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"
28 #include "block_int.h"
30 #include "qemu-objects.h"
33 #include <sys/types.h>
35 #include <sys/ioctl.h>
36 #include <sys/queue.h>
46 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
47 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
48 BlockDriverCompletionFunc *cb, void *opaque);
49 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
50 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
51 BlockDriverCompletionFunc *cb, void *opaque);
52 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
53 BlockDriverCompletionFunc *cb, void *opaque);
54 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
55 BlockDriverCompletionFunc *cb, void *opaque);
56 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
57 uint8_t *buf, int nb_sectors);
58 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
59 const uint8_t *buf, int nb_sectors);
61 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(bdrv_states);
64 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
65 QLIST_HEAD_INITIALIZER(bdrv_drivers);
67 /* The device to use for VM snapshots */
68 static BlockDriverState *bs_snapshots;
70 /* If non-zero, use only whitelisted block drivers */
71 static int use_bdrv_whitelist;
74 static int is_windows_drive_prefix(const char *filename)
76 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
77 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
81 int is_windows_drive(const char *filename)
83 if (is_windows_drive_prefix(filename) &&
86 if (strstart(filename, "\\\\.\\", NULL) ||
87 strstart(filename, "//./", NULL))
93 /* check if the path starts with "<protocol>:" */
94 static int path_has_protocol(const char *path)
97 if (is_windows_drive(path) ||
98 is_windows_drive_prefix(path)) {
103 return strchr(path, ':') != NULL;
106 int path_is_absolute(const char *path)
110 /* specific case for names like: "\\.\d:" */
111 if (*path == '/' || *path == '\\')
114 p = strchr(path, ':');
120 return (*p == '/' || *p == '\\');
126 /* if filename is absolute, just copy it to dest. Otherwise, build a
127 path to it by considering it is relative to base_path. URL are
129 void path_combine(char *dest, int dest_size,
130 const char *base_path,
131 const char *filename)
138 if (path_is_absolute(filename)) {
139 pstrcpy(dest, dest_size, filename);
141 p = strchr(base_path, ':');
146 p1 = strrchr(base_path, '/');
150 p2 = strrchr(base_path, '\\');
162 if (len > dest_size - 1)
164 memcpy(dest, base_path, len);
166 pstrcat(dest, dest_size, filename);
170 void bdrv_register(BlockDriver *bdrv)
172 if (!bdrv->bdrv_aio_readv) {
173 /* add AIO emulation layer */
174 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
175 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
176 } else if (!bdrv->bdrv_read) {
177 /* add synchronous IO emulation layer */
178 bdrv->bdrv_read = bdrv_read_em;
179 bdrv->bdrv_write = bdrv_write_em;
182 if (!bdrv->bdrv_aio_flush)
183 bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
185 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
188 /* create a new block device (by default it is empty) */
189 BlockDriverState *bdrv_new(const char *device_name)
191 BlockDriverState *bs;
193 bs = qemu_mallocz(sizeof(BlockDriverState));
194 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
195 if (device_name[0] != '\0') {
196 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
201 BlockDriver *bdrv_find_format(const char *format_name)
204 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
205 if (!strcmp(drv1->format_name, format_name)) {
212 static int bdrv_is_whitelisted(BlockDriver *drv)
214 static const char *whitelist[] = {
215 CONFIG_BDRV_WHITELIST
220 return 1; /* no whitelist, anything goes */
222 for (p = whitelist; *p; p++) {
223 if (!strcmp(drv->format_name, *p)) {
230 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
232 BlockDriver *drv = bdrv_find_format(format_name);
233 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
236 int bdrv_create(BlockDriver *drv, const char* filename,
237 QEMUOptionParameter *options)
239 if (!drv->bdrv_create)
242 return drv->bdrv_create(filename, options);
245 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
249 drv = bdrv_find_protocol(filename);
254 return bdrv_create(drv, filename, options);
258 void get_tmp_filename(char *filename, int size)
260 char temp_dir[MAX_PATH];
262 GetTempPath(MAX_PATH, temp_dir);
263 GetTempFileName(temp_dir, "qem", 0, filename);
266 void get_tmp_filename(char *filename, int size)
270 /* XXX: race condition possible */
271 tmpdir = getenv("TMPDIR");
274 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
275 fd = mkstemp(filename);
281 * Detect host devices. By convention, /dev/cdrom[N] is always
282 * recognized as a host CDROM.
284 static BlockDriver *find_hdev_driver(const char *filename)
286 int score_max = 0, score;
287 BlockDriver *drv = NULL, *d;
289 QLIST_FOREACH(d, &bdrv_drivers, list) {
290 if (d->bdrv_probe_device) {
291 score = d->bdrv_probe_device(filename);
292 if (score > score_max) {
302 BlockDriver *bdrv_find_protocol(const char *filename)
309 /* TODO Drivers without bdrv_file_open must be specified explicitly */
312 * XXX(hch): we really should not let host device detection
313 * override an explicit protocol specification, but moving this
314 * later breaks access to device names with colons in them.
315 * Thanks to the brain-dead persistent naming schemes on udev-
316 * based Linux systems those actually are quite common.
318 drv1 = find_hdev_driver(filename);
323 if (!path_has_protocol(filename)) {
324 return bdrv_find_format("file");
326 p = strchr(filename, ':');
329 if (len > sizeof(protocol) - 1)
330 len = sizeof(protocol) - 1;
331 memcpy(protocol, filename, len);
332 protocol[len] = '\0';
333 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
334 if (drv1->protocol_name &&
335 !strcmp(drv1->protocol_name, protocol)) {
342 static int find_image_format(const char *filename, BlockDriver **pdrv)
344 int ret, score, score_max;
345 BlockDriver *drv1, *drv;
347 BlockDriverState *bs;
349 ret = bdrv_file_open(&bs, filename, 0);
355 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
356 if (bs->sg || !bdrv_is_inserted(bs)) {
358 drv = bdrv_find_format("raw");
366 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
375 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
376 if (drv1->bdrv_probe) {
377 score = drv1->bdrv_probe(buf, ret, filename);
378 if (score > score_max) {
392 * Set the current 'total_sectors' value
394 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
396 BlockDriver *drv = bs->drv;
398 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
402 /* query actual device if possible, otherwise just trust the hint */
403 if (drv->bdrv_getlength) {
404 int64_t length = drv->bdrv_getlength(bs);
408 hint = length >> BDRV_SECTOR_BITS;
411 bs->total_sectors = hint;
416 * Common part for opening disk images and files
418 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
419 int flags, BlockDriver *drv)
426 bs->total_sectors = 0;
429 bs->open_flags = flags;
430 /* buffer_alignment defaulted to 512, drivers can change this value */
431 bs->buffer_alignment = 512;
433 pstrcpy(bs->filename, sizeof(bs->filename), filename);
435 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
440 bs->opaque = qemu_mallocz(drv->instance_size);
443 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
444 * write cache to the guest. We do need the fdatasync to flush
445 * out transactions for block allocations, and we maybe have a
446 * volatile write cache in our backing device to deal with.
448 if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
449 bs->enable_write_cache = 1;
452 * Clear flags that are internal to the block layer before opening the
455 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
458 * Snapshots should be writeable.
460 if (bs->is_temporary) {
461 open_flags |= BDRV_O_RDWR;
464 /* Open the image, either directly or using a protocol */
465 if (drv->bdrv_file_open) {
466 ret = drv->bdrv_file_open(bs, filename, open_flags);
468 ret = bdrv_file_open(&bs->file, filename, open_flags);
470 ret = drv->bdrv_open(bs, open_flags);
478 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
480 ret = refresh_total_sectors(bs, bs->total_sectors);
486 if (bs->is_temporary) {
494 bdrv_delete(bs->file);
497 qemu_free(bs->opaque);
504 * Opens a file using a protocol (file, host_device, nbd, ...)
506 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
508 BlockDriverState *bs;
512 drv = bdrv_find_protocol(filename);
518 ret = bdrv_open_common(bs, filename, flags, drv);
529 * Opens a disk image (raw, qcow2, vmdk, ...)
531 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
536 if (flags & BDRV_O_SNAPSHOT) {
537 BlockDriverState *bs1;
540 BlockDriver *bdrv_qcow2;
541 QEMUOptionParameter *options;
542 char tmp_filename[PATH_MAX];
543 char backing_filename[PATH_MAX];
545 /* if snapshot, we create a temporary backing file and open it
546 instead of opening 'filename' directly */
548 /* if there is a backing file, use it */
550 ret = bdrv_open(bs1, filename, 0, drv);
555 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
557 if (bs1->drv && bs1->drv->protocol_name)
562 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
564 /* Real path is meaningless for protocols */
566 snprintf(backing_filename, sizeof(backing_filename),
568 else if (!realpath(filename, backing_filename))
571 bdrv_qcow2 = bdrv_find_format("qcow2");
572 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
574 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
575 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
577 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
581 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
582 free_option_parameters(options);
587 filename = tmp_filename;
589 bs->is_temporary = 1;
592 /* Find the right image format driver */
594 ret = find_image_format(filename, &drv);
598 goto unlink_and_fail;
602 ret = bdrv_open_common(bs, filename, flags, drv);
604 goto unlink_and_fail;
607 /* If there is a backing file, use it */
608 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
609 char backing_filename[PATH_MAX];
611 BlockDriver *back_drv = NULL;
613 bs->backing_hd = bdrv_new("");
615 if (path_has_protocol(bs->backing_file)) {
616 pstrcpy(backing_filename, sizeof(backing_filename),
619 path_combine(backing_filename, sizeof(backing_filename),
620 filename, bs->backing_file);
623 if (bs->backing_format[0] != '\0') {
624 back_drv = bdrv_find_format(bs->backing_format);
627 /* backing files always opened read-only */
629 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
631 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
636 if (bs->is_temporary) {
637 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
639 /* base image inherits from "parent" */
640 bs->backing_hd->keep_read_only = bs->keep_read_only;
644 if (!bdrv_key_required(bs)) {
645 /* call the change callback */
646 bs->media_changed = 1;
648 bs->change_cb(bs->change_opaque);
654 if (bs->is_temporary) {
660 void bdrv_close(BlockDriverState *bs)
663 if (bs == bs_snapshots) {
666 if (bs->backing_hd) {
667 bdrv_delete(bs->backing_hd);
668 bs->backing_hd = NULL;
670 bs->drv->bdrv_close(bs);
671 qemu_free(bs->opaque);
673 if (bs->is_temporary) {
674 unlink(bs->filename);
680 if (bs->file != NULL) {
681 bdrv_close(bs->file);
684 /* call the change callback */
685 bs->media_changed = 1;
687 bs->change_cb(bs->change_opaque);
691 void bdrv_close_all(void)
693 BlockDriverState *bs;
695 QTAILQ_FOREACH(bs, &bdrv_states, list) {
700 void bdrv_delete(BlockDriverState *bs)
704 /* remove from list, if necessary */
705 if (bs->device_name[0] != '\0') {
706 QTAILQ_REMOVE(&bdrv_states, bs, list);
710 if (bs->file != NULL) {
711 bdrv_delete(bs->file);
714 assert(bs != bs_snapshots);
718 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
727 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
729 assert(bs->peer == qdev);
733 DeviceState *bdrv_get_attached(BlockDriverState *bs)
739 * Run consistency checks on an image
741 * Returns 0 if the check could be completed (it doesn't mean that the image is
742 * free of errors) or -errno when an internal error occured. The results of the
743 * check are stored in res.
745 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
747 if (bs->drv->bdrv_check == NULL) {
751 memset(res, 0, sizeof(*res));
752 return bs->drv->bdrv_check(bs, res);
755 #define COMMIT_BUF_SECTORS 2048
757 /* commit COW file into the raw image */
758 int bdrv_commit(BlockDriverState *bs)
760 BlockDriver *drv = bs->drv;
761 BlockDriver *backing_drv;
762 int64_t sector, total_sectors;
763 int n, ro, open_flags;
764 int ret = 0, rw_ret = 0;
767 BlockDriverState *bs_rw, *bs_ro;
772 if (!bs->backing_hd) {
776 if (bs->backing_hd->keep_read_only) {
780 backing_drv = bs->backing_hd->drv;
781 ro = bs->backing_hd->read_only;
782 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
783 open_flags = bs->backing_hd->open_flags;
787 bdrv_delete(bs->backing_hd);
788 bs->backing_hd = NULL;
789 bs_rw = bdrv_new("");
790 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
794 /* try to re-open read-only */
795 bs_ro = bdrv_new("");
796 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
800 /* drive not functional anymore */
804 bs->backing_hd = bs_ro;
807 bs->backing_hd = bs_rw;
810 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
811 buf = qemu_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
813 for (sector = 0; sector < total_sectors; sector += n) {
814 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
816 if (bdrv_read(bs, sector, buf, n) != 0) {
821 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
828 if (drv->bdrv_make_empty) {
829 ret = drv->bdrv_make_empty(bs);
834 * Make sure all data we wrote to the backing device is actually
838 bdrv_flush(bs->backing_hd);
845 bdrv_delete(bs->backing_hd);
846 bs->backing_hd = NULL;
847 bs_ro = bdrv_new("");
848 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
852 /* drive not functional anymore */
856 bs->backing_hd = bs_ro;
857 bs->backing_hd->keep_read_only = 0;
863 void bdrv_commit_all(void)
865 BlockDriverState *bs;
867 QTAILQ_FOREACH(bs, &bdrv_states, list) {
875 * -EINVAL - backing format specified, but no file
876 * -ENOSPC - can't update the backing file because no space is left in the
878 * -ENOTSUP - format driver doesn't support changing the backing file
880 int bdrv_change_backing_file(BlockDriverState *bs,
881 const char *backing_file, const char *backing_fmt)
883 BlockDriver *drv = bs->drv;
885 if (drv->bdrv_change_backing_file != NULL) {
886 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
892 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
897 if (!bdrv_is_inserted(bs))
903 len = bdrv_getlength(bs);
908 if ((offset > len) || (len - offset < size))
914 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
917 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
918 nb_sectors * BDRV_SECTOR_SIZE);
921 /* return < 0 if error. See bdrv_write() for the return codes */
922 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
923 uint8_t *buf, int nb_sectors)
925 BlockDriver *drv = bs->drv;
929 if (bdrv_check_request(bs, sector_num, nb_sectors))
932 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
935 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
936 int nb_sectors, int dirty)
939 unsigned long val, idx, bit;
941 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
942 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
944 for (; start <= end; start++) {
945 idx = start / (sizeof(unsigned long) * 8);
946 bit = start % (sizeof(unsigned long) * 8);
947 val = bs->dirty_bitmap[idx];
949 if (!(val & (1UL << bit))) {
954 if (val & (1UL << bit)) {
956 val &= ~(1UL << bit);
959 bs->dirty_bitmap[idx] = val;
963 /* Return < 0 if error. Important errors are:
964 -EIO generic I/O error (may happen for all errors)
965 -ENOMEDIUM No media inserted.
966 -EINVAL Invalid sector number or nb_sectors
967 -EACCES Trying to write a read-only device
969 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
970 const uint8_t *buf, int nb_sectors)
972 BlockDriver *drv = bs->drv;
977 if (bdrv_check_request(bs, sector_num, nb_sectors))
980 if (bs->dirty_bitmap) {
981 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
984 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
985 bs->wr_highest_sector = sector_num + nb_sectors - 1;
988 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
991 int bdrv_pread(BlockDriverState *bs, int64_t offset,
992 void *buf, int count1)
994 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
995 int len, nb_sectors, count;
1000 /* first read to align to sector start */
1001 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1004 sector_num = offset >> BDRV_SECTOR_BITS;
1006 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1008 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1016 /* read the sectors "in place" */
1017 nb_sectors = count >> BDRV_SECTOR_BITS;
1018 if (nb_sectors > 0) {
1019 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1021 sector_num += nb_sectors;
1022 len = nb_sectors << BDRV_SECTOR_BITS;
1027 /* add data from the last sector */
1029 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1031 memcpy(buf, tmp_buf, count);
1036 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1037 const void *buf, int count1)
1039 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1040 int len, nb_sectors, count;
1045 /* first write to align to sector start */
1046 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1049 sector_num = offset >> BDRV_SECTOR_BITS;
1051 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1053 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1054 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1063 /* write the sectors "in place" */
1064 nb_sectors = count >> BDRV_SECTOR_BITS;
1065 if (nb_sectors > 0) {
1066 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1068 sector_num += nb_sectors;
1069 len = nb_sectors << BDRV_SECTOR_BITS;
1074 /* add data from the last sector */
1076 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1078 memcpy(tmp_buf, buf, count);
1079 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1086 * Writes to the file and ensures that no writes are reordered across this
1087 * request (acts as a barrier)
1089 * Returns 0 on success, -errno in error cases.
1091 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1092 const void *buf, int count)
1096 ret = bdrv_pwrite(bs, offset, buf, count);
1101 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1102 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1110 * Writes to the file and ensures that no writes are reordered across this
1111 * request (acts as a barrier)
1113 * Returns 0 on success, -errno in error cases.
1115 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1116 const uint8_t *buf, int nb_sectors)
1118 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1119 buf, BDRV_SECTOR_SIZE * nb_sectors);
1123 * Truncate file to 'offset' bytes (needed only for file protocols)
1125 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1127 BlockDriver *drv = bs->drv;
1131 if (!drv->bdrv_truncate)
1135 ret = drv->bdrv_truncate(bs, offset);
1137 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1143 * Length of a file in bytes. Return < 0 if error or unknown.
1145 int64_t bdrv_getlength(BlockDriverState *bs)
1147 BlockDriver *drv = bs->drv;
1151 /* Fixed size devices use the total_sectors value for speed instead of
1152 issuing a length query (like lseek) on each call. Also, legacy block
1153 drivers don't provide a bdrv_getlength function and must use
1155 if (!bs->growable || !drv->bdrv_getlength) {
1156 return bs->total_sectors * BDRV_SECTOR_SIZE;
1158 return drv->bdrv_getlength(bs);
1161 /* return 0 as number of sectors if no device present or error */
1162 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1165 length = bdrv_getlength(bs);
1169 length = length >> BDRV_SECTOR_BITS;
1170 *nb_sectors_ptr = length;
1174 uint8_t boot_ind; /* 0x80 - active */
1175 uint8_t head; /* starting head */
1176 uint8_t sector; /* starting sector */
1177 uint8_t cyl; /* starting cylinder */
1178 uint8_t sys_ind; /* What partition type */
1179 uint8_t end_head; /* end head */
1180 uint8_t end_sector; /* end sector */
1181 uint8_t end_cyl; /* end cylinder */
1182 uint32_t start_sect; /* starting sector counting from 0 */
1183 uint32_t nr_sects; /* nr of sectors in partition */
1184 } __attribute__((packed));
1186 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1187 static int guess_disk_lchs(BlockDriverState *bs,
1188 int *pcylinders, int *pheads, int *psectors)
1190 uint8_t buf[BDRV_SECTOR_SIZE];
1191 int ret, i, heads, sectors, cylinders;
1192 struct partition *p;
1194 uint64_t nb_sectors;
1196 bdrv_get_geometry(bs, &nb_sectors);
1198 ret = bdrv_read(bs, 0, buf, 1);
1201 /* test msdos magic */
1202 if (buf[510] != 0x55 || buf[511] != 0xaa)
1204 for(i = 0; i < 4; i++) {
1205 p = ((struct partition *)(buf + 0x1be)) + i;
1206 nr_sects = le32_to_cpu(p->nr_sects);
1207 if (nr_sects && p->end_head) {
1208 /* We make the assumption that the partition terminates on
1209 a cylinder boundary */
1210 heads = p->end_head + 1;
1211 sectors = p->end_sector & 63;
1214 cylinders = nb_sectors / (heads * sectors);
1215 if (cylinders < 1 || cylinders > 16383)
1218 *psectors = sectors;
1219 *pcylinders = cylinders;
1221 printf("guessed geometry: LCHS=%d %d %d\n",
1222 cylinders, heads, sectors);
1230 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1232 int translation, lba_detected = 0;
1233 int cylinders, heads, secs;
1234 uint64_t nb_sectors;
1236 /* if a geometry hint is available, use it */
1237 bdrv_get_geometry(bs, &nb_sectors);
1238 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1239 translation = bdrv_get_translation_hint(bs);
1240 if (cylinders != 0) {
1245 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1247 /* if heads > 16, it means that a BIOS LBA
1248 translation was active, so the default
1249 hardware geometry is OK */
1251 goto default_geometry;
1256 /* disable any translation to be in sync with
1257 the logical geometry */
1258 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1259 bdrv_set_translation_hint(bs,
1260 BIOS_ATA_TRANSLATION_NONE);
1265 /* if no geometry, use a standard physical disk geometry */
1266 cylinders = nb_sectors / (16 * 63);
1268 if (cylinders > 16383)
1270 else if (cylinders < 2)
1275 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1276 if ((*pcyls * *pheads) <= 131072) {
1277 bdrv_set_translation_hint(bs,
1278 BIOS_ATA_TRANSLATION_LARGE);
1280 bdrv_set_translation_hint(bs,
1281 BIOS_ATA_TRANSLATION_LBA);
1285 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1289 void bdrv_set_geometry_hint(BlockDriverState *bs,
1290 int cyls, int heads, int secs)
1297 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1300 bs->removable = ((type == BDRV_TYPE_CDROM ||
1301 type == BDRV_TYPE_FLOPPY));
1304 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1306 bs->translation = translation;
1309 void bdrv_get_geometry_hint(BlockDriverState *bs,
1310 int *pcyls, int *pheads, int *psecs)
1313 *pheads = bs->heads;
1317 int bdrv_get_type_hint(BlockDriverState *bs)
1322 int bdrv_get_translation_hint(BlockDriverState *bs)
1324 return bs->translation;
1327 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1328 BlockErrorAction on_write_error)
1330 bs->on_read_error = on_read_error;
1331 bs->on_write_error = on_write_error;
1334 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1336 return is_read ? bs->on_read_error : bs->on_write_error;
1339 void bdrv_set_removable(BlockDriverState *bs, int removable)
1341 bs->removable = removable;
1342 if (removable && bs == bs_snapshots) {
1343 bs_snapshots = NULL;
1347 int bdrv_is_removable(BlockDriverState *bs)
1349 return bs->removable;
1352 int bdrv_is_read_only(BlockDriverState *bs)
1354 return bs->read_only;
1357 int bdrv_is_sg(BlockDriverState *bs)
1362 int bdrv_enable_write_cache(BlockDriverState *bs)
1364 return bs->enable_write_cache;
1367 /* XXX: no longer used */
1368 void bdrv_set_change_cb(BlockDriverState *bs,
1369 void (*change_cb)(void *opaque), void *opaque)
1371 bs->change_cb = change_cb;
1372 bs->change_opaque = opaque;
1375 int bdrv_is_encrypted(BlockDriverState *bs)
1377 if (bs->backing_hd && bs->backing_hd->encrypted)
1379 return bs->encrypted;
1382 int bdrv_key_required(BlockDriverState *bs)
1384 BlockDriverState *backing_hd = bs->backing_hd;
1386 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1388 return (bs->encrypted && !bs->valid_key);
1391 int bdrv_set_key(BlockDriverState *bs, const char *key)
1394 if (bs->backing_hd && bs->backing_hd->encrypted) {
1395 ret = bdrv_set_key(bs->backing_hd, key);
1401 if (!bs->encrypted) {
1403 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1406 ret = bs->drv->bdrv_set_key(bs, key);
1409 } else if (!bs->valid_key) {
1411 /* call the change callback now, we skipped it on open */
1412 bs->media_changed = 1;
1414 bs->change_cb(bs->change_opaque);
1419 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1424 pstrcpy(buf, buf_size, bs->drv->format_name);
1428 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1433 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1434 it(opaque, drv->format_name);
1438 BlockDriverState *bdrv_find(const char *name)
1440 BlockDriverState *bs;
1442 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1443 if (!strcmp(name, bs->device_name)) {
1450 BlockDriverState *bdrv_next(BlockDriverState *bs)
1453 return QTAILQ_FIRST(&bdrv_states);
1455 return QTAILQ_NEXT(bs, list);
1458 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1460 BlockDriverState *bs;
1462 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1467 const char *bdrv_get_device_name(BlockDriverState *bs)
1469 return bs->device_name;
1472 int bdrv_flush(BlockDriverState *bs)
1474 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1478 if (bs->drv && bs->drv->bdrv_flush) {
1479 return bs->drv->bdrv_flush(bs);
1483 * Some block drivers always operate in either writethrough or unsafe mode
1484 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1485 * the server works (because the behaviour is hardcoded or depends on
1486 * server-side configuration), so we can't ensure that everything is safe
1487 * on disk. Returning an error doesn't work because that would break guests
1488 * even if the server operates in writethrough mode.
1490 * Let's hope the user knows what he's doing.
1495 void bdrv_flush_all(void)
1497 BlockDriverState *bs;
1499 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1500 if (bs->drv && !bdrv_is_read_only(bs) &&
1501 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1507 int bdrv_has_zero_init(BlockDriverState *bs)
1511 if (bs->drv->bdrv_has_zero_init) {
1512 return bs->drv->bdrv_has_zero_init(bs);
1518 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
1523 if (!bs->drv->bdrv_discard) {
1526 return bs->drv->bdrv_discard(bs, sector_num, nb_sectors);
1530 * Returns true iff the specified sector is present in the disk image. Drivers
1531 * not implementing the functionality are assumed to not support backing files,
1532 * hence all their sectors are reported as allocated.
1534 * 'pnum' is set to the number of sectors (including and immediately following
1535 * the specified sector) that are known to be in the same
1536 * allocated/unallocated state.
1538 * 'nb_sectors' is the max value 'pnum' should be set to.
1540 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1544 if (!bs->drv->bdrv_is_allocated) {
1545 if (sector_num >= bs->total_sectors) {
1549 n = bs->total_sectors - sector_num;
1550 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1553 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1556 void bdrv_mon_event(const BlockDriverState *bdrv,
1557 BlockMonEventAction action, int is_read)
1560 const char *action_str;
1563 case BDRV_ACTION_REPORT:
1564 action_str = "report";
1566 case BDRV_ACTION_IGNORE:
1567 action_str = "ignore";
1569 case BDRV_ACTION_STOP:
1570 action_str = "stop";
1576 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1579 is_read ? "read" : "write");
1580 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1582 qobject_decref(data);
1585 static void bdrv_print_dict(QObject *obj, void *opaque)
1588 Monitor *mon = opaque;
1590 bs_dict = qobject_to_qdict(obj);
1592 monitor_printf(mon, "%s: type=%s removable=%d",
1593 qdict_get_str(bs_dict, "device"),
1594 qdict_get_str(bs_dict, "type"),
1595 qdict_get_bool(bs_dict, "removable"));
1597 if (qdict_get_bool(bs_dict, "removable")) {
1598 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1601 if (qdict_haskey(bs_dict, "inserted")) {
1602 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1604 monitor_printf(mon, " file=");
1605 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1606 if (qdict_haskey(qdict, "backing_file")) {
1607 monitor_printf(mon, " backing_file=");
1608 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1610 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1611 qdict_get_bool(qdict, "ro"),
1612 qdict_get_str(qdict, "drv"),
1613 qdict_get_bool(qdict, "encrypted"));
1615 monitor_printf(mon, " [not inserted]");
1618 monitor_printf(mon, "\n");
1621 void bdrv_info_print(Monitor *mon, const QObject *data)
1623 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1626 void bdrv_info(Monitor *mon, QObject **ret_data)
1629 BlockDriverState *bs;
1631 bs_list = qlist_new();
1633 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1635 const char *type = "unknown";
1641 case BDRV_TYPE_CDROM:
1644 case BDRV_TYPE_FLOPPY:
1649 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1650 "'removable': %i, 'locked': %i }",
1651 bs->device_name, type, bs->removable,
1656 QDict *bs_dict = qobject_to_qdict(bs_obj);
1658 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1659 "'encrypted': %i }",
1660 bs->filename, bs->read_only,
1661 bs->drv->format_name,
1662 bdrv_is_encrypted(bs));
1663 if (bs->backing_file[0] != '\0') {
1664 QDict *qdict = qobject_to_qdict(obj);
1665 qdict_put(qdict, "backing_file",
1666 qstring_from_str(bs->backing_file));
1669 qdict_put_obj(bs_dict, "inserted", obj);
1671 qlist_append_obj(bs_list, bs_obj);
1674 *ret_data = QOBJECT(bs_list);
1677 static void bdrv_stats_iter(QObject *data, void *opaque)
1680 Monitor *mon = opaque;
1682 qdict = qobject_to_qdict(data);
1683 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1685 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1686 monitor_printf(mon, " rd_bytes=%" PRId64
1687 " wr_bytes=%" PRId64
1688 " rd_operations=%" PRId64
1689 " wr_operations=%" PRId64
1691 qdict_get_int(qdict, "rd_bytes"),
1692 qdict_get_int(qdict, "wr_bytes"),
1693 qdict_get_int(qdict, "rd_operations"),
1694 qdict_get_int(qdict, "wr_operations"));
1697 void bdrv_stats_print(Monitor *mon, const QObject *data)
1699 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1702 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1707 res = qobject_from_jsonf("{ 'stats': {"
1708 "'rd_bytes': %" PRId64 ","
1709 "'wr_bytes': %" PRId64 ","
1710 "'rd_operations': %" PRId64 ","
1711 "'wr_operations': %" PRId64 ","
1712 "'wr_highest_offset': %" PRId64
1714 bs->rd_bytes, bs->wr_bytes,
1715 bs->rd_ops, bs->wr_ops,
1716 bs->wr_highest_sector *
1717 (uint64_t)BDRV_SECTOR_SIZE);
1718 dict = qobject_to_qdict(res);
1720 if (*bs->device_name) {
1721 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1725 QObject *parent = bdrv_info_stats_bs(bs->file);
1726 qdict_put_obj(dict, "parent", parent);
1732 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1736 BlockDriverState *bs;
1738 devices = qlist_new();
1740 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1741 obj = bdrv_info_stats_bs(bs);
1742 qlist_append_obj(devices, obj);
1745 *ret_data = QOBJECT(devices);
1748 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1750 if (bs->backing_hd && bs->backing_hd->encrypted)
1751 return bs->backing_file;
1752 else if (bs->encrypted)
1753 return bs->filename;
1758 void bdrv_get_backing_filename(BlockDriverState *bs,
1759 char *filename, int filename_size)
1761 if (!bs->backing_file) {
1762 pstrcpy(filename, filename_size, "");
1764 pstrcpy(filename, filename_size, bs->backing_file);
1768 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1769 const uint8_t *buf, int nb_sectors)
1771 BlockDriver *drv = bs->drv;
1774 if (!drv->bdrv_write_compressed)
1776 if (bdrv_check_request(bs, sector_num, nb_sectors))
1779 if (bs->dirty_bitmap) {
1780 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1783 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1786 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1788 BlockDriver *drv = bs->drv;
1791 if (!drv->bdrv_get_info)
1793 memset(bdi, 0, sizeof(*bdi));
1794 return drv->bdrv_get_info(bs, bdi);
1797 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1798 int64_t pos, int size)
1800 BlockDriver *drv = bs->drv;
1803 if (drv->bdrv_save_vmstate)
1804 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1806 return bdrv_save_vmstate(bs->file, buf, pos, size);
1810 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1811 int64_t pos, int size)
1813 BlockDriver *drv = bs->drv;
1816 if (drv->bdrv_load_vmstate)
1817 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1819 return bdrv_load_vmstate(bs->file, buf, pos, size);
1823 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1825 BlockDriver *drv = bs->drv;
1827 if (!drv || !drv->bdrv_debug_event) {
1831 return drv->bdrv_debug_event(bs, event);
1835 /**************************************************************/
1836 /* handling of snapshots */
1838 int bdrv_can_snapshot(BlockDriverState *bs)
1840 BlockDriver *drv = bs->drv;
1841 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1845 if (!drv->bdrv_snapshot_create) {
1846 if (bs->file != NULL) {
1847 return bdrv_can_snapshot(bs->file);
1855 int bdrv_is_snapshot(BlockDriverState *bs)
1857 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1860 BlockDriverState *bdrv_snapshots(void)
1862 BlockDriverState *bs;
1865 return bs_snapshots;
1869 while ((bs = bdrv_next(bs))) {
1870 if (bdrv_can_snapshot(bs)) {
1878 int bdrv_snapshot_create(BlockDriverState *bs,
1879 QEMUSnapshotInfo *sn_info)
1881 BlockDriver *drv = bs->drv;
1884 if (drv->bdrv_snapshot_create)
1885 return drv->bdrv_snapshot_create(bs, sn_info);
1887 return bdrv_snapshot_create(bs->file, sn_info);
1891 int bdrv_snapshot_goto(BlockDriverState *bs,
1892 const char *snapshot_id)
1894 BlockDriver *drv = bs->drv;
1899 if (drv->bdrv_snapshot_goto)
1900 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1903 drv->bdrv_close(bs);
1904 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1905 open_ret = drv->bdrv_open(bs, bs->open_flags);
1907 bdrv_delete(bs->file);
1917 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1919 BlockDriver *drv = bs->drv;
1922 if (drv->bdrv_snapshot_delete)
1923 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1925 return bdrv_snapshot_delete(bs->file, snapshot_id);
1929 int bdrv_snapshot_list(BlockDriverState *bs,
1930 QEMUSnapshotInfo **psn_info)
1932 BlockDriver *drv = bs->drv;
1935 if (drv->bdrv_snapshot_list)
1936 return drv->bdrv_snapshot_list(bs, psn_info);
1938 return bdrv_snapshot_list(bs->file, psn_info);
1942 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
1943 const char *snapshot_name)
1945 BlockDriver *drv = bs->drv;
1949 if (!bs->read_only) {
1952 if (drv->bdrv_snapshot_load_tmp) {
1953 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
1958 #define NB_SUFFIXES 4
1960 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1962 static const char suffixes[NB_SUFFIXES] = "KMGT";
1967 snprintf(buf, buf_size, "%" PRId64, size);
1970 for(i = 0; i < NB_SUFFIXES; i++) {
1971 if (size < (10 * base)) {
1972 snprintf(buf, buf_size, "%0.1f%c",
1973 (double)size / base,
1976 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1977 snprintf(buf, buf_size, "%" PRId64 "%c",
1978 ((size + (base >> 1)) / base),
1988 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1990 char buf1[128], date_buf[128], clock_buf[128];
2000 snprintf(buf, buf_size,
2001 "%-10s%-20s%7s%20s%15s",
2002 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2006 ptm = localtime(&ti);
2007 strftime(date_buf, sizeof(date_buf),
2008 "%Y-%m-%d %H:%M:%S", ptm);
2010 localtime_r(&ti, &tm);
2011 strftime(date_buf, sizeof(date_buf),
2012 "%Y-%m-%d %H:%M:%S", &tm);
2014 secs = sn->vm_clock_nsec / 1000000000;
2015 snprintf(clock_buf, sizeof(clock_buf),
2016 "%02d:%02d:%02d.%03d",
2018 (int)((secs / 60) % 60),
2020 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2021 snprintf(buf, buf_size,
2022 "%-10s%-20s%7s%20s%15s",
2023 sn->id_str, sn->name,
2024 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2032 /**************************************************************/
2035 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2036 QEMUIOVector *qiov, int nb_sectors,
2037 BlockDriverCompletionFunc *cb, void *opaque)
2039 BlockDriver *drv = bs->drv;
2040 BlockDriverAIOCB *ret;
2042 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2046 if (bdrv_check_request(bs, sector_num, nb_sectors))
2049 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2053 /* Update stats even though technically transfer has not happened. */
2054 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2061 typedef struct BlockCompleteData {
2062 BlockDriverCompletionFunc *cb;
2064 BlockDriverState *bs;
2067 } BlockCompleteData;
2069 static void block_complete_cb(void *opaque, int ret)
2071 BlockCompleteData *b = opaque;
2073 if (b->bs->dirty_bitmap) {
2074 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2076 b->cb(b->opaque, ret);
2080 static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2083 BlockDriverCompletionFunc *cb,
2086 BlockCompleteData *blkdata = qemu_mallocz(sizeof(BlockCompleteData));
2090 blkdata->opaque = opaque;
2091 blkdata->sector_num = sector_num;
2092 blkdata->nb_sectors = nb_sectors;
2097 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2098 QEMUIOVector *qiov, int nb_sectors,
2099 BlockDriverCompletionFunc *cb, void *opaque)
2101 BlockDriver *drv = bs->drv;
2102 BlockDriverAIOCB *ret;
2103 BlockCompleteData *blk_cb_data;
2105 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2111 if (bdrv_check_request(bs, sector_num, nb_sectors))
2114 if (bs->dirty_bitmap) {
2115 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2117 cb = &block_complete_cb;
2118 opaque = blk_cb_data;
2121 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2125 /* Update stats even though technically transfer has not happened. */
2126 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2128 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2129 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2137 typedef struct MultiwriteCB {
2142 BlockDriverCompletionFunc *cb;
2144 QEMUIOVector *free_qiov;
2149 static void multiwrite_user_cb(MultiwriteCB *mcb)
2153 for (i = 0; i < mcb->num_callbacks; i++) {
2154 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2155 if (mcb->callbacks[i].free_qiov) {
2156 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2158 qemu_free(mcb->callbacks[i].free_qiov);
2159 qemu_vfree(mcb->callbacks[i].free_buf);
2163 static void multiwrite_cb(void *opaque, int ret)
2165 MultiwriteCB *mcb = opaque;
2167 trace_multiwrite_cb(mcb, ret);
2169 if (ret < 0 && !mcb->error) {
2173 mcb->num_requests--;
2174 if (mcb->num_requests == 0) {
2175 multiwrite_user_cb(mcb);
2180 static int multiwrite_req_compare(const void *a, const void *b)
2182 const BlockRequest *req1 = a, *req2 = b;
2185 * Note that we can't simply subtract req2->sector from req1->sector
2186 * here as that could overflow the return value.
2188 if (req1->sector > req2->sector) {
2190 } else if (req1->sector < req2->sector) {
2198 * Takes a bunch of requests and tries to merge them. Returns the number of
2199 * requests that remain after merging.
2201 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2202 int num_reqs, MultiwriteCB *mcb)
2206 // Sort requests by start sector
2207 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2209 // Check if adjacent requests touch the same clusters. If so, combine them,
2210 // filling up gaps with zero sectors.
2212 for (i = 1; i < num_reqs; i++) {
2214 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2216 // This handles the cases that are valid for all block drivers, namely
2217 // exactly sequential writes and overlapping writes.
2218 if (reqs[i].sector <= oldreq_last) {
2222 // The block driver may decide that it makes sense to combine requests
2223 // even if there is a gap of some sectors between them. In this case,
2224 // the gap is filled with zeros (therefore only applicable for yet
2225 // unused space in format like qcow2).
2226 if (!merge && bs->drv->bdrv_merge_requests) {
2227 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2230 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2236 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2237 qemu_iovec_init(qiov,
2238 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2240 // Add the first request to the merged one. If the requests are
2241 // overlapping, drop the last sectors of the first request.
2242 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2243 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2245 // We might need to add some zeros between the two requests
2246 if (reqs[i].sector > oldreq_last) {
2247 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2248 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2249 memset(buf, 0, zero_bytes);
2250 qemu_iovec_add(qiov, buf, zero_bytes);
2251 mcb->callbacks[i].free_buf = buf;
2254 // Add the second request
2255 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2257 reqs[outidx].nb_sectors = qiov->size >> 9;
2258 reqs[outidx].qiov = qiov;
2260 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2263 reqs[outidx].sector = reqs[i].sector;
2264 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2265 reqs[outidx].qiov = reqs[i].qiov;
2273 * Submit multiple AIO write requests at once.
2275 * On success, the function returns 0 and all requests in the reqs array have
2276 * been submitted. In error case this function returns -1, and any of the
2277 * requests may or may not be submitted yet. In particular, this means that the
2278 * callback will be called for some of the requests, for others it won't. The
2279 * caller must check the error field of the BlockRequest to wait for the right
2280 * callbacks (if error != 0, no callback will be called).
2282 * The implementation may modify the contents of the reqs array, e.g. to merge
2283 * requests. However, the fields opaque and error are left unmodified as they
2284 * are used to signal failure for a single request to the caller.
2286 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2288 BlockDriverAIOCB *acb;
2292 if (num_reqs == 0) {
2296 // Create MultiwriteCB structure
2297 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2298 mcb->num_requests = 0;
2299 mcb->num_callbacks = num_reqs;
2301 for (i = 0; i < num_reqs; i++) {
2302 mcb->callbacks[i].cb = reqs[i].cb;
2303 mcb->callbacks[i].opaque = reqs[i].opaque;
2306 // Check for mergable requests
2307 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2309 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2312 * Run the aio requests. As soon as one request can't be submitted
2313 * successfully, fail all requests that are not yet submitted (we must
2314 * return failure for all requests anyway)
2316 * num_requests cannot be set to the right value immediately: If
2317 * bdrv_aio_writev fails for some request, num_requests would be too high
2318 * and therefore multiwrite_cb() would never recognize the multiwrite
2319 * request as completed. We also cannot use the loop variable i to set it
2320 * when the first request fails because the callback may already have been
2321 * called for previously submitted requests. Thus, num_requests must be
2322 * incremented for each request that is submitted.
2324 * The problem that callbacks may be called early also means that we need
2325 * to take care that num_requests doesn't become 0 before all requests are
2326 * submitted - multiwrite_cb() would consider the multiwrite request
2327 * completed. A dummy request that is "completed" by a manual call to
2328 * multiwrite_cb() takes care of this.
2330 mcb->num_requests = 1;
2332 // Run the aio requests
2333 for (i = 0; i < num_reqs; i++) {
2334 mcb->num_requests++;
2335 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2336 reqs[i].nb_sectors, multiwrite_cb, mcb);
2339 // We can only fail the whole thing if no request has been
2340 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2341 // complete and report the error in the callback.
2343 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2346 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2347 multiwrite_cb(mcb, -EIO);
2353 /* Complete the dummy request */
2354 multiwrite_cb(mcb, 0);
2359 for (i = 0; i < mcb->num_callbacks; i++) {
2360 reqs[i].error = -EIO;
2366 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2367 BlockDriverCompletionFunc *cb, void *opaque)
2369 BlockDriver *drv = bs->drv;
2371 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2372 return bdrv_aio_noop_em(bs, cb, opaque);
2377 return drv->bdrv_aio_flush(bs, cb, opaque);
2380 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2382 acb->pool->cancel(acb);
2386 /**************************************************************/
2387 /* async block device emulation */
2389 typedef struct BlockDriverAIOCBSync {
2390 BlockDriverAIOCB common;
2393 /* vector translation state */
2397 } BlockDriverAIOCBSync;
2399 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2401 BlockDriverAIOCBSync *acb =
2402 container_of(blockacb, BlockDriverAIOCBSync, common);
2403 qemu_bh_delete(acb->bh);
2405 qemu_aio_release(acb);
2408 static AIOPool bdrv_em_aio_pool = {
2409 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2410 .cancel = bdrv_aio_cancel_em,
2413 static void bdrv_aio_bh_cb(void *opaque)
2415 BlockDriverAIOCBSync *acb = opaque;
2418 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2419 qemu_vfree(acb->bounce);
2420 acb->common.cb(acb->common.opaque, acb->ret);
2421 qemu_bh_delete(acb->bh);
2423 qemu_aio_release(acb);
2426 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2430 BlockDriverCompletionFunc *cb,
2435 BlockDriverAIOCBSync *acb;
2437 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2438 acb->is_write = is_write;
2440 acb->bounce = qemu_blockalign(bs, qiov->size);
2443 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2446 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2447 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2449 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2452 qemu_bh_schedule(acb->bh);
2454 return &acb->common;
2457 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2458 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2459 BlockDriverCompletionFunc *cb, void *opaque)
2461 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2464 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2465 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2466 BlockDriverCompletionFunc *cb, void *opaque)
2468 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2471 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2472 BlockDriverCompletionFunc *cb, void *opaque)
2474 BlockDriverAIOCBSync *acb;
2476 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2477 acb->is_write = 1; /* don't bounce in the completion hadler */
2483 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2486 qemu_bh_schedule(acb->bh);
2487 return &acb->common;
2490 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2491 BlockDriverCompletionFunc *cb, void *opaque)
2493 BlockDriverAIOCBSync *acb;
2495 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2496 acb->is_write = 1; /* don't bounce in the completion handler */
2502 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2505 qemu_bh_schedule(acb->bh);
2506 return &acb->common;
2509 /**************************************************************/
2510 /* sync block device emulation */
2512 static void bdrv_rw_em_cb(void *opaque, int ret)
2514 *(int *)opaque = ret;
2517 #define NOT_DONE 0x7fffffff
2519 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2520 uint8_t *buf, int nb_sectors)
2523 BlockDriverAIOCB *acb;
2527 async_context_push();
2529 async_ret = NOT_DONE;
2530 iov.iov_base = (void *)buf;
2531 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2532 qemu_iovec_init_external(&qiov, &iov, 1);
2533 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2534 bdrv_rw_em_cb, &async_ret);
2540 while (async_ret == NOT_DONE) {
2546 async_context_pop();
2550 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2551 const uint8_t *buf, int nb_sectors)
2554 BlockDriverAIOCB *acb;
2558 async_context_push();
2560 async_ret = NOT_DONE;
2561 iov.iov_base = (void *)buf;
2562 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2563 qemu_iovec_init_external(&qiov, &iov, 1);
2564 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2565 bdrv_rw_em_cb, &async_ret);
2570 while (async_ret == NOT_DONE) {
2575 async_context_pop();
2579 void bdrv_init(void)
2581 module_call_init(MODULE_INIT_BLOCK);
2584 void bdrv_init_with_whitelist(void)
2586 use_bdrv_whitelist = 1;
2590 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2591 BlockDriverCompletionFunc *cb, void *opaque)
2593 BlockDriverAIOCB *acb;
2595 if (pool->free_aiocb) {
2596 acb = pool->free_aiocb;
2597 pool->free_aiocb = acb->next;
2599 acb = qemu_mallocz(pool->aiocb_size);
2604 acb->opaque = opaque;
2608 void qemu_aio_release(void *p)
2610 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2611 AIOPool *pool = acb->pool;
2612 acb->next = pool->free_aiocb;
2613 pool->free_aiocb = acb;
2616 /**************************************************************/
2617 /* removable device support */
2620 * Return TRUE if the media is present
2622 int bdrv_is_inserted(BlockDriverState *bs)
2624 BlockDriver *drv = bs->drv;
2628 if (!drv->bdrv_is_inserted)
2629 return !bs->tray_open;
2630 ret = drv->bdrv_is_inserted(bs);
2635 * Return TRUE if the media changed since the last call to this
2636 * function. It is currently only used for floppy disks
2638 int bdrv_media_changed(BlockDriverState *bs)
2640 BlockDriver *drv = bs->drv;
2643 if (!drv || !drv->bdrv_media_changed)
2646 ret = drv->bdrv_media_changed(bs);
2647 if (ret == -ENOTSUP)
2648 ret = bs->media_changed;
2649 bs->media_changed = 0;
2654 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2656 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2658 BlockDriver *drv = bs->drv;
2665 if (!drv || !drv->bdrv_eject) {
2668 ret = drv->bdrv_eject(bs, eject_flag);
2670 if (ret == -ENOTSUP) {
2674 bs->tray_open = eject_flag;
2680 int bdrv_is_locked(BlockDriverState *bs)
2686 * Lock or unlock the media (if it is locked, the user won't be able
2687 * to eject it manually).
2689 void bdrv_set_locked(BlockDriverState *bs, int locked)
2691 BlockDriver *drv = bs->drv;
2693 bs->locked = locked;
2694 if (drv && drv->bdrv_set_locked) {
2695 drv->bdrv_set_locked(bs, locked);
2699 /* needed for generic scsi interface */
2701 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2703 BlockDriver *drv = bs->drv;
2705 if (drv && drv->bdrv_ioctl)
2706 return drv->bdrv_ioctl(bs, req, buf);
2710 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2711 unsigned long int req, void *buf,
2712 BlockDriverCompletionFunc *cb, void *opaque)
2714 BlockDriver *drv = bs->drv;
2716 if (drv && drv->bdrv_aio_ioctl)
2717 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2723 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2725 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2728 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2730 int64_t bitmap_size;
2732 bs->dirty_count = 0;
2734 if (!bs->dirty_bitmap) {
2735 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2736 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2737 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2739 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2742 if (bs->dirty_bitmap) {
2743 qemu_free(bs->dirty_bitmap);
2744 bs->dirty_bitmap = NULL;
2749 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2751 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2753 if (bs->dirty_bitmap &&
2754 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2755 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2756 (1UL << (chunk % (sizeof(unsigned long) * 8))));
2762 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2765 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2768 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2770 return bs->dirty_count;
2773 int bdrv_img_create(const char *filename, const char *fmt,
2774 const char *base_filename, const char *base_fmt,
2775 char *options, uint64_t img_size, int flags)
2777 QEMUOptionParameter *param = NULL, *create_options = NULL;
2778 QEMUOptionParameter *backing_fmt, *backing_file;
2779 BlockDriverState *bs = NULL;
2780 BlockDriver *drv, *proto_drv;
2783 /* Find driver and parse its options */
2784 drv = bdrv_find_format(fmt);
2786 error_report("Unknown file format '%s'", fmt);
2791 proto_drv = bdrv_find_protocol(filename);
2793 error_report("Unknown protocol '%s'", filename);
2798 create_options = append_option_parameters(create_options,
2799 drv->create_options);
2800 create_options = append_option_parameters(create_options,
2801 proto_drv->create_options);
2803 /* Create parameter list with default values */
2804 param = parse_option_parameters("", create_options, param);
2806 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
2808 /* Parse -o options */
2810 param = parse_option_parameters(options, create_options, param);
2811 if (param == NULL) {
2812 error_report("Invalid options for file format '%s'.", fmt);
2818 if (base_filename) {
2819 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
2821 error_report("Backing file not supported for file format '%s'",
2829 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
2830 error_report("Backing file format not supported for file "
2831 "format '%s'", fmt);
2837 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
2838 if (backing_file && backing_file->value.s) {
2839 if (!strcmp(filename, backing_file->value.s)) {
2840 error_report("Error: Trying to create an image with the "
2841 "same filename as the backing file");
2847 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
2848 if (backing_fmt && backing_fmt->value.s) {
2849 if (!bdrv_find_format(backing_fmt->value.s)) {
2850 error_report("Unknown backing file format '%s'",
2851 backing_fmt->value.s);
2857 // The size for the image must always be specified, with one exception:
2858 // If we are using a backing file, we can obtain the size from there
2859 if (get_option_parameter(param, BLOCK_OPT_SIZE)->value.n == -1) {
2860 if (backing_file && backing_file->value.s) {
2866 ret = bdrv_open(bs, backing_file->value.s, flags, drv);
2868 error_report("Could not open '%s'", filename);
2871 bdrv_get_geometry(bs, &size);
2874 snprintf(buf, sizeof(buf), "%" PRId64, size);
2875 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
2877 error_report("Image creation needs a size parameter");
2883 printf("Formatting '%s', fmt=%s ", filename, fmt);
2884 print_option_parameters(param);
2887 ret = bdrv_create(drv, filename, param);
2890 if (ret == -ENOTSUP) {
2891 error_report("Formatting or formatting option not supported for "
2892 "file format '%s'", fmt);
2893 } else if (ret == -EFBIG) {
2894 error_report("The image size is too large for file format '%s'",
2897 error_report("%s: error while creating %s: %s", filename, fmt,
2903 free_option_parameters(create_options);
2904 free_option_parameters(param);