block/backup: hoist bitmap check into QMP interface
[qemu/ar7.git] / block.c
blob3e698e9cabdd2da302697f2ffa5f9f31d12de8cf
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
50 #include "qemu/id.h"
52 #ifdef CONFIG_BSD
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
55 #ifndef __DragonFly__
56 #include <sys/disk.h>
57 #endif
58 #endif
60 #ifdef _WIN32
61 #include <windows.h>
62 #endif
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
66 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
69 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
72 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
75 static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const BdrvChildRole *child_role,
80 Error **errp);
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist;
85 #ifdef _WIN32
86 static int is_windows_drive_prefix(const char *filename)
88 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
89 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
90 filename[1] == ':');
93 int is_windows_drive(const char *filename)
95 if (is_windows_drive_prefix(filename) &&
96 filename[2] == '\0')
97 return 1;
98 if (strstart(filename, "\\\\.\\", NULL) ||
99 strstart(filename, "//./", NULL))
100 return 1;
101 return 0;
103 #endif
105 size_t bdrv_opt_mem_align(BlockDriverState *bs)
107 if (!bs || !bs->drv) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, getpagesize());
112 return bs->bl.opt_mem_alignment;
115 size_t bdrv_min_mem_align(BlockDriverState *bs)
117 if (!bs || !bs->drv) {
118 /* page size or 4k (hdd sector size) should be on the safe side */
119 return MAX(4096, getpagesize());
122 return bs->bl.min_mem_alignment;
125 /* check if the path starts with "<protocol>:" */
126 int path_has_protocol(const char *path)
128 const char *p;
130 #ifdef _WIN32
131 if (is_windows_drive(path) ||
132 is_windows_drive_prefix(path)) {
133 return 0;
135 p = path + strcspn(path, ":/\\");
136 #else
137 p = path + strcspn(path, ":/");
138 #endif
140 return *p == ':';
143 int path_is_absolute(const char *path)
145 #ifdef _WIN32
146 /* specific case for names like: "\\.\d:" */
147 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
148 return 1;
150 return (*path == '/' || *path == '\\');
151 #else
152 return (*path == '/');
153 #endif
156 /* if filename is absolute, just return its duplicate. Otherwise, build a
157 path to it by considering it is relative to base_path. URL are
158 supported. */
159 char *path_combine(const char *base_path, const char *filename)
161 const char *protocol_stripped = NULL;
162 const char *p, *p1;
163 char *result;
164 int len;
166 if (path_is_absolute(filename)) {
167 return g_strdup(filename);
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;
187 #endif
188 if (p1) {
189 p1++;
190 } else {
191 p1 = base_path;
193 if (p1 > p) {
194 p = p1;
196 len = p - base_path;
198 result = g_malloc(len + strlen(filename) + 1);
199 memcpy(result, base_path, len);
200 strcpy(result + len, filename);
202 return result;
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
210 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
211 QDict *options)
213 if (strstart(filename, prefix, &filename)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename)) {
217 QString *fat_filename;
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename));
223 /* And we can thus fix the protocol detection issue by prefixing it
224 * by "./" */
225 fat_filename = qstring_from_str("./");
226 qstring_append(fat_filename, filename);
228 assert(!path_has_protocol(qstring_get_str(fat_filename)));
230 qdict_put(options, "filename", fat_filename);
231 } else {
232 /* If no protocol prefix was detected, we can use the shortened
233 * filename as-is */
234 qdict_put_str(options, "filename", filename);
240 /* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
243 bool bdrv_is_read_only(BlockDriverState *bs)
245 return bs->read_only;
248 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
249 bool ignore_allow_rdw, Error **errp)
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs->copy_on_read && read_only) {
253 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs));
255 return -EINVAL;
258 /* Do not clear read_only if it is prohibited */
259 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
260 !ignore_allow_rdw)
262 error_setg(errp, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs));
264 return -EPERM;
267 return 0;
271 * Called by a driver that can only provide a read-only image.
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
280 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
281 Error **errp)
283 int ret = 0;
285 if (!(bs->open_flags & BDRV_O_RDWR)) {
286 return 0;
288 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
289 goto fail;
292 ret = bdrv_can_set_read_only(bs, true, false, NULL);
293 if (ret < 0) {
294 goto fail;
297 bs->read_only = true;
298 bs->open_flags &= ~BDRV_O_RDWR;
300 return 0;
302 fail:
303 error_setg(errp, "%s", errmsg ?: "Image is read-only");
304 return -EACCES;
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
310 * set.
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
316 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
317 const char *backing,
318 Error **errp)
320 if (backing[0] == '\0') {
321 return NULL;
322 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
323 return g_strdup(backing);
324 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
325 error_setg(errp, "Cannot use relative backing file names for '%s'",
326 backed);
327 return NULL;
328 } else {
329 return path_combine(backed, backing);
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
336 * @errp set.
338 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
339 const char *filename, Error **errp)
341 char *dir, *full_name;
343 if (!filename || filename[0] == '\0') {
344 return NULL;
345 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
346 return g_strdup(filename);
349 dir = bdrv_dirname(relative_to, errp);
350 if (!dir) {
351 return NULL;
354 full_name = g_strconcat(dir, filename, NULL);
355 g_free(dir);
356 return full_name;
359 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
361 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
364 void bdrv_register(BlockDriver *bdrv)
366 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
369 BlockDriverState *bdrv_new(void)
371 BlockDriverState *bs;
372 int i;
374 bs = g_new0(BlockDriverState, 1);
375 QLIST_INIT(&bs->dirty_bitmaps);
376 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
377 QLIST_INIT(&bs->op_blockers[i]);
379 notifier_with_return_list_init(&bs->before_write_notifiers);
380 qemu_co_mutex_init(&bs->reqs_lock);
381 qemu_mutex_init(&bs->dirty_bitmap_mutex);
382 bs->refcnt = 1;
383 bs->aio_context = qemu_get_aio_context();
385 qemu_co_queue_init(&bs->flush_queue);
387 for (i = 0; i < bdrv_drain_all_count; i++) {
388 bdrv_drained_begin(bs);
391 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
393 return bs;
396 static BlockDriver *bdrv_do_find_format(const char *format_name)
398 BlockDriver *drv1;
400 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
401 if (!strcmp(drv1->format_name, format_name)) {
402 return drv1;
406 return NULL;
409 BlockDriver *bdrv_find_format(const char *format_name)
411 BlockDriver *drv1;
412 int i;
414 drv1 = bdrv_do_find_format(format_name);
415 if (drv1) {
416 return drv1;
419 /* The driver isn't registered, maybe we need to load a module */
420 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
421 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
422 block_module_load_one(block_driver_modules[i].library_name);
423 break;
427 return bdrv_do_find_format(format_name);
430 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
432 static const char *whitelist_rw[] = {
433 CONFIG_BDRV_RW_WHITELIST
435 static const char *whitelist_ro[] = {
436 CONFIG_BDRV_RO_WHITELIST
438 const char **p;
440 if (!whitelist_rw[0] && !whitelist_ro[0]) {
441 return 1; /* no whitelist, anything goes */
444 for (p = whitelist_rw; *p; p++) {
445 if (!strcmp(format_name, *p)) {
446 return 1;
449 if (read_only) {
450 for (p = whitelist_ro; *p; p++) {
451 if (!strcmp(format_name, *p)) {
452 return 1;
456 return 0;
459 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
461 return bdrv_format_is_whitelisted(drv->format_name, read_only);
464 bool bdrv_uses_whitelist(void)
466 return use_bdrv_whitelist;
469 typedef struct CreateCo {
470 BlockDriver *drv;
471 char *filename;
472 QemuOpts *opts;
473 int ret;
474 Error *err;
475 } CreateCo;
477 static void coroutine_fn bdrv_create_co_entry(void *opaque)
479 Error *local_err = NULL;
480 int ret;
482 CreateCo *cco = opaque;
483 assert(cco->drv);
485 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
486 error_propagate(&cco->err, local_err);
487 cco->ret = ret;
490 int bdrv_create(BlockDriver *drv, const char* filename,
491 QemuOpts *opts, Error **errp)
493 int ret;
495 Coroutine *co;
496 CreateCo cco = {
497 .drv = drv,
498 .filename = g_strdup(filename),
499 .opts = opts,
500 .ret = NOT_DONE,
501 .err = NULL,
504 if (!drv->bdrv_co_create_opts) {
505 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
506 ret = -ENOTSUP;
507 goto out;
510 if (qemu_in_coroutine()) {
511 /* Fast-path if already in coroutine context */
512 bdrv_create_co_entry(&cco);
513 } else {
514 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
515 qemu_coroutine_enter(co);
516 while (cco.ret == NOT_DONE) {
517 aio_poll(qemu_get_aio_context(), true);
521 ret = cco.ret;
522 if (ret < 0) {
523 if (cco.err) {
524 error_propagate(errp, cco.err);
525 } else {
526 error_setg_errno(errp, -ret, "Could not create image");
530 out:
531 g_free(cco.filename);
532 return ret;
535 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
537 BlockDriver *drv;
538 Error *local_err = NULL;
539 int ret;
541 drv = bdrv_find_protocol(filename, true, errp);
542 if (drv == NULL) {
543 return -ENOENT;
546 ret = bdrv_create(drv, filename, opts, &local_err);
547 error_propagate(errp, local_err);
548 return ret;
552 * Try to get @bs's logical and physical block size.
553 * On success, store them in @bsz struct and return 0.
554 * On failure return -errno.
555 * @bs must not be empty.
557 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
559 BlockDriver *drv = bs->drv;
561 if (drv && drv->bdrv_probe_blocksizes) {
562 return drv->bdrv_probe_blocksizes(bs, bsz);
563 } else if (drv && drv->is_filter && bs->file) {
564 return bdrv_probe_blocksizes(bs->file->bs, bsz);
567 return -ENOTSUP;
571 * Try to get @bs's geometry (cyls, heads, sectors).
572 * On success, store them in @geo struct and return 0.
573 * On failure return -errno.
574 * @bs must not be empty.
576 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
578 BlockDriver *drv = bs->drv;
580 if (drv && drv->bdrv_probe_geometry) {
581 return drv->bdrv_probe_geometry(bs, geo);
582 } else if (drv && drv->is_filter && bs->file) {
583 return bdrv_probe_geometry(bs->file->bs, geo);
586 return -ENOTSUP;
590 * Create a uniquely-named empty temporary file.
591 * Return 0 upon success, otherwise a negative errno value.
593 int get_tmp_filename(char *filename, int size)
595 #ifdef _WIN32
596 char temp_dir[MAX_PATH];
597 /* GetTempFileName requires that its output buffer (4th param)
598 have length MAX_PATH or greater. */
599 assert(size >= MAX_PATH);
600 return (GetTempPath(MAX_PATH, temp_dir)
601 && GetTempFileName(temp_dir, "qem", 0, filename)
602 ? 0 : -GetLastError());
603 #else
604 int fd;
605 const char *tmpdir;
606 tmpdir = getenv("TMPDIR");
607 if (!tmpdir) {
608 tmpdir = "/var/tmp";
610 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
611 return -EOVERFLOW;
613 fd = mkstemp(filename);
614 if (fd < 0) {
615 return -errno;
617 if (close(fd) != 0) {
618 unlink(filename);
619 return -errno;
621 return 0;
622 #endif
626 * Detect host devices. By convention, /dev/cdrom[N] is always
627 * recognized as a host CDROM.
629 static BlockDriver *find_hdev_driver(const char *filename)
631 int score_max = 0, score;
632 BlockDriver *drv = NULL, *d;
634 QLIST_FOREACH(d, &bdrv_drivers, list) {
635 if (d->bdrv_probe_device) {
636 score = d->bdrv_probe_device(filename);
637 if (score > score_max) {
638 score_max = score;
639 drv = d;
644 return drv;
647 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
649 BlockDriver *drv1;
651 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
652 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
653 return drv1;
657 return NULL;
660 BlockDriver *bdrv_find_protocol(const char *filename,
661 bool allow_protocol_prefix,
662 Error **errp)
664 BlockDriver *drv1;
665 char protocol[128];
666 int len;
667 const char *p;
668 int i;
670 /* TODO Drivers without bdrv_file_open must be specified explicitly */
673 * XXX(hch): we really should not let host device detection
674 * override an explicit protocol specification, but moving this
675 * later breaks access to device names with colons in them.
676 * Thanks to the brain-dead persistent naming schemes on udev-
677 * based Linux systems those actually are quite common.
679 drv1 = find_hdev_driver(filename);
680 if (drv1) {
681 return drv1;
684 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
685 return &bdrv_file;
688 p = strchr(filename, ':');
689 assert(p != NULL);
690 len = p - filename;
691 if (len > sizeof(protocol) - 1)
692 len = sizeof(protocol) - 1;
693 memcpy(protocol, filename, len);
694 protocol[len] = '\0';
696 drv1 = bdrv_do_find_protocol(protocol);
697 if (drv1) {
698 return drv1;
701 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
702 if (block_driver_modules[i].protocol_name &&
703 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
704 block_module_load_one(block_driver_modules[i].library_name);
705 break;
709 drv1 = bdrv_do_find_protocol(protocol);
710 if (!drv1) {
711 error_setg(errp, "Unknown protocol '%s'", protocol);
713 return drv1;
717 * Guess image format by probing its contents.
718 * This is not a good idea when your image is raw (CVE-2008-2004), but
719 * we do it anyway for backward compatibility.
721 * @buf contains the image's first @buf_size bytes.
722 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
723 * but can be smaller if the image file is smaller)
724 * @filename is its filename.
726 * For all block drivers, call the bdrv_probe() method to get its
727 * probing score.
728 * Return the first block driver with the highest probing score.
730 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
731 const char *filename)
733 int score_max = 0, score;
734 BlockDriver *drv = NULL, *d;
736 QLIST_FOREACH(d, &bdrv_drivers, list) {
737 if (d->bdrv_probe) {
738 score = d->bdrv_probe(buf, buf_size, filename);
739 if (score > score_max) {
740 score_max = score;
741 drv = d;
746 return drv;
749 static int find_image_format(BlockBackend *file, const char *filename,
750 BlockDriver **pdrv, Error **errp)
752 BlockDriver *drv;
753 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
754 int ret = 0;
756 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
757 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
758 *pdrv = &bdrv_raw;
759 return ret;
762 ret = blk_pread(file, 0, buf, sizeof(buf));
763 if (ret < 0) {
764 error_setg_errno(errp, -ret, "Could not read image for determining its "
765 "format");
766 *pdrv = NULL;
767 return ret;
770 drv = bdrv_probe_all(buf, ret, filename);
771 if (!drv) {
772 error_setg(errp, "Could not determine image format: No compatible "
773 "driver found");
774 ret = -ENOENT;
776 *pdrv = drv;
777 return ret;
781 * Set the current 'total_sectors' value
782 * Return 0 on success, -errno on error.
784 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
786 BlockDriver *drv = bs->drv;
788 if (!drv) {
789 return -ENOMEDIUM;
792 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
793 if (bdrv_is_sg(bs))
794 return 0;
796 /* query actual device if possible, otherwise just trust the hint */
797 if (drv->bdrv_getlength) {
798 int64_t length = drv->bdrv_getlength(bs);
799 if (length < 0) {
800 return length;
802 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
805 bs->total_sectors = hint;
806 return 0;
810 * Combines a QDict of new block driver @options with any missing options taken
811 * from @old_options, so that leaving out an option defaults to its old value.
813 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
814 QDict *old_options)
816 if (bs->drv && bs->drv->bdrv_join_options) {
817 bs->drv->bdrv_join_options(options, old_options);
818 } else {
819 qdict_join(options, old_options, false);
823 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
824 int open_flags,
825 Error **errp)
827 Error *local_err = NULL;
828 char *value = qemu_opt_get_del(opts, "detect-zeroes");
829 BlockdevDetectZeroesOptions detect_zeroes =
830 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
831 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
832 g_free(value);
833 if (local_err) {
834 error_propagate(errp, local_err);
835 return detect_zeroes;
838 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
839 !(open_flags & BDRV_O_UNMAP))
841 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
842 "without setting discard operation to unmap");
845 return detect_zeroes;
849 * Set open flags for a given discard mode
851 * Return 0 on success, -1 if the discard mode was invalid.
853 int bdrv_parse_discard_flags(const char *mode, int *flags)
855 *flags &= ~BDRV_O_UNMAP;
857 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
858 /* do nothing */
859 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
860 *flags |= BDRV_O_UNMAP;
861 } else {
862 return -1;
865 return 0;
869 * Set open flags for a given cache mode
871 * Return 0 on success, -1 if the cache mode was invalid.
873 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
875 *flags &= ~BDRV_O_CACHE_MASK;
877 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
878 *writethrough = false;
879 *flags |= BDRV_O_NOCACHE;
880 } else if (!strcmp(mode, "directsync")) {
881 *writethrough = true;
882 *flags |= BDRV_O_NOCACHE;
883 } else if (!strcmp(mode, "writeback")) {
884 *writethrough = false;
885 } else if (!strcmp(mode, "unsafe")) {
886 *writethrough = false;
887 *flags |= BDRV_O_NO_FLUSH;
888 } else if (!strcmp(mode, "writethrough")) {
889 *writethrough = true;
890 } else {
891 return -1;
894 return 0;
897 static char *bdrv_child_get_parent_desc(BdrvChild *c)
899 BlockDriverState *parent = c->opaque;
900 return g_strdup(bdrv_get_device_or_node_name(parent));
903 static void bdrv_child_cb_drained_begin(BdrvChild *child)
905 BlockDriverState *bs = child->opaque;
906 bdrv_do_drained_begin_quiesce(bs, NULL, false);
909 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
911 BlockDriverState *bs = child->opaque;
912 return bdrv_drain_poll(bs, false, NULL, false);
915 static void bdrv_child_cb_drained_end(BdrvChild *child,
916 int *drained_end_counter)
918 BlockDriverState *bs = child->opaque;
919 bdrv_drained_end_no_poll(bs, drained_end_counter);
922 static void bdrv_child_cb_attach(BdrvChild *child)
924 BlockDriverState *bs = child->opaque;
925 bdrv_apply_subtree_drain(child, bs);
928 static void bdrv_child_cb_detach(BdrvChild *child)
930 BlockDriverState *bs = child->opaque;
931 bdrv_unapply_subtree_drain(child, bs);
934 static int bdrv_child_cb_inactivate(BdrvChild *child)
936 BlockDriverState *bs = child->opaque;
937 assert(bs->open_flags & BDRV_O_INACTIVE);
938 return 0;
941 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
942 GSList **ignore, Error **errp)
944 BlockDriverState *bs = child->opaque;
945 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
948 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
949 GSList **ignore)
951 BlockDriverState *bs = child->opaque;
952 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
956 * Returns the options and flags that a temporary snapshot should get, based on
957 * the originally requested flags (the originally requested image will have
958 * flags like a backing file)
960 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
961 int parent_flags, QDict *parent_options)
963 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
965 /* For temporary files, unconditional cache=unsafe is fine */
966 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
967 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
969 /* Copy the read-only and discard options from the parent */
970 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
971 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
973 /* aio=native doesn't work for cache.direct=off, so disable it for the
974 * temporary snapshot */
975 *child_flags &= ~BDRV_O_NATIVE_AIO;
979 * Returns the options and flags that bs->file should get if a protocol driver
980 * is expected, based on the given options and flags for the parent BDS
982 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
983 int parent_flags, QDict *parent_options)
985 int flags = parent_flags;
987 /* Enable protocol handling, disable format probing for bs->file */
988 flags |= BDRV_O_PROTOCOL;
990 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
991 * the parent. */
992 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
993 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
994 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
996 /* Inherit the read-only option from the parent if it's not set */
997 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
998 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
1000 /* Our block drivers take care to send flushes and respect unmap policy,
1001 * so we can default to enable both on lower layers regardless of the
1002 * corresponding parent options. */
1003 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1005 /* Clear flags that only apply to the top layer */
1006 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
1007 BDRV_O_NO_IO);
1009 *child_flags = flags;
1012 const BdrvChildRole child_file = {
1013 .parent_is_bds = true,
1014 .get_parent_desc = bdrv_child_get_parent_desc,
1015 .inherit_options = bdrv_inherited_options,
1016 .drained_begin = bdrv_child_cb_drained_begin,
1017 .drained_poll = bdrv_child_cb_drained_poll,
1018 .drained_end = bdrv_child_cb_drained_end,
1019 .attach = bdrv_child_cb_attach,
1020 .detach = bdrv_child_cb_detach,
1021 .inactivate = bdrv_child_cb_inactivate,
1022 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1023 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1027 * Returns the options and flags that bs->file should get if the use of formats
1028 * (and not only protocols) is permitted for it, based on the given options and
1029 * flags for the parent BDS
1031 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1032 int parent_flags, QDict *parent_options)
1034 child_file.inherit_options(child_flags, child_options,
1035 parent_flags, parent_options);
1037 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
1040 const BdrvChildRole child_format = {
1041 .parent_is_bds = true,
1042 .get_parent_desc = bdrv_child_get_parent_desc,
1043 .inherit_options = bdrv_inherited_fmt_options,
1044 .drained_begin = bdrv_child_cb_drained_begin,
1045 .drained_poll = bdrv_child_cb_drained_poll,
1046 .drained_end = bdrv_child_cb_drained_end,
1047 .attach = bdrv_child_cb_attach,
1048 .detach = bdrv_child_cb_detach,
1049 .inactivate = bdrv_child_cb_inactivate,
1050 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1051 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1054 static void bdrv_backing_attach(BdrvChild *c)
1056 BlockDriverState *parent = c->opaque;
1057 BlockDriverState *backing_hd = c->bs;
1059 assert(!parent->backing_blocker);
1060 error_setg(&parent->backing_blocker,
1061 "node is used as backing hd of '%s'",
1062 bdrv_get_device_or_node_name(parent));
1064 bdrv_refresh_filename(backing_hd);
1066 parent->open_flags &= ~BDRV_O_NO_BACKING;
1067 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1068 backing_hd->filename);
1069 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1070 backing_hd->drv ? backing_hd->drv->format_name : "");
1072 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1073 /* Otherwise we won't be able to commit or stream */
1074 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1075 parent->backing_blocker);
1076 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1077 parent->backing_blocker);
1079 * We do backup in 3 ways:
1080 * 1. drive backup
1081 * The target bs is new opened, and the source is top BDS
1082 * 2. blockdev backup
1083 * Both the source and the target are top BDSes.
1084 * 3. internal backup(used for block replication)
1085 * Both the source and the target are backing file
1087 * In case 1 and 2, neither the source nor the target is the backing file.
1088 * In case 3, we will block the top BDS, so there is only one block job
1089 * for the top BDS and its backing chain.
1091 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1092 parent->backing_blocker);
1093 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1094 parent->backing_blocker);
1096 bdrv_child_cb_attach(c);
1099 static void bdrv_backing_detach(BdrvChild *c)
1101 BlockDriverState *parent = c->opaque;
1103 assert(parent->backing_blocker);
1104 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1105 error_free(parent->backing_blocker);
1106 parent->backing_blocker = NULL;
1108 bdrv_child_cb_detach(c);
1112 * Returns the options and flags that bs->backing should get, based on the
1113 * given options and flags for the parent BDS
1115 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1116 int parent_flags, QDict *parent_options)
1118 int flags = parent_flags;
1120 /* The cache mode is inherited unmodified for backing files; except WCE,
1121 * which is only applied on the top level (BlockBackend) */
1122 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1123 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1124 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1126 /* backing files always opened read-only */
1127 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1128 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1129 flags &= ~BDRV_O_COPY_ON_READ;
1131 /* snapshot=on is handled on the top layer */
1132 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1134 *child_flags = flags;
1137 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1138 const char *filename, Error **errp)
1140 BlockDriverState *parent = c->opaque;
1141 bool read_only = bdrv_is_read_only(parent);
1142 int ret;
1144 if (read_only) {
1145 ret = bdrv_reopen_set_read_only(parent, false, errp);
1146 if (ret < 0) {
1147 return ret;
1151 ret = bdrv_change_backing_file(parent, filename,
1152 base->drv ? base->drv->format_name : "");
1153 if (ret < 0) {
1154 error_setg_errno(errp, -ret, "Could not update backing file link");
1157 if (read_only) {
1158 bdrv_reopen_set_read_only(parent, true, NULL);
1161 return ret;
1164 const BdrvChildRole child_backing = {
1165 .parent_is_bds = true,
1166 .get_parent_desc = bdrv_child_get_parent_desc,
1167 .attach = bdrv_backing_attach,
1168 .detach = bdrv_backing_detach,
1169 .inherit_options = bdrv_backing_options,
1170 .drained_begin = bdrv_child_cb_drained_begin,
1171 .drained_poll = bdrv_child_cb_drained_poll,
1172 .drained_end = bdrv_child_cb_drained_end,
1173 .inactivate = bdrv_child_cb_inactivate,
1174 .update_filename = bdrv_backing_update_filename,
1175 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1176 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1179 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1181 int open_flags = flags;
1184 * Clear flags that are internal to the block layer before opening the
1185 * image.
1187 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1189 return open_flags;
1192 static void update_flags_from_options(int *flags, QemuOpts *opts)
1194 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1196 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1197 *flags |= BDRV_O_NO_FLUSH;
1200 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1201 *flags |= BDRV_O_NOCACHE;
1204 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1205 *flags |= BDRV_O_RDWR;
1208 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1209 *flags |= BDRV_O_AUTO_RDONLY;
1213 static void update_options_from_flags(QDict *options, int flags)
1215 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1216 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1218 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1219 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1220 flags & BDRV_O_NO_FLUSH);
1222 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1223 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1225 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1226 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1227 flags & BDRV_O_AUTO_RDONLY);
1231 static void bdrv_assign_node_name(BlockDriverState *bs,
1232 const char *node_name,
1233 Error **errp)
1235 char *gen_node_name = NULL;
1237 if (!node_name) {
1238 node_name = gen_node_name = id_generate(ID_BLOCK);
1239 } else if (!id_wellformed(node_name)) {
1241 * Check for empty string or invalid characters, but not if it is
1242 * generated (generated names use characters not available to the user)
1244 error_setg(errp, "Invalid node name");
1245 return;
1248 /* takes care of avoiding namespaces collisions */
1249 if (blk_by_name(node_name)) {
1250 error_setg(errp, "node-name=%s is conflicting with a device id",
1251 node_name);
1252 goto out;
1255 /* takes care of avoiding duplicates node names */
1256 if (bdrv_find_node(node_name)) {
1257 error_setg(errp, "Duplicate node name");
1258 goto out;
1261 /* Make sure that the node name isn't truncated */
1262 if (strlen(node_name) >= sizeof(bs->node_name)) {
1263 error_setg(errp, "Node name too long");
1264 goto out;
1267 /* copy node name into the bs and insert it into the graph list */
1268 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1269 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1270 out:
1271 g_free(gen_node_name);
1274 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1275 const char *node_name, QDict *options,
1276 int open_flags, Error **errp)
1278 Error *local_err = NULL;
1279 int i, ret;
1281 bdrv_assign_node_name(bs, node_name, &local_err);
1282 if (local_err) {
1283 error_propagate(errp, local_err);
1284 return -EINVAL;
1287 bs->drv = drv;
1288 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1289 bs->opaque = g_malloc0(drv->instance_size);
1291 if (drv->bdrv_file_open) {
1292 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1293 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1294 } else if (drv->bdrv_open) {
1295 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1296 } else {
1297 ret = 0;
1300 if (ret < 0) {
1301 if (local_err) {
1302 error_propagate(errp, local_err);
1303 } else if (bs->filename[0]) {
1304 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1305 } else {
1306 error_setg_errno(errp, -ret, "Could not open image");
1308 goto open_failed;
1311 ret = refresh_total_sectors(bs, bs->total_sectors);
1312 if (ret < 0) {
1313 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1314 return ret;
1317 bdrv_refresh_limits(bs, &local_err);
1318 if (local_err) {
1319 error_propagate(errp, local_err);
1320 return -EINVAL;
1323 assert(bdrv_opt_mem_align(bs) != 0);
1324 assert(bdrv_min_mem_align(bs) != 0);
1325 assert(is_power_of_2(bs->bl.request_alignment));
1327 for (i = 0; i < bs->quiesce_counter; i++) {
1328 if (drv->bdrv_co_drain_begin) {
1329 drv->bdrv_co_drain_begin(bs);
1333 return 0;
1334 open_failed:
1335 bs->drv = NULL;
1336 if (bs->file != NULL) {
1337 bdrv_unref_child(bs, bs->file);
1338 bs->file = NULL;
1340 g_free(bs->opaque);
1341 bs->opaque = NULL;
1342 return ret;
1345 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1346 int flags, Error **errp)
1348 BlockDriverState *bs;
1349 int ret;
1351 bs = bdrv_new();
1352 bs->open_flags = flags;
1353 bs->explicit_options = qdict_new();
1354 bs->options = qdict_new();
1355 bs->opaque = NULL;
1357 update_options_from_flags(bs->options, flags);
1359 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1360 if (ret < 0) {
1361 qobject_unref(bs->explicit_options);
1362 bs->explicit_options = NULL;
1363 qobject_unref(bs->options);
1364 bs->options = NULL;
1365 bdrv_unref(bs);
1366 return NULL;
1369 return bs;
1372 QemuOptsList bdrv_runtime_opts = {
1373 .name = "bdrv_common",
1374 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1375 .desc = {
1377 .name = "node-name",
1378 .type = QEMU_OPT_STRING,
1379 .help = "Node name of the block device node",
1382 .name = "driver",
1383 .type = QEMU_OPT_STRING,
1384 .help = "Block driver to use for the node",
1387 .name = BDRV_OPT_CACHE_DIRECT,
1388 .type = QEMU_OPT_BOOL,
1389 .help = "Bypass software writeback cache on the host",
1392 .name = BDRV_OPT_CACHE_NO_FLUSH,
1393 .type = QEMU_OPT_BOOL,
1394 .help = "Ignore flush requests",
1397 .name = BDRV_OPT_READ_ONLY,
1398 .type = QEMU_OPT_BOOL,
1399 .help = "Node is opened in read-only mode",
1402 .name = BDRV_OPT_AUTO_READ_ONLY,
1403 .type = QEMU_OPT_BOOL,
1404 .help = "Node can become read-only if opening read-write fails",
1407 .name = "detect-zeroes",
1408 .type = QEMU_OPT_STRING,
1409 .help = "try to optimize zero writes (off, on, unmap)",
1412 .name = BDRV_OPT_DISCARD,
1413 .type = QEMU_OPT_STRING,
1414 .help = "discard operation (ignore/off, unmap/on)",
1417 .name = BDRV_OPT_FORCE_SHARE,
1418 .type = QEMU_OPT_BOOL,
1419 .help = "always accept other writers (default: off)",
1421 { /* end of list */ }
1426 * Common part for opening disk images and files
1428 * Removes all processed options from *options.
1430 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1431 QDict *options, Error **errp)
1433 int ret, open_flags;
1434 const char *filename;
1435 const char *driver_name = NULL;
1436 const char *node_name = NULL;
1437 const char *discard;
1438 QemuOpts *opts;
1439 BlockDriver *drv;
1440 Error *local_err = NULL;
1442 assert(bs->file == NULL);
1443 assert(options != NULL && bs->options != options);
1445 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1446 qemu_opts_absorb_qdict(opts, options, &local_err);
1447 if (local_err) {
1448 error_propagate(errp, local_err);
1449 ret = -EINVAL;
1450 goto fail_opts;
1453 update_flags_from_options(&bs->open_flags, opts);
1455 driver_name = qemu_opt_get(opts, "driver");
1456 drv = bdrv_find_format(driver_name);
1457 assert(drv != NULL);
1459 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1461 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1462 error_setg(errp,
1463 BDRV_OPT_FORCE_SHARE
1464 "=on can only be used with read-only images");
1465 ret = -EINVAL;
1466 goto fail_opts;
1469 if (file != NULL) {
1470 bdrv_refresh_filename(blk_bs(file));
1471 filename = blk_bs(file)->filename;
1472 } else {
1474 * Caution: while qdict_get_try_str() is fine, getting
1475 * non-string types would require more care. When @options
1476 * come from -blockdev or blockdev_add, its members are typed
1477 * according to the QAPI schema, but when they come from
1478 * -drive, they're all QString.
1480 filename = qdict_get_try_str(options, "filename");
1483 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1484 error_setg(errp, "The '%s' block driver requires a file name",
1485 drv->format_name);
1486 ret = -EINVAL;
1487 goto fail_opts;
1490 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1491 drv->format_name);
1493 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1495 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1496 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1497 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1498 } else {
1499 ret = -ENOTSUP;
1501 if (ret < 0) {
1502 error_setg(errp,
1503 !bs->read_only && bdrv_is_whitelisted(drv, true)
1504 ? "Driver '%s' can only be used for read-only devices"
1505 : "Driver '%s' is not whitelisted",
1506 drv->format_name);
1507 goto fail_opts;
1511 /* bdrv_new() and bdrv_close() make it so */
1512 assert(atomic_read(&bs->copy_on_read) == 0);
1514 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1515 if (!bs->read_only) {
1516 bdrv_enable_copy_on_read(bs);
1517 } else {
1518 error_setg(errp, "Can't use copy-on-read on read-only device");
1519 ret = -EINVAL;
1520 goto fail_opts;
1524 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1525 if (discard != NULL) {
1526 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1527 error_setg(errp, "Invalid discard option");
1528 ret = -EINVAL;
1529 goto fail_opts;
1533 bs->detect_zeroes =
1534 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1535 if (local_err) {
1536 error_propagate(errp, local_err);
1537 ret = -EINVAL;
1538 goto fail_opts;
1541 if (filename != NULL) {
1542 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1543 } else {
1544 bs->filename[0] = '\0';
1546 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1548 /* Open the image, either directly or using a protocol */
1549 open_flags = bdrv_open_flags(bs, bs->open_flags);
1550 node_name = qemu_opt_get(opts, "node-name");
1552 assert(!drv->bdrv_file_open || file == NULL);
1553 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1554 if (ret < 0) {
1555 goto fail_opts;
1558 qemu_opts_del(opts);
1559 return 0;
1561 fail_opts:
1562 qemu_opts_del(opts);
1563 return ret;
1566 static QDict *parse_json_filename(const char *filename, Error **errp)
1568 QObject *options_obj;
1569 QDict *options;
1570 int ret;
1572 ret = strstart(filename, "json:", &filename);
1573 assert(ret);
1575 options_obj = qobject_from_json(filename, errp);
1576 if (!options_obj) {
1577 error_prepend(errp, "Could not parse the JSON options: ");
1578 return NULL;
1581 options = qobject_to(QDict, options_obj);
1582 if (!options) {
1583 qobject_unref(options_obj);
1584 error_setg(errp, "Invalid JSON object given");
1585 return NULL;
1588 qdict_flatten(options);
1590 return options;
1593 static void parse_json_protocol(QDict *options, const char **pfilename,
1594 Error **errp)
1596 QDict *json_options;
1597 Error *local_err = NULL;
1599 /* Parse json: pseudo-protocol */
1600 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1601 return;
1604 json_options = parse_json_filename(*pfilename, &local_err);
1605 if (local_err) {
1606 error_propagate(errp, local_err);
1607 return;
1610 /* Options given in the filename have lower priority than options
1611 * specified directly */
1612 qdict_join(options, json_options, false);
1613 qobject_unref(json_options);
1614 *pfilename = NULL;
1618 * Fills in default options for opening images and converts the legacy
1619 * filename/flags pair to option QDict entries.
1620 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1621 * block driver has been specified explicitly.
1623 static int bdrv_fill_options(QDict **options, const char *filename,
1624 int *flags, Error **errp)
1626 const char *drvname;
1627 bool protocol = *flags & BDRV_O_PROTOCOL;
1628 bool parse_filename = false;
1629 BlockDriver *drv = NULL;
1630 Error *local_err = NULL;
1633 * Caution: while qdict_get_try_str() is fine, getting non-string
1634 * types would require more care. When @options come from
1635 * -blockdev or blockdev_add, its members are typed according to
1636 * the QAPI schema, but when they come from -drive, they're all
1637 * QString.
1639 drvname = qdict_get_try_str(*options, "driver");
1640 if (drvname) {
1641 drv = bdrv_find_format(drvname);
1642 if (!drv) {
1643 error_setg(errp, "Unknown driver '%s'", drvname);
1644 return -ENOENT;
1646 /* If the user has explicitly specified the driver, this choice should
1647 * override the BDRV_O_PROTOCOL flag */
1648 protocol = drv->bdrv_file_open;
1651 if (protocol) {
1652 *flags |= BDRV_O_PROTOCOL;
1653 } else {
1654 *flags &= ~BDRV_O_PROTOCOL;
1657 /* Translate cache options from flags into options */
1658 update_options_from_flags(*options, *flags);
1660 /* Fetch the file name from the options QDict if necessary */
1661 if (protocol && filename) {
1662 if (!qdict_haskey(*options, "filename")) {
1663 qdict_put_str(*options, "filename", filename);
1664 parse_filename = true;
1665 } else {
1666 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1667 "the same time");
1668 return -EINVAL;
1672 /* Find the right block driver */
1673 /* See cautionary note on accessing @options above */
1674 filename = qdict_get_try_str(*options, "filename");
1676 if (!drvname && protocol) {
1677 if (filename) {
1678 drv = bdrv_find_protocol(filename, parse_filename, errp);
1679 if (!drv) {
1680 return -EINVAL;
1683 drvname = drv->format_name;
1684 qdict_put_str(*options, "driver", drvname);
1685 } else {
1686 error_setg(errp, "Must specify either driver or file");
1687 return -EINVAL;
1691 assert(drv || !protocol);
1693 /* Driver-specific filename parsing */
1694 if (drv && drv->bdrv_parse_filename && parse_filename) {
1695 drv->bdrv_parse_filename(filename, *options, &local_err);
1696 if (local_err) {
1697 error_propagate(errp, local_err);
1698 return -EINVAL;
1701 if (!drv->bdrv_needs_filename) {
1702 qdict_del(*options, "filename");
1706 return 0;
1709 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1710 uint64_t perm, uint64_t shared,
1711 GSList *ignore_children,
1712 bool *tighten_restrictions, Error **errp);
1713 static void bdrv_child_abort_perm_update(BdrvChild *c);
1714 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1715 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1716 uint64_t *shared_perm);
1718 typedef struct BlockReopenQueueEntry {
1719 bool prepared;
1720 bool perms_checked;
1721 BDRVReopenState state;
1722 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1723 } BlockReopenQueueEntry;
1726 * Return the flags that @bs will have after the reopens in @q have
1727 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1728 * return the current flags.
1730 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1732 BlockReopenQueueEntry *entry;
1734 if (q != NULL) {
1735 QSIMPLEQ_FOREACH(entry, q, entry) {
1736 if (entry->state.bs == bs) {
1737 return entry->state.flags;
1742 return bs->open_flags;
1745 /* Returns whether the image file can be written to after the reopen queue @q
1746 * has been successfully applied, or right now if @q is NULL. */
1747 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1748 BlockReopenQueue *q)
1750 int flags = bdrv_reopen_get_flags(q, bs);
1752 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1756 * Return whether the BDS can be written to. This is not necessarily
1757 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1758 * be written to but do not count as read-only images.
1760 bool bdrv_is_writable(BlockDriverState *bs)
1762 return bdrv_is_writable_after_reopen(bs, NULL);
1765 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1766 BdrvChild *c, const BdrvChildRole *role,
1767 BlockReopenQueue *reopen_queue,
1768 uint64_t parent_perm, uint64_t parent_shared,
1769 uint64_t *nperm, uint64_t *nshared)
1771 assert(bs->drv && bs->drv->bdrv_child_perm);
1772 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1773 parent_perm, parent_shared,
1774 nperm, nshared);
1775 /* TODO Take force_share from reopen_queue */
1776 if (child_bs && child_bs->force_share) {
1777 *nshared = BLK_PERM_ALL;
1782 * Check whether permissions on this node can be changed in a way that
1783 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1784 * permissions of all its parents. This involves checking whether all necessary
1785 * permission changes to child nodes can be performed.
1787 * Will set *tighten_restrictions to true if and only if new permissions have to
1788 * be taken or currently shared permissions are to be unshared. Otherwise,
1789 * errors are not fatal as long as the caller accepts that the restrictions
1790 * remain tighter than they need to be. The caller still has to abort the
1791 * transaction.
1792 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1793 * encounter fatal errors even though no restrictions are to be tightened. For
1794 * example, changing a node from RW to RO will fail if the WRITE permission is
1795 * to be kept.
1797 * A call to this function must always be followed by a call to bdrv_set_perm()
1798 * or bdrv_abort_perm_update().
1800 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1801 uint64_t cumulative_perms,
1802 uint64_t cumulative_shared_perms,
1803 GSList *ignore_children,
1804 bool *tighten_restrictions, Error **errp)
1806 BlockDriver *drv = bs->drv;
1807 BdrvChild *c;
1808 int ret;
1810 assert(!q || !tighten_restrictions);
1812 if (tighten_restrictions) {
1813 uint64_t current_perms, current_shared;
1814 uint64_t added_perms, removed_shared_perms;
1816 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1818 added_perms = cumulative_perms & ~current_perms;
1819 removed_shared_perms = current_shared & ~cumulative_shared_perms;
1821 *tighten_restrictions = added_perms || removed_shared_perms;
1824 /* Write permissions never work with read-only images */
1825 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1826 !bdrv_is_writable_after_reopen(bs, q))
1828 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
1829 error_setg(errp, "Block node is read-only");
1830 } else {
1831 uint64_t current_perms, current_shared;
1832 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1833 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
1834 error_setg(errp, "Cannot make block node read-only, there is "
1835 "a writer on it");
1836 } else {
1837 error_setg(errp, "Cannot make block node read-only and create "
1838 "a writer on it");
1842 return -EPERM;
1845 /* Check this node */
1846 if (!drv) {
1847 return 0;
1850 if (drv->bdrv_check_perm) {
1851 return drv->bdrv_check_perm(bs, cumulative_perms,
1852 cumulative_shared_perms, errp);
1855 /* Drivers that never have children can omit .bdrv_child_perm() */
1856 if (!drv->bdrv_child_perm) {
1857 assert(QLIST_EMPTY(&bs->children));
1858 return 0;
1861 /* Check all children */
1862 QLIST_FOREACH(c, &bs->children, next) {
1863 uint64_t cur_perm, cur_shared;
1864 bool child_tighten_restr;
1866 bdrv_child_perm(bs, c->bs, c, c->role, q,
1867 cumulative_perms, cumulative_shared_perms,
1868 &cur_perm, &cur_shared);
1869 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
1870 tighten_restrictions ? &child_tighten_restr
1871 : NULL,
1872 errp);
1873 if (tighten_restrictions) {
1874 *tighten_restrictions |= child_tighten_restr;
1876 if (ret < 0) {
1877 return ret;
1881 return 0;
1885 * Notifies drivers that after a previous bdrv_check_perm() call, the
1886 * permission update is not performed and any preparations made for it (e.g.
1887 * taken file locks) need to be undone.
1889 * This function recursively notifies all child nodes.
1891 static void bdrv_abort_perm_update(BlockDriverState *bs)
1893 BlockDriver *drv = bs->drv;
1894 BdrvChild *c;
1896 if (!drv) {
1897 return;
1900 if (drv->bdrv_abort_perm_update) {
1901 drv->bdrv_abort_perm_update(bs);
1904 QLIST_FOREACH(c, &bs->children, next) {
1905 bdrv_child_abort_perm_update(c);
1909 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1910 uint64_t cumulative_shared_perms)
1912 BlockDriver *drv = bs->drv;
1913 BdrvChild *c;
1915 if (!drv) {
1916 return;
1919 /* Update this node */
1920 if (drv->bdrv_set_perm) {
1921 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1924 /* Drivers that never have children can omit .bdrv_child_perm() */
1925 if (!drv->bdrv_child_perm) {
1926 assert(QLIST_EMPTY(&bs->children));
1927 return;
1930 /* Update all children */
1931 QLIST_FOREACH(c, &bs->children, next) {
1932 uint64_t cur_perm, cur_shared;
1933 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1934 cumulative_perms, cumulative_shared_perms,
1935 &cur_perm, &cur_shared);
1936 bdrv_child_set_perm(c, cur_perm, cur_shared);
1940 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1941 uint64_t *shared_perm)
1943 BdrvChild *c;
1944 uint64_t cumulative_perms = 0;
1945 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1947 QLIST_FOREACH(c, &bs->parents, next_parent) {
1948 cumulative_perms |= c->perm;
1949 cumulative_shared_perms &= c->shared_perm;
1952 *perm = cumulative_perms;
1953 *shared_perm = cumulative_shared_perms;
1956 static char *bdrv_child_user_desc(BdrvChild *c)
1958 if (c->role->get_parent_desc) {
1959 return c->role->get_parent_desc(c);
1962 return g_strdup("another user");
1965 char *bdrv_perm_names(uint64_t perm)
1967 struct perm_name {
1968 uint64_t perm;
1969 const char *name;
1970 } permissions[] = {
1971 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1972 { BLK_PERM_WRITE, "write" },
1973 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1974 { BLK_PERM_RESIZE, "resize" },
1975 { BLK_PERM_GRAPH_MOD, "change children" },
1976 { 0, NULL }
1979 char *result = g_strdup("");
1980 struct perm_name *p;
1982 for (p = permissions; p->name; p++) {
1983 if (perm & p->perm) {
1984 char *old = result;
1985 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1986 g_free(old);
1990 return result;
1994 * Checks whether a new reference to @bs can be added if the new user requires
1995 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1996 * set, the BdrvChild objects in this list are ignored in the calculations;
1997 * this allows checking permission updates for an existing reference.
1999 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2001 * Needs to be followed by a call to either bdrv_set_perm() or
2002 * bdrv_abort_perm_update(). */
2003 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2004 uint64_t new_used_perm,
2005 uint64_t new_shared_perm,
2006 GSList *ignore_children,
2007 bool *tighten_restrictions,
2008 Error **errp)
2010 BdrvChild *c;
2011 uint64_t cumulative_perms = new_used_perm;
2012 uint64_t cumulative_shared_perms = new_shared_perm;
2014 assert(!q || !tighten_restrictions);
2016 /* There is no reason why anyone couldn't tolerate write_unchanged */
2017 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2019 QLIST_FOREACH(c, &bs->parents, next_parent) {
2020 if (g_slist_find(ignore_children, c)) {
2021 continue;
2024 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2025 char *user = bdrv_child_user_desc(c);
2026 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2028 if (tighten_restrictions) {
2029 *tighten_restrictions = true;
2032 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2033 "allow '%s' on %s",
2034 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2035 g_free(user);
2036 g_free(perm_names);
2037 return -EPERM;
2040 if ((c->perm & new_shared_perm) != c->perm) {
2041 char *user = bdrv_child_user_desc(c);
2042 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2044 if (tighten_restrictions) {
2045 *tighten_restrictions = true;
2048 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2049 "'%s' on %s",
2050 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2051 g_free(user);
2052 g_free(perm_names);
2053 return -EPERM;
2056 cumulative_perms |= c->perm;
2057 cumulative_shared_perms &= c->shared_perm;
2060 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2061 ignore_children, tighten_restrictions, errp);
2064 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2065 * bdrv_child_abort_perm_update(). */
2066 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2067 uint64_t perm, uint64_t shared,
2068 GSList *ignore_children,
2069 bool *tighten_restrictions, Error **errp)
2071 int ret;
2073 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2074 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2075 tighten_restrictions, errp);
2076 g_slist_free(ignore_children);
2078 if (ret < 0) {
2079 return ret;
2082 if (!c->has_backup_perm) {
2083 c->has_backup_perm = true;
2084 c->backup_perm = c->perm;
2085 c->backup_shared_perm = c->shared_perm;
2088 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2089 * same child twice during check_perm procedure
2092 c->perm = perm;
2093 c->shared_perm = shared;
2095 return 0;
2098 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2100 uint64_t cumulative_perms, cumulative_shared_perms;
2102 c->has_backup_perm = false;
2104 c->perm = perm;
2105 c->shared_perm = shared;
2107 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2108 &cumulative_shared_perms);
2109 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2112 static void bdrv_child_abort_perm_update(BdrvChild *c)
2114 if (c->has_backup_perm) {
2115 c->perm = c->backup_perm;
2116 c->shared_perm = c->backup_shared_perm;
2117 c->has_backup_perm = false;
2120 bdrv_abort_perm_update(c->bs);
2123 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2124 Error **errp)
2126 Error *local_err = NULL;
2127 int ret;
2128 bool tighten_restrictions;
2130 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2131 &tighten_restrictions, &local_err);
2132 if (ret < 0) {
2133 bdrv_child_abort_perm_update(c);
2134 if (tighten_restrictions) {
2135 error_propagate(errp, local_err);
2136 } else {
2138 * Our caller may intend to only loosen restrictions and
2139 * does not expect this function to fail. Errors are not
2140 * fatal in such a case, so we can just hide them from our
2141 * caller.
2143 error_free(local_err);
2144 ret = 0;
2146 return ret;
2149 bdrv_child_set_perm(c, perm, shared);
2151 return 0;
2154 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2156 uint64_t parent_perms, parent_shared;
2157 uint64_t perms, shared;
2159 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2160 bdrv_child_perm(bs, c->bs, c, c->role, NULL, parent_perms, parent_shared,
2161 &perms, &shared);
2163 return bdrv_child_try_set_perm(c, perms, shared, errp);
2166 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2167 const BdrvChildRole *role,
2168 BlockReopenQueue *reopen_queue,
2169 uint64_t perm, uint64_t shared,
2170 uint64_t *nperm, uint64_t *nshared)
2172 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2173 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2176 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2177 const BdrvChildRole *role,
2178 BlockReopenQueue *reopen_queue,
2179 uint64_t perm, uint64_t shared,
2180 uint64_t *nperm, uint64_t *nshared)
2182 bool backing = (role == &child_backing);
2183 assert(role == &child_backing || role == &child_file);
2185 if (!backing) {
2186 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2188 /* Apart from the modifications below, the same permissions are
2189 * forwarded and left alone as for filters */
2190 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2191 &perm, &shared);
2193 /* Format drivers may touch metadata even if the guest doesn't write */
2194 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2195 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2198 /* bs->file always needs to be consistent because of the metadata. We
2199 * can never allow other users to resize or write to it. */
2200 if (!(flags & BDRV_O_NO_IO)) {
2201 perm |= BLK_PERM_CONSISTENT_READ;
2203 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2204 } else {
2205 /* We want consistent read from backing files if the parent needs it.
2206 * No other operations are performed on backing files. */
2207 perm &= BLK_PERM_CONSISTENT_READ;
2209 /* If the parent can deal with changing data, we're okay with a
2210 * writable and resizable backing file. */
2211 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2212 if (shared & BLK_PERM_WRITE) {
2213 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2214 } else {
2215 shared = 0;
2218 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2219 BLK_PERM_WRITE_UNCHANGED;
2222 if (bs->open_flags & BDRV_O_INACTIVE) {
2223 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2226 *nperm = perm;
2227 *nshared = shared;
2230 static void bdrv_replace_child_noperm(BdrvChild *child,
2231 BlockDriverState *new_bs)
2233 BlockDriverState *old_bs = child->bs;
2234 int new_bs_quiesce_counter;
2235 int drain_saldo;
2237 assert(!child->frozen);
2239 if (old_bs && new_bs) {
2240 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2243 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2244 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2247 * If the new child node is drained but the old one was not, flush
2248 * all outstanding requests to the old child node.
2250 while (drain_saldo > 0 && child->role->drained_begin) {
2251 bdrv_parent_drained_begin_single(child, true);
2252 drain_saldo--;
2255 if (old_bs) {
2256 /* Detach first so that the recursive drain sections coming from @child
2257 * are already gone and we only end the drain sections that came from
2258 * elsewhere. */
2259 if (child->role->detach) {
2260 child->role->detach(child);
2262 QLIST_REMOVE(child, next_parent);
2265 child->bs = new_bs;
2267 if (new_bs) {
2268 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2271 * Detaching the old node may have led to the new node's
2272 * quiesce_counter having been decreased. Not a problem, we
2273 * just need to recognize this here and then invoke
2274 * drained_end appropriately more often.
2276 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2277 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2279 /* Attach only after starting new drained sections, so that recursive
2280 * drain sections coming from @child don't get an extra .drained_begin
2281 * callback. */
2282 if (child->role->attach) {
2283 child->role->attach(child);
2288 * If the old child node was drained but the new one is not, allow
2289 * requests to come in only after the new node has been attached.
2291 while (drain_saldo < 0 && child->role->drained_end) {
2292 bdrv_parent_drained_end_single(child);
2293 drain_saldo++;
2298 * Updates @child to change its reference to point to @new_bs, including
2299 * checking and applying the necessary permisson updates both to the old node
2300 * and to @new_bs.
2302 * NULL is passed as @new_bs for removing the reference before freeing @child.
2304 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2305 * function uses bdrv_set_perm() to update the permissions according to the new
2306 * reference that @new_bs gets.
2308 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2310 BlockDriverState *old_bs = child->bs;
2311 uint64_t perm, shared_perm;
2313 bdrv_replace_child_noperm(child, new_bs);
2316 * Start with the new node's permissions. If @new_bs is a (direct
2317 * or indirect) child of @old_bs, we must complete the permission
2318 * update on @new_bs before we loosen the restrictions on @old_bs.
2319 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2320 * updating the permissions of @new_bs, and thus not purely loosen
2321 * restrictions.
2323 if (new_bs) {
2324 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2325 bdrv_set_perm(new_bs, perm, shared_perm);
2328 if (old_bs) {
2329 /* Update permissions for old node. This is guaranteed to succeed
2330 * because we're just taking a parent away, so we're loosening
2331 * restrictions. */
2332 bool tighten_restrictions;
2333 int ret;
2335 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2336 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2337 &tighten_restrictions, NULL);
2338 assert(tighten_restrictions == false);
2339 if (ret < 0) {
2340 /* We only tried to loosen restrictions, so errors are not fatal */
2341 bdrv_abort_perm_update(old_bs);
2342 } else {
2343 bdrv_set_perm(old_bs, perm, shared_perm);
2346 /* When the parent requiring a non-default AioContext is removed, the
2347 * node moves back to the main AioContext */
2348 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2353 * This function steals the reference to child_bs from the caller.
2354 * That reference is later dropped by bdrv_root_unref_child().
2356 * On failure NULL is returned, errp is set and the reference to
2357 * child_bs is also dropped.
2359 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2360 * (unless @child_bs is already in @ctx).
2362 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2363 const char *child_name,
2364 const BdrvChildRole *child_role,
2365 AioContext *ctx,
2366 uint64_t perm, uint64_t shared_perm,
2367 void *opaque, Error **errp)
2369 BdrvChild *child;
2370 Error *local_err = NULL;
2371 int ret;
2373 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2374 errp);
2375 if (ret < 0) {
2376 bdrv_abort_perm_update(child_bs);
2377 bdrv_unref(child_bs);
2378 return NULL;
2381 child = g_new(BdrvChild, 1);
2382 *child = (BdrvChild) {
2383 .bs = NULL,
2384 .name = g_strdup(child_name),
2385 .role = child_role,
2386 .perm = perm,
2387 .shared_perm = shared_perm,
2388 .opaque = opaque,
2391 /* If the AioContexts don't match, first try to move the subtree of
2392 * child_bs into the AioContext of the new parent. If this doesn't work,
2393 * try moving the parent into the AioContext of child_bs instead. */
2394 if (bdrv_get_aio_context(child_bs) != ctx) {
2395 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2396 if (ret < 0 && child_role->can_set_aio_ctx) {
2397 GSList *ignore = g_slist_prepend(NULL, child);;
2398 ctx = bdrv_get_aio_context(child_bs);
2399 if (child_role->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2400 error_free(local_err);
2401 ret = 0;
2402 g_slist_free(ignore);
2403 ignore = g_slist_prepend(NULL, child);;
2404 child_role->set_aio_ctx(child, ctx, &ignore);
2406 g_slist_free(ignore);
2408 if (ret < 0) {
2409 error_propagate(errp, local_err);
2410 g_free(child);
2411 bdrv_abort_perm_update(child_bs);
2412 return NULL;
2416 /* This performs the matching bdrv_set_perm() for the above check. */
2417 bdrv_replace_child(child, child_bs);
2419 return child;
2423 * This function transfers the reference to child_bs from the caller
2424 * to parent_bs. That reference is later dropped by parent_bs on
2425 * bdrv_close() or if someone calls bdrv_unref_child().
2427 * On failure NULL is returned, errp is set and the reference to
2428 * child_bs is also dropped.
2430 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2431 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2433 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2434 BlockDriverState *child_bs,
2435 const char *child_name,
2436 const BdrvChildRole *child_role,
2437 Error **errp)
2439 BdrvChild *child;
2440 uint64_t perm, shared_perm;
2442 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2444 assert(parent_bs->drv);
2445 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2446 perm, shared_perm, &perm, &shared_perm);
2448 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2449 bdrv_get_aio_context(parent_bs),
2450 perm, shared_perm, parent_bs, errp);
2451 if (child == NULL) {
2452 return NULL;
2455 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2456 return child;
2459 static void bdrv_detach_child(BdrvChild *child)
2461 if (child->next.le_prev) {
2462 QLIST_REMOVE(child, next);
2463 child->next.le_prev = NULL;
2466 bdrv_replace_child(child, NULL);
2468 g_free(child->name);
2469 g_free(child);
2472 void bdrv_root_unref_child(BdrvChild *child)
2474 BlockDriverState *child_bs;
2476 child_bs = child->bs;
2477 bdrv_detach_child(child);
2478 bdrv_unref(child_bs);
2482 * Clear all inherits_from pointers from children and grandchildren of
2483 * @root that point to @root, where necessary.
2485 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2487 BdrvChild *c;
2489 if (child->bs->inherits_from == root) {
2491 * Remove inherits_from only when the last reference between root and
2492 * child->bs goes away.
2494 QLIST_FOREACH(c, &root->children, next) {
2495 if (c != child && c->bs == child->bs) {
2496 break;
2499 if (c == NULL) {
2500 child->bs->inherits_from = NULL;
2504 QLIST_FOREACH(c, &child->bs->children, next) {
2505 bdrv_unset_inherits_from(root, c);
2509 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2511 if (child == NULL) {
2512 return;
2515 bdrv_unset_inherits_from(parent, child);
2516 bdrv_root_unref_child(child);
2520 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2522 BdrvChild *c;
2523 QLIST_FOREACH(c, &bs->parents, next_parent) {
2524 if (c->role->change_media) {
2525 c->role->change_media(c, load);
2530 /* Return true if you can reach parent going through child->inherits_from
2531 * recursively. If parent or child are NULL, return false */
2532 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2533 BlockDriverState *parent)
2535 while (child && child != parent) {
2536 child = child->inherits_from;
2539 return child != NULL;
2543 * Sets the backing file link of a BDS. A new reference is created; callers
2544 * which don't need their own reference any more must call bdrv_unref().
2546 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2547 Error **errp)
2549 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2550 bdrv_inherits_from_recursive(backing_hd, bs);
2552 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2553 return;
2556 if (backing_hd) {
2557 bdrv_ref(backing_hd);
2560 if (bs->backing) {
2561 bdrv_unref_child(bs, bs->backing);
2564 if (!backing_hd) {
2565 bs->backing = NULL;
2566 goto out;
2569 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2570 errp);
2571 /* If backing_hd was already part of bs's backing chain, and
2572 * inherits_from pointed recursively to bs then let's update it to
2573 * point directly to bs (else it will become NULL). */
2574 if (bs->backing && update_inherits_from) {
2575 backing_hd->inherits_from = bs;
2578 out:
2579 bdrv_refresh_limits(bs, NULL);
2583 * Opens the backing file for a BlockDriverState if not yet open
2585 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2586 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2587 * itself, all options starting with "${bdref_key}." are considered part of the
2588 * BlockdevRef.
2590 * TODO Can this be unified with bdrv_open_image()?
2592 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2593 const char *bdref_key, Error **errp)
2595 char *backing_filename = NULL;
2596 char *bdref_key_dot;
2597 const char *reference = NULL;
2598 int ret = 0;
2599 bool implicit_backing = false;
2600 BlockDriverState *backing_hd;
2601 QDict *options;
2602 QDict *tmp_parent_options = NULL;
2603 Error *local_err = NULL;
2605 if (bs->backing != NULL) {
2606 goto free_exit;
2609 /* NULL means an empty set of options */
2610 if (parent_options == NULL) {
2611 tmp_parent_options = qdict_new();
2612 parent_options = tmp_parent_options;
2615 bs->open_flags &= ~BDRV_O_NO_BACKING;
2617 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2618 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2619 g_free(bdref_key_dot);
2622 * Caution: while qdict_get_try_str() is fine, getting non-string
2623 * types would require more care. When @parent_options come from
2624 * -blockdev or blockdev_add, its members are typed according to
2625 * the QAPI schema, but when they come from -drive, they're all
2626 * QString.
2628 reference = qdict_get_try_str(parent_options, bdref_key);
2629 if (reference || qdict_haskey(options, "file.filename")) {
2630 /* keep backing_filename NULL */
2631 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2632 qobject_unref(options);
2633 goto free_exit;
2634 } else {
2635 if (qdict_size(options) == 0) {
2636 /* If the user specifies options that do not modify the
2637 * backing file's behavior, we might still consider it the
2638 * implicit backing file. But it's easier this way, and
2639 * just specifying some of the backing BDS's options is
2640 * only possible with -drive anyway (otherwise the QAPI
2641 * schema forces the user to specify everything). */
2642 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2645 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2646 if (local_err) {
2647 ret = -EINVAL;
2648 error_propagate(errp, local_err);
2649 qobject_unref(options);
2650 goto free_exit;
2654 if (!bs->drv || !bs->drv->supports_backing) {
2655 ret = -EINVAL;
2656 error_setg(errp, "Driver doesn't support backing files");
2657 qobject_unref(options);
2658 goto free_exit;
2661 if (!reference &&
2662 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2663 qdict_put_str(options, "driver", bs->backing_format);
2666 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2667 &child_backing, errp);
2668 if (!backing_hd) {
2669 bs->open_flags |= BDRV_O_NO_BACKING;
2670 error_prepend(errp, "Could not open backing file: ");
2671 ret = -EINVAL;
2672 goto free_exit;
2675 if (implicit_backing) {
2676 bdrv_refresh_filename(backing_hd);
2677 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2678 backing_hd->filename);
2681 /* Hook up the backing file link; drop our reference, bs owns the
2682 * backing_hd reference now */
2683 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2684 bdrv_unref(backing_hd);
2685 if (local_err) {
2686 error_propagate(errp, local_err);
2687 ret = -EINVAL;
2688 goto free_exit;
2691 qdict_del(parent_options, bdref_key);
2693 free_exit:
2694 g_free(backing_filename);
2695 qobject_unref(tmp_parent_options);
2696 return ret;
2699 static BlockDriverState *
2700 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2701 BlockDriverState *parent, const BdrvChildRole *child_role,
2702 bool allow_none, Error **errp)
2704 BlockDriverState *bs = NULL;
2705 QDict *image_options;
2706 char *bdref_key_dot;
2707 const char *reference;
2709 assert(child_role != NULL);
2711 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2712 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2713 g_free(bdref_key_dot);
2716 * Caution: while qdict_get_try_str() is fine, getting non-string
2717 * types would require more care. When @options come from
2718 * -blockdev or blockdev_add, its members are typed according to
2719 * the QAPI schema, but when they come from -drive, they're all
2720 * QString.
2722 reference = qdict_get_try_str(options, bdref_key);
2723 if (!filename && !reference && !qdict_size(image_options)) {
2724 if (!allow_none) {
2725 error_setg(errp, "A block device must be specified for \"%s\"",
2726 bdref_key);
2728 qobject_unref(image_options);
2729 goto done;
2732 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2733 parent, child_role, errp);
2734 if (!bs) {
2735 goto done;
2738 done:
2739 qdict_del(options, bdref_key);
2740 return bs;
2744 * Opens a disk image whose options are given as BlockdevRef in another block
2745 * device's options.
2747 * If allow_none is true, no image will be opened if filename is false and no
2748 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2750 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2751 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2752 * itself, all options starting with "${bdref_key}." are considered part of the
2753 * BlockdevRef.
2755 * The BlockdevRef will be removed from the options QDict.
2757 BdrvChild *bdrv_open_child(const char *filename,
2758 QDict *options, const char *bdref_key,
2759 BlockDriverState *parent,
2760 const BdrvChildRole *child_role,
2761 bool allow_none, Error **errp)
2763 BlockDriverState *bs;
2765 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2766 allow_none, errp);
2767 if (bs == NULL) {
2768 return NULL;
2771 return bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2774 /* TODO Future callers may need to specify parent/child_role in order for
2775 * option inheritance to work. Existing callers use it for the root node. */
2776 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2778 BlockDriverState *bs = NULL;
2779 Error *local_err = NULL;
2780 QObject *obj = NULL;
2781 QDict *qdict = NULL;
2782 const char *reference = NULL;
2783 Visitor *v = NULL;
2785 if (ref->type == QTYPE_QSTRING) {
2786 reference = ref->u.reference;
2787 } else {
2788 BlockdevOptions *options = &ref->u.definition;
2789 assert(ref->type == QTYPE_QDICT);
2791 v = qobject_output_visitor_new(&obj);
2792 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2793 if (local_err) {
2794 error_propagate(errp, local_err);
2795 goto fail;
2797 visit_complete(v, &obj);
2799 qdict = qobject_to(QDict, obj);
2800 qdict_flatten(qdict);
2802 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2803 * compatibility with other callers) rather than what we want as the
2804 * real defaults. Apply the defaults here instead. */
2805 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2806 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2807 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2808 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2812 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2813 obj = NULL;
2815 fail:
2816 qobject_unref(obj);
2817 visit_free(v);
2818 return bs;
2821 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2822 int flags,
2823 QDict *snapshot_options,
2824 Error **errp)
2826 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2827 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2828 int64_t total_size;
2829 QemuOpts *opts = NULL;
2830 BlockDriverState *bs_snapshot = NULL;
2831 Error *local_err = NULL;
2832 int ret;
2834 /* if snapshot, we create a temporary backing file and open it
2835 instead of opening 'filename' directly */
2837 /* Get the required size from the image */
2838 total_size = bdrv_getlength(bs);
2839 if (total_size < 0) {
2840 error_setg_errno(errp, -total_size, "Could not get image size");
2841 goto out;
2844 /* Create the temporary image */
2845 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2846 if (ret < 0) {
2847 error_setg_errno(errp, -ret, "Could not get temporary filename");
2848 goto out;
2851 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2852 &error_abort);
2853 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2854 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2855 qemu_opts_del(opts);
2856 if (ret < 0) {
2857 error_prepend(errp, "Could not create temporary overlay '%s': ",
2858 tmp_filename);
2859 goto out;
2862 /* Prepare options QDict for the temporary file */
2863 qdict_put_str(snapshot_options, "file.driver", "file");
2864 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2865 qdict_put_str(snapshot_options, "driver", "qcow2");
2867 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2868 snapshot_options = NULL;
2869 if (!bs_snapshot) {
2870 goto out;
2873 /* bdrv_append() consumes a strong reference to bs_snapshot
2874 * (i.e. it will call bdrv_unref() on it) even on error, so in
2875 * order to be able to return one, we have to increase
2876 * bs_snapshot's refcount here */
2877 bdrv_ref(bs_snapshot);
2878 bdrv_append(bs_snapshot, bs, &local_err);
2879 if (local_err) {
2880 error_propagate(errp, local_err);
2881 bs_snapshot = NULL;
2882 goto out;
2885 out:
2886 qobject_unref(snapshot_options);
2887 g_free(tmp_filename);
2888 return bs_snapshot;
2892 * Opens a disk image (raw, qcow2, vmdk, ...)
2894 * options is a QDict of options to pass to the block drivers, or NULL for an
2895 * empty set of options. The reference to the QDict belongs to the block layer
2896 * after the call (even on failure), so if the caller intends to reuse the
2897 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2899 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2900 * If it is not NULL, the referenced BDS will be reused.
2902 * The reference parameter may be used to specify an existing block device which
2903 * should be opened. If specified, neither options nor a filename may be given,
2904 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2906 static BlockDriverState *bdrv_open_inherit(const char *filename,
2907 const char *reference,
2908 QDict *options, int flags,
2909 BlockDriverState *parent,
2910 const BdrvChildRole *child_role,
2911 Error **errp)
2913 int ret;
2914 BlockBackend *file = NULL;
2915 BlockDriverState *bs;
2916 BlockDriver *drv = NULL;
2917 BdrvChild *child;
2918 const char *drvname;
2919 const char *backing;
2920 Error *local_err = NULL;
2921 QDict *snapshot_options = NULL;
2922 int snapshot_flags = 0;
2924 assert(!child_role || !flags);
2925 assert(!child_role == !parent);
2927 if (reference) {
2928 bool options_non_empty = options ? qdict_size(options) : false;
2929 qobject_unref(options);
2931 if (filename || options_non_empty) {
2932 error_setg(errp, "Cannot reference an existing block device with "
2933 "additional options or a new filename");
2934 return NULL;
2937 bs = bdrv_lookup_bs(reference, reference, errp);
2938 if (!bs) {
2939 return NULL;
2942 bdrv_ref(bs);
2943 return bs;
2946 bs = bdrv_new();
2948 /* NULL means an empty set of options */
2949 if (options == NULL) {
2950 options = qdict_new();
2953 /* json: syntax counts as explicit options, as if in the QDict */
2954 parse_json_protocol(options, &filename, &local_err);
2955 if (local_err) {
2956 goto fail;
2959 bs->explicit_options = qdict_clone_shallow(options);
2961 if (child_role) {
2962 bs->inherits_from = parent;
2963 child_role->inherit_options(&flags, options,
2964 parent->open_flags, parent->options);
2967 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2968 if (local_err) {
2969 goto fail;
2973 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2974 * Caution: getting a boolean member of @options requires care.
2975 * When @options come from -blockdev or blockdev_add, members are
2976 * typed according to the QAPI schema, but when they come from
2977 * -drive, they're all QString.
2979 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2980 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2981 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2982 } else {
2983 flags &= ~BDRV_O_RDWR;
2986 if (flags & BDRV_O_SNAPSHOT) {
2987 snapshot_options = qdict_new();
2988 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2989 flags, options);
2990 /* Let bdrv_backing_options() override "read-only" */
2991 qdict_del(options, BDRV_OPT_READ_ONLY);
2992 bdrv_backing_options(&flags, options, flags, options);
2995 bs->open_flags = flags;
2996 bs->options = options;
2997 options = qdict_clone_shallow(options);
2999 /* Find the right image format driver */
3000 /* See cautionary note on accessing @options above */
3001 drvname = qdict_get_try_str(options, "driver");
3002 if (drvname) {
3003 drv = bdrv_find_format(drvname);
3004 if (!drv) {
3005 error_setg(errp, "Unknown driver: '%s'", drvname);
3006 goto fail;
3010 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3012 /* See cautionary note on accessing @options above */
3013 backing = qdict_get_try_str(options, "backing");
3014 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3015 (backing && *backing == '\0'))
3017 if (backing) {
3018 warn_report("Use of \"backing\": \"\" is deprecated; "
3019 "use \"backing\": null instead");
3021 flags |= BDRV_O_NO_BACKING;
3022 qdict_del(options, "backing");
3025 /* Open image file without format layer. This BlockBackend is only used for
3026 * probing, the block drivers will do their own bdrv_open_child() for the
3027 * same BDS, which is why we put the node name back into options. */
3028 if ((flags & BDRV_O_PROTOCOL) == 0) {
3029 BlockDriverState *file_bs;
3031 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3032 &child_file, true, &local_err);
3033 if (local_err) {
3034 goto fail;
3036 if (file_bs != NULL) {
3037 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3038 * looking at the header to guess the image format. This works even
3039 * in cases where a guest would not see a consistent state. */
3040 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3041 blk_insert_bs(file, file_bs, &local_err);
3042 bdrv_unref(file_bs);
3043 if (local_err) {
3044 goto fail;
3047 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3051 /* Image format probing */
3052 bs->probed = !drv;
3053 if (!drv && file) {
3054 ret = find_image_format(file, filename, &drv, &local_err);
3055 if (ret < 0) {
3056 goto fail;
3059 * This option update would logically belong in bdrv_fill_options(),
3060 * but we first need to open bs->file for the probing to work, while
3061 * opening bs->file already requires the (mostly) final set of options
3062 * so that cache mode etc. can be inherited.
3064 * Adding the driver later is somewhat ugly, but it's not an option
3065 * that would ever be inherited, so it's correct. We just need to make
3066 * sure to update both bs->options (which has the full effective
3067 * options for bs) and options (which has file.* already removed).
3069 qdict_put_str(bs->options, "driver", drv->format_name);
3070 qdict_put_str(options, "driver", drv->format_name);
3071 } else if (!drv) {
3072 error_setg(errp, "Must specify either driver or file");
3073 goto fail;
3076 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3077 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3078 /* file must be NULL if a protocol BDS is about to be created
3079 * (the inverse results in an error message from bdrv_open_common()) */
3080 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3082 /* Open the image */
3083 ret = bdrv_open_common(bs, file, options, &local_err);
3084 if (ret < 0) {
3085 goto fail;
3088 if (file) {
3089 blk_unref(file);
3090 file = NULL;
3093 /* If there is a backing file, use it */
3094 if ((flags & BDRV_O_NO_BACKING) == 0) {
3095 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3096 if (ret < 0) {
3097 goto close_and_fail;
3101 /* Remove all children options and references
3102 * from bs->options and bs->explicit_options */
3103 QLIST_FOREACH(child, &bs->children, next) {
3104 char *child_key_dot;
3105 child_key_dot = g_strdup_printf("%s.", child->name);
3106 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3107 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3108 qdict_del(bs->explicit_options, child->name);
3109 qdict_del(bs->options, child->name);
3110 g_free(child_key_dot);
3113 /* Check if any unknown options were used */
3114 if (qdict_size(options) != 0) {
3115 const QDictEntry *entry = qdict_first(options);
3116 if (flags & BDRV_O_PROTOCOL) {
3117 error_setg(errp, "Block protocol '%s' doesn't support the option "
3118 "'%s'", drv->format_name, entry->key);
3119 } else {
3120 error_setg(errp,
3121 "Block format '%s' does not support the option '%s'",
3122 drv->format_name, entry->key);
3125 goto close_and_fail;
3128 bdrv_parent_cb_change_media(bs, true);
3130 qobject_unref(options);
3131 options = NULL;
3133 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3134 * temporary snapshot afterwards. */
3135 if (snapshot_flags) {
3136 BlockDriverState *snapshot_bs;
3137 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3138 snapshot_options, &local_err);
3139 snapshot_options = NULL;
3140 if (local_err) {
3141 goto close_and_fail;
3143 /* We are not going to return bs but the overlay on top of it
3144 * (snapshot_bs); thus, we have to drop the strong reference to bs
3145 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3146 * though, because the overlay still has a reference to it. */
3147 bdrv_unref(bs);
3148 bs = snapshot_bs;
3151 return bs;
3153 fail:
3154 blk_unref(file);
3155 qobject_unref(snapshot_options);
3156 qobject_unref(bs->explicit_options);
3157 qobject_unref(bs->options);
3158 qobject_unref(options);
3159 bs->options = NULL;
3160 bs->explicit_options = NULL;
3161 bdrv_unref(bs);
3162 error_propagate(errp, local_err);
3163 return NULL;
3165 close_and_fail:
3166 bdrv_unref(bs);
3167 qobject_unref(snapshot_options);
3168 qobject_unref(options);
3169 error_propagate(errp, local_err);
3170 return NULL;
3173 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3174 QDict *options, int flags, Error **errp)
3176 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3177 NULL, errp);
3180 /* Return true if the NULL-terminated @list contains @str */
3181 static bool is_str_in_list(const char *str, const char *const *list)
3183 if (str && list) {
3184 int i;
3185 for (i = 0; list[i] != NULL; i++) {
3186 if (!strcmp(str, list[i])) {
3187 return true;
3191 return false;
3195 * Check that every option set in @bs->options is also set in
3196 * @new_opts.
3198 * Options listed in the common_options list and in
3199 * @bs->drv->mutable_opts are skipped.
3201 * Return 0 on success, otherwise return -EINVAL and set @errp.
3203 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3204 const QDict *new_opts, Error **errp)
3206 const QDictEntry *e;
3207 /* These options are common to all block drivers and are handled
3208 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3209 const char *const common_options[] = {
3210 "node-name", "discard", "cache.direct", "cache.no-flush",
3211 "read-only", "auto-read-only", "detect-zeroes", NULL
3214 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3215 if (!qdict_haskey(new_opts, e->key) &&
3216 !is_str_in_list(e->key, common_options) &&
3217 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3218 error_setg(errp, "Option '%s' cannot be reset "
3219 "to its default value", e->key);
3220 return -EINVAL;
3224 return 0;
3228 * Returns true if @child can be reached recursively from @bs
3230 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3231 BlockDriverState *child)
3233 BdrvChild *c;
3235 if (bs == child) {
3236 return true;
3239 QLIST_FOREACH(c, &bs->children, next) {
3240 if (bdrv_recurse_has_child(c->bs, child)) {
3241 return true;
3245 return false;
3249 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3250 * reopen of multiple devices.
3252 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
3253 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3254 * be created and initialized. This newly created BlockReopenQueue should be
3255 * passed back in for subsequent calls that are intended to be of the same
3256 * atomic 'set'.
3258 * bs is the BlockDriverState to add to the reopen queue.
3260 * options contains the changed options for the associated bs
3261 * (the BlockReopenQueue takes ownership)
3263 * flags contains the open flags for the associated bs
3265 * returns a pointer to bs_queue, which is either the newly allocated
3266 * bs_queue, or the existing bs_queue being used.
3268 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3270 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3271 BlockDriverState *bs,
3272 QDict *options,
3273 const BdrvChildRole *role,
3274 QDict *parent_options,
3275 int parent_flags,
3276 bool keep_old_opts)
3278 assert(bs != NULL);
3280 BlockReopenQueueEntry *bs_entry;
3281 BdrvChild *child;
3282 QDict *old_options, *explicit_options, *options_copy;
3283 int flags;
3284 QemuOpts *opts;
3286 /* Make sure that the caller remembered to use a drained section. This is
3287 * important to avoid graph changes between the recursive queuing here and
3288 * bdrv_reopen_multiple(). */
3289 assert(bs->quiesce_counter > 0);
3291 if (bs_queue == NULL) {
3292 bs_queue = g_new0(BlockReopenQueue, 1);
3293 QSIMPLEQ_INIT(bs_queue);
3296 if (!options) {
3297 options = qdict_new();
3300 /* Check if this BlockDriverState is already in the queue */
3301 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3302 if (bs == bs_entry->state.bs) {
3303 break;
3308 * Precedence of options:
3309 * 1. Explicitly passed in options (highest)
3310 * 2. Retained from explicitly set options of bs
3311 * 3. Inherited from parent node
3312 * 4. Retained from effective options of bs
3315 /* Old explicitly set values (don't overwrite by inherited value) */
3316 if (bs_entry || keep_old_opts) {
3317 old_options = qdict_clone_shallow(bs_entry ?
3318 bs_entry->state.explicit_options :
3319 bs->explicit_options);
3320 bdrv_join_options(bs, options, old_options);
3321 qobject_unref(old_options);
3324 explicit_options = qdict_clone_shallow(options);
3326 /* Inherit from parent node */
3327 if (parent_options) {
3328 flags = 0;
3329 role->inherit_options(&flags, options, parent_flags, parent_options);
3330 } else {
3331 flags = bdrv_get_flags(bs);
3334 if (keep_old_opts) {
3335 /* Old values are used for options that aren't set yet */
3336 old_options = qdict_clone_shallow(bs->options);
3337 bdrv_join_options(bs, options, old_options);
3338 qobject_unref(old_options);
3341 /* We have the final set of options so let's update the flags */
3342 options_copy = qdict_clone_shallow(options);
3343 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3344 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3345 update_flags_from_options(&flags, opts);
3346 qemu_opts_del(opts);
3347 qobject_unref(options_copy);
3349 /* bdrv_open_inherit() sets and clears some additional flags internally */
3350 flags &= ~BDRV_O_PROTOCOL;
3351 if (flags & BDRV_O_RDWR) {
3352 flags |= BDRV_O_ALLOW_RDWR;
3355 if (!bs_entry) {
3356 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3357 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3358 } else {
3359 qobject_unref(bs_entry->state.options);
3360 qobject_unref(bs_entry->state.explicit_options);
3363 bs_entry->state.bs = bs;
3364 bs_entry->state.options = options;
3365 bs_entry->state.explicit_options = explicit_options;
3366 bs_entry->state.flags = flags;
3368 /* This needs to be overwritten in bdrv_reopen_prepare() */
3369 bs_entry->state.perm = UINT64_MAX;
3370 bs_entry->state.shared_perm = 0;
3373 * If keep_old_opts is false then it means that unspecified
3374 * options must be reset to their original value. We don't allow
3375 * resetting 'backing' but we need to know if the option is
3376 * missing in order to decide if we have to return an error.
3378 if (!keep_old_opts) {
3379 bs_entry->state.backing_missing =
3380 !qdict_haskey(options, "backing") &&
3381 !qdict_haskey(options, "backing.driver");
3384 QLIST_FOREACH(child, &bs->children, next) {
3385 QDict *new_child_options = NULL;
3386 bool child_keep_old = keep_old_opts;
3388 /* reopen can only change the options of block devices that were
3389 * implicitly created and inherited options. For other (referenced)
3390 * block devices, a syntax like "backing.foo" results in an error. */
3391 if (child->bs->inherits_from != bs) {
3392 continue;
3395 /* Check if the options contain a child reference */
3396 if (qdict_haskey(options, child->name)) {
3397 const char *childref = qdict_get_try_str(options, child->name);
3399 * The current child must not be reopened if the child
3400 * reference is null or points to a different node.
3402 if (g_strcmp0(childref, child->bs->node_name)) {
3403 continue;
3406 * If the child reference points to the current child then
3407 * reopen it with its existing set of options (note that
3408 * it can still inherit new options from the parent).
3410 child_keep_old = true;
3411 } else {
3412 /* Extract child options ("child-name.*") */
3413 char *child_key_dot = g_strdup_printf("%s.", child->name);
3414 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3415 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3416 g_free(child_key_dot);
3419 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3420 child->role, options, flags, child_keep_old);
3423 return bs_queue;
3426 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3427 BlockDriverState *bs,
3428 QDict *options, bool keep_old_opts)
3430 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0,
3431 keep_old_opts);
3435 * Reopen multiple BlockDriverStates atomically & transactionally.
3437 * The queue passed in (bs_queue) must have been built up previous
3438 * via bdrv_reopen_queue().
3440 * Reopens all BDS specified in the queue, with the appropriate
3441 * flags. All devices are prepared for reopen, and failure of any
3442 * device will cause all device changes to be abandoned, and intermediate
3443 * data cleaned up.
3445 * If all devices prepare successfully, then the changes are committed
3446 * to all devices.
3448 * All affected nodes must be drained between bdrv_reopen_queue() and
3449 * bdrv_reopen_multiple().
3451 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
3453 int ret = -1;
3454 BlockReopenQueueEntry *bs_entry, *next;
3456 assert(bs_queue != NULL);
3458 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3459 assert(bs_entry->state.bs->quiesce_counter > 0);
3460 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
3461 goto cleanup;
3463 bs_entry->prepared = true;
3466 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3467 BDRVReopenState *state = &bs_entry->state;
3468 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
3469 state->shared_perm, NULL, NULL, errp);
3470 if (ret < 0) {
3471 goto cleanup_perm;
3473 /* Check if new_backing_bs would accept the new permissions */
3474 if (state->replace_backing_bs && state->new_backing_bs) {
3475 uint64_t nperm, nshared;
3476 bdrv_child_perm(state->bs, state->new_backing_bs,
3477 NULL, &child_backing, bs_queue,
3478 state->perm, state->shared_perm,
3479 &nperm, &nshared);
3480 ret = bdrv_check_update_perm(state->new_backing_bs, NULL,
3481 nperm, nshared, NULL, NULL, errp);
3482 if (ret < 0) {
3483 goto cleanup_perm;
3486 bs_entry->perms_checked = true;
3489 /* If we reach this point, we have success and just need to apply the
3490 * changes
3492 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3493 bdrv_reopen_commit(&bs_entry->state);
3496 ret = 0;
3497 cleanup_perm:
3498 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3499 BDRVReopenState *state = &bs_entry->state;
3501 if (!bs_entry->perms_checked) {
3502 continue;
3505 if (ret == 0) {
3506 bdrv_set_perm(state->bs, state->perm, state->shared_perm);
3507 } else {
3508 bdrv_abort_perm_update(state->bs);
3509 if (state->replace_backing_bs && state->new_backing_bs) {
3510 bdrv_abort_perm_update(state->new_backing_bs);
3514 cleanup:
3515 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3516 if (ret) {
3517 if (bs_entry->prepared) {
3518 bdrv_reopen_abort(&bs_entry->state);
3520 qobject_unref(bs_entry->state.explicit_options);
3521 qobject_unref(bs_entry->state.options);
3523 if (bs_entry->state.new_backing_bs) {
3524 bdrv_unref(bs_entry->state.new_backing_bs);
3526 g_free(bs_entry);
3528 g_free(bs_queue);
3530 return ret;
3533 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3534 Error **errp)
3536 int ret;
3537 BlockReopenQueue *queue;
3538 QDict *opts = qdict_new();
3540 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3542 bdrv_subtree_drained_begin(bs);
3543 queue = bdrv_reopen_queue(NULL, bs, opts, true);
3544 ret = bdrv_reopen_multiple(queue, errp);
3545 bdrv_subtree_drained_end(bs);
3547 return ret;
3550 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3551 BdrvChild *c)
3553 BlockReopenQueueEntry *entry;
3555 QSIMPLEQ_FOREACH(entry, q, entry) {
3556 BlockDriverState *bs = entry->state.bs;
3557 BdrvChild *child;
3559 QLIST_FOREACH(child, &bs->children, next) {
3560 if (child == c) {
3561 return entry;
3566 return NULL;
3569 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3570 uint64_t *perm, uint64_t *shared)
3572 BdrvChild *c;
3573 BlockReopenQueueEntry *parent;
3574 uint64_t cumulative_perms = 0;
3575 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3577 QLIST_FOREACH(c, &bs->parents, next_parent) {
3578 parent = find_parent_in_reopen_queue(q, c);
3579 if (!parent) {
3580 cumulative_perms |= c->perm;
3581 cumulative_shared_perms &= c->shared_perm;
3582 } else {
3583 uint64_t nperm, nshared;
3585 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3586 parent->state.perm, parent->state.shared_perm,
3587 &nperm, &nshared);
3589 cumulative_perms |= nperm;
3590 cumulative_shared_perms &= nshared;
3593 *perm = cumulative_perms;
3594 *shared = cumulative_shared_perms;
3598 * Take a BDRVReopenState and check if the value of 'backing' in the
3599 * reopen_state->options QDict is valid or not.
3601 * If 'backing' is missing from the QDict then return 0.
3603 * If 'backing' contains the node name of the backing file of
3604 * reopen_state->bs then return 0.
3606 * If 'backing' contains a different node name (or is null) then check
3607 * whether the current backing file can be replaced with the new one.
3608 * If that's the case then reopen_state->replace_backing_bs is set to
3609 * true and reopen_state->new_backing_bs contains a pointer to the new
3610 * backing BlockDriverState (or NULL).
3612 * Return 0 on success, otherwise return < 0 and set @errp.
3614 static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
3615 Error **errp)
3617 BlockDriverState *bs = reopen_state->bs;
3618 BlockDriverState *overlay_bs, *new_backing_bs;
3619 QObject *value;
3620 const char *str;
3622 value = qdict_get(reopen_state->options, "backing");
3623 if (value == NULL) {
3624 return 0;
3627 switch (qobject_type(value)) {
3628 case QTYPE_QNULL:
3629 new_backing_bs = NULL;
3630 break;
3631 case QTYPE_QSTRING:
3632 str = qobject_get_try_str(value);
3633 new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
3634 if (new_backing_bs == NULL) {
3635 return -EINVAL;
3636 } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
3637 error_setg(errp, "Making '%s' a backing file of '%s' "
3638 "would create a cycle", str, bs->node_name);
3639 return -EINVAL;
3641 break;
3642 default:
3643 /* 'backing' does not allow any other data type */
3644 g_assert_not_reached();
3648 * TODO: before removing the x- prefix from x-blockdev-reopen we
3649 * should move the new backing file into the right AioContext
3650 * instead of returning an error.
3652 if (new_backing_bs) {
3653 if (bdrv_get_aio_context(new_backing_bs) != bdrv_get_aio_context(bs)) {
3654 error_setg(errp, "Cannot use a new backing file "
3655 "with a different AioContext");
3656 return -EINVAL;
3661 * Find the "actual" backing file by skipping all links that point
3662 * to an implicit node, if any (e.g. a commit filter node).
3664 overlay_bs = bs;
3665 while (backing_bs(overlay_bs) && backing_bs(overlay_bs)->implicit) {
3666 overlay_bs = backing_bs(overlay_bs);
3669 /* If we want to replace the backing file we need some extra checks */
3670 if (new_backing_bs != backing_bs(overlay_bs)) {
3671 /* Check for implicit nodes between bs and its backing file */
3672 if (bs != overlay_bs) {
3673 error_setg(errp, "Cannot change backing link if '%s' has "
3674 "an implicit backing file", bs->node_name);
3675 return -EPERM;
3677 /* Check if the backing link that we want to replace is frozen */
3678 if (bdrv_is_backing_chain_frozen(overlay_bs, backing_bs(overlay_bs),
3679 errp)) {
3680 return -EPERM;
3682 reopen_state->replace_backing_bs = true;
3683 if (new_backing_bs) {
3684 bdrv_ref(new_backing_bs);
3685 reopen_state->new_backing_bs = new_backing_bs;
3689 return 0;
3693 * Prepares a BlockDriverState for reopen. All changes are staged in the
3694 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3695 * the block driver layer .bdrv_reopen_prepare()
3697 * bs is the BlockDriverState to reopen
3698 * flags are the new open flags
3699 * queue is the reopen queue
3701 * Returns 0 on success, non-zero on error. On error errp will be set
3702 * as well.
3704 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3705 * It is the responsibility of the caller to then call the abort() or
3706 * commit() for any other BDS that have been left in a prepare() state
3709 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3710 Error **errp)
3712 int ret = -1;
3713 int old_flags;
3714 Error *local_err = NULL;
3715 BlockDriver *drv;
3716 QemuOpts *opts;
3717 QDict *orig_reopen_opts;
3718 char *discard = NULL;
3719 bool read_only;
3720 bool drv_prepared = false;
3722 assert(reopen_state != NULL);
3723 assert(reopen_state->bs->drv != NULL);
3724 drv = reopen_state->bs->drv;
3726 /* This function and each driver's bdrv_reopen_prepare() remove
3727 * entries from reopen_state->options as they are processed, so
3728 * we need to make a copy of the original QDict. */
3729 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3731 /* Process generic block layer options */
3732 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3733 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3734 if (local_err) {
3735 error_propagate(errp, local_err);
3736 ret = -EINVAL;
3737 goto error;
3740 /* This was already called in bdrv_reopen_queue_child() so the flags
3741 * are up-to-date. This time we simply want to remove the options from
3742 * QemuOpts in order to indicate that they have been processed. */
3743 old_flags = reopen_state->flags;
3744 update_flags_from_options(&reopen_state->flags, opts);
3745 assert(old_flags == reopen_state->flags);
3747 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
3748 if (discard != NULL) {
3749 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3750 error_setg(errp, "Invalid discard option");
3751 ret = -EINVAL;
3752 goto error;
3756 reopen_state->detect_zeroes =
3757 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3758 if (local_err) {
3759 error_propagate(errp, local_err);
3760 ret = -EINVAL;
3761 goto error;
3764 /* All other options (including node-name and driver) must be unchanged.
3765 * Put them back into the QDict, so that they are checked at the end
3766 * of this function. */
3767 qemu_opts_to_qdict(opts, reopen_state->options);
3769 /* If we are to stay read-only, do not allow permission change
3770 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3771 * not set, or if the BDS still has copy_on_read enabled */
3772 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3773 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3774 if (local_err) {
3775 error_propagate(errp, local_err);
3776 goto error;
3779 /* Calculate required permissions after reopening */
3780 bdrv_reopen_perm(queue, reopen_state->bs,
3781 &reopen_state->perm, &reopen_state->shared_perm);
3783 ret = bdrv_flush(reopen_state->bs);
3784 if (ret) {
3785 error_setg_errno(errp, -ret, "Error flushing drive");
3786 goto error;
3789 if (drv->bdrv_reopen_prepare) {
3791 * If a driver-specific option is missing, it means that we
3792 * should reset it to its default value.
3793 * But not all options allow that, so we need to check it first.
3795 ret = bdrv_reset_options_allowed(reopen_state->bs,
3796 reopen_state->options, errp);
3797 if (ret) {
3798 goto error;
3801 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3802 if (ret) {
3803 if (local_err != NULL) {
3804 error_propagate(errp, local_err);
3805 } else {
3806 bdrv_refresh_filename(reopen_state->bs);
3807 error_setg(errp, "failed while preparing to reopen image '%s'",
3808 reopen_state->bs->filename);
3810 goto error;
3812 } else {
3813 /* It is currently mandatory to have a bdrv_reopen_prepare()
3814 * handler for each supported drv. */
3815 error_setg(errp, "Block format '%s' used by node '%s' "
3816 "does not support reopening files", drv->format_name,
3817 bdrv_get_device_or_node_name(reopen_state->bs));
3818 ret = -1;
3819 goto error;
3822 drv_prepared = true;
3825 * We must provide the 'backing' option if the BDS has a backing
3826 * file or if the image file has a backing file name as part of
3827 * its metadata. Otherwise the 'backing' option can be omitted.
3829 if (drv->supports_backing && reopen_state->backing_missing &&
3830 (backing_bs(reopen_state->bs) || reopen_state->bs->backing_file[0])) {
3831 error_setg(errp, "backing is missing for '%s'",
3832 reopen_state->bs->node_name);
3833 ret = -EINVAL;
3834 goto error;
3838 * Allow changing the 'backing' option. The new value can be
3839 * either a reference to an existing node (using its node name)
3840 * or NULL to simply detach the current backing file.
3842 ret = bdrv_reopen_parse_backing(reopen_state, errp);
3843 if (ret < 0) {
3844 goto error;
3846 qdict_del(reopen_state->options, "backing");
3848 /* Options that are not handled are only okay if they are unchanged
3849 * compared to the old state. It is expected that some options are only
3850 * used for the initial open, but not reopen (e.g. filename) */
3851 if (qdict_size(reopen_state->options)) {
3852 const QDictEntry *entry = qdict_first(reopen_state->options);
3854 do {
3855 QObject *new = entry->value;
3856 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3858 /* Allow child references (child_name=node_name) as long as they
3859 * point to the current child (i.e. everything stays the same). */
3860 if (qobject_type(new) == QTYPE_QSTRING) {
3861 BdrvChild *child;
3862 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
3863 if (!strcmp(child->name, entry->key)) {
3864 break;
3868 if (child) {
3869 const char *str = qobject_get_try_str(new);
3870 if (!strcmp(child->bs->node_name, str)) {
3871 continue; /* Found child with this name, skip option */
3877 * TODO: When using -drive to specify blockdev options, all values
3878 * will be strings; however, when using -blockdev, blockdev-add or
3879 * filenames using the json:{} pseudo-protocol, they will be
3880 * correctly typed.
3881 * In contrast, reopening options are (currently) always strings
3882 * (because you can only specify them through qemu-io; all other
3883 * callers do not specify any options).
3884 * Therefore, when using anything other than -drive to create a BDS,
3885 * this cannot detect non-string options as unchanged, because
3886 * qobject_is_equal() always returns false for objects of different
3887 * type. In the future, this should be remedied by correctly typing
3888 * all options. For now, this is not too big of an issue because
3889 * the user can simply omit options which cannot be changed anyway,
3890 * so they will stay unchanged.
3892 if (!qobject_is_equal(new, old)) {
3893 error_setg(errp, "Cannot change the option '%s'", entry->key);
3894 ret = -EINVAL;
3895 goto error;
3897 } while ((entry = qdict_next(reopen_state->options, entry)));
3900 ret = 0;
3902 /* Restore the original reopen_state->options QDict */
3903 qobject_unref(reopen_state->options);
3904 reopen_state->options = qobject_ref(orig_reopen_opts);
3906 error:
3907 if (ret < 0 && drv_prepared) {
3908 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3909 * call drv->bdrv_reopen_abort() before signaling an error
3910 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3911 * when the respective bdrv_reopen_prepare() has failed) */
3912 if (drv->bdrv_reopen_abort) {
3913 drv->bdrv_reopen_abort(reopen_state);
3916 qemu_opts_del(opts);
3917 qobject_unref(orig_reopen_opts);
3918 g_free(discard);
3919 return ret;
3923 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3924 * makes them final by swapping the staging BlockDriverState contents into
3925 * the active BlockDriverState contents.
3927 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3929 BlockDriver *drv;
3930 BlockDriverState *bs;
3931 BdrvChild *child;
3932 bool old_can_write, new_can_write;
3934 assert(reopen_state != NULL);
3935 bs = reopen_state->bs;
3936 drv = bs->drv;
3937 assert(drv != NULL);
3939 old_can_write =
3940 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3942 /* If there are any driver level actions to take */
3943 if (drv->bdrv_reopen_commit) {
3944 drv->bdrv_reopen_commit(reopen_state);
3947 /* set BDS specific flags now */
3948 qobject_unref(bs->explicit_options);
3949 qobject_unref(bs->options);
3951 bs->explicit_options = reopen_state->explicit_options;
3952 bs->options = reopen_state->options;
3953 bs->open_flags = reopen_state->flags;
3954 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3955 bs->detect_zeroes = reopen_state->detect_zeroes;
3957 if (reopen_state->replace_backing_bs) {
3958 qdict_del(bs->explicit_options, "backing");
3959 qdict_del(bs->options, "backing");
3962 /* Remove child references from bs->options and bs->explicit_options.
3963 * Child options were already removed in bdrv_reopen_queue_child() */
3964 QLIST_FOREACH(child, &bs->children, next) {
3965 qdict_del(bs->explicit_options, child->name);
3966 qdict_del(bs->options, child->name);
3970 * Change the backing file if a new one was specified. We do this
3971 * after updating bs->options, so bdrv_refresh_filename() (called
3972 * from bdrv_set_backing_hd()) has the new values.
3974 if (reopen_state->replace_backing_bs) {
3975 BlockDriverState *old_backing_bs = backing_bs(bs);
3976 assert(!old_backing_bs || !old_backing_bs->implicit);
3977 /* Abort the permission update on the backing bs we're detaching */
3978 if (old_backing_bs) {
3979 bdrv_abort_perm_update(old_backing_bs);
3981 bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort);
3984 bdrv_refresh_limits(bs, NULL);
3986 new_can_write =
3987 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3988 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3989 Error *local_err = NULL;
3990 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3991 /* This is not fatal, bitmaps just left read-only, so all following
3992 * writes will fail. User can remove read-only bitmaps to unblock
3993 * writes.
3995 error_reportf_err(local_err,
3996 "%s: Failed to make dirty bitmaps writable: ",
3997 bdrv_get_node_name(bs));
4003 * Abort the reopen, and delete and free the staged changes in
4004 * reopen_state
4006 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4008 BlockDriver *drv;
4010 assert(reopen_state != NULL);
4011 drv = reopen_state->bs->drv;
4012 assert(drv != NULL);
4014 if (drv->bdrv_reopen_abort) {
4015 drv->bdrv_reopen_abort(reopen_state);
4020 static void bdrv_close(BlockDriverState *bs)
4022 BdrvAioNotifier *ban, *ban_next;
4023 BdrvChild *child, *next;
4025 assert(!bs->refcnt);
4027 bdrv_drained_begin(bs); /* complete I/O */
4028 bdrv_flush(bs);
4029 bdrv_drain(bs); /* in case flush left pending I/O */
4031 if (bs->drv) {
4032 if (bs->drv->bdrv_close) {
4033 bs->drv->bdrv_close(bs);
4035 bs->drv = NULL;
4038 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
4039 bdrv_unref_child(bs, child);
4042 bs->backing = NULL;
4043 bs->file = NULL;
4044 g_free(bs->opaque);
4045 bs->opaque = NULL;
4046 atomic_set(&bs->copy_on_read, 0);
4047 bs->backing_file[0] = '\0';
4048 bs->backing_format[0] = '\0';
4049 bs->total_sectors = 0;
4050 bs->encrypted = false;
4051 bs->sg = false;
4052 qobject_unref(bs->options);
4053 qobject_unref(bs->explicit_options);
4054 bs->options = NULL;
4055 bs->explicit_options = NULL;
4056 qobject_unref(bs->full_open_options);
4057 bs->full_open_options = NULL;
4059 bdrv_release_named_dirty_bitmaps(bs);
4060 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4062 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4063 g_free(ban);
4065 QLIST_INIT(&bs->aio_notifiers);
4066 bdrv_drained_end(bs);
4069 void bdrv_close_all(void)
4071 assert(job_next(NULL) == NULL);
4072 nbd_export_close_all();
4074 /* Drop references from requests still in flight, such as canceled block
4075 * jobs whose AIO context has not been polled yet */
4076 bdrv_drain_all();
4078 blk_remove_all_bs();
4079 blockdev_close_all_bdrv_states();
4081 assert(QTAILQ_EMPTY(&all_bdrv_states));
4084 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4086 GQueue *queue;
4087 GHashTable *found;
4088 bool ret;
4090 if (c->role->stay_at_node) {
4091 return false;
4094 /* If the child @c belongs to the BDS @to, replacing the current
4095 * c->bs by @to would mean to create a loop.
4097 * Such a case occurs when appending a BDS to a backing chain.
4098 * For instance, imagine the following chain:
4100 * guest device -> node A -> further backing chain...
4102 * Now we create a new BDS B which we want to put on top of this
4103 * chain, so we first attach A as its backing node:
4105 * node B
4108 * guest device -> node A -> further backing chain...
4110 * Finally we want to replace A by B. When doing that, we want to
4111 * replace all pointers to A by pointers to B -- except for the
4112 * pointer from B because (1) that would create a loop, and (2)
4113 * that pointer should simply stay intact:
4115 * guest device -> node B
4118 * node A -> further backing chain...
4120 * In general, when replacing a node A (c->bs) by a node B (@to),
4121 * if A is a child of B, that means we cannot replace A by B there
4122 * because that would create a loop. Silently detaching A from B
4123 * is also not really an option. So overall just leaving A in
4124 * place there is the most sensible choice.
4126 * We would also create a loop in any cases where @c is only
4127 * indirectly referenced by @to. Prevent this by returning false
4128 * if @c is found (by breadth-first search) anywhere in the whole
4129 * subtree of @to.
4132 ret = true;
4133 found = g_hash_table_new(NULL, NULL);
4134 g_hash_table_add(found, to);
4135 queue = g_queue_new();
4136 g_queue_push_tail(queue, to);
4138 while (!g_queue_is_empty(queue)) {
4139 BlockDriverState *v = g_queue_pop_head(queue);
4140 BdrvChild *c2;
4142 QLIST_FOREACH(c2, &v->children, next) {
4143 if (c2 == c) {
4144 ret = false;
4145 break;
4148 if (g_hash_table_contains(found, c2->bs)) {
4149 continue;
4152 g_queue_push_tail(queue, c2->bs);
4153 g_hash_table_add(found, c2->bs);
4157 g_queue_free(queue);
4158 g_hash_table_destroy(found);
4160 return ret;
4163 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
4164 Error **errp)
4166 BdrvChild *c, *next;
4167 GSList *list = NULL, *p;
4168 uint64_t old_perm, old_shared;
4169 uint64_t perm = 0, shared = BLK_PERM_ALL;
4170 int ret;
4172 /* Make sure that @from doesn't go away until we have successfully attached
4173 * all of its parents to @to. */
4174 bdrv_ref(from);
4176 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4177 bdrv_drained_begin(from);
4179 /* Put all parents into @list and calculate their cumulative permissions */
4180 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4181 assert(c->bs == from);
4182 if (!should_update_child(c, to)) {
4183 continue;
4185 if (c->frozen) {
4186 error_setg(errp, "Cannot change '%s' link to '%s'",
4187 c->name, from->node_name);
4188 goto out;
4190 list = g_slist_prepend(list, c);
4191 perm |= c->perm;
4192 shared &= c->shared_perm;
4195 /* Check whether the required permissions can be granted on @to, ignoring
4196 * all BdrvChild in @list so that they can't block themselves. */
4197 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, NULL, errp);
4198 if (ret < 0) {
4199 bdrv_abort_perm_update(to);
4200 goto out;
4203 /* Now actually perform the change. We performed the permission check for
4204 * all elements of @list at once, so set the permissions all at once at the
4205 * very end. */
4206 for (p = list; p != NULL; p = p->next) {
4207 c = p->data;
4209 bdrv_ref(to);
4210 bdrv_replace_child_noperm(c, to);
4211 bdrv_unref(from);
4214 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
4215 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
4217 out:
4218 g_slist_free(list);
4219 bdrv_drained_end(from);
4220 bdrv_unref(from);
4224 * Add new bs contents at the top of an image chain while the chain is
4225 * live, while keeping required fields on the top layer.
4227 * This will modify the BlockDriverState fields, and swap contents
4228 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4230 * bs_new must not be attached to a BlockBackend.
4232 * This function does not create any image files.
4234 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4235 * that's what the callers commonly need. bs_new will be referenced by the old
4236 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4237 * reference of its own, it must call bdrv_ref().
4239 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
4240 Error **errp)
4242 Error *local_err = NULL;
4244 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
4245 if (local_err) {
4246 error_propagate(errp, local_err);
4247 goto out;
4250 bdrv_replace_node(bs_top, bs_new, &local_err);
4251 if (local_err) {
4252 error_propagate(errp, local_err);
4253 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
4254 goto out;
4257 /* bs_new is now referenced by its new parents, we don't need the
4258 * additional reference any more. */
4259 out:
4260 bdrv_unref(bs_new);
4263 static void bdrv_delete(BlockDriverState *bs)
4265 assert(bdrv_op_blocker_is_empty(bs));
4266 assert(!bs->refcnt);
4268 /* remove from list, if necessary */
4269 if (bs->node_name[0] != '\0') {
4270 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
4272 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
4274 bdrv_close(bs);
4276 g_free(bs);
4280 * Run consistency checks on an image
4282 * Returns 0 if the check could be completed (it doesn't mean that the image is
4283 * free of errors) or -errno when an internal error occurred. The results of the
4284 * check are stored in res.
4286 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
4287 BdrvCheckResult *res, BdrvCheckMode fix)
4289 if (bs->drv == NULL) {
4290 return -ENOMEDIUM;
4292 if (bs->drv->bdrv_co_check == NULL) {
4293 return -ENOTSUP;
4296 memset(res, 0, sizeof(*res));
4297 return bs->drv->bdrv_co_check(bs, res, fix);
4300 typedef struct CheckCo {
4301 BlockDriverState *bs;
4302 BdrvCheckResult *res;
4303 BdrvCheckMode fix;
4304 int ret;
4305 } CheckCo;
4307 static void coroutine_fn bdrv_check_co_entry(void *opaque)
4309 CheckCo *cco = opaque;
4310 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
4311 aio_wait_kick();
4314 int bdrv_check(BlockDriverState *bs,
4315 BdrvCheckResult *res, BdrvCheckMode fix)
4317 Coroutine *co;
4318 CheckCo cco = {
4319 .bs = bs,
4320 .res = res,
4321 .ret = -EINPROGRESS,
4322 .fix = fix,
4325 if (qemu_in_coroutine()) {
4326 /* Fast-path if already in coroutine context */
4327 bdrv_check_co_entry(&cco);
4328 } else {
4329 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
4330 bdrv_coroutine_enter(bs, co);
4331 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
4334 return cco.ret;
4338 * Return values:
4339 * 0 - success
4340 * -EINVAL - backing format specified, but no file
4341 * -ENOSPC - can't update the backing file because no space is left in the
4342 * image file header
4343 * -ENOTSUP - format driver doesn't support changing the backing file
4345 int bdrv_change_backing_file(BlockDriverState *bs,
4346 const char *backing_file, const char *backing_fmt)
4348 BlockDriver *drv = bs->drv;
4349 int ret;
4351 if (!drv) {
4352 return -ENOMEDIUM;
4355 /* Backing file format doesn't make sense without a backing file */
4356 if (backing_fmt && !backing_file) {
4357 return -EINVAL;
4360 if (drv->bdrv_change_backing_file != NULL) {
4361 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
4362 } else {
4363 ret = -ENOTSUP;
4366 if (ret == 0) {
4367 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
4368 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
4369 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
4370 backing_file ?: "");
4372 return ret;
4376 * Finds the image layer in the chain that has 'bs' as its backing file.
4378 * active is the current topmost image.
4380 * Returns NULL if bs is not found in active's image chain,
4381 * or if active == bs.
4383 * Returns the bottommost base image if bs == NULL.
4385 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
4386 BlockDriverState *bs)
4388 while (active && bs != backing_bs(active)) {
4389 active = backing_bs(active);
4392 return active;
4395 /* Given a BDS, searches for the base layer. */
4396 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4398 return bdrv_find_overlay(bs, NULL);
4402 * Return true if at least one of the backing links between @bs and
4403 * @base is frozen. @errp is set if that's the case.
4404 * @base must be reachable from @bs, or NULL.
4406 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
4407 Error **errp)
4409 BlockDriverState *i;
4411 for (i = bs; i != base; i = backing_bs(i)) {
4412 if (i->backing && i->backing->frozen) {
4413 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4414 i->backing->name, i->node_name,
4415 backing_bs(i)->node_name);
4416 return true;
4420 return false;
4424 * Freeze all backing links between @bs and @base.
4425 * If any of the links is already frozen the operation is aborted and
4426 * none of the links are modified.
4427 * @base must be reachable from @bs, or NULL.
4428 * Returns 0 on success. On failure returns < 0 and sets @errp.
4430 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4431 Error **errp)
4433 BlockDriverState *i;
4435 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4436 return -EPERM;
4439 for (i = bs; i != base; i = backing_bs(i)) {
4440 if (i->backing && backing_bs(i)->never_freeze) {
4441 error_setg(errp, "Cannot freeze '%s' link to '%s'",
4442 i->backing->name, backing_bs(i)->node_name);
4443 return -EPERM;
4447 for (i = bs; i != base; i = backing_bs(i)) {
4448 if (i->backing) {
4449 i->backing->frozen = true;
4453 return 0;
4457 * Unfreeze all backing links between @bs and @base. The caller must
4458 * ensure that all links are frozen before using this function.
4459 * @base must be reachable from @bs, or NULL.
4461 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4463 BlockDriverState *i;
4465 for (i = bs; i != base; i = backing_bs(i)) {
4466 if (i->backing) {
4467 assert(i->backing->frozen);
4468 i->backing->frozen = false;
4474 * Drops images above 'base' up to and including 'top', and sets the image
4475 * above 'top' to have base as its backing file.
4477 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4478 * information in 'bs' can be properly updated.
4480 * E.g., this will convert the following chain:
4481 * bottom <- base <- intermediate <- top <- active
4483 * to
4485 * bottom <- base <- active
4487 * It is allowed for bottom==base, in which case it converts:
4489 * base <- intermediate <- top <- active
4491 * to
4493 * base <- active
4495 * If backing_file_str is non-NULL, it will be used when modifying top's
4496 * overlay image metadata.
4498 * Error conditions:
4499 * if active == top, that is considered an error
4502 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4503 const char *backing_file_str)
4505 BlockDriverState *explicit_top = top;
4506 bool update_inherits_from;
4507 BdrvChild *c, *next;
4508 Error *local_err = NULL;
4509 int ret = -EIO;
4511 bdrv_ref(top);
4512 bdrv_subtree_drained_begin(top);
4514 if (!top->drv || !base->drv) {
4515 goto exit;
4518 /* Make sure that base is in the backing chain of top */
4519 if (!bdrv_chain_contains(top, base)) {
4520 goto exit;
4523 /* This function changes all links that point to top and makes
4524 * them point to base. Check that none of them is frozen. */
4525 QLIST_FOREACH(c, &top->parents, next_parent) {
4526 if (c->frozen) {
4527 goto exit;
4531 /* If 'base' recursively inherits from 'top' then we should set
4532 * base->inherits_from to top->inherits_from after 'top' and all
4533 * other intermediate nodes have been dropped.
4534 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4535 * it because no one inherits from it. We use explicit_top for that. */
4536 while (explicit_top && explicit_top->implicit) {
4537 explicit_top = backing_bs(explicit_top);
4539 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4541 /* success - we can delete the intermediate states, and link top->base */
4542 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4543 * we've figured out how they should work. */
4544 if (!backing_file_str) {
4545 bdrv_refresh_filename(base);
4546 backing_file_str = base->filename;
4549 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
4550 /* Check whether we are allowed to switch c from top to base */
4551 GSList *ignore_children = g_slist_prepend(NULL, c);
4552 ret = bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
4553 ignore_children, NULL, &local_err);
4554 g_slist_free(ignore_children);
4555 if (ret < 0) {
4556 error_report_err(local_err);
4557 goto exit;
4560 /* If so, update the backing file path in the image file */
4561 if (c->role->update_filename) {
4562 ret = c->role->update_filename(c, base, backing_file_str,
4563 &local_err);
4564 if (ret < 0) {
4565 bdrv_abort_perm_update(base);
4566 error_report_err(local_err);
4567 goto exit;
4571 /* Do the actual switch in the in-memory graph.
4572 * Completes bdrv_check_update_perm() transaction internally. */
4573 bdrv_ref(base);
4574 bdrv_replace_child(c, base);
4575 bdrv_unref(top);
4578 if (update_inherits_from) {
4579 base->inherits_from = explicit_top->inherits_from;
4582 ret = 0;
4583 exit:
4584 bdrv_subtree_drained_end(top);
4585 bdrv_unref(top);
4586 return ret;
4590 * Length of a allocated file in bytes. Sparse files are counted by actual
4591 * allocated space. Return < 0 if error or unknown.
4593 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
4595 BlockDriver *drv = bs->drv;
4596 if (!drv) {
4597 return -ENOMEDIUM;
4599 if (drv->bdrv_get_allocated_file_size) {
4600 return drv->bdrv_get_allocated_file_size(bs);
4602 if (bs->file) {
4603 return bdrv_get_allocated_file_size(bs->file->bs);
4605 return -ENOTSUP;
4609 * bdrv_measure:
4610 * @drv: Format driver
4611 * @opts: Creation options for new image
4612 * @in_bs: Existing image containing data for new image (may be NULL)
4613 * @errp: Error object
4614 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4615 * or NULL on error
4617 * Calculate file size required to create a new image.
4619 * If @in_bs is given then space for allocated clusters and zero clusters
4620 * from that image are included in the calculation. If @opts contains a
4621 * backing file that is shared by @in_bs then backing clusters may be omitted
4622 * from the calculation.
4624 * If @in_bs is NULL then the calculation includes no allocated clusters
4625 * unless a preallocation option is given in @opts.
4627 * Note that @in_bs may use a different BlockDriver from @drv.
4629 * If an error occurs the @errp pointer is set.
4631 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
4632 BlockDriverState *in_bs, Error **errp)
4634 if (!drv->bdrv_measure) {
4635 error_setg(errp, "Block driver '%s' does not support size measurement",
4636 drv->format_name);
4637 return NULL;
4640 return drv->bdrv_measure(opts, in_bs, errp);
4644 * Return number of sectors on success, -errno on error.
4646 int64_t bdrv_nb_sectors(BlockDriverState *bs)
4648 BlockDriver *drv = bs->drv;
4650 if (!drv)
4651 return -ENOMEDIUM;
4653 if (drv->has_variable_length) {
4654 int ret = refresh_total_sectors(bs, bs->total_sectors);
4655 if (ret < 0) {
4656 return ret;
4659 return bs->total_sectors;
4663 * Return length in bytes on success, -errno on error.
4664 * The length is always a multiple of BDRV_SECTOR_SIZE.
4666 int64_t bdrv_getlength(BlockDriverState *bs)
4668 int64_t ret = bdrv_nb_sectors(bs);
4670 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
4671 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
4674 /* return 0 as number of sectors if no device present or error */
4675 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
4677 int64_t nb_sectors = bdrv_nb_sectors(bs);
4679 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
4682 bool bdrv_is_sg(BlockDriverState *bs)
4684 return bs->sg;
4687 bool bdrv_is_encrypted(BlockDriverState *bs)
4689 if (bs->backing && bs->backing->bs->encrypted) {
4690 return true;
4692 return bs->encrypted;
4695 const char *bdrv_get_format_name(BlockDriverState *bs)
4697 return bs->drv ? bs->drv->format_name : NULL;
4700 static int qsort_strcmp(const void *a, const void *b)
4702 return strcmp(*(char *const *)a, *(char *const *)b);
4705 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
4706 void *opaque, bool read_only)
4708 BlockDriver *drv;
4709 int count = 0;
4710 int i;
4711 const char **formats = NULL;
4713 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4714 if (drv->format_name) {
4715 bool found = false;
4716 int i = count;
4718 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
4719 continue;
4722 while (formats && i && !found) {
4723 found = !strcmp(formats[--i], drv->format_name);
4726 if (!found) {
4727 formats = g_renew(const char *, formats, count + 1);
4728 formats[count++] = drv->format_name;
4733 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4734 const char *format_name = block_driver_modules[i].format_name;
4736 if (format_name) {
4737 bool found = false;
4738 int j = count;
4740 if (use_bdrv_whitelist &&
4741 !bdrv_format_is_whitelisted(format_name, read_only)) {
4742 continue;
4745 while (formats && j && !found) {
4746 found = !strcmp(formats[--j], format_name);
4749 if (!found) {
4750 formats = g_renew(const char *, formats, count + 1);
4751 formats[count++] = format_name;
4756 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
4758 for (i = 0; i < count; i++) {
4759 it(opaque, formats[i]);
4762 g_free(formats);
4765 /* This function is to find a node in the bs graph */
4766 BlockDriverState *bdrv_find_node(const char *node_name)
4768 BlockDriverState *bs;
4770 assert(node_name);
4772 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4773 if (!strcmp(node_name, bs->node_name)) {
4774 return bs;
4777 return NULL;
4780 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4781 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
4783 BlockDeviceInfoList *list, *entry;
4784 BlockDriverState *bs;
4786 list = NULL;
4787 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4788 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
4789 if (!info) {
4790 qapi_free_BlockDeviceInfoList(list);
4791 return NULL;
4793 entry = g_malloc0(sizeof(*entry));
4794 entry->value = info;
4795 entry->next = list;
4796 list = entry;
4799 return list;
4802 #define QAPI_LIST_ADD(list, element) do { \
4803 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4804 _tmp->value = (element); \
4805 _tmp->next = (list); \
4806 (list) = _tmp; \
4807 } while (0)
4809 typedef struct XDbgBlockGraphConstructor {
4810 XDbgBlockGraph *graph;
4811 GHashTable *graph_nodes;
4812 } XDbgBlockGraphConstructor;
4814 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
4816 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
4818 gr->graph = g_new0(XDbgBlockGraph, 1);
4819 gr->graph_nodes = g_hash_table_new(NULL, NULL);
4821 return gr;
4824 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
4826 XDbgBlockGraph *graph = gr->graph;
4828 g_hash_table_destroy(gr->graph_nodes);
4829 g_free(gr);
4831 return graph;
4834 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
4836 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
4838 if (ret != 0) {
4839 return ret;
4843 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4844 * answer of g_hash_table_lookup.
4846 ret = g_hash_table_size(gr->graph_nodes) + 1;
4847 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
4849 return ret;
4852 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
4853 XDbgBlockGraphNodeType type, const char *name)
4855 XDbgBlockGraphNode *n;
4857 n = g_new0(XDbgBlockGraphNode, 1);
4859 n->id = xdbg_graph_node_num(gr, node);
4860 n->type = type;
4861 n->name = g_strdup(name);
4863 QAPI_LIST_ADD(gr->graph->nodes, n);
4866 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
4867 const BdrvChild *child)
4869 typedef struct {
4870 unsigned int flag;
4871 BlockPermission num;
4872 } PermissionMap;
4874 static const PermissionMap permissions[] = {
4875 { BLK_PERM_CONSISTENT_READ, BLOCK_PERMISSION_CONSISTENT_READ },
4876 { BLK_PERM_WRITE, BLOCK_PERMISSION_WRITE },
4877 { BLK_PERM_WRITE_UNCHANGED, BLOCK_PERMISSION_WRITE_UNCHANGED },
4878 { BLK_PERM_RESIZE, BLOCK_PERMISSION_RESIZE },
4879 { BLK_PERM_GRAPH_MOD, BLOCK_PERMISSION_GRAPH_MOD },
4880 { 0, 0 }
4882 const PermissionMap *p;
4883 XDbgBlockGraphEdge *edge;
4885 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions) - 1) != BLK_PERM_ALL + 1);
4887 edge = g_new0(XDbgBlockGraphEdge, 1);
4889 edge->parent = xdbg_graph_node_num(gr, parent);
4890 edge->child = xdbg_graph_node_num(gr, child->bs);
4891 edge->name = g_strdup(child->name);
4893 for (p = permissions; p->flag; p++) {
4894 if (p->flag & child->perm) {
4895 QAPI_LIST_ADD(edge->perm, p->num);
4897 if (p->flag & child->shared_perm) {
4898 QAPI_LIST_ADD(edge->shared_perm, p->num);
4902 QAPI_LIST_ADD(gr->graph->edges, edge);
4906 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
4908 BlockBackend *blk;
4909 BlockJob *job;
4910 BlockDriverState *bs;
4911 BdrvChild *child;
4912 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
4914 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
4915 char *allocated_name = NULL;
4916 const char *name = blk_name(blk);
4918 if (!*name) {
4919 name = allocated_name = blk_get_attached_dev_id(blk);
4921 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
4922 name);
4923 g_free(allocated_name);
4924 if (blk_root(blk)) {
4925 xdbg_graph_add_edge(gr, blk, blk_root(blk));
4929 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4930 GSList *el;
4932 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
4933 job->job.id);
4934 for (el = job->nodes; el; el = el->next) {
4935 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
4939 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4940 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
4941 bs->node_name);
4942 QLIST_FOREACH(child, &bs->children, next) {
4943 xdbg_graph_add_edge(gr, bs, child);
4947 return xdbg_graph_finalize(gr);
4950 BlockDriverState *bdrv_lookup_bs(const char *device,
4951 const char *node_name,
4952 Error **errp)
4954 BlockBackend *blk;
4955 BlockDriverState *bs;
4957 if (device) {
4958 blk = blk_by_name(device);
4960 if (blk) {
4961 bs = blk_bs(blk);
4962 if (!bs) {
4963 error_setg(errp, "Device '%s' has no medium", device);
4966 return bs;
4970 if (node_name) {
4971 bs = bdrv_find_node(node_name);
4973 if (bs) {
4974 return bs;
4978 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4979 device ? device : "",
4980 node_name ? node_name : "");
4981 return NULL;
4984 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4985 * return false. If either argument is NULL, return false. */
4986 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4988 while (top && top != base) {
4989 top = backing_bs(top);
4992 return top != NULL;
4995 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4997 if (!bs) {
4998 return QTAILQ_FIRST(&graph_bdrv_states);
5000 return QTAILQ_NEXT(bs, node_list);
5003 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5005 if (!bs) {
5006 return QTAILQ_FIRST(&all_bdrv_states);
5008 return QTAILQ_NEXT(bs, bs_list);
5011 const char *bdrv_get_node_name(const BlockDriverState *bs)
5013 return bs->node_name;
5016 const char *bdrv_get_parent_name(const BlockDriverState *bs)
5018 BdrvChild *c;
5019 const char *name;
5021 /* If multiple parents have a name, just pick the first one. */
5022 QLIST_FOREACH(c, &bs->parents, next_parent) {
5023 if (c->role->get_name) {
5024 name = c->role->get_name(c);
5025 if (name && *name) {
5026 return name;
5031 return NULL;
5034 /* TODO check what callers really want: bs->node_name or blk_name() */
5035 const char *bdrv_get_device_name(const BlockDriverState *bs)
5037 return bdrv_get_parent_name(bs) ?: "";
5040 /* This can be used to identify nodes that might not have a device
5041 * name associated. Since node and device names live in the same
5042 * namespace, the result is unambiguous. The exception is if both are
5043 * absent, then this returns an empty (non-null) string. */
5044 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
5046 return bdrv_get_parent_name(bs) ?: bs->node_name;
5049 int bdrv_get_flags(BlockDriverState *bs)
5051 return bs->open_flags;
5054 int bdrv_has_zero_init_1(BlockDriverState *bs)
5056 return 1;
5059 int bdrv_has_zero_init(BlockDriverState *bs)
5061 if (!bs->drv) {
5062 return 0;
5065 /* If BS is a copy on write image, it is initialized to
5066 the contents of the base image, which may not be zeroes. */
5067 if (bs->backing) {
5068 return 0;
5070 if (bs->drv->bdrv_has_zero_init) {
5071 return bs->drv->bdrv_has_zero_init(bs);
5073 if (bs->file && bs->drv->is_filter) {
5074 return bdrv_has_zero_init(bs->file->bs);
5077 /* safe default */
5078 return 0;
5081 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
5083 BlockDriverInfo bdi;
5085 if (bs->backing) {
5086 return false;
5089 if (bdrv_get_info(bs, &bdi) == 0) {
5090 return bdi.unallocated_blocks_are_zero;
5093 return false;
5096 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
5098 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5099 return false;
5102 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
5105 void bdrv_get_backing_filename(BlockDriverState *bs,
5106 char *filename, int filename_size)
5108 pstrcpy(filename, filename_size, bs->backing_file);
5111 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5113 BlockDriver *drv = bs->drv;
5114 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5115 if (!drv) {
5116 return -ENOMEDIUM;
5118 if (!drv->bdrv_get_info) {
5119 if (bs->file && drv->is_filter) {
5120 return bdrv_get_info(bs->file->bs, bdi);
5122 return -ENOTSUP;
5124 memset(bdi, 0, sizeof(*bdi));
5125 return drv->bdrv_get_info(bs, bdi);
5128 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
5129 Error **errp)
5131 BlockDriver *drv = bs->drv;
5132 if (drv && drv->bdrv_get_specific_info) {
5133 return drv->bdrv_get_specific_info(bs, errp);
5135 return NULL;
5138 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
5140 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
5141 return;
5144 bs->drv->bdrv_debug_event(bs, event);
5147 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
5148 const char *tag)
5150 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
5151 bs = bs->file ? bs->file->bs : NULL;
5154 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
5155 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
5158 return -ENOTSUP;
5161 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
5163 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
5164 bs = bs->file ? bs->file->bs : NULL;
5167 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
5168 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
5171 return -ENOTSUP;
5174 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
5176 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
5177 bs = bs->file ? bs->file->bs : NULL;
5180 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
5181 return bs->drv->bdrv_debug_resume(bs, tag);
5184 return -ENOTSUP;
5187 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
5189 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
5190 bs = bs->file ? bs->file->bs : NULL;
5193 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
5194 return bs->drv->bdrv_debug_is_suspended(bs, tag);
5197 return false;
5200 /* backing_file can either be relative, or absolute, or a protocol. If it is
5201 * relative, it must be relative to the chain. So, passing in bs->filename
5202 * from a BDS as backing_file should not be done, as that may be relative to
5203 * the CWD rather than the chain. */
5204 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
5205 const char *backing_file)
5207 char *filename_full = NULL;
5208 char *backing_file_full = NULL;
5209 char *filename_tmp = NULL;
5210 int is_protocol = 0;
5211 BlockDriverState *curr_bs = NULL;
5212 BlockDriverState *retval = NULL;
5214 if (!bs || !bs->drv || !backing_file) {
5215 return NULL;
5218 filename_full = g_malloc(PATH_MAX);
5219 backing_file_full = g_malloc(PATH_MAX);
5221 is_protocol = path_has_protocol(backing_file);
5223 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
5225 /* If either of the filename paths is actually a protocol, then
5226 * compare unmodified paths; otherwise make paths relative */
5227 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
5228 char *backing_file_full_ret;
5230 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
5231 retval = curr_bs->backing->bs;
5232 break;
5234 /* Also check against the full backing filename for the image */
5235 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
5236 NULL);
5237 if (backing_file_full_ret) {
5238 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
5239 g_free(backing_file_full_ret);
5240 if (equal) {
5241 retval = curr_bs->backing->bs;
5242 break;
5245 } else {
5246 /* If not an absolute filename path, make it relative to the current
5247 * image's filename path */
5248 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
5249 NULL);
5250 /* We are going to compare canonicalized absolute pathnames */
5251 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
5252 g_free(filename_tmp);
5253 continue;
5255 g_free(filename_tmp);
5257 /* We need to make sure the backing filename we are comparing against
5258 * is relative to the current image filename (or absolute) */
5259 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
5260 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
5261 g_free(filename_tmp);
5262 continue;
5264 g_free(filename_tmp);
5266 if (strcmp(backing_file_full, filename_full) == 0) {
5267 retval = curr_bs->backing->bs;
5268 break;
5273 g_free(filename_full);
5274 g_free(backing_file_full);
5275 return retval;
5278 void bdrv_init(void)
5280 module_call_init(MODULE_INIT_BLOCK);
5283 void bdrv_init_with_whitelist(void)
5285 use_bdrv_whitelist = 1;
5286 bdrv_init();
5289 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
5290 Error **errp)
5292 BdrvChild *child, *parent;
5293 uint64_t perm, shared_perm;
5294 Error *local_err = NULL;
5295 int ret;
5296 BdrvDirtyBitmap *bm;
5298 if (!bs->drv) {
5299 return;
5302 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
5303 return;
5306 QLIST_FOREACH(child, &bs->children, next) {
5307 bdrv_co_invalidate_cache(child->bs, &local_err);
5308 if (local_err) {
5309 error_propagate(errp, local_err);
5310 return;
5315 * Update permissions, they may differ for inactive nodes.
5317 * Note that the required permissions of inactive images are always a
5318 * subset of the permissions required after activating the image. This
5319 * allows us to just get the permissions upfront without restricting
5320 * drv->bdrv_invalidate_cache().
5322 * It also means that in error cases, we don't have to try and revert to
5323 * the old permissions (which is an operation that could fail, too). We can
5324 * just keep the extended permissions for the next time that an activation
5325 * of the image is tried.
5327 bs->open_flags &= ~BDRV_O_INACTIVE;
5328 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5329 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, NULL, &local_err);
5330 if (ret < 0) {
5331 bs->open_flags |= BDRV_O_INACTIVE;
5332 error_propagate(errp, local_err);
5333 return;
5335 bdrv_set_perm(bs, perm, shared_perm);
5337 if (bs->drv->bdrv_co_invalidate_cache) {
5338 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
5339 if (local_err) {
5340 bs->open_flags |= BDRV_O_INACTIVE;
5341 error_propagate(errp, local_err);
5342 return;
5346 for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm;
5347 bm = bdrv_dirty_bitmap_next(bs, bm))
5349 bdrv_dirty_bitmap_skip_store(bm, false);
5352 ret = refresh_total_sectors(bs, bs->total_sectors);
5353 if (ret < 0) {
5354 bs->open_flags |= BDRV_O_INACTIVE;
5355 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5356 return;
5359 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5360 if (parent->role->activate) {
5361 parent->role->activate(parent, &local_err);
5362 if (local_err) {
5363 bs->open_flags |= BDRV_O_INACTIVE;
5364 error_propagate(errp, local_err);
5365 return;
5371 typedef struct InvalidateCacheCo {
5372 BlockDriverState *bs;
5373 Error **errp;
5374 bool done;
5375 } InvalidateCacheCo;
5377 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
5379 InvalidateCacheCo *ico = opaque;
5380 bdrv_co_invalidate_cache(ico->bs, ico->errp);
5381 ico->done = true;
5382 aio_wait_kick();
5385 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
5387 Coroutine *co;
5388 InvalidateCacheCo ico = {
5389 .bs = bs,
5390 .done = false,
5391 .errp = errp
5394 if (qemu_in_coroutine()) {
5395 /* Fast-path if already in coroutine context */
5396 bdrv_invalidate_cache_co_entry(&ico);
5397 } else {
5398 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
5399 bdrv_coroutine_enter(bs, co);
5400 BDRV_POLL_WHILE(bs, !ico.done);
5404 void bdrv_invalidate_cache_all(Error **errp)
5406 BlockDriverState *bs;
5407 Error *local_err = NULL;
5408 BdrvNextIterator it;
5410 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5411 AioContext *aio_context = bdrv_get_aio_context(bs);
5413 aio_context_acquire(aio_context);
5414 bdrv_invalidate_cache(bs, &local_err);
5415 aio_context_release(aio_context);
5416 if (local_err) {
5417 error_propagate(errp, local_err);
5418 bdrv_next_cleanup(&it);
5419 return;
5424 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5426 BdrvChild *parent;
5428 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5429 if (parent->role->parent_is_bds) {
5430 BlockDriverState *parent_bs = parent->opaque;
5431 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5432 return true;
5437 return false;
5440 static int bdrv_inactivate_recurse(BlockDriverState *bs)
5442 BdrvChild *child, *parent;
5443 bool tighten_restrictions;
5444 uint64_t perm, shared_perm;
5445 int ret;
5447 if (!bs->drv) {
5448 return -ENOMEDIUM;
5451 /* Make sure that we don't inactivate a child before its parent.
5452 * It will be covered by recursion from the yet active parent. */
5453 if (bdrv_has_bds_parent(bs, true)) {
5454 return 0;
5457 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5459 /* Inactivate this node */
5460 if (bs->drv->bdrv_inactivate) {
5461 ret = bs->drv->bdrv_inactivate(bs);
5462 if (ret < 0) {
5463 return ret;
5467 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5468 if (parent->role->inactivate) {
5469 ret = parent->role->inactivate(parent);
5470 if (ret < 0) {
5471 return ret;
5476 bs->open_flags |= BDRV_O_INACTIVE;
5478 /* Update permissions, they may differ for inactive nodes */
5479 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5480 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL,
5481 &tighten_restrictions, NULL);
5482 assert(tighten_restrictions == false);
5483 if (ret < 0) {
5484 /* We only tried to loosen restrictions, so errors are not fatal */
5485 bdrv_abort_perm_update(bs);
5486 } else {
5487 bdrv_set_perm(bs, perm, shared_perm);
5491 /* Recursively inactivate children */
5492 QLIST_FOREACH(child, &bs->children, next) {
5493 ret = bdrv_inactivate_recurse(child->bs);
5494 if (ret < 0) {
5495 return ret;
5499 return 0;
5502 int bdrv_inactivate_all(void)
5504 BlockDriverState *bs = NULL;
5505 BdrvNextIterator it;
5506 int ret = 0;
5507 GSList *aio_ctxs = NULL, *ctx;
5509 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5510 AioContext *aio_context = bdrv_get_aio_context(bs);
5512 if (!g_slist_find(aio_ctxs, aio_context)) {
5513 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5514 aio_context_acquire(aio_context);
5518 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5519 /* Nodes with BDS parents are covered by recursion from the last
5520 * parent that gets inactivated. Don't inactivate them a second
5521 * time if that has already happened. */
5522 if (bdrv_has_bds_parent(bs, false)) {
5523 continue;
5525 ret = bdrv_inactivate_recurse(bs);
5526 if (ret < 0) {
5527 bdrv_next_cleanup(&it);
5528 goto out;
5532 out:
5533 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5534 AioContext *aio_context = ctx->data;
5535 aio_context_release(aio_context);
5537 g_slist_free(aio_ctxs);
5539 return ret;
5542 /**************************************************************/
5543 /* removable device support */
5546 * Return TRUE if the media is present
5548 bool bdrv_is_inserted(BlockDriverState *bs)
5550 BlockDriver *drv = bs->drv;
5551 BdrvChild *child;
5553 if (!drv) {
5554 return false;
5556 if (drv->bdrv_is_inserted) {
5557 return drv->bdrv_is_inserted(bs);
5559 QLIST_FOREACH(child, &bs->children, next) {
5560 if (!bdrv_is_inserted(child->bs)) {
5561 return false;
5564 return true;
5568 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5570 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5572 BlockDriver *drv = bs->drv;
5574 if (drv && drv->bdrv_eject) {
5575 drv->bdrv_eject(bs, eject_flag);
5580 * Lock or unlock the media (if it is locked, the user won't be able
5581 * to eject it manually).
5583 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5585 BlockDriver *drv = bs->drv;
5587 trace_bdrv_lock_medium(bs, locked);
5589 if (drv && drv->bdrv_lock_medium) {
5590 drv->bdrv_lock_medium(bs, locked);
5594 /* Get a reference to bs */
5595 void bdrv_ref(BlockDriverState *bs)
5597 bs->refcnt++;
5600 /* Release a previously grabbed reference to bs.
5601 * If after releasing, reference count is zero, the BlockDriverState is
5602 * deleted. */
5603 void bdrv_unref(BlockDriverState *bs)
5605 if (!bs) {
5606 return;
5608 assert(bs->refcnt > 0);
5609 if (--bs->refcnt == 0) {
5610 bdrv_delete(bs);
5614 struct BdrvOpBlocker {
5615 Error *reason;
5616 QLIST_ENTRY(BdrvOpBlocker) list;
5619 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5621 BdrvOpBlocker *blocker;
5622 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5623 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5624 blocker = QLIST_FIRST(&bs->op_blockers[op]);
5625 error_propagate_prepend(errp, error_copy(blocker->reason),
5626 "Node '%s' is busy: ",
5627 bdrv_get_device_or_node_name(bs));
5628 return true;
5630 return false;
5633 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5635 BdrvOpBlocker *blocker;
5636 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5638 blocker = g_new0(BdrvOpBlocker, 1);
5639 blocker->reason = reason;
5640 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5643 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5645 BdrvOpBlocker *blocker, *next;
5646 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5647 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5648 if (blocker->reason == reason) {
5649 QLIST_REMOVE(blocker, list);
5650 g_free(blocker);
5655 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5657 int i;
5658 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5659 bdrv_op_block(bs, i, reason);
5663 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5665 int i;
5666 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5667 bdrv_op_unblock(bs, i, reason);
5671 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5673 int i;
5675 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5676 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5677 return false;
5680 return true;
5683 void bdrv_img_create(const char *filename, const char *fmt,
5684 const char *base_filename, const char *base_fmt,
5685 char *options, uint64_t img_size, int flags, bool quiet,
5686 Error **errp)
5688 QemuOptsList *create_opts = NULL;
5689 QemuOpts *opts = NULL;
5690 const char *backing_fmt, *backing_file;
5691 int64_t size;
5692 BlockDriver *drv, *proto_drv;
5693 Error *local_err = NULL;
5694 int ret = 0;
5696 /* Find driver and parse its options */
5697 drv = bdrv_find_format(fmt);
5698 if (!drv) {
5699 error_setg(errp, "Unknown file format '%s'", fmt);
5700 return;
5703 proto_drv = bdrv_find_protocol(filename, true, errp);
5704 if (!proto_drv) {
5705 return;
5708 if (!drv->create_opts) {
5709 error_setg(errp, "Format driver '%s' does not support image creation",
5710 drv->format_name);
5711 return;
5714 if (!proto_drv->create_opts) {
5715 error_setg(errp, "Protocol driver '%s' does not support image creation",
5716 proto_drv->format_name);
5717 return;
5720 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5721 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5723 /* Create parameter list with default values */
5724 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5725 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5727 /* Parse -o options */
5728 if (options) {
5729 qemu_opts_do_parse(opts, options, NULL, &local_err);
5730 if (local_err) {
5731 goto out;
5735 if (base_filename) {
5736 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
5737 if (local_err) {
5738 error_setg(errp, "Backing file not supported for file format '%s'",
5739 fmt);
5740 goto out;
5744 if (base_fmt) {
5745 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
5746 if (local_err) {
5747 error_setg(errp, "Backing file format not supported for file "
5748 "format '%s'", fmt);
5749 goto out;
5753 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5754 if (backing_file) {
5755 if (!strcmp(filename, backing_file)) {
5756 error_setg(errp, "Error: Trying to create an image with the "
5757 "same filename as the backing file");
5758 goto out;
5762 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5764 /* The size for the image must always be specified, unless we have a backing
5765 * file and we have not been forbidden from opening it. */
5766 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
5767 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
5768 BlockDriverState *bs;
5769 char *full_backing;
5770 int back_flags;
5771 QDict *backing_options = NULL;
5773 full_backing =
5774 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
5775 &local_err);
5776 if (local_err) {
5777 goto out;
5779 assert(full_backing);
5781 /* backing files always opened read-only */
5782 back_flags = flags;
5783 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5785 backing_options = qdict_new();
5786 if (backing_fmt) {
5787 qdict_put_str(backing_options, "driver", backing_fmt);
5789 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
5791 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
5792 &local_err);
5793 g_free(full_backing);
5794 if (!bs && size != -1) {
5795 /* Couldn't open BS, but we have a size, so it's nonfatal */
5796 warn_reportf_err(local_err,
5797 "Could not verify backing image. "
5798 "This may become an error in future versions.\n");
5799 local_err = NULL;
5800 } else if (!bs) {
5801 /* Couldn't open bs, do not have size */
5802 error_append_hint(&local_err,
5803 "Could not open backing image to determine size.\n");
5804 goto out;
5805 } else {
5806 if (size == -1) {
5807 /* Opened BS, have no size */
5808 size = bdrv_getlength(bs);
5809 if (size < 0) {
5810 error_setg_errno(errp, -size, "Could not get size of '%s'",
5811 backing_file);
5812 bdrv_unref(bs);
5813 goto out;
5815 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
5817 bdrv_unref(bs);
5819 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5821 if (size == -1) {
5822 error_setg(errp, "Image creation needs a size parameter");
5823 goto out;
5826 if (!quiet) {
5827 printf("Formatting '%s', fmt=%s ", filename, fmt);
5828 qemu_opts_print(opts, " ");
5829 puts("");
5832 ret = bdrv_create(drv, filename, opts, &local_err);
5834 if (ret == -EFBIG) {
5835 /* This is generally a better message than whatever the driver would
5836 * deliver (especially because of the cluster_size_hint), since that
5837 * is most probably not much different from "image too large". */
5838 const char *cluster_size_hint = "";
5839 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5840 cluster_size_hint = " (try using a larger cluster size)";
5842 error_setg(errp, "The image size is too large for file format '%s'"
5843 "%s", fmt, cluster_size_hint);
5844 error_free(local_err);
5845 local_err = NULL;
5848 out:
5849 qemu_opts_del(opts);
5850 qemu_opts_free(create_opts);
5851 error_propagate(errp, local_err);
5854 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5856 return bs ? bs->aio_context : qemu_get_aio_context();
5859 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
5861 aio_co_enter(bdrv_get_aio_context(bs), co);
5864 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
5866 QLIST_REMOVE(ban, list);
5867 g_free(ban);
5870 static void bdrv_detach_aio_context(BlockDriverState *bs)
5872 BdrvAioNotifier *baf, *baf_tmp;
5874 assert(!bs->walking_aio_notifiers);
5875 bs->walking_aio_notifiers = true;
5876 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
5877 if (baf->deleted) {
5878 bdrv_do_remove_aio_context_notifier(baf);
5879 } else {
5880 baf->detach_aio_context(baf->opaque);
5883 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5884 * remove remaining aio notifiers if we aren't called again.
5886 bs->walking_aio_notifiers = false;
5888 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
5889 bs->drv->bdrv_detach_aio_context(bs);
5892 if (bs->quiesce_counter) {
5893 aio_enable_external(bs->aio_context);
5895 bs->aio_context = NULL;
5898 static void bdrv_attach_aio_context(BlockDriverState *bs,
5899 AioContext *new_context)
5901 BdrvAioNotifier *ban, *ban_tmp;
5903 if (bs->quiesce_counter) {
5904 aio_disable_external(new_context);
5907 bs->aio_context = new_context;
5909 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
5910 bs->drv->bdrv_attach_aio_context(bs, new_context);
5913 assert(!bs->walking_aio_notifiers);
5914 bs->walking_aio_notifiers = true;
5915 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
5916 if (ban->deleted) {
5917 bdrv_do_remove_aio_context_notifier(ban);
5918 } else {
5919 ban->attached_aio_context(new_context, ban->opaque);
5922 bs->walking_aio_notifiers = false;
5926 * Changes the AioContext used for fd handlers, timers, and BHs by this
5927 * BlockDriverState and all its children and parents.
5929 * Must be called from the main AioContext.
5931 * The caller must own the AioContext lock for the old AioContext of bs, but it
5932 * must not own the AioContext lock for new_context (unless new_context is the
5933 * same as the current context of bs).
5935 * @ignore will accumulate all visited BdrvChild object. The caller is
5936 * responsible for freeing the list afterwards.
5938 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
5939 AioContext *new_context, GSList **ignore)
5941 AioContext *old_context = bdrv_get_aio_context(bs);
5942 BdrvChild *child;
5944 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5946 if (old_context == new_context) {
5947 return;
5950 bdrv_drained_begin(bs);
5952 QLIST_FOREACH(child, &bs->children, next) {
5953 if (g_slist_find(*ignore, child)) {
5954 continue;
5956 *ignore = g_slist_prepend(*ignore, child);
5957 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
5959 QLIST_FOREACH(child, &bs->parents, next_parent) {
5960 if (g_slist_find(*ignore, child)) {
5961 continue;
5963 assert(child->role->set_aio_ctx);
5964 *ignore = g_slist_prepend(*ignore, child);
5965 child->role->set_aio_ctx(child, new_context, ignore);
5968 bdrv_detach_aio_context(bs);
5970 /* Acquire the new context, if necessary */
5971 if (qemu_get_aio_context() != new_context) {
5972 aio_context_acquire(new_context);
5975 bdrv_attach_aio_context(bs, new_context);
5978 * If this function was recursively called from
5979 * bdrv_set_aio_context_ignore(), there may be nodes in the
5980 * subtree that have not yet been moved to the new AioContext.
5981 * Release the old one so bdrv_drained_end() can poll them.
5983 if (qemu_get_aio_context() != old_context) {
5984 aio_context_release(old_context);
5987 bdrv_drained_end(bs);
5989 if (qemu_get_aio_context() != old_context) {
5990 aio_context_acquire(old_context);
5992 if (qemu_get_aio_context() != new_context) {
5993 aio_context_release(new_context);
5997 static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
5998 GSList **ignore, Error **errp)
6000 if (g_slist_find(*ignore, c)) {
6001 return true;
6003 *ignore = g_slist_prepend(*ignore, c);
6005 /* A BdrvChildRole that doesn't handle AioContext changes cannot
6006 * tolerate any AioContext changes */
6007 if (!c->role->can_set_aio_ctx) {
6008 char *user = bdrv_child_user_desc(c);
6009 error_setg(errp, "Changing iothreads is not supported by %s", user);
6010 g_free(user);
6011 return false;
6013 if (!c->role->can_set_aio_ctx(c, ctx, ignore, errp)) {
6014 assert(!errp || *errp);
6015 return false;
6017 return true;
6020 bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6021 GSList **ignore, Error **errp)
6023 if (g_slist_find(*ignore, c)) {
6024 return true;
6026 *ignore = g_slist_prepend(*ignore, c);
6027 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6030 /* @ignore will accumulate all visited BdrvChild object. The caller is
6031 * responsible for freeing the list afterwards. */
6032 bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6033 GSList **ignore, Error **errp)
6035 BdrvChild *c;
6037 if (bdrv_get_aio_context(bs) == ctx) {
6038 return true;
6041 QLIST_FOREACH(c, &bs->parents, next_parent) {
6042 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
6043 return false;
6046 QLIST_FOREACH(c, &bs->children, next) {
6047 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
6048 return false;
6052 return true;
6055 int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6056 BdrvChild *ignore_child, Error **errp)
6058 GSList *ignore;
6059 bool ret;
6061 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6062 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
6063 g_slist_free(ignore);
6065 if (!ret) {
6066 return -EPERM;
6069 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6070 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
6071 g_slist_free(ignore);
6073 return 0;
6076 int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6077 Error **errp)
6079 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
6082 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
6083 void (*attached_aio_context)(AioContext *new_context, void *opaque),
6084 void (*detach_aio_context)(void *opaque), void *opaque)
6086 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
6087 *ban = (BdrvAioNotifier){
6088 .attached_aio_context = attached_aio_context,
6089 .detach_aio_context = detach_aio_context,
6090 .opaque = opaque
6093 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
6096 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
6097 void (*attached_aio_context)(AioContext *,
6098 void *),
6099 void (*detach_aio_context)(void *),
6100 void *opaque)
6102 BdrvAioNotifier *ban, *ban_next;
6104 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
6105 if (ban->attached_aio_context == attached_aio_context &&
6106 ban->detach_aio_context == detach_aio_context &&
6107 ban->opaque == opaque &&
6108 ban->deleted == false)
6110 if (bs->walking_aio_notifiers) {
6111 ban->deleted = true;
6112 } else {
6113 bdrv_do_remove_aio_context_notifier(ban);
6115 return;
6119 abort();
6122 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
6123 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
6124 Error **errp)
6126 if (!bs->drv) {
6127 error_setg(errp, "Node is ejected");
6128 return -ENOMEDIUM;
6130 if (!bs->drv->bdrv_amend_options) {
6131 error_setg(errp, "Block driver '%s' does not support option amendment",
6132 bs->drv->format_name);
6133 return -ENOTSUP;
6135 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
6138 /* This function will be called by the bdrv_recurse_is_first_non_filter method
6139 * of block filter and by bdrv_is_first_non_filter.
6140 * It is used to test if the given bs is the candidate or recurse more in the
6141 * node graph.
6143 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
6144 BlockDriverState *candidate)
6146 /* return false if basic checks fails */
6147 if (!bs || !bs->drv) {
6148 return false;
6151 /* the code reached a non block filter driver -> check if the bs is
6152 * the same as the candidate. It's the recursion termination condition.
6154 if (!bs->drv->is_filter) {
6155 return bs == candidate;
6157 /* Down this path the driver is a block filter driver */
6159 /* If the block filter recursion method is defined use it to recurse down
6160 * the node graph.
6162 if (bs->drv->bdrv_recurse_is_first_non_filter) {
6163 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
6166 /* the driver is a block filter but don't allow to recurse -> return false
6168 return false;
6171 /* This function checks if the candidate is the first non filter bs down it's
6172 * bs chain. Since we don't have pointers to parents it explore all bs chains
6173 * from the top. Some filters can choose not to pass down the recursion.
6175 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
6177 BlockDriverState *bs;
6178 BdrvNextIterator it;
6180 /* walk down the bs forest recursively */
6181 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6182 bool perm;
6184 /* try to recurse in this top level bs */
6185 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
6187 /* candidate is the first non filter */
6188 if (perm) {
6189 bdrv_next_cleanup(&it);
6190 return true;
6194 return false;
6197 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
6198 const char *node_name, Error **errp)
6200 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
6201 AioContext *aio_context;
6203 if (!to_replace_bs) {
6204 error_setg(errp, "Node name '%s' not found", node_name);
6205 return NULL;
6208 aio_context = bdrv_get_aio_context(to_replace_bs);
6209 aio_context_acquire(aio_context);
6211 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
6212 to_replace_bs = NULL;
6213 goto out;
6216 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6217 * most non filter in order to prevent data corruption.
6218 * Another benefit is that this tests exclude backing files which are
6219 * blocked by the backing blockers.
6221 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
6222 error_setg(errp, "Only top most non filter can be replaced");
6223 to_replace_bs = NULL;
6224 goto out;
6227 out:
6228 aio_context_release(aio_context);
6229 return to_replace_bs;
6233 * Iterates through the list of runtime option keys that are said to
6234 * be "strong" for a BDS. An option is called "strong" if it changes
6235 * a BDS's data. For example, the null block driver's "size" and
6236 * "read-zeroes" options are strong, but its "latency-ns" option is
6237 * not.
6239 * If a key returned by this function ends with a dot, all options
6240 * starting with that prefix are strong.
6242 static const char *const *strong_options(BlockDriverState *bs,
6243 const char *const *curopt)
6245 static const char *const global_options[] = {
6246 "driver", "filename", NULL
6249 if (!curopt) {
6250 return &global_options[0];
6253 curopt++;
6254 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
6255 curopt = bs->drv->strong_runtime_opts;
6258 return (curopt && *curopt) ? curopt : NULL;
6262 * Copies all strong runtime options from bs->options to the given
6263 * QDict. The set of strong option keys is determined by invoking
6264 * strong_options().
6266 * Returns true iff any strong option was present in bs->options (and
6267 * thus copied to the target QDict) with the exception of "filename"
6268 * and "driver". The caller is expected to use this value to decide
6269 * whether the existence of strong options prevents the generation of
6270 * a plain filename.
6272 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
6274 bool found_any = false;
6275 const char *const *option_name = NULL;
6277 if (!bs->drv) {
6278 return false;
6281 while ((option_name = strong_options(bs, option_name))) {
6282 bool option_given = false;
6284 assert(strlen(*option_name) > 0);
6285 if ((*option_name)[strlen(*option_name) - 1] != '.') {
6286 QObject *entry = qdict_get(bs->options, *option_name);
6287 if (!entry) {
6288 continue;
6291 qdict_put_obj(d, *option_name, qobject_ref(entry));
6292 option_given = true;
6293 } else {
6294 const QDictEntry *entry;
6295 for (entry = qdict_first(bs->options); entry;
6296 entry = qdict_next(bs->options, entry))
6298 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
6299 qdict_put_obj(d, qdict_entry_key(entry),
6300 qobject_ref(qdict_entry_value(entry)));
6301 option_given = true;
6306 /* While "driver" and "filename" need to be included in a JSON filename,
6307 * their existence does not prohibit generation of a plain filename. */
6308 if (!found_any && option_given &&
6309 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
6311 found_any = true;
6315 if (!qdict_haskey(d, "driver")) {
6316 /* Drivers created with bdrv_new_open_driver() may not have a
6317 * @driver option. Add it here. */
6318 qdict_put_str(d, "driver", bs->drv->format_name);
6321 return found_any;
6324 /* Note: This function may return false positives; it may return true
6325 * even if opening the backing file specified by bs's image header
6326 * would result in exactly bs->backing. */
6327 static bool bdrv_backing_overridden(BlockDriverState *bs)
6329 if (bs->backing) {
6330 return strcmp(bs->auto_backing_file,
6331 bs->backing->bs->filename);
6332 } else {
6333 /* No backing BDS, so if the image header reports any backing
6334 * file, it must have been suppressed */
6335 return bs->auto_backing_file[0] != '\0';
6339 /* Updates the following BDS fields:
6340 * - exact_filename: A filename which may be used for opening a block device
6341 * which (mostly) equals the given BDS (even without any
6342 * other options; so reading and writing must return the same
6343 * results, but caching etc. may be different)
6344 * - full_open_options: Options which, when given when opening a block device
6345 * (without a filename), result in a BDS (mostly)
6346 * equalling the given one
6347 * - filename: If exact_filename is set, it is copied here. Otherwise,
6348 * full_open_options is converted to a JSON object, prefixed with
6349 * "json:" (for use through the JSON pseudo protocol) and put here.
6351 void bdrv_refresh_filename(BlockDriverState *bs)
6353 BlockDriver *drv = bs->drv;
6354 BdrvChild *child;
6355 QDict *opts;
6356 bool backing_overridden;
6357 bool generate_json_filename; /* Whether our default implementation should
6358 fill exact_filename (false) or not (true) */
6360 if (!drv) {
6361 return;
6364 /* This BDS's file name may depend on any of its children's file names, so
6365 * refresh those first */
6366 QLIST_FOREACH(child, &bs->children, next) {
6367 bdrv_refresh_filename(child->bs);
6370 if (bs->implicit) {
6371 /* For implicit nodes, just copy everything from the single child */
6372 child = QLIST_FIRST(&bs->children);
6373 assert(QLIST_NEXT(child, next) == NULL);
6375 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
6376 child->bs->exact_filename);
6377 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
6379 bs->full_open_options = qobject_ref(child->bs->full_open_options);
6381 return;
6384 backing_overridden = bdrv_backing_overridden(bs);
6386 if (bs->open_flags & BDRV_O_NO_IO) {
6387 /* Without I/O, the backing file does not change anything.
6388 * Therefore, in such a case (primarily qemu-img), we can
6389 * pretend the backing file has not been overridden even if
6390 * it technically has been. */
6391 backing_overridden = false;
6394 /* Gather the options QDict */
6395 opts = qdict_new();
6396 generate_json_filename = append_strong_runtime_options(opts, bs);
6397 generate_json_filename |= backing_overridden;
6399 if (drv->bdrv_gather_child_options) {
6400 /* Some block drivers may not want to present all of their children's
6401 * options, or name them differently from BdrvChild.name */
6402 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
6403 } else {
6404 QLIST_FOREACH(child, &bs->children, next) {
6405 if (child->role == &child_backing && !backing_overridden) {
6406 /* We can skip the backing BDS if it has not been overridden */
6407 continue;
6410 qdict_put(opts, child->name,
6411 qobject_ref(child->bs->full_open_options));
6414 if (backing_overridden && !bs->backing) {
6415 /* Force no backing file */
6416 qdict_put_null(opts, "backing");
6420 qobject_unref(bs->full_open_options);
6421 bs->full_open_options = opts;
6423 if (drv->bdrv_refresh_filename) {
6424 /* Obsolete information is of no use here, so drop the old file name
6425 * information before refreshing it */
6426 bs->exact_filename[0] = '\0';
6428 drv->bdrv_refresh_filename(bs);
6429 } else if (bs->file) {
6430 /* Try to reconstruct valid information from the underlying file */
6432 bs->exact_filename[0] = '\0';
6435 * We can use the underlying file's filename if:
6436 * - it has a filename,
6437 * - the file is a protocol BDS, and
6438 * - opening that file (as this BDS's format) will automatically create
6439 * the BDS tree we have right now, that is:
6440 * - the user did not significantly change this BDS's behavior with
6441 * some explicit (strong) options
6442 * - no non-file child of this BDS has been overridden by the user
6443 * Both of these conditions are represented by generate_json_filename.
6445 if (bs->file->bs->exact_filename[0] &&
6446 bs->file->bs->drv->bdrv_file_open &&
6447 !generate_json_filename)
6449 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
6453 if (bs->exact_filename[0]) {
6454 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
6455 } else {
6456 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
6457 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
6458 qstring_get_str(json));
6459 qobject_unref(json);
6463 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
6465 BlockDriver *drv = bs->drv;
6467 if (!drv) {
6468 error_setg(errp, "Node '%s' is ejected", bs->node_name);
6469 return NULL;
6472 if (drv->bdrv_dirname) {
6473 return drv->bdrv_dirname(bs, errp);
6476 if (bs->file) {
6477 return bdrv_dirname(bs->file->bs, errp);
6480 bdrv_refresh_filename(bs);
6481 if (bs->exact_filename[0] != '\0') {
6482 return path_combine(bs->exact_filename, "");
6485 error_setg(errp, "Cannot generate a base directory for %s nodes",
6486 drv->format_name);
6487 return NULL;
6491 * Hot add/remove a BDS's child. So the user can take a child offline when
6492 * it is broken and take a new child online
6494 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
6495 Error **errp)
6498 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
6499 error_setg(errp, "The node %s does not support adding a child",
6500 bdrv_get_device_or_node_name(parent_bs));
6501 return;
6504 if (!QLIST_EMPTY(&child_bs->parents)) {
6505 error_setg(errp, "The node %s already has a parent",
6506 child_bs->node_name);
6507 return;
6510 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
6513 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
6515 BdrvChild *tmp;
6517 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
6518 error_setg(errp, "The node %s does not support removing a child",
6519 bdrv_get_device_or_node_name(parent_bs));
6520 return;
6523 QLIST_FOREACH(tmp, &parent_bs->children, next) {
6524 if (tmp == child) {
6525 break;
6529 if (!tmp) {
6530 error_setg(errp, "The node %s does not have a child named %s",
6531 bdrv_get_device_or_node_name(parent_bs),
6532 bdrv_get_device_or_node_name(child->bs));
6533 return;
6536 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
6539 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
6540 uint32_t granularity, Error **errp)
6542 BlockDriver *drv = bs->drv;
6544 if (!drv) {
6545 error_setg_errno(errp, ENOMEDIUM,
6546 "Can't store persistent bitmaps to %s",
6547 bdrv_get_device_or_node_name(bs));
6548 return false;
6551 if (!drv->bdrv_can_store_new_dirty_bitmap) {
6552 error_setg_errno(errp, ENOTSUP,
6553 "Can't store persistent bitmaps to %s",
6554 bdrv_get_device_or_node_name(bs));
6555 return false;
6558 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);