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
24 #include "config-host.h"
25 #include "qemu-common.h"
27 #include "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "qmp-commands.h"
36 #include "qemu/timer.h"
39 #include <sys/types.h>
41 #include <sys/ioctl.h>
42 #include <sys/queue.h>
52 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
55 BDRV_REQ_COPY_ON_READ = 0x1,
56 BDRV_REQ_ZERO_WRITE = 0x2,
59 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
60 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
61 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
62 BlockDriverCompletionFunc *cb, void *opaque);
63 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
64 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
65 BlockDriverCompletionFunc *cb, void *opaque);
66 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
67 int64_t sector_num, int nb_sectors,
69 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors,
72 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
74 BdrvRequestFlags flags);
75 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
76 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
77 BdrvRequestFlags flags);
78 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
82 BlockDriverCompletionFunc *cb,
85 static void coroutine_fn bdrv_co_do_rw(void *opaque);
86 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
87 int64_t sector_num, int nb_sectors);
89 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
90 QTAILQ_HEAD_INITIALIZER(bdrv_states);
92 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
93 QLIST_HEAD_INITIALIZER(bdrv_drivers);
95 /* If non-zero, use only whitelisted block drivers */
96 static int use_bdrv_whitelist;
99 static int is_windows_drive_prefix(const char *filename)
101 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
102 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
106 int is_windows_drive(const char *filename)
108 if (is_windows_drive_prefix(filename) &&
111 if (strstart(filename, "\\\\.\\", NULL) ||
112 strstart(filename, "//./", NULL))
118 /* throttling disk I/O limits */
119 void bdrv_set_io_limits(BlockDriverState *bs,
124 throttle_config(&bs->throttle_state, cfg);
126 for (i = 0; i < 2; i++) {
127 qemu_co_enter_next(&bs->throttled_reqs[i]);
131 /* this function drain all the throttled IOs */
132 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
134 bool drained = false;
135 bool enabled = bs->io_limits_enabled;
138 bs->io_limits_enabled = false;
140 for (i = 0; i < 2; i++) {
141 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
146 bs->io_limits_enabled = enabled;
151 void bdrv_io_limits_disable(BlockDriverState *bs)
153 bs->io_limits_enabled = false;
155 bdrv_start_throttled_reqs(bs);
157 throttle_destroy(&bs->throttle_state);
160 static void bdrv_throttle_read_timer_cb(void *opaque)
162 BlockDriverState *bs = opaque;
163 qemu_co_enter_next(&bs->throttled_reqs[0]);
166 static void bdrv_throttle_write_timer_cb(void *opaque)
168 BlockDriverState *bs = opaque;
169 qemu_co_enter_next(&bs->throttled_reqs[1]);
172 /* should be called before bdrv_set_io_limits if a limit is set */
173 void bdrv_io_limits_enable(BlockDriverState *bs)
175 assert(!bs->io_limits_enabled);
176 throttle_init(&bs->throttle_state,
178 bdrv_throttle_read_timer_cb,
179 bdrv_throttle_write_timer_cb,
181 bs->io_limits_enabled = true;
184 /* This function makes an IO wait if needed
186 * @nb_sectors: the number of sectors of the IO
187 * @is_write: is the IO a write
189 static void bdrv_io_limits_intercept(BlockDriverState *bs,
193 /* does this io must wait */
194 bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
196 /* if must wait or any request of this type throttled queue the IO */
198 !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
199 qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
202 /* the IO will be executed, do the accounting */
203 throttle_account(&bs->throttle_state,
205 nb_sectors * BDRV_SECTOR_SIZE);
207 /* if the next request must wait -> do nothing */
208 if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
212 /* else queue next request for execution */
213 qemu_co_queue_next(&bs->throttled_reqs[is_write]);
216 /* check if the path starts with "<protocol>:" */
217 static int path_has_protocol(const char *path)
222 if (is_windows_drive(path) ||
223 is_windows_drive_prefix(path)) {
226 p = path + strcspn(path, ":/\\");
228 p = path + strcspn(path, ":/");
234 int path_is_absolute(const char *path)
237 /* specific case for names like: "\\.\d:" */
238 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
241 return (*path == '/' || *path == '\\');
243 return (*path == '/');
247 /* if filename is absolute, just copy it to dest. Otherwise, build a
248 path to it by considering it is relative to base_path. URL are
250 void path_combine(char *dest, int dest_size,
251 const char *base_path,
252 const char *filename)
259 if (path_is_absolute(filename)) {
260 pstrcpy(dest, dest_size, filename);
262 p = strchr(base_path, ':');
267 p1 = strrchr(base_path, '/');
271 p2 = strrchr(base_path, '\\');
283 if (len > dest_size - 1)
285 memcpy(dest, base_path, len);
287 pstrcat(dest, dest_size, filename);
291 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
293 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
294 pstrcpy(dest, sz, bs->backing_file);
296 path_combine(dest, sz, bs->filename, bs->backing_file);
300 void bdrv_register(BlockDriver *bdrv)
302 /* Block drivers without coroutine functions need emulation */
303 if (!bdrv->bdrv_co_readv) {
304 bdrv->bdrv_co_readv = bdrv_co_readv_em;
305 bdrv->bdrv_co_writev = bdrv_co_writev_em;
307 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
308 * the block driver lacks aio we need to emulate that too.
310 if (!bdrv->bdrv_aio_readv) {
311 /* add AIO emulation layer */
312 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
313 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
317 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
320 /* create a new block device (by default it is empty) */
321 BlockDriverState *bdrv_new(const char *device_name)
323 BlockDriverState *bs;
325 bs = g_malloc0(sizeof(BlockDriverState));
326 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
327 if (device_name[0] != '\0') {
328 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
330 bdrv_iostatus_disable(bs);
331 notifier_list_init(&bs->close_notifiers);
332 notifier_with_return_list_init(&bs->before_write_notifiers);
333 qemu_co_queue_init(&bs->throttled_reqs[0]);
334 qemu_co_queue_init(&bs->throttled_reqs[1]);
340 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
342 notifier_list_add(&bs->close_notifiers, notify);
345 BlockDriver *bdrv_find_format(const char *format_name)
348 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
349 if (!strcmp(drv1->format_name, format_name)) {
356 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
358 static const char *whitelist_rw[] = {
359 CONFIG_BDRV_RW_WHITELIST
361 static const char *whitelist_ro[] = {
362 CONFIG_BDRV_RO_WHITELIST
366 if (!whitelist_rw[0] && !whitelist_ro[0]) {
367 return 1; /* no whitelist, anything goes */
370 for (p = whitelist_rw; *p; p++) {
371 if (!strcmp(drv->format_name, *p)) {
376 for (p = whitelist_ro; *p; p++) {
377 if (!strcmp(drv->format_name, *p)) {
385 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
388 BlockDriver *drv = bdrv_find_format(format_name);
389 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
392 typedef struct CreateCo {
395 QEMUOptionParameter *options;
400 static void coroutine_fn bdrv_create_co_entry(void *opaque)
402 Error *local_err = NULL;
405 CreateCo *cco = opaque;
408 ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err);
409 if (error_is_set(&local_err)) {
410 error_propagate(&cco->err, local_err);
415 int bdrv_create(BlockDriver *drv, const char* filename,
416 QEMUOptionParameter *options, Error **errp)
423 .filename = g_strdup(filename),
429 if (!drv->bdrv_create) {
430 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
435 if (qemu_in_coroutine()) {
436 /* Fast-path if already in coroutine context */
437 bdrv_create_co_entry(&cco);
439 co = qemu_coroutine_create(bdrv_create_co_entry);
440 qemu_coroutine_enter(co, &cco);
441 while (cco.ret == NOT_DONE) {
448 if (error_is_set(&cco.err)) {
449 error_propagate(errp, cco.err);
451 error_setg_errno(errp, -ret, "Could not create image");
456 g_free(cco.filename);
460 int bdrv_create_file(const char* filename, QEMUOptionParameter *options,
464 Error *local_err = NULL;
467 drv = bdrv_find_protocol(filename, true);
469 error_setg(errp, "Could not find protocol for file '%s'", filename);
473 ret = bdrv_create(drv, filename, options, &local_err);
474 if (error_is_set(&local_err)) {
475 error_propagate(errp, local_err);
481 * Create a uniquely-named empty temporary file.
482 * Return 0 upon success, otherwise a negative errno value.
484 int get_tmp_filename(char *filename, int size)
487 char temp_dir[MAX_PATH];
488 /* GetTempFileName requires that its output buffer (4th param)
489 have length MAX_PATH or greater. */
490 assert(size >= MAX_PATH);
491 return (GetTempPath(MAX_PATH, temp_dir)
492 && GetTempFileName(temp_dir, "qem", 0, filename)
493 ? 0 : -GetLastError());
497 tmpdir = getenv("TMPDIR");
500 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
503 fd = mkstemp(filename);
507 if (close(fd) != 0) {
516 * Detect host devices. By convention, /dev/cdrom[N] is always
517 * recognized as a host CDROM.
519 static BlockDriver *find_hdev_driver(const char *filename)
521 int score_max = 0, score;
522 BlockDriver *drv = NULL, *d;
524 QLIST_FOREACH(d, &bdrv_drivers, list) {
525 if (d->bdrv_probe_device) {
526 score = d->bdrv_probe_device(filename);
527 if (score > score_max) {
537 BlockDriver *bdrv_find_protocol(const char *filename,
538 bool allow_protocol_prefix)
545 /* TODO Drivers without bdrv_file_open must be specified explicitly */
548 * XXX(hch): we really should not let host device detection
549 * override an explicit protocol specification, but moving this
550 * later breaks access to device names with colons in them.
551 * Thanks to the brain-dead persistent naming schemes on udev-
552 * based Linux systems those actually are quite common.
554 drv1 = find_hdev_driver(filename);
559 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
560 return bdrv_find_format("file");
563 p = strchr(filename, ':');
566 if (len > sizeof(protocol) - 1)
567 len = sizeof(protocol) - 1;
568 memcpy(protocol, filename, len);
569 protocol[len] = '\0';
570 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
571 if (drv1->protocol_name &&
572 !strcmp(drv1->protocol_name, protocol)) {
579 static int find_image_format(BlockDriverState *bs, const char *filename,
580 BlockDriver **pdrv, Error **errp)
582 int score, score_max;
583 BlockDriver *drv1, *drv;
587 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
588 if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
589 drv = bdrv_find_format("raw");
591 error_setg(errp, "Could not find raw image format");
598 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
600 error_setg_errno(errp, -ret, "Could not read image for determining its "
608 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
609 if (drv1->bdrv_probe) {
610 score = drv1->bdrv_probe(buf, ret, filename);
611 if (score > score_max) {
618 error_setg(errp, "Could not determine image format: No compatible "
627 * Set the current 'total_sectors' value
629 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
631 BlockDriver *drv = bs->drv;
633 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
637 /* query actual device if possible, otherwise just trust the hint */
638 if (drv->bdrv_getlength) {
639 int64_t length = drv->bdrv_getlength(bs);
643 hint = length >> BDRV_SECTOR_BITS;
646 bs->total_sectors = hint;
651 * Set open flags for a given discard mode
653 * Return 0 on success, -1 if the discard mode was invalid.
655 int bdrv_parse_discard_flags(const char *mode, int *flags)
657 *flags &= ~BDRV_O_UNMAP;
659 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
661 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
662 *flags |= BDRV_O_UNMAP;
671 * Set open flags for a given cache mode
673 * Return 0 on success, -1 if the cache mode was invalid.
675 int bdrv_parse_cache_flags(const char *mode, int *flags)
677 *flags &= ~BDRV_O_CACHE_MASK;
679 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
680 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
681 } else if (!strcmp(mode, "directsync")) {
682 *flags |= BDRV_O_NOCACHE;
683 } else if (!strcmp(mode, "writeback")) {
684 *flags |= BDRV_O_CACHE_WB;
685 } else if (!strcmp(mode, "unsafe")) {
686 *flags |= BDRV_O_CACHE_WB;
687 *flags |= BDRV_O_NO_FLUSH;
688 } else if (!strcmp(mode, "writethrough")) {
689 /* this is the default */
698 * The copy-on-read flag is actually a reference count so multiple users may
699 * use the feature without worrying about clobbering its previous state.
700 * Copy-on-read stays enabled until all users have called to disable it.
702 void bdrv_enable_copy_on_read(BlockDriverState *bs)
707 void bdrv_disable_copy_on_read(BlockDriverState *bs)
709 assert(bs->copy_on_read > 0);
713 static int bdrv_open_flags(BlockDriverState *bs, int flags)
715 int open_flags = flags | BDRV_O_CACHE_WB;
718 * Clear flags that are internal to the block layer before opening the
721 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
724 * Snapshots should be writable.
726 if (bs->is_temporary) {
727 open_flags |= BDRV_O_RDWR;
734 * Common part for opening disk images and files
736 * Removes all processed options from *options.
738 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
739 QDict *options, int flags, BlockDriver *drv, Error **errp)
742 const char *filename;
743 Error *local_err = NULL;
746 assert(bs->file == NULL);
747 assert(options != NULL && bs->options != options);
750 filename = file->filename;
752 filename = qdict_get_try_str(options, "filename");
755 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
757 /* bdrv_open() with directly using a protocol as drv. This layer is already
758 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
759 * and return immediately. */
760 if (file != NULL && drv->bdrv_file_open) {
765 bs->open_flags = flags;
766 bs->buffer_alignment = 512;
767 bs->zero_beyond_eof = true;
768 open_flags = bdrv_open_flags(bs, flags);
769 bs->read_only = !(open_flags & BDRV_O_RDWR);
771 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
772 error_setg(errp, "Driver '%s' is not whitelisted", drv->format_name);
776 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
777 if (!bs->read_only && (flags & BDRV_O_COPY_ON_READ)) {
778 bdrv_enable_copy_on_read(bs);
781 if (filename != NULL) {
782 pstrcpy(bs->filename, sizeof(bs->filename), filename);
784 bs->filename[0] = '\0';
788 bs->opaque = g_malloc0(drv->instance_size);
790 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
792 /* Open the image, either directly or using a protocol */
793 if (drv->bdrv_file_open) {
794 assert(file == NULL);
795 assert(!drv->bdrv_needs_filename || filename != NULL);
796 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
799 error_setg(errp, "Can't use '%s' as a block driver for the "
800 "protocol level", drv->format_name);
805 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
809 if (error_is_set(&local_err)) {
810 error_propagate(errp, local_err);
811 } else if (bs->filename[0]) {
812 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
814 error_setg_errno(errp, -ret, "Could not open image");
819 ret = refresh_total_sectors(bs, bs->total_sectors);
821 error_setg_errno(errp, -ret, "Could not refresh total sector count");
826 if (bs->is_temporary) {
827 assert(bs->filename[0] != '\0');
828 unlink(bs->filename);
842 * Opens a file using a protocol (file, host_device, nbd, ...)
844 * options is a QDict of options to pass to the block drivers, or NULL for an
845 * empty set of options. The reference to the QDict belongs to the block layer
846 * after the call (even on failure), so if the caller intends to reuse the
847 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
849 int bdrv_file_open(BlockDriverState **pbs, const char *filename,
850 QDict *options, int flags, Error **errp)
852 BlockDriverState *bs;
855 bool allow_protocol_prefix = false;
856 Error *local_err = NULL;
859 /* NULL means an empty set of options */
860 if (options == NULL) {
861 options = qdict_new();
865 bs->options = options;
866 options = qdict_clone_shallow(options);
868 /* Fetch the file name from the options QDict if necessary */
870 filename = qdict_get_try_str(options, "filename");
871 } else if (filename && !qdict_haskey(options, "filename")) {
872 qdict_put(options, "filename", qstring_from_str(filename));
873 allow_protocol_prefix = true;
875 error_setg(errp, "Can't specify 'file' and 'filename' options at the "
881 /* Find the right block driver */
882 drvname = qdict_get_try_str(options, "driver");
884 drv = bdrv_find_whitelisted_format(drvname, !(flags & BDRV_O_RDWR));
886 error_setg(errp, "Unknown driver '%s'", drvname);
888 qdict_del(options, "driver");
889 } else if (filename) {
890 drv = bdrv_find_protocol(filename, allow_protocol_prefix);
892 error_setg(errp, "Unknown protocol");
895 error_setg(errp, "Must specify either driver or file");
900 /* errp has been set already */
905 /* Parse the filename and open it */
906 if (drv->bdrv_parse_filename && filename) {
907 drv->bdrv_parse_filename(filename, options, &local_err);
908 if (error_is_set(&local_err)) {
909 error_propagate(errp, local_err);
913 qdict_del(options, "filename");
914 } else if (drv->bdrv_needs_filename && !filename) {
915 error_setg(errp, "The '%s' block driver requires a file name",
921 ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err);
923 error_propagate(errp, local_err);
927 /* Check if any unknown options were used */
928 if (qdict_size(options) != 0) {
929 const QDictEntry *entry = qdict_first(options);
930 error_setg(errp, "Block protocol '%s' doesn't support the option '%s'",
931 drv->format_name, entry->key);
944 QDECREF(bs->options);
951 * Opens the backing file for a BlockDriverState if not yet open
953 * options is a QDict of options to pass to the block drivers, or NULL for an
954 * empty set of options. The reference to the QDict is transferred to this
955 * function (even on failure), so if the caller intends to reuse the dictionary,
956 * it needs to use QINCREF() before calling bdrv_file_open.
958 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
960 char backing_filename[PATH_MAX];
962 BlockDriver *back_drv = NULL;
963 Error *local_err = NULL;
965 if (bs->backing_hd != NULL) {
970 /* NULL means an empty set of options */
971 if (options == NULL) {
972 options = qdict_new();
975 bs->open_flags &= ~BDRV_O_NO_BACKING;
976 if (qdict_haskey(options, "file.filename")) {
977 backing_filename[0] = '\0';
978 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
982 bdrv_get_full_backing_filename(bs, backing_filename,
983 sizeof(backing_filename));
986 bs->backing_hd = bdrv_new("");
988 if (bs->backing_format[0] != '\0') {
989 back_drv = bdrv_find_format(bs->backing_format);
992 /* backing files always opened read-only */
993 back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT);
995 ret = bdrv_open(bs->backing_hd,
996 *backing_filename ? backing_filename : NULL, options,
997 back_flags, back_drv, &local_err);
998 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
999 bs->backing_hd->file->filename);
1001 bdrv_unref(bs->backing_hd);
1002 bs->backing_hd = NULL;
1003 bs->open_flags |= BDRV_O_NO_BACKING;
1004 error_propagate(errp, local_err);
1011 * Opens a disk image (raw, qcow2, vmdk, ...)
1013 * options is a QDict of options to pass to the block drivers, or NULL for an
1014 * empty set of options. The reference to the QDict belongs to the block layer
1015 * after the call (even on failure), so if the caller intends to reuse the
1016 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1018 int bdrv_open(BlockDriverState *bs, const char *filename, QDict *options,
1019 int flags, BlockDriver *drv, Error **errp)
1022 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1023 char tmp_filename[PATH_MAX + 1];
1024 BlockDriverState *file = NULL;
1025 QDict *file_options = NULL;
1026 const char *drvname;
1027 Error *local_err = NULL;
1029 /* NULL means an empty set of options */
1030 if (options == NULL) {
1031 options = qdict_new();
1034 bs->options = options;
1035 options = qdict_clone_shallow(options);
1037 /* For snapshot=on, create a temporary qcow2 overlay */
1038 if (flags & BDRV_O_SNAPSHOT) {
1039 BlockDriverState *bs1;
1041 BlockDriver *bdrv_qcow2;
1042 QEMUOptionParameter *create_options;
1043 char backing_filename[PATH_MAX];
1045 if (qdict_size(options) != 0) {
1046 error_setg(errp, "Can't use snapshot=on with driver-specific options");
1050 assert(filename != NULL);
1052 /* if snapshot, we create a temporary backing file and open it
1053 instead of opening 'filename' directly */
1055 /* if there is a backing file, use it */
1057 ret = bdrv_open(bs1, filename, NULL, 0, drv, &local_err);
1062 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
1066 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
1068 error_setg_errno(errp, -ret, "Could not get temporary filename");
1072 /* Real path is meaningless for protocols */
1073 if (path_has_protocol(filename)) {
1074 snprintf(backing_filename, sizeof(backing_filename),
1076 } else if (!realpath(filename, backing_filename)) {
1077 error_setg_errno(errp, errno, "Could not resolve path '%s'", filename);
1082 bdrv_qcow2 = bdrv_find_format("qcow2");
1083 create_options = parse_option_parameters("", bdrv_qcow2->create_options,
1086 set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size);
1087 set_option_parameter(create_options, BLOCK_OPT_BACKING_FILE,
1090 set_option_parameter(create_options, BLOCK_OPT_BACKING_FMT,
1094 ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err);
1095 free_option_parameters(create_options);
1097 error_setg_errno(errp, -ret, "Could not create temporary overlay "
1098 "'%s': %s", tmp_filename,
1099 error_get_pretty(local_err));
1100 error_free(local_err);
1105 filename = tmp_filename;
1107 bs->is_temporary = 1;
1110 /* Open image file without format layer */
1111 if (flags & BDRV_O_RDWR) {
1112 flags |= BDRV_O_ALLOW_RDWR;
1115 qdict_extract_subqdict(options, &file_options, "file.");
1117 ret = bdrv_file_open(&file, filename, file_options,
1118 bdrv_open_flags(bs, flags | BDRV_O_UNMAP), &local_err);
1123 /* Find the right image format driver */
1124 drvname = qdict_get_try_str(options, "driver");
1126 drv = bdrv_find_whitelisted_format(drvname, !(flags & BDRV_O_RDWR));
1127 qdict_del(options, "driver");
1131 ret = find_image_format(file, filename, &drv, &local_err);
1135 goto unlink_and_fail;
1138 /* Open the image */
1139 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1141 goto unlink_and_fail;
1144 if (bs->file != file) {
1149 /* If there is a backing file, use it */
1150 if ((flags & BDRV_O_NO_BACKING) == 0) {
1151 QDict *backing_options;
1153 qdict_extract_subqdict(options, &backing_options, "backing.");
1154 ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1156 goto close_and_fail;
1160 /* Check if any unknown options were used */
1161 if (qdict_size(options) != 0) {
1162 const QDictEntry *entry = qdict_first(options);
1163 error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1164 "support the option '%s'", drv->format_name, bs->device_name,
1168 goto close_and_fail;
1172 if (!bdrv_key_required(bs)) {
1173 bdrv_dev_change_media_cb(bs, true);
1182 if (bs->is_temporary) {
1186 QDECREF(bs->options);
1189 if (error_is_set(&local_err)) {
1190 error_propagate(errp, local_err);
1197 if (error_is_set(&local_err)) {
1198 error_propagate(errp, local_err);
1203 typedef struct BlockReopenQueueEntry {
1205 BDRVReopenState state;
1206 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1207 } BlockReopenQueueEntry;
1210 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1211 * reopen of multiple devices.
1213 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1214 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1215 * be created and initialized. This newly created BlockReopenQueue should be
1216 * passed back in for subsequent calls that are intended to be of the same
1219 * bs is the BlockDriverState to add to the reopen queue.
1221 * flags contains the open flags for the associated bs
1223 * returns a pointer to bs_queue, which is either the newly allocated
1224 * bs_queue, or the existing bs_queue being used.
1227 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1228 BlockDriverState *bs, int flags)
1232 BlockReopenQueueEntry *bs_entry;
1233 if (bs_queue == NULL) {
1234 bs_queue = g_new0(BlockReopenQueue, 1);
1235 QSIMPLEQ_INIT(bs_queue);
1239 bdrv_reopen_queue(bs_queue, bs->file, flags);
1242 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1243 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1245 bs_entry->state.bs = bs;
1246 bs_entry->state.flags = flags;
1252 * Reopen multiple BlockDriverStates atomically & transactionally.
1254 * The queue passed in (bs_queue) must have been built up previous
1255 * via bdrv_reopen_queue().
1257 * Reopens all BDS specified in the queue, with the appropriate
1258 * flags. All devices are prepared for reopen, and failure of any
1259 * device will cause all device changes to be abandonded, and intermediate
1262 * If all devices prepare successfully, then the changes are committed
1266 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1269 BlockReopenQueueEntry *bs_entry, *next;
1270 Error *local_err = NULL;
1272 assert(bs_queue != NULL);
1276 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1277 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1278 error_propagate(errp, local_err);
1281 bs_entry->prepared = true;
1284 /* If we reach this point, we have success and just need to apply the
1287 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1288 bdrv_reopen_commit(&bs_entry->state);
1294 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1295 if (ret && bs_entry->prepared) {
1296 bdrv_reopen_abort(&bs_entry->state);
1305 /* Reopen a single BlockDriverState with the specified flags. */
1306 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1309 Error *local_err = NULL;
1310 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1312 ret = bdrv_reopen_multiple(queue, &local_err);
1313 if (local_err != NULL) {
1314 error_propagate(errp, local_err);
1321 * Prepares a BlockDriverState for reopen. All changes are staged in the
1322 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1323 * the block driver layer .bdrv_reopen_prepare()
1325 * bs is the BlockDriverState to reopen
1326 * flags are the new open flags
1327 * queue is the reopen queue
1329 * Returns 0 on success, non-zero on error. On error errp will be set
1332 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1333 * It is the responsibility of the caller to then call the abort() or
1334 * commit() for any other BDS that have been left in a prepare() state
1337 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1341 Error *local_err = NULL;
1344 assert(reopen_state != NULL);
1345 assert(reopen_state->bs->drv != NULL);
1346 drv = reopen_state->bs->drv;
1348 /* if we are to stay read-only, do not allow permission change
1350 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1351 reopen_state->flags & BDRV_O_RDWR) {
1352 error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1353 reopen_state->bs->device_name);
1358 ret = bdrv_flush(reopen_state->bs);
1360 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1365 if (drv->bdrv_reopen_prepare) {
1366 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1368 if (local_err != NULL) {
1369 error_propagate(errp, local_err);
1371 error_setg(errp, "failed while preparing to reopen image '%s'",
1372 reopen_state->bs->filename);
1377 /* It is currently mandatory to have a bdrv_reopen_prepare()
1378 * handler for each supported drv. */
1379 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1380 drv->format_name, reopen_state->bs->device_name,
1381 "reopening of file");
1393 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1394 * makes them final by swapping the staging BlockDriverState contents into
1395 * the active BlockDriverState contents.
1397 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1401 assert(reopen_state != NULL);
1402 drv = reopen_state->bs->drv;
1403 assert(drv != NULL);
1405 /* If there are any driver level actions to take */
1406 if (drv->bdrv_reopen_commit) {
1407 drv->bdrv_reopen_commit(reopen_state);
1410 /* set BDS specific flags now */
1411 reopen_state->bs->open_flags = reopen_state->flags;
1412 reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1414 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1418 * Abort the reopen, and delete and free the staged changes in
1421 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1425 assert(reopen_state != NULL);
1426 drv = reopen_state->bs->drv;
1427 assert(drv != NULL);
1429 if (drv->bdrv_reopen_abort) {
1430 drv->bdrv_reopen_abort(reopen_state);
1435 void bdrv_close(BlockDriverState *bs)
1438 block_job_cancel_sync(bs->job);
1440 bdrv_drain_all(); /* complete I/O */
1442 bdrv_drain_all(); /* in case flush left pending I/O */
1443 notifier_list_notify(&bs->close_notifiers, bs);
1446 if (bs->backing_hd) {
1447 bdrv_unref(bs->backing_hd);
1448 bs->backing_hd = NULL;
1450 bs->drv->bdrv_close(bs);
1453 if (bs->is_temporary) {
1454 unlink(bs->filename);
1459 bs->copy_on_read = 0;
1460 bs->backing_file[0] = '\0';
1461 bs->backing_format[0] = '\0';
1462 bs->total_sectors = 0;
1467 bs->zero_beyond_eof = false;
1468 QDECREF(bs->options);
1471 if (bs->file != NULL) {
1472 bdrv_unref(bs->file);
1477 bdrv_dev_change_media_cb(bs, false);
1479 /*throttling disk I/O limits*/
1480 if (bs->io_limits_enabled) {
1481 bdrv_io_limits_disable(bs);
1485 void bdrv_close_all(void)
1487 BlockDriverState *bs;
1489 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1494 /* Check if any requests are in-flight (including throttled requests) */
1495 static bool bdrv_requests_pending(BlockDriverState *bs)
1497 if (!QLIST_EMPTY(&bs->tracked_requests)) {
1500 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1503 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1506 if (bs->file && bdrv_requests_pending(bs->file)) {
1509 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1515 static bool bdrv_requests_pending_all(void)
1517 BlockDriverState *bs;
1518 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1519 if (bdrv_requests_pending(bs)) {
1527 * Wait for pending requests to complete across all BlockDriverStates
1529 * This function does not flush data to disk, use bdrv_flush_all() for that
1530 * after calling this function.
1532 * Note that completion of an asynchronous I/O operation can trigger any
1533 * number of other I/O operations on other devices---for example a coroutine
1534 * can be arbitrarily complex and a constant flow of I/O can come until the
1535 * coroutine is complete. Because of this, it is not possible to have a
1536 * function to drain a single device's I/O queue.
1538 void bdrv_drain_all(void)
1540 /* Always run first iteration so any pending completion BHs run */
1542 BlockDriverState *bs;
1545 /* FIXME: We do not have timer support here, so this is effectively
1548 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1549 if (bdrv_start_throttled_reqs(bs)) {
1554 busy = bdrv_requests_pending_all();
1555 busy |= aio_poll(qemu_get_aio_context(), busy);
1559 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1560 Also, NULL terminate the device_name to prevent double remove */
1561 void bdrv_make_anon(BlockDriverState *bs)
1563 if (bs->device_name[0] != '\0') {
1564 QTAILQ_REMOVE(&bdrv_states, bs, list);
1566 bs->device_name[0] = '\0';
1569 static void bdrv_rebind(BlockDriverState *bs)
1571 if (bs->drv && bs->drv->bdrv_rebind) {
1572 bs->drv->bdrv_rebind(bs);
1576 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1577 BlockDriverState *bs_src)
1579 /* move some fields that need to stay attached to the device */
1580 bs_dest->open_flags = bs_src->open_flags;
1583 bs_dest->dev_ops = bs_src->dev_ops;
1584 bs_dest->dev_opaque = bs_src->dev_opaque;
1585 bs_dest->dev = bs_src->dev;
1586 bs_dest->buffer_alignment = bs_src->buffer_alignment;
1587 bs_dest->copy_on_read = bs_src->copy_on_read;
1589 bs_dest->enable_write_cache = bs_src->enable_write_cache;
1591 /* i/o throttled req */
1592 memcpy(&bs_dest->throttle_state,
1593 &bs_src->throttle_state,
1594 sizeof(ThrottleState));
1595 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
1596 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
1597 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1600 bs_dest->on_read_error = bs_src->on_read_error;
1601 bs_dest->on_write_error = bs_src->on_write_error;
1604 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1605 bs_dest->iostatus = bs_src->iostatus;
1608 bs_dest->dirty_bitmap = bs_src->dirty_bitmap;
1610 /* reference count */
1611 bs_dest->refcnt = bs_src->refcnt;
1614 bs_dest->in_use = bs_src->in_use;
1615 bs_dest->job = bs_src->job;
1617 /* keep the same entry in bdrv_states */
1618 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1619 bs_src->device_name);
1620 bs_dest->list = bs_src->list;
1624 * Swap bs contents for two image chains while they are live,
1625 * while keeping required fields on the BlockDriverState that is
1626 * actually attached to a device.
1628 * This will modify the BlockDriverState fields, and swap contents
1629 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1631 * bs_new is required to be anonymous.
1633 * This function does not create any image files.
1635 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
1637 BlockDriverState tmp;
1639 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1640 assert(bs_new->device_name[0] == '\0');
1641 assert(bs_new->dirty_bitmap == NULL);
1642 assert(bs_new->job == NULL);
1643 assert(bs_new->dev == NULL);
1644 assert(bs_new->in_use == 0);
1645 assert(bs_new->io_limits_enabled == false);
1646 assert(!throttle_have_timer(&bs_new->throttle_state));
1652 /* there are some fields that should not be swapped, move them back */
1653 bdrv_move_feature_fields(&tmp, bs_old);
1654 bdrv_move_feature_fields(bs_old, bs_new);
1655 bdrv_move_feature_fields(bs_new, &tmp);
1657 /* bs_new shouldn't be in bdrv_states even after the swap! */
1658 assert(bs_new->device_name[0] == '\0');
1660 /* Check a few fields that should remain attached to the device */
1661 assert(bs_new->dev == NULL);
1662 assert(bs_new->job == NULL);
1663 assert(bs_new->in_use == 0);
1664 assert(bs_new->io_limits_enabled == false);
1665 assert(!throttle_have_timer(&bs_new->throttle_state));
1667 bdrv_rebind(bs_new);
1668 bdrv_rebind(bs_old);
1672 * Add new bs contents at the top of an image chain while the chain is
1673 * live, while keeping required fields on the top layer.
1675 * This will modify the BlockDriverState fields, and swap contents
1676 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1678 * bs_new is required to be anonymous.
1680 * This function does not create any image files.
1682 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
1684 bdrv_swap(bs_new, bs_top);
1686 /* The contents of 'tmp' will become bs_top, as we are
1687 * swapping bs_new and bs_top contents. */
1688 bs_top->backing_hd = bs_new;
1689 bs_top->open_flags &= ~BDRV_O_NO_BACKING;
1690 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
1692 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
1693 bs_new->drv ? bs_new->drv->format_name : "");
1696 static void bdrv_delete(BlockDriverState *bs)
1700 assert(!bs->in_use);
1701 assert(!bs->refcnt);
1705 /* remove from list, if necessary */
1711 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1712 /* TODO change to DeviceState *dev when all users are qdevified */
1718 bdrv_iostatus_reset(bs);
1722 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1723 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1725 if (bdrv_attach_dev(bs, dev) < 0) {
1730 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1731 /* TODO change to DeviceState *dev when all users are qdevified */
1733 assert(bs->dev == dev);
1736 bs->dev_opaque = NULL;
1737 bs->buffer_alignment = 512;
1740 /* TODO change to return DeviceState * when all users are qdevified */
1741 void *bdrv_get_attached_dev(BlockDriverState *bs)
1746 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1750 bs->dev_opaque = opaque;
1753 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1754 enum MonitorEvent ev,
1755 BlockErrorAction action, bool is_read)
1758 const char *action_str;
1761 case BDRV_ACTION_REPORT:
1762 action_str = "report";
1764 case BDRV_ACTION_IGNORE:
1765 action_str = "ignore";
1767 case BDRV_ACTION_STOP:
1768 action_str = "stop";
1774 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1777 is_read ? "read" : "write");
1778 monitor_protocol_event(ev, data);
1780 qobject_decref(data);
1783 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1787 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1788 bdrv_get_device_name(bs), ejected);
1789 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1791 qobject_decref(data);
1794 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1796 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1797 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1798 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1799 if (tray_was_closed) {
1801 bdrv_emit_qmp_eject_event(bs, true);
1805 bdrv_emit_qmp_eject_event(bs, false);
1810 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1812 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1815 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1817 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1818 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1822 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1824 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1825 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1830 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1832 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1833 bs->dev_ops->resize_cb(bs->dev_opaque);
1837 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1839 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1840 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1846 * Run consistency checks on an image
1848 * Returns 0 if the check could be completed (it doesn't mean that the image is
1849 * free of errors) or -errno when an internal error occurred. The results of the
1850 * check are stored in res.
1852 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
1854 if (bs->drv->bdrv_check == NULL) {
1858 memset(res, 0, sizeof(*res));
1859 return bs->drv->bdrv_check(bs, res, fix);
1862 #define COMMIT_BUF_SECTORS 2048
1864 /* commit COW file into the raw image */
1865 int bdrv_commit(BlockDriverState *bs)
1867 BlockDriver *drv = bs->drv;
1868 int64_t sector, total_sectors;
1869 int n, ro, open_flags;
1872 char filename[PATH_MAX];
1877 if (!bs->backing_hd) {
1881 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1885 ro = bs->backing_hd->read_only;
1886 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
1887 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
1888 open_flags = bs->backing_hd->open_flags;
1891 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
1896 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1897 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1899 for (sector = 0; sector < total_sectors; sector += n) {
1900 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
1905 if (bdrv_read(bs, sector, buf, n) != 0) {
1910 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1917 if (drv->bdrv_make_empty) {
1918 ret = drv->bdrv_make_empty(bs);
1923 * Make sure all data we wrote to the backing device is actually
1927 bdrv_flush(bs->backing_hd);
1933 /* ignoring error return here */
1934 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
1940 int bdrv_commit_all(void)
1942 BlockDriverState *bs;
1944 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1945 if (bs->drv && bs->backing_hd) {
1946 int ret = bdrv_commit(bs);
1956 * Remove an active request from the tracked requests list
1958 * This function should be called when a tracked request is completing.
1960 static void tracked_request_end(BdrvTrackedRequest *req)
1962 QLIST_REMOVE(req, list);
1963 qemu_co_queue_restart_all(&req->wait_queue);
1967 * Add an active request to the tracked requests list
1969 static void tracked_request_begin(BdrvTrackedRequest *req,
1970 BlockDriverState *bs,
1972 int nb_sectors, bool is_write)
1974 *req = (BdrvTrackedRequest){
1976 .sector_num = sector_num,
1977 .nb_sectors = nb_sectors,
1978 .is_write = is_write,
1979 .co = qemu_coroutine_self(),
1982 qemu_co_queue_init(&req->wait_queue);
1984 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1988 * Round a region to cluster boundaries
1990 void bdrv_round_to_clusters(BlockDriverState *bs,
1991 int64_t sector_num, int nb_sectors,
1992 int64_t *cluster_sector_num,
1993 int *cluster_nb_sectors)
1995 BlockDriverInfo bdi;
1997 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1998 *cluster_sector_num = sector_num;
1999 *cluster_nb_sectors = nb_sectors;
2001 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2002 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2003 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2008 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2009 int64_t sector_num, int nb_sectors) {
2011 if (sector_num >= req->sector_num + req->nb_sectors) {
2015 if (req->sector_num >= sector_num + nb_sectors) {
2021 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
2022 int64_t sector_num, int nb_sectors)
2024 BdrvTrackedRequest *req;
2025 int64_t cluster_sector_num;
2026 int cluster_nb_sectors;
2029 /* If we touch the same cluster it counts as an overlap. This guarantees
2030 * that allocating writes will be serialized and not race with each other
2031 * for the same cluster. For example, in copy-on-read it ensures that the
2032 * CoR read and write operations are atomic and guest writes cannot
2033 * interleave between them.
2035 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2036 &cluster_sector_num, &cluster_nb_sectors);
2040 QLIST_FOREACH(req, &bs->tracked_requests, list) {
2041 if (tracked_request_overlaps(req, cluster_sector_num,
2042 cluster_nb_sectors)) {
2043 /* Hitting this means there was a reentrant request, for
2044 * example, a block driver issuing nested requests. This must
2045 * never happen since it means deadlock.
2047 assert(qemu_coroutine_self() != req->co);
2049 qemu_co_queue_wait(&req->wait_queue);
2060 * -EINVAL - backing format specified, but no file
2061 * -ENOSPC - can't update the backing file because no space is left in the
2063 * -ENOTSUP - format driver doesn't support changing the backing file
2065 int bdrv_change_backing_file(BlockDriverState *bs,
2066 const char *backing_file, const char *backing_fmt)
2068 BlockDriver *drv = bs->drv;
2071 /* Backing file format doesn't make sense without a backing file */
2072 if (backing_fmt && !backing_file) {
2076 if (drv->bdrv_change_backing_file != NULL) {
2077 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2083 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2084 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2090 * Finds the image layer in the chain that has 'bs' as its backing file.
2092 * active is the current topmost image.
2094 * Returns NULL if bs is not found in active's image chain,
2095 * or if active == bs.
2097 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2098 BlockDriverState *bs)
2100 BlockDriverState *overlay = NULL;
2101 BlockDriverState *intermediate;
2103 assert(active != NULL);
2106 /* if bs is the same as active, then by definition it has no overlay
2112 intermediate = active;
2113 while (intermediate->backing_hd) {
2114 if (intermediate->backing_hd == bs) {
2115 overlay = intermediate;
2118 intermediate = intermediate->backing_hd;
2124 typedef struct BlkIntermediateStates {
2125 BlockDriverState *bs;
2126 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2127 } BlkIntermediateStates;
2131 * Drops images above 'base' up to and including 'top', and sets the image
2132 * above 'top' to have base as its backing file.
2134 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2135 * information in 'bs' can be properly updated.
2137 * E.g., this will convert the following chain:
2138 * bottom <- base <- intermediate <- top <- active
2142 * bottom <- base <- active
2144 * It is allowed for bottom==base, in which case it converts:
2146 * base <- intermediate <- top <- active
2153 * if active == top, that is considered an error
2156 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2157 BlockDriverState *base)
2159 BlockDriverState *intermediate;
2160 BlockDriverState *base_bs = NULL;
2161 BlockDriverState *new_top_bs = NULL;
2162 BlkIntermediateStates *intermediate_state, *next;
2165 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2166 QSIMPLEQ_INIT(&states_to_delete);
2168 if (!top->drv || !base->drv) {
2172 new_top_bs = bdrv_find_overlay(active, top);
2174 if (new_top_bs == NULL) {
2175 /* we could not find the image above 'top', this is an error */
2179 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2180 * to do, no intermediate images */
2181 if (new_top_bs->backing_hd == base) {
2188 /* now we will go down through the list, and add each BDS we find
2189 * into our deletion queue, until we hit the 'base'
2191 while (intermediate) {
2192 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2193 intermediate_state->bs = intermediate;
2194 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2196 if (intermediate->backing_hd == base) {
2197 base_bs = intermediate->backing_hd;
2200 intermediate = intermediate->backing_hd;
2202 if (base_bs == NULL) {
2203 /* something went wrong, we did not end at the base. safely
2204 * unravel everything, and exit with error */
2208 /* success - we can delete the intermediate states, and link top->base */
2209 ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2210 base_bs->drv ? base_bs->drv->format_name : "");
2214 new_top_bs->backing_hd = base_bs;
2217 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2218 /* so that bdrv_close() does not recursively close the chain */
2219 intermediate_state->bs->backing_hd = NULL;
2220 bdrv_unref(intermediate_state->bs);
2225 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2226 g_free(intermediate_state);
2232 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2237 if (!bdrv_is_inserted(bs))
2243 len = bdrv_getlength(bs);
2248 if ((offset > len) || (len - offset < size))
2254 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2257 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2258 nb_sectors * BDRV_SECTOR_SIZE);
2261 typedef struct RwCo {
2262 BlockDriverState *bs;
2268 BdrvRequestFlags flags;
2271 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2273 RwCo *rwco = opaque;
2275 if (!rwco->is_write) {
2276 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
2277 rwco->nb_sectors, rwco->qiov,
2280 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
2281 rwco->nb_sectors, rwco->qiov,
2287 * Process a vectored synchronous request using coroutines
2289 static int bdrv_rwv_co(BlockDriverState *bs, int64_t sector_num,
2290 QEMUIOVector *qiov, bool is_write,
2291 BdrvRequestFlags flags)
2296 .sector_num = sector_num,
2297 .nb_sectors = qiov->size >> BDRV_SECTOR_BITS,
2299 .is_write = is_write,
2303 assert((qiov->size & (BDRV_SECTOR_SIZE - 1)) == 0);
2306 * In sync call context, when the vcpu is blocked, this throttling timer
2307 * will not fire; so the I/O throttling function has to be disabled here
2308 * if it has been enabled.
2310 if (bs->io_limits_enabled) {
2311 fprintf(stderr, "Disabling I/O throttling on '%s' due "
2312 "to synchronous I/O.\n", bdrv_get_device_name(bs));
2313 bdrv_io_limits_disable(bs);
2316 if (qemu_in_coroutine()) {
2317 /* Fast-path if already in coroutine context */
2318 bdrv_rw_co_entry(&rwco);
2320 co = qemu_coroutine_create(bdrv_rw_co_entry);
2321 qemu_coroutine_enter(co, &rwco);
2322 while (rwco.ret == NOT_DONE) {
2330 * Process a synchronous request using coroutines
2332 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2333 int nb_sectors, bool is_write, BdrvRequestFlags flags)
2336 struct iovec iov = {
2337 .iov_base = (void *)buf,
2338 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2341 qemu_iovec_init_external(&qiov, &iov, 1);
2342 return bdrv_rwv_co(bs, sector_num, &qiov, is_write, flags);
2345 /* return < 0 if error. See bdrv_write() for the return codes */
2346 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2347 uint8_t *buf, int nb_sectors)
2349 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2352 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2353 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2354 uint8_t *buf, int nb_sectors)
2359 enabled = bs->io_limits_enabled;
2360 bs->io_limits_enabled = false;
2361 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2362 bs->io_limits_enabled = enabled;
2366 /* Return < 0 if error. Important errors are:
2367 -EIO generic I/O error (may happen for all errors)
2368 -ENOMEDIUM No media inserted.
2369 -EINVAL Invalid sector number or nb_sectors
2370 -EACCES Trying to write a read-only device
2372 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2373 const uint8_t *buf, int nb_sectors)
2375 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2378 int bdrv_writev(BlockDriverState *bs, int64_t sector_num, QEMUIOVector *qiov)
2380 return bdrv_rwv_co(bs, sector_num, qiov, true, 0);
2383 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
2385 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2386 BDRV_REQ_ZERO_WRITE);
2389 int bdrv_pread(BlockDriverState *bs, int64_t offset,
2390 void *buf, int count1)
2392 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
2393 int len, nb_sectors, count;
2398 /* first read to align to sector start */
2399 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
2402 sector_num = offset >> BDRV_SECTOR_BITS;
2404 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2406 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
2414 /* read the sectors "in place" */
2415 nb_sectors = count >> BDRV_SECTOR_BITS;
2416 if (nb_sectors > 0) {
2417 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
2419 sector_num += nb_sectors;
2420 len = nb_sectors << BDRV_SECTOR_BITS;
2425 /* add data from the last sector */
2427 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2429 memcpy(buf, tmp_buf, count);
2434 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2436 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
2437 int len, nb_sectors, count;
2443 /* first write to align to sector start */
2444 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
2447 sector_num = offset >> BDRV_SECTOR_BITS;
2449 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2451 qemu_iovec_to_buf(qiov, 0, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)),
2453 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
2461 /* write the sectors "in place" */
2462 nb_sectors = count >> BDRV_SECTOR_BITS;
2463 if (nb_sectors > 0) {
2464 QEMUIOVector qiov_inplace;
2466 qemu_iovec_init(&qiov_inplace, qiov->niov);
2467 qemu_iovec_concat(&qiov_inplace, qiov, len,
2468 nb_sectors << BDRV_SECTOR_BITS);
2469 ret = bdrv_writev(bs, sector_num, &qiov_inplace);
2470 qemu_iovec_destroy(&qiov_inplace);
2475 sector_num += nb_sectors;
2476 len = nb_sectors << BDRV_SECTOR_BITS;
2480 /* add data from the last sector */
2482 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2484 qemu_iovec_to_buf(qiov, qiov->size - count, tmp_buf, count);
2485 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
2491 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2492 const void *buf, int count1)
2495 struct iovec iov = {
2496 .iov_base = (void *) buf,
2500 qemu_iovec_init_external(&qiov, &iov, 1);
2501 return bdrv_pwritev(bs, offset, &qiov);
2505 * Writes to the file and ensures that no writes are reordered across this
2506 * request (acts as a barrier)
2508 * Returns 0 on success, -errno in error cases.
2510 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2511 const void *buf, int count)
2515 ret = bdrv_pwrite(bs, offset, buf, count);
2520 /* No flush needed for cache modes that already do it */
2521 if (bs->enable_write_cache) {
2528 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2529 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2531 /* Perform I/O through a temporary buffer so that users who scribble over
2532 * their read buffer while the operation is in progress do not end up
2533 * modifying the image file. This is critical for zero-copy guest I/O
2534 * where anything might happen inside guest memory.
2536 void *bounce_buffer;
2538 BlockDriver *drv = bs->drv;
2540 QEMUIOVector bounce_qiov;
2541 int64_t cluster_sector_num;
2542 int cluster_nb_sectors;
2546 /* Cover entire cluster so no additional backing file I/O is required when
2547 * allocating cluster in the image file.
2549 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2550 &cluster_sector_num, &cluster_nb_sectors);
2552 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
2553 cluster_sector_num, cluster_nb_sectors);
2555 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
2556 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
2557 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
2559 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
2565 if (drv->bdrv_co_write_zeroes &&
2566 buffer_is_zero(bounce_buffer, iov.iov_len)) {
2567 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
2568 cluster_nb_sectors);
2570 /* This does not change the data on the disk, it is not necessary
2571 * to flush even in cache=writethrough mode.
2573 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
2578 /* It might be okay to ignore write errors for guest requests. If this
2579 * is a deliberate copy-on-read then we don't want to ignore the error.
2580 * Simply report it in all cases.
2585 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
2586 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
2587 nb_sectors * BDRV_SECTOR_SIZE);
2590 qemu_vfree(bounce_buffer);
2595 * Handle a read request in coroutine context
2597 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
2598 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
2599 BdrvRequestFlags flags)
2601 BlockDriver *drv = bs->drv;
2602 BdrvTrackedRequest req;
2608 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
2612 if (bs->copy_on_read) {
2613 flags |= BDRV_REQ_COPY_ON_READ;
2615 if (flags & BDRV_REQ_COPY_ON_READ) {
2616 bs->copy_on_read_in_flight++;
2619 if (bs->copy_on_read_in_flight) {
2620 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
2623 /* throttling disk I/O */
2624 if (bs->io_limits_enabled) {
2625 bdrv_io_limits_intercept(bs, nb_sectors, false);
2628 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
2630 if (flags & BDRV_REQ_COPY_ON_READ) {
2633 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
2638 if (!ret || pnum != nb_sectors) {
2639 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
2644 if (!(bs->zero_beyond_eof && bs->growable)) {
2645 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
2647 /* Read zeros after EOF of growable BDSes */
2648 int64_t len, total_sectors, max_nb_sectors;
2650 len = bdrv_getlength(bs);
2656 total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
2657 max_nb_sectors = MAX(0, total_sectors - sector_num);
2658 if (max_nb_sectors > 0) {
2659 ret = drv->bdrv_co_readv(bs, sector_num,
2660 MIN(nb_sectors, max_nb_sectors), qiov);
2665 /* Reading beyond end of file is supposed to produce zeroes */
2666 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
2667 uint64_t offset = MAX(0, total_sectors - sector_num);
2668 uint64_t bytes = (sector_num + nb_sectors - offset) *
2670 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
2675 tracked_request_end(&req);
2677 if (flags & BDRV_REQ_COPY_ON_READ) {
2678 bs->copy_on_read_in_flight--;
2684 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
2685 int nb_sectors, QEMUIOVector *qiov)
2687 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
2689 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
2692 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
2693 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2695 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
2697 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
2698 BDRV_REQ_COPY_ON_READ);
2701 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
2702 int64_t sector_num, int nb_sectors)
2704 BlockDriver *drv = bs->drv;
2709 /* TODO Emulate only part of misaligned requests instead of letting block
2710 * drivers return -ENOTSUP and emulate everything */
2712 /* First try the efficient write zeroes operation */
2713 if (drv->bdrv_co_write_zeroes) {
2714 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
2715 if (ret != -ENOTSUP) {
2720 /* Fall back to bounce buffer if write zeroes is unsupported */
2721 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2722 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
2723 memset(iov.iov_base, 0, iov.iov_len);
2724 qemu_iovec_init_external(&qiov, &iov, 1);
2726 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
2728 qemu_vfree(iov.iov_base);
2733 * Handle a write request in coroutine context
2735 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
2736 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
2737 BdrvRequestFlags flags)
2739 BlockDriver *drv = bs->drv;
2740 BdrvTrackedRequest req;
2746 if (bs->read_only) {
2749 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
2753 if (bs->copy_on_read_in_flight) {
2754 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
2757 /* throttling disk I/O */
2758 if (bs->io_limits_enabled) {
2759 bdrv_io_limits_intercept(bs, nb_sectors, true);
2762 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
2764 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req);
2767 /* Do nothing, write notifier decided to fail this request */
2768 } else if (flags & BDRV_REQ_ZERO_WRITE) {
2769 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
2771 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
2774 if (ret == 0 && !bs->enable_write_cache) {
2775 ret = bdrv_co_flush(bs);
2778 if (bs->dirty_bitmap) {
2779 bdrv_set_dirty(bs, sector_num, nb_sectors);
2782 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2783 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2785 if (bs->growable && ret >= 0) {
2786 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
2789 tracked_request_end(&req);
2794 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
2795 int nb_sectors, QEMUIOVector *qiov)
2797 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
2799 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
2802 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
2803 int64_t sector_num, int nb_sectors)
2805 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
2807 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
2808 BDRV_REQ_ZERO_WRITE);
2812 * Truncate file to 'offset' bytes (needed only for file protocols)
2814 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2816 BlockDriver *drv = bs->drv;
2820 if (!drv->bdrv_truncate)
2824 if (bdrv_in_use(bs))
2826 ret = drv->bdrv_truncate(bs, offset);
2828 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2829 bdrv_dev_resize_cb(bs);
2835 * Length of a allocated file in bytes. Sparse files are counted by actual
2836 * allocated space. Return < 0 if error or unknown.
2838 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2840 BlockDriver *drv = bs->drv;
2844 if (drv->bdrv_get_allocated_file_size) {
2845 return drv->bdrv_get_allocated_file_size(bs);
2848 return bdrv_get_allocated_file_size(bs->file);
2854 * Length of a file in bytes. Return < 0 if error or unknown.
2856 int64_t bdrv_getlength(BlockDriverState *bs)
2858 BlockDriver *drv = bs->drv;
2862 if (bdrv_dev_has_removable_media(bs)) {
2863 if (drv->bdrv_getlength) {
2864 return drv->bdrv_getlength(bs);
2867 return bs->total_sectors * BDRV_SECTOR_SIZE;
2870 /* return 0 as number of sectors if no device present or error */
2871 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2874 length = bdrv_getlength(bs);
2878 length = length >> BDRV_SECTOR_BITS;
2879 *nb_sectors_ptr = length;
2882 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
2883 BlockdevOnError on_write_error)
2885 bs->on_read_error = on_read_error;
2886 bs->on_write_error = on_write_error;
2889 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
2891 return is_read ? bs->on_read_error : bs->on_write_error;
2894 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
2896 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
2899 case BLOCKDEV_ON_ERROR_ENOSPC:
2900 return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT;
2901 case BLOCKDEV_ON_ERROR_STOP:
2902 return BDRV_ACTION_STOP;
2903 case BLOCKDEV_ON_ERROR_REPORT:
2904 return BDRV_ACTION_REPORT;
2905 case BLOCKDEV_ON_ERROR_IGNORE:
2906 return BDRV_ACTION_IGNORE;
2912 /* This is done by device models because, while the block layer knows
2913 * about the error, it does not know whether an operation comes from
2914 * the device or the block layer (from a job, for example).
2916 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
2917 bool is_read, int error)
2920 bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read);
2921 if (action == BDRV_ACTION_STOP) {
2922 vm_stop(RUN_STATE_IO_ERROR);
2923 bdrv_iostatus_set_err(bs, error);
2927 int bdrv_is_read_only(BlockDriverState *bs)
2929 return bs->read_only;
2932 int bdrv_is_sg(BlockDriverState *bs)
2937 int bdrv_enable_write_cache(BlockDriverState *bs)
2939 return bs->enable_write_cache;
2942 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
2944 bs->enable_write_cache = wce;
2946 /* so a reopen() will preserve wce */
2948 bs->open_flags |= BDRV_O_CACHE_WB;
2950 bs->open_flags &= ~BDRV_O_CACHE_WB;
2954 int bdrv_is_encrypted(BlockDriverState *bs)
2956 if (bs->backing_hd && bs->backing_hd->encrypted)
2958 return bs->encrypted;
2961 int bdrv_key_required(BlockDriverState *bs)
2963 BlockDriverState *backing_hd = bs->backing_hd;
2965 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2967 return (bs->encrypted && !bs->valid_key);
2970 int bdrv_set_key(BlockDriverState *bs, const char *key)
2973 if (bs->backing_hd && bs->backing_hd->encrypted) {
2974 ret = bdrv_set_key(bs->backing_hd, key);
2980 if (!bs->encrypted) {
2982 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2985 ret = bs->drv->bdrv_set_key(bs, key);
2988 } else if (!bs->valid_key) {
2990 /* call the change callback now, we skipped it on open */
2991 bdrv_dev_change_media_cb(bs, true);
2996 const char *bdrv_get_format_name(BlockDriverState *bs)
2998 return bs->drv ? bs->drv->format_name : NULL;
3001 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3006 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3007 it(opaque, drv->format_name);
3011 BlockDriverState *bdrv_find(const char *name)
3013 BlockDriverState *bs;
3015 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3016 if (!strcmp(name, bs->device_name)) {
3023 BlockDriverState *bdrv_next(BlockDriverState *bs)
3026 return QTAILQ_FIRST(&bdrv_states);
3028 return QTAILQ_NEXT(bs, list);
3031 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3033 BlockDriverState *bs;
3035 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3040 const char *bdrv_get_device_name(BlockDriverState *bs)
3042 return bs->device_name;
3045 int bdrv_get_flags(BlockDriverState *bs)
3047 return bs->open_flags;
3050 int bdrv_flush_all(void)
3052 BlockDriverState *bs;
3055 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3056 int ret = bdrv_flush(bs);
3057 if (ret < 0 && !result) {
3065 int bdrv_has_zero_init_1(BlockDriverState *bs)
3070 int bdrv_has_zero_init(BlockDriverState *bs)
3074 /* If BS is a copy on write image, it is initialized to
3075 the contents of the base image, which may not be zeroes. */
3076 if (bs->backing_hd) {
3079 if (bs->drv->bdrv_has_zero_init) {
3080 return bs->drv->bdrv_has_zero_init(bs);
3087 typedef struct BdrvCoGetBlockStatusData {
3088 BlockDriverState *bs;
3089 BlockDriverState *base;
3095 } BdrvCoGetBlockStatusData;
3098 * Returns true iff the specified sector is present in the disk image. Drivers
3099 * not implementing the functionality are assumed to not support backing files,
3100 * hence all their sectors are reported as allocated.
3102 * If 'sector_num' is beyond the end of the disk image the return value is 0
3103 * and 'pnum' is set to 0.
3105 * 'pnum' is set to the number of sectors (including and immediately following
3106 * the specified sector) that are known to be in the same
3107 * allocated/unallocated state.
3109 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3110 * beyond the end of the disk image it will be clamped.
3112 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3114 int nb_sectors, int *pnum)
3120 length = bdrv_getlength(bs);
3125 if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3130 n = bs->total_sectors - sector_num;
3131 if (n < nb_sectors) {
3135 if (!bs->drv->bdrv_co_get_block_status) {
3137 ret = BDRV_BLOCK_DATA;
3138 if (bs->drv->protocol_name) {
3139 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3144 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3150 if (!(ret & BDRV_BLOCK_DATA)) {
3151 if (bdrv_has_zero_init(bs)) {
3152 ret |= BDRV_BLOCK_ZERO;
3153 } else if (bs->backing_hd) {
3154 BlockDriverState *bs2 = bs->backing_hd;
3155 int64_t length2 = bdrv_getlength(bs2);
3156 if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
3157 ret |= BDRV_BLOCK_ZERO;
3163 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
3164 (ret & BDRV_BLOCK_OFFSET_VALID)) {
3165 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3168 /* Ignore errors. This is just providing extra information, it
3169 * is useful but not necessary.
3171 ret |= (ret2 & BDRV_BLOCK_ZERO);
3178 /* Coroutine wrapper for bdrv_get_block_status() */
3179 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
3181 BdrvCoGetBlockStatusData *data = opaque;
3182 BlockDriverState *bs = data->bs;
3184 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
3190 * Synchronous wrapper around bdrv_co_get_block_status().
3192 * See bdrv_co_get_block_status() for details.
3194 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
3195 int nb_sectors, int *pnum)
3198 BdrvCoGetBlockStatusData data = {
3200 .sector_num = sector_num,
3201 .nb_sectors = nb_sectors,
3206 if (qemu_in_coroutine()) {
3207 /* Fast-path if already in coroutine context */
3208 bdrv_get_block_status_co_entry(&data);
3210 co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
3211 qemu_coroutine_enter(co, &data);
3212 while (!data.done) {
3219 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
3220 int nb_sectors, int *pnum)
3222 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
3227 (ret & BDRV_BLOCK_DATA) ||
3228 ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs));
3232 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
3234 * Return true if the given sector is allocated in any image between
3235 * BASE and TOP (inclusive). BASE can be NULL to check if the given
3236 * sector is allocated in any image of the chain. Return false otherwise.
3238 * 'pnum' is set to the number of sectors (including and immediately following
3239 * the specified sector) that are known to be in the same
3240 * allocated/unallocated state.
3243 int bdrv_is_allocated_above(BlockDriverState *top,
3244 BlockDriverState *base,
3246 int nb_sectors, int *pnum)
3248 BlockDriverState *intermediate;
3249 int ret, n = nb_sectors;
3252 while (intermediate && intermediate != base) {
3254 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
3264 * [sector_num, nb_sectors] is unallocated on top but intermediate
3267 * [sector_num+x, nr_sectors] allocated.
3269 if (n > pnum_inter &&
3270 (intermediate == top ||
3271 sector_num + pnum_inter < intermediate->total_sectors)) {
3275 intermediate = intermediate->backing_hd;
3282 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3284 if (bs->backing_hd && bs->backing_hd->encrypted)
3285 return bs->backing_file;
3286 else if (bs->encrypted)
3287 return bs->filename;
3292 void bdrv_get_backing_filename(BlockDriverState *bs,
3293 char *filename, int filename_size)
3295 pstrcpy(filename, filename_size, bs->backing_file);
3298 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
3299 const uint8_t *buf, int nb_sectors)
3301 BlockDriver *drv = bs->drv;
3304 if (!drv->bdrv_write_compressed)
3306 if (bdrv_check_request(bs, sector_num, nb_sectors))
3309 assert(!bs->dirty_bitmap);
3311 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
3314 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3316 BlockDriver *drv = bs->drv;
3319 if (!drv->bdrv_get_info)
3321 memset(bdi, 0, sizeof(*bdi));
3322 return drv->bdrv_get_info(bs, bdi);
3325 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
3326 int64_t pos, int size)
3329 struct iovec iov = {
3330 .iov_base = (void *) buf,
3334 qemu_iovec_init_external(&qiov, &iov, 1);
3335 return bdrv_writev_vmstate(bs, &qiov, pos);
3338 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
3340 BlockDriver *drv = bs->drv;
3344 } else if (drv->bdrv_save_vmstate) {
3345 return drv->bdrv_save_vmstate(bs, qiov, pos);
3346 } else if (bs->file) {
3347 return bdrv_writev_vmstate(bs->file, qiov, pos);
3353 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
3354 int64_t pos, int size)
3356 BlockDriver *drv = bs->drv;
3359 if (drv->bdrv_load_vmstate)
3360 return drv->bdrv_load_vmstate(bs, buf, pos, size);
3362 return bdrv_load_vmstate(bs->file, buf, pos, size);
3366 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
3368 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3372 bs->drv->bdrv_debug_event(bs, event);
3375 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3378 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3382 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3383 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3389 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3391 while (bs && bs->drv && !bs->drv->bdrv_debug_resume) {
3395 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3396 return bs->drv->bdrv_debug_resume(bs, tag);
3402 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3404 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3408 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3409 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3415 int bdrv_is_snapshot(BlockDriverState *bs)
3417 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
3420 /* backing_file can either be relative, or absolute, or a protocol. If it is
3421 * relative, it must be relative to the chain. So, passing in bs->filename
3422 * from a BDS as backing_file should not be done, as that may be relative to
3423 * the CWD rather than the chain. */
3424 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3425 const char *backing_file)
3427 char *filename_full = NULL;
3428 char *backing_file_full = NULL;
3429 char *filename_tmp = NULL;
3430 int is_protocol = 0;
3431 BlockDriverState *curr_bs = NULL;
3432 BlockDriverState *retval = NULL;
3434 if (!bs || !bs->drv || !backing_file) {
3438 filename_full = g_malloc(PATH_MAX);
3439 backing_file_full = g_malloc(PATH_MAX);
3440 filename_tmp = g_malloc(PATH_MAX);
3442 is_protocol = path_has_protocol(backing_file);
3444 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
3446 /* If either of the filename paths is actually a protocol, then
3447 * compare unmodified paths; otherwise make paths relative */
3448 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3449 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3450 retval = curr_bs->backing_hd;
3454 /* If not an absolute filename path, make it relative to the current
3455 * image's filename path */
3456 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3459 /* We are going to compare absolute pathnames */
3460 if (!realpath(filename_tmp, filename_full)) {
3464 /* We need to make sure the backing filename we are comparing against
3465 * is relative to the current image filename (or absolute) */
3466 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3467 curr_bs->backing_file);
3469 if (!realpath(filename_tmp, backing_file_full)) {
3473 if (strcmp(backing_file_full, filename_full) == 0) {
3474 retval = curr_bs->backing_hd;
3480 g_free(filename_full);
3481 g_free(backing_file_full);
3482 g_free(filename_tmp);
3486 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3492 if (!bs->backing_hd) {
3496 return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
3499 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3501 BlockDriverState *curr_bs = NULL;
3509 while (curr_bs->backing_hd) {
3510 curr_bs = curr_bs->backing_hd;
3515 /**************************************************************/
3518 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
3519 QEMUIOVector *qiov, int nb_sectors,
3520 BlockDriverCompletionFunc *cb, void *opaque)
3522 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
3524 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
3528 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
3529 QEMUIOVector *qiov, int nb_sectors,
3530 BlockDriverCompletionFunc *cb, void *opaque)
3532 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
3534 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
3539 typedef struct MultiwriteCB {
3544 BlockDriverCompletionFunc *cb;
3546 QEMUIOVector *free_qiov;
3550 static void multiwrite_user_cb(MultiwriteCB *mcb)
3554 for (i = 0; i < mcb->num_callbacks; i++) {
3555 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
3556 if (mcb->callbacks[i].free_qiov) {
3557 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
3559 g_free(mcb->callbacks[i].free_qiov);
3563 static void multiwrite_cb(void *opaque, int ret)
3565 MultiwriteCB *mcb = opaque;
3567 trace_multiwrite_cb(mcb, ret);
3569 if (ret < 0 && !mcb->error) {
3573 mcb->num_requests--;
3574 if (mcb->num_requests == 0) {
3575 multiwrite_user_cb(mcb);
3580 static int multiwrite_req_compare(const void *a, const void *b)
3582 const BlockRequest *req1 = a, *req2 = b;
3585 * Note that we can't simply subtract req2->sector from req1->sector
3586 * here as that could overflow the return value.
3588 if (req1->sector > req2->sector) {
3590 } else if (req1->sector < req2->sector) {
3598 * Takes a bunch of requests and tries to merge them. Returns the number of
3599 * requests that remain after merging.
3601 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
3602 int num_reqs, MultiwriteCB *mcb)
3606 // Sort requests by start sector
3607 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
3609 // Check if adjacent requests touch the same clusters. If so, combine them,
3610 // filling up gaps with zero sectors.
3612 for (i = 1; i < num_reqs; i++) {
3614 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
3616 // Handle exactly sequential writes and overlapping writes.
3617 if (reqs[i].sector <= oldreq_last) {
3621 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3627 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3628 qemu_iovec_init(qiov,
3629 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3631 // Add the first request to the merged one. If the requests are
3632 // overlapping, drop the last sectors of the first request.
3633 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3634 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
3636 // We should need to add any zeros between the two requests
3637 assert (reqs[i].sector <= oldreq_last);
3639 // Add the second request
3640 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
3642 reqs[outidx].nb_sectors = qiov->size >> 9;
3643 reqs[outidx].qiov = qiov;
3645 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3648 reqs[outidx].sector = reqs[i].sector;
3649 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3650 reqs[outidx].qiov = reqs[i].qiov;
3658 * Submit multiple AIO write requests at once.
3660 * On success, the function returns 0 and all requests in the reqs array have
3661 * been submitted. In error case this function returns -1, and any of the
3662 * requests may or may not be submitted yet. In particular, this means that the
3663 * callback will be called for some of the requests, for others it won't. The
3664 * caller must check the error field of the BlockRequest to wait for the right
3665 * callbacks (if error != 0, no callback will be called).
3667 * The implementation may modify the contents of the reqs array, e.g. to merge
3668 * requests. However, the fields opaque and error are left unmodified as they
3669 * are used to signal failure for a single request to the caller.
3671 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3676 /* don't submit writes if we don't have a medium */
3677 if (bs->drv == NULL) {
3678 for (i = 0; i < num_reqs; i++) {
3679 reqs[i].error = -ENOMEDIUM;
3684 if (num_reqs == 0) {
3688 // Create MultiwriteCB structure
3689 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3690 mcb->num_requests = 0;
3691 mcb->num_callbacks = num_reqs;
3693 for (i = 0; i < num_reqs; i++) {
3694 mcb->callbacks[i].cb = reqs[i].cb;
3695 mcb->callbacks[i].opaque = reqs[i].opaque;
3698 // Check for mergable requests
3699 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3701 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3703 /* Run the aio requests. */
3704 mcb->num_requests = num_reqs;
3705 for (i = 0; i < num_reqs; i++) {
3706 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3707 reqs[i].nb_sectors, multiwrite_cb, mcb);
3713 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3715 acb->aiocb_info->cancel(acb);
3718 /**************************************************************/
3719 /* async block device emulation */
3721 typedef struct BlockDriverAIOCBSync {
3722 BlockDriverAIOCB common;
3725 /* vector translation state */
3729 } BlockDriverAIOCBSync;
3731 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3733 BlockDriverAIOCBSync *acb =
3734 container_of(blockacb, BlockDriverAIOCBSync, common);
3735 qemu_bh_delete(acb->bh);
3737 qemu_aio_release(acb);
3740 static const AIOCBInfo bdrv_em_aiocb_info = {
3741 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3742 .cancel = bdrv_aio_cancel_em,
3745 static void bdrv_aio_bh_cb(void *opaque)
3747 BlockDriverAIOCBSync *acb = opaque;
3750 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
3751 qemu_vfree(acb->bounce);
3752 acb->common.cb(acb->common.opaque, acb->ret);
3753 qemu_bh_delete(acb->bh);
3755 qemu_aio_release(acb);
3758 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3762 BlockDriverCompletionFunc *cb,
3767 BlockDriverAIOCBSync *acb;
3769 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
3770 acb->is_write = is_write;
3772 acb->bounce = qemu_blockalign(bs, qiov->size);
3773 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3776 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
3777 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3779 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3782 qemu_bh_schedule(acb->bh);
3784 return &acb->common;
3787 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3788 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3789 BlockDriverCompletionFunc *cb, void *opaque)
3791 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3794 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3795 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3796 BlockDriverCompletionFunc *cb, void *opaque)
3798 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3802 typedef struct BlockDriverAIOCBCoroutine {
3803 BlockDriverAIOCB common;
3808 } BlockDriverAIOCBCoroutine;
3810 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3812 BlockDriverAIOCBCoroutine *acb =
3813 container_of(blockacb, BlockDriverAIOCBCoroutine, common);
3822 static const AIOCBInfo bdrv_em_co_aiocb_info = {
3823 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3824 .cancel = bdrv_aio_co_cancel_em,
3827 static void bdrv_co_em_bh(void *opaque)
3829 BlockDriverAIOCBCoroutine *acb = opaque;
3831 acb->common.cb(acb->common.opaque, acb->req.error);
3837 qemu_bh_delete(acb->bh);
3838 qemu_aio_release(acb);
3841 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3842 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3844 BlockDriverAIOCBCoroutine *acb = opaque;
3845 BlockDriverState *bs = acb->common.bs;
3847 if (!acb->is_write) {
3848 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3849 acb->req.nb_sectors, acb->req.qiov, 0);
3851 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3852 acb->req.nb_sectors, acb->req.qiov, 0);
3855 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3856 qemu_bh_schedule(acb->bh);
3859 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3863 BlockDriverCompletionFunc *cb,
3868 BlockDriverAIOCBCoroutine *acb;
3870 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
3871 acb->req.sector = sector_num;
3872 acb->req.nb_sectors = nb_sectors;
3873 acb->req.qiov = qiov;
3874 acb->is_write = is_write;
3877 co = qemu_coroutine_create(bdrv_co_do_rw);
3878 qemu_coroutine_enter(co, acb);
3880 return &acb->common;
3883 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3885 BlockDriverAIOCBCoroutine *acb = opaque;
3886 BlockDriverState *bs = acb->common.bs;
3888 acb->req.error = bdrv_co_flush(bs);
3889 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3890 qemu_bh_schedule(acb->bh);
3893 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3894 BlockDriverCompletionFunc *cb, void *opaque)
3896 trace_bdrv_aio_flush(bs, opaque);
3899 BlockDriverAIOCBCoroutine *acb;
3901 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
3904 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3905 qemu_coroutine_enter(co, acb);
3907 return &acb->common;
3910 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3912 BlockDriverAIOCBCoroutine *acb = opaque;
3913 BlockDriverState *bs = acb->common.bs;
3915 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3916 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3917 qemu_bh_schedule(acb->bh);
3920 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3921 int64_t sector_num, int nb_sectors,
3922 BlockDriverCompletionFunc *cb, void *opaque)
3925 BlockDriverAIOCBCoroutine *acb;
3927 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3929 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
3930 acb->req.sector = sector_num;
3931 acb->req.nb_sectors = nb_sectors;
3933 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3934 qemu_coroutine_enter(co, acb);
3936 return &acb->common;
3939 void bdrv_init(void)
3941 module_call_init(MODULE_INIT_BLOCK);
3944 void bdrv_init_with_whitelist(void)
3946 use_bdrv_whitelist = 1;
3950 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
3951 BlockDriverCompletionFunc *cb, void *opaque)
3953 BlockDriverAIOCB *acb;
3955 acb = g_slice_alloc(aiocb_info->aiocb_size);
3956 acb->aiocb_info = aiocb_info;
3959 acb->opaque = opaque;
3963 void qemu_aio_release(void *p)
3965 BlockDriverAIOCB *acb = p;
3966 g_slice_free1(acb->aiocb_info->aiocb_size, acb);
3969 /**************************************************************/
3970 /* Coroutine block device emulation */
3972 typedef struct CoroutineIOCompletion {
3973 Coroutine *coroutine;
3975 } CoroutineIOCompletion;
3977 static void bdrv_co_io_em_complete(void *opaque, int ret)
3979 CoroutineIOCompletion *co = opaque;
3982 qemu_coroutine_enter(co->coroutine, NULL);
3985 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3986 int nb_sectors, QEMUIOVector *iov,
3989 CoroutineIOCompletion co = {
3990 .coroutine = qemu_coroutine_self(),
3992 BlockDriverAIOCB *acb;
3995 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3996 bdrv_co_io_em_complete, &co);
3998 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3999 bdrv_co_io_em_complete, &co);
4002 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4006 qemu_coroutine_yield();
4011 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4012 int64_t sector_num, int nb_sectors,
4015 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4018 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4019 int64_t sector_num, int nb_sectors,
4022 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4025 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4027 RwCo *rwco = opaque;
4029 rwco->ret = bdrv_co_flush(rwco->bs);
4032 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4036 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4040 /* Write back cached data to the OS even with cache=unsafe */
4041 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4042 if (bs->drv->bdrv_co_flush_to_os) {
4043 ret = bs->drv->bdrv_co_flush_to_os(bs);
4049 /* But don't actually force it to the disk with cache=unsafe */
4050 if (bs->open_flags & BDRV_O_NO_FLUSH) {
4054 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4055 if (bs->drv->bdrv_co_flush_to_disk) {
4056 ret = bs->drv->bdrv_co_flush_to_disk(bs);
4057 } else if (bs->drv->bdrv_aio_flush) {
4058 BlockDriverAIOCB *acb;
4059 CoroutineIOCompletion co = {
4060 .coroutine = qemu_coroutine_self(),
4063 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4067 qemu_coroutine_yield();
4072 * Some block drivers always operate in either writethrough or unsafe
4073 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4074 * know how the server works (because the behaviour is hardcoded or
4075 * depends on server-side configuration), so we can't ensure that
4076 * everything is safe on disk. Returning an error doesn't work because
4077 * that would break guests even if the server operates in writethrough
4080 * Let's hope the user knows what he's doing.
4088 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4089 * in the case of cache=unsafe, so there are no useless flushes.
4092 return bdrv_co_flush(bs->file);
4095 void bdrv_invalidate_cache(BlockDriverState *bs)
4097 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
4098 bs->drv->bdrv_invalidate_cache(bs);
4102 void bdrv_invalidate_cache_all(void)
4104 BlockDriverState *bs;
4106 QTAILQ_FOREACH(bs, &bdrv_states, list) {
4107 bdrv_invalidate_cache(bs);
4111 void bdrv_clear_incoming_migration_all(void)
4113 BlockDriverState *bs;
4115 QTAILQ_FOREACH(bs, &bdrv_states, list) {
4116 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
4120 int bdrv_flush(BlockDriverState *bs)
4128 if (qemu_in_coroutine()) {
4129 /* Fast-path if already in coroutine context */
4130 bdrv_flush_co_entry(&rwco);
4132 co = qemu_coroutine_create(bdrv_flush_co_entry);
4133 qemu_coroutine_enter(co, &rwco);
4134 while (rwco.ret == NOT_DONE) {
4142 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
4144 RwCo *rwco = opaque;
4146 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
4149 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
4154 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
4156 } else if (bs->read_only) {
4160 if (bs->dirty_bitmap) {
4161 bdrv_reset_dirty(bs, sector_num, nb_sectors);
4164 /* Do nothing if disabled. */
4165 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4169 if (bs->drv->bdrv_co_discard) {
4170 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
4171 } else if (bs->drv->bdrv_aio_discard) {
4172 BlockDriverAIOCB *acb;
4173 CoroutineIOCompletion co = {
4174 .coroutine = qemu_coroutine_self(),
4177 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
4178 bdrv_co_io_em_complete, &co);
4182 qemu_coroutine_yield();
4190 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
4195 .sector_num = sector_num,
4196 .nb_sectors = nb_sectors,
4200 if (qemu_in_coroutine()) {
4201 /* Fast-path if already in coroutine context */
4202 bdrv_discard_co_entry(&rwco);
4204 co = qemu_coroutine_create(bdrv_discard_co_entry);
4205 qemu_coroutine_enter(co, &rwco);
4206 while (rwco.ret == NOT_DONE) {
4214 /**************************************************************/
4215 /* removable device support */
4218 * Return TRUE if the media is present
4220 int bdrv_is_inserted(BlockDriverState *bs)
4222 BlockDriver *drv = bs->drv;
4226 if (!drv->bdrv_is_inserted)
4228 return drv->bdrv_is_inserted(bs);
4232 * Return whether the media changed since the last call to this
4233 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4235 int bdrv_media_changed(BlockDriverState *bs)
4237 BlockDriver *drv = bs->drv;
4239 if (drv && drv->bdrv_media_changed) {
4240 return drv->bdrv_media_changed(bs);
4246 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4248 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4250 BlockDriver *drv = bs->drv;
4252 if (drv && drv->bdrv_eject) {
4253 drv->bdrv_eject(bs, eject_flag);
4256 if (bs->device_name[0] != '\0') {
4257 bdrv_emit_qmp_eject_event(bs, eject_flag);
4262 * Lock or unlock the media (if it is locked, the user won't be able
4263 * to eject it manually).
4265 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4267 BlockDriver *drv = bs->drv;
4269 trace_bdrv_lock_medium(bs, locked);
4271 if (drv && drv->bdrv_lock_medium) {
4272 drv->bdrv_lock_medium(bs, locked);
4276 /* needed for generic scsi interface */
4278 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
4280 BlockDriver *drv = bs->drv;
4282 if (drv && drv->bdrv_ioctl)
4283 return drv->bdrv_ioctl(bs, req, buf);
4287 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
4288 unsigned long int req, void *buf,
4289 BlockDriverCompletionFunc *cb, void *opaque)
4291 BlockDriver *drv = bs->drv;
4293 if (drv && drv->bdrv_aio_ioctl)
4294 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
4298 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
4300 bs->buffer_alignment = align;
4303 void *qemu_blockalign(BlockDriverState *bs, size_t size)
4305 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
4309 * Check if all memory in this vector is sector aligned.
4311 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
4315 for (i = 0; i < qiov->niov; i++) {
4316 if ((uintptr_t) qiov->iov[i].iov_base % bs->buffer_alignment) {
4324 void bdrv_set_dirty_tracking(BlockDriverState *bs, int granularity)
4326 int64_t bitmap_size;
4328 assert((granularity & (granularity - 1)) == 0);
4331 granularity >>= BDRV_SECTOR_BITS;
4332 assert(!bs->dirty_bitmap);
4333 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS);
4334 bs->dirty_bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
4336 if (bs->dirty_bitmap) {
4337 hbitmap_free(bs->dirty_bitmap);
4338 bs->dirty_bitmap = NULL;
4343 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
4345 if (bs->dirty_bitmap) {
4346 return hbitmap_get(bs->dirty_bitmap, sector);
4352 void bdrv_dirty_iter_init(BlockDriverState *bs, HBitmapIter *hbi)
4354 hbitmap_iter_init(hbi, bs->dirty_bitmap, 0);
4357 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
4360 hbitmap_set(bs->dirty_bitmap, cur_sector, nr_sectors);
4363 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
4366 hbitmap_reset(bs->dirty_bitmap, cur_sector, nr_sectors);
4369 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
4371 if (bs->dirty_bitmap) {
4372 return hbitmap_count(bs->dirty_bitmap);
4378 /* Get a reference to bs */
4379 void bdrv_ref(BlockDriverState *bs)
4384 /* Release a previously grabbed reference to bs.
4385 * If after releasing, reference count is zero, the BlockDriverState is
4387 void bdrv_unref(BlockDriverState *bs)
4389 assert(bs->refcnt > 0);
4390 if (--bs->refcnt == 0) {
4395 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
4397 assert(bs->in_use != in_use);
4398 bs->in_use = in_use;
4401 int bdrv_in_use(BlockDriverState *bs)
4406 void bdrv_iostatus_enable(BlockDriverState *bs)
4408 bs->iostatus_enabled = true;
4409 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4412 /* The I/O status is only enabled if the drive explicitly
4413 * enables it _and_ the VM is configured to stop on errors */
4414 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
4416 return (bs->iostatus_enabled &&
4417 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
4418 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
4419 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
4422 void bdrv_iostatus_disable(BlockDriverState *bs)
4424 bs->iostatus_enabled = false;
4427 void bdrv_iostatus_reset(BlockDriverState *bs)
4429 if (bdrv_iostatus_is_enabled(bs)) {
4430 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4432 block_job_iostatus_reset(bs->job);
4437 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
4439 assert(bdrv_iostatus_is_enabled(bs));
4440 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
4441 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
4442 BLOCK_DEVICE_IO_STATUS_FAILED;
4447 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
4448 enum BlockAcctType type)
4450 assert(type < BDRV_MAX_IOTYPE);
4452 cookie->bytes = bytes;
4453 cookie->start_time_ns = get_clock();
4454 cookie->type = type;
4458 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
4460 assert(cookie->type < BDRV_MAX_IOTYPE);
4462 bs->nr_bytes[cookie->type] += cookie->bytes;
4463 bs->nr_ops[cookie->type]++;
4464 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
4467 void bdrv_img_create(const char *filename, const char *fmt,
4468 const char *base_filename, const char *base_fmt,
4469 char *options, uint64_t img_size, int flags,
4470 Error **errp, bool quiet)
4472 QEMUOptionParameter *param = NULL, *create_options = NULL;
4473 QEMUOptionParameter *backing_fmt, *backing_file, *size;
4474 BlockDriverState *bs = NULL;
4475 BlockDriver *drv, *proto_drv;
4476 BlockDriver *backing_drv = NULL;
4477 Error *local_err = NULL;
4480 /* Find driver and parse its options */
4481 drv = bdrv_find_format(fmt);
4483 error_setg(errp, "Unknown file format '%s'", fmt);
4487 proto_drv = bdrv_find_protocol(filename, true);
4489 error_setg(errp, "Unknown protocol '%s'", filename);
4493 create_options = append_option_parameters(create_options,
4494 drv->create_options);
4495 create_options = append_option_parameters(create_options,
4496 proto_drv->create_options);
4498 /* Create parameter list with default values */
4499 param = parse_option_parameters("", create_options, param);
4501 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
4503 /* Parse -o options */
4505 param = parse_option_parameters(options, create_options, param);
4506 if (param == NULL) {
4507 error_setg(errp, "Invalid options for file format '%s'.", fmt);
4512 if (base_filename) {
4513 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
4515 error_setg(errp, "Backing file not supported for file format '%s'",
4522 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4523 error_setg(errp, "Backing file format not supported for file "
4524 "format '%s'", fmt);
4529 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4530 if (backing_file && backing_file->value.s) {
4531 if (!strcmp(filename, backing_file->value.s)) {
4532 error_setg(errp, "Error: Trying to create an image with the "
4533 "same filename as the backing file");
4538 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4539 if (backing_fmt && backing_fmt->value.s) {
4540 backing_drv = bdrv_find_format(backing_fmt->value.s);
4542 error_setg(errp, "Unknown backing file format '%s'",
4543 backing_fmt->value.s);
4548 // The size for the image must always be specified, with one exception:
4549 // If we are using a backing file, we can obtain the size from there
4550 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4551 if (size && size->value.n == -1) {
4552 if (backing_file && backing_file->value.s) {
4557 /* backing files always opened read-only */
4559 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4563 ret = bdrv_open(bs, backing_file->value.s, NULL, back_flags,
4564 backing_drv, &local_err);
4566 error_setg_errno(errp, -ret, "Could not open '%s': %s",
4567 backing_file->value.s,
4568 error_get_pretty(local_err));
4569 error_free(local_err);
4573 bdrv_get_geometry(bs, &size);
4576 snprintf(buf, sizeof(buf), "%" PRId64, size);
4577 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4579 error_setg(errp, "Image creation needs a size parameter");
4585 printf("Formatting '%s', fmt=%s ", filename, fmt);
4586 print_option_parameters(param);
4589 ret = bdrv_create(drv, filename, param, &local_err);
4590 if (ret == -EFBIG) {
4591 /* This is generally a better message than whatever the driver would
4592 * deliver (especially because of the cluster_size_hint), since that
4593 * is most probably not much different from "image too large". */
4594 const char *cluster_size_hint = "";
4595 if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) {
4596 cluster_size_hint = " (try using a larger cluster size)";
4598 error_setg(errp, "The image size is too large for file format '%s'"
4599 "%s", fmt, cluster_size_hint);
4600 error_free(local_err);
4605 free_option_parameters(create_options);
4606 free_option_parameters(param);
4611 if (error_is_set(&local_err)) {
4612 error_propagate(errp, local_err);
4616 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4618 /* Currently BlockDriverState always uses the main loop AioContext */
4619 return qemu_get_aio_context();
4622 void bdrv_add_before_write_notifier(BlockDriverState *bs,
4623 NotifierWithReturn *notifier)
4625 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
4628 int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options)
4630 if (bs->drv->bdrv_amend_options == NULL) {
4633 return bs->drv->bdrv_amend_options(bs, options);