linux-user: Use correct alignment for long long on i386 guests
[qemu/cris-port.git] / block.c
blob30d64e6ca5aba6d4a91f33187f620460c897c304
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
24 #include "qemu/osdep.h"
25 #include "trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "qemu/error-report.h"
29 #include "qemu/module.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qapi/qmp/qbool.h"
32 #include "qapi/qmp/qjson.h"
33 #include "sysemu/block-backend.h"
34 #include "sysemu/sysemu.h"
35 #include "qemu/notify.h"
36 #include "qemu/coroutine.h"
37 #include "block/qapi.h"
38 #include "qmp-commands.h"
39 #include "qemu/timer.h"
40 #include "qapi-event.h"
41 #include "qemu/cutils.h"
42 #include "qemu/id.h"
44 #ifdef CONFIG_BSD
45 #include <sys/ioctl.h>
46 #include <sys/queue.h>
47 #ifndef __DragonFly__
48 #include <sys/disk.h>
49 #endif
50 #endif
52 #ifdef _WIN32
53 #include <windows.h>
54 #endif
56 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
58 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
59 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
61 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
64 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
65 QLIST_HEAD_INITIALIZER(bdrv_drivers);
67 static BlockDriverState *bdrv_open_inherit(const char *filename,
68 const char *reference,
69 QDict *options, int flags,
70 BlockDriverState *parent,
71 const BdrvChildRole *child_role,
72 Error **errp);
74 /* If non-zero, use only whitelisted block drivers */
75 static int use_bdrv_whitelist;
77 #ifdef _WIN32
78 static int is_windows_drive_prefix(const char *filename)
80 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
81 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
82 filename[1] == ':');
85 int is_windows_drive(const char *filename)
87 if (is_windows_drive_prefix(filename) &&
88 filename[2] == '\0')
89 return 1;
90 if (strstart(filename, "\\\\.\\", NULL) ||
91 strstart(filename, "//./", NULL))
92 return 1;
93 return 0;
95 #endif
97 size_t bdrv_opt_mem_align(BlockDriverState *bs)
99 if (!bs || !bs->drv) {
100 /* page size or 4k (hdd sector size) should be on the safe side */
101 return MAX(4096, getpagesize());
104 return bs->bl.opt_mem_alignment;
107 size_t bdrv_min_mem_align(BlockDriverState *bs)
109 if (!bs || !bs->drv) {
110 /* page size or 4k (hdd sector size) should be on the safe side */
111 return MAX(4096, getpagesize());
114 return bs->bl.min_mem_alignment;
117 /* check if the path starts with "<protocol>:" */
118 int path_has_protocol(const char *path)
120 const char *p;
122 #ifdef _WIN32
123 if (is_windows_drive(path) ||
124 is_windows_drive_prefix(path)) {
125 return 0;
127 p = path + strcspn(path, ":/\\");
128 #else
129 p = path + strcspn(path, ":/");
130 #endif
132 return *p == ':';
135 int path_is_absolute(const char *path)
137 #ifdef _WIN32
138 /* specific case for names like: "\\.\d:" */
139 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
140 return 1;
142 return (*path == '/' || *path == '\\');
143 #else
144 return (*path == '/');
145 #endif
148 /* if filename is absolute, just copy it to dest. Otherwise, build a
149 path to it by considering it is relative to base_path. URL are
150 supported. */
151 void path_combine(char *dest, int dest_size,
152 const char *base_path,
153 const char *filename)
155 const char *p, *p1;
156 int len;
158 if (dest_size <= 0)
159 return;
160 if (path_is_absolute(filename)) {
161 pstrcpy(dest, dest_size, filename);
162 } else {
163 p = strchr(base_path, ':');
164 if (p)
165 p++;
166 else
167 p = base_path;
168 p1 = strrchr(base_path, '/');
169 #ifdef _WIN32
171 const char *p2;
172 p2 = strrchr(base_path, '\\');
173 if (!p1 || p2 > p1)
174 p1 = p2;
176 #endif
177 if (p1)
178 p1++;
179 else
180 p1 = base_path;
181 if (p1 > p)
182 p = p1;
183 len = p - base_path;
184 if (len > dest_size - 1)
185 len = dest_size - 1;
186 memcpy(dest, base_path, len);
187 dest[len] = '\0';
188 pstrcat(dest, dest_size, filename);
192 void bdrv_get_full_backing_filename_from_filename(const char *backed,
193 const char *backing,
194 char *dest, size_t sz,
195 Error **errp)
197 if (backing[0] == '\0' || path_has_protocol(backing) ||
198 path_is_absolute(backing))
200 pstrcpy(dest, sz, backing);
201 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
202 error_setg(errp, "Cannot use relative backing file names for '%s'",
203 backed);
204 } else {
205 path_combine(dest, sz, backed, backing);
209 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
210 Error **errp)
212 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
214 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
215 dest, sz, errp);
218 void bdrv_register(BlockDriver *bdrv)
220 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
223 BlockDriverState *bdrv_new(void)
225 BlockDriverState *bs;
226 int i;
228 bs = g_new0(BlockDriverState, 1);
229 QLIST_INIT(&bs->dirty_bitmaps);
230 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
231 QLIST_INIT(&bs->op_blockers[i]);
233 notifier_with_return_list_init(&bs->before_write_notifiers);
234 bs->refcnt = 1;
235 bs->aio_context = qemu_get_aio_context();
237 qemu_co_queue_init(&bs->flush_queue);
239 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
241 return bs;
244 BlockDriver *bdrv_find_format(const char *format_name)
246 BlockDriver *drv1;
247 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
248 if (!strcmp(drv1->format_name, format_name)) {
249 return drv1;
252 return NULL;
255 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
257 static const char *whitelist_rw[] = {
258 CONFIG_BDRV_RW_WHITELIST
260 static const char *whitelist_ro[] = {
261 CONFIG_BDRV_RO_WHITELIST
263 const char **p;
265 if (!whitelist_rw[0] && !whitelist_ro[0]) {
266 return 1; /* no whitelist, anything goes */
269 for (p = whitelist_rw; *p; p++) {
270 if (!strcmp(drv->format_name, *p)) {
271 return 1;
274 if (read_only) {
275 for (p = whitelist_ro; *p; p++) {
276 if (!strcmp(drv->format_name, *p)) {
277 return 1;
281 return 0;
284 bool bdrv_uses_whitelist(void)
286 return use_bdrv_whitelist;
289 typedef struct CreateCo {
290 BlockDriver *drv;
291 char *filename;
292 QemuOpts *opts;
293 int ret;
294 Error *err;
295 } CreateCo;
297 static void coroutine_fn bdrv_create_co_entry(void *opaque)
299 Error *local_err = NULL;
300 int ret;
302 CreateCo *cco = opaque;
303 assert(cco->drv);
305 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
306 error_propagate(&cco->err, local_err);
307 cco->ret = ret;
310 int bdrv_create(BlockDriver *drv, const char* filename,
311 QemuOpts *opts, Error **errp)
313 int ret;
315 Coroutine *co;
316 CreateCo cco = {
317 .drv = drv,
318 .filename = g_strdup(filename),
319 .opts = opts,
320 .ret = NOT_DONE,
321 .err = NULL,
324 if (!drv->bdrv_create) {
325 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
326 ret = -ENOTSUP;
327 goto out;
330 if (qemu_in_coroutine()) {
331 /* Fast-path if already in coroutine context */
332 bdrv_create_co_entry(&cco);
333 } else {
334 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
335 qemu_coroutine_enter(co);
336 while (cco.ret == NOT_DONE) {
337 aio_poll(qemu_get_aio_context(), true);
341 ret = cco.ret;
342 if (ret < 0) {
343 if (cco.err) {
344 error_propagate(errp, cco.err);
345 } else {
346 error_setg_errno(errp, -ret, "Could not create image");
350 out:
351 g_free(cco.filename);
352 return ret;
355 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
357 BlockDriver *drv;
358 Error *local_err = NULL;
359 int ret;
361 drv = bdrv_find_protocol(filename, true, errp);
362 if (drv == NULL) {
363 return -ENOENT;
366 ret = bdrv_create(drv, filename, opts, &local_err);
367 error_propagate(errp, local_err);
368 return ret;
372 * Try to get @bs's logical and physical block size.
373 * On success, store them in @bsz struct and return 0.
374 * On failure return -errno.
375 * @bs must not be empty.
377 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
379 BlockDriver *drv = bs->drv;
381 if (drv && drv->bdrv_probe_blocksizes) {
382 return drv->bdrv_probe_blocksizes(bs, bsz);
385 return -ENOTSUP;
389 * Try to get @bs's geometry (cyls, heads, sectors).
390 * On success, store them in @geo struct and return 0.
391 * On failure return -errno.
392 * @bs must not be empty.
394 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
396 BlockDriver *drv = bs->drv;
398 if (drv && drv->bdrv_probe_geometry) {
399 return drv->bdrv_probe_geometry(bs, geo);
402 return -ENOTSUP;
406 * Create a uniquely-named empty temporary file.
407 * Return 0 upon success, otherwise a negative errno value.
409 int get_tmp_filename(char *filename, int size)
411 #ifdef _WIN32
412 char temp_dir[MAX_PATH];
413 /* GetTempFileName requires that its output buffer (4th param)
414 have length MAX_PATH or greater. */
415 assert(size >= MAX_PATH);
416 return (GetTempPath(MAX_PATH, temp_dir)
417 && GetTempFileName(temp_dir, "qem", 0, filename)
418 ? 0 : -GetLastError());
419 #else
420 int fd;
421 const char *tmpdir;
422 tmpdir = getenv("TMPDIR");
423 if (!tmpdir) {
424 tmpdir = "/var/tmp";
426 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
427 return -EOVERFLOW;
429 fd = mkstemp(filename);
430 if (fd < 0) {
431 return -errno;
433 if (close(fd) != 0) {
434 unlink(filename);
435 return -errno;
437 return 0;
438 #endif
442 * Detect host devices. By convention, /dev/cdrom[N] is always
443 * recognized as a host CDROM.
445 static BlockDriver *find_hdev_driver(const char *filename)
447 int score_max = 0, score;
448 BlockDriver *drv = NULL, *d;
450 QLIST_FOREACH(d, &bdrv_drivers, list) {
451 if (d->bdrv_probe_device) {
452 score = d->bdrv_probe_device(filename);
453 if (score > score_max) {
454 score_max = score;
455 drv = d;
460 return drv;
463 BlockDriver *bdrv_find_protocol(const char *filename,
464 bool allow_protocol_prefix,
465 Error **errp)
467 BlockDriver *drv1;
468 char protocol[128];
469 int len;
470 const char *p;
472 /* TODO Drivers without bdrv_file_open must be specified explicitly */
475 * XXX(hch): we really should not let host device detection
476 * override an explicit protocol specification, but moving this
477 * later breaks access to device names with colons in them.
478 * Thanks to the brain-dead persistent naming schemes on udev-
479 * based Linux systems those actually are quite common.
481 drv1 = find_hdev_driver(filename);
482 if (drv1) {
483 return drv1;
486 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
487 return &bdrv_file;
490 p = strchr(filename, ':');
491 assert(p != NULL);
492 len = p - filename;
493 if (len > sizeof(protocol) - 1)
494 len = sizeof(protocol) - 1;
495 memcpy(protocol, filename, len);
496 protocol[len] = '\0';
497 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
498 if (drv1->protocol_name &&
499 !strcmp(drv1->protocol_name, protocol)) {
500 return drv1;
504 error_setg(errp, "Unknown protocol '%s'", protocol);
505 return NULL;
509 * Guess image format by probing its contents.
510 * This is not a good idea when your image is raw (CVE-2008-2004), but
511 * we do it anyway for backward compatibility.
513 * @buf contains the image's first @buf_size bytes.
514 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
515 * but can be smaller if the image file is smaller)
516 * @filename is its filename.
518 * For all block drivers, call the bdrv_probe() method to get its
519 * probing score.
520 * Return the first block driver with the highest probing score.
522 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
523 const char *filename)
525 int score_max = 0, score;
526 BlockDriver *drv = NULL, *d;
528 QLIST_FOREACH(d, &bdrv_drivers, list) {
529 if (d->bdrv_probe) {
530 score = d->bdrv_probe(buf, buf_size, filename);
531 if (score > score_max) {
532 score_max = score;
533 drv = d;
538 return drv;
541 static int find_image_format(BdrvChild *file, const char *filename,
542 BlockDriver **pdrv, Error **errp)
544 BlockDriverState *bs = file->bs;
545 BlockDriver *drv;
546 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
547 int ret = 0;
549 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
550 if (bdrv_is_sg(bs) || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
551 *pdrv = &bdrv_raw;
552 return ret;
555 ret = bdrv_pread(file, 0, buf, sizeof(buf));
556 if (ret < 0) {
557 error_setg_errno(errp, -ret, "Could not read image for determining its "
558 "format");
559 *pdrv = NULL;
560 return ret;
563 drv = bdrv_probe_all(buf, ret, filename);
564 if (!drv) {
565 error_setg(errp, "Could not determine image format: No compatible "
566 "driver found");
567 ret = -ENOENT;
569 *pdrv = drv;
570 return ret;
574 * Set the current 'total_sectors' value
575 * Return 0 on success, -errno on error.
577 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
579 BlockDriver *drv = bs->drv;
581 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
582 if (bdrv_is_sg(bs))
583 return 0;
585 /* query actual device if possible, otherwise just trust the hint */
586 if (drv->bdrv_getlength) {
587 int64_t length = drv->bdrv_getlength(bs);
588 if (length < 0) {
589 return length;
591 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
594 bs->total_sectors = hint;
595 return 0;
599 * Combines a QDict of new block driver @options with any missing options taken
600 * from @old_options, so that leaving out an option defaults to its old value.
602 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
603 QDict *old_options)
605 if (bs->drv && bs->drv->bdrv_join_options) {
606 bs->drv->bdrv_join_options(options, old_options);
607 } else {
608 qdict_join(options, old_options, false);
613 * Set open flags for a given discard mode
615 * Return 0 on success, -1 if the discard mode was invalid.
617 int bdrv_parse_discard_flags(const char *mode, int *flags)
619 *flags &= ~BDRV_O_UNMAP;
621 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
622 /* do nothing */
623 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
624 *flags |= BDRV_O_UNMAP;
625 } else {
626 return -1;
629 return 0;
633 * Set open flags for a given cache mode
635 * Return 0 on success, -1 if the cache mode was invalid.
637 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
639 *flags &= ~BDRV_O_CACHE_MASK;
641 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
642 *writethrough = false;
643 *flags |= BDRV_O_NOCACHE;
644 } else if (!strcmp(mode, "directsync")) {
645 *writethrough = true;
646 *flags |= BDRV_O_NOCACHE;
647 } else if (!strcmp(mode, "writeback")) {
648 *writethrough = false;
649 } else if (!strcmp(mode, "unsafe")) {
650 *writethrough = false;
651 *flags |= BDRV_O_NO_FLUSH;
652 } else if (!strcmp(mode, "writethrough")) {
653 *writethrough = true;
654 } else {
655 return -1;
658 return 0;
661 static void bdrv_child_cb_drained_begin(BdrvChild *child)
663 BlockDriverState *bs = child->opaque;
664 bdrv_drained_begin(bs);
667 static void bdrv_child_cb_drained_end(BdrvChild *child)
669 BlockDriverState *bs = child->opaque;
670 bdrv_drained_end(bs);
674 * Returns the options and flags that a temporary snapshot should get, based on
675 * the originally requested flags (the originally requested image will have
676 * flags like a backing file)
678 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
679 int parent_flags, QDict *parent_options)
681 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
683 /* For temporary files, unconditional cache=unsafe is fine */
684 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
685 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
687 /* aio=native doesn't work for cache.direct=off, so disable it for the
688 * temporary snapshot */
689 *child_flags &= ~BDRV_O_NATIVE_AIO;
693 * Returns the options and flags that bs->file should get if a protocol driver
694 * is expected, based on the given options and flags for the parent BDS
696 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
697 int parent_flags, QDict *parent_options)
699 int flags = parent_flags;
701 /* Enable protocol handling, disable format probing for bs->file */
702 flags |= BDRV_O_PROTOCOL;
704 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
705 * the parent. */
706 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
707 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
709 /* Our block drivers take care to send flushes and respect unmap policy,
710 * so we can default to enable both on lower layers regardless of the
711 * corresponding parent options. */
712 flags |= BDRV_O_UNMAP;
714 /* Clear flags that only apply to the top layer */
715 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
716 BDRV_O_NO_IO);
718 *child_flags = flags;
721 const BdrvChildRole child_file = {
722 .inherit_options = bdrv_inherited_options,
723 .drained_begin = bdrv_child_cb_drained_begin,
724 .drained_end = bdrv_child_cb_drained_end,
728 * Returns the options and flags that bs->file should get if the use of formats
729 * (and not only protocols) is permitted for it, based on the given options and
730 * flags for the parent BDS
732 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
733 int parent_flags, QDict *parent_options)
735 child_file.inherit_options(child_flags, child_options,
736 parent_flags, parent_options);
738 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
741 const BdrvChildRole child_format = {
742 .inherit_options = bdrv_inherited_fmt_options,
743 .drained_begin = bdrv_child_cb_drained_begin,
744 .drained_end = bdrv_child_cb_drained_end,
748 * Returns the options and flags that bs->backing should get, based on the
749 * given options and flags for the parent BDS
751 static void bdrv_backing_options(int *child_flags, QDict *child_options,
752 int parent_flags, QDict *parent_options)
754 int flags = parent_flags;
756 /* The cache mode is inherited unmodified for backing files; except WCE,
757 * which is only applied on the top level (BlockBackend) */
758 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
759 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
761 /* backing files always opened read-only */
762 flags &= ~(BDRV_O_RDWR | BDRV_O_COPY_ON_READ);
764 /* snapshot=on is handled on the top layer */
765 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
767 *child_flags = flags;
770 static const BdrvChildRole child_backing = {
771 .inherit_options = bdrv_backing_options,
772 .drained_begin = bdrv_child_cb_drained_begin,
773 .drained_end = bdrv_child_cb_drained_end,
776 static int bdrv_open_flags(BlockDriverState *bs, int flags)
778 int open_flags = flags;
781 * Clear flags that are internal to the block layer before opening the
782 * image.
784 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
787 * Snapshots should be writable.
789 if (flags & BDRV_O_TEMPORARY) {
790 open_flags |= BDRV_O_RDWR;
793 return open_flags;
796 static void update_flags_from_options(int *flags, QemuOpts *opts)
798 *flags &= ~BDRV_O_CACHE_MASK;
800 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
801 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
802 *flags |= BDRV_O_NO_FLUSH;
805 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
806 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
807 *flags |= BDRV_O_NOCACHE;
811 static void update_options_from_flags(QDict *options, int flags)
813 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
814 qdict_put(options, BDRV_OPT_CACHE_DIRECT,
815 qbool_from_bool(flags & BDRV_O_NOCACHE));
817 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
818 qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH,
819 qbool_from_bool(flags & BDRV_O_NO_FLUSH));
823 static void bdrv_assign_node_name(BlockDriverState *bs,
824 const char *node_name,
825 Error **errp)
827 char *gen_node_name = NULL;
829 if (!node_name) {
830 node_name = gen_node_name = id_generate(ID_BLOCK);
831 } else if (!id_wellformed(node_name)) {
833 * Check for empty string or invalid characters, but not if it is
834 * generated (generated names use characters not available to the user)
836 error_setg(errp, "Invalid node name");
837 return;
840 /* takes care of avoiding namespaces collisions */
841 if (blk_by_name(node_name)) {
842 error_setg(errp, "node-name=%s is conflicting with a device id",
843 node_name);
844 goto out;
847 /* takes care of avoiding duplicates node names */
848 if (bdrv_find_node(node_name)) {
849 error_setg(errp, "Duplicate node name");
850 goto out;
853 /* copy node name into the bs and insert it into the graph list */
854 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
855 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
856 out:
857 g_free(gen_node_name);
860 static QemuOptsList bdrv_runtime_opts = {
861 .name = "bdrv_common",
862 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
863 .desc = {
865 .name = "node-name",
866 .type = QEMU_OPT_STRING,
867 .help = "Node name of the block device node",
870 .name = "driver",
871 .type = QEMU_OPT_STRING,
872 .help = "Block driver to use for the node",
875 .name = BDRV_OPT_CACHE_DIRECT,
876 .type = QEMU_OPT_BOOL,
877 .help = "Bypass software writeback cache on the host",
880 .name = BDRV_OPT_CACHE_NO_FLUSH,
881 .type = QEMU_OPT_BOOL,
882 .help = "Ignore flush requests",
884 { /* end of list */ }
889 * Common part for opening disk images and files
891 * Removes all processed options from *options.
893 static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
894 QDict *options, Error **errp)
896 int ret, open_flags;
897 const char *filename;
898 const char *driver_name = NULL;
899 const char *node_name = NULL;
900 QemuOpts *opts;
901 BlockDriver *drv;
902 Error *local_err = NULL;
904 assert(bs->file == NULL);
905 assert(options != NULL && bs->options != options);
907 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
908 qemu_opts_absorb_qdict(opts, options, &local_err);
909 if (local_err) {
910 error_propagate(errp, local_err);
911 ret = -EINVAL;
912 goto fail_opts;
915 driver_name = qemu_opt_get(opts, "driver");
916 drv = bdrv_find_format(driver_name);
917 assert(drv != NULL);
919 if (file != NULL) {
920 filename = file->bs->filename;
921 } else {
922 filename = qdict_get_try_str(options, "filename");
925 if (drv->bdrv_needs_filename && !filename) {
926 error_setg(errp, "The '%s' block driver requires a file name",
927 drv->format_name);
928 ret = -EINVAL;
929 goto fail_opts;
932 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
933 drv->format_name);
935 node_name = qemu_opt_get(opts, "node-name");
936 bdrv_assign_node_name(bs, node_name, &local_err);
937 if (local_err) {
938 error_propagate(errp, local_err);
939 ret = -EINVAL;
940 goto fail_opts;
943 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
945 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
946 error_setg(errp,
947 !bs->read_only && bdrv_is_whitelisted(drv, true)
948 ? "Driver '%s' can only be used for read-only devices"
949 : "Driver '%s' is not whitelisted",
950 drv->format_name);
951 ret = -ENOTSUP;
952 goto fail_opts;
955 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
956 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
957 if (!bs->read_only) {
958 bdrv_enable_copy_on_read(bs);
959 } else {
960 error_setg(errp, "Can't use copy-on-read on read-only device");
961 ret = -EINVAL;
962 goto fail_opts;
966 if (filename != NULL) {
967 pstrcpy(bs->filename, sizeof(bs->filename), filename);
968 } else {
969 bs->filename[0] = '\0';
971 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
973 bs->drv = drv;
974 bs->opaque = g_malloc0(drv->instance_size);
976 /* Apply cache mode options */
977 update_flags_from_options(&bs->open_flags, opts);
979 /* Open the image, either directly or using a protocol */
980 open_flags = bdrv_open_flags(bs, bs->open_flags);
981 if (drv->bdrv_file_open) {
982 assert(file == NULL);
983 assert(!drv->bdrv_needs_filename || filename != NULL);
984 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
985 } else {
986 if (file == NULL) {
987 error_setg(errp, "Can't use '%s' as a block driver for the "
988 "protocol level", drv->format_name);
989 ret = -EINVAL;
990 goto free_and_fail;
992 bs->file = file;
993 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
996 if (ret < 0) {
997 if (local_err) {
998 error_propagate(errp, local_err);
999 } else if (bs->filename[0]) {
1000 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1001 } else {
1002 error_setg_errno(errp, -ret, "Could not open image");
1004 goto free_and_fail;
1007 ret = refresh_total_sectors(bs, bs->total_sectors);
1008 if (ret < 0) {
1009 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1010 goto free_and_fail;
1013 bdrv_refresh_limits(bs, &local_err);
1014 if (local_err) {
1015 error_propagate(errp, local_err);
1016 ret = -EINVAL;
1017 goto free_and_fail;
1020 assert(bdrv_opt_mem_align(bs) != 0);
1021 assert(bdrv_min_mem_align(bs) != 0);
1022 assert(is_power_of_2(bs->bl.request_alignment));
1024 qemu_opts_del(opts);
1025 return 0;
1027 free_and_fail:
1028 bs->file = NULL;
1029 g_free(bs->opaque);
1030 bs->opaque = NULL;
1031 bs->drv = NULL;
1032 fail_opts:
1033 qemu_opts_del(opts);
1034 return ret;
1037 static QDict *parse_json_filename(const char *filename, Error **errp)
1039 QObject *options_obj;
1040 QDict *options;
1041 int ret;
1043 ret = strstart(filename, "json:", &filename);
1044 assert(ret);
1046 options_obj = qobject_from_json(filename);
1047 if (!options_obj) {
1048 error_setg(errp, "Could not parse the JSON options");
1049 return NULL;
1052 if (qobject_type(options_obj) != QTYPE_QDICT) {
1053 qobject_decref(options_obj);
1054 error_setg(errp, "Invalid JSON object given");
1055 return NULL;
1058 options = qobject_to_qdict(options_obj);
1059 qdict_flatten(options);
1061 return options;
1064 static void parse_json_protocol(QDict *options, const char **pfilename,
1065 Error **errp)
1067 QDict *json_options;
1068 Error *local_err = NULL;
1070 /* Parse json: pseudo-protocol */
1071 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1072 return;
1075 json_options = parse_json_filename(*pfilename, &local_err);
1076 if (local_err) {
1077 error_propagate(errp, local_err);
1078 return;
1081 /* Options given in the filename have lower priority than options
1082 * specified directly */
1083 qdict_join(options, json_options, false);
1084 QDECREF(json_options);
1085 *pfilename = NULL;
1089 * Fills in default options for opening images and converts the legacy
1090 * filename/flags pair to option QDict entries.
1091 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1092 * block driver has been specified explicitly.
1094 static int bdrv_fill_options(QDict **options, const char *filename,
1095 int *flags, Error **errp)
1097 const char *drvname;
1098 bool protocol = *flags & BDRV_O_PROTOCOL;
1099 bool parse_filename = false;
1100 BlockDriver *drv = NULL;
1101 Error *local_err = NULL;
1103 drvname = qdict_get_try_str(*options, "driver");
1104 if (drvname) {
1105 drv = bdrv_find_format(drvname);
1106 if (!drv) {
1107 error_setg(errp, "Unknown driver '%s'", drvname);
1108 return -ENOENT;
1110 /* If the user has explicitly specified the driver, this choice should
1111 * override the BDRV_O_PROTOCOL flag */
1112 protocol = drv->bdrv_file_open;
1115 if (protocol) {
1116 *flags |= BDRV_O_PROTOCOL;
1117 } else {
1118 *flags &= ~BDRV_O_PROTOCOL;
1121 /* Translate cache options from flags into options */
1122 update_options_from_flags(*options, *flags);
1124 /* Fetch the file name from the options QDict if necessary */
1125 if (protocol && filename) {
1126 if (!qdict_haskey(*options, "filename")) {
1127 qdict_put(*options, "filename", qstring_from_str(filename));
1128 parse_filename = true;
1129 } else {
1130 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1131 "the same time");
1132 return -EINVAL;
1136 /* Find the right block driver */
1137 filename = qdict_get_try_str(*options, "filename");
1139 if (!drvname && protocol) {
1140 if (filename) {
1141 drv = bdrv_find_protocol(filename, parse_filename, errp);
1142 if (!drv) {
1143 return -EINVAL;
1146 drvname = drv->format_name;
1147 qdict_put(*options, "driver", qstring_from_str(drvname));
1148 } else {
1149 error_setg(errp, "Must specify either driver or file");
1150 return -EINVAL;
1154 assert(drv || !protocol);
1156 /* Driver-specific filename parsing */
1157 if (drv && drv->bdrv_parse_filename && parse_filename) {
1158 drv->bdrv_parse_filename(filename, *options, &local_err);
1159 if (local_err) {
1160 error_propagate(errp, local_err);
1161 return -EINVAL;
1164 if (!drv->bdrv_needs_filename) {
1165 qdict_del(*options, "filename");
1169 return 0;
1172 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1174 BlockDriverState *old_bs = child->bs;
1176 if (old_bs) {
1177 if (old_bs->quiesce_counter && child->role->drained_end) {
1178 child->role->drained_end(child);
1180 QLIST_REMOVE(child, next_parent);
1183 child->bs = new_bs;
1185 if (new_bs) {
1186 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1187 if (new_bs->quiesce_counter && child->role->drained_begin) {
1188 child->role->drained_begin(child);
1193 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1194 const char *child_name,
1195 const BdrvChildRole *child_role,
1196 void *opaque)
1198 BdrvChild *child = g_new(BdrvChild, 1);
1199 *child = (BdrvChild) {
1200 .bs = NULL,
1201 .name = g_strdup(child_name),
1202 .role = child_role,
1203 .opaque = opaque,
1206 bdrv_replace_child(child, child_bs);
1208 return child;
1211 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1212 BlockDriverState *child_bs,
1213 const char *child_name,
1214 const BdrvChildRole *child_role)
1216 BdrvChild *child = bdrv_root_attach_child(child_bs, child_name, child_role,
1217 parent_bs);
1218 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1219 return child;
1222 static void bdrv_detach_child(BdrvChild *child)
1224 if (child->next.le_prev) {
1225 QLIST_REMOVE(child, next);
1226 child->next.le_prev = NULL;
1229 bdrv_replace_child(child, NULL);
1231 g_free(child->name);
1232 g_free(child);
1235 void bdrv_root_unref_child(BdrvChild *child)
1237 BlockDriverState *child_bs;
1239 child_bs = child->bs;
1240 bdrv_detach_child(child);
1241 bdrv_unref(child_bs);
1244 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1246 if (child == NULL) {
1247 return;
1250 if (child->bs->inherits_from == parent) {
1251 child->bs->inherits_from = NULL;
1254 bdrv_root_unref_child(child);
1258 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1260 BdrvChild *c;
1261 QLIST_FOREACH(c, &bs->parents, next_parent) {
1262 if (c->role->change_media) {
1263 c->role->change_media(c, load);
1268 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1270 BdrvChild *c;
1271 QLIST_FOREACH(c, &bs->parents, next_parent) {
1272 if (c->role->resize) {
1273 c->role->resize(c);
1279 * Sets the backing file link of a BDS. A new reference is created; callers
1280 * which don't need their own reference any more must call bdrv_unref().
1282 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1284 if (backing_hd) {
1285 bdrv_ref(backing_hd);
1288 if (bs->backing) {
1289 assert(bs->backing_blocker);
1290 bdrv_op_unblock_all(bs->backing->bs, bs->backing_blocker);
1291 bdrv_unref_child(bs, bs->backing);
1292 } else if (backing_hd) {
1293 error_setg(&bs->backing_blocker,
1294 "node is used as backing hd of '%s'",
1295 bdrv_get_device_or_node_name(bs));
1298 if (!backing_hd) {
1299 error_free(bs->backing_blocker);
1300 bs->backing_blocker = NULL;
1301 bs->backing = NULL;
1302 goto out;
1304 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing);
1305 bs->open_flags &= ~BDRV_O_NO_BACKING;
1306 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1307 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1308 backing_hd->drv ? backing_hd->drv->format_name : "");
1310 bdrv_op_block_all(backing_hd, bs->backing_blocker);
1311 /* Otherwise we won't be able to commit due to check in bdrv_commit */
1312 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1313 bs->backing_blocker);
1314 out:
1315 bdrv_refresh_limits(bs, NULL);
1319 * Opens the backing file for a BlockDriverState if not yet open
1321 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
1322 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1323 * itself, all options starting with "${bdref_key}." are considered part of the
1324 * BlockdevRef.
1326 * TODO Can this be unified with bdrv_open_image()?
1328 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
1329 const char *bdref_key, Error **errp)
1331 char *backing_filename = g_malloc0(PATH_MAX);
1332 char *bdref_key_dot;
1333 const char *reference = NULL;
1334 int ret = 0;
1335 BlockDriverState *backing_hd;
1336 QDict *options;
1337 QDict *tmp_parent_options = NULL;
1338 Error *local_err = NULL;
1340 if (bs->backing != NULL) {
1341 goto free_exit;
1344 /* NULL means an empty set of options */
1345 if (parent_options == NULL) {
1346 tmp_parent_options = qdict_new();
1347 parent_options = tmp_parent_options;
1350 bs->open_flags &= ~BDRV_O_NO_BACKING;
1352 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1353 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
1354 g_free(bdref_key_dot);
1356 reference = qdict_get_try_str(parent_options, bdref_key);
1357 if (reference || qdict_haskey(options, "file.filename")) {
1358 backing_filename[0] = '\0';
1359 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1360 QDECREF(options);
1361 goto free_exit;
1362 } else {
1363 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
1364 &local_err);
1365 if (local_err) {
1366 ret = -EINVAL;
1367 error_propagate(errp, local_err);
1368 QDECREF(options);
1369 goto free_exit;
1373 if (!bs->drv || !bs->drv->supports_backing) {
1374 ret = -EINVAL;
1375 error_setg(errp, "Driver doesn't support backing files");
1376 QDECREF(options);
1377 goto free_exit;
1380 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
1381 qdict_put(options, "driver", qstring_from_str(bs->backing_format));
1384 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
1385 reference, options, 0, bs, &child_backing,
1386 errp);
1387 if (!backing_hd) {
1388 bs->open_flags |= BDRV_O_NO_BACKING;
1389 error_prepend(errp, "Could not open backing file: ");
1390 ret = -EINVAL;
1391 goto free_exit;
1394 /* Hook up the backing file link; drop our reference, bs owns the
1395 * backing_hd reference now */
1396 bdrv_set_backing_hd(bs, backing_hd);
1397 bdrv_unref(backing_hd);
1399 qdict_del(parent_options, bdref_key);
1401 free_exit:
1402 g_free(backing_filename);
1403 QDECREF(tmp_parent_options);
1404 return ret;
1408 * Opens a disk image whose options are given as BlockdevRef in another block
1409 * device's options.
1411 * If allow_none is true, no image will be opened if filename is false and no
1412 * BlockdevRef is given. NULL will be returned, but errp remains unset.
1414 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1415 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1416 * itself, all options starting with "${bdref_key}." are considered part of the
1417 * BlockdevRef.
1419 * The BlockdevRef will be removed from the options QDict.
1421 BdrvChild *bdrv_open_child(const char *filename,
1422 QDict *options, const char *bdref_key,
1423 BlockDriverState* parent,
1424 const BdrvChildRole *child_role,
1425 bool allow_none, Error **errp)
1427 BdrvChild *c = NULL;
1428 BlockDriverState *bs;
1429 QDict *image_options;
1430 char *bdref_key_dot;
1431 const char *reference;
1433 assert(child_role != NULL);
1435 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1436 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1437 g_free(bdref_key_dot);
1439 reference = qdict_get_try_str(options, bdref_key);
1440 if (!filename && !reference && !qdict_size(image_options)) {
1441 if (!allow_none) {
1442 error_setg(errp, "A block device must be specified for \"%s\"",
1443 bdref_key);
1445 QDECREF(image_options);
1446 goto done;
1449 bs = bdrv_open_inherit(filename, reference, image_options, 0,
1450 parent, child_role, errp);
1451 if (!bs) {
1452 goto done;
1455 c = bdrv_attach_child(parent, bs, bdref_key, child_role);
1457 done:
1458 qdict_del(options, bdref_key);
1459 return c;
1462 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
1463 int flags,
1464 QDict *snapshot_options,
1465 Error **errp)
1467 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1468 char *tmp_filename = g_malloc0(PATH_MAX + 1);
1469 int64_t total_size;
1470 QemuOpts *opts = NULL;
1471 BlockDriverState *bs_snapshot;
1472 int ret;
1474 /* if snapshot, we create a temporary backing file and open it
1475 instead of opening 'filename' directly */
1477 /* Get the required size from the image */
1478 total_size = bdrv_getlength(bs);
1479 if (total_size < 0) {
1480 error_setg_errno(errp, -total_size, "Could not get image size");
1481 goto out;
1484 /* Create the temporary image */
1485 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1486 if (ret < 0) {
1487 error_setg_errno(errp, -ret, "Could not get temporary filename");
1488 goto out;
1491 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
1492 &error_abort);
1493 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
1494 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
1495 qemu_opts_del(opts);
1496 if (ret < 0) {
1497 error_prepend(errp, "Could not create temporary overlay '%s': ",
1498 tmp_filename);
1499 goto out;
1502 /* Prepare options QDict for the temporary file */
1503 qdict_put(snapshot_options, "file.driver",
1504 qstring_from_str("file"));
1505 qdict_put(snapshot_options, "file.filename",
1506 qstring_from_str(tmp_filename));
1507 qdict_put(snapshot_options, "driver",
1508 qstring_from_str("qcow2"));
1510 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
1511 snapshot_options = NULL;
1512 if (!bs_snapshot) {
1513 ret = -EINVAL;
1514 goto out;
1517 /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will
1518 * call bdrv_unref() on it), so in order to be able to return one, we have
1519 * to increase bs_snapshot's refcount here */
1520 bdrv_ref(bs_snapshot);
1521 bdrv_append(bs_snapshot, bs);
1523 g_free(tmp_filename);
1524 return bs_snapshot;
1526 out:
1527 QDECREF(snapshot_options);
1528 g_free(tmp_filename);
1529 return NULL;
1533 * Opens a disk image (raw, qcow2, vmdk, ...)
1535 * options is a QDict of options to pass to the block drivers, or NULL for an
1536 * empty set of options. The reference to the QDict belongs to the block layer
1537 * after the call (even on failure), so if the caller intends to reuse the
1538 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1540 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1541 * If it is not NULL, the referenced BDS will be reused.
1543 * The reference parameter may be used to specify an existing block device which
1544 * should be opened. If specified, neither options nor a filename may be given,
1545 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1547 static BlockDriverState *bdrv_open_inherit(const char *filename,
1548 const char *reference,
1549 QDict *options, int flags,
1550 BlockDriverState *parent,
1551 const BdrvChildRole *child_role,
1552 Error **errp)
1554 int ret;
1555 BdrvChild *file = NULL;
1556 BlockDriverState *bs;
1557 BlockDriver *drv = NULL;
1558 const char *drvname;
1559 const char *backing;
1560 Error *local_err = NULL;
1561 QDict *snapshot_options = NULL;
1562 int snapshot_flags = 0;
1564 assert(!child_role || !flags);
1565 assert(!child_role == !parent);
1567 if (reference) {
1568 bool options_non_empty = options ? qdict_size(options) : false;
1569 QDECREF(options);
1571 if (filename || options_non_empty) {
1572 error_setg(errp, "Cannot reference an existing block device with "
1573 "additional options or a new filename");
1574 return NULL;
1577 bs = bdrv_lookup_bs(reference, reference, errp);
1578 if (!bs) {
1579 return NULL;
1582 bdrv_ref(bs);
1583 return bs;
1586 bs = bdrv_new();
1588 /* NULL means an empty set of options */
1589 if (options == NULL) {
1590 options = qdict_new();
1593 /* json: syntax counts as explicit options, as if in the QDict */
1594 parse_json_protocol(options, &filename, &local_err);
1595 if (local_err) {
1596 goto fail;
1599 bs->explicit_options = qdict_clone_shallow(options);
1601 if (child_role) {
1602 bs->inherits_from = parent;
1603 child_role->inherit_options(&flags, options,
1604 parent->open_flags, parent->options);
1607 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
1608 if (local_err) {
1609 goto fail;
1612 bs->open_flags = flags;
1613 bs->options = options;
1614 options = qdict_clone_shallow(options);
1616 /* Find the right image format driver */
1617 drvname = qdict_get_try_str(options, "driver");
1618 if (drvname) {
1619 drv = bdrv_find_format(drvname);
1620 if (!drv) {
1621 error_setg(errp, "Unknown driver: '%s'", drvname);
1622 goto fail;
1626 assert(drvname || !(flags & BDRV_O_PROTOCOL));
1628 backing = qdict_get_try_str(options, "backing");
1629 if (backing && *backing == '\0') {
1630 flags |= BDRV_O_NO_BACKING;
1631 qdict_del(options, "backing");
1634 /* Open image file without format layer */
1635 if ((flags & BDRV_O_PROTOCOL) == 0) {
1636 if (flags & BDRV_O_RDWR) {
1637 flags |= BDRV_O_ALLOW_RDWR;
1639 if (flags & BDRV_O_SNAPSHOT) {
1640 snapshot_options = qdict_new();
1641 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
1642 flags, options);
1643 bdrv_backing_options(&flags, options, flags, options);
1646 bs->open_flags = flags;
1648 file = bdrv_open_child(filename, options, "file", bs,
1649 &child_file, true, &local_err);
1650 if (local_err) {
1651 goto fail;
1655 /* Image format probing */
1656 bs->probed = !drv;
1657 if (!drv && file) {
1658 ret = find_image_format(file, filename, &drv, &local_err);
1659 if (ret < 0) {
1660 goto fail;
1663 * This option update would logically belong in bdrv_fill_options(),
1664 * but we first need to open bs->file for the probing to work, while
1665 * opening bs->file already requires the (mostly) final set of options
1666 * so that cache mode etc. can be inherited.
1668 * Adding the driver later is somewhat ugly, but it's not an option
1669 * that would ever be inherited, so it's correct. We just need to make
1670 * sure to update both bs->options (which has the full effective
1671 * options for bs) and options (which has file.* already removed).
1673 qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
1674 qdict_put(options, "driver", qstring_from_str(drv->format_name));
1675 } else if (!drv) {
1676 error_setg(errp, "Must specify either driver or file");
1677 goto fail;
1680 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
1681 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
1682 /* file must be NULL if a protocol BDS is about to be created
1683 * (the inverse results in an error message from bdrv_open_common()) */
1684 assert(!(flags & BDRV_O_PROTOCOL) || !file);
1686 /* Open the image */
1687 ret = bdrv_open_common(bs, file, options, &local_err);
1688 if (ret < 0) {
1689 goto fail;
1692 if (file && (bs->file != file)) {
1693 bdrv_unref_child(bs, file);
1694 file = NULL;
1697 /* If there is a backing file, use it */
1698 if ((flags & BDRV_O_NO_BACKING) == 0) {
1699 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
1700 if (ret < 0) {
1701 goto close_and_fail;
1705 bdrv_refresh_filename(bs);
1707 /* Check if any unknown options were used */
1708 if (options && (qdict_size(options) != 0)) {
1709 const QDictEntry *entry = qdict_first(options);
1710 if (flags & BDRV_O_PROTOCOL) {
1711 error_setg(errp, "Block protocol '%s' doesn't support the option "
1712 "'%s'", drv->format_name, entry->key);
1713 } else {
1714 error_setg(errp,
1715 "Block format '%s' does not support the option '%s'",
1716 drv->format_name, entry->key);
1719 goto close_and_fail;
1722 if (!bdrv_key_required(bs)) {
1723 bdrv_parent_cb_change_media(bs, true);
1724 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1725 && !runstate_check(RUN_STATE_INMIGRATE)
1726 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1727 error_setg(errp,
1728 "Guest must be stopped for opening of encrypted image");
1729 goto close_and_fail;
1732 QDECREF(options);
1734 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1735 * temporary snapshot afterwards. */
1736 if (snapshot_flags) {
1737 BlockDriverState *snapshot_bs;
1738 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
1739 snapshot_options, &local_err);
1740 snapshot_options = NULL;
1741 if (local_err) {
1742 goto close_and_fail;
1744 /* We are not going to return bs but the overlay on top of it
1745 * (snapshot_bs); thus, we have to drop the strong reference to bs
1746 * (which we obtained by calling bdrv_new()). bs will not be deleted,
1747 * though, because the overlay still has a reference to it. */
1748 bdrv_unref(bs);
1749 bs = snapshot_bs;
1752 return bs;
1754 fail:
1755 if (file != NULL) {
1756 bdrv_unref_child(bs, file);
1758 QDECREF(snapshot_options);
1759 QDECREF(bs->explicit_options);
1760 QDECREF(bs->options);
1761 QDECREF(options);
1762 bs->options = NULL;
1763 bdrv_unref(bs);
1764 error_propagate(errp, local_err);
1765 return NULL;
1767 close_and_fail:
1768 bdrv_unref(bs);
1769 QDECREF(snapshot_options);
1770 QDECREF(options);
1771 error_propagate(errp, local_err);
1772 return NULL;
1775 BlockDriverState *bdrv_open(const char *filename, const char *reference,
1776 QDict *options, int flags, Error **errp)
1778 return bdrv_open_inherit(filename, reference, options, flags, NULL,
1779 NULL, errp);
1782 typedef struct BlockReopenQueueEntry {
1783 bool prepared;
1784 BDRVReopenState state;
1785 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1786 } BlockReopenQueueEntry;
1789 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1790 * reopen of multiple devices.
1792 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1793 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1794 * be created and initialized. This newly created BlockReopenQueue should be
1795 * passed back in for subsequent calls that are intended to be of the same
1796 * atomic 'set'.
1798 * bs is the BlockDriverState to add to the reopen queue.
1800 * options contains the changed options for the associated bs
1801 * (the BlockReopenQueue takes ownership)
1803 * flags contains the open flags for the associated bs
1805 * returns a pointer to bs_queue, which is either the newly allocated
1806 * bs_queue, or the existing bs_queue being used.
1809 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
1810 BlockDriverState *bs,
1811 QDict *options,
1812 int flags,
1813 const BdrvChildRole *role,
1814 QDict *parent_options,
1815 int parent_flags)
1817 assert(bs != NULL);
1819 BlockReopenQueueEntry *bs_entry;
1820 BdrvChild *child;
1821 QDict *old_options, *explicit_options;
1823 if (bs_queue == NULL) {
1824 bs_queue = g_new0(BlockReopenQueue, 1);
1825 QSIMPLEQ_INIT(bs_queue);
1828 if (!options) {
1829 options = qdict_new();
1833 * Precedence of options:
1834 * 1. Explicitly passed in options (highest)
1835 * 2. Set in flags (only for top level)
1836 * 3. Retained from explicitly set options of bs
1837 * 4. Inherited from parent node
1838 * 5. Retained from effective options of bs
1841 if (!parent_options) {
1843 * Any setting represented by flags is always updated. If the
1844 * corresponding QDict option is set, it takes precedence. Otherwise
1845 * the flag is translated into a QDict option. The old setting of bs is
1846 * not considered.
1848 update_options_from_flags(options, flags);
1851 /* Old explicitly set values (don't overwrite by inherited value) */
1852 old_options = qdict_clone_shallow(bs->explicit_options);
1853 bdrv_join_options(bs, options, old_options);
1854 QDECREF(old_options);
1856 explicit_options = qdict_clone_shallow(options);
1858 /* Inherit from parent node */
1859 if (parent_options) {
1860 assert(!flags);
1861 role->inherit_options(&flags, options, parent_flags, parent_options);
1864 /* Old values are used for options that aren't set yet */
1865 old_options = qdict_clone_shallow(bs->options);
1866 bdrv_join_options(bs, options, old_options);
1867 QDECREF(old_options);
1869 /* bdrv_open() masks this flag out */
1870 flags &= ~BDRV_O_PROTOCOL;
1872 QLIST_FOREACH(child, &bs->children, next) {
1873 QDict *new_child_options;
1874 char *child_key_dot;
1876 /* reopen can only change the options of block devices that were
1877 * implicitly created and inherited options. For other (referenced)
1878 * block devices, a syntax like "backing.foo" results in an error. */
1879 if (child->bs->inherits_from != bs) {
1880 continue;
1883 child_key_dot = g_strdup_printf("%s.", child->name);
1884 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
1885 g_free(child_key_dot);
1887 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
1888 child->role, options, flags);
1891 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1892 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1894 bs_entry->state.bs = bs;
1895 bs_entry->state.options = options;
1896 bs_entry->state.explicit_options = explicit_options;
1897 bs_entry->state.flags = flags;
1899 return bs_queue;
1902 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1903 BlockDriverState *bs,
1904 QDict *options, int flags)
1906 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
1907 NULL, NULL, 0);
1911 * Reopen multiple BlockDriverStates atomically & transactionally.
1913 * The queue passed in (bs_queue) must have been built up previous
1914 * via bdrv_reopen_queue().
1916 * Reopens all BDS specified in the queue, with the appropriate
1917 * flags. All devices are prepared for reopen, and failure of any
1918 * device will cause all device changes to be abandonded, and intermediate
1919 * data cleaned up.
1921 * If all devices prepare successfully, then the changes are committed
1922 * to all devices.
1925 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1927 int ret = -1;
1928 BlockReopenQueueEntry *bs_entry, *next;
1929 Error *local_err = NULL;
1931 assert(bs_queue != NULL);
1933 bdrv_drain_all();
1935 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1936 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1937 error_propagate(errp, local_err);
1938 goto cleanup;
1940 bs_entry->prepared = true;
1943 /* If we reach this point, we have success and just need to apply the
1944 * changes
1946 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1947 bdrv_reopen_commit(&bs_entry->state);
1950 ret = 0;
1952 cleanup:
1953 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1954 if (ret && bs_entry->prepared) {
1955 bdrv_reopen_abort(&bs_entry->state);
1956 } else if (ret) {
1957 QDECREF(bs_entry->state.explicit_options);
1959 QDECREF(bs_entry->state.options);
1960 g_free(bs_entry);
1962 g_free(bs_queue);
1963 return ret;
1967 /* Reopen a single BlockDriverState with the specified flags. */
1968 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1970 int ret = -1;
1971 Error *local_err = NULL;
1972 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
1974 ret = bdrv_reopen_multiple(queue, &local_err);
1975 if (local_err != NULL) {
1976 error_propagate(errp, local_err);
1978 return ret;
1983 * Prepares a BlockDriverState for reopen. All changes are staged in the
1984 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1985 * the block driver layer .bdrv_reopen_prepare()
1987 * bs is the BlockDriverState to reopen
1988 * flags are the new open flags
1989 * queue is the reopen queue
1991 * Returns 0 on success, non-zero on error. On error errp will be set
1992 * as well.
1994 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1995 * It is the responsibility of the caller to then call the abort() or
1996 * commit() for any other BDS that have been left in a prepare() state
1999 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2000 Error **errp)
2002 int ret = -1;
2003 Error *local_err = NULL;
2004 BlockDriver *drv;
2005 QemuOpts *opts;
2006 const char *value;
2008 assert(reopen_state != NULL);
2009 assert(reopen_state->bs->drv != NULL);
2010 drv = reopen_state->bs->drv;
2012 /* Process generic block layer options */
2013 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2014 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2015 if (local_err) {
2016 error_propagate(errp, local_err);
2017 ret = -EINVAL;
2018 goto error;
2021 update_flags_from_options(&reopen_state->flags, opts);
2023 /* node-name and driver must be unchanged. Put them back into the QDict, so
2024 * that they are checked at the end of this function. */
2025 value = qemu_opt_get(opts, "node-name");
2026 if (value) {
2027 qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
2030 value = qemu_opt_get(opts, "driver");
2031 if (value) {
2032 qdict_put(reopen_state->options, "driver", qstring_from_str(value));
2035 /* if we are to stay read-only, do not allow permission change
2036 * to r/w */
2037 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
2038 reopen_state->flags & BDRV_O_RDWR) {
2039 error_setg(errp, "Node '%s' is read only",
2040 bdrv_get_device_or_node_name(reopen_state->bs));
2041 goto error;
2045 ret = bdrv_flush(reopen_state->bs);
2046 if (ret) {
2047 error_setg_errno(errp, -ret, "Error flushing drive");
2048 goto error;
2051 if (drv->bdrv_reopen_prepare) {
2052 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2053 if (ret) {
2054 if (local_err != NULL) {
2055 error_propagate(errp, local_err);
2056 } else {
2057 error_setg(errp, "failed while preparing to reopen image '%s'",
2058 reopen_state->bs->filename);
2060 goto error;
2062 } else {
2063 /* It is currently mandatory to have a bdrv_reopen_prepare()
2064 * handler for each supported drv. */
2065 error_setg(errp, "Block format '%s' used by node '%s' "
2066 "does not support reopening files", drv->format_name,
2067 bdrv_get_device_or_node_name(reopen_state->bs));
2068 ret = -1;
2069 goto error;
2072 /* Options that are not handled are only okay if they are unchanged
2073 * compared to the old state. It is expected that some options are only
2074 * used for the initial open, but not reopen (e.g. filename) */
2075 if (qdict_size(reopen_state->options)) {
2076 const QDictEntry *entry = qdict_first(reopen_state->options);
2078 do {
2079 QString *new_obj = qobject_to_qstring(entry->value);
2080 const char *new = qstring_get_str(new_obj);
2081 const char *old = qdict_get_try_str(reopen_state->bs->options,
2082 entry->key);
2084 if (!old || strcmp(new, old)) {
2085 error_setg(errp, "Cannot change the option '%s'", entry->key);
2086 ret = -EINVAL;
2087 goto error;
2089 } while ((entry = qdict_next(reopen_state->options, entry)));
2092 ret = 0;
2094 error:
2095 qemu_opts_del(opts);
2096 return ret;
2100 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2101 * makes them final by swapping the staging BlockDriverState contents into
2102 * the active BlockDriverState contents.
2104 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2106 BlockDriver *drv;
2108 assert(reopen_state != NULL);
2109 drv = reopen_state->bs->drv;
2110 assert(drv != NULL);
2112 /* If there are any driver level actions to take */
2113 if (drv->bdrv_reopen_commit) {
2114 drv->bdrv_reopen_commit(reopen_state);
2117 /* set BDS specific flags now */
2118 QDECREF(reopen_state->bs->explicit_options);
2120 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2121 reopen_state->bs->open_flags = reopen_state->flags;
2122 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2124 bdrv_refresh_limits(reopen_state->bs, NULL);
2128 * Abort the reopen, and delete and free the staged changes in
2129 * reopen_state
2131 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2133 BlockDriver *drv;
2135 assert(reopen_state != NULL);
2136 drv = reopen_state->bs->drv;
2137 assert(drv != NULL);
2139 if (drv->bdrv_reopen_abort) {
2140 drv->bdrv_reopen_abort(reopen_state);
2143 QDECREF(reopen_state->explicit_options);
2147 static void bdrv_close(BlockDriverState *bs)
2149 BdrvAioNotifier *ban, *ban_next;
2151 assert(!bs->job);
2152 assert(!bs->refcnt);
2154 bdrv_drained_begin(bs); /* complete I/O */
2155 bdrv_flush(bs);
2156 bdrv_drain(bs); /* in case flush left pending I/O */
2158 bdrv_release_named_dirty_bitmaps(bs);
2159 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2161 if (bs->drv) {
2162 BdrvChild *child, *next;
2164 bs->drv->bdrv_close(bs);
2165 bs->drv = NULL;
2167 bdrv_set_backing_hd(bs, NULL);
2169 if (bs->file != NULL) {
2170 bdrv_unref_child(bs, bs->file);
2171 bs->file = NULL;
2174 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2175 /* TODO Remove bdrv_unref() from drivers' close function and use
2176 * bdrv_unref_child() here */
2177 if (child->bs->inherits_from == bs) {
2178 child->bs->inherits_from = NULL;
2180 bdrv_detach_child(child);
2183 g_free(bs->opaque);
2184 bs->opaque = NULL;
2185 bs->copy_on_read = 0;
2186 bs->backing_file[0] = '\0';
2187 bs->backing_format[0] = '\0';
2188 bs->total_sectors = 0;
2189 bs->encrypted = false;
2190 bs->valid_key = false;
2191 bs->sg = false;
2192 QDECREF(bs->options);
2193 QDECREF(bs->explicit_options);
2194 bs->options = NULL;
2195 QDECREF(bs->full_open_options);
2196 bs->full_open_options = NULL;
2199 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2200 g_free(ban);
2202 QLIST_INIT(&bs->aio_notifiers);
2203 bdrv_drained_end(bs);
2206 void bdrv_close_all(void)
2208 block_job_cancel_sync_all();
2210 /* Drop references from requests still in flight, such as canceled block
2211 * jobs whose AIO context has not been polled yet */
2212 bdrv_drain_all();
2214 blk_remove_all_bs();
2215 blockdev_close_all_bdrv_states();
2217 assert(QTAILQ_EMPTY(&all_bdrv_states));
2220 static void change_parent_backing_link(BlockDriverState *from,
2221 BlockDriverState *to)
2223 BdrvChild *c, *next, *to_c;
2225 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
2226 if (c->role == &child_backing) {
2227 /* @from is generally not allowed to be a backing file, except for
2228 * when @to is the overlay. In that case, @from may not be replaced
2229 * by @to as @to's backing node. */
2230 QLIST_FOREACH(to_c, &to->children, next) {
2231 if (to_c == c) {
2232 break;
2235 if (to_c) {
2236 continue;
2240 assert(c->role != &child_backing);
2241 bdrv_ref(to);
2242 bdrv_replace_child(c, to);
2243 bdrv_unref(from);
2248 * Add new bs contents at the top of an image chain while the chain is
2249 * live, while keeping required fields on the top layer.
2251 * This will modify the BlockDriverState fields, and swap contents
2252 * between bs_new and bs_top. Both bs_new and bs_top are modified.
2254 * bs_new must not be attached to a BlockBackend.
2256 * This function does not create any image files.
2258 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
2259 * that's what the callers commonly need. bs_new will be referenced by the old
2260 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
2261 * reference of its own, it must call bdrv_ref().
2263 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2265 assert(!bdrv_requests_pending(bs_top));
2266 assert(!bdrv_requests_pending(bs_new));
2268 bdrv_ref(bs_top);
2270 change_parent_backing_link(bs_top, bs_new);
2271 bdrv_set_backing_hd(bs_new, bs_top);
2272 bdrv_unref(bs_top);
2274 /* bs_new is now referenced by its new parents, we don't need the
2275 * additional reference any more. */
2276 bdrv_unref(bs_new);
2279 void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new)
2281 assert(!bdrv_requests_pending(old));
2282 assert(!bdrv_requests_pending(new));
2284 bdrv_ref(old);
2286 change_parent_backing_link(old, new);
2288 bdrv_unref(old);
2291 static void bdrv_delete(BlockDriverState *bs)
2293 assert(!bs->job);
2294 assert(bdrv_op_blocker_is_empty(bs));
2295 assert(!bs->refcnt);
2297 bdrv_close(bs);
2299 /* remove from list, if necessary */
2300 if (bs->node_name[0] != '\0') {
2301 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
2303 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
2305 g_free(bs);
2309 * Run consistency checks on an image
2311 * Returns 0 if the check could be completed (it doesn't mean that the image is
2312 * free of errors) or -errno when an internal error occurred. The results of the
2313 * check are stored in res.
2315 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2317 if (bs->drv == NULL) {
2318 return -ENOMEDIUM;
2320 if (bs->drv->bdrv_check == NULL) {
2321 return -ENOTSUP;
2324 memset(res, 0, sizeof(*res));
2325 return bs->drv->bdrv_check(bs, res, fix);
2329 * Return values:
2330 * 0 - success
2331 * -EINVAL - backing format specified, but no file
2332 * -ENOSPC - can't update the backing file because no space is left in the
2333 * image file header
2334 * -ENOTSUP - format driver doesn't support changing the backing file
2336 int bdrv_change_backing_file(BlockDriverState *bs,
2337 const char *backing_file, const char *backing_fmt)
2339 BlockDriver *drv = bs->drv;
2340 int ret;
2342 /* Backing file format doesn't make sense without a backing file */
2343 if (backing_fmt && !backing_file) {
2344 return -EINVAL;
2347 if (drv->bdrv_change_backing_file != NULL) {
2348 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2349 } else {
2350 ret = -ENOTSUP;
2353 if (ret == 0) {
2354 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2355 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2357 return ret;
2361 * Finds the image layer in the chain that has 'bs' as its backing file.
2363 * active is the current topmost image.
2365 * Returns NULL if bs is not found in active's image chain,
2366 * or if active == bs.
2368 * Returns the bottommost base image if bs == NULL.
2370 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2371 BlockDriverState *bs)
2373 while (active && bs != backing_bs(active)) {
2374 active = backing_bs(active);
2377 return active;
2380 /* Given a BDS, searches for the base layer. */
2381 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
2383 return bdrv_find_overlay(bs, NULL);
2387 * Drops images above 'base' up to and including 'top', and sets the image
2388 * above 'top' to have base as its backing file.
2390 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2391 * information in 'bs' can be properly updated.
2393 * E.g., this will convert the following chain:
2394 * bottom <- base <- intermediate <- top <- active
2396 * to
2398 * bottom <- base <- active
2400 * It is allowed for bottom==base, in which case it converts:
2402 * base <- intermediate <- top <- active
2404 * to
2406 * base <- active
2408 * If backing_file_str is non-NULL, it will be used when modifying top's
2409 * overlay image metadata.
2411 * Error conditions:
2412 * if active == top, that is considered an error
2415 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2416 BlockDriverState *base, const char *backing_file_str)
2418 BlockDriverState *new_top_bs = NULL;
2419 int ret = -EIO;
2421 if (!top->drv || !base->drv) {
2422 goto exit;
2425 new_top_bs = bdrv_find_overlay(active, top);
2427 if (new_top_bs == NULL) {
2428 /* we could not find the image above 'top', this is an error */
2429 goto exit;
2432 /* special case of new_top_bs->backing->bs already pointing to base - nothing
2433 * to do, no intermediate images */
2434 if (backing_bs(new_top_bs) == base) {
2435 ret = 0;
2436 goto exit;
2439 /* Make sure that base is in the backing chain of top */
2440 if (!bdrv_chain_contains(top, base)) {
2441 goto exit;
2444 /* success - we can delete the intermediate states, and link top->base */
2445 backing_file_str = backing_file_str ? backing_file_str : base->filename;
2446 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
2447 base->drv ? base->drv->format_name : "");
2448 if (ret) {
2449 goto exit;
2451 bdrv_set_backing_hd(new_top_bs, base);
2453 ret = 0;
2454 exit:
2455 return ret;
2459 * Truncate file to 'offset' bytes (needed only for file protocols)
2461 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2463 BlockDriver *drv = bs->drv;
2464 int ret;
2465 if (!drv)
2466 return -ENOMEDIUM;
2467 if (!drv->bdrv_truncate)
2468 return -ENOTSUP;
2469 if (bs->read_only)
2470 return -EACCES;
2472 ret = drv->bdrv_truncate(bs, offset);
2473 if (ret == 0) {
2474 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2475 bdrv_dirty_bitmap_truncate(bs);
2476 bdrv_parent_cb_resize(bs);
2477 ++bs->write_gen;
2479 return ret;
2483 * Length of a allocated file in bytes. Sparse files are counted by actual
2484 * allocated space. Return < 0 if error or unknown.
2486 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2488 BlockDriver *drv = bs->drv;
2489 if (!drv) {
2490 return -ENOMEDIUM;
2492 if (drv->bdrv_get_allocated_file_size) {
2493 return drv->bdrv_get_allocated_file_size(bs);
2495 if (bs->file) {
2496 return bdrv_get_allocated_file_size(bs->file->bs);
2498 return -ENOTSUP;
2502 * Return number of sectors on success, -errno on error.
2504 int64_t bdrv_nb_sectors(BlockDriverState *bs)
2506 BlockDriver *drv = bs->drv;
2508 if (!drv)
2509 return -ENOMEDIUM;
2511 if (drv->has_variable_length) {
2512 int ret = refresh_total_sectors(bs, bs->total_sectors);
2513 if (ret < 0) {
2514 return ret;
2517 return bs->total_sectors;
2521 * Return length in bytes on success, -errno on error.
2522 * The length is always a multiple of BDRV_SECTOR_SIZE.
2524 int64_t bdrv_getlength(BlockDriverState *bs)
2526 int64_t ret = bdrv_nb_sectors(bs);
2528 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
2529 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
2532 /* return 0 as number of sectors if no device present or error */
2533 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2535 int64_t nb_sectors = bdrv_nb_sectors(bs);
2537 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
2540 bool bdrv_is_read_only(BlockDriverState *bs)
2542 return bs->read_only;
2545 bool bdrv_is_sg(BlockDriverState *bs)
2547 return bs->sg;
2550 bool bdrv_is_encrypted(BlockDriverState *bs)
2552 if (bs->backing && bs->backing->bs->encrypted) {
2553 return true;
2555 return bs->encrypted;
2558 bool bdrv_key_required(BlockDriverState *bs)
2560 BdrvChild *backing = bs->backing;
2562 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
2563 return true;
2565 return (bs->encrypted && !bs->valid_key);
2568 int bdrv_set_key(BlockDriverState *bs, const char *key)
2570 int ret;
2571 if (bs->backing && bs->backing->bs->encrypted) {
2572 ret = bdrv_set_key(bs->backing->bs, key);
2573 if (ret < 0)
2574 return ret;
2575 if (!bs->encrypted)
2576 return 0;
2578 if (!bs->encrypted) {
2579 return -EINVAL;
2580 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2581 return -ENOMEDIUM;
2583 ret = bs->drv->bdrv_set_key(bs, key);
2584 if (ret < 0) {
2585 bs->valid_key = false;
2586 } else if (!bs->valid_key) {
2587 /* call the change callback now, we skipped it on open */
2588 bs->valid_key = true;
2589 bdrv_parent_cb_change_media(bs, true);
2591 return ret;
2595 * Provide an encryption key for @bs.
2596 * If @key is non-null:
2597 * If @bs is not encrypted, fail.
2598 * Else if the key is invalid, fail.
2599 * Else set @bs's key to @key, replacing the existing key, if any.
2600 * If @key is null:
2601 * If @bs is encrypted and still lacks a key, fail.
2602 * Else do nothing.
2603 * On failure, store an error object through @errp if non-null.
2605 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
2607 if (key) {
2608 if (!bdrv_is_encrypted(bs)) {
2609 error_setg(errp, "Node '%s' is not encrypted",
2610 bdrv_get_device_or_node_name(bs));
2611 } else if (bdrv_set_key(bs, key) < 0) {
2612 error_setg(errp, QERR_INVALID_PASSWORD);
2614 } else {
2615 if (bdrv_key_required(bs)) {
2616 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
2617 "'%s' (%s) is encrypted",
2618 bdrv_get_device_or_node_name(bs),
2619 bdrv_get_encrypted_filename(bs));
2624 const char *bdrv_get_format_name(BlockDriverState *bs)
2626 return bs->drv ? bs->drv->format_name : NULL;
2629 static int qsort_strcmp(const void *a, const void *b)
2631 return strcmp(a, b);
2634 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2635 void *opaque)
2637 BlockDriver *drv;
2638 int count = 0;
2639 int i;
2640 const char **formats = NULL;
2642 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2643 if (drv->format_name) {
2644 bool found = false;
2645 int i = count;
2646 while (formats && i && !found) {
2647 found = !strcmp(formats[--i], drv->format_name);
2650 if (!found) {
2651 formats = g_renew(const char *, formats, count + 1);
2652 formats[count++] = drv->format_name;
2657 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
2659 for (i = 0; i < count; i++) {
2660 it(opaque, formats[i]);
2663 g_free(formats);
2666 /* This function is to find a node in the bs graph */
2667 BlockDriverState *bdrv_find_node(const char *node_name)
2669 BlockDriverState *bs;
2671 assert(node_name);
2673 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2674 if (!strcmp(node_name, bs->node_name)) {
2675 return bs;
2678 return NULL;
2681 /* Put this QMP function here so it can access the static graph_bdrv_states. */
2682 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
2684 BlockDeviceInfoList *list, *entry;
2685 BlockDriverState *bs;
2687 list = NULL;
2688 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2689 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
2690 if (!info) {
2691 qapi_free_BlockDeviceInfoList(list);
2692 return NULL;
2694 entry = g_malloc0(sizeof(*entry));
2695 entry->value = info;
2696 entry->next = list;
2697 list = entry;
2700 return list;
2703 BlockDriverState *bdrv_lookup_bs(const char *device,
2704 const char *node_name,
2705 Error **errp)
2707 BlockBackend *blk;
2708 BlockDriverState *bs;
2710 if (device) {
2711 blk = blk_by_name(device);
2713 if (blk) {
2714 bs = blk_bs(blk);
2715 if (!bs) {
2716 error_setg(errp, "Device '%s' has no medium", device);
2719 return bs;
2723 if (node_name) {
2724 bs = bdrv_find_node(node_name);
2726 if (bs) {
2727 return bs;
2731 error_setg(errp, "Cannot find device=%s nor node_name=%s",
2732 device ? device : "",
2733 node_name ? node_name : "");
2734 return NULL;
2737 /* If 'base' is in the same chain as 'top', return true. Otherwise,
2738 * return false. If either argument is NULL, return false. */
2739 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
2741 while (top && top != base) {
2742 top = backing_bs(top);
2745 return top != NULL;
2748 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
2750 if (!bs) {
2751 return QTAILQ_FIRST(&graph_bdrv_states);
2753 return QTAILQ_NEXT(bs, node_list);
2756 const char *bdrv_get_node_name(const BlockDriverState *bs)
2758 return bs->node_name;
2761 const char *bdrv_get_parent_name(const BlockDriverState *bs)
2763 BdrvChild *c;
2764 const char *name;
2766 /* If multiple parents have a name, just pick the first one. */
2767 QLIST_FOREACH(c, &bs->parents, next_parent) {
2768 if (c->role->get_name) {
2769 name = c->role->get_name(c);
2770 if (name && *name) {
2771 return name;
2776 return NULL;
2779 /* TODO check what callers really want: bs->node_name or blk_name() */
2780 const char *bdrv_get_device_name(const BlockDriverState *bs)
2782 return bdrv_get_parent_name(bs) ?: "";
2785 /* This can be used to identify nodes that might not have a device
2786 * name associated. Since node and device names live in the same
2787 * namespace, the result is unambiguous. The exception is if both are
2788 * absent, then this returns an empty (non-null) string. */
2789 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
2791 return bdrv_get_parent_name(bs) ?: bs->node_name;
2794 int bdrv_get_flags(BlockDriverState *bs)
2796 return bs->open_flags;
2799 int bdrv_has_zero_init_1(BlockDriverState *bs)
2801 return 1;
2804 int bdrv_has_zero_init(BlockDriverState *bs)
2806 assert(bs->drv);
2808 /* If BS is a copy on write image, it is initialized to
2809 the contents of the base image, which may not be zeroes. */
2810 if (bs->backing) {
2811 return 0;
2813 if (bs->drv->bdrv_has_zero_init) {
2814 return bs->drv->bdrv_has_zero_init(bs);
2817 /* safe default */
2818 return 0;
2821 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
2823 BlockDriverInfo bdi;
2825 if (bs->backing) {
2826 return false;
2829 if (bdrv_get_info(bs, &bdi) == 0) {
2830 return bdi.unallocated_blocks_are_zero;
2833 return false;
2836 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
2838 BlockDriverInfo bdi;
2840 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2841 return false;
2844 if (bdrv_get_info(bs, &bdi) == 0) {
2845 return bdi.can_write_zeroes_with_unmap;
2848 return false;
2851 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2853 if (bs->backing && bs->backing->bs->encrypted)
2854 return bs->backing_file;
2855 else if (bs->encrypted)
2856 return bs->filename;
2857 else
2858 return NULL;
2861 void bdrv_get_backing_filename(BlockDriverState *bs,
2862 char *filename, int filename_size)
2864 pstrcpy(filename, filename_size, bs->backing_file);
2867 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2869 BlockDriver *drv = bs->drv;
2870 if (!drv)
2871 return -ENOMEDIUM;
2872 if (!drv->bdrv_get_info)
2873 return -ENOTSUP;
2874 memset(bdi, 0, sizeof(*bdi));
2875 return drv->bdrv_get_info(bs, bdi);
2878 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
2880 BlockDriver *drv = bs->drv;
2881 if (drv && drv->bdrv_get_specific_info) {
2882 return drv->bdrv_get_specific_info(bs);
2884 return NULL;
2887 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
2889 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
2890 return;
2893 bs->drv->bdrv_debug_event(bs, event);
2896 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
2897 const char *tag)
2899 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
2900 bs = bs->file ? bs->file->bs : NULL;
2903 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
2904 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
2907 return -ENOTSUP;
2910 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
2912 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
2913 bs = bs->file ? bs->file->bs : NULL;
2916 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
2917 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
2920 return -ENOTSUP;
2923 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
2925 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
2926 bs = bs->file ? bs->file->bs : NULL;
2929 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
2930 return bs->drv->bdrv_debug_resume(bs, tag);
2933 return -ENOTSUP;
2936 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
2938 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
2939 bs = bs->file ? bs->file->bs : NULL;
2942 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
2943 return bs->drv->bdrv_debug_is_suspended(bs, tag);
2946 return false;
2949 int bdrv_is_snapshot(BlockDriverState *bs)
2951 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2954 /* backing_file can either be relative, or absolute, or a protocol. If it is
2955 * relative, it must be relative to the chain. So, passing in bs->filename
2956 * from a BDS as backing_file should not be done, as that may be relative to
2957 * the CWD rather than the chain. */
2958 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2959 const char *backing_file)
2961 char *filename_full = NULL;
2962 char *backing_file_full = NULL;
2963 char *filename_tmp = NULL;
2964 int is_protocol = 0;
2965 BlockDriverState *curr_bs = NULL;
2966 BlockDriverState *retval = NULL;
2968 if (!bs || !bs->drv || !backing_file) {
2969 return NULL;
2972 filename_full = g_malloc(PATH_MAX);
2973 backing_file_full = g_malloc(PATH_MAX);
2974 filename_tmp = g_malloc(PATH_MAX);
2976 is_protocol = path_has_protocol(backing_file);
2978 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
2980 /* If either of the filename paths is actually a protocol, then
2981 * compare unmodified paths; otherwise make paths relative */
2982 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
2983 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
2984 retval = curr_bs->backing->bs;
2985 break;
2987 } else {
2988 /* If not an absolute filename path, make it relative to the current
2989 * image's filename path */
2990 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
2991 backing_file);
2993 /* We are going to compare absolute pathnames */
2994 if (!realpath(filename_tmp, filename_full)) {
2995 continue;
2998 /* We need to make sure the backing filename we are comparing against
2999 * is relative to the current image filename (or absolute) */
3000 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3001 curr_bs->backing_file);
3003 if (!realpath(filename_tmp, backing_file_full)) {
3004 continue;
3007 if (strcmp(backing_file_full, filename_full) == 0) {
3008 retval = curr_bs->backing->bs;
3009 break;
3014 g_free(filename_full);
3015 g_free(backing_file_full);
3016 g_free(filename_tmp);
3017 return retval;
3020 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3022 if (!bs->drv) {
3023 return 0;
3026 if (!bs->backing) {
3027 return 0;
3030 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3033 void bdrv_init(void)
3035 module_call_init(MODULE_INIT_BLOCK);
3038 void bdrv_init_with_whitelist(void)
3040 use_bdrv_whitelist = 1;
3041 bdrv_init();
3044 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3046 BdrvChild *child;
3047 Error *local_err = NULL;
3048 int ret;
3050 if (!bs->drv) {
3051 return;
3054 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3055 return;
3057 bs->open_flags &= ~BDRV_O_INACTIVE;
3059 if (bs->drv->bdrv_invalidate_cache) {
3060 bs->drv->bdrv_invalidate_cache(bs, &local_err);
3061 if (local_err) {
3062 bs->open_flags |= BDRV_O_INACTIVE;
3063 error_propagate(errp, local_err);
3064 return;
3068 QLIST_FOREACH(child, &bs->children, next) {
3069 bdrv_invalidate_cache(child->bs, &local_err);
3070 if (local_err) {
3071 bs->open_flags |= BDRV_O_INACTIVE;
3072 error_propagate(errp, local_err);
3073 return;
3077 ret = refresh_total_sectors(bs, bs->total_sectors);
3078 if (ret < 0) {
3079 bs->open_flags |= BDRV_O_INACTIVE;
3080 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3081 return;
3085 void bdrv_invalidate_cache_all(Error **errp)
3087 BlockDriverState *bs;
3088 Error *local_err = NULL;
3089 BdrvNextIterator it;
3091 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3092 AioContext *aio_context = bdrv_get_aio_context(bs);
3094 aio_context_acquire(aio_context);
3095 bdrv_invalidate_cache(bs, &local_err);
3096 aio_context_release(aio_context);
3097 if (local_err) {
3098 error_propagate(errp, local_err);
3099 return;
3104 static int bdrv_inactivate_recurse(BlockDriverState *bs,
3105 bool setting_flag)
3107 BdrvChild *child;
3108 int ret;
3110 if (!setting_flag && bs->drv->bdrv_inactivate) {
3111 ret = bs->drv->bdrv_inactivate(bs);
3112 if (ret < 0) {
3113 return ret;
3117 QLIST_FOREACH(child, &bs->children, next) {
3118 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
3119 if (ret < 0) {
3120 return ret;
3124 if (setting_flag) {
3125 bs->open_flags |= BDRV_O_INACTIVE;
3127 return 0;
3130 int bdrv_inactivate_all(void)
3132 BlockDriverState *bs = NULL;
3133 BdrvNextIterator it;
3134 int ret = 0;
3135 int pass;
3137 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3138 aio_context_acquire(bdrv_get_aio_context(bs));
3141 /* We do two passes of inactivation. The first pass calls to drivers'
3142 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
3143 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
3144 * is allowed. */
3145 for (pass = 0; pass < 2; pass++) {
3146 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3147 ret = bdrv_inactivate_recurse(bs, pass);
3148 if (ret < 0) {
3149 goto out;
3154 out:
3155 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3156 aio_context_release(bdrv_get_aio_context(bs));
3159 return ret;
3162 /**************************************************************/
3163 /* removable device support */
3166 * Return TRUE if the media is present
3168 bool bdrv_is_inserted(BlockDriverState *bs)
3170 BlockDriver *drv = bs->drv;
3171 BdrvChild *child;
3173 if (!drv) {
3174 return false;
3176 if (drv->bdrv_is_inserted) {
3177 return drv->bdrv_is_inserted(bs);
3179 QLIST_FOREACH(child, &bs->children, next) {
3180 if (!bdrv_is_inserted(child->bs)) {
3181 return false;
3184 return true;
3188 * Return whether the media changed since the last call to this
3189 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3191 int bdrv_media_changed(BlockDriverState *bs)
3193 BlockDriver *drv = bs->drv;
3195 if (drv && drv->bdrv_media_changed) {
3196 return drv->bdrv_media_changed(bs);
3198 return -ENOTSUP;
3202 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3204 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3206 BlockDriver *drv = bs->drv;
3207 const char *device_name;
3209 if (drv && drv->bdrv_eject) {
3210 drv->bdrv_eject(bs, eject_flag);
3213 device_name = bdrv_get_device_name(bs);
3214 if (device_name[0] != '\0') {
3215 qapi_event_send_device_tray_moved(device_name,
3216 eject_flag, &error_abort);
3221 * Lock or unlock the media (if it is locked, the user won't be able
3222 * to eject it manually).
3224 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3226 BlockDriver *drv = bs->drv;
3228 trace_bdrv_lock_medium(bs, locked);
3230 if (drv && drv->bdrv_lock_medium) {
3231 drv->bdrv_lock_medium(bs, locked);
3235 /* Get a reference to bs */
3236 void bdrv_ref(BlockDriverState *bs)
3238 bs->refcnt++;
3241 /* Release a previously grabbed reference to bs.
3242 * If after releasing, reference count is zero, the BlockDriverState is
3243 * deleted. */
3244 void bdrv_unref(BlockDriverState *bs)
3246 if (!bs) {
3247 return;
3249 assert(bs->refcnt > 0);
3250 if (--bs->refcnt == 0) {
3251 bdrv_delete(bs);
3255 struct BdrvOpBlocker {
3256 Error *reason;
3257 QLIST_ENTRY(BdrvOpBlocker) list;
3260 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
3262 BdrvOpBlocker *blocker;
3263 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3264 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
3265 blocker = QLIST_FIRST(&bs->op_blockers[op]);
3266 if (errp) {
3267 *errp = error_copy(blocker->reason);
3268 error_prepend(errp, "Node '%s' is busy: ",
3269 bdrv_get_device_or_node_name(bs));
3271 return true;
3273 return false;
3276 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
3278 BdrvOpBlocker *blocker;
3279 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3281 blocker = g_new0(BdrvOpBlocker, 1);
3282 blocker->reason = reason;
3283 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
3286 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
3288 BdrvOpBlocker *blocker, *next;
3289 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3290 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
3291 if (blocker->reason == reason) {
3292 QLIST_REMOVE(blocker, list);
3293 g_free(blocker);
3298 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
3300 int i;
3301 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3302 bdrv_op_block(bs, i, reason);
3306 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
3308 int i;
3309 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3310 bdrv_op_unblock(bs, i, reason);
3314 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
3316 int i;
3318 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3319 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
3320 return false;
3323 return true;
3326 void bdrv_img_create(const char *filename, const char *fmt,
3327 const char *base_filename, const char *base_fmt,
3328 char *options, uint64_t img_size, int flags,
3329 Error **errp, bool quiet)
3331 QemuOptsList *create_opts = NULL;
3332 QemuOpts *opts = NULL;
3333 const char *backing_fmt, *backing_file;
3334 int64_t size;
3335 BlockDriver *drv, *proto_drv;
3336 Error *local_err = NULL;
3337 int ret = 0;
3339 /* Find driver and parse its options */
3340 drv = bdrv_find_format(fmt);
3341 if (!drv) {
3342 error_setg(errp, "Unknown file format '%s'", fmt);
3343 return;
3346 proto_drv = bdrv_find_protocol(filename, true, errp);
3347 if (!proto_drv) {
3348 return;
3351 if (!drv->create_opts) {
3352 error_setg(errp, "Format driver '%s' does not support image creation",
3353 drv->format_name);
3354 return;
3357 if (!proto_drv->create_opts) {
3358 error_setg(errp, "Protocol driver '%s' does not support image creation",
3359 proto_drv->format_name);
3360 return;
3363 create_opts = qemu_opts_append(create_opts, drv->create_opts);
3364 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
3366 /* Create parameter list with default values */
3367 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
3368 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
3370 /* Parse -o options */
3371 if (options) {
3372 qemu_opts_do_parse(opts, options, NULL, &local_err);
3373 if (local_err) {
3374 error_report_err(local_err);
3375 local_err = NULL;
3376 error_setg(errp, "Invalid options for file format '%s'", fmt);
3377 goto out;
3381 if (base_filename) {
3382 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
3383 if (local_err) {
3384 error_setg(errp, "Backing file not supported for file format '%s'",
3385 fmt);
3386 goto out;
3390 if (base_fmt) {
3391 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
3392 if (local_err) {
3393 error_setg(errp, "Backing file format not supported for file "
3394 "format '%s'", fmt);
3395 goto out;
3399 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3400 if (backing_file) {
3401 if (!strcmp(filename, backing_file)) {
3402 error_setg(errp, "Error: Trying to create an image with the "
3403 "same filename as the backing file");
3404 goto out;
3408 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3410 // The size for the image must always be specified, with one exception:
3411 // If we are using a backing file, we can obtain the size from there
3412 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3413 if (size == -1) {
3414 if (backing_file) {
3415 BlockDriverState *bs;
3416 char *full_backing = g_new0(char, PATH_MAX);
3417 int64_t size;
3418 int back_flags;
3419 QDict *backing_options = NULL;
3421 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
3422 full_backing, PATH_MAX,
3423 &local_err);
3424 if (local_err) {
3425 g_free(full_backing);
3426 goto out;
3429 /* backing files always opened read-only */
3430 back_flags = flags;
3431 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
3433 if (backing_fmt) {
3434 backing_options = qdict_new();
3435 qdict_put(backing_options, "driver",
3436 qstring_from_str(backing_fmt));
3439 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
3440 &local_err);
3441 g_free(full_backing);
3442 if (!bs) {
3443 goto out;
3445 size = bdrv_getlength(bs);
3446 if (size < 0) {
3447 error_setg_errno(errp, -size, "Could not get size of '%s'",
3448 backing_file);
3449 bdrv_unref(bs);
3450 goto out;
3453 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
3455 bdrv_unref(bs);
3456 } else {
3457 error_setg(errp, "Image creation needs a size parameter");
3458 goto out;
3462 if (!quiet) {
3463 printf("Formatting '%s', fmt=%s ", filename, fmt);
3464 qemu_opts_print(opts, " ");
3465 puts("");
3468 ret = bdrv_create(drv, filename, opts, &local_err);
3470 if (ret == -EFBIG) {
3471 /* This is generally a better message than whatever the driver would
3472 * deliver (especially because of the cluster_size_hint), since that
3473 * is most probably not much different from "image too large". */
3474 const char *cluster_size_hint = "";
3475 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
3476 cluster_size_hint = " (try using a larger cluster size)";
3478 error_setg(errp, "The image size is too large for file format '%s'"
3479 "%s", fmt, cluster_size_hint);
3480 error_free(local_err);
3481 local_err = NULL;
3484 out:
3485 qemu_opts_del(opts);
3486 qemu_opts_free(create_opts);
3487 error_propagate(errp, local_err);
3490 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
3492 return bs->aio_context;
3495 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
3497 QLIST_REMOVE(ban, list);
3498 g_free(ban);
3501 void bdrv_detach_aio_context(BlockDriverState *bs)
3503 BdrvAioNotifier *baf, *baf_tmp;
3504 BdrvChild *child;
3506 if (!bs->drv) {
3507 return;
3510 assert(!bs->walking_aio_notifiers);
3511 bs->walking_aio_notifiers = true;
3512 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
3513 if (baf->deleted) {
3514 bdrv_do_remove_aio_context_notifier(baf);
3515 } else {
3516 baf->detach_aio_context(baf->opaque);
3519 /* Never mind iterating again to check for ->deleted. bdrv_close() will
3520 * remove remaining aio notifiers if we aren't called again.
3522 bs->walking_aio_notifiers = false;
3524 if (bs->drv->bdrv_detach_aio_context) {
3525 bs->drv->bdrv_detach_aio_context(bs);
3527 QLIST_FOREACH(child, &bs->children, next) {
3528 bdrv_detach_aio_context(child->bs);
3531 bs->aio_context = NULL;
3534 void bdrv_attach_aio_context(BlockDriverState *bs,
3535 AioContext *new_context)
3537 BdrvAioNotifier *ban, *ban_tmp;
3538 BdrvChild *child;
3540 if (!bs->drv) {
3541 return;
3544 bs->aio_context = new_context;
3546 QLIST_FOREACH(child, &bs->children, next) {
3547 bdrv_attach_aio_context(child->bs, new_context);
3549 if (bs->drv->bdrv_attach_aio_context) {
3550 bs->drv->bdrv_attach_aio_context(bs, new_context);
3553 assert(!bs->walking_aio_notifiers);
3554 bs->walking_aio_notifiers = true;
3555 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
3556 if (ban->deleted) {
3557 bdrv_do_remove_aio_context_notifier(ban);
3558 } else {
3559 ban->attached_aio_context(new_context, ban->opaque);
3562 bs->walking_aio_notifiers = false;
3565 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
3567 bdrv_drain(bs); /* ensure there are no in-flight requests */
3569 bdrv_detach_aio_context(bs);
3571 /* This function executes in the old AioContext so acquire the new one in
3572 * case it runs in a different thread.
3574 aio_context_acquire(new_context);
3575 bdrv_attach_aio_context(bs, new_context);
3576 aio_context_release(new_context);
3579 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
3580 void (*attached_aio_context)(AioContext *new_context, void *opaque),
3581 void (*detach_aio_context)(void *opaque), void *opaque)
3583 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
3584 *ban = (BdrvAioNotifier){
3585 .attached_aio_context = attached_aio_context,
3586 .detach_aio_context = detach_aio_context,
3587 .opaque = opaque
3590 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
3593 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
3594 void (*attached_aio_context)(AioContext *,
3595 void *),
3596 void (*detach_aio_context)(void *),
3597 void *opaque)
3599 BdrvAioNotifier *ban, *ban_next;
3601 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3602 if (ban->attached_aio_context == attached_aio_context &&
3603 ban->detach_aio_context == detach_aio_context &&
3604 ban->opaque == opaque &&
3605 ban->deleted == false)
3607 if (bs->walking_aio_notifiers) {
3608 ban->deleted = true;
3609 } else {
3610 bdrv_do_remove_aio_context_notifier(ban);
3612 return;
3616 abort();
3619 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
3620 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
3622 if (!bs->drv->bdrv_amend_options) {
3623 return -ENOTSUP;
3625 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
3628 /* This function will be called by the bdrv_recurse_is_first_non_filter method
3629 * of block filter and by bdrv_is_first_non_filter.
3630 * It is used to test if the given bs is the candidate or recurse more in the
3631 * node graph.
3633 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
3634 BlockDriverState *candidate)
3636 /* return false if basic checks fails */
3637 if (!bs || !bs->drv) {
3638 return false;
3641 /* the code reached a non block filter driver -> check if the bs is
3642 * the same as the candidate. It's the recursion termination condition.
3644 if (!bs->drv->is_filter) {
3645 return bs == candidate;
3647 /* Down this path the driver is a block filter driver */
3649 /* If the block filter recursion method is defined use it to recurse down
3650 * the node graph.
3652 if (bs->drv->bdrv_recurse_is_first_non_filter) {
3653 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
3656 /* the driver is a block filter but don't allow to recurse -> return false
3658 return false;
3661 /* This function checks if the candidate is the first non filter bs down it's
3662 * bs chain. Since we don't have pointers to parents it explore all bs chains
3663 * from the top. Some filters can choose not to pass down the recursion.
3665 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
3667 BlockDriverState *bs;
3668 BdrvNextIterator it;
3670 /* walk down the bs forest recursively */
3671 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3672 bool perm;
3674 /* try to recurse in this top level bs */
3675 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
3677 /* candidate is the first non filter */
3678 if (perm) {
3679 return true;
3683 return false;
3686 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
3687 const char *node_name, Error **errp)
3689 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
3690 AioContext *aio_context;
3692 if (!to_replace_bs) {
3693 error_setg(errp, "Node name '%s' not found", node_name);
3694 return NULL;
3697 aio_context = bdrv_get_aio_context(to_replace_bs);
3698 aio_context_acquire(aio_context);
3700 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
3701 to_replace_bs = NULL;
3702 goto out;
3705 /* We don't want arbitrary node of the BDS chain to be replaced only the top
3706 * most non filter in order to prevent data corruption.
3707 * Another benefit is that this tests exclude backing files which are
3708 * blocked by the backing blockers.
3710 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
3711 error_setg(errp, "Only top most non filter can be replaced");
3712 to_replace_bs = NULL;
3713 goto out;
3716 out:
3717 aio_context_release(aio_context);
3718 return to_replace_bs;
3721 static bool append_open_options(QDict *d, BlockDriverState *bs)
3723 const QDictEntry *entry;
3724 QemuOptDesc *desc;
3725 BdrvChild *child;
3726 bool found_any = false;
3727 const char *p;
3729 for (entry = qdict_first(bs->options); entry;
3730 entry = qdict_next(bs->options, entry))
3732 /* Exclude options for children */
3733 QLIST_FOREACH(child, &bs->children, next) {
3734 if (strstart(qdict_entry_key(entry), child->name, &p)
3735 && (!*p || *p == '.'))
3737 break;
3740 if (child) {
3741 continue;
3744 /* And exclude all non-driver-specific options */
3745 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
3746 if (!strcmp(qdict_entry_key(entry), desc->name)) {
3747 break;
3750 if (desc->name) {
3751 continue;
3754 qobject_incref(qdict_entry_value(entry));
3755 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
3756 found_any = true;
3759 return found_any;
3762 /* Updates the following BDS fields:
3763 * - exact_filename: A filename which may be used for opening a block device
3764 * which (mostly) equals the given BDS (even without any
3765 * other options; so reading and writing must return the same
3766 * results, but caching etc. may be different)
3767 * - full_open_options: Options which, when given when opening a block device
3768 * (without a filename), result in a BDS (mostly)
3769 * equalling the given one
3770 * - filename: If exact_filename is set, it is copied here. Otherwise,
3771 * full_open_options is converted to a JSON object, prefixed with
3772 * "json:" (for use through the JSON pseudo protocol) and put here.
3774 void bdrv_refresh_filename(BlockDriverState *bs)
3776 BlockDriver *drv = bs->drv;
3777 QDict *opts;
3779 if (!drv) {
3780 return;
3783 /* This BDS's file name will most probably depend on its file's name, so
3784 * refresh that first */
3785 if (bs->file) {
3786 bdrv_refresh_filename(bs->file->bs);
3789 if (drv->bdrv_refresh_filename) {
3790 /* Obsolete information is of no use here, so drop the old file name
3791 * information before refreshing it */
3792 bs->exact_filename[0] = '\0';
3793 if (bs->full_open_options) {
3794 QDECREF(bs->full_open_options);
3795 bs->full_open_options = NULL;
3798 opts = qdict_new();
3799 append_open_options(opts, bs);
3800 drv->bdrv_refresh_filename(bs, opts);
3801 QDECREF(opts);
3802 } else if (bs->file) {
3803 /* Try to reconstruct valid information from the underlying file */
3804 bool has_open_options;
3806 bs->exact_filename[0] = '\0';
3807 if (bs->full_open_options) {
3808 QDECREF(bs->full_open_options);
3809 bs->full_open_options = NULL;
3812 opts = qdict_new();
3813 has_open_options = append_open_options(opts, bs);
3815 /* If no specific options have been given for this BDS, the filename of
3816 * the underlying file should suffice for this one as well */
3817 if (bs->file->bs->exact_filename[0] && !has_open_options) {
3818 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
3820 /* Reconstructing the full options QDict is simple for most format block
3821 * drivers, as long as the full options are known for the underlying
3822 * file BDS. The full options QDict of that file BDS should somehow
3823 * contain a representation of the filename, therefore the following
3824 * suffices without querying the (exact_)filename of this BDS. */
3825 if (bs->file->bs->full_open_options) {
3826 qdict_put_obj(opts, "driver",
3827 QOBJECT(qstring_from_str(drv->format_name)));
3828 QINCREF(bs->file->bs->full_open_options);
3829 qdict_put_obj(opts, "file",
3830 QOBJECT(bs->file->bs->full_open_options));
3832 bs->full_open_options = opts;
3833 } else {
3834 QDECREF(opts);
3836 } else if (!bs->full_open_options && qdict_size(bs->options)) {
3837 /* There is no underlying file BDS (at least referenced by BDS.file),
3838 * so the full options QDict should be equal to the options given
3839 * specifically for this block device when it was opened (plus the
3840 * driver specification).
3841 * Because those options don't change, there is no need to update
3842 * full_open_options when it's already set. */
3844 opts = qdict_new();
3845 append_open_options(opts, bs);
3846 qdict_put_obj(opts, "driver",
3847 QOBJECT(qstring_from_str(drv->format_name)));
3849 if (bs->exact_filename[0]) {
3850 /* This may not work for all block protocol drivers (some may
3851 * require this filename to be parsed), but we have to find some
3852 * default solution here, so just include it. If some block driver
3853 * does not support pure options without any filename at all or
3854 * needs some special format of the options QDict, it needs to
3855 * implement the driver-specific bdrv_refresh_filename() function.
3857 qdict_put_obj(opts, "filename",
3858 QOBJECT(qstring_from_str(bs->exact_filename)));
3861 bs->full_open_options = opts;
3864 if (bs->exact_filename[0]) {
3865 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
3866 } else if (bs->full_open_options) {
3867 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
3868 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
3869 qstring_get_str(json));
3870 QDECREF(json);
3875 * Hot add/remove a BDS's child. So the user can take a child offline when
3876 * it is broken and take a new child online
3878 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
3879 Error **errp)
3882 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
3883 error_setg(errp, "The node %s does not support adding a child",
3884 bdrv_get_device_or_node_name(parent_bs));
3885 return;
3888 if (!QLIST_EMPTY(&child_bs->parents)) {
3889 error_setg(errp, "The node %s already has a parent",
3890 child_bs->node_name);
3891 return;
3894 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
3897 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
3899 BdrvChild *tmp;
3901 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
3902 error_setg(errp, "The node %s does not support removing a child",
3903 bdrv_get_device_or_node_name(parent_bs));
3904 return;
3907 QLIST_FOREACH(tmp, &parent_bs->children, next) {
3908 if (tmp == child) {
3909 break;
3913 if (!tmp) {
3914 error_setg(errp, "The node %s does not have a child named %s",
3915 bdrv_get_device_or_node_name(parent_bs),
3916 bdrv_get_device_or_node_name(child->bs));
3917 return;
3920 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);