tests/migration: Add migration-test header file
[qemu/ar7.git] / block.c
blobc298ca6a197ac3635a81986fa322c7b5172530d7
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/module.h"
34 #include "qapi/error.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qjson.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-output-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "sysemu/block-backend.h"
42 #include "sysemu/sysemu.h"
43 #include "qemu/notify.h"
44 #include "qemu/option.h"
45 #include "qemu/coroutine.h"
46 #include "block/qapi.h"
47 #include "qemu/timer.h"
48 #include "qemu/cutils.h"
49 #include "qemu/id.h"
51 #ifdef CONFIG_BSD
52 #include <sys/ioctl.h>
53 #include <sys/queue.h>
54 #ifndef __DragonFly__
55 #include <sys/disk.h>
56 #endif
57 #endif
59 #ifdef _WIN32
60 #include <windows.h>
61 #endif
63 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
72 QLIST_HEAD_INITIALIZER(bdrv_drivers);
74 static BlockDriverState *bdrv_open_inherit(const char *filename,
75 const char *reference,
76 QDict *options, int flags,
77 BlockDriverState *parent,
78 const BdrvChildRole *child_role,
79 Error **errp);
81 /* If non-zero, use only whitelisted block drivers */
82 static int use_bdrv_whitelist;
84 #ifdef _WIN32
85 static int is_windows_drive_prefix(const char *filename)
87 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
88 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
89 filename[1] == ':');
92 int is_windows_drive(const char *filename)
94 if (is_windows_drive_prefix(filename) &&
95 filename[2] == '\0')
96 return 1;
97 if (strstart(filename, "\\\\.\\", NULL) ||
98 strstart(filename, "//./", NULL))
99 return 1;
100 return 0;
102 #endif
104 size_t bdrv_opt_mem_align(BlockDriverState *bs)
106 if (!bs || !bs->drv) {
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
111 return bs->bl.opt_mem_alignment;
114 size_t bdrv_min_mem_align(BlockDriverState *bs)
116 if (!bs || !bs->drv) {
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
121 return bs->bl.min_mem_alignment;
124 /* check if the path starts with "<protocol>:" */
125 int path_has_protocol(const char *path)
127 const char *p;
129 #ifdef _WIN32
130 if (is_windows_drive(path) ||
131 is_windows_drive_prefix(path)) {
132 return 0;
134 p = path + strcspn(path, ":/\\");
135 #else
136 p = path + strcspn(path, ":/");
137 #endif
139 return *p == ':';
142 int path_is_absolute(const char *path)
144 #ifdef _WIN32
145 /* specific case for names like: "\\.\d:" */
146 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
147 return 1;
149 return (*path == '/' || *path == '\\');
150 #else
151 return (*path == '/');
152 #endif
155 /* if filename is absolute, just copy it to dest. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
157 supported. */
158 void path_combine(char *dest, int dest_size,
159 const char *base_path,
160 const char *filename)
162 const char *p, *p1;
163 int len;
165 if (dest_size <= 0)
166 return;
167 if (path_is_absolute(filename)) {
168 pstrcpy(dest, dest_size, filename);
169 } else {
170 const char *protocol_stripped = NULL;
172 if (path_has_protocol(base_path)) {
173 protocol_stripped = strchr(base_path, ':');
174 if (protocol_stripped) {
175 protocol_stripped++;
178 p = protocol_stripped ?: base_path;
180 p1 = strrchr(base_path, '/');
181 #ifdef _WIN32
183 const char *p2;
184 p2 = strrchr(base_path, '\\');
185 if (!p1 || p2 > p1)
186 p1 = p2;
188 #endif
189 if (p1)
190 p1++;
191 else
192 p1 = base_path;
193 if (p1 > p)
194 p = p1;
195 len = p - base_path;
196 if (len > dest_size - 1)
197 len = dest_size - 1;
198 memcpy(dest, base_path, len);
199 dest[len] = '\0';
200 pstrcat(dest, dest_size, filename);
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
209 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
210 QDict *options)
212 if (strstart(filename, prefix, &filename)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename)) {
216 QString *fat_filename;
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename));
222 /* And we can thus fix the protocol detection issue by prefixing it
223 * by "./" */
224 fat_filename = qstring_from_str("./");
225 qstring_append(fat_filename, filename);
227 assert(!path_has_protocol(qstring_get_str(fat_filename)));
229 qdict_put(options, "filename", fat_filename);
230 } else {
231 /* If no protocol prefix was detected, we can use the shortened
232 * filename as-is */
233 qdict_put_str(options, "filename", filename);
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState *bs)
244 return bs->read_only;
247 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
248 bool ignore_allow_rdw, Error **errp)
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs->copy_on_read && read_only) {
252 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs));
254 return -EINVAL;
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
259 !ignore_allow_rdw)
261 error_setg(errp, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs));
263 return -EPERM;
266 return 0;
269 /* TODO Remove (deprecated since 2.11)
270 * Block drivers are not supposed to automatically change bs->read_only.
271 * Instead, they should just check whether they can provide what the user
272 * explicitly requested and error out if read-write is requested, but they can
273 * only provide read-only access. */
274 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
276 int ret = 0;
278 ret = bdrv_can_set_read_only(bs, read_only, false, errp);
279 if (ret < 0) {
280 return ret;
283 bs->read_only = read_only;
284 return 0;
287 void bdrv_get_full_backing_filename_from_filename(const char *backed,
288 const char *backing,
289 char *dest, size_t sz,
290 Error **errp)
292 if (backing[0] == '\0' || path_has_protocol(backing) ||
293 path_is_absolute(backing))
295 pstrcpy(dest, sz, backing);
296 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
297 error_setg(errp, "Cannot use relative backing file names for '%s'",
298 backed);
299 } else {
300 path_combine(dest, sz, backed, backing);
304 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
305 Error **errp)
307 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
309 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
310 dest, sz, errp);
313 void bdrv_register(BlockDriver *bdrv)
315 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
318 BlockDriverState *bdrv_new(void)
320 BlockDriverState *bs;
321 int i;
323 bs = g_new0(BlockDriverState, 1);
324 QLIST_INIT(&bs->dirty_bitmaps);
325 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
326 QLIST_INIT(&bs->op_blockers[i]);
328 notifier_with_return_list_init(&bs->before_write_notifiers);
329 qemu_co_mutex_init(&bs->reqs_lock);
330 qemu_mutex_init(&bs->dirty_bitmap_mutex);
331 bs->refcnt = 1;
332 bs->aio_context = qemu_get_aio_context();
334 qemu_co_queue_init(&bs->flush_queue);
336 for (i = 0; i < bdrv_drain_all_count; i++) {
337 bdrv_drained_begin(bs);
340 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
342 return bs;
345 static BlockDriver *bdrv_do_find_format(const char *format_name)
347 BlockDriver *drv1;
349 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
350 if (!strcmp(drv1->format_name, format_name)) {
351 return drv1;
355 return NULL;
358 BlockDriver *bdrv_find_format(const char *format_name)
360 BlockDriver *drv1;
361 int i;
363 drv1 = bdrv_do_find_format(format_name);
364 if (drv1) {
365 return drv1;
368 /* The driver isn't registered, maybe we need to load a module */
369 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
370 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
371 block_module_load_one(block_driver_modules[i].library_name);
372 break;
376 return bdrv_do_find_format(format_name);
379 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
381 static const char *whitelist_rw[] = {
382 CONFIG_BDRV_RW_WHITELIST
384 static const char *whitelist_ro[] = {
385 CONFIG_BDRV_RO_WHITELIST
387 const char **p;
389 if (!whitelist_rw[0] && !whitelist_ro[0]) {
390 return 1; /* no whitelist, anything goes */
393 for (p = whitelist_rw; *p; p++) {
394 if (!strcmp(drv->format_name, *p)) {
395 return 1;
398 if (read_only) {
399 for (p = whitelist_ro; *p; p++) {
400 if (!strcmp(drv->format_name, *p)) {
401 return 1;
405 return 0;
408 bool bdrv_uses_whitelist(void)
410 return use_bdrv_whitelist;
413 typedef struct CreateCo {
414 BlockDriver *drv;
415 char *filename;
416 QemuOpts *opts;
417 int ret;
418 Error *err;
419 } CreateCo;
421 static void coroutine_fn bdrv_create_co_entry(void *opaque)
423 Error *local_err = NULL;
424 int ret;
426 CreateCo *cco = opaque;
427 assert(cco->drv);
429 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
430 error_propagate(&cco->err, local_err);
431 cco->ret = ret;
434 int bdrv_create(BlockDriver *drv, const char* filename,
435 QemuOpts *opts, Error **errp)
437 int ret;
439 Coroutine *co;
440 CreateCo cco = {
441 .drv = drv,
442 .filename = g_strdup(filename),
443 .opts = opts,
444 .ret = NOT_DONE,
445 .err = NULL,
448 if (!drv->bdrv_co_create_opts) {
449 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
450 ret = -ENOTSUP;
451 goto out;
454 if (qemu_in_coroutine()) {
455 /* Fast-path if already in coroutine context */
456 bdrv_create_co_entry(&cco);
457 } else {
458 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
459 qemu_coroutine_enter(co);
460 while (cco.ret == NOT_DONE) {
461 aio_poll(qemu_get_aio_context(), true);
465 ret = cco.ret;
466 if (ret < 0) {
467 if (cco.err) {
468 error_propagate(errp, cco.err);
469 } else {
470 error_setg_errno(errp, -ret, "Could not create image");
474 out:
475 g_free(cco.filename);
476 return ret;
479 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
481 BlockDriver *drv;
482 Error *local_err = NULL;
483 int ret;
485 drv = bdrv_find_protocol(filename, true, errp);
486 if (drv == NULL) {
487 return -ENOENT;
490 ret = bdrv_create(drv, filename, opts, &local_err);
491 error_propagate(errp, local_err);
492 return ret;
496 * Try to get @bs's logical and physical block size.
497 * On success, store them in @bsz struct and return 0.
498 * On failure return -errno.
499 * @bs must not be empty.
501 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
503 BlockDriver *drv = bs->drv;
505 if (drv && drv->bdrv_probe_blocksizes) {
506 return drv->bdrv_probe_blocksizes(bs, bsz);
507 } else if (drv && drv->is_filter && bs->file) {
508 return bdrv_probe_blocksizes(bs->file->bs, bsz);
511 return -ENOTSUP;
515 * Try to get @bs's geometry (cyls, heads, sectors).
516 * On success, store them in @geo struct and return 0.
517 * On failure return -errno.
518 * @bs must not be empty.
520 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
522 BlockDriver *drv = bs->drv;
524 if (drv && drv->bdrv_probe_geometry) {
525 return drv->bdrv_probe_geometry(bs, geo);
526 } else if (drv && drv->is_filter && bs->file) {
527 return bdrv_probe_geometry(bs->file->bs, geo);
530 return -ENOTSUP;
534 * Create a uniquely-named empty temporary file.
535 * Return 0 upon success, otherwise a negative errno value.
537 int get_tmp_filename(char *filename, int size)
539 #ifdef _WIN32
540 char temp_dir[MAX_PATH];
541 /* GetTempFileName requires that its output buffer (4th param)
542 have length MAX_PATH or greater. */
543 assert(size >= MAX_PATH);
544 return (GetTempPath(MAX_PATH, temp_dir)
545 && GetTempFileName(temp_dir, "qem", 0, filename)
546 ? 0 : -GetLastError());
547 #else
548 int fd;
549 const char *tmpdir;
550 tmpdir = getenv("TMPDIR");
551 if (!tmpdir) {
552 tmpdir = "/var/tmp";
554 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
555 return -EOVERFLOW;
557 fd = mkstemp(filename);
558 if (fd < 0) {
559 return -errno;
561 if (close(fd) != 0) {
562 unlink(filename);
563 return -errno;
565 return 0;
566 #endif
570 * Detect host devices. By convention, /dev/cdrom[N] is always
571 * recognized as a host CDROM.
573 static BlockDriver *find_hdev_driver(const char *filename)
575 int score_max = 0, score;
576 BlockDriver *drv = NULL, *d;
578 QLIST_FOREACH(d, &bdrv_drivers, list) {
579 if (d->bdrv_probe_device) {
580 score = d->bdrv_probe_device(filename);
581 if (score > score_max) {
582 score_max = score;
583 drv = d;
588 return drv;
591 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
593 BlockDriver *drv1;
595 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
596 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
597 return drv1;
601 return NULL;
604 BlockDriver *bdrv_find_protocol(const char *filename,
605 bool allow_protocol_prefix,
606 Error **errp)
608 BlockDriver *drv1;
609 char protocol[128];
610 int len;
611 const char *p;
612 int i;
614 /* TODO Drivers without bdrv_file_open must be specified explicitly */
617 * XXX(hch): we really should not let host device detection
618 * override an explicit protocol specification, but moving this
619 * later breaks access to device names with colons in them.
620 * Thanks to the brain-dead persistent naming schemes on udev-
621 * based Linux systems those actually are quite common.
623 drv1 = find_hdev_driver(filename);
624 if (drv1) {
625 return drv1;
628 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
629 return &bdrv_file;
632 p = strchr(filename, ':');
633 assert(p != NULL);
634 len = p - filename;
635 if (len > sizeof(protocol) - 1)
636 len = sizeof(protocol) - 1;
637 memcpy(protocol, filename, len);
638 protocol[len] = '\0';
640 drv1 = bdrv_do_find_protocol(protocol);
641 if (drv1) {
642 return drv1;
645 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
646 if (block_driver_modules[i].protocol_name &&
647 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
648 block_module_load_one(block_driver_modules[i].library_name);
649 break;
653 drv1 = bdrv_do_find_protocol(protocol);
654 if (!drv1) {
655 error_setg(errp, "Unknown protocol '%s'", protocol);
657 return drv1;
661 * Guess image format by probing its contents.
662 * This is not a good idea when your image is raw (CVE-2008-2004), but
663 * we do it anyway for backward compatibility.
665 * @buf contains the image's first @buf_size bytes.
666 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
667 * but can be smaller if the image file is smaller)
668 * @filename is its filename.
670 * For all block drivers, call the bdrv_probe() method to get its
671 * probing score.
672 * Return the first block driver with the highest probing score.
674 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
675 const char *filename)
677 int score_max = 0, score;
678 BlockDriver *drv = NULL, *d;
680 QLIST_FOREACH(d, &bdrv_drivers, list) {
681 if (d->bdrv_probe) {
682 score = d->bdrv_probe(buf, buf_size, filename);
683 if (score > score_max) {
684 score_max = score;
685 drv = d;
690 return drv;
693 static int find_image_format(BlockBackend *file, const char *filename,
694 BlockDriver **pdrv, Error **errp)
696 BlockDriver *drv;
697 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
698 int ret = 0;
700 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
701 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
702 *pdrv = &bdrv_raw;
703 return ret;
706 ret = blk_pread(file, 0, buf, sizeof(buf));
707 if (ret < 0) {
708 error_setg_errno(errp, -ret, "Could not read image for determining its "
709 "format");
710 *pdrv = NULL;
711 return ret;
714 drv = bdrv_probe_all(buf, ret, filename);
715 if (!drv) {
716 error_setg(errp, "Could not determine image format: No compatible "
717 "driver found");
718 ret = -ENOENT;
720 *pdrv = drv;
721 return ret;
725 * Set the current 'total_sectors' value
726 * Return 0 on success, -errno on error.
728 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
730 BlockDriver *drv = bs->drv;
732 if (!drv) {
733 return -ENOMEDIUM;
736 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
737 if (bdrv_is_sg(bs))
738 return 0;
740 /* query actual device if possible, otherwise just trust the hint */
741 if (drv->bdrv_getlength) {
742 int64_t length = drv->bdrv_getlength(bs);
743 if (length < 0) {
744 return length;
746 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
749 bs->total_sectors = hint;
750 return 0;
754 * Combines a QDict of new block driver @options with any missing options taken
755 * from @old_options, so that leaving out an option defaults to its old value.
757 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
758 QDict *old_options)
760 if (bs->drv && bs->drv->bdrv_join_options) {
761 bs->drv->bdrv_join_options(options, old_options);
762 } else {
763 qdict_join(options, old_options, false);
768 * Set open flags for a given discard mode
770 * Return 0 on success, -1 if the discard mode was invalid.
772 int bdrv_parse_discard_flags(const char *mode, int *flags)
774 *flags &= ~BDRV_O_UNMAP;
776 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
777 /* do nothing */
778 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
779 *flags |= BDRV_O_UNMAP;
780 } else {
781 return -1;
784 return 0;
788 * Set open flags for a given cache mode
790 * Return 0 on success, -1 if the cache mode was invalid.
792 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
794 *flags &= ~BDRV_O_CACHE_MASK;
796 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
797 *writethrough = false;
798 *flags |= BDRV_O_NOCACHE;
799 } else if (!strcmp(mode, "directsync")) {
800 *writethrough = true;
801 *flags |= BDRV_O_NOCACHE;
802 } else if (!strcmp(mode, "writeback")) {
803 *writethrough = false;
804 } else if (!strcmp(mode, "unsafe")) {
805 *writethrough = false;
806 *flags |= BDRV_O_NO_FLUSH;
807 } else if (!strcmp(mode, "writethrough")) {
808 *writethrough = true;
809 } else {
810 return -1;
813 return 0;
816 static char *bdrv_child_get_parent_desc(BdrvChild *c)
818 BlockDriverState *parent = c->opaque;
819 return g_strdup(bdrv_get_device_or_node_name(parent));
822 static void bdrv_child_cb_drained_begin(BdrvChild *child)
824 BlockDriverState *bs = child->opaque;
825 bdrv_do_drained_begin_quiesce(bs, NULL, false);
828 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
830 BlockDriverState *bs = child->opaque;
831 return bdrv_drain_poll(bs, false, NULL, false);
834 static void bdrv_child_cb_drained_end(BdrvChild *child)
836 BlockDriverState *bs = child->opaque;
837 bdrv_drained_end(bs);
840 static void bdrv_child_cb_attach(BdrvChild *child)
842 BlockDriverState *bs = child->opaque;
843 bdrv_apply_subtree_drain(child, bs);
846 static void bdrv_child_cb_detach(BdrvChild *child)
848 BlockDriverState *bs = child->opaque;
849 bdrv_unapply_subtree_drain(child, bs);
852 static int bdrv_child_cb_inactivate(BdrvChild *child)
854 BlockDriverState *bs = child->opaque;
855 assert(bs->open_flags & BDRV_O_INACTIVE);
856 return 0;
860 * Returns the options and flags that a temporary snapshot should get, based on
861 * the originally requested flags (the originally requested image will have
862 * flags like a backing file)
864 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
865 int parent_flags, QDict *parent_options)
867 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
869 /* For temporary files, unconditional cache=unsafe is fine */
870 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
871 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
873 /* Copy the read-only option from the parent */
874 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
876 /* aio=native doesn't work for cache.direct=off, so disable it for the
877 * temporary snapshot */
878 *child_flags &= ~BDRV_O_NATIVE_AIO;
882 * Returns the options and flags that bs->file should get if a protocol driver
883 * is expected, based on the given options and flags for the parent BDS
885 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
886 int parent_flags, QDict *parent_options)
888 int flags = parent_flags;
890 /* Enable protocol handling, disable format probing for bs->file */
891 flags |= BDRV_O_PROTOCOL;
893 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
894 * the parent. */
895 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
896 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
897 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
899 /* Inherit the read-only option from the parent if it's not set */
900 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
902 /* Our block drivers take care to send flushes and respect unmap policy,
903 * so we can default to enable both on lower layers regardless of the
904 * corresponding parent options. */
905 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
907 /* Clear flags that only apply to the top layer */
908 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
909 BDRV_O_NO_IO);
911 *child_flags = flags;
914 const BdrvChildRole child_file = {
915 .parent_is_bds = true,
916 .get_parent_desc = bdrv_child_get_parent_desc,
917 .inherit_options = bdrv_inherited_options,
918 .drained_begin = bdrv_child_cb_drained_begin,
919 .drained_poll = bdrv_child_cb_drained_poll,
920 .drained_end = bdrv_child_cb_drained_end,
921 .attach = bdrv_child_cb_attach,
922 .detach = bdrv_child_cb_detach,
923 .inactivate = bdrv_child_cb_inactivate,
927 * Returns the options and flags that bs->file should get if the use of formats
928 * (and not only protocols) is permitted for it, based on the given options and
929 * flags for the parent BDS
931 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
932 int parent_flags, QDict *parent_options)
934 child_file.inherit_options(child_flags, child_options,
935 parent_flags, parent_options);
937 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
940 const BdrvChildRole child_format = {
941 .parent_is_bds = true,
942 .get_parent_desc = bdrv_child_get_parent_desc,
943 .inherit_options = bdrv_inherited_fmt_options,
944 .drained_begin = bdrv_child_cb_drained_begin,
945 .drained_poll = bdrv_child_cb_drained_poll,
946 .drained_end = bdrv_child_cb_drained_end,
947 .attach = bdrv_child_cb_attach,
948 .detach = bdrv_child_cb_detach,
949 .inactivate = bdrv_child_cb_inactivate,
952 static void bdrv_backing_attach(BdrvChild *c)
954 BlockDriverState *parent = c->opaque;
955 BlockDriverState *backing_hd = c->bs;
957 assert(!parent->backing_blocker);
958 error_setg(&parent->backing_blocker,
959 "node is used as backing hd of '%s'",
960 bdrv_get_device_or_node_name(parent));
962 parent->open_flags &= ~BDRV_O_NO_BACKING;
963 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
964 backing_hd->filename);
965 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
966 backing_hd->drv ? backing_hd->drv->format_name : "");
968 bdrv_op_block_all(backing_hd, parent->backing_blocker);
969 /* Otherwise we won't be able to commit or stream */
970 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
971 parent->backing_blocker);
972 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
973 parent->backing_blocker);
975 * We do backup in 3 ways:
976 * 1. drive backup
977 * The target bs is new opened, and the source is top BDS
978 * 2. blockdev backup
979 * Both the source and the target are top BDSes.
980 * 3. internal backup(used for block replication)
981 * Both the source and the target are backing file
983 * In case 1 and 2, neither the source nor the target is the backing file.
984 * In case 3, we will block the top BDS, so there is only one block job
985 * for the top BDS and its backing chain.
987 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
988 parent->backing_blocker);
989 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
990 parent->backing_blocker);
992 bdrv_child_cb_attach(c);
995 static void bdrv_backing_detach(BdrvChild *c)
997 BlockDriverState *parent = c->opaque;
999 assert(parent->backing_blocker);
1000 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1001 error_free(parent->backing_blocker);
1002 parent->backing_blocker = NULL;
1004 bdrv_child_cb_detach(c);
1008 * Returns the options and flags that bs->backing should get, based on the
1009 * given options and flags for the parent BDS
1011 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1012 int parent_flags, QDict *parent_options)
1014 int flags = parent_flags;
1016 /* The cache mode is inherited unmodified for backing files; except WCE,
1017 * which is only applied on the top level (BlockBackend) */
1018 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1019 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1020 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1022 /* backing files always opened read-only */
1023 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1024 flags &= ~BDRV_O_COPY_ON_READ;
1026 /* snapshot=on is handled on the top layer */
1027 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1029 *child_flags = flags;
1032 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1033 const char *filename, Error **errp)
1035 BlockDriverState *parent = c->opaque;
1036 int orig_flags = bdrv_get_flags(parent);
1037 int ret;
1039 if (!(orig_flags & BDRV_O_RDWR)) {
1040 ret = bdrv_reopen(parent, orig_flags | BDRV_O_RDWR, errp);
1041 if (ret < 0) {
1042 return ret;
1046 ret = bdrv_change_backing_file(parent, filename,
1047 base->drv ? base->drv->format_name : "");
1048 if (ret < 0) {
1049 error_setg_errno(errp, -ret, "Could not update backing file link");
1052 if (!(orig_flags & BDRV_O_RDWR)) {
1053 bdrv_reopen(parent, orig_flags, NULL);
1056 return ret;
1059 const BdrvChildRole child_backing = {
1060 .parent_is_bds = true,
1061 .get_parent_desc = bdrv_child_get_parent_desc,
1062 .attach = bdrv_backing_attach,
1063 .detach = bdrv_backing_detach,
1064 .inherit_options = bdrv_backing_options,
1065 .drained_begin = bdrv_child_cb_drained_begin,
1066 .drained_poll = bdrv_child_cb_drained_poll,
1067 .drained_end = bdrv_child_cb_drained_end,
1068 .inactivate = bdrv_child_cb_inactivate,
1069 .update_filename = bdrv_backing_update_filename,
1072 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1074 int open_flags = flags;
1077 * Clear flags that are internal to the block layer before opening the
1078 * image.
1080 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1083 * Snapshots should be writable.
1085 if (flags & BDRV_O_TEMPORARY) {
1086 open_flags |= BDRV_O_RDWR;
1089 return open_flags;
1092 static void update_flags_from_options(int *flags, QemuOpts *opts)
1094 *flags &= ~BDRV_O_CACHE_MASK;
1096 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1097 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1098 *flags |= BDRV_O_NO_FLUSH;
1101 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1102 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1103 *flags |= BDRV_O_NOCACHE;
1106 *flags &= ~BDRV_O_RDWR;
1108 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1109 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1110 *flags |= BDRV_O_RDWR;
1115 static void update_options_from_flags(QDict *options, int flags)
1117 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1118 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1120 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1121 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1122 flags & BDRV_O_NO_FLUSH);
1124 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1125 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1129 static void bdrv_assign_node_name(BlockDriverState *bs,
1130 const char *node_name,
1131 Error **errp)
1133 char *gen_node_name = NULL;
1135 if (!node_name) {
1136 node_name = gen_node_name = id_generate(ID_BLOCK);
1137 } else if (!id_wellformed(node_name)) {
1139 * Check for empty string or invalid characters, but not if it is
1140 * generated (generated names use characters not available to the user)
1142 error_setg(errp, "Invalid node name");
1143 return;
1146 /* takes care of avoiding namespaces collisions */
1147 if (blk_by_name(node_name)) {
1148 error_setg(errp, "node-name=%s is conflicting with a device id",
1149 node_name);
1150 goto out;
1153 /* takes care of avoiding duplicates node names */
1154 if (bdrv_find_node(node_name)) {
1155 error_setg(errp, "Duplicate node name");
1156 goto out;
1159 /* Make sure that the node name isn't truncated */
1160 if (strlen(node_name) >= sizeof(bs->node_name)) {
1161 error_setg(errp, "Node name too long");
1162 goto out;
1165 /* copy node name into the bs and insert it into the graph list */
1166 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1167 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1168 out:
1169 g_free(gen_node_name);
1172 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1173 const char *node_name, QDict *options,
1174 int open_flags, Error **errp)
1176 Error *local_err = NULL;
1177 int i, ret;
1179 bdrv_assign_node_name(bs, node_name, &local_err);
1180 if (local_err) {
1181 error_propagate(errp, local_err);
1182 return -EINVAL;
1185 bs->drv = drv;
1186 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1187 bs->opaque = g_malloc0(drv->instance_size);
1189 if (drv->bdrv_file_open) {
1190 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1191 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1192 } else if (drv->bdrv_open) {
1193 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1194 } else {
1195 ret = 0;
1198 if (ret < 0) {
1199 if (local_err) {
1200 error_propagate(errp, local_err);
1201 } else if (bs->filename[0]) {
1202 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1203 } else {
1204 error_setg_errno(errp, -ret, "Could not open image");
1206 goto open_failed;
1209 ret = refresh_total_sectors(bs, bs->total_sectors);
1210 if (ret < 0) {
1211 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1212 return ret;
1215 bdrv_refresh_limits(bs, &local_err);
1216 if (local_err) {
1217 error_propagate(errp, local_err);
1218 return -EINVAL;
1221 assert(bdrv_opt_mem_align(bs) != 0);
1222 assert(bdrv_min_mem_align(bs) != 0);
1223 assert(is_power_of_2(bs->bl.request_alignment));
1225 for (i = 0; i < bs->quiesce_counter; i++) {
1226 if (drv->bdrv_co_drain_begin) {
1227 drv->bdrv_co_drain_begin(bs);
1231 return 0;
1232 open_failed:
1233 bs->drv = NULL;
1234 if (bs->file != NULL) {
1235 bdrv_unref_child(bs, bs->file);
1236 bs->file = NULL;
1238 g_free(bs->opaque);
1239 bs->opaque = NULL;
1240 return ret;
1243 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1244 int flags, Error **errp)
1246 BlockDriverState *bs;
1247 int ret;
1249 bs = bdrv_new();
1250 bs->open_flags = flags;
1251 bs->explicit_options = qdict_new();
1252 bs->options = qdict_new();
1253 bs->opaque = NULL;
1255 update_options_from_flags(bs->options, flags);
1257 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1258 if (ret < 0) {
1259 qobject_unref(bs->explicit_options);
1260 bs->explicit_options = NULL;
1261 qobject_unref(bs->options);
1262 bs->options = NULL;
1263 bdrv_unref(bs);
1264 return NULL;
1267 return bs;
1270 QemuOptsList bdrv_runtime_opts = {
1271 .name = "bdrv_common",
1272 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1273 .desc = {
1275 .name = "node-name",
1276 .type = QEMU_OPT_STRING,
1277 .help = "Node name of the block device node",
1280 .name = "driver",
1281 .type = QEMU_OPT_STRING,
1282 .help = "Block driver to use for the node",
1285 .name = BDRV_OPT_CACHE_DIRECT,
1286 .type = QEMU_OPT_BOOL,
1287 .help = "Bypass software writeback cache on the host",
1290 .name = BDRV_OPT_CACHE_NO_FLUSH,
1291 .type = QEMU_OPT_BOOL,
1292 .help = "Ignore flush requests",
1295 .name = BDRV_OPT_READ_ONLY,
1296 .type = QEMU_OPT_BOOL,
1297 .help = "Node is opened in read-only mode",
1300 .name = "detect-zeroes",
1301 .type = QEMU_OPT_STRING,
1302 .help = "try to optimize zero writes (off, on, unmap)",
1305 .name = "discard",
1306 .type = QEMU_OPT_STRING,
1307 .help = "discard operation (ignore/off, unmap/on)",
1310 .name = BDRV_OPT_FORCE_SHARE,
1311 .type = QEMU_OPT_BOOL,
1312 .help = "always accept other writers (default: off)",
1314 { /* end of list */ }
1319 * Common part for opening disk images and files
1321 * Removes all processed options from *options.
1323 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1324 QDict *options, Error **errp)
1326 int ret, open_flags;
1327 const char *filename;
1328 const char *driver_name = NULL;
1329 const char *node_name = NULL;
1330 const char *discard;
1331 const char *detect_zeroes;
1332 QemuOpts *opts;
1333 BlockDriver *drv;
1334 Error *local_err = NULL;
1336 assert(bs->file == NULL);
1337 assert(options != NULL && bs->options != options);
1339 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1340 qemu_opts_absorb_qdict(opts, options, &local_err);
1341 if (local_err) {
1342 error_propagate(errp, local_err);
1343 ret = -EINVAL;
1344 goto fail_opts;
1347 update_flags_from_options(&bs->open_flags, opts);
1349 driver_name = qemu_opt_get(opts, "driver");
1350 drv = bdrv_find_format(driver_name);
1351 assert(drv != NULL);
1353 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1355 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1356 error_setg(errp,
1357 BDRV_OPT_FORCE_SHARE
1358 "=on can only be used with read-only images");
1359 ret = -EINVAL;
1360 goto fail_opts;
1363 if (file != NULL) {
1364 filename = blk_bs(file)->filename;
1365 } else {
1367 * Caution: while qdict_get_try_str() is fine, getting
1368 * non-string types would require more care. When @options
1369 * come from -blockdev or blockdev_add, its members are typed
1370 * according to the QAPI schema, but when they come from
1371 * -drive, they're all QString.
1373 filename = qdict_get_try_str(options, "filename");
1376 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1377 error_setg(errp, "The '%s' block driver requires a file name",
1378 drv->format_name);
1379 ret = -EINVAL;
1380 goto fail_opts;
1383 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1384 drv->format_name);
1386 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1388 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1389 error_setg(errp,
1390 !bs->read_only && bdrv_is_whitelisted(drv, true)
1391 ? "Driver '%s' can only be used for read-only devices"
1392 : "Driver '%s' is not whitelisted",
1393 drv->format_name);
1394 ret = -ENOTSUP;
1395 goto fail_opts;
1398 /* bdrv_new() and bdrv_close() make it so */
1399 assert(atomic_read(&bs->copy_on_read) == 0);
1401 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1402 if (!bs->read_only) {
1403 bdrv_enable_copy_on_read(bs);
1404 } else {
1405 error_setg(errp, "Can't use copy-on-read on read-only device");
1406 ret = -EINVAL;
1407 goto fail_opts;
1411 discard = qemu_opt_get(opts, "discard");
1412 if (discard != NULL) {
1413 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1414 error_setg(errp, "Invalid discard option");
1415 ret = -EINVAL;
1416 goto fail_opts;
1420 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1421 if (detect_zeroes) {
1422 BlockdevDetectZeroesOptions value =
1423 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
1424 detect_zeroes,
1425 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1426 &local_err);
1427 if (local_err) {
1428 error_propagate(errp, local_err);
1429 ret = -EINVAL;
1430 goto fail_opts;
1433 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1434 !(bs->open_flags & BDRV_O_UNMAP))
1436 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1437 "without setting discard operation to unmap");
1438 ret = -EINVAL;
1439 goto fail_opts;
1442 bs->detect_zeroes = value;
1445 if (filename != NULL) {
1446 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1447 } else {
1448 bs->filename[0] = '\0';
1450 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1452 /* Open the image, either directly or using a protocol */
1453 open_flags = bdrv_open_flags(bs, bs->open_flags);
1454 node_name = qemu_opt_get(opts, "node-name");
1456 assert(!drv->bdrv_file_open || file == NULL);
1457 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1458 if (ret < 0) {
1459 goto fail_opts;
1462 qemu_opts_del(opts);
1463 return 0;
1465 fail_opts:
1466 qemu_opts_del(opts);
1467 return ret;
1470 static QDict *parse_json_filename(const char *filename, Error **errp)
1472 QObject *options_obj;
1473 QDict *options;
1474 int ret;
1476 ret = strstart(filename, "json:", &filename);
1477 assert(ret);
1479 options_obj = qobject_from_json(filename, errp);
1480 if (!options_obj) {
1481 error_prepend(errp, "Could not parse the JSON options: ");
1482 return NULL;
1485 options = qobject_to(QDict, options_obj);
1486 if (!options) {
1487 qobject_unref(options_obj);
1488 error_setg(errp, "Invalid JSON object given");
1489 return NULL;
1492 qdict_flatten(options);
1494 return options;
1497 static void parse_json_protocol(QDict *options, const char **pfilename,
1498 Error **errp)
1500 QDict *json_options;
1501 Error *local_err = NULL;
1503 /* Parse json: pseudo-protocol */
1504 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1505 return;
1508 json_options = parse_json_filename(*pfilename, &local_err);
1509 if (local_err) {
1510 error_propagate(errp, local_err);
1511 return;
1514 /* Options given in the filename have lower priority than options
1515 * specified directly */
1516 qdict_join(options, json_options, false);
1517 qobject_unref(json_options);
1518 *pfilename = NULL;
1522 * Fills in default options for opening images and converts the legacy
1523 * filename/flags pair to option QDict entries.
1524 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1525 * block driver has been specified explicitly.
1527 static int bdrv_fill_options(QDict **options, const char *filename,
1528 int *flags, Error **errp)
1530 const char *drvname;
1531 bool protocol = *flags & BDRV_O_PROTOCOL;
1532 bool parse_filename = false;
1533 BlockDriver *drv = NULL;
1534 Error *local_err = NULL;
1537 * Caution: while qdict_get_try_str() is fine, getting non-string
1538 * types would require more care. When @options come from
1539 * -blockdev or blockdev_add, its members are typed according to
1540 * the QAPI schema, but when they come from -drive, they're all
1541 * QString.
1543 drvname = qdict_get_try_str(*options, "driver");
1544 if (drvname) {
1545 drv = bdrv_find_format(drvname);
1546 if (!drv) {
1547 error_setg(errp, "Unknown driver '%s'", drvname);
1548 return -ENOENT;
1550 /* If the user has explicitly specified the driver, this choice should
1551 * override the BDRV_O_PROTOCOL flag */
1552 protocol = drv->bdrv_file_open;
1555 if (protocol) {
1556 *flags |= BDRV_O_PROTOCOL;
1557 } else {
1558 *flags &= ~BDRV_O_PROTOCOL;
1561 /* Translate cache options from flags into options */
1562 update_options_from_flags(*options, *flags);
1564 /* Fetch the file name from the options QDict if necessary */
1565 if (protocol && filename) {
1566 if (!qdict_haskey(*options, "filename")) {
1567 qdict_put_str(*options, "filename", filename);
1568 parse_filename = true;
1569 } else {
1570 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1571 "the same time");
1572 return -EINVAL;
1576 /* Find the right block driver */
1577 /* See cautionary note on accessing @options above */
1578 filename = qdict_get_try_str(*options, "filename");
1580 if (!drvname && protocol) {
1581 if (filename) {
1582 drv = bdrv_find_protocol(filename, parse_filename, errp);
1583 if (!drv) {
1584 return -EINVAL;
1587 drvname = drv->format_name;
1588 qdict_put_str(*options, "driver", drvname);
1589 } else {
1590 error_setg(errp, "Must specify either driver or file");
1591 return -EINVAL;
1595 assert(drv || !protocol);
1597 /* Driver-specific filename parsing */
1598 if (drv && drv->bdrv_parse_filename && parse_filename) {
1599 drv->bdrv_parse_filename(filename, *options, &local_err);
1600 if (local_err) {
1601 error_propagate(errp, local_err);
1602 return -EINVAL;
1605 if (!drv->bdrv_needs_filename) {
1606 qdict_del(*options, "filename");
1610 return 0;
1613 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1614 uint64_t perm, uint64_t shared,
1615 GSList *ignore_children, Error **errp);
1616 static void bdrv_child_abort_perm_update(BdrvChild *c);
1617 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1619 typedef struct BlockReopenQueueEntry {
1620 bool prepared;
1621 BDRVReopenState state;
1622 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1623 } BlockReopenQueueEntry;
1626 * Return the flags that @bs will have after the reopens in @q have
1627 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1628 * return the current flags.
1630 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1632 BlockReopenQueueEntry *entry;
1634 if (q != NULL) {
1635 QSIMPLEQ_FOREACH(entry, q, entry) {
1636 if (entry->state.bs == bs) {
1637 return entry->state.flags;
1642 return bs->open_flags;
1645 /* Returns whether the image file can be written to after the reopen queue @q
1646 * has been successfully applied, or right now if @q is NULL. */
1647 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1648 BlockReopenQueue *q)
1650 int flags = bdrv_reopen_get_flags(q, bs);
1652 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1656 * Return whether the BDS can be written to. This is not necessarily
1657 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1658 * be written to but do not count as read-only images.
1660 bool bdrv_is_writable(BlockDriverState *bs)
1662 return bdrv_is_writable_after_reopen(bs, NULL);
1665 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1666 BdrvChild *c, const BdrvChildRole *role,
1667 BlockReopenQueue *reopen_queue,
1668 uint64_t parent_perm, uint64_t parent_shared,
1669 uint64_t *nperm, uint64_t *nshared)
1671 if (bs->drv && bs->drv->bdrv_child_perm) {
1672 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1673 parent_perm, parent_shared,
1674 nperm, nshared);
1676 /* TODO Take force_share from reopen_queue */
1677 if (child_bs && child_bs->force_share) {
1678 *nshared = BLK_PERM_ALL;
1683 * Check whether permissions on this node can be changed in a way that
1684 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1685 * permissions of all its parents. This involves checking whether all necessary
1686 * permission changes to child nodes can be performed.
1688 * A call to this function must always be followed by a call to bdrv_set_perm()
1689 * or bdrv_abort_perm_update().
1691 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1692 uint64_t cumulative_perms,
1693 uint64_t cumulative_shared_perms,
1694 GSList *ignore_children, Error **errp)
1696 BlockDriver *drv = bs->drv;
1697 BdrvChild *c;
1698 int ret;
1700 /* Write permissions never work with read-only images */
1701 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1702 !bdrv_is_writable_after_reopen(bs, q))
1704 error_setg(errp, "Block node is read-only");
1705 return -EPERM;
1708 /* Check this node */
1709 if (!drv) {
1710 return 0;
1713 if (drv->bdrv_check_perm) {
1714 return drv->bdrv_check_perm(bs, cumulative_perms,
1715 cumulative_shared_perms, errp);
1718 /* Drivers that never have children can omit .bdrv_child_perm() */
1719 if (!drv->bdrv_child_perm) {
1720 assert(QLIST_EMPTY(&bs->children));
1721 return 0;
1724 /* Check all children */
1725 QLIST_FOREACH(c, &bs->children, next) {
1726 uint64_t cur_perm, cur_shared;
1727 bdrv_child_perm(bs, c->bs, c, c->role, q,
1728 cumulative_perms, cumulative_shared_perms,
1729 &cur_perm, &cur_shared);
1730 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1731 ignore_children, errp);
1732 if (ret < 0) {
1733 return ret;
1737 return 0;
1741 * Notifies drivers that after a previous bdrv_check_perm() call, the
1742 * permission update is not performed and any preparations made for it (e.g.
1743 * taken file locks) need to be undone.
1745 * This function recursively notifies all child nodes.
1747 static void bdrv_abort_perm_update(BlockDriverState *bs)
1749 BlockDriver *drv = bs->drv;
1750 BdrvChild *c;
1752 if (!drv) {
1753 return;
1756 if (drv->bdrv_abort_perm_update) {
1757 drv->bdrv_abort_perm_update(bs);
1760 QLIST_FOREACH(c, &bs->children, next) {
1761 bdrv_child_abort_perm_update(c);
1765 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1766 uint64_t cumulative_shared_perms)
1768 BlockDriver *drv = bs->drv;
1769 BdrvChild *c;
1771 if (!drv) {
1772 return;
1775 /* Update this node */
1776 if (drv->bdrv_set_perm) {
1777 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1780 /* Drivers that never have children can omit .bdrv_child_perm() */
1781 if (!drv->bdrv_child_perm) {
1782 assert(QLIST_EMPTY(&bs->children));
1783 return;
1786 /* Update all children */
1787 QLIST_FOREACH(c, &bs->children, next) {
1788 uint64_t cur_perm, cur_shared;
1789 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1790 cumulative_perms, cumulative_shared_perms,
1791 &cur_perm, &cur_shared);
1792 bdrv_child_set_perm(c, cur_perm, cur_shared);
1796 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1797 uint64_t *shared_perm)
1799 BdrvChild *c;
1800 uint64_t cumulative_perms = 0;
1801 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1803 QLIST_FOREACH(c, &bs->parents, next_parent) {
1804 cumulative_perms |= c->perm;
1805 cumulative_shared_perms &= c->shared_perm;
1808 *perm = cumulative_perms;
1809 *shared_perm = cumulative_shared_perms;
1812 static char *bdrv_child_user_desc(BdrvChild *c)
1814 if (c->role->get_parent_desc) {
1815 return c->role->get_parent_desc(c);
1818 return g_strdup("another user");
1821 char *bdrv_perm_names(uint64_t perm)
1823 struct perm_name {
1824 uint64_t perm;
1825 const char *name;
1826 } permissions[] = {
1827 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1828 { BLK_PERM_WRITE, "write" },
1829 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1830 { BLK_PERM_RESIZE, "resize" },
1831 { BLK_PERM_GRAPH_MOD, "change children" },
1832 { 0, NULL }
1835 char *result = g_strdup("");
1836 struct perm_name *p;
1838 for (p = permissions; p->name; p++) {
1839 if (perm & p->perm) {
1840 char *old = result;
1841 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1842 g_free(old);
1846 return result;
1850 * Checks whether a new reference to @bs can be added if the new user requires
1851 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1852 * set, the BdrvChild objects in this list are ignored in the calculations;
1853 * this allows checking permission updates for an existing reference.
1855 * Needs to be followed by a call to either bdrv_set_perm() or
1856 * bdrv_abort_perm_update(). */
1857 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1858 uint64_t new_used_perm,
1859 uint64_t new_shared_perm,
1860 GSList *ignore_children, Error **errp)
1862 BdrvChild *c;
1863 uint64_t cumulative_perms = new_used_perm;
1864 uint64_t cumulative_shared_perms = new_shared_perm;
1866 /* There is no reason why anyone couldn't tolerate write_unchanged */
1867 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1869 QLIST_FOREACH(c, &bs->parents, next_parent) {
1870 if (g_slist_find(ignore_children, c)) {
1871 continue;
1874 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1875 char *user = bdrv_child_user_desc(c);
1876 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1877 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1878 "allow '%s' on %s",
1879 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1880 g_free(user);
1881 g_free(perm_names);
1882 return -EPERM;
1885 if ((c->perm & new_shared_perm) != c->perm) {
1886 char *user = bdrv_child_user_desc(c);
1887 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1888 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1889 "'%s' on %s",
1890 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1891 g_free(user);
1892 g_free(perm_names);
1893 return -EPERM;
1896 cumulative_perms |= c->perm;
1897 cumulative_shared_perms &= c->shared_perm;
1900 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1901 ignore_children, errp);
1904 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1905 * bdrv_child_abort_perm_update(). */
1906 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1907 uint64_t perm, uint64_t shared,
1908 GSList *ignore_children, Error **errp)
1910 int ret;
1912 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1913 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1914 g_slist_free(ignore_children);
1916 return ret;
1919 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1921 uint64_t cumulative_perms, cumulative_shared_perms;
1923 c->perm = perm;
1924 c->shared_perm = shared;
1926 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1927 &cumulative_shared_perms);
1928 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1931 static void bdrv_child_abort_perm_update(BdrvChild *c)
1933 bdrv_abort_perm_update(c->bs);
1936 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1937 Error **errp)
1939 int ret;
1941 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1942 if (ret < 0) {
1943 bdrv_child_abort_perm_update(c);
1944 return ret;
1947 bdrv_child_set_perm(c, perm, shared);
1949 return 0;
1952 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1953 const BdrvChildRole *role,
1954 BlockReopenQueue *reopen_queue,
1955 uint64_t perm, uint64_t shared,
1956 uint64_t *nperm, uint64_t *nshared)
1958 if (c == NULL) {
1959 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1960 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1961 return;
1964 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1965 (c->perm & DEFAULT_PERM_UNCHANGED);
1966 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1967 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1970 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1971 const BdrvChildRole *role,
1972 BlockReopenQueue *reopen_queue,
1973 uint64_t perm, uint64_t shared,
1974 uint64_t *nperm, uint64_t *nshared)
1976 bool backing = (role == &child_backing);
1977 assert(role == &child_backing || role == &child_file);
1979 if (!backing) {
1980 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
1982 /* Apart from the modifications below, the same permissions are
1983 * forwarded and left alone as for filters */
1984 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
1985 &perm, &shared);
1987 /* Format drivers may touch metadata even if the guest doesn't write */
1988 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
1989 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1992 /* bs->file always needs to be consistent because of the metadata. We
1993 * can never allow other users to resize or write to it. */
1994 if (!(flags & BDRV_O_NO_IO)) {
1995 perm |= BLK_PERM_CONSISTENT_READ;
1997 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1998 } else {
1999 /* We want consistent read from backing files if the parent needs it.
2000 * No other operations are performed on backing files. */
2001 perm &= BLK_PERM_CONSISTENT_READ;
2003 /* If the parent can deal with changing data, we're okay with a
2004 * writable and resizable backing file. */
2005 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2006 if (shared & BLK_PERM_WRITE) {
2007 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2008 } else {
2009 shared = 0;
2012 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2013 BLK_PERM_WRITE_UNCHANGED;
2016 if (bs->open_flags & BDRV_O_INACTIVE) {
2017 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2020 *nperm = perm;
2021 *nshared = shared;
2024 static void bdrv_replace_child_noperm(BdrvChild *child,
2025 BlockDriverState *new_bs)
2027 BlockDriverState *old_bs = child->bs;
2028 int i;
2030 if (old_bs && new_bs) {
2031 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2033 if (old_bs) {
2034 /* Detach first so that the recursive drain sections coming from @child
2035 * are already gone and we only end the drain sections that came from
2036 * elsewhere. */
2037 if (child->role->detach) {
2038 child->role->detach(child);
2040 if (old_bs->quiesce_counter && child->role->drained_end) {
2041 int num = old_bs->quiesce_counter;
2042 if (child->role->parent_is_bds) {
2043 num -= bdrv_drain_all_count;
2045 assert(num >= 0);
2046 for (i = 0; i < num; i++) {
2047 child->role->drained_end(child);
2050 QLIST_REMOVE(child, next_parent);
2053 child->bs = new_bs;
2055 if (new_bs) {
2056 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2057 if (new_bs->quiesce_counter && child->role->drained_begin) {
2058 int num = new_bs->quiesce_counter;
2059 if (child->role->parent_is_bds) {
2060 num -= bdrv_drain_all_count;
2062 assert(num >= 0);
2063 for (i = 0; i < num; i++) {
2064 bdrv_parent_drained_begin_single(child, true);
2068 /* Attach only after starting new drained sections, so that recursive
2069 * drain sections coming from @child don't get an extra .drained_begin
2070 * callback. */
2071 if (child->role->attach) {
2072 child->role->attach(child);
2078 * Updates @child to change its reference to point to @new_bs, including
2079 * checking and applying the necessary permisson updates both to the old node
2080 * and to @new_bs.
2082 * NULL is passed as @new_bs for removing the reference before freeing @child.
2084 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2085 * function uses bdrv_set_perm() to update the permissions according to the new
2086 * reference that @new_bs gets.
2088 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2090 BlockDriverState *old_bs = child->bs;
2091 uint64_t perm, shared_perm;
2093 bdrv_replace_child_noperm(child, new_bs);
2095 if (old_bs) {
2096 /* Update permissions for old node. This is guaranteed to succeed
2097 * because we're just taking a parent away, so we're loosening
2098 * restrictions. */
2099 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2100 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2101 bdrv_set_perm(old_bs, perm, shared_perm);
2104 if (new_bs) {
2105 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2106 bdrv_set_perm(new_bs, perm, shared_perm);
2110 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2111 const char *child_name,
2112 const BdrvChildRole *child_role,
2113 uint64_t perm, uint64_t shared_perm,
2114 void *opaque, Error **errp)
2116 BdrvChild *child;
2117 int ret;
2119 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2120 if (ret < 0) {
2121 bdrv_abort_perm_update(child_bs);
2122 return NULL;
2125 child = g_new(BdrvChild, 1);
2126 *child = (BdrvChild) {
2127 .bs = NULL,
2128 .name = g_strdup(child_name),
2129 .role = child_role,
2130 .perm = perm,
2131 .shared_perm = shared_perm,
2132 .opaque = opaque,
2135 /* This performs the matching bdrv_set_perm() for the above check. */
2136 bdrv_replace_child(child, child_bs);
2138 return child;
2141 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2142 BlockDriverState *child_bs,
2143 const char *child_name,
2144 const BdrvChildRole *child_role,
2145 Error **errp)
2147 BdrvChild *child;
2148 uint64_t perm, shared_perm;
2150 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2152 assert(parent_bs->drv);
2153 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2154 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2155 perm, shared_perm, &perm, &shared_perm);
2157 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2158 perm, shared_perm, parent_bs, errp);
2159 if (child == NULL) {
2160 return NULL;
2163 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2164 return child;
2167 static void bdrv_detach_child(BdrvChild *child)
2169 if (child->next.le_prev) {
2170 QLIST_REMOVE(child, next);
2171 child->next.le_prev = NULL;
2174 bdrv_replace_child(child, NULL);
2176 g_free(child->name);
2177 g_free(child);
2180 void bdrv_root_unref_child(BdrvChild *child)
2182 BlockDriverState *child_bs;
2184 child_bs = child->bs;
2185 bdrv_detach_child(child);
2186 bdrv_unref(child_bs);
2189 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2191 if (child == NULL) {
2192 return;
2195 if (child->bs->inherits_from == parent) {
2196 BdrvChild *c;
2198 /* Remove inherits_from only when the last reference between parent and
2199 * child->bs goes away. */
2200 QLIST_FOREACH(c, &parent->children, next) {
2201 if (c != child && c->bs == child->bs) {
2202 break;
2205 if (c == NULL) {
2206 child->bs->inherits_from = NULL;
2210 bdrv_root_unref_child(child);
2214 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2216 BdrvChild *c;
2217 QLIST_FOREACH(c, &bs->parents, next_parent) {
2218 if (c->role->change_media) {
2219 c->role->change_media(c, load);
2225 * Sets the backing file link of a BDS. A new reference is created; callers
2226 * which don't need their own reference any more must call bdrv_unref().
2228 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2229 Error **errp)
2231 if (backing_hd) {
2232 bdrv_ref(backing_hd);
2235 if (bs->backing) {
2236 bdrv_unref_child(bs, bs->backing);
2239 if (!backing_hd) {
2240 bs->backing = NULL;
2241 goto out;
2244 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2245 errp);
2246 if (!bs->backing) {
2247 bdrv_unref(backing_hd);
2250 bdrv_refresh_filename(bs);
2252 out:
2253 bdrv_refresh_limits(bs, NULL);
2257 * Opens the backing file for a BlockDriverState if not yet open
2259 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2260 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2261 * itself, all options starting with "${bdref_key}." are considered part of the
2262 * BlockdevRef.
2264 * TODO Can this be unified with bdrv_open_image()?
2266 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2267 const char *bdref_key, Error **errp)
2269 char *backing_filename = g_malloc0(PATH_MAX);
2270 char *bdref_key_dot;
2271 const char *reference = NULL;
2272 int ret = 0;
2273 BlockDriverState *backing_hd;
2274 QDict *options;
2275 QDict *tmp_parent_options = NULL;
2276 Error *local_err = NULL;
2278 if (bs->backing != NULL) {
2279 goto free_exit;
2282 /* NULL means an empty set of options */
2283 if (parent_options == NULL) {
2284 tmp_parent_options = qdict_new();
2285 parent_options = tmp_parent_options;
2288 bs->open_flags &= ~BDRV_O_NO_BACKING;
2290 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2291 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2292 g_free(bdref_key_dot);
2295 * Caution: while qdict_get_try_str() is fine, getting non-string
2296 * types would require more care. When @parent_options come from
2297 * -blockdev or blockdev_add, its members are typed according to
2298 * the QAPI schema, but when they come from -drive, they're all
2299 * QString.
2301 reference = qdict_get_try_str(parent_options, bdref_key);
2302 if (reference || qdict_haskey(options, "file.filename")) {
2303 backing_filename[0] = '\0';
2304 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2305 qobject_unref(options);
2306 goto free_exit;
2307 } else {
2308 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2309 &local_err);
2310 if (local_err) {
2311 ret = -EINVAL;
2312 error_propagate(errp, local_err);
2313 qobject_unref(options);
2314 goto free_exit;
2318 if (!bs->drv || !bs->drv->supports_backing) {
2319 ret = -EINVAL;
2320 error_setg(errp, "Driver doesn't support backing files");
2321 qobject_unref(options);
2322 goto free_exit;
2325 if (!reference &&
2326 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2327 qdict_put_str(options, "driver", bs->backing_format);
2330 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2331 reference, options, 0, bs, &child_backing,
2332 errp);
2333 if (!backing_hd) {
2334 bs->open_flags |= BDRV_O_NO_BACKING;
2335 error_prepend(errp, "Could not open backing file: ");
2336 ret = -EINVAL;
2337 goto free_exit;
2339 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2341 /* Hook up the backing file link; drop our reference, bs owns the
2342 * backing_hd reference now */
2343 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2344 bdrv_unref(backing_hd);
2345 if (local_err) {
2346 error_propagate(errp, local_err);
2347 ret = -EINVAL;
2348 goto free_exit;
2351 qdict_del(parent_options, bdref_key);
2353 free_exit:
2354 g_free(backing_filename);
2355 qobject_unref(tmp_parent_options);
2356 return ret;
2359 static BlockDriverState *
2360 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2361 BlockDriverState *parent, const BdrvChildRole *child_role,
2362 bool allow_none, Error **errp)
2364 BlockDriverState *bs = NULL;
2365 QDict *image_options;
2366 char *bdref_key_dot;
2367 const char *reference;
2369 assert(child_role != NULL);
2371 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2372 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2373 g_free(bdref_key_dot);
2376 * Caution: while qdict_get_try_str() is fine, getting non-string
2377 * types would require more care. When @options come from
2378 * -blockdev or blockdev_add, its members are typed according to
2379 * the QAPI schema, but when they come from -drive, they're all
2380 * QString.
2382 reference = qdict_get_try_str(options, bdref_key);
2383 if (!filename && !reference && !qdict_size(image_options)) {
2384 if (!allow_none) {
2385 error_setg(errp, "A block device must be specified for \"%s\"",
2386 bdref_key);
2388 qobject_unref(image_options);
2389 goto done;
2392 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2393 parent, child_role, errp);
2394 if (!bs) {
2395 goto done;
2398 done:
2399 qdict_del(options, bdref_key);
2400 return bs;
2404 * Opens a disk image whose options are given as BlockdevRef in another block
2405 * device's options.
2407 * If allow_none is true, no image will be opened if filename is false and no
2408 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2410 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2411 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2412 * itself, all options starting with "${bdref_key}." are considered part of the
2413 * BlockdevRef.
2415 * The BlockdevRef will be removed from the options QDict.
2417 BdrvChild *bdrv_open_child(const char *filename,
2418 QDict *options, const char *bdref_key,
2419 BlockDriverState *parent,
2420 const BdrvChildRole *child_role,
2421 bool allow_none, Error **errp)
2423 BdrvChild *c;
2424 BlockDriverState *bs;
2426 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2427 allow_none, errp);
2428 if (bs == NULL) {
2429 return NULL;
2432 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2433 if (!c) {
2434 bdrv_unref(bs);
2435 return NULL;
2438 return c;
2441 /* TODO Future callers may need to specify parent/child_role in order for
2442 * option inheritance to work. Existing callers use it for the root node. */
2443 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2445 BlockDriverState *bs = NULL;
2446 Error *local_err = NULL;
2447 QObject *obj = NULL;
2448 QDict *qdict = NULL;
2449 const char *reference = NULL;
2450 Visitor *v = NULL;
2452 if (ref->type == QTYPE_QSTRING) {
2453 reference = ref->u.reference;
2454 } else {
2455 BlockdevOptions *options = &ref->u.definition;
2456 assert(ref->type == QTYPE_QDICT);
2458 v = qobject_output_visitor_new(&obj);
2459 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2460 if (local_err) {
2461 error_propagate(errp, local_err);
2462 goto fail;
2464 visit_complete(v, &obj);
2466 qdict = qobject_to(QDict, obj);
2467 qdict_flatten(qdict);
2469 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2470 * compatibility with other callers) rather than what we want as the
2471 * real defaults. Apply the defaults here instead. */
2472 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2473 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2474 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2477 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2478 obj = NULL;
2480 fail:
2481 qobject_unref(obj);
2482 visit_free(v);
2483 return bs;
2486 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2487 int flags,
2488 QDict *snapshot_options,
2489 Error **errp)
2491 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2492 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2493 int64_t total_size;
2494 QemuOpts *opts = NULL;
2495 BlockDriverState *bs_snapshot = NULL;
2496 Error *local_err = NULL;
2497 int ret;
2499 /* if snapshot, we create a temporary backing file and open it
2500 instead of opening 'filename' directly */
2502 /* Get the required size from the image */
2503 total_size = bdrv_getlength(bs);
2504 if (total_size < 0) {
2505 error_setg_errno(errp, -total_size, "Could not get image size");
2506 goto out;
2509 /* Create the temporary image */
2510 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2511 if (ret < 0) {
2512 error_setg_errno(errp, -ret, "Could not get temporary filename");
2513 goto out;
2516 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2517 &error_abort);
2518 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2519 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2520 qemu_opts_del(opts);
2521 if (ret < 0) {
2522 error_prepend(errp, "Could not create temporary overlay '%s': ",
2523 tmp_filename);
2524 goto out;
2527 /* Prepare options QDict for the temporary file */
2528 qdict_put_str(snapshot_options, "file.driver", "file");
2529 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2530 qdict_put_str(snapshot_options, "driver", "qcow2");
2532 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2533 snapshot_options = NULL;
2534 if (!bs_snapshot) {
2535 goto out;
2538 /* bdrv_append() consumes a strong reference to bs_snapshot
2539 * (i.e. it will call bdrv_unref() on it) even on error, so in
2540 * order to be able to return one, we have to increase
2541 * bs_snapshot's refcount here */
2542 bdrv_ref(bs_snapshot);
2543 bdrv_append(bs_snapshot, bs, &local_err);
2544 if (local_err) {
2545 error_propagate(errp, local_err);
2546 bs_snapshot = NULL;
2547 goto out;
2550 out:
2551 qobject_unref(snapshot_options);
2552 g_free(tmp_filename);
2553 return bs_snapshot;
2557 * Opens a disk image (raw, qcow2, vmdk, ...)
2559 * options is a QDict of options to pass to the block drivers, or NULL for an
2560 * empty set of options. The reference to the QDict belongs to the block layer
2561 * after the call (even on failure), so if the caller intends to reuse the
2562 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2564 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2565 * If it is not NULL, the referenced BDS will be reused.
2567 * The reference parameter may be used to specify an existing block device which
2568 * should be opened. If specified, neither options nor a filename may be given,
2569 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2571 static BlockDriverState *bdrv_open_inherit(const char *filename,
2572 const char *reference,
2573 QDict *options, int flags,
2574 BlockDriverState *parent,
2575 const BdrvChildRole *child_role,
2576 Error **errp)
2578 int ret;
2579 BlockBackend *file = NULL;
2580 BlockDriverState *bs;
2581 BlockDriver *drv = NULL;
2582 BdrvChild *child;
2583 const char *drvname;
2584 const char *backing;
2585 Error *local_err = NULL;
2586 QDict *snapshot_options = NULL;
2587 int snapshot_flags = 0;
2589 assert(!child_role || !flags);
2590 assert(!child_role == !parent);
2592 if (reference) {
2593 bool options_non_empty = options ? qdict_size(options) : false;
2594 qobject_unref(options);
2596 if (filename || options_non_empty) {
2597 error_setg(errp, "Cannot reference an existing block device with "
2598 "additional options or a new filename");
2599 return NULL;
2602 bs = bdrv_lookup_bs(reference, reference, errp);
2603 if (!bs) {
2604 return NULL;
2607 bdrv_ref(bs);
2608 return bs;
2611 bs = bdrv_new();
2613 /* NULL means an empty set of options */
2614 if (options == NULL) {
2615 options = qdict_new();
2618 /* json: syntax counts as explicit options, as if in the QDict */
2619 parse_json_protocol(options, &filename, &local_err);
2620 if (local_err) {
2621 goto fail;
2624 bs->explicit_options = qdict_clone_shallow(options);
2626 if (child_role) {
2627 bs->inherits_from = parent;
2628 child_role->inherit_options(&flags, options,
2629 parent->open_flags, parent->options);
2632 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2633 if (local_err) {
2634 goto fail;
2638 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2639 * Caution: getting a boolean member of @options requires care.
2640 * When @options come from -blockdev or blockdev_add, members are
2641 * typed according to the QAPI schema, but when they come from
2642 * -drive, they're all QString.
2644 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2645 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2646 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2647 } else {
2648 flags &= ~BDRV_O_RDWR;
2651 if (flags & BDRV_O_SNAPSHOT) {
2652 snapshot_options = qdict_new();
2653 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2654 flags, options);
2655 /* Let bdrv_backing_options() override "read-only" */
2656 qdict_del(options, BDRV_OPT_READ_ONLY);
2657 bdrv_backing_options(&flags, options, flags, options);
2660 bs->open_flags = flags;
2661 bs->options = options;
2662 options = qdict_clone_shallow(options);
2664 /* Find the right image format driver */
2665 /* See cautionary note on accessing @options above */
2666 drvname = qdict_get_try_str(options, "driver");
2667 if (drvname) {
2668 drv = bdrv_find_format(drvname);
2669 if (!drv) {
2670 error_setg(errp, "Unknown driver: '%s'", drvname);
2671 goto fail;
2675 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2677 /* See cautionary note on accessing @options above */
2678 backing = qdict_get_try_str(options, "backing");
2679 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2680 (backing && *backing == '\0'))
2682 if (backing) {
2683 warn_report("Use of \"backing\": \"\" is deprecated; "
2684 "use \"backing\": null instead");
2686 flags |= BDRV_O_NO_BACKING;
2687 qdict_del(options, "backing");
2690 /* Open image file without format layer. This BlockBackend is only used for
2691 * probing, the block drivers will do their own bdrv_open_child() for the
2692 * same BDS, which is why we put the node name back into options. */
2693 if ((flags & BDRV_O_PROTOCOL) == 0) {
2694 BlockDriverState *file_bs;
2696 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2697 &child_file, true, &local_err);
2698 if (local_err) {
2699 goto fail;
2701 if (file_bs != NULL) {
2702 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2703 * looking at the header to guess the image format. This works even
2704 * in cases where a guest would not see a consistent state. */
2705 file = blk_new(0, BLK_PERM_ALL);
2706 blk_insert_bs(file, file_bs, &local_err);
2707 bdrv_unref(file_bs);
2708 if (local_err) {
2709 goto fail;
2712 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2716 /* Image format probing */
2717 bs->probed = !drv;
2718 if (!drv && file) {
2719 ret = find_image_format(file, filename, &drv, &local_err);
2720 if (ret < 0) {
2721 goto fail;
2724 * This option update would logically belong in bdrv_fill_options(),
2725 * but we first need to open bs->file for the probing to work, while
2726 * opening bs->file already requires the (mostly) final set of options
2727 * so that cache mode etc. can be inherited.
2729 * Adding the driver later is somewhat ugly, but it's not an option
2730 * that would ever be inherited, so it's correct. We just need to make
2731 * sure to update both bs->options (which has the full effective
2732 * options for bs) and options (which has file.* already removed).
2734 qdict_put_str(bs->options, "driver", drv->format_name);
2735 qdict_put_str(options, "driver", drv->format_name);
2736 } else if (!drv) {
2737 error_setg(errp, "Must specify either driver or file");
2738 goto fail;
2741 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2742 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2743 /* file must be NULL if a protocol BDS is about to be created
2744 * (the inverse results in an error message from bdrv_open_common()) */
2745 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2747 /* Open the image */
2748 ret = bdrv_open_common(bs, file, options, &local_err);
2749 if (ret < 0) {
2750 goto fail;
2753 if (file) {
2754 blk_unref(file);
2755 file = NULL;
2758 /* If there is a backing file, use it */
2759 if ((flags & BDRV_O_NO_BACKING) == 0) {
2760 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2761 if (ret < 0) {
2762 goto close_and_fail;
2766 /* Remove all children options from bs->options and bs->explicit_options */
2767 QLIST_FOREACH(child, &bs->children, next) {
2768 char *child_key_dot;
2769 child_key_dot = g_strdup_printf("%s.", child->name);
2770 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
2771 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
2772 g_free(child_key_dot);
2775 bdrv_refresh_filename(bs);
2777 /* Check if any unknown options were used */
2778 if (qdict_size(options) != 0) {
2779 const QDictEntry *entry = qdict_first(options);
2780 if (flags & BDRV_O_PROTOCOL) {
2781 error_setg(errp, "Block protocol '%s' doesn't support the option "
2782 "'%s'", drv->format_name, entry->key);
2783 } else {
2784 error_setg(errp,
2785 "Block format '%s' does not support the option '%s'",
2786 drv->format_name, entry->key);
2789 goto close_and_fail;
2792 bdrv_parent_cb_change_media(bs, true);
2794 qobject_unref(options);
2795 options = NULL;
2797 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2798 * temporary snapshot afterwards. */
2799 if (snapshot_flags) {
2800 BlockDriverState *snapshot_bs;
2801 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2802 snapshot_options, &local_err);
2803 snapshot_options = NULL;
2804 if (local_err) {
2805 goto close_and_fail;
2807 /* We are not going to return bs but the overlay on top of it
2808 * (snapshot_bs); thus, we have to drop the strong reference to bs
2809 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2810 * though, because the overlay still has a reference to it. */
2811 bdrv_unref(bs);
2812 bs = snapshot_bs;
2815 return bs;
2817 fail:
2818 blk_unref(file);
2819 qobject_unref(snapshot_options);
2820 qobject_unref(bs->explicit_options);
2821 qobject_unref(bs->options);
2822 qobject_unref(options);
2823 bs->options = NULL;
2824 bs->explicit_options = NULL;
2825 bdrv_unref(bs);
2826 error_propagate(errp, local_err);
2827 return NULL;
2829 close_and_fail:
2830 bdrv_unref(bs);
2831 qobject_unref(snapshot_options);
2832 qobject_unref(options);
2833 error_propagate(errp, local_err);
2834 return NULL;
2837 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2838 QDict *options, int flags, Error **errp)
2840 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2841 NULL, errp);
2845 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2846 * reopen of multiple devices.
2848 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2849 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2850 * be created and initialized. This newly created BlockReopenQueue should be
2851 * passed back in for subsequent calls that are intended to be of the same
2852 * atomic 'set'.
2854 * bs is the BlockDriverState to add to the reopen queue.
2856 * options contains the changed options for the associated bs
2857 * (the BlockReopenQueue takes ownership)
2859 * flags contains the open flags for the associated bs
2861 * returns a pointer to bs_queue, which is either the newly allocated
2862 * bs_queue, or the existing bs_queue being used.
2864 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2866 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2867 BlockDriverState *bs,
2868 QDict *options,
2869 int flags,
2870 const BdrvChildRole *role,
2871 QDict *parent_options,
2872 int parent_flags)
2874 assert(bs != NULL);
2876 BlockReopenQueueEntry *bs_entry;
2877 BdrvChild *child;
2878 QDict *old_options, *explicit_options;
2880 /* Make sure that the caller remembered to use a drained section. This is
2881 * important to avoid graph changes between the recursive queuing here and
2882 * bdrv_reopen_multiple(). */
2883 assert(bs->quiesce_counter > 0);
2885 if (bs_queue == NULL) {
2886 bs_queue = g_new0(BlockReopenQueue, 1);
2887 QSIMPLEQ_INIT(bs_queue);
2890 if (!options) {
2891 options = qdict_new();
2894 /* Check if this BlockDriverState is already in the queue */
2895 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2896 if (bs == bs_entry->state.bs) {
2897 break;
2902 * Precedence of options:
2903 * 1. Explicitly passed in options (highest)
2904 * 2. Set in flags (only for top level)
2905 * 3. Retained from explicitly set options of bs
2906 * 4. Inherited from parent node
2907 * 5. Retained from effective options of bs
2910 if (!parent_options) {
2912 * Any setting represented by flags is always updated. If the
2913 * corresponding QDict option is set, it takes precedence. Otherwise
2914 * the flag is translated into a QDict option. The old setting of bs is
2915 * not considered.
2917 update_options_from_flags(options, flags);
2920 /* Old explicitly set values (don't overwrite by inherited value) */
2921 if (bs_entry) {
2922 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2923 } else {
2924 old_options = qdict_clone_shallow(bs->explicit_options);
2926 bdrv_join_options(bs, options, old_options);
2927 qobject_unref(old_options);
2929 explicit_options = qdict_clone_shallow(options);
2931 /* Inherit from parent node */
2932 if (parent_options) {
2933 QemuOpts *opts;
2934 QDict *options_copy;
2935 assert(!flags);
2936 role->inherit_options(&flags, options, parent_flags, parent_options);
2937 options_copy = qdict_clone_shallow(options);
2938 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2939 qemu_opts_absorb_qdict(opts, options_copy, NULL);
2940 update_flags_from_options(&flags, opts);
2941 qemu_opts_del(opts);
2942 qobject_unref(options_copy);
2945 /* Old values are used for options that aren't set yet */
2946 old_options = qdict_clone_shallow(bs->options);
2947 bdrv_join_options(bs, options, old_options);
2948 qobject_unref(old_options);
2950 /* bdrv_open_inherit() sets and clears some additional flags internally */
2951 flags &= ~BDRV_O_PROTOCOL;
2952 if (flags & BDRV_O_RDWR) {
2953 flags |= BDRV_O_ALLOW_RDWR;
2956 if (!bs_entry) {
2957 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2958 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2959 } else {
2960 qobject_unref(bs_entry->state.options);
2961 qobject_unref(bs_entry->state.explicit_options);
2964 bs_entry->state.bs = bs;
2965 bs_entry->state.options = options;
2966 bs_entry->state.explicit_options = explicit_options;
2967 bs_entry->state.flags = flags;
2969 /* This needs to be overwritten in bdrv_reopen_prepare() */
2970 bs_entry->state.perm = UINT64_MAX;
2971 bs_entry->state.shared_perm = 0;
2973 QLIST_FOREACH(child, &bs->children, next) {
2974 QDict *new_child_options;
2975 char *child_key_dot;
2977 /* reopen can only change the options of block devices that were
2978 * implicitly created and inherited options. For other (referenced)
2979 * block devices, a syntax like "backing.foo" results in an error. */
2980 if (child->bs->inherits_from != bs) {
2981 continue;
2984 child_key_dot = g_strdup_printf("%s.", child->name);
2985 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
2986 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2987 g_free(child_key_dot);
2989 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2990 child->role, options, flags);
2993 return bs_queue;
2996 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2997 BlockDriverState *bs,
2998 QDict *options, int flags)
3000 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
3001 NULL, NULL, 0);
3005 * Reopen multiple BlockDriverStates atomically & transactionally.
3007 * The queue passed in (bs_queue) must have been built up previous
3008 * via bdrv_reopen_queue().
3010 * Reopens all BDS specified in the queue, with the appropriate
3011 * flags. All devices are prepared for reopen, and failure of any
3012 * device will cause all device changes to be abandoned, and intermediate
3013 * data cleaned up.
3015 * If all devices prepare successfully, then the changes are committed
3016 * to all devices.
3018 * All affected nodes must be drained between bdrv_reopen_queue() and
3019 * bdrv_reopen_multiple().
3021 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
3023 int ret = -1;
3024 BlockReopenQueueEntry *bs_entry, *next;
3025 Error *local_err = NULL;
3027 assert(bs_queue != NULL);
3029 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3030 assert(bs_entry->state.bs->quiesce_counter > 0);
3031 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
3032 error_propagate(errp, local_err);
3033 goto cleanup;
3035 bs_entry->prepared = true;
3038 /* If we reach this point, we have success and just need to apply the
3039 * changes
3041 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3042 bdrv_reopen_commit(&bs_entry->state);
3045 ret = 0;
3047 cleanup:
3048 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3049 if (ret) {
3050 if (bs_entry->prepared) {
3051 bdrv_reopen_abort(&bs_entry->state);
3053 qobject_unref(bs_entry->state.explicit_options);
3054 qobject_unref(bs_entry->state.options);
3056 g_free(bs_entry);
3058 g_free(bs_queue);
3060 return ret;
3064 /* Reopen a single BlockDriverState with the specified flags. */
3065 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
3067 int ret = -1;
3068 Error *local_err = NULL;
3069 BlockReopenQueue *queue;
3071 bdrv_subtree_drained_begin(bs);
3073 queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
3074 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
3075 if (local_err != NULL) {
3076 error_propagate(errp, local_err);
3079 bdrv_subtree_drained_end(bs);
3081 return ret;
3084 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3085 BdrvChild *c)
3087 BlockReopenQueueEntry *entry;
3089 QSIMPLEQ_FOREACH(entry, q, entry) {
3090 BlockDriverState *bs = entry->state.bs;
3091 BdrvChild *child;
3093 QLIST_FOREACH(child, &bs->children, next) {
3094 if (child == c) {
3095 return entry;
3100 return NULL;
3103 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3104 uint64_t *perm, uint64_t *shared)
3106 BdrvChild *c;
3107 BlockReopenQueueEntry *parent;
3108 uint64_t cumulative_perms = 0;
3109 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3111 QLIST_FOREACH(c, &bs->parents, next_parent) {
3112 parent = find_parent_in_reopen_queue(q, c);
3113 if (!parent) {
3114 cumulative_perms |= c->perm;
3115 cumulative_shared_perms &= c->shared_perm;
3116 } else {
3117 uint64_t nperm, nshared;
3119 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3120 parent->state.perm, parent->state.shared_perm,
3121 &nperm, &nshared);
3123 cumulative_perms |= nperm;
3124 cumulative_shared_perms &= nshared;
3127 *perm = cumulative_perms;
3128 *shared = cumulative_shared_perms;
3132 * Prepares a BlockDriverState for reopen. All changes are staged in the
3133 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3134 * the block driver layer .bdrv_reopen_prepare()
3136 * bs is the BlockDriverState to reopen
3137 * flags are the new open flags
3138 * queue is the reopen queue
3140 * Returns 0 on success, non-zero on error. On error errp will be set
3141 * as well.
3143 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3144 * It is the responsibility of the caller to then call the abort() or
3145 * commit() for any other BDS that have been left in a prepare() state
3148 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3149 Error **errp)
3151 int ret = -1;
3152 Error *local_err = NULL;
3153 BlockDriver *drv;
3154 QemuOpts *opts;
3155 QDict *orig_reopen_opts;
3156 const char *value;
3157 bool read_only;
3159 assert(reopen_state != NULL);
3160 assert(reopen_state->bs->drv != NULL);
3161 drv = reopen_state->bs->drv;
3163 /* This function and each driver's bdrv_reopen_prepare() remove
3164 * entries from reopen_state->options as they are processed, so
3165 * we need to make a copy of the original QDict. */
3166 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3168 /* Process generic block layer options */
3169 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3170 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3171 if (local_err) {
3172 error_propagate(errp, local_err);
3173 ret = -EINVAL;
3174 goto error;
3177 update_flags_from_options(&reopen_state->flags, opts);
3179 /* node-name and driver must be unchanged. Put them back into the QDict, so
3180 * that they are checked at the end of this function. */
3181 value = qemu_opt_get(opts, "node-name");
3182 if (value) {
3183 qdict_put_str(reopen_state->options, "node-name", value);
3186 value = qemu_opt_get(opts, "driver");
3187 if (value) {
3188 qdict_put_str(reopen_state->options, "driver", value);
3191 /* If we are to stay read-only, do not allow permission change
3192 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3193 * not set, or if the BDS still has copy_on_read enabled */
3194 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3195 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3196 if (local_err) {
3197 error_propagate(errp, local_err);
3198 goto error;
3201 /* Calculate required permissions after reopening */
3202 bdrv_reopen_perm(queue, reopen_state->bs,
3203 &reopen_state->perm, &reopen_state->shared_perm);
3205 ret = bdrv_flush(reopen_state->bs);
3206 if (ret) {
3207 error_setg_errno(errp, -ret, "Error flushing drive");
3208 goto error;
3211 if (drv->bdrv_reopen_prepare) {
3212 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3213 if (ret) {
3214 if (local_err != NULL) {
3215 error_propagate(errp, local_err);
3216 } else {
3217 error_setg(errp, "failed while preparing to reopen image '%s'",
3218 reopen_state->bs->filename);
3220 goto error;
3222 } else {
3223 /* It is currently mandatory to have a bdrv_reopen_prepare()
3224 * handler for each supported drv. */
3225 error_setg(errp, "Block format '%s' used by node '%s' "
3226 "does not support reopening files", drv->format_name,
3227 bdrv_get_device_or_node_name(reopen_state->bs));
3228 ret = -1;
3229 goto error;
3232 /* Options that are not handled are only okay if they are unchanged
3233 * compared to the old state. It is expected that some options are only
3234 * used for the initial open, but not reopen (e.g. filename) */
3235 if (qdict_size(reopen_state->options)) {
3236 const QDictEntry *entry = qdict_first(reopen_state->options);
3238 do {
3239 QObject *new = entry->value;
3240 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3243 * TODO: When using -drive to specify blockdev options, all values
3244 * will be strings; however, when using -blockdev, blockdev-add or
3245 * filenames using the json:{} pseudo-protocol, they will be
3246 * correctly typed.
3247 * In contrast, reopening options are (currently) always strings
3248 * (because you can only specify them through qemu-io; all other
3249 * callers do not specify any options).
3250 * Therefore, when using anything other than -drive to create a BDS,
3251 * this cannot detect non-string options as unchanged, because
3252 * qobject_is_equal() always returns false for objects of different
3253 * type. In the future, this should be remedied by correctly typing
3254 * all options. For now, this is not too big of an issue because
3255 * the user can simply omit options which cannot be changed anyway,
3256 * so they will stay unchanged.
3258 if (!qobject_is_equal(new, old)) {
3259 error_setg(errp, "Cannot change the option '%s'", entry->key);
3260 ret = -EINVAL;
3261 goto error;
3263 } while ((entry = qdict_next(reopen_state->options, entry)));
3266 ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3267 reopen_state->shared_perm, NULL, errp);
3268 if (ret < 0) {
3269 goto error;
3272 ret = 0;
3274 /* Restore the original reopen_state->options QDict */
3275 qobject_unref(reopen_state->options);
3276 reopen_state->options = qobject_ref(orig_reopen_opts);
3278 error:
3279 qemu_opts_del(opts);
3280 qobject_unref(orig_reopen_opts);
3281 return ret;
3285 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3286 * makes them final by swapping the staging BlockDriverState contents into
3287 * the active BlockDriverState contents.
3289 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3291 BlockDriver *drv;
3292 BlockDriverState *bs;
3293 bool old_can_write, new_can_write;
3295 assert(reopen_state != NULL);
3296 bs = reopen_state->bs;
3297 drv = bs->drv;
3298 assert(drv != NULL);
3300 old_can_write =
3301 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3303 /* If there are any driver level actions to take */
3304 if (drv->bdrv_reopen_commit) {
3305 drv->bdrv_reopen_commit(reopen_state);
3308 /* set BDS specific flags now */
3309 qobject_unref(bs->explicit_options);
3310 qobject_unref(bs->options);
3312 bs->explicit_options = reopen_state->explicit_options;
3313 bs->options = reopen_state->options;
3314 bs->open_flags = reopen_state->flags;
3315 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3317 bdrv_refresh_limits(bs, NULL);
3319 bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3320 reopen_state->shared_perm);
3322 new_can_write =
3323 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3324 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3325 Error *local_err = NULL;
3326 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3327 /* This is not fatal, bitmaps just left read-only, so all following
3328 * writes will fail. User can remove read-only bitmaps to unblock
3329 * writes.
3331 error_reportf_err(local_err,
3332 "%s: Failed to make dirty bitmaps writable: ",
3333 bdrv_get_node_name(bs));
3339 * Abort the reopen, and delete and free the staged changes in
3340 * reopen_state
3342 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3344 BlockDriver *drv;
3346 assert(reopen_state != NULL);
3347 drv = reopen_state->bs->drv;
3348 assert(drv != NULL);
3350 if (drv->bdrv_reopen_abort) {
3351 drv->bdrv_reopen_abort(reopen_state);
3354 bdrv_abort_perm_update(reopen_state->bs);
3358 static void bdrv_close(BlockDriverState *bs)
3360 BdrvAioNotifier *ban, *ban_next;
3361 BdrvChild *child, *next;
3363 assert(!bs->job);
3364 assert(!bs->refcnt);
3366 bdrv_drained_begin(bs); /* complete I/O */
3367 bdrv_flush(bs);
3368 bdrv_drain(bs); /* in case flush left pending I/O */
3370 if (bs->drv) {
3371 if (bs->drv->bdrv_close) {
3372 bs->drv->bdrv_close(bs);
3374 bs->drv = NULL;
3377 bdrv_set_backing_hd(bs, NULL, &error_abort);
3379 if (bs->file != NULL) {
3380 bdrv_unref_child(bs, bs->file);
3381 bs->file = NULL;
3384 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3385 /* TODO Remove bdrv_unref() from drivers' close function and use
3386 * bdrv_unref_child() here */
3387 if (child->bs->inherits_from == bs) {
3388 child->bs->inherits_from = NULL;
3390 bdrv_detach_child(child);
3393 g_free(bs->opaque);
3394 bs->opaque = NULL;
3395 atomic_set(&bs->copy_on_read, 0);
3396 bs->backing_file[0] = '\0';
3397 bs->backing_format[0] = '\0';
3398 bs->total_sectors = 0;
3399 bs->encrypted = false;
3400 bs->sg = false;
3401 qobject_unref(bs->options);
3402 qobject_unref(bs->explicit_options);
3403 bs->options = NULL;
3404 bs->explicit_options = NULL;
3405 qobject_unref(bs->full_open_options);
3406 bs->full_open_options = NULL;
3408 bdrv_release_named_dirty_bitmaps(bs);
3409 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3411 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3412 g_free(ban);
3414 QLIST_INIT(&bs->aio_notifiers);
3415 bdrv_drained_end(bs);
3418 void bdrv_close_all(void)
3420 assert(job_next(NULL) == NULL);
3421 nbd_export_close_all();
3423 /* Drop references from requests still in flight, such as canceled block
3424 * jobs whose AIO context has not been polled yet */
3425 bdrv_drain_all();
3427 blk_remove_all_bs();
3428 blockdev_close_all_bdrv_states();
3430 assert(QTAILQ_EMPTY(&all_bdrv_states));
3433 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3435 BdrvChild *to_c;
3437 if (c->role->stay_at_node) {
3438 return false;
3441 /* If the child @c belongs to the BDS @to, replacing the current
3442 * c->bs by @to would mean to create a loop.
3444 * Such a case occurs when appending a BDS to a backing chain.
3445 * For instance, imagine the following chain:
3447 * guest device -> node A -> further backing chain...
3449 * Now we create a new BDS B which we want to put on top of this
3450 * chain, so we first attach A as its backing node:
3452 * node B
3455 * guest device -> node A -> further backing chain...
3457 * Finally we want to replace A by B. When doing that, we want to
3458 * replace all pointers to A by pointers to B -- except for the
3459 * pointer from B because (1) that would create a loop, and (2)
3460 * that pointer should simply stay intact:
3462 * guest device -> node B
3465 * node A -> further backing chain...
3467 * In general, when replacing a node A (c->bs) by a node B (@to),
3468 * if A is a child of B, that means we cannot replace A by B there
3469 * because that would create a loop. Silently detaching A from B
3470 * is also not really an option. So overall just leaving A in
3471 * place there is the most sensible choice. */
3472 QLIST_FOREACH(to_c, &to->children, next) {
3473 if (to_c == c) {
3474 return false;
3478 return true;
3481 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3482 Error **errp)
3484 BdrvChild *c, *next;
3485 GSList *list = NULL, *p;
3486 uint64_t old_perm, old_shared;
3487 uint64_t perm = 0, shared = BLK_PERM_ALL;
3488 int ret;
3490 assert(!atomic_read(&from->in_flight));
3491 assert(!atomic_read(&to->in_flight));
3493 /* Make sure that @from doesn't go away until we have successfully attached
3494 * all of its parents to @to. */
3495 bdrv_ref(from);
3497 /* Put all parents into @list and calculate their cumulative permissions */
3498 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3499 assert(c->bs == from);
3500 if (!should_update_child(c, to)) {
3501 continue;
3503 list = g_slist_prepend(list, c);
3504 perm |= c->perm;
3505 shared &= c->shared_perm;
3508 /* Check whether the required permissions can be granted on @to, ignoring
3509 * all BdrvChild in @list so that they can't block themselves. */
3510 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3511 if (ret < 0) {
3512 bdrv_abort_perm_update(to);
3513 goto out;
3516 /* Now actually perform the change. We performed the permission check for
3517 * all elements of @list at once, so set the permissions all at once at the
3518 * very end. */
3519 for (p = list; p != NULL; p = p->next) {
3520 c = p->data;
3522 bdrv_ref(to);
3523 bdrv_replace_child_noperm(c, to);
3524 bdrv_unref(from);
3527 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3528 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3530 out:
3531 g_slist_free(list);
3532 bdrv_unref(from);
3536 * Add new bs contents at the top of an image chain while the chain is
3537 * live, while keeping required fields on the top layer.
3539 * This will modify the BlockDriverState fields, and swap contents
3540 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3542 * bs_new must not be attached to a BlockBackend.
3544 * This function does not create any image files.
3546 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3547 * that's what the callers commonly need. bs_new will be referenced by the old
3548 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3549 * reference of its own, it must call bdrv_ref().
3551 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3552 Error **errp)
3554 Error *local_err = NULL;
3556 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3557 if (local_err) {
3558 error_propagate(errp, local_err);
3559 goto out;
3562 bdrv_replace_node(bs_top, bs_new, &local_err);
3563 if (local_err) {
3564 error_propagate(errp, local_err);
3565 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3566 goto out;
3569 /* bs_new is now referenced by its new parents, we don't need the
3570 * additional reference any more. */
3571 out:
3572 bdrv_unref(bs_new);
3575 static void bdrv_delete(BlockDriverState *bs)
3577 assert(!bs->job);
3578 assert(bdrv_op_blocker_is_empty(bs));
3579 assert(!bs->refcnt);
3581 bdrv_close(bs);
3583 /* remove from list, if necessary */
3584 if (bs->node_name[0] != '\0') {
3585 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3587 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3589 g_free(bs);
3593 * Run consistency checks on an image
3595 * Returns 0 if the check could be completed (it doesn't mean that the image is
3596 * free of errors) or -errno when an internal error occurred. The results of the
3597 * check are stored in res.
3599 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3600 BdrvCheckResult *res, BdrvCheckMode fix)
3602 if (bs->drv == NULL) {
3603 return -ENOMEDIUM;
3605 if (bs->drv->bdrv_co_check == NULL) {
3606 return -ENOTSUP;
3609 memset(res, 0, sizeof(*res));
3610 return bs->drv->bdrv_co_check(bs, res, fix);
3613 typedef struct CheckCo {
3614 BlockDriverState *bs;
3615 BdrvCheckResult *res;
3616 BdrvCheckMode fix;
3617 int ret;
3618 } CheckCo;
3620 static void bdrv_check_co_entry(void *opaque)
3622 CheckCo *cco = opaque;
3623 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3626 int bdrv_check(BlockDriverState *bs,
3627 BdrvCheckResult *res, BdrvCheckMode fix)
3629 Coroutine *co;
3630 CheckCo cco = {
3631 .bs = bs,
3632 .res = res,
3633 .ret = -EINPROGRESS,
3634 .fix = fix,
3637 if (qemu_in_coroutine()) {
3638 /* Fast-path if already in coroutine context */
3639 bdrv_check_co_entry(&cco);
3640 } else {
3641 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3642 qemu_coroutine_enter(co);
3643 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3646 return cco.ret;
3650 * Return values:
3651 * 0 - success
3652 * -EINVAL - backing format specified, but no file
3653 * -ENOSPC - can't update the backing file because no space is left in the
3654 * image file header
3655 * -ENOTSUP - format driver doesn't support changing the backing file
3657 int bdrv_change_backing_file(BlockDriverState *bs,
3658 const char *backing_file, const char *backing_fmt)
3660 BlockDriver *drv = bs->drv;
3661 int ret;
3663 if (!drv) {
3664 return -ENOMEDIUM;
3667 /* Backing file format doesn't make sense without a backing file */
3668 if (backing_fmt && !backing_file) {
3669 return -EINVAL;
3672 if (drv->bdrv_change_backing_file != NULL) {
3673 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3674 } else {
3675 ret = -ENOTSUP;
3678 if (ret == 0) {
3679 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3680 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3682 return ret;
3686 * Finds the image layer in the chain that has 'bs' as its backing file.
3688 * active is the current topmost image.
3690 * Returns NULL if bs is not found in active's image chain,
3691 * or if active == bs.
3693 * Returns the bottommost base image if bs == NULL.
3695 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3696 BlockDriverState *bs)
3698 while (active && bs != backing_bs(active)) {
3699 active = backing_bs(active);
3702 return active;
3705 /* Given a BDS, searches for the base layer. */
3706 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3708 return bdrv_find_overlay(bs, NULL);
3712 * Drops images above 'base' up to and including 'top', and sets the image
3713 * above 'top' to have base as its backing file.
3715 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3716 * information in 'bs' can be properly updated.
3718 * E.g., this will convert the following chain:
3719 * bottom <- base <- intermediate <- top <- active
3721 * to
3723 * bottom <- base <- active
3725 * It is allowed for bottom==base, in which case it converts:
3727 * base <- intermediate <- top <- active
3729 * to
3731 * base <- active
3733 * If backing_file_str is non-NULL, it will be used when modifying top's
3734 * overlay image metadata.
3736 * Error conditions:
3737 * if active == top, that is considered an error
3740 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3741 const char *backing_file_str)
3743 BdrvChild *c, *next;
3744 Error *local_err = NULL;
3745 int ret = -EIO;
3747 bdrv_ref(top);
3749 if (!top->drv || !base->drv) {
3750 goto exit;
3753 /* Make sure that base is in the backing chain of top */
3754 if (!bdrv_chain_contains(top, base)) {
3755 goto exit;
3758 /* success - we can delete the intermediate states, and link top->base */
3759 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3760 * we've figured out how they should work. */
3761 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3763 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3764 /* Check whether we are allowed to switch c from top to base */
3765 GSList *ignore_children = g_slist_prepend(NULL, c);
3766 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3767 ignore_children, &local_err);
3768 g_slist_free(ignore_children);
3769 if (local_err) {
3770 ret = -EPERM;
3771 error_report_err(local_err);
3772 goto exit;
3775 /* If so, update the backing file path in the image file */
3776 if (c->role->update_filename) {
3777 ret = c->role->update_filename(c, base, backing_file_str,
3778 &local_err);
3779 if (ret < 0) {
3780 bdrv_abort_perm_update(base);
3781 error_report_err(local_err);
3782 goto exit;
3786 /* Do the actual switch in the in-memory graph.
3787 * Completes bdrv_check_update_perm() transaction internally. */
3788 bdrv_ref(base);
3789 bdrv_replace_child(c, base);
3790 bdrv_unref(top);
3793 ret = 0;
3794 exit:
3795 bdrv_unref(top);
3796 return ret;
3800 * Length of a allocated file in bytes. Sparse files are counted by actual
3801 * allocated space. Return < 0 if error or unknown.
3803 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3805 BlockDriver *drv = bs->drv;
3806 if (!drv) {
3807 return -ENOMEDIUM;
3809 if (drv->bdrv_get_allocated_file_size) {
3810 return drv->bdrv_get_allocated_file_size(bs);
3812 if (bs->file) {
3813 return bdrv_get_allocated_file_size(bs->file->bs);
3815 return -ENOTSUP;
3819 * bdrv_measure:
3820 * @drv: Format driver
3821 * @opts: Creation options for new image
3822 * @in_bs: Existing image containing data for new image (may be NULL)
3823 * @errp: Error object
3824 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3825 * or NULL on error
3827 * Calculate file size required to create a new image.
3829 * If @in_bs is given then space for allocated clusters and zero clusters
3830 * from that image are included in the calculation. If @opts contains a
3831 * backing file that is shared by @in_bs then backing clusters may be omitted
3832 * from the calculation.
3834 * If @in_bs is NULL then the calculation includes no allocated clusters
3835 * unless a preallocation option is given in @opts.
3837 * Note that @in_bs may use a different BlockDriver from @drv.
3839 * If an error occurs the @errp pointer is set.
3841 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3842 BlockDriverState *in_bs, Error **errp)
3844 if (!drv->bdrv_measure) {
3845 error_setg(errp, "Block driver '%s' does not support size measurement",
3846 drv->format_name);
3847 return NULL;
3850 return drv->bdrv_measure(opts, in_bs, errp);
3854 * Return number of sectors on success, -errno on error.
3856 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3858 BlockDriver *drv = bs->drv;
3860 if (!drv)
3861 return -ENOMEDIUM;
3863 if (drv->has_variable_length) {
3864 int ret = refresh_total_sectors(bs, bs->total_sectors);
3865 if (ret < 0) {
3866 return ret;
3869 return bs->total_sectors;
3873 * Return length in bytes on success, -errno on error.
3874 * The length is always a multiple of BDRV_SECTOR_SIZE.
3876 int64_t bdrv_getlength(BlockDriverState *bs)
3878 int64_t ret = bdrv_nb_sectors(bs);
3880 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3881 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3884 /* return 0 as number of sectors if no device present or error */
3885 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3887 int64_t nb_sectors = bdrv_nb_sectors(bs);
3889 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3892 bool bdrv_is_sg(BlockDriverState *bs)
3894 return bs->sg;
3897 bool bdrv_is_encrypted(BlockDriverState *bs)
3899 if (bs->backing && bs->backing->bs->encrypted) {
3900 return true;
3902 return bs->encrypted;
3905 const char *bdrv_get_format_name(BlockDriverState *bs)
3907 return bs->drv ? bs->drv->format_name : NULL;
3910 static int qsort_strcmp(const void *a, const void *b)
3912 return strcmp(*(char *const *)a, *(char *const *)b);
3915 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3916 void *opaque)
3918 BlockDriver *drv;
3919 int count = 0;
3920 int i;
3921 const char **formats = NULL;
3923 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3924 if (drv->format_name) {
3925 bool found = false;
3926 int i = count;
3927 while (formats && i && !found) {
3928 found = !strcmp(formats[--i], drv->format_name);
3931 if (!found) {
3932 formats = g_renew(const char *, formats, count + 1);
3933 formats[count++] = drv->format_name;
3938 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3939 const char *format_name = block_driver_modules[i].format_name;
3941 if (format_name) {
3942 bool found = false;
3943 int j = count;
3945 while (formats && j && !found) {
3946 found = !strcmp(formats[--j], format_name);
3949 if (!found) {
3950 formats = g_renew(const char *, formats, count + 1);
3951 formats[count++] = format_name;
3956 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3958 for (i = 0; i < count; i++) {
3959 it(opaque, formats[i]);
3962 g_free(formats);
3965 /* This function is to find a node in the bs graph */
3966 BlockDriverState *bdrv_find_node(const char *node_name)
3968 BlockDriverState *bs;
3970 assert(node_name);
3972 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3973 if (!strcmp(node_name, bs->node_name)) {
3974 return bs;
3977 return NULL;
3980 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3981 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3983 BlockDeviceInfoList *list, *entry;
3984 BlockDriverState *bs;
3986 list = NULL;
3987 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3988 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3989 if (!info) {
3990 qapi_free_BlockDeviceInfoList(list);
3991 return NULL;
3993 entry = g_malloc0(sizeof(*entry));
3994 entry->value = info;
3995 entry->next = list;
3996 list = entry;
3999 return list;
4002 BlockDriverState *bdrv_lookup_bs(const char *device,
4003 const char *node_name,
4004 Error **errp)
4006 BlockBackend *blk;
4007 BlockDriverState *bs;
4009 if (device) {
4010 blk = blk_by_name(device);
4012 if (blk) {
4013 bs = blk_bs(blk);
4014 if (!bs) {
4015 error_setg(errp, "Device '%s' has no medium", device);
4018 return bs;
4022 if (node_name) {
4023 bs = bdrv_find_node(node_name);
4025 if (bs) {
4026 return bs;
4030 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4031 device ? device : "",
4032 node_name ? node_name : "");
4033 return NULL;
4036 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4037 * return false. If either argument is NULL, return false. */
4038 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4040 while (top && top != base) {
4041 top = backing_bs(top);
4044 return top != NULL;
4047 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4049 if (!bs) {
4050 return QTAILQ_FIRST(&graph_bdrv_states);
4052 return QTAILQ_NEXT(bs, node_list);
4055 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
4057 if (!bs) {
4058 return QTAILQ_FIRST(&all_bdrv_states);
4060 return QTAILQ_NEXT(bs, bs_list);
4063 const char *bdrv_get_node_name(const BlockDriverState *bs)
4065 return bs->node_name;
4068 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4070 BdrvChild *c;
4071 const char *name;
4073 /* If multiple parents have a name, just pick the first one. */
4074 QLIST_FOREACH(c, &bs->parents, next_parent) {
4075 if (c->role->get_name) {
4076 name = c->role->get_name(c);
4077 if (name && *name) {
4078 return name;
4083 return NULL;
4086 /* TODO check what callers really want: bs->node_name or blk_name() */
4087 const char *bdrv_get_device_name(const BlockDriverState *bs)
4089 return bdrv_get_parent_name(bs) ?: "";
4092 /* This can be used to identify nodes that might not have a device
4093 * name associated. Since node and device names live in the same
4094 * namespace, the result is unambiguous. The exception is if both are
4095 * absent, then this returns an empty (non-null) string. */
4096 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4098 return bdrv_get_parent_name(bs) ?: bs->node_name;
4101 int bdrv_get_flags(BlockDriverState *bs)
4103 return bs->open_flags;
4106 int bdrv_has_zero_init_1(BlockDriverState *bs)
4108 return 1;
4111 int bdrv_has_zero_init(BlockDriverState *bs)
4113 if (!bs->drv) {
4114 return 0;
4117 /* If BS is a copy on write image, it is initialized to
4118 the contents of the base image, which may not be zeroes. */
4119 if (bs->backing) {
4120 return 0;
4122 if (bs->drv->bdrv_has_zero_init) {
4123 return bs->drv->bdrv_has_zero_init(bs);
4125 if (bs->file && bs->drv->is_filter) {
4126 return bdrv_has_zero_init(bs->file->bs);
4129 /* safe default */
4130 return 0;
4133 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4135 BlockDriverInfo bdi;
4137 if (bs->backing) {
4138 return false;
4141 if (bdrv_get_info(bs, &bdi) == 0) {
4142 return bdi.unallocated_blocks_are_zero;
4145 return false;
4148 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4150 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4151 return false;
4154 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4157 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4159 if (bs->backing && bs->backing->bs->encrypted)
4160 return bs->backing_file;
4161 else if (bs->encrypted)
4162 return bs->filename;
4163 else
4164 return NULL;
4167 void bdrv_get_backing_filename(BlockDriverState *bs,
4168 char *filename, int filename_size)
4170 pstrcpy(filename, filename_size, bs->backing_file);
4173 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4175 BlockDriver *drv = bs->drv;
4176 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4177 if (!drv) {
4178 return -ENOMEDIUM;
4180 if (!drv->bdrv_get_info) {
4181 if (bs->file && drv->is_filter) {
4182 return bdrv_get_info(bs->file->bs, bdi);
4184 return -ENOTSUP;
4186 memset(bdi, 0, sizeof(*bdi));
4187 return drv->bdrv_get_info(bs, bdi);
4190 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4192 BlockDriver *drv = bs->drv;
4193 if (drv && drv->bdrv_get_specific_info) {
4194 return drv->bdrv_get_specific_info(bs);
4196 return NULL;
4199 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4201 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4202 return;
4205 bs->drv->bdrv_debug_event(bs, event);
4208 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4209 const char *tag)
4211 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4212 bs = bs->file ? bs->file->bs : NULL;
4215 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4216 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4219 return -ENOTSUP;
4222 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4224 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4225 bs = bs->file ? bs->file->bs : NULL;
4228 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4229 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4232 return -ENOTSUP;
4235 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4237 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4238 bs = bs->file ? bs->file->bs : NULL;
4241 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4242 return bs->drv->bdrv_debug_resume(bs, tag);
4245 return -ENOTSUP;
4248 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4250 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4251 bs = bs->file ? bs->file->bs : NULL;
4254 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4255 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4258 return false;
4261 /* backing_file can either be relative, or absolute, or a protocol. If it is
4262 * relative, it must be relative to the chain. So, passing in bs->filename
4263 * from a BDS as backing_file should not be done, as that may be relative to
4264 * the CWD rather than the chain. */
4265 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4266 const char *backing_file)
4268 char *filename_full = NULL;
4269 char *backing_file_full = NULL;
4270 char *filename_tmp = NULL;
4271 int is_protocol = 0;
4272 BlockDriverState *curr_bs = NULL;
4273 BlockDriverState *retval = NULL;
4274 Error *local_error = NULL;
4276 if (!bs || !bs->drv || !backing_file) {
4277 return NULL;
4280 filename_full = g_malloc(PATH_MAX);
4281 backing_file_full = g_malloc(PATH_MAX);
4282 filename_tmp = g_malloc(PATH_MAX);
4284 is_protocol = path_has_protocol(backing_file);
4286 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4288 /* If either of the filename paths is actually a protocol, then
4289 * compare unmodified paths; otherwise make paths relative */
4290 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4291 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4292 retval = curr_bs->backing->bs;
4293 break;
4295 /* Also check against the full backing filename for the image */
4296 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4297 &local_error);
4298 if (local_error == NULL) {
4299 if (strcmp(backing_file, backing_file_full) == 0) {
4300 retval = curr_bs->backing->bs;
4301 break;
4303 } else {
4304 error_free(local_error);
4305 local_error = NULL;
4307 } else {
4308 /* If not an absolute filename path, make it relative to the current
4309 * image's filename path */
4310 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4311 backing_file);
4313 /* We are going to compare absolute pathnames */
4314 if (!realpath(filename_tmp, filename_full)) {
4315 continue;
4318 /* We need to make sure the backing filename we are comparing against
4319 * is relative to the current image filename (or absolute) */
4320 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4321 curr_bs->backing_file);
4323 if (!realpath(filename_tmp, backing_file_full)) {
4324 continue;
4327 if (strcmp(backing_file_full, filename_full) == 0) {
4328 retval = curr_bs->backing->bs;
4329 break;
4334 g_free(filename_full);
4335 g_free(backing_file_full);
4336 g_free(filename_tmp);
4337 return retval;
4340 void bdrv_init(void)
4342 module_call_init(MODULE_INIT_BLOCK);
4345 void bdrv_init_with_whitelist(void)
4347 use_bdrv_whitelist = 1;
4348 bdrv_init();
4351 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4352 Error **errp)
4354 BdrvChild *child, *parent;
4355 uint64_t perm, shared_perm;
4356 Error *local_err = NULL;
4357 int ret;
4359 if (!bs->drv) {
4360 return;
4363 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4364 return;
4367 QLIST_FOREACH(child, &bs->children, next) {
4368 bdrv_co_invalidate_cache(child->bs, &local_err);
4369 if (local_err) {
4370 error_propagate(errp, local_err);
4371 return;
4376 * Update permissions, they may differ for inactive nodes.
4378 * Note that the required permissions of inactive images are always a
4379 * subset of the permissions required after activating the image. This
4380 * allows us to just get the permissions upfront without restricting
4381 * drv->bdrv_invalidate_cache().
4383 * It also means that in error cases, we don't have to try and revert to
4384 * the old permissions (which is an operation that could fail, too). We can
4385 * just keep the extended permissions for the next time that an activation
4386 * of the image is tried.
4388 bs->open_flags &= ~BDRV_O_INACTIVE;
4389 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4390 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4391 if (ret < 0) {
4392 bs->open_flags |= BDRV_O_INACTIVE;
4393 error_propagate(errp, local_err);
4394 return;
4396 bdrv_set_perm(bs, perm, shared_perm);
4398 if (bs->drv->bdrv_co_invalidate_cache) {
4399 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4400 if (local_err) {
4401 bs->open_flags |= BDRV_O_INACTIVE;
4402 error_propagate(errp, local_err);
4403 return;
4407 ret = refresh_total_sectors(bs, bs->total_sectors);
4408 if (ret < 0) {
4409 bs->open_flags |= BDRV_O_INACTIVE;
4410 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4411 return;
4414 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4415 if (parent->role->activate) {
4416 parent->role->activate(parent, &local_err);
4417 if (local_err) {
4418 error_propagate(errp, local_err);
4419 return;
4425 typedef struct InvalidateCacheCo {
4426 BlockDriverState *bs;
4427 Error **errp;
4428 bool done;
4429 } InvalidateCacheCo;
4431 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4433 InvalidateCacheCo *ico = opaque;
4434 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4435 ico->done = true;
4438 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4440 Coroutine *co;
4441 InvalidateCacheCo ico = {
4442 .bs = bs,
4443 .done = false,
4444 .errp = errp
4447 if (qemu_in_coroutine()) {
4448 /* Fast-path if already in coroutine context */
4449 bdrv_invalidate_cache_co_entry(&ico);
4450 } else {
4451 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4452 qemu_coroutine_enter(co);
4453 BDRV_POLL_WHILE(bs, !ico.done);
4457 void bdrv_invalidate_cache_all(Error **errp)
4459 BlockDriverState *bs;
4460 Error *local_err = NULL;
4461 BdrvNextIterator it;
4463 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4464 AioContext *aio_context = bdrv_get_aio_context(bs);
4466 aio_context_acquire(aio_context);
4467 bdrv_invalidate_cache(bs, &local_err);
4468 aio_context_release(aio_context);
4469 if (local_err) {
4470 error_propagate(errp, local_err);
4471 bdrv_next_cleanup(&it);
4472 return;
4477 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4478 bool setting_flag)
4480 BdrvChild *child, *parent;
4481 int ret;
4483 if (!bs->drv) {
4484 return -ENOMEDIUM;
4487 if (!setting_flag && bs->drv->bdrv_inactivate) {
4488 ret = bs->drv->bdrv_inactivate(bs);
4489 if (ret < 0) {
4490 return ret;
4494 if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
4495 uint64_t perm, shared_perm;
4497 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4498 if (parent->role->inactivate) {
4499 ret = parent->role->inactivate(parent);
4500 if (ret < 0) {
4501 return ret;
4506 bs->open_flags |= BDRV_O_INACTIVE;
4508 /* Update permissions, they may differ for inactive nodes */
4509 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4510 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4511 bdrv_set_perm(bs, perm, shared_perm);
4514 QLIST_FOREACH(child, &bs->children, next) {
4515 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4516 if (ret < 0) {
4517 return ret;
4521 /* At this point persistent bitmaps should be already stored by the format
4522 * driver */
4523 bdrv_release_persistent_dirty_bitmaps(bs);
4525 return 0;
4528 int bdrv_inactivate_all(void)
4530 BlockDriverState *bs = NULL;
4531 BdrvNextIterator it;
4532 int ret = 0;
4533 int pass;
4534 GSList *aio_ctxs = NULL, *ctx;
4536 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4537 AioContext *aio_context = bdrv_get_aio_context(bs);
4539 if (!g_slist_find(aio_ctxs, aio_context)) {
4540 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4541 aio_context_acquire(aio_context);
4545 /* We do two passes of inactivation. The first pass calls to drivers'
4546 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4547 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4548 * is allowed. */
4549 for (pass = 0; pass < 2; pass++) {
4550 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4551 ret = bdrv_inactivate_recurse(bs, pass);
4552 if (ret < 0) {
4553 bdrv_next_cleanup(&it);
4554 goto out;
4559 out:
4560 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4561 AioContext *aio_context = ctx->data;
4562 aio_context_release(aio_context);
4564 g_slist_free(aio_ctxs);
4566 return ret;
4569 /**************************************************************/
4570 /* removable device support */
4573 * Return TRUE if the media is present
4575 bool bdrv_is_inserted(BlockDriverState *bs)
4577 BlockDriver *drv = bs->drv;
4578 BdrvChild *child;
4580 if (!drv) {
4581 return false;
4583 if (drv->bdrv_is_inserted) {
4584 return drv->bdrv_is_inserted(bs);
4586 QLIST_FOREACH(child, &bs->children, next) {
4587 if (!bdrv_is_inserted(child->bs)) {
4588 return false;
4591 return true;
4595 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4597 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4599 BlockDriver *drv = bs->drv;
4601 if (drv && drv->bdrv_eject) {
4602 drv->bdrv_eject(bs, eject_flag);
4607 * Lock or unlock the media (if it is locked, the user won't be able
4608 * to eject it manually).
4610 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4612 BlockDriver *drv = bs->drv;
4614 trace_bdrv_lock_medium(bs, locked);
4616 if (drv && drv->bdrv_lock_medium) {
4617 drv->bdrv_lock_medium(bs, locked);
4621 /* Get a reference to bs */
4622 void bdrv_ref(BlockDriverState *bs)
4624 bs->refcnt++;
4627 /* Release a previously grabbed reference to bs.
4628 * If after releasing, reference count is zero, the BlockDriverState is
4629 * deleted. */
4630 void bdrv_unref(BlockDriverState *bs)
4632 if (!bs) {
4633 return;
4635 assert(bs->refcnt > 0);
4636 if (--bs->refcnt == 0) {
4637 bdrv_delete(bs);
4641 struct BdrvOpBlocker {
4642 Error *reason;
4643 QLIST_ENTRY(BdrvOpBlocker) list;
4646 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4648 BdrvOpBlocker *blocker;
4649 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4650 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4651 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4652 error_propagate(errp, error_copy(blocker->reason));
4653 error_prepend(errp, "Node '%s' is busy: ",
4654 bdrv_get_device_or_node_name(bs));
4655 return true;
4657 return false;
4660 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4662 BdrvOpBlocker *blocker;
4663 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4665 blocker = g_new0(BdrvOpBlocker, 1);
4666 blocker->reason = reason;
4667 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4670 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4672 BdrvOpBlocker *blocker, *next;
4673 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4674 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4675 if (blocker->reason == reason) {
4676 QLIST_REMOVE(blocker, list);
4677 g_free(blocker);
4682 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4684 int i;
4685 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4686 bdrv_op_block(bs, i, reason);
4690 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4692 int i;
4693 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4694 bdrv_op_unblock(bs, i, reason);
4698 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4700 int i;
4702 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4703 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4704 return false;
4707 return true;
4710 void bdrv_img_create(const char *filename, const char *fmt,
4711 const char *base_filename, const char *base_fmt,
4712 char *options, uint64_t img_size, int flags, bool quiet,
4713 Error **errp)
4715 QemuOptsList *create_opts = NULL;
4716 QemuOpts *opts = NULL;
4717 const char *backing_fmt, *backing_file;
4718 int64_t size;
4719 BlockDriver *drv, *proto_drv;
4720 Error *local_err = NULL;
4721 int ret = 0;
4723 /* Find driver and parse its options */
4724 drv = bdrv_find_format(fmt);
4725 if (!drv) {
4726 error_setg(errp, "Unknown file format '%s'", fmt);
4727 return;
4730 proto_drv = bdrv_find_protocol(filename, true, errp);
4731 if (!proto_drv) {
4732 return;
4735 if (!drv->create_opts) {
4736 error_setg(errp, "Format driver '%s' does not support image creation",
4737 drv->format_name);
4738 return;
4741 if (!proto_drv->create_opts) {
4742 error_setg(errp, "Protocol driver '%s' does not support image creation",
4743 proto_drv->format_name);
4744 return;
4747 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4748 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4750 /* Create parameter list with default values */
4751 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4752 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4754 /* Parse -o options */
4755 if (options) {
4756 qemu_opts_do_parse(opts, options, NULL, &local_err);
4757 if (local_err) {
4758 error_report_err(local_err);
4759 local_err = NULL;
4760 error_setg(errp, "Invalid options for file format '%s'", fmt);
4761 goto out;
4765 if (base_filename) {
4766 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4767 if (local_err) {
4768 error_setg(errp, "Backing file not supported for file format '%s'",
4769 fmt);
4770 goto out;
4774 if (base_fmt) {
4775 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4776 if (local_err) {
4777 error_setg(errp, "Backing file format not supported for file "
4778 "format '%s'", fmt);
4779 goto out;
4783 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4784 if (backing_file) {
4785 if (!strcmp(filename, backing_file)) {
4786 error_setg(errp, "Error: Trying to create an image with the "
4787 "same filename as the backing file");
4788 goto out;
4792 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4794 /* The size for the image must always be specified, unless we have a backing
4795 * file and we have not been forbidden from opening it. */
4796 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4797 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4798 BlockDriverState *bs;
4799 char *full_backing = g_new0(char, PATH_MAX);
4800 int back_flags;
4801 QDict *backing_options = NULL;
4803 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4804 full_backing, PATH_MAX,
4805 &local_err);
4806 if (local_err) {
4807 g_free(full_backing);
4808 goto out;
4811 /* backing files always opened read-only */
4812 back_flags = flags;
4813 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4815 backing_options = qdict_new();
4816 if (backing_fmt) {
4817 qdict_put_str(backing_options, "driver", backing_fmt);
4819 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
4821 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4822 &local_err);
4823 g_free(full_backing);
4824 if (!bs && size != -1) {
4825 /* Couldn't open BS, but we have a size, so it's nonfatal */
4826 warn_reportf_err(local_err,
4827 "Could not verify backing image. "
4828 "This may become an error in future versions.\n");
4829 local_err = NULL;
4830 } else if (!bs) {
4831 /* Couldn't open bs, do not have size */
4832 error_append_hint(&local_err,
4833 "Could not open backing image to determine size.\n");
4834 goto out;
4835 } else {
4836 if (size == -1) {
4837 /* Opened BS, have no size */
4838 size = bdrv_getlength(bs);
4839 if (size < 0) {
4840 error_setg_errno(errp, -size, "Could not get size of '%s'",
4841 backing_file);
4842 bdrv_unref(bs);
4843 goto out;
4845 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4847 bdrv_unref(bs);
4849 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4851 if (size == -1) {
4852 error_setg(errp, "Image creation needs a size parameter");
4853 goto out;
4856 if (!quiet) {
4857 printf("Formatting '%s', fmt=%s ", filename, fmt);
4858 qemu_opts_print(opts, " ");
4859 puts("");
4862 ret = bdrv_create(drv, filename, opts, &local_err);
4864 if (ret == -EFBIG) {
4865 /* This is generally a better message than whatever the driver would
4866 * deliver (especially because of the cluster_size_hint), since that
4867 * is most probably not much different from "image too large". */
4868 const char *cluster_size_hint = "";
4869 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4870 cluster_size_hint = " (try using a larger cluster size)";
4872 error_setg(errp, "The image size is too large for file format '%s'"
4873 "%s", fmt, cluster_size_hint);
4874 error_free(local_err);
4875 local_err = NULL;
4878 out:
4879 qemu_opts_del(opts);
4880 qemu_opts_free(create_opts);
4881 error_propagate(errp, local_err);
4884 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4886 return bs ? bs->aio_context : qemu_get_aio_context();
4889 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4891 aio_co_enter(bdrv_get_aio_context(bs), co);
4894 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4896 QLIST_REMOVE(ban, list);
4897 g_free(ban);
4900 void bdrv_detach_aio_context(BlockDriverState *bs)
4902 BdrvAioNotifier *baf, *baf_tmp;
4903 BdrvChild *child;
4905 if (!bs->drv) {
4906 return;
4909 assert(!bs->walking_aio_notifiers);
4910 bs->walking_aio_notifiers = true;
4911 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4912 if (baf->deleted) {
4913 bdrv_do_remove_aio_context_notifier(baf);
4914 } else {
4915 baf->detach_aio_context(baf->opaque);
4918 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4919 * remove remaining aio notifiers if we aren't called again.
4921 bs->walking_aio_notifiers = false;
4923 if (bs->drv->bdrv_detach_aio_context) {
4924 bs->drv->bdrv_detach_aio_context(bs);
4926 QLIST_FOREACH(child, &bs->children, next) {
4927 bdrv_detach_aio_context(child->bs);
4930 bs->aio_context = NULL;
4933 void bdrv_attach_aio_context(BlockDriverState *bs,
4934 AioContext *new_context)
4936 BdrvAioNotifier *ban, *ban_tmp;
4937 BdrvChild *child;
4939 if (!bs->drv) {
4940 return;
4943 bs->aio_context = new_context;
4945 QLIST_FOREACH(child, &bs->children, next) {
4946 bdrv_attach_aio_context(child->bs, new_context);
4948 if (bs->drv->bdrv_attach_aio_context) {
4949 bs->drv->bdrv_attach_aio_context(bs, new_context);
4952 assert(!bs->walking_aio_notifiers);
4953 bs->walking_aio_notifiers = true;
4954 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4955 if (ban->deleted) {
4956 bdrv_do_remove_aio_context_notifier(ban);
4957 } else {
4958 ban->attached_aio_context(new_context, ban->opaque);
4961 bs->walking_aio_notifiers = false;
4964 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4966 AioContext *ctx = bdrv_get_aio_context(bs);
4968 aio_disable_external(ctx);
4969 bdrv_parent_drained_begin(bs, NULL, false);
4970 bdrv_drain(bs); /* ensure there are no in-flight requests */
4972 while (aio_poll(ctx, false)) {
4973 /* wait for all bottom halves to execute */
4976 bdrv_detach_aio_context(bs);
4978 /* This function executes in the old AioContext so acquire the new one in
4979 * case it runs in a different thread.
4981 aio_context_acquire(new_context);
4982 bdrv_attach_aio_context(bs, new_context);
4983 bdrv_parent_drained_end(bs, NULL, false);
4984 aio_enable_external(ctx);
4985 aio_context_release(new_context);
4988 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4989 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4990 void (*detach_aio_context)(void *opaque), void *opaque)
4992 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4993 *ban = (BdrvAioNotifier){
4994 .attached_aio_context = attached_aio_context,
4995 .detach_aio_context = detach_aio_context,
4996 .opaque = opaque
4999 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
5002 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
5003 void (*attached_aio_context)(AioContext *,
5004 void *),
5005 void (*detach_aio_context)(void *),
5006 void *opaque)
5008 BdrvAioNotifier *ban, *ban_next;
5010 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5011 if (ban->attached_aio_context == attached_aio_context &&
5012 ban->detach_aio_context == detach_aio_context &&
5013 ban->opaque == opaque &&
5014 ban->deleted == false)
5016 if (bs->walking_aio_notifiers) {
5017 ban->deleted = true;
5018 } else {
5019 bdrv_do_remove_aio_context_notifier(ban);
5021 return;
5025 abort();
5028 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
5029 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5030 Error **errp)
5032 if (!bs->drv) {
5033 error_setg(errp, "Node is ejected");
5034 return -ENOMEDIUM;
5036 if (!bs->drv->bdrv_amend_options) {
5037 error_setg(errp, "Block driver '%s' does not support option amendment",
5038 bs->drv->format_name);
5039 return -ENOTSUP;
5041 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
5044 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5045 * of block filter and by bdrv_is_first_non_filter.
5046 * It is used to test if the given bs is the candidate or recurse more in the
5047 * node graph.
5049 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5050 BlockDriverState *candidate)
5052 /* return false if basic checks fails */
5053 if (!bs || !bs->drv) {
5054 return false;
5057 /* the code reached a non block filter driver -> check if the bs is
5058 * the same as the candidate. It's the recursion termination condition.
5060 if (!bs->drv->is_filter) {
5061 return bs == candidate;
5063 /* Down this path the driver is a block filter driver */
5065 /* If the block filter recursion method is defined use it to recurse down
5066 * the node graph.
5068 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5069 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5072 /* the driver is a block filter but don't allow to recurse -> return false
5074 return false;
5077 /* This function checks if the candidate is the first non filter bs down it's
5078 * bs chain. Since we don't have pointers to parents it explore all bs chains
5079 * from the top. Some filters can choose not to pass down the recursion.
5081 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5083 BlockDriverState *bs;
5084 BdrvNextIterator it;
5086 /* walk down the bs forest recursively */
5087 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5088 bool perm;
5090 /* try to recurse in this top level bs */
5091 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5093 /* candidate is the first non filter */
5094 if (perm) {
5095 bdrv_next_cleanup(&it);
5096 return true;
5100 return false;
5103 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5104 const char *node_name, Error **errp)
5106 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5107 AioContext *aio_context;
5109 if (!to_replace_bs) {
5110 error_setg(errp, "Node name '%s' not found", node_name);
5111 return NULL;
5114 aio_context = bdrv_get_aio_context(to_replace_bs);
5115 aio_context_acquire(aio_context);
5117 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5118 to_replace_bs = NULL;
5119 goto out;
5122 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5123 * most non filter in order to prevent data corruption.
5124 * Another benefit is that this tests exclude backing files which are
5125 * blocked by the backing blockers.
5127 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5128 error_setg(errp, "Only top most non filter can be replaced");
5129 to_replace_bs = NULL;
5130 goto out;
5133 out:
5134 aio_context_release(aio_context);
5135 return to_replace_bs;
5138 static bool append_open_options(QDict *d, BlockDriverState *bs)
5140 const QDictEntry *entry;
5141 QemuOptDesc *desc;
5142 BdrvChild *child;
5143 bool found_any = false;
5145 for (entry = qdict_first(bs->options); entry;
5146 entry = qdict_next(bs->options, entry))
5148 /* Exclude node-name references to children */
5149 QLIST_FOREACH(child, &bs->children, next) {
5150 if (!strcmp(entry->key, child->name)) {
5151 break;
5154 if (child) {
5155 continue;
5158 /* And exclude all non-driver-specific options */
5159 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
5160 if (!strcmp(qdict_entry_key(entry), desc->name)) {
5161 break;
5164 if (desc->name) {
5165 continue;
5168 qdict_put_obj(d, qdict_entry_key(entry),
5169 qobject_ref(qdict_entry_value(entry)));
5170 found_any = true;
5173 return found_any;
5176 /* Updates the following BDS fields:
5177 * - exact_filename: A filename which may be used for opening a block device
5178 * which (mostly) equals the given BDS (even without any
5179 * other options; so reading and writing must return the same
5180 * results, but caching etc. may be different)
5181 * - full_open_options: Options which, when given when opening a block device
5182 * (without a filename), result in a BDS (mostly)
5183 * equalling the given one
5184 * - filename: If exact_filename is set, it is copied here. Otherwise,
5185 * full_open_options is converted to a JSON object, prefixed with
5186 * "json:" (for use through the JSON pseudo protocol) and put here.
5188 void bdrv_refresh_filename(BlockDriverState *bs)
5190 BlockDriver *drv = bs->drv;
5191 QDict *opts;
5193 if (!drv) {
5194 return;
5197 /* This BDS's file name will most probably depend on its file's name, so
5198 * refresh that first */
5199 if (bs->file) {
5200 bdrv_refresh_filename(bs->file->bs);
5203 if (drv->bdrv_refresh_filename) {
5204 /* Obsolete information is of no use here, so drop the old file name
5205 * information before refreshing it */
5206 bs->exact_filename[0] = '\0';
5207 if (bs->full_open_options) {
5208 qobject_unref(bs->full_open_options);
5209 bs->full_open_options = NULL;
5212 opts = qdict_new();
5213 append_open_options(opts, bs);
5214 drv->bdrv_refresh_filename(bs, opts);
5215 qobject_unref(opts);
5216 } else if (bs->file) {
5217 /* Try to reconstruct valid information from the underlying file */
5218 bool has_open_options;
5220 bs->exact_filename[0] = '\0';
5221 if (bs->full_open_options) {
5222 qobject_unref(bs->full_open_options);
5223 bs->full_open_options = NULL;
5226 opts = qdict_new();
5227 has_open_options = append_open_options(opts, bs);
5229 /* If no specific options have been given for this BDS, the filename of
5230 * the underlying file should suffice for this one as well */
5231 if (bs->file->bs->exact_filename[0] && !has_open_options) {
5232 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5234 /* Reconstructing the full options QDict is simple for most format block
5235 * drivers, as long as the full options are known for the underlying
5236 * file BDS. The full options QDict of that file BDS should somehow
5237 * contain a representation of the filename, therefore the following
5238 * suffices without querying the (exact_)filename of this BDS. */
5239 if (bs->file->bs->full_open_options) {
5240 qdict_put_str(opts, "driver", drv->format_name);
5241 qdict_put(opts, "file",
5242 qobject_ref(bs->file->bs->full_open_options));
5244 bs->full_open_options = opts;
5245 } else {
5246 qobject_unref(opts);
5248 } else if (!bs->full_open_options && qdict_size(bs->options)) {
5249 /* There is no underlying file BDS (at least referenced by BDS.file),
5250 * so the full options QDict should be equal to the options given
5251 * specifically for this block device when it was opened (plus the
5252 * driver specification).
5253 * Because those options don't change, there is no need to update
5254 * full_open_options when it's already set. */
5256 opts = qdict_new();
5257 append_open_options(opts, bs);
5258 qdict_put_str(opts, "driver", drv->format_name);
5260 if (bs->exact_filename[0]) {
5261 /* This may not work for all block protocol drivers (some may
5262 * require this filename to be parsed), but we have to find some
5263 * default solution here, so just include it. If some block driver
5264 * does not support pure options without any filename at all or
5265 * needs some special format of the options QDict, it needs to
5266 * implement the driver-specific bdrv_refresh_filename() function.
5268 qdict_put_str(opts, "filename", bs->exact_filename);
5271 bs->full_open_options = opts;
5274 if (bs->exact_filename[0]) {
5275 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5276 } else if (bs->full_open_options) {
5277 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5278 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5279 qstring_get_str(json));
5280 qobject_unref(json);
5285 * Hot add/remove a BDS's child. So the user can take a child offline when
5286 * it is broken and take a new child online
5288 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5289 Error **errp)
5292 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5293 error_setg(errp, "The node %s does not support adding a child",
5294 bdrv_get_device_or_node_name(parent_bs));
5295 return;
5298 if (!QLIST_EMPTY(&child_bs->parents)) {
5299 error_setg(errp, "The node %s already has a parent",
5300 child_bs->node_name);
5301 return;
5304 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5307 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5309 BdrvChild *tmp;
5311 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5312 error_setg(errp, "The node %s does not support removing a child",
5313 bdrv_get_device_or_node_name(parent_bs));
5314 return;
5317 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5318 if (tmp == child) {
5319 break;
5323 if (!tmp) {
5324 error_setg(errp, "The node %s does not have a child named %s",
5325 bdrv_get_device_or_node_name(parent_bs),
5326 bdrv_get_device_or_node_name(child->bs));
5327 return;
5330 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5333 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5334 uint32_t granularity, Error **errp)
5336 BlockDriver *drv = bs->drv;
5338 if (!drv) {
5339 error_setg_errno(errp, ENOMEDIUM,
5340 "Can't store persistent bitmaps to %s",
5341 bdrv_get_device_or_node_name(bs));
5342 return false;
5345 if (!drv->bdrv_can_store_new_dirty_bitmap) {
5346 error_setg_errno(errp, ENOTSUP,
5347 "Can't store persistent bitmaps to %s",
5348 bdrv_get_device_or_node_name(bs));
5349 return false;
5352 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);