virtio-net: support RSC v4/v6 tcp traffic for Windows HCK
[qemu/ar7.git] / block.c
blob4f5ff2cc1208fc2dd022c2487e59df6e7a3eb261
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/module.h"
34 #include "qapi/error.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qjson.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-output-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "sysemu/block-backend.h"
42 #include "sysemu/sysemu.h"
43 #include "qemu/notify.h"
44 #include "qemu/option.h"
45 #include "qemu/coroutine.h"
46 #include "block/qapi.h"
47 #include "qemu/timer.h"
48 #include "qemu/cutils.h"
49 #include "qemu/id.h"
51 #ifdef CONFIG_BSD
52 #include <sys/ioctl.h>
53 #include <sys/queue.h>
54 #ifndef __DragonFly__
55 #include <sys/disk.h>
56 #endif
57 #endif
59 #ifdef _WIN32
60 #include <windows.h>
61 #endif
63 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
72 QLIST_HEAD_INITIALIZER(bdrv_drivers);
74 static BlockDriverState *bdrv_open_inherit(const char *filename,
75 const char *reference,
76 QDict *options, int flags,
77 BlockDriverState *parent,
78 const BdrvChildRole *child_role,
79 Error **errp);
81 /* If non-zero, use only whitelisted block drivers */
82 static int use_bdrv_whitelist;
84 #ifdef _WIN32
85 static int is_windows_drive_prefix(const char *filename)
87 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
88 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
89 filename[1] == ':');
92 int is_windows_drive(const char *filename)
94 if (is_windows_drive_prefix(filename) &&
95 filename[2] == '\0')
96 return 1;
97 if (strstart(filename, "\\\\.\\", NULL) ||
98 strstart(filename, "//./", NULL))
99 return 1;
100 return 0;
102 #endif
104 size_t bdrv_opt_mem_align(BlockDriverState *bs)
106 if (!bs || !bs->drv) {
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
111 return bs->bl.opt_mem_alignment;
114 size_t bdrv_min_mem_align(BlockDriverState *bs)
116 if (!bs || !bs->drv) {
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
121 return bs->bl.min_mem_alignment;
124 /* check if the path starts with "<protocol>:" */
125 int path_has_protocol(const char *path)
127 const char *p;
129 #ifdef _WIN32
130 if (is_windows_drive(path) ||
131 is_windows_drive_prefix(path)) {
132 return 0;
134 p = path + strcspn(path, ":/\\");
135 #else
136 p = path + strcspn(path, ":/");
137 #endif
139 return *p == ':';
142 int path_is_absolute(const char *path)
144 #ifdef _WIN32
145 /* specific case for names like: "\\.\d:" */
146 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
147 return 1;
149 return (*path == '/' || *path == '\\');
150 #else
151 return (*path == '/');
152 #endif
155 /* if filename is absolute, just copy it to dest. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
157 supported. */
158 void path_combine(char *dest, int dest_size,
159 const char *base_path,
160 const char *filename)
162 const char *p, *p1;
163 int len;
165 if (dest_size <= 0)
166 return;
167 if (path_is_absolute(filename)) {
168 pstrcpy(dest, dest_size, filename);
169 } else {
170 const char *protocol_stripped = NULL;
172 if (path_has_protocol(base_path)) {
173 protocol_stripped = strchr(base_path, ':');
174 if (protocol_stripped) {
175 protocol_stripped++;
178 p = protocol_stripped ?: base_path;
180 p1 = strrchr(base_path, '/');
181 #ifdef _WIN32
183 const char *p2;
184 p2 = strrchr(base_path, '\\');
185 if (!p1 || p2 > p1)
186 p1 = p2;
188 #endif
189 if (p1)
190 p1++;
191 else
192 p1 = base_path;
193 if (p1 > p)
194 p = p1;
195 len = p - base_path;
196 if (len > dest_size - 1)
197 len = dest_size - 1;
198 memcpy(dest, base_path, len);
199 dest[len] = '\0';
200 pstrcat(dest, dest_size, filename);
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
209 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
210 QDict *options)
212 if (strstart(filename, prefix, &filename)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename)) {
216 QString *fat_filename;
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename));
222 /* And we can thus fix the protocol detection issue by prefixing it
223 * by "./" */
224 fat_filename = qstring_from_str("./");
225 qstring_append(fat_filename, filename);
227 assert(!path_has_protocol(qstring_get_str(fat_filename)));
229 qdict_put(options, "filename", fat_filename);
230 } else {
231 /* If no protocol prefix was detected, we can use the shortened
232 * filename as-is */
233 qdict_put_str(options, "filename", filename);
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState *bs)
244 return bs->read_only;
247 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
248 bool ignore_allow_rdw, Error **errp)
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs->copy_on_read && read_only) {
252 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs));
254 return -EINVAL;
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
259 !ignore_allow_rdw)
261 error_setg(errp, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs));
263 return -EPERM;
266 return 0;
270 * Called by a driver that can only provide a read-only image.
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
279 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
280 Error **errp)
282 int ret = 0;
284 if (!(bs->open_flags & BDRV_O_RDWR)) {
285 return 0;
287 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
288 goto fail;
291 ret = bdrv_can_set_read_only(bs, true, false, NULL);
292 if (ret < 0) {
293 goto fail;
296 bs->read_only = true;
297 bs->open_flags &= ~BDRV_O_RDWR;
299 return 0;
301 fail:
302 error_setg(errp, "%s", errmsg ?: "Image is read-only");
303 return -EACCES;
306 void bdrv_get_full_backing_filename_from_filename(const char *backed,
307 const char *backing,
308 char *dest, size_t sz,
309 Error **errp)
311 if (backing[0] == '\0' || path_has_protocol(backing) ||
312 path_is_absolute(backing))
314 pstrcpy(dest, sz, backing);
315 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
316 error_setg(errp, "Cannot use relative backing file names for '%s'",
317 backed);
318 } else {
319 path_combine(dest, sz, backed, backing);
323 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
324 Error **errp)
326 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
328 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
329 dest, sz, errp);
332 void bdrv_register(BlockDriver *bdrv)
334 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
337 BlockDriverState *bdrv_new(void)
339 BlockDriverState *bs;
340 int i;
342 bs = g_new0(BlockDriverState, 1);
343 QLIST_INIT(&bs->dirty_bitmaps);
344 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
345 QLIST_INIT(&bs->op_blockers[i]);
347 notifier_with_return_list_init(&bs->before_write_notifiers);
348 qemu_co_mutex_init(&bs->reqs_lock);
349 qemu_mutex_init(&bs->dirty_bitmap_mutex);
350 bs->refcnt = 1;
351 bs->aio_context = qemu_get_aio_context();
353 qemu_co_queue_init(&bs->flush_queue);
355 for (i = 0; i < bdrv_drain_all_count; i++) {
356 bdrv_drained_begin(bs);
359 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
361 return bs;
364 static BlockDriver *bdrv_do_find_format(const char *format_name)
366 BlockDriver *drv1;
368 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
369 if (!strcmp(drv1->format_name, format_name)) {
370 return drv1;
374 return NULL;
377 BlockDriver *bdrv_find_format(const char *format_name)
379 BlockDriver *drv1;
380 int i;
382 drv1 = bdrv_do_find_format(format_name);
383 if (drv1) {
384 return drv1;
387 /* The driver isn't registered, maybe we need to load a module */
388 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
389 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
390 block_module_load_one(block_driver_modules[i].library_name);
391 break;
395 return bdrv_do_find_format(format_name);
398 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
400 static const char *whitelist_rw[] = {
401 CONFIG_BDRV_RW_WHITELIST
403 static const char *whitelist_ro[] = {
404 CONFIG_BDRV_RO_WHITELIST
406 const char **p;
408 if (!whitelist_rw[0] && !whitelist_ro[0]) {
409 return 1; /* no whitelist, anything goes */
412 for (p = whitelist_rw; *p; p++) {
413 if (!strcmp(drv->format_name, *p)) {
414 return 1;
417 if (read_only) {
418 for (p = whitelist_ro; *p; p++) {
419 if (!strcmp(drv->format_name, *p)) {
420 return 1;
424 return 0;
427 bool bdrv_uses_whitelist(void)
429 return use_bdrv_whitelist;
432 typedef struct CreateCo {
433 BlockDriver *drv;
434 char *filename;
435 QemuOpts *opts;
436 int ret;
437 Error *err;
438 } CreateCo;
440 static void coroutine_fn bdrv_create_co_entry(void *opaque)
442 Error *local_err = NULL;
443 int ret;
445 CreateCo *cco = opaque;
446 assert(cco->drv);
448 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
449 error_propagate(&cco->err, local_err);
450 cco->ret = ret;
453 int bdrv_create(BlockDriver *drv, const char* filename,
454 QemuOpts *opts, Error **errp)
456 int ret;
458 Coroutine *co;
459 CreateCo cco = {
460 .drv = drv,
461 .filename = g_strdup(filename),
462 .opts = opts,
463 .ret = NOT_DONE,
464 .err = NULL,
467 if (!drv->bdrv_co_create_opts) {
468 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
469 ret = -ENOTSUP;
470 goto out;
473 if (qemu_in_coroutine()) {
474 /* Fast-path if already in coroutine context */
475 bdrv_create_co_entry(&cco);
476 } else {
477 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
478 qemu_coroutine_enter(co);
479 while (cco.ret == NOT_DONE) {
480 aio_poll(qemu_get_aio_context(), true);
484 ret = cco.ret;
485 if (ret < 0) {
486 if (cco.err) {
487 error_propagate(errp, cco.err);
488 } else {
489 error_setg_errno(errp, -ret, "Could not create image");
493 out:
494 g_free(cco.filename);
495 return ret;
498 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
500 BlockDriver *drv;
501 Error *local_err = NULL;
502 int ret;
504 drv = bdrv_find_protocol(filename, true, errp);
505 if (drv == NULL) {
506 return -ENOENT;
509 ret = bdrv_create(drv, filename, opts, &local_err);
510 error_propagate(errp, local_err);
511 return ret;
515 * Try to get @bs's logical and physical block size.
516 * On success, store them in @bsz struct and return 0.
517 * On failure return -errno.
518 * @bs must not be empty.
520 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
522 BlockDriver *drv = bs->drv;
524 if (drv && drv->bdrv_probe_blocksizes) {
525 return drv->bdrv_probe_blocksizes(bs, bsz);
526 } else if (drv && drv->is_filter && bs->file) {
527 return bdrv_probe_blocksizes(bs->file->bs, bsz);
530 return -ENOTSUP;
534 * Try to get @bs's geometry (cyls, heads, sectors).
535 * On success, store them in @geo struct and return 0.
536 * On failure return -errno.
537 * @bs must not be empty.
539 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
541 BlockDriver *drv = bs->drv;
543 if (drv && drv->bdrv_probe_geometry) {
544 return drv->bdrv_probe_geometry(bs, geo);
545 } else if (drv && drv->is_filter && bs->file) {
546 return bdrv_probe_geometry(bs->file->bs, geo);
549 return -ENOTSUP;
553 * Create a uniquely-named empty temporary file.
554 * Return 0 upon success, otherwise a negative errno value.
556 int get_tmp_filename(char *filename, int size)
558 #ifdef _WIN32
559 char temp_dir[MAX_PATH];
560 /* GetTempFileName requires that its output buffer (4th param)
561 have length MAX_PATH or greater. */
562 assert(size >= MAX_PATH);
563 return (GetTempPath(MAX_PATH, temp_dir)
564 && GetTempFileName(temp_dir, "qem", 0, filename)
565 ? 0 : -GetLastError());
566 #else
567 int fd;
568 const char *tmpdir;
569 tmpdir = getenv("TMPDIR");
570 if (!tmpdir) {
571 tmpdir = "/var/tmp";
573 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
574 return -EOVERFLOW;
576 fd = mkstemp(filename);
577 if (fd < 0) {
578 return -errno;
580 if (close(fd) != 0) {
581 unlink(filename);
582 return -errno;
584 return 0;
585 #endif
589 * Detect host devices. By convention, /dev/cdrom[N] is always
590 * recognized as a host CDROM.
592 static BlockDriver *find_hdev_driver(const char *filename)
594 int score_max = 0, score;
595 BlockDriver *drv = NULL, *d;
597 QLIST_FOREACH(d, &bdrv_drivers, list) {
598 if (d->bdrv_probe_device) {
599 score = d->bdrv_probe_device(filename);
600 if (score > score_max) {
601 score_max = score;
602 drv = d;
607 return drv;
610 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
612 BlockDriver *drv1;
614 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
615 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
616 return drv1;
620 return NULL;
623 BlockDriver *bdrv_find_protocol(const char *filename,
624 bool allow_protocol_prefix,
625 Error **errp)
627 BlockDriver *drv1;
628 char protocol[128];
629 int len;
630 const char *p;
631 int i;
633 /* TODO Drivers without bdrv_file_open must be specified explicitly */
636 * XXX(hch): we really should not let host device detection
637 * override an explicit protocol specification, but moving this
638 * later breaks access to device names with colons in them.
639 * Thanks to the brain-dead persistent naming schemes on udev-
640 * based Linux systems those actually are quite common.
642 drv1 = find_hdev_driver(filename);
643 if (drv1) {
644 return drv1;
647 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
648 return &bdrv_file;
651 p = strchr(filename, ':');
652 assert(p != NULL);
653 len = p - filename;
654 if (len > sizeof(protocol) - 1)
655 len = sizeof(protocol) - 1;
656 memcpy(protocol, filename, len);
657 protocol[len] = '\0';
659 drv1 = bdrv_do_find_protocol(protocol);
660 if (drv1) {
661 return drv1;
664 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
665 if (block_driver_modules[i].protocol_name &&
666 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
667 block_module_load_one(block_driver_modules[i].library_name);
668 break;
672 drv1 = bdrv_do_find_protocol(protocol);
673 if (!drv1) {
674 error_setg(errp, "Unknown protocol '%s'", protocol);
676 return drv1;
680 * Guess image format by probing its contents.
681 * This is not a good idea when your image is raw (CVE-2008-2004), but
682 * we do it anyway for backward compatibility.
684 * @buf contains the image's first @buf_size bytes.
685 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
686 * but can be smaller if the image file is smaller)
687 * @filename is its filename.
689 * For all block drivers, call the bdrv_probe() method to get its
690 * probing score.
691 * Return the first block driver with the highest probing score.
693 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
694 const char *filename)
696 int score_max = 0, score;
697 BlockDriver *drv = NULL, *d;
699 QLIST_FOREACH(d, &bdrv_drivers, list) {
700 if (d->bdrv_probe) {
701 score = d->bdrv_probe(buf, buf_size, filename);
702 if (score > score_max) {
703 score_max = score;
704 drv = d;
709 return drv;
712 static int find_image_format(BlockBackend *file, const char *filename,
713 BlockDriver **pdrv, Error **errp)
715 BlockDriver *drv;
716 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
717 int ret = 0;
719 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
720 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
721 *pdrv = &bdrv_raw;
722 return ret;
725 ret = blk_pread(file, 0, buf, sizeof(buf));
726 if (ret < 0) {
727 error_setg_errno(errp, -ret, "Could not read image for determining its "
728 "format");
729 *pdrv = NULL;
730 return ret;
733 drv = bdrv_probe_all(buf, ret, filename);
734 if (!drv) {
735 error_setg(errp, "Could not determine image format: No compatible "
736 "driver found");
737 ret = -ENOENT;
739 *pdrv = drv;
740 return ret;
744 * Set the current 'total_sectors' value
745 * Return 0 on success, -errno on error.
747 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
749 BlockDriver *drv = bs->drv;
751 if (!drv) {
752 return -ENOMEDIUM;
755 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
756 if (bdrv_is_sg(bs))
757 return 0;
759 /* query actual device if possible, otherwise just trust the hint */
760 if (drv->bdrv_getlength) {
761 int64_t length = drv->bdrv_getlength(bs);
762 if (length < 0) {
763 return length;
765 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
768 bs->total_sectors = hint;
769 return 0;
773 * Combines a QDict of new block driver @options with any missing options taken
774 * from @old_options, so that leaving out an option defaults to its old value.
776 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
777 QDict *old_options)
779 if (bs->drv && bs->drv->bdrv_join_options) {
780 bs->drv->bdrv_join_options(options, old_options);
781 } else {
782 qdict_join(options, old_options, false);
786 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
787 int open_flags,
788 Error **errp)
790 Error *local_err = NULL;
791 char *value = qemu_opt_get_del(opts, "detect-zeroes");
792 BlockdevDetectZeroesOptions detect_zeroes =
793 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
794 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
795 g_free(value);
796 if (local_err) {
797 error_propagate(errp, local_err);
798 return detect_zeroes;
801 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
802 !(open_flags & BDRV_O_UNMAP))
804 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
805 "without setting discard operation to unmap");
808 return detect_zeroes;
812 * Set open flags for a given discard mode
814 * Return 0 on success, -1 if the discard mode was invalid.
816 int bdrv_parse_discard_flags(const char *mode, int *flags)
818 *flags &= ~BDRV_O_UNMAP;
820 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
821 /* do nothing */
822 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
823 *flags |= BDRV_O_UNMAP;
824 } else {
825 return -1;
828 return 0;
832 * Set open flags for a given cache mode
834 * Return 0 on success, -1 if the cache mode was invalid.
836 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
838 *flags &= ~BDRV_O_CACHE_MASK;
840 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
841 *writethrough = false;
842 *flags |= BDRV_O_NOCACHE;
843 } else if (!strcmp(mode, "directsync")) {
844 *writethrough = true;
845 *flags |= BDRV_O_NOCACHE;
846 } else if (!strcmp(mode, "writeback")) {
847 *writethrough = false;
848 } else if (!strcmp(mode, "unsafe")) {
849 *writethrough = false;
850 *flags |= BDRV_O_NO_FLUSH;
851 } else if (!strcmp(mode, "writethrough")) {
852 *writethrough = true;
853 } else {
854 return -1;
857 return 0;
860 static char *bdrv_child_get_parent_desc(BdrvChild *c)
862 BlockDriverState *parent = c->opaque;
863 return g_strdup(bdrv_get_device_or_node_name(parent));
866 static void bdrv_child_cb_drained_begin(BdrvChild *child)
868 BlockDriverState *bs = child->opaque;
869 bdrv_do_drained_begin_quiesce(bs, NULL, false);
872 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
874 BlockDriverState *bs = child->opaque;
875 return bdrv_drain_poll(bs, false, NULL, false);
878 static void bdrv_child_cb_drained_end(BdrvChild *child)
880 BlockDriverState *bs = child->opaque;
881 bdrv_drained_end(bs);
884 static void bdrv_child_cb_attach(BdrvChild *child)
886 BlockDriverState *bs = child->opaque;
887 bdrv_apply_subtree_drain(child, bs);
890 static void bdrv_child_cb_detach(BdrvChild *child)
892 BlockDriverState *bs = child->opaque;
893 bdrv_unapply_subtree_drain(child, bs);
896 static int bdrv_child_cb_inactivate(BdrvChild *child)
898 BlockDriverState *bs = child->opaque;
899 assert(bs->open_flags & BDRV_O_INACTIVE);
900 return 0;
904 * Returns the options and flags that a temporary snapshot should get, based on
905 * the originally requested flags (the originally requested image will have
906 * flags like a backing file)
908 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
909 int parent_flags, QDict *parent_options)
911 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
913 /* For temporary files, unconditional cache=unsafe is fine */
914 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
915 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
917 /* Copy the read-only option from the parent */
918 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
920 /* aio=native doesn't work for cache.direct=off, so disable it for the
921 * temporary snapshot */
922 *child_flags &= ~BDRV_O_NATIVE_AIO;
926 * Returns the options and flags that bs->file should get if a protocol driver
927 * is expected, based on the given options and flags for the parent BDS
929 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
930 int parent_flags, QDict *parent_options)
932 int flags = parent_flags;
934 /* Enable protocol handling, disable format probing for bs->file */
935 flags |= BDRV_O_PROTOCOL;
937 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
938 * the parent. */
939 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
940 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
941 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
943 /* Inherit the read-only option from the parent if it's not set */
944 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
945 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
947 /* Our block drivers take care to send flushes and respect unmap policy,
948 * so we can default to enable both on lower layers regardless of the
949 * corresponding parent options. */
950 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
952 /* Clear flags that only apply to the top layer */
953 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
954 BDRV_O_NO_IO);
956 *child_flags = flags;
959 const BdrvChildRole child_file = {
960 .parent_is_bds = true,
961 .get_parent_desc = bdrv_child_get_parent_desc,
962 .inherit_options = bdrv_inherited_options,
963 .drained_begin = bdrv_child_cb_drained_begin,
964 .drained_poll = bdrv_child_cb_drained_poll,
965 .drained_end = bdrv_child_cb_drained_end,
966 .attach = bdrv_child_cb_attach,
967 .detach = bdrv_child_cb_detach,
968 .inactivate = bdrv_child_cb_inactivate,
972 * Returns the options and flags that bs->file should get if the use of formats
973 * (and not only protocols) is permitted for it, based on the given options and
974 * flags for the parent BDS
976 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
977 int parent_flags, QDict *parent_options)
979 child_file.inherit_options(child_flags, child_options,
980 parent_flags, parent_options);
982 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
985 const BdrvChildRole child_format = {
986 .parent_is_bds = true,
987 .get_parent_desc = bdrv_child_get_parent_desc,
988 .inherit_options = bdrv_inherited_fmt_options,
989 .drained_begin = bdrv_child_cb_drained_begin,
990 .drained_poll = bdrv_child_cb_drained_poll,
991 .drained_end = bdrv_child_cb_drained_end,
992 .attach = bdrv_child_cb_attach,
993 .detach = bdrv_child_cb_detach,
994 .inactivate = bdrv_child_cb_inactivate,
997 static void bdrv_backing_attach(BdrvChild *c)
999 BlockDriverState *parent = c->opaque;
1000 BlockDriverState *backing_hd = c->bs;
1002 assert(!parent->backing_blocker);
1003 error_setg(&parent->backing_blocker,
1004 "node is used as backing hd of '%s'",
1005 bdrv_get_device_or_node_name(parent));
1007 parent->open_flags &= ~BDRV_O_NO_BACKING;
1008 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1009 backing_hd->filename);
1010 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1011 backing_hd->drv ? backing_hd->drv->format_name : "");
1013 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1014 /* Otherwise we won't be able to commit or stream */
1015 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1016 parent->backing_blocker);
1017 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1018 parent->backing_blocker);
1020 * We do backup in 3 ways:
1021 * 1. drive backup
1022 * The target bs is new opened, and the source is top BDS
1023 * 2. blockdev backup
1024 * Both the source and the target are top BDSes.
1025 * 3. internal backup(used for block replication)
1026 * Both the source and the target are backing file
1028 * In case 1 and 2, neither the source nor the target is the backing file.
1029 * In case 3, we will block the top BDS, so there is only one block job
1030 * for the top BDS and its backing chain.
1032 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1033 parent->backing_blocker);
1034 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1035 parent->backing_blocker);
1037 bdrv_child_cb_attach(c);
1040 static void bdrv_backing_detach(BdrvChild *c)
1042 BlockDriverState *parent = c->opaque;
1044 assert(parent->backing_blocker);
1045 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1046 error_free(parent->backing_blocker);
1047 parent->backing_blocker = NULL;
1049 bdrv_child_cb_detach(c);
1053 * Returns the options and flags that bs->backing should get, based on the
1054 * given options and flags for the parent BDS
1056 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1057 int parent_flags, QDict *parent_options)
1059 int flags = parent_flags;
1061 /* The cache mode is inherited unmodified for backing files; except WCE,
1062 * which is only applied on the top level (BlockBackend) */
1063 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1064 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1065 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1067 /* backing files always opened read-only */
1068 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1069 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1070 flags &= ~BDRV_O_COPY_ON_READ;
1072 /* snapshot=on is handled on the top layer */
1073 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1075 *child_flags = flags;
1078 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1079 const char *filename, Error **errp)
1081 BlockDriverState *parent = c->opaque;
1082 bool read_only = bdrv_is_read_only(parent);
1083 int ret;
1085 if (read_only) {
1086 ret = bdrv_reopen_set_read_only(parent, false, errp);
1087 if (ret < 0) {
1088 return ret;
1092 ret = bdrv_change_backing_file(parent, filename,
1093 base->drv ? base->drv->format_name : "");
1094 if (ret < 0) {
1095 error_setg_errno(errp, -ret, "Could not update backing file link");
1098 if (read_only) {
1099 bdrv_reopen_set_read_only(parent, true, NULL);
1102 return ret;
1105 const BdrvChildRole child_backing = {
1106 .parent_is_bds = true,
1107 .get_parent_desc = bdrv_child_get_parent_desc,
1108 .attach = bdrv_backing_attach,
1109 .detach = bdrv_backing_detach,
1110 .inherit_options = bdrv_backing_options,
1111 .drained_begin = bdrv_child_cb_drained_begin,
1112 .drained_poll = bdrv_child_cb_drained_poll,
1113 .drained_end = bdrv_child_cb_drained_end,
1114 .inactivate = bdrv_child_cb_inactivate,
1115 .update_filename = bdrv_backing_update_filename,
1118 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1120 int open_flags = flags;
1123 * Clear flags that are internal to the block layer before opening the
1124 * image.
1126 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1129 * Snapshots should be writable.
1131 if (flags & BDRV_O_TEMPORARY) {
1132 open_flags |= BDRV_O_RDWR;
1135 return open_flags;
1138 static void update_flags_from_options(int *flags, QemuOpts *opts)
1140 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1142 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1143 *flags |= BDRV_O_NO_FLUSH;
1146 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1147 *flags |= BDRV_O_NOCACHE;
1150 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1151 *flags |= BDRV_O_RDWR;
1154 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1155 *flags |= BDRV_O_AUTO_RDONLY;
1159 static void update_options_from_flags(QDict *options, int flags)
1161 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1162 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1164 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1165 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1166 flags & BDRV_O_NO_FLUSH);
1168 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1169 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1171 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1172 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1173 flags & BDRV_O_AUTO_RDONLY);
1177 static void bdrv_assign_node_name(BlockDriverState *bs,
1178 const char *node_name,
1179 Error **errp)
1181 char *gen_node_name = NULL;
1183 if (!node_name) {
1184 node_name = gen_node_name = id_generate(ID_BLOCK);
1185 } else if (!id_wellformed(node_name)) {
1187 * Check for empty string or invalid characters, but not if it is
1188 * generated (generated names use characters not available to the user)
1190 error_setg(errp, "Invalid node name");
1191 return;
1194 /* takes care of avoiding namespaces collisions */
1195 if (blk_by_name(node_name)) {
1196 error_setg(errp, "node-name=%s is conflicting with a device id",
1197 node_name);
1198 goto out;
1201 /* takes care of avoiding duplicates node names */
1202 if (bdrv_find_node(node_name)) {
1203 error_setg(errp, "Duplicate node name");
1204 goto out;
1207 /* Make sure that the node name isn't truncated */
1208 if (strlen(node_name) >= sizeof(bs->node_name)) {
1209 error_setg(errp, "Node name too long");
1210 goto out;
1213 /* copy node name into the bs and insert it into the graph list */
1214 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1215 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1216 out:
1217 g_free(gen_node_name);
1220 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1221 const char *node_name, QDict *options,
1222 int open_flags, Error **errp)
1224 Error *local_err = NULL;
1225 int i, ret;
1227 bdrv_assign_node_name(bs, node_name, &local_err);
1228 if (local_err) {
1229 error_propagate(errp, local_err);
1230 return -EINVAL;
1233 bs->drv = drv;
1234 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1235 bs->opaque = g_malloc0(drv->instance_size);
1237 if (drv->bdrv_file_open) {
1238 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1239 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1240 } else if (drv->bdrv_open) {
1241 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1242 } else {
1243 ret = 0;
1246 if (ret < 0) {
1247 if (local_err) {
1248 error_propagate(errp, local_err);
1249 } else if (bs->filename[0]) {
1250 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1251 } else {
1252 error_setg_errno(errp, -ret, "Could not open image");
1254 goto open_failed;
1257 ret = refresh_total_sectors(bs, bs->total_sectors);
1258 if (ret < 0) {
1259 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1260 return ret;
1263 bdrv_refresh_limits(bs, &local_err);
1264 if (local_err) {
1265 error_propagate(errp, local_err);
1266 return -EINVAL;
1269 assert(bdrv_opt_mem_align(bs) != 0);
1270 assert(bdrv_min_mem_align(bs) != 0);
1271 assert(is_power_of_2(bs->bl.request_alignment));
1273 for (i = 0; i < bs->quiesce_counter; i++) {
1274 if (drv->bdrv_co_drain_begin) {
1275 drv->bdrv_co_drain_begin(bs);
1279 return 0;
1280 open_failed:
1281 bs->drv = NULL;
1282 if (bs->file != NULL) {
1283 bdrv_unref_child(bs, bs->file);
1284 bs->file = NULL;
1286 g_free(bs->opaque);
1287 bs->opaque = NULL;
1288 return ret;
1291 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1292 int flags, Error **errp)
1294 BlockDriverState *bs;
1295 int ret;
1297 bs = bdrv_new();
1298 bs->open_flags = flags;
1299 bs->explicit_options = qdict_new();
1300 bs->options = qdict_new();
1301 bs->opaque = NULL;
1303 update_options_from_flags(bs->options, flags);
1305 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1306 if (ret < 0) {
1307 qobject_unref(bs->explicit_options);
1308 bs->explicit_options = NULL;
1309 qobject_unref(bs->options);
1310 bs->options = NULL;
1311 bdrv_unref(bs);
1312 return NULL;
1315 return bs;
1318 QemuOptsList bdrv_runtime_opts = {
1319 .name = "bdrv_common",
1320 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1321 .desc = {
1323 .name = "node-name",
1324 .type = QEMU_OPT_STRING,
1325 .help = "Node name of the block device node",
1328 .name = "driver",
1329 .type = QEMU_OPT_STRING,
1330 .help = "Block driver to use for the node",
1333 .name = BDRV_OPT_CACHE_DIRECT,
1334 .type = QEMU_OPT_BOOL,
1335 .help = "Bypass software writeback cache on the host",
1338 .name = BDRV_OPT_CACHE_NO_FLUSH,
1339 .type = QEMU_OPT_BOOL,
1340 .help = "Ignore flush requests",
1343 .name = BDRV_OPT_READ_ONLY,
1344 .type = QEMU_OPT_BOOL,
1345 .help = "Node is opened in read-only mode",
1348 .name = BDRV_OPT_AUTO_READ_ONLY,
1349 .type = QEMU_OPT_BOOL,
1350 .help = "Node can become read-only if opening read-write fails",
1353 .name = "detect-zeroes",
1354 .type = QEMU_OPT_STRING,
1355 .help = "try to optimize zero writes (off, on, unmap)",
1358 .name = BDRV_OPT_DISCARD,
1359 .type = QEMU_OPT_STRING,
1360 .help = "discard operation (ignore/off, unmap/on)",
1363 .name = BDRV_OPT_FORCE_SHARE,
1364 .type = QEMU_OPT_BOOL,
1365 .help = "always accept other writers (default: off)",
1367 { /* end of list */ }
1372 * Common part for opening disk images and files
1374 * Removes all processed options from *options.
1376 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1377 QDict *options, Error **errp)
1379 int ret, open_flags;
1380 const char *filename;
1381 const char *driver_name = NULL;
1382 const char *node_name = NULL;
1383 const char *discard;
1384 QemuOpts *opts;
1385 BlockDriver *drv;
1386 Error *local_err = NULL;
1388 assert(bs->file == NULL);
1389 assert(options != NULL && bs->options != options);
1391 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1392 qemu_opts_absorb_qdict(opts, options, &local_err);
1393 if (local_err) {
1394 error_propagate(errp, local_err);
1395 ret = -EINVAL;
1396 goto fail_opts;
1399 update_flags_from_options(&bs->open_flags, opts);
1401 driver_name = qemu_opt_get(opts, "driver");
1402 drv = bdrv_find_format(driver_name);
1403 assert(drv != NULL);
1405 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1407 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1408 error_setg(errp,
1409 BDRV_OPT_FORCE_SHARE
1410 "=on can only be used with read-only images");
1411 ret = -EINVAL;
1412 goto fail_opts;
1415 if (file != NULL) {
1416 filename = blk_bs(file)->filename;
1417 } else {
1419 * Caution: while qdict_get_try_str() is fine, getting
1420 * non-string types would require more care. When @options
1421 * come from -blockdev or blockdev_add, its members are typed
1422 * according to the QAPI schema, but when they come from
1423 * -drive, they're all QString.
1425 filename = qdict_get_try_str(options, "filename");
1428 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1429 error_setg(errp, "The '%s' block driver requires a file name",
1430 drv->format_name);
1431 ret = -EINVAL;
1432 goto fail_opts;
1435 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1436 drv->format_name);
1438 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1440 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1441 error_setg(errp,
1442 !bs->read_only && bdrv_is_whitelisted(drv, true)
1443 ? "Driver '%s' can only be used for read-only devices"
1444 : "Driver '%s' is not whitelisted",
1445 drv->format_name);
1446 ret = -ENOTSUP;
1447 goto fail_opts;
1450 /* bdrv_new() and bdrv_close() make it so */
1451 assert(atomic_read(&bs->copy_on_read) == 0);
1453 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1454 if (!bs->read_only) {
1455 bdrv_enable_copy_on_read(bs);
1456 } else {
1457 error_setg(errp, "Can't use copy-on-read on read-only device");
1458 ret = -EINVAL;
1459 goto fail_opts;
1463 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1464 if (discard != NULL) {
1465 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1466 error_setg(errp, "Invalid discard option");
1467 ret = -EINVAL;
1468 goto fail_opts;
1472 bs->detect_zeroes =
1473 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1474 if (local_err) {
1475 error_propagate(errp, local_err);
1476 ret = -EINVAL;
1477 goto fail_opts;
1480 if (filename != NULL) {
1481 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1482 } else {
1483 bs->filename[0] = '\0';
1485 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1487 /* Open the image, either directly or using a protocol */
1488 open_flags = bdrv_open_flags(bs, bs->open_flags);
1489 node_name = qemu_opt_get(opts, "node-name");
1491 assert(!drv->bdrv_file_open || file == NULL);
1492 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1493 if (ret < 0) {
1494 goto fail_opts;
1497 qemu_opts_del(opts);
1498 return 0;
1500 fail_opts:
1501 qemu_opts_del(opts);
1502 return ret;
1505 static QDict *parse_json_filename(const char *filename, Error **errp)
1507 QObject *options_obj;
1508 QDict *options;
1509 int ret;
1511 ret = strstart(filename, "json:", &filename);
1512 assert(ret);
1514 options_obj = qobject_from_json(filename, errp);
1515 if (!options_obj) {
1516 error_prepend(errp, "Could not parse the JSON options: ");
1517 return NULL;
1520 options = qobject_to(QDict, options_obj);
1521 if (!options) {
1522 qobject_unref(options_obj);
1523 error_setg(errp, "Invalid JSON object given");
1524 return NULL;
1527 qdict_flatten(options);
1529 return options;
1532 static void parse_json_protocol(QDict *options, const char **pfilename,
1533 Error **errp)
1535 QDict *json_options;
1536 Error *local_err = NULL;
1538 /* Parse json: pseudo-protocol */
1539 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1540 return;
1543 json_options = parse_json_filename(*pfilename, &local_err);
1544 if (local_err) {
1545 error_propagate(errp, local_err);
1546 return;
1549 /* Options given in the filename have lower priority than options
1550 * specified directly */
1551 qdict_join(options, json_options, false);
1552 qobject_unref(json_options);
1553 *pfilename = NULL;
1557 * Fills in default options for opening images and converts the legacy
1558 * filename/flags pair to option QDict entries.
1559 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1560 * block driver has been specified explicitly.
1562 static int bdrv_fill_options(QDict **options, const char *filename,
1563 int *flags, Error **errp)
1565 const char *drvname;
1566 bool protocol = *flags & BDRV_O_PROTOCOL;
1567 bool parse_filename = false;
1568 BlockDriver *drv = NULL;
1569 Error *local_err = NULL;
1572 * Caution: while qdict_get_try_str() is fine, getting non-string
1573 * types would require more care. When @options come from
1574 * -blockdev or blockdev_add, its members are typed according to
1575 * the QAPI schema, but when they come from -drive, they're all
1576 * QString.
1578 drvname = qdict_get_try_str(*options, "driver");
1579 if (drvname) {
1580 drv = bdrv_find_format(drvname);
1581 if (!drv) {
1582 error_setg(errp, "Unknown driver '%s'", drvname);
1583 return -ENOENT;
1585 /* If the user has explicitly specified the driver, this choice should
1586 * override the BDRV_O_PROTOCOL flag */
1587 protocol = drv->bdrv_file_open;
1590 if (protocol) {
1591 *flags |= BDRV_O_PROTOCOL;
1592 } else {
1593 *flags &= ~BDRV_O_PROTOCOL;
1596 /* Translate cache options from flags into options */
1597 update_options_from_flags(*options, *flags);
1599 /* Fetch the file name from the options QDict if necessary */
1600 if (protocol && filename) {
1601 if (!qdict_haskey(*options, "filename")) {
1602 qdict_put_str(*options, "filename", filename);
1603 parse_filename = true;
1604 } else {
1605 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1606 "the same time");
1607 return -EINVAL;
1611 /* Find the right block driver */
1612 /* See cautionary note on accessing @options above */
1613 filename = qdict_get_try_str(*options, "filename");
1615 if (!drvname && protocol) {
1616 if (filename) {
1617 drv = bdrv_find_protocol(filename, parse_filename, errp);
1618 if (!drv) {
1619 return -EINVAL;
1622 drvname = drv->format_name;
1623 qdict_put_str(*options, "driver", drvname);
1624 } else {
1625 error_setg(errp, "Must specify either driver or file");
1626 return -EINVAL;
1630 assert(drv || !protocol);
1632 /* Driver-specific filename parsing */
1633 if (drv && drv->bdrv_parse_filename && parse_filename) {
1634 drv->bdrv_parse_filename(filename, *options, &local_err);
1635 if (local_err) {
1636 error_propagate(errp, local_err);
1637 return -EINVAL;
1640 if (!drv->bdrv_needs_filename) {
1641 qdict_del(*options, "filename");
1645 return 0;
1648 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1649 uint64_t perm, uint64_t shared,
1650 GSList *ignore_children, Error **errp);
1651 static void bdrv_child_abort_perm_update(BdrvChild *c);
1652 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1654 typedef struct BlockReopenQueueEntry {
1655 bool prepared;
1656 BDRVReopenState state;
1657 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1658 } BlockReopenQueueEntry;
1661 * Return the flags that @bs will have after the reopens in @q have
1662 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1663 * return the current flags.
1665 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1667 BlockReopenQueueEntry *entry;
1669 if (q != NULL) {
1670 QSIMPLEQ_FOREACH(entry, q, entry) {
1671 if (entry->state.bs == bs) {
1672 return entry->state.flags;
1677 return bs->open_flags;
1680 /* Returns whether the image file can be written to after the reopen queue @q
1681 * has been successfully applied, or right now if @q is NULL. */
1682 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1683 BlockReopenQueue *q)
1685 int flags = bdrv_reopen_get_flags(q, bs);
1687 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1691 * Return whether the BDS can be written to. This is not necessarily
1692 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1693 * be written to but do not count as read-only images.
1695 bool bdrv_is_writable(BlockDriverState *bs)
1697 return bdrv_is_writable_after_reopen(bs, NULL);
1700 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1701 BdrvChild *c, const BdrvChildRole *role,
1702 BlockReopenQueue *reopen_queue,
1703 uint64_t parent_perm, uint64_t parent_shared,
1704 uint64_t *nperm, uint64_t *nshared)
1706 if (bs->drv && bs->drv->bdrv_child_perm) {
1707 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1708 parent_perm, parent_shared,
1709 nperm, nshared);
1711 /* TODO Take force_share from reopen_queue */
1712 if (child_bs && child_bs->force_share) {
1713 *nshared = BLK_PERM_ALL;
1718 * Check whether permissions on this node can be changed in a way that
1719 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1720 * permissions of all its parents. This involves checking whether all necessary
1721 * permission changes to child nodes can be performed.
1723 * A call to this function must always be followed by a call to bdrv_set_perm()
1724 * or bdrv_abort_perm_update().
1726 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1727 uint64_t cumulative_perms,
1728 uint64_t cumulative_shared_perms,
1729 GSList *ignore_children, Error **errp)
1731 BlockDriver *drv = bs->drv;
1732 BdrvChild *c;
1733 int ret;
1735 /* Write permissions never work with read-only images */
1736 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1737 !bdrv_is_writable_after_reopen(bs, q))
1739 error_setg(errp, "Block node is read-only");
1740 return -EPERM;
1743 /* Check this node */
1744 if (!drv) {
1745 return 0;
1748 if (drv->bdrv_check_perm) {
1749 return drv->bdrv_check_perm(bs, cumulative_perms,
1750 cumulative_shared_perms, errp);
1753 /* Drivers that never have children can omit .bdrv_child_perm() */
1754 if (!drv->bdrv_child_perm) {
1755 assert(QLIST_EMPTY(&bs->children));
1756 return 0;
1759 /* Check all children */
1760 QLIST_FOREACH(c, &bs->children, next) {
1761 uint64_t cur_perm, cur_shared;
1762 bdrv_child_perm(bs, c->bs, c, c->role, q,
1763 cumulative_perms, cumulative_shared_perms,
1764 &cur_perm, &cur_shared);
1765 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1766 ignore_children, errp);
1767 if (ret < 0) {
1768 return ret;
1772 return 0;
1776 * Notifies drivers that after a previous bdrv_check_perm() call, the
1777 * permission update is not performed and any preparations made for it (e.g.
1778 * taken file locks) need to be undone.
1780 * This function recursively notifies all child nodes.
1782 static void bdrv_abort_perm_update(BlockDriverState *bs)
1784 BlockDriver *drv = bs->drv;
1785 BdrvChild *c;
1787 if (!drv) {
1788 return;
1791 if (drv->bdrv_abort_perm_update) {
1792 drv->bdrv_abort_perm_update(bs);
1795 QLIST_FOREACH(c, &bs->children, next) {
1796 bdrv_child_abort_perm_update(c);
1800 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1801 uint64_t cumulative_shared_perms)
1803 BlockDriver *drv = bs->drv;
1804 BdrvChild *c;
1806 if (!drv) {
1807 return;
1810 /* Update this node */
1811 if (drv->bdrv_set_perm) {
1812 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1815 /* Drivers that never have children can omit .bdrv_child_perm() */
1816 if (!drv->bdrv_child_perm) {
1817 assert(QLIST_EMPTY(&bs->children));
1818 return;
1821 /* Update all children */
1822 QLIST_FOREACH(c, &bs->children, next) {
1823 uint64_t cur_perm, cur_shared;
1824 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1825 cumulative_perms, cumulative_shared_perms,
1826 &cur_perm, &cur_shared);
1827 bdrv_child_set_perm(c, cur_perm, cur_shared);
1831 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1832 uint64_t *shared_perm)
1834 BdrvChild *c;
1835 uint64_t cumulative_perms = 0;
1836 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1838 QLIST_FOREACH(c, &bs->parents, next_parent) {
1839 cumulative_perms |= c->perm;
1840 cumulative_shared_perms &= c->shared_perm;
1843 *perm = cumulative_perms;
1844 *shared_perm = cumulative_shared_perms;
1847 static char *bdrv_child_user_desc(BdrvChild *c)
1849 if (c->role->get_parent_desc) {
1850 return c->role->get_parent_desc(c);
1853 return g_strdup("another user");
1856 char *bdrv_perm_names(uint64_t perm)
1858 struct perm_name {
1859 uint64_t perm;
1860 const char *name;
1861 } permissions[] = {
1862 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1863 { BLK_PERM_WRITE, "write" },
1864 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1865 { BLK_PERM_RESIZE, "resize" },
1866 { BLK_PERM_GRAPH_MOD, "change children" },
1867 { 0, NULL }
1870 char *result = g_strdup("");
1871 struct perm_name *p;
1873 for (p = permissions; p->name; p++) {
1874 if (perm & p->perm) {
1875 char *old = result;
1876 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1877 g_free(old);
1881 return result;
1885 * Checks whether a new reference to @bs can be added if the new user requires
1886 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1887 * set, the BdrvChild objects in this list are ignored in the calculations;
1888 * this allows checking permission updates for an existing reference.
1890 * Needs to be followed by a call to either bdrv_set_perm() or
1891 * bdrv_abort_perm_update(). */
1892 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1893 uint64_t new_used_perm,
1894 uint64_t new_shared_perm,
1895 GSList *ignore_children, Error **errp)
1897 BdrvChild *c;
1898 uint64_t cumulative_perms = new_used_perm;
1899 uint64_t cumulative_shared_perms = new_shared_perm;
1901 /* There is no reason why anyone couldn't tolerate write_unchanged */
1902 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1904 QLIST_FOREACH(c, &bs->parents, next_parent) {
1905 if (g_slist_find(ignore_children, c)) {
1906 continue;
1909 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1910 char *user = bdrv_child_user_desc(c);
1911 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1912 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1913 "allow '%s' on %s",
1914 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1915 g_free(user);
1916 g_free(perm_names);
1917 return -EPERM;
1920 if ((c->perm & new_shared_perm) != c->perm) {
1921 char *user = bdrv_child_user_desc(c);
1922 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1923 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1924 "'%s' on %s",
1925 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1926 g_free(user);
1927 g_free(perm_names);
1928 return -EPERM;
1931 cumulative_perms |= c->perm;
1932 cumulative_shared_perms &= c->shared_perm;
1935 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1936 ignore_children, errp);
1939 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1940 * bdrv_child_abort_perm_update(). */
1941 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1942 uint64_t perm, uint64_t shared,
1943 GSList *ignore_children, Error **errp)
1945 int ret;
1947 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1948 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1949 g_slist_free(ignore_children);
1951 return ret;
1954 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1956 uint64_t cumulative_perms, cumulative_shared_perms;
1958 c->perm = perm;
1959 c->shared_perm = shared;
1961 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1962 &cumulative_shared_perms);
1963 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1966 static void bdrv_child_abort_perm_update(BdrvChild *c)
1968 bdrv_abort_perm_update(c->bs);
1971 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1972 Error **errp)
1974 int ret;
1976 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1977 if (ret < 0) {
1978 bdrv_child_abort_perm_update(c);
1979 return ret;
1982 bdrv_child_set_perm(c, perm, shared);
1984 return 0;
1987 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1988 const BdrvChildRole *role,
1989 BlockReopenQueue *reopen_queue,
1990 uint64_t perm, uint64_t shared,
1991 uint64_t *nperm, uint64_t *nshared)
1993 if (c == NULL) {
1994 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1995 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1996 return;
1999 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
2000 (c->perm & DEFAULT_PERM_UNCHANGED);
2001 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
2002 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
2005 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2006 const BdrvChildRole *role,
2007 BlockReopenQueue *reopen_queue,
2008 uint64_t perm, uint64_t shared,
2009 uint64_t *nperm, uint64_t *nshared)
2011 bool backing = (role == &child_backing);
2012 assert(role == &child_backing || role == &child_file);
2014 if (!backing) {
2015 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2017 /* Apart from the modifications below, the same permissions are
2018 * forwarded and left alone as for filters */
2019 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2020 &perm, &shared);
2022 /* Format drivers may touch metadata even if the guest doesn't write */
2023 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2024 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2027 /* bs->file always needs to be consistent because of the metadata. We
2028 * can never allow other users to resize or write to it. */
2029 if (!(flags & BDRV_O_NO_IO)) {
2030 perm |= BLK_PERM_CONSISTENT_READ;
2032 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2033 } else {
2034 /* We want consistent read from backing files if the parent needs it.
2035 * No other operations are performed on backing files. */
2036 perm &= BLK_PERM_CONSISTENT_READ;
2038 /* If the parent can deal with changing data, we're okay with a
2039 * writable and resizable backing file. */
2040 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2041 if (shared & BLK_PERM_WRITE) {
2042 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2043 } else {
2044 shared = 0;
2047 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2048 BLK_PERM_WRITE_UNCHANGED;
2051 if (bs->open_flags & BDRV_O_INACTIVE) {
2052 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2055 *nperm = perm;
2056 *nshared = shared;
2059 static void bdrv_replace_child_noperm(BdrvChild *child,
2060 BlockDriverState *new_bs)
2062 BlockDriverState *old_bs = child->bs;
2063 int i;
2065 if (old_bs && new_bs) {
2066 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2068 if (old_bs) {
2069 /* Detach first so that the recursive drain sections coming from @child
2070 * are already gone and we only end the drain sections that came from
2071 * elsewhere. */
2072 if (child->role->detach) {
2073 child->role->detach(child);
2075 if (old_bs->quiesce_counter && child->role->drained_end) {
2076 int num = old_bs->quiesce_counter;
2077 if (child->role->parent_is_bds) {
2078 num -= bdrv_drain_all_count;
2080 assert(num >= 0);
2081 for (i = 0; i < num; i++) {
2082 child->role->drained_end(child);
2085 QLIST_REMOVE(child, next_parent);
2088 child->bs = new_bs;
2090 if (new_bs) {
2091 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2092 if (new_bs->quiesce_counter && child->role->drained_begin) {
2093 int num = new_bs->quiesce_counter;
2094 if (child->role->parent_is_bds) {
2095 num -= bdrv_drain_all_count;
2097 assert(num >= 0);
2098 for (i = 0; i < num; i++) {
2099 bdrv_parent_drained_begin_single(child, true);
2103 /* Attach only after starting new drained sections, so that recursive
2104 * drain sections coming from @child don't get an extra .drained_begin
2105 * callback. */
2106 if (child->role->attach) {
2107 child->role->attach(child);
2113 * Updates @child to change its reference to point to @new_bs, including
2114 * checking and applying the necessary permisson updates both to the old node
2115 * and to @new_bs.
2117 * NULL is passed as @new_bs for removing the reference before freeing @child.
2119 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2120 * function uses bdrv_set_perm() to update the permissions according to the new
2121 * reference that @new_bs gets.
2123 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2125 BlockDriverState *old_bs = child->bs;
2126 uint64_t perm, shared_perm;
2128 bdrv_replace_child_noperm(child, new_bs);
2130 if (old_bs) {
2131 /* Update permissions for old node. This is guaranteed to succeed
2132 * because we're just taking a parent away, so we're loosening
2133 * restrictions. */
2134 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2135 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2136 bdrv_set_perm(old_bs, perm, shared_perm);
2139 if (new_bs) {
2140 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2141 bdrv_set_perm(new_bs, perm, shared_perm);
2145 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2146 const char *child_name,
2147 const BdrvChildRole *child_role,
2148 uint64_t perm, uint64_t shared_perm,
2149 void *opaque, Error **errp)
2151 BdrvChild *child;
2152 int ret;
2154 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2155 if (ret < 0) {
2156 bdrv_abort_perm_update(child_bs);
2157 return NULL;
2160 child = g_new(BdrvChild, 1);
2161 *child = (BdrvChild) {
2162 .bs = NULL,
2163 .name = g_strdup(child_name),
2164 .role = child_role,
2165 .perm = perm,
2166 .shared_perm = shared_perm,
2167 .opaque = opaque,
2170 /* This performs the matching bdrv_set_perm() for the above check. */
2171 bdrv_replace_child(child, child_bs);
2173 return child;
2176 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2177 BlockDriverState *child_bs,
2178 const char *child_name,
2179 const BdrvChildRole *child_role,
2180 Error **errp)
2182 BdrvChild *child;
2183 uint64_t perm, shared_perm;
2185 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2187 assert(parent_bs->drv);
2188 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2189 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2190 perm, shared_perm, &perm, &shared_perm);
2192 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2193 perm, shared_perm, parent_bs, errp);
2194 if (child == NULL) {
2195 return NULL;
2198 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2199 return child;
2202 static void bdrv_detach_child(BdrvChild *child)
2204 if (child->next.le_prev) {
2205 QLIST_REMOVE(child, next);
2206 child->next.le_prev = NULL;
2209 bdrv_replace_child(child, NULL);
2211 g_free(child->name);
2212 g_free(child);
2215 void bdrv_root_unref_child(BdrvChild *child)
2217 BlockDriverState *child_bs;
2219 child_bs = child->bs;
2220 bdrv_detach_child(child);
2221 bdrv_unref(child_bs);
2224 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2226 if (child == NULL) {
2227 return;
2230 if (child->bs->inherits_from == parent) {
2231 BdrvChild *c;
2233 /* Remove inherits_from only when the last reference between parent and
2234 * child->bs goes away. */
2235 QLIST_FOREACH(c, &parent->children, next) {
2236 if (c != child && c->bs == child->bs) {
2237 break;
2240 if (c == NULL) {
2241 child->bs->inherits_from = NULL;
2245 bdrv_root_unref_child(child);
2249 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2251 BdrvChild *c;
2252 QLIST_FOREACH(c, &bs->parents, next_parent) {
2253 if (c->role->change_media) {
2254 c->role->change_media(c, load);
2259 /* Return true if you can reach parent going through child->inherits_from
2260 * recursively. If parent or child are NULL, return false */
2261 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2262 BlockDriverState *parent)
2264 while (child && child != parent) {
2265 child = child->inherits_from;
2268 return child != NULL;
2272 * Sets the backing file link of a BDS. A new reference is created; callers
2273 * which don't need their own reference any more must call bdrv_unref().
2275 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2276 Error **errp)
2278 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2279 bdrv_inherits_from_recursive(backing_hd, bs);
2281 if (backing_hd) {
2282 bdrv_ref(backing_hd);
2285 if (bs->backing) {
2286 bdrv_unref_child(bs, bs->backing);
2289 if (!backing_hd) {
2290 bs->backing = NULL;
2291 goto out;
2294 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2295 errp);
2296 /* If backing_hd was already part of bs's backing chain, and
2297 * inherits_from pointed recursively to bs then let's update it to
2298 * point directly to bs (else it will become NULL). */
2299 if (update_inherits_from) {
2300 backing_hd->inherits_from = bs;
2302 if (!bs->backing) {
2303 bdrv_unref(backing_hd);
2306 bdrv_refresh_filename(bs);
2308 out:
2309 bdrv_refresh_limits(bs, NULL);
2313 * Opens the backing file for a BlockDriverState if not yet open
2315 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2316 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2317 * itself, all options starting with "${bdref_key}." are considered part of the
2318 * BlockdevRef.
2320 * TODO Can this be unified with bdrv_open_image()?
2322 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2323 const char *bdref_key, Error **errp)
2325 char *backing_filename = g_malloc0(PATH_MAX);
2326 char *bdref_key_dot;
2327 const char *reference = NULL;
2328 int ret = 0;
2329 BlockDriverState *backing_hd;
2330 QDict *options;
2331 QDict *tmp_parent_options = NULL;
2332 Error *local_err = NULL;
2334 if (bs->backing != NULL) {
2335 goto free_exit;
2338 /* NULL means an empty set of options */
2339 if (parent_options == NULL) {
2340 tmp_parent_options = qdict_new();
2341 parent_options = tmp_parent_options;
2344 bs->open_flags &= ~BDRV_O_NO_BACKING;
2346 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2347 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2348 g_free(bdref_key_dot);
2351 * Caution: while qdict_get_try_str() is fine, getting non-string
2352 * types would require more care. When @parent_options come from
2353 * -blockdev or blockdev_add, its members are typed according to
2354 * the QAPI schema, but when they come from -drive, they're all
2355 * QString.
2357 reference = qdict_get_try_str(parent_options, bdref_key);
2358 if (reference || qdict_haskey(options, "file.filename")) {
2359 backing_filename[0] = '\0';
2360 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2361 qobject_unref(options);
2362 goto free_exit;
2363 } else {
2364 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2365 &local_err);
2366 if (local_err) {
2367 ret = -EINVAL;
2368 error_propagate(errp, local_err);
2369 qobject_unref(options);
2370 goto free_exit;
2374 if (!bs->drv || !bs->drv->supports_backing) {
2375 ret = -EINVAL;
2376 error_setg(errp, "Driver doesn't support backing files");
2377 qobject_unref(options);
2378 goto free_exit;
2381 if (!reference &&
2382 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2383 qdict_put_str(options, "driver", bs->backing_format);
2386 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2387 reference, options, 0, bs, &child_backing,
2388 errp);
2389 if (!backing_hd) {
2390 bs->open_flags |= BDRV_O_NO_BACKING;
2391 error_prepend(errp, "Could not open backing file: ");
2392 ret = -EINVAL;
2393 goto free_exit;
2395 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2397 /* Hook up the backing file link; drop our reference, bs owns the
2398 * backing_hd reference now */
2399 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2400 bdrv_unref(backing_hd);
2401 if (local_err) {
2402 error_propagate(errp, local_err);
2403 ret = -EINVAL;
2404 goto free_exit;
2407 qdict_del(parent_options, bdref_key);
2409 free_exit:
2410 g_free(backing_filename);
2411 qobject_unref(tmp_parent_options);
2412 return ret;
2415 static BlockDriverState *
2416 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2417 BlockDriverState *parent, const BdrvChildRole *child_role,
2418 bool allow_none, Error **errp)
2420 BlockDriverState *bs = NULL;
2421 QDict *image_options;
2422 char *bdref_key_dot;
2423 const char *reference;
2425 assert(child_role != NULL);
2427 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2428 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2429 g_free(bdref_key_dot);
2432 * Caution: while qdict_get_try_str() is fine, getting non-string
2433 * types would require more care. When @options come from
2434 * -blockdev or blockdev_add, its members are typed according to
2435 * the QAPI schema, but when they come from -drive, they're all
2436 * QString.
2438 reference = qdict_get_try_str(options, bdref_key);
2439 if (!filename && !reference && !qdict_size(image_options)) {
2440 if (!allow_none) {
2441 error_setg(errp, "A block device must be specified for \"%s\"",
2442 bdref_key);
2444 qobject_unref(image_options);
2445 goto done;
2448 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2449 parent, child_role, errp);
2450 if (!bs) {
2451 goto done;
2454 done:
2455 qdict_del(options, bdref_key);
2456 return bs;
2460 * Opens a disk image whose options are given as BlockdevRef in another block
2461 * device's options.
2463 * If allow_none is true, no image will be opened if filename is false and no
2464 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2466 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2467 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2468 * itself, all options starting with "${bdref_key}." are considered part of the
2469 * BlockdevRef.
2471 * The BlockdevRef will be removed from the options QDict.
2473 BdrvChild *bdrv_open_child(const char *filename,
2474 QDict *options, const char *bdref_key,
2475 BlockDriverState *parent,
2476 const BdrvChildRole *child_role,
2477 bool allow_none, Error **errp)
2479 BdrvChild *c;
2480 BlockDriverState *bs;
2482 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2483 allow_none, errp);
2484 if (bs == NULL) {
2485 return NULL;
2488 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2489 if (!c) {
2490 bdrv_unref(bs);
2491 return NULL;
2494 return c;
2497 /* TODO Future callers may need to specify parent/child_role in order for
2498 * option inheritance to work. Existing callers use it for the root node. */
2499 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2501 BlockDriverState *bs = NULL;
2502 Error *local_err = NULL;
2503 QObject *obj = NULL;
2504 QDict *qdict = NULL;
2505 const char *reference = NULL;
2506 Visitor *v = NULL;
2508 if (ref->type == QTYPE_QSTRING) {
2509 reference = ref->u.reference;
2510 } else {
2511 BlockdevOptions *options = &ref->u.definition;
2512 assert(ref->type == QTYPE_QDICT);
2514 v = qobject_output_visitor_new(&obj);
2515 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2516 if (local_err) {
2517 error_propagate(errp, local_err);
2518 goto fail;
2520 visit_complete(v, &obj);
2522 qdict = qobject_to(QDict, obj);
2523 qdict_flatten(qdict);
2525 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2526 * compatibility with other callers) rather than what we want as the
2527 * real defaults. Apply the defaults here instead. */
2528 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2529 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2530 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2531 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2535 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2536 obj = NULL;
2538 fail:
2539 qobject_unref(obj);
2540 visit_free(v);
2541 return bs;
2544 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2545 int flags,
2546 QDict *snapshot_options,
2547 Error **errp)
2549 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2550 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2551 int64_t total_size;
2552 QemuOpts *opts = NULL;
2553 BlockDriverState *bs_snapshot = NULL;
2554 Error *local_err = NULL;
2555 int ret;
2557 /* if snapshot, we create a temporary backing file and open it
2558 instead of opening 'filename' directly */
2560 /* Get the required size from the image */
2561 total_size = bdrv_getlength(bs);
2562 if (total_size < 0) {
2563 error_setg_errno(errp, -total_size, "Could not get image size");
2564 goto out;
2567 /* Create the temporary image */
2568 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2569 if (ret < 0) {
2570 error_setg_errno(errp, -ret, "Could not get temporary filename");
2571 goto out;
2574 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2575 &error_abort);
2576 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2577 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2578 qemu_opts_del(opts);
2579 if (ret < 0) {
2580 error_prepend(errp, "Could not create temporary overlay '%s': ",
2581 tmp_filename);
2582 goto out;
2585 /* Prepare options QDict for the temporary file */
2586 qdict_put_str(snapshot_options, "file.driver", "file");
2587 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2588 qdict_put_str(snapshot_options, "driver", "qcow2");
2590 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2591 snapshot_options = NULL;
2592 if (!bs_snapshot) {
2593 goto out;
2596 /* bdrv_append() consumes a strong reference to bs_snapshot
2597 * (i.e. it will call bdrv_unref() on it) even on error, so in
2598 * order to be able to return one, we have to increase
2599 * bs_snapshot's refcount here */
2600 bdrv_ref(bs_snapshot);
2601 bdrv_append(bs_snapshot, bs, &local_err);
2602 if (local_err) {
2603 error_propagate(errp, local_err);
2604 bs_snapshot = NULL;
2605 goto out;
2608 out:
2609 qobject_unref(snapshot_options);
2610 g_free(tmp_filename);
2611 return bs_snapshot;
2615 * Opens a disk image (raw, qcow2, vmdk, ...)
2617 * options is a QDict of options to pass to the block drivers, or NULL for an
2618 * empty set of options. The reference to the QDict belongs to the block layer
2619 * after the call (even on failure), so if the caller intends to reuse the
2620 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2622 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2623 * If it is not NULL, the referenced BDS will be reused.
2625 * The reference parameter may be used to specify an existing block device which
2626 * should be opened. If specified, neither options nor a filename may be given,
2627 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2629 static BlockDriverState *bdrv_open_inherit(const char *filename,
2630 const char *reference,
2631 QDict *options, int flags,
2632 BlockDriverState *parent,
2633 const BdrvChildRole *child_role,
2634 Error **errp)
2636 int ret;
2637 BlockBackend *file = NULL;
2638 BlockDriverState *bs;
2639 BlockDriver *drv = NULL;
2640 BdrvChild *child;
2641 const char *drvname;
2642 const char *backing;
2643 Error *local_err = NULL;
2644 QDict *snapshot_options = NULL;
2645 int snapshot_flags = 0;
2647 assert(!child_role || !flags);
2648 assert(!child_role == !parent);
2650 if (reference) {
2651 bool options_non_empty = options ? qdict_size(options) : false;
2652 qobject_unref(options);
2654 if (filename || options_non_empty) {
2655 error_setg(errp, "Cannot reference an existing block device with "
2656 "additional options or a new filename");
2657 return NULL;
2660 bs = bdrv_lookup_bs(reference, reference, errp);
2661 if (!bs) {
2662 return NULL;
2665 bdrv_ref(bs);
2666 return bs;
2669 bs = bdrv_new();
2671 /* NULL means an empty set of options */
2672 if (options == NULL) {
2673 options = qdict_new();
2676 /* json: syntax counts as explicit options, as if in the QDict */
2677 parse_json_protocol(options, &filename, &local_err);
2678 if (local_err) {
2679 goto fail;
2682 bs->explicit_options = qdict_clone_shallow(options);
2684 if (child_role) {
2685 bs->inherits_from = parent;
2686 child_role->inherit_options(&flags, options,
2687 parent->open_flags, parent->options);
2690 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2691 if (local_err) {
2692 goto fail;
2696 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2697 * Caution: getting a boolean member of @options requires care.
2698 * When @options come from -blockdev or blockdev_add, members are
2699 * typed according to the QAPI schema, but when they come from
2700 * -drive, they're all QString.
2702 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2703 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2704 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2705 } else {
2706 flags &= ~BDRV_O_RDWR;
2709 if (flags & BDRV_O_SNAPSHOT) {
2710 snapshot_options = qdict_new();
2711 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2712 flags, options);
2713 /* Let bdrv_backing_options() override "read-only" */
2714 qdict_del(options, BDRV_OPT_READ_ONLY);
2715 bdrv_backing_options(&flags, options, flags, options);
2718 bs->open_flags = flags;
2719 bs->options = options;
2720 options = qdict_clone_shallow(options);
2722 /* Find the right image format driver */
2723 /* See cautionary note on accessing @options above */
2724 drvname = qdict_get_try_str(options, "driver");
2725 if (drvname) {
2726 drv = bdrv_find_format(drvname);
2727 if (!drv) {
2728 error_setg(errp, "Unknown driver: '%s'", drvname);
2729 goto fail;
2733 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2735 /* See cautionary note on accessing @options above */
2736 backing = qdict_get_try_str(options, "backing");
2737 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2738 (backing && *backing == '\0'))
2740 if (backing) {
2741 warn_report("Use of \"backing\": \"\" is deprecated; "
2742 "use \"backing\": null instead");
2744 flags |= BDRV_O_NO_BACKING;
2745 qdict_del(options, "backing");
2748 /* Open image file without format layer. This BlockBackend is only used for
2749 * probing, the block drivers will do their own bdrv_open_child() for the
2750 * same BDS, which is why we put the node name back into options. */
2751 if ((flags & BDRV_O_PROTOCOL) == 0) {
2752 BlockDriverState *file_bs;
2754 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2755 &child_file, true, &local_err);
2756 if (local_err) {
2757 goto fail;
2759 if (file_bs != NULL) {
2760 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2761 * looking at the header to guess the image format. This works even
2762 * in cases where a guest would not see a consistent state. */
2763 file = blk_new(0, BLK_PERM_ALL);
2764 blk_insert_bs(file, file_bs, &local_err);
2765 bdrv_unref(file_bs);
2766 if (local_err) {
2767 goto fail;
2770 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2774 /* Image format probing */
2775 bs->probed = !drv;
2776 if (!drv && file) {
2777 ret = find_image_format(file, filename, &drv, &local_err);
2778 if (ret < 0) {
2779 goto fail;
2782 * This option update would logically belong in bdrv_fill_options(),
2783 * but we first need to open bs->file for the probing to work, while
2784 * opening bs->file already requires the (mostly) final set of options
2785 * so that cache mode etc. can be inherited.
2787 * Adding the driver later is somewhat ugly, but it's not an option
2788 * that would ever be inherited, so it's correct. We just need to make
2789 * sure to update both bs->options (which has the full effective
2790 * options for bs) and options (which has file.* already removed).
2792 qdict_put_str(bs->options, "driver", drv->format_name);
2793 qdict_put_str(options, "driver", drv->format_name);
2794 } else if (!drv) {
2795 error_setg(errp, "Must specify either driver or file");
2796 goto fail;
2799 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2800 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2801 /* file must be NULL if a protocol BDS is about to be created
2802 * (the inverse results in an error message from bdrv_open_common()) */
2803 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2805 /* Open the image */
2806 ret = bdrv_open_common(bs, file, options, &local_err);
2807 if (ret < 0) {
2808 goto fail;
2811 if (file) {
2812 blk_unref(file);
2813 file = NULL;
2816 /* If there is a backing file, use it */
2817 if ((flags & BDRV_O_NO_BACKING) == 0) {
2818 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2819 if (ret < 0) {
2820 goto close_and_fail;
2824 /* Remove all children options and references
2825 * from bs->options and bs->explicit_options */
2826 QLIST_FOREACH(child, &bs->children, next) {
2827 char *child_key_dot;
2828 child_key_dot = g_strdup_printf("%s.", child->name);
2829 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
2830 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
2831 qdict_del(bs->explicit_options, child->name);
2832 qdict_del(bs->options, child->name);
2833 g_free(child_key_dot);
2836 bdrv_refresh_filename(bs);
2838 /* Check if any unknown options were used */
2839 if (qdict_size(options) != 0) {
2840 const QDictEntry *entry = qdict_first(options);
2841 if (flags & BDRV_O_PROTOCOL) {
2842 error_setg(errp, "Block protocol '%s' doesn't support the option "
2843 "'%s'", drv->format_name, entry->key);
2844 } else {
2845 error_setg(errp,
2846 "Block format '%s' does not support the option '%s'",
2847 drv->format_name, entry->key);
2850 goto close_and_fail;
2853 bdrv_parent_cb_change_media(bs, true);
2855 qobject_unref(options);
2856 options = NULL;
2858 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2859 * temporary snapshot afterwards. */
2860 if (snapshot_flags) {
2861 BlockDriverState *snapshot_bs;
2862 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2863 snapshot_options, &local_err);
2864 snapshot_options = NULL;
2865 if (local_err) {
2866 goto close_and_fail;
2868 /* We are not going to return bs but the overlay on top of it
2869 * (snapshot_bs); thus, we have to drop the strong reference to bs
2870 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2871 * though, because the overlay still has a reference to it. */
2872 bdrv_unref(bs);
2873 bs = snapshot_bs;
2876 return bs;
2878 fail:
2879 blk_unref(file);
2880 qobject_unref(snapshot_options);
2881 qobject_unref(bs->explicit_options);
2882 qobject_unref(bs->options);
2883 qobject_unref(options);
2884 bs->options = NULL;
2885 bs->explicit_options = NULL;
2886 bdrv_unref(bs);
2887 error_propagate(errp, local_err);
2888 return NULL;
2890 close_and_fail:
2891 bdrv_unref(bs);
2892 qobject_unref(snapshot_options);
2893 qobject_unref(options);
2894 error_propagate(errp, local_err);
2895 return NULL;
2898 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2899 QDict *options, int flags, Error **errp)
2901 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2902 NULL, errp);
2906 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2907 * reopen of multiple devices.
2909 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2910 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2911 * be created and initialized. This newly created BlockReopenQueue should be
2912 * passed back in for subsequent calls that are intended to be of the same
2913 * atomic 'set'.
2915 * bs is the BlockDriverState to add to the reopen queue.
2917 * options contains the changed options for the associated bs
2918 * (the BlockReopenQueue takes ownership)
2920 * flags contains the open flags for the associated bs
2922 * returns a pointer to bs_queue, which is either the newly allocated
2923 * bs_queue, or the existing bs_queue being used.
2925 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2927 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2928 BlockDriverState *bs,
2929 QDict *options,
2930 const BdrvChildRole *role,
2931 QDict *parent_options,
2932 int parent_flags)
2934 assert(bs != NULL);
2936 BlockReopenQueueEntry *bs_entry;
2937 BdrvChild *child;
2938 QDict *old_options, *explicit_options, *options_copy;
2939 int flags;
2940 QemuOpts *opts;
2942 /* Make sure that the caller remembered to use a drained section. This is
2943 * important to avoid graph changes between the recursive queuing here and
2944 * bdrv_reopen_multiple(). */
2945 assert(bs->quiesce_counter > 0);
2947 if (bs_queue == NULL) {
2948 bs_queue = g_new0(BlockReopenQueue, 1);
2949 QSIMPLEQ_INIT(bs_queue);
2952 if (!options) {
2953 options = qdict_new();
2956 /* Check if this BlockDriverState is already in the queue */
2957 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2958 if (bs == bs_entry->state.bs) {
2959 break;
2964 * Precedence of options:
2965 * 1. Explicitly passed in options (highest)
2966 * 2. Retained from explicitly set options of bs
2967 * 3. Inherited from parent node
2968 * 4. Retained from effective options of bs
2971 /* Old explicitly set values (don't overwrite by inherited value) */
2972 if (bs_entry) {
2973 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2974 } else {
2975 old_options = qdict_clone_shallow(bs->explicit_options);
2977 bdrv_join_options(bs, options, old_options);
2978 qobject_unref(old_options);
2980 explicit_options = qdict_clone_shallow(options);
2982 /* Inherit from parent node */
2983 if (parent_options) {
2984 flags = 0;
2985 role->inherit_options(&flags, options, parent_flags, parent_options);
2986 } else {
2987 flags = bdrv_get_flags(bs);
2990 /* Old values are used for options that aren't set yet */
2991 old_options = qdict_clone_shallow(bs->options);
2992 bdrv_join_options(bs, options, old_options);
2993 qobject_unref(old_options);
2995 /* We have the final set of options so let's update the flags */
2996 options_copy = qdict_clone_shallow(options);
2997 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2998 qemu_opts_absorb_qdict(opts, options_copy, NULL);
2999 update_flags_from_options(&flags, opts);
3000 qemu_opts_del(opts);
3001 qobject_unref(options_copy);
3003 /* bdrv_open_inherit() sets and clears some additional flags internally */
3004 flags &= ~BDRV_O_PROTOCOL;
3005 if (flags & BDRV_O_RDWR) {
3006 flags |= BDRV_O_ALLOW_RDWR;
3009 if (!bs_entry) {
3010 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3011 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3012 } else {
3013 qobject_unref(bs_entry->state.options);
3014 qobject_unref(bs_entry->state.explicit_options);
3017 bs_entry->state.bs = bs;
3018 bs_entry->state.options = options;
3019 bs_entry->state.explicit_options = explicit_options;
3020 bs_entry->state.flags = flags;
3022 /* This needs to be overwritten in bdrv_reopen_prepare() */
3023 bs_entry->state.perm = UINT64_MAX;
3024 bs_entry->state.shared_perm = 0;
3026 QLIST_FOREACH(child, &bs->children, next) {
3027 QDict *new_child_options;
3028 char *child_key_dot;
3030 /* reopen can only change the options of block devices that were
3031 * implicitly created and inherited options. For other (referenced)
3032 * block devices, a syntax like "backing.foo" results in an error. */
3033 if (child->bs->inherits_from != bs) {
3034 continue;
3037 child_key_dot = g_strdup_printf("%s.", child->name);
3038 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3039 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3040 g_free(child_key_dot);
3042 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3043 child->role, options, flags);
3046 return bs_queue;
3049 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3050 BlockDriverState *bs,
3051 QDict *options)
3053 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0);
3057 * Reopen multiple BlockDriverStates atomically & transactionally.
3059 * The queue passed in (bs_queue) must have been built up previous
3060 * via bdrv_reopen_queue().
3062 * Reopens all BDS specified in the queue, with the appropriate
3063 * flags. All devices are prepared for reopen, and failure of any
3064 * device will cause all device changes to be abandoned, and intermediate
3065 * data cleaned up.
3067 * If all devices prepare successfully, then the changes are committed
3068 * to all devices.
3070 * All affected nodes must be drained between bdrv_reopen_queue() and
3071 * bdrv_reopen_multiple().
3073 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
3075 int ret = -1;
3076 BlockReopenQueueEntry *bs_entry, *next;
3077 Error *local_err = NULL;
3079 assert(bs_queue != NULL);
3081 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3082 assert(bs_entry->state.bs->quiesce_counter > 0);
3083 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
3084 error_propagate(errp, local_err);
3085 goto cleanup;
3087 bs_entry->prepared = true;
3090 /* If we reach this point, we have success and just need to apply the
3091 * changes
3093 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3094 bdrv_reopen_commit(&bs_entry->state);
3097 ret = 0;
3099 cleanup:
3100 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3101 if (ret) {
3102 if (bs_entry->prepared) {
3103 bdrv_reopen_abort(&bs_entry->state);
3105 qobject_unref(bs_entry->state.explicit_options);
3106 qobject_unref(bs_entry->state.options);
3108 g_free(bs_entry);
3110 g_free(bs_queue);
3112 return ret;
3115 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3116 Error **errp)
3118 int ret;
3119 BlockReopenQueue *queue;
3120 QDict *opts = qdict_new();
3122 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3124 bdrv_subtree_drained_begin(bs);
3125 queue = bdrv_reopen_queue(NULL, bs, opts);
3126 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, errp);
3127 bdrv_subtree_drained_end(bs);
3129 return ret;
3132 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3133 BdrvChild *c)
3135 BlockReopenQueueEntry *entry;
3137 QSIMPLEQ_FOREACH(entry, q, entry) {
3138 BlockDriverState *bs = entry->state.bs;
3139 BdrvChild *child;
3141 QLIST_FOREACH(child, &bs->children, next) {
3142 if (child == c) {
3143 return entry;
3148 return NULL;
3151 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3152 uint64_t *perm, uint64_t *shared)
3154 BdrvChild *c;
3155 BlockReopenQueueEntry *parent;
3156 uint64_t cumulative_perms = 0;
3157 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3159 QLIST_FOREACH(c, &bs->parents, next_parent) {
3160 parent = find_parent_in_reopen_queue(q, c);
3161 if (!parent) {
3162 cumulative_perms |= c->perm;
3163 cumulative_shared_perms &= c->shared_perm;
3164 } else {
3165 uint64_t nperm, nshared;
3167 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3168 parent->state.perm, parent->state.shared_perm,
3169 &nperm, &nshared);
3171 cumulative_perms |= nperm;
3172 cumulative_shared_perms &= nshared;
3175 *perm = cumulative_perms;
3176 *shared = cumulative_shared_perms;
3180 * Prepares a BlockDriverState for reopen. All changes are staged in the
3181 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3182 * the block driver layer .bdrv_reopen_prepare()
3184 * bs is the BlockDriverState to reopen
3185 * flags are the new open flags
3186 * queue is the reopen queue
3188 * Returns 0 on success, non-zero on error. On error errp will be set
3189 * as well.
3191 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3192 * It is the responsibility of the caller to then call the abort() or
3193 * commit() for any other BDS that have been left in a prepare() state
3196 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3197 Error **errp)
3199 int ret = -1;
3200 int old_flags;
3201 Error *local_err = NULL;
3202 BlockDriver *drv;
3203 QemuOpts *opts;
3204 QDict *orig_reopen_opts;
3205 char *discard = NULL;
3206 bool read_only;
3207 bool drv_prepared = false;
3209 assert(reopen_state != NULL);
3210 assert(reopen_state->bs->drv != NULL);
3211 drv = reopen_state->bs->drv;
3213 /* This function and each driver's bdrv_reopen_prepare() remove
3214 * entries from reopen_state->options as they are processed, so
3215 * we need to make a copy of the original QDict. */
3216 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3218 /* Process generic block layer options */
3219 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3220 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3221 if (local_err) {
3222 error_propagate(errp, local_err);
3223 ret = -EINVAL;
3224 goto error;
3227 /* This was already called in bdrv_reopen_queue_child() so the flags
3228 * are up-to-date. This time we simply want to remove the options from
3229 * QemuOpts in order to indicate that they have been processed. */
3230 old_flags = reopen_state->flags;
3231 update_flags_from_options(&reopen_state->flags, opts);
3232 assert(old_flags == reopen_state->flags);
3234 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
3235 if (discard != NULL) {
3236 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3237 error_setg(errp, "Invalid discard option");
3238 ret = -EINVAL;
3239 goto error;
3243 reopen_state->detect_zeroes =
3244 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3245 if (local_err) {
3246 error_propagate(errp, local_err);
3247 ret = -EINVAL;
3248 goto error;
3251 /* All other options (including node-name and driver) must be unchanged.
3252 * Put them back into the QDict, so that they are checked at the end
3253 * of this function. */
3254 qemu_opts_to_qdict(opts, reopen_state->options);
3256 /* If we are to stay read-only, do not allow permission change
3257 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3258 * not set, or if the BDS still has copy_on_read enabled */
3259 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3260 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3261 if (local_err) {
3262 error_propagate(errp, local_err);
3263 goto error;
3266 /* Calculate required permissions after reopening */
3267 bdrv_reopen_perm(queue, reopen_state->bs,
3268 &reopen_state->perm, &reopen_state->shared_perm);
3270 ret = bdrv_flush(reopen_state->bs);
3271 if (ret) {
3272 error_setg_errno(errp, -ret, "Error flushing drive");
3273 goto error;
3276 if (drv->bdrv_reopen_prepare) {
3277 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3278 if (ret) {
3279 if (local_err != NULL) {
3280 error_propagate(errp, local_err);
3281 } else {
3282 error_setg(errp, "failed while preparing to reopen image '%s'",
3283 reopen_state->bs->filename);
3285 goto error;
3287 } else {
3288 /* It is currently mandatory to have a bdrv_reopen_prepare()
3289 * handler for each supported drv. */
3290 error_setg(errp, "Block format '%s' used by node '%s' "
3291 "does not support reopening files", drv->format_name,
3292 bdrv_get_device_or_node_name(reopen_state->bs));
3293 ret = -1;
3294 goto error;
3297 drv_prepared = true;
3299 /* Options that are not handled are only okay if they are unchanged
3300 * compared to the old state. It is expected that some options are only
3301 * used for the initial open, but not reopen (e.g. filename) */
3302 if (qdict_size(reopen_state->options)) {
3303 const QDictEntry *entry = qdict_first(reopen_state->options);
3305 do {
3306 QObject *new = entry->value;
3307 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3309 /* Allow child references (child_name=node_name) as long as they
3310 * point to the current child (i.e. everything stays the same). */
3311 if (qobject_type(new) == QTYPE_QSTRING) {
3312 BdrvChild *child;
3313 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
3314 if (!strcmp(child->name, entry->key)) {
3315 break;
3319 if (child) {
3320 const char *str = qobject_get_try_str(new);
3321 if (!strcmp(child->bs->node_name, str)) {
3322 continue; /* Found child with this name, skip option */
3328 * TODO: When using -drive to specify blockdev options, all values
3329 * will be strings; however, when using -blockdev, blockdev-add or
3330 * filenames using the json:{} pseudo-protocol, they will be
3331 * correctly typed.
3332 * In contrast, reopening options are (currently) always strings
3333 * (because you can only specify them through qemu-io; all other
3334 * callers do not specify any options).
3335 * Therefore, when using anything other than -drive to create a BDS,
3336 * this cannot detect non-string options as unchanged, because
3337 * qobject_is_equal() always returns false for objects of different
3338 * type. In the future, this should be remedied by correctly typing
3339 * all options. For now, this is not too big of an issue because
3340 * the user can simply omit options which cannot be changed anyway,
3341 * so they will stay unchanged.
3343 if (!qobject_is_equal(new, old)) {
3344 error_setg(errp, "Cannot change the option '%s'", entry->key);
3345 ret = -EINVAL;
3346 goto error;
3348 } while ((entry = qdict_next(reopen_state->options, entry)));
3351 ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3352 reopen_state->shared_perm, NULL, errp);
3353 if (ret < 0) {
3354 goto error;
3357 ret = 0;
3359 /* Restore the original reopen_state->options QDict */
3360 qobject_unref(reopen_state->options);
3361 reopen_state->options = qobject_ref(orig_reopen_opts);
3363 error:
3364 if (ret < 0 && drv_prepared) {
3365 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3366 * call drv->bdrv_reopen_abort() before signaling an error
3367 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3368 * when the respective bdrv_reopen_prepare() has failed) */
3369 if (drv->bdrv_reopen_abort) {
3370 drv->bdrv_reopen_abort(reopen_state);
3373 qemu_opts_del(opts);
3374 qobject_unref(orig_reopen_opts);
3375 g_free(discard);
3376 return ret;
3380 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3381 * makes them final by swapping the staging BlockDriverState contents into
3382 * the active BlockDriverState contents.
3384 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3386 BlockDriver *drv;
3387 BlockDriverState *bs;
3388 BdrvChild *child;
3389 bool old_can_write, new_can_write;
3391 assert(reopen_state != NULL);
3392 bs = reopen_state->bs;
3393 drv = bs->drv;
3394 assert(drv != NULL);
3396 old_can_write =
3397 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3399 /* If there are any driver level actions to take */
3400 if (drv->bdrv_reopen_commit) {
3401 drv->bdrv_reopen_commit(reopen_state);
3404 /* set BDS specific flags now */
3405 qobject_unref(bs->explicit_options);
3406 qobject_unref(bs->options);
3408 bs->explicit_options = reopen_state->explicit_options;
3409 bs->options = reopen_state->options;
3410 bs->open_flags = reopen_state->flags;
3411 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3412 bs->detect_zeroes = reopen_state->detect_zeroes;
3414 /* Remove child references from bs->options and bs->explicit_options.
3415 * Child options were already removed in bdrv_reopen_queue_child() */
3416 QLIST_FOREACH(child, &bs->children, next) {
3417 qdict_del(bs->explicit_options, child->name);
3418 qdict_del(bs->options, child->name);
3421 bdrv_refresh_limits(bs, NULL);
3423 bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3424 reopen_state->shared_perm);
3426 new_can_write =
3427 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3428 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3429 Error *local_err = NULL;
3430 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3431 /* This is not fatal, bitmaps just left read-only, so all following
3432 * writes will fail. User can remove read-only bitmaps to unblock
3433 * writes.
3435 error_reportf_err(local_err,
3436 "%s: Failed to make dirty bitmaps writable: ",
3437 bdrv_get_node_name(bs));
3443 * Abort the reopen, and delete and free the staged changes in
3444 * reopen_state
3446 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3448 BlockDriver *drv;
3450 assert(reopen_state != NULL);
3451 drv = reopen_state->bs->drv;
3452 assert(drv != NULL);
3454 if (drv->bdrv_reopen_abort) {
3455 drv->bdrv_reopen_abort(reopen_state);
3458 bdrv_abort_perm_update(reopen_state->bs);
3462 static void bdrv_close(BlockDriverState *bs)
3464 BdrvAioNotifier *ban, *ban_next;
3465 BdrvChild *child, *next;
3467 assert(!bs->job);
3468 assert(!bs->refcnt);
3470 bdrv_drained_begin(bs); /* complete I/O */
3471 bdrv_flush(bs);
3472 bdrv_drain(bs); /* in case flush left pending I/O */
3474 if (bs->drv) {
3475 if (bs->drv->bdrv_close) {
3476 bs->drv->bdrv_close(bs);
3478 bs->drv = NULL;
3481 bdrv_set_backing_hd(bs, NULL, &error_abort);
3483 if (bs->file != NULL) {
3484 bdrv_unref_child(bs, bs->file);
3485 bs->file = NULL;
3488 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3489 /* TODO Remove bdrv_unref() from drivers' close function and use
3490 * bdrv_unref_child() here */
3491 if (child->bs->inherits_from == bs) {
3492 child->bs->inherits_from = NULL;
3494 bdrv_detach_child(child);
3497 g_free(bs->opaque);
3498 bs->opaque = NULL;
3499 atomic_set(&bs->copy_on_read, 0);
3500 bs->backing_file[0] = '\0';
3501 bs->backing_format[0] = '\0';
3502 bs->total_sectors = 0;
3503 bs->encrypted = false;
3504 bs->sg = false;
3505 qobject_unref(bs->options);
3506 qobject_unref(bs->explicit_options);
3507 bs->options = NULL;
3508 bs->explicit_options = NULL;
3509 qobject_unref(bs->full_open_options);
3510 bs->full_open_options = NULL;
3512 bdrv_release_named_dirty_bitmaps(bs);
3513 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3515 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3516 g_free(ban);
3518 QLIST_INIT(&bs->aio_notifiers);
3519 bdrv_drained_end(bs);
3522 void bdrv_close_all(void)
3524 assert(job_next(NULL) == NULL);
3525 nbd_export_close_all();
3527 /* Drop references from requests still in flight, such as canceled block
3528 * jobs whose AIO context has not been polled yet */
3529 bdrv_drain_all();
3531 blk_remove_all_bs();
3532 blockdev_close_all_bdrv_states();
3534 assert(QTAILQ_EMPTY(&all_bdrv_states));
3537 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3539 BdrvChild *to_c;
3541 if (c->role->stay_at_node) {
3542 return false;
3545 /* If the child @c belongs to the BDS @to, replacing the current
3546 * c->bs by @to would mean to create a loop.
3548 * Such a case occurs when appending a BDS to a backing chain.
3549 * For instance, imagine the following chain:
3551 * guest device -> node A -> further backing chain...
3553 * Now we create a new BDS B which we want to put on top of this
3554 * chain, so we first attach A as its backing node:
3556 * node B
3559 * guest device -> node A -> further backing chain...
3561 * Finally we want to replace A by B. When doing that, we want to
3562 * replace all pointers to A by pointers to B -- except for the
3563 * pointer from B because (1) that would create a loop, and (2)
3564 * that pointer should simply stay intact:
3566 * guest device -> node B
3569 * node A -> further backing chain...
3571 * In general, when replacing a node A (c->bs) by a node B (@to),
3572 * if A is a child of B, that means we cannot replace A by B there
3573 * because that would create a loop. Silently detaching A from B
3574 * is also not really an option. So overall just leaving A in
3575 * place there is the most sensible choice. */
3576 QLIST_FOREACH(to_c, &to->children, next) {
3577 if (to_c == c) {
3578 return false;
3582 return true;
3585 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3586 Error **errp)
3588 BdrvChild *c, *next;
3589 GSList *list = NULL, *p;
3590 uint64_t old_perm, old_shared;
3591 uint64_t perm = 0, shared = BLK_PERM_ALL;
3592 int ret;
3594 assert(!atomic_read(&from->in_flight));
3595 assert(!atomic_read(&to->in_flight));
3597 /* Make sure that @from doesn't go away until we have successfully attached
3598 * all of its parents to @to. */
3599 bdrv_ref(from);
3601 /* Put all parents into @list and calculate their cumulative permissions */
3602 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3603 assert(c->bs == from);
3604 if (!should_update_child(c, to)) {
3605 continue;
3607 list = g_slist_prepend(list, c);
3608 perm |= c->perm;
3609 shared &= c->shared_perm;
3612 /* Check whether the required permissions can be granted on @to, ignoring
3613 * all BdrvChild in @list so that they can't block themselves. */
3614 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3615 if (ret < 0) {
3616 bdrv_abort_perm_update(to);
3617 goto out;
3620 /* Now actually perform the change. We performed the permission check for
3621 * all elements of @list at once, so set the permissions all at once at the
3622 * very end. */
3623 for (p = list; p != NULL; p = p->next) {
3624 c = p->data;
3626 bdrv_ref(to);
3627 bdrv_replace_child_noperm(c, to);
3628 bdrv_unref(from);
3631 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3632 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3634 out:
3635 g_slist_free(list);
3636 bdrv_unref(from);
3640 * Add new bs contents at the top of an image chain while the chain is
3641 * live, while keeping required fields on the top layer.
3643 * This will modify the BlockDriverState fields, and swap contents
3644 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3646 * bs_new must not be attached to a BlockBackend.
3648 * This function does not create any image files.
3650 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3651 * that's what the callers commonly need. bs_new will be referenced by the old
3652 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3653 * reference of its own, it must call bdrv_ref().
3655 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3656 Error **errp)
3658 Error *local_err = NULL;
3660 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3661 if (local_err) {
3662 error_propagate(errp, local_err);
3663 goto out;
3666 bdrv_replace_node(bs_top, bs_new, &local_err);
3667 if (local_err) {
3668 error_propagate(errp, local_err);
3669 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3670 goto out;
3673 /* bs_new is now referenced by its new parents, we don't need the
3674 * additional reference any more. */
3675 out:
3676 bdrv_unref(bs_new);
3679 static void bdrv_delete(BlockDriverState *bs)
3681 assert(!bs->job);
3682 assert(bdrv_op_blocker_is_empty(bs));
3683 assert(!bs->refcnt);
3685 bdrv_close(bs);
3687 /* remove from list, if necessary */
3688 if (bs->node_name[0] != '\0') {
3689 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3691 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3693 g_free(bs);
3697 * Run consistency checks on an image
3699 * Returns 0 if the check could be completed (it doesn't mean that the image is
3700 * free of errors) or -errno when an internal error occurred. The results of the
3701 * check are stored in res.
3703 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3704 BdrvCheckResult *res, BdrvCheckMode fix)
3706 if (bs->drv == NULL) {
3707 return -ENOMEDIUM;
3709 if (bs->drv->bdrv_co_check == NULL) {
3710 return -ENOTSUP;
3713 memset(res, 0, sizeof(*res));
3714 return bs->drv->bdrv_co_check(bs, res, fix);
3717 typedef struct CheckCo {
3718 BlockDriverState *bs;
3719 BdrvCheckResult *res;
3720 BdrvCheckMode fix;
3721 int ret;
3722 } CheckCo;
3724 static void bdrv_check_co_entry(void *opaque)
3726 CheckCo *cco = opaque;
3727 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3730 int bdrv_check(BlockDriverState *bs,
3731 BdrvCheckResult *res, BdrvCheckMode fix)
3733 Coroutine *co;
3734 CheckCo cco = {
3735 .bs = bs,
3736 .res = res,
3737 .ret = -EINPROGRESS,
3738 .fix = fix,
3741 if (qemu_in_coroutine()) {
3742 /* Fast-path if already in coroutine context */
3743 bdrv_check_co_entry(&cco);
3744 } else {
3745 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3746 qemu_coroutine_enter(co);
3747 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3750 return cco.ret;
3754 * Return values:
3755 * 0 - success
3756 * -EINVAL - backing format specified, but no file
3757 * -ENOSPC - can't update the backing file because no space is left in the
3758 * image file header
3759 * -ENOTSUP - format driver doesn't support changing the backing file
3761 int bdrv_change_backing_file(BlockDriverState *bs,
3762 const char *backing_file, const char *backing_fmt)
3764 BlockDriver *drv = bs->drv;
3765 int ret;
3767 if (!drv) {
3768 return -ENOMEDIUM;
3771 /* Backing file format doesn't make sense without a backing file */
3772 if (backing_fmt && !backing_file) {
3773 return -EINVAL;
3776 if (drv->bdrv_change_backing_file != NULL) {
3777 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3778 } else {
3779 ret = -ENOTSUP;
3782 if (ret == 0) {
3783 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3784 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3786 return ret;
3790 * Finds the image layer in the chain that has 'bs' as its backing file.
3792 * active is the current topmost image.
3794 * Returns NULL if bs is not found in active's image chain,
3795 * or if active == bs.
3797 * Returns the bottommost base image if bs == NULL.
3799 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3800 BlockDriverState *bs)
3802 while (active && bs != backing_bs(active)) {
3803 active = backing_bs(active);
3806 return active;
3809 /* Given a BDS, searches for the base layer. */
3810 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3812 return bdrv_find_overlay(bs, NULL);
3816 * Drops images above 'base' up to and including 'top', and sets the image
3817 * above 'top' to have base as its backing file.
3819 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3820 * information in 'bs' can be properly updated.
3822 * E.g., this will convert the following chain:
3823 * bottom <- base <- intermediate <- top <- active
3825 * to
3827 * bottom <- base <- active
3829 * It is allowed for bottom==base, in which case it converts:
3831 * base <- intermediate <- top <- active
3833 * to
3835 * base <- active
3837 * If backing_file_str is non-NULL, it will be used when modifying top's
3838 * overlay image metadata.
3840 * Error conditions:
3841 * if active == top, that is considered an error
3844 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3845 const char *backing_file_str)
3847 BlockDriverState *explicit_top = top;
3848 bool update_inherits_from;
3849 BdrvChild *c, *next;
3850 Error *local_err = NULL;
3851 int ret = -EIO;
3853 bdrv_ref(top);
3855 if (!top->drv || !base->drv) {
3856 goto exit;
3859 /* Make sure that base is in the backing chain of top */
3860 if (!bdrv_chain_contains(top, base)) {
3861 goto exit;
3864 /* If 'base' recursively inherits from 'top' then we should set
3865 * base->inherits_from to top->inherits_from after 'top' and all
3866 * other intermediate nodes have been dropped.
3867 * If 'top' is an implicit node (e.g. "commit_top") we should skip
3868 * it because no one inherits from it. We use explicit_top for that. */
3869 while (explicit_top && explicit_top->implicit) {
3870 explicit_top = backing_bs(explicit_top);
3872 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
3874 /* success - we can delete the intermediate states, and link top->base */
3875 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3876 * we've figured out how they should work. */
3877 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3879 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3880 /* Check whether we are allowed to switch c from top to base */
3881 GSList *ignore_children = g_slist_prepend(NULL, c);
3882 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3883 ignore_children, &local_err);
3884 g_slist_free(ignore_children);
3885 if (local_err) {
3886 ret = -EPERM;
3887 error_report_err(local_err);
3888 goto exit;
3891 /* If so, update the backing file path in the image file */
3892 if (c->role->update_filename) {
3893 ret = c->role->update_filename(c, base, backing_file_str,
3894 &local_err);
3895 if (ret < 0) {
3896 bdrv_abort_perm_update(base);
3897 error_report_err(local_err);
3898 goto exit;
3902 /* Do the actual switch in the in-memory graph.
3903 * Completes bdrv_check_update_perm() transaction internally. */
3904 bdrv_ref(base);
3905 bdrv_replace_child(c, base);
3906 bdrv_unref(top);
3909 if (update_inherits_from) {
3910 base->inherits_from = explicit_top->inherits_from;
3913 ret = 0;
3914 exit:
3915 bdrv_unref(top);
3916 return ret;
3920 * Length of a allocated file in bytes. Sparse files are counted by actual
3921 * allocated space. Return < 0 if error or unknown.
3923 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3925 BlockDriver *drv = bs->drv;
3926 if (!drv) {
3927 return -ENOMEDIUM;
3929 if (drv->bdrv_get_allocated_file_size) {
3930 return drv->bdrv_get_allocated_file_size(bs);
3932 if (bs->file) {
3933 return bdrv_get_allocated_file_size(bs->file->bs);
3935 return -ENOTSUP;
3939 * bdrv_measure:
3940 * @drv: Format driver
3941 * @opts: Creation options for new image
3942 * @in_bs: Existing image containing data for new image (may be NULL)
3943 * @errp: Error object
3944 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3945 * or NULL on error
3947 * Calculate file size required to create a new image.
3949 * If @in_bs is given then space for allocated clusters and zero clusters
3950 * from that image are included in the calculation. If @opts contains a
3951 * backing file that is shared by @in_bs then backing clusters may be omitted
3952 * from the calculation.
3954 * If @in_bs is NULL then the calculation includes no allocated clusters
3955 * unless a preallocation option is given in @opts.
3957 * Note that @in_bs may use a different BlockDriver from @drv.
3959 * If an error occurs the @errp pointer is set.
3961 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3962 BlockDriverState *in_bs, Error **errp)
3964 if (!drv->bdrv_measure) {
3965 error_setg(errp, "Block driver '%s' does not support size measurement",
3966 drv->format_name);
3967 return NULL;
3970 return drv->bdrv_measure(opts, in_bs, errp);
3974 * Return number of sectors on success, -errno on error.
3976 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3978 BlockDriver *drv = bs->drv;
3980 if (!drv)
3981 return -ENOMEDIUM;
3983 if (drv->has_variable_length) {
3984 int ret = refresh_total_sectors(bs, bs->total_sectors);
3985 if (ret < 0) {
3986 return ret;
3989 return bs->total_sectors;
3993 * Return length in bytes on success, -errno on error.
3994 * The length is always a multiple of BDRV_SECTOR_SIZE.
3996 int64_t bdrv_getlength(BlockDriverState *bs)
3998 int64_t ret = bdrv_nb_sectors(bs);
4000 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
4001 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
4004 /* return 0 as number of sectors if no device present or error */
4005 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
4007 int64_t nb_sectors = bdrv_nb_sectors(bs);
4009 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
4012 bool bdrv_is_sg(BlockDriverState *bs)
4014 return bs->sg;
4017 bool bdrv_is_encrypted(BlockDriverState *bs)
4019 if (bs->backing && bs->backing->bs->encrypted) {
4020 return true;
4022 return bs->encrypted;
4025 const char *bdrv_get_format_name(BlockDriverState *bs)
4027 return bs->drv ? bs->drv->format_name : NULL;
4030 static int qsort_strcmp(const void *a, const void *b)
4032 return strcmp(*(char *const *)a, *(char *const *)b);
4035 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
4036 void *opaque)
4038 BlockDriver *drv;
4039 int count = 0;
4040 int i;
4041 const char **formats = NULL;
4043 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4044 if (drv->format_name) {
4045 bool found = false;
4046 int i = count;
4047 while (formats && i && !found) {
4048 found = !strcmp(formats[--i], drv->format_name);
4051 if (!found) {
4052 formats = g_renew(const char *, formats, count + 1);
4053 formats[count++] = drv->format_name;
4058 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4059 const char *format_name = block_driver_modules[i].format_name;
4061 if (format_name) {
4062 bool found = false;
4063 int j = count;
4065 while (formats && j && !found) {
4066 found = !strcmp(formats[--j], format_name);
4069 if (!found) {
4070 formats = g_renew(const char *, formats, count + 1);
4071 formats[count++] = format_name;
4076 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
4078 for (i = 0; i < count; i++) {
4079 it(opaque, formats[i]);
4082 g_free(formats);
4085 /* This function is to find a node in the bs graph */
4086 BlockDriverState *bdrv_find_node(const char *node_name)
4088 BlockDriverState *bs;
4090 assert(node_name);
4092 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4093 if (!strcmp(node_name, bs->node_name)) {
4094 return bs;
4097 return NULL;
4100 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4101 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
4103 BlockDeviceInfoList *list, *entry;
4104 BlockDriverState *bs;
4106 list = NULL;
4107 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4108 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
4109 if (!info) {
4110 qapi_free_BlockDeviceInfoList(list);
4111 return NULL;
4113 entry = g_malloc0(sizeof(*entry));
4114 entry->value = info;
4115 entry->next = list;
4116 list = entry;
4119 return list;
4122 BlockDriverState *bdrv_lookup_bs(const char *device,
4123 const char *node_name,
4124 Error **errp)
4126 BlockBackend *blk;
4127 BlockDriverState *bs;
4129 if (device) {
4130 blk = blk_by_name(device);
4132 if (blk) {
4133 bs = blk_bs(blk);
4134 if (!bs) {
4135 error_setg(errp, "Device '%s' has no medium", device);
4138 return bs;
4142 if (node_name) {
4143 bs = bdrv_find_node(node_name);
4145 if (bs) {
4146 return bs;
4150 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4151 device ? device : "",
4152 node_name ? node_name : "");
4153 return NULL;
4156 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4157 * return false. If either argument is NULL, return false. */
4158 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4160 while (top && top != base) {
4161 top = backing_bs(top);
4164 return top != NULL;
4167 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4169 if (!bs) {
4170 return QTAILQ_FIRST(&graph_bdrv_states);
4172 return QTAILQ_NEXT(bs, node_list);
4175 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
4177 if (!bs) {
4178 return QTAILQ_FIRST(&all_bdrv_states);
4180 return QTAILQ_NEXT(bs, bs_list);
4183 const char *bdrv_get_node_name(const BlockDriverState *bs)
4185 return bs->node_name;
4188 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4190 BdrvChild *c;
4191 const char *name;
4193 /* If multiple parents have a name, just pick the first one. */
4194 QLIST_FOREACH(c, &bs->parents, next_parent) {
4195 if (c->role->get_name) {
4196 name = c->role->get_name(c);
4197 if (name && *name) {
4198 return name;
4203 return NULL;
4206 /* TODO check what callers really want: bs->node_name or blk_name() */
4207 const char *bdrv_get_device_name(const BlockDriverState *bs)
4209 return bdrv_get_parent_name(bs) ?: "";
4212 /* This can be used to identify nodes that might not have a device
4213 * name associated. Since node and device names live in the same
4214 * namespace, the result is unambiguous. The exception is if both are
4215 * absent, then this returns an empty (non-null) string. */
4216 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4218 return bdrv_get_parent_name(bs) ?: bs->node_name;
4221 int bdrv_get_flags(BlockDriverState *bs)
4223 return bs->open_flags;
4226 int bdrv_has_zero_init_1(BlockDriverState *bs)
4228 return 1;
4231 int bdrv_has_zero_init(BlockDriverState *bs)
4233 if (!bs->drv) {
4234 return 0;
4237 /* If BS is a copy on write image, it is initialized to
4238 the contents of the base image, which may not be zeroes. */
4239 if (bs->backing) {
4240 return 0;
4242 if (bs->drv->bdrv_has_zero_init) {
4243 return bs->drv->bdrv_has_zero_init(bs);
4245 if (bs->file && bs->drv->is_filter) {
4246 return bdrv_has_zero_init(bs->file->bs);
4249 /* safe default */
4250 return 0;
4253 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4255 BlockDriverInfo bdi;
4257 if (bs->backing) {
4258 return false;
4261 if (bdrv_get_info(bs, &bdi) == 0) {
4262 return bdi.unallocated_blocks_are_zero;
4265 return false;
4268 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4270 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4271 return false;
4274 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4277 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4279 if (bs->backing && bs->backing->bs->encrypted)
4280 return bs->backing_file;
4281 else if (bs->encrypted)
4282 return bs->filename;
4283 else
4284 return NULL;
4287 void bdrv_get_backing_filename(BlockDriverState *bs,
4288 char *filename, int filename_size)
4290 pstrcpy(filename, filename_size, bs->backing_file);
4293 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4295 BlockDriver *drv = bs->drv;
4296 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4297 if (!drv) {
4298 return -ENOMEDIUM;
4300 if (!drv->bdrv_get_info) {
4301 if (bs->file && drv->is_filter) {
4302 return bdrv_get_info(bs->file->bs, bdi);
4304 return -ENOTSUP;
4306 memset(bdi, 0, sizeof(*bdi));
4307 return drv->bdrv_get_info(bs, bdi);
4310 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4312 BlockDriver *drv = bs->drv;
4313 if (drv && drv->bdrv_get_specific_info) {
4314 return drv->bdrv_get_specific_info(bs);
4316 return NULL;
4319 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4321 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4322 return;
4325 bs->drv->bdrv_debug_event(bs, event);
4328 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4329 const char *tag)
4331 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4332 bs = bs->file ? bs->file->bs : NULL;
4335 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4336 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4339 return -ENOTSUP;
4342 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4344 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4345 bs = bs->file ? bs->file->bs : NULL;
4348 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4349 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4352 return -ENOTSUP;
4355 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4357 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4358 bs = bs->file ? bs->file->bs : NULL;
4361 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4362 return bs->drv->bdrv_debug_resume(bs, tag);
4365 return -ENOTSUP;
4368 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4370 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4371 bs = bs->file ? bs->file->bs : NULL;
4374 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4375 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4378 return false;
4381 /* backing_file can either be relative, or absolute, or a protocol. If it is
4382 * relative, it must be relative to the chain. So, passing in bs->filename
4383 * from a BDS as backing_file should not be done, as that may be relative to
4384 * the CWD rather than the chain. */
4385 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4386 const char *backing_file)
4388 char *filename_full = NULL;
4389 char *backing_file_full = NULL;
4390 char *filename_tmp = NULL;
4391 int is_protocol = 0;
4392 BlockDriverState *curr_bs = NULL;
4393 BlockDriverState *retval = NULL;
4394 Error *local_error = NULL;
4396 if (!bs || !bs->drv || !backing_file) {
4397 return NULL;
4400 filename_full = g_malloc(PATH_MAX);
4401 backing_file_full = g_malloc(PATH_MAX);
4402 filename_tmp = g_malloc(PATH_MAX);
4404 is_protocol = path_has_protocol(backing_file);
4406 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4408 /* If either of the filename paths is actually a protocol, then
4409 * compare unmodified paths; otherwise make paths relative */
4410 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4411 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4412 retval = curr_bs->backing->bs;
4413 break;
4415 /* Also check against the full backing filename for the image */
4416 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4417 &local_error);
4418 if (local_error == NULL) {
4419 if (strcmp(backing_file, backing_file_full) == 0) {
4420 retval = curr_bs->backing->bs;
4421 break;
4423 } else {
4424 error_free(local_error);
4425 local_error = NULL;
4427 } else {
4428 /* If not an absolute filename path, make it relative to the current
4429 * image's filename path */
4430 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4431 backing_file);
4433 /* We are going to compare absolute pathnames */
4434 if (!realpath(filename_tmp, filename_full)) {
4435 continue;
4438 /* We need to make sure the backing filename we are comparing against
4439 * is relative to the current image filename (or absolute) */
4440 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4441 curr_bs->backing_file);
4443 if (!realpath(filename_tmp, backing_file_full)) {
4444 continue;
4447 if (strcmp(backing_file_full, filename_full) == 0) {
4448 retval = curr_bs->backing->bs;
4449 break;
4454 g_free(filename_full);
4455 g_free(backing_file_full);
4456 g_free(filename_tmp);
4457 return retval;
4460 void bdrv_init(void)
4462 module_call_init(MODULE_INIT_BLOCK);
4465 void bdrv_init_with_whitelist(void)
4467 use_bdrv_whitelist = 1;
4468 bdrv_init();
4471 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4472 Error **errp)
4474 BdrvChild *child, *parent;
4475 uint64_t perm, shared_perm;
4476 Error *local_err = NULL;
4477 int ret;
4478 BdrvDirtyBitmap *bm;
4480 if (!bs->drv) {
4481 return;
4484 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4485 return;
4488 QLIST_FOREACH(child, &bs->children, next) {
4489 bdrv_co_invalidate_cache(child->bs, &local_err);
4490 if (local_err) {
4491 error_propagate(errp, local_err);
4492 return;
4497 * Update permissions, they may differ for inactive nodes.
4499 * Note that the required permissions of inactive images are always a
4500 * subset of the permissions required after activating the image. This
4501 * allows us to just get the permissions upfront without restricting
4502 * drv->bdrv_invalidate_cache().
4504 * It also means that in error cases, we don't have to try and revert to
4505 * the old permissions (which is an operation that could fail, too). We can
4506 * just keep the extended permissions for the next time that an activation
4507 * of the image is tried.
4509 bs->open_flags &= ~BDRV_O_INACTIVE;
4510 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4511 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4512 if (ret < 0) {
4513 bs->open_flags |= BDRV_O_INACTIVE;
4514 error_propagate(errp, local_err);
4515 return;
4517 bdrv_set_perm(bs, perm, shared_perm);
4519 if (bs->drv->bdrv_co_invalidate_cache) {
4520 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4521 if (local_err) {
4522 bs->open_flags |= BDRV_O_INACTIVE;
4523 error_propagate(errp, local_err);
4524 return;
4528 for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm;
4529 bm = bdrv_dirty_bitmap_next(bs, bm))
4531 bdrv_dirty_bitmap_set_migration(bm, false);
4534 ret = refresh_total_sectors(bs, bs->total_sectors);
4535 if (ret < 0) {
4536 bs->open_flags |= BDRV_O_INACTIVE;
4537 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4538 return;
4541 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4542 if (parent->role->activate) {
4543 parent->role->activate(parent, &local_err);
4544 if (local_err) {
4545 error_propagate(errp, local_err);
4546 return;
4552 typedef struct InvalidateCacheCo {
4553 BlockDriverState *bs;
4554 Error **errp;
4555 bool done;
4556 } InvalidateCacheCo;
4558 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4560 InvalidateCacheCo *ico = opaque;
4561 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4562 ico->done = true;
4565 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4567 Coroutine *co;
4568 InvalidateCacheCo ico = {
4569 .bs = bs,
4570 .done = false,
4571 .errp = errp
4574 if (qemu_in_coroutine()) {
4575 /* Fast-path if already in coroutine context */
4576 bdrv_invalidate_cache_co_entry(&ico);
4577 } else {
4578 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4579 qemu_coroutine_enter(co);
4580 BDRV_POLL_WHILE(bs, !ico.done);
4584 void bdrv_invalidate_cache_all(Error **errp)
4586 BlockDriverState *bs;
4587 Error *local_err = NULL;
4588 BdrvNextIterator it;
4590 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4591 AioContext *aio_context = bdrv_get_aio_context(bs);
4593 aio_context_acquire(aio_context);
4594 bdrv_invalidate_cache(bs, &local_err);
4595 aio_context_release(aio_context);
4596 if (local_err) {
4597 error_propagate(errp, local_err);
4598 bdrv_next_cleanup(&it);
4599 return;
4604 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
4606 BdrvChild *parent;
4608 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4609 if (parent->role->parent_is_bds) {
4610 BlockDriverState *parent_bs = parent->opaque;
4611 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
4612 return true;
4617 return false;
4620 static int bdrv_inactivate_recurse(BlockDriverState *bs)
4622 BdrvChild *child, *parent;
4623 uint64_t perm, shared_perm;
4624 int ret;
4626 if (!bs->drv) {
4627 return -ENOMEDIUM;
4630 /* Make sure that we don't inactivate a child before its parent.
4631 * It will be covered by recursion from the yet active parent. */
4632 if (bdrv_has_bds_parent(bs, true)) {
4633 return 0;
4636 assert(!(bs->open_flags & BDRV_O_INACTIVE));
4638 /* Inactivate this node */
4639 if (bs->drv->bdrv_inactivate) {
4640 ret = bs->drv->bdrv_inactivate(bs);
4641 if (ret < 0) {
4642 return ret;
4646 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4647 if (parent->role->inactivate) {
4648 ret = parent->role->inactivate(parent);
4649 if (ret < 0) {
4650 return ret;
4655 bs->open_flags |= BDRV_O_INACTIVE;
4657 /* Update permissions, they may differ for inactive nodes */
4658 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4659 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4660 bdrv_set_perm(bs, perm, shared_perm);
4663 /* Recursively inactivate children */
4664 QLIST_FOREACH(child, &bs->children, next) {
4665 ret = bdrv_inactivate_recurse(child->bs);
4666 if (ret < 0) {
4667 return ret;
4671 return 0;
4674 int bdrv_inactivate_all(void)
4676 BlockDriverState *bs = NULL;
4677 BdrvNextIterator it;
4678 int ret = 0;
4679 GSList *aio_ctxs = NULL, *ctx;
4681 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4682 AioContext *aio_context = bdrv_get_aio_context(bs);
4684 if (!g_slist_find(aio_ctxs, aio_context)) {
4685 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4686 aio_context_acquire(aio_context);
4690 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4691 /* Nodes with BDS parents are covered by recursion from the last
4692 * parent that gets inactivated. Don't inactivate them a second
4693 * time if that has already happened. */
4694 if (bdrv_has_bds_parent(bs, false)) {
4695 continue;
4697 ret = bdrv_inactivate_recurse(bs);
4698 if (ret < 0) {
4699 bdrv_next_cleanup(&it);
4700 goto out;
4704 out:
4705 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4706 AioContext *aio_context = ctx->data;
4707 aio_context_release(aio_context);
4709 g_slist_free(aio_ctxs);
4711 return ret;
4714 /**************************************************************/
4715 /* removable device support */
4718 * Return TRUE if the media is present
4720 bool bdrv_is_inserted(BlockDriverState *bs)
4722 BlockDriver *drv = bs->drv;
4723 BdrvChild *child;
4725 if (!drv) {
4726 return false;
4728 if (drv->bdrv_is_inserted) {
4729 return drv->bdrv_is_inserted(bs);
4731 QLIST_FOREACH(child, &bs->children, next) {
4732 if (!bdrv_is_inserted(child->bs)) {
4733 return false;
4736 return true;
4740 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4742 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4744 BlockDriver *drv = bs->drv;
4746 if (drv && drv->bdrv_eject) {
4747 drv->bdrv_eject(bs, eject_flag);
4752 * Lock or unlock the media (if it is locked, the user won't be able
4753 * to eject it manually).
4755 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4757 BlockDriver *drv = bs->drv;
4759 trace_bdrv_lock_medium(bs, locked);
4761 if (drv && drv->bdrv_lock_medium) {
4762 drv->bdrv_lock_medium(bs, locked);
4766 /* Get a reference to bs */
4767 void bdrv_ref(BlockDriverState *bs)
4769 bs->refcnt++;
4772 /* Release a previously grabbed reference to bs.
4773 * If after releasing, reference count is zero, the BlockDriverState is
4774 * deleted. */
4775 void bdrv_unref(BlockDriverState *bs)
4777 if (!bs) {
4778 return;
4780 assert(bs->refcnt > 0);
4781 if (--bs->refcnt == 0) {
4782 bdrv_delete(bs);
4786 struct BdrvOpBlocker {
4787 Error *reason;
4788 QLIST_ENTRY(BdrvOpBlocker) list;
4791 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4793 BdrvOpBlocker *blocker;
4794 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4795 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4796 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4797 error_propagate_prepend(errp, error_copy(blocker->reason),
4798 "Node '%s' is busy: ",
4799 bdrv_get_device_or_node_name(bs));
4800 return true;
4802 return false;
4805 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4807 BdrvOpBlocker *blocker;
4808 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4810 blocker = g_new0(BdrvOpBlocker, 1);
4811 blocker->reason = reason;
4812 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4815 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4817 BdrvOpBlocker *blocker, *next;
4818 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4819 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4820 if (blocker->reason == reason) {
4821 QLIST_REMOVE(blocker, list);
4822 g_free(blocker);
4827 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4829 int i;
4830 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4831 bdrv_op_block(bs, i, reason);
4835 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4837 int i;
4838 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4839 bdrv_op_unblock(bs, i, reason);
4843 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4845 int i;
4847 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4848 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4849 return false;
4852 return true;
4855 void bdrv_img_create(const char *filename, const char *fmt,
4856 const char *base_filename, const char *base_fmt,
4857 char *options, uint64_t img_size, int flags, bool quiet,
4858 Error **errp)
4860 QemuOptsList *create_opts = NULL;
4861 QemuOpts *opts = NULL;
4862 const char *backing_fmt, *backing_file;
4863 int64_t size;
4864 BlockDriver *drv, *proto_drv;
4865 Error *local_err = NULL;
4866 int ret = 0;
4868 /* Find driver and parse its options */
4869 drv = bdrv_find_format(fmt);
4870 if (!drv) {
4871 error_setg(errp, "Unknown file format '%s'", fmt);
4872 return;
4875 proto_drv = bdrv_find_protocol(filename, true, errp);
4876 if (!proto_drv) {
4877 return;
4880 if (!drv->create_opts) {
4881 error_setg(errp, "Format driver '%s' does not support image creation",
4882 drv->format_name);
4883 return;
4886 if (!proto_drv->create_opts) {
4887 error_setg(errp, "Protocol driver '%s' does not support image creation",
4888 proto_drv->format_name);
4889 return;
4892 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4893 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4895 /* Create parameter list with default values */
4896 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4897 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4899 /* Parse -o options */
4900 if (options) {
4901 qemu_opts_do_parse(opts, options, NULL, &local_err);
4902 if (local_err) {
4903 goto out;
4907 if (base_filename) {
4908 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4909 if (local_err) {
4910 error_setg(errp, "Backing file not supported for file format '%s'",
4911 fmt);
4912 goto out;
4916 if (base_fmt) {
4917 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4918 if (local_err) {
4919 error_setg(errp, "Backing file format not supported for file "
4920 "format '%s'", fmt);
4921 goto out;
4925 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4926 if (backing_file) {
4927 if (!strcmp(filename, backing_file)) {
4928 error_setg(errp, "Error: Trying to create an image with the "
4929 "same filename as the backing file");
4930 goto out;
4934 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4936 /* The size for the image must always be specified, unless we have a backing
4937 * file and we have not been forbidden from opening it. */
4938 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4939 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4940 BlockDriverState *bs;
4941 char *full_backing = g_new0(char, PATH_MAX);
4942 int back_flags;
4943 QDict *backing_options = NULL;
4945 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4946 full_backing, PATH_MAX,
4947 &local_err);
4948 if (local_err) {
4949 g_free(full_backing);
4950 goto out;
4953 /* backing files always opened read-only */
4954 back_flags = flags;
4955 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4957 backing_options = qdict_new();
4958 if (backing_fmt) {
4959 qdict_put_str(backing_options, "driver", backing_fmt);
4961 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
4963 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4964 &local_err);
4965 g_free(full_backing);
4966 if (!bs && size != -1) {
4967 /* Couldn't open BS, but we have a size, so it's nonfatal */
4968 warn_reportf_err(local_err,
4969 "Could not verify backing image. "
4970 "This may become an error in future versions.\n");
4971 local_err = NULL;
4972 } else if (!bs) {
4973 /* Couldn't open bs, do not have size */
4974 error_append_hint(&local_err,
4975 "Could not open backing image to determine size.\n");
4976 goto out;
4977 } else {
4978 if (size == -1) {
4979 /* Opened BS, have no size */
4980 size = bdrv_getlength(bs);
4981 if (size < 0) {
4982 error_setg_errno(errp, -size, "Could not get size of '%s'",
4983 backing_file);
4984 bdrv_unref(bs);
4985 goto out;
4987 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4989 bdrv_unref(bs);
4991 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4993 if (size == -1) {
4994 error_setg(errp, "Image creation needs a size parameter");
4995 goto out;
4998 if (!quiet) {
4999 printf("Formatting '%s', fmt=%s ", filename, fmt);
5000 qemu_opts_print(opts, " ");
5001 puts("");
5004 ret = bdrv_create(drv, filename, opts, &local_err);
5006 if (ret == -EFBIG) {
5007 /* This is generally a better message than whatever the driver would
5008 * deliver (especially because of the cluster_size_hint), since that
5009 * is most probably not much different from "image too large". */
5010 const char *cluster_size_hint = "";
5011 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5012 cluster_size_hint = " (try using a larger cluster size)";
5014 error_setg(errp, "The image size is too large for file format '%s'"
5015 "%s", fmt, cluster_size_hint);
5016 error_free(local_err);
5017 local_err = NULL;
5020 out:
5021 qemu_opts_del(opts);
5022 qemu_opts_free(create_opts);
5023 error_propagate(errp, local_err);
5026 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5028 return bs ? bs->aio_context : qemu_get_aio_context();
5031 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
5033 aio_co_enter(bdrv_get_aio_context(bs), co);
5036 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
5038 QLIST_REMOVE(ban, list);
5039 g_free(ban);
5042 void bdrv_detach_aio_context(BlockDriverState *bs)
5044 BdrvAioNotifier *baf, *baf_tmp;
5045 BdrvChild *child;
5047 if (!bs->drv) {
5048 return;
5051 assert(!bs->walking_aio_notifiers);
5052 bs->walking_aio_notifiers = true;
5053 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
5054 if (baf->deleted) {
5055 bdrv_do_remove_aio_context_notifier(baf);
5056 } else {
5057 baf->detach_aio_context(baf->opaque);
5060 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5061 * remove remaining aio notifiers if we aren't called again.
5063 bs->walking_aio_notifiers = false;
5065 if (bs->drv->bdrv_detach_aio_context) {
5066 bs->drv->bdrv_detach_aio_context(bs);
5068 QLIST_FOREACH(child, &bs->children, next) {
5069 bdrv_detach_aio_context(child->bs);
5072 bs->aio_context = NULL;
5075 void bdrv_attach_aio_context(BlockDriverState *bs,
5076 AioContext *new_context)
5078 BdrvAioNotifier *ban, *ban_tmp;
5079 BdrvChild *child;
5081 if (!bs->drv) {
5082 return;
5085 bs->aio_context = new_context;
5087 QLIST_FOREACH(child, &bs->children, next) {
5088 bdrv_attach_aio_context(child->bs, new_context);
5090 if (bs->drv->bdrv_attach_aio_context) {
5091 bs->drv->bdrv_attach_aio_context(bs, new_context);
5094 assert(!bs->walking_aio_notifiers);
5095 bs->walking_aio_notifiers = true;
5096 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
5097 if (ban->deleted) {
5098 bdrv_do_remove_aio_context_notifier(ban);
5099 } else {
5100 ban->attached_aio_context(new_context, ban->opaque);
5103 bs->walking_aio_notifiers = false;
5106 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5108 AioContext *ctx = bdrv_get_aio_context(bs);
5110 aio_disable_external(ctx);
5111 bdrv_parent_drained_begin(bs, NULL, false);
5112 bdrv_drain(bs); /* ensure there are no in-flight requests */
5114 while (aio_poll(ctx, false)) {
5115 /* wait for all bottom halves to execute */
5118 bdrv_detach_aio_context(bs);
5120 /* This function executes in the old AioContext so acquire the new one in
5121 * case it runs in a different thread.
5123 aio_context_acquire(new_context);
5124 bdrv_attach_aio_context(bs, new_context);
5125 bdrv_parent_drained_end(bs, NULL, false);
5126 aio_enable_external(ctx);
5127 aio_context_release(new_context);
5130 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
5131 void (*attached_aio_context)(AioContext *new_context, void *opaque),
5132 void (*detach_aio_context)(void *opaque), void *opaque)
5134 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
5135 *ban = (BdrvAioNotifier){
5136 .attached_aio_context = attached_aio_context,
5137 .detach_aio_context = detach_aio_context,
5138 .opaque = opaque
5141 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
5144 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
5145 void (*attached_aio_context)(AioContext *,
5146 void *),
5147 void (*detach_aio_context)(void *),
5148 void *opaque)
5150 BdrvAioNotifier *ban, *ban_next;
5152 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5153 if (ban->attached_aio_context == attached_aio_context &&
5154 ban->detach_aio_context == detach_aio_context &&
5155 ban->opaque == opaque &&
5156 ban->deleted == false)
5158 if (bs->walking_aio_notifiers) {
5159 ban->deleted = true;
5160 } else {
5161 bdrv_do_remove_aio_context_notifier(ban);
5163 return;
5167 abort();
5170 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
5171 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5172 Error **errp)
5174 if (!bs->drv) {
5175 error_setg(errp, "Node is ejected");
5176 return -ENOMEDIUM;
5178 if (!bs->drv->bdrv_amend_options) {
5179 error_setg(errp, "Block driver '%s' does not support option amendment",
5180 bs->drv->format_name);
5181 return -ENOTSUP;
5183 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
5186 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5187 * of block filter and by bdrv_is_first_non_filter.
5188 * It is used to test if the given bs is the candidate or recurse more in the
5189 * node graph.
5191 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5192 BlockDriverState *candidate)
5194 /* return false if basic checks fails */
5195 if (!bs || !bs->drv) {
5196 return false;
5199 /* the code reached a non block filter driver -> check if the bs is
5200 * the same as the candidate. It's the recursion termination condition.
5202 if (!bs->drv->is_filter) {
5203 return bs == candidate;
5205 /* Down this path the driver is a block filter driver */
5207 /* If the block filter recursion method is defined use it to recurse down
5208 * the node graph.
5210 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5211 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5214 /* the driver is a block filter but don't allow to recurse -> return false
5216 return false;
5219 /* This function checks if the candidate is the first non filter bs down it's
5220 * bs chain. Since we don't have pointers to parents it explore all bs chains
5221 * from the top. Some filters can choose not to pass down the recursion.
5223 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5225 BlockDriverState *bs;
5226 BdrvNextIterator it;
5228 /* walk down the bs forest recursively */
5229 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5230 bool perm;
5232 /* try to recurse in this top level bs */
5233 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5235 /* candidate is the first non filter */
5236 if (perm) {
5237 bdrv_next_cleanup(&it);
5238 return true;
5242 return false;
5245 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5246 const char *node_name, Error **errp)
5248 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5249 AioContext *aio_context;
5251 if (!to_replace_bs) {
5252 error_setg(errp, "Node name '%s' not found", node_name);
5253 return NULL;
5256 aio_context = bdrv_get_aio_context(to_replace_bs);
5257 aio_context_acquire(aio_context);
5259 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5260 to_replace_bs = NULL;
5261 goto out;
5264 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5265 * most non filter in order to prevent data corruption.
5266 * Another benefit is that this tests exclude backing files which are
5267 * blocked by the backing blockers.
5269 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5270 error_setg(errp, "Only top most non filter can be replaced");
5271 to_replace_bs = NULL;
5272 goto out;
5275 out:
5276 aio_context_release(aio_context);
5277 return to_replace_bs;
5280 static bool append_open_options(QDict *d, BlockDriverState *bs)
5282 const QDictEntry *entry;
5283 QemuOptDesc *desc;
5284 bool found_any = false;
5286 for (entry = qdict_first(bs->options); entry;
5287 entry = qdict_next(bs->options, entry))
5289 /* Exclude all non-driver-specific options */
5290 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
5291 if (!strcmp(qdict_entry_key(entry), desc->name)) {
5292 break;
5295 if (desc->name) {
5296 continue;
5299 qdict_put_obj(d, qdict_entry_key(entry),
5300 qobject_ref(qdict_entry_value(entry)));
5301 found_any = true;
5304 return found_any;
5307 /* Updates the following BDS fields:
5308 * - exact_filename: A filename which may be used for opening a block device
5309 * which (mostly) equals the given BDS (even without any
5310 * other options; so reading and writing must return the same
5311 * results, but caching etc. may be different)
5312 * - full_open_options: Options which, when given when opening a block device
5313 * (without a filename), result in a BDS (mostly)
5314 * equalling the given one
5315 * - filename: If exact_filename is set, it is copied here. Otherwise,
5316 * full_open_options is converted to a JSON object, prefixed with
5317 * "json:" (for use through the JSON pseudo protocol) and put here.
5319 void bdrv_refresh_filename(BlockDriverState *bs)
5321 BlockDriver *drv = bs->drv;
5322 QDict *opts;
5324 if (!drv) {
5325 return;
5328 /* This BDS's file name will most probably depend on its file's name, so
5329 * refresh that first */
5330 if (bs->file) {
5331 bdrv_refresh_filename(bs->file->bs);
5334 if (drv->bdrv_refresh_filename) {
5335 /* Obsolete information is of no use here, so drop the old file name
5336 * information before refreshing it */
5337 bs->exact_filename[0] = '\0';
5338 if (bs->full_open_options) {
5339 qobject_unref(bs->full_open_options);
5340 bs->full_open_options = NULL;
5343 opts = qdict_new();
5344 append_open_options(opts, bs);
5345 drv->bdrv_refresh_filename(bs, opts);
5346 qobject_unref(opts);
5347 } else if (bs->file) {
5348 /* Try to reconstruct valid information from the underlying file */
5349 bool has_open_options;
5351 bs->exact_filename[0] = '\0';
5352 if (bs->full_open_options) {
5353 qobject_unref(bs->full_open_options);
5354 bs->full_open_options = NULL;
5357 opts = qdict_new();
5358 has_open_options = append_open_options(opts, bs);
5360 /* If no specific options have been given for this BDS, the filename of
5361 * the underlying file should suffice for this one as well */
5362 if (bs->file->bs->exact_filename[0] && !has_open_options) {
5363 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5365 /* Reconstructing the full options QDict is simple for most format block
5366 * drivers, as long as the full options are known for the underlying
5367 * file BDS. The full options QDict of that file BDS should somehow
5368 * contain a representation of the filename, therefore the following
5369 * suffices without querying the (exact_)filename of this BDS. */
5370 if (bs->file->bs->full_open_options) {
5371 qdict_put_str(opts, "driver", drv->format_name);
5372 qdict_put(opts, "file",
5373 qobject_ref(bs->file->bs->full_open_options));
5375 bs->full_open_options = opts;
5376 } else {
5377 qobject_unref(opts);
5379 } else if (!bs->full_open_options && qdict_size(bs->options)) {
5380 /* There is no underlying file BDS (at least referenced by BDS.file),
5381 * so the full options QDict should be equal to the options given
5382 * specifically for this block device when it was opened (plus the
5383 * driver specification).
5384 * Because those options don't change, there is no need to update
5385 * full_open_options when it's already set. */
5387 opts = qdict_new();
5388 append_open_options(opts, bs);
5389 qdict_put_str(opts, "driver", drv->format_name);
5391 if (bs->exact_filename[0]) {
5392 /* This may not work for all block protocol drivers (some may
5393 * require this filename to be parsed), but we have to find some
5394 * default solution here, so just include it. If some block driver
5395 * does not support pure options without any filename at all or
5396 * needs some special format of the options QDict, it needs to
5397 * implement the driver-specific bdrv_refresh_filename() function.
5399 qdict_put_str(opts, "filename", bs->exact_filename);
5402 bs->full_open_options = opts;
5405 if (bs->exact_filename[0]) {
5406 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5407 } else if (bs->full_open_options) {
5408 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5409 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5410 qstring_get_str(json));
5411 qobject_unref(json);
5416 * Hot add/remove a BDS's child. So the user can take a child offline when
5417 * it is broken and take a new child online
5419 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5420 Error **errp)
5423 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5424 error_setg(errp, "The node %s does not support adding a child",
5425 bdrv_get_device_or_node_name(parent_bs));
5426 return;
5429 if (!QLIST_EMPTY(&child_bs->parents)) {
5430 error_setg(errp, "The node %s already has a parent",
5431 child_bs->node_name);
5432 return;
5435 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5438 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5440 BdrvChild *tmp;
5442 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5443 error_setg(errp, "The node %s does not support removing a child",
5444 bdrv_get_device_or_node_name(parent_bs));
5445 return;
5448 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5449 if (tmp == child) {
5450 break;
5454 if (!tmp) {
5455 error_setg(errp, "The node %s does not have a child named %s",
5456 bdrv_get_device_or_node_name(parent_bs),
5457 bdrv_get_device_or_node_name(child->bs));
5458 return;
5461 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5464 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5465 uint32_t granularity, Error **errp)
5467 BlockDriver *drv = bs->drv;
5469 if (!drv) {
5470 error_setg_errno(errp, ENOMEDIUM,
5471 "Can't store persistent bitmaps to %s",
5472 bdrv_get_device_or_node_name(bs));
5473 return false;
5476 if (!drv->bdrv_can_store_new_dirty_bitmap) {
5477 error_setg_errno(errp, ENOTSUP,
5478 "Can't store persistent bitmaps to %s",
5479 bdrv_get_device_or_node_name(bs));
5480 return false;
5483 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);