PPC: E500: Update u-boot to v2019.01
[qemu.git] / block.c
blobccf008c177c87549bfe9b71fa585d003559f924c
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 return its duplicate. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
157 supported. */
158 char *path_combine(const char *base_path, const char *filename)
160 const char *protocol_stripped = NULL;
161 const char *p, *p1;
162 char *result;
163 int len;
165 if (path_is_absolute(filename)) {
166 return g_strdup(filename);
169 if (path_has_protocol(base_path)) {
170 protocol_stripped = strchr(base_path, ':');
171 if (protocol_stripped) {
172 protocol_stripped++;
175 p = protocol_stripped ?: base_path;
177 p1 = strrchr(base_path, '/');
178 #ifdef _WIN32
180 const char *p2;
181 p2 = strrchr(base_path, '\\');
182 if (!p1 || p2 > p1) {
183 p1 = p2;
186 #endif
187 if (p1) {
188 p1++;
189 } else {
190 p1 = base_path;
192 if (p1 > p) {
193 p = p1;
195 len = p - base_path;
197 result = g_malloc(len + strlen(filename) + 1);
198 memcpy(result, base_path, len);
199 strcpy(result + len, filename);
201 return result;
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;
307 * If @backing is empty, this function returns NULL without setting
308 * @errp. In all other cases, NULL will only be returned with @errp
309 * set.
311 * Therefore, a return value of NULL without @errp set means that
312 * there is no backing file; if @errp is set, there is one but its
313 * absolute filename cannot be generated.
315 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
316 const char *backing,
317 Error **errp)
319 if (backing[0] == '\0') {
320 return NULL;
321 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
322 return g_strdup(backing);
323 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
324 error_setg(errp, "Cannot use relative backing file names for '%s'",
325 backed);
326 return NULL;
327 } else {
328 return path_combine(backed, backing);
333 * If @filename is empty or NULL, this function returns NULL without
334 * setting @errp. In all other cases, NULL will only be returned with
335 * @errp set.
337 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
338 const char *filename, Error **errp)
340 char *dir, *full_name;
342 if (!filename || filename[0] == '\0') {
343 return NULL;
344 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
345 return g_strdup(filename);
348 dir = bdrv_dirname(relative_to, errp);
349 if (!dir) {
350 return NULL;
353 full_name = g_strconcat(dir, filename, NULL);
354 g_free(dir);
355 return full_name;
358 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
360 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
363 void bdrv_register(BlockDriver *bdrv)
365 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
368 BlockDriverState *bdrv_new(void)
370 BlockDriverState *bs;
371 int i;
373 bs = g_new0(BlockDriverState, 1);
374 QLIST_INIT(&bs->dirty_bitmaps);
375 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
376 QLIST_INIT(&bs->op_blockers[i]);
378 notifier_with_return_list_init(&bs->before_write_notifiers);
379 qemu_co_mutex_init(&bs->reqs_lock);
380 qemu_mutex_init(&bs->dirty_bitmap_mutex);
381 bs->refcnt = 1;
382 bs->aio_context = qemu_get_aio_context();
384 qemu_co_queue_init(&bs->flush_queue);
386 for (i = 0; i < bdrv_drain_all_count; i++) {
387 bdrv_drained_begin(bs);
390 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
392 return bs;
395 static BlockDriver *bdrv_do_find_format(const char *format_name)
397 BlockDriver *drv1;
399 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
400 if (!strcmp(drv1->format_name, format_name)) {
401 return drv1;
405 return NULL;
408 BlockDriver *bdrv_find_format(const char *format_name)
410 BlockDriver *drv1;
411 int i;
413 drv1 = bdrv_do_find_format(format_name);
414 if (drv1) {
415 return drv1;
418 /* The driver isn't registered, maybe we need to load a module */
419 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
420 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
421 block_module_load_one(block_driver_modules[i].library_name);
422 break;
426 return bdrv_do_find_format(format_name);
429 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
431 static const char *whitelist_rw[] = {
432 CONFIG_BDRV_RW_WHITELIST
434 static const char *whitelist_ro[] = {
435 CONFIG_BDRV_RO_WHITELIST
437 const char **p;
439 if (!whitelist_rw[0] && !whitelist_ro[0]) {
440 return 1; /* no whitelist, anything goes */
443 for (p = whitelist_rw; *p; p++) {
444 if (!strcmp(format_name, *p)) {
445 return 1;
448 if (read_only) {
449 for (p = whitelist_ro; *p; p++) {
450 if (!strcmp(format_name, *p)) {
451 return 1;
455 return 0;
458 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
460 return bdrv_format_is_whitelisted(drv->format_name, read_only);
463 bool bdrv_uses_whitelist(void)
465 return use_bdrv_whitelist;
468 typedef struct CreateCo {
469 BlockDriver *drv;
470 char *filename;
471 QemuOpts *opts;
472 int ret;
473 Error *err;
474 } CreateCo;
476 static void coroutine_fn bdrv_create_co_entry(void *opaque)
478 Error *local_err = NULL;
479 int ret;
481 CreateCo *cco = opaque;
482 assert(cco->drv);
484 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
485 error_propagate(&cco->err, local_err);
486 cco->ret = ret;
489 int bdrv_create(BlockDriver *drv, const char* filename,
490 QemuOpts *opts, Error **errp)
492 int ret;
494 Coroutine *co;
495 CreateCo cco = {
496 .drv = drv,
497 .filename = g_strdup(filename),
498 .opts = opts,
499 .ret = NOT_DONE,
500 .err = NULL,
503 if (!drv->bdrv_co_create_opts) {
504 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
505 ret = -ENOTSUP;
506 goto out;
509 if (qemu_in_coroutine()) {
510 /* Fast-path if already in coroutine context */
511 bdrv_create_co_entry(&cco);
512 } else {
513 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
514 qemu_coroutine_enter(co);
515 while (cco.ret == NOT_DONE) {
516 aio_poll(qemu_get_aio_context(), true);
520 ret = cco.ret;
521 if (ret < 0) {
522 if (cco.err) {
523 error_propagate(errp, cco.err);
524 } else {
525 error_setg_errno(errp, -ret, "Could not create image");
529 out:
530 g_free(cco.filename);
531 return ret;
534 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
536 BlockDriver *drv;
537 Error *local_err = NULL;
538 int ret;
540 drv = bdrv_find_protocol(filename, true, errp);
541 if (drv == NULL) {
542 return -ENOENT;
545 ret = bdrv_create(drv, filename, opts, &local_err);
546 error_propagate(errp, local_err);
547 return ret;
551 * Try to get @bs's logical and physical block size.
552 * On success, store them in @bsz struct and return 0.
553 * On failure return -errno.
554 * @bs must not be empty.
556 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
558 BlockDriver *drv = bs->drv;
560 if (drv && drv->bdrv_probe_blocksizes) {
561 return drv->bdrv_probe_blocksizes(bs, bsz);
562 } else if (drv && drv->is_filter && bs->file) {
563 return bdrv_probe_blocksizes(bs->file->bs, bsz);
566 return -ENOTSUP;
570 * Try to get @bs's geometry (cyls, heads, sectors).
571 * On success, store them in @geo struct and return 0.
572 * On failure return -errno.
573 * @bs must not be empty.
575 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
577 BlockDriver *drv = bs->drv;
579 if (drv && drv->bdrv_probe_geometry) {
580 return drv->bdrv_probe_geometry(bs, geo);
581 } else if (drv && drv->is_filter && bs->file) {
582 return bdrv_probe_geometry(bs->file->bs, geo);
585 return -ENOTSUP;
589 * Create a uniquely-named empty temporary file.
590 * Return 0 upon success, otherwise a negative errno value.
592 int get_tmp_filename(char *filename, int size)
594 #ifdef _WIN32
595 char temp_dir[MAX_PATH];
596 /* GetTempFileName requires that its output buffer (4th param)
597 have length MAX_PATH or greater. */
598 assert(size >= MAX_PATH);
599 return (GetTempPath(MAX_PATH, temp_dir)
600 && GetTempFileName(temp_dir, "qem", 0, filename)
601 ? 0 : -GetLastError());
602 #else
603 int fd;
604 const char *tmpdir;
605 tmpdir = getenv("TMPDIR");
606 if (!tmpdir) {
607 tmpdir = "/var/tmp";
609 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
610 return -EOVERFLOW;
612 fd = mkstemp(filename);
613 if (fd < 0) {
614 return -errno;
616 if (close(fd) != 0) {
617 unlink(filename);
618 return -errno;
620 return 0;
621 #endif
625 * Detect host devices. By convention, /dev/cdrom[N] is always
626 * recognized as a host CDROM.
628 static BlockDriver *find_hdev_driver(const char *filename)
630 int score_max = 0, score;
631 BlockDriver *drv = NULL, *d;
633 QLIST_FOREACH(d, &bdrv_drivers, list) {
634 if (d->bdrv_probe_device) {
635 score = d->bdrv_probe_device(filename);
636 if (score > score_max) {
637 score_max = score;
638 drv = d;
643 return drv;
646 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
648 BlockDriver *drv1;
650 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
651 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
652 return drv1;
656 return NULL;
659 BlockDriver *bdrv_find_protocol(const char *filename,
660 bool allow_protocol_prefix,
661 Error **errp)
663 BlockDriver *drv1;
664 char protocol[128];
665 int len;
666 const char *p;
667 int i;
669 /* TODO Drivers without bdrv_file_open must be specified explicitly */
672 * XXX(hch): we really should not let host device detection
673 * override an explicit protocol specification, but moving this
674 * later breaks access to device names with colons in them.
675 * Thanks to the brain-dead persistent naming schemes on udev-
676 * based Linux systems those actually are quite common.
678 drv1 = find_hdev_driver(filename);
679 if (drv1) {
680 return drv1;
683 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
684 return &bdrv_file;
687 p = strchr(filename, ':');
688 assert(p != NULL);
689 len = p - filename;
690 if (len > sizeof(protocol) - 1)
691 len = sizeof(protocol) - 1;
692 memcpy(protocol, filename, len);
693 protocol[len] = '\0';
695 drv1 = bdrv_do_find_protocol(protocol);
696 if (drv1) {
697 return drv1;
700 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
701 if (block_driver_modules[i].protocol_name &&
702 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
703 block_module_load_one(block_driver_modules[i].library_name);
704 break;
708 drv1 = bdrv_do_find_protocol(protocol);
709 if (!drv1) {
710 error_setg(errp, "Unknown protocol '%s'", protocol);
712 return drv1;
716 * Guess image format by probing its contents.
717 * This is not a good idea when your image is raw (CVE-2008-2004), but
718 * we do it anyway for backward compatibility.
720 * @buf contains the image's first @buf_size bytes.
721 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
722 * but can be smaller if the image file is smaller)
723 * @filename is its filename.
725 * For all block drivers, call the bdrv_probe() method to get its
726 * probing score.
727 * Return the first block driver with the highest probing score.
729 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
730 const char *filename)
732 int score_max = 0, score;
733 BlockDriver *drv = NULL, *d;
735 QLIST_FOREACH(d, &bdrv_drivers, list) {
736 if (d->bdrv_probe) {
737 score = d->bdrv_probe(buf, buf_size, filename);
738 if (score > score_max) {
739 score_max = score;
740 drv = d;
745 return drv;
748 static int find_image_format(BlockBackend *file, const char *filename,
749 BlockDriver **pdrv, Error **errp)
751 BlockDriver *drv;
752 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
753 int ret = 0;
755 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
756 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
757 *pdrv = &bdrv_raw;
758 return ret;
761 ret = blk_pread(file, 0, buf, sizeof(buf));
762 if (ret < 0) {
763 error_setg_errno(errp, -ret, "Could not read image for determining its "
764 "format");
765 *pdrv = NULL;
766 return ret;
769 drv = bdrv_probe_all(buf, ret, filename);
770 if (!drv) {
771 error_setg(errp, "Could not determine image format: No compatible "
772 "driver found");
773 ret = -ENOENT;
775 *pdrv = drv;
776 return ret;
780 * Set the current 'total_sectors' value
781 * Return 0 on success, -errno on error.
783 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
785 BlockDriver *drv = bs->drv;
787 if (!drv) {
788 return -ENOMEDIUM;
791 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
792 if (bdrv_is_sg(bs))
793 return 0;
795 /* query actual device if possible, otherwise just trust the hint */
796 if (drv->bdrv_getlength) {
797 int64_t length = drv->bdrv_getlength(bs);
798 if (length < 0) {
799 return length;
801 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
804 bs->total_sectors = hint;
805 return 0;
809 * Combines a QDict of new block driver @options with any missing options taken
810 * from @old_options, so that leaving out an option defaults to its old value.
812 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
813 QDict *old_options)
815 if (bs->drv && bs->drv->bdrv_join_options) {
816 bs->drv->bdrv_join_options(options, old_options);
817 } else {
818 qdict_join(options, old_options, false);
822 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
823 int open_flags,
824 Error **errp)
826 Error *local_err = NULL;
827 char *value = qemu_opt_get_del(opts, "detect-zeroes");
828 BlockdevDetectZeroesOptions detect_zeroes =
829 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
830 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
831 g_free(value);
832 if (local_err) {
833 error_propagate(errp, local_err);
834 return detect_zeroes;
837 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
838 !(open_flags & BDRV_O_UNMAP))
840 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
841 "without setting discard operation to unmap");
844 return detect_zeroes;
848 * Set open flags for a given discard mode
850 * Return 0 on success, -1 if the discard mode was invalid.
852 int bdrv_parse_discard_flags(const char *mode, int *flags)
854 *flags &= ~BDRV_O_UNMAP;
856 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
857 /* do nothing */
858 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
859 *flags |= BDRV_O_UNMAP;
860 } else {
861 return -1;
864 return 0;
868 * Set open flags for a given cache mode
870 * Return 0 on success, -1 if the cache mode was invalid.
872 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
874 *flags &= ~BDRV_O_CACHE_MASK;
876 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
877 *writethrough = false;
878 *flags |= BDRV_O_NOCACHE;
879 } else if (!strcmp(mode, "directsync")) {
880 *writethrough = true;
881 *flags |= BDRV_O_NOCACHE;
882 } else if (!strcmp(mode, "writeback")) {
883 *writethrough = false;
884 } else if (!strcmp(mode, "unsafe")) {
885 *writethrough = false;
886 *flags |= BDRV_O_NO_FLUSH;
887 } else if (!strcmp(mode, "writethrough")) {
888 *writethrough = true;
889 } else {
890 return -1;
893 return 0;
896 static char *bdrv_child_get_parent_desc(BdrvChild *c)
898 BlockDriverState *parent = c->opaque;
899 return g_strdup(bdrv_get_device_or_node_name(parent));
902 static void bdrv_child_cb_drained_begin(BdrvChild *child)
904 BlockDriverState *bs = child->opaque;
905 bdrv_do_drained_begin_quiesce(bs, NULL, false);
908 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
910 BlockDriverState *bs = child->opaque;
911 return bdrv_drain_poll(bs, false, NULL, false);
914 static void bdrv_child_cb_drained_end(BdrvChild *child)
916 BlockDriverState *bs = child->opaque;
917 bdrv_drained_end(bs);
920 static void bdrv_child_cb_attach(BdrvChild *child)
922 BlockDriverState *bs = child->opaque;
923 bdrv_apply_subtree_drain(child, bs);
926 static void bdrv_child_cb_detach(BdrvChild *child)
928 BlockDriverState *bs = child->opaque;
929 bdrv_unapply_subtree_drain(child, bs);
932 static int bdrv_child_cb_inactivate(BdrvChild *child)
934 BlockDriverState *bs = child->opaque;
935 assert(bs->open_flags & BDRV_O_INACTIVE);
936 return 0;
940 * Returns the options and flags that a temporary snapshot should get, based on
941 * the originally requested flags (the originally requested image will have
942 * flags like a backing file)
944 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
945 int parent_flags, QDict *parent_options)
947 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
949 /* For temporary files, unconditional cache=unsafe is fine */
950 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
951 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
953 /* Copy the read-only option from the parent */
954 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
956 /* aio=native doesn't work for cache.direct=off, so disable it for the
957 * temporary snapshot */
958 *child_flags &= ~BDRV_O_NATIVE_AIO;
962 * Returns the options and flags that bs->file should get if a protocol driver
963 * is expected, based on the given options and flags for the parent BDS
965 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
966 int parent_flags, QDict *parent_options)
968 int flags = parent_flags;
970 /* Enable protocol handling, disable format probing for bs->file */
971 flags |= BDRV_O_PROTOCOL;
973 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
974 * the parent. */
975 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
976 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
977 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
979 /* Inherit the read-only option from the parent if it's not set */
980 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
981 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
983 /* Our block drivers take care to send flushes and respect unmap policy,
984 * so we can default to enable both on lower layers regardless of the
985 * corresponding parent options. */
986 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
988 /* Clear flags that only apply to the top layer */
989 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
990 BDRV_O_NO_IO);
992 *child_flags = flags;
995 const BdrvChildRole child_file = {
996 .parent_is_bds = true,
997 .get_parent_desc = bdrv_child_get_parent_desc,
998 .inherit_options = bdrv_inherited_options,
999 .drained_begin = bdrv_child_cb_drained_begin,
1000 .drained_poll = bdrv_child_cb_drained_poll,
1001 .drained_end = bdrv_child_cb_drained_end,
1002 .attach = bdrv_child_cb_attach,
1003 .detach = bdrv_child_cb_detach,
1004 .inactivate = bdrv_child_cb_inactivate,
1008 * Returns the options and flags that bs->file should get if the use of formats
1009 * (and not only protocols) is permitted for it, based on the given options and
1010 * flags for the parent BDS
1012 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1013 int parent_flags, QDict *parent_options)
1015 child_file.inherit_options(child_flags, child_options,
1016 parent_flags, parent_options);
1018 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
1021 const BdrvChildRole child_format = {
1022 .parent_is_bds = true,
1023 .get_parent_desc = bdrv_child_get_parent_desc,
1024 .inherit_options = bdrv_inherited_fmt_options,
1025 .drained_begin = bdrv_child_cb_drained_begin,
1026 .drained_poll = bdrv_child_cb_drained_poll,
1027 .drained_end = bdrv_child_cb_drained_end,
1028 .attach = bdrv_child_cb_attach,
1029 .detach = bdrv_child_cb_detach,
1030 .inactivate = bdrv_child_cb_inactivate,
1033 static void bdrv_backing_attach(BdrvChild *c)
1035 BlockDriverState *parent = c->opaque;
1036 BlockDriverState *backing_hd = c->bs;
1038 assert(!parent->backing_blocker);
1039 error_setg(&parent->backing_blocker,
1040 "node is used as backing hd of '%s'",
1041 bdrv_get_device_or_node_name(parent));
1043 bdrv_refresh_filename(backing_hd);
1045 parent->open_flags &= ~BDRV_O_NO_BACKING;
1046 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1047 backing_hd->filename);
1048 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1049 backing_hd->drv ? backing_hd->drv->format_name : "");
1051 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1052 /* Otherwise we won't be able to commit or stream */
1053 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1054 parent->backing_blocker);
1055 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1056 parent->backing_blocker);
1058 * We do backup in 3 ways:
1059 * 1. drive backup
1060 * The target bs is new opened, and the source is top BDS
1061 * 2. blockdev backup
1062 * Both the source and the target are top BDSes.
1063 * 3. internal backup(used for block replication)
1064 * Both the source and the target are backing file
1066 * In case 1 and 2, neither the source nor the target is the backing file.
1067 * In case 3, we will block the top BDS, so there is only one block job
1068 * for the top BDS and its backing chain.
1070 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1071 parent->backing_blocker);
1072 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1073 parent->backing_blocker);
1075 bdrv_child_cb_attach(c);
1078 static void bdrv_backing_detach(BdrvChild *c)
1080 BlockDriverState *parent = c->opaque;
1082 assert(parent->backing_blocker);
1083 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1084 error_free(parent->backing_blocker);
1085 parent->backing_blocker = NULL;
1087 bdrv_child_cb_detach(c);
1091 * Returns the options and flags that bs->backing should get, based on the
1092 * given options and flags for the parent BDS
1094 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1095 int parent_flags, QDict *parent_options)
1097 int flags = parent_flags;
1099 /* The cache mode is inherited unmodified for backing files; except WCE,
1100 * which is only applied on the top level (BlockBackend) */
1101 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1102 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1103 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1105 /* backing files always opened read-only */
1106 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1107 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1108 flags &= ~BDRV_O_COPY_ON_READ;
1110 /* snapshot=on is handled on the top layer */
1111 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1113 *child_flags = flags;
1116 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1117 const char *filename, Error **errp)
1119 BlockDriverState *parent = c->opaque;
1120 bool read_only = bdrv_is_read_only(parent);
1121 int ret;
1123 if (read_only) {
1124 ret = bdrv_reopen_set_read_only(parent, false, errp);
1125 if (ret < 0) {
1126 return ret;
1130 ret = bdrv_change_backing_file(parent, filename,
1131 base->drv ? base->drv->format_name : "");
1132 if (ret < 0) {
1133 error_setg_errno(errp, -ret, "Could not update backing file link");
1136 if (read_only) {
1137 bdrv_reopen_set_read_only(parent, true, NULL);
1140 return ret;
1143 const BdrvChildRole child_backing = {
1144 .parent_is_bds = true,
1145 .get_parent_desc = bdrv_child_get_parent_desc,
1146 .attach = bdrv_backing_attach,
1147 .detach = bdrv_backing_detach,
1148 .inherit_options = bdrv_backing_options,
1149 .drained_begin = bdrv_child_cb_drained_begin,
1150 .drained_poll = bdrv_child_cb_drained_poll,
1151 .drained_end = bdrv_child_cb_drained_end,
1152 .inactivate = bdrv_child_cb_inactivate,
1153 .update_filename = bdrv_backing_update_filename,
1156 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1158 int open_flags = flags;
1161 * Clear flags that are internal to the block layer before opening the
1162 * image.
1164 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1167 * Snapshots should be writable.
1169 if (flags & BDRV_O_TEMPORARY) {
1170 open_flags |= BDRV_O_RDWR;
1173 return open_flags;
1176 static void update_flags_from_options(int *flags, QemuOpts *opts)
1178 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1180 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1181 *flags |= BDRV_O_NO_FLUSH;
1184 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1185 *flags |= BDRV_O_NOCACHE;
1188 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1189 *flags |= BDRV_O_RDWR;
1192 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1193 *flags |= BDRV_O_AUTO_RDONLY;
1197 static void update_options_from_flags(QDict *options, int flags)
1199 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1200 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1202 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1203 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1204 flags & BDRV_O_NO_FLUSH);
1206 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1207 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1209 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1210 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1211 flags & BDRV_O_AUTO_RDONLY);
1215 static void bdrv_assign_node_name(BlockDriverState *bs,
1216 const char *node_name,
1217 Error **errp)
1219 char *gen_node_name = NULL;
1221 if (!node_name) {
1222 node_name = gen_node_name = id_generate(ID_BLOCK);
1223 } else if (!id_wellformed(node_name)) {
1225 * Check for empty string or invalid characters, but not if it is
1226 * generated (generated names use characters not available to the user)
1228 error_setg(errp, "Invalid node name");
1229 return;
1232 /* takes care of avoiding namespaces collisions */
1233 if (blk_by_name(node_name)) {
1234 error_setg(errp, "node-name=%s is conflicting with a device id",
1235 node_name);
1236 goto out;
1239 /* takes care of avoiding duplicates node names */
1240 if (bdrv_find_node(node_name)) {
1241 error_setg(errp, "Duplicate node name");
1242 goto out;
1245 /* Make sure that the node name isn't truncated */
1246 if (strlen(node_name) >= sizeof(bs->node_name)) {
1247 error_setg(errp, "Node name too long");
1248 goto out;
1251 /* copy node name into the bs and insert it into the graph list */
1252 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1253 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1254 out:
1255 g_free(gen_node_name);
1258 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1259 const char *node_name, QDict *options,
1260 int open_flags, Error **errp)
1262 Error *local_err = NULL;
1263 int i, ret;
1265 bdrv_assign_node_name(bs, node_name, &local_err);
1266 if (local_err) {
1267 error_propagate(errp, local_err);
1268 return -EINVAL;
1271 bs->drv = drv;
1272 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1273 bs->opaque = g_malloc0(drv->instance_size);
1275 if (drv->bdrv_file_open) {
1276 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1277 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1278 } else if (drv->bdrv_open) {
1279 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1280 } else {
1281 ret = 0;
1284 if (ret < 0) {
1285 if (local_err) {
1286 error_propagate(errp, local_err);
1287 } else if (bs->filename[0]) {
1288 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1289 } else {
1290 error_setg_errno(errp, -ret, "Could not open image");
1292 goto open_failed;
1295 ret = refresh_total_sectors(bs, bs->total_sectors);
1296 if (ret < 0) {
1297 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1298 return ret;
1301 bdrv_refresh_limits(bs, &local_err);
1302 if (local_err) {
1303 error_propagate(errp, local_err);
1304 return -EINVAL;
1307 assert(bdrv_opt_mem_align(bs) != 0);
1308 assert(bdrv_min_mem_align(bs) != 0);
1309 assert(is_power_of_2(bs->bl.request_alignment));
1311 for (i = 0; i < bs->quiesce_counter; i++) {
1312 if (drv->bdrv_co_drain_begin) {
1313 drv->bdrv_co_drain_begin(bs);
1317 return 0;
1318 open_failed:
1319 bs->drv = NULL;
1320 if (bs->file != NULL) {
1321 bdrv_unref_child(bs, bs->file);
1322 bs->file = NULL;
1324 g_free(bs->opaque);
1325 bs->opaque = NULL;
1326 return ret;
1329 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1330 int flags, Error **errp)
1332 BlockDriverState *bs;
1333 int ret;
1335 bs = bdrv_new();
1336 bs->open_flags = flags;
1337 bs->explicit_options = qdict_new();
1338 bs->options = qdict_new();
1339 bs->opaque = NULL;
1341 update_options_from_flags(bs->options, flags);
1343 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1344 if (ret < 0) {
1345 qobject_unref(bs->explicit_options);
1346 bs->explicit_options = NULL;
1347 qobject_unref(bs->options);
1348 bs->options = NULL;
1349 bdrv_unref(bs);
1350 return NULL;
1353 return bs;
1356 QemuOptsList bdrv_runtime_opts = {
1357 .name = "bdrv_common",
1358 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1359 .desc = {
1361 .name = "node-name",
1362 .type = QEMU_OPT_STRING,
1363 .help = "Node name of the block device node",
1366 .name = "driver",
1367 .type = QEMU_OPT_STRING,
1368 .help = "Block driver to use for the node",
1371 .name = BDRV_OPT_CACHE_DIRECT,
1372 .type = QEMU_OPT_BOOL,
1373 .help = "Bypass software writeback cache on the host",
1376 .name = BDRV_OPT_CACHE_NO_FLUSH,
1377 .type = QEMU_OPT_BOOL,
1378 .help = "Ignore flush requests",
1381 .name = BDRV_OPT_READ_ONLY,
1382 .type = QEMU_OPT_BOOL,
1383 .help = "Node is opened in read-only mode",
1386 .name = BDRV_OPT_AUTO_READ_ONLY,
1387 .type = QEMU_OPT_BOOL,
1388 .help = "Node can become read-only if opening read-write fails",
1391 .name = "detect-zeroes",
1392 .type = QEMU_OPT_STRING,
1393 .help = "try to optimize zero writes (off, on, unmap)",
1396 .name = BDRV_OPT_DISCARD,
1397 .type = QEMU_OPT_STRING,
1398 .help = "discard operation (ignore/off, unmap/on)",
1401 .name = BDRV_OPT_FORCE_SHARE,
1402 .type = QEMU_OPT_BOOL,
1403 .help = "always accept other writers (default: off)",
1405 { /* end of list */ }
1410 * Common part for opening disk images and files
1412 * Removes all processed options from *options.
1414 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1415 QDict *options, Error **errp)
1417 int ret, open_flags;
1418 const char *filename;
1419 const char *driver_name = NULL;
1420 const char *node_name = NULL;
1421 const char *discard;
1422 QemuOpts *opts;
1423 BlockDriver *drv;
1424 Error *local_err = NULL;
1426 assert(bs->file == NULL);
1427 assert(options != NULL && bs->options != options);
1429 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1430 qemu_opts_absorb_qdict(opts, options, &local_err);
1431 if (local_err) {
1432 error_propagate(errp, local_err);
1433 ret = -EINVAL;
1434 goto fail_opts;
1437 update_flags_from_options(&bs->open_flags, opts);
1439 driver_name = qemu_opt_get(opts, "driver");
1440 drv = bdrv_find_format(driver_name);
1441 assert(drv != NULL);
1443 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1445 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1446 error_setg(errp,
1447 BDRV_OPT_FORCE_SHARE
1448 "=on can only be used with read-only images");
1449 ret = -EINVAL;
1450 goto fail_opts;
1453 if (file != NULL) {
1454 bdrv_refresh_filename(blk_bs(file));
1455 filename = blk_bs(file)->filename;
1456 } else {
1458 * Caution: while qdict_get_try_str() is fine, getting
1459 * non-string types would require more care. When @options
1460 * come from -blockdev or blockdev_add, its members are typed
1461 * according to the QAPI schema, but when they come from
1462 * -drive, they're all QString.
1464 filename = qdict_get_try_str(options, "filename");
1467 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1468 error_setg(errp, "The '%s' block driver requires a file name",
1469 drv->format_name);
1470 ret = -EINVAL;
1471 goto fail_opts;
1474 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1475 drv->format_name);
1477 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1479 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1480 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1481 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1482 } else {
1483 ret = -ENOTSUP;
1485 if (ret < 0) {
1486 error_setg(errp,
1487 !bs->read_only && bdrv_is_whitelisted(drv, true)
1488 ? "Driver '%s' can only be used for read-only devices"
1489 : "Driver '%s' is not whitelisted",
1490 drv->format_name);
1491 goto fail_opts;
1495 /* bdrv_new() and bdrv_close() make it so */
1496 assert(atomic_read(&bs->copy_on_read) == 0);
1498 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1499 if (!bs->read_only) {
1500 bdrv_enable_copy_on_read(bs);
1501 } else {
1502 error_setg(errp, "Can't use copy-on-read on read-only device");
1503 ret = -EINVAL;
1504 goto fail_opts;
1508 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1509 if (discard != NULL) {
1510 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1511 error_setg(errp, "Invalid discard option");
1512 ret = -EINVAL;
1513 goto fail_opts;
1517 bs->detect_zeroes =
1518 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1519 if (local_err) {
1520 error_propagate(errp, local_err);
1521 ret = -EINVAL;
1522 goto fail_opts;
1525 if (filename != NULL) {
1526 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1527 } else {
1528 bs->filename[0] = '\0';
1530 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1532 /* Open the image, either directly or using a protocol */
1533 open_flags = bdrv_open_flags(bs, bs->open_flags);
1534 node_name = qemu_opt_get(opts, "node-name");
1536 assert(!drv->bdrv_file_open || file == NULL);
1537 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1538 if (ret < 0) {
1539 goto fail_opts;
1542 qemu_opts_del(opts);
1543 return 0;
1545 fail_opts:
1546 qemu_opts_del(opts);
1547 return ret;
1550 static QDict *parse_json_filename(const char *filename, Error **errp)
1552 QObject *options_obj;
1553 QDict *options;
1554 int ret;
1556 ret = strstart(filename, "json:", &filename);
1557 assert(ret);
1559 options_obj = qobject_from_json(filename, errp);
1560 if (!options_obj) {
1561 error_prepend(errp, "Could not parse the JSON options: ");
1562 return NULL;
1565 options = qobject_to(QDict, options_obj);
1566 if (!options) {
1567 qobject_unref(options_obj);
1568 error_setg(errp, "Invalid JSON object given");
1569 return NULL;
1572 qdict_flatten(options);
1574 return options;
1577 static void parse_json_protocol(QDict *options, const char **pfilename,
1578 Error **errp)
1580 QDict *json_options;
1581 Error *local_err = NULL;
1583 /* Parse json: pseudo-protocol */
1584 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1585 return;
1588 json_options = parse_json_filename(*pfilename, &local_err);
1589 if (local_err) {
1590 error_propagate(errp, local_err);
1591 return;
1594 /* Options given in the filename have lower priority than options
1595 * specified directly */
1596 qdict_join(options, json_options, false);
1597 qobject_unref(json_options);
1598 *pfilename = NULL;
1602 * Fills in default options for opening images and converts the legacy
1603 * filename/flags pair to option QDict entries.
1604 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1605 * block driver has been specified explicitly.
1607 static int bdrv_fill_options(QDict **options, const char *filename,
1608 int *flags, Error **errp)
1610 const char *drvname;
1611 bool protocol = *flags & BDRV_O_PROTOCOL;
1612 bool parse_filename = false;
1613 BlockDriver *drv = NULL;
1614 Error *local_err = NULL;
1617 * Caution: while qdict_get_try_str() is fine, getting non-string
1618 * types would require more care. When @options come from
1619 * -blockdev or blockdev_add, its members are typed according to
1620 * the QAPI schema, but when they come from -drive, they're all
1621 * QString.
1623 drvname = qdict_get_try_str(*options, "driver");
1624 if (drvname) {
1625 drv = bdrv_find_format(drvname);
1626 if (!drv) {
1627 error_setg(errp, "Unknown driver '%s'", drvname);
1628 return -ENOENT;
1630 /* If the user has explicitly specified the driver, this choice should
1631 * override the BDRV_O_PROTOCOL flag */
1632 protocol = drv->bdrv_file_open;
1635 if (protocol) {
1636 *flags |= BDRV_O_PROTOCOL;
1637 } else {
1638 *flags &= ~BDRV_O_PROTOCOL;
1641 /* Translate cache options from flags into options */
1642 update_options_from_flags(*options, *flags);
1644 /* Fetch the file name from the options QDict if necessary */
1645 if (protocol && filename) {
1646 if (!qdict_haskey(*options, "filename")) {
1647 qdict_put_str(*options, "filename", filename);
1648 parse_filename = true;
1649 } else {
1650 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1651 "the same time");
1652 return -EINVAL;
1656 /* Find the right block driver */
1657 /* See cautionary note on accessing @options above */
1658 filename = qdict_get_try_str(*options, "filename");
1660 if (!drvname && protocol) {
1661 if (filename) {
1662 drv = bdrv_find_protocol(filename, parse_filename, errp);
1663 if (!drv) {
1664 return -EINVAL;
1667 drvname = drv->format_name;
1668 qdict_put_str(*options, "driver", drvname);
1669 } else {
1670 error_setg(errp, "Must specify either driver or file");
1671 return -EINVAL;
1675 assert(drv || !protocol);
1677 /* Driver-specific filename parsing */
1678 if (drv && drv->bdrv_parse_filename && parse_filename) {
1679 drv->bdrv_parse_filename(filename, *options, &local_err);
1680 if (local_err) {
1681 error_propagate(errp, local_err);
1682 return -EINVAL;
1685 if (!drv->bdrv_needs_filename) {
1686 qdict_del(*options, "filename");
1690 return 0;
1693 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1694 uint64_t perm, uint64_t shared,
1695 GSList *ignore_children, Error **errp);
1696 static void bdrv_child_abort_perm_update(BdrvChild *c);
1697 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1699 typedef struct BlockReopenQueueEntry {
1700 bool prepared;
1701 BDRVReopenState state;
1702 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1703 } BlockReopenQueueEntry;
1706 * Return the flags that @bs will have after the reopens in @q have
1707 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1708 * return the current flags.
1710 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1712 BlockReopenQueueEntry *entry;
1714 if (q != NULL) {
1715 QSIMPLEQ_FOREACH(entry, q, entry) {
1716 if (entry->state.bs == bs) {
1717 return entry->state.flags;
1722 return bs->open_flags;
1725 /* Returns whether the image file can be written to after the reopen queue @q
1726 * has been successfully applied, or right now if @q is NULL. */
1727 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1728 BlockReopenQueue *q)
1730 int flags = bdrv_reopen_get_flags(q, bs);
1732 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1736 * Return whether the BDS can be written to. This is not necessarily
1737 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1738 * be written to but do not count as read-only images.
1740 bool bdrv_is_writable(BlockDriverState *bs)
1742 return bdrv_is_writable_after_reopen(bs, NULL);
1745 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1746 BdrvChild *c, const BdrvChildRole *role,
1747 BlockReopenQueue *reopen_queue,
1748 uint64_t parent_perm, uint64_t parent_shared,
1749 uint64_t *nperm, uint64_t *nshared)
1751 if (bs->drv && bs->drv->bdrv_child_perm) {
1752 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1753 parent_perm, parent_shared,
1754 nperm, nshared);
1756 /* TODO Take force_share from reopen_queue */
1757 if (child_bs && child_bs->force_share) {
1758 *nshared = BLK_PERM_ALL;
1763 * Check whether permissions on this node can be changed in a way that
1764 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1765 * permissions of all its parents. This involves checking whether all necessary
1766 * permission changes to child nodes can be performed.
1768 * A call to this function must always be followed by a call to bdrv_set_perm()
1769 * or bdrv_abort_perm_update().
1771 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1772 uint64_t cumulative_perms,
1773 uint64_t cumulative_shared_perms,
1774 GSList *ignore_children, Error **errp)
1776 BlockDriver *drv = bs->drv;
1777 BdrvChild *c;
1778 int ret;
1780 /* Write permissions never work with read-only images */
1781 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1782 !bdrv_is_writable_after_reopen(bs, q))
1784 error_setg(errp, "Block node is read-only");
1785 return -EPERM;
1788 /* Check this node */
1789 if (!drv) {
1790 return 0;
1793 if (drv->bdrv_check_perm) {
1794 return drv->bdrv_check_perm(bs, cumulative_perms,
1795 cumulative_shared_perms, errp);
1798 /* Drivers that never have children can omit .bdrv_child_perm() */
1799 if (!drv->bdrv_child_perm) {
1800 assert(QLIST_EMPTY(&bs->children));
1801 return 0;
1804 /* Check all children */
1805 QLIST_FOREACH(c, &bs->children, next) {
1806 uint64_t cur_perm, cur_shared;
1807 bdrv_child_perm(bs, c->bs, c, c->role, q,
1808 cumulative_perms, cumulative_shared_perms,
1809 &cur_perm, &cur_shared);
1810 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1811 ignore_children, errp);
1812 if (ret < 0) {
1813 return ret;
1817 return 0;
1821 * Notifies drivers that after a previous bdrv_check_perm() call, the
1822 * permission update is not performed and any preparations made for it (e.g.
1823 * taken file locks) need to be undone.
1825 * This function recursively notifies all child nodes.
1827 static void bdrv_abort_perm_update(BlockDriverState *bs)
1829 BlockDriver *drv = bs->drv;
1830 BdrvChild *c;
1832 if (!drv) {
1833 return;
1836 if (drv->bdrv_abort_perm_update) {
1837 drv->bdrv_abort_perm_update(bs);
1840 QLIST_FOREACH(c, &bs->children, next) {
1841 bdrv_child_abort_perm_update(c);
1845 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1846 uint64_t cumulative_shared_perms)
1848 BlockDriver *drv = bs->drv;
1849 BdrvChild *c;
1851 if (!drv) {
1852 return;
1855 /* Update this node */
1856 if (drv->bdrv_set_perm) {
1857 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1860 /* Drivers that never have children can omit .bdrv_child_perm() */
1861 if (!drv->bdrv_child_perm) {
1862 assert(QLIST_EMPTY(&bs->children));
1863 return;
1866 /* Update all children */
1867 QLIST_FOREACH(c, &bs->children, next) {
1868 uint64_t cur_perm, cur_shared;
1869 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1870 cumulative_perms, cumulative_shared_perms,
1871 &cur_perm, &cur_shared);
1872 bdrv_child_set_perm(c, cur_perm, cur_shared);
1876 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1877 uint64_t *shared_perm)
1879 BdrvChild *c;
1880 uint64_t cumulative_perms = 0;
1881 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1883 QLIST_FOREACH(c, &bs->parents, next_parent) {
1884 cumulative_perms |= c->perm;
1885 cumulative_shared_perms &= c->shared_perm;
1888 *perm = cumulative_perms;
1889 *shared_perm = cumulative_shared_perms;
1892 static char *bdrv_child_user_desc(BdrvChild *c)
1894 if (c->role->get_parent_desc) {
1895 return c->role->get_parent_desc(c);
1898 return g_strdup("another user");
1901 char *bdrv_perm_names(uint64_t perm)
1903 struct perm_name {
1904 uint64_t perm;
1905 const char *name;
1906 } permissions[] = {
1907 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1908 { BLK_PERM_WRITE, "write" },
1909 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1910 { BLK_PERM_RESIZE, "resize" },
1911 { BLK_PERM_GRAPH_MOD, "change children" },
1912 { 0, NULL }
1915 char *result = g_strdup("");
1916 struct perm_name *p;
1918 for (p = permissions; p->name; p++) {
1919 if (perm & p->perm) {
1920 char *old = result;
1921 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1922 g_free(old);
1926 return result;
1930 * Checks whether a new reference to @bs can be added if the new user requires
1931 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1932 * set, the BdrvChild objects in this list are ignored in the calculations;
1933 * this allows checking permission updates for an existing reference.
1935 * Needs to be followed by a call to either bdrv_set_perm() or
1936 * bdrv_abort_perm_update(). */
1937 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1938 uint64_t new_used_perm,
1939 uint64_t new_shared_perm,
1940 GSList *ignore_children, Error **errp)
1942 BdrvChild *c;
1943 uint64_t cumulative_perms = new_used_perm;
1944 uint64_t cumulative_shared_perms = new_shared_perm;
1946 /* There is no reason why anyone couldn't tolerate write_unchanged */
1947 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1949 QLIST_FOREACH(c, &bs->parents, next_parent) {
1950 if (g_slist_find(ignore_children, c)) {
1951 continue;
1954 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1955 char *user = bdrv_child_user_desc(c);
1956 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1957 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1958 "allow '%s' on %s",
1959 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1960 g_free(user);
1961 g_free(perm_names);
1962 return -EPERM;
1965 if ((c->perm & new_shared_perm) != c->perm) {
1966 char *user = bdrv_child_user_desc(c);
1967 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1968 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1969 "'%s' on %s",
1970 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1971 g_free(user);
1972 g_free(perm_names);
1973 return -EPERM;
1976 cumulative_perms |= c->perm;
1977 cumulative_shared_perms &= c->shared_perm;
1980 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1981 ignore_children, errp);
1984 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1985 * bdrv_child_abort_perm_update(). */
1986 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1987 uint64_t perm, uint64_t shared,
1988 GSList *ignore_children, Error **errp)
1990 int ret;
1992 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1993 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1994 g_slist_free(ignore_children);
1996 if (ret < 0) {
1997 return ret;
2000 if (!c->has_backup_perm) {
2001 c->has_backup_perm = true;
2002 c->backup_perm = c->perm;
2003 c->backup_shared_perm = c->shared_perm;
2006 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2007 * same child twice during check_perm procedure
2010 c->perm = perm;
2011 c->shared_perm = shared;
2013 return 0;
2016 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2018 uint64_t cumulative_perms, cumulative_shared_perms;
2020 c->has_backup_perm = false;
2022 c->perm = perm;
2023 c->shared_perm = shared;
2025 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2026 &cumulative_shared_perms);
2027 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2030 static void bdrv_child_abort_perm_update(BdrvChild *c)
2032 if (c->has_backup_perm) {
2033 c->perm = c->backup_perm;
2034 c->shared_perm = c->backup_shared_perm;
2035 c->has_backup_perm = false;
2038 bdrv_abort_perm_update(c->bs);
2041 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2042 Error **errp)
2044 int ret;
2046 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
2047 if (ret < 0) {
2048 bdrv_child_abort_perm_update(c);
2049 return ret;
2052 bdrv_child_set_perm(c, perm, shared);
2054 return 0;
2057 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2058 const BdrvChildRole *role,
2059 BlockReopenQueue *reopen_queue,
2060 uint64_t perm, uint64_t shared,
2061 uint64_t *nperm, uint64_t *nshared)
2063 if (c == NULL) {
2064 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2065 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2066 return;
2069 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
2070 (c->perm & DEFAULT_PERM_UNCHANGED);
2071 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
2072 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
2075 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2076 const BdrvChildRole *role,
2077 BlockReopenQueue *reopen_queue,
2078 uint64_t perm, uint64_t shared,
2079 uint64_t *nperm, uint64_t *nshared)
2081 bool backing = (role == &child_backing);
2082 assert(role == &child_backing || role == &child_file);
2084 if (!backing) {
2085 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2087 /* Apart from the modifications below, the same permissions are
2088 * forwarded and left alone as for filters */
2089 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2090 &perm, &shared);
2092 /* Format drivers may touch metadata even if the guest doesn't write */
2093 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2094 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2097 /* bs->file always needs to be consistent because of the metadata. We
2098 * can never allow other users to resize or write to it. */
2099 if (!(flags & BDRV_O_NO_IO)) {
2100 perm |= BLK_PERM_CONSISTENT_READ;
2102 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2103 } else {
2104 /* We want consistent read from backing files if the parent needs it.
2105 * No other operations are performed on backing files. */
2106 perm &= BLK_PERM_CONSISTENT_READ;
2108 /* If the parent can deal with changing data, we're okay with a
2109 * writable and resizable backing file. */
2110 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2111 if (shared & BLK_PERM_WRITE) {
2112 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2113 } else {
2114 shared = 0;
2117 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2118 BLK_PERM_WRITE_UNCHANGED;
2121 if (bs->open_flags & BDRV_O_INACTIVE) {
2122 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2125 *nperm = perm;
2126 *nshared = shared;
2129 static void bdrv_replace_child_noperm(BdrvChild *child,
2130 BlockDriverState *new_bs)
2132 BlockDriverState *old_bs = child->bs;
2133 int i;
2135 if (old_bs && new_bs) {
2136 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2138 if (old_bs) {
2139 /* Detach first so that the recursive drain sections coming from @child
2140 * are already gone and we only end the drain sections that came from
2141 * elsewhere. */
2142 if (child->role->detach) {
2143 child->role->detach(child);
2145 if (old_bs->quiesce_counter && child->role->drained_end) {
2146 int num = old_bs->quiesce_counter;
2147 if (child->role->parent_is_bds) {
2148 num -= bdrv_drain_all_count;
2150 assert(num >= 0);
2151 for (i = 0; i < num; i++) {
2152 child->role->drained_end(child);
2155 QLIST_REMOVE(child, next_parent);
2158 child->bs = new_bs;
2160 if (new_bs) {
2161 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2162 if (new_bs->quiesce_counter && child->role->drained_begin) {
2163 int num = new_bs->quiesce_counter;
2164 if (child->role->parent_is_bds) {
2165 num -= bdrv_drain_all_count;
2167 assert(num >= 0);
2168 for (i = 0; i < num; i++) {
2169 bdrv_parent_drained_begin_single(child, true);
2173 /* Attach only after starting new drained sections, so that recursive
2174 * drain sections coming from @child don't get an extra .drained_begin
2175 * callback. */
2176 if (child->role->attach) {
2177 child->role->attach(child);
2183 * Updates @child to change its reference to point to @new_bs, including
2184 * checking and applying the necessary permisson updates both to the old node
2185 * and to @new_bs.
2187 * NULL is passed as @new_bs for removing the reference before freeing @child.
2189 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2190 * function uses bdrv_set_perm() to update the permissions according to the new
2191 * reference that @new_bs gets.
2193 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2195 BlockDriverState *old_bs = child->bs;
2196 uint64_t perm, shared_perm;
2198 bdrv_replace_child_noperm(child, new_bs);
2200 if (old_bs) {
2201 /* Update permissions for old node. This is guaranteed to succeed
2202 * because we're just taking a parent away, so we're loosening
2203 * restrictions. */
2204 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2205 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2206 bdrv_set_perm(old_bs, perm, shared_perm);
2209 if (new_bs) {
2210 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2211 bdrv_set_perm(new_bs, perm, shared_perm);
2215 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2216 const char *child_name,
2217 const BdrvChildRole *child_role,
2218 uint64_t perm, uint64_t shared_perm,
2219 void *opaque, Error **errp)
2221 BdrvChild *child;
2222 int ret;
2224 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2225 if (ret < 0) {
2226 bdrv_abort_perm_update(child_bs);
2227 return NULL;
2230 child = g_new(BdrvChild, 1);
2231 *child = (BdrvChild) {
2232 .bs = NULL,
2233 .name = g_strdup(child_name),
2234 .role = child_role,
2235 .perm = perm,
2236 .shared_perm = shared_perm,
2237 .opaque = opaque,
2240 /* This performs the matching bdrv_set_perm() for the above check. */
2241 bdrv_replace_child(child, child_bs);
2243 return child;
2246 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2247 BlockDriverState *child_bs,
2248 const char *child_name,
2249 const BdrvChildRole *child_role,
2250 Error **errp)
2252 BdrvChild *child;
2253 uint64_t perm, shared_perm;
2255 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2257 assert(parent_bs->drv);
2258 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2259 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2260 perm, shared_perm, &perm, &shared_perm);
2262 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2263 perm, shared_perm, parent_bs, errp);
2264 if (child == NULL) {
2265 return NULL;
2268 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2269 return child;
2272 static void bdrv_detach_child(BdrvChild *child)
2274 if (child->next.le_prev) {
2275 QLIST_REMOVE(child, next);
2276 child->next.le_prev = NULL;
2279 bdrv_replace_child(child, NULL);
2281 g_free(child->name);
2282 g_free(child);
2285 void bdrv_root_unref_child(BdrvChild *child)
2287 BlockDriverState *child_bs;
2289 child_bs = child->bs;
2290 bdrv_detach_child(child);
2291 bdrv_unref(child_bs);
2294 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2296 if (child == NULL) {
2297 return;
2300 if (child->bs->inherits_from == parent) {
2301 BdrvChild *c;
2303 /* Remove inherits_from only when the last reference between parent and
2304 * child->bs goes away. */
2305 QLIST_FOREACH(c, &parent->children, next) {
2306 if (c != child && c->bs == child->bs) {
2307 break;
2310 if (c == NULL) {
2311 child->bs->inherits_from = NULL;
2315 bdrv_root_unref_child(child);
2319 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2321 BdrvChild *c;
2322 QLIST_FOREACH(c, &bs->parents, next_parent) {
2323 if (c->role->change_media) {
2324 c->role->change_media(c, load);
2329 /* Return true if you can reach parent going through child->inherits_from
2330 * recursively. If parent or child are NULL, return false */
2331 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2332 BlockDriverState *parent)
2334 while (child && child != parent) {
2335 child = child->inherits_from;
2338 return child != NULL;
2342 * Sets the backing file link of a BDS. A new reference is created; callers
2343 * which don't need their own reference any more must call bdrv_unref().
2345 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2346 Error **errp)
2348 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2349 bdrv_inherits_from_recursive(backing_hd, bs);
2351 if (backing_hd) {
2352 bdrv_ref(backing_hd);
2355 if (bs->backing) {
2356 bdrv_unref_child(bs, bs->backing);
2359 if (!backing_hd) {
2360 bs->backing = NULL;
2361 goto out;
2364 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2365 errp);
2366 /* If backing_hd was already part of bs's backing chain, and
2367 * inherits_from pointed recursively to bs then let's update it to
2368 * point directly to bs (else it will become NULL). */
2369 if (update_inherits_from) {
2370 backing_hd->inherits_from = bs;
2372 if (!bs->backing) {
2373 bdrv_unref(backing_hd);
2376 out:
2377 bdrv_refresh_limits(bs, NULL);
2381 * Opens the backing file for a BlockDriverState if not yet open
2383 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2384 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2385 * itself, all options starting with "${bdref_key}." are considered part of the
2386 * BlockdevRef.
2388 * TODO Can this be unified with bdrv_open_image()?
2390 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2391 const char *bdref_key, Error **errp)
2393 char *backing_filename = NULL;
2394 char *bdref_key_dot;
2395 const char *reference = NULL;
2396 int ret = 0;
2397 bool implicit_backing = false;
2398 BlockDriverState *backing_hd;
2399 QDict *options;
2400 QDict *tmp_parent_options = NULL;
2401 Error *local_err = NULL;
2403 if (bs->backing != NULL) {
2404 goto free_exit;
2407 /* NULL means an empty set of options */
2408 if (parent_options == NULL) {
2409 tmp_parent_options = qdict_new();
2410 parent_options = tmp_parent_options;
2413 bs->open_flags &= ~BDRV_O_NO_BACKING;
2415 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2416 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2417 g_free(bdref_key_dot);
2420 * Caution: while qdict_get_try_str() is fine, getting non-string
2421 * types would require more care. When @parent_options come from
2422 * -blockdev or blockdev_add, its members are typed according to
2423 * the QAPI schema, but when they come from -drive, they're all
2424 * QString.
2426 reference = qdict_get_try_str(parent_options, bdref_key);
2427 if (reference || qdict_haskey(options, "file.filename")) {
2428 /* keep backing_filename NULL */
2429 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2430 qobject_unref(options);
2431 goto free_exit;
2432 } else {
2433 if (qdict_size(options) == 0) {
2434 /* If the user specifies options that do not modify the
2435 * backing file's behavior, we might still consider it the
2436 * implicit backing file. But it's easier this way, and
2437 * just specifying some of the backing BDS's options is
2438 * only possible with -drive anyway (otherwise the QAPI
2439 * schema forces the user to specify everything). */
2440 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2443 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2444 if (local_err) {
2445 ret = -EINVAL;
2446 error_propagate(errp, local_err);
2447 qobject_unref(options);
2448 goto free_exit;
2452 if (!bs->drv || !bs->drv->supports_backing) {
2453 ret = -EINVAL;
2454 error_setg(errp, "Driver doesn't support backing files");
2455 qobject_unref(options);
2456 goto free_exit;
2459 if (!reference &&
2460 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2461 qdict_put_str(options, "driver", bs->backing_format);
2464 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2465 &child_backing, errp);
2466 if (!backing_hd) {
2467 bs->open_flags |= BDRV_O_NO_BACKING;
2468 error_prepend(errp, "Could not open backing file: ");
2469 ret = -EINVAL;
2470 goto free_exit;
2472 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2474 if (implicit_backing) {
2475 bdrv_refresh_filename(backing_hd);
2476 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2477 backing_hd->filename);
2480 /* Hook up the backing file link; drop our reference, bs owns the
2481 * backing_hd reference now */
2482 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2483 bdrv_unref(backing_hd);
2484 if (local_err) {
2485 error_propagate(errp, local_err);
2486 ret = -EINVAL;
2487 goto free_exit;
2490 qdict_del(parent_options, bdref_key);
2492 free_exit:
2493 g_free(backing_filename);
2494 qobject_unref(tmp_parent_options);
2495 return ret;
2498 static BlockDriverState *
2499 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2500 BlockDriverState *parent, const BdrvChildRole *child_role,
2501 bool allow_none, Error **errp)
2503 BlockDriverState *bs = NULL;
2504 QDict *image_options;
2505 char *bdref_key_dot;
2506 const char *reference;
2508 assert(child_role != NULL);
2510 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2511 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2512 g_free(bdref_key_dot);
2515 * Caution: while qdict_get_try_str() is fine, getting non-string
2516 * types would require more care. When @options come from
2517 * -blockdev or blockdev_add, its members are typed according to
2518 * the QAPI schema, but when they come from -drive, they're all
2519 * QString.
2521 reference = qdict_get_try_str(options, bdref_key);
2522 if (!filename && !reference && !qdict_size(image_options)) {
2523 if (!allow_none) {
2524 error_setg(errp, "A block device must be specified for \"%s\"",
2525 bdref_key);
2527 qobject_unref(image_options);
2528 goto done;
2531 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2532 parent, child_role, errp);
2533 if (!bs) {
2534 goto done;
2537 done:
2538 qdict_del(options, bdref_key);
2539 return bs;
2543 * Opens a disk image whose options are given as BlockdevRef in another block
2544 * device's options.
2546 * If allow_none is true, no image will be opened if filename is false and no
2547 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2549 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2550 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2551 * itself, all options starting with "${bdref_key}." are considered part of the
2552 * BlockdevRef.
2554 * The BlockdevRef will be removed from the options QDict.
2556 BdrvChild *bdrv_open_child(const char *filename,
2557 QDict *options, const char *bdref_key,
2558 BlockDriverState *parent,
2559 const BdrvChildRole *child_role,
2560 bool allow_none, Error **errp)
2562 BdrvChild *c;
2563 BlockDriverState *bs;
2565 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2566 allow_none, errp);
2567 if (bs == NULL) {
2568 return NULL;
2571 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2572 if (!c) {
2573 bdrv_unref(bs);
2574 return NULL;
2577 return c;
2580 /* TODO Future callers may need to specify parent/child_role in order for
2581 * option inheritance to work. Existing callers use it for the root node. */
2582 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2584 BlockDriverState *bs = NULL;
2585 Error *local_err = NULL;
2586 QObject *obj = NULL;
2587 QDict *qdict = NULL;
2588 const char *reference = NULL;
2589 Visitor *v = NULL;
2591 if (ref->type == QTYPE_QSTRING) {
2592 reference = ref->u.reference;
2593 } else {
2594 BlockdevOptions *options = &ref->u.definition;
2595 assert(ref->type == QTYPE_QDICT);
2597 v = qobject_output_visitor_new(&obj);
2598 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2599 if (local_err) {
2600 error_propagate(errp, local_err);
2601 goto fail;
2603 visit_complete(v, &obj);
2605 qdict = qobject_to(QDict, obj);
2606 qdict_flatten(qdict);
2608 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2609 * compatibility with other callers) rather than what we want as the
2610 * real defaults. Apply the defaults here instead. */
2611 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2612 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2613 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2614 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2618 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2619 obj = NULL;
2621 fail:
2622 qobject_unref(obj);
2623 visit_free(v);
2624 return bs;
2627 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2628 int flags,
2629 QDict *snapshot_options,
2630 Error **errp)
2632 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2633 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2634 int64_t total_size;
2635 QemuOpts *opts = NULL;
2636 BlockDriverState *bs_snapshot = NULL;
2637 Error *local_err = NULL;
2638 int ret;
2640 /* if snapshot, we create a temporary backing file and open it
2641 instead of opening 'filename' directly */
2643 /* Get the required size from the image */
2644 total_size = bdrv_getlength(bs);
2645 if (total_size < 0) {
2646 error_setg_errno(errp, -total_size, "Could not get image size");
2647 goto out;
2650 /* Create the temporary image */
2651 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2652 if (ret < 0) {
2653 error_setg_errno(errp, -ret, "Could not get temporary filename");
2654 goto out;
2657 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2658 &error_abort);
2659 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2660 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2661 qemu_opts_del(opts);
2662 if (ret < 0) {
2663 error_prepend(errp, "Could not create temporary overlay '%s': ",
2664 tmp_filename);
2665 goto out;
2668 /* Prepare options QDict for the temporary file */
2669 qdict_put_str(snapshot_options, "file.driver", "file");
2670 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2671 qdict_put_str(snapshot_options, "driver", "qcow2");
2673 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2674 snapshot_options = NULL;
2675 if (!bs_snapshot) {
2676 goto out;
2679 /* bdrv_append() consumes a strong reference to bs_snapshot
2680 * (i.e. it will call bdrv_unref() on it) even on error, so in
2681 * order to be able to return one, we have to increase
2682 * bs_snapshot's refcount here */
2683 bdrv_ref(bs_snapshot);
2684 bdrv_append(bs_snapshot, bs, &local_err);
2685 if (local_err) {
2686 error_propagate(errp, local_err);
2687 bs_snapshot = NULL;
2688 goto out;
2691 out:
2692 qobject_unref(snapshot_options);
2693 g_free(tmp_filename);
2694 return bs_snapshot;
2698 * Opens a disk image (raw, qcow2, vmdk, ...)
2700 * options is a QDict of options to pass to the block drivers, or NULL for an
2701 * empty set of options. The reference to the QDict belongs to the block layer
2702 * after the call (even on failure), so if the caller intends to reuse the
2703 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2705 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2706 * If it is not NULL, the referenced BDS will be reused.
2708 * The reference parameter may be used to specify an existing block device which
2709 * should be opened. If specified, neither options nor a filename may be given,
2710 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2712 static BlockDriverState *bdrv_open_inherit(const char *filename,
2713 const char *reference,
2714 QDict *options, int flags,
2715 BlockDriverState *parent,
2716 const BdrvChildRole *child_role,
2717 Error **errp)
2719 int ret;
2720 BlockBackend *file = NULL;
2721 BlockDriverState *bs;
2722 BlockDriver *drv = NULL;
2723 BdrvChild *child;
2724 const char *drvname;
2725 const char *backing;
2726 Error *local_err = NULL;
2727 QDict *snapshot_options = NULL;
2728 int snapshot_flags = 0;
2730 assert(!child_role || !flags);
2731 assert(!child_role == !parent);
2733 if (reference) {
2734 bool options_non_empty = options ? qdict_size(options) : false;
2735 qobject_unref(options);
2737 if (filename || options_non_empty) {
2738 error_setg(errp, "Cannot reference an existing block device with "
2739 "additional options or a new filename");
2740 return NULL;
2743 bs = bdrv_lookup_bs(reference, reference, errp);
2744 if (!bs) {
2745 return NULL;
2748 bdrv_ref(bs);
2749 return bs;
2752 bs = bdrv_new();
2754 /* NULL means an empty set of options */
2755 if (options == NULL) {
2756 options = qdict_new();
2759 /* json: syntax counts as explicit options, as if in the QDict */
2760 parse_json_protocol(options, &filename, &local_err);
2761 if (local_err) {
2762 goto fail;
2765 bs->explicit_options = qdict_clone_shallow(options);
2767 if (child_role) {
2768 bs->inherits_from = parent;
2769 child_role->inherit_options(&flags, options,
2770 parent->open_flags, parent->options);
2773 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2774 if (local_err) {
2775 goto fail;
2779 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2780 * Caution: getting a boolean member of @options requires care.
2781 * When @options come from -blockdev or blockdev_add, members are
2782 * typed according to the QAPI schema, but when they come from
2783 * -drive, they're all QString.
2785 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2786 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2787 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2788 } else {
2789 flags &= ~BDRV_O_RDWR;
2792 if (flags & BDRV_O_SNAPSHOT) {
2793 snapshot_options = qdict_new();
2794 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2795 flags, options);
2796 /* Let bdrv_backing_options() override "read-only" */
2797 qdict_del(options, BDRV_OPT_READ_ONLY);
2798 bdrv_backing_options(&flags, options, flags, options);
2801 bs->open_flags = flags;
2802 bs->options = options;
2803 options = qdict_clone_shallow(options);
2805 /* Find the right image format driver */
2806 /* See cautionary note on accessing @options above */
2807 drvname = qdict_get_try_str(options, "driver");
2808 if (drvname) {
2809 drv = bdrv_find_format(drvname);
2810 if (!drv) {
2811 error_setg(errp, "Unknown driver: '%s'", drvname);
2812 goto fail;
2816 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2818 /* See cautionary note on accessing @options above */
2819 backing = qdict_get_try_str(options, "backing");
2820 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2821 (backing && *backing == '\0'))
2823 if (backing) {
2824 warn_report("Use of \"backing\": \"\" is deprecated; "
2825 "use \"backing\": null instead");
2827 flags |= BDRV_O_NO_BACKING;
2828 qdict_del(options, "backing");
2831 /* Open image file without format layer. This BlockBackend is only used for
2832 * probing, the block drivers will do their own bdrv_open_child() for the
2833 * same BDS, which is why we put the node name back into options. */
2834 if ((flags & BDRV_O_PROTOCOL) == 0) {
2835 BlockDriverState *file_bs;
2837 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2838 &child_file, true, &local_err);
2839 if (local_err) {
2840 goto fail;
2842 if (file_bs != NULL) {
2843 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2844 * looking at the header to guess the image format. This works even
2845 * in cases where a guest would not see a consistent state. */
2846 file = blk_new(0, BLK_PERM_ALL);
2847 blk_insert_bs(file, file_bs, &local_err);
2848 bdrv_unref(file_bs);
2849 if (local_err) {
2850 goto fail;
2853 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2857 /* Image format probing */
2858 bs->probed = !drv;
2859 if (!drv && file) {
2860 ret = find_image_format(file, filename, &drv, &local_err);
2861 if (ret < 0) {
2862 goto fail;
2865 * This option update would logically belong in bdrv_fill_options(),
2866 * but we first need to open bs->file for the probing to work, while
2867 * opening bs->file already requires the (mostly) final set of options
2868 * so that cache mode etc. can be inherited.
2870 * Adding the driver later is somewhat ugly, but it's not an option
2871 * that would ever be inherited, so it's correct. We just need to make
2872 * sure to update both bs->options (which has the full effective
2873 * options for bs) and options (which has file.* already removed).
2875 qdict_put_str(bs->options, "driver", drv->format_name);
2876 qdict_put_str(options, "driver", drv->format_name);
2877 } else if (!drv) {
2878 error_setg(errp, "Must specify either driver or file");
2879 goto fail;
2882 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2883 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2884 /* file must be NULL if a protocol BDS is about to be created
2885 * (the inverse results in an error message from bdrv_open_common()) */
2886 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2888 /* Open the image */
2889 ret = bdrv_open_common(bs, file, options, &local_err);
2890 if (ret < 0) {
2891 goto fail;
2894 if (file) {
2895 blk_unref(file);
2896 file = NULL;
2899 /* If there is a backing file, use it */
2900 if ((flags & BDRV_O_NO_BACKING) == 0) {
2901 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2902 if (ret < 0) {
2903 goto close_and_fail;
2907 /* Remove all children options and references
2908 * from bs->options and bs->explicit_options */
2909 QLIST_FOREACH(child, &bs->children, next) {
2910 char *child_key_dot;
2911 child_key_dot = g_strdup_printf("%s.", child->name);
2912 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
2913 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
2914 qdict_del(bs->explicit_options, child->name);
2915 qdict_del(bs->options, child->name);
2916 g_free(child_key_dot);
2919 /* Check if any unknown options were used */
2920 if (qdict_size(options) != 0) {
2921 const QDictEntry *entry = qdict_first(options);
2922 if (flags & BDRV_O_PROTOCOL) {
2923 error_setg(errp, "Block protocol '%s' doesn't support the option "
2924 "'%s'", drv->format_name, entry->key);
2925 } else {
2926 error_setg(errp,
2927 "Block format '%s' does not support the option '%s'",
2928 drv->format_name, entry->key);
2931 goto close_and_fail;
2934 bdrv_parent_cb_change_media(bs, true);
2936 qobject_unref(options);
2937 options = NULL;
2939 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2940 * temporary snapshot afterwards. */
2941 if (snapshot_flags) {
2942 BlockDriverState *snapshot_bs;
2943 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2944 snapshot_options, &local_err);
2945 snapshot_options = NULL;
2946 if (local_err) {
2947 goto close_and_fail;
2949 /* We are not going to return bs but the overlay on top of it
2950 * (snapshot_bs); thus, we have to drop the strong reference to bs
2951 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2952 * though, because the overlay still has a reference to it. */
2953 bdrv_unref(bs);
2954 bs = snapshot_bs;
2957 return bs;
2959 fail:
2960 blk_unref(file);
2961 qobject_unref(snapshot_options);
2962 qobject_unref(bs->explicit_options);
2963 qobject_unref(bs->options);
2964 qobject_unref(options);
2965 bs->options = NULL;
2966 bs->explicit_options = NULL;
2967 bdrv_unref(bs);
2968 error_propagate(errp, local_err);
2969 return NULL;
2971 close_and_fail:
2972 bdrv_unref(bs);
2973 qobject_unref(snapshot_options);
2974 qobject_unref(options);
2975 error_propagate(errp, local_err);
2976 return NULL;
2979 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2980 QDict *options, int flags, Error **errp)
2982 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2983 NULL, errp);
2987 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2988 * reopen of multiple devices.
2990 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2991 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2992 * be created and initialized. This newly created BlockReopenQueue should be
2993 * passed back in for subsequent calls that are intended to be of the same
2994 * atomic 'set'.
2996 * bs is the BlockDriverState to add to the reopen queue.
2998 * options contains the changed options for the associated bs
2999 * (the BlockReopenQueue takes ownership)
3001 * flags contains the open flags for the associated bs
3003 * returns a pointer to bs_queue, which is either the newly allocated
3004 * bs_queue, or the existing bs_queue being used.
3006 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3008 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3009 BlockDriverState *bs,
3010 QDict *options,
3011 const BdrvChildRole *role,
3012 QDict *parent_options,
3013 int parent_flags)
3015 assert(bs != NULL);
3017 BlockReopenQueueEntry *bs_entry;
3018 BdrvChild *child;
3019 QDict *old_options, *explicit_options, *options_copy;
3020 int flags;
3021 QemuOpts *opts;
3023 /* Make sure that the caller remembered to use a drained section. This is
3024 * important to avoid graph changes between the recursive queuing here and
3025 * bdrv_reopen_multiple(). */
3026 assert(bs->quiesce_counter > 0);
3028 if (bs_queue == NULL) {
3029 bs_queue = g_new0(BlockReopenQueue, 1);
3030 QSIMPLEQ_INIT(bs_queue);
3033 if (!options) {
3034 options = qdict_new();
3037 /* Check if this BlockDriverState is already in the queue */
3038 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3039 if (bs == bs_entry->state.bs) {
3040 break;
3045 * Precedence of options:
3046 * 1. Explicitly passed in options (highest)
3047 * 2. Retained from explicitly set options of bs
3048 * 3. Inherited from parent node
3049 * 4. Retained from effective options of bs
3052 /* Old explicitly set values (don't overwrite by inherited value) */
3053 if (bs_entry) {
3054 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
3055 } else {
3056 old_options = qdict_clone_shallow(bs->explicit_options);
3058 bdrv_join_options(bs, options, old_options);
3059 qobject_unref(old_options);
3061 explicit_options = qdict_clone_shallow(options);
3063 /* Inherit from parent node */
3064 if (parent_options) {
3065 flags = 0;
3066 role->inherit_options(&flags, options, parent_flags, parent_options);
3067 } else {
3068 flags = bdrv_get_flags(bs);
3071 /* Old values are used for options that aren't set yet */
3072 old_options = qdict_clone_shallow(bs->options);
3073 bdrv_join_options(bs, options, old_options);
3074 qobject_unref(old_options);
3076 /* We have the final set of options so let's update the flags */
3077 options_copy = qdict_clone_shallow(options);
3078 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3079 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3080 update_flags_from_options(&flags, opts);
3081 qemu_opts_del(opts);
3082 qobject_unref(options_copy);
3084 /* bdrv_open_inherit() sets and clears some additional flags internally */
3085 flags &= ~BDRV_O_PROTOCOL;
3086 if (flags & BDRV_O_RDWR) {
3087 flags |= BDRV_O_ALLOW_RDWR;
3090 if (!bs_entry) {
3091 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3092 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3093 } else {
3094 qobject_unref(bs_entry->state.options);
3095 qobject_unref(bs_entry->state.explicit_options);
3098 bs_entry->state.bs = bs;
3099 bs_entry->state.options = options;
3100 bs_entry->state.explicit_options = explicit_options;
3101 bs_entry->state.flags = flags;
3103 /* This needs to be overwritten in bdrv_reopen_prepare() */
3104 bs_entry->state.perm = UINT64_MAX;
3105 bs_entry->state.shared_perm = 0;
3107 QLIST_FOREACH(child, &bs->children, next) {
3108 QDict *new_child_options;
3109 char *child_key_dot;
3111 /* reopen can only change the options of block devices that were
3112 * implicitly created and inherited options. For other (referenced)
3113 * block devices, a syntax like "backing.foo" results in an error. */
3114 if (child->bs->inherits_from != bs) {
3115 continue;
3118 child_key_dot = g_strdup_printf("%s.", child->name);
3119 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3120 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3121 g_free(child_key_dot);
3123 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3124 child->role, options, flags);
3127 return bs_queue;
3130 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3131 BlockDriverState *bs,
3132 QDict *options)
3134 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0);
3138 * Reopen multiple BlockDriverStates atomically & transactionally.
3140 * The queue passed in (bs_queue) must have been built up previous
3141 * via bdrv_reopen_queue().
3143 * Reopens all BDS specified in the queue, with the appropriate
3144 * flags. All devices are prepared for reopen, and failure of any
3145 * device will cause all device changes to be abandoned, and intermediate
3146 * data cleaned up.
3148 * If all devices prepare successfully, then the changes are committed
3149 * to all devices.
3151 * All affected nodes must be drained between bdrv_reopen_queue() and
3152 * bdrv_reopen_multiple().
3154 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
3156 int ret = -1;
3157 BlockReopenQueueEntry *bs_entry, *next;
3158 Error *local_err = NULL;
3160 assert(bs_queue != NULL);
3162 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3163 assert(bs_entry->state.bs->quiesce_counter > 0);
3164 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
3165 error_propagate(errp, local_err);
3166 goto cleanup;
3168 bs_entry->prepared = true;
3171 /* If we reach this point, we have success and just need to apply the
3172 * changes
3174 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3175 bdrv_reopen_commit(&bs_entry->state);
3178 ret = 0;
3180 cleanup:
3181 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3182 if (ret) {
3183 if (bs_entry->prepared) {
3184 bdrv_reopen_abort(&bs_entry->state);
3186 qobject_unref(bs_entry->state.explicit_options);
3187 qobject_unref(bs_entry->state.options);
3189 g_free(bs_entry);
3191 g_free(bs_queue);
3193 return ret;
3196 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3197 Error **errp)
3199 int ret;
3200 BlockReopenQueue *queue;
3201 QDict *opts = qdict_new();
3203 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3205 bdrv_subtree_drained_begin(bs);
3206 queue = bdrv_reopen_queue(NULL, bs, opts);
3207 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, errp);
3208 bdrv_subtree_drained_end(bs);
3210 return ret;
3213 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3214 BdrvChild *c)
3216 BlockReopenQueueEntry *entry;
3218 QSIMPLEQ_FOREACH(entry, q, entry) {
3219 BlockDriverState *bs = entry->state.bs;
3220 BdrvChild *child;
3222 QLIST_FOREACH(child, &bs->children, next) {
3223 if (child == c) {
3224 return entry;
3229 return NULL;
3232 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3233 uint64_t *perm, uint64_t *shared)
3235 BdrvChild *c;
3236 BlockReopenQueueEntry *parent;
3237 uint64_t cumulative_perms = 0;
3238 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3240 QLIST_FOREACH(c, &bs->parents, next_parent) {
3241 parent = find_parent_in_reopen_queue(q, c);
3242 if (!parent) {
3243 cumulative_perms |= c->perm;
3244 cumulative_shared_perms &= c->shared_perm;
3245 } else {
3246 uint64_t nperm, nshared;
3248 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3249 parent->state.perm, parent->state.shared_perm,
3250 &nperm, &nshared);
3252 cumulative_perms |= nperm;
3253 cumulative_shared_perms &= nshared;
3256 *perm = cumulative_perms;
3257 *shared = cumulative_shared_perms;
3261 * Prepares a BlockDriverState for reopen. All changes are staged in the
3262 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3263 * the block driver layer .bdrv_reopen_prepare()
3265 * bs is the BlockDriverState to reopen
3266 * flags are the new open flags
3267 * queue is the reopen queue
3269 * Returns 0 on success, non-zero on error. On error errp will be set
3270 * as well.
3272 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3273 * It is the responsibility of the caller to then call the abort() or
3274 * commit() for any other BDS that have been left in a prepare() state
3277 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3278 Error **errp)
3280 int ret = -1;
3281 int old_flags;
3282 Error *local_err = NULL;
3283 BlockDriver *drv;
3284 QemuOpts *opts;
3285 QDict *orig_reopen_opts;
3286 char *discard = NULL;
3287 bool read_only;
3288 bool drv_prepared = false;
3290 assert(reopen_state != NULL);
3291 assert(reopen_state->bs->drv != NULL);
3292 drv = reopen_state->bs->drv;
3294 /* This function and each driver's bdrv_reopen_prepare() remove
3295 * entries from reopen_state->options as they are processed, so
3296 * we need to make a copy of the original QDict. */
3297 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3299 /* Process generic block layer options */
3300 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3301 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3302 if (local_err) {
3303 error_propagate(errp, local_err);
3304 ret = -EINVAL;
3305 goto error;
3308 /* This was already called in bdrv_reopen_queue_child() so the flags
3309 * are up-to-date. This time we simply want to remove the options from
3310 * QemuOpts in order to indicate that they have been processed. */
3311 old_flags = reopen_state->flags;
3312 update_flags_from_options(&reopen_state->flags, opts);
3313 assert(old_flags == reopen_state->flags);
3315 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
3316 if (discard != NULL) {
3317 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3318 error_setg(errp, "Invalid discard option");
3319 ret = -EINVAL;
3320 goto error;
3324 reopen_state->detect_zeroes =
3325 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3326 if (local_err) {
3327 error_propagate(errp, local_err);
3328 ret = -EINVAL;
3329 goto error;
3332 /* All other options (including node-name and driver) must be unchanged.
3333 * Put them back into the QDict, so that they are checked at the end
3334 * of this function. */
3335 qemu_opts_to_qdict(opts, reopen_state->options);
3337 /* If we are to stay read-only, do not allow permission change
3338 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3339 * not set, or if the BDS still has copy_on_read enabled */
3340 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3341 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3342 if (local_err) {
3343 error_propagate(errp, local_err);
3344 goto error;
3347 /* Calculate required permissions after reopening */
3348 bdrv_reopen_perm(queue, reopen_state->bs,
3349 &reopen_state->perm, &reopen_state->shared_perm);
3351 ret = bdrv_flush(reopen_state->bs);
3352 if (ret) {
3353 error_setg_errno(errp, -ret, "Error flushing drive");
3354 goto error;
3357 if (drv->bdrv_reopen_prepare) {
3358 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3359 if (ret) {
3360 if (local_err != NULL) {
3361 error_propagate(errp, local_err);
3362 } else {
3363 bdrv_refresh_filename(reopen_state->bs);
3364 error_setg(errp, "failed while preparing to reopen image '%s'",
3365 reopen_state->bs->filename);
3367 goto error;
3369 } else {
3370 /* It is currently mandatory to have a bdrv_reopen_prepare()
3371 * handler for each supported drv. */
3372 error_setg(errp, "Block format '%s' used by node '%s' "
3373 "does not support reopening files", drv->format_name,
3374 bdrv_get_device_or_node_name(reopen_state->bs));
3375 ret = -1;
3376 goto error;
3379 drv_prepared = true;
3381 /* Options that are not handled are only okay if they are unchanged
3382 * compared to the old state. It is expected that some options are only
3383 * used for the initial open, but not reopen (e.g. filename) */
3384 if (qdict_size(reopen_state->options)) {
3385 const QDictEntry *entry = qdict_first(reopen_state->options);
3387 do {
3388 QObject *new = entry->value;
3389 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3391 /* Allow child references (child_name=node_name) as long as they
3392 * point to the current child (i.e. everything stays the same). */
3393 if (qobject_type(new) == QTYPE_QSTRING) {
3394 BdrvChild *child;
3395 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
3396 if (!strcmp(child->name, entry->key)) {
3397 break;
3401 if (child) {
3402 const char *str = qobject_get_try_str(new);
3403 if (!strcmp(child->bs->node_name, str)) {
3404 continue; /* Found child with this name, skip option */
3410 * TODO: When using -drive to specify blockdev options, all values
3411 * will be strings; however, when using -blockdev, blockdev-add or
3412 * filenames using the json:{} pseudo-protocol, they will be
3413 * correctly typed.
3414 * In contrast, reopening options are (currently) always strings
3415 * (because you can only specify them through qemu-io; all other
3416 * callers do not specify any options).
3417 * Therefore, when using anything other than -drive to create a BDS,
3418 * this cannot detect non-string options as unchanged, because
3419 * qobject_is_equal() always returns false for objects of different
3420 * type. In the future, this should be remedied by correctly typing
3421 * all options. For now, this is not too big of an issue because
3422 * the user can simply omit options which cannot be changed anyway,
3423 * so they will stay unchanged.
3425 if (!qobject_is_equal(new, old)) {
3426 error_setg(errp, "Cannot change the option '%s'", entry->key);
3427 ret = -EINVAL;
3428 goto error;
3430 } while ((entry = qdict_next(reopen_state->options, entry)));
3433 ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3434 reopen_state->shared_perm, NULL, errp);
3435 if (ret < 0) {
3436 goto error;
3439 ret = 0;
3441 /* Restore the original reopen_state->options QDict */
3442 qobject_unref(reopen_state->options);
3443 reopen_state->options = qobject_ref(orig_reopen_opts);
3445 error:
3446 if (ret < 0 && drv_prepared) {
3447 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3448 * call drv->bdrv_reopen_abort() before signaling an error
3449 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3450 * when the respective bdrv_reopen_prepare() has failed) */
3451 if (drv->bdrv_reopen_abort) {
3452 drv->bdrv_reopen_abort(reopen_state);
3455 qemu_opts_del(opts);
3456 qobject_unref(orig_reopen_opts);
3457 g_free(discard);
3458 return ret;
3462 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3463 * makes them final by swapping the staging BlockDriverState contents into
3464 * the active BlockDriverState contents.
3466 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3468 BlockDriver *drv;
3469 BlockDriverState *bs;
3470 BdrvChild *child;
3471 bool old_can_write, new_can_write;
3473 assert(reopen_state != NULL);
3474 bs = reopen_state->bs;
3475 drv = bs->drv;
3476 assert(drv != NULL);
3478 old_can_write =
3479 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3481 /* If there are any driver level actions to take */
3482 if (drv->bdrv_reopen_commit) {
3483 drv->bdrv_reopen_commit(reopen_state);
3486 /* set BDS specific flags now */
3487 qobject_unref(bs->explicit_options);
3488 qobject_unref(bs->options);
3490 bs->explicit_options = reopen_state->explicit_options;
3491 bs->options = reopen_state->options;
3492 bs->open_flags = reopen_state->flags;
3493 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3494 bs->detect_zeroes = reopen_state->detect_zeroes;
3496 /* Remove child references from bs->options and bs->explicit_options.
3497 * Child options were already removed in bdrv_reopen_queue_child() */
3498 QLIST_FOREACH(child, &bs->children, next) {
3499 qdict_del(bs->explicit_options, child->name);
3500 qdict_del(bs->options, child->name);
3503 bdrv_refresh_limits(bs, NULL);
3505 bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3506 reopen_state->shared_perm);
3508 new_can_write =
3509 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3510 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3511 Error *local_err = NULL;
3512 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3513 /* This is not fatal, bitmaps just left read-only, so all following
3514 * writes will fail. User can remove read-only bitmaps to unblock
3515 * writes.
3517 error_reportf_err(local_err,
3518 "%s: Failed to make dirty bitmaps writable: ",
3519 bdrv_get_node_name(bs));
3525 * Abort the reopen, and delete and free the staged changes in
3526 * reopen_state
3528 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3530 BlockDriver *drv;
3532 assert(reopen_state != NULL);
3533 drv = reopen_state->bs->drv;
3534 assert(drv != NULL);
3536 if (drv->bdrv_reopen_abort) {
3537 drv->bdrv_reopen_abort(reopen_state);
3540 bdrv_abort_perm_update(reopen_state->bs);
3544 static void bdrv_close(BlockDriverState *bs)
3546 BdrvAioNotifier *ban, *ban_next;
3547 BdrvChild *child, *next;
3549 assert(!bs->job);
3550 assert(!bs->refcnt);
3552 bdrv_drained_begin(bs); /* complete I/O */
3553 bdrv_flush(bs);
3554 bdrv_drain(bs); /* in case flush left pending I/O */
3556 if (bs->drv) {
3557 if (bs->drv->bdrv_close) {
3558 bs->drv->bdrv_close(bs);
3560 bs->drv = NULL;
3563 bdrv_set_backing_hd(bs, NULL, &error_abort);
3565 if (bs->file != NULL) {
3566 bdrv_unref_child(bs, bs->file);
3567 bs->file = NULL;
3570 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3571 /* TODO Remove bdrv_unref() from drivers' close function and use
3572 * bdrv_unref_child() here */
3573 if (child->bs->inherits_from == bs) {
3574 child->bs->inherits_from = NULL;
3576 bdrv_detach_child(child);
3579 g_free(bs->opaque);
3580 bs->opaque = NULL;
3581 atomic_set(&bs->copy_on_read, 0);
3582 bs->backing_file[0] = '\0';
3583 bs->backing_format[0] = '\0';
3584 bs->total_sectors = 0;
3585 bs->encrypted = false;
3586 bs->sg = false;
3587 qobject_unref(bs->options);
3588 qobject_unref(bs->explicit_options);
3589 bs->options = NULL;
3590 bs->explicit_options = NULL;
3591 qobject_unref(bs->full_open_options);
3592 bs->full_open_options = NULL;
3594 bdrv_release_named_dirty_bitmaps(bs);
3595 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3597 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3598 g_free(ban);
3600 QLIST_INIT(&bs->aio_notifiers);
3601 bdrv_drained_end(bs);
3604 void bdrv_close_all(void)
3606 assert(job_next(NULL) == NULL);
3607 nbd_export_close_all();
3609 /* Drop references from requests still in flight, such as canceled block
3610 * jobs whose AIO context has not been polled yet */
3611 bdrv_drain_all();
3613 blk_remove_all_bs();
3614 blockdev_close_all_bdrv_states();
3616 assert(QTAILQ_EMPTY(&all_bdrv_states));
3619 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3621 GQueue *queue;
3622 GHashTable *found;
3623 bool ret;
3625 if (c->role->stay_at_node) {
3626 return false;
3629 /* If the child @c belongs to the BDS @to, replacing the current
3630 * c->bs by @to would mean to create a loop.
3632 * Such a case occurs when appending a BDS to a backing chain.
3633 * For instance, imagine the following chain:
3635 * guest device -> node A -> further backing chain...
3637 * Now we create a new BDS B which we want to put on top of this
3638 * chain, so we first attach A as its backing node:
3640 * node B
3643 * guest device -> node A -> further backing chain...
3645 * Finally we want to replace A by B. When doing that, we want to
3646 * replace all pointers to A by pointers to B -- except for the
3647 * pointer from B because (1) that would create a loop, and (2)
3648 * that pointer should simply stay intact:
3650 * guest device -> node B
3653 * node A -> further backing chain...
3655 * In general, when replacing a node A (c->bs) by a node B (@to),
3656 * if A is a child of B, that means we cannot replace A by B there
3657 * because that would create a loop. Silently detaching A from B
3658 * is also not really an option. So overall just leaving A in
3659 * place there is the most sensible choice.
3661 * We would also create a loop in any cases where @c is only
3662 * indirectly referenced by @to. Prevent this by returning false
3663 * if @c is found (by breadth-first search) anywhere in the whole
3664 * subtree of @to.
3667 ret = true;
3668 found = g_hash_table_new(NULL, NULL);
3669 g_hash_table_add(found, to);
3670 queue = g_queue_new();
3671 g_queue_push_tail(queue, to);
3673 while (!g_queue_is_empty(queue)) {
3674 BlockDriverState *v = g_queue_pop_head(queue);
3675 BdrvChild *c2;
3677 QLIST_FOREACH(c2, &v->children, next) {
3678 if (c2 == c) {
3679 ret = false;
3680 break;
3683 if (g_hash_table_contains(found, c2->bs)) {
3684 continue;
3687 g_queue_push_tail(queue, c2->bs);
3688 g_hash_table_add(found, c2->bs);
3692 g_queue_free(queue);
3693 g_hash_table_destroy(found);
3695 return ret;
3698 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3699 Error **errp)
3701 BdrvChild *c, *next;
3702 GSList *list = NULL, *p;
3703 uint64_t old_perm, old_shared;
3704 uint64_t perm = 0, shared = BLK_PERM_ALL;
3705 int ret;
3707 assert(!atomic_read(&from->in_flight));
3708 assert(!atomic_read(&to->in_flight));
3710 /* Make sure that @from doesn't go away until we have successfully attached
3711 * all of its parents to @to. */
3712 bdrv_ref(from);
3714 /* Put all parents into @list and calculate their cumulative permissions */
3715 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3716 assert(c->bs == from);
3717 if (!should_update_child(c, to)) {
3718 continue;
3720 list = g_slist_prepend(list, c);
3721 perm |= c->perm;
3722 shared &= c->shared_perm;
3725 /* Check whether the required permissions can be granted on @to, ignoring
3726 * all BdrvChild in @list so that they can't block themselves. */
3727 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3728 if (ret < 0) {
3729 bdrv_abort_perm_update(to);
3730 goto out;
3733 /* Now actually perform the change. We performed the permission check for
3734 * all elements of @list at once, so set the permissions all at once at the
3735 * very end. */
3736 for (p = list; p != NULL; p = p->next) {
3737 c = p->data;
3739 bdrv_ref(to);
3740 bdrv_replace_child_noperm(c, to);
3741 bdrv_unref(from);
3744 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3745 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3747 out:
3748 g_slist_free(list);
3749 bdrv_unref(from);
3753 * Add new bs contents at the top of an image chain while the chain is
3754 * live, while keeping required fields on the top layer.
3756 * This will modify the BlockDriverState fields, and swap contents
3757 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3759 * bs_new must not be attached to a BlockBackend.
3761 * This function does not create any image files.
3763 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3764 * that's what the callers commonly need. bs_new will be referenced by the old
3765 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3766 * reference of its own, it must call bdrv_ref().
3768 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3769 Error **errp)
3771 Error *local_err = NULL;
3773 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3774 if (local_err) {
3775 error_propagate(errp, local_err);
3776 goto out;
3779 bdrv_replace_node(bs_top, bs_new, &local_err);
3780 if (local_err) {
3781 error_propagate(errp, local_err);
3782 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3783 goto out;
3786 /* bs_new is now referenced by its new parents, we don't need the
3787 * additional reference any more. */
3788 out:
3789 bdrv_unref(bs_new);
3792 static void bdrv_delete(BlockDriverState *bs)
3794 assert(!bs->job);
3795 assert(bdrv_op_blocker_is_empty(bs));
3796 assert(!bs->refcnt);
3798 bdrv_close(bs);
3800 /* remove from list, if necessary */
3801 if (bs->node_name[0] != '\0') {
3802 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3804 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3806 g_free(bs);
3810 * Run consistency checks on an image
3812 * Returns 0 if the check could be completed (it doesn't mean that the image is
3813 * free of errors) or -errno when an internal error occurred. The results of the
3814 * check are stored in res.
3816 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3817 BdrvCheckResult *res, BdrvCheckMode fix)
3819 if (bs->drv == NULL) {
3820 return -ENOMEDIUM;
3822 if (bs->drv->bdrv_co_check == NULL) {
3823 return -ENOTSUP;
3826 memset(res, 0, sizeof(*res));
3827 return bs->drv->bdrv_co_check(bs, res, fix);
3830 typedef struct CheckCo {
3831 BlockDriverState *bs;
3832 BdrvCheckResult *res;
3833 BdrvCheckMode fix;
3834 int ret;
3835 } CheckCo;
3837 static void bdrv_check_co_entry(void *opaque)
3839 CheckCo *cco = opaque;
3840 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3841 aio_wait_kick();
3844 int bdrv_check(BlockDriverState *bs,
3845 BdrvCheckResult *res, BdrvCheckMode fix)
3847 Coroutine *co;
3848 CheckCo cco = {
3849 .bs = bs,
3850 .res = res,
3851 .ret = -EINPROGRESS,
3852 .fix = fix,
3855 if (qemu_in_coroutine()) {
3856 /* Fast-path if already in coroutine context */
3857 bdrv_check_co_entry(&cco);
3858 } else {
3859 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3860 bdrv_coroutine_enter(bs, co);
3861 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3864 return cco.ret;
3868 * Return values:
3869 * 0 - success
3870 * -EINVAL - backing format specified, but no file
3871 * -ENOSPC - can't update the backing file because no space is left in the
3872 * image file header
3873 * -ENOTSUP - format driver doesn't support changing the backing file
3875 int bdrv_change_backing_file(BlockDriverState *bs,
3876 const char *backing_file, const char *backing_fmt)
3878 BlockDriver *drv = bs->drv;
3879 int ret;
3881 if (!drv) {
3882 return -ENOMEDIUM;
3885 /* Backing file format doesn't make sense without a backing file */
3886 if (backing_fmt && !backing_file) {
3887 return -EINVAL;
3890 if (drv->bdrv_change_backing_file != NULL) {
3891 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3892 } else {
3893 ret = -ENOTSUP;
3896 if (ret == 0) {
3897 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3898 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3899 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3900 backing_file ?: "");
3902 return ret;
3906 * Finds the image layer in the chain that has 'bs' as its backing file.
3908 * active is the current topmost image.
3910 * Returns NULL if bs is not found in active's image chain,
3911 * or if active == bs.
3913 * Returns the bottommost base image if bs == NULL.
3915 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3916 BlockDriverState *bs)
3918 while (active && bs != backing_bs(active)) {
3919 active = backing_bs(active);
3922 return active;
3925 /* Given a BDS, searches for the base layer. */
3926 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3928 return bdrv_find_overlay(bs, NULL);
3932 * Drops images above 'base' up to and including 'top', and sets the image
3933 * above 'top' to have base as its backing file.
3935 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3936 * information in 'bs' can be properly updated.
3938 * E.g., this will convert the following chain:
3939 * bottom <- base <- intermediate <- top <- active
3941 * to
3943 * bottom <- base <- active
3945 * It is allowed for bottom==base, in which case it converts:
3947 * base <- intermediate <- top <- active
3949 * to
3951 * base <- active
3953 * If backing_file_str is non-NULL, it will be used when modifying top's
3954 * overlay image metadata.
3956 * Error conditions:
3957 * if active == top, that is considered an error
3960 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3961 const char *backing_file_str)
3963 BlockDriverState *explicit_top = top;
3964 bool update_inherits_from;
3965 BdrvChild *c, *next;
3966 Error *local_err = NULL;
3967 int ret = -EIO;
3969 bdrv_ref(top);
3971 if (!top->drv || !base->drv) {
3972 goto exit;
3975 /* Make sure that base is in the backing chain of top */
3976 if (!bdrv_chain_contains(top, base)) {
3977 goto exit;
3980 /* If 'base' recursively inherits from 'top' then we should set
3981 * base->inherits_from to top->inherits_from after 'top' and all
3982 * other intermediate nodes have been dropped.
3983 * If 'top' is an implicit node (e.g. "commit_top") we should skip
3984 * it because no one inherits from it. We use explicit_top for that. */
3985 while (explicit_top && explicit_top->implicit) {
3986 explicit_top = backing_bs(explicit_top);
3988 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
3990 /* success - we can delete the intermediate states, and link top->base */
3991 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3992 * we've figured out how they should work. */
3993 if (!backing_file_str) {
3994 bdrv_refresh_filename(base);
3995 backing_file_str = base->filename;
3998 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3999 /* Check whether we are allowed to switch c from top to base */
4000 GSList *ignore_children = g_slist_prepend(NULL, c);
4001 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
4002 ignore_children, &local_err);
4003 g_slist_free(ignore_children);
4004 if (local_err) {
4005 ret = -EPERM;
4006 error_report_err(local_err);
4007 goto exit;
4010 /* If so, update the backing file path in the image file */
4011 if (c->role->update_filename) {
4012 ret = c->role->update_filename(c, base, backing_file_str,
4013 &local_err);
4014 if (ret < 0) {
4015 bdrv_abort_perm_update(base);
4016 error_report_err(local_err);
4017 goto exit;
4021 /* Do the actual switch in the in-memory graph.
4022 * Completes bdrv_check_update_perm() transaction internally. */
4023 bdrv_ref(base);
4024 bdrv_replace_child(c, base);
4025 bdrv_unref(top);
4028 if (update_inherits_from) {
4029 base->inherits_from = explicit_top->inherits_from;
4032 ret = 0;
4033 exit:
4034 bdrv_unref(top);
4035 return ret;
4039 * Length of a allocated file in bytes. Sparse files are counted by actual
4040 * allocated space. Return < 0 if error or unknown.
4042 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
4044 BlockDriver *drv = bs->drv;
4045 if (!drv) {
4046 return -ENOMEDIUM;
4048 if (drv->bdrv_get_allocated_file_size) {
4049 return drv->bdrv_get_allocated_file_size(bs);
4051 if (bs->file) {
4052 return bdrv_get_allocated_file_size(bs->file->bs);
4054 return -ENOTSUP;
4058 * bdrv_measure:
4059 * @drv: Format driver
4060 * @opts: Creation options for new image
4061 * @in_bs: Existing image containing data for new image (may be NULL)
4062 * @errp: Error object
4063 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4064 * or NULL on error
4066 * Calculate file size required to create a new image.
4068 * If @in_bs is given then space for allocated clusters and zero clusters
4069 * from that image are included in the calculation. If @opts contains a
4070 * backing file that is shared by @in_bs then backing clusters may be omitted
4071 * from the calculation.
4073 * If @in_bs is NULL then the calculation includes no allocated clusters
4074 * unless a preallocation option is given in @opts.
4076 * Note that @in_bs may use a different BlockDriver from @drv.
4078 * If an error occurs the @errp pointer is set.
4080 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
4081 BlockDriverState *in_bs, Error **errp)
4083 if (!drv->bdrv_measure) {
4084 error_setg(errp, "Block driver '%s' does not support size measurement",
4085 drv->format_name);
4086 return NULL;
4089 return drv->bdrv_measure(opts, in_bs, errp);
4093 * Return number of sectors on success, -errno on error.
4095 int64_t bdrv_nb_sectors(BlockDriverState *bs)
4097 BlockDriver *drv = bs->drv;
4099 if (!drv)
4100 return -ENOMEDIUM;
4102 if (drv->has_variable_length) {
4103 int ret = refresh_total_sectors(bs, bs->total_sectors);
4104 if (ret < 0) {
4105 return ret;
4108 return bs->total_sectors;
4112 * Return length in bytes on success, -errno on error.
4113 * The length is always a multiple of BDRV_SECTOR_SIZE.
4115 int64_t bdrv_getlength(BlockDriverState *bs)
4117 int64_t ret = bdrv_nb_sectors(bs);
4119 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
4120 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
4123 /* return 0 as number of sectors if no device present or error */
4124 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
4126 int64_t nb_sectors = bdrv_nb_sectors(bs);
4128 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
4131 bool bdrv_is_sg(BlockDriverState *bs)
4133 return bs->sg;
4136 bool bdrv_is_encrypted(BlockDriverState *bs)
4138 if (bs->backing && bs->backing->bs->encrypted) {
4139 return true;
4141 return bs->encrypted;
4144 const char *bdrv_get_format_name(BlockDriverState *bs)
4146 return bs->drv ? bs->drv->format_name : NULL;
4149 static int qsort_strcmp(const void *a, const void *b)
4151 return strcmp(*(char *const *)a, *(char *const *)b);
4154 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
4155 void *opaque, bool read_only)
4157 BlockDriver *drv;
4158 int count = 0;
4159 int i;
4160 const char **formats = NULL;
4162 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4163 if (drv->format_name) {
4164 bool found = false;
4165 int i = count;
4167 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
4168 continue;
4171 while (formats && i && !found) {
4172 found = !strcmp(formats[--i], drv->format_name);
4175 if (!found) {
4176 formats = g_renew(const char *, formats, count + 1);
4177 formats[count++] = drv->format_name;
4182 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4183 const char *format_name = block_driver_modules[i].format_name;
4185 if (format_name) {
4186 bool found = false;
4187 int j = count;
4189 if (use_bdrv_whitelist &&
4190 !bdrv_format_is_whitelisted(format_name, read_only)) {
4191 continue;
4194 while (formats && j && !found) {
4195 found = !strcmp(formats[--j], format_name);
4198 if (!found) {
4199 formats = g_renew(const char *, formats, count + 1);
4200 formats[count++] = format_name;
4205 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
4207 for (i = 0; i < count; i++) {
4208 it(opaque, formats[i]);
4211 g_free(formats);
4214 /* This function is to find a node in the bs graph */
4215 BlockDriverState *bdrv_find_node(const char *node_name)
4217 BlockDriverState *bs;
4219 assert(node_name);
4221 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4222 if (!strcmp(node_name, bs->node_name)) {
4223 return bs;
4226 return NULL;
4229 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4230 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
4232 BlockDeviceInfoList *list, *entry;
4233 BlockDriverState *bs;
4235 list = NULL;
4236 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4237 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
4238 if (!info) {
4239 qapi_free_BlockDeviceInfoList(list);
4240 return NULL;
4242 entry = g_malloc0(sizeof(*entry));
4243 entry->value = info;
4244 entry->next = list;
4245 list = entry;
4248 return list;
4251 #define QAPI_LIST_ADD(list, element) do { \
4252 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4253 _tmp->value = (element); \
4254 _tmp->next = (list); \
4255 (list) = _tmp; \
4256 } while (0)
4258 typedef struct XDbgBlockGraphConstructor {
4259 XDbgBlockGraph *graph;
4260 GHashTable *graph_nodes;
4261 } XDbgBlockGraphConstructor;
4263 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
4265 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
4267 gr->graph = g_new0(XDbgBlockGraph, 1);
4268 gr->graph_nodes = g_hash_table_new(NULL, NULL);
4270 return gr;
4273 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
4275 XDbgBlockGraph *graph = gr->graph;
4277 g_hash_table_destroy(gr->graph_nodes);
4278 g_free(gr);
4280 return graph;
4283 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
4285 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
4287 if (ret != 0) {
4288 return ret;
4292 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4293 * answer of g_hash_table_lookup.
4295 ret = g_hash_table_size(gr->graph_nodes) + 1;
4296 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
4298 return ret;
4301 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
4302 XDbgBlockGraphNodeType type, const char *name)
4304 XDbgBlockGraphNode *n;
4306 n = g_new0(XDbgBlockGraphNode, 1);
4308 n->id = xdbg_graph_node_num(gr, node);
4309 n->type = type;
4310 n->name = g_strdup(name);
4312 QAPI_LIST_ADD(gr->graph->nodes, n);
4315 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
4316 const BdrvChild *child)
4318 typedef struct {
4319 unsigned int flag;
4320 BlockPermission num;
4321 } PermissionMap;
4323 static const PermissionMap permissions[] = {
4324 { BLK_PERM_CONSISTENT_READ, BLOCK_PERMISSION_CONSISTENT_READ },
4325 { BLK_PERM_WRITE, BLOCK_PERMISSION_WRITE },
4326 { BLK_PERM_WRITE_UNCHANGED, BLOCK_PERMISSION_WRITE_UNCHANGED },
4327 { BLK_PERM_RESIZE, BLOCK_PERMISSION_RESIZE },
4328 { BLK_PERM_GRAPH_MOD, BLOCK_PERMISSION_GRAPH_MOD },
4329 { 0, 0 }
4331 const PermissionMap *p;
4332 XDbgBlockGraphEdge *edge;
4334 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions) - 1) != BLK_PERM_ALL + 1);
4336 edge = g_new0(XDbgBlockGraphEdge, 1);
4338 edge->parent = xdbg_graph_node_num(gr, parent);
4339 edge->child = xdbg_graph_node_num(gr, child->bs);
4340 edge->name = g_strdup(child->name);
4342 for (p = permissions; p->flag; p++) {
4343 if (p->flag & child->perm) {
4344 QAPI_LIST_ADD(edge->perm, p->num);
4346 if (p->flag & child->shared_perm) {
4347 QAPI_LIST_ADD(edge->shared_perm, p->num);
4351 QAPI_LIST_ADD(gr->graph->edges, edge);
4355 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
4357 BlockBackend *blk;
4358 BlockJob *job;
4359 BlockDriverState *bs;
4360 BdrvChild *child;
4361 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
4363 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
4364 char *allocated_name = NULL;
4365 const char *name = blk_name(blk);
4367 if (!*name) {
4368 name = allocated_name = blk_get_attached_dev_id(blk);
4370 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
4371 name);
4372 g_free(allocated_name);
4373 if (blk_root(blk)) {
4374 xdbg_graph_add_edge(gr, blk, blk_root(blk));
4378 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4379 GSList *el;
4381 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
4382 job->job.id);
4383 for (el = job->nodes; el; el = el->next) {
4384 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
4388 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4389 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
4390 bs->node_name);
4391 QLIST_FOREACH(child, &bs->children, next) {
4392 xdbg_graph_add_edge(gr, bs, child);
4396 return xdbg_graph_finalize(gr);
4399 BlockDriverState *bdrv_lookup_bs(const char *device,
4400 const char *node_name,
4401 Error **errp)
4403 BlockBackend *blk;
4404 BlockDriverState *bs;
4406 if (device) {
4407 blk = blk_by_name(device);
4409 if (blk) {
4410 bs = blk_bs(blk);
4411 if (!bs) {
4412 error_setg(errp, "Device '%s' has no medium", device);
4415 return bs;
4419 if (node_name) {
4420 bs = bdrv_find_node(node_name);
4422 if (bs) {
4423 return bs;
4427 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4428 device ? device : "",
4429 node_name ? node_name : "");
4430 return NULL;
4433 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4434 * return false. If either argument is NULL, return false. */
4435 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4437 while (top && top != base) {
4438 top = backing_bs(top);
4441 return top != NULL;
4444 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4446 if (!bs) {
4447 return QTAILQ_FIRST(&graph_bdrv_states);
4449 return QTAILQ_NEXT(bs, node_list);
4452 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
4454 if (!bs) {
4455 return QTAILQ_FIRST(&all_bdrv_states);
4457 return QTAILQ_NEXT(bs, bs_list);
4460 const char *bdrv_get_node_name(const BlockDriverState *bs)
4462 return bs->node_name;
4465 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4467 BdrvChild *c;
4468 const char *name;
4470 /* If multiple parents have a name, just pick the first one. */
4471 QLIST_FOREACH(c, &bs->parents, next_parent) {
4472 if (c->role->get_name) {
4473 name = c->role->get_name(c);
4474 if (name && *name) {
4475 return name;
4480 return NULL;
4483 /* TODO check what callers really want: bs->node_name or blk_name() */
4484 const char *bdrv_get_device_name(const BlockDriverState *bs)
4486 return bdrv_get_parent_name(bs) ?: "";
4489 /* This can be used to identify nodes that might not have a device
4490 * name associated. Since node and device names live in the same
4491 * namespace, the result is unambiguous. The exception is if both are
4492 * absent, then this returns an empty (non-null) string. */
4493 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4495 return bdrv_get_parent_name(bs) ?: bs->node_name;
4498 int bdrv_get_flags(BlockDriverState *bs)
4500 return bs->open_flags;
4503 int bdrv_has_zero_init_1(BlockDriverState *bs)
4505 return 1;
4508 int bdrv_has_zero_init(BlockDriverState *bs)
4510 if (!bs->drv) {
4511 return 0;
4514 /* If BS is a copy on write image, it is initialized to
4515 the contents of the base image, which may not be zeroes. */
4516 if (bs->backing) {
4517 return 0;
4519 if (bs->drv->bdrv_has_zero_init) {
4520 return bs->drv->bdrv_has_zero_init(bs);
4522 if (bs->file && bs->drv->is_filter) {
4523 return bdrv_has_zero_init(bs->file->bs);
4526 /* safe default */
4527 return 0;
4530 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4532 BlockDriverInfo bdi;
4534 if (bs->backing) {
4535 return false;
4538 if (bdrv_get_info(bs, &bdi) == 0) {
4539 return bdi.unallocated_blocks_are_zero;
4542 return false;
4545 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4547 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4548 return false;
4551 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4554 void bdrv_get_backing_filename(BlockDriverState *bs,
4555 char *filename, int filename_size)
4557 pstrcpy(filename, filename_size, bs->backing_file);
4560 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4562 BlockDriver *drv = bs->drv;
4563 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4564 if (!drv) {
4565 return -ENOMEDIUM;
4567 if (!drv->bdrv_get_info) {
4568 if (bs->file && drv->is_filter) {
4569 return bdrv_get_info(bs->file->bs, bdi);
4571 return -ENOTSUP;
4573 memset(bdi, 0, sizeof(*bdi));
4574 return drv->bdrv_get_info(bs, bdi);
4577 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
4578 Error **errp)
4580 BlockDriver *drv = bs->drv;
4581 if (drv && drv->bdrv_get_specific_info) {
4582 return drv->bdrv_get_specific_info(bs, errp);
4584 return NULL;
4587 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4589 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4590 return;
4593 bs->drv->bdrv_debug_event(bs, event);
4596 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4597 const char *tag)
4599 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4600 bs = bs->file ? bs->file->bs : NULL;
4603 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4604 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4607 return -ENOTSUP;
4610 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4612 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4613 bs = bs->file ? bs->file->bs : NULL;
4616 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4617 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4620 return -ENOTSUP;
4623 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4625 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4626 bs = bs->file ? bs->file->bs : NULL;
4629 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4630 return bs->drv->bdrv_debug_resume(bs, tag);
4633 return -ENOTSUP;
4636 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4638 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4639 bs = bs->file ? bs->file->bs : NULL;
4642 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4643 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4646 return false;
4649 /* backing_file can either be relative, or absolute, or a protocol. If it is
4650 * relative, it must be relative to the chain. So, passing in bs->filename
4651 * from a BDS as backing_file should not be done, as that may be relative to
4652 * the CWD rather than the chain. */
4653 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4654 const char *backing_file)
4656 char *filename_full = NULL;
4657 char *backing_file_full = NULL;
4658 char *filename_tmp = NULL;
4659 int is_protocol = 0;
4660 BlockDriverState *curr_bs = NULL;
4661 BlockDriverState *retval = NULL;
4663 if (!bs || !bs->drv || !backing_file) {
4664 return NULL;
4667 filename_full = g_malloc(PATH_MAX);
4668 backing_file_full = g_malloc(PATH_MAX);
4670 is_protocol = path_has_protocol(backing_file);
4672 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4674 /* If either of the filename paths is actually a protocol, then
4675 * compare unmodified paths; otherwise make paths relative */
4676 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4677 char *backing_file_full_ret;
4679 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4680 retval = curr_bs->backing->bs;
4681 break;
4683 /* Also check against the full backing filename for the image */
4684 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
4685 NULL);
4686 if (backing_file_full_ret) {
4687 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
4688 g_free(backing_file_full_ret);
4689 if (equal) {
4690 retval = curr_bs->backing->bs;
4691 break;
4694 } else {
4695 /* If not an absolute filename path, make it relative to the current
4696 * image's filename path */
4697 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
4698 NULL);
4699 /* We are going to compare canonicalized absolute pathnames */
4700 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
4701 g_free(filename_tmp);
4702 continue;
4704 g_free(filename_tmp);
4706 /* We need to make sure the backing filename we are comparing against
4707 * is relative to the current image filename (or absolute) */
4708 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
4709 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
4710 g_free(filename_tmp);
4711 continue;
4713 g_free(filename_tmp);
4715 if (strcmp(backing_file_full, filename_full) == 0) {
4716 retval = curr_bs->backing->bs;
4717 break;
4722 g_free(filename_full);
4723 g_free(backing_file_full);
4724 return retval;
4727 void bdrv_init(void)
4729 module_call_init(MODULE_INIT_BLOCK);
4732 void bdrv_init_with_whitelist(void)
4734 use_bdrv_whitelist = 1;
4735 bdrv_init();
4738 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4739 Error **errp)
4741 BdrvChild *child, *parent;
4742 uint64_t perm, shared_perm;
4743 Error *local_err = NULL;
4744 int ret;
4745 BdrvDirtyBitmap *bm;
4747 if (!bs->drv) {
4748 return;
4751 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4752 return;
4755 QLIST_FOREACH(child, &bs->children, next) {
4756 bdrv_co_invalidate_cache(child->bs, &local_err);
4757 if (local_err) {
4758 error_propagate(errp, local_err);
4759 return;
4764 * Update permissions, they may differ for inactive nodes.
4766 * Note that the required permissions of inactive images are always a
4767 * subset of the permissions required after activating the image. This
4768 * allows us to just get the permissions upfront without restricting
4769 * drv->bdrv_invalidate_cache().
4771 * It also means that in error cases, we don't have to try and revert to
4772 * the old permissions (which is an operation that could fail, too). We can
4773 * just keep the extended permissions for the next time that an activation
4774 * of the image is tried.
4776 bs->open_flags &= ~BDRV_O_INACTIVE;
4777 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4778 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4779 if (ret < 0) {
4780 bs->open_flags |= BDRV_O_INACTIVE;
4781 error_propagate(errp, local_err);
4782 return;
4784 bdrv_set_perm(bs, perm, shared_perm);
4786 if (bs->drv->bdrv_co_invalidate_cache) {
4787 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4788 if (local_err) {
4789 bs->open_flags |= BDRV_O_INACTIVE;
4790 error_propagate(errp, local_err);
4791 return;
4795 for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm;
4796 bm = bdrv_dirty_bitmap_next(bs, bm))
4798 bdrv_dirty_bitmap_set_migration(bm, false);
4801 ret = refresh_total_sectors(bs, bs->total_sectors);
4802 if (ret < 0) {
4803 bs->open_flags |= BDRV_O_INACTIVE;
4804 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4805 return;
4808 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4809 if (parent->role->activate) {
4810 parent->role->activate(parent, &local_err);
4811 if (local_err) {
4812 bs->open_flags |= BDRV_O_INACTIVE;
4813 error_propagate(errp, local_err);
4814 return;
4820 typedef struct InvalidateCacheCo {
4821 BlockDriverState *bs;
4822 Error **errp;
4823 bool done;
4824 } InvalidateCacheCo;
4826 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4828 InvalidateCacheCo *ico = opaque;
4829 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4830 ico->done = true;
4831 aio_wait_kick();
4834 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4836 Coroutine *co;
4837 InvalidateCacheCo ico = {
4838 .bs = bs,
4839 .done = false,
4840 .errp = errp
4843 if (qemu_in_coroutine()) {
4844 /* Fast-path if already in coroutine context */
4845 bdrv_invalidate_cache_co_entry(&ico);
4846 } else {
4847 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4848 bdrv_coroutine_enter(bs, co);
4849 BDRV_POLL_WHILE(bs, !ico.done);
4853 void bdrv_invalidate_cache_all(Error **errp)
4855 BlockDriverState *bs;
4856 Error *local_err = NULL;
4857 BdrvNextIterator it;
4859 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4860 AioContext *aio_context = bdrv_get_aio_context(bs);
4862 aio_context_acquire(aio_context);
4863 bdrv_invalidate_cache(bs, &local_err);
4864 aio_context_release(aio_context);
4865 if (local_err) {
4866 error_propagate(errp, local_err);
4867 bdrv_next_cleanup(&it);
4868 return;
4873 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
4875 BdrvChild *parent;
4877 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4878 if (parent->role->parent_is_bds) {
4879 BlockDriverState *parent_bs = parent->opaque;
4880 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
4881 return true;
4886 return false;
4889 static int bdrv_inactivate_recurse(BlockDriverState *bs)
4891 BdrvChild *child, *parent;
4892 uint64_t perm, shared_perm;
4893 int ret;
4895 if (!bs->drv) {
4896 return -ENOMEDIUM;
4899 /* Make sure that we don't inactivate a child before its parent.
4900 * It will be covered by recursion from the yet active parent. */
4901 if (bdrv_has_bds_parent(bs, true)) {
4902 return 0;
4905 assert(!(bs->open_flags & BDRV_O_INACTIVE));
4907 /* Inactivate this node */
4908 if (bs->drv->bdrv_inactivate) {
4909 ret = bs->drv->bdrv_inactivate(bs);
4910 if (ret < 0) {
4911 return ret;
4915 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4916 if (parent->role->inactivate) {
4917 ret = parent->role->inactivate(parent);
4918 if (ret < 0) {
4919 return ret;
4924 bs->open_flags |= BDRV_O_INACTIVE;
4926 /* Update permissions, they may differ for inactive nodes */
4927 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4928 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4929 bdrv_set_perm(bs, perm, shared_perm);
4932 /* Recursively inactivate children */
4933 QLIST_FOREACH(child, &bs->children, next) {
4934 ret = bdrv_inactivate_recurse(child->bs);
4935 if (ret < 0) {
4936 return ret;
4940 return 0;
4943 int bdrv_inactivate_all(void)
4945 BlockDriverState *bs = NULL;
4946 BdrvNextIterator it;
4947 int ret = 0;
4948 GSList *aio_ctxs = NULL, *ctx;
4950 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4951 AioContext *aio_context = bdrv_get_aio_context(bs);
4953 if (!g_slist_find(aio_ctxs, aio_context)) {
4954 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4955 aio_context_acquire(aio_context);
4959 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4960 /* Nodes with BDS parents are covered by recursion from the last
4961 * parent that gets inactivated. Don't inactivate them a second
4962 * time if that has already happened. */
4963 if (bdrv_has_bds_parent(bs, false)) {
4964 continue;
4966 ret = bdrv_inactivate_recurse(bs);
4967 if (ret < 0) {
4968 bdrv_next_cleanup(&it);
4969 goto out;
4973 out:
4974 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4975 AioContext *aio_context = ctx->data;
4976 aio_context_release(aio_context);
4978 g_slist_free(aio_ctxs);
4980 return ret;
4983 /**************************************************************/
4984 /* removable device support */
4987 * Return TRUE if the media is present
4989 bool bdrv_is_inserted(BlockDriverState *bs)
4991 BlockDriver *drv = bs->drv;
4992 BdrvChild *child;
4994 if (!drv) {
4995 return false;
4997 if (drv->bdrv_is_inserted) {
4998 return drv->bdrv_is_inserted(bs);
5000 QLIST_FOREACH(child, &bs->children, next) {
5001 if (!bdrv_is_inserted(child->bs)) {
5002 return false;
5005 return true;
5009 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5011 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5013 BlockDriver *drv = bs->drv;
5015 if (drv && drv->bdrv_eject) {
5016 drv->bdrv_eject(bs, eject_flag);
5021 * Lock or unlock the media (if it is locked, the user won't be able
5022 * to eject it manually).
5024 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5026 BlockDriver *drv = bs->drv;
5028 trace_bdrv_lock_medium(bs, locked);
5030 if (drv && drv->bdrv_lock_medium) {
5031 drv->bdrv_lock_medium(bs, locked);
5035 /* Get a reference to bs */
5036 void bdrv_ref(BlockDriverState *bs)
5038 bs->refcnt++;
5041 /* Release a previously grabbed reference to bs.
5042 * If after releasing, reference count is zero, the BlockDriverState is
5043 * deleted. */
5044 void bdrv_unref(BlockDriverState *bs)
5046 if (!bs) {
5047 return;
5049 assert(bs->refcnt > 0);
5050 if (--bs->refcnt == 0) {
5051 bdrv_delete(bs);
5055 struct BdrvOpBlocker {
5056 Error *reason;
5057 QLIST_ENTRY(BdrvOpBlocker) list;
5060 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5062 BdrvOpBlocker *blocker;
5063 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5064 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5065 blocker = QLIST_FIRST(&bs->op_blockers[op]);
5066 error_propagate_prepend(errp, error_copy(blocker->reason),
5067 "Node '%s' is busy: ",
5068 bdrv_get_device_or_node_name(bs));
5069 return true;
5071 return false;
5074 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5076 BdrvOpBlocker *blocker;
5077 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5079 blocker = g_new0(BdrvOpBlocker, 1);
5080 blocker->reason = reason;
5081 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5084 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5086 BdrvOpBlocker *blocker, *next;
5087 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5088 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5089 if (blocker->reason == reason) {
5090 QLIST_REMOVE(blocker, list);
5091 g_free(blocker);
5096 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5098 int i;
5099 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5100 bdrv_op_block(bs, i, reason);
5104 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5106 int i;
5107 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5108 bdrv_op_unblock(bs, i, reason);
5112 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5114 int i;
5116 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5117 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5118 return false;
5121 return true;
5124 void bdrv_img_create(const char *filename, const char *fmt,
5125 const char *base_filename, const char *base_fmt,
5126 char *options, uint64_t img_size, int flags, bool quiet,
5127 Error **errp)
5129 QemuOptsList *create_opts = NULL;
5130 QemuOpts *opts = NULL;
5131 const char *backing_fmt, *backing_file;
5132 int64_t size;
5133 BlockDriver *drv, *proto_drv;
5134 Error *local_err = NULL;
5135 int ret = 0;
5137 /* Find driver and parse its options */
5138 drv = bdrv_find_format(fmt);
5139 if (!drv) {
5140 error_setg(errp, "Unknown file format '%s'", fmt);
5141 return;
5144 proto_drv = bdrv_find_protocol(filename, true, errp);
5145 if (!proto_drv) {
5146 return;
5149 if (!drv->create_opts) {
5150 error_setg(errp, "Format driver '%s' does not support image creation",
5151 drv->format_name);
5152 return;
5155 if (!proto_drv->create_opts) {
5156 error_setg(errp, "Protocol driver '%s' does not support image creation",
5157 proto_drv->format_name);
5158 return;
5161 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5162 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5164 /* Create parameter list with default values */
5165 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5166 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5168 /* Parse -o options */
5169 if (options) {
5170 qemu_opts_do_parse(opts, options, NULL, &local_err);
5171 if (local_err) {
5172 goto out;
5176 if (base_filename) {
5177 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
5178 if (local_err) {
5179 error_setg(errp, "Backing file not supported for file format '%s'",
5180 fmt);
5181 goto out;
5185 if (base_fmt) {
5186 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
5187 if (local_err) {
5188 error_setg(errp, "Backing file format not supported for file "
5189 "format '%s'", fmt);
5190 goto out;
5194 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5195 if (backing_file) {
5196 if (!strcmp(filename, backing_file)) {
5197 error_setg(errp, "Error: Trying to create an image with the "
5198 "same filename as the backing file");
5199 goto out;
5203 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5205 /* The size for the image must always be specified, unless we have a backing
5206 * file and we have not been forbidden from opening it. */
5207 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
5208 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
5209 BlockDriverState *bs;
5210 char *full_backing;
5211 int back_flags;
5212 QDict *backing_options = NULL;
5214 full_backing =
5215 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
5216 &local_err);
5217 if (local_err) {
5218 goto out;
5220 assert(full_backing);
5222 /* backing files always opened read-only */
5223 back_flags = flags;
5224 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5226 backing_options = qdict_new();
5227 if (backing_fmt) {
5228 qdict_put_str(backing_options, "driver", backing_fmt);
5230 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
5232 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
5233 &local_err);
5234 g_free(full_backing);
5235 if (!bs && size != -1) {
5236 /* Couldn't open BS, but we have a size, so it's nonfatal */
5237 warn_reportf_err(local_err,
5238 "Could not verify backing image. "
5239 "This may become an error in future versions.\n");
5240 local_err = NULL;
5241 } else if (!bs) {
5242 /* Couldn't open bs, do not have size */
5243 error_append_hint(&local_err,
5244 "Could not open backing image to determine size.\n");
5245 goto out;
5246 } else {
5247 if (size == -1) {
5248 /* Opened BS, have no size */
5249 size = bdrv_getlength(bs);
5250 if (size < 0) {
5251 error_setg_errno(errp, -size, "Could not get size of '%s'",
5252 backing_file);
5253 bdrv_unref(bs);
5254 goto out;
5256 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
5258 bdrv_unref(bs);
5260 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5262 if (size == -1) {
5263 error_setg(errp, "Image creation needs a size parameter");
5264 goto out;
5267 if (!quiet) {
5268 printf("Formatting '%s', fmt=%s ", filename, fmt);
5269 qemu_opts_print(opts, " ");
5270 puts("");
5273 ret = bdrv_create(drv, filename, opts, &local_err);
5275 if (ret == -EFBIG) {
5276 /* This is generally a better message than whatever the driver would
5277 * deliver (especially because of the cluster_size_hint), since that
5278 * is most probably not much different from "image too large". */
5279 const char *cluster_size_hint = "";
5280 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5281 cluster_size_hint = " (try using a larger cluster size)";
5283 error_setg(errp, "The image size is too large for file format '%s'"
5284 "%s", fmt, cluster_size_hint);
5285 error_free(local_err);
5286 local_err = NULL;
5289 out:
5290 qemu_opts_del(opts);
5291 qemu_opts_free(create_opts);
5292 error_propagate(errp, local_err);
5295 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5297 return bs ? bs->aio_context : qemu_get_aio_context();
5300 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
5302 aio_co_enter(bdrv_get_aio_context(bs), co);
5305 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
5307 QLIST_REMOVE(ban, list);
5308 g_free(ban);
5311 void bdrv_detach_aio_context(BlockDriverState *bs)
5313 BdrvAioNotifier *baf, *baf_tmp;
5314 BdrvChild *child;
5316 if (!bs->drv) {
5317 return;
5320 assert(!bs->walking_aio_notifiers);
5321 bs->walking_aio_notifiers = true;
5322 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
5323 if (baf->deleted) {
5324 bdrv_do_remove_aio_context_notifier(baf);
5325 } else {
5326 baf->detach_aio_context(baf->opaque);
5329 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5330 * remove remaining aio notifiers if we aren't called again.
5332 bs->walking_aio_notifiers = false;
5334 if (bs->drv->bdrv_detach_aio_context) {
5335 bs->drv->bdrv_detach_aio_context(bs);
5337 QLIST_FOREACH(child, &bs->children, next) {
5338 bdrv_detach_aio_context(child->bs);
5341 if (bs->quiesce_counter) {
5342 aio_enable_external(bs->aio_context);
5344 bs->aio_context = NULL;
5347 void bdrv_attach_aio_context(BlockDriverState *bs,
5348 AioContext *new_context)
5350 BdrvAioNotifier *ban, *ban_tmp;
5351 BdrvChild *child;
5353 if (!bs->drv) {
5354 return;
5357 if (bs->quiesce_counter) {
5358 aio_disable_external(new_context);
5361 bs->aio_context = new_context;
5363 QLIST_FOREACH(child, &bs->children, next) {
5364 bdrv_attach_aio_context(child->bs, new_context);
5366 if (bs->drv->bdrv_attach_aio_context) {
5367 bs->drv->bdrv_attach_aio_context(bs, new_context);
5370 assert(!bs->walking_aio_notifiers);
5371 bs->walking_aio_notifiers = true;
5372 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
5373 if (ban->deleted) {
5374 bdrv_do_remove_aio_context_notifier(ban);
5375 } else {
5376 ban->attached_aio_context(new_context, ban->opaque);
5379 bs->walking_aio_notifiers = false;
5382 /* The caller must own the AioContext lock for the old AioContext of bs, but it
5383 * must not own the AioContext lock for new_context (unless new_context is
5384 * the same as the current context of bs). */
5385 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5387 if (bdrv_get_aio_context(bs) == new_context) {
5388 return;
5391 bdrv_drained_begin(bs);
5392 bdrv_detach_aio_context(bs);
5394 /* This function executes in the old AioContext so acquire the new one in
5395 * case it runs in a different thread.
5397 aio_context_acquire(new_context);
5398 bdrv_attach_aio_context(bs, new_context);
5399 bdrv_drained_end(bs);
5400 aio_context_release(new_context);
5403 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
5404 void (*attached_aio_context)(AioContext *new_context, void *opaque),
5405 void (*detach_aio_context)(void *opaque), void *opaque)
5407 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
5408 *ban = (BdrvAioNotifier){
5409 .attached_aio_context = attached_aio_context,
5410 .detach_aio_context = detach_aio_context,
5411 .opaque = opaque
5414 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
5417 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
5418 void (*attached_aio_context)(AioContext *,
5419 void *),
5420 void (*detach_aio_context)(void *),
5421 void *opaque)
5423 BdrvAioNotifier *ban, *ban_next;
5425 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5426 if (ban->attached_aio_context == attached_aio_context &&
5427 ban->detach_aio_context == detach_aio_context &&
5428 ban->opaque == opaque &&
5429 ban->deleted == false)
5431 if (bs->walking_aio_notifiers) {
5432 ban->deleted = true;
5433 } else {
5434 bdrv_do_remove_aio_context_notifier(ban);
5436 return;
5440 abort();
5443 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
5444 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5445 Error **errp)
5447 if (!bs->drv) {
5448 error_setg(errp, "Node is ejected");
5449 return -ENOMEDIUM;
5451 if (!bs->drv->bdrv_amend_options) {
5452 error_setg(errp, "Block driver '%s' does not support option amendment",
5453 bs->drv->format_name);
5454 return -ENOTSUP;
5456 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
5459 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5460 * of block filter and by bdrv_is_first_non_filter.
5461 * It is used to test if the given bs is the candidate or recurse more in the
5462 * node graph.
5464 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5465 BlockDriverState *candidate)
5467 /* return false if basic checks fails */
5468 if (!bs || !bs->drv) {
5469 return false;
5472 /* the code reached a non block filter driver -> check if the bs is
5473 * the same as the candidate. It's the recursion termination condition.
5475 if (!bs->drv->is_filter) {
5476 return bs == candidate;
5478 /* Down this path the driver is a block filter driver */
5480 /* If the block filter recursion method is defined use it to recurse down
5481 * the node graph.
5483 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5484 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5487 /* the driver is a block filter but don't allow to recurse -> return false
5489 return false;
5492 /* This function checks if the candidate is the first non filter bs down it's
5493 * bs chain. Since we don't have pointers to parents it explore all bs chains
5494 * from the top. Some filters can choose not to pass down the recursion.
5496 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5498 BlockDriverState *bs;
5499 BdrvNextIterator it;
5501 /* walk down the bs forest recursively */
5502 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5503 bool perm;
5505 /* try to recurse in this top level bs */
5506 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5508 /* candidate is the first non filter */
5509 if (perm) {
5510 bdrv_next_cleanup(&it);
5511 return true;
5515 return false;
5518 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5519 const char *node_name, Error **errp)
5521 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5522 AioContext *aio_context;
5524 if (!to_replace_bs) {
5525 error_setg(errp, "Node name '%s' not found", node_name);
5526 return NULL;
5529 aio_context = bdrv_get_aio_context(to_replace_bs);
5530 aio_context_acquire(aio_context);
5532 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5533 to_replace_bs = NULL;
5534 goto out;
5537 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5538 * most non filter in order to prevent data corruption.
5539 * Another benefit is that this tests exclude backing files which are
5540 * blocked by the backing blockers.
5542 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5543 error_setg(errp, "Only top most non filter can be replaced");
5544 to_replace_bs = NULL;
5545 goto out;
5548 out:
5549 aio_context_release(aio_context);
5550 return to_replace_bs;
5554 * Iterates through the list of runtime option keys that are said to
5555 * be "strong" for a BDS. An option is called "strong" if it changes
5556 * a BDS's data. For example, the null block driver's "size" and
5557 * "read-zeroes" options are strong, but its "latency-ns" option is
5558 * not.
5560 * If a key returned by this function ends with a dot, all options
5561 * starting with that prefix are strong.
5563 static const char *const *strong_options(BlockDriverState *bs,
5564 const char *const *curopt)
5566 static const char *const global_options[] = {
5567 "driver", "filename", NULL
5570 if (!curopt) {
5571 return &global_options[0];
5574 curopt++;
5575 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
5576 curopt = bs->drv->strong_runtime_opts;
5579 return (curopt && *curopt) ? curopt : NULL;
5583 * Copies all strong runtime options from bs->options to the given
5584 * QDict. The set of strong option keys is determined by invoking
5585 * strong_options().
5587 * Returns true iff any strong option was present in bs->options (and
5588 * thus copied to the target QDict) with the exception of "filename"
5589 * and "driver". The caller is expected to use this value to decide
5590 * whether the existence of strong options prevents the generation of
5591 * a plain filename.
5593 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
5595 bool found_any = false;
5596 const char *const *option_name = NULL;
5598 if (!bs->drv) {
5599 return false;
5602 while ((option_name = strong_options(bs, option_name))) {
5603 bool option_given = false;
5605 assert(strlen(*option_name) > 0);
5606 if ((*option_name)[strlen(*option_name) - 1] != '.') {
5607 QObject *entry = qdict_get(bs->options, *option_name);
5608 if (!entry) {
5609 continue;
5612 qdict_put_obj(d, *option_name, qobject_ref(entry));
5613 option_given = true;
5614 } else {
5615 const QDictEntry *entry;
5616 for (entry = qdict_first(bs->options); entry;
5617 entry = qdict_next(bs->options, entry))
5619 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
5620 qdict_put_obj(d, qdict_entry_key(entry),
5621 qobject_ref(qdict_entry_value(entry)));
5622 option_given = true;
5627 /* While "driver" and "filename" need to be included in a JSON filename,
5628 * their existence does not prohibit generation of a plain filename. */
5629 if (!found_any && option_given &&
5630 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
5632 found_any = true;
5636 if (!qdict_haskey(d, "driver")) {
5637 /* Drivers created with bdrv_new_open_driver() may not have a
5638 * @driver option. Add it here. */
5639 qdict_put_str(d, "driver", bs->drv->format_name);
5642 return found_any;
5645 /* Note: This function may return false positives; it may return true
5646 * even if opening the backing file specified by bs's image header
5647 * would result in exactly bs->backing. */
5648 static bool bdrv_backing_overridden(BlockDriverState *bs)
5650 if (bs->backing) {
5651 return strcmp(bs->auto_backing_file,
5652 bs->backing->bs->filename);
5653 } else {
5654 /* No backing BDS, so if the image header reports any backing
5655 * file, it must have been suppressed */
5656 return bs->auto_backing_file[0] != '\0';
5660 /* Updates the following BDS fields:
5661 * - exact_filename: A filename which may be used for opening a block device
5662 * which (mostly) equals the given BDS (even without any
5663 * other options; so reading and writing must return the same
5664 * results, but caching etc. may be different)
5665 * - full_open_options: Options which, when given when opening a block device
5666 * (without a filename), result in a BDS (mostly)
5667 * equalling the given one
5668 * - filename: If exact_filename is set, it is copied here. Otherwise,
5669 * full_open_options is converted to a JSON object, prefixed with
5670 * "json:" (for use through the JSON pseudo protocol) and put here.
5672 void bdrv_refresh_filename(BlockDriverState *bs)
5674 BlockDriver *drv = bs->drv;
5675 BdrvChild *child;
5676 QDict *opts;
5677 bool backing_overridden;
5678 bool generate_json_filename; /* Whether our default implementation should
5679 fill exact_filename (false) or not (true) */
5681 if (!drv) {
5682 return;
5685 /* This BDS's file name may depend on any of its children's file names, so
5686 * refresh those first */
5687 QLIST_FOREACH(child, &bs->children, next) {
5688 bdrv_refresh_filename(child->bs);
5691 if (bs->implicit) {
5692 /* For implicit nodes, just copy everything from the single child */
5693 child = QLIST_FIRST(&bs->children);
5694 assert(QLIST_NEXT(child, next) == NULL);
5696 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
5697 child->bs->exact_filename);
5698 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
5700 bs->full_open_options = qobject_ref(child->bs->full_open_options);
5702 return;
5705 backing_overridden = bdrv_backing_overridden(bs);
5707 if (bs->open_flags & BDRV_O_NO_IO) {
5708 /* Without I/O, the backing file does not change anything.
5709 * Therefore, in such a case (primarily qemu-img), we can
5710 * pretend the backing file has not been overridden even if
5711 * it technically has been. */
5712 backing_overridden = false;
5715 /* Gather the options QDict */
5716 opts = qdict_new();
5717 generate_json_filename = append_strong_runtime_options(opts, bs);
5718 generate_json_filename |= backing_overridden;
5720 if (drv->bdrv_gather_child_options) {
5721 /* Some block drivers may not want to present all of their children's
5722 * options, or name them differently from BdrvChild.name */
5723 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
5724 } else {
5725 QLIST_FOREACH(child, &bs->children, next) {
5726 if (child->role == &child_backing && !backing_overridden) {
5727 /* We can skip the backing BDS if it has not been overridden */
5728 continue;
5731 qdict_put(opts, child->name,
5732 qobject_ref(child->bs->full_open_options));
5735 if (backing_overridden && !bs->backing) {
5736 /* Force no backing file */
5737 qdict_put_null(opts, "backing");
5741 qobject_unref(bs->full_open_options);
5742 bs->full_open_options = opts;
5744 if (drv->bdrv_refresh_filename) {
5745 /* Obsolete information is of no use here, so drop the old file name
5746 * information before refreshing it */
5747 bs->exact_filename[0] = '\0';
5749 drv->bdrv_refresh_filename(bs);
5750 } else if (bs->file) {
5751 /* Try to reconstruct valid information from the underlying file */
5753 bs->exact_filename[0] = '\0';
5756 * We can use the underlying file's filename if:
5757 * - it has a filename,
5758 * - the file is a protocol BDS, and
5759 * - opening that file (as this BDS's format) will automatically create
5760 * the BDS tree we have right now, that is:
5761 * - the user did not significantly change this BDS's behavior with
5762 * some explicit (strong) options
5763 * - no non-file child of this BDS has been overridden by the user
5764 * Both of these conditions are represented by generate_json_filename.
5766 if (bs->file->bs->exact_filename[0] &&
5767 bs->file->bs->drv->bdrv_file_open &&
5768 !generate_json_filename)
5770 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5774 if (bs->exact_filename[0]) {
5775 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5776 } else {
5777 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5778 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5779 qstring_get_str(json));
5780 qobject_unref(json);
5784 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
5786 BlockDriver *drv = bs->drv;
5788 if (!drv) {
5789 error_setg(errp, "Node '%s' is ejected", bs->node_name);
5790 return NULL;
5793 if (drv->bdrv_dirname) {
5794 return drv->bdrv_dirname(bs, errp);
5797 if (bs->file) {
5798 return bdrv_dirname(bs->file->bs, errp);
5801 bdrv_refresh_filename(bs);
5802 if (bs->exact_filename[0] != '\0') {
5803 return path_combine(bs->exact_filename, "");
5806 error_setg(errp, "Cannot generate a base directory for %s nodes",
5807 drv->format_name);
5808 return NULL;
5812 * Hot add/remove a BDS's child. So the user can take a child offline when
5813 * it is broken and take a new child online
5815 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5816 Error **errp)
5819 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5820 error_setg(errp, "The node %s does not support adding a child",
5821 bdrv_get_device_or_node_name(parent_bs));
5822 return;
5825 if (!QLIST_EMPTY(&child_bs->parents)) {
5826 error_setg(errp, "The node %s already has a parent",
5827 child_bs->node_name);
5828 return;
5831 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5834 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5836 BdrvChild *tmp;
5838 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5839 error_setg(errp, "The node %s does not support removing a child",
5840 bdrv_get_device_or_node_name(parent_bs));
5841 return;
5844 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5845 if (tmp == child) {
5846 break;
5850 if (!tmp) {
5851 error_setg(errp, "The node %s does not have a child named %s",
5852 bdrv_get_device_or_node_name(parent_bs),
5853 bdrv_get_device_or_node_name(child->bs));
5854 return;
5857 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5860 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5861 uint32_t granularity, Error **errp)
5863 BlockDriver *drv = bs->drv;
5865 if (!drv) {
5866 error_setg_errno(errp, ENOMEDIUM,
5867 "Can't store persistent bitmaps to %s",
5868 bdrv_get_device_or_node_name(bs));
5869 return false;
5872 if (!drv->bdrv_can_store_new_dirty_bitmap) {
5873 error_setg_errno(errp, ENOTSUP,
5874 "Can't store persistent bitmaps to %s",
5875 bdrv_get_device_or_node_name(bs));
5876 return false;
5879 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);