hw/sd/bcm2835_sdhost: Add tracepoints
[qemu/kevin.git] / block.c
bloba2caadf0a07ac8ac8e5abe35bac603a3aae9d3fd
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "qemu/error-report.h"
31 #include "module_block.h"
32 #include "qemu/module.h"
33 #include "qapi/error.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qjson.h"
36 #include "qapi/qmp/qnull.h"
37 #include "qapi/qmp/qstring.h"
38 #include "qapi/qobject-output-visitor.h"
39 #include "qapi/qapi-visit-block-core.h"
40 #include "sysemu/block-backend.h"
41 #include "sysemu/sysemu.h"
42 #include "qemu/notify.h"
43 #include "qemu/option.h"
44 #include "qemu/coroutine.h"
45 #include "block/qapi.h"
46 #include "qemu/timer.h"
47 #include "qemu/cutils.h"
48 #include "qemu/id.h"
50 #ifdef CONFIG_BSD
51 #include <sys/ioctl.h>
52 #include <sys/queue.h>
53 #ifndef __DragonFly__
54 #include <sys/disk.h>
55 #endif
56 #endif
58 #ifdef _WIN32
59 #include <windows.h>
60 #endif
62 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
64 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
65 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
67 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
68 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
70 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
71 QLIST_HEAD_INITIALIZER(bdrv_drivers);
73 static BlockDriverState *bdrv_open_inherit(const char *filename,
74 const char *reference,
75 QDict *options, int flags,
76 BlockDriverState *parent,
77 const BdrvChildRole *child_role,
78 Error **errp);
80 /* If non-zero, use only whitelisted block drivers */
81 static int use_bdrv_whitelist;
83 #ifdef _WIN32
84 static int is_windows_drive_prefix(const char *filename)
86 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
87 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
88 filename[1] == ':');
91 int is_windows_drive(const char *filename)
93 if (is_windows_drive_prefix(filename) &&
94 filename[2] == '\0')
95 return 1;
96 if (strstart(filename, "\\\\.\\", NULL) ||
97 strstart(filename, "//./", NULL))
98 return 1;
99 return 0;
101 #endif
103 size_t bdrv_opt_mem_align(BlockDriverState *bs)
105 if (!bs || !bs->drv) {
106 /* page size or 4k (hdd sector size) should be on the safe side */
107 return MAX(4096, getpagesize());
110 return bs->bl.opt_mem_alignment;
113 size_t bdrv_min_mem_align(BlockDriverState *bs)
115 if (!bs || !bs->drv) {
116 /* page size or 4k (hdd sector size) should be on the safe side */
117 return MAX(4096, getpagesize());
120 return bs->bl.min_mem_alignment;
123 /* check if the path starts with "<protocol>:" */
124 int path_has_protocol(const char *path)
126 const char *p;
128 #ifdef _WIN32
129 if (is_windows_drive(path) ||
130 is_windows_drive_prefix(path)) {
131 return 0;
133 p = path + strcspn(path, ":/\\");
134 #else
135 p = path + strcspn(path, ":/");
136 #endif
138 return *p == ':';
141 int path_is_absolute(const char *path)
143 #ifdef _WIN32
144 /* specific case for names like: "\\.\d:" */
145 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
146 return 1;
148 return (*path == '/' || *path == '\\');
149 #else
150 return (*path == '/');
151 #endif
154 /* if filename is absolute, just copy it to dest. Otherwise, build a
155 path to it by considering it is relative to base_path. URL are
156 supported. */
157 void path_combine(char *dest, int dest_size,
158 const char *base_path,
159 const char *filename)
161 const char *p, *p1;
162 int len;
164 if (dest_size <= 0)
165 return;
166 if (path_is_absolute(filename)) {
167 pstrcpy(dest, dest_size, filename);
168 } else {
169 const char *protocol_stripped = NULL;
171 if (path_has_protocol(base_path)) {
172 protocol_stripped = strchr(base_path, ':');
173 if (protocol_stripped) {
174 protocol_stripped++;
177 p = protocol_stripped ?: base_path;
179 p1 = strrchr(base_path, '/');
180 #ifdef _WIN32
182 const char *p2;
183 p2 = strrchr(base_path, '\\');
184 if (!p1 || p2 > p1)
185 p1 = p2;
187 #endif
188 if (p1)
189 p1++;
190 else
191 p1 = base_path;
192 if (p1 > p)
193 p = p1;
194 len = p - base_path;
195 if (len > dest_size - 1)
196 len = dest_size - 1;
197 memcpy(dest, base_path, len);
198 dest[len] = '\0';
199 pstrcat(dest, dest_size, filename);
204 * Helper function for bdrv_parse_filename() implementations to remove optional
205 * protocol prefixes (especially "file:") from a filename and for putting the
206 * stripped filename into the options QDict if there is such a prefix.
208 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
209 QDict *options)
211 if (strstart(filename, prefix, &filename)) {
212 /* Stripping the explicit protocol prefix may result in a protocol
213 * prefix being (wrongly) detected (if the filename contains a colon) */
214 if (path_has_protocol(filename)) {
215 QString *fat_filename;
217 /* This means there is some colon before the first slash; therefore,
218 * this cannot be an absolute path */
219 assert(!path_is_absolute(filename));
221 /* And we can thus fix the protocol detection issue by prefixing it
222 * by "./" */
223 fat_filename = qstring_from_str("./");
224 qstring_append(fat_filename, filename);
226 assert(!path_has_protocol(qstring_get_str(fat_filename)));
228 qdict_put(options, "filename", fat_filename);
229 } else {
230 /* If no protocol prefix was detected, we can use the shortened
231 * filename as-is */
232 qdict_put_str(options, "filename", filename);
238 /* Returns whether the image file is opened as read-only. Note that this can
239 * return false and writing to the image file is still not possible because the
240 * image is inactivated. */
241 bool bdrv_is_read_only(BlockDriverState *bs)
243 return bs->read_only;
246 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
247 bool ignore_allow_rdw, Error **errp)
249 /* Do not set read_only if copy_on_read is enabled */
250 if (bs->copy_on_read && read_only) {
251 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
252 bdrv_get_device_or_node_name(bs));
253 return -EINVAL;
256 /* Do not clear read_only if it is prohibited */
257 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
258 !ignore_allow_rdw)
260 error_setg(errp, "Node '%s' is read only",
261 bdrv_get_device_or_node_name(bs));
262 return -EPERM;
265 return 0;
268 /* TODO Remove (deprecated since 2.11)
269 * Block drivers are not supposed to automatically change bs->read_only.
270 * Instead, they should just check whether they can provide what the user
271 * explicitly requested and error out if read-write is requested, but they can
272 * only provide read-only access. */
273 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
275 int ret = 0;
277 ret = bdrv_can_set_read_only(bs, read_only, false, errp);
278 if (ret < 0) {
279 return ret;
282 bs->read_only = read_only;
283 return 0;
286 void bdrv_get_full_backing_filename_from_filename(const char *backed,
287 const char *backing,
288 char *dest, size_t sz,
289 Error **errp)
291 if (backing[0] == '\0' || path_has_protocol(backing) ||
292 path_is_absolute(backing))
294 pstrcpy(dest, sz, backing);
295 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
296 error_setg(errp, "Cannot use relative backing file names for '%s'",
297 backed);
298 } else {
299 path_combine(dest, sz, backed, backing);
303 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
304 Error **errp)
306 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
308 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
309 dest, sz, errp);
312 void bdrv_register(BlockDriver *bdrv)
314 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
317 BlockDriverState *bdrv_new(void)
319 BlockDriverState *bs;
320 int i;
322 bs = g_new0(BlockDriverState, 1);
323 QLIST_INIT(&bs->dirty_bitmaps);
324 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
325 QLIST_INIT(&bs->op_blockers[i]);
327 notifier_with_return_list_init(&bs->before_write_notifiers);
328 qemu_co_mutex_init(&bs->reqs_lock);
329 qemu_mutex_init(&bs->dirty_bitmap_mutex);
330 bs->refcnt = 1;
331 bs->aio_context = qemu_get_aio_context();
333 qemu_co_queue_init(&bs->flush_queue);
335 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
337 return bs;
340 static BlockDriver *bdrv_do_find_format(const char *format_name)
342 BlockDriver *drv1;
344 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
345 if (!strcmp(drv1->format_name, format_name)) {
346 return drv1;
350 return NULL;
353 BlockDriver *bdrv_find_format(const char *format_name)
355 BlockDriver *drv1;
356 int i;
358 drv1 = bdrv_do_find_format(format_name);
359 if (drv1) {
360 return drv1;
363 /* The driver isn't registered, maybe we need to load a module */
364 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
365 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
366 block_module_load_one(block_driver_modules[i].library_name);
367 break;
371 return bdrv_do_find_format(format_name);
374 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
376 static const char *whitelist_rw[] = {
377 CONFIG_BDRV_RW_WHITELIST
379 static const char *whitelist_ro[] = {
380 CONFIG_BDRV_RO_WHITELIST
382 const char **p;
384 if (!whitelist_rw[0] && !whitelist_ro[0]) {
385 return 1; /* no whitelist, anything goes */
388 for (p = whitelist_rw; *p; p++) {
389 if (!strcmp(drv->format_name, *p)) {
390 return 1;
393 if (read_only) {
394 for (p = whitelist_ro; *p; p++) {
395 if (!strcmp(drv->format_name, *p)) {
396 return 1;
400 return 0;
403 bool bdrv_uses_whitelist(void)
405 return use_bdrv_whitelist;
408 typedef struct CreateCo {
409 BlockDriver *drv;
410 char *filename;
411 QemuOpts *opts;
412 int ret;
413 Error *err;
414 } CreateCo;
416 static void coroutine_fn bdrv_create_co_entry(void *opaque)
418 Error *local_err = NULL;
419 int ret;
421 CreateCo *cco = opaque;
422 assert(cco->drv);
424 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
425 error_propagate(&cco->err, local_err);
426 cco->ret = ret;
429 int bdrv_create(BlockDriver *drv, const char* filename,
430 QemuOpts *opts, Error **errp)
432 int ret;
434 Coroutine *co;
435 CreateCo cco = {
436 .drv = drv,
437 .filename = g_strdup(filename),
438 .opts = opts,
439 .ret = NOT_DONE,
440 .err = NULL,
443 if (!drv->bdrv_co_create_opts) {
444 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
445 ret = -ENOTSUP;
446 goto out;
449 if (qemu_in_coroutine()) {
450 /* Fast-path if already in coroutine context */
451 bdrv_create_co_entry(&cco);
452 } else {
453 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
454 qemu_coroutine_enter(co);
455 while (cco.ret == NOT_DONE) {
456 aio_poll(qemu_get_aio_context(), true);
460 ret = cco.ret;
461 if (ret < 0) {
462 if (cco.err) {
463 error_propagate(errp, cco.err);
464 } else {
465 error_setg_errno(errp, -ret, "Could not create image");
469 out:
470 g_free(cco.filename);
471 return ret;
474 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
476 BlockDriver *drv;
477 Error *local_err = NULL;
478 int ret;
480 drv = bdrv_find_protocol(filename, true, errp);
481 if (drv == NULL) {
482 return -ENOENT;
485 ret = bdrv_create(drv, filename, opts, &local_err);
486 error_propagate(errp, local_err);
487 return ret;
491 * Try to get @bs's logical and physical block size.
492 * On success, store them in @bsz struct and return 0.
493 * On failure return -errno.
494 * @bs must not be empty.
496 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
498 BlockDriver *drv = bs->drv;
500 if (drv && drv->bdrv_probe_blocksizes) {
501 return drv->bdrv_probe_blocksizes(bs, bsz);
502 } else if (drv && drv->is_filter && bs->file) {
503 return bdrv_probe_blocksizes(bs->file->bs, bsz);
506 return -ENOTSUP;
510 * Try to get @bs's geometry (cyls, heads, sectors).
511 * On success, store them in @geo struct and return 0.
512 * On failure return -errno.
513 * @bs must not be empty.
515 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
517 BlockDriver *drv = bs->drv;
519 if (drv && drv->bdrv_probe_geometry) {
520 return drv->bdrv_probe_geometry(bs, geo);
521 } else if (drv && drv->is_filter && bs->file) {
522 return bdrv_probe_geometry(bs->file->bs, geo);
525 return -ENOTSUP;
529 * Create a uniquely-named empty temporary file.
530 * Return 0 upon success, otherwise a negative errno value.
532 int get_tmp_filename(char *filename, int size)
534 #ifdef _WIN32
535 char temp_dir[MAX_PATH];
536 /* GetTempFileName requires that its output buffer (4th param)
537 have length MAX_PATH or greater. */
538 assert(size >= MAX_PATH);
539 return (GetTempPath(MAX_PATH, temp_dir)
540 && GetTempFileName(temp_dir, "qem", 0, filename)
541 ? 0 : -GetLastError());
542 #else
543 int fd;
544 const char *tmpdir;
545 tmpdir = getenv("TMPDIR");
546 if (!tmpdir) {
547 tmpdir = "/var/tmp";
549 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
550 return -EOVERFLOW;
552 fd = mkstemp(filename);
553 if (fd < 0) {
554 return -errno;
556 if (close(fd) != 0) {
557 unlink(filename);
558 return -errno;
560 return 0;
561 #endif
565 * Detect host devices. By convention, /dev/cdrom[N] is always
566 * recognized as a host CDROM.
568 static BlockDriver *find_hdev_driver(const char *filename)
570 int score_max = 0, score;
571 BlockDriver *drv = NULL, *d;
573 QLIST_FOREACH(d, &bdrv_drivers, list) {
574 if (d->bdrv_probe_device) {
575 score = d->bdrv_probe_device(filename);
576 if (score > score_max) {
577 score_max = score;
578 drv = d;
583 return drv;
586 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
588 BlockDriver *drv1;
590 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
591 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
592 return drv1;
596 return NULL;
599 BlockDriver *bdrv_find_protocol(const char *filename,
600 bool allow_protocol_prefix,
601 Error **errp)
603 BlockDriver *drv1;
604 char protocol[128];
605 int len;
606 const char *p;
607 int i;
609 /* TODO Drivers without bdrv_file_open must be specified explicitly */
612 * XXX(hch): we really should not let host device detection
613 * override an explicit protocol specification, but moving this
614 * later breaks access to device names with colons in them.
615 * Thanks to the brain-dead persistent naming schemes on udev-
616 * based Linux systems those actually are quite common.
618 drv1 = find_hdev_driver(filename);
619 if (drv1) {
620 return drv1;
623 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
624 return &bdrv_file;
627 p = strchr(filename, ':');
628 assert(p != NULL);
629 len = p - filename;
630 if (len > sizeof(protocol) - 1)
631 len = sizeof(protocol) - 1;
632 memcpy(protocol, filename, len);
633 protocol[len] = '\0';
635 drv1 = bdrv_do_find_protocol(protocol);
636 if (drv1) {
637 return drv1;
640 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
641 if (block_driver_modules[i].protocol_name &&
642 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
643 block_module_load_one(block_driver_modules[i].library_name);
644 break;
648 drv1 = bdrv_do_find_protocol(protocol);
649 if (!drv1) {
650 error_setg(errp, "Unknown protocol '%s'", protocol);
652 return drv1;
656 * Guess image format by probing its contents.
657 * This is not a good idea when your image is raw (CVE-2008-2004), but
658 * we do it anyway for backward compatibility.
660 * @buf contains the image's first @buf_size bytes.
661 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
662 * but can be smaller if the image file is smaller)
663 * @filename is its filename.
665 * For all block drivers, call the bdrv_probe() method to get its
666 * probing score.
667 * Return the first block driver with the highest probing score.
669 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
670 const char *filename)
672 int score_max = 0, score;
673 BlockDriver *drv = NULL, *d;
675 QLIST_FOREACH(d, &bdrv_drivers, list) {
676 if (d->bdrv_probe) {
677 score = d->bdrv_probe(buf, buf_size, filename);
678 if (score > score_max) {
679 score_max = score;
680 drv = d;
685 return drv;
688 static int find_image_format(BlockBackend *file, const char *filename,
689 BlockDriver **pdrv, Error **errp)
691 BlockDriver *drv;
692 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
693 int ret = 0;
695 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
696 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
697 *pdrv = &bdrv_raw;
698 return ret;
701 ret = blk_pread(file, 0, buf, sizeof(buf));
702 if (ret < 0) {
703 error_setg_errno(errp, -ret, "Could not read image for determining its "
704 "format");
705 *pdrv = NULL;
706 return ret;
709 drv = bdrv_probe_all(buf, ret, filename);
710 if (!drv) {
711 error_setg(errp, "Could not determine image format: No compatible "
712 "driver found");
713 ret = -ENOENT;
715 *pdrv = drv;
716 return ret;
720 * Set the current 'total_sectors' value
721 * Return 0 on success, -errno on error.
723 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
725 BlockDriver *drv = bs->drv;
727 if (!drv) {
728 return -ENOMEDIUM;
731 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
732 if (bdrv_is_sg(bs))
733 return 0;
735 /* query actual device if possible, otherwise just trust the hint */
736 if (drv->bdrv_getlength) {
737 int64_t length = drv->bdrv_getlength(bs);
738 if (length < 0) {
739 return length;
741 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
744 bs->total_sectors = hint;
745 return 0;
749 * Combines a QDict of new block driver @options with any missing options taken
750 * from @old_options, so that leaving out an option defaults to its old value.
752 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
753 QDict *old_options)
755 if (bs->drv && bs->drv->bdrv_join_options) {
756 bs->drv->bdrv_join_options(options, old_options);
757 } else {
758 qdict_join(options, old_options, false);
763 * Set open flags for a given discard mode
765 * Return 0 on success, -1 if the discard mode was invalid.
767 int bdrv_parse_discard_flags(const char *mode, int *flags)
769 *flags &= ~BDRV_O_UNMAP;
771 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
772 /* do nothing */
773 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
774 *flags |= BDRV_O_UNMAP;
775 } else {
776 return -1;
779 return 0;
783 * Set open flags for a given cache mode
785 * Return 0 on success, -1 if the cache mode was invalid.
787 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
789 *flags &= ~BDRV_O_CACHE_MASK;
791 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
792 *writethrough = false;
793 *flags |= BDRV_O_NOCACHE;
794 } else if (!strcmp(mode, "directsync")) {
795 *writethrough = true;
796 *flags |= BDRV_O_NOCACHE;
797 } else if (!strcmp(mode, "writeback")) {
798 *writethrough = false;
799 } else if (!strcmp(mode, "unsafe")) {
800 *writethrough = false;
801 *flags |= BDRV_O_NO_FLUSH;
802 } else if (!strcmp(mode, "writethrough")) {
803 *writethrough = true;
804 } else {
805 return -1;
808 return 0;
811 static char *bdrv_child_get_parent_desc(BdrvChild *c)
813 BlockDriverState *parent = c->opaque;
814 return g_strdup(bdrv_get_device_or_node_name(parent));
817 static void bdrv_child_cb_drained_begin(BdrvChild *child)
819 BlockDriverState *bs = child->opaque;
820 bdrv_drained_begin(bs);
823 static void bdrv_child_cb_drained_end(BdrvChild *child)
825 BlockDriverState *bs = child->opaque;
826 bdrv_drained_end(bs);
829 static void bdrv_child_cb_attach(BdrvChild *child)
831 BlockDriverState *bs = child->opaque;
832 bdrv_apply_subtree_drain(child, bs);
835 static void bdrv_child_cb_detach(BdrvChild *child)
837 BlockDriverState *bs = child->opaque;
838 bdrv_unapply_subtree_drain(child, bs);
841 static int bdrv_child_cb_inactivate(BdrvChild *child)
843 BlockDriverState *bs = child->opaque;
844 assert(bs->open_flags & BDRV_O_INACTIVE);
845 return 0;
849 * Returns the options and flags that a temporary snapshot should get, based on
850 * the originally requested flags (the originally requested image will have
851 * flags like a backing file)
853 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
854 int parent_flags, QDict *parent_options)
856 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
858 /* For temporary files, unconditional cache=unsafe is fine */
859 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
860 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
862 /* Copy the read-only option from the parent */
863 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
865 /* aio=native doesn't work for cache.direct=off, so disable it for the
866 * temporary snapshot */
867 *child_flags &= ~BDRV_O_NATIVE_AIO;
871 * Returns the options and flags that bs->file should get if a protocol driver
872 * is expected, based on the given options and flags for the parent BDS
874 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
875 int parent_flags, QDict *parent_options)
877 int flags = parent_flags;
879 /* Enable protocol handling, disable format probing for bs->file */
880 flags |= BDRV_O_PROTOCOL;
882 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
883 * the parent. */
884 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
885 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
886 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
888 /* Inherit the read-only option from the parent if it's not set */
889 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
891 /* Our block drivers take care to send flushes and respect unmap policy,
892 * so we can default to enable both on lower layers regardless of the
893 * corresponding parent options. */
894 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
896 /* Clear flags that only apply to the top layer */
897 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
898 BDRV_O_NO_IO);
900 *child_flags = flags;
903 const BdrvChildRole child_file = {
904 .get_parent_desc = bdrv_child_get_parent_desc,
905 .inherit_options = bdrv_inherited_options,
906 .drained_begin = bdrv_child_cb_drained_begin,
907 .drained_end = bdrv_child_cb_drained_end,
908 .attach = bdrv_child_cb_attach,
909 .detach = bdrv_child_cb_detach,
910 .inactivate = bdrv_child_cb_inactivate,
914 * Returns the options and flags that bs->file should get if the use of formats
915 * (and not only protocols) is permitted for it, based on the given options and
916 * flags for the parent BDS
918 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
919 int parent_flags, QDict *parent_options)
921 child_file.inherit_options(child_flags, child_options,
922 parent_flags, parent_options);
924 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
927 const BdrvChildRole child_format = {
928 .get_parent_desc = bdrv_child_get_parent_desc,
929 .inherit_options = bdrv_inherited_fmt_options,
930 .drained_begin = bdrv_child_cb_drained_begin,
931 .drained_end = bdrv_child_cb_drained_end,
932 .attach = bdrv_child_cb_attach,
933 .detach = bdrv_child_cb_detach,
934 .inactivate = bdrv_child_cb_inactivate,
937 static void bdrv_backing_attach(BdrvChild *c)
939 BlockDriverState *parent = c->opaque;
940 BlockDriverState *backing_hd = c->bs;
942 assert(!parent->backing_blocker);
943 error_setg(&parent->backing_blocker,
944 "node is used as backing hd of '%s'",
945 bdrv_get_device_or_node_name(parent));
947 parent->open_flags &= ~BDRV_O_NO_BACKING;
948 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
949 backing_hd->filename);
950 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
951 backing_hd->drv ? backing_hd->drv->format_name : "");
953 bdrv_op_block_all(backing_hd, parent->backing_blocker);
954 /* Otherwise we won't be able to commit or stream */
955 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
956 parent->backing_blocker);
957 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
958 parent->backing_blocker);
960 * We do backup in 3 ways:
961 * 1. drive backup
962 * The target bs is new opened, and the source is top BDS
963 * 2. blockdev backup
964 * Both the source and the target are top BDSes.
965 * 3. internal backup(used for block replication)
966 * Both the source and the target are backing file
968 * In case 1 and 2, neither the source nor the target is the backing file.
969 * In case 3, we will block the top BDS, so there is only one block job
970 * for the top BDS and its backing chain.
972 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
973 parent->backing_blocker);
974 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
975 parent->backing_blocker);
977 bdrv_child_cb_attach(c);
980 static void bdrv_backing_detach(BdrvChild *c)
982 BlockDriverState *parent = c->opaque;
984 assert(parent->backing_blocker);
985 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
986 error_free(parent->backing_blocker);
987 parent->backing_blocker = NULL;
989 bdrv_child_cb_detach(c);
993 * Returns the options and flags that bs->backing should get, based on the
994 * given options and flags for the parent BDS
996 static void bdrv_backing_options(int *child_flags, QDict *child_options,
997 int parent_flags, QDict *parent_options)
999 int flags = parent_flags;
1001 /* The cache mode is inherited unmodified for backing files; except WCE,
1002 * which is only applied on the top level (BlockBackend) */
1003 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1004 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1005 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1007 /* backing files always opened read-only */
1008 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1009 flags &= ~BDRV_O_COPY_ON_READ;
1011 /* snapshot=on is handled on the top layer */
1012 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1014 *child_flags = flags;
1017 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1018 const char *filename, Error **errp)
1020 BlockDriverState *parent = c->opaque;
1021 int orig_flags = bdrv_get_flags(parent);
1022 int ret;
1024 if (!(orig_flags & BDRV_O_RDWR)) {
1025 ret = bdrv_reopen(parent, orig_flags | BDRV_O_RDWR, errp);
1026 if (ret < 0) {
1027 return ret;
1031 ret = bdrv_change_backing_file(parent, filename,
1032 base->drv ? base->drv->format_name : "");
1033 if (ret < 0) {
1034 error_setg_errno(errp, -ret, "Could not update backing file link");
1037 if (!(orig_flags & BDRV_O_RDWR)) {
1038 bdrv_reopen(parent, orig_flags, NULL);
1041 return ret;
1044 const BdrvChildRole child_backing = {
1045 .get_parent_desc = bdrv_child_get_parent_desc,
1046 .attach = bdrv_backing_attach,
1047 .detach = bdrv_backing_detach,
1048 .inherit_options = bdrv_backing_options,
1049 .drained_begin = bdrv_child_cb_drained_begin,
1050 .drained_end = bdrv_child_cb_drained_end,
1051 .inactivate = bdrv_child_cb_inactivate,
1052 .update_filename = bdrv_backing_update_filename,
1055 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1057 int open_flags = flags;
1060 * Clear flags that are internal to the block layer before opening the
1061 * image.
1063 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1066 * Snapshots should be writable.
1068 if (flags & BDRV_O_TEMPORARY) {
1069 open_flags |= BDRV_O_RDWR;
1072 return open_flags;
1075 static void update_flags_from_options(int *flags, QemuOpts *opts)
1077 *flags &= ~BDRV_O_CACHE_MASK;
1079 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1080 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1081 *flags |= BDRV_O_NO_FLUSH;
1084 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1085 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1086 *flags |= BDRV_O_NOCACHE;
1089 *flags &= ~BDRV_O_RDWR;
1091 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1092 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1093 *flags |= BDRV_O_RDWR;
1098 static void update_options_from_flags(QDict *options, int flags)
1100 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1101 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1103 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1104 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1105 flags & BDRV_O_NO_FLUSH);
1107 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1108 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1112 static void bdrv_assign_node_name(BlockDriverState *bs,
1113 const char *node_name,
1114 Error **errp)
1116 char *gen_node_name = NULL;
1118 if (!node_name) {
1119 node_name = gen_node_name = id_generate(ID_BLOCK);
1120 } else if (!id_wellformed(node_name)) {
1122 * Check for empty string or invalid characters, but not if it is
1123 * generated (generated names use characters not available to the user)
1125 error_setg(errp, "Invalid node name");
1126 return;
1129 /* takes care of avoiding namespaces collisions */
1130 if (blk_by_name(node_name)) {
1131 error_setg(errp, "node-name=%s is conflicting with a device id",
1132 node_name);
1133 goto out;
1136 /* takes care of avoiding duplicates node names */
1137 if (bdrv_find_node(node_name)) {
1138 error_setg(errp, "Duplicate node name");
1139 goto out;
1142 /* copy node name into the bs and insert it into the graph list */
1143 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1144 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1145 out:
1146 g_free(gen_node_name);
1149 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1150 const char *node_name, QDict *options,
1151 int open_flags, Error **errp)
1153 Error *local_err = NULL;
1154 int ret;
1156 bdrv_assign_node_name(bs, node_name, &local_err);
1157 if (local_err) {
1158 error_propagate(errp, local_err);
1159 return -EINVAL;
1162 bs->drv = drv;
1163 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1164 bs->opaque = g_malloc0(drv->instance_size);
1166 if (drv->bdrv_file_open) {
1167 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1168 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1169 } else if (drv->bdrv_open) {
1170 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1171 } else {
1172 ret = 0;
1175 if (ret < 0) {
1176 if (local_err) {
1177 error_propagate(errp, local_err);
1178 } else if (bs->filename[0]) {
1179 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1180 } else {
1181 error_setg_errno(errp, -ret, "Could not open image");
1183 goto open_failed;
1186 ret = refresh_total_sectors(bs, bs->total_sectors);
1187 if (ret < 0) {
1188 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1189 return ret;
1192 bdrv_refresh_limits(bs, &local_err);
1193 if (local_err) {
1194 error_propagate(errp, local_err);
1195 return -EINVAL;
1198 assert(bdrv_opt_mem_align(bs) != 0);
1199 assert(bdrv_min_mem_align(bs) != 0);
1200 assert(is_power_of_2(bs->bl.request_alignment));
1202 return 0;
1203 open_failed:
1204 bs->drv = NULL;
1205 if (bs->file != NULL) {
1206 bdrv_unref_child(bs, bs->file);
1207 bs->file = NULL;
1209 g_free(bs->opaque);
1210 bs->opaque = NULL;
1211 return ret;
1214 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1215 int flags, Error **errp)
1217 BlockDriverState *bs;
1218 int ret;
1220 bs = bdrv_new();
1221 bs->open_flags = flags;
1222 bs->explicit_options = qdict_new();
1223 bs->options = qdict_new();
1224 bs->opaque = NULL;
1226 update_options_from_flags(bs->options, flags);
1228 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1229 if (ret < 0) {
1230 QDECREF(bs->explicit_options);
1231 bs->explicit_options = NULL;
1232 QDECREF(bs->options);
1233 bs->options = NULL;
1234 bdrv_unref(bs);
1235 return NULL;
1238 return bs;
1241 QemuOptsList bdrv_runtime_opts = {
1242 .name = "bdrv_common",
1243 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1244 .desc = {
1246 .name = "node-name",
1247 .type = QEMU_OPT_STRING,
1248 .help = "Node name of the block device node",
1251 .name = "driver",
1252 .type = QEMU_OPT_STRING,
1253 .help = "Block driver to use for the node",
1256 .name = BDRV_OPT_CACHE_DIRECT,
1257 .type = QEMU_OPT_BOOL,
1258 .help = "Bypass software writeback cache on the host",
1261 .name = BDRV_OPT_CACHE_NO_FLUSH,
1262 .type = QEMU_OPT_BOOL,
1263 .help = "Ignore flush requests",
1266 .name = BDRV_OPT_READ_ONLY,
1267 .type = QEMU_OPT_BOOL,
1268 .help = "Node is opened in read-only mode",
1271 .name = "detect-zeroes",
1272 .type = QEMU_OPT_STRING,
1273 .help = "try to optimize zero writes (off, on, unmap)",
1276 .name = "discard",
1277 .type = QEMU_OPT_STRING,
1278 .help = "discard operation (ignore/off, unmap/on)",
1281 .name = BDRV_OPT_FORCE_SHARE,
1282 .type = QEMU_OPT_BOOL,
1283 .help = "always accept other writers (default: off)",
1285 { /* end of list */ }
1290 * Common part for opening disk images and files
1292 * Removes all processed options from *options.
1294 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1295 QDict *options, Error **errp)
1297 int ret, open_flags;
1298 const char *filename;
1299 const char *driver_name = NULL;
1300 const char *node_name = NULL;
1301 const char *discard;
1302 const char *detect_zeroes;
1303 QemuOpts *opts;
1304 BlockDriver *drv;
1305 Error *local_err = NULL;
1307 assert(bs->file == NULL);
1308 assert(options != NULL && bs->options != options);
1310 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1311 qemu_opts_absorb_qdict(opts, options, &local_err);
1312 if (local_err) {
1313 error_propagate(errp, local_err);
1314 ret = -EINVAL;
1315 goto fail_opts;
1318 update_flags_from_options(&bs->open_flags, opts);
1320 driver_name = qemu_opt_get(opts, "driver");
1321 drv = bdrv_find_format(driver_name);
1322 assert(drv != NULL);
1324 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1326 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1327 error_setg(errp,
1328 BDRV_OPT_FORCE_SHARE
1329 "=on can only be used with read-only images");
1330 ret = -EINVAL;
1331 goto fail_opts;
1334 if (file != NULL) {
1335 filename = blk_bs(file)->filename;
1336 } else {
1338 * Caution: while qdict_get_try_str() is fine, getting
1339 * non-string types would require more care. When @options
1340 * come from -blockdev or blockdev_add, its members are typed
1341 * according to the QAPI schema, but when they come from
1342 * -drive, they're all QString.
1344 filename = qdict_get_try_str(options, "filename");
1347 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1348 error_setg(errp, "The '%s' block driver requires a file name",
1349 drv->format_name);
1350 ret = -EINVAL;
1351 goto fail_opts;
1354 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1355 drv->format_name);
1357 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1359 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1360 error_setg(errp,
1361 !bs->read_only && bdrv_is_whitelisted(drv, true)
1362 ? "Driver '%s' can only be used for read-only devices"
1363 : "Driver '%s' is not whitelisted",
1364 drv->format_name);
1365 ret = -ENOTSUP;
1366 goto fail_opts;
1369 /* bdrv_new() and bdrv_close() make it so */
1370 assert(atomic_read(&bs->copy_on_read) == 0);
1372 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1373 if (!bs->read_only) {
1374 bdrv_enable_copy_on_read(bs);
1375 } else {
1376 error_setg(errp, "Can't use copy-on-read on read-only device");
1377 ret = -EINVAL;
1378 goto fail_opts;
1382 discard = qemu_opt_get(opts, "discard");
1383 if (discard != NULL) {
1384 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1385 error_setg(errp, "Invalid discard option");
1386 ret = -EINVAL;
1387 goto fail_opts;
1391 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1392 if (detect_zeroes) {
1393 BlockdevDetectZeroesOptions value =
1394 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
1395 detect_zeroes,
1396 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1397 &local_err);
1398 if (local_err) {
1399 error_propagate(errp, local_err);
1400 ret = -EINVAL;
1401 goto fail_opts;
1404 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1405 !(bs->open_flags & BDRV_O_UNMAP))
1407 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1408 "without setting discard operation to unmap");
1409 ret = -EINVAL;
1410 goto fail_opts;
1413 bs->detect_zeroes = value;
1416 if (filename != NULL) {
1417 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1418 } else {
1419 bs->filename[0] = '\0';
1421 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1423 /* Open the image, either directly or using a protocol */
1424 open_flags = bdrv_open_flags(bs, bs->open_flags);
1425 node_name = qemu_opt_get(opts, "node-name");
1427 assert(!drv->bdrv_file_open || file == NULL);
1428 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1429 if (ret < 0) {
1430 goto fail_opts;
1433 qemu_opts_del(opts);
1434 return 0;
1436 fail_opts:
1437 qemu_opts_del(opts);
1438 return ret;
1441 static QDict *parse_json_filename(const char *filename, Error **errp)
1443 QObject *options_obj;
1444 QDict *options;
1445 int ret;
1447 ret = strstart(filename, "json:", &filename);
1448 assert(ret);
1450 options_obj = qobject_from_json(filename, errp);
1451 if (!options_obj) {
1452 /* Work around qobject_from_json() lossage TODO fix that */
1453 if (errp && !*errp) {
1454 error_setg(errp, "Could not parse the JSON options");
1455 return NULL;
1457 error_prepend(errp, "Could not parse the JSON options: ");
1458 return NULL;
1461 options = qobject_to(QDict, options_obj);
1462 if (!options) {
1463 qobject_decref(options_obj);
1464 error_setg(errp, "Invalid JSON object given");
1465 return NULL;
1468 qdict_flatten(options);
1470 return options;
1473 static void parse_json_protocol(QDict *options, const char **pfilename,
1474 Error **errp)
1476 QDict *json_options;
1477 Error *local_err = NULL;
1479 /* Parse json: pseudo-protocol */
1480 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1481 return;
1484 json_options = parse_json_filename(*pfilename, &local_err);
1485 if (local_err) {
1486 error_propagate(errp, local_err);
1487 return;
1490 /* Options given in the filename have lower priority than options
1491 * specified directly */
1492 qdict_join(options, json_options, false);
1493 QDECREF(json_options);
1494 *pfilename = NULL;
1498 * Fills in default options for opening images and converts the legacy
1499 * filename/flags pair to option QDict entries.
1500 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1501 * block driver has been specified explicitly.
1503 static int bdrv_fill_options(QDict **options, const char *filename,
1504 int *flags, Error **errp)
1506 const char *drvname;
1507 bool protocol = *flags & BDRV_O_PROTOCOL;
1508 bool parse_filename = false;
1509 BlockDriver *drv = NULL;
1510 Error *local_err = NULL;
1513 * Caution: while qdict_get_try_str() is fine, getting non-string
1514 * types would require more care. When @options come from
1515 * -blockdev or blockdev_add, its members are typed according to
1516 * the QAPI schema, but when they come from -drive, they're all
1517 * QString.
1519 drvname = qdict_get_try_str(*options, "driver");
1520 if (drvname) {
1521 drv = bdrv_find_format(drvname);
1522 if (!drv) {
1523 error_setg(errp, "Unknown driver '%s'", drvname);
1524 return -ENOENT;
1526 /* If the user has explicitly specified the driver, this choice should
1527 * override the BDRV_O_PROTOCOL flag */
1528 protocol = drv->bdrv_file_open;
1531 if (protocol) {
1532 *flags |= BDRV_O_PROTOCOL;
1533 } else {
1534 *flags &= ~BDRV_O_PROTOCOL;
1537 /* Translate cache options from flags into options */
1538 update_options_from_flags(*options, *flags);
1540 /* Fetch the file name from the options QDict if necessary */
1541 if (protocol && filename) {
1542 if (!qdict_haskey(*options, "filename")) {
1543 qdict_put_str(*options, "filename", filename);
1544 parse_filename = true;
1545 } else {
1546 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1547 "the same time");
1548 return -EINVAL;
1552 /* Find the right block driver */
1553 /* See cautionary note on accessing @options above */
1554 filename = qdict_get_try_str(*options, "filename");
1556 if (!drvname && protocol) {
1557 if (filename) {
1558 drv = bdrv_find_protocol(filename, parse_filename, errp);
1559 if (!drv) {
1560 return -EINVAL;
1563 drvname = drv->format_name;
1564 qdict_put_str(*options, "driver", drvname);
1565 } else {
1566 error_setg(errp, "Must specify either driver or file");
1567 return -EINVAL;
1571 assert(drv || !protocol);
1573 /* Driver-specific filename parsing */
1574 if (drv && drv->bdrv_parse_filename && parse_filename) {
1575 drv->bdrv_parse_filename(filename, *options, &local_err);
1576 if (local_err) {
1577 error_propagate(errp, local_err);
1578 return -EINVAL;
1581 if (!drv->bdrv_needs_filename) {
1582 qdict_del(*options, "filename");
1586 return 0;
1589 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1590 uint64_t perm, uint64_t shared,
1591 GSList *ignore_children, Error **errp);
1592 static void bdrv_child_abort_perm_update(BdrvChild *c);
1593 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1595 typedef struct BlockReopenQueueEntry {
1596 bool prepared;
1597 BDRVReopenState state;
1598 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1599 } BlockReopenQueueEntry;
1602 * Return the flags that @bs will have after the reopens in @q have
1603 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1604 * return the current flags.
1606 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1608 BlockReopenQueueEntry *entry;
1610 if (q != NULL) {
1611 QSIMPLEQ_FOREACH(entry, q, entry) {
1612 if (entry->state.bs == bs) {
1613 return entry->state.flags;
1618 return bs->open_flags;
1621 /* Returns whether the image file can be written to after the reopen queue @q
1622 * has been successfully applied, or right now if @q is NULL. */
1623 static bool bdrv_is_writable(BlockDriverState *bs, BlockReopenQueue *q)
1625 int flags = bdrv_reopen_get_flags(q, bs);
1627 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1630 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1631 BdrvChild *c, const BdrvChildRole *role,
1632 BlockReopenQueue *reopen_queue,
1633 uint64_t parent_perm, uint64_t parent_shared,
1634 uint64_t *nperm, uint64_t *nshared)
1636 if (bs->drv && bs->drv->bdrv_child_perm) {
1637 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1638 parent_perm, parent_shared,
1639 nperm, nshared);
1641 /* TODO Take force_share from reopen_queue */
1642 if (child_bs && child_bs->force_share) {
1643 *nshared = BLK_PERM_ALL;
1648 * Check whether permissions on this node can be changed in a way that
1649 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1650 * permissions of all its parents. This involves checking whether all necessary
1651 * permission changes to child nodes can be performed.
1653 * A call to this function must always be followed by a call to bdrv_set_perm()
1654 * or bdrv_abort_perm_update().
1656 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1657 uint64_t cumulative_perms,
1658 uint64_t cumulative_shared_perms,
1659 GSList *ignore_children, Error **errp)
1661 BlockDriver *drv = bs->drv;
1662 BdrvChild *c;
1663 int ret;
1665 /* Write permissions never work with read-only images */
1666 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1667 !bdrv_is_writable(bs, q))
1669 error_setg(errp, "Block node is read-only");
1670 return -EPERM;
1673 /* Check this node */
1674 if (!drv) {
1675 return 0;
1678 if (drv->bdrv_check_perm) {
1679 return drv->bdrv_check_perm(bs, cumulative_perms,
1680 cumulative_shared_perms, errp);
1683 /* Drivers that never have children can omit .bdrv_child_perm() */
1684 if (!drv->bdrv_child_perm) {
1685 assert(QLIST_EMPTY(&bs->children));
1686 return 0;
1689 /* Check all children */
1690 QLIST_FOREACH(c, &bs->children, next) {
1691 uint64_t cur_perm, cur_shared;
1692 bdrv_child_perm(bs, c->bs, c, c->role, q,
1693 cumulative_perms, cumulative_shared_perms,
1694 &cur_perm, &cur_shared);
1695 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1696 ignore_children, errp);
1697 if (ret < 0) {
1698 return ret;
1702 return 0;
1706 * Notifies drivers that after a previous bdrv_check_perm() call, the
1707 * permission update is not performed and any preparations made for it (e.g.
1708 * taken file locks) need to be undone.
1710 * This function recursively notifies all child nodes.
1712 static void bdrv_abort_perm_update(BlockDriverState *bs)
1714 BlockDriver *drv = bs->drv;
1715 BdrvChild *c;
1717 if (!drv) {
1718 return;
1721 if (drv->bdrv_abort_perm_update) {
1722 drv->bdrv_abort_perm_update(bs);
1725 QLIST_FOREACH(c, &bs->children, next) {
1726 bdrv_child_abort_perm_update(c);
1730 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1731 uint64_t cumulative_shared_perms)
1733 BlockDriver *drv = bs->drv;
1734 BdrvChild *c;
1736 if (!drv) {
1737 return;
1740 /* Update this node */
1741 if (drv->bdrv_set_perm) {
1742 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1745 /* Drivers that never have children can omit .bdrv_child_perm() */
1746 if (!drv->bdrv_child_perm) {
1747 assert(QLIST_EMPTY(&bs->children));
1748 return;
1751 /* Update all children */
1752 QLIST_FOREACH(c, &bs->children, next) {
1753 uint64_t cur_perm, cur_shared;
1754 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1755 cumulative_perms, cumulative_shared_perms,
1756 &cur_perm, &cur_shared);
1757 bdrv_child_set_perm(c, cur_perm, cur_shared);
1761 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1762 uint64_t *shared_perm)
1764 BdrvChild *c;
1765 uint64_t cumulative_perms = 0;
1766 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1768 QLIST_FOREACH(c, &bs->parents, next_parent) {
1769 cumulative_perms |= c->perm;
1770 cumulative_shared_perms &= c->shared_perm;
1773 *perm = cumulative_perms;
1774 *shared_perm = cumulative_shared_perms;
1777 static char *bdrv_child_user_desc(BdrvChild *c)
1779 if (c->role->get_parent_desc) {
1780 return c->role->get_parent_desc(c);
1783 return g_strdup("another user");
1786 char *bdrv_perm_names(uint64_t perm)
1788 struct perm_name {
1789 uint64_t perm;
1790 const char *name;
1791 } permissions[] = {
1792 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1793 { BLK_PERM_WRITE, "write" },
1794 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1795 { BLK_PERM_RESIZE, "resize" },
1796 { BLK_PERM_GRAPH_MOD, "change children" },
1797 { 0, NULL }
1800 char *result = g_strdup("");
1801 struct perm_name *p;
1803 for (p = permissions; p->name; p++) {
1804 if (perm & p->perm) {
1805 char *old = result;
1806 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1807 g_free(old);
1811 return result;
1815 * Checks whether a new reference to @bs can be added if the new user requires
1816 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1817 * set, the BdrvChild objects in this list are ignored in the calculations;
1818 * this allows checking permission updates for an existing reference.
1820 * Needs to be followed by a call to either bdrv_set_perm() or
1821 * bdrv_abort_perm_update(). */
1822 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1823 uint64_t new_used_perm,
1824 uint64_t new_shared_perm,
1825 GSList *ignore_children, Error **errp)
1827 BdrvChild *c;
1828 uint64_t cumulative_perms = new_used_perm;
1829 uint64_t cumulative_shared_perms = new_shared_perm;
1831 /* There is no reason why anyone couldn't tolerate write_unchanged */
1832 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1834 QLIST_FOREACH(c, &bs->parents, next_parent) {
1835 if (g_slist_find(ignore_children, c)) {
1836 continue;
1839 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1840 char *user = bdrv_child_user_desc(c);
1841 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1842 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1843 "allow '%s' on %s",
1844 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1845 g_free(user);
1846 g_free(perm_names);
1847 return -EPERM;
1850 if ((c->perm & new_shared_perm) != c->perm) {
1851 char *user = bdrv_child_user_desc(c);
1852 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1853 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1854 "'%s' on %s",
1855 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1856 g_free(user);
1857 g_free(perm_names);
1858 return -EPERM;
1861 cumulative_perms |= c->perm;
1862 cumulative_shared_perms &= c->shared_perm;
1865 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1866 ignore_children, errp);
1869 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1870 * bdrv_child_abort_perm_update(). */
1871 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1872 uint64_t perm, uint64_t shared,
1873 GSList *ignore_children, Error **errp)
1875 int ret;
1877 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1878 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1879 g_slist_free(ignore_children);
1881 return ret;
1884 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1886 uint64_t cumulative_perms, cumulative_shared_perms;
1888 c->perm = perm;
1889 c->shared_perm = shared;
1891 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1892 &cumulative_shared_perms);
1893 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1896 static void bdrv_child_abort_perm_update(BdrvChild *c)
1898 bdrv_abort_perm_update(c->bs);
1901 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1902 Error **errp)
1904 int ret;
1906 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1907 if (ret < 0) {
1908 bdrv_child_abort_perm_update(c);
1909 return ret;
1912 bdrv_child_set_perm(c, perm, shared);
1914 return 0;
1917 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1918 | BLK_PERM_WRITE \
1919 | BLK_PERM_WRITE_UNCHANGED \
1920 | BLK_PERM_RESIZE)
1921 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1923 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1924 const BdrvChildRole *role,
1925 BlockReopenQueue *reopen_queue,
1926 uint64_t perm, uint64_t shared,
1927 uint64_t *nperm, uint64_t *nshared)
1929 if (c == NULL) {
1930 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1931 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1932 return;
1935 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1936 (c->perm & DEFAULT_PERM_UNCHANGED);
1937 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1938 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1941 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1942 const BdrvChildRole *role,
1943 BlockReopenQueue *reopen_queue,
1944 uint64_t perm, uint64_t shared,
1945 uint64_t *nperm, uint64_t *nshared)
1947 bool backing = (role == &child_backing);
1948 assert(role == &child_backing || role == &child_file);
1950 if (!backing) {
1951 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
1953 /* Apart from the modifications below, the same permissions are
1954 * forwarded and left alone as for filters */
1955 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
1956 &perm, &shared);
1958 /* Format drivers may touch metadata even if the guest doesn't write */
1959 if (bdrv_is_writable(bs, reopen_queue)) {
1960 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1963 /* bs->file always needs to be consistent because of the metadata. We
1964 * can never allow other users to resize or write to it. */
1965 if (!(flags & BDRV_O_NO_IO)) {
1966 perm |= BLK_PERM_CONSISTENT_READ;
1968 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1969 } else {
1970 /* We want consistent read from backing files if the parent needs it.
1971 * No other operations are performed on backing files. */
1972 perm &= BLK_PERM_CONSISTENT_READ;
1974 /* If the parent can deal with changing data, we're okay with a
1975 * writable and resizable backing file. */
1976 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1977 if (shared & BLK_PERM_WRITE) {
1978 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1979 } else {
1980 shared = 0;
1983 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1984 BLK_PERM_WRITE_UNCHANGED;
1987 if (bs->open_flags & BDRV_O_INACTIVE) {
1988 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1991 *nperm = perm;
1992 *nshared = shared;
1995 static void bdrv_replace_child_noperm(BdrvChild *child,
1996 BlockDriverState *new_bs)
1998 BlockDriverState *old_bs = child->bs;
1999 int i;
2001 if (old_bs && new_bs) {
2002 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2004 if (old_bs) {
2005 /* Detach first so that the recursive drain sections coming from @child
2006 * are already gone and we only end the drain sections that came from
2007 * elsewhere. */
2008 if (child->role->detach) {
2009 child->role->detach(child);
2011 if (old_bs->quiesce_counter && child->role->drained_end) {
2012 for (i = 0; i < old_bs->quiesce_counter; i++) {
2013 child->role->drained_end(child);
2016 QLIST_REMOVE(child, next_parent);
2019 child->bs = new_bs;
2021 if (new_bs) {
2022 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2023 if (new_bs->quiesce_counter && child->role->drained_begin) {
2024 for (i = 0; i < new_bs->quiesce_counter; i++) {
2025 child->role->drained_begin(child);
2029 /* Attach only after starting new drained sections, so that recursive
2030 * drain sections coming from @child don't get an extra .drained_begin
2031 * callback. */
2032 if (child->role->attach) {
2033 child->role->attach(child);
2039 * Updates @child to change its reference to point to @new_bs, including
2040 * checking and applying the necessary permisson updates both to the old node
2041 * and to @new_bs.
2043 * NULL is passed as @new_bs for removing the reference before freeing @child.
2045 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2046 * function uses bdrv_set_perm() to update the permissions according to the new
2047 * reference that @new_bs gets.
2049 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2051 BlockDriverState *old_bs = child->bs;
2052 uint64_t perm, shared_perm;
2054 bdrv_replace_child_noperm(child, new_bs);
2056 if (old_bs) {
2057 /* Update permissions for old node. This is guaranteed to succeed
2058 * because we're just taking a parent away, so we're loosening
2059 * restrictions. */
2060 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2061 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2062 bdrv_set_perm(old_bs, perm, shared_perm);
2065 if (new_bs) {
2066 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2067 bdrv_set_perm(new_bs, perm, shared_perm);
2071 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2072 const char *child_name,
2073 const BdrvChildRole *child_role,
2074 uint64_t perm, uint64_t shared_perm,
2075 void *opaque, Error **errp)
2077 BdrvChild *child;
2078 int ret;
2080 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2081 if (ret < 0) {
2082 bdrv_abort_perm_update(child_bs);
2083 return NULL;
2086 child = g_new(BdrvChild, 1);
2087 *child = (BdrvChild) {
2088 .bs = NULL,
2089 .name = g_strdup(child_name),
2090 .role = child_role,
2091 .perm = perm,
2092 .shared_perm = shared_perm,
2093 .opaque = opaque,
2096 /* This performs the matching bdrv_set_perm() for the above check. */
2097 bdrv_replace_child(child, child_bs);
2099 return child;
2102 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2103 BlockDriverState *child_bs,
2104 const char *child_name,
2105 const BdrvChildRole *child_role,
2106 Error **errp)
2108 BdrvChild *child;
2109 uint64_t perm, shared_perm;
2111 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2113 assert(parent_bs->drv);
2114 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2115 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2116 perm, shared_perm, &perm, &shared_perm);
2118 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2119 perm, shared_perm, parent_bs, errp);
2120 if (child == NULL) {
2121 return NULL;
2124 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2125 return child;
2128 static void bdrv_detach_child(BdrvChild *child)
2130 if (child->next.le_prev) {
2131 QLIST_REMOVE(child, next);
2132 child->next.le_prev = NULL;
2135 bdrv_replace_child(child, NULL);
2137 g_free(child->name);
2138 g_free(child);
2141 void bdrv_root_unref_child(BdrvChild *child)
2143 BlockDriverState *child_bs;
2145 child_bs = child->bs;
2146 bdrv_detach_child(child);
2147 bdrv_unref(child_bs);
2150 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2152 if (child == NULL) {
2153 return;
2156 if (child->bs->inherits_from == parent) {
2157 BdrvChild *c;
2159 /* Remove inherits_from only when the last reference between parent and
2160 * child->bs goes away. */
2161 QLIST_FOREACH(c, &parent->children, next) {
2162 if (c != child && c->bs == child->bs) {
2163 break;
2166 if (c == NULL) {
2167 child->bs->inherits_from = NULL;
2171 bdrv_root_unref_child(child);
2175 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2177 BdrvChild *c;
2178 QLIST_FOREACH(c, &bs->parents, next_parent) {
2179 if (c->role->change_media) {
2180 c->role->change_media(c, load);
2185 static void bdrv_parent_cb_resize(BlockDriverState *bs)
2187 BdrvChild *c;
2188 QLIST_FOREACH(c, &bs->parents, next_parent) {
2189 if (c->role->resize) {
2190 c->role->resize(c);
2196 * Sets the backing file link of a BDS. A new reference is created; callers
2197 * which don't need their own reference any more must call bdrv_unref().
2199 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2200 Error **errp)
2202 if (backing_hd) {
2203 bdrv_ref(backing_hd);
2206 if (bs->backing) {
2207 bdrv_unref_child(bs, bs->backing);
2210 if (!backing_hd) {
2211 bs->backing = NULL;
2212 goto out;
2215 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2216 errp);
2217 if (!bs->backing) {
2218 bdrv_unref(backing_hd);
2221 bdrv_refresh_filename(bs);
2223 out:
2224 bdrv_refresh_limits(bs, NULL);
2228 * Opens the backing file for a BlockDriverState if not yet open
2230 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2231 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2232 * itself, all options starting with "${bdref_key}." are considered part of the
2233 * BlockdevRef.
2235 * TODO Can this be unified with bdrv_open_image()?
2237 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2238 const char *bdref_key, Error **errp)
2240 char *backing_filename = g_malloc0(PATH_MAX);
2241 char *bdref_key_dot;
2242 const char *reference = NULL;
2243 int ret = 0;
2244 BlockDriverState *backing_hd;
2245 QDict *options;
2246 QDict *tmp_parent_options = NULL;
2247 Error *local_err = NULL;
2249 if (bs->backing != NULL) {
2250 goto free_exit;
2253 /* NULL means an empty set of options */
2254 if (parent_options == NULL) {
2255 tmp_parent_options = qdict_new();
2256 parent_options = tmp_parent_options;
2259 bs->open_flags &= ~BDRV_O_NO_BACKING;
2261 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2262 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2263 g_free(bdref_key_dot);
2266 * Caution: while qdict_get_try_str() is fine, getting non-string
2267 * types would require more care. When @parent_options come from
2268 * -blockdev or blockdev_add, its members are typed according to
2269 * the QAPI schema, but when they come from -drive, they're all
2270 * QString.
2272 reference = qdict_get_try_str(parent_options, bdref_key);
2273 if (reference || qdict_haskey(options, "file.filename")) {
2274 backing_filename[0] = '\0';
2275 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2276 QDECREF(options);
2277 goto free_exit;
2278 } else {
2279 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2280 &local_err);
2281 if (local_err) {
2282 ret = -EINVAL;
2283 error_propagate(errp, local_err);
2284 QDECREF(options);
2285 goto free_exit;
2289 if (!bs->drv || !bs->drv->supports_backing) {
2290 ret = -EINVAL;
2291 error_setg(errp, "Driver doesn't support backing files");
2292 QDECREF(options);
2293 goto free_exit;
2296 if (!reference &&
2297 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2298 qdict_put_str(options, "driver", bs->backing_format);
2301 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2302 reference, options, 0, bs, &child_backing,
2303 errp);
2304 if (!backing_hd) {
2305 bs->open_flags |= BDRV_O_NO_BACKING;
2306 error_prepend(errp, "Could not open backing file: ");
2307 ret = -EINVAL;
2308 goto free_exit;
2310 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2312 /* Hook up the backing file link; drop our reference, bs owns the
2313 * backing_hd reference now */
2314 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2315 bdrv_unref(backing_hd);
2316 if (local_err) {
2317 error_propagate(errp, local_err);
2318 ret = -EINVAL;
2319 goto free_exit;
2322 qdict_del(parent_options, bdref_key);
2324 free_exit:
2325 g_free(backing_filename);
2326 QDECREF(tmp_parent_options);
2327 return ret;
2330 static BlockDriverState *
2331 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2332 BlockDriverState *parent, const BdrvChildRole *child_role,
2333 bool allow_none, Error **errp)
2335 BlockDriverState *bs = NULL;
2336 QDict *image_options;
2337 char *bdref_key_dot;
2338 const char *reference;
2340 assert(child_role != NULL);
2342 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2343 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2344 g_free(bdref_key_dot);
2347 * Caution: while qdict_get_try_str() is fine, getting non-string
2348 * types would require more care. When @options come from
2349 * -blockdev or blockdev_add, its members are typed according to
2350 * the QAPI schema, but when they come from -drive, they're all
2351 * QString.
2353 reference = qdict_get_try_str(options, bdref_key);
2354 if (!filename && !reference && !qdict_size(image_options)) {
2355 if (!allow_none) {
2356 error_setg(errp, "A block device must be specified for \"%s\"",
2357 bdref_key);
2359 QDECREF(image_options);
2360 goto done;
2363 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2364 parent, child_role, errp);
2365 if (!bs) {
2366 goto done;
2369 done:
2370 qdict_del(options, bdref_key);
2371 return bs;
2375 * Opens a disk image whose options are given as BlockdevRef in another block
2376 * device's options.
2378 * If allow_none is true, no image will be opened if filename is false and no
2379 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2381 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2382 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2383 * itself, all options starting with "${bdref_key}." are considered part of the
2384 * BlockdevRef.
2386 * The BlockdevRef will be removed from the options QDict.
2388 BdrvChild *bdrv_open_child(const char *filename,
2389 QDict *options, const char *bdref_key,
2390 BlockDriverState *parent,
2391 const BdrvChildRole *child_role,
2392 bool allow_none, Error **errp)
2394 BdrvChild *c;
2395 BlockDriverState *bs;
2397 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2398 allow_none, errp);
2399 if (bs == NULL) {
2400 return NULL;
2403 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2404 if (!c) {
2405 bdrv_unref(bs);
2406 return NULL;
2409 return c;
2412 /* TODO Future callers may need to specify parent/child_role in order for
2413 * option inheritance to work. Existing callers use it for the root node. */
2414 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2416 BlockDriverState *bs = NULL;
2417 Error *local_err = NULL;
2418 QObject *obj = NULL;
2419 QDict *qdict = NULL;
2420 const char *reference = NULL;
2421 Visitor *v = NULL;
2423 if (ref->type == QTYPE_QSTRING) {
2424 reference = ref->u.reference;
2425 } else {
2426 BlockdevOptions *options = &ref->u.definition;
2427 assert(ref->type == QTYPE_QDICT);
2429 v = qobject_output_visitor_new(&obj);
2430 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2431 if (local_err) {
2432 error_propagate(errp, local_err);
2433 goto fail;
2435 visit_complete(v, &obj);
2437 qdict = qobject_to(QDict, obj);
2438 qdict_flatten(qdict);
2440 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2441 * compatibility with other callers) rather than what we want as the
2442 * real defaults. Apply the defaults here instead. */
2443 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2444 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2445 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2448 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2449 obj = NULL;
2451 fail:
2452 qobject_decref(obj);
2453 visit_free(v);
2454 return bs;
2457 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2458 int flags,
2459 QDict *snapshot_options,
2460 Error **errp)
2462 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2463 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2464 int64_t total_size;
2465 QemuOpts *opts = NULL;
2466 BlockDriverState *bs_snapshot = NULL;
2467 Error *local_err = NULL;
2468 int ret;
2470 /* if snapshot, we create a temporary backing file and open it
2471 instead of opening 'filename' directly */
2473 /* Get the required size from the image */
2474 total_size = bdrv_getlength(bs);
2475 if (total_size < 0) {
2476 error_setg_errno(errp, -total_size, "Could not get image size");
2477 goto out;
2480 /* Create the temporary image */
2481 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2482 if (ret < 0) {
2483 error_setg_errno(errp, -ret, "Could not get temporary filename");
2484 goto out;
2487 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2488 &error_abort);
2489 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2490 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2491 qemu_opts_del(opts);
2492 if (ret < 0) {
2493 error_prepend(errp, "Could not create temporary overlay '%s': ",
2494 tmp_filename);
2495 goto out;
2498 /* Prepare options QDict for the temporary file */
2499 qdict_put_str(snapshot_options, "file.driver", "file");
2500 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2501 qdict_put_str(snapshot_options, "driver", "qcow2");
2503 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2504 snapshot_options = NULL;
2505 if (!bs_snapshot) {
2506 goto out;
2509 /* bdrv_append() consumes a strong reference to bs_snapshot
2510 * (i.e. it will call bdrv_unref() on it) even on error, so in
2511 * order to be able to return one, we have to increase
2512 * bs_snapshot's refcount here */
2513 bdrv_ref(bs_snapshot);
2514 bdrv_append(bs_snapshot, bs, &local_err);
2515 if (local_err) {
2516 error_propagate(errp, local_err);
2517 bs_snapshot = NULL;
2518 goto out;
2521 out:
2522 QDECREF(snapshot_options);
2523 g_free(tmp_filename);
2524 return bs_snapshot;
2528 * Opens a disk image (raw, qcow2, vmdk, ...)
2530 * options is a QDict of options to pass to the block drivers, or NULL for an
2531 * empty set of options. The reference to the QDict belongs to the block layer
2532 * after the call (even on failure), so if the caller intends to reuse the
2533 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2535 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2536 * If it is not NULL, the referenced BDS will be reused.
2538 * The reference parameter may be used to specify an existing block device which
2539 * should be opened. If specified, neither options nor a filename may be given,
2540 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2542 static BlockDriverState *bdrv_open_inherit(const char *filename,
2543 const char *reference,
2544 QDict *options, int flags,
2545 BlockDriverState *parent,
2546 const BdrvChildRole *child_role,
2547 Error **errp)
2549 int ret;
2550 BlockBackend *file = NULL;
2551 BlockDriverState *bs;
2552 BlockDriver *drv = NULL;
2553 const char *drvname;
2554 const char *backing;
2555 Error *local_err = NULL;
2556 QDict *snapshot_options = NULL;
2557 int snapshot_flags = 0;
2559 assert(!child_role || !flags);
2560 assert(!child_role == !parent);
2562 if (reference) {
2563 bool options_non_empty = options ? qdict_size(options) : false;
2564 QDECREF(options);
2566 if (filename || options_non_empty) {
2567 error_setg(errp, "Cannot reference an existing block device with "
2568 "additional options or a new filename");
2569 return NULL;
2572 bs = bdrv_lookup_bs(reference, reference, errp);
2573 if (!bs) {
2574 return NULL;
2577 bdrv_ref(bs);
2578 return bs;
2581 bs = bdrv_new();
2583 /* NULL means an empty set of options */
2584 if (options == NULL) {
2585 options = qdict_new();
2588 /* json: syntax counts as explicit options, as if in the QDict */
2589 parse_json_protocol(options, &filename, &local_err);
2590 if (local_err) {
2591 goto fail;
2594 bs->explicit_options = qdict_clone_shallow(options);
2596 if (child_role) {
2597 bs->inherits_from = parent;
2598 child_role->inherit_options(&flags, options,
2599 parent->open_flags, parent->options);
2602 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2603 if (local_err) {
2604 goto fail;
2608 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2609 * Caution: getting a boolean member of @options requires care.
2610 * When @options come from -blockdev or blockdev_add, members are
2611 * typed according to the QAPI schema, but when they come from
2612 * -drive, they're all QString.
2614 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2615 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2616 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2617 } else {
2618 flags &= ~BDRV_O_RDWR;
2621 if (flags & BDRV_O_SNAPSHOT) {
2622 snapshot_options = qdict_new();
2623 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2624 flags, options);
2625 /* Let bdrv_backing_options() override "read-only" */
2626 qdict_del(options, BDRV_OPT_READ_ONLY);
2627 bdrv_backing_options(&flags, options, flags, options);
2630 bs->open_flags = flags;
2631 bs->options = options;
2632 options = qdict_clone_shallow(options);
2634 /* Find the right image format driver */
2635 /* See cautionary note on accessing @options above */
2636 drvname = qdict_get_try_str(options, "driver");
2637 if (drvname) {
2638 drv = bdrv_find_format(drvname);
2639 if (!drv) {
2640 error_setg(errp, "Unknown driver: '%s'", drvname);
2641 goto fail;
2645 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2647 /* See cautionary note on accessing @options above */
2648 backing = qdict_get_try_str(options, "backing");
2649 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2650 (backing && *backing == '\0'))
2652 if (backing) {
2653 warn_report("Use of \"backing\": \"\" is deprecated; "
2654 "use \"backing\": null instead");
2656 flags |= BDRV_O_NO_BACKING;
2657 qdict_del(options, "backing");
2660 /* Open image file without format layer. This BlockBackend is only used for
2661 * probing, the block drivers will do their own bdrv_open_child() for the
2662 * same BDS, which is why we put the node name back into options. */
2663 if ((flags & BDRV_O_PROTOCOL) == 0) {
2664 BlockDriverState *file_bs;
2666 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2667 &child_file, true, &local_err);
2668 if (local_err) {
2669 goto fail;
2671 if (file_bs != NULL) {
2672 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2673 * looking at the header to guess the image format. This works even
2674 * in cases where a guest would not see a consistent state. */
2675 file = blk_new(0, BLK_PERM_ALL);
2676 blk_insert_bs(file, file_bs, &local_err);
2677 bdrv_unref(file_bs);
2678 if (local_err) {
2679 goto fail;
2682 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2686 /* Image format probing */
2687 bs->probed = !drv;
2688 if (!drv && file) {
2689 ret = find_image_format(file, filename, &drv, &local_err);
2690 if (ret < 0) {
2691 goto fail;
2694 * This option update would logically belong in bdrv_fill_options(),
2695 * but we first need to open bs->file for the probing to work, while
2696 * opening bs->file already requires the (mostly) final set of options
2697 * so that cache mode etc. can be inherited.
2699 * Adding the driver later is somewhat ugly, but it's not an option
2700 * that would ever be inherited, so it's correct. We just need to make
2701 * sure to update both bs->options (which has the full effective
2702 * options for bs) and options (which has file.* already removed).
2704 qdict_put_str(bs->options, "driver", drv->format_name);
2705 qdict_put_str(options, "driver", drv->format_name);
2706 } else if (!drv) {
2707 error_setg(errp, "Must specify either driver or file");
2708 goto fail;
2711 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2712 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2713 /* file must be NULL if a protocol BDS is about to be created
2714 * (the inverse results in an error message from bdrv_open_common()) */
2715 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2717 /* Open the image */
2718 ret = bdrv_open_common(bs, file, options, &local_err);
2719 if (ret < 0) {
2720 goto fail;
2723 if (file) {
2724 blk_unref(file);
2725 file = NULL;
2728 /* If there is a backing file, use it */
2729 if ((flags & BDRV_O_NO_BACKING) == 0) {
2730 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2731 if (ret < 0) {
2732 goto close_and_fail;
2736 bdrv_refresh_filename(bs);
2738 /* Check if any unknown options were used */
2739 if (qdict_size(options) != 0) {
2740 const QDictEntry *entry = qdict_first(options);
2741 if (flags & BDRV_O_PROTOCOL) {
2742 error_setg(errp, "Block protocol '%s' doesn't support the option "
2743 "'%s'", drv->format_name, entry->key);
2744 } else {
2745 error_setg(errp,
2746 "Block format '%s' does not support the option '%s'",
2747 drv->format_name, entry->key);
2750 goto close_and_fail;
2753 bdrv_parent_cb_change_media(bs, true);
2755 QDECREF(options);
2757 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2758 * temporary snapshot afterwards. */
2759 if (snapshot_flags) {
2760 BlockDriverState *snapshot_bs;
2761 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2762 snapshot_options, &local_err);
2763 snapshot_options = NULL;
2764 if (local_err) {
2765 goto close_and_fail;
2767 /* We are not going to return bs but the overlay on top of it
2768 * (snapshot_bs); thus, we have to drop the strong reference to bs
2769 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2770 * though, because the overlay still has a reference to it. */
2771 bdrv_unref(bs);
2772 bs = snapshot_bs;
2775 return bs;
2777 fail:
2778 blk_unref(file);
2779 QDECREF(snapshot_options);
2780 QDECREF(bs->explicit_options);
2781 QDECREF(bs->options);
2782 QDECREF(options);
2783 bs->options = NULL;
2784 bs->explicit_options = NULL;
2785 bdrv_unref(bs);
2786 error_propagate(errp, local_err);
2787 return NULL;
2789 close_and_fail:
2790 bdrv_unref(bs);
2791 QDECREF(snapshot_options);
2792 QDECREF(options);
2793 error_propagate(errp, local_err);
2794 return NULL;
2797 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2798 QDict *options, int flags, Error **errp)
2800 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2801 NULL, errp);
2805 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2806 * reopen of multiple devices.
2808 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2809 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2810 * be created and initialized. This newly created BlockReopenQueue should be
2811 * passed back in for subsequent calls that are intended to be of the same
2812 * atomic 'set'.
2814 * bs is the BlockDriverState to add to the reopen queue.
2816 * options contains the changed options for the associated bs
2817 * (the BlockReopenQueue takes ownership)
2819 * flags contains the open flags for the associated bs
2821 * returns a pointer to bs_queue, which is either the newly allocated
2822 * bs_queue, or the existing bs_queue being used.
2824 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2826 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2827 BlockDriverState *bs,
2828 QDict *options,
2829 int flags,
2830 const BdrvChildRole *role,
2831 QDict *parent_options,
2832 int parent_flags)
2834 assert(bs != NULL);
2836 BlockReopenQueueEntry *bs_entry;
2837 BdrvChild *child;
2838 QDict *old_options, *explicit_options;
2840 /* Make sure that the caller remembered to use a drained section. This is
2841 * important to avoid graph changes between the recursive queuing here and
2842 * bdrv_reopen_multiple(). */
2843 assert(bs->quiesce_counter > 0);
2845 if (bs_queue == NULL) {
2846 bs_queue = g_new0(BlockReopenQueue, 1);
2847 QSIMPLEQ_INIT(bs_queue);
2850 if (!options) {
2851 options = qdict_new();
2854 /* Check if this BlockDriverState is already in the queue */
2855 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2856 if (bs == bs_entry->state.bs) {
2857 break;
2862 * Precedence of options:
2863 * 1. Explicitly passed in options (highest)
2864 * 2. Set in flags (only for top level)
2865 * 3. Retained from explicitly set options of bs
2866 * 4. Inherited from parent node
2867 * 5. Retained from effective options of bs
2870 if (!parent_options) {
2872 * Any setting represented by flags is always updated. If the
2873 * corresponding QDict option is set, it takes precedence. Otherwise
2874 * the flag is translated into a QDict option. The old setting of bs is
2875 * not considered.
2877 update_options_from_flags(options, flags);
2880 /* Old explicitly set values (don't overwrite by inherited value) */
2881 if (bs_entry) {
2882 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2883 } else {
2884 old_options = qdict_clone_shallow(bs->explicit_options);
2886 bdrv_join_options(bs, options, old_options);
2887 QDECREF(old_options);
2889 explicit_options = qdict_clone_shallow(options);
2891 /* Inherit from parent node */
2892 if (parent_options) {
2893 QemuOpts *opts;
2894 QDict *options_copy;
2895 assert(!flags);
2896 role->inherit_options(&flags, options, parent_flags, parent_options);
2897 options_copy = qdict_clone_shallow(options);
2898 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2899 qemu_opts_absorb_qdict(opts, options_copy, NULL);
2900 update_flags_from_options(&flags, opts);
2901 qemu_opts_del(opts);
2902 QDECREF(options_copy);
2905 /* Old values are used for options that aren't set yet */
2906 old_options = qdict_clone_shallow(bs->options);
2907 bdrv_join_options(bs, options, old_options);
2908 QDECREF(old_options);
2910 /* bdrv_open_inherit() sets and clears some additional flags internally */
2911 flags &= ~BDRV_O_PROTOCOL;
2912 if (flags & BDRV_O_RDWR) {
2913 flags |= BDRV_O_ALLOW_RDWR;
2916 if (!bs_entry) {
2917 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2918 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2919 } else {
2920 QDECREF(bs_entry->state.options);
2921 QDECREF(bs_entry->state.explicit_options);
2924 bs_entry->state.bs = bs;
2925 bs_entry->state.options = options;
2926 bs_entry->state.explicit_options = explicit_options;
2927 bs_entry->state.flags = flags;
2929 /* This needs to be overwritten in bdrv_reopen_prepare() */
2930 bs_entry->state.perm = UINT64_MAX;
2931 bs_entry->state.shared_perm = 0;
2933 QLIST_FOREACH(child, &bs->children, next) {
2934 QDict *new_child_options;
2935 char *child_key_dot;
2937 /* reopen can only change the options of block devices that were
2938 * implicitly created and inherited options. For other (referenced)
2939 * block devices, a syntax like "backing.foo" results in an error. */
2940 if (child->bs->inherits_from != bs) {
2941 continue;
2944 child_key_dot = g_strdup_printf("%s.", child->name);
2945 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2946 g_free(child_key_dot);
2948 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2949 child->role, options, flags);
2952 return bs_queue;
2955 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2956 BlockDriverState *bs,
2957 QDict *options, int flags)
2959 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2960 NULL, NULL, 0);
2964 * Reopen multiple BlockDriverStates atomically & transactionally.
2966 * The queue passed in (bs_queue) must have been built up previous
2967 * via bdrv_reopen_queue().
2969 * Reopens all BDS specified in the queue, with the appropriate
2970 * flags. All devices are prepared for reopen, and failure of any
2971 * device will cause all device changes to be abandonded, and intermediate
2972 * data cleaned up.
2974 * If all devices prepare successfully, then the changes are committed
2975 * to all devices.
2977 * All affected nodes must be drained between bdrv_reopen_queue() and
2978 * bdrv_reopen_multiple().
2980 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2982 int ret = -1;
2983 BlockReopenQueueEntry *bs_entry, *next;
2984 Error *local_err = NULL;
2986 assert(bs_queue != NULL);
2988 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2989 assert(bs_entry->state.bs->quiesce_counter > 0);
2990 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2991 error_propagate(errp, local_err);
2992 goto cleanup;
2994 bs_entry->prepared = true;
2997 /* If we reach this point, we have success and just need to apply the
2998 * changes
3000 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3001 bdrv_reopen_commit(&bs_entry->state);
3004 ret = 0;
3006 cleanup:
3007 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3008 if (ret && bs_entry->prepared) {
3009 bdrv_reopen_abort(&bs_entry->state);
3010 } else if (ret) {
3011 QDECREF(bs_entry->state.explicit_options);
3013 QDECREF(bs_entry->state.options);
3014 g_free(bs_entry);
3016 g_free(bs_queue);
3018 return ret;
3022 /* Reopen a single BlockDriverState with the specified flags. */
3023 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
3025 int ret = -1;
3026 Error *local_err = NULL;
3027 BlockReopenQueue *queue;
3029 bdrv_subtree_drained_begin(bs);
3031 queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
3032 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
3033 if (local_err != NULL) {
3034 error_propagate(errp, local_err);
3037 bdrv_subtree_drained_end(bs);
3039 return ret;
3042 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3043 BdrvChild *c)
3045 BlockReopenQueueEntry *entry;
3047 QSIMPLEQ_FOREACH(entry, q, entry) {
3048 BlockDriverState *bs = entry->state.bs;
3049 BdrvChild *child;
3051 QLIST_FOREACH(child, &bs->children, next) {
3052 if (child == c) {
3053 return entry;
3058 return NULL;
3061 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3062 uint64_t *perm, uint64_t *shared)
3064 BdrvChild *c;
3065 BlockReopenQueueEntry *parent;
3066 uint64_t cumulative_perms = 0;
3067 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3069 QLIST_FOREACH(c, &bs->parents, next_parent) {
3070 parent = find_parent_in_reopen_queue(q, c);
3071 if (!parent) {
3072 cumulative_perms |= c->perm;
3073 cumulative_shared_perms &= c->shared_perm;
3074 } else {
3075 uint64_t nperm, nshared;
3077 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3078 parent->state.perm, parent->state.shared_perm,
3079 &nperm, &nshared);
3081 cumulative_perms |= nperm;
3082 cumulative_shared_perms &= nshared;
3085 *perm = cumulative_perms;
3086 *shared = cumulative_shared_perms;
3090 * Prepares a BlockDriverState for reopen. All changes are staged in the
3091 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3092 * the block driver layer .bdrv_reopen_prepare()
3094 * bs is the BlockDriverState to reopen
3095 * flags are the new open flags
3096 * queue is the reopen queue
3098 * Returns 0 on success, non-zero on error. On error errp will be set
3099 * as well.
3101 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3102 * It is the responsibility of the caller to then call the abort() or
3103 * commit() for any other BDS that have been left in a prepare() state
3106 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3107 Error **errp)
3109 int ret = -1;
3110 Error *local_err = NULL;
3111 BlockDriver *drv;
3112 QemuOpts *opts;
3113 const char *value;
3114 bool read_only;
3116 assert(reopen_state != NULL);
3117 assert(reopen_state->bs->drv != NULL);
3118 drv = reopen_state->bs->drv;
3120 /* Process generic block layer options */
3121 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3122 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3123 if (local_err) {
3124 error_propagate(errp, local_err);
3125 ret = -EINVAL;
3126 goto error;
3129 update_flags_from_options(&reopen_state->flags, opts);
3131 /* node-name and driver must be unchanged. Put them back into the QDict, so
3132 * that they are checked at the end of this function. */
3133 value = qemu_opt_get(opts, "node-name");
3134 if (value) {
3135 qdict_put_str(reopen_state->options, "node-name", value);
3138 value = qemu_opt_get(opts, "driver");
3139 if (value) {
3140 qdict_put_str(reopen_state->options, "driver", value);
3143 /* If we are to stay read-only, do not allow permission change
3144 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3145 * not set, or if the BDS still has copy_on_read enabled */
3146 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3147 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3148 if (local_err) {
3149 error_propagate(errp, local_err);
3150 goto error;
3153 /* Calculate required permissions after reopening */
3154 bdrv_reopen_perm(queue, reopen_state->bs,
3155 &reopen_state->perm, &reopen_state->shared_perm);
3157 ret = bdrv_flush(reopen_state->bs);
3158 if (ret) {
3159 error_setg_errno(errp, -ret, "Error flushing drive");
3160 goto error;
3163 if (drv->bdrv_reopen_prepare) {
3164 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3165 if (ret) {
3166 if (local_err != NULL) {
3167 error_propagate(errp, local_err);
3168 } else {
3169 error_setg(errp, "failed while preparing to reopen image '%s'",
3170 reopen_state->bs->filename);
3172 goto error;
3174 } else {
3175 /* It is currently mandatory to have a bdrv_reopen_prepare()
3176 * handler for each supported drv. */
3177 error_setg(errp, "Block format '%s' used by node '%s' "
3178 "does not support reopening files", drv->format_name,
3179 bdrv_get_device_or_node_name(reopen_state->bs));
3180 ret = -1;
3181 goto error;
3184 /* Options that are not handled are only okay if they are unchanged
3185 * compared to the old state. It is expected that some options are only
3186 * used for the initial open, but not reopen (e.g. filename) */
3187 if (qdict_size(reopen_state->options)) {
3188 const QDictEntry *entry = qdict_first(reopen_state->options);
3190 do {
3191 QObject *new = entry->value;
3192 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3195 * TODO: When using -drive to specify blockdev options, all values
3196 * will be strings; however, when using -blockdev, blockdev-add or
3197 * filenames using the json:{} pseudo-protocol, they will be
3198 * correctly typed.
3199 * In contrast, reopening options are (currently) always strings
3200 * (because you can only specify them through qemu-io; all other
3201 * callers do not specify any options).
3202 * Therefore, when using anything other than -drive to create a BDS,
3203 * this cannot detect non-string options as unchanged, because
3204 * qobject_is_equal() always returns false for objects of different
3205 * type. In the future, this should be remedied by correctly typing
3206 * all options. For now, this is not too big of an issue because
3207 * the user can simply omit options which cannot be changed anyway,
3208 * so they will stay unchanged.
3210 if (!qobject_is_equal(new, old)) {
3211 error_setg(errp, "Cannot change the option '%s'", entry->key);
3212 ret = -EINVAL;
3213 goto error;
3215 } while ((entry = qdict_next(reopen_state->options, entry)));
3218 ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3219 reopen_state->shared_perm, NULL, errp);
3220 if (ret < 0) {
3221 goto error;
3224 ret = 0;
3226 error:
3227 qemu_opts_del(opts);
3228 return ret;
3232 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3233 * makes them final by swapping the staging BlockDriverState contents into
3234 * the active BlockDriverState contents.
3236 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3238 BlockDriver *drv;
3239 BlockDriverState *bs;
3240 bool old_can_write, new_can_write;
3242 assert(reopen_state != NULL);
3243 bs = reopen_state->bs;
3244 drv = bs->drv;
3245 assert(drv != NULL);
3247 old_can_write =
3248 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3250 /* If there are any driver level actions to take */
3251 if (drv->bdrv_reopen_commit) {
3252 drv->bdrv_reopen_commit(reopen_state);
3255 /* set BDS specific flags now */
3256 QDECREF(bs->explicit_options);
3258 bs->explicit_options = reopen_state->explicit_options;
3259 bs->open_flags = reopen_state->flags;
3260 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3262 bdrv_refresh_limits(bs, NULL);
3264 bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3265 reopen_state->shared_perm);
3267 new_can_write =
3268 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3269 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3270 Error *local_err = NULL;
3271 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3272 /* This is not fatal, bitmaps just left read-only, so all following
3273 * writes will fail. User can remove read-only bitmaps to unblock
3274 * writes.
3276 error_reportf_err(local_err,
3277 "%s: Failed to make dirty bitmaps writable: ",
3278 bdrv_get_node_name(bs));
3284 * Abort the reopen, and delete and free the staged changes in
3285 * reopen_state
3287 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3289 BlockDriver *drv;
3291 assert(reopen_state != NULL);
3292 drv = reopen_state->bs->drv;
3293 assert(drv != NULL);
3295 if (drv->bdrv_reopen_abort) {
3296 drv->bdrv_reopen_abort(reopen_state);
3299 QDECREF(reopen_state->explicit_options);
3301 bdrv_abort_perm_update(reopen_state->bs);
3305 static void bdrv_close(BlockDriverState *bs)
3307 BdrvAioNotifier *ban, *ban_next;
3308 BdrvChild *child, *next;
3310 assert(!bs->job);
3311 assert(!bs->refcnt);
3313 bdrv_drained_begin(bs); /* complete I/O */
3314 bdrv_flush(bs);
3315 bdrv_drain(bs); /* in case flush left pending I/O */
3317 if (bs->drv) {
3318 bs->drv->bdrv_close(bs);
3319 bs->drv = NULL;
3322 bdrv_set_backing_hd(bs, NULL, &error_abort);
3324 if (bs->file != NULL) {
3325 bdrv_unref_child(bs, bs->file);
3326 bs->file = NULL;
3329 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3330 /* TODO Remove bdrv_unref() from drivers' close function and use
3331 * bdrv_unref_child() here */
3332 if (child->bs->inherits_from == bs) {
3333 child->bs->inherits_from = NULL;
3335 bdrv_detach_child(child);
3338 g_free(bs->opaque);
3339 bs->opaque = NULL;
3340 atomic_set(&bs->copy_on_read, 0);
3341 bs->backing_file[0] = '\0';
3342 bs->backing_format[0] = '\0';
3343 bs->total_sectors = 0;
3344 bs->encrypted = false;
3345 bs->sg = false;
3346 QDECREF(bs->options);
3347 QDECREF(bs->explicit_options);
3348 bs->options = NULL;
3349 bs->explicit_options = NULL;
3350 QDECREF(bs->full_open_options);
3351 bs->full_open_options = NULL;
3353 bdrv_release_named_dirty_bitmaps(bs);
3354 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3356 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3357 g_free(ban);
3359 QLIST_INIT(&bs->aio_notifiers);
3360 bdrv_drained_end(bs);
3363 void bdrv_close_all(void)
3365 block_job_cancel_sync_all();
3366 nbd_export_close_all();
3368 /* Drop references from requests still in flight, such as canceled block
3369 * jobs whose AIO context has not been polled yet */
3370 bdrv_drain_all();
3372 blk_remove_all_bs();
3373 blockdev_close_all_bdrv_states();
3375 assert(QTAILQ_EMPTY(&all_bdrv_states));
3378 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3380 BdrvChild *to_c;
3382 if (c->role->stay_at_node) {
3383 return false;
3386 if (c->role == &child_backing) {
3387 /* If @from is a backing file of @to, ignore the child to avoid
3388 * creating a loop. We only want to change the pointer of other
3389 * parents. */
3390 QLIST_FOREACH(to_c, &to->children, next) {
3391 if (to_c == c) {
3392 break;
3395 if (to_c) {
3396 return false;
3400 return true;
3403 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3404 Error **errp)
3406 BdrvChild *c, *next;
3407 GSList *list = NULL, *p;
3408 uint64_t old_perm, old_shared;
3409 uint64_t perm = 0, shared = BLK_PERM_ALL;
3410 int ret;
3412 assert(!atomic_read(&from->in_flight));
3413 assert(!atomic_read(&to->in_flight));
3415 /* Make sure that @from doesn't go away until we have successfully attached
3416 * all of its parents to @to. */
3417 bdrv_ref(from);
3419 /* Put all parents into @list and calculate their cumulative permissions */
3420 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3421 if (!should_update_child(c, to)) {
3422 continue;
3424 list = g_slist_prepend(list, c);
3425 perm |= c->perm;
3426 shared &= c->shared_perm;
3429 /* Check whether the required permissions can be granted on @to, ignoring
3430 * all BdrvChild in @list so that they can't block themselves. */
3431 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3432 if (ret < 0) {
3433 bdrv_abort_perm_update(to);
3434 goto out;
3437 /* Now actually perform the change. We performed the permission check for
3438 * all elements of @list at once, so set the permissions all at once at the
3439 * very end. */
3440 for (p = list; p != NULL; p = p->next) {
3441 c = p->data;
3443 bdrv_ref(to);
3444 bdrv_replace_child_noperm(c, to);
3445 bdrv_unref(from);
3448 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3449 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3451 out:
3452 g_slist_free(list);
3453 bdrv_unref(from);
3457 * Add new bs contents at the top of an image chain while the chain is
3458 * live, while keeping required fields on the top layer.
3460 * This will modify the BlockDriverState fields, and swap contents
3461 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3463 * bs_new must not be attached to a BlockBackend.
3465 * This function does not create any image files.
3467 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3468 * that's what the callers commonly need. bs_new will be referenced by the old
3469 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3470 * reference of its own, it must call bdrv_ref().
3472 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3473 Error **errp)
3475 Error *local_err = NULL;
3477 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3478 if (local_err) {
3479 error_propagate(errp, local_err);
3480 goto out;
3483 bdrv_replace_node(bs_top, bs_new, &local_err);
3484 if (local_err) {
3485 error_propagate(errp, local_err);
3486 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3487 goto out;
3490 /* bs_new is now referenced by its new parents, we don't need the
3491 * additional reference any more. */
3492 out:
3493 bdrv_unref(bs_new);
3496 static void bdrv_delete(BlockDriverState *bs)
3498 assert(!bs->job);
3499 assert(bdrv_op_blocker_is_empty(bs));
3500 assert(!bs->refcnt);
3502 bdrv_close(bs);
3504 /* remove from list, if necessary */
3505 if (bs->node_name[0] != '\0') {
3506 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3508 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3510 g_free(bs);
3514 * Run consistency checks on an image
3516 * Returns 0 if the check could be completed (it doesn't mean that the image is
3517 * free of errors) or -errno when an internal error occurred. The results of the
3518 * check are stored in res.
3520 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3521 BdrvCheckResult *res, BdrvCheckMode fix)
3523 if (bs->drv == NULL) {
3524 return -ENOMEDIUM;
3526 if (bs->drv->bdrv_co_check == NULL) {
3527 return -ENOTSUP;
3530 memset(res, 0, sizeof(*res));
3531 return bs->drv->bdrv_co_check(bs, res, fix);
3534 typedef struct CheckCo {
3535 BlockDriverState *bs;
3536 BdrvCheckResult *res;
3537 BdrvCheckMode fix;
3538 int ret;
3539 } CheckCo;
3541 static void bdrv_check_co_entry(void *opaque)
3543 CheckCo *cco = opaque;
3544 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3547 int bdrv_check(BlockDriverState *bs,
3548 BdrvCheckResult *res, BdrvCheckMode fix)
3550 Coroutine *co;
3551 CheckCo cco = {
3552 .bs = bs,
3553 .res = res,
3554 .ret = -EINPROGRESS,
3555 .fix = fix,
3558 if (qemu_in_coroutine()) {
3559 /* Fast-path if already in coroutine context */
3560 bdrv_check_co_entry(&cco);
3561 } else {
3562 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3563 qemu_coroutine_enter(co);
3564 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3567 return cco.ret;
3571 * Return values:
3572 * 0 - success
3573 * -EINVAL - backing format specified, but no file
3574 * -ENOSPC - can't update the backing file because no space is left in the
3575 * image file header
3576 * -ENOTSUP - format driver doesn't support changing the backing file
3578 int bdrv_change_backing_file(BlockDriverState *bs,
3579 const char *backing_file, const char *backing_fmt)
3581 BlockDriver *drv = bs->drv;
3582 int ret;
3584 if (!drv) {
3585 return -ENOMEDIUM;
3588 /* Backing file format doesn't make sense without a backing file */
3589 if (backing_fmt && !backing_file) {
3590 return -EINVAL;
3593 if (drv->bdrv_change_backing_file != NULL) {
3594 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3595 } else {
3596 ret = -ENOTSUP;
3599 if (ret == 0) {
3600 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3601 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3603 return ret;
3607 * Finds the image layer in the chain that has 'bs' as its backing file.
3609 * active is the current topmost image.
3611 * Returns NULL if bs is not found in active's image chain,
3612 * or if active == bs.
3614 * Returns the bottommost base image if bs == NULL.
3616 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3617 BlockDriverState *bs)
3619 while (active && bs != backing_bs(active)) {
3620 active = backing_bs(active);
3623 return active;
3626 /* Given a BDS, searches for the base layer. */
3627 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3629 return bdrv_find_overlay(bs, NULL);
3633 * Drops images above 'base' up to and including 'top', and sets the image
3634 * above 'top' to have base as its backing file.
3636 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3637 * information in 'bs' can be properly updated.
3639 * E.g., this will convert the following chain:
3640 * bottom <- base <- intermediate <- top <- active
3642 * to
3644 * bottom <- base <- active
3646 * It is allowed for bottom==base, in which case it converts:
3648 * base <- intermediate <- top <- active
3650 * to
3652 * base <- active
3654 * If backing_file_str is non-NULL, it will be used when modifying top's
3655 * overlay image metadata.
3657 * Error conditions:
3658 * if active == top, that is considered an error
3661 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3662 const char *backing_file_str)
3664 BdrvChild *c, *next;
3665 Error *local_err = NULL;
3666 int ret = -EIO;
3668 bdrv_ref(top);
3670 if (!top->drv || !base->drv) {
3671 goto exit;
3674 /* Make sure that base is in the backing chain of top */
3675 if (!bdrv_chain_contains(top, base)) {
3676 goto exit;
3679 /* success - we can delete the intermediate states, and link top->base */
3680 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3681 * we've figured out how they should work. */
3682 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3684 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3685 /* Check whether we are allowed to switch c from top to base */
3686 GSList *ignore_children = g_slist_prepend(NULL, c);
3687 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3688 ignore_children, &local_err);
3689 g_slist_free(ignore_children);
3690 if (local_err) {
3691 ret = -EPERM;
3692 error_report_err(local_err);
3693 goto exit;
3696 /* If so, update the backing file path in the image file */
3697 if (c->role->update_filename) {
3698 ret = c->role->update_filename(c, base, backing_file_str,
3699 &local_err);
3700 if (ret < 0) {
3701 bdrv_abort_perm_update(base);
3702 error_report_err(local_err);
3703 goto exit;
3707 /* Do the actual switch in the in-memory graph.
3708 * Completes bdrv_check_update_perm() transaction internally. */
3709 bdrv_ref(base);
3710 bdrv_replace_child(c, base);
3711 bdrv_unref(top);
3714 ret = 0;
3715 exit:
3716 bdrv_unref(top);
3717 return ret;
3721 * Truncate file to 'offset' bytes (needed only for file protocols)
3723 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
3724 Error **errp)
3726 BlockDriverState *bs = child->bs;
3727 BlockDriver *drv = bs->drv;
3728 int ret;
3730 assert(child->perm & BLK_PERM_RESIZE);
3732 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3733 if (!drv) {
3734 error_setg(errp, "No medium inserted");
3735 return -ENOMEDIUM;
3737 if (offset < 0) {
3738 error_setg(errp, "Image size cannot be negative");
3739 return -EINVAL;
3742 if (!drv->bdrv_truncate) {
3743 if (bs->file && drv->is_filter) {
3744 return bdrv_truncate(bs->file, offset, prealloc, errp);
3746 error_setg(errp, "Image format driver does not support resize");
3747 return -ENOTSUP;
3749 if (bs->read_only) {
3750 error_setg(errp, "Image is read-only");
3751 return -EACCES;
3754 assert(!(bs->open_flags & BDRV_O_INACTIVE));
3756 ret = drv->bdrv_truncate(bs, offset, prealloc, errp);
3757 if (ret < 0) {
3758 return ret;
3760 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3761 if (ret < 0) {
3762 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3763 } else {
3764 offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3766 bdrv_dirty_bitmap_truncate(bs, offset);
3767 bdrv_parent_cb_resize(bs);
3768 atomic_inc(&bs->write_gen);
3769 return ret;
3773 * Length of a allocated file in bytes. Sparse files are counted by actual
3774 * allocated space. Return < 0 if error or unknown.
3776 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3778 BlockDriver *drv = bs->drv;
3779 if (!drv) {
3780 return -ENOMEDIUM;
3782 if (drv->bdrv_get_allocated_file_size) {
3783 return drv->bdrv_get_allocated_file_size(bs);
3785 if (bs->file) {
3786 return bdrv_get_allocated_file_size(bs->file->bs);
3788 return -ENOTSUP;
3792 * bdrv_measure:
3793 * @drv: Format driver
3794 * @opts: Creation options for new image
3795 * @in_bs: Existing image containing data for new image (may be NULL)
3796 * @errp: Error object
3797 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3798 * or NULL on error
3800 * Calculate file size required to create a new image.
3802 * If @in_bs is given then space for allocated clusters and zero clusters
3803 * from that image are included in the calculation. If @opts contains a
3804 * backing file that is shared by @in_bs then backing clusters may be omitted
3805 * from the calculation.
3807 * If @in_bs is NULL then the calculation includes no allocated clusters
3808 * unless a preallocation option is given in @opts.
3810 * Note that @in_bs may use a different BlockDriver from @drv.
3812 * If an error occurs the @errp pointer is set.
3814 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3815 BlockDriverState *in_bs, Error **errp)
3817 if (!drv->bdrv_measure) {
3818 error_setg(errp, "Block driver '%s' does not support size measurement",
3819 drv->format_name);
3820 return NULL;
3823 return drv->bdrv_measure(opts, in_bs, errp);
3827 * Return number of sectors on success, -errno on error.
3829 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3831 BlockDriver *drv = bs->drv;
3833 if (!drv)
3834 return -ENOMEDIUM;
3836 if (drv->has_variable_length) {
3837 int ret = refresh_total_sectors(bs, bs->total_sectors);
3838 if (ret < 0) {
3839 return ret;
3842 return bs->total_sectors;
3846 * Return length in bytes on success, -errno on error.
3847 * The length is always a multiple of BDRV_SECTOR_SIZE.
3849 int64_t bdrv_getlength(BlockDriverState *bs)
3851 int64_t ret = bdrv_nb_sectors(bs);
3853 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3854 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3857 /* return 0 as number of sectors if no device present or error */
3858 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3860 int64_t nb_sectors = bdrv_nb_sectors(bs);
3862 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3865 bool bdrv_is_sg(BlockDriverState *bs)
3867 return bs->sg;
3870 bool bdrv_is_encrypted(BlockDriverState *bs)
3872 if (bs->backing && bs->backing->bs->encrypted) {
3873 return true;
3875 return bs->encrypted;
3878 const char *bdrv_get_format_name(BlockDriverState *bs)
3880 return bs->drv ? bs->drv->format_name : NULL;
3883 static int qsort_strcmp(const void *a, const void *b)
3885 return strcmp(*(char *const *)a, *(char *const *)b);
3888 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3889 void *opaque)
3891 BlockDriver *drv;
3892 int count = 0;
3893 int i;
3894 const char **formats = NULL;
3896 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3897 if (drv->format_name) {
3898 bool found = false;
3899 int i = count;
3900 while (formats && i && !found) {
3901 found = !strcmp(formats[--i], drv->format_name);
3904 if (!found) {
3905 formats = g_renew(const char *, formats, count + 1);
3906 formats[count++] = drv->format_name;
3911 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3912 const char *format_name = block_driver_modules[i].format_name;
3914 if (format_name) {
3915 bool found = false;
3916 int j = count;
3918 while (formats && j && !found) {
3919 found = !strcmp(formats[--j], format_name);
3922 if (!found) {
3923 formats = g_renew(const char *, formats, count + 1);
3924 formats[count++] = format_name;
3929 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3931 for (i = 0; i < count; i++) {
3932 it(opaque, formats[i]);
3935 g_free(formats);
3938 /* This function is to find a node in the bs graph */
3939 BlockDriverState *bdrv_find_node(const char *node_name)
3941 BlockDriverState *bs;
3943 assert(node_name);
3945 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3946 if (!strcmp(node_name, bs->node_name)) {
3947 return bs;
3950 return NULL;
3953 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3954 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3956 BlockDeviceInfoList *list, *entry;
3957 BlockDriverState *bs;
3959 list = NULL;
3960 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3961 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3962 if (!info) {
3963 qapi_free_BlockDeviceInfoList(list);
3964 return NULL;
3966 entry = g_malloc0(sizeof(*entry));
3967 entry->value = info;
3968 entry->next = list;
3969 list = entry;
3972 return list;
3975 BlockDriverState *bdrv_lookup_bs(const char *device,
3976 const char *node_name,
3977 Error **errp)
3979 BlockBackend *blk;
3980 BlockDriverState *bs;
3982 if (device) {
3983 blk = blk_by_name(device);
3985 if (blk) {
3986 bs = blk_bs(blk);
3987 if (!bs) {
3988 error_setg(errp, "Device '%s' has no medium", device);
3991 return bs;
3995 if (node_name) {
3996 bs = bdrv_find_node(node_name);
3998 if (bs) {
3999 return bs;
4003 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4004 device ? device : "",
4005 node_name ? node_name : "");
4006 return NULL;
4009 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4010 * return false. If either argument is NULL, return false. */
4011 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4013 while (top && top != base) {
4014 top = backing_bs(top);
4017 return top != NULL;
4020 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4022 if (!bs) {
4023 return QTAILQ_FIRST(&graph_bdrv_states);
4025 return QTAILQ_NEXT(bs, node_list);
4028 const char *bdrv_get_node_name(const BlockDriverState *bs)
4030 return bs->node_name;
4033 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4035 BdrvChild *c;
4036 const char *name;
4038 /* If multiple parents have a name, just pick the first one. */
4039 QLIST_FOREACH(c, &bs->parents, next_parent) {
4040 if (c->role->get_name) {
4041 name = c->role->get_name(c);
4042 if (name && *name) {
4043 return name;
4048 return NULL;
4051 /* TODO check what callers really want: bs->node_name or blk_name() */
4052 const char *bdrv_get_device_name(const BlockDriverState *bs)
4054 return bdrv_get_parent_name(bs) ?: "";
4057 /* This can be used to identify nodes that might not have a device
4058 * name associated. Since node and device names live in the same
4059 * namespace, the result is unambiguous. The exception is if both are
4060 * absent, then this returns an empty (non-null) string. */
4061 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4063 return bdrv_get_parent_name(bs) ?: bs->node_name;
4066 int bdrv_get_flags(BlockDriverState *bs)
4068 return bs->open_flags;
4071 int bdrv_has_zero_init_1(BlockDriverState *bs)
4073 return 1;
4076 int bdrv_has_zero_init(BlockDriverState *bs)
4078 if (!bs->drv) {
4079 return 0;
4082 /* If BS is a copy on write image, it is initialized to
4083 the contents of the base image, which may not be zeroes. */
4084 if (bs->backing) {
4085 return 0;
4087 if (bs->drv->bdrv_has_zero_init) {
4088 return bs->drv->bdrv_has_zero_init(bs);
4090 if (bs->file && bs->drv->is_filter) {
4091 return bdrv_has_zero_init(bs->file->bs);
4094 /* safe default */
4095 return 0;
4098 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4100 BlockDriverInfo bdi;
4102 if (bs->backing) {
4103 return false;
4106 if (bdrv_get_info(bs, &bdi) == 0) {
4107 return bdi.unallocated_blocks_are_zero;
4110 return false;
4113 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4115 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4116 return false;
4119 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4122 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4124 if (bs->backing && bs->backing->bs->encrypted)
4125 return bs->backing_file;
4126 else if (bs->encrypted)
4127 return bs->filename;
4128 else
4129 return NULL;
4132 void bdrv_get_backing_filename(BlockDriverState *bs,
4133 char *filename, int filename_size)
4135 pstrcpy(filename, filename_size, bs->backing_file);
4138 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4140 BlockDriver *drv = bs->drv;
4141 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4142 if (!drv) {
4143 return -ENOMEDIUM;
4145 if (!drv->bdrv_get_info) {
4146 if (bs->file && drv->is_filter) {
4147 return bdrv_get_info(bs->file->bs, bdi);
4149 return -ENOTSUP;
4151 memset(bdi, 0, sizeof(*bdi));
4152 return drv->bdrv_get_info(bs, bdi);
4155 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4157 BlockDriver *drv = bs->drv;
4158 if (drv && drv->bdrv_get_specific_info) {
4159 return drv->bdrv_get_specific_info(bs);
4161 return NULL;
4164 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4166 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4167 return;
4170 bs->drv->bdrv_debug_event(bs, event);
4173 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4174 const char *tag)
4176 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4177 bs = bs->file ? bs->file->bs : NULL;
4180 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4181 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4184 return -ENOTSUP;
4187 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4189 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4190 bs = bs->file ? bs->file->bs : NULL;
4193 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4194 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4197 return -ENOTSUP;
4200 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4202 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4203 bs = bs->file ? bs->file->bs : NULL;
4206 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4207 return bs->drv->bdrv_debug_resume(bs, tag);
4210 return -ENOTSUP;
4213 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4215 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4216 bs = bs->file ? bs->file->bs : NULL;
4219 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4220 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4223 return false;
4226 /* backing_file can either be relative, or absolute, or a protocol. If it is
4227 * relative, it must be relative to the chain. So, passing in bs->filename
4228 * from a BDS as backing_file should not be done, as that may be relative to
4229 * the CWD rather than the chain. */
4230 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4231 const char *backing_file)
4233 char *filename_full = NULL;
4234 char *backing_file_full = NULL;
4235 char *filename_tmp = NULL;
4236 int is_protocol = 0;
4237 BlockDriverState *curr_bs = NULL;
4238 BlockDriverState *retval = NULL;
4239 Error *local_error = NULL;
4241 if (!bs || !bs->drv || !backing_file) {
4242 return NULL;
4245 filename_full = g_malloc(PATH_MAX);
4246 backing_file_full = g_malloc(PATH_MAX);
4247 filename_tmp = g_malloc(PATH_MAX);
4249 is_protocol = path_has_protocol(backing_file);
4251 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4253 /* If either of the filename paths is actually a protocol, then
4254 * compare unmodified paths; otherwise make paths relative */
4255 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4256 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4257 retval = curr_bs->backing->bs;
4258 break;
4260 /* Also check against the full backing filename for the image */
4261 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4262 &local_error);
4263 if (local_error == NULL) {
4264 if (strcmp(backing_file, backing_file_full) == 0) {
4265 retval = curr_bs->backing->bs;
4266 break;
4268 } else {
4269 error_free(local_error);
4270 local_error = NULL;
4272 } else {
4273 /* If not an absolute filename path, make it relative to the current
4274 * image's filename path */
4275 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4276 backing_file);
4278 /* We are going to compare absolute pathnames */
4279 if (!realpath(filename_tmp, filename_full)) {
4280 continue;
4283 /* We need to make sure the backing filename we are comparing against
4284 * is relative to the current image filename (or absolute) */
4285 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4286 curr_bs->backing_file);
4288 if (!realpath(filename_tmp, backing_file_full)) {
4289 continue;
4292 if (strcmp(backing_file_full, filename_full) == 0) {
4293 retval = curr_bs->backing->bs;
4294 break;
4299 g_free(filename_full);
4300 g_free(backing_file_full);
4301 g_free(filename_tmp);
4302 return retval;
4305 void bdrv_init(void)
4307 module_call_init(MODULE_INIT_BLOCK);
4310 void bdrv_init_with_whitelist(void)
4312 use_bdrv_whitelist = 1;
4313 bdrv_init();
4316 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4317 Error **errp)
4319 BdrvChild *child, *parent;
4320 uint64_t perm, shared_perm;
4321 Error *local_err = NULL;
4322 int ret;
4324 if (!bs->drv) {
4325 return;
4328 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4329 return;
4332 QLIST_FOREACH(child, &bs->children, next) {
4333 bdrv_co_invalidate_cache(child->bs, &local_err);
4334 if (local_err) {
4335 error_propagate(errp, local_err);
4336 return;
4341 * Update permissions, they may differ for inactive nodes.
4343 * Note that the required permissions of inactive images are always a
4344 * subset of the permissions required after activating the image. This
4345 * allows us to just get the permissions upfront without restricting
4346 * drv->bdrv_invalidate_cache().
4348 * It also means that in error cases, we don't have to try and revert to
4349 * the old permissions (which is an operation that could fail, too). We can
4350 * just keep the extended permissions for the next time that an activation
4351 * of the image is tried.
4353 bs->open_flags &= ~BDRV_O_INACTIVE;
4354 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4355 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4356 if (ret < 0) {
4357 bs->open_flags |= BDRV_O_INACTIVE;
4358 error_propagate(errp, local_err);
4359 return;
4361 bdrv_set_perm(bs, perm, shared_perm);
4363 if (bs->drv->bdrv_co_invalidate_cache) {
4364 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4365 if (local_err) {
4366 bs->open_flags |= BDRV_O_INACTIVE;
4367 error_propagate(errp, local_err);
4368 return;
4372 ret = refresh_total_sectors(bs, bs->total_sectors);
4373 if (ret < 0) {
4374 bs->open_flags |= BDRV_O_INACTIVE;
4375 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4376 return;
4379 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4380 if (parent->role->activate) {
4381 parent->role->activate(parent, &local_err);
4382 if (local_err) {
4383 error_propagate(errp, local_err);
4384 return;
4390 typedef struct InvalidateCacheCo {
4391 BlockDriverState *bs;
4392 Error **errp;
4393 bool done;
4394 } InvalidateCacheCo;
4396 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4398 InvalidateCacheCo *ico = opaque;
4399 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4400 ico->done = true;
4403 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4405 Coroutine *co;
4406 InvalidateCacheCo ico = {
4407 .bs = bs,
4408 .done = false,
4409 .errp = errp
4412 if (qemu_in_coroutine()) {
4413 /* Fast-path if already in coroutine context */
4414 bdrv_invalidate_cache_co_entry(&ico);
4415 } else {
4416 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4417 qemu_coroutine_enter(co);
4418 BDRV_POLL_WHILE(bs, !ico.done);
4422 void bdrv_invalidate_cache_all(Error **errp)
4424 BlockDriverState *bs;
4425 Error *local_err = NULL;
4426 BdrvNextIterator it;
4428 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4429 AioContext *aio_context = bdrv_get_aio_context(bs);
4431 aio_context_acquire(aio_context);
4432 bdrv_invalidate_cache(bs, &local_err);
4433 aio_context_release(aio_context);
4434 if (local_err) {
4435 error_propagate(errp, local_err);
4436 bdrv_next_cleanup(&it);
4437 return;
4442 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4443 bool setting_flag)
4445 BdrvChild *child, *parent;
4446 int ret;
4448 if (!bs->drv) {
4449 return -ENOMEDIUM;
4452 if (!setting_flag && bs->drv->bdrv_inactivate) {
4453 ret = bs->drv->bdrv_inactivate(bs);
4454 if (ret < 0) {
4455 return ret;
4459 if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
4460 uint64_t perm, shared_perm;
4462 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4463 if (parent->role->inactivate) {
4464 ret = parent->role->inactivate(parent);
4465 if (ret < 0) {
4466 return ret;
4471 bs->open_flags |= BDRV_O_INACTIVE;
4473 /* Update permissions, they may differ for inactive nodes */
4474 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4475 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4476 bdrv_set_perm(bs, perm, shared_perm);
4479 QLIST_FOREACH(child, &bs->children, next) {
4480 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4481 if (ret < 0) {
4482 return ret;
4486 /* At this point persistent bitmaps should be already stored by the format
4487 * driver */
4488 bdrv_release_persistent_dirty_bitmaps(bs);
4490 return 0;
4493 int bdrv_inactivate_all(void)
4495 BlockDriverState *bs = NULL;
4496 BdrvNextIterator it;
4497 int ret = 0;
4498 int pass;
4499 GSList *aio_ctxs = NULL, *ctx;
4501 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4502 AioContext *aio_context = bdrv_get_aio_context(bs);
4504 if (!g_slist_find(aio_ctxs, aio_context)) {
4505 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4506 aio_context_acquire(aio_context);
4510 /* We do two passes of inactivation. The first pass calls to drivers'
4511 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4512 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4513 * is allowed. */
4514 for (pass = 0; pass < 2; pass++) {
4515 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4516 ret = bdrv_inactivate_recurse(bs, pass);
4517 if (ret < 0) {
4518 bdrv_next_cleanup(&it);
4519 goto out;
4524 out:
4525 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4526 AioContext *aio_context = ctx->data;
4527 aio_context_release(aio_context);
4529 g_slist_free(aio_ctxs);
4531 return ret;
4534 /**************************************************************/
4535 /* removable device support */
4538 * Return TRUE if the media is present
4540 bool bdrv_is_inserted(BlockDriverState *bs)
4542 BlockDriver *drv = bs->drv;
4543 BdrvChild *child;
4545 if (!drv) {
4546 return false;
4548 if (drv->bdrv_is_inserted) {
4549 return drv->bdrv_is_inserted(bs);
4551 QLIST_FOREACH(child, &bs->children, next) {
4552 if (!bdrv_is_inserted(child->bs)) {
4553 return false;
4556 return true;
4560 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4562 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4564 BlockDriver *drv = bs->drv;
4566 if (drv && drv->bdrv_eject) {
4567 drv->bdrv_eject(bs, eject_flag);
4572 * Lock or unlock the media (if it is locked, the user won't be able
4573 * to eject it manually).
4575 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4577 BlockDriver *drv = bs->drv;
4579 trace_bdrv_lock_medium(bs, locked);
4581 if (drv && drv->bdrv_lock_medium) {
4582 drv->bdrv_lock_medium(bs, locked);
4586 /* Get a reference to bs */
4587 void bdrv_ref(BlockDriverState *bs)
4589 bs->refcnt++;
4592 /* Release a previously grabbed reference to bs.
4593 * If after releasing, reference count is zero, the BlockDriverState is
4594 * deleted. */
4595 void bdrv_unref(BlockDriverState *bs)
4597 if (!bs) {
4598 return;
4600 assert(bs->refcnt > 0);
4601 if (--bs->refcnt == 0) {
4602 bdrv_delete(bs);
4606 struct BdrvOpBlocker {
4607 Error *reason;
4608 QLIST_ENTRY(BdrvOpBlocker) list;
4611 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4613 BdrvOpBlocker *blocker;
4614 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4615 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4616 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4617 error_propagate(errp, error_copy(blocker->reason));
4618 error_prepend(errp, "Node '%s' is busy: ",
4619 bdrv_get_device_or_node_name(bs));
4620 return true;
4622 return false;
4625 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4627 BdrvOpBlocker *blocker;
4628 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4630 blocker = g_new0(BdrvOpBlocker, 1);
4631 blocker->reason = reason;
4632 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4635 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4637 BdrvOpBlocker *blocker, *next;
4638 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4639 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4640 if (blocker->reason == reason) {
4641 QLIST_REMOVE(blocker, list);
4642 g_free(blocker);
4647 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4649 int i;
4650 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4651 bdrv_op_block(bs, i, reason);
4655 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4657 int i;
4658 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4659 bdrv_op_unblock(bs, i, reason);
4663 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4665 int i;
4667 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4668 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4669 return false;
4672 return true;
4675 void bdrv_img_create(const char *filename, const char *fmt,
4676 const char *base_filename, const char *base_fmt,
4677 char *options, uint64_t img_size, int flags, bool quiet,
4678 Error **errp)
4680 QemuOptsList *create_opts = NULL;
4681 QemuOpts *opts = NULL;
4682 const char *backing_fmt, *backing_file;
4683 int64_t size;
4684 BlockDriver *drv, *proto_drv;
4685 Error *local_err = NULL;
4686 int ret = 0;
4688 /* Find driver and parse its options */
4689 drv = bdrv_find_format(fmt);
4690 if (!drv) {
4691 error_setg(errp, "Unknown file format '%s'", fmt);
4692 return;
4695 proto_drv = bdrv_find_protocol(filename, true, errp);
4696 if (!proto_drv) {
4697 return;
4700 if (!drv->create_opts) {
4701 error_setg(errp, "Format driver '%s' does not support image creation",
4702 drv->format_name);
4703 return;
4706 if (!proto_drv->create_opts) {
4707 error_setg(errp, "Protocol driver '%s' does not support image creation",
4708 proto_drv->format_name);
4709 return;
4712 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4713 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4715 /* Create parameter list with default values */
4716 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4717 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4719 /* Parse -o options */
4720 if (options) {
4721 qemu_opts_do_parse(opts, options, NULL, &local_err);
4722 if (local_err) {
4723 error_report_err(local_err);
4724 local_err = NULL;
4725 error_setg(errp, "Invalid options for file format '%s'", fmt);
4726 goto out;
4730 if (base_filename) {
4731 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4732 if (local_err) {
4733 error_setg(errp, "Backing file not supported for file format '%s'",
4734 fmt);
4735 goto out;
4739 if (base_fmt) {
4740 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4741 if (local_err) {
4742 error_setg(errp, "Backing file format not supported for file "
4743 "format '%s'", fmt);
4744 goto out;
4748 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4749 if (backing_file) {
4750 if (!strcmp(filename, backing_file)) {
4751 error_setg(errp, "Error: Trying to create an image with the "
4752 "same filename as the backing file");
4753 goto out;
4757 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4759 /* The size for the image must always be specified, unless we have a backing
4760 * file and we have not been forbidden from opening it. */
4761 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4762 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4763 BlockDriverState *bs;
4764 char *full_backing = g_new0(char, PATH_MAX);
4765 int back_flags;
4766 QDict *backing_options = NULL;
4768 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4769 full_backing, PATH_MAX,
4770 &local_err);
4771 if (local_err) {
4772 g_free(full_backing);
4773 goto out;
4776 /* backing files always opened read-only */
4777 back_flags = flags;
4778 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4780 backing_options = qdict_new();
4781 if (backing_fmt) {
4782 qdict_put_str(backing_options, "driver", backing_fmt);
4784 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
4786 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4787 &local_err);
4788 g_free(full_backing);
4789 if (!bs && size != -1) {
4790 /* Couldn't open BS, but we have a size, so it's nonfatal */
4791 warn_reportf_err(local_err,
4792 "Could not verify backing image. "
4793 "This may become an error in future versions.\n");
4794 local_err = NULL;
4795 } else if (!bs) {
4796 /* Couldn't open bs, do not have size */
4797 error_append_hint(&local_err,
4798 "Could not open backing image to determine size.\n");
4799 goto out;
4800 } else {
4801 if (size == -1) {
4802 /* Opened BS, have no size */
4803 size = bdrv_getlength(bs);
4804 if (size < 0) {
4805 error_setg_errno(errp, -size, "Could not get size of '%s'",
4806 backing_file);
4807 bdrv_unref(bs);
4808 goto out;
4810 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4812 bdrv_unref(bs);
4814 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4816 if (size == -1) {
4817 error_setg(errp, "Image creation needs a size parameter");
4818 goto out;
4821 if (!quiet) {
4822 printf("Formatting '%s', fmt=%s ", filename, fmt);
4823 qemu_opts_print(opts, " ");
4824 puts("");
4827 ret = bdrv_create(drv, filename, opts, &local_err);
4829 if (ret == -EFBIG) {
4830 /* This is generally a better message than whatever the driver would
4831 * deliver (especially because of the cluster_size_hint), since that
4832 * is most probably not much different from "image too large". */
4833 const char *cluster_size_hint = "";
4834 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4835 cluster_size_hint = " (try using a larger cluster size)";
4837 error_setg(errp, "The image size is too large for file format '%s'"
4838 "%s", fmt, cluster_size_hint);
4839 error_free(local_err);
4840 local_err = NULL;
4843 out:
4844 qemu_opts_del(opts);
4845 qemu_opts_free(create_opts);
4846 error_propagate(errp, local_err);
4849 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4851 return bs ? bs->aio_context : qemu_get_aio_context();
4854 AioWait *bdrv_get_aio_wait(BlockDriverState *bs)
4856 return bs ? &bs->wait : NULL;
4859 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4861 aio_co_enter(bdrv_get_aio_context(bs), co);
4864 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4866 QLIST_REMOVE(ban, list);
4867 g_free(ban);
4870 void bdrv_detach_aio_context(BlockDriverState *bs)
4872 BdrvAioNotifier *baf, *baf_tmp;
4873 BdrvChild *child;
4875 if (!bs->drv) {
4876 return;
4879 assert(!bs->walking_aio_notifiers);
4880 bs->walking_aio_notifiers = true;
4881 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4882 if (baf->deleted) {
4883 bdrv_do_remove_aio_context_notifier(baf);
4884 } else {
4885 baf->detach_aio_context(baf->opaque);
4888 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4889 * remove remaining aio notifiers if we aren't called again.
4891 bs->walking_aio_notifiers = false;
4893 if (bs->drv->bdrv_detach_aio_context) {
4894 bs->drv->bdrv_detach_aio_context(bs);
4896 QLIST_FOREACH(child, &bs->children, next) {
4897 bdrv_detach_aio_context(child->bs);
4900 bs->aio_context = NULL;
4903 void bdrv_attach_aio_context(BlockDriverState *bs,
4904 AioContext *new_context)
4906 BdrvAioNotifier *ban, *ban_tmp;
4907 BdrvChild *child;
4909 if (!bs->drv) {
4910 return;
4913 bs->aio_context = new_context;
4915 QLIST_FOREACH(child, &bs->children, next) {
4916 bdrv_attach_aio_context(child->bs, new_context);
4918 if (bs->drv->bdrv_attach_aio_context) {
4919 bs->drv->bdrv_attach_aio_context(bs, new_context);
4922 assert(!bs->walking_aio_notifiers);
4923 bs->walking_aio_notifiers = true;
4924 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4925 if (ban->deleted) {
4926 bdrv_do_remove_aio_context_notifier(ban);
4927 } else {
4928 ban->attached_aio_context(new_context, ban->opaque);
4931 bs->walking_aio_notifiers = false;
4934 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4936 AioContext *ctx = bdrv_get_aio_context(bs);
4938 aio_disable_external(ctx);
4939 bdrv_parent_drained_begin(bs, NULL);
4940 bdrv_drain(bs); /* ensure there are no in-flight requests */
4942 while (aio_poll(ctx, false)) {
4943 /* wait for all bottom halves to execute */
4946 bdrv_detach_aio_context(bs);
4948 /* This function executes in the old AioContext so acquire the new one in
4949 * case it runs in a different thread.
4951 aio_context_acquire(new_context);
4952 bdrv_attach_aio_context(bs, new_context);
4953 bdrv_parent_drained_end(bs, NULL);
4954 aio_enable_external(ctx);
4955 aio_context_release(new_context);
4958 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4959 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4960 void (*detach_aio_context)(void *opaque), void *opaque)
4962 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4963 *ban = (BdrvAioNotifier){
4964 .attached_aio_context = attached_aio_context,
4965 .detach_aio_context = detach_aio_context,
4966 .opaque = opaque
4969 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4972 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4973 void (*attached_aio_context)(AioContext *,
4974 void *),
4975 void (*detach_aio_context)(void *),
4976 void *opaque)
4978 BdrvAioNotifier *ban, *ban_next;
4980 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4981 if (ban->attached_aio_context == attached_aio_context &&
4982 ban->detach_aio_context == detach_aio_context &&
4983 ban->opaque == opaque &&
4984 ban->deleted == false)
4986 if (bs->walking_aio_notifiers) {
4987 ban->deleted = true;
4988 } else {
4989 bdrv_do_remove_aio_context_notifier(ban);
4991 return;
4995 abort();
4998 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4999 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
5001 if (!bs->drv) {
5002 return -ENOMEDIUM;
5004 if (!bs->drv->bdrv_amend_options) {
5005 return -ENOTSUP;
5007 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
5010 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5011 * of block filter and by bdrv_is_first_non_filter.
5012 * It is used to test if the given bs is the candidate or recurse more in the
5013 * node graph.
5015 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5016 BlockDriverState *candidate)
5018 /* return false if basic checks fails */
5019 if (!bs || !bs->drv) {
5020 return false;
5023 /* the code reached a non block filter driver -> check if the bs is
5024 * the same as the candidate. It's the recursion termination condition.
5026 if (!bs->drv->is_filter) {
5027 return bs == candidate;
5029 /* Down this path the driver is a block filter driver */
5031 /* If the block filter recursion method is defined use it to recurse down
5032 * the node graph.
5034 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5035 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5038 /* the driver is a block filter but don't allow to recurse -> return false
5040 return false;
5043 /* This function checks if the candidate is the first non filter bs down it's
5044 * bs chain. Since we don't have pointers to parents it explore all bs chains
5045 * from the top. Some filters can choose not to pass down the recursion.
5047 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5049 BlockDriverState *bs;
5050 BdrvNextIterator it;
5052 /* walk down the bs forest recursively */
5053 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5054 bool perm;
5056 /* try to recurse in this top level bs */
5057 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5059 /* candidate is the first non filter */
5060 if (perm) {
5061 bdrv_next_cleanup(&it);
5062 return true;
5066 return false;
5069 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5070 const char *node_name, Error **errp)
5072 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5073 AioContext *aio_context;
5075 if (!to_replace_bs) {
5076 error_setg(errp, "Node name '%s' not found", node_name);
5077 return NULL;
5080 aio_context = bdrv_get_aio_context(to_replace_bs);
5081 aio_context_acquire(aio_context);
5083 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5084 to_replace_bs = NULL;
5085 goto out;
5088 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5089 * most non filter in order to prevent data corruption.
5090 * Another benefit is that this tests exclude backing files which are
5091 * blocked by the backing blockers.
5093 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5094 error_setg(errp, "Only top most non filter can be replaced");
5095 to_replace_bs = NULL;
5096 goto out;
5099 out:
5100 aio_context_release(aio_context);
5101 return to_replace_bs;
5104 static bool append_open_options(QDict *d, BlockDriverState *bs)
5106 const QDictEntry *entry;
5107 QemuOptDesc *desc;
5108 BdrvChild *child;
5109 bool found_any = false;
5110 const char *p;
5112 for (entry = qdict_first(bs->options); entry;
5113 entry = qdict_next(bs->options, entry))
5115 /* Exclude options for children */
5116 QLIST_FOREACH(child, &bs->children, next) {
5117 if (strstart(qdict_entry_key(entry), child->name, &p)
5118 && (!*p || *p == '.'))
5120 break;
5123 if (child) {
5124 continue;
5127 /* And exclude all non-driver-specific options */
5128 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
5129 if (!strcmp(qdict_entry_key(entry), desc->name)) {
5130 break;
5133 if (desc->name) {
5134 continue;
5137 qobject_incref(qdict_entry_value(entry));
5138 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
5139 found_any = true;
5142 return found_any;
5145 /* Updates the following BDS fields:
5146 * - exact_filename: A filename which may be used for opening a block device
5147 * which (mostly) equals the given BDS (even without any
5148 * other options; so reading and writing must return the same
5149 * results, but caching etc. may be different)
5150 * - full_open_options: Options which, when given when opening a block device
5151 * (without a filename), result in a BDS (mostly)
5152 * equalling the given one
5153 * - filename: If exact_filename is set, it is copied here. Otherwise,
5154 * full_open_options is converted to a JSON object, prefixed with
5155 * "json:" (for use through the JSON pseudo protocol) and put here.
5157 void bdrv_refresh_filename(BlockDriverState *bs)
5159 BlockDriver *drv = bs->drv;
5160 QDict *opts;
5162 if (!drv) {
5163 return;
5166 /* This BDS's file name will most probably depend on its file's name, so
5167 * refresh that first */
5168 if (bs->file) {
5169 bdrv_refresh_filename(bs->file->bs);
5172 if (drv->bdrv_refresh_filename) {
5173 /* Obsolete information is of no use here, so drop the old file name
5174 * information before refreshing it */
5175 bs->exact_filename[0] = '\0';
5176 if (bs->full_open_options) {
5177 QDECREF(bs->full_open_options);
5178 bs->full_open_options = NULL;
5181 opts = qdict_new();
5182 append_open_options(opts, bs);
5183 drv->bdrv_refresh_filename(bs, opts);
5184 QDECREF(opts);
5185 } else if (bs->file) {
5186 /* Try to reconstruct valid information from the underlying file */
5187 bool has_open_options;
5189 bs->exact_filename[0] = '\0';
5190 if (bs->full_open_options) {
5191 QDECREF(bs->full_open_options);
5192 bs->full_open_options = NULL;
5195 opts = qdict_new();
5196 has_open_options = append_open_options(opts, bs);
5198 /* If no specific options have been given for this BDS, the filename of
5199 * the underlying file should suffice for this one as well */
5200 if (bs->file->bs->exact_filename[0] && !has_open_options) {
5201 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5203 /* Reconstructing the full options QDict is simple for most format block
5204 * drivers, as long as the full options are known for the underlying
5205 * file BDS. The full options QDict of that file BDS should somehow
5206 * contain a representation of the filename, therefore the following
5207 * suffices without querying the (exact_)filename of this BDS. */
5208 if (bs->file->bs->full_open_options) {
5209 qdict_put_str(opts, "driver", drv->format_name);
5210 QINCREF(bs->file->bs->full_open_options);
5211 qdict_put(opts, "file", bs->file->bs->full_open_options);
5213 bs->full_open_options = opts;
5214 } else {
5215 QDECREF(opts);
5217 } else if (!bs->full_open_options && qdict_size(bs->options)) {
5218 /* There is no underlying file BDS (at least referenced by BDS.file),
5219 * so the full options QDict should be equal to the options given
5220 * specifically for this block device when it was opened (plus the
5221 * driver specification).
5222 * Because those options don't change, there is no need to update
5223 * full_open_options when it's already set. */
5225 opts = qdict_new();
5226 append_open_options(opts, bs);
5227 qdict_put_str(opts, "driver", drv->format_name);
5229 if (bs->exact_filename[0]) {
5230 /* This may not work for all block protocol drivers (some may
5231 * require this filename to be parsed), but we have to find some
5232 * default solution here, so just include it. If some block driver
5233 * does not support pure options without any filename at all or
5234 * needs some special format of the options QDict, it needs to
5235 * implement the driver-specific bdrv_refresh_filename() function.
5237 qdict_put_str(opts, "filename", bs->exact_filename);
5240 bs->full_open_options = opts;
5243 if (bs->exact_filename[0]) {
5244 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5245 } else if (bs->full_open_options) {
5246 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5247 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5248 qstring_get_str(json));
5249 QDECREF(json);
5254 * Hot add/remove a BDS's child. So the user can take a child offline when
5255 * it is broken and take a new child online
5257 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5258 Error **errp)
5261 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5262 error_setg(errp, "The node %s does not support adding a child",
5263 bdrv_get_device_or_node_name(parent_bs));
5264 return;
5267 if (!QLIST_EMPTY(&child_bs->parents)) {
5268 error_setg(errp, "The node %s already has a parent",
5269 child_bs->node_name);
5270 return;
5273 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5276 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5278 BdrvChild *tmp;
5280 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5281 error_setg(errp, "The node %s does not support removing a child",
5282 bdrv_get_device_or_node_name(parent_bs));
5283 return;
5286 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5287 if (tmp == child) {
5288 break;
5292 if (!tmp) {
5293 error_setg(errp, "The node %s does not have a child named %s",
5294 bdrv_get_device_or_node_name(parent_bs),
5295 bdrv_get_device_or_node_name(child->bs));
5296 return;
5299 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5302 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5303 uint32_t granularity, Error **errp)
5305 BlockDriver *drv = bs->drv;
5307 if (!drv) {
5308 error_setg_errno(errp, ENOMEDIUM,
5309 "Can't store persistent bitmaps to %s",
5310 bdrv_get_device_or_node_name(bs));
5311 return false;
5314 if (!drv->bdrv_can_store_new_dirty_bitmap) {
5315 error_setg_errno(errp, ENOTSUP,
5316 "Can't store persistent bitmaps to %s",
5317 bdrv_get_device_or_node_name(bs));
5318 return false;
5321 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);