hw/arm/virt: explicitly allocate cpu_index for cpus
[qemu/kevin.git] / block.c
bloba45b9b5c29b12a297e239f846ad4235cda68c752
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
44 #include "qemu/id.h"
45 #include "qapi/util.h"
47 #ifdef CONFIG_BSD
48 #include <sys/ioctl.h>
49 #include <sys/queue.h>
50 #ifndef __DragonFly__
51 #include <sys/disk.h>
52 #endif
53 #endif
55 #ifdef _WIN32
56 #include <windows.h>
57 #endif
59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
61 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
64 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
65 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
67 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
68 QLIST_HEAD_INITIALIZER(bdrv_drivers);
70 static BlockDriverState *bdrv_open_inherit(const char *filename,
71 const char *reference,
72 QDict *options, int flags,
73 BlockDriverState *parent,
74 const BdrvChildRole *child_role,
75 Error **errp);
77 /* If non-zero, use only whitelisted block drivers */
78 static int use_bdrv_whitelist;
80 #ifdef _WIN32
81 static int is_windows_drive_prefix(const char *filename)
83 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
84 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
85 filename[1] == ':');
88 int is_windows_drive(const char *filename)
90 if (is_windows_drive_prefix(filename) &&
91 filename[2] == '\0')
92 return 1;
93 if (strstart(filename, "\\\\.\\", NULL) ||
94 strstart(filename, "//./", NULL))
95 return 1;
96 return 0;
98 #endif
100 size_t bdrv_opt_mem_align(BlockDriverState *bs)
102 if (!bs || !bs->drv) {
103 /* page size or 4k (hdd sector size) should be on the safe side */
104 return MAX(4096, getpagesize());
107 return bs->bl.opt_mem_alignment;
110 size_t bdrv_min_mem_align(BlockDriverState *bs)
112 if (!bs || !bs->drv) {
113 /* page size or 4k (hdd sector size) should be on the safe side */
114 return MAX(4096, getpagesize());
117 return bs->bl.min_mem_alignment;
120 /* check if the path starts with "<protocol>:" */
121 int path_has_protocol(const char *path)
123 const char *p;
125 #ifdef _WIN32
126 if (is_windows_drive(path) ||
127 is_windows_drive_prefix(path)) {
128 return 0;
130 p = path + strcspn(path, ":/\\");
131 #else
132 p = path + strcspn(path, ":/");
133 #endif
135 return *p == ':';
138 int path_is_absolute(const char *path)
140 #ifdef _WIN32
141 /* specific case for names like: "\\.\d:" */
142 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
143 return 1;
145 return (*path == '/' || *path == '\\');
146 #else
147 return (*path == '/');
148 #endif
151 /* if filename is absolute, just copy it to dest. Otherwise, build a
152 path to it by considering it is relative to base_path. URL are
153 supported. */
154 void path_combine(char *dest, int dest_size,
155 const char *base_path,
156 const char *filename)
158 const char *p, *p1;
159 int len;
161 if (dest_size <= 0)
162 return;
163 if (path_is_absolute(filename)) {
164 pstrcpy(dest, dest_size, filename);
165 } else {
166 p = strchr(base_path, ':');
167 if (p)
168 p++;
169 else
170 p = base_path;
171 p1 = strrchr(base_path, '/');
172 #ifdef _WIN32
174 const char *p2;
175 p2 = strrchr(base_path, '\\');
176 if (!p1 || p2 > p1)
177 p1 = p2;
179 #endif
180 if (p1)
181 p1++;
182 else
183 p1 = base_path;
184 if (p1 > p)
185 p = p1;
186 len = p - base_path;
187 if (len > dest_size - 1)
188 len = dest_size - 1;
189 memcpy(dest, base_path, len);
190 dest[len] = '\0';
191 pstrcat(dest, dest_size, filename);
195 bool bdrv_is_read_only(BlockDriverState *bs)
197 return bs->read_only;
200 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
202 /* Do not set read_only if copy_on_read is enabled */
203 if (bs->copy_on_read && read_only) {
204 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
205 bdrv_get_device_or_node_name(bs));
206 return -EINVAL;
209 /* Do not clear read_only if it is prohibited */
210 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR)) {
211 error_setg(errp, "Node '%s' is read only",
212 bdrv_get_device_or_node_name(bs));
213 return -EPERM;
216 return 0;
219 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
221 int ret = 0;
223 ret = bdrv_can_set_read_only(bs, read_only, errp);
224 if (ret < 0) {
225 return ret;
228 bs->read_only = read_only;
229 return 0;
232 void bdrv_get_full_backing_filename_from_filename(const char *backed,
233 const char *backing,
234 char *dest, size_t sz,
235 Error **errp)
237 if (backing[0] == '\0' || path_has_protocol(backing) ||
238 path_is_absolute(backing))
240 pstrcpy(dest, sz, backing);
241 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
242 error_setg(errp, "Cannot use relative backing file names for '%s'",
243 backed);
244 } else {
245 path_combine(dest, sz, backed, backing);
249 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
250 Error **errp)
252 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
254 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
255 dest, sz, errp);
258 void bdrv_register(BlockDriver *bdrv)
260 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
263 BlockDriverState *bdrv_new(void)
265 BlockDriverState *bs;
266 int i;
268 bs = g_new0(BlockDriverState, 1);
269 QLIST_INIT(&bs->dirty_bitmaps);
270 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
271 QLIST_INIT(&bs->op_blockers[i]);
273 notifier_with_return_list_init(&bs->before_write_notifiers);
274 bs->refcnt = 1;
275 bs->aio_context = qemu_get_aio_context();
277 qemu_co_queue_init(&bs->flush_queue);
279 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
281 return bs;
284 static BlockDriver *bdrv_do_find_format(const char *format_name)
286 BlockDriver *drv1;
288 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
289 if (!strcmp(drv1->format_name, format_name)) {
290 return drv1;
294 return NULL;
297 BlockDriver *bdrv_find_format(const char *format_name)
299 BlockDriver *drv1;
300 int i;
302 drv1 = bdrv_do_find_format(format_name);
303 if (drv1) {
304 return drv1;
307 /* The driver isn't registered, maybe we need to load a module */
308 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
309 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
310 block_module_load_one(block_driver_modules[i].library_name);
311 break;
315 return bdrv_do_find_format(format_name);
318 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
320 static const char *whitelist_rw[] = {
321 CONFIG_BDRV_RW_WHITELIST
323 static const char *whitelist_ro[] = {
324 CONFIG_BDRV_RO_WHITELIST
326 const char **p;
328 if (!whitelist_rw[0] && !whitelist_ro[0]) {
329 return 1; /* no whitelist, anything goes */
332 for (p = whitelist_rw; *p; p++) {
333 if (!strcmp(drv->format_name, *p)) {
334 return 1;
337 if (read_only) {
338 for (p = whitelist_ro; *p; p++) {
339 if (!strcmp(drv->format_name, *p)) {
340 return 1;
344 return 0;
347 bool bdrv_uses_whitelist(void)
349 return use_bdrv_whitelist;
352 typedef struct CreateCo {
353 BlockDriver *drv;
354 char *filename;
355 QemuOpts *opts;
356 int ret;
357 Error *err;
358 } CreateCo;
360 static void coroutine_fn bdrv_create_co_entry(void *opaque)
362 Error *local_err = NULL;
363 int ret;
365 CreateCo *cco = opaque;
366 assert(cco->drv);
368 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
369 error_propagate(&cco->err, local_err);
370 cco->ret = ret;
373 int bdrv_create(BlockDriver *drv, const char* filename,
374 QemuOpts *opts, Error **errp)
376 int ret;
378 Coroutine *co;
379 CreateCo cco = {
380 .drv = drv,
381 .filename = g_strdup(filename),
382 .opts = opts,
383 .ret = NOT_DONE,
384 .err = NULL,
387 if (!drv->bdrv_create) {
388 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
389 ret = -ENOTSUP;
390 goto out;
393 if (qemu_in_coroutine()) {
394 /* Fast-path if already in coroutine context */
395 bdrv_create_co_entry(&cco);
396 } else {
397 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
398 qemu_coroutine_enter(co);
399 while (cco.ret == NOT_DONE) {
400 aio_poll(qemu_get_aio_context(), true);
404 ret = cco.ret;
405 if (ret < 0) {
406 if (cco.err) {
407 error_propagate(errp, cco.err);
408 } else {
409 error_setg_errno(errp, -ret, "Could not create image");
413 out:
414 g_free(cco.filename);
415 return ret;
418 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
420 BlockDriver *drv;
421 Error *local_err = NULL;
422 int ret;
424 drv = bdrv_find_protocol(filename, true, errp);
425 if (drv == NULL) {
426 return -ENOENT;
429 ret = bdrv_create(drv, filename, opts, &local_err);
430 error_propagate(errp, local_err);
431 return ret;
435 * Try to get @bs's logical and physical block size.
436 * On success, store them in @bsz struct and return 0.
437 * On failure return -errno.
438 * @bs must not be empty.
440 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
442 BlockDriver *drv = bs->drv;
444 if (drv && drv->bdrv_probe_blocksizes) {
445 return drv->bdrv_probe_blocksizes(bs, bsz);
448 return -ENOTSUP;
452 * Try to get @bs's geometry (cyls, heads, sectors).
453 * On success, store them in @geo struct and return 0.
454 * On failure return -errno.
455 * @bs must not be empty.
457 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
459 BlockDriver *drv = bs->drv;
461 if (drv && drv->bdrv_probe_geometry) {
462 return drv->bdrv_probe_geometry(bs, geo);
465 return -ENOTSUP;
469 * Create a uniquely-named empty temporary file.
470 * Return 0 upon success, otherwise a negative errno value.
472 int get_tmp_filename(char *filename, int size)
474 #ifdef _WIN32
475 char temp_dir[MAX_PATH];
476 /* GetTempFileName requires that its output buffer (4th param)
477 have length MAX_PATH or greater. */
478 assert(size >= MAX_PATH);
479 return (GetTempPath(MAX_PATH, temp_dir)
480 && GetTempFileName(temp_dir, "qem", 0, filename)
481 ? 0 : -GetLastError());
482 #else
483 int fd;
484 const char *tmpdir;
485 tmpdir = getenv("TMPDIR");
486 if (!tmpdir) {
487 tmpdir = "/var/tmp";
489 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
490 return -EOVERFLOW;
492 fd = mkstemp(filename);
493 if (fd < 0) {
494 return -errno;
496 if (close(fd) != 0) {
497 unlink(filename);
498 return -errno;
500 return 0;
501 #endif
505 * Detect host devices. By convention, /dev/cdrom[N] is always
506 * recognized as a host CDROM.
508 static BlockDriver *find_hdev_driver(const char *filename)
510 int score_max = 0, score;
511 BlockDriver *drv = NULL, *d;
513 QLIST_FOREACH(d, &bdrv_drivers, list) {
514 if (d->bdrv_probe_device) {
515 score = d->bdrv_probe_device(filename);
516 if (score > score_max) {
517 score_max = score;
518 drv = d;
523 return drv;
526 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
528 BlockDriver *drv1;
530 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
531 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
532 return drv1;
536 return NULL;
539 BlockDriver *bdrv_find_protocol(const char *filename,
540 bool allow_protocol_prefix,
541 Error **errp)
543 BlockDriver *drv1;
544 char protocol[128];
545 int len;
546 const char *p;
547 int i;
549 /* TODO Drivers without bdrv_file_open must be specified explicitly */
552 * XXX(hch): we really should not let host device detection
553 * override an explicit protocol specification, but moving this
554 * later breaks access to device names with colons in them.
555 * Thanks to the brain-dead persistent naming schemes on udev-
556 * based Linux systems those actually are quite common.
558 drv1 = find_hdev_driver(filename);
559 if (drv1) {
560 return drv1;
563 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
564 return &bdrv_file;
567 p = strchr(filename, ':');
568 assert(p != NULL);
569 len = p - filename;
570 if (len > sizeof(protocol) - 1)
571 len = sizeof(protocol) - 1;
572 memcpy(protocol, filename, len);
573 protocol[len] = '\0';
575 drv1 = bdrv_do_find_protocol(protocol);
576 if (drv1) {
577 return drv1;
580 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
581 if (block_driver_modules[i].protocol_name &&
582 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
583 block_module_load_one(block_driver_modules[i].library_name);
584 break;
588 drv1 = bdrv_do_find_protocol(protocol);
589 if (!drv1) {
590 error_setg(errp, "Unknown protocol '%s'", protocol);
592 return drv1;
596 * Guess image format by probing its contents.
597 * This is not a good idea when your image is raw (CVE-2008-2004), but
598 * we do it anyway for backward compatibility.
600 * @buf contains the image's first @buf_size bytes.
601 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
602 * but can be smaller if the image file is smaller)
603 * @filename is its filename.
605 * For all block drivers, call the bdrv_probe() method to get its
606 * probing score.
607 * Return the first block driver with the highest probing score.
609 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
610 const char *filename)
612 int score_max = 0, score;
613 BlockDriver *drv = NULL, *d;
615 QLIST_FOREACH(d, &bdrv_drivers, list) {
616 if (d->bdrv_probe) {
617 score = d->bdrv_probe(buf, buf_size, filename);
618 if (score > score_max) {
619 score_max = score;
620 drv = d;
625 return drv;
628 static int find_image_format(BlockBackend *file, const char *filename,
629 BlockDriver **pdrv, Error **errp)
631 BlockDriver *drv;
632 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
633 int ret = 0;
635 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
636 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
637 *pdrv = &bdrv_raw;
638 return ret;
641 ret = blk_pread(file, 0, buf, sizeof(buf));
642 if (ret < 0) {
643 error_setg_errno(errp, -ret, "Could not read image for determining its "
644 "format");
645 *pdrv = NULL;
646 return ret;
649 drv = bdrv_probe_all(buf, ret, filename);
650 if (!drv) {
651 error_setg(errp, "Could not determine image format: No compatible "
652 "driver found");
653 ret = -ENOENT;
655 *pdrv = drv;
656 return ret;
660 * Set the current 'total_sectors' value
661 * Return 0 on success, -errno on error.
663 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
665 BlockDriver *drv = bs->drv;
667 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
668 if (bdrv_is_sg(bs))
669 return 0;
671 /* query actual device if possible, otherwise just trust the hint */
672 if (drv->bdrv_getlength) {
673 int64_t length = drv->bdrv_getlength(bs);
674 if (length < 0) {
675 return length;
677 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
680 bs->total_sectors = hint;
681 return 0;
685 * Combines a QDict of new block driver @options with any missing options taken
686 * from @old_options, so that leaving out an option defaults to its old value.
688 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
689 QDict *old_options)
691 if (bs->drv && bs->drv->bdrv_join_options) {
692 bs->drv->bdrv_join_options(options, old_options);
693 } else {
694 qdict_join(options, old_options, false);
699 * Set open flags for a given discard mode
701 * Return 0 on success, -1 if the discard mode was invalid.
703 int bdrv_parse_discard_flags(const char *mode, int *flags)
705 *flags &= ~BDRV_O_UNMAP;
707 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
708 /* do nothing */
709 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
710 *flags |= BDRV_O_UNMAP;
711 } else {
712 return -1;
715 return 0;
719 * Set open flags for a given cache mode
721 * Return 0 on success, -1 if the cache mode was invalid.
723 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
725 *flags &= ~BDRV_O_CACHE_MASK;
727 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
728 *writethrough = false;
729 *flags |= BDRV_O_NOCACHE;
730 } else if (!strcmp(mode, "directsync")) {
731 *writethrough = true;
732 *flags |= BDRV_O_NOCACHE;
733 } else if (!strcmp(mode, "writeback")) {
734 *writethrough = false;
735 } else if (!strcmp(mode, "unsafe")) {
736 *writethrough = false;
737 *flags |= BDRV_O_NO_FLUSH;
738 } else if (!strcmp(mode, "writethrough")) {
739 *writethrough = true;
740 } else {
741 return -1;
744 return 0;
747 static char *bdrv_child_get_parent_desc(BdrvChild *c)
749 BlockDriverState *parent = c->opaque;
750 return g_strdup(bdrv_get_device_or_node_name(parent));
753 static void bdrv_child_cb_drained_begin(BdrvChild *child)
755 BlockDriverState *bs = child->opaque;
756 bdrv_drained_begin(bs);
759 static void bdrv_child_cb_drained_end(BdrvChild *child)
761 BlockDriverState *bs = child->opaque;
762 bdrv_drained_end(bs);
766 * Returns the options and flags that a temporary snapshot should get, based on
767 * the originally requested flags (the originally requested image will have
768 * flags like a backing file)
770 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
771 int parent_flags, QDict *parent_options)
773 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
775 /* For temporary files, unconditional cache=unsafe is fine */
776 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
777 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
779 /* Copy the read-only option from the parent */
780 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
782 /* aio=native doesn't work for cache.direct=off, so disable it for the
783 * temporary snapshot */
784 *child_flags &= ~BDRV_O_NATIVE_AIO;
788 * Returns the options and flags that bs->file should get if a protocol driver
789 * is expected, based on the given options and flags for the parent BDS
791 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
792 int parent_flags, QDict *parent_options)
794 int flags = parent_flags;
796 /* Enable protocol handling, disable format probing for bs->file */
797 flags |= BDRV_O_PROTOCOL;
799 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
800 * the parent. */
801 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
802 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
804 /* Inherit the read-only option from the parent if it's not set */
805 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
807 /* Our block drivers take care to send flushes and respect unmap policy,
808 * so we can default to enable both on lower layers regardless of the
809 * corresponding parent options. */
810 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
812 /* Clear flags that only apply to the top layer */
813 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
814 BDRV_O_NO_IO);
816 *child_flags = flags;
819 const BdrvChildRole child_file = {
820 .get_parent_desc = bdrv_child_get_parent_desc,
821 .inherit_options = bdrv_inherited_options,
822 .drained_begin = bdrv_child_cb_drained_begin,
823 .drained_end = bdrv_child_cb_drained_end,
827 * Returns the options and flags that bs->file should get if the use of formats
828 * (and not only protocols) is permitted for it, based on the given options and
829 * flags for the parent BDS
831 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
832 int parent_flags, QDict *parent_options)
834 child_file.inherit_options(child_flags, child_options,
835 parent_flags, parent_options);
837 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
840 const BdrvChildRole child_format = {
841 .get_parent_desc = bdrv_child_get_parent_desc,
842 .inherit_options = bdrv_inherited_fmt_options,
843 .drained_begin = bdrv_child_cb_drained_begin,
844 .drained_end = bdrv_child_cb_drained_end,
847 static void bdrv_backing_attach(BdrvChild *c)
849 BlockDriverState *parent = c->opaque;
850 BlockDriverState *backing_hd = c->bs;
852 assert(!parent->backing_blocker);
853 error_setg(&parent->backing_blocker,
854 "node is used as backing hd of '%s'",
855 bdrv_get_device_or_node_name(parent));
857 parent->open_flags &= ~BDRV_O_NO_BACKING;
858 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
859 backing_hd->filename);
860 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
861 backing_hd->drv ? backing_hd->drv->format_name : "");
863 bdrv_op_block_all(backing_hd, parent->backing_blocker);
864 /* Otherwise we won't be able to commit or stream */
865 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
866 parent->backing_blocker);
867 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
868 parent->backing_blocker);
870 * We do backup in 3 ways:
871 * 1. drive backup
872 * The target bs is new opened, and the source is top BDS
873 * 2. blockdev backup
874 * Both the source and the target are top BDSes.
875 * 3. internal backup(used for block replication)
876 * Both the source and the target are backing file
878 * In case 1 and 2, neither the source nor the target is the backing file.
879 * In case 3, we will block the top BDS, so there is only one block job
880 * for the top BDS and its backing chain.
882 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
883 parent->backing_blocker);
884 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
885 parent->backing_blocker);
888 static void bdrv_backing_detach(BdrvChild *c)
890 BlockDriverState *parent = c->opaque;
892 assert(parent->backing_blocker);
893 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
894 error_free(parent->backing_blocker);
895 parent->backing_blocker = NULL;
899 * Returns the options and flags that bs->backing should get, based on the
900 * given options and flags for the parent BDS
902 static void bdrv_backing_options(int *child_flags, QDict *child_options,
903 int parent_flags, QDict *parent_options)
905 int flags = parent_flags;
907 /* The cache mode is inherited unmodified for backing files; except WCE,
908 * which is only applied on the top level (BlockBackend) */
909 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
910 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
912 /* backing files always opened read-only */
913 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
914 flags &= ~BDRV_O_COPY_ON_READ;
916 /* snapshot=on is handled on the top layer */
917 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
919 *child_flags = flags;
922 const BdrvChildRole child_backing = {
923 .get_parent_desc = bdrv_child_get_parent_desc,
924 .attach = bdrv_backing_attach,
925 .detach = bdrv_backing_detach,
926 .inherit_options = bdrv_backing_options,
927 .drained_begin = bdrv_child_cb_drained_begin,
928 .drained_end = bdrv_child_cb_drained_end,
931 static int bdrv_open_flags(BlockDriverState *bs, int flags)
933 int open_flags = flags;
936 * Clear flags that are internal to the block layer before opening the
937 * image.
939 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
942 * Snapshots should be writable.
944 if (flags & BDRV_O_TEMPORARY) {
945 open_flags |= BDRV_O_RDWR;
948 return open_flags;
951 static void update_flags_from_options(int *flags, QemuOpts *opts)
953 *flags &= ~BDRV_O_CACHE_MASK;
955 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
956 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
957 *flags |= BDRV_O_NO_FLUSH;
960 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
961 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
962 *flags |= BDRV_O_NOCACHE;
965 *flags &= ~BDRV_O_RDWR;
967 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
968 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
969 *flags |= BDRV_O_RDWR;
974 static void update_options_from_flags(QDict *options, int flags)
976 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
977 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
979 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
980 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
981 flags & BDRV_O_NO_FLUSH);
983 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
984 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
988 static void bdrv_assign_node_name(BlockDriverState *bs,
989 const char *node_name,
990 Error **errp)
992 char *gen_node_name = NULL;
994 if (!node_name) {
995 node_name = gen_node_name = id_generate(ID_BLOCK);
996 } else if (!id_wellformed(node_name)) {
998 * Check for empty string or invalid characters, but not if it is
999 * generated (generated names use characters not available to the user)
1001 error_setg(errp, "Invalid node name");
1002 return;
1005 /* takes care of avoiding namespaces collisions */
1006 if (blk_by_name(node_name)) {
1007 error_setg(errp, "node-name=%s is conflicting with a device id",
1008 node_name);
1009 goto out;
1012 /* takes care of avoiding duplicates node names */
1013 if (bdrv_find_node(node_name)) {
1014 error_setg(errp, "Duplicate node name");
1015 goto out;
1018 /* copy node name into the bs and insert it into the graph list */
1019 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1020 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1021 out:
1022 g_free(gen_node_name);
1025 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1026 const char *node_name, QDict *options,
1027 int open_flags, Error **errp)
1029 Error *local_err = NULL;
1030 int ret;
1032 bdrv_assign_node_name(bs, node_name, &local_err);
1033 if (local_err) {
1034 error_propagate(errp, local_err);
1035 return -EINVAL;
1038 bs->drv = drv;
1039 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1040 bs->opaque = g_malloc0(drv->instance_size);
1042 if (drv->bdrv_file_open) {
1043 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1044 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1045 } else if (drv->bdrv_open) {
1046 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1047 } else {
1048 ret = 0;
1051 if (ret < 0) {
1052 if (local_err) {
1053 error_propagate(errp, local_err);
1054 } else if (bs->filename[0]) {
1055 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1056 } else {
1057 error_setg_errno(errp, -ret, "Could not open image");
1059 goto free_and_fail;
1062 ret = refresh_total_sectors(bs, bs->total_sectors);
1063 if (ret < 0) {
1064 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1065 goto free_and_fail;
1068 bdrv_refresh_limits(bs, &local_err);
1069 if (local_err) {
1070 error_propagate(errp, local_err);
1071 ret = -EINVAL;
1072 goto free_and_fail;
1075 assert(bdrv_opt_mem_align(bs) != 0);
1076 assert(bdrv_min_mem_align(bs) != 0);
1077 assert(is_power_of_2(bs->bl.request_alignment));
1079 return 0;
1081 free_and_fail:
1082 /* FIXME Close bs first if already opened*/
1083 g_free(bs->opaque);
1084 bs->opaque = NULL;
1085 bs->drv = NULL;
1086 return ret;
1089 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1090 int flags, Error **errp)
1092 BlockDriverState *bs;
1093 int ret;
1095 bs = bdrv_new();
1096 bs->open_flags = flags;
1097 bs->explicit_options = qdict_new();
1098 bs->options = qdict_new();
1099 bs->opaque = NULL;
1101 update_options_from_flags(bs->options, flags);
1103 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1104 if (ret < 0) {
1105 QDECREF(bs->explicit_options);
1106 QDECREF(bs->options);
1107 bdrv_unref(bs);
1108 return NULL;
1111 return bs;
1114 QemuOptsList bdrv_runtime_opts = {
1115 .name = "bdrv_common",
1116 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1117 .desc = {
1119 .name = "node-name",
1120 .type = QEMU_OPT_STRING,
1121 .help = "Node name of the block device node",
1124 .name = "driver",
1125 .type = QEMU_OPT_STRING,
1126 .help = "Block driver to use for the node",
1129 .name = BDRV_OPT_CACHE_DIRECT,
1130 .type = QEMU_OPT_BOOL,
1131 .help = "Bypass software writeback cache on the host",
1134 .name = BDRV_OPT_CACHE_NO_FLUSH,
1135 .type = QEMU_OPT_BOOL,
1136 .help = "Ignore flush requests",
1139 .name = BDRV_OPT_READ_ONLY,
1140 .type = QEMU_OPT_BOOL,
1141 .help = "Node is opened in read-only mode",
1144 .name = "detect-zeroes",
1145 .type = QEMU_OPT_STRING,
1146 .help = "try to optimize zero writes (off, on, unmap)",
1149 .name = "discard",
1150 .type = QEMU_OPT_STRING,
1151 .help = "discard operation (ignore/off, unmap/on)",
1153 { /* end of list */ }
1158 * Common part for opening disk images and files
1160 * Removes all processed options from *options.
1162 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1163 QDict *options, Error **errp)
1165 int ret, open_flags;
1166 const char *filename;
1167 const char *driver_name = NULL;
1168 const char *node_name = NULL;
1169 const char *discard;
1170 const char *detect_zeroes;
1171 QemuOpts *opts;
1172 BlockDriver *drv;
1173 Error *local_err = NULL;
1175 assert(bs->file == NULL);
1176 assert(options != NULL && bs->options != options);
1178 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1179 qemu_opts_absorb_qdict(opts, options, &local_err);
1180 if (local_err) {
1181 error_propagate(errp, local_err);
1182 ret = -EINVAL;
1183 goto fail_opts;
1186 update_flags_from_options(&bs->open_flags, opts);
1188 driver_name = qemu_opt_get(opts, "driver");
1189 drv = bdrv_find_format(driver_name);
1190 assert(drv != NULL);
1192 if (file != NULL) {
1193 filename = blk_bs(file)->filename;
1194 } else {
1196 * Caution: while qdict_get_try_str() is fine, getting
1197 * non-string types would require more care. When @options
1198 * come from -blockdev or blockdev_add, its members are typed
1199 * according to the QAPI schema, but when they come from
1200 * -drive, they're all QString.
1202 filename = qdict_get_try_str(options, "filename");
1205 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1206 error_setg(errp, "The '%s' block driver requires a file name",
1207 drv->format_name);
1208 ret = -EINVAL;
1209 goto fail_opts;
1212 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1213 drv->format_name);
1215 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1217 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1218 error_setg(errp,
1219 !bs->read_only && bdrv_is_whitelisted(drv, true)
1220 ? "Driver '%s' can only be used for read-only devices"
1221 : "Driver '%s' is not whitelisted",
1222 drv->format_name);
1223 ret = -ENOTSUP;
1224 goto fail_opts;
1227 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
1228 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1229 if (!bs->read_only) {
1230 bdrv_enable_copy_on_read(bs);
1231 } else {
1232 error_setg(errp, "Can't use copy-on-read on read-only device");
1233 ret = -EINVAL;
1234 goto fail_opts;
1238 discard = qemu_opt_get(opts, "discard");
1239 if (discard != NULL) {
1240 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1241 error_setg(errp, "Invalid discard option");
1242 ret = -EINVAL;
1243 goto fail_opts;
1247 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1248 if (detect_zeroes) {
1249 BlockdevDetectZeroesOptions value =
1250 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1251 detect_zeroes,
1252 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1253 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1254 &local_err);
1255 if (local_err) {
1256 error_propagate(errp, local_err);
1257 ret = -EINVAL;
1258 goto fail_opts;
1261 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1262 !(bs->open_flags & BDRV_O_UNMAP))
1264 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1265 "without setting discard operation to unmap");
1266 ret = -EINVAL;
1267 goto fail_opts;
1270 bs->detect_zeroes = value;
1273 if (filename != NULL) {
1274 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1275 } else {
1276 bs->filename[0] = '\0';
1278 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1280 /* Open the image, either directly or using a protocol */
1281 open_flags = bdrv_open_flags(bs, bs->open_flags);
1282 node_name = qemu_opt_get(opts, "node-name");
1284 assert(!drv->bdrv_file_open || file == NULL);
1285 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1286 if (ret < 0) {
1287 goto fail_opts;
1290 qemu_opts_del(opts);
1291 return 0;
1293 fail_opts:
1294 qemu_opts_del(opts);
1295 return ret;
1298 static QDict *parse_json_filename(const char *filename, Error **errp)
1300 QObject *options_obj;
1301 QDict *options;
1302 int ret;
1304 ret = strstart(filename, "json:", &filename);
1305 assert(ret);
1307 options_obj = qobject_from_json(filename, errp);
1308 if (!options_obj) {
1309 /* Work around qobject_from_json() lossage TODO fix that */
1310 if (errp && !*errp) {
1311 error_setg(errp, "Could not parse the JSON options");
1312 return NULL;
1314 error_prepend(errp, "Could not parse the JSON options: ");
1315 return NULL;
1318 options = qobject_to_qdict(options_obj);
1319 if (!options) {
1320 qobject_decref(options_obj);
1321 error_setg(errp, "Invalid JSON object given");
1322 return NULL;
1325 qdict_flatten(options);
1327 return options;
1330 static void parse_json_protocol(QDict *options, const char **pfilename,
1331 Error **errp)
1333 QDict *json_options;
1334 Error *local_err = NULL;
1336 /* Parse json: pseudo-protocol */
1337 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1338 return;
1341 json_options = parse_json_filename(*pfilename, &local_err);
1342 if (local_err) {
1343 error_propagate(errp, local_err);
1344 return;
1347 /* Options given in the filename have lower priority than options
1348 * specified directly */
1349 qdict_join(options, json_options, false);
1350 QDECREF(json_options);
1351 *pfilename = NULL;
1355 * Fills in default options for opening images and converts the legacy
1356 * filename/flags pair to option QDict entries.
1357 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1358 * block driver has been specified explicitly.
1360 static int bdrv_fill_options(QDict **options, const char *filename,
1361 int *flags, Error **errp)
1363 const char *drvname;
1364 bool protocol = *flags & BDRV_O_PROTOCOL;
1365 bool parse_filename = false;
1366 BlockDriver *drv = NULL;
1367 Error *local_err = NULL;
1370 * Caution: while qdict_get_try_str() is fine, getting non-string
1371 * types would require more care. When @options come from
1372 * -blockdev or blockdev_add, its members are typed according to
1373 * the QAPI schema, but when they come from -drive, they're all
1374 * QString.
1376 drvname = qdict_get_try_str(*options, "driver");
1377 if (drvname) {
1378 drv = bdrv_find_format(drvname);
1379 if (!drv) {
1380 error_setg(errp, "Unknown driver '%s'", drvname);
1381 return -ENOENT;
1383 /* If the user has explicitly specified the driver, this choice should
1384 * override the BDRV_O_PROTOCOL flag */
1385 protocol = drv->bdrv_file_open;
1388 if (protocol) {
1389 *flags |= BDRV_O_PROTOCOL;
1390 } else {
1391 *flags &= ~BDRV_O_PROTOCOL;
1394 /* Translate cache options from flags into options */
1395 update_options_from_flags(*options, *flags);
1397 /* Fetch the file name from the options QDict if necessary */
1398 if (protocol && filename) {
1399 if (!qdict_haskey(*options, "filename")) {
1400 qdict_put_str(*options, "filename", filename);
1401 parse_filename = true;
1402 } else {
1403 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1404 "the same time");
1405 return -EINVAL;
1409 /* Find the right block driver */
1410 /* See cautionary note on accessing @options above */
1411 filename = qdict_get_try_str(*options, "filename");
1413 if (!drvname && protocol) {
1414 if (filename) {
1415 drv = bdrv_find_protocol(filename, parse_filename, errp);
1416 if (!drv) {
1417 return -EINVAL;
1420 drvname = drv->format_name;
1421 qdict_put_str(*options, "driver", drvname);
1422 } else {
1423 error_setg(errp, "Must specify either driver or file");
1424 return -EINVAL;
1428 assert(drv || !protocol);
1430 /* Driver-specific filename parsing */
1431 if (drv && drv->bdrv_parse_filename && parse_filename) {
1432 drv->bdrv_parse_filename(filename, *options, &local_err);
1433 if (local_err) {
1434 error_propagate(errp, local_err);
1435 return -EINVAL;
1438 if (!drv->bdrv_needs_filename) {
1439 qdict_del(*options, "filename");
1443 return 0;
1446 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1447 GSList *ignore_children, Error **errp);
1448 static void bdrv_child_abort_perm_update(BdrvChild *c);
1449 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1452 * Check whether permissions on this node can be changed in a way that
1453 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1454 * permissions of all its parents. This involves checking whether all necessary
1455 * permission changes to child nodes can be performed.
1457 * A call to this function must always be followed by a call to bdrv_set_perm()
1458 * or bdrv_abort_perm_update().
1460 static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1461 uint64_t cumulative_shared_perms,
1462 GSList *ignore_children, Error **errp)
1464 BlockDriver *drv = bs->drv;
1465 BdrvChild *c;
1466 int ret;
1468 /* Write permissions never work with read-only images */
1469 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1470 bdrv_is_read_only(bs))
1472 error_setg(errp, "Block node is read-only");
1473 return -EPERM;
1476 /* Check this node */
1477 if (!drv) {
1478 return 0;
1481 if (drv->bdrv_check_perm) {
1482 return drv->bdrv_check_perm(bs, cumulative_perms,
1483 cumulative_shared_perms, errp);
1486 /* Drivers that never have children can omit .bdrv_child_perm() */
1487 if (!drv->bdrv_child_perm) {
1488 assert(QLIST_EMPTY(&bs->children));
1489 return 0;
1492 /* Check all children */
1493 QLIST_FOREACH(c, &bs->children, next) {
1494 uint64_t cur_perm, cur_shared;
1495 drv->bdrv_child_perm(bs, c, c->role,
1496 cumulative_perms, cumulative_shared_perms,
1497 &cur_perm, &cur_shared);
1498 ret = bdrv_child_check_perm(c, cur_perm, cur_shared, ignore_children,
1499 errp);
1500 if (ret < 0) {
1501 return ret;
1505 return 0;
1509 * Notifies drivers that after a previous bdrv_check_perm() call, the
1510 * permission update is not performed and any preparations made for it (e.g.
1511 * taken file locks) need to be undone.
1513 * This function recursively notifies all child nodes.
1515 static void bdrv_abort_perm_update(BlockDriverState *bs)
1517 BlockDriver *drv = bs->drv;
1518 BdrvChild *c;
1520 if (!drv) {
1521 return;
1524 if (drv->bdrv_abort_perm_update) {
1525 drv->bdrv_abort_perm_update(bs);
1528 QLIST_FOREACH(c, &bs->children, next) {
1529 bdrv_child_abort_perm_update(c);
1533 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1534 uint64_t cumulative_shared_perms)
1536 BlockDriver *drv = bs->drv;
1537 BdrvChild *c;
1539 if (!drv) {
1540 return;
1543 /* Update this node */
1544 if (drv->bdrv_set_perm) {
1545 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1548 /* Drivers that never have children can omit .bdrv_child_perm() */
1549 if (!drv->bdrv_child_perm) {
1550 assert(QLIST_EMPTY(&bs->children));
1551 return;
1554 /* Update all children */
1555 QLIST_FOREACH(c, &bs->children, next) {
1556 uint64_t cur_perm, cur_shared;
1557 drv->bdrv_child_perm(bs, c, c->role,
1558 cumulative_perms, cumulative_shared_perms,
1559 &cur_perm, &cur_shared);
1560 bdrv_child_set_perm(c, cur_perm, cur_shared);
1564 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1565 uint64_t *shared_perm)
1567 BdrvChild *c;
1568 uint64_t cumulative_perms = 0;
1569 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1571 QLIST_FOREACH(c, &bs->parents, next_parent) {
1572 cumulative_perms |= c->perm;
1573 cumulative_shared_perms &= c->shared_perm;
1576 *perm = cumulative_perms;
1577 *shared_perm = cumulative_shared_perms;
1580 static char *bdrv_child_user_desc(BdrvChild *c)
1582 if (c->role->get_parent_desc) {
1583 return c->role->get_parent_desc(c);
1586 return g_strdup("another user");
1589 static char *bdrv_perm_names(uint64_t perm)
1591 struct perm_name {
1592 uint64_t perm;
1593 const char *name;
1594 } permissions[] = {
1595 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1596 { BLK_PERM_WRITE, "write" },
1597 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1598 { BLK_PERM_RESIZE, "resize" },
1599 { BLK_PERM_GRAPH_MOD, "change children" },
1600 { 0, NULL }
1603 char *result = g_strdup("");
1604 struct perm_name *p;
1606 for (p = permissions; p->name; p++) {
1607 if (perm & p->perm) {
1608 char *old = result;
1609 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1610 g_free(old);
1614 return result;
1618 * Checks whether a new reference to @bs can be added if the new user requires
1619 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1620 * set, the BdrvChild objects in this list are ignored in the calculations;
1621 * this allows checking permission updates for an existing reference.
1623 * Needs to be followed by a call to either bdrv_set_perm() or
1624 * bdrv_abort_perm_update(). */
1625 static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm,
1626 uint64_t new_shared_perm,
1627 GSList *ignore_children, Error **errp)
1629 BdrvChild *c;
1630 uint64_t cumulative_perms = new_used_perm;
1631 uint64_t cumulative_shared_perms = new_shared_perm;
1633 /* There is no reason why anyone couldn't tolerate write_unchanged */
1634 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1636 QLIST_FOREACH(c, &bs->parents, next_parent) {
1637 if (g_slist_find(ignore_children, c)) {
1638 continue;
1641 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1642 char *user = bdrv_child_user_desc(c);
1643 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1644 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1645 "allow '%s' on %s",
1646 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1647 g_free(user);
1648 g_free(perm_names);
1649 return -EPERM;
1652 if ((c->perm & new_shared_perm) != c->perm) {
1653 char *user = bdrv_child_user_desc(c);
1654 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1655 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1656 "'%s' on %s",
1657 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1658 g_free(user);
1659 g_free(perm_names);
1660 return -EPERM;
1663 cumulative_perms |= c->perm;
1664 cumulative_shared_perms &= c->shared_perm;
1667 return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms,
1668 ignore_children, errp);
1671 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1672 * bdrv_child_abort_perm_update(). */
1673 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1674 GSList *ignore_children, Error **errp)
1676 int ret;
1678 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1679 ret = bdrv_check_update_perm(c->bs, perm, shared, ignore_children, errp);
1680 g_slist_free(ignore_children);
1682 return ret;
1685 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1687 uint64_t cumulative_perms, cumulative_shared_perms;
1689 c->perm = perm;
1690 c->shared_perm = shared;
1692 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1693 &cumulative_shared_perms);
1694 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1697 static void bdrv_child_abort_perm_update(BdrvChild *c)
1699 bdrv_abort_perm_update(c->bs);
1702 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1703 Error **errp)
1705 int ret;
1707 ret = bdrv_child_check_perm(c, perm, shared, NULL, errp);
1708 if (ret < 0) {
1709 bdrv_child_abort_perm_update(c);
1710 return ret;
1713 bdrv_child_set_perm(c, perm, shared);
1715 return 0;
1718 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1719 | BLK_PERM_WRITE \
1720 | BLK_PERM_WRITE_UNCHANGED \
1721 | BLK_PERM_RESIZE)
1722 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1724 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1725 const BdrvChildRole *role,
1726 uint64_t perm, uint64_t shared,
1727 uint64_t *nperm, uint64_t *nshared)
1729 if (c == NULL) {
1730 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1731 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1732 return;
1735 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1736 (c->perm & DEFAULT_PERM_UNCHANGED);
1737 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1738 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1741 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1742 const BdrvChildRole *role,
1743 uint64_t perm, uint64_t shared,
1744 uint64_t *nperm, uint64_t *nshared)
1746 bool backing = (role == &child_backing);
1747 assert(role == &child_backing || role == &child_file);
1749 if (!backing) {
1750 /* Apart from the modifications below, the same permissions are
1751 * forwarded and left alone as for filters */
1752 bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared);
1754 /* Format drivers may touch metadata even if the guest doesn't write */
1755 if (!bdrv_is_read_only(bs)) {
1756 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1759 /* bs->file always needs to be consistent because of the metadata. We
1760 * can never allow other users to resize or write to it. */
1761 perm |= BLK_PERM_CONSISTENT_READ;
1762 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1763 } else {
1764 /* We want consistent read from backing files if the parent needs it.
1765 * No other operations are performed on backing files. */
1766 perm &= BLK_PERM_CONSISTENT_READ;
1768 /* If the parent can deal with changing data, we're okay with a
1769 * writable and resizable backing file. */
1770 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1771 if (shared & BLK_PERM_WRITE) {
1772 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1773 } else {
1774 shared = 0;
1777 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1778 BLK_PERM_WRITE_UNCHANGED;
1781 *nperm = perm;
1782 *nshared = shared;
1785 static void bdrv_replace_child_noperm(BdrvChild *child,
1786 BlockDriverState *new_bs)
1788 BlockDriverState *old_bs = child->bs;
1790 if (old_bs && new_bs) {
1791 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1793 if (old_bs) {
1794 if (old_bs->quiesce_counter && child->role->drained_end) {
1795 child->role->drained_end(child);
1797 if (child->role->detach) {
1798 child->role->detach(child);
1800 QLIST_REMOVE(child, next_parent);
1803 child->bs = new_bs;
1805 if (new_bs) {
1806 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1807 if (new_bs->quiesce_counter && child->role->drained_begin) {
1808 child->role->drained_begin(child);
1811 if (child->role->attach) {
1812 child->role->attach(child);
1818 * Updates @child to change its reference to point to @new_bs, including
1819 * checking and applying the necessary permisson updates both to the old node
1820 * and to @new_bs.
1822 * NULL is passed as @new_bs for removing the reference before freeing @child.
1824 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1825 * function uses bdrv_set_perm() to update the permissions according to the new
1826 * reference that @new_bs gets.
1828 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1830 BlockDriverState *old_bs = child->bs;
1831 uint64_t perm, shared_perm;
1833 if (old_bs) {
1834 /* Update permissions for old node. This is guaranteed to succeed
1835 * because we're just taking a parent away, so we're loosening
1836 * restrictions. */
1837 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1838 bdrv_check_perm(old_bs, perm, shared_perm, NULL, &error_abort);
1839 bdrv_set_perm(old_bs, perm, shared_perm);
1842 bdrv_replace_child_noperm(child, new_bs);
1844 if (new_bs) {
1845 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
1846 bdrv_set_perm(new_bs, perm, shared_perm);
1850 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1851 const char *child_name,
1852 const BdrvChildRole *child_role,
1853 uint64_t perm, uint64_t shared_perm,
1854 void *opaque, Error **errp)
1856 BdrvChild *child;
1857 int ret;
1859 ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp);
1860 if (ret < 0) {
1861 bdrv_abort_perm_update(child_bs);
1862 return NULL;
1865 child = g_new(BdrvChild, 1);
1866 *child = (BdrvChild) {
1867 .bs = NULL,
1868 .name = g_strdup(child_name),
1869 .role = child_role,
1870 .perm = perm,
1871 .shared_perm = shared_perm,
1872 .opaque = opaque,
1875 /* This performs the matching bdrv_set_perm() for the above check. */
1876 bdrv_replace_child(child, child_bs);
1878 return child;
1881 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1882 BlockDriverState *child_bs,
1883 const char *child_name,
1884 const BdrvChildRole *child_role,
1885 Error **errp)
1887 BdrvChild *child;
1888 uint64_t perm, shared_perm;
1890 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
1892 assert(parent_bs->drv);
1893 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
1894 parent_bs->drv->bdrv_child_perm(parent_bs, NULL, child_role,
1895 perm, shared_perm, &perm, &shared_perm);
1897 child = bdrv_root_attach_child(child_bs, child_name, child_role,
1898 perm, shared_perm, parent_bs, errp);
1899 if (child == NULL) {
1900 return NULL;
1903 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1904 return child;
1907 static void bdrv_detach_child(BdrvChild *child)
1909 if (child->next.le_prev) {
1910 QLIST_REMOVE(child, next);
1911 child->next.le_prev = NULL;
1914 bdrv_replace_child(child, NULL);
1916 g_free(child->name);
1917 g_free(child);
1920 void bdrv_root_unref_child(BdrvChild *child)
1922 BlockDriverState *child_bs;
1924 child_bs = child->bs;
1925 bdrv_detach_child(child);
1926 bdrv_unref(child_bs);
1929 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1931 if (child == NULL) {
1932 return;
1935 if (child->bs->inherits_from == parent) {
1936 BdrvChild *c;
1938 /* Remove inherits_from only when the last reference between parent and
1939 * child->bs goes away. */
1940 QLIST_FOREACH(c, &parent->children, next) {
1941 if (c != child && c->bs == child->bs) {
1942 break;
1945 if (c == NULL) {
1946 child->bs->inherits_from = NULL;
1950 bdrv_root_unref_child(child);
1954 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1956 BdrvChild *c;
1957 QLIST_FOREACH(c, &bs->parents, next_parent) {
1958 if (c->role->change_media) {
1959 c->role->change_media(c, load);
1964 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1966 BdrvChild *c;
1967 QLIST_FOREACH(c, &bs->parents, next_parent) {
1968 if (c->role->resize) {
1969 c->role->resize(c);
1975 * Sets the backing file link of a BDS. A new reference is created; callers
1976 * which don't need their own reference any more must call bdrv_unref().
1978 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
1979 Error **errp)
1981 if (backing_hd) {
1982 bdrv_ref(backing_hd);
1985 if (bs->backing) {
1986 bdrv_unref_child(bs, bs->backing);
1989 if (!backing_hd) {
1990 bs->backing = NULL;
1991 goto out;
1994 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
1995 errp);
1996 if (!bs->backing) {
1997 bdrv_unref(backing_hd);
2000 bdrv_refresh_filename(bs);
2002 out:
2003 bdrv_refresh_limits(bs, NULL);
2007 * Opens the backing file for a BlockDriverState if not yet open
2009 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2010 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2011 * itself, all options starting with "${bdref_key}." are considered part of the
2012 * BlockdevRef.
2014 * TODO Can this be unified with bdrv_open_image()?
2016 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2017 const char *bdref_key, Error **errp)
2019 char *backing_filename = g_malloc0(PATH_MAX);
2020 char *bdref_key_dot;
2021 const char *reference = NULL;
2022 int ret = 0;
2023 BlockDriverState *backing_hd;
2024 QDict *options;
2025 QDict *tmp_parent_options = NULL;
2026 Error *local_err = NULL;
2028 if (bs->backing != NULL) {
2029 goto free_exit;
2032 /* NULL means an empty set of options */
2033 if (parent_options == NULL) {
2034 tmp_parent_options = qdict_new();
2035 parent_options = tmp_parent_options;
2038 bs->open_flags &= ~BDRV_O_NO_BACKING;
2040 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2041 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2042 g_free(bdref_key_dot);
2045 * Caution: while qdict_get_try_str() is fine, getting non-string
2046 * types would require more care. When @parent_options come from
2047 * -blockdev or blockdev_add, its members are typed according to
2048 * the QAPI schema, but when they come from -drive, they're all
2049 * QString.
2051 reference = qdict_get_try_str(parent_options, bdref_key);
2052 if (reference || qdict_haskey(options, "file.filename")) {
2053 backing_filename[0] = '\0';
2054 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2055 QDECREF(options);
2056 goto free_exit;
2057 } else {
2058 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2059 &local_err);
2060 if (local_err) {
2061 ret = -EINVAL;
2062 error_propagate(errp, local_err);
2063 QDECREF(options);
2064 goto free_exit;
2068 if (!bs->drv || !bs->drv->supports_backing) {
2069 ret = -EINVAL;
2070 error_setg(errp, "Driver doesn't support backing files");
2071 QDECREF(options);
2072 goto free_exit;
2075 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2076 qdict_put_str(options, "driver", bs->backing_format);
2079 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2080 reference, options, 0, bs, &child_backing,
2081 errp);
2082 if (!backing_hd) {
2083 bs->open_flags |= BDRV_O_NO_BACKING;
2084 error_prepend(errp, "Could not open backing file: ");
2085 ret = -EINVAL;
2086 goto free_exit;
2089 /* Hook up the backing file link; drop our reference, bs owns the
2090 * backing_hd reference now */
2091 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2092 bdrv_unref(backing_hd);
2093 if (local_err) {
2094 error_propagate(errp, local_err);
2095 ret = -EINVAL;
2096 goto free_exit;
2099 qdict_del(parent_options, bdref_key);
2101 free_exit:
2102 g_free(backing_filename);
2103 QDECREF(tmp_parent_options);
2104 return ret;
2107 static BlockDriverState *
2108 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2109 BlockDriverState *parent, const BdrvChildRole *child_role,
2110 bool allow_none, Error **errp)
2112 BlockDriverState *bs = NULL;
2113 QDict *image_options;
2114 char *bdref_key_dot;
2115 const char *reference;
2117 assert(child_role != NULL);
2119 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2120 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2121 g_free(bdref_key_dot);
2124 * Caution: while qdict_get_try_str() is fine, getting non-string
2125 * types would require more care. When @options come from
2126 * -blockdev or blockdev_add, its members are typed according to
2127 * the QAPI schema, but when they come from -drive, they're all
2128 * QString.
2130 reference = qdict_get_try_str(options, bdref_key);
2131 if (!filename && !reference && !qdict_size(image_options)) {
2132 if (!allow_none) {
2133 error_setg(errp, "A block device must be specified for \"%s\"",
2134 bdref_key);
2136 QDECREF(image_options);
2137 goto done;
2140 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2141 parent, child_role, errp);
2142 if (!bs) {
2143 goto done;
2146 done:
2147 qdict_del(options, bdref_key);
2148 return bs;
2152 * Opens a disk image whose options are given as BlockdevRef in another block
2153 * device's options.
2155 * If allow_none is true, no image will be opened if filename is false and no
2156 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2158 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2159 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2160 * itself, all options starting with "${bdref_key}." are considered part of the
2161 * BlockdevRef.
2163 * The BlockdevRef will be removed from the options QDict.
2165 BdrvChild *bdrv_open_child(const char *filename,
2166 QDict *options, const char *bdref_key,
2167 BlockDriverState *parent,
2168 const BdrvChildRole *child_role,
2169 bool allow_none, Error **errp)
2171 BdrvChild *c;
2172 BlockDriverState *bs;
2174 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2175 allow_none, errp);
2176 if (bs == NULL) {
2177 return NULL;
2180 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2181 if (!c) {
2182 bdrv_unref(bs);
2183 return NULL;
2186 return c;
2189 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2190 int flags,
2191 QDict *snapshot_options,
2192 Error **errp)
2194 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2195 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2196 int64_t total_size;
2197 QemuOpts *opts = NULL;
2198 BlockDriverState *bs_snapshot = NULL;
2199 Error *local_err = NULL;
2200 int ret;
2202 /* if snapshot, we create a temporary backing file and open it
2203 instead of opening 'filename' directly */
2205 /* Get the required size from the image */
2206 total_size = bdrv_getlength(bs);
2207 if (total_size < 0) {
2208 error_setg_errno(errp, -total_size, "Could not get image size");
2209 goto out;
2212 /* Create the temporary image */
2213 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2214 if (ret < 0) {
2215 error_setg_errno(errp, -ret, "Could not get temporary filename");
2216 goto out;
2219 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2220 &error_abort);
2221 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2222 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2223 qemu_opts_del(opts);
2224 if (ret < 0) {
2225 error_prepend(errp, "Could not create temporary overlay '%s': ",
2226 tmp_filename);
2227 goto out;
2230 /* Prepare options QDict for the temporary file */
2231 qdict_put_str(snapshot_options, "file.driver", "file");
2232 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2233 qdict_put_str(snapshot_options, "driver", "qcow2");
2235 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2236 snapshot_options = NULL;
2237 if (!bs_snapshot) {
2238 goto out;
2241 /* bdrv_append() consumes a strong reference to bs_snapshot
2242 * (i.e. it will call bdrv_unref() on it) even on error, so in
2243 * order to be able to return one, we have to increase
2244 * bs_snapshot's refcount here */
2245 bdrv_ref(bs_snapshot);
2246 bdrv_append(bs_snapshot, bs, &local_err);
2247 if (local_err) {
2248 error_propagate(errp, local_err);
2249 bs_snapshot = NULL;
2250 goto out;
2253 out:
2254 QDECREF(snapshot_options);
2255 g_free(tmp_filename);
2256 return bs_snapshot;
2260 * Opens a disk image (raw, qcow2, vmdk, ...)
2262 * options is a QDict of options to pass to the block drivers, or NULL for an
2263 * empty set of options. The reference to the QDict belongs to the block layer
2264 * after the call (even on failure), so if the caller intends to reuse the
2265 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2267 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2268 * If it is not NULL, the referenced BDS will be reused.
2270 * The reference parameter may be used to specify an existing block device which
2271 * should be opened. If specified, neither options nor a filename may be given,
2272 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2274 static BlockDriverState *bdrv_open_inherit(const char *filename,
2275 const char *reference,
2276 QDict *options, int flags,
2277 BlockDriverState *parent,
2278 const BdrvChildRole *child_role,
2279 Error **errp)
2281 int ret;
2282 BlockBackend *file = NULL;
2283 BlockDriverState *bs;
2284 BlockDriver *drv = NULL;
2285 const char *drvname;
2286 const char *backing;
2287 Error *local_err = NULL;
2288 QDict *snapshot_options = NULL;
2289 int snapshot_flags = 0;
2291 assert(!child_role || !flags);
2292 assert(!child_role == !parent);
2294 if (reference) {
2295 bool options_non_empty = options ? qdict_size(options) : false;
2296 QDECREF(options);
2298 if (filename || options_non_empty) {
2299 error_setg(errp, "Cannot reference an existing block device with "
2300 "additional options or a new filename");
2301 return NULL;
2304 bs = bdrv_lookup_bs(reference, reference, errp);
2305 if (!bs) {
2306 return NULL;
2309 bdrv_ref(bs);
2310 return bs;
2313 bs = bdrv_new();
2315 /* NULL means an empty set of options */
2316 if (options == NULL) {
2317 options = qdict_new();
2320 /* json: syntax counts as explicit options, as if in the QDict */
2321 parse_json_protocol(options, &filename, &local_err);
2322 if (local_err) {
2323 goto fail;
2326 bs->explicit_options = qdict_clone_shallow(options);
2328 if (child_role) {
2329 bs->inherits_from = parent;
2330 child_role->inherit_options(&flags, options,
2331 parent->open_flags, parent->options);
2334 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2335 if (local_err) {
2336 goto fail;
2340 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2341 * Caution: getting a boolean member of @options requires care.
2342 * When @options come from -blockdev or blockdev_add, members are
2343 * typed according to the QAPI schema, but when they come from
2344 * -drive, they're all QString.
2346 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2347 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2348 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2349 } else {
2350 flags &= ~BDRV_O_RDWR;
2353 if (flags & BDRV_O_SNAPSHOT) {
2354 snapshot_options = qdict_new();
2355 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2356 flags, options);
2357 /* Let bdrv_backing_options() override "read-only" */
2358 qdict_del(options, BDRV_OPT_READ_ONLY);
2359 bdrv_backing_options(&flags, options, flags, options);
2362 bs->open_flags = flags;
2363 bs->options = options;
2364 options = qdict_clone_shallow(options);
2366 /* Find the right image format driver */
2367 /* See cautionary note on accessing @options above */
2368 drvname = qdict_get_try_str(options, "driver");
2369 if (drvname) {
2370 drv = bdrv_find_format(drvname);
2371 if (!drv) {
2372 error_setg(errp, "Unknown driver: '%s'", drvname);
2373 goto fail;
2377 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2379 /* See cautionary note on accessing @options above */
2380 backing = qdict_get_try_str(options, "backing");
2381 if (backing && *backing == '\0') {
2382 flags |= BDRV_O_NO_BACKING;
2383 qdict_del(options, "backing");
2386 /* Open image file without format layer. This BlockBackend is only used for
2387 * probing, the block drivers will do their own bdrv_open_child() for the
2388 * same BDS, which is why we put the node name back into options. */
2389 if ((flags & BDRV_O_PROTOCOL) == 0) {
2390 BlockDriverState *file_bs;
2392 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2393 &child_file, true, &local_err);
2394 if (local_err) {
2395 goto fail;
2397 if (file_bs != NULL) {
2398 file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
2399 blk_insert_bs(file, file_bs, &local_err);
2400 bdrv_unref(file_bs);
2401 if (local_err) {
2402 goto fail;
2405 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2409 /* Image format probing */
2410 bs->probed = !drv;
2411 if (!drv && file) {
2412 ret = find_image_format(file, filename, &drv, &local_err);
2413 if (ret < 0) {
2414 goto fail;
2417 * This option update would logically belong in bdrv_fill_options(),
2418 * but we first need to open bs->file for the probing to work, while
2419 * opening bs->file already requires the (mostly) final set of options
2420 * so that cache mode etc. can be inherited.
2422 * Adding the driver later is somewhat ugly, but it's not an option
2423 * that would ever be inherited, so it's correct. We just need to make
2424 * sure to update both bs->options (which has the full effective
2425 * options for bs) and options (which has file.* already removed).
2427 qdict_put_str(bs->options, "driver", drv->format_name);
2428 qdict_put_str(options, "driver", drv->format_name);
2429 } else if (!drv) {
2430 error_setg(errp, "Must specify either driver or file");
2431 goto fail;
2434 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2435 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2436 /* file must be NULL if a protocol BDS is about to be created
2437 * (the inverse results in an error message from bdrv_open_common()) */
2438 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2440 /* Open the image */
2441 ret = bdrv_open_common(bs, file, options, &local_err);
2442 if (ret < 0) {
2443 goto fail;
2446 if (file) {
2447 blk_unref(file);
2448 file = NULL;
2451 /* If there is a backing file, use it */
2452 if ((flags & BDRV_O_NO_BACKING) == 0) {
2453 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2454 if (ret < 0) {
2455 goto close_and_fail;
2459 bdrv_refresh_filename(bs);
2461 /* Check if any unknown options were used */
2462 if (qdict_size(options) != 0) {
2463 const QDictEntry *entry = qdict_first(options);
2464 if (flags & BDRV_O_PROTOCOL) {
2465 error_setg(errp, "Block protocol '%s' doesn't support the option "
2466 "'%s'", drv->format_name, entry->key);
2467 } else {
2468 error_setg(errp,
2469 "Block format '%s' does not support the option '%s'",
2470 drv->format_name, entry->key);
2473 goto close_and_fail;
2476 if (!bdrv_key_required(bs)) {
2477 bdrv_parent_cb_change_media(bs, true);
2478 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
2479 && !runstate_check(RUN_STATE_INMIGRATE)
2480 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
2481 error_setg(errp,
2482 "Guest must be stopped for opening of encrypted image");
2483 goto close_and_fail;
2486 QDECREF(options);
2488 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2489 * temporary snapshot afterwards. */
2490 if (snapshot_flags) {
2491 BlockDriverState *snapshot_bs;
2492 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2493 snapshot_options, &local_err);
2494 snapshot_options = NULL;
2495 if (local_err) {
2496 goto close_and_fail;
2498 /* We are not going to return bs but the overlay on top of it
2499 * (snapshot_bs); thus, we have to drop the strong reference to bs
2500 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2501 * though, because the overlay still has a reference to it. */
2502 bdrv_unref(bs);
2503 bs = snapshot_bs;
2506 return bs;
2508 fail:
2509 blk_unref(file);
2510 if (bs->file != NULL) {
2511 bdrv_unref_child(bs, bs->file);
2513 QDECREF(snapshot_options);
2514 QDECREF(bs->explicit_options);
2515 QDECREF(bs->options);
2516 QDECREF(options);
2517 bs->options = NULL;
2518 bdrv_unref(bs);
2519 error_propagate(errp, local_err);
2520 return NULL;
2522 close_and_fail:
2523 bdrv_unref(bs);
2524 QDECREF(snapshot_options);
2525 QDECREF(options);
2526 error_propagate(errp, local_err);
2527 return NULL;
2530 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2531 QDict *options, int flags, Error **errp)
2533 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2534 NULL, errp);
2537 typedef struct BlockReopenQueueEntry {
2538 bool prepared;
2539 BDRVReopenState state;
2540 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2541 } BlockReopenQueueEntry;
2544 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2545 * reopen of multiple devices.
2547 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2548 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2549 * be created and initialized. This newly created BlockReopenQueue should be
2550 * passed back in for subsequent calls that are intended to be of the same
2551 * atomic 'set'.
2553 * bs is the BlockDriverState to add to the reopen queue.
2555 * options contains the changed options for the associated bs
2556 * (the BlockReopenQueue takes ownership)
2558 * flags contains the open flags for the associated bs
2560 * returns a pointer to bs_queue, which is either the newly allocated
2561 * bs_queue, or the existing bs_queue being used.
2564 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2565 BlockDriverState *bs,
2566 QDict *options,
2567 int flags,
2568 const BdrvChildRole *role,
2569 QDict *parent_options,
2570 int parent_flags)
2572 assert(bs != NULL);
2574 BlockReopenQueueEntry *bs_entry;
2575 BdrvChild *child;
2576 QDict *old_options, *explicit_options;
2578 if (bs_queue == NULL) {
2579 bs_queue = g_new0(BlockReopenQueue, 1);
2580 QSIMPLEQ_INIT(bs_queue);
2583 if (!options) {
2584 options = qdict_new();
2587 /* Check if this BlockDriverState is already in the queue */
2588 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2589 if (bs == bs_entry->state.bs) {
2590 break;
2595 * Precedence of options:
2596 * 1. Explicitly passed in options (highest)
2597 * 2. Set in flags (only for top level)
2598 * 3. Retained from explicitly set options of bs
2599 * 4. Inherited from parent node
2600 * 5. Retained from effective options of bs
2603 if (!parent_options) {
2605 * Any setting represented by flags is always updated. If the
2606 * corresponding QDict option is set, it takes precedence. Otherwise
2607 * the flag is translated into a QDict option. The old setting of bs is
2608 * not considered.
2610 update_options_from_flags(options, flags);
2613 /* Old explicitly set values (don't overwrite by inherited value) */
2614 if (bs_entry) {
2615 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2616 } else {
2617 old_options = qdict_clone_shallow(bs->explicit_options);
2619 bdrv_join_options(bs, options, old_options);
2620 QDECREF(old_options);
2622 explicit_options = qdict_clone_shallow(options);
2624 /* Inherit from parent node */
2625 if (parent_options) {
2626 assert(!flags);
2627 role->inherit_options(&flags, options, parent_flags, parent_options);
2630 /* Old values are used for options that aren't set yet */
2631 old_options = qdict_clone_shallow(bs->options);
2632 bdrv_join_options(bs, options, old_options);
2633 QDECREF(old_options);
2635 /* bdrv_open() masks this flag out */
2636 flags &= ~BDRV_O_PROTOCOL;
2638 QLIST_FOREACH(child, &bs->children, next) {
2639 QDict *new_child_options;
2640 char *child_key_dot;
2642 /* reopen can only change the options of block devices that were
2643 * implicitly created and inherited options. For other (referenced)
2644 * block devices, a syntax like "backing.foo" results in an error. */
2645 if (child->bs->inherits_from != bs) {
2646 continue;
2649 child_key_dot = g_strdup_printf("%s.", child->name);
2650 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2651 g_free(child_key_dot);
2653 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2654 child->role, options, flags);
2657 if (!bs_entry) {
2658 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2659 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2660 } else {
2661 QDECREF(bs_entry->state.options);
2662 QDECREF(bs_entry->state.explicit_options);
2665 bs_entry->state.bs = bs;
2666 bs_entry->state.options = options;
2667 bs_entry->state.explicit_options = explicit_options;
2668 bs_entry->state.flags = flags;
2670 return bs_queue;
2673 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2674 BlockDriverState *bs,
2675 QDict *options, int flags)
2677 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2678 NULL, NULL, 0);
2682 * Reopen multiple BlockDriverStates atomically & transactionally.
2684 * The queue passed in (bs_queue) must have been built up previous
2685 * via bdrv_reopen_queue().
2687 * Reopens all BDS specified in the queue, with the appropriate
2688 * flags. All devices are prepared for reopen, and failure of any
2689 * device will cause all device changes to be abandonded, and intermediate
2690 * data cleaned up.
2692 * If all devices prepare successfully, then the changes are committed
2693 * to all devices.
2696 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2698 int ret = -1;
2699 BlockReopenQueueEntry *bs_entry, *next;
2700 Error *local_err = NULL;
2702 assert(bs_queue != NULL);
2704 aio_context_release(ctx);
2705 bdrv_drain_all_begin();
2706 aio_context_acquire(ctx);
2708 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2709 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2710 error_propagate(errp, local_err);
2711 goto cleanup;
2713 bs_entry->prepared = true;
2716 /* If we reach this point, we have success and just need to apply the
2717 * changes
2719 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2720 bdrv_reopen_commit(&bs_entry->state);
2723 ret = 0;
2725 cleanup:
2726 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2727 if (ret && bs_entry->prepared) {
2728 bdrv_reopen_abort(&bs_entry->state);
2729 } else if (ret) {
2730 QDECREF(bs_entry->state.explicit_options);
2732 QDECREF(bs_entry->state.options);
2733 g_free(bs_entry);
2735 g_free(bs_queue);
2737 bdrv_drain_all_end();
2739 return ret;
2743 /* Reopen a single BlockDriverState with the specified flags. */
2744 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2746 int ret = -1;
2747 Error *local_err = NULL;
2748 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2750 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2751 if (local_err != NULL) {
2752 error_propagate(errp, local_err);
2754 return ret;
2759 * Prepares a BlockDriverState for reopen. All changes are staged in the
2760 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2761 * the block driver layer .bdrv_reopen_prepare()
2763 * bs is the BlockDriverState to reopen
2764 * flags are the new open flags
2765 * queue is the reopen queue
2767 * Returns 0 on success, non-zero on error. On error errp will be set
2768 * as well.
2770 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2771 * It is the responsibility of the caller to then call the abort() or
2772 * commit() for any other BDS that have been left in a prepare() state
2775 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2776 Error **errp)
2778 int ret = -1;
2779 Error *local_err = NULL;
2780 BlockDriver *drv;
2781 QemuOpts *opts;
2782 const char *value;
2783 bool read_only;
2785 assert(reopen_state != NULL);
2786 assert(reopen_state->bs->drv != NULL);
2787 drv = reopen_state->bs->drv;
2789 /* Process generic block layer options */
2790 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2791 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2792 if (local_err) {
2793 error_propagate(errp, local_err);
2794 ret = -EINVAL;
2795 goto error;
2798 update_flags_from_options(&reopen_state->flags, opts);
2800 /* node-name and driver must be unchanged. Put them back into the QDict, so
2801 * that they are checked at the end of this function. */
2802 value = qemu_opt_get(opts, "node-name");
2803 if (value) {
2804 qdict_put_str(reopen_state->options, "node-name", value);
2807 value = qemu_opt_get(opts, "driver");
2808 if (value) {
2809 qdict_put_str(reopen_state->options, "driver", value);
2812 /* If we are to stay read-only, do not allow permission change
2813 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
2814 * not set, or if the BDS still has copy_on_read enabled */
2815 read_only = !(reopen_state->flags & BDRV_O_RDWR);
2816 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, &local_err);
2817 if (local_err) {
2818 error_propagate(errp, local_err);
2819 goto error;
2823 ret = bdrv_flush(reopen_state->bs);
2824 if (ret) {
2825 error_setg_errno(errp, -ret, "Error flushing drive");
2826 goto error;
2829 if (drv->bdrv_reopen_prepare) {
2830 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2831 if (ret) {
2832 if (local_err != NULL) {
2833 error_propagate(errp, local_err);
2834 } else {
2835 error_setg(errp, "failed while preparing to reopen image '%s'",
2836 reopen_state->bs->filename);
2838 goto error;
2840 } else {
2841 /* It is currently mandatory to have a bdrv_reopen_prepare()
2842 * handler for each supported drv. */
2843 error_setg(errp, "Block format '%s' used by node '%s' "
2844 "does not support reopening files", drv->format_name,
2845 bdrv_get_device_or_node_name(reopen_state->bs));
2846 ret = -1;
2847 goto error;
2850 /* Options that are not handled are only okay if they are unchanged
2851 * compared to the old state. It is expected that some options are only
2852 * used for the initial open, but not reopen (e.g. filename) */
2853 if (qdict_size(reopen_state->options)) {
2854 const QDictEntry *entry = qdict_first(reopen_state->options);
2856 do {
2857 QString *new_obj = qobject_to_qstring(entry->value);
2858 const char *new = qstring_get_str(new_obj);
2860 * Caution: while qdict_get_try_str() is fine, getting
2861 * non-string types would require more care. When
2862 * bs->options come from -blockdev or blockdev_add, its
2863 * members are typed according to the QAPI schema, but
2864 * when they come from -drive, they're all QString.
2866 const char *old = qdict_get_try_str(reopen_state->bs->options,
2867 entry->key);
2869 if (!old || strcmp(new, old)) {
2870 error_setg(errp, "Cannot change the option '%s'", entry->key);
2871 ret = -EINVAL;
2872 goto error;
2874 } while ((entry = qdict_next(reopen_state->options, entry)));
2877 ret = 0;
2879 error:
2880 qemu_opts_del(opts);
2881 return ret;
2885 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2886 * makes them final by swapping the staging BlockDriverState contents into
2887 * the active BlockDriverState contents.
2889 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2891 BlockDriver *drv;
2893 assert(reopen_state != NULL);
2894 drv = reopen_state->bs->drv;
2895 assert(drv != NULL);
2897 /* If there are any driver level actions to take */
2898 if (drv->bdrv_reopen_commit) {
2899 drv->bdrv_reopen_commit(reopen_state);
2902 /* set BDS specific flags now */
2903 QDECREF(reopen_state->bs->explicit_options);
2905 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2906 reopen_state->bs->open_flags = reopen_state->flags;
2907 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2909 bdrv_refresh_limits(reopen_state->bs, NULL);
2913 * Abort the reopen, and delete and free the staged changes in
2914 * reopen_state
2916 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2918 BlockDriver *drv;
2920 assert(reopen_state != NULL);
2921 drv = reopen_state->bs->drv;
2922 assert(drv != NULL);
2924 if (drv->bdrv_reopen_abort) {
2925 drv->bdrv_reopen_abort(reopen_state);
2928 QDECREF(reopen_state->explicit_options);
2932 static void bdrv_close(BlockDriverState *bs)
2934 BdrvAioNotifier *ban, *ban_next;
2936 assert(!bs->job);
2937 assert(!bs->refcnt);
2939 bdrv_drained_begin(bs); /* complete I/O */
2940 bdrv_flush(bs);
2941 bdrv_drain(bs); /* in case flush left pending I/O */
2943 bdrv_release_named_dirty_bitmaps(bs);
2944 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2946 if (bs->drv) {
2947 BdrvChild *child, *next;
2949 bs->drv->bdrv_close(bs);
2950 bs->drv = NULL;
2952 bdrv_set_backing_hd(bs, NULL, &error_abort);
2954 if (bs->file != NULL) {
2955 bdrv_unref_child(bs, bs->file);
2956 bs->file = NULL;
2959 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2960 /* TODO Remove bdrv_unref() from drivers' close function and use
2961 * bdrv_unref_child() here */
2962 if (child->bs->inherits_from == bs) {
2963 child->bs->inherits_from = NULL;
2965 bdrv_detach_child(child);
2968 g_free(bs->opaque);
2969 bs->opaque = NULL;
2970 bs->copy_on_read = 0;
2971 bs->backing_file[0] = '\0';
2972 bs->backing_format[0] = '\0';
2973 bs->total_sectors = 0;
2974 bs->encrypted = false;
2975 bs->valid_key = false;
2976 bs->sg = false;
2977 QDECREF(bs->options);
2978 QDECREF(bs->explicit_options);
2979 bs->options = NULL;
2980 QDECREF(bs->full_open_options);
2981 bs->full_open_options = NULL;
2984 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2985 g_free(ban);
2987 QLIST_INIT(&bs->aio_notifiers);
2988 bdrv_drained_end(bs);
2991 void bdrv_close_all(void)
2993 block_job_cancel_sync_all();
2994 nbd_export_close_all();
2996 /* Drop references from requests still in flight, such as canceled block
2997 * jobs whose AIO context has not been polled yet */
2998 bdrv_drain_all();
3000 blk_remove_all_bs();
3001 blockdev_close_all_bdrv_states();
3003 assert(QTAILQ_EMPTY(&all_bdrv_states));
3006 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3008 BdrvChild *to_c;
3010 if (c->role->stay_at_node) {
3011 return false;
3014 if (c->role == &child_backing) {
3015 /* If @from is a backing file of @to, ignore the child to avoid
3016 * creating a loop. We only want to change the pointer of other
3017 * parents. */
3018 QLIST_FOREACH(to_c, &to->children, next) {
3019 if (to_c == c) {
3020 break;
3023 if (to_c) {
3024 return false;
3028 return true;
3031 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3032 Error **errp)
3034 BdrvChild *c, *next;
3035 GSList *list = NULL, *p;
3036 uint64_t old_perm, old_shared;
3037 uint64_t perm = 0, shared = BLK_PERM_ALL;
3038 int ret;
3040 assert(!atomic_read(&from->in_flight));
3041 assert(!atomic_read(&to->in_flight));
3043 /* Make sure that @from doesn't go away until we have successfully attached
3044 * all of its parents to @to. */
3045 bdrv_ref(from);
3047 /* Put all parents into @list and calculate their cumulative permissions */
3048 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3049 if (!should_update_child(c, to)) {
3050 continue;
3052 list = g_slist_prepend(list, c);
3053 perm |= c->perm;
3054 shared &= c->shared_perm;
3057 /* Check whether the required permissions can be granted on @to, ignoring
3058 * all BdrvChild in @list so that they can't block themselves. */
3059 ret = bdrv_check_update_perm(to, perm, shared, list, errp);
3060 if (ret < 0) {
3061 bdrv_abort_perm_update(to);
3062 goto out;
3065 /* Now actually perform the change. We performed the permission check for
3066 * all elements of @list at once, so set the permissions all at once at the
3067 * very end. */
3068 for (p = list; p != NULL; p = p->next) {
3069 c = p->data;
3071 bdrv_ref(to);
3072 bdrv_replace_child_noperm(c, to);
3073 bdrv_unref(from);
3076 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3077 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3079 out:
3080 g_slist_free(list);
3081 bdrv_unref(from);
3085 * Add new bs contents at the top of an image chain while the chain is
3086 * live, while keeping required fields on the top layer.
3088 * This will modify the BlockDriverState fields, and swap contents
3089 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3091 * bs_new must not be attached to a BlockBackend.
3093 * This function does not create any image files.
3095 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3096 * that's what the callers commonly need. bs_new will be referenced by the old
3097 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3098 * reference of its own, it must call bdrv_ref().
3100 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3101 Error **errp)
3103 Error *local_err = NULL;
3105 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3106 if (local_err) {
3107 error_propagate(errp, local_err);
3108 goto out;
3111 bdrv_replace_node(bs_top, bs_new, &local_err);
3112 if (local_err) {
3113 error_propagate(errp, local_err);
3114 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3115 goto out;
3118 /* bs_new is now referenced by its new parents, we don't need the
3119 * additional reference any more. */
3120 out:
3121 bdrv_unref(bs_new);
3124 static void bdrv_delete(BlockDriverState *bs)
3126 assert(!bs->job);
3127 assert(bdrv_op_blocker_is_empty(bs));
3128 assert(!bs->refcnt);
3130 bdrv_close(bs);
3132 /* remove from list, if necessary */
3133 if (bs->node_name[0] != '\0') {
3134 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3136 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3138 g_free(bs);
3142 * Run consistency checks on an image
3144 * Returns 0 if the check could be completed (it doesn't mean that the image is
3145 * free of errors) or -errno when an internal error occurred. The results of the
3146 * check are stored in res.
3148 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3150 if (bs->drv == NULL) {
3151 return -ENOMEDIUM;
3153 if (bs->drv->bdrv_check == NULL) {
3154 return -ENOTSUP;
3157 memset(res, 0, sizeof(*res));
3158 return bs->drv->bdrv_check(bs, res, fix);
3162 * Return values:
3163 * 0 - success
3164 * -EINVAL - backing format specified, but no file
3165 * -ENOSPC - can't update the backing file because no space is left in the
3166 * image file header
3167 * -ENOTSUP - format driver doesn't support changing the backing file
3169 int bdrv_change_backing_file(BlockDriverState *bs,
3170 const char *backing_file, const char *backing_fmt)
3172 BlockDriver *drv = bs->drv;
3173 int ret;
3175 /* Backing file format doesn't make sense without a backing file */
3176 if (backing_fmt && !backing_file) {
3177 return -EINVAL;
3180 if (drv->bdrv_change_backing_file != NULL) {
3181 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3182 } else {
3183 ret = -ENOTSUP;
3186 if (ret == 0) {
3187 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3188 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3190 return ret;
3194 * Finds the image layer in the chain that has 'bs' as its backing file.
3196 * active is the current topmost image.
3198 * Returns NULL if bs is not found in active's image chain,
3199 * or if active == bs.
3201 * Returns the bottommost base image if bs == NULL.
3203 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3204 BlockDriverState *bs)
3206 while (active && bs != backing_bs(active)) {
3207 active = backing_bs(active);
3210 return active;
3213 /* Given a BDS, searches for the base layer. */
3214 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3216 return bdrv_find_overlay(bs, NULL);
3220 * Drops images above 'base' up to and including 'top', and sets the image
3221 * above 'top' to have base as its backing file.
3223 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3224 * information in 'bs' can be properly updated.
3226 * E.g., this will convert the following chain:
3227 * bottom <- base <- intermediate <- top <- active
3229 * to
3231 * bottom <- base <- active
3233 * It is allowed for bottom==base, in which case it converts:
3235 * base <- intermediate <- top <- active
3237 * to
3239 * base <- active
3241 * If backing_file_str is non-NULL, it will be used when modifying top's
3242 * overlay image metadata.
3244 * Error conditions:
3245 * if active == top, that is considered an error
3248 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
3249 BlockDriverState *base, const char *backing_file_str)
3251 BlockDriverState *new_top_bs = NULL;
3252 Error *local_err = NULL;
3253 int ret = -EIO;
3255 if (!top->drv || !base->drv) {
3256 goto exit;
3259 new_top_bs = bdrv_find_overlay(active, top);
3261 if (new_top_bs == NULL) {
3262 /* we could not find the image above 'top', this is an error */
3263 goto exit;
3266 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3267 * to do, no intermediate images */
3268 if (backing_bs(new_top_bs) == base) {
3269 ret = 0;
3270 goto exit;
3273 /* Make sure that base is in the backing chain of top */
3274 if (!bdrv_chain_contains(top, base)) {
3275 goto exit;
3278 /* success - we can delete the intermediate states, and link top->base */
3279 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3280 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
3281 base->drv ? base->drv->format_name : "");
3282 if (ret) {
3283 goto exit;
3286 bdrv_set_backing_hd(new_top_bs, base, &local_err);
3287 if (local_err) {
3288 ret = -EPERM;
3289 error_report_err(local_err);
3290 goto exit;
3293 ret = 0;
3294 exit:
3295 return ret;
3299 * Truncate file to 'offset' bytes (needed only for file protocols)
3301 int bdrv_truncate(BdrvChild *child, int64_t offset, Error **errp)
3303 BlockDriverState *bs = child->bs;
3304 BlockDriver *drv = bs->drv;
3305 int ret;
3307 assert(child->perm & BLK_PERM_RESIZE);
3309 if (!drv) {
3310 error_setg(errp, "No medium inserted");
3311 return -ENOMEDIUM;
3313 if (!drv->bdrv_truncate) {
3314 error_setg(errp, "Image format driver does not support resize");
3315 return -ENOTSUP;
3317 if (bs->read_only) {
3318 error_setg(errp, "Image is read-only");
3319 return -EACCES;
3322 assert(!(bs->open_flags & BDRV_O_INACTIVE));
3324 ret = drv->bdrv_truncate(bs, offset, errp);
3325 if (ret == 0) {
3326 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3327 bdrv_dirty_bitmap_truncate(bs);
3328 bdrv_parent_cb_resize(bs);
3329 ++bs->write_gen;
3331 return ret;
3335 * Length of a allocated file in bytes. Sparse files are counted by actual
3336 * allocated space. Return < 0 if error or unknown.
3338 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3340 BlockDriver *drv = bs->drv;
3341 if (!drv) {
3342 return -ENOMEDIUM;
3344 if (drv->bdrv_get_allocated_file_size) {
3345 return drv->bdrv_get_allocated_file_size(bs);
3347 if (bs->file) {
3348 return bdrv_get_allocated_file_size(bs->file->bs);
3350 return -ENOTSUP;
3354 * Return number of sectors on success, -errno on error.
3356 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3358 BlockDriver *drv = bs->drv;
3360 if (!drv)
3361 return -ENOMEDIUM;
3363 if (drv->has_variable_length) {
3364 int ret = refresh_total_sectors(bs, bs->total_sectors);
3365 if (ret < 0) {
3366 return ret;
3369 return bs->total_sectors;
3373 * Return length in bytes on success, -errno on error.
3374 * The length is always a multiple of BDRV_SECTOR_SIZE.
3376 int64_t bdrv_getlength(BlockDriverState *bs)
3378 int64_t ret = bdrv_nb_sectors(bs);
3380 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3381 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3384 /* return 0 as number of sectors if no device present or error */
3385 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3387 int64_t nb_sectors = bdrv_nb_sectors(bs);
3389 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3392 bool bdrv_is_sg(BlockDriverState *bs)
3394 return bs->sg;
3397 bool bdrv_is_encrypted(BlockDriverState *bs)
3399 if (bs->backing && bs->backing->bs->encrypted) {
3400 return true;
3402 return bs->encrypted;
3405 bool bdrv_key_required(BlockDriverState *bs)
3407 BdrvChild *backing = bs->backing;
3409 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
3410 return true;
3412 return (bs->encrypted && !bs->valid_key);
3415 int bdrv_set_key(BlockDriverState *bs, const char *key)
3417 int ret;
3418 if (bs->backing && bs->backing->bs->encrypted) {
3419 ret = bdrv_set_key(bs->backing->bs, key);
3420 if (ret < 0)
3421 return ret;
3422 if (!bs->encrypted)
3423 return 0;
3425 if (!bs->encrypted) {
3426 return -EINVAL;
3427 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3428 return -ENOMEDIUM;
3430 ret = bs->drv->bdrv_set_key(bs, key);
3431 if (ret < 0) {
3432 bs->valid_key = false;
3433 } else if (!bs->valid_key) {
3434 /* call the change callback now, we skipped it on open */
3435 bs->valid_key = true;
3436 bdrv_parent_cb_change_media(bs, true);
3438 return ret;
3442 * Provide an encryption key for @bs.
3443 * If @key is non-null:
3444 * If @bs is not encrypted, fail.
3445 * Else if the key is invalid, fail.
3446 * Else set @bs's key to @key, replacing the existing key, if any.
3447 * If @key is null:
3448 * If @bs is encrypted and still lacks a key, fail.
3449 * Else do nothing.
3450 * On failure, store an error object through @errp if non-null.
3452 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
3454 if (key) {
3455 if (!bdrv_is_encrypted(bs)) {
3456 error_setg(errp, "Node '%s' is not encrypted",
3457 bdrv_get_device_or_node_name(bs));
3458 } else if (bdrv_set_key(bs, key) < 0) {
3459 error_setg(errp, QERR_INVALID_PASSWORD);
3461 } else {
3462 if (bdrv_key_required(bs)) {
3463 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
3464 "'%s' (%s) is encrypted",
3465 bdrv_get_device_or_node_name(bs),
3466 bdrv_get_encrypted_filename(bs));
3471 const char *bdrv_get_format_name(BlockDriverState *bs)
3473 return bs->drv ? bs->drv->format_name : NULL;
3476 static int qsort_strcmp(const void *a, const void *b)
3478 return strcmp(*(char *const *)a, *(char *const *)b);
3481 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3482 void *opaque)
3484 BlockDriver *drv;
3485 int count = 0;
3486 int i;
3487 const char **formats = NULL;
3489 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3490 if (drv->format_name) {
3491 bool found = false;
3492 int i = count;
3493 while (formats && i && !found) {
3494 found = !strcmp(formats[--i], drv->format_name);
3497 if (!found) {
3498 formats = g_renew(const char *, formats, count + 1);
3499 formats[count++] = drv->format_name;
3504 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3505 const char *format_name = block_driver_modules[i].format_name;
3507 if (format_name) {
3508 bool found = false;
3509 int j = count;
3511 while (formats && j && !found) {
3512 found = !strcmp(formats[--j], format_name);
3515 if (!found) {
3516 formats = g_renew(const char *, formats, count + 1);
3517 formats[count++] = format_name;
3522 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3524 for (i = 0; i < count; i++) {
3525 it(opaque, formats[i]);
3528 g_free(formats);
3531 /* This function is to find a node in the bs graph */
3532 BlockDriverState *bdrv_find_node(const char *node_name)
3534 BlockDriverState *bs;
3536 assert(node_name);
3538 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3539 if (!strcmp(node_name, bs->node_name)) {
3540 return bs;
3543 return NULL;
3546 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3547 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3549 BlockDeviceInfoList *list, *entry;
3550 BlockDriverState *bs;
3552 list = NULL;
3553 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3554 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3555 if (!info) {
3556 qapi_free_BlockDeviceInfoList(list);
3557 return NULL;
3559 entry = g_malloc0(sizeof(*entry));
3560 entry->value = info;
3561 entry->next = list;
3562 list = entry;
3565 return list;
3568 BlockDriverState *bdrv_lookup_bs(const char *device,
3569 const char *node_name,
3570 Error **errp)
3572 BlockBackend *blk;
3573 BlockDriverState *bs;
3575 if (device) {
3576 blk = blk_by_name(device);
3578 if (blk) {
3579 bs = blk_bs(blk);
3580 if (!bs) {
3581 error_setg(errp, "Device '%s' has no medium", device);
3584 return bs;
3588 if (node_name) {
3589 bs = bdrv_find_node(node_name);
3591 if (bs) {
3592 return bs;
3596 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3597 device ? device : "",
3598 node_name ? node_name : "");
3599 return NULL;
3602 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3603 * return false. If either argument is NULL, return false. */
3604 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3606 while (top && top != base) {
3607 top = backing_bs(top);
3610 return top != NULL;
3613 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3615 if (!bs) {
3616 return QTAILQ_FIRST(&graph_bdrv_states);
3618 return QTAILQ_NEXT(bs, node_list);
3621 const char *bdrv_get_node_name(const BlockDriverState *bs)
3623 return bs->node_name;
3626 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3628 BdrvChild *c;
3629 const char *name;
3631 /* If multiple parents have a name, just pick the first one. */
3632 QLIST_FOREACH(c, &bs->parents, next_parent) {
3633 if (c->role->get_name) {
3634 name = c->role->get_name(c);
3635 if (name && *name) {
3636 return name;
3641 return NULL;
3644 /* TODO check what callers really want: bs->node_name or blk_name() */
3645 const char *bdrv_get_device_name(const BlockDriverState *bs)
3647 return bdrv_get_parent_name(bs) ?: "";
3650 /* This can be used to identify nodes that might not have a device
3651 * name associated. Since node and device names live in the same
3652 * namespace, the result is unambiguous. The exception is if both are
3653 * absent, then this returns an empty (non-null) string. */
3654 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3656 return bdrv_get_parent_name(bs) ?: bs->node_name;
3659 int bdrv_get_flags(BlockDriverState *bs)
3661 return bs->open_flags;
3664 int bdrv_has_zero_init_1(BlockDriverState *bs)
3666 return 1;
3669 int bdrv_has_zero_init(BlockDriverState *bs)
3671 assert(bs->drv);
3673 /* If BS is a copy on write image, it is initialized to
3674 the contents of the base image, which may not be zeroes. */
3675 if (bs->backing) {
3676 return 0;
3678 if (bs->drv->bdrv_has_zero_init) {
3679 return bs->drv->bdrv_has_zero_init(bs);
3682 /* safe default */
3683 return 0;
3686 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3688 BlockDriverInfo bdi;
3690 if (bs->backing) {
3691 return false;
3694 if (bdrv_get_info(bs, &bdi) == 0) {
3695 return bdi.unallocated_blocks_are_zero;
3698 return false;
3701 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3703 BlockDriverInfo bdi;
3705 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3706 return false;
3709 if (bdrv_get_info(bs, &bdi) == 0) {
3710 return bdi.can_write_zeroes_with_unmap;
3713 return false;
3716 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3718 if (bs->backing && bs->backing->bs->encrypted)
3719 return bs->backing_file;
3720 else if (bs->encrypted)
3721 return bs->filename;
3722 else
3723 return NULL;
3726 void bdrv_get_backing_filename(BlockDriverState *bs,
3727 char *filename, int filename_size)
3729 pstrcpy(filename, filename_size, bs->backing_file);
3732 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3734 BlockDriver *drv = bs->drv;
3735 if (!drv)
3736 return -ENOMEDIUM;
3737 if (!drv->bdrv_get_info)
3738 return -ENOTSUP;
3739 memset(bdi, 0, sizeof(*bdi));
3740 return drv->bdrv_get_info(bs, bdi);
3743 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3745 BlockDriver *drv = bs->drv;
3746 if (drv && drv->bdrv_get_specific_info) {
3747 return drv->bdrv_get_specific_info(bs);
3749 return NULL;
3752 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3754 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3755 return;
3758 bs->drv->bdrv_debug_event(bs, event);
3761 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3762 const char *tag)
3764 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3765 bs = bs->file ? bs->file->bs : NULL;
3768 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3769 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3772 return -ENOTSUP;
3775 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3777 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3778 bs = bs->file ? bs->file->bs : NULL;
3781 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3782 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3785 return -ENOTSUP;
3788 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3790 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3791 bs = bs->file ? bs->file->bs : NULL;
3794 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3795 return bs->drv->bdrv_debug_resume(bs, tag);
3798 return -ENOTSUP;
3801 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3803 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3804 bs = bs->file ? bs->file->bs : NULL;
3807 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3808 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3811 return false;
3814 /* backing_file can either be relative, or absolute, or a protocol. If it is
3815 * relative, it must be relative to the chain. So, passing in bs->filename
3816 * from a BDS as backing_file should not be done, as that may be relative to
3817 * the CWD rather than the chain. */
3818 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3819 const char *backing_file)
3821 char *filename_full = NULL;
3822 char *backing_file_full = NULL;
3823 char *filename_tmp = NULL;
3824 int is_protocol = 0;
3825 BlockDriverState *curr_bs = NULL;
3826 BlockDriverState *retval = NULL;
3827 Error *local_error = NULL;
3829 if (!bs || !bs->drv || !backing_file) {
3830 return NULL;
3833 filename_full = g_malloc(PATH_MAX);
3834 backing_file_full = g_malloc(PATH_MAX);
3835 filename_tmp = g_malloc(PATH_MAX);
3837 is_protocol = path_has_protocol(backing_file);
3839 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3841 /* If either of the filename paths is actually a protocol, then
3842 * compare unmodified paths; otherwise make paths relative */
3843 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3844 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3845 retval = curr_bs->backing->bs;
3846 break;
3848 /* Also check against the full backing filename for the image */
3849 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
3850 &local_error);
3851 if (local_error == NULL) {
3852 if (strcmp(backing_file, backing_file_full) == 0) {
3853 retval = curr_bs->backing->bs;
3854 break;
3856 } else {
3857 error_free(local_error);
3858 local_error = NULL;
3860 } else {
3861 /* If not an absolute filename path, make it relative to the current
3862 * image's filename path */
3863 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3864 backing_file);
3866 /* We are going to compare absolute pathnames */
3867 if (!realpath(filename_tmp, filename_full)) {
3868 continue;
3871 /* We need to make sure the backing filename we are comparing against
3872 * is relative to the current image filename (or absolute) */
3873 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3874 curr_bs->backing_file);
3876 if (!realpath(filename_tmp, backing_file_full)) {
3877 continue;
3880 if (strcmp(backing_file_full, filename_full) == 0) {
3881 retval = curr_bs->backing->bs;
3882 break;
3887 g_free(filename_full);
3888 g_free(backing_file_full);
3889 g_free(filename_tmp);
3890 return retval;
3893 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3895 if (!bs->drv) {
3896 return 0;
3899 if (!bs->backing) {
3900 return 0;
3903 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3906 void bdrv_init(void)
3908 module_call_init(MODULE_INIT_BLOCK);
3911 void bdrv_init_with_whitelist(void)
3913 use_bdrv_whitelist = 1;
3914 bdrv_init();
3917 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3919 BdrvChild *child;
3920 Error *local_err = NULL;
3921 int ret;
3923 if (!bs->drv) {
3924 return;
3927 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3928 return;
3931 QLIST_FOREACH(child, &bs->children, next) {
3932 bdrv_invalidate_cache(child->bs, &local_err);
3933 if (local_err) {
3934 error_propagate(errp, local_err);
3935 return;
3939 bs->open_flags &= ~BDRV_O_INACTIVE;
3940 if (bs->drv->bdrv_invalidate_cache) {
3941 bs->drv->bdrv_invalidate_cache(bs, &local_err);
3942 if (local_err) {
3943 bs->open_flags |= BDRV_O_INACTIVE;
3944 error_propagate(errp, local_err);
3945 return;
3949 ret = refresh_total_sectors(bs, bs->total_sectors);
3950 if (ret < 0) {
3951 bs->open_flags |= BDRV_O_INACTIVE;
3952 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3953 return;
3957 void bdrv_invalidate_cache_all(Error **errp)
3959 BlockDriverState *bs;
3960 Error *local_err = NULL;
3961 BdrvNextIterator it;
3963 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3964 AioContext *aio_context = bdrv_get_aio_context(bs);
3966 aio_context_acquire(aio_context);
3967 bdrv_invalidate_cache(bs, &local_err);
3968 aio_context_release(aio_context);
3969 if (local_err) {
3970 error_propagate(errp, local_err);
3971 return;
3976 static int bdrv_inactivate_recurse(BlockDriverState *bs,
3977 bool setting_flag)
3979 BdrvChild *child;
3980 int ret;
3982 if (!setting_flag && bs->drv->bdrv_inactivate) {
3983 ret = bs->drv->bdrv_inactivate(bs);
3984 if (ret < 0) {
3985 return ret;
3989 QLIST_FOREACH(child, &bs->children, next) {
3990 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
3991 if (ret < 0) {
3992 return ret;
3996 if (setting_flag) {
3997 bs->open_flags |= BDRV_O_INACTIVE;
3999 return 0;
4002 int bdrv_inactivate_all(void)
4004 BlockDriverState *bs = NULL;
4005 BdrvNextIterator it;
4006 int ret = 0;
4007 int pass;
4009 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4010 aio_context_acquire(bdrv_get_aio_context(bs));
4013 /* We do two passes of inactivation. The first pass calls to drivers'
4014 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4015 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4016 * is allowed. */
4017 for (pass = 0; pass < 2; pass++) {
4018 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4019 ret = bdrv_inactivate_recurse(bs, pass);
4020 if (ret < 0) {
4021 goto out;
4026 out:
4027 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4028 aio_context_release(bdrv_get_aio_context(bs));
4031 return ret;
4034 /**************************************************************/
4035 /* removable device support */
4038 * Return TRUE if the media is present
4040 bool bdrv_is_inserted(BlockDriverState *bs)
4042 BlockDriver *drv = bs->drv;
4043 BdrvChild *child;
4045 if (!drv) {
4046 return false;
4048 if (drv->bdrv_is_inserted) {
4049 return drv->bdrv_is_inserted(bs);
4051 QLIST_FOREACH(child, &bs->children, next) {
4052 if (!bdrv_is_inserted(child->bs)) {
4053 return false;
4056 return true;
4060 * Return whether the media changed since the last call to this
4061 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4063 int bdrv_media_changed(BlockDriverState *bs)
4065 BlockDriver *drv = bs->drv;
4067 if (drv && drv->bdrv_media_changed) {
4068 return drv->bdrv_media_changed(bs);
4070 return -ENOTSUP;
4074 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4076 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4078 BlockDriver *drv = bs->drv;
4080 if (drv && drv->bdrv_eject) {
4081 drv->bdrv_eject(bs, eject_flag);
4086 * Lock or unlock the media (if it is locked, the user won't be able
4087 * to eject it manually).
4089 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4091 BlockDriver *drv = bs->drv;
4093 trace_bdrv_lock_medium(bs, locked);
4095 if (drv && drv->bdrv_lock_medium) {
4096 drv->bdrv_lock_medium(bs, locked);
4100 /* Get a reference to bs */
4101 void bdrv_ref(BlockDriverState *bs)
4103 bs->refcnt++;
4106 /* Release a previously grabbed reference to bs.
4107 * If after releasing, reference count is zero, the BlockDriverState is
4108 * deleted. */
4109 void bdrv_unref(BlockDriverState *bs)
4111 if (!bs) {
4112 return;
4114 assert(bs->refcnt > 0);
4115 if (--bs->refcnt == 0) {
4116 bdrv_delete(bs);
4120 struct BdrvOpBlocker {
4121 Error *reason;
4122 QLIST_ENTRY(BdrvOpBlocker) list;
4125 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4127 BdrvOpBlocker *blocker;
4128 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4129 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4130 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4131 if (errp) {
4132 *errp = error_copy(blocker->reason);
4133 error_prepend(errp, "Node '%s' is busy: ",
4134 bdrv_get_device_or_node_name(bs));
4136 return true;
4138 return false;
4141 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4143 BdrvOpBlocker *blocker;
4144 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4146 blocker = g_new0(BdrvOpBlocker, 1);
4147 blocker->reason = reason;
4148 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4151 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4153 BdrvOpBlocker *blocker, *next;
4154 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4155 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4156 if (blocker->reason == reason) {
4157 QLIST_REMOVE(blocker, list);
4158 g_free(blocker);
4163 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4165 int i;
4166 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4167 bdrv_op_block(bs, i, reason);
4171 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4173 int i;
4174 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4175 bdrv_op_unblock(bs, i, reason);
4179 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4181 int i;
4183 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4184 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4185 return false;
4188 return true;
4191 void bdrv_img_create(const char *filename, const char *fmt,
4192 const char *base_filename, const char *base_fmt,
4193 char *options, uint64_t img_size, int flags, bool quiet,
4194 Error **errp)
4196 QemuOptsList *create_opts = NULL;
4197 QemuOpts *opts = NULL;
4198 const char *backing_fmt, *backing_file;
4199 int64_t size;
4200 BlockDriver *drv, *proto_drv;
4201 Error *local_err = NULL;
4202 int ret = 0;
4204 /* Find driver and parse its options */
4205 drv = bdrv_find_format(fmt);
4206 if (!drv) {
4207 error_setg(errp, "Unknown file format '%s'", fmt);
4208 return;
4211 proto_drv = bdrv_find_protocol(filename, true, errp);
4212 if (!proto_drv) {
4213 return;
4216 if (!drv->create_opts) {
4217 error_setg(errp, "Format driver '%s' does not support image creation",
4218 drv->format_name);
4219 return;
4222 if (!proto_drv->create_opts) {
4223 error_setg(errp, "Protocol driver '%s' does not support image creation",
4224 proto_drv->format_name);
4225 return;
4228 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4229 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4231 /* Create parameter list with default values */
4232 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4233 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4235 /* Parse -o options */
4236 if (options) {
4237 qemu_opts_do_parse(opts, options, NULL, &local_err);
4238 if (local_err) {
4239 error_report_err(local_err);
4240 local_err = NULL;
4241 error_setg(errp, "Invalid options for file format '%s'", fmt);
4242 goto out;
4246 if (base_filename) {
4247 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4248 if (local_err) {
4249 error_setg(errp, "Backing file not supported for file format '%s'",
4250 fmt);
4251 goto out;
4255 if (base_fmt) {
4256 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4257 if (local_err) {
4258 error_setg(errp, "Backing file format not supported for file "
4259 "format '%s'", fmt);
4260 goto out;
4264 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4265 if (backing_file) {
4266 if (!strcmp(filename, backing_file)) {
4267 error_setg(errp, "Error: Trying to create an image with the "
4268 "same filename as the backing file");
4269 goto out;
4273 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4275 // The size for the image must always be specified, with one exception:
4276 // If we are using a backing file, we can obtain the size from there
4277 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4278 if (size == -1) {
4279 if (backing_file) {
4280 BlockDriverState *bs;
4281 char *full_backing = g_new0(char, PATH_MAX);
4282 int64_t size;
4283 int back_flags;
4284 QDict *backing_options = NULL;
4286 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4287 full_backing, PATH_MAX,
4288 &local_err);
4289 if (local_err) {
4290 g_free(full_backing);
4291 goto out;
4294 /* backing files always opened read-only */
4295 back_flags = flags;
4296 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4298 if (backing_fmt) {
4299 backing_options = qdict_new();
4300 qdict_put_str(backing_options, "driver", backing_fmt);
4303 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4304 &local_err);
4305 g_free(full_backing);
4306 if (!bs) {
4307 goto out;
4309 size = bdrv_getlength(bs);
4310 if (size < 0) {
4311 error_setg_errno(errp, -size, "Could not get size of '%s'",
4312 backing_file);
4313 bdrv_unref(bs);
4314 goto out;
4317 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4319 bdrv_unref(bs);
4320 } else {
4321 error_setg(errp, "Image creation needs a size parameter");
4322 goto out;
4326 if (!quiet) {
4327 printf("Formatting '%s', fmt=%s ", filename, fmt);
4328 qemu_opts_print(opts, " ");
4329 puts("");
4332 ret = bdrv_create(drv, filename, opts, &local_err);
4334 if (ret == -EFBIG) {
4335 /* This is generally a better message than whatever the driver would
4336 * deliver (especially because of the cluster_size_hint), since that
4337 * is most probably not much different from "image too large". */
4338 const char *cluster_size_hint = "";
4339 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4340 cluster_size_hint = " (try using a larger cluster size)";
4342 error_setg(errp, "The image size is too large for file format '%s'"
4343 "%s", fmt, cluster_size_hint);
4344 error_free(local_err);
4345 local_err = NULL;
4348 out:
4349 qemu_opts_del(opts);
4350 qemu_opts_free(create_opts);
4351 error_propagate(errp, local_err);
4354 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4356 return bs->aio_context;
4359 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4361 aio_co_enter(bdrv_get_aio_context(bs), co);
4364 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4366 QLIST_REMOVE(ban, list);
4367 g_free(ban);
4370 void bdrv_detach_aio_context(BlockDriverState *bs)
4372 BdrvAioNotifier *baf, *baf_tmp;
4373 BdrvChild *child;
4375 if (!bs->drv) {
4376 return;
4379 assert(!bs->walking_aio_notifiers);
4380 bs->walking_aio_notifiers = true;
4381 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4382 if (baf->deleted) {
4383 bdrv_do_remove_aio_context_notifier(baf);
4384 } else {
4385 baf->detach_aio_context(baf->opaque);
4388 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4389 * remove remaining aio notifiers if we aren't called again.
4391 bs->walking_aio_notifiers = false;
4393 if (bs->drv->bdrv_detach_aio_context) {
4394 bs->drv->bdrv_detach_aio_context(bs);
4396 QLIST_FOREACH(child, &bs->children, next) {
4397 bdrv_detach_aio_context(child->bs);
4400 bs->aio_context = NULL;
4403 void bdrv_attach_aio_context(BlockDriverState *bs,
4404 AioContext *new_context)
4406 BdrvAioNotifier *ban, *ban_tmp;
4407 BdrvChild *child;
4409 if (!bs->drv) {
4410 return;
4413 bs->aio_context = new_context;
4415 QLIST_FOREACH(child, &bs->children, next) {
4416 bdrv_attach_aio_context(child->bs, new_context);
4418 if (bs->drv->bdrv_attach_aio_context) {
4419 bs->drv->bdrv_attach_aio_context(bs, new_context);
4422 assert(!bs->walking_aio_notifiers);
4423 bs->walking_aio_notifiers = true;
4424 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4425 if (ban->deleted) {
4426 bdrv_do_remove_aio_context_notifier(ban);
4427 } else {
4428 ban->attached_aio_context(new_context, ban->opaque);
4431 bs->walking_aio_notifiers = false;
4434 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4436 AioContext *ctx = bdrv_get_aio_context(bs);
4438 aio_disable_external(ctx);
4439 bdrv_parent_drained_begin(bs);
4440 bdrv_drain(bs); /* ensure there are no in-flight requests */
4442 while (aio_poll(ctx, false)) {
4443 /* wait for all bottom halves to execute */
4446 bdrv_detach_aio_context(bs);
4448 /* This function executes in the old AioContext so acquire the new one in
4449 * case it runs in a different thread.
4451 aio_context_acquire(new_context);
4452 bdrv_attach_aio_context(bs, new_context);
4453 bdrv_parent_drained_end(bs);
4454 aio_enable_external(ctx);
4455 aio_context_release(new_context);
4458 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4459 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4460 void (*detach_aio_context)(void *opaque), void *opaque)
4462 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4463 *ban = (BdrvAioNotifier){
4464 .attached_aio_context = attached_aio_context,
4465 .detach_aio_context = detach_aio_context,
4466 .opaque = opaque
4469 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4472 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4473 void (*attached_aio_context)(AioContext *,
4474 void *),
4475 void (*detach_aio_context)(void *),
4476 void *opaque)
4478 BdrvAioNotifier *ban, *ban_next;
4480 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4481 if (ban->attached_aio_context == attached_aio_context &&
4482 ban->detach_aio_context == detach_aio_context &&
4483 ban->opaque == opaque &&
4484 ban->deleted == false)
4486 if (bs->walking_aio_notifiers) {
4487 ban->deleted = true;
4488 } else {
4489 bdrv_do_remove_aio_context_notifier(ban);
4491 return;
4495 abort();
4498 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4499 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4501 if (!bs->drv->bdrv_amend_options) {
4502 return -ENOTSUP;
4504 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
4507 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4508 * of block filter and by bdrv_is_first_non_filter.
4509 * It is used to test if the given bs is the candidate or recurse more in the
4510 * node graph.
4512 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
4513 BlockDriverState *candidate)
4515 /* return false if basic checks fails */
4516 if (!bs || !bs->drv) {
4517 return false;
4520 /* the code reached a non block filter driver -> check if the bs is
4521 * the same as the candidate. It's the recursion termination condition.
4523 if (!bs->drv->is_filter) {
4524 return bs == candidate;
4526 /* Down this path the driver is a block filter driver */
4528 /* If the block filter recursion method is defined use it to recurse down
4529 * the node graph.
4531 if (bs->drv->bdrv_recurse_is_first_non_filter) {
4532 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
4535 /* the driver is a block filter but don't allow to recurse -> return false
4537 return false;
4540 /* This function checks if the candidate is the first non filter bs down it's
4541 * bs chain. Since we don't have pointers to parents it explore all bs chains
4542 * from the top. Some filters can choose not to pass down the recursion.
4544 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
4546 BlockDriverState *bs;
4547 BdrvNextIterator it;
4549 /* walk down the bs forest recursively */
4550 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4551 bool perm;
4553 /* try to recurse in this top level bs */
4554 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
4556 /* candidate is the first non filter */
4557 if (perm) {
4558 return true;
4562 return false;
4565 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
4566 const char *node_name, Error **errp)
4568 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
4569 AioContext *aio_context;
4571 if (!to_replace_bs) {
4572 error_setg(errp, "Node name '%s' not found", node_name);
4573 return NULL;
4576 aio_context = bdrv_get_aio_context(to_replace_bs);
4577 aio_context_acquire(aio_context);
4579 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
4580 to_replace_bs = NULL;
4581 goto out;
4584 /* We don't want arbitrary node of the BDS chain to be replaced only the top
4585 * most non filter in order to prevent data corruption.
4586 * Another benefit is that this tests exclude backing files which are
4587 * blocked by the backing blockers.
4589 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
4590 error_setg(errp, "Only top most non filter can be replaced");
4591 to_replace_bs = NULL;
4592 goto out;
4595 out:
4596 aio_context_release(aio_context);
4597 return to_replace_bs;
4600 static bool append_open_options(QDict *d, BlockDriverState *bs)
4602 const QDictEntry *entry;
4603 QemuOptDesc *desc;
4604 BdrvChild *child;
4605 bool found_any = false;
4606 const char *p;
4608 for (entry = qdict_first(bs->options); entry;
4609 entry = qdict_next(bs->options, entry))
4611 /* Exclude options for children */
4612 QLIST_FOREACH(child, &bs->children, next) {
4613 if (strstart(qdict_entry_key(entry), child->name, &p)
4614 && (!*p || *p == '.'))
4616 break;
4619 if (child) {
4620 continue;
4623 /* And exclude all non-driver-specific options */
4624 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4625 if (!strcmp(qdict_entry_key(entry), desc->name)) {
4626 break;
4629 if (desc->name) {
4630 continue;
4633 qobject_incref(qdict_entry_value(entry));
4634 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
4635 found_any = true;
4638 return found_any;
4641 /* Updates the following BDS fields:
4642 * - exact_filename: A filename which may be used for opening a block device
4643 * which (mostly) equals the given BDS (even without any
4644 * other options; so reading and writing must return the same
4645 * results, but caching etc. may be different)
4646 * - full_open_options: Options which, when given when opening a block device
4647 * (without a filename), result in a BDS (mostly)
4648 * equalling the given one
4649 * - filename: If exact_filename is set, it is copied here. Otherwise,
4650 * full_open_options is converted to a JSON object, prefixed with
4651 * "json:" (for use through the JSON pseudo protocol) and put here.
4653 void bdrv_refresh_filename(BlockDriverState *bs)
4655 BlockDriver *drv = bs->drv;
4656 QDict *opts;
4658 if (!drv) {
4659 return;
4662 /* This BDS's file name will most probably depend on its file's name, so
4663 * refresh that first */
4664 if (bs->file) {
4665 bdrv_refresh_filename(bs->file->bs);
4668 if (drv->bdrv_refresh_filename) {
4669 /* Obsolete information is of no use here, so drop the old file name
4670 * information before refreshing it */
4671 bs->exact_filename[0] = '\0';
4672 if (bs->full_open_options) {
4673 QDECREF(bs->full_open_options);
4674 bs->full_open_options = NULL;
4677 opts = qdict_new();
4678 append_open_options(opts, bs);
4679 drv->bdrv_refresh_filename(bs, opts);
4680 QDECREF(opts);
4681 } else if (bs->file) {
4682 /* Try to reconstruct valid information from the underlying file */
4683 bool has_open_options;
4685 bs->exact_filename[0] = '\0';
4686 if (bs->full_open_options) {
4687 QDECREF(bs->full_open_options);
4688 bs->full_open_options = NULL;
4691 opts = qdict_new();
4692 has_open_options = append_open_options(opts, bs);
4694 /* If no specific options have been given for this BDS, the filename of
4695 * the underlying file should suffice for this one as well */
4696 if (bs->file->bs->exact_filename[0] && !has_open_options) {
4697 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
4699 /* Reconstructing the full options QDict is simple for most format block
4700 * drivers, as long as the full options are known for the underlying
4701 * file BDS. The full options QDict of that file BDS should somehow
4702 * contain a representation of the filename, therefore the following
4703 * suffices without querying the (exact_)filename of this BDS. */
4704 if (bs->file->bs->full_open_options) {
4705 qdict_put_str(opts, "driver", drv->format_name);
4706 QINCREF(bs->file->bs->full_open_options);
4707 qdict_put(opts, "file", bs->file->bs->full_open_options);
4709 bs->full_open_options = opts;
4710 } else {
4711 QDECREF(opts);
4713 } else if (!bs->full_open_options && qdict_size(bs->options)) {
4714 /* There is no underlying file BDS (at least referenced by BDS.file),
4715 * so the full options QDict should be equal to the options given
4716 * specifically for this block device when it was opened (plus the
4717 * driver specification).
4718 * Because those options don't change, there is no need to update
4719 * full_open_options when it's already set. */
4721 opts = qdict_new();
4722 append_open_options(opts, bs);
4723 qdict_put_str(opts, "driver", drv->format_name);
4725 if (bs->exact_filename[0]) {
4726 /* This may not work for all block protocol drivers (some may
4727 * require this filename to be parsed), but we have to find some
4728 * default solution here, so just include it. If some block driver
4729 * does not support pure options without any filename at all or
4730 * needs some special format of the options QDict, it needs to
4731 * implement the driver-specific bdrv_refresh_filename() function.
4733 qdict_put_str(opts, "filename", bs->exact_filename);
4736 bs->full_open_options = opts;
4739 if (bs->exact_filename[0]) {
4740 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
4741 } else if (bs->full_open_options) {
4742 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
4743 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
4744 qstring_get_str(json));
4745 QDECREF(json);
4750 * Hot add/remove a BDS's child. So the user can take a child offline when
4751 * it is broken and take a new child online
4753 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
4754 Error **errp)
4757 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
4758 error_setg(errp, "The node %s does not support adding a child",
4759 bdrv_get_device_or_node_name(parent_bs));
4760 return;
4763 if (!QLIST_EMPTY(&child_bs->parents)) {
4764 error_setg(errp, "The node %s already has a parent",
4765 child_bs->node_name);
4766 return;
4769 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
4772 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
4774 BdrvChild *tmp;
4776 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
4777 error_setg(errp, "The node %s does not support removing a child",
4778 bdrv_get_device_or_node_name(parent_bs));
4779 return;
4782 QLIST_FOREACH(tmp, &parent_bs->children, next) {
4783 if (tmp == child) {
4784 break;
4788 if (!tmp) {
4789 error_setg(errp, "The node %s does not have a child named %s",
4790 bdrv_get_device_or_node_name(parent_bs),
4791 bdrv_get_device_or_node_name(child->bs));
4792 return;
4795 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);