state->options);
2912 do {
2913 QString *new_obj = qobject_to_qstring(entry->value);
2914 const char *new = qstring_get_str(new_obj);
2916 * Caution: while qdict_get_try_str() is fine, getting
2917 * non-string types would require more care. When
2918 * bs->options come from -blockdev or blockdev_add, its
2919 * members are typed according to the QAPI schema, but
2920 * when they come from -drive, they're all QString.
2922 const char *old = qdict_get_try_str(reopen_state->bs->options,
2923 entry->key);
2925 if (!old || strcmp(new, old)) {
2926 error_setg(errp, "Cannot change the option '%s'", entry->key);
2927 ret = -EINVAL;
2928 goto error;
2930 } while ((entry = qdict_next(reopen_state->options, entry)));
2933 ret = 0;
2935 error:
2936 qemu_opts_del(opts);
2937 return ret;
2941 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2942 * makes them final by swapping the staging BlockDriverState contents into
2943 * the active BlockDriverState contents.
2945 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2947 BlockDriver *drv;
2949 assert(reopen_state != NULL);
2950 drv = reopen_state->bs->drv;
2951 assert(drv != NULL);
2953 /* If there are any driver level actions to take */
2954 if (drv->bdrv_reopen_commit) {
2955 drv->bdrv_reopen_commit(reopen_state);
2958 /* set BDS specific flags now */
2959 QDECREF(reopen_state->bs->explicit_options);
2961 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2962 reopen_state->bs->open_flags = reopen_state->flags;
2963 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2965 bdrv_refresh_limits(reopen_state->bs, NULL);
2969 * Abort the reopen, and delete and free the staged changes in
2970 * reopen_state
2972 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2974 BlockDriver *drv;
2976 assert(reopen_state != NULL);
2977 drv = reopen_state->bs->drv;
2978 assert(drv != NULL);
2980 if (drv->bdrv_reopen_abort) {
2981 drv->bdrv_reopen_abort(reopen_state);
2984 QDECREF(reopen_state->explicit_options);
2988 static void bdrv_close(BlockDriverState *bs)
2990 BdrvAioNotifier *ban, *ban_next;
2992 assert(!bs->job);
2993 assert(!bs->refcnt);
2995 bdrv_drained_begin(bs); /* complete I/O */
2996 bdrv_flush(bs);
2997 bdrv_drain(bs); /* in case flush left pending I/O */
2999 bdrv_release_named_dirty_bitmaps(bs);
3000 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3002 if (bs->drv) {
3003 BdrvChild *child, *next;
3005 bs->drv->bdrv_close(bs);
3006 bs->drv = NULL;
3008 bdrv_set_backing_hd(bs, NULL, &error_abort);
3010 if (bs->file != NULL) {
3011 bdrv_unref_child(bs, bs->file);
3012 bs->file = NULL;
3015 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3016 /* TODO Remove bdrv_unref() from drivers' close function and use
3017 * bdrv_unref_child() here */
3018 if (child->bs->inherits_from == bs) {
3019 child->bs->inherits_from = NULL;
3021 bdrv_detach_child(child);
3024 g_free(bs->opaque);
3025 bs->opaque = NULL;
3026 bs->copy_on_read = 0;
3027 bs->backing_file[0] = '\0';
3028 bs->backing_format[0] = '\0';
3029 bs->total_sectors = 0;
3030 bs->encrypted = false;
3031 bs->valid_key = false;
3032 bs->sg = false;
3033 QDECREF(bs->options);
3034 QDECREF(bs->explicit_options);
3035 bs->options = NULL;
3036 QDECREF(bs->full_open_options);
3037 bs->full_open_options = NULL;
3040 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3041 g_free(ban);
3043 QLIST_INIT(&bs->aio_notifiers);
3044 bdrv_drained_end(bs);
3047 void bdrv_close_all(void)
3049 block_job_cancel_sync_all();
3050 nbd_export_close_all();
3052 /* Drop references from requests still in flight, such as canceled block
3053 * jobs whose AIO context has not been polled yet */
3054 bdrv_drain_all();
3056 blk_remove_all_bs();
3057 blockdev_close_all_bdrv_states();
3059 assert(QTAILQ_EMPTY(&all_bdrv_states));
3062 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3064 BdrvChild *to_c;
3066 if (c->role->stay_at_node) {
3067 return false;
3070 if (c->role == &child_backing) {
3071 /* If @from is a backing file of @to, ignore the child to avoid
3072 * creating a loop. We only want to change the pointer of other
3073 * parents. */
3074 QLIST_FOREACH(to_c, &to->children, next) {
3075 if (to_c == c) {
3076 break;
3079 if (to_c) {
3080 return false;
3084 return true;
3087 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3088 Error **errp)
3090 BdrvChild *c, *next;
3091 GSList *list = NULL, *p;
3092 uint64_t old_perm, old_shared;
3093 uint64_t perm = 0, shared = BLK_PERM_ALL;
3094 int ret;
3096 assert(!atomic_read(&from->in_flight));
3097 assert(!atomic_read(&to->in_flight));
3099 /* Make sure that @from doesn't go away until we have successfully attached
3100 * all of its parents to @to. */
3101 bdrv_ref(from);
3103 /* Put all parents into @list and calculate their cumulative permissions */
3104 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3105 if (!should_update_child(c, to)) {
3106 continue;
3108 list = g_slist_prepend(list, c);
3109 perm |= c->perm;
3110 shared &= c->shared_perm;
3113 /* Check whether the required permissions can be granted on @to, ignoring
3114 * all BdrvChild in @list so that they can't block themselves. */
3115 ret = bdrv_check_update_perm(to, perm, shared, list, errp);
3116 if (ret < 0) {
3117 bdrv_abort_perm_update(to);
3118 goto out;
3121 /* Now actually perform the change. We performed the permission check for
3122 * all elements of @list at once, so set the permissions all at once at the
3123 * very end. */
3124 for (p = list; p != NULL; p = p->next) {
3125 c = p->data;
3127 bdrv_ref(to);
3128 bdrv_replace_child_noperm(c, to);
3129 bdrv_unref(from);
3132 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3133 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3135 out:
3136 g_slist_free(list);
3137 bdrv_unref(from);
3141 * Add new bs contents at the top of an image chain while the chain is
3142 * live, while keeping required fields on the top layer.
3144 * This will modify the BlockDriverState fields, and swap contents
3145 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3147 * bs_new must not be attached to a BlockBackend.
3149 * This function does not create any image files.
3151 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3152 * that's what the callers commonly need. bs_new will be referenced by the old
3153 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3154 * reference of its own, it must call bdrv_ref().
3156 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3157 Error **errp)
3159 Error *local_err = NULL;
3161 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3162 if (local_err) {
3163 error_propagate(errp, local_err);
3164 goto out;
3167 bdrv_replace_node(bs_top, bs_new, &local_err);
3168 if (local_err) {
3169 error_propagate(errp, local_err);
3170 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3171 goto out;
3174 /* bs_new is now referenced by its new parents, we don't need the
3175 * additional reference any more. */
3176 out:
3177 bdrv_unref(bs_new);
3180 static void bdrv_delete(BlockDriverState *bs)
3182 assert(!bs->job);
3183 assert(bdrv_op_blocker_is_empty(bs));
3184 assert(!bs->refcnt);
3186 bdrv_close(bs);
3188 /* remove from list, if necessary */
3189 if (bs->node_name[0] != '\0') {
3190 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3192 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3194 g_free(bs);
3198 * Run consistency checks on an image
3200 * Returns 0 if the check could be completed (it doesn't mean that the image is
3201 * free of errors) or -errno when an internal error occurred. The results of the
3202 * check are stored in res.
3204 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3206 if (bs->drv == NULL) {
3207 return -ENOMEDIUM;
3209 if (bs->drv->bdrv_check == NULL) {
3210 return -ENOTSUP;
3213 memset(res, 0, sizeof(*res));
3214 return bs->drv->bdrv_check(bs, res, fix);
3218 * Return values:
3219 * 0 - success
3220 * -EINVAL - backing format specified, but no file
3221 * -ENOSPC - can't update the backing file because no space is left in the
3222 * image file header
3223 * -ENOTSUP - format driver doesn't support changing the backing file
3225 int bdrv_change_backing_file(BlockDriverState *bs,
3226 const char *backing_file, const char *backing_fmt)
3228 BlockDriver *drv = bs->drv;
3229 int ret;
3231 /* Backing file format doesn't make sense without a backing file */
3232 if (backing_fmt && !backing_file) {
3233 return -EINVAL;
3236 if (drv->bdrv_change_backing_file != NULL) {
3237 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3238 } else {
3239 ret = -ENOTSUP;
3242 if (ret == 0) {
3243 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3244 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3246 return ret;
3250 * Finds the image layer in the chain that has 'bs' as its backing file.
3252 * active is the current topmost image.
3254 * Returns NULL if bs is not found in active's image chain,
3255 * or if active == bs.
3257 * Returns the bottommost base image if bs == NULL.
3259 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3260 BlockDriverState *bs)
3262 while (active && bs != backing_bs(active)) {
3263 active = backing_bs(active);
3266 return active;
3269 /* Given a BDS, searches for the base layer. */
3270 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3272 return bdrv_find_overlay(bs, NULL);
3276 * Drops images above 'base' up to and including 'top', and sets the image
3277 * above 'top' to have base as its backing file.
3279 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3280 * information in 'bs' can be properly updated.
3282 * E.g., this will convert the following chain:
3283 * bottom <- base <- intermediate <- top <- active
3285 * to
3287 * bottom <- base <- active
3289 * It is allowed for bottom==base, in which case it converts:
3291 * base <- intermediate <- top <- active
3293 * to
3295 * base <- active
3297 * If backing_file_str is non-NULL, it will be used when modifying top's
3298 * overlay image metadata.
3300 * Error conditions:
3301 * if active == top, that is considered an error
3304 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
3305 BlockDriverState *base, const char *backing_file_str)
3307 BlockDriverState *new_top_bs = NULL;
3308 Error *local_err = NULL;
3309 int ret = -EIO;
3311 if (!top->drv || !base->drv) {
3312 goto exit;
3315 new_top_bs = bdrv_find_overlay(active, top);
3317 if (new_top_bs == NULL) {
3318 /* we could not find the image above 'top', this is an error */
3319 goto exit;
3322 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3323 * to do, no intermediate images */
3324 if (backing_bs(new_top_bs) == base) {
3325 ret = 0;
3326 goto exit;
3329 /* Make sure that base is in the backing chain of top */
3330 if (!bdrv_chain_contains(top, base)) {
3331 goto exit;
3334 /* success - we can delete the intermediate states, and link top->base */
3335 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3336 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
3337 base->drv ? base->drv->format_name : "");
3338 if (ret) {
3339 goto exit;
3342 bdrv_set_backing_hd(new_top_bs, base, &local_err);
3343 if (local_err) {
3344 ret = -EPERM;
3345 error_report_err(local_err);
3346 goto exit;
3349 ret = 0;
3350 exit:
3351 return ret;
3355 * Truncate file to 'offset' bytes (needed only for file protocols)
3357 int bdrv_truncate(BdrvChild *child, int64_t offset, Error **errp)
3359 BlockDriverState *bs = child->bs;
3360 BlockDriver *drv = bs->drv;
3361 int ret;
3363 assert(child->perm & BLK_PERM_RESIZE);
3365 if (!drv) {
3366 error_setg(errp, "No medium inserted");
3367 return -ENOMEDIUM;
3369 if (!drv->bdrv_truncate) {
3370 error_setg(errp, "Image format driver does not support resize");
3371 return -ENOTSUP;
3373 if (bs->read_only) {
3374 error_setg(errp, "Image is read-only");
3375 return -EACCES;
3378 assert(!(bs->open_flags & BDRV_O_INACTIVE));
3380 ret = drv->bdrv_truncate(bs, offset, errp);
3381 if (ret == 0) {
3382 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3383 bdrv_dirty_bitmap_truncate(bs);
3384 bdrv_parent_cb_resize(bs);
3385 ++bs->write_gen;
3387 return ret;
3391 * Length of a allocated file in bytes. Sparse files are counted by actual
3392 * allocated space. Return < 0 if error or unknown.
3394 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3396 BlockDriver *drv = bs->drv;
3397 if (!drv) {
3398 return -ENOMEDIUM;
3400 if (drv->bdrv_get_allocated_file_size) {
3401 return drv->bdrv_get_allocated_file_size(bs);
3403 if (bs->file) {
3404 return bdrv_get_allocated_file_size(bs->file->bs);
3406 return -ENOTSUP;
3410 * Return number of sectors on success, -errno on error.
3412 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3414 BlockDriver *drv = bs->drv;
3416 if (!drv)
3417 return -ENOMEDIUM;
3419 if (drv->has_variable_length) {
3420 int ret = refresh_total_sectors(bs, bs->total_sectors);
3421 if (ret < 0) {
3422 return ret;
3425 return bs->total_sectors;
3429 * Return length in bytes on success, -errno on error.
3430 * The length is always a multiple of BDRV_SECTOR_SIZE.
3432 int64_t bdrv_getlength(BlockDriverState *bs)
3434 int64_t ret = bdrv_nb_sectors(bs);
3436 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3437 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3440 /* return 0 as number of sectors if no device present or error */
3441 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3443 int64_t nb_sectors = bdrv_nb_sectors(bs);
3445 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3448 bool bdrv_is_sg(BlockDriverState *bs)
3450 return bs->sg;
3453 bool bdrv_is_encrypted(BlockDriverState *bs)
3455 if (bs->backing && bs->backing->bs->encrypted) {
3456 return true;
3458 return bs->encrypted;
3461 bool bdrv_key_required(BlockDriverState *bs)
3463 BdrvChild *backing = bs->backing;
3465 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
3466 return true;
3468 return (bs->encrypted && !bs->valid_key);
3471 int bdrv_set_key(BlockDriverState *bs, const char *key)
3473 int ret;
3474 if (bs->backing && bs->backing->bs->encrypted) {
3475 ret = bdrv_set_key(bs->backing->bs, key);
3476 if (ret < 0)
3477 return ret;
3478 if (!bs->encrypted)
3479 return 0;
3481 if (!bs->encrypted) {
3482 return -EINVAL;
3483 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3484 return -ENOMEDIUM;
3486 ret = bs->drv->bdrv_set_key(bs, key);
3487 if (ret < 0) {
3488 bs->valid_key = false;
3489 } else if (!bs->valid_key) {
3490 /* call the change callback now, we skipped it on open */
3491 bs->valid_key = true;
3492 bdrv_parent_cb_change_media(bs, true);
3494 return ret;
3498 * Provide an encryption key for @bs.
3499 * If @key is non-null:
3500 * If @bs is not encrypted, fail.
3501 * Else if the key is invalid, fail.
3502 * Else set @bs's key to @key, replacing the existing key, if any.
3503 * If @key is null:
3504 * If @bs is encrypted and still lacks a key, fail.
3505 * Else do nothing.
3506 * On failure, store an error object through @errp if non-null.
3508 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
3510 if (key) {
3511 if (!bdrv_is_encrypted(bs)) {
3512 error_setg(errp, "Node '%s' is not encrypted",
3513 bdrv_get_device_or_node_name(bs));
3514 } else if (bdrv_set_key(bs, key) < 0) {
3515 error_setg(errp, QERR_INVALID_PASSWORD);
3517 } else {
3518 if (bdrv_key_required(bs)) {
3519 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
3520 "'%s' (%s) is encrypted",
3521 bdrv_get_device_or_node_name(bs),
3522 bdrv_get_encrypted_filename(bs));
3527 const char *bdrv_get_format_name(BlockDriverState *bs)
3529 return bs->drv ? bs->drv->format_name : NULL;
3532 static int qsort_strcmp(const void *a, const void *b)
3534 return strcmp(*(char *const *)a, *(char *const *)b);
3537 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3538 void *opaque)
3540 BlockDriver *drv;
3541 int count = 0;
3542 int i;
3543 const char **formats = NULL;
3545 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3546 if (drv->format_name) {
3547 bool found = false;
3548 int i = count;
3549 while (formats && i && !found) {
3550 found = !strcmp(formats[--i], drv->format_name);
3553 if (!found) {
3554 formats = g_renew(const char *, formats, count + 1);
3555 formats[count++] = drv->format_name;
3560 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3561 const char *format_name = block_driver_modules[i].format_name;
3563 if (format_name) {
3564 bool found = false;
3565 int j = count;
3567 while (formats && j && !found) {
3568 found = !strcmp(formats[--j], format_name);
3571 if (!found) {
3572 formats = g_renew(const char *, formats, count + 1);
3573 formats[count++] = format_name;
3578 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3580 for (i = 0; i < count; i++) {
3581 it(opaque, formats[i]);
3584 g_free(formats);
3587 /* This function is to find a node in the bs graph */
3588 BlockDriverState *bdrv_find_node(const char *node_name)
3590 BlockDriverState *bs;
3592 assert(node_name);
3594 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3595 if (!strcmp(node_name, bs->node_name)) {
3596 return bs;
3599 return NULL;
3602 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3603 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3605 BlockDeviceInfoList *list, *entry;
3606 BlockDriverState *bs;
3608 list = NULL;
3609 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3610 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3611 if (!info) {
3612 qapi_free_BlockDeviceInfoList(list);
3613 return NULL;
3615 entry = g_malloc0(sizeof(*entry));
3616 entry->value = info;
3617 entry->next = list;
3618 list = entry;
3621 return list;
3624 BlockDriverState *bdrv_lookup_bs(const char *device,
3625 const char *node_name,
3626 Error **errp)
3628 BlockBackend *blk;
3629 BlockDriverState *bs;
3631 if (device) {
3632 blk = blk_by_name(device);
3634 if (blk) {
3635 bs = blk_bs(blk);
3636 if (!bs) {
3637 error_setg(errp, "Device '%s' has no medium", device);
3640 return bs;
3644 if (node_name) {
3645 bs = bdrv_find_node(node_name);
3647 if (bs) {
3648 return bs;
3652 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3653 device ? device : "",
3654 node_name ? node_name : "");
3655 return NULL;
3658 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3659 * return false. If either argument is NULL, return false. */
3660 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3662 while (top && top != base) {
3663 top = backing_bs(top);
3666 return top != NULL;
3669 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3671 if (!bs) {
3672 return QTAILQ_FIRST(&graph_bdrv_states);
3674 return QTAILQ_NEXT(bs, node_list);
3677 const char *bdrv_get_node_name(const BlockDriverState *bs)
3679 return bs->node_name;
3682 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3684 BdrvChild *c;
3685 const char *name;
3687 /* If multiple parents have a name, just pick the first one. */
3688 QLIST_FOREACH(c, &bs->parents, next_parent) {
3689 if (c->role->get_name) {
3690 name = c->role->get_name(c);
3691 if (name && *name) {
3692 return name;
3697 return NULL;
3700 /* TODO check what callers really want: bs->node_name or blk_name() */
3701 const char *bdrv_get_device_name(const BlockDriverState *bs)
3703 return bdrv_get_parent_name(bs) ?: "";
3706 /* This can be used to identify nodes that might not have a device
3707 * name associated. Since node and device names live in the same
3708 * namespace, the result is unambiguous. The exception is if both are
3709 * absent, then this returns an empty (non-null) string. */
3710 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3712 return bdrv_get_parent_name(bs) ?: bs->node_name;
3715 int bdrv_get_flags(BlockDriverState *bs)
3717 return bs->open_flags;
3720 int bdrv_has_zero_init_1(BlockDriverState *bs)
3722 return 1;
3725 int bdrv_has_zero_init(BlockDriverState *bs)
3727 assert(bs->drv);
3729 /* If BS is a copy on write image, it is initialized to
3730 the contents of the base image, which may not be zeroes. */
3731 if (bs->backing) {
3732 return 0;
3734 if (bs->drv->bdrv_has_zero_init) {
3735 return bs->drv->bdrv_has_zero_init(bs);
3738 /* safe default */
3739 return 0;
3742 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3744 BlockDriverInfo bdi;
3746 if (bs->backing) {
3747 return false;
3750 if (bdrv_get_info(bs, &bdi) == 0) {
3751 return bdi.unallocated_blocks_are_zero;
3754 return false;
3757 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3759 BlockDriverInfo bdi;
3761 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3762 return false;
3765 if (bdrv_get_info(bs, &bdi) == 0) {
3766 return bdi.can_write_zeroes_with_unmap;
3769 return false;
3772 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3774 if (bs->backing && bs->backing->bs->encrypted)
3775 return bs->backing_file;
3776 else if (bs->encrypted)
3777 return bs->filename;
3778 else
3779 return NULL;
3782 void bdrv_get_backing_filename(BlockDriverState *bs,
3783 char *filename, int filename_size)
3785 pstrcpy(filename, filename_size, bs->backing_file);
3788 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3790 BlockDriver *drv = bs->drv;
3791 if (!drv)
3792 return -ENOMEDIUM;
3793 if (!drv->bdrv_get_info)
3794 return -ENOTSUP;
3795 memset(bdi, 0, sizeof(*bdi));
3796 return drv->bdrv_get_info(bs, bdi);
3799 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3801 BlockDriver *drv = bs->drv;
3802 if (drv && drv->bdrv_get_specific_info) {
3803 return drv->bdrv_get_specific_info(bs);
3805 return NULL;
3808 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3810 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3811 return;
3814 bs->drv->bdrv_debug_event(bs, event);
3817 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3818 const char *tag)
3820 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3821 bs = bs->file ? bs->file->bs : NULL;
3824 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3825 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3828 return -ENOTSUP;
3831 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3833 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3834 bs = bs->file ? bs->file->bs : NULL;
3837 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3838 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3841 return -ENOTSUP;
3844 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3846 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3847 bs = bs->file ? bs->file->bs : NULL;
3850 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3851 return bs->drv->bdrv_debug_resume(bs, tag);
3854 return -ENOTSUP;
3857 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3859 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3860 bs = bs->file ? bs->file->bs : NULL;
3863 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3864 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3867 return false;
3870 /* backing_file can either be relative, or absolute, or a protocol. If it is
3871 * relative, it must be relative to the chain. So, passing in bs->filename
3872 * from a BDS as backing_file should not be done, as that may be relative to
3873 * the CWD rather than the chain. */
3874 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3875 const char *backing_file)
3877 char *filename_full = NULL;
3878 char *backing_file_full = NULL;
3879 char *filename_tmp = NULL;
3880 int is_protocol = 0;
3881 BlockDriverState *curr_bs = NULL;
3882 BlockDriverState *retval = NULL;
3883 Error *local_error = NULL;
3885 if (!bs || !bs->drv || !backing_file) {
3886 return NULL;
3889 filename_full = g_malloc(PATH_MAX);
3890 backing_file_full = g_malloc(PATH_MAX);
3891 filename_tmp = g_malloc(PATH_MAX);
3893 is_protocol = path_has_protocol(backing_file);
3895 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3897 /* If either of the filename paths is actually a protocol, then
3898 * compare unmodified paths; otherwise make paths relative */
3899 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3900 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3901 retval = curr_bs->backing->bs;
3902 break;
3904 /* Also check against the full backing filename for the image */
3905 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
3906 &local_error);
3907 if (local_error == NULL) {
3908 if (strcmp(backing_file, backing_file_full) == 0) {
3909 retval = curr_bs->backing->bs;
3910 break;
3912 } else {
3913 error_free(local_error);
3914 local_error = NULL;
3916 } else {
3917 /* If not an absolute filename path, make it relative to the current
3918 * image's filename path */
3919 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3920 backing_file);
3922 /* We are going to compare absolute pathnames */
3923 if (!realpath(filename_tmp, filename_full)) {
3924 continue;
3927 /* We need to make sure the backing filename we are comparing against
3928 * is relative to the current image filename (or absolute) */
3929 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3930 curr_bs->backing_file);
3932 if (!realpath(filename_tmp, backing_file_full)) {
3933 continue;
3936 if (strcmp(backing_file_full, filename_full) == 0) {
3937 retval = curr_bs->backing->bs;
3938 break;
3943 g_free(filename_full);
3944 g_free(backing_file_full);
3945 g_free(filename_tmp);
3946 return retval;
3949 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3951 if (!bs->drv) {
3952 return 0;
3955 if (!bs->backing) {
3956 return 0;
3959 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3962 void bdrv_init(void)
3964 module_call_init(MODULE_INIT_BLOCK);
3967 void bdrv_init_with_whitelist(void)
3969 use_bdrv_whitelist = 1;
3970 bdrv_init();
3973 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3975 BdrvChild *child, *parent;
3976 uint64_t perm, shared_perm;
3977 Error *local_err = NULL;
3978 int ret;
3980 if (!bs->drv) {
3981 return;
3984 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3985 return;
3988 QLIST_FOREACH(child, &bs->children, next) {
3989 bdrv_invalidate_cache(child->bs, &local_err);
3990 if (local_err) {
3991 error_propagate(errp, local_err);
3992 return;
3996 bs->open_flags &= ~BDRV_O_INACTIVE;
3997 if (bs->drv->bdrv_invalidate_cache) {
3998 bs->drv->bdrv_invalidate_cache(bs, &local_err);
3999 if (local_err) {
4000 bs->open_flags |= BDRV_O_INACTIVE;
4001 error_propagate(errp, local_err);
4002 return;
4006 ret = refresh_total_sectors(bs, bs->total_sectors);
4007 if (ret < 0) {
4008 bs->open_flags |= BDRV_O_INACTIVE;
4009 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4010 return;
4013 /* Update permissions, they may differ for inactive nodes */
4014 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4015 ret = bdrv_check_perm(bs, perm, shared_perm, NULL, &local_err);
4016 if (ret < 0) {
4017 bs->open_flags |= BDRV_O_INACTIVE;
4018 error_propagate(errp, local_err);
4019 return;
4021 bdrv_set_perm(bs, perm, shared_perm);
4023 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4024 if (parent->role->activate) {
4025 parent->role->activate(parent, &local_err);
4026 if (local_err) {
4027 error_propagate(errp, local_err);
4028 return;
4034 void bdrv_invalidate_cache_all(Error **errp)
4036 BlockDriverState *bs;
4037 Error *local_err = NULL;
4038 BdrvNextIterator it;
4040 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4041 AioContext *aio_context = bdrv_get_aio_context(bs);
4043 aio_context_acquire(aio_context);
4044 bdrv_invalidate_cache(bs, &local_err);
4045 aio_context_release(aio_context);
4046 if (local_err) {
4047 error_propagate(errp, local_err);
4048 return;
4053 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4054 bool setting_flag)
4056 BdrvChild *child, *parent;
4057 int ret;
4059 if (!setting_flag && bs->drv->bdrv_inactivate) {
4060 ret = bs->drv->bdrv_inactivate(bs);
4061 if (ret < 0) {
4062 return ret;
4066 if (setting_flag) {
4067 uint64_t perm, shared_perm;
4069 bs->open_flags |= BDRV_O_INACTIVE;
4071 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4072 if (parent->role->inactivate) {
4073 ret = parent->role->inactivate(parent);
4074 if (ret < 0) {
4075 bs->open_flags &= ~BDRV_O_INACTIVE;
4076 return ret;
4081 /* Update permissions, they may differ for inactive nodes */
4082 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4083 bdrv_check_perm(bs, perm, shared_perm, NULL, &error_abort);
4084 bdrv_set_perm(bs, perm, shared_perm);
4087 QLIST_FOREACH(child, &bs->children, next) {
4088 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4089 if (ret < 0) {
4090 return ret;
4094 return 0;
4097 int bdrv_inactivate_all(void)
4099 BlockDriverState *bs = NULL;
4100 BdrvNextIterator it;
4101 int ret = 0;
4102 int pass;
4104 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4105 aio_context_acquire(bdrv_get_aio_context(bs));
4108 /* We do two passes of inactivation. The first pass calls to drivers'
4109 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4110 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4111 * is allowed. */
4112 for (pass = 0; pass < 2; pass++) {
4113 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4114 ret = bdrv_inactivate_recurse(bs, pass);
4115 if (ret < 0) {
4116 goto out;
4121 out:
4122 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4123 aio_context_release(bdrv_get_aio_context(bs));
4126 return ret;
4129 /**************************************************************/
4130 /* removable device support */
4133 * Return TRUE if the media is present
4135 bool bdrv_is_inserted(BlockDriverState *bs)
4137 BlockDriver *drv = bs->drv;
4138 BdrvChild *child;
4140 if (!drv) {
4141 return false;
4143 if (drv->bdrv_is_inserted) {
4144 return drv->bdrv_is_inserted(bs);
4146 QLIST_FOREACH(child, &bs->children, next) {
4147 if (!bdrv_is_inserted(child->bs)) {
4148 return false;
4151 return true;
4155 * Return whether the media changed since the last call to this
4156 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4158 int bdrv_media_changed(BlockDriverState *bs)
4160 BlockDriver *drv = bs->drv;
4162 if (drv && drv->bdrv_media_changed) {
4163 return drv->bdrv_media_changed(bs);
4165 return -ENOTSUP;
4169 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4171 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4173 BlockDriver *drv = bs->drv;
4175 if (drv && drv->bdrv_eject) {
4176 drv->bdrv_eject(bs, eject_flag);
4181 * Lock or unlock the media (if it is locked, the user won't be able
4182 * to eject it manually).
4184 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4186 BlockDriver *drv = bs->drv;
4188 trace_bdrv_lock_medium(bs, locked);
4190 if (drv && drv->bdrv_lock_medium) {
4191 drv->bdrv_lock_medium(bs, locked);
4195 /* Get a reference to bs */
4196 void bdrv_ref(BlockDriverState *bs)
4198 bs->refcnt++;
4201 /* Release a previously grabbed reference to bs.
4202 * If after releasing, reference count is zero, the BlockDriverState is
4203 * deleted. */
4204 void bdrv_unref(BlockDriverState *bs)
4206 if (!bs) {
4207 return;
4209 assert(bs->refcnt > 0);
4210 if (--bs->refcnt == 0) {
4211 bdrv_delete(bs);
4215 struct BdrvOpBlocker {
4216 Error *reason;
4217 QLIST_ENTRY(BdrvOpBlocker) list;
4220 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4222 BdrvOpBlocker *blocker;
4223 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4224 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4225 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4226 if (errp) {
4227 *errp = error_copy(blocker->reason);
4228 error_prepend(errp, "Node '%s' is busy: ",
4229 bdrv_get_device_or_node_name(bs));
4231 return true;
4233 return false;
4236 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4238 BdrvOpBlocker *blocker;
4239 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4241 blocker = g_new0(BdrvOpBlocker, 1);
4242 blocker->reason = reason;
4243 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4246 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4248 BdrvOpBlocker *blocker, *next;
4249 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4250 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4251 if (blocker->reason == reason) {
4252 QLIST_REMOVE(blocker, list);
4253 g_free(blocker);
4258 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4260 int i;
4261 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4262 bdrv_op_block(bs, i, reason);
4266 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4268 int i;
4269 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4270 bdrv_op_unblock(bs, i, reason);
4274 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4276 int i;
4278 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4279 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4280 return false;
4283 return true;
4286 void bdrv_img_create(const char *filename, const char *fmt,
4287 const char *base_filename, const char *base_fmt,
4288 char *options, uint64_t img_size, int flags, bool quiet,
4289 Error **errp)
4291 QemuOptsList *create_opts = NULL;
4292 QemuOpts *opts = NULL;
4293 const char *backing_fmt, *backing_file;
4294 int64_t size;
4295 BlockDriver *drv, *proto_drv;
4296 Error *local_err = NULL;
4297 int ret = 0;
4299 /* Find driver and parse its options */
4300 drv = bdrv_find_format(fmt);
4301 if (!drv) {
4302 error_setg(errp, "Unknown file format '%s'", fmt);
4303 return;
4306 proto_drv = bdrv_find_protocol(filename, true, errp);
4307 if (!proto_drv) {
4308 return;
4311 if (!drv->create_opts) {
4312 error_setg(errp, "Format driver '%s' does not support image creation",
4313 drv->format_name);
4314 return;
4317 if (!proto_drv->create_opts) {
4318 error_setg(errp, "Protocol driver '%s' does not support image creation",
4319 proto_drv->format_name);
4320 return;
4323 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4324 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4326 /* Create parameter list with default values */
4327 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4328 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4330 /* Parse -o options */
4331 if (options) {
4332 qemu_opts_do_parse(opts, options, NULL, &local_err);
4333 if (local_err) {
4334 error_report_err(local_err);
4335 local_err = NULL;
4336 error_setg(errp, "Invalid options for file format '%s'", fmt);
4337 goto out;
4341 if (base_filename) {
4342 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4343 if (local_err) {
4344 error_setg(errp, "Backing file not supported for file format '%s'",
4345 fmt);
4346 goto out;
4350 if (base_fmt) {
4351 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4352 if (local_err) {
4353 error_setg(errp, "Backing file format not supported for file "
4354 "format '%s'", fmt);
4355 goto out;
4359 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4360 if (backing_file) {
4361 if (!strcmp(filename, backing_file)) {
4362 error_setg(errp, "Error: Trying to create an image with the "
4363 "same filename as the backing file");
4364 goto out;
4368 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4370 // The size for the image must always be specified, with one exception:
4371 // If we are using a backing file, we can obtain the size from there
4372 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4373 if (size == -1) {
4374 if (backing_file) {
4375 BlockDriverState *bs;
4376 char *full_backing = g_new0(char, PATH_MAX);
4377 int64_t size;
4378 int back_flags;
4379 QDict *backing_options = NULL;
4381 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4382 full_backing, PATH_MAX,
4383 &local_err);
4384 if (local_err) {
4385 g_free(full_backing);
4386 goto out;
4389 /* backing files always opened read-only */
4390 back_flags = flags;
4391 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4393 if (backing_fmt) {
4394 backing_options = qdict_new();
4395 qdict_put_str(backing_options, "driver", backing_fmt);
4398 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4399 &local_err);
4400 g_free(full_backing);
4401 if (!bs) {
4402 goto out;
4404 size = bdrv_getlength(bs);
4405 if (size < 0) {
4406 error_setg_errno(errp, -size, "Could not get size of '%s'",
4407 backing_file);
4408 bdrv_unref(bs);
4409 goto out;
4412 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4414 bdrv_unref(bs);
4415 } else {
4416 error_setg(errp, "Image creation needs a size parameter");
4417 goto out;
4421 if (!quiet) {
4422 printf("Formatting '%s', fmt=%s ", filename, fmt);
4423 qemu_opts_print(opts, " ");
4424 puts("");
4427 ret = bdrv_create(drv, filename, opts, &local_err);
4429 if (ret == -EFBIG) {
4430 /* This is generally a better message than whatever the driver would
4431 * deliver (especially because of the cluster_size_hint), since that
4432 * is most probably not much different from "image too large". */
4433 const char *cluster_size_hint = "";
4434 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4435 cluster_size_hint = " (try using a larger cluster size)";
4437 error_setg(errp, "The image size is too large for file format '%s'"
4438 "%s", fmt, cluster_size_hint);
4439 error_free(local_err);
4440 local_err = NULL;
4443 out:
4444 qemu_opts_del(opts);
4445 qemu_opts_free(create_opts);
4446 error_propagate(errp, local_err);
4449 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4451 return bs->aio_context;
4454 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4456 aio_co_enter(bdrv_get_aio_context(bs), co);
4459 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4461 QLIST_REMOVE(ban, list);
4462 g_free(ban);
4465 void bdrv_detach_aio_context(BlockDriverState *bs)
4467 BdrvAioNotifier *baf, *baf_tmp;
4468 BdrvChild *child;
4470 if (!bs->drv) {
4471 return;
4474 assert(!bs->walking_aio_notifiers);
4475 bs->walking_aio_notifiers = true;
4476 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4477 if (baf->deleted) {
4478 bdrv_do_remove_aio_context_notifier(baf);
4479 } else {
4480 baf->detach_aio_context(baf->opaque);
4483 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4484 * remove remaining aio notifiers if we aren't called again.
4486 bs->walking_aio_notifiers = false;
4488 if (bs->drv->bdrv_detach_aio_context) {
4489 bs->drv->bdrv_detach_aio_context(bs);
4491 QLIST_FOREACH(child, &bs->children, next) {
4492 bdrv_detach_aio_context(child->bs);
4495 bs->aio_context = NULL;
4498 void bdrv_attach_aio_context(BlockDriverState *bs,
4499 AioContext *new_context)
4501 BdrvAioNotifier *ban, *ban_tmp;
4502 BdrvChild *child;
4504 if (!bs->drv) {
4505 return;
4508 bs->aio_context = new_context;
4510 QLIST_FOREACH(child, &bs->children, next) {
4511 bdrv_attach_aio_context(child->bs, new_context);
4513 if (bs->drv->bdrv_attach_aio_context) {
4514 bs->drv->bdrv_attach_aio_context(bs, new_context);
4517 assert(!bs->walking_aio_notifiers);
4518 bs->walking_aio_notifiers = true;
4519 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4520 if (ban->deleted) {
4521 bdrv_do_remove_aio_context_notifier(ban);
4522 } else {
4523 ban->attached_aio_context(new_context, ban->opaque);
4526 bs->walking_aio_notifiers = false;
4529 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4531 AioContext *ctx = bdrv_get_aio_context(bs);
4533 aio_disable_external(ctx);
4534 bdrv_parent_drained_begin(bs);
4535 bdrv_drain(bs); /* ensure there are no in-flight requests */
4537 while (aio_poll(ctx, false)) {
4538 /* wait for all bottom halves to execute */
4541 bdrv_detach_aio_context(bs);
4543 /* This function executes in the old AioContext so acquire the new one in
4544 * case it runs in a different thread.
4546 aio_context_acquire(new_context);
4547 bdrv_attach_aio_context(bs, new_context);
4548 bdrv_parent_drained_end(bs);
4549 aio_enable_external(ctx);
4550 aio_context_release(new_context);
4553 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4554 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4555 void (*detach_aio_context)(void *opaque), void *opaque)
4557 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4558 *ban = (BdrvAioNotifier){
4559 .attached_aio_context = attached_aio_context,
4560 .detach_aio_context = detach_aio_context,
4561 .opaque = opaque
4564 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4567 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4568 void (*attached_aio_context)(AioContext *,
4569 void *),
4570 void (*detach_aio_context)(void *),
4571 void *opaque)
4573 BdrvAioNotifier *ban, *ban_next;
4575 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4576 if (ban->attached_aio_context == attached_aio_context &&
4577 ban->detach_aio_context == detach_aio_context &&
4578 ban->opaque == opaque &&
4579 ban->deleted == false)
4581 if (bs->walking_aio_notifiers) {
4582 ban->deleted = true;
4583 } else {
4584 bdrv_do_remove_aio_context_notifier(ban);
4586 return;
4590 abort();
4593 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4594 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4596 if (!bs->drv->bdrv_amend_options) {
4597 return -ENOTSUP;
4599 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
4602 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4603 * of block filter and by bdrv_is_first_non_filter.
4604 * It is used to test if the given bs is the candidate or recurse more in the
4605 * node graph.
4607 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
4608 BlockDriverState *candidate)
4610 /* return false if basic checks fails */
4611 if (!bs || !bs->drv) {
4612 return false;
4615 /* the code reached a non block filter driver -> check if the bs is
4616 * the same as the candidate. It's the recursion termination condition.
4618 if (!bs->drv->is_filter) {
4619 return bs == candidate;
4621 /* Down this path the driver is a block filter driver */
4623 /* If the block filter recursion method is defined use it to recurse down
4624 * the node graph.
4626 if (bs->drv->bdrv_recurse_is_first_non_filter) {
4627 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
4630 /* the driver is a block filter but don't allow to recurse -> return false
4632 return false;
4635 /* This function checks if the candidate is the first non filter bs down it's
4636 * bs chain. Since we don't have pointers to parents it explore all bs chains
4637 * from the top. Some filters can choose not to pass down the recursion.
4639 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
4641 BlockDriverState *bs;
4642 BdrvNextIterator it;
4644 /* walk down the bs forest recursively */
4645 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4646 bool perm;
4648 /* try to recurse in this top level bs */
4649 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
4651 /* candidate is the first non filter */
4652 if (perm) {
4653 return true;
4657 return false;
4660 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
4661 const char *node_name, Error **errp)
4663 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
4664 AioContext *aio_context;
4666 if (!to_replace_bs) {
4667 error_setg(errp, "Node name '%s' not found", node_name);
4668 return NULL;
4671 aio_context = bdrv_get_aio_context(to_replace_bs);
4672 aio_context_acquire(aio_context);
4674 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
4675 to_replace_bs = NULL;
4676 goto out;
4679 /* We don't want arbitrary node of the BDS chain to be replaced only the top
4680 * most non filter in order to prevent data corruption.
4681 * Another benefit is that this tests exclude backing files which are
4682 * blocked by the backing blockers.
4684 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
4685 error_setg(errp, "Only top most non filter can be replaced");
4686 to_replace_bs = NULL;
4687 goto out;
4690 out:
4691 aio_context_release(aio_context);
4692 return to_replace_bs;
4695 static bool append_open_options(QDict *d, BlockDriverState *bs)
4697 const QDictEntry *entry;
4698 QemuOptDesc *desc;
4699 BdrvChild *child;
4700 bool found_any = false;
4701 const char *p;
4703 for (entry = qdict_first(bs->options); entry;
4704 entry = qdict_next(bs->options, entry))
4706 /* Exclude options for children */
4707 QLIST_FOREACH(child, &bs->children, next) {
4708 if (strstart(qdict_entry_key(entry), child->name, &p)
4709 && (!*p || *p == '.'))
4711 break;
4714 if (child) {
4715 continue;
4718 /* And exclude all non-driver-specific options */
4719 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4720 if (!strcmp(qdict_entry_key(entry), desc->name)) {
4721 break;
4724 if (desc->name) {
4725 continue;
4728 qobject_incref(qdict_entry_value(entry));
4729 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
4730 found_any = true;
4733 return found_any;
4736 /* Updates the following BDS fields:
4737 * - exact_filename: A filename which may be used for opening a block device
4738 * which (mostly) equals the given BDS (even without any
4739 * other options; so reading and writing must return the same
4740 * results, but caching etc. may be different)
4741 * - full_open_options: Options which, when given when opening a block device
4742 * (without a filename), result in a BDS (mostly)
4743 * equalling the given one
4744 * - filename: If exact_filename is set, it is copied here. Otherwise,
4745 * full_open_options is converted to a JSON object, prefixed with
4746 * "json:" (for use through the JSON pseudo protocol) and put here.
4748 void bdrv_refresh_filename(BlockDriverState *bs)
4750 BlockDriver *drv = bs->drv;
4751 QDict *opts;
4753 if (!drv) {
4754 return;
4757 /* This BDS's file name will most probably depend on its file's name, so
4758 * refresh that first */
4759 if (bs->file) {
4760 bdrv_refresh_filename(bs->file->bs);
4763 if (drv->bdrv_refresh_filename) {
4764 /* Obsolete information is of no use here, so drop the old file name
4765 * information before refreshing it */
4766 bs->exact_filename[0] = '\0';
4767 if (bs->full_open_options) {
4768 QDECREF(bs->full_open_options);
4769 bs->full_open_options = NULL;
4772 opts = qdict_new();
4773 append_open_options(opts, bs);
4774 drv->bdrv_refresh_filename(bs, opts);
4775 QDECREF(opts);
4776 } else if (bs->file) {
4777 /* Try to reconstruct valid information from the underlying file */
4778 bool has_open_options;
4780 bs->exact_filename[0] = '\0';
4781 if (bs->full_open_options) {
4782 QDECREF(bs->full_open_options);
4783 bs->full_open_options = NULL;
4786 opts = qdict_new();
4787 has_open_options = append_open_options(opts, bs);
4789 /* If no specific options have been given for this BDS, the filename of
4790 * the underlying file should suffice for this one as well */
4791 if (bs->file->bs->exact_filename[0] && !has_open_options) {
4792 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
4794 /* Reconstructing the full options QDict is simple for most format block
4795 * drivers, as long as the full options are known for the underlying
4796 * file BDS. The full options QDict of that file BDS should somehow
4797 * contain a representation of the filename, therefore the following
4798 * suffices without querying the (exact_)filename of this BDS. */
4799 if (bs->file->bs->full_open_options) {
4800 qdict_put_str(opts, "driver", drv->format_name);
4801 QINCREF(bs->file->bs->full_open_options);
4802 qdict_put(opts, "file", bs->file->bs->full_open_options);
4804 bs->full_open_options = opts;
4805 } else {
4806 QDECREF(opts);
4808 } else if (!bs->full_open_options && qdict_size(bs->options)) {
4809 /* There is no underlying file BDS (at least referenced by BDS.file),
4810 * so the full options QDict should be equal to the options given
4811 * specifically for this block device when it was opened (plus the
4812 * driver specification).
4813 * Because those options don't change, there is no need to update
4814 * full_open_options when it's already set. */
4816 opts = qdict_new();
4817 append_open_options(opts, bs);
4818 qdict_put_str(opts, "driver", drv->format_name);
4820 if (bs->exact_filename[0]) {
4821 /* This may not work for all block protocol drivers (some may
4822 * require this filename to be parsed), but we have to find some
4823 * default solution here, so just include it. If some block driver
4824 * does not support pure options without any filename at all or
4825 * needs some special format of the options QDict, it needs to
4826 * implement the driver-specific bdrv_refresh_filename() function.
4828 qdict_put_str(opts, "filename", bs->exact_filename);
4831 bs->full_open_options = opts;
4834 if (bs->exact_filename[0]) {
4835 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
4836 } else if (bs->full_open_options) {
4837 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
4838 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
4839 qstring_get_str(json));
4840 QDECREF(json);
4845 * Hot add/remove a BDS's child. So the user can take a child offline when
4846 * it is broken and take a new child online
4848 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
4849 Error **errp)
4852 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
4853 error_setg(errp, "The node %s does not support adding a child",
4854 bdrv_get_device_or_node_name(parent_bs));
4855 return;
4858 if (!QLIST_EMPTY(&child_bs->parents)) {
4859 error_setg(errp, "The node %s already has a parent",
4860 child_bs->node_name);
4861 return;
4864 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
4867 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
4869 BdrvChild *tmp;
4871 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
4872 error_setg(errp, "The node %s does not support removing a child",
4873 bdrv_get_device_or_node_name(parent_bs));
4874 return;
4877 QLIST_FOREACH(tmp, &parent_bs->children, next) {
4878 if (tmp == child) {
4879 break;
4883 if (!tmp) {
4884 error_setg(errp, "The node %s does not have a child named %s",
4885 bdrv_get_device_or_node_name(parent_bs),
4886 bdrv_get_device_or_node_name(child->bs));
4887 return;
4890 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);