hw/arm/raspi: Provide spin-loop code for AArch64 CPUs
[qemu.git] / block.c
blobf7f9d8eca74d6b7a000d7299767d289fb4e59bc4
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "qemu/error-report.h"
31 #include "module_block.h"
32 #include "qemu/module.h"
33 #include "qapi/error.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qjson.h"
36 #include "qapi/qmp/qstring.h"
37 #include "qapi/qobject-output-visitor.h"
38 #include "qapi/qapi-visit-block-core.h"
39 #include "sysemu/block-backend.h"
40 #include "sysemu/sysemu.h"
41 #include "qemu/notify.h"
42 #include "qemu/option.h"
43 #include "qemu/coroutine.h"
44 #include "block/qapi.h"
45 #include "qemu/timer.h"
46 #include "qemu/cutils.h"
47 #include "qemu/id.h"
49 #ifdef CONFIG_BSD
50 #include <sys/ioctl.h>
51 #include <sys/queue.h>
52 #ifndef __DragonFly__
53 #include <sys/disk.h>
54 #endif
55 #endif
57 #ifdef _WIN32
58 #include <windows.h>
59 #endif
61 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
63 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
64 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
66 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
69 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
70 QLIST_HEAD_INITIALIZER(bdrv_drivers);
72 static BlockDriverState *bdrv_open_inherit(const char *filename,
73 const char *reference,
74 QDict *options, int flags,
75 BlockDriverState *parent,
76 const BdrvChildRole *child_role,
77 Error **errp);
79 /* If non-zero, use only whitelisted block drivers */
80 static int use_bdrv_whitelist;
82 #ifdef _WIN32
83 static int is_windows_drive_prefix(const char *filename)
85 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
86 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
87 filename[1] == ':');
90 int is_windows_drive(const char *filename)
92 if (is_windows_drive_prefix(filename) &&
93 filename[2] == '\0')
94 return 1;
95 if (strstart(filename, "\\\\.\\", NULL) ||
96 strstart(filename, "//./", NULL))
97 return 1;
98 return 0;
100 #endif
102 size_t bdrv_opt_mem_align(BlockDriverState *bs)
104 if (!bs || !bs->drv) {
105 /* page size or 4k (hdd sector size) should be on the safe side */
106 return MAX(4096, getpagesize());
109 return bs->bl.opt_mem_alignment;
112 size_t bdrv_min_mem_align(BlockDriverState *bs)
114 if (!bs || !bs->drv) {
115 /* page size or 4k (hdd sector size) should be on the safe side */
116 return MAX(4096, getpagesize());
119 return bs->bl.min_mem_alignment;
122 /* check if the path starts with "<protocol>:" */
123 int path_has_protocol(const char *path)
125 const char *p;
127 #ifdef _WIN32
128 if (is_windows_drive(path) ||
129 is_windows_drive_prefix(path)) {
130 return 0;
132 p = path + strcspn(path, ":/\\");
133 #else
134 p = path + strcspn(path, ":/");
135 #endif
137 return *p == ':';
140 int path_is_absolute(const char *path)
142 #ifdef _WIN32
143 /* specific case for names like: "\\.\d:" */
144 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
145 return 1;
147 return (*path == '/' || *path == '\\');
148 #else
149 return (*path == '/');
150 #endif
153 /* if filename is absolute, just copy it to dest. Otherwise, build a
154 path to it by considering it is relative to base_path. URL are
155 supported. */
156 void path_combine(char *dest, int dest_size,
157 const char *base_path,
158 const char *filename)
160 const char *p, *p1;
161 int len;
163 if (dest_size <= 0)
164 return;
165 if (path_is_absolute(filename)) {
166 pstrcpy(dest, dest_size, filename);
167 } else {
168 const char *protocol_stripped = NULL;
170 if (path_has_protocol(base_path)) {
171 protocol_stripped = strchr(base_path, ':');
172 if (protocol_stripped) {
173 protocol_stripped++;
176 p = protocol_stripped ?: base_path;
178 p1 = strrchr(base_path, '/');
179 #ifdef _WIN32
181 const char *p2;
182 p2 = strrchr(base_path, '\\');
183 if (!p1 || p2 > p1)
184 p1 = p2;
186 #endif
187 if (p1)
188 p1++;
189 else
190 p1 = base_path;
191 if (p1 > p)
192 p = p1;
193 len = p - base_path;
194 if (len > dest_size - 1)
195 len = dest_size - 1;
196 memcpy(dest, base_path, len);
197 dest[len] = '\0';
198 pstrcat(dest, dest_size, filename);
203 * Helper function for bdrv_parse_filename() implementations to remove optional
204 * protocol prefixes (especially "file:") from a filename and for putting the
205 * stripped filename into the options QDict if there is such a prefix.
207 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
208 QDict *options)
210 if (strstart(filename, prefix, &filename)) {
211 /* Stripping the explicit protocol prefix may result in a protocol
212 * prefix being (wrongly) detected (if the filename contains a colon) */
213 if (path_has_protocol(filename)) {
214 QString *fat_filename;
216 /* This means there is some colon before the first slash; therefore,
217 * this cannot be an absolute path */
218 assert(!path_is_absolute(filename));
220 /* And we can thus fix the protocol detection issue by prefixing it
221 * by "./" */
222 fat_filename = qstring_from_str("./");
223 qstring_append(fat_filename, filename);
225 assert(!path_has_protocol(qstring_get_str(fat_filename)));
227 qdict_put(options, "filename", fat_filename);
228 } else {
229 /* If no protocol prefix was detected, we can use the shortened
230 * filename as-is */
231 qdict_put_str(options, "filename", filename);
237 /* Returns whether the image file is opened as read-only. Note that this can
238 * return false and writing to the image file is still not possible because the
239 * image is inactivated. */
240 bool bdrv_is_read_only(BlockDriverState *bs)
242 return bs->read_only;
245 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
246 bool ignore_allow_rdw, Error **errp)
248 /* Do not set read_only if copy_on_read is enabled */
249 if (bs->copy_on_read && read_only) {
250 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
251 bdrv_get_device_or_node_name(bs));
252 return -EINVAL;
255 /* Do not clear read_only if it is prohibited */
256 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
257 !ignore_allow_rdw)
259 error_setg(errp, "Node '%s' is read only",
260 bdrv_get_device_or_node_name(bs));
261 return -EPERM;
264 return 0;
267 /* TODO Remove (deprecated since 2.11)
268 * Block drivers are not supposed to automatically change bs->read_only.
269 * Instead, they should just check whether they can provide what the user
270 * explicitly requested and error out if read-write is requested, but they can
271 * only provide read-only access. */
272 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
274 int ret = 0;
276 ret = bdrv_can_set_read_only(bs, read_only, false, errp);
277 if (ret < 0) {
278 return ret;
281 bs->read_only = read_only;
282 return 0;
285 void bdrv_get_full_backing_filename_from_filename(const char *backed,
286 const char *backing,
287 char *dest, size_t sz,
288 Error **errp)
290 if (backing[0] == '\0' || path_has_protocol(backing) ||
291 path_is_absolute(backing))
293 pstrcpy(dest, sz, backing);
294 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
295 error_setg(errp, "Cannot use relative backing file names for '%s'",
296 backed);
297 } else {
298 path_combine(dest, sz, backed, backing);
302 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
303 Error **errp)
305 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
307 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
308 dest, sz, errp);
311 void bdrv_register(BlockDriver *bdrv)
313 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
316 BlockDriverState *bdrv_new(void)
318 BlockDriverState *bs;
319 int i;
321 bs = g_new0(BlockDriverState, 1);
322 QLIST_INIT(&bs->dirty_bitmaps);
323 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
324 QLIST_INIT(&bs->op_blockers[i]);
326 notifier_with_return_list_init(&bs->before_write_notifiers);
327 qemu_co_mutex_init(&bs->reqs_lock);
328 qemu_mutex_init(&bs->dirty_bitmap_mutex);
329 bs->refcnt = 1;
330 bs->aio_context = qemu_get_aio_context();
332 qemu_co_queue_init(&bs->flush_queue);
334 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
336 return bs;
339 static BlockDriver *bdrv_do_find_format(const char *format_name)
341 BlockDriver *drv1;
343 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
344 if (!strcmp(drv1->format_name, format_name)) {
345 return drv1;
349 return NULL;
352 BlockDriver *bdrv_find_format(const char *format_name)
354 BlockDriver *drv1;
355 int i;
357 drv1 = bdrv_do_find_format(format_name);
358 if (drv1) {
359 return drv1;
362 /* The driver isn't registered, maybe we need to load a module */
363 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
364 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
365 block_module_load_one(block_driver_modules[i].library_name);
366 break;
370 return bdrv_do_find_format(format_name);
373 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
375 static const char *whitelist_rw[] = {
376 CONFIG_BDRV_RW_WHITELIST
378 static const char *whitelist_ro[] = {
379 CONFIG_BDRV_RO_WHITELIST
381 const char **p;
383 if (!whitelist_rw[0] && !whitelist_ro[0]) {
384 return 1; /* no whitelist, anything goes */
387 for (p = whitelist_rw; *p; p++) {
388 if (!strcmp(drv->format_name, *p)) {
389 return 1;
392 if (read_only) {
393 for (p = whitelist_ro; *p; p++) {
394 if (!strcmp(drv->format_name, *p)) {
395 return 1;
399 return 0;
402 bool bdrv_uses_whitelist(void)
404 return use_bdrv_whitelist;
407 typedef struct CreateCo {
408 BlockDriver *drv;
409 char *filename;
410 QemuOpts *opts;
411 int ret;
412 Error *err;
413 } CreateCo;
415 static void coroutine_fn bdrv_create_co_entry(void *opaque)
417 Error *local_err = NULL;
418 int ret;
420 CreateCo *cco = opaque;
421 assert(cco->drv);
423 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
424 error_propagate(&cco->err, local_err);
425 cco->ret = ret;
428 int bdrv_create(BlockDriver *drv, const char* filename,
429 QemuOpts *opts, Error **errp)
431 int ret;
433 Coroutine *co;
434 CreateCo cco = {
435 .drv = drv,
436 .filename = g_strdup(filename),
437 .opts = opts,
438 .ret = NOT_DONE,
439 .err = NULL,
442 if (!drv->bdrv_co_create_opts) {
443 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
444 ret = -ENOTSUP;
445 goto out;
448 if (qemu_in_coroutine()) {
449 /* Fast-path if already in coroutine context */
450 bdrv_create_co_entry(&cco);
451 } else {
452 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
453 qemu_coroutine_enter(co);
454 while (cco.ret == NOT_DONE) {
455 aio_poll(qemu_get_aio_context(), true);
459 ret = cco.ret;
460 if (ret < 0) {
461 if (cco.err) {
462 error_propagate(errp, cco.err);
463 } else {
464 error_setg_errno(errp, -ret, "Could not create image");
468 out:
469 g_free(cco.filename);
470 return ret;
473 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
475 BlockDriver *drv;
476 Error *local_err = NULL;
477 int ret;
479 drv = bdrv_find_protocol(filename, true, errp);
480 if (drv == NULL) {
481 return -ENOENT;
484 ret = bdrv_create(drv, filename, opts, &local_err);
485 error_propagate(errp, local_err);
486 return ret;
490 * Try to get @bs's logical and physical block size.
491 * On success, store them in @bsz struct and return 0.
492 * On failure return -errno.
493 * @bs must not be empty.
495 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
497 BlockDriver *drv = bs->drv;
499 if (drv && drv->bdrv_probe_blocksizes) {
500 return drv->bdrv_probe_blocksizes(bs, bsz);
501 } else if (drv && drv->is_filter && bs->file) {
502 return bdrv_probe_blocksizes(bs->file->bs, bsz);
505 return -ENOTSUP;
509 * Try to get @bs's geometry (cyls, heads, sectors).
510 * On success, store them in @geo struct and return 0.
511 * On failure return -errno.
512 * @bs must not be empty.
514 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
516 BlockDriver *drv = bs->drv;
518 if (drv && drv->bdrv_probe_geometry) {
519 return drv->bdrv_probe_geometry(bs, geo);
520 } else if (drv && drv->is_filter && bs->file) {
521 return bdrv_probe_geometry(bs->file->bs, geo);
524 return -ENOTSUP;
528 * Create a uniquely-named empty temporary file.
529 * Return 0 upon success, otherwise a negative errno value.
531 int get_tmp_filename(char *filename, int size)
533 #ifdef _WIN32
534 char temp_dir[MAX_PATH];
535 /* GetTempFileName requires that its output buffer (4th param)
536 have length MAX_PATH or greater. */
537 assert(size >= MAX_PATH);
538 return (GetTempPath(MAX_PATH, temp_dir)
539 && GetTempFileName(temp_dir, "qem", 0, filename)
540 ? 0 : -GetLastError());
541 #else
542 int fd;
543 const char *tmpdir;
544 tmpdir = getenv("TMPDIR");
545 if (!tmpdir) {
546 tmpdir = "/var/tmp";
548 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
549 return -EOVERFLOW;
551 fd = mkstemp(filename);
552 if (fd < 0) {
553 return -errno;
555 if (close(fd) != 0) {
556 unlink(filename);
557 return -errno;
559 return 0;
560 #endif
564 * Detect host devices. By convention, /dev/cdrom[N] is always
565 * recognized as a host CDROM.
567 static BlockDriver *find_hdev_driver(const char *filename)
569 int score_max = 0, score;
570 BlockDriver *drv = NULL, *d;
572 QLIST_FOREACH(d, &bdrv_drivers, list) {
573 if (d->bdrv_probe_device) {
574 score = d->bdrv_probe_device(filename);
575 if (score > score_max) {
576 score_max = score;
577 drv = d;
582 return drv;
585 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
587 BlockDriver *drv1;
589 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
590 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
591 return drv1;
595 return NULL;
598 BlockDriver *bdrv_find_protocol(const char *filename,
599 bool allow_protocol_prefix,
600 Error **errp)
602 BlockDriver *drv1;
603 char protocol[128];
604 int len;
605 const char *p;
606 int i;
608 /* TODO Drivers without bdrv_file_open must be specified explicitly */
611 * XXX(hch): we really should not let host device detection
612 * override an explicit protocol specification, but moving this
613 * later breaks access to device names with colons in them.
614 * Thanks to the brain-dead persistent naming schemes on udev-
615 * based Linux systems those actually are quite common.
617 drv1 = find_hdev_driver(filename);
618 if (drv1) {
619 return drv1;
622 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
623 return &bdrv_file;
626 p = strchr(filename, ':');
627 assert(p != NULL);
628 len = p - filename;
629 if (len > sizeof(protocol) - 1)
630 len = sizeof(protocol) - 1;
631 memcpy(protocol, filename, len);
632 protocol[len] = '\0';
634 drv1 = bdrv_do_find_protocol(protocol);
635 if (drv1) {
636 return drv1;
639 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
640 if (block_driver_modules[i].protocol_name &&
641 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
642 block_module_load_one(block_driver_modules[i].library_name);
643 break;
647 drv1 = bdrv_do_find_protocol(protocol);
648 if (!drv1) {
649 error_setg(errp, "Unknown protocol '%s'", protocol);
651 return drv1;
655 * Guess image format by probing its contents.
656 * This is not a good idea when your image is raw (CVE-2008-2004), but
657 * we do it anyway for backward compatibility.
659 * @buf contains the image's first @buf_size bytes.
660 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
661 * but can be smaller if the image file is smaller)
662 * @filename is its filename.
664 * For all block drivers, call the bdrv_probe() method to get its
665 * probing score.
666 * Return the first block driver with the highest probing score.
668 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
669 const char *filename)
671 int score_max = 0, score;
672 BlockDriver *drv = NULL, *d;
674 QLIST_FOREACH(d, &bdrv_drivers, list) {
675 if (d->bdrv_probe) {
676 score = d->bdrv_probe(buf, buf_size, filename);
677 if (score > score_max) {
678 score_max = score;
679 drv = d;
684 return drv;
687 static int find_image_format(BlockBackend *file, const char *filename,
688 BlockDriver **pdrv, Error **errp)
690 BlockDriver *drv;
691 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
692 int ret = 0;
694 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
695 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
696 *pdrv = &bdrv_raw;
697 return ret;
700 ret = blk_pread(file, 0, buf, sizeof(buf));
701 if (ret < 0) {
702 error_setg_errno(errp, -ret, "Could not read image for determining its "
703 "format");
704 *pdrv = NULL;
705 return ret;
708 drv = bdrv_probe_all(buf, ret, filename);
709 if (!drv) {
710 error_setg(errp, "Could not determine image format: No compatible "
711 "driver found");
712 ret = -ENOENT;
714 *pdrv = drv;
715 return ret;
719 * Set the current 'total_sectors' value
720 * Return 0 on success, -errno on error.
722 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
724 BlockDriver *drv = bs->drv;
726 if (!drv) {
727 return -ENOMEDIUM;
730 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
731 if (bdrv_is_sg(bs))
732 return 0;
734 /* query actual device if possible, otherwise just trust the hint */
735 if (drv->bdrv_getlength) {
736 int64_t length = drv->bdrv_getlength(bs);
737 if (length < 0) {
738 return length;
740 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
743 bs->total_sectors = hint;
744 return 0;
748 * Combines a QDict of new block driver @options with any missing options taken
749 * from @old_options, so that leaving out an option defaults to its old value.
751 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
752 QDict *old_options)
754 if (bs->drv && bs->drv->bdrv_join_options) {
755 bs->drv->bdrv_join_options(options, old_options);
756 } else {
757 qdict_join(options, old_options, false);
762 * Set open flags for a given discard mode
764 * Return 0 on success, -1 if the discard mode was invalid.
766 int bdrv_parse_discard_flags(const char *mode, int *flags)
768 *flags &= ~BDRV_O_UNMAP;
770 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
771 /* do nothing */
772 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
773 *flags |= BDRV_O_UNMAP;
774 } else {
775 return -1;
778 return 0;
782 * Set open flags for a given cache mode
784 * Return 0 on success, -1 if the cache mode was invalid.
786 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
788 *flags &= ~BDRV_O_CACHE_MASK;
790 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
791 *writethrough = false;
792 *flags |= BDRV_O_NOCACHE;
793 } else if (!strcmp(mode, "directsync")) {
794 *writethrough = true;
795 *flags |= BDRV_O_NOCACHE;
796 } else if (!strcmp(mode, "writeback")) {
797 *writethrough = false;
798 } else if (!strcmp(mode, "unsafe")) {
799 *writethrough = false;
800 *flags |= BDRV_O_NO_FLUSH;
801 } else if (!strcmp(mode, "writethrough")) {
802 *writethrough = true;
803 } else {
804 return -1;
807 return 0;
810 static char *bdrv_child_get_parent_desc(BdrvChild *c)
812 BlockDriverState *parent = c->opaque;
813 return g_strdup(bdrv_get_device_or_node_name(parent));
816 static void bdrv_child_cb_drained_begin(BdrvChild *child)
818 BlockDriverState *bs = child->opaque;
819 bdrv_drained_begin(bs);
822 static void bdrv_child_cb_drained_end(BdrvChild *child)
824 BlockDriverState *bs = child->opaque;
825 bdrv_drained_end(bs);
828 static void bdrv_child_cb_attach(BdrvChild *child)
830 BlockDriverState *bs = child->opaque;
831 bdrv_apply_subtree_drain(child, bs);
834 static void bdrv_child_cb_detach(BdrvChild *child)
836 BlockDriverState *bs = child->opaque;
837 bdrv_unapply_subtree_drain(child, bs);
840 static int bdrv_child_cb_inactivate(BdrvChild *child)
842 BlockDriverState *bs = child->opaque;
843 assert(bs->open_flags & BDRV_O_INACTIVE);
844 return 0;
848 * Returns the options and flags that a temporary snapshot should get, based on
849 * the originally requested flags (the originally requested image will have
850 * flags like a backing file)
852 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
853 int parent_flags, QDict *parent_options)
855 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
857 /* For temporary files, unconditional cache=unsafe is fine */
858 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
859 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
861 /* Copy the read-only option from the parent */
862 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
864 /* aio=native doesn't work for cache.direct=off, so disable it for the
865 * temporary snapshot */
866 *child_flags &= ~BDRV_O_NATIVE_AIO;
870 * Returns the options and flags that bs->file should get if a protocol driver
871 * is expected, based on the given options and flags for the parent BDS
873 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
874 int parent_flags, QDict *parent_options)
876 int flags = parent_flags;
878 /* Enable protocol handling, disable format probing for bs->file */
879 flags |= BDRV_O_PROTOCOL;
881 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
882 * the parent. */
883 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
884 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
885 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
887 /* Inherit the read-only option from the parent if it's not set */
888 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
890 /* Our block drivers take care to send flushes and respect unmap policy,
891 * so we can default to enable both on lower layers regardless of the
892 * corresponding parent options. */
893 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
895 /* Clear flags that only apply to the top layer */
896 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
897 BDRV_O_NO_IO);
899 *child_flags = flags;
902 const BdrvChildRole child_file = {
903 .get_parent_desc = bdrv_child_get_parent_desc,
904 .inherit_options = bdrv_inherited_options,
905 .drained_begin = bdrv_child_cb_drained_begin,
906 .drained_end = bdrv_child_cb_drained_end,
907 .attach = bdrv_child_cb_attach,
908 .detach = bdrv_child_cb_detach,
909 .inactivate = bdrv_child_cb_inactivate,
913 * Returns the options and flags that bs->file should get if the use of formats
914 * (and not only protocols) is permitted for it, based on the given options and
915 * flags for the parent BDS
917 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
918 int parent_flags, QDict *parent_options)
920 child_file.inherit_options(child_flags, child_options,
921 parent_flags, parent_options);
923 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
926 const BdrvChildRole child_format = {
927 .get_parent_desc = bdrv_child_get_parent_desc,
928 .inherit_options = bdrv_inherited_fmt_options,
929 .drained_begin = bdrv_child_cb_drained_begin,
930 .drained_end = bdrv_child_cb_drained_end,
931 .attach = bdrv_child_cb_attach,
932 .detach = bdrv_child_cb_detach,
933 .inactivate = bdrv_child_cb_inactivate,
936 static void bdrv_backing_attach(BdrvChild *c)
938 BlockDriverState *parent = c->opaque;
939 BlockDriverState *backing_hd = c->bs;
941 assert(!parent->backing_blocker);
942 error_setg(&parent->backing_blocker,
943 "node is used as backing hd of '%s'",
944 bdrv_get_device_or_node_name(parent));
946 parent->open_flags &= ~BDRV_O_NO_BACKING;
947 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
948 backing_hd->filename);
949 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
950 backing_hd->drv ? backing_hd->drv->format_name : "");
952 bdrv_op_block_all(backing_hd, parent->backing_blocker);
953 /* Otherwise we won't be able to commit or stream */
954 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
955 parent->backing_blocker);
956 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
957 parent->backing_blocker);
959 * We do backup in 3 ways:
960 * 1. drive backup
961 * The target bs is new opened, and the source is top BDS
962 * 2. blockdev backup
963 * Both the source and the target are top BDSes.
964 * 3. internal backup(used for block replication)
965 * Both the source and the target are backing file
967 * In case 1 and 2, neither the source nor the target is the backing file.
968 * In case 3, we will block the top BDS, so there is only one block job
969 * for the top BDS and its backing chain.
971 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
972 parent->backing_blocker);
973 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
974 parent->backing_blocker);
976 bdrv_child_cb_attach(c);
979 static void bdrv_backing_detach(BdrvChild *c)
981 BlockDriverState *parent = c->opaque;
983 assert(parent->backing_blocker);
984 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
985 error_free(parent->backing_blocker);
986 parent->backing_blocker = NULL;
988 bdrv_child_cb_detach(c);
992 * Returns the options and flags that bs->backing should get, based on the
993 * given options and flags for the parent BDS
995 static void bdrv_backing_options(int *child_flags, QDict *child_options,
996 int parent_flags, QDict *parent_options)
998 int flags = parent_flags;
1000 /* The cache mode is inherited unmodified for backing files; except WCE,
1001 * which is only applied on the top level (BlockBackend) */
1002 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1003 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1004 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1006 /* backing files always opened read-only */
1007 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1008 flags &= ~BDRV_O_COPY_ON_READ;
1010 /* snapshot=on is handled on the top layer */
1011 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1013 *child_flags = flags;
1016 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1017 const char *filename, Error **errp)
1019 BlockDriverState *parent = c->opaque;
1020 int orig_flags = bdrv_get_flags(parent);
1021 int ret;
1023 if (!(orig_flags & BDRV_O_RDWR)) {
1024 ret = bdrv_reopen(parent, orig_flags | BDRV_O_RDWR, errp);
1025 if (ret < 0) {
1026 return ret;
1030 ret = bdrv_change_backing_file(parent, filename,
1031 base->drv ? base->drv->format_name : "");
1032 if (ret < 0) {
1033 error_setg_errno(errp, -ret, "Could not update backing file link");
1036 if (!(orig_flags & BDRV_O_RDWR)) {
1037 bdrv_reopen(parent, orig_flags, NULL);
1040 return ret;
1043 const BdrvChildRole child_backing = {
1044 .get_parent_desc = bdrv_child_get_parent_desc,
1045 .attach = bdrv_backing_attach,
1046 .detach = bdrv_backing_detach,
1047 .inherit_options = bdrv_backing_options,
1048 .drained_begin = bdrv_child_cb_drained_begin,
1049 .drained_end = bdrv_child_cb_drained_end,
1050 .inactivate = bdrv_child_cb_inactivate,
1051 .update_filename = bdrv_backing_update_filename,
1054 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1056 int open_flags = flags;
1059 * Clear flags that are internal to the block layer before opening the
1060 * image.
1062 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1065 * Snapshots should be writable.
1067 if (flags & BDRV_O_TEMPORARY) {
1068 open_flags |= BDRV_O_RDWR;
1071 return open_flags;
1074 static void update_flags_from_options(int *flags, QemuOpts *opts)
1076 *flags &= ~BDRV_O_CACHE_MASK;
1078 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1079 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1080 *flags |= BDRV_O_NO_FLUSH;
1083 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1084 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1085 *flags |= BDRV_O_NOCACHE;
1088 *flags &= ~BDRV_O_RDWR;
1090 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1091 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1092 *flags |= BDRV_O_RDWR;
1097 static void update_options_from_flags(QDict *options, int flags)
1099 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1100 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1102 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1103 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1104 flags & BDRV_O_NO_FLUSH);
1106 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1107 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1111 static void bdrv_assign_node_name(BlockDriverState *bs,
1112 const char *node_name,
1113 Error **errp)
1115 char *gen_node_name = NULL;
1117 if (!node_name) {
1118 node_name = gen_node_name = id_generate(ID_BLOCK);
1119 } else if (!id_wellformed(node_name)) {
1121 * Check for empty string or invalid characters, but not if it is
1122 * generated (generated names use characters not available to the user)
1124 error_setg(errp, "Invalid node name");
1125 return;
1128 /* takes care of avoiding namespaces collisions */
1129 if (blk_by_name(node_name)) {
1130 error_setg(errp, "node-name=%s is conflicting with a device id",
1131 node_name);
1132 goto out;
1135 /* takes care of avoiding duplicates node names */
1136 if (bdrv_find_node(node_name)) {
1137 error_setg(errp, "Duplicate node name");
1138 goto out;
1141 /* copy node name into the bs and insert it into the graph list */
1142 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1143 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1144 out:
1145 g_free(gen_node_name);
1148 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1149 const char *node_name, QDict *options,
1150 int open_flags, Error **errp)
1152 Error *local_err = NULL;
1153 int ret;
1155 bdrv_assign_node_name(bs, node_name, &local_err);
1156 if (local_err) {
1157 error_propagate(errp, local_err);
1158 return -EINVAL;
1161 bs->drv = drv;
1162 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1163 bs->opaque = g_malloc0(drv->instance_size);
1165 if (drv->bdrv_file_open) {
1166 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1167 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1168 } else if (drv->bdrv_open) {
1169 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1170 } else {
1171 ret = 0;
1174 if (ret < 0) {
1175 if (local_err) {
1176 error_propagate(errp, local_err);
1177 } else if (bs->filename[0]) {
1178 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1179 } else {
1180 error_setg_errno(errp, -ret, "Could not open image");
1182 goto open_failed;
1185 ret = refresh_total_sectors(bs, bs->total_sectors);
1186 if (ret < 0) {
1187 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1188 return ret;
1191 bdrv_refresh_limits(bs, &local_err);
1192 if (local_err) {
1193 error_propagate(errp, local_err);
1194 return -EINVAL;
1197 assert(bdrv_opt_mem_align(bs) != 0);
1198 assert(bdrv_min_mem_align(bs) != 0);
1199 assert(is_power_of_2(bs->bl.request_alignment));
1201 return 0;
1202 open_failed:
1203 bs->drv = NULL;
1204 if (bs->file != NULL) {
1205 bdrv_unref_child(bs, bs->file);
1206 bs->file = NULL;
1208 g_free(bs->opaque);
1209 bs->opaque = NULL;
1210 return ret;
1213 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1214 int flags, Error **errp)
1216 BlockDriverState *bs;
1217 int ret;
1219 bs = bdrv_new();
1220 bs->open_flags = flags;
1221 bs->explicit_options = qdict_new();
1222 bs->options = qdict_new();
1223 bs->opaque = NULL;
1225 update_options_from_flags(bs->options, flags);
1227 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1228 if (ret < 0) {
1229 QDECREF(bs->explicit_options);
1230 bs->explicit_options = NULL;
1231 QDECREF(bs->options);
1232 bs->options = NULL;
1233 bdrv_unref(bs);
1234 return NULL;
1237 return bs;
1240 QemuOptsList bdrv_runtime_opts = {
1241 .name = "bdrv_common",
1242 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1243 .desc = {
1245 .name = "node-name",
1246 .type = QEMU_OPT_STRING,
1247 .help = "Node name of the block device node",
1250 .name = "driver",
1251 .type = QEMU_OPT_STRING,
1252 .help = "Block driver to use for the node",
1255 .name = BDRV_OPT_CACHE_DIRECT,
1256 .type = QEMU_OPT_BOOL,
1257 .help = "Bypass software writeback cache on the host",
1260 .name = BDRV_OPT_CACHE_NO_FLUSH,
1261 .type = QEMU_OPT_BOOL,
1262 .help = "Ignore flush requests",
1265 .name = BDRV_OPT_READ_ONLY,
1266 .type = QEMU_OPT_BOOL,
1267 .help = "Node is opened in read-only mode",
1270 .name = "detect-zeroes",
1271 .type = QEMU_OPT_STRING,
1272 .help = "try to optimize zero writes (off, on, unmap)",
1275 .name = "discard",
1276 .type = QEMU_OPT_STRING,
1277 .help = "discard operation (ignore/off, unmap/on)",
1280 .name = BDRV_OPT_FORCE_SHARE,
1281 .type = QEMU_OPT_BOOL,
1282 .help = "always accept other writers (default: off)",
1284 { /* end of list */ }
1289 * Common part for opening disk images and files
1291 * Removes all processed options from *options.
1293 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1294 QDict *options, Error **errp)
1296 int ret, open_flags;
1297 const char *filename;
1298 const char *driver_name = NULL;
1299 const char *node_name = NULL;
1300 const char *discard;
1301 const char *detect_zeroes;
1302 QemuOpts *opts;
1303 BlockDriver *drv;
1304 Error *local_err = NULL;
1306 assert(bs->file == NULL);
1307 assert(options != NULL && bs->options != options);
1309 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1310 qemu_opts_absorb_qdict(opts, options, &local_err);
1311 if (local_err) {
1312 error_propagate(errp, local_err);
1313 ret = -EINVAL;
1314 goto fail_opts;
1317 update_flags_from_options(&bs->open_flags, opts);
1319 driver_name = qemu_opt_get(opts, "driver");
1320 drv = bdrv_find_format(driver_name);
1321 assert(drv != NULL);
1323 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1325 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1326 error_setg(errp,
1327 BDRV_OPT_FORCE_SHARE
1328 "=on can only be used with read-only images");
1329 ret = -EINVAL;
1330 goto fail_opts;
1333 if (file != NULL) {
1334 filename = blk_bs(file)->filename;
1335 } else {
1337 * Caution: while qdict_get_try_str() is fine, getting
1338 * non-string types would require more care. When @options
1339 * come from -blockdev or blockdev_add, its members are typed
1340 * according to the QAPI schema, but when they come from
1341 * -drive, they're all QString.
1343 filename = qdict_get_try_str(options, "filename");
1346 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1347 error_setg(errp, "The '%s' block driver requires a file name",
1348 drv->format_name);
1349 ret = -EINVAL;
1350 goto fail_opts;
1353 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1354 drv->format_name);
1356 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1358 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1359 error_setg(errp,
1360 !bs->read_only && bdrv_is_whitelisted(drv, true)
1361 ? "Driver '%s' can only be used for read-only devices"
1362 : "Driver '%s' is not whitelisted",
1363 drv->format_name);
1364 ret = -ENOTSUP;
1365 goto fail_opts;
1368 /* bdrv_new() and bdrv_close() make it so */
1369 assert(atomic_read(&bs->copy_on_read) == 0);
1371 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1372 if (!bs->read_only) {
1373 bdrv_enable_copy_on_read(bs);
1374 } else {
1375 error_setg(errp, "Can't use copy-on-read on read-only device");
1376 ret = -EINVAL;
1377 goto fail_opts;
1381 discard = qemu_opt_get(opts, "discard");
1382 if (discard != NULL) {
1383 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1384 error_setg(errp, "Invalid discard option");
1385 ret = -EINVAL;
1386 goto fail_opts;
1390 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1391 if (detect_zeroes) {
1392 BlockdevDetectZeroesOptions value =
1393 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
1394 detect_zeroes,
1395 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1396 &local_err);
1397 if (local_err) {
1398 error_propagate(errp, local_err);
1399 ret = -EINVAL;
1400 goto fail_opts;
1403 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1404 !(bs->open_flags & BDRV_O_UNMAP))
1406 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1407 "without setting discard operation to unmap");
1408 ret = -EINVAL;
1409 goto fail_opts;
1412 bs->detect_zeroes = value;
1415 if (filename != NULL) {
1416 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1417 } else {
1418 bs->filename[0] = '\0';
1420 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1422 /* Open the image, either directly or using a protocol */
1423 open_flags = bdrv_open_flags(bs, bs->open_flags);
1424 node_name = qemu_opt_get(opts, "node-name");
1426 assert(!drv->bdrv_file_open || file == NULL);
1427 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1428 if (ret < 0) {
1429 goto fail_opts;
1432 qemu_opts_del(opts);
1433 return 0;
1435 fail_opts:
1436 qemu_opts_del(opts);
1437 return ret;
1440 static QDict *parse_json_filename(const char *filename, Error **errp)
1442 QObject *options_obj;
1443 QDict *options;
1444 int ret;
1446 ret = strstart(filename, "json:", &filename);
1447 assert(ret);
1449 options_obj = qobject_from_json(filename, errp);
1450 if (!options_obj) {
1451 /* Work around qobject_from_json() lossage TODO fix that */
1452 if (errp && !*errp) {
1453 error_setg(errp, "Could not parse the JSON options");
1454 return NULL;
1456 error_prepend(errp, "Could not parse the JSON options: ");
1457 return NULL;
1460 options = qobject_to_qdict(options_obj);
1461 if (!options) {
1462 qobject_decref(options_obj);
1463 error_setg(errp, "Invalid JSON object given");
1464 return NULL;
1467 qdict_flatten(options);
1469 return options;
1472 static void parse_json_protocol(QDict *options, const char **pfilename,
1473 Error **errp)
1475 QDict *json_options;
1476 Error *local_err = NULL;
1478 /* Parse json: pseudo-protocol */
1479 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1480 return;
1483 json_options = parse_json_filename(*pfilename, &local_err);
1484 if (local_err) {
1485 error_propagate(errp, local_err);
1486 return;
1489 /* Options given in the filename have lower priority than options
1490 * specified directly */
1491 qdict_join(options, json_options, false);
1492 QDECREF(json_options);
1493 *pfilename = NULL;
1497 * Fills in default options for opening images and converts the legacy
1498 * filename/flags pair to option QDict entries.
1499 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1500 * block driver has been specified explicitly.
1502 static int bdrv_fill_options(QDict **options, const char *filename,
1503 int *flags, Error **errp)
1505 const char *drvname;
1506 bool protocol = *flags & BDRV_O_PROTOCOL;
1507 bool parse_filename = false;
1508 BlockDriver *drv = NULL;
1509 Error *local_err = NULL;
1512 * Caution: while qdict_get_try_str() is fine, getting non-string
1513 * types would require more care. When @options come from
1514 * -blockdev or blockdev_add, its members are typed according to
1515 * the QAPI schema, but when they come from -drive, they're all
1516 * QString.
1518 drvname = qdict_get_try_str(*options, "driver");
1519 if (drvname) {
1520 drv = bdrv_find_format(drvname);
1521 if (!drv) {
1522 error_setg(errp, "Unknown driver '%s'", drvname);
1523 return -ENOENT;
1525 /* If the user has explicitly specified the driver, this choice should
1526 * override the BDRV_O_PROTOCOL flag */
1527 protocol = drv->bdrv_file_open;
1530 if (protocol) {
1531 *flags |= BDRV_O_PROTOCOL;
1532 } else {
1533 *flags &= ~BDRV_O_PROTOCOL;
1536 /* Translate cache options from flags into options */
1537 update_options_from_flags(*options, *flags);
1539 /* Fetch the file name from the options QDict if necessary */
1540 if (protocol && filename) {
1541 if (!qdict_haskey(*options, "filename")) {
1542 qdict_put_str(*options, "filename", filename);
1543 parse_filename = true;
1544 } else {
1545 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1546 "the same time");
1547 return -EINVAL;
1551 /* Find the right block driver */
1552 /* See cautionary note on accessing @options above */
1553 filename = qdict_get_try_str(*options, "filename");
1555 if (!drvname && protocol) {
1556 if (filename) {
1557 drv = bdrv_find_protocol(filename, parse_filename, errp);
1558 if (!drv) {
1559 return -EINVAL;
1562 drvname = drv->format_name;
1563 qdict_put_str(*options, "driver", drvname);
1564 } else {
1565 error_setg(errp, "Must specify either driver or file");
1566 return -EINVAL;
1570 assert(drv || !protocol);
1572 /* Driver-specific filename parsing */
1573 if (drv && drv->bdrv_parse_filename && parse_filename) {
1574 drv->bdrv_parse_filename(filename, *options, &local_err);
1575 if (local_err) {
1576 error_propagate(errp, local_err);
1577 return -EINVAL;
1580 if (!drv->bdrv_needs_filename) {
1581 qdict_del(*options, "filename");
1585 return 0;
1588 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1589 uint64_t perm, uint64_t shared,
1590 GSList *ignore_children, Error **errp);
1591 static void bdrv_child_abort_perm_update(BdrvChild *c);
1592 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1594 typedef struct BlockReopenQueueEntry {
1595 bool prepared;
1596 BDRVReopenState state;
1597 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1598 } BlockReopenQueueEntry;
1601 * Return the flags that @bs will have after the reopens in @q have
1602 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1603 * return the current flags.
1605 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1607 BlockReopenQueueEntry *entry;
1609 if (q != NULL) {
1610 QSIMPLEQ_FOREACH(entry, q, entry) {
1611 if (entry->state.bs == bs) {
1612 return entry->state.flags;
1617 return bs->open_flags;
1620 /* Returns whether the image file can be written to after the reopen queue @q
1621 * has been successfully applied, or right now if @q is NULL. */
1622 static bool bdrv_is_writable(BlockDriverState *bs, BlockReopenQueue *q)
1624 int flags = bdrv_reopen_get_flags(q, bs);
1626 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1629 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1630 BdrvChild *c, const BdrvChildRole *role,
1631 BlockReopenQueue *reopen_queue,
1632 uint64_t parent_perm, uint64_t parent_shared,
1633 uint64_t *nperm, uint64_t *nshared)
1635 if (bs->drv && bs->drv->bdrv_child_perm) {
1636 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1637 parent_perm, parent_shared,
1638 nperm, nshared);
1640 /* TODO Take force_share from reopen_queue */
1641 if (child_bs && child_bs->force_share) {
1642 *nshared = BLK_PERM_ALL;
1647 * Check whether permissions on this node can be changed in a way that
1648 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1649 * permissions of all its parents. This involves checking whether all necessary
1650 * permission changes to child nodes can be performed.
1652 * A call to this function must always be followed by a call to bdrv_set_perm()
1653 * or bdrv_abort_perm_update().
1655 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1656 uint64_t cumulative_perms,
1657 uint64_t cumulative_shared_perms,
1658 GSList *ignore_children, Error **errp)
1660 BlockDriver *drv = bs->drv;
1661 BdrvChild *c;
1662 int ret;
1664 /* Write permissions never work with read-only images */
1665 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1666 !bdrv_is_writable(bs, q))
1668 error_setg(errp, "Block node is read-only");
1669 return -EPERM;
1672 /* Check this node */
1673 if (!drv) {
1674 return 0;
1677 if (drv->bdrv_check_perm) {
1678 return drv->bdrv_check_perm(bs, cumulative_perms,
1679 cumulative_shared_perms, errp);
1682 /* Drivers that never have children can omit .bdrv_child_perm() */
1683 if (!drv->bdrv_child_perm) {
1684 assert(QLIST_EMPTY(&bs->children));
1685 return 0;
1688 /* Check all children */
1689 QLIST_FOREACH(c, &bs->children, next) {
1690 uint64_t cur_perm, cur_shared;
1691 bdrv_child_perm(bs, c->bs, c, c->role, q,
1692 cumulative_perms, cumulative_shared_perms,
1693 &cur_perm, &cur_shared);
1694 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1695 ignore_children, errp);
1696 if (ret < 0) {
1697 return ret;
1701 return 0;
1705 * Notifies drivers that after a previous bdrv_check_perm() call, the
1706 * permission update is not performed and any preparations made for it (e.g.
1707 * taken file locks) need to be undone.
1709 * This function recursively notifies all child nodes.
1711 static void bdrv_abort_perm_update(BlockDriverState *bs)
1713 BlockDriver *drv = bs->drv;
1714 BdrvChild *c;
1716 if (!drv) {
1717 return;
1720 if (drv->bdrv_abort_perm_update) {
1721 drv->bdrv_abort_perm_update(bs);
1724 QLIST_FOREACH(c, &bs->children, next) {
1725 bdrv_child_abort_perm_update(c);
1729 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1730 uint64_t cumulative_shared_perms)
1732 BlockDriver *drv = bs->drv;
1733 BdrvChild *c;
1735 if (!drv) {
1736 return;
1739 /* Update this node */
1740 if (drv->bdrv_set_perm) {
1741 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1744 /* Drivers that never have children can omit .bdrv_child_perm() */
1745 if (!drv->bdrv_child_perm) {
1746 assert(QLIST_EMPTY(&bs->children));
1747 return;
1750 /* Update all children */
1751 QLIST_FOREACH(c, &bs->children, next) {
1752 uint64_t cur_perm, cur_shared;
1753 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1754 cumulative_perms, cumulative_shared_perms,
1755 &cur_perm, &cur_shared);
1756 bdrv_child_set_perm(c, cur_perm, cur_shared);
1760 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1761 uint64_t *shared_perm)
1763 BdrvChild *c;
1764 uint64_t cumulative_perms = 0;
1765 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1767 QLIST_FOREACH(c, &bs->parents, next_parent) {
1768 cumulative_perms |= c->perm;
1769 cumulative_shared_perms &= c->shared_perm;
1772 *perm = cumulative_perms;
1773 *shared_perm = cumulative_shared_perms;
1776 static char *bdrv_child_user_desc(BdrvChild *c)
1778 if (c->role->get_parent_desc) {
1779 return c->role->get_parent_desc(c);
1782 return g_strdup("another user");
1785 char *bdrv_perm_names(uint64_t perm)
1787 struct perm_name {
1788 uint64_t perm;
1789 const char *name;
1790 } permissions[] = {
1791 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1792 { BLK_PERM_WRITE, "write" },
1793 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1794 { BLK_PERM_RESIZE, "resize" },
1795 { BLK_PERM_GRAPH_MOD, "change children" },
1796 { 0, NULL }
1799 char *result = g_strdup("");
1800 struct perm_name *p;
1802 for (p = permissions; p->name; p++) {
1803 if (perm & p->perm) {
1804 char *old = result;
1805 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1806 g_free(old);
1810 return result;
1814 * Checks whether a new reference to @bs can be added if the new user requires
1815 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1816 * set, the BdrvChild objects in this list are ignored in the calculations;
1817 * this allows checking permission updates for an existing reference.
1819 * Needs to be followed by a call to either bdrv_set_perm() or
1820 * bdrv_abort_perm_update(). */
1821 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1822 uint64_t new_used_perm,
1823 uint64_t new_shared_perm,
1824 GSList *ignore_children, Error **errp)
1826 BdrvChild *c;
1827 uint64_t cumulative_perms = new_used_perm;
1828 uint64_t cumulative_shared_perms = new_shared_perm;
1830 /* There is no reason why anyone couldn't tolerate write_unchanged */
1831 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1833 QLIST_FOREACH(c, &bs->parents, next_parent) {
1834 if (g_slist_find(ignore_children, c)) {
1835 continue;
1838 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1839 char *user = bdrv_child_user_desc(c);
1840 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1841 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1842 "allow '%s' on %s",
1843 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1844 g_free(user);
1845 g_free(perm_names);
1846 return -EPERM;
1849 if ((c->perm & new_shared_perm) != c->perm) {
1850 char *user = bdrv_child_user_desc(c);
1851 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1852 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1853 "'%s' on %s",
1854 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1855 g_free(user);
1856 g_free(perm_names);
1857 return -EPERM;
1860 cumulative_perms |= c->perm;
1861 cumulative_shared_perms &= c->shared_perm;
1864 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1865 ignore_children, errp);
1868 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1869 * bdrv_child_abort_perm_update(). */
1870 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1871 uint64_t perm, uint64_t shared,
1872 GSList *ignore_children, Error **errp)
1874 int ret;
1876 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1877 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1878 g_slist_free(ignore_children);
1880 return ret;
1883 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1885 uint64_t cumulative_perms, cumulative_shared_perms;
1887 c->perm = perm;
1888 c->shared_perm = shared;
1890 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1891 &cumulative_shared_perms);
1892 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1895 static void bdrv_child_abort_perm_update(BdrvChild *c)
1897 bdrv_abort_perm_update(c->bs);
1900 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1901 Error **errp)
1903 int ret;
1905 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1906 if (ret < 0) {
1907 bdrv_child_abort_perm_update(c);
1908 return ret;
1911 bdrv_child_set_perm(c, perm, shared);
1913 return 0;
1916 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1917 | BLK_PERM_WRITE \
1918 | BLK_PERM_WRITE_UNCHANGED \
1919 | BLK_PERM_RESIZE)
1920 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1922 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1923 const BdrvChildRole *role,
1924 BlockReopenQueue *reopen_queue,
1925 uint64_t perm, uint64_t shared,
1926 uint64_t *nperm, uint64_t *nshared)
1928 if (c == NULL) {
1929 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1930 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1931 return;
1934 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1935 (c->perm & DEFAULT_PERM_UNCHANGED);
1936 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1937 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1940 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1941 const BdrvChildRole *role,
1942 BlockReopenQueue *reopen_queue,
1943 uint64_t perm, uint64_t shared,
1944 uint64_t *nperm, uint64_t *nshared)
1946 bool backing = (role == &child_backing);
1947 assert(role == &child_backing || role == &child_file);
1949 if (!backing) {
1950 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
1952 /* Apart from the modifications below, the same permissions are
1953 * forwarded and left alone as for filters */
1954 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
1955 &perm, &shared);
1957 /* Format drivers may touch metadata even if the guest doesn't write */
1958 if (bdrv_is_writable(bs, reopen_queue)) {
1959 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1962 /* bs->file always needs to be consistent because of the metadata. We
1963 * can never allow other users to resize or write to it. */
1964 if (!(flags & BDRV_O_NO_IO)) {
1965 perm |= BLK_PERM_CONSISTENT_READ;
1967 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1968 } else {
1969 /* We want consistent read from backing files if the parent needs it.
1970 * No other operations are performed on backing files. */
1971 perm &= BLK_PERM_CONSISTENT_READ;
1973 /* If the parent can deal with changing data, we're okay with a
1974 * writable and resizable backing file. */
1975 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1976 if (shared & BLK_PERM_WRITE) {
1977 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1978 } else {
1979 shared = 0;
1982 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1983 BLK_PERM_WRITE_UNCHANGED;
1986 if (bs->open_flags & BDRV_O_INACTIVE) {
1987 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1990 *nperm = perm;
1991 *nshared = shared;
1994 static void bdrv_replace_child_noperm(BdrvChild *child,
1995 BlockDriverState *new_bs)
1997 BlockDriverState *old_bs = child->bs;
1998 int i;
2000 if (old_bs && new_bs) {
2001 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2003 if (old_bs) {
2004 /* Detach first so that the recursive drain sections coming from @child
2005 * are already gone and we only end the drain sections that came from
2006 * elsewhere. */
2007 if (child->role->detach) {
2008 child->role->detach(child);
2010 if (old_bs->quiesce_counter && child->role->drained_end) {
2011 for (i = 0; i < old_bs->quiesce_counter; i++) {
2012 child->role->drained_end(child);
2015 QLIST_REMOVE(child, next_parent);
2018 child->bs = new_bs;
2020 if (new_bs) {
2021 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2022 if (new_bs->quiesce_counter && child->role->drained_begin) {
2023 for (i = 0; i < new_bs->quiesce_counter; i++) {
2024 child->role->drained_begin(child);
2028 /* Attach only after starting new drained sections, so that recursive
2029 * drain sections coming from @child don't get an extra .drained_begin
2030 * callback. */
2031 if (child->role->attach) {
2032 child->role->attach(child);
2038 * Updates @child to change its reference to point to @new_bs, including
2039 * checking and applying the necessary permisson updates both to the old node
2040 * and to @new_bs.
2042 * NULL is passed as @new_bs for removing the reference before freeing @child.
2044 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2045 * function uses bdrv_set_perm() to update the permissions according to the new
2046 * reference that @new_bs gets.
2048 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2050 BlockDriverState *old_bs = child->bs;
2051 uint64_t perm, shared_perm;
2053 bdrv_replace_child_noperm(child, new_bs);
2055 if (old_bs) {
2056 /* Update permissions for old node. This is guaranteed to succeed
2057 * because we're just taking a parent away, so we're loosening
2058 * restrictions. */
2059 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2060 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2061 bdrv_set_perm(old_bs, perm, shared_perm);
2064 if (new_bs) {
2065 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2066 bdrv_set_perm(new_bs, perm, shared_perm);
2070 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2071 const char *child_name,
2072 const BdrvChildRole *child_role,
2073 uint64_t perm, uint64_t shared_perm,
2074 void *opaque, Error **errp)
2076 BdrvChild *child;
2077 int ret;
2079 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2080 if (ret < 0) {
2081 bdrv_abort_perm_update(child_bs);
2082 return NULL;
2085 child = g_new(BdrvChild, 1);
2086 *child = (BdrvChild) {
2087 .bs = NULL,
2088 .name = g_strdup(child_name),
2089 .role = child_role,
2090 .perm = perm,
2091 .shared_perm = shared_perm,
2092 .opaque = opaque,
2095 /* This performs the matching bdrv_set_perm() for the above check. */
2096 bdrv_replace_child(child, child_bs);
2098 return child;
2101 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2102 BlockDriverState *child_bs,
2103 const char *child_name,
2104 const BdrvChildRole *child_role,
2105 Error **errp)
2107 BdrvChild *child;
2108 uint64_t perm, shared_perm;
2110 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2112 assert(parent_bs->drv);
2113 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2114 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2115 perm, shared_perm, &perm, &shared_perm);
2117 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2118 perm, shared_perm, parent_bs, errp);
2119 if (child == NULL) {
2120 return NULL;
2123 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2124 return child;
2127 static void bdrv_detach_child(BdrvChild *child)
2129 if (child->next.le_prev) {
2130 QLIST_REMOVE(child, next);
2131 child->next.le_prev = NULL;
2134 bdrv_replace_child(child, NULL);
2136 g_free(child->name);
2137 g_free(child);
2140 void bdrv_root_unref_child(BdrvChild *child)
2142 BlockDriverState *child_bs;
2144 child_bs = child->bs;
2145 bdrv_detach_child(child);
2146 bdrv_unref(child_bs);
2149 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2151 if (child == NULL) {
2152 return;
2155 if (child->bs->inherits_from == parent) {
2156 BdrvChild *c;
2158 /* Remove inherits_from only when the last reference between parent and
2159 * child->bs goes away. */
2160 QLIST_FOREACH(c, &parent->children, next) {
2161 if (c != child && c->bs == child->bs) {
2162 break;
2165 if (c == NULL) {
2166 child->bs->inherits_from = NULL;
2170 bdrv_root_unref_child(child);
2174 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2176 BdrvChild *c;
2177 QLIST_FOREACH(c, &bs->parents, next_parent) {
2178 if (c->role->change_media) {
2179 c->role->change_media(c, load);
2184 static void bdrv_parent_cb_resize(BlockDriverState *bs)
2186 BdrvChild *c;
2187 QLIST_FOREACH(c, &bs->parents, next_parent) {
2188 if (c->role->resize) {
2189 c->role->resize(c);
2195 * Sets the backing file link of a BDS. A new reference is created; callers
2196 * which don't need their own reference any more must call bdrv_unref().
2198 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2199 Error **errp)
2201 if (backing_hd) {
2202 bdrv_ref(backing_hd);
2205 if (bs->backing) {
2206 bdrv_unref_child(bs, bs->backing);
2209 if (!backing_hd) {
2210 bs->backing = NULL;
2211 goto out;
2214 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2215 errp);
2216 if (!bs->backing) {
2217 bdrv_unref(backing_hd);
2220 bdrv_refresh_filename(bs);
2222 out:
2223 bdrv_refresh_limits(bs, NULL);
2227 * Opens the backing file for a BlockDriverState if not yet open
2229 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2230 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2231 * itself, all options starting with "${bdref_key}." are considered part of the
2232 * BlockdevRef.
2234 * TODO Can this be unified with bdrv_open_image()?
2236 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2237 const char *bdref_key, Error **errp)
2239 char *backing_filename = g_malloc0(PATH_MAX);
2240 char *bdref_key_dot;
2241 const char *reference = NULL;
2242 int ret = 0;
2243 BlockDriverState *backing_hd;
2244 QDict *options;
2245 QDict *tmp_parent_options = NULL;
2246 Error *local_err = NULL;
2248 if (bs->backing != NULL) {
2249 goto free_exit;
2252 /* NULL means an empty set of options */
2253 if (parent_options == NULL) {
2254 tmp_parent_options = qdict_new();
2255 parent_options = tmp_parent_options;
2258 bs->open_flags &= ~BDRV_O_NO_BACKING;
2260 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2261 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2262 g_free(bdref_key_dot);
2265 * Caution: while qdict_get_try_str() is fine, getting non-string
2266 * types would require more care. When @parent_options come from
2267 * -blockdev or blockdev_add, its members are typed according to
2268 * the QAPI schema, but when they come from -drive, they're all
2269 * QString.
2271 reference = qdict_get_try_str(parent_options, bdref_key);
2272 if (reference || qdict_haskey(options, "file.filename")) {
2273 backing_filename[0] = '\0';
2274 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2275 QDECREF(options);
2276 goto free_exit;
2277 } else {
2278 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2279 &local_err);
2280 if (local_err) {
2281 ret = -EINVAL;
2282 error_propagate(errp, local_err);
2283 QDECREF(options);
2284 goto free_exit;
2288 if (!bs->drv || !bs->drv->supports_backing) {
2289 ret = -EINVAL;
2290 error_setg(errp, "Driver doesn't support backing files");
2291 QDECREF(options);
2292 goto free_exit;
2295 if (!reference &&
2296 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2297 qdict_put_str(options, "driver", bs->backing_format);
2300 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2301 reference, options, 0, bs, &child_backing,
2302 errp);
2303 if (!backing_hd) {
2304 bs->open_flags |= BDRV_O_NO_BACKING;
2305 error_prepend(errp, "Could not open backing file: ");
2306 ret = -EINVAL;
2307 goto free_exit;
2309 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2311 /* Hook up the backing file link; drop our reference, bs owns the
2312 * backing_hd reference now */
2313 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2314 bdrv_unref(backing_hd);
2315 if (local_err) {
2316 error_propagate(errp, local_err);
2317 ret = -EINVAL;
2318 goto free_exit;
2321 qdict_del(parent_options, bdref_key);
2323 free_exit:
2324 g_free(backing_filename);
2325 QDECREF(tmp_parent_options);
2326 return ret;
2329 static BlockDriverState *
2330 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2331 BlockDriverState *parent, const BdrvChildRole *child_role,
2332 bool allow_none, Error **errp)
2334 BlockDriverState *bs = NULL;
2335 QDict *image_options;
2336 char *bdref_key_dot;
2337 const char *reference;
2339 assert(child_role != NULL);
2341 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2342 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2343 g_free(bdref_key_dot);
2346 * Caution: while qdict_get_try_str() is fine, getting non-string
2347 * types would require more care. When @options come from
2348 * -blockdev or blockdev_add, its members are typed according to
2349 * the QAPI schema, but when they come from -drive, they're all
2350 * QString.
2352 reference = qdict_get_try_str(options, bdref_key);
2353 if (!filename && !reference && !qdict_size(image_options)) {
2354 if (!allow_none) {
2355 error_setg(errp, "A block device must be specified for \"%s\"",
2356 bdref_key);
2358 QDECREF(image_options);
2359 goto done;
2362 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2363 parent, child_role, errp);
2364 if (!bs) {
2365 goto done;
2368 done:
2369 qdict_del(options, bdref_key);
2370 return bs;
2374 * Opens a disk image whose options are given as BlockdevRef in another block
2375 * device's options.
2377 * If allow_none is true, no image will be opened if filename is false and no
2378 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2380 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2381 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2382 * itself, all options starting with "${bdref_key}." are considered part of the
2383 * BlockdevRef.
2385 * The BlockdevRef will be removed from the options QDict.
2387 BdrvChild *bdrv_open_child(const char *filename,
2388 QDict *options, const char *bdref_key,
2389 BlockDriverState *parent,
2390 const BdrvChildRole *child_role,
2391 bool allow_none, Error **errp)
2393 BdrvChild *c;
2394 BlockDriverState *bs;
2396 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2397 allow_none, errp);
2398 if (bs == NULL) {
2399 return NULL;
2402 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2403 if (!c) {
2404 bdrv_unref(bs);
2405 return NULL;
2408 return c;
2411 /* TODO Future callers may need to specify parent/child_role in order for
2412 * option inheritance to work. Existing callers use it for the root node. */
2413 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2415 BlockDriverState *bs = NULL;
2416 Error *local_err = NULL;
2417 QObject *obj = NULL;
2418 QDict *qdict = NULL;
2419 const char *reference = NULL;
2420 Visitor *v = NULL;
2422 if (ref->type == QTYPE_QSTRING) {
2423 reference = ref->u.reference;
2424 } else {
2425 BlockdevOptions *options = &ref->u.definition;
2426 assert(ref->type == QTYPE_QDICT);
2428 v = qobject_output_visitor_new(&obj);
2429 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2430 if (local_err) {
2431 error_propagate(errp, local_err);
2432 goto fail;
2434 visit_complete(v, &obj);
2436 qdict = qobject_to_qdict(obj);
2437 qdict_flatten(qdict);
2439 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2440 * compatibility with other callers) rather than what we want as the
2441 * real defaults. Apply the defaults here instead. */
2442 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2443 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2444 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2447 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2448 obj = NULL;
2450 fail:
2451 qobject_decref(obj);
2452 visit_free(v);
2453 return bs;
2456 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2457 int flags,
2458 QDict *snapshot_options,
2459 Error **errp)
2461 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2462 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2463 int64_t total_size;
2464 QemuOpts *opts = NULL;
2465 BlockDriverState *bs_snapshot = NULL;
2466 Error *local_err = NULL;
2467 int ret;
2469 /* if snapshot, we create a temporary backing file and open it
2470 instead of opening 'filename' directly */
2472 /* Get the required size from the image */
2473 total_size = bdrv_getlength(bs);
2474 if (total_size < 0) {
2475 error_setg_errno(errp, -total_size, "Could not get image size");
2476 goto out;
2479 /* Create the temporary image */
2480 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2481 if (ret < 0) {
2482 error_setg_errno(errp, -ret, "Could not get temporary filename");
2483 goto out;
2486 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2487 &error_abort);
2488 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2489 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2490 qemu_opts_del(opts);
2491 if (ret < 0) {
2492 error_prepend(errp, "Could not create temporary overlay '%s': ",
2493 tmp_filename);
2494 goto out;
2497 /* Prepare options QDict for the temporary file */
2498 qdict_put_str(snapshot_options, "file.driver", "file");
2499 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2500 qdict_put_str(snapshot_options, "driver", "qcow2");
2502 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2503 snapshot_options = NULL;
2504 if (!bs_snapshot) {
2505 goto out;
2508 /* bdrv_append() consumes a strong reference to bs_snapshot
2509 * (i.e. it will call bdrv_unref() on it) even on error, so in
2510 * order to be able to return one, we have to increase
2511 * bs_snapshot's refcount here */
2512 bdrv_ref(bs_snapshot);
2513 bdrv_append(bs_snapshot, bs, &local_err);
2514 if (local_err) {
2515 error_propagate(errp, local_err);
2516 bs_snapshot = NULL;
2517 goto out;
2520 out:
2521 QDECREF(snapshot_options);
2522 g_free(tmp_filename);
2523 return bs_snapshot;
2527 * Opens a disk image (raw, qcow2, vmdk, ...)
2529 * options is a QDict of options to pass to the block drivers, or NULL for an
2530 * empty set of options. The reference to the QDict belongs to the block layer
2531 * after the call (even on failure), so if the caller intends to reuse the
2532 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2534 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2535 * If it is not NULL, the referenced BDS will be reused.
2537 * The reference parameter may be used to specify an existing block device which
2538 * should be opened. If specified, neither options nor a filename may be given,
2539 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2541 static BlockDriverState *bdrv_open_inherit(const char *filename,
2542 const char *reference,
2543 QDict *options, int flags,
2544 BlockDriverState *parent,
2545 const BdrvChildRole *child_role,
2546 Error **errp)
2548 int ret;
2549 BlockBackend *file = NULL;
2550 BlockDriverState *bs;
2551 BlockDriver *drv = NULL;
2552 const char *drvname;
2553 const char *backing;
2554 Error *local_err = NULL;
2555 QDict *snapshot_options = NULL;
2556 int snapshot_flags = 0;
2558 assert(!child_role || !flags);
2559 assert(!child_role == !parent);
2561 if (reference) {
2562 bool options_non_empty = options ? qdict_size(options) : false;
2563 QDECREF(options);
2565 if (filename || options_non_empty) {
2566 error_setg(errp, "Cannot reference an existing block device with "
2567 "additional options or a new filename");
2568 return NULL;
2571 bs = bdrv_lookup_bs(reference, reference, errp);
2572 if (!bs) {
2573 return NULL;
2576 bdrv_ref(bs);
2577 return bs;
2580 bs = bdrv_new();
2582 /* NULL means an empty set of options */
2583 if (options == NULL) {
2584 options = qdict_new();
2587 /* json: syntax counts as explicit options, as if in the QDict */
2588 parse_json_protocol(options, &filename, &local_err);
2589 if (local_err) {
2590 goto fail;
2593 bs->explicit_options = qdict_clone_shallow(options);
2595 if (child_role) {
2596 bs->inherits_from = parent;
2597 child_role->inherit_options(&flags, options,
2598 parent->open_flags, parent->options);
2601 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2602 if (local_err) {
2603 goto fail;
2607 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2608 * Caution: getting a boolean member of @options requires care.
2609 * When @options come from -blockdev or blockdev_add, members are
2610 * typed according to the QAPI schema, but when they come from
2611 * -drive, they're all QString.
2613 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2614 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2615 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2616 } else {
2617 flags &= ~BDRV_O_RDWR;
2620 if (flags & BDRV_O_SNAPSHOT) {
2621 snapshot_options = qdict_new();
2622 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2623 flags, options);
2624 /* Let bdrv_backing_options() override "read-only" */
2625 qdict_del(options, BDRV_OPT_READ_ONLY);
2626 bdrv_backing_options(&flags, options, flags, options);
2629 bs->open_flags = flags;
2630 bs->options = options;
2631 options = qdict_clone_shallow(options);
2633 /* Find the right image format driver */
2634 /* See cautionary note on accessing @options above */
2635 drvname = qdict_get_try_str(options, "driver");
2636 if (drvname) {
2637 drv = bdrv_find_format(drvname);
2638 if (!drv) {
2639 error_setg(errp, "Unknown driver: '%s'", drvname);
2640 goto fail;
2644 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2646 /* See cautionary note on accessing @options above */
2647 backing = qdict_get_try_str(options, "backing");
2648 if (backing && *backing == '\0') {
2649 flags |= BDRV_O_NO_BACKING;
2650 qdict_del(options, "backing");
2653 /* Open image file without format layer. This BlockBackend is only used for
2654 * probing, the block drivers will do their own bdrv_open_child() for the
2655 * same BDS, which is why we put the node name back into options. */
2656 if ((flags & BDRV_O_PROTOCOL) == 0) {
2657 BlockDriverState *file_bs;
2659 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2660 &child_file, true, &local_err);
2661 if (local_err) {
2662 goto fail;
2664 if (file_bs != NULL) {
2665 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2666 * looking at the header to guess the image format. This works even
2667 * in cases where a guest would not see a consistent state. */
2668 file = blk_new(0, BLK_PERM_ALL);
2669 blk_insert_bs(file, file_bs, &local_err);
2670 bdrv_unref(file_bs);
2671 if (local_err) {
2672 goto fail;
2675 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2679 /* Image format probing */
2680 bs->probed = !drv;
2681 if (!drv && file) {
2682 ret = find_image_format(file, filename, &drv, &local_err);
2683 if (ret < 0) {
2684 goto fail;
2687 * This option update would logically belong in bdrv_fill_options(),
2688 * but we first need to open bs->file for the probing to work, while
2689 * opening bs->file already requires the (mostly) final set of options
2690 * so that cache mode etc. can be inherited.
2692 * Adding the driver later is somewhat ugly, but it's not an option
2693 * that would ever be inherited, so it's correct. We just need to make
2694 * sure to update both bs->options (which has the full effective
2695 * options for bs) and options (which has file.* already removed).
2697 qdict_put_str(bs->options, "driver", drv->format_name);
2698 qdict_put_str(options, "driver", drv->format_name);
2699 } else if (!drv) {
2700 error_setg(errp, "Must specify either driver or file");
2701 goto fail;
2704 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2705 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2706 /* file must be NULL if a protocol BDS is about to be created
2707 * (the inverse results in an error message from bdrv_open_common()) */
2708 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2710 /* Open the image */
2711 ret = bdrv_open_common(bs, file, options, &local_err);
2712 if (ret < 0) {
2713 goto fail;
2716 if (file) {
2717 blk_unref(file);
2718 file = NULL;
2721 /* If there is a backing file, use it */
2722 if ((flags & BDRV_O_NO_BACKING) == 0) {
2723 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2724 if (ret < 0) {
2725 goto close_and_fail;
2729 bdrv_refresh_filename(bs);
2731 /* Check if any unknown options were used */
2732 if (qdict_size(options) != 0) {
2733 const QDictEntry *entry = qdict_first(options);
2734 if (flags & BDRV_O_PROTOCOL) {
2735 error_setg(errp, "Block protocol '%s' doesn't support the option "
2736 "'%s'", drv->format_name, entry->key);
2737 } else {
2738 error_setg(errp,
2739 "Block format '%s' does not support the option '%s'",
2740 drv->format_name, entry->key);
2743 goto close_and_fail;
2746 bdrv_parent_cb_change_media(bs, true);
2748 QDECREF(options);
2750 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2751 * temporary snapshot afterwards. */
2752 if (snapshot_flags) {
2753 BlockDriverState *snapshot_bs;
2754 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2755 snapshot_options, &local_err);
2756 snapshot_options = NULL;
2757 if (local_err) {
2758 goto close_and_fail;
2760 /* We are not going to return bs but the overlay on top of it
2761 * (snapshot_bs); thus, we have to drop the strong reference to bs
2762 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2763 * though, because the overlay still has a reference to it. */
2764 bdrv_unref(bs);
2765 bs = snapshot_bs;
2768 return bs;
2770 fail:
2771 blk_unref(file);
2772 QDECREF(snapshot_options);
2773 QDECREF(bs->explicit_options);
2774 QDECREF(bs->options);
2775 QDECREF(options);
2776 bs->options = NULL;
2777 bs->explicit_options = NULL;
2778 bdrv_unref(bs);
2779 error_propagate(errp, local_err);
2780 return NULL;
2782 close_and_fail:
2783 bdrv_unref(bs);
2784 QDECREF(snapshot_options);
2785 QDECREF(options);
2786 error_propagate(errp, local_err);
2787 return NULL;
2790 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2791 QDict *options, int flags, Error **errp)
2793 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2794 NULL, errp);
2798 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2799 * reopen of multiple devices.
2801 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2802 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2803 * be created and initialized. This newly created BlockReopenQueue should be
2804 * passed back in for subsequent calls that are intended to be of the same
2805 * atomic 'set'.
2807 * bs is the BlockDriverState to add to the reopen queue.
2809 * options contains the changed options for the associated bs
2810 * (the BlockReopenQueue takes ownership)
2812 * flags contains the open flags for the associated bs
2814 * returns a pointer to bs_queue, which is either the newly allocated
2815 * bs_queue, or the existing bs_queue being used.
2817 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2819 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2820 BlockDriverState *bs,
2821 QDict *options,
2822 int flags,
2823 const BdrvChildRole *role,
2824 QDict *parent_options,
2825 int parent_flags)
2827 assert(bs != NULL);
2829 BlockReopenQueueEntry *bs_entry;
2830 BdrvChild *child;
2831 QDict *old_options, *explicit_options;
2833 /* Make sure that the caller remembered to use a drained section. This is
2834 * important to avoid graph changes between the recursive queuing here and
2835 * bdrv_reopen_multiple(). */
2836 assert(bs->quiesce_counter > 0);
2838 if (bs_queue == NULL) {
2839 bs_queue = g_new0(BlockReopenQueue, 1);
2840 QSIMPLEQ_INIT(bs_queue);
2843 if (!options) {
2844 options = qdict_new();
2847 /* Check if this BlockDriverState is already in the queue */
2848 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2849 if (bs == bs_entry->state.bs) {
2850 break;
2855 * Precedence of options:
2856 * 1. Explicitly passed in options (highest)
2857 * 2. Set in flags (only for top level)
2858 * 3. Retained from explicitly set options of bs
2859 * 4. Inherited from parent node
2860 * 5. Retained from effective options of bs
2863 if (!parent_options) {
2865 * Any setting represented by flags is always updated. If the
2866 * corresponding QDict option is set, it takes precedence. Otherwise
2867 * the flag is translated into a QDict option. The old setting of bs is
2868 * not considered.
2870 update_options_from_flags(options, flags);
2873 /* Old explicitly set values (don't overwrite by inherited value) */
2874 if (bs_entry) {
2875 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2876 } else {
2877 old_options = qdict_clone_shallow(bs->explicit_options);
2879 bdrv_join_options(bs, options, old_options);
2880 QDECREF(old_options);
2882 explicit_options = qdict_clone_shallow(options);
2884 /* Inherit from parent node */
2885 if (parent_options) {
2886 QemuOpts *opts;
2887 QDict *options_copy;
2888 assert(!flags);
2889 role->inherit_options(&flags, options, parent_flags, parent_options);
2890 options_copy = qdict_clone_shallow(options);
2891 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2892 qemu_opts_absorb_qdict(opts, options_copy, NULL);
2893 update_flags_from_options(&flags, opts);
2894 qemu_opts_del(opts);
2895 QDECREF(options_copy);
2898 /* Old values are used for options that aren't set yet */
2899 old_options = qdict_clone_shallow(bs->options);
2900 bdrv_join_options(bs, options, old_options);
2901 QDECREF(old_options);
2903 /* bdrv_open_inherit() sets and clears some additional flags internally */
2904 flags &= ~BDRV_O_PROTOCOL;
2905 if (flags & BDRV_O_RDWR) {
2906 flags |= BDRV_O_ALLOW_RDWR;
2909 if (!bs_entry) {
2910 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2911 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2912 } else {
2913 QDECREF(bs_entry->state.options);
2914 QDECREF(bs_entry->state.explicit_options);
2917 bs_entry->state.bs = bs;
2918 bs_entry->state.options = options;
2919 bs_entry->state.explicit_options = explicit_options;
2920 bs_entry->state.flags = flags;
2922 /* This needs to be overwritten in bdrv_reopen_prepare() */
2923 bs_entry->state.perm = UINT64_MAX;
2924 bs_entry->state.shared_perm = 0;
2926 QLIST_FOREACH(child, &bs->children, next) {
2927 QDict *new_child_options;
2928 char *child_key_dot;
2930 /* reopen can only change the options of block devices that were
2931 * implicitly created and inherited options. For other (referenced)
2932 * block devices, a syntax like "backing.foo" results in an error. */
2933 if (child->bs->inherits_from != bs) {
2934 continue;
2937 child_key_dot = g_strdup_printf("%s.", child->name);
2938 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2939 g_free(child_key_dot);
2941 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2942 child->role, options, flags);
2945 return bs_queue;
2948 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2949 BlockDriverState *bs,
2950 QDict *options, int flags)
2952 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2953 NULL, NULL, 0);
2957 * Reopen multiple BlockDriverStates atomically & transactionally.
2959 * The queue passed in (bs_queue) must have been built up previous
2960 * via bdrv_reopen_queue().
2962 * Reopens all BDS specified in the queue, with the appropriate
2963 * flags. All devices are prepared for reopen, and failure of any
2964 * device will cause all device changes to be abandonded, and intermediate
2965 * data cleaned up.
2967 * If all devices prepare successfully, then the changes are committed
2968 * to all devices.
2970 * All affected nodes must be drained between bdrv_reopen_queue() and
2971 * bdrv_reopen_multiple().
2973 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2975 int ret = -1;
2976 BlockReopenQueueEntry *bs_entry, *next;
2977 Error *local_err = NULL;
2979 assert(bs_queue != NULL);
2981 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2982 assert(bs_entry->state.bs->quiesce_counter > 0);
2983 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2984 error_propagate(errp, local_err);
2985 goto cleanup;
2987 bs_entry->prepared = true;
2990 /* If we reach this point, we have success and just need to apply the
2991 * changes
2993 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2994 bdrv_reopen_commit(&bs_entry->state);
2997 ret = 0;
2999 cleanup:
3000 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3001 if (ret && bs_entry->prepared) {
3002 bdrv_reopen_abort(&bs_entry->state);
3003 } else if (ret) {
3004 QDECREF(bs_entry->state.explicit_options);
3006 QDECREF(bs_entry->state.options);
3007 g_free(bs_entry);
3009 g_free(bs_queue);
3011 return ret;
3015 /* Reopen a single BlockDriverState with the specified flags. */
3016 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
3018 int ret = -1;
3019 Error *local_err = NULL;
3020 BlockReopenQueue *queue;
3022 bdrv_subtree_drained_begin(bs);
3024 queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
3025 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
3026 if (local_err != NULL) {
3027 error_propagate(errp, local_err);
3030 bdrv_subtree_drained_end(bs);
3032 return ret;
3035 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3036 BdrvChild *c)
3038 BlockReopenQueueEntry *entry;
3040 QSIMPLEQ_FOREACH(entry, q, entry) {
3041 BlockDriverState *bs = entry->state.bs;
3042 BdrvChild *child;
3044 QLIST_FOREACH(child, &bs->children, next) {
3045 if (child == c) {
3046 return entry;
3051 return NULL;
3054 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3055 uint64_t *perm, uint64_t *shared)
3057 BdrvChild *c;
3058 BlockReopenQueueEntry *parent;
3059 uint64_t cumulative_perms = 0;
3060 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3062 QLIST_FOREACH(c, &bs->parents, next_parent) {
3063 parent = find_parent_in_reopen_queue(q, c);
3064 if (!parent) {
3065 cumulative_perms |= c->perm;
3066 cumulative_shared_perms &= c->shared_perm;
3067 } else {
3068 uint64_t nperm, nshared;
3070 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3071 parent->state.perm, parent->state.shared_perm,
3072 &nperm, &nshared);
3074 cumulative_perms |= nperm;
3075 cumulative_shared_perms &= nshared;
3078 *perm = cumulative_perms;
3079 *shared = cumulative_shared_perms;
3083 * Prepares a BlockDriverState for reopen. All changes are staged in the
3084 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3085 * the block driver layer .bdrv_reopen_prepare()
3087 * bs is the BlockDriverState to reopen
3088 * flags are the new open flags
3089 * queue is the reopen queue
3091 * Returns 0 on success, non-zero on error. On error errp will be set
3092 * as well.
3094 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3095 * It is the responsibility of the caller to then call the abort() or
3096 * commit() for any other BDS that have been left in a prepare() state
3099 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3100 Error **errp)
3102 int ret = -1;
3103 Error *local_err = NULL;
3104 BlockDriver *drv;
3105 QemuOpts *opts;
3106 const char *value;
3107 bool read_only;
3109 assert(reopen_state != NULL);
3110 assert(reopen_state->bs->drv != NULL);
3111 drv = reopen_state->bs->drv;
3113 /* Process generic block layer options */
3114 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3115 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3116 if (local_err) {
3117 error_propagate(errp, local_err);
3118 ret = -EINVAL;
3119 goto error;
3122 update_flags_from_options(&reopen_state->flags, opts);
3124 /* node-name and driver must be unchanged. Put them back into the QDict, so
3125 * that they are checked at the end of this function. */
3126 value = qemu_opt_get(opts, "node-name");
3127 if (value) {
3128 qdict_put_str(reopen_state->options, "node-name", value);
3131 value = qemu_opt_get(opts, "driver");
3132 if (value) {
3133 qdict_put_str(reopen_state->options, "driver", value);
3136 /* If we are to stay read-only, do not allow permission change
3137 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3138 * not set, or if the BDS still has copy_on_read enabled */
3139 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3140 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3141 if (local_err) {
3142 error_propagate(errp, local_err);
3143 goto error;
3146 /* Calculate required permissions after reopening */
3147 bdrv_reopen_perm(queue, reopen_state->bs,
3148 &reopen_state->perm, &reopen_state->shared_perm);
3150 ret = bdrv_flush(reopen_state->bs);
3151 if (ret) {
3152 error_setg_errno(errp, -ret, "Error flushing drive");
3153 goto error;
3156 if (drv->bdrv_reopen_prepare) {
3157 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3158 if (ret) {
3159 if (local_err != NULL) {
3160 error_propagate(errp, local_err);
3161 } else {
3162 error_setg(errp, "failed while preparing to reopen image '%s'",
3163 reopen_state->bs->filename);
3165 goto error;
3167 } else {
3168 /* It is currently mandatory to have a bdrv_reopen_prepare()
3169 * handler for each supported drv. */
3170 error_setg(errp, "Block format '%s' used by node '%s' "
3171 "does not support reopening files", drv->format_name,
3172 bdrv_get_device_or_node_name(reopen_state->bs));
3173 ret = -1;
3174 goto error;
3177 /* Options that are not handled are only okay if they are unchanged
3178 * compared to the old state. It is expected that some options are only
3179 * used for the initial open, but not reopen (e.g. filename) */
3180 if (qdict_size(reopen_state->options)) {
3181 const QDictEntry *entry = qdict_first(reopen_state->options);
3183 do {
3184 QObject *new = entry->value;
3185 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3188 * TODO: When using -drive to specify blockdev options, all values
3189 * will be strings; however, when using -blockdev, blockdev-add or
3190 * filenames using the json:{} pseudo-protocol, they will be
3191 * correctly typed.
3192 * In contrast, reopening options are (currently) always strings
3193 * (because you can only specify them through qemu-io; all other
3194 * callers do not specify any options).
3195 * Therefore, when using anything other than -drive to create a BDS,
3196 * this cannot detect non-string options as unchanged, because
3197 * qobject_is_equal() always returns false for objects of different
3198 * type. In the future, this should be remedied by correctly typing
3199 * all options. For now, this is not too big of an issue because
3200 * the user can simply omit options which cannot be changed anyway,
3201 * so they will stay unchanged.
3203 if (!qobject_is_equal(new, old)) {
3204 error_setg(errp, "Cannot change the option '%s'", entry->key);
3205 ret = -EINVAL;
3206 goto error;
3208 } while ((entry = qdict_next(reopen_state->options, entry)));
3211 ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3212 reopen_state->shared_perm, NULL, errp);
3213 if (ret < 0) {
3214 goto error;
3217 ret = 0;
3219 error:
3220 qemu_opts_del(opts);
3221 return ret;
3225 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3226 * makes them final by swapping the staging BlockDriverState contents into
3227 * the active BlockDriverState contents.
3229 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3231 BlockDriver *drv;
3232 BlockDriverState *bs;
3233 bool old_can_write, new_can_write;
3235 assert(reopen_state != NULL);
3236 bs = reopen_state->bs;
3237 drv = bs->drv;
3238 assert(drv != NULL);
3240 old_can_write =
3241 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3243 /* If there are any driver level actions to take */
3244 if (drv->bdrv_reopen_commit) {
3245 drv->bdrv_reopen_commit(reopen_state);
3248 /* set BDS specific flags now */
3249 QDECREF(bs->explicit_options);
3251 bs->explicit_options = reopen_state->explicit_options;
3252 bs->open_flags = reopen_state->flags;
3253 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3255 bdrv_refresh_limits(bs, NULL);
3257 bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3258 reopen_state->shared_perm);
3260 new_can_write =
3261 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3262 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3263 Error *local_err = NULL;
3264 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3265 /* This is not fatal, bitmaps just left read-only, so all following
3266 * writes will fail. User can remove read-only bitmaps to unblock
3267 * writes.
3269 error_reportf_err(local_err,
3270 "%s: Failed to make dirty bitmaps writable: ",
3271 bdrv_get_node_name(bs));
3277 * Abort the reopen, and delete and free the staged changes in
3278 * reopen_state
3280 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3282 BlockDriver *drv;
3284 assert(reopen_state != NULL);
3285 drv = reopen_state->bs->drv;
3286 assert(drv != NULL);
3288 if (drv->bdrv_reopen_abort) {
3289 drv->bdrv_reopen_abort(reopen_state);
3292 QDECREF(reopen_state->explicit_options);
3294 bdrv_abort_perm_update(reopen_state->bs);
3298 static void bdrv_close(BlockDriverState *bs)
3300 BdrvAioNotifier *ban, *ban_next;
3301 BdrvChild *child, *next;
3303 assert(!bs->job);
3304 assert(!bs->refcnt);
3306 bdrv_drained_begin(bs); /* complete I/O */
3307 bdrv_flush(bs);
3308 bdrv_drain(bs); /* in case flush left pending I/O */
3310 if (bs->drv) {
3311 bs->drv->bdrv_close(bs);
3312 bs->drv = NULL;
3315 bdrv_set_backing_hd(bs, NULL, &error_abort);
3317 if (bs->file != NULL) {
3318 bdrv_unref_child(bs, bs->file);
3319 bs->file = NULL;
3322 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3323 /* TODO Remove bdrv_unref() from drivers' close function and use
3324 * bdrv_unref_child() here */
3325 if (child->bs->inherits_from == bs) {
3326 child->bs->inherits_from = NULL;
3328 bdrv_detach_child(child);
3331 g_free(bs->opaque);
3332 bs->opaque = NULL;
3333 atomic_set(&bs->copy_on_read, 0);
3334 bs->backing_file[0] = '\0';
3335 bs->backing_format[0] = '\0';
3336 bs->total_sectors = 0;
3337 bs->encrypted = false;
3338 bs->sg = false;
3339 QDECREF(bs->options);
3340 QDECREF(bs->explicit_options);
3341 bs->options = NULL;
3342 bs->explicit_options = NULL;
3343 QDECREF(bs->full_open_options);
3344 bs->full_open_options = NULL;
3346 bdrv_release_named_dirty_bitmaps(bs);
3347 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3349 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3350 g_free(ban);
3352 QLIST_INIT(&bs->aio_notifiers);
3353 bdrv_drained_end(bs);
3356 void bdrv_close_all(void)
3358 block_job_cancel_sync_all();
3359 nbd_export_close_all();
3361 /* Drop references from requests still in flight, such as canceled block
3362 * jobs whose AIO context has not been polled yet */
3363 bdrv_drain_all();
3365 blk_remove_all_bs();
3366 blockdev_close_all_bdrv_states();
3368 assert(QTAILQ_EMPTY(&all_bdrv_states));
3371 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3373 BdrvChild *to_c;
3375 if (c->role->stay_at_node) {
3376 return false;
3379 if (c->role == &child_backing) {
3380 /* If @from is a backing file of @to, ignore the child to avoid
3381 * creating a loop. We only want to change the pointer of other
3382 * parents. */
3383 QLIST_FOREACH(to_c, &to->children, next) {
3384 if (to_c == c) {
3385 break;
3388 if (to_c) {
3389 return false;
3393 return true;
3396 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3397 Error **errp)
3399 BdrvChild *c, *next;
3400 GSList *list = NULL, *p;
3401 uint64_t old_perm, old_shared;
3402 uint64_t perm = 0, shared = BLK_PERM_ALL;
3403 int ret;
3405 assert(!atomic_read(&from->in_flight));
3406 assert(!atomic_read(&to->in_flight));
3408 /* Make sure that @from doesn't go away until we have successfully attached
3409 * all of its parents to @to. */
3410 bdrv_ref(from);
3412 /* Put all parents into @list and calculate their cumulative permissions */
3413 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3414 if (!should_update_child(c, to)) {
3415 continue;
3417 list = g_slist_prepend(list, c);
3418 perm |= c->perm;
3419 shared &= c->shared_perm;
3422 /* Check whether the required permissions can be granted on @to, ignoring
3423 * all BdrvChild in @list so that they can't block themselves. */
3424 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3425 if (ret < 0) {
3426 bdrv_abort_perm_update(to);
3427 goto out;
3430 /* Now actually perform the change. We performed the permission check for
3431 * all elements of @list at once, so set the permissions all at once at the
3432 * very end. */
3433 for (p = list; p != NULL; p = p->next) {
3434 c = p->data;
3436 bdrv_ref(to);
3437 bdrv_replace_child_noperm(c, to);
3438 bdrv_unref(from);
3441 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3442 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3444 out:
3445 g_slist_free(list);
3446 bdrv_unref(from);
3450 * Add new bs contents at the top of an image chain while the chain is
3451 * live, while keeping required fields on the top layer.
3453 * This will modify the BlockDriverState fields, and swap contents
3454 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3456 * bs_new must not be attached to a BlockBackend.
3458 * This function does not create any image files.
3460 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3461 * that's what the callers commonly need. bs_new will be referenced by the old
3462 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3463 * reference of its own, it must call bdrv_ref().
3465 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3466 Error **errp)
3468 Error *local_err = NULL;
3470 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3471 if (local_err) {
3472 error_propagate(errp, local_err);
3473 goto out;
3476 bdrv_replace_node(bs_top, bs_new, &local_err);
3477 if (local_err) {
3478 error_propagate(errp, local_err);
3479 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3480 goto out;
3483 /* bs_new is now referenced by its new parents, we don't need the
3484 * additional reference any more. */
3485 out:
3486 bdrv_unref(bs_new);
3489 static void bdrv_delete(BlockDriverState *bs)
3491 assert(!bs->job);
3492 assert(bdrv_op_blocker_is_empty(bs));
3493 assert(!bs->refcnt);
3495 bdrv_close(bs);
3497 /* remove from list, if necessary */
3498 if (bs->node_name[0] != '\0') {
3499 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3501 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3503 g_free(bs);
3507 * Run consistency checks on an image
3509 * Returns 0 if the check could be completed (it doesn't mean that the image is
3510 * free of errors) or -errno when an internal error occurred. The results of the
3511 * check are stored in res.
3513 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3514 BdrvCheckResult *res, BdrvCheckMode fix)
3516 if (bs->drv == NULL) {
3517 return -ENOMEDIUM;
3519 if (bs->drv->bdrv_co_check == NULL) {
3520 return -ENOTSUP;
3523 memset(res, 0, sizeof(*res));
3524 return bs->drv->bdrv_co_check(bs, res, fix);
3527 typedef struct CheckCo {
3528 BlockDriverState *bs;
3529 BdrvCheckResult *res;
3530 BdrvCheckMode fix;
3531 int ret;
3532 } CheckCo;
3534 static void bdrv_check_co_entry(void *opaque)
3536 CheckCo *cco = opaque;
3537 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3540 int bdrv_check(BlockDriverState *bs,
3541 BdrvCheckResult *res, BdrvCheckMode fix)
3543 Coroutine *co;
3544 CheckCo cco = {
3545 .bs = bs,
3546 .res = res,
3547 .ret = -EINPROGRESS,
3548 .fix = fix,
3551 if (qemu_in_coroutine()) {
3552 /* Fast-path if already in coroutine context */
3553 bdrv_check_co_entry(&cco);
3554 } else {
3555 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3556 qemu_coroutine_enter(co);
3557 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3560 return cco.ret;
3564 * Return values:
3565 * 0 - success
3566 * -EINVAL - backing format specified, but no file
3567 * -ENOSPC - can't update the backing file because no space is left in the
3568 * image file header
3569 * -ENOTSUP - format driver doesn't support changing the backing file
3571 int bdrv_change_backing_file(BlockDriverState *bs,
3572 const char *backing_file, const char *backing_fmt)
3574 BlockDriver *drv = bs->drv;
3575 int ret;
3577 if (!drv) {
3578 return -ENOMEDIUM;
3581 /* Backing file format doesn't make sense without a backing file */
3582 if (backing_fmt && !backing_file) {
3583 return -EINVAL;
3586 if (drv->bdrv_change_backing_file != NULL) {
3587 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3588 } else {
3589 ret = -ENOTSUP;
3592 if (ret == 0) {
3593 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3594 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3596 return ret;
3600 * Finds the image layer in the chain that has 'bs' as its backing file.
3602 * active is the current topmost image.
3604 * Returns NULL if bs is not found in active's image chain,
3605 * or if active == bs.
3607 * Returns the bottommost base image if bs == NULL.
3609 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3610 BlockDriverState *bs)
3612 while (active && bs != backing_bs(active)) {
3613 active = backing_bs(active);
3616 return active;
3619 /* Given a BDS, searches for the base layer. */
3620 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3622 return bdrv_find_overlay(bs, NULL);
3626 * Drops images above 'base' up to and including 'top', and sets the image
3627 * above 'top' to have base as its backing file.
3629 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3630 * information in 'bs' can be properly updated.
3632 * E.g., this will convert the following chain:
3633 * bottom <- base <- intermediate <- top <- active
3635 * to
3637 * bottom <- base <- active
3639 * It is allowed for bottom==base, in which case it converts:
3641 * base <- intermediate <- top <- active
3643 * to
3645 * base <- active
3647 * If backing_file_str is non-NULL, it will be used when modifying top's
3648 * overlay image metadata.
3650 * Error conditions:
3651 * if active == top, that is considered an error
3654 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3655 const char *backing_file_str)
3657 BdrvChild *c, *next;
3658 Error *local_err = NULL;
3659 int ret = -EIO;
3661 bdrv_ref(top);
3663 if (!top->drv || !base->drv) {
3664 goto exit;
3667 /* Make sure that base is in the backing chain of top */
3668 if (!bdrv_chain_contains(top, base)) {
3669 goto exit;
3672 /* success - we can delete the intermediate states, and link top->base */
3673 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3674 * we've figured out how they should work. */
3675 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3677 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3678 /* Check whether we are allowed to switch c from top to base */
3679 GSList *ignore_children = g_slist_prepend(NULL, c);
3680 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3681 ignore_children, &local_err);
3682 g_slist_free(ignore_children);
3683 if (local_err) {
3684 ret = -EPERM;
3685 error_report_err(local_err);
3686 goto exit;
3689 /* If so, update the backing file path in the image file */
3690 if (c->role->update_filename) {
3691 ret = c->role->update_filename(c, base, backing_file_str,
3692 &local_err);
3693 if (ret < 0) {
3694 bdrv_abort_perm_update(base);
3695 error_report_err(local_err);
3696 goto exit;
3700 /* Do the actual switch in the in-memory graph.
3701 * Completes bdrv_check_update_perm() transaction internally. */
3702 bdrv_ref(base);
3703 bdrv_replace_child(c, base);
3704 bdrv_unref(top);
3707 ret = 0;
3708 exit:
3709 bdrv_unref(top);
3710 return ret;
3714 * Truncate file to 'offset' bytes (needed only for file protocols)
3716 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
3717 Error **errp)
3719 BlockDriverState *bs = child->bs;
3720 BlockDriver *drv = bs->drv;
3721 int ret;
3723 assert(child->perm & BLK_PERM_RESIZE);
3725 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3726 if (!drv) {
3727 error_setg(errp, "No medium inserted");
3728 return -ENOMEDIUM;
3730 if (offset < 0) {
3731 error_setg(errp, "Image size cannot be negative");
3732 return -EINVAL;
3735 if (!drv->bdrv_truncate) {
3736 if (bs->file && drv->is_filter) {
3737 return bdrv_truncate(bs->file, offset, prealloc, errp);
3739 error_setg(errp, "Image format driver does not support resize");
3740 return -ENOTSUP;
3742 if (bs->read_only) {
3743 error_setg(errp, "Image is read-only");
3744 return -EACCES;
3747 assert(!(bs->open_flags & BDRV_O_INACTIVE));
3749 ret = drv->bdrv_truncate(bs, offset, prealloc, errp);
3750 if (ret < 0) {
3751 return ret;
3753 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3754 if (ret < 0) {
3755 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3756 } else {
3757 offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3759 bdrv_dirty_bitmap_truncate(bs, offset);
3760 bdrv_parent_cb_resize(bs);
3761 atomic_inc(&bs->write_gen);
3762 return ret;
3766 * Length of a allocated file in bytes. Sparse files are counted by actual
3767 * allocated space. Return < 0 if error or unknown.
3769 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3771 BlockDriver *drv = bs->drv;
3772 if (!drv) {
3773 return -ENOMEDIUM;
3775 if (drv->bdrv_get_allocated_file_size) {
3776 return drv->bdrv_get_allocated_file_size(bs);
3778 if (bs->file) {
3779 return bdrv_get_allocated_file_size(bs->file->bs);
3781 return -ENOTSUP;
3785 * bdrv_measure:
3786 * @drv: Format driver
3787 * @opts: Creation options for new image
3788 * @in_bs: Existing image containing data for new image (may be NULL)
3789 * @errp: Error object
3790 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3791 * or NULL on error
3793 * Calculate file size required to create a new image.
3795 * If @in_bs is given then space for allocated clusters and zero clusters
3796 * from that image are included in the calculation. If @opts contains a
3797 * backing file that is shared by @in_bs then backing clusters may be omitted
3798 * from the calculation.
3800 * If @in_bs is NULL then the calculation includes no allocated clusters
3801 * unless a preallocation option is given in @opts.
3803 * Note that @in_bs may use a different BlockDriver from @drv.
3805 * If an error occurs the @errp pointer is set.
3807 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3808 BlockDriverState *in_bs, Error **errp)
3810 if (!drv->bdrv_measure) {
3811 error_setg(errp, "Block driver '%s' does not support size measurement",
3812 drv->format_name);
3813 return NULL;
3816 return drv->bdrv_measure(opts, in_bs, errp);
3820 * Return number of sectors on success, -errno on error.
3822 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3824 BlockDriver *drv = bs->drv;
3826 if (!drv)
3827 return -ENOMEDIUM;
3829 if (drv->has_variable_length) {
3830 int ret = refresh_total_sectors(bs, bs->total_sectors);
3831 if (ret < 0) {
3832 return ret;
3835 return bs->total_sectors;
3839 * Return length in bytes on success, -errno on error.
3840 * The length is always a multiple of BDRV_SECTOR_SIZE.
3842 int64_t bdrv_getlength(BlockDriverState *bs)
3844 int64_t ret = bdrv_nb_sectors(bs);
3846 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3847 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3850 /* return 0 as number of sectors if no device present or error */
3851 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3853 int64_t nb_sectors = bdrv_nb_sectors(bs);
3855 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3858 bool bdrv_is_sg(BlockDriverState *bs)
3860 return bs->sg;
3863 bool bdrv_is_encrypted(BlockDriverState *bs)
3865 if (bs->backing && bs->backing->bs->encrypted) {
3866 return true;
3868 return bs->encrypted;
3871 const char *bdrv_get_format_name(BlockDriverState *bs)
3873 return bs->drv ? bs->drv->format_name : NULL;
3876 static int qsort_strcmp(const void *a, const void *b)
3878 return strcmp(*(char *const *)a, *(char *const *)b);
3881 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3882 void *opaque)
3884 BlockDriver *drv;
3885 int count = 0;
3886 int i;
3887 const char **formats = NULL;
3889 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3890 if (drv->format_name) {
3891 bool found = false;
3892 int i = count;
3893 while (formats && i && !found) {
3894 found = !strcmp(formats[--i], drv->format_name);
3897 if (!found) {
3898 formats = g_renew(const char *, formats, count + 1);
3899 formats[count++] = drv->format_name;
3904 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3905 const char *format_name = block_driver_modules[i].format_name;
3907 if (format_name) {
3908 bool found = false;
3909 int j = count;
3911 while (formats && j && !found) {
3912 found = !strcmp(formats[--j], format_name);
3915 if (!found) {
3916 formats = g_renew(const char *, formats, count + 1);
3917 formats[count++] = format_name;
3922 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3924 for (i = 0; i < count; i++) {
3925 it(opaque, formats[i]);
3928 g_free(formats);
3931 /* This function is to find a node in the bs graph */
3932 BlockDriverState *bdrv_find_node(const char *node_name)
3934 BlockDriverState *bs;
3936 assert(node_name);
3938 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3939 if (!strcmp(node_name, bs->node_name)) {
3940 return bs;
3943 return NULL;
3946 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3947 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3949 BlockDeviceInfoList *list, *entry;
3950 BlockDriverState *bs;
3952 list = NULL;
3953 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3954 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3955 if (!info) {
3956 qapi_free_BlockDeviceInfoList(list);
3957 return NULL;
3959 entry = g_malloc0(sizeof(*entry));
3960 entry->value = info;
3961 entry->next = list;
3962 list = entry;
3965 return list;
3968 BlockDriverState *bdrv_lookup_bs(const char *device,
3969 const char *node_name,
3970 Error **errp)
3972 BlockBackend *blk;
3973 BlockDriverState *bs;
3975 if (device) {
3976 blk = blk_by_name(device);
3978 if (blk) {
3979 bs = blk_bs(blk);
3980 if (!bs) {
3981 error_setg(errp, "Device '%s' has no medium", device);
3984 return bs;
3988 if (node_name) {
3989 bs = bdrv_find_node(node_name);
3991 if (bs) {
3992 return bs;
3996 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3997 device ? device : "",
3998 node_name ? node_name : "");
3999 return NULL;
4002 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4003 * return false. If either argument is NULL, return false. */
4004 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4006 while (top && top != base) {
4007 top = backing_bs(top);
4010 return top != NULL;
4013 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4015 if (!bs) {
4016 return QTAILQ_FIRST(&graph_bdrv_states);
4018 return QTAILQ_NEXT(bs, node_list);
4021 const char *bdrv_get_node_name(const BlockDriverState *bs)
4023 return bs->node_name;
4026 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4028 BdrvChild *c;
4029 const char *name;
4031 /* If multiple parents have a name, just pick the first one. */
4032 QLIST_FOREACH(c, &bs->parents, next_parent) {
4033 if (c->role->get_name) {
4034 name = c->role->get_name(c);
4035 if (name && *name) {
4036 return name;
4041 return NULL;
4044 /* TODO check what callers really want: bs->node_name or blk_name() */
4045 const char *bdrv_get_device_name(const BlockDriverState *bs)
4047 return bdrv_get_parent_name(bs) ?: "";
4050 /* This can be used to identify nodes that might not have a device
4051 * name associated. Since node and device names live in the same
4052 * namespace, the result is unambiguous. The exception is if both are
4053 * absent, then this returns an empty (non-null) string. */
4054 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4056 return bdrv_get_parent_name(bs) ?: bs->node_name;
4059 int bdrv_get_flags(BlockDriverState *bs)
4061 return bs->open_flags;
4064 int bdrv_has_zero_init_1(BlockDriverState *bs)
4066 return 1;
4069 int bdrv_has_zero_init(BlockDriverState *bs)
4071 if (!bs->drv) {
4072 return 0;
4075 /* If BS is a copy on write image, it is initialized to
4076 the contents of the base image, which may not be zeroes. */
4077 if (bs->backing) {
4078 return 0;
4080 if (bs->drv->bdrv_has_zero_init) {
4081 return bs->drv->bdrv_has_zero_init(bs);
4083 if (bs->file && bs->drv->is_filter) {
4084 return bdrv_has_zero_init(bs->file->bs);
4087 /* safe default */
4088 return 0;
4091 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4093 BlockDriverInfo bdi;
4095 if (bs->backing) {
4096 return false;
4099 if (bdrv_get_info(bs, &bdi) == 0) {
4100 return bdi.unallocated_blocks_are_zero;
4103 return false;
4106 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4108 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4109 return false;
4112 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4115 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4117 if (bs->backing && bs->backing->bs->encrypted)
4118 return bs->backing_file;
4119 else if (bs->encrypted)
4120 return bs->filename;
4121 else
4122 return NULL;
4125 void bdrv_get_backing_filename(BlockDriverState *bs,
4126 char *filename, int filename_size)
4128 pstrcpy(filename, filename_size, bs->backing_file);
4131 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4133 BlockDriver *drv = bs->drv;
4134 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4135 if (!drv) {
4136 return -ENOMEDIUM;
4138 if (!drv->bdrv_get_info) {
4139 if (bs->file && drv->is_filter) {
4140 return bdrv_get_info(bs->file->bs, bdi);
4142 return -ENOTSUP;
4144 memset(bdi, 0, sizeof(*bdi));
4145 return drv->bdrv_get_info(bs, bdi);
4148 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4150 BlockDriver *drv = bs->drv;
4151 if (drv && drv->bdrv_get_specific_info) {
4152 return drv->bdrv_get_specific_info(bs);
4154 return NULL;
4157 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4159 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4160 return;
4163 bs->drv->bdrv_debug_event(bs, event);
4166 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4167 const char *tag)
4169 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4170 bs = bs->file ? bs->file->bs : NULL;
4173 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4174 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4177 return -ENOTSUP;
4180 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4182 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4183 bs = bs->file ? bs->file->bs : NULL;
4186 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4187 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4190 return -ENOTSUP;
4193 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4195 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4196 bs = bs->file ? bs->file->bs : NULL;
4199 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4200 return bs->drv->bdrv_debug_resume(bs, tag);
4203 return -ENOTSUP;
4206 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4208 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4209 bs = bs->file ? bs->file->bs : NULL;
4212 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4213 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4216 return false;
4219 /* backing_file can either be relative, or absolute, or a protocol. If it is
4220 * relative, it must be relative to the chain. So, passing in bs->filename
4221 * from a BDS as backing_file should not be done, as that may be relative to
4222 * the CWD rather than the chain. */
4223 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4224 const char *backing_file)
4226 char *filename_full = NULL;
4227 char *backing_file_full = NULL;
4228 char *filename_tmp = NULL;
4229 int is_protocol = 0;
4230 BlockDriverState *curr_bs = NULL;
4231 BlockDriverState *retval = NULL;
4232 Error *local_error = NULL;
4234 if (!bs || !bs->drv || !backing_file) {
4235 return NULL;
4238 filename_full = g_malloc(PATH_MAX);
4239 backing_file_full = g_malloc(PATH_MAX);
4240 filename_tmp = g_malloc(PATH_MAX);
4242 is_protocol = path_has_protocol(backing_file);
4244 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4246 /* If either of the filename paths is actually a protocol, then
4247 * compare unmodified paths; otherwise make paths relative */
4248 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4249 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4250 retval = curr_bs->backing->bs;
4251 break;
4253 /* Also check against the full backing filename for the image */
4254 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4255 &local_error);
4256 if (local_error == NULL) {
4257 if (strcmp(backing_file, backing_file_full) == 0) {
4258 retval = curr_bs->backing->bs;
4259 break;
4261 } else {
4262 error_free(local_error);
4263 local_error = NULL;
4265 } else {
4266 /* If not an absolute filename path, make it relative to the current
4267 * image's filename path */
4268 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4269 backing_file);
4271 /* We are going to compare absolute pathnames */
4272 if (!realpath(filename_tmp, filename_full)) {
4273 continue;
4276 /* We need to make sure the backing filename we are comparing against
4277 * is relative to the current image filename (or absolute) */
4278 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4279 curr_bs->backing_file);
4281 if (!realpath(filename_tmp, backing_file_full)) {
4282 continue;
4285 if (strcmp(backing_file_full, filename_full) == 0) {
4286 retval = curr_bs->backing->bs;
4287 break;
4292 g_free(filename_full);
4293 g_free(backing_file_full);
4294 g_free(filename_tmp);
4295 return retval;
4298 void bdrv_init(void)
4300 module_call_init(MODULE_INIT_BLOCK);
4303 void bdrv_init_with_whitelist(void)
4305 use_bdrv_whitelist = 1;
4306 bdrv_init();
4309 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4310 Error **errp)
4312 BdrvChild *child, *parent;
4313 uint64_t perm, shared_perm;
4314 Error *local_err = NULL;
4315 int ret;
4317 if (!bs->drv) {
4318 return;
4321 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4322 return;
4325 QLIST_FOREACH(child, &bs->children, next) {
4326 bdrv_co_invalidate_cache(child->bs, &local_err);
4327 if (local_err) {
4328 error_propagate(errp, local_err);
4329 return;
4334 * Update permissions, they may differ for inactive nodes.
4336 * Note that the required permissions of inactive images are always a
4337 * subset of the permissions required after activating the image. This
4338 * allows us to just get the permissions upfront without restricting
4339 * drv->bdrv_invalidate_cache().
4341 * It also means that in error cases, we don't have to try and revert to
4342 * the old permissions (which is an operation that could fail, too). We can
4343 * just keep the extended permissions for the next time that an activation
4344 * of the image is tried.
4346 bs->open_flags &= ~BDRV_O_INACTIVE;
4347 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4348 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4349 if (ret < 0) {
4350 bs->open_flags |= BDRV_O_INACTIVE;
4351 error_propagate(errp, local_err);
4352 return;
4354 bdrv_set_perm(bs, perm, shared_perm);
4356 if (bs->drv->bdrv_co_invalidate_cache) {
4357 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4358 if (local_err) {
4359 bs->open_flags |= BDRV_O_INACTIVE;
4360 error_propagate(errp, local_err);
4361 return;
4365 ret = refresh_total_sectors(bs, bs->total_sectors);
4366 if (ret < 0) {
4367 bs->open_flags |= BDRV_O_INACTIVE;
4368 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4369 return;
4372 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4373 if (parent->role->activate) {
4374 parent->role->activate(parent, &local_err);
4375 if (local_err) {
4376 error_propagate(errp, local_err);
4377 return;
4383 typedef struct InvalidateCacheCo {
4384 BlockDriverState *bs;
4385 Error **errp;
4386 bool done;
4387 } InvalidateCacheCo;
4389 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4391 InvalidateCacheCo *ico = opaque;
4392 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4393 ico->done = true;
4396 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4398 Coroutine *co;
4399 InvalidateCacheCo ico = {
4400 .bs = bs,
4401 .done = false,
4402 .errp = errp
4405 if (qemu_in_coroutine()) {
4406 /* Fast-path if already in coroutine context */
4407 bdrv_invalidate_cache_co_entry(&ico);
4408 } else {
4409 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4410 qemu_coroutine_enter(co);
4411 BDRV_POLL_WHILE(bs, !ico.done);
4415 void bdrv_invalidate_cache_all(Error **errp)
4417 BlockDriverState *bs;
4418 Error *local_err = NULL;
4419 BdrvNextIterator it;
4421 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4422 AioContext *aio_context = bdrv_get_aio_context(bs);
4424 aio_context_acquire(aio_context);
4425 bdrv_invalidate_cache(bs, &local_err);
4426 aio_context_release(aio_context);
4427 if (local_err) {
4428 error_propagate(errp, local_err);
4429 bdrv_next_cleanup(&it);
4430 return;
4435 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4436 bool setting_flag)
4438 BdrvChild *child, *parent;
4439 int ret;
4441 if (!bs->drv) {
4442 return -ENOMEDIUM;
4445 if (!setting_flag && bs->drv->bdrv_inactivate) {
4446 ret = bs->drv->bdrv_inactivate(bs);
4447 if (ret < 0) {
4448 return ret;
4452 if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
4453 uint64_t perm, shared_perm;
4455 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4456 if (parent->role->inactivate) {
4457 ret = parent->role->inactivate(parent);
4458 if (ret < 0) {
4459 return ret;
4464 bs->open_flags |= BDRV_O_INACTIVE;
4466 /* Update permissions, they may differ for inactive nodes */
4467 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4468 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4469 bdrv_set_perm(bs, perm, shared_perm);
4472 QLIST_FOREACH(child, &bs->children, next) {
4473 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4474 if (ret < 0) {
4475 return ret;
4479 /* At this point persistent bitmaps should be already stored by the format
4480 * driver */
4481 bdrv_release_persistent_dirty_bitmaps(bs);
4483 return 0;
4486 int bdrv_inactivate_all(void)
4488 BlockDriverState *bs = NULL;
4489 BdrvNextIterator it;
4490 int ret = 0;
4491 int pass;
4492 GSList *aio_ctxs = NULL, *ctx;
4494 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4495 AioContext *aio_context = bdrv_get_aio_context(bs);
4497 if (!g_slist_find(aio_ctxs, aio_context)) {
4498 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4499 aio_context_acquire(aio_context);
4503 /* We do two passes of inactivation. The first pass calls to drivers'
4504 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4505 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4506 * is allowed. */
4507 for (pass = 0; pass < 2; pass++) {
4508 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4509 ret = bdrv_inactivate_recurse(bs, pass);
4510 if (ret < 0) {
4511 bdrv_next_cleanup(&it);
4512 goto out;
4517 out:
4518 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4519 AioContext *aio_context = ctx->data;
4520 aio_context_release(aio_context);
4522 g_slist_free(aio_ctxs);
4524 return ret;
4527 /**************************************************************/
4528 /* removable device support */
4531 * Return TRUE if the media is present
4533 bool bdrv_is_inserted(BlockDriverState *bs)
4535 BlockDriver *drv = bs->drv;
4536 BdrvChild *child;
4538 if (!drv) {
4539 return false;
4541 if (drv->bdrv_is_inserted) {
4542 return drv->bdrv_is_inserted(bs);
4544 QLIST_FOREACH(child, &bs->children, next) {
4545 if (!bdrv_is_inserted(child->bs)) {
4546 return false;
4549 return true;
4553 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4555 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4557 BlockDriver *drv = bs->drv;
4559 if (drv && drv->bdrv_eject) {
4560 drv->bdrv_eject(bs, eject_flag);
4565 * Lock or unlock the media (if it is locked, the user won't be able
4566 * to eject it manually).
4568 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4570 BlockDriver *drv = bs->drv;
4572 trace_bdrv_lock_medium(bs, locked);
4574 if (drv && drv->bdrv_lock_medium) {
4575 drv->bdrv_lock_medium(bs, locked);
4579 /* Get a reference to bs */
4580 void bdrv_ref(BlockDriverState *bs)
4582 bs->refcnt++;
4585 /* Release a previously grabbed reference to bs.
4586 * If after releasing, reference count is zero, the BlockDriverState is
4587 * deleted. */
4588 void bdrv_unref(BlockDriverState *bs)
4590 if (!bs) {
4591 return;
4593 assert(bs->refcnt > 0);
4594 if (--bs->refcnt == 0) {
4595 bdrv_delete(bs);
4599 struct BdrvOpBlocker {
4600 Error *reason;
4601 QLIST_ENTRY(BdrvOpBlocker) list;
4604 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4606 BdrvOpBlocker *blocker;
4607 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4608 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4609 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4610 error_propagate(errp, error_copy(blocker->reason));
4611 error_prepend(errp, "Node '%s' is busy: ",
4612 bdrv_get_device_or_node_name(bs));
4613 return true;
4615 return false;
4618 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4620 BdrvOpBlocker *blocker;
4621 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4623 blocker = g_new0(BdrvOpBlocker, 1);
4624 blocker->reason = reason;
4625 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4628 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4630 BdrvOpBlocker *blocker, *next;
4631 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4632 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4633 if (blocker->reason == reason) {
4634 QLIST_REMOVE(blocker, list);
4635 g_free(blocker);
4640 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4642 int i;
4643 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4644 bdrv_op_block(bs, i, reason);
4648 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4650 int i;
4651 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4652 bdrv_op_unblock(bs, i, reason);
4656 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4658 int i;
4660 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4661 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4662 return false;
4665 return true;
4668 void bdrv_img_create(const char *filename, const char *fmt,
4669 const char *base_filename, const char *base_fmt,
4670 char *options, uint64_t img_size, int flags, bool quiet,
4671 Error **errp)
4673 QemuOptsList *create_opts = NULL;
4674 QemuOpts *opts = NULL;
4675 const char *backing_fmt, *backing_file;
4676 int64_t size;
4677 BlockDriver *drv, *proto_drv;
4678 Error *local_err = NULL;
4679 int ret = 0;
4681 /* Find driver and parse its options */
4682 drv = bdrv_find_format(fmt);
4683 if (!drv) {
4684 error_setg(errp, "Unknown file format '%s'", fmt);
4685 return;
4688 proto_drv = bdrv_find_protocol(filename, true, errp);
4689 if (!proto_drv) {
4690 return;
4693 if (!drv->create_opts) {
4694 error_setg(errp, "Format driver '%s' does not support image creation",
4695 drv->format_name);
4696 return;
4699 if (!proto_drv->create_opts) {
4700 error_setg(errp, "Protocol driver '%s' does not support image creation",
4701 proto_drv->format_name);
4702 return;
4705 create_opts = qemu_opts_append(create_opts, drv->create_opts);
4706 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4708 /* Create parameter list with default values */
4709 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4710 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4712 /* Parse -o options */
4713 if (options) {
4714 qemu_opts_do_parse(opts, options, NULL, &local_err);
4715 if (local_err) {
4716 error_report_err(local_err);
4717 local_err = NULL;
4718 error_setg(errp, "Invalid options for file format '%s'", fmt);
4719 goto out;
4723 if (base_filename) {
4724 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4725 if (local_err) {
4726 error_setg(errp, "Backing file not supported for file format '%s'",
4727 fmt);
4728 goto out;
4732 if (base_fmt) {
4733 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4734 if (local_err) {
4735 error_setg(errp, "Backing file format not supported for file "
4736 "format '%s'", fmt);
4737 goto out;
4741 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4742 if (backing_file) {
4743 if (!strcmp(filename, backing_file)) {
4744 error_setg(errp, "Error: Trying to create an image with the "
4745 "same filename as the backing file");
4746 goto out;
4750 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4752 /* The size for the image must always be specified, unless we have a backing
4753 * file and we have not been forbidden from opening it. */
4754 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4755 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4756 BlockDriverState *bs;
4757 char *full_backing = g_new0(char, PATH_MAX);
4758 int back_flags;
4759 QDict *backing_options = NULL;
4761 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4762 full_backing, PATH_MAX,
4763 &local_err);
4764 if (local_err) {
4765 g_free(full_backing);
4766 goto out;
4769 /* backing files always opened read-only */
4770 back_flags = flags;
4771 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4773 backing_options = qdict_new();
4774 if (backing_fmt) {
4775 qdict_put_str(backing_options, "driver", backing_fmt);
4777 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
4779 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4780 &local_err);
4781 g_free(full_backing);
4782 if (!bs && size != -1) {
4783 /* Couldn't open BS, but we have a size, so it's nonfatal */
4784 warn_reportf_err(local_err,
4785 "Could not verify backing image. "
4786 "This may become an error in future versions.\n");
4787 local_err = NULL;
4788 } else if (!bs) {
4789 /* Couldn't open bs, do not have size */
4790 error_append_hint(&local_err,
4791 "Could not open backing image to determine size.\n");
4792 goto out;
4793 } else {
4794 if (size == -1) {
4795 /* Opened BS, have no size */
4796 size = bdrv_getlength(bs);
4797 if (size < 0) {
4798 error_setg_errno(errp, -size, "Could not get size of '%s'",
4799 backing_file);
4800 bdrv_unref(bs);
4801 goto out;
4803 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4805 bdrv_unref(bs);
4807 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4809 if (size == -1) {
4810 error_setg(errp, "Image creation needs a size parameter");
4811 goto out;
4814 if (!quiet) {
4815 printf("Formatting '%s', fmt=%s ", filename, fmt);
4816 qemu_opts_print(opts, " ");
4817 puts("");
4820 ret = bdrv_create(drv, filename, opts, &local_err);
4822 if (ret == -EFBIG) {
4823 /* This is generally a better message than whatever the driver would
4824 * deliver (especially because of the cluster_size_hint), since that
4825 * is most probably not much different from "image too large". */
4826 const char *cluster_size_hint = "";
4827 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4828 cluster_size_hint = " (try using a larger cluster size)";
4830 error_setg(errp, "The image size is too large for file format '%s'"
4831 "%s", fmt, cluster_size_hint);
4832 error_free(local_err);
4833 local_err = NULL;
4836 out:
4837 qemu_opts_del(opts);
4838 qemu_opts_free(create_opts);
4839 error_propagate(errp, local_err);
4842 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4844 return bs ? bs->aio_context : qemu_get_aio_context();
4847 AioWait *bdrv_get_aio_wait(BlockDriverState *bs)
4849 return bs ? &bs->wait : NULL;
4852 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4854 aio_co_enter(bdrv_get_aio_context(bs), co);
4857 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4859 QLIST_REMOVE(ban, list);
4860 g_free(ban);
4863 void bdrv_detach_aio_context(BlockDriverState *bs)
4865 BdrvAioNotifier *baf, *baf_tmp;
4866 BdrvChild *child;
4868 if (!bs->drv) {
4869 return;
4872 assert(!bs->walking_aio_notifiers);
4873 bs->walking_aio_notifiers = true;
4874 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4875 if (baf->deleted) {
4876 bdrv_do_remove_aio_context_notifier(baf);
4877 } else {
4878 baf->detach_aio_context(baf->opaque);
4881 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4882 * remove remaining aio notifiers if we aren't called again.
4884 bs->walking_aio_notifiers = false;
4886 if (bs->drv->bdrv_detach_aio_context) {
4887 bs->drv->bdrv_detach_aio_context(bs);
4889 QLIST_FOREACH(child, &bs->children, next) {
4890 bdrv_detach_aio_context(child->bs);
4893 bs->aio_context = NULL;
4896 void bdrv_attach_aio_context(BlockDriverState *bs,
4897 AioContext *new_context)
4899 BdrvAioNotifier *ban, *ban_tmp;
4900 BdrvChild *child;
4902 if (!bs->drv) {
4903 return;
4906 bs->aio_context = new_context;
4908 QLIST_FOREACH(child, &bs->children, next) {
4909 bdrv_attach_aio_context(child->bs, new_context);
4911 if (bs->drv->bdrv_attach_aio_context) {
4912 bs->drv->bdrv_attach_aio_context(bs, new_context);
4915 assert(!bs->walking_aio_notifiers);
4916 bs->walking_aio_notifiers = true;
4917 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4918 if (ban->deleted) {
4919 bdrv_do_remove_aio_context_notifier(ban);
4920 } else {
4921 ban->attached_aio_context(new_context, ban->opaque);
4924 bs->walking_aio_notifiers = false;
4927 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4929 AioContext *ctx = bdrv_get_aio_context(bs);
4931 aio_disable_external(ctx);
4932 bdrv_parent_drained_begin(bs, NULL);
4933 bdrv_drain(bs); /* ensure there are no in-flight requests */
4935 while (aio_poll(ctx, false)) {
4936 /* wait for all bottom halves to execute */
4939 bdrv_detach_aio_context(bs);
4941 /* This function executes in the old AioContext so acquire the new one in
4942 * case it runs in a different thread.
4944 aio_context_acquire(new_context);
4945 bdrv_attach_aio_context(bs, new_context);
4946 bdrv_parent_drained_end(bs, NULL);
4947 aio_enable_external(ctx);
4948 aio_context_release(new_context);
4951 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4952 void (*attached_aio_context)(AioContext *new_context, void *opaque),
4953 void (*detach_aio_context)(void *opaque), void *opaque)
4955 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4956 *ban = (BdrvAioNotifier){
4957 .attached_aio_context = attached_aio_context,
4958 .detach_aio_context = detach_aio_context,
4959 .opaque = opaque
4962 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4965 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4966 void (*attached_aio_context)(AioContext *,
4967 void *),
4968 void (*detach_aio_context)(void *),
4969 void *opaque)
4971 BdrvAioNotifier *ban, *ban_next;
4973 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4974 if (ban->attached_aio_context == attached_aio_context &&
4975 ban->detach_aio_context == detach_aio_context &&
4976 ban->opaque == opaque &&
4977 ban->deleted == false)
4979 if (bs->walking_aio_notifiers) {
4980 ban->deleted = true;
4981 } else {
4982 bdrv_do_remove_aio_context_notifier(ban);
4984 return;
4988 abort();
4991 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4992 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4994 if (!bs->drv) {
4995 return -ENOMEDIUM;
4997 if (!bs->drv->bdrv_amend_options) {
4998 return -ENOTSUP;
5000 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
5003 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5004 * of block filter and by bdrv_is_first_non_filter.
5005 * It is used to test if the given bs is the candidate or recurse more in the
5006 * node graph.
5008 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5009 BlockDriverState *candidate)
5011 /* return false if basic checks fails */
5012 if (!bs || !bs->drv) {
5013 return false;
5016 /* the code reached a non block filter driver -> check if the bs is
5017 * the same as the candidate. It's the recursion termination condition.
5019 if (!bs->drv->is_filter) {
5020 return bs == candidate;
5022 /* Down this path the driver is a block filter driver */
5024 /* If the block filter recursion method is defined use it to recurse down
5025 * the node graph.
5027 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5028 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5031 /* the driver is a block filter but don't allow to recurse -> return false
5033 return false;
5036 /* This function checks if the candidate is the first non filter bs down it's
5037 * bs chain. Since we don't have pointers to parents it explore all bs chains
5038 * from the top. Some filters can choose not to pass down the recursion.
5040 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5042 BlockDriverState *bs;
5043 BdrvNextIterator it;
5045 /* walk down the bs forest recursively */
5046 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5047 bool perm;
5049 /* try to recurse in this top level bs */
5050 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5052 /* candidate is the first non filter */
5053 if (perm) {
5054 bdrv_next_cleanup(&it);
5055 return true;
5059 return false;
5062 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5063 const char *node_name, Error **errp)
5065 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5066 AioContext *aio_context;
5068 if (!to_replace_bs) {
5069 error_setg(errp, "Node name '%s' not found", node_name);
5070 return NULL;
5073 aio_context = bdrv_get_aio_context(to_replace_bs);
5074 aio_context_acquire(aio_context);
5076 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5077 to_replace_bs = NULL;
5078 goto out;
5081 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5082 * most non filter in order to prevent data corruption.
5083 * Another benefit is that this tests exclude backing files which are
5084 * blocked by the backing blockers.
5086 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5087 error_setg(errp, "Only top most non filter can be replaced");
5088 to_replace_bs = NULL;
5089 goto out;
5092 out:
5093 aio_context_release(aio_context);
5094 return to_replace_bs;
5097 static bool append_open_options(QDict *d, BlockDriverState *bs)
5099 const QDictEntry *entry;
5100 QemuOptDesc *desc;
5101 BdrvChild *child;
5102 bool found_any = false;
5103 const char *p;
5105 for (entry = qdict_first(bs->options); entry;
5106 entry = qdict_next(bs->options, entry))
5108 /* Exclude options for children */
5109 QLIST_FOREACH(child, &bs->children, next) {
5110 if (strstart(qdict_entry_key(entry), child->name, &p)
5111 && (!*p || *p == '.'))
5113 break;
5116 if (child) {
5117 continue;
5120 /* And exclude all non-driver-specific options */
5121 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
5122 if (!strcmp(qdict_entry_key(entry), desc->name)) {
5123 break;
5126 if (desc->name) {
5127 continue;
5130 qobject_incref(qdict_entry_value(entry));
5131 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
5132 found_any = true;
5135 return found_any;
5138 /* Updates the following BDS fields:
5139 * - exact_filename: A filename which may be used for opening a block device
5140 * which (mostly) equals the given BDS (even without any
5141 * other options; so reading and writing must return the same
5142 * results, but caching etc. may be different)
5143 * - full_open_options: Options which, when given when opening a block device
5144 * (without a filename), result in a BDS (mostly)
5145 * equalling the given one
5146 * - filename: If exact_filename is set, it is copied here. Otherwise,
5147 * full_open_options is converted to a JSON object, prefixed with
5148 * "json:" (for use through the JSON pseudo protocol) and put here.
5150 void bdrv_refresh_filename(BlockDriverState *bs)
5152 BlockDriver *drv = bs->drv;
5153 QDict *opts;
5155 if (!drv) {
5156 return;
5159 /* This BDS's file name will most probably depend on its file's name, so
5160 * refresh that first */
5161 if (bs->file) {
5162 bdrv_refresh_filename(bs->file->bs);
5165 if (drv->bdrv_refresh_filename) {
5166 /* Obsolete information is of no use here, so drop the old file name
5167 * information before refreshing it */
5168 bs->exact_filename[0] = '\0';
5169 if (bs->full_open_options) {
5170 QDECREF(bs->full_open_options);
5171 bs->full_open_options = NULL;
5174 opts = qdict_new();
5175 append_open_options(opts, bs);
5176 drv->bdrv_refresh_filename(bs, opts);
5177 QDECREF(opts);
5178 } else if (bs->file) {
5179 /* Try to reconstruct valid information from the underlying file */
5180 bool has_open_options;
5182 bs->exact_filename[0] = '\0';
5183 if (bs->full_open_options) {
5184 QDECREF(bs->full_open_options);
5185 bs->full_open_options = NULL;
5188 opts = qdict_new();
5189 has_open_options = append_open_options(opts, bs);
5191 /* If no specific options have been given for this BDS, the filename of
5192 * the underlying file should suffice for this one as well */
5193 if (bs->file->bs->exact_filename[0] && !has_open_options) {
5194 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5196 /* Reconstructing the full options QDict is simple for most format block
5197 * drivers, as long as the full options are known for the underlying
5198 * file BDS. The full options QDict of that file BDS should somehow
5199 * contain a representation of the filename, therefore the following
5200 * suffices without querying the (exact_)filename of this BDS. */
5201 if (bs->file->bs->full_open_options) {
5202 qdict_put_str(opts, "driver", drv->format_name);
5203 QINCREF(bs->file->bs->full_open_options);
5204 qdict_put(opts, "file", bs->file->bs->full_open_options);
5206 bs->full_open_options = opts;
5207 } else {
5208 QDECREF(opts);
5210 } else if (!bs->full_open_options && qdict_size(bs->options)) {
5211 /* There is no underlying file BDS (at least referenced by BDS.file),
5212 * so the full options QDict should be equal to the options given
5213 * specifically for this block device when it was opened (plus the
5214 * driver specification).
5215 * Because those options don't change, there is no need to update
5216 * full_open_options when it's already set. */
5218 opts = qdict_new();
5219 append_open_options(opts, bs);
5220 qdict_put_str(opts, "driver", drv->format_name);
5222 if (bs->exact_filename[0]) {
5223 /* This may not work for all block protocol drivers (some may
5224 * require this filename to be parsed), but we have to find some
5225 * default solution here, so just include it. If some block driver
5226 * does not support pure options without any filename at all or
5227 * needs some special format of the options QDict, it needs to
5228 * implement the driver-specific bdrv_refresh_filename() function.
5230 qdict_put_str(opts, "filename", bs->exact_filename);
5233 bs->full_open_options = opts;
5236 if (bs->exact_filename[0]) {
5237 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5238 } else if (bs->full_open_options) {
5239 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5240 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5241 qstring_get_str(json));
5242 QDECREF(json);
5247 * Hot add/remove a BDS's child. So the user can take a child offline when
5248 * it is broken and take a new child online
5250 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5251 Error **errp)
5254 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5255 error_setg(errp, "The node %s does not support adding a child",
5256 bdrv_get_device_or_node_name(parent_bs));
5257 return;
5260 if (!QLIST_EMPTY(&child_bs->parents)) {
5261 error_setg(errp, "The node %s already has a parent",
5262 child_bs->node_name);
5263 return;
5266 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5269 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5271 BdrvChild *tmp;
5273 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5274 error_setg(errp, "The node %s does not support removing a child",
5275 bdrv_get_device_or_node_name(parent_bs));
5276 return;
5279 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5280 if (tmp == child) {
5281 break;
5285 if (!tmp) {
5286 error_setg(errp, "The node %s does not have a child named %s",
5287 bdrv_get_device_or_node_name(parent_bs),
5288 bdrv_get_device_or_node_name(child->bs));
5289 return;
5292 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5295 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5296 uint32_t granularity, Error **errp)
5298 BlockDriver *drv = bs->drv;
5300 if (!drv) {
5301 error_setg_errno(errp, ENOMEDIUM,
5302 "Can't store persistent bitmaps to %s",
5303 bdrv_get_device_or_node_name(bs));
5304 return false;
5307 if (!drv->bdrv_can_store_new_dirty_bitmap) {
5308 error_setg_errno(errp, ENOTSUP,
5309 "Can't store persistent bitmaps to %s",
5310 bdrv_get_device_or_node_name(bs));
5311 return false;
5314 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);