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 "block_int.h"
29 #include "qemu-objects.h"
32 #include <sys/types.h>
34 #include <sys/ioctl.h>
35 #include <sys/queue.h>
45 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
46 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
47 BlockDriverCompletionFunc *cb, void *opaque);
48 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
49 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
50 BlockDriverCompletionFunc *cb, void *opaque);
51 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
52 BlockDriverCompletionFunc *cb, void *opaque);
53 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
54 BlockDriverCompletionFunc *cb, void *opaque);
55 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
56 uint8_t *buf, int nb_sectors);
57 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
58 const uint8_t *buf, int nb_sectors);
60 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
61 QTAILQ_HEAD_INITIALIZER(bdrv_states);
63 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
64 QLIST_HEAD_INITIALIZER(bdrv_drivers);
66 /* The device to use for VM snapshots */
67 static BlockDriverState *bs_snapshots;
69 /* If non-zero, use only whitelisted block drivers */
70 static int use_bdrv_whitelist;
72 int path_is_absolute(const char *path)
76 /* specific case for names like: "\\.\d:" */
77 if (*path == '/' || *path == '\\')
80 p = strchr(path, ':');
86 return (*p == '/' || *p == '\\');
92 /* if filename is absolute, just copy it to dest. Otherwise, build a
93 path to it by considering it is relative to base_path. URL are
95 void path_combine(char *dest, int dest_size,
96 const char *base_path,
104 if (path_is_absolute(filename)) {
105 pstrcpy(dest, dest_size, filename);
107 p = strchr(base_path, ':');
112 p1 = strrchr(base_path, '/');
116 p2 = strrchr(base_path, '\\');
128 if (len > dest_size - 1)
130 memcpy(dest, base_path, len);
132 pstrcat(dest, dest_size, filename);
136 void bdrv_register(BlockDriver *bdrv)
138 if (!bdrv->bdrv_aio_readv) {
139 /* add AIO emulation layer */
140 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
141 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
142 } else if (!bdrv->bdrv_read) {
143 /* add synchronous IO emulation layer */
144 bdrv->bdrv_read = bdrv_read_em;
145 bdrv->bdrv_write = bdrv_write_em;
148 if (!bdrv->bdrv_aio_flush)
149 bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
151 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
154 /* create a new block device (by default it is empty) */
155 BlockDriverState *bdrv_new(const char *device_name)
157 BlockDriverState *bs;
159 bs = qemu_mallocz(sizeof(BlockDriverState));
160 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
161 if (device_name[0] != '\0') {
162 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
167 BlockDriver *bdrv_find_format(const char *format_name)
170 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
171 if (!strcmp(drv1->format_name, format_name)) {
178 static int bdrv_is_whitelisted(BlockDriver *drv)
180 static const char *whitelist[] = {
181 CONFIG_BDRV_WHITELIST
186 return 1; /* no whitelist, anything goes */
188 for (p = whitelist; *p; p++) {
189 if (!strcmp(drv->format_name, *p)) {
196 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
198 BlockDriver *drv = bdrv_find_format(format_name);
199 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
202 int bdrv_create(BlockDriver *drv, const char* filename,
203 QEMUOptionParameter *options)
205 if (!drv->bdrv_create)
208 return drv->bdrv_create(filename, options);
211 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
215 drv = bdrv_find_protocol(filename);
217 drv = bdrv_find_format("file");
220 return bdrv_create(drv, filename, options);
224 void get_tmp_filename(char *filename, int size)
226 char temp_dir[MAX_PATH];
228 GetTempPath(MAX_PATH, temp_dir);
229 GetTempFileName(temp_dir, "qem", 0, filename);
232 void get_tmp_filename(char *filename, int size)
236 /* XXX: race condition possible */
237 tmpdir = getenv("TMPDIR");
240 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
241 fd = mkstemp(filename);
247 static int is_windows_drive_prefix(const char *filename)
249 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
250 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
254 int is_windows_drive(const char *filename)
256 if (is_windows_drive_prefix(filename) &&
259 if (strstart(filename, "\\\\.\\", NULL) ||
260 strstart(filename, "//./", NULL))
267 * Detect host devices. By convention, /dev/cdrom[N] is always
268 * recognized as a host CDROM.
270 static BlockDriver *find_hdev_driver(const char *filename)
272 int score_max = 0, score;
273 BlockDriver *drv = NULL, *d;
275 QLIST_FOREACH(d, &bdrv_drivers, list) {
276 if (d->bdrv_probe_device) {
277 score = d->bdrv_probe_device(filename);
278 if (score > score_max) {
288 BlockDriver *bdrv_find_protocol(const char *filename)
295 /* TODO Drivers without bdrv_file_open must be specified explicitly */
298 * XXX(hch): we really should not let host device detection
299 * override an explicit protocol specification, but moving this
300 * later breaks access to device names with colons in them.
301 * Thanks to the brain-dead persistent naming schemes on udev-
302 * based Linux systems those actually are quite common.
304 drv1 = find_hdev_driver(filename);
310 if (is_windows_drive(filename) ||
311 is_windows_drive_prefix(filename))
312 return bdrv_find_format("file");
315 p = strchr(filename, ':');
317 return bdrv_find_format("file");
320 if (len > sizeof(protocol) - 1)
321 len = sizeof(protocol) - 1;
322 memcpy(protocol, filename, len);
323 protocol[len] = '\0';
324 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
325 if (drv1->protocol_name &&
326 !strcmp(drv1->protocol_name, protocol)) {
333 static int find_image_format(const char *filename, BlockDriver **pdrv)
335 int ret, score, score_max;
336 BlockDriver *drv1, *drv;
338 BlockDriverState *bs;
340 ret = bdrv_file_open(&bs, filename, 0);
346 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
347 if (bs->sg || !bdrv_is_inserted(bs)) {
349 drv = bdrv_find_format("raw");
357 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
366 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
367 if (drv1->bdrv_probe) {
368 score = drv1->bdrv_probe(buf, ret, filename);
369 if (score > score_max) {
383 * Set the current 'total_sectors' value
385 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
387 BlockDriver *drv = bs->drv;
389 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
393 /* query actual device if possible, otherwise just trust the hint */
394 if (drv->bdrv_getlength) {
395 int64_t length = drv->bdrv_getlength(bs);
399 hint = length >> BDRV_SECTOR_BITS;
402 bs->total_sectors = hint;
407 * Common part for opening disk images and files
409 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
410 int flags, BlockDriver *drv)
417 bs->total_sectors = 0;
420 bs->open_flags = flags;
421 /* buffer_alignment defaulted to 512, drivers can change this value */
422 bs->buffer_alignment = 512;
424 pstrcpy(bs->filename, sizeof(bs->filename), filename);
426 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
431 bs->opaque = qemu_mallocz(drv->instance_size);
434 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
435 * write cache to the guest. We do need the fdatasync to flush
436 * out transactions for block allocations, and we maybe have a
437 * volatile write cache in our backing device to deal with.
439 if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
440 bs->enable_write_cache = 1;
443 * Clear flags that are internal to the block layer before opening the
446 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
449 * Snapshots should be writeable.
451 if (bs->is_temporary) {
452 open_flags |= BDRV_O_RDWR;
455 /* Open the image, either directly or using a protocol */
456 if (drv->bdrv_file_open) {
457 ret = drv->bdrv_file_open(bs, filename, open_flags);
459 ret = bdrv_file_open(&bs->file, filename, open_flags);
461 ret = drv->bdrv_open(bs, open_flags);
469 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
471 ret = refresh_total_sectors(bs, bs->total_sectors);
477 if (bs->is_temporary) {
485 bdrv_delete(bs->file);
488 qemu_free(bs->opaque);
495 * Opens a file using a protocol (file, host_device, nbd, ...)
497 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
499 BlockDriverState *bs;
503 drv = bdrv_find_protocol(filename);
509 ret = bdrv_open_common(bs, filename, flags, drv);
520 * Opens a disk image (raw, qcow2, vmdk, ...)
522 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
527 if (flags & BDRV_O_SNAPSHOT) {
528 BlockDriverState *bs1;
531 BlockDriver *bdrv_qcow2;
532 QEMUOptionParameter *options;
533 char tmp_filename[PATH_MAX];
534 char backing_filename[PATH_MAX];
536 /* if snapshot, we create a temporary backing file and open it
537 instead of opening 'filename' directly */
539 /* if there is a backing file, use it */
541 ret = bdrv_open(bs1, filename, 0, drv);
546 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
548 if (bs1->drv && bs1->drv->protocol_name)
553 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
555 /* Real path is meaningless for protocols */
557 snprintf(backing_filename, sizeof(backing_filename),
559 else if (!realpath(filename, backing_filename))
562 bdrv_qcow2 = bdrv_find_format("qcow2");
563 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
565 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
566 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
568 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
572 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
573 free_option_parameters(options);
578 filename = tmp_filename;
580 bs->is_temporary = 1;
583 /* Find the right image format driver */
585 ret = find_image_format(filename, &drv);
589 goto unlink_and_fail;
593 ret = bdrv_open_common(bs, filename, flags, drv);
595 goto unlink_and_fail;
598 /* If there is a backing file, use it */
599 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
600 char backing_filename[PATH_MAX];
602 BlockDriver *back_drv = NULL;
604 bs->backing_hd = bdrv_new("");
605 path_combine(backing_filename, sizeof(backing_filename),
606 filename, bs->backing_file);
607 if (bs->backing_format[0] != '\0')
608 back_drv = bdrv_find_format(bs->backing_format);
610 /* backing files always opened read-only */
612 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
614 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
619 if (bs->is_temporary) {
620 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
622 /* base image inherits from "parent" */
623 bs->backing_hd->keep_read_only = bs->keep_read_only;
627 if (!bdrv_key_required(bs)) {
628 /* call the change callback */
629 bs->media_changed = 1;
631 bs->change_cb(bs->change_opaque);
637 if (bs->is_temporary) {
643 void bdrv_close(BlockDriverState *bs)
646 if (bs == bs_snapshots) {
649 if (bs->backing_hd) {
650 bdrv_delete(bs->backing_hd);
651 bs->backing_hd = NULL;
653 bs->drv->bdrv_close(bs);
654 qemu_free(bs->opaque);
656 if (bs->is_temporary) {
657 unlink(bs->filename);
663 if (bs->file != NULL) {
664 bdrv_close(bs->file);
667 /* call the change callback */
668 bs->media_changed = 1;
670 bs->change_cb(bs->change_opaque);
674 void bdrv_close_all(void)
676 BlockDriverState *bs;
678 QTAILQ_FOREACH(bs, &bdrv_states, list) {
683 void bdrv_delete(BlockDriverState *bs)
687 /* remove from list, if necessary */
688 if (bs->device_name[0] != '\0') {
689 QTAILQ_REMOVE(&bdrv_states, bs, list);
693 if (bs->file != NULL) {
694 bdrv_delete(bs->file);
697 assert(bs != bs_snapshots);
701 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
710 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
712 assert(bs->peer == qdev);
716 DeviceState *bdrv_get_attached(BlockDriverState *bs)
722 * Run consistency checks on an image
724 * Returns 0 if the check could be completed (it doesn't mean that the image is
725 * free of errors) or -errno when an internal error occured. The results of the
726 * check are stored in res.
728 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
730 if (bs->drv->bdrv_check == NULL) {
734 memset(res, 0, sizeof(*res));
735 return bs->drv->bdrv_check(bs, res);
738 #define COMMIT_BUF_SECTORS 2048
740 /* commit COW file into the raw image */
741 int bdrv_commit(BlockDriverState *bs)
743 BlockDriver *drv = bs->drv;
744 BlockDriver *backing_drv;
745 int64_t sector, total_sectors;
746 int n, ro, open_flags;
747 int ret = 0, rw_ret = 0;
750 BlockDriverState *bs_rw, *bs_ro;
755 if (!bs->backing_hd) {
759 if (bs->backing_hd->keep_read_only) {
763 backing_drv = bs->backing_hd->drv;
764 ro = bs->backing_hd->read_only;
765 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
766 open_flags = bs->backing_hd->open_flags;
770 bdrv_delete(bs->backing_hd);
771 bs->backing_hd = NULL;
772 bs_rw = bdrv_new("");
773 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
777 /* try to re-open read-only */
778 bs_ro = bdrv_new("");
779 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
783 /* drive not functional anymore */
787 bs->backing_hd = bs_ro;
790 bs->backing_hd = bs_rw;
793 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
794 buf = qemu_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
796 for (sector = 0; sector < total_sectors; sector += n) {
797 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
799 if (bdrv_read(bs, sector, buf, n) != 0) {
804 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
811 if (drv->bdrv_make_empty) {
812 ret = drv->bdrv_make_empty(bs);
817 * Make sure all data we wrote to the backing device is actually
821 bdrv_flush(bs->backing_hd);
828 bdrv_delete(bs->backing_hd);
829 bs->backing_hd = NULL;
830 bs_ro = bdrv_new("");
831 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
835 /* drive not functional anymore */
839 bs->backing_hd = bs_ro;
840 bs->backing_hd->keep_read_only = 0;
846 void bdrv_commit_all(void)
848 BlockDriverState *bs;
850 QTAILQ_FOREACH(bs, &bdrv_states, list) {
858 * -EINVAL - backing format specified, but no file
859 * -ENOSPC - can't update the backing file because no space is left in the
861 * -ENOTSUP - format driver doesn't support changing the backing file
863 int bdrv_change_backing_file(BlockDriverState *bs,
864 const char *backing_file, const char *backing_fmt)
866 BlockDriver *drv = bs->drv;
868 if (drv->bdrv_change_backing_file != NULL) {
869 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
875 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
880 if (!bdrv_is_inserted(bs))
886 len = bdrv_getlength(bs);
891 if ((offset > len) || (len - offset < size))
897 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
900 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
901 nb_sectors * BDRV_SECTOR_SIZE);
904 /* return < 0 if error. See bdrv_write() for the return codes */
905 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
906 uint8_t *buf, int nb_sectors)
908 BlockDriver *drv = bs->drv;
912 if (bdrv_check_request(bs, sector_num, nb_sectors))
915 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
918 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
919 int nb_sectors, int dirty)
922 unsigned long val, idx, bit;
924 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
925 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
927 for (; start <= end; start++) {
928 idx = start / (sizeof(unsigned long) * 8);
929 bit = start % (sizeof(unsigned long) * 8);
930 val = bs->dirty_bitmap[idx];
932 if (!(val & (1 << bit))) {
937 if (val & (1 << bit)) {
942 bs->dirty_bitmap[idx] = val;
946 /* Return < 0 if error. Important errors are:
947 -EIO generic I/O error (may happen for all errors)
948 -ENOMEDIUM No media inserted.
949 -EINVAL Invalid sector number or nb_sectors
950 -EACCES Trying to write a read-only device
952 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
953 const uint8_t *buf, int nb_sectors)
955 BlockDriver *drv = bs->drv;
960 if (bdrv_check_request(bs, sector_num, nb_sectors))
963 if (bs->dirty_bitmap) {
964 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
967 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
968 bs->wr_highest_sector = sector_num + nb_sectors - 1;
971 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
974 int bdrv_pread(BlockDriverState *bs, int64_t offset,
975 void *buf, int count1)
977 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
978 int len, nb_sectors, count;
983 /* first read to align to sector start */
984 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
987 sector_num = offset >> BDRV_SECTOR_BITS;
989 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
991 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
999 /* read the sectors "in place" */
1000 nb_sectors = count >> BDRV_SECTOR_BITS;
1001 if (nb_sectors > 0) {
1002 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1004 sector_num += nb_sectors;
1005 len = nb_sectors << BDRV_SECTOR_BITS;
1010 /* add data from the last sector */
1012 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1014 memcpy(buf, tmp_buf, count);
1019 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1020 const void *buf, int count1)
1022 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1023 int len, nb_sectors, count;
1028 /* first write to align to sector start */
1029 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1032 sector_num = offset >> BDRV_SECTOR_BITS;
1034 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1036 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1037 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1046 /* write the sectors "in place" */
1047 nb_sectors = count >> BDRV_SECTOR_BITS;
1048 if (nb_sectors > 0) {
1049 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1051 sector_num += nb_sectors;
1052 len = nb_sectors << BDRV_SECTOR_BITS;
1057 /* add data from the last sector */
1059 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1061 memcpy(tmp_buf, buf, count);
1062 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1069 * Writes to the file and ensures that no writes are reordered across this
1070 * request (acts as a barrier)
1072 * Returns 0 on success, -errno in error cases.
1074 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1075 const void *buf, int count)
1079 ret = bdrv_pwrite(bs, offset, buf, count);
1084 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1085 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1093 * Writes to the file and ensures that no writes are reordered across this
1094 * request (acts as a barrier)
1096 * Returns 0 on success, -errno in error cases.
1098 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1099 const uint8_t *buf, int nb_sectors)
1101 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1102 buf, BDRV_SECTOR_SIZE * nb_sectors);
1106 * Truncate file to 'offset' bytes (needed only for file protocols)
1108 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1110 BlockDriver *drv = bs->drv;
1114 if (!drv->bdrv_truncate)
1118 ret = drv->bdrv_truncate(bs, offset);
1120 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1126 * Length of a file in bytes. Return < 0 if error or unknown.
1128 int64_t bdrv_getlength(BlockDriverState *bs)
1130 BlockDriver *drv = bs->drv;
1134 /* Fixed size devices use the total_sectors value for speed instead of
1135 issuing a length query (like lseek) on each call. Also, legacy block
1136 drivers don't provide a bdrv_getlength function and must use
1138 if (!bs->growable || !drv->bdrv_getlength) {
1139 return bs->total_sectors * BDRV_SECTOR_SIZE;
1141 return drv->bdrv_getlength(bs);
1144 /* return 0 as number of sectors if no device present or error */
1145 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1148 length = bdrv_getlength(bs);
1152 length = length >> BDRV_SECTOR_BITS;
1153 *nb_sectors_ptr = length;
1157 uint8_t boot_ind; /* 0x80 - active */
1158 uint8_t head; /* starting head */
1159 uint8_t sector; /* starting sector */
1160 uint8_t cyl; /* starting cylinder */
1161 uint8_t sys_ind; /* What partition type */
1162 uint8_t end_head; /* end head */
1163 uint8_t end_sector; /* end sector */
1164 uint8_t end_cyl; /* end cylinder */
1165 uint32_t start_sect; /* starting sector counting from 0 */
1166 uint32_t nr_sects; /* nr of sectors in partition */
1167 } __attribute__((packed));
1169 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1170 static int guess_disk_lchs(BlockDriverState *bs,
1171 int *pcylinders, int *pheads, int *psectors)
1173 uint8_t buf[BDRV_SECTOR_SIZE];
1174 int ret, i, heads, sectors, cylinders;
1175 struct partition *p;
1177 uint64_t nb_sectors;
1179 bdrv_get_geometry(bs, &nb_sectors);
1181 ret = bdrv_read(bs, 0, buf, 1);
1184 /* test msdos magic */
1185 if (buf[510] != 0x55 || buf[511] != 0xaa)
1187 for(i = 0; i < 4; i++) {
1188 p = ((struct partition *)(buf + 0x1be)) + i;
1189 nr_sects = le32_to_cpu(p->nr_sects);
1190 if (nr_sects && p->end_head) {
1191 /* We make the assumption that the partition terminates on
1192 a cylinder boundary */
1193 heads = p->end_head + 1;
1194 sectors = p->end_sector & 63;
1197 cylinders = nb_sectors / (heads * sectors);
1198 if (cylinders < 1 || cylinders > 16383)
1201 *psectors = sectors;
1202 *pcylinders = cylinders;
1204 printf("guessed geometry: LCHS=%d %d %d\n",
1205 cylinders, heads, sectors);
1213 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1215 int translation, lba_detected = 0;
1216 int cylinders, heads, secs;
1217 uint64_t nb_sectors;
1219 /* if a geometry hint is available, use it */
1220 bdrv_get_geometry(bs, &nb_sectors);
1221 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1222 translation = bdrv_get_translation_hint(bs);
1223 if (cylinders != 0) {
1228 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1230 /* if heads > 16, it means that a BIOS LBA
1231 translation was active, so the default
1232 hardware geometry is OK */
1234 goto default_geometry;
1239 /* disable any translation to be in sync with
1240 the logical geometry */
1241 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1242 bdrv_set_translation_hint(bs,
1243 BIOS_ATA_TRANSLATION_NONE);
1248 /* if no geometry, use a standard physical disk geometry */
1249 cylinders = nb_sectors / (16 * 63);
1251 if (cylinders > 16383)
1253 else if (cylinders < 2)
1258 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1259 if ((*pcyls * *pheads) <= 131072) {
1260 bdrv_set_translation_hint(bs,
1261 BIOS_ATA_TRANSLATION_LARGE);
1263 bdrv_set_translation_hint(bs,
1264 BIOS_ATA_TRANSLATION_LBA);
1268 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1272 void bdrv_set_geometry_hint(BlockDriverState *bs,
1273 int cyls, int heads, int secs)
1280 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1283 bs->removable = ((type == BDRV_TYPE_CDROM ||
1284 type == BDRV_TYPE_FLOPPY));
1287 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1289 bs->translation = translation;
1292 void bdrv_get_geometry_hint(BlockDriverState *bs,
1293 int *pcyls, int *pheads, int *psecs)
1296 *pheads = bs->heads;
1300 int bdrv_get_type_hint(BlockDriverState *bs)
1305 int bdrv_get_translation_hint(BlockDriverState *bs)
1307 return bs->translation;
1310 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1311 BlockErrorAction on_write_error)
1313 bs->on_read_error = on_read_error;
1314 bs->on_write_error = on_write_error;
1317 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1319 return is_read ? bs->on_read_error : bs->on_write_error;
1322 void bdrv_set_removable(BlockDriverState *bs, int removable)
1324 bs->removable = removable;
1325 if (removable && bs == bs_snapshots) {
1326 bs_snapshots = NULL;
1330 int bdrv_is_removable(BlockDriverState *bs)
1332 return bs->removable;
1335 int bdrv_is_read_only(BlockDriverState *bs)
1337 return bs->read_only;
1340 int bdrv_is_sg(BlockDriverState *bs)
1345 int bdrv_enable_write_cache(BlockDriverState *bs)
1347 return bs->enable_write_cache;
1350 /* XXX: no longer used */
1351 void bdrv_set_change_cb(BlockDriverState *bs,
1352 void (*change_cb)(void *opaque), void *opaque)
1354 bs->change_cb = change_cb;
1355 bs->change_opaque = opaque;
1358 int bdrv_is_encrypted(BlockDriverState *bs)
1360 if (bs->backing_hd && bs->backing_hd->encrypted)
1362 return bs->encrypted;
1365 int bdrv_key_required(BlockDriverState *bs)
1367 BlockDriverState *backing_hd = bs->backing_hd;
1369 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1371 return (bs->encrypted && !bs->valid_key);
1374 int bdrv_set_key(BlockDriverState *bs, const char *key)
1377 if (bs->backing_hd && bs->backing_hd->encrypted) {
1378 ret = bdrv_set_key(bs->backing_hd, key);
1384 if (!bs->encrypted) {
1386 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1389 ret = bs->drv->bdrv_set_key(bs, key);
1392 } else if (!bs->valid_key) {
1394 /* call the change callback now, we skipped it on open */
1395 bs->media_changed = 1;
1397 bs->change_cb(bs->change_opaque);
1402 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1407 pstrcpy(buf, buf_size, bs->drv->format_name);
1411 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1416 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1417 it(opaque, drv->format_name);
1421 BlockDriverState *bdrv_find(const char *name)
1423 BlockDriverState *bs;
1425 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1426 if (!strcmp(name, bs->device_name)) {
1433 BlockDriverState *bdrv_next(BlockDriverState *bs)
1436 return QTAILQ_FIRST(&bdrv_states);
1438 return QTAILQ_NEXT(bs, list);
1441 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1443 BlockDriverState *bs;
1445 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1450 const char *bdrv_get_device_name(BlockDriverState *bs)
1452 return bs->device_name;
1455 void bdrv_flush(BlockDriverState *bs)
1457 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1461 if (bs->drv && bs->drv->bdrv_flush)
1462 bs->drv->bdrv_flush(bs);
1465 void bdrv_flush_all(void)
1467 BlockDriverState *bs;
1469 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1470 if (bs->drv && !bdrv_is_read_only(bs) &&
1471 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1477 int bdrv_has_zero_init(BlockDriverState *bs)
1481 if (bs->drv->bdrv_has_zero_init) {
1482 return bs->drv->bdrv_has_zero_init(bs);
1489 * Returns true iff the specified sector is present in the disk image. Drivers
1490 * not implementing the functionality are assumed to not support backing files,
1491 * hence all their sectors are reported as allocated.
1493 * 'pnum' is set to the number of sectors (including and immediately following
1494 * the specified sector) that are known to be in the same
1495 * allocated/unallocated state.
1497 * 'nb_sectors' is the max value 'pnum' should be set to.
1499 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1503 if (!bs->drv->bdrv_is_allocated) {
1504 if (sector_num >= bs->total_sectors) {
1508 n = bs->total_sectors - sector_num;
1509 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1512 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1515 void bdrv_mon_event(const BlockDriverState *bdrv,
1516 BlockMonEventAction action, int is_read)
1519 const char *action_str;
1522 case BDRV_ACTION_REPORT:
1523 action_str = "report";
1525 case BDRV_ACTION_IGNORE:
1526 action_str = "ignore";
1528 case BDRV_ACTION_STOP:
1529 action_str = "stop";
1535 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1538 is_read ? "read" : "write");
1539 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1541 qobject_decref(data);
1544 static void bdrv_print_dict(QObject *obj, void *opaque)
1547 Monitor *mon = opaque;
1549 bs_dict = qobject_to_qdict(obj);
1551 monitor_printf(mon, "%s: type=%s removable=%d",
1552 qdict_get_str(bs_dict, "device"),
1553 qdict_get_str(bs_dict, "type"),
1554 qdict_get_bool(bs_dict, "removable"));
1556 if (qdict_get_bool(bs_dict, "removable")) {
1557 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1560 if (qdict_haskey(bs_dict, "inserted")) {
1561 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1563 monitor_printf(mon, " file=");
1564 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1565 if (qdict_haskey(qdict, "backing_file")) {
1566 monitor_printf(mon, " backing_file=");
1567 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1569 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1570 qdict_get_bool(qdict, "ro"),
1571 qdict_get_str(qdict, "drv"),
1572 qdict_get_bool(qdict, "encrypted"));
1574 monitor_printf(mon, " [not inserted]");
1577 monitor_printf(mon, "\n");
1580 void bdrv_info_print(Monitor *mon, const QObject *data)
1582 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1585 void bdrv_info(Monitor *mon, QObject **ret_data)
1588 BlockDriverState *bs;
1590 bs_list = qlist_new();
1592 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1594 const char *type = "unknown";
1600 case BDRV_TYPE_CDROM:
1603 case BDRV_TYPE_FLOPPY:
1608 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1609 "'removable': %i, 'locked': %i }",
1610 bs->device_name, type, bs->removable,
1615 QDict *bs_dict = qobject_to_qdict(bs_obj);
1617 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1618 "'encrypted': %i }",
1619 bs->filename, bs->read_only,
1620 bs->drv->format_name,
1621 bdrv_is_encrypted(bs));
1622 if (bs->backing_file[0] != '\0') {
1623 QDict *qdict = qobject_to_qdict(obj);
1624 qdict_put(qdict, "backing_file",
1625 qstring_from_str(bs->backing_file));
1628 qdict_put_obj(bs_dict, "inserted", obj);
1630 qlist_append_obj(bs_list, bs_obj);
1633 *ret_data = QOBJECT(bs_list);
1636 static void bdrv_stats_iter(QObject *data, void *opaque)
1639 Monitor *mon = opaque;
1641 qdict = qobject_to_qdict(data);
1642 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1644 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1645 monitor_printf(mon, " rd_bytes=%" PRId64
1646 " wr_bytes=%" PRId64
1647 " rd_operations=%" PRId64
1648 " wr_operations=%" PRId64
1650 qdict_get_int(qdict, "rd_bytes"),
1651 qdict_get_int(qdict, "wr_bytes"),
1652 qdict_get_int(qdict, "rd_operations"),
1653 qdict_get_int(qdict, "wr_operations"));
1656 void bdrv_stats_print(Monitor *mon, const QObject *data)
1658 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1661 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1666 res = qobject_from_jsonf("{ 'stats': {"
1667 "'rd_bytes': %" PRId64 ","
1668 "'wr_bytes': %" PRId64 ","
1669 "'rd_operations': %" PRId64 ","
1670 "'wr_operations': %" PRId64 ","
1671 "'wr_highest_offset': %" PRId64
1673 bs->rd_bytes, bs->wr_bytes,
1674 bs->rd_ops, bs->wr_ops,
1675 bs->wr_highest_sector *
1676 (uint64_t)BDRV_SECTOR_SIZE);
1677 dict = qobject_to_qdict(res);
1679 if (*bs->device_name) {
1680 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1684 QObject *parent = bdrv_info_stats_bs(bs->file);
1685 qdict_put_obj(dict, "parent", parent);
1691 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1695 BlockDriverState *bs;
1697 devices = qlist_new();
1699 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1700 obj = bdrv_info_stats_bs(bs);
1701 qlist_append_obj(devices, obj);
1704 *ret_data = QOBJECT(devices);
1707 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1709 if (bs->backing_hd && bs->backing_hd->encrypted)
1710 return bs->backing_file;
1711 else if (bs->encrypted)
1712 return bs->filename;
1717 void bdrv_get_backing_filename(BlockDriverState *bs,
1718 char *filename, int filename_size)
1720 if (!bs->backing_file) {
1721 pstrcpy(filename, filename_size, "");
1723 pstrcpy(filename, filename_size, bs->backing_file);
1727 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1728 const uint8_t *buf, int nb_sectors)
1730 BlockDriver *drv = bs->drv;
1733 if (!drv->bdrv_write_compressed)
1735 if (bdrv_check_request(bs, sector_num, nb_sectors))
1738 if (bs->dirty_bitmap) {
1739 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1742 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1745 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1747 BlockDriver *drv = bs->drv;
1750 if (!drv->bdrv_get_info)
1752 memset(bdi, 0, sizeof(*bdi));
1753 return drv->bdrv_get_info(bs, bdi);
1756 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1757 int64_t pos, int size)
1759 BlockDriver *drv = bs->drv;
1762 if (drv->bdrv_save_vmstate)
1763 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1765 return bdrv_save_vmstate(bs->file, buf, pos, size);
1769 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1770 int64_t pos, int size)
1772 BlockDriver *drv = bs->drv;
1775 if (drv->bdrv_load_vmstate)
1776 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1778 return bdrv_load_vmstate(bs->file, buf, pos, size);
1782 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1784 BlockDriver *drv = bs->drv;
1786 if (!drv || !drv->bdrv_debug_event) {
1790 return drv->bdrv_debug_event(bs, event);
1794 /**************************************************************/
1795 /* handling of snapshots */
1797 int bdrv_can_snapshot(BlockDriverState *bs)
1799 BlockDriver *drv = bs->drv;
1800 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1804 if (!drv->bdrv_snapshot_create) {
1805 if (bs->file != NULL) {
1806 return bdrv_can_snapshot(bs->file);
1814 int bdrv_is_snapshot(BlockDriverState *bs)
1816 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1819 BlockDriverState *bdrv_snapshots(void)
1821 BlockDriverState *bs;
1824 return bs_snapshots;
1828 while ((bs = bdrv_next(bs))) {
1829 if (bdrv_can_snapshot(bs)) {
1837 int bdrv_snapshot_create(BlockDriverState *bs,
1838 QEMUSnapshotInfo *sn_info)
1840 BlockDriver *drv = bs->drv;
1843 if (drv->bdrv_snapshot_create)
1844 return drv->bdrv_snapshot_create(bs, sn_info);
1846 return bdrv_snapshot_create(bs->file, sn_info);
1850 int bdrv_snapshot_goto(BlockDriverState *bs,
1851 const char *snapshot_id)
1853 BlockDriver *drv = bs->drv;
1858 if (drv->bdrv_snapshot_goto)
1859 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1862 drv->bdrv_close(bs);
1863 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1864 open_ret = drv->bdrv_open(bs, bs->open_flags);
1866 bdrv_delete(bs->file);
1876 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1878 BlockDriver *drv = bs->drv;
1881 if (drv->bdrv_snapshot_delete)
1882 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1884 return bdrv_snapshot_delete(bs->file, snapshot_id);
1888 int bdrv_snapshot_list(BlockDriverState *bs,
1889 QEMUSnapshotInfo **psn_info)
1891 BlockDriver *drv = bs->drv;
1894 if (drv->bdrv_snapshot_list)
1895 return drv->bdrv_snapshot_list(bs, psn_info);
1897 return bdrv_snapshot_list(bs->file, psn_info);
1901 #define NB_SUFFIXES 4
1903 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1905 static const char suffixes[NB_SUFFIXES] = "KMGT";
1910 snprintf(buf, buf_size, "%" PRId64, size);
1913 for(i = 0; i < NB_SUFFIXES; i++) {
1914 if (size < (10 * base)) {
1915 snprintf(buf, buf_size, "%0.1f%c",
1916 (double)size / base,
1919 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1920 snprintf(buf, buf_size, "%" PRId64 "%c",
1921 ((size + (base >> 1)) / base),
1931 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1933 char buf1[128], date_buf[128], clock_buf[128];
1943 snprintf(buf, buf_size,
1944 "%-10s%-20s%7s%20s%15s",
1945 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1949 ptm = localtime(&ti);
1950 strftime(date_buf, sizeof(date_buf),
1951 "%Y-%m-%d %H:%M:%S", ptm);
1953 localtime_r(&ti, &tm);
1954 strftime(date_buf, sizeof(date_buf),
1955 "%Y-%m-%d %H:%M:%S", &tm);
1957 secs = sn->vm_clock_nsec / 1000000000;
1958 snprintf(clock_buf, sizeof(clock_buf),
1959 "%02d:%02d:%02d.%03d",
1961 (int)((secs / 60) % 60),
1963 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1964 snprintf(buf, buf_size,
1965 "%-10s%-20s%7s%20s%15s",
1966 sn->id_str, sn->name,
1967 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1975 /**************************************************************/
1978 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1979 QEMUIOVector *qiov, int nb_sectors,
1980 BlockDriverCompletionFunc *cb, void *opaque)
1982 BlockDriver *drv = bs->drv;
1983 BlockDriverAIOCB *ret;
1987 if (bdrv_check_request(bs, sector_num, nb_sectors))
1990 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1994 /* Update stats even though technically transfer has not happened. */
1995 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2002 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2003 QEMUIOVector *qiov, int nb_sectors,
2004 BlockDriverCompletionFunc *cb, void *opaque)
2006 BlockDriver *drv = bs->drv;
2007 BlockDriverAIOCB *ret;
2013 if (bdrv_check_request(bs, sector_num, nb_sectors))
2016 if (bs->dirty_bitmap) {
2017 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2020 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2024 /* Update stats even though technically transfer has not happened. */
2025 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2027 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2028 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2036 typedef struct MultiwriteCB {
2041 BlockDriverCompletionFunc *cb;
2043 QEMUIOVector *free_qiov;
2048 static void multiwrite_user_cb(MultiwriteCB *mcb)
2052 for (i = 0; i < mcb->num_callbacks; i++) {
2053 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2054 if (mcb->callbacks[i].free_qiov) {
2055 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2057 qemu_free(mcb->callbacks[i].free_qiov);
2058 qemu_vfree(mcb->callbacks[i].free_buf);
2062 static void multiwrite_cb(void *opaque, int ret)
2064 MultiwriteCB *mcb = opaque;
2066 if (ret < 0 && !mcb->error) {
2070 mcb->num_requests--;
2071 if (mcb->num_requests == 0) {
2072 multiwrite_user_cb(mcb);
2077 static int multiwrite_req_compare(const void *a, const void *b)
2079 const BlockRequest *req1 = a, *req2 = b;
2082 * Note that we can't simply subtract req2->sector from req1->sector
2083 * here as that could overflow the return value.
2085 if (req1->sector > req2->sector) {
2087 } else if (req1->sector < req2->sector) {
2095 * Takes a bunch of requests and tries to merge them. Returns the number of
2096 * requests that remain after merging.
2098 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2099 int num_reqs, MultiwriteCB *mcb)
2103 // Sort requests by start sector
2104 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2106 // Check if adjacent requests touch the same clusters. If so, combine them,
2107 // filling up gaps with zero sectors.
2109 for (i = 1; i < num_reqs; i++) {
2111 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2113 // This handles the cases that are valid for all block drivers, namely
2114 // exactly sequential writes and overlapping writes.
2115 if (reqs[i].sector <= oldreq_last) {
2119 // The block driver may decide that it makes sense to combine requests
2120 // even if there is a gap of some sectors between them. In this case,
2121 // the gap is filled with zeros (therefore only applicable for yet
2122 // unused space in format like qcow2).
2123 if (!merge && bs->drv->bdrv_merge_requests) {
2124 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2127 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2133 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2134 qemu_iovec_init(qiov,
2135 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2137 // Add the first request to the merged one. If the requests are
2138 // overlapping, drop the last sectors of the first request.
2139 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2140 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2142 // We might need to add some zeros between the two requests
2143 if (reqs[i].sector > oldreq_last) {
2144 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2145 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2146 memset(buf, 0, zero_bytes);
2147 qemu_iovec_add(qiov, buf, zero_bytes);
2148 mcb->callbacks[i].free_buf = buf;
2151 // Add the second request
2152 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2154 reqs[outidx].nb_sectors = qiov->size >> 9;
2155 reqs[outidx].qiov = qiov;
2157 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2160 reqs[outidx].sector = reqs[i].sector;
2161 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2162 reqs[outidx].qiov = reqs[i].qiov;
2170 * Submit multiple AIO write requests at once.
2172 * On success, the function returns 0 and all requests in the reqs array have
2173 * been submitted. In error case this function returns -1, and any of the
2174 * requests may or may not be submitted yet. In particular, this means that the
2175 * callback will be called for some of the requests, for others it won't. The
2176 * caller must check the error field of the BlockRequest to wait for the right
2177 * callbacks (if error != 0, no callback will be called).
2179 * The implementation may modify the contents of the reqs array, e.g. to merge
2180 * requests. However, the fields opaque and error are left unmodified as they
2181 * are used to signal failure for a single request to the caller.
2183 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2185 BlockDriverAIOCB *acb;
2189 if (num_reqs == 0) {
2193 // Create MultiwriteCB structure
2194 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2195 mcb->num_requests = 0;
2196 mcb->num_callbacks = num_reqs;
2198 for (i = 0; i < num_reqs; i++) {
2199 mcb->callbacks[i].cb = reqs[i].cb;
2200 mcb->callbacks[i].opaque = reqs[i].opaque;
2203 // Check for mergable requests
2204 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2207 * Run the aio requests. As soon as one request can't be submitted
2208 * successfully, fail all requests that are not yet submitted (we must
2209 * return failure for all requests anyway)
2211 * num_requests cannot be set to the right value immediately: If
2212 * bdrv_aio_writev fails for some request, num_requests would be too high
2213 * and therefore multiwrite_cb() would never recognize the multiwrite
2214 * request as completed. We also cannot use the loop variable i to set it
2215 * when the first request fails because the callback may already have been
2216 * called for previously submitted requests. Thus, num_requests must be
2217 * incremented for each request that is submitted.
2219 * The problem that callbacks may be called early also means that we need
2220 * to take care that num_requests doesn't become 0 before all requests are
2221 * submitted - multiwrite_cb() would consider the multiwrite request
2222 * completed. A dummy request that is "completed" by a manual call to
2223 * multiwrite_cb() takes care of this.
2225 mcb->num_requests = 1;
2227 for (i = 0; i < num_reqs; i++) {
2228 mcb->num_requests++;
2229 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2230 reqs[i].nb_sectors, multiwrite_cb, mcb);
2233 // We can only fail the whole thing if no request has been
2234 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2235 // complete and report the error in the callback.
2239 multiwrite_cb(mcb, -EIO);
2245 /* Complete the dummy request */
2246 multiwrite_cb(mcb, 0);
2251 for (i = 0; i < mcb->num_callbacks; i++) {
2252 reqs[i].error = -EIO;
2258 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2259 BlockDriverCompletionFunc *cb, void *opaque)
2261 BlockDriver *drv = bs->drv;
2263 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2264 return bdrv_aio_noop_em(bs, cb, opaque);
2269 return drv->bdrv_aio_flush(bs, cb, opaque);
2272 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2274 acb->pool->cancel(acb);
2278 /**************************************************************/
2279 /* async block device emulation */
2281 typedef struct BlockDriverAIOCBSync {
2282 BlockDriverAIOCB common;
2285 /* vector translation state */
2289 } BlockDriverAIOCBSync;
2291 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2293 BlockDriverAIOCBSync *acb =
2294 container_of(blockacb, BlockDriverAIOCBSync, common);
2295 qemu_bh_delete(acb->bh);
2297 qemu_aio_release(acb);
2300 static AIOPool bdrv_em_aio_pool = {
2301 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2302 .cancel = bdrv_aio_cancel_em,
2305 static void bdrv_aio_bh_cb(void *opaque)
2307 BlockDriverAIOCBSync *acb = opaque;
2310 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2311 qemu_vfree(acb->bounce);
2312 acb->common.cb(acb->common.opaque, acb->ret);
2313 qemu_bh_delete(acb->bh);
2315 qemu_aio_release(acb);
2318 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2322 BlockDriverCompletionFunc *cb,
2327 BlockDriverAIOCBSync *acb;
2329 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2330 acb->is_write = is_write;
2332 acb->bounce = qemu_blockalign(bs, qiov->size);
2335 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2338 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2339 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2341 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2344 qemu_bh_schedule(acb->bh);
2346 return &acb->common;
2349 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2350 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2351 BlockDriverCompletionFunc *cb, void *opaque)
2353 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2356 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2357 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2358 BlockDriverCompletionFunc *cb, void *opaque)
2360 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2363 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2364 BlockDriverCompletionFunc *cb, void *opaque)
2366 BlockDriverAIOCBSync *acb;
2368 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2369 acb->is_write = 1; /* don't bounce in the completion hadler */
2375 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2378 qemu_bh_schedule(acb->bh);
2379 return &acb->common;
2382 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2383 BlockDriverCompletionFunc *cb, void *opaque)
2385 BlockDriverAIOCBSync *acb;
2387 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2388 acb->is_write = 1; /* don't bounce in the completion handler */
2394 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2397 qemu_bh_schedule(acb->bh);
2398 return &acb->common;
2401 /**************************************************************/
2402 /* sync block device emulation */
2404 static void bdrv_rw_em_cb(void *opaque, int ret)
2406 *(int *)opaque = ret;
2409 #define NOT_DONE 0x7fffffff
2411 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2412 uint8_t *buf, int nb_sectors)
2415 BlockDriverAIOCB *acb;
2419 async_context_push();
2421 async_ret = NOT_DONE;
2422 iov.iov_base = (void *)buf;
2423 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2424 qemu_iovec_init_external(&qiov, &iov, 1);
2425 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2426 bdrv_rw_em_cb, &async_ret);
2432 while (async_ret == NOT_DONE) {
2438 async_context_pop();
2442 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2443 const uint8_t *buf, int nb_sectors)
2446 BlockDriverAIOCB *acb;
2450 async_context_push();
2452 async_ret = NOT_DONE;
2453 iov.iov_base = (void *)buf;
2454 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2455 qemu_iovec_init_external(&qiov, &iov, 1);
2456 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2457 bdrv_rw_em_cb, &async_ret);
2462 while (async_ret == NOT_DONE) {
2467 async_context_pop();
2471 void bdrv_init(void)
2473 module_call_init(MODULE_INIT_BLOCK);
2476 void bdrv_init_with_whitelist(void)
2478 use_bdrv_whitelist = 1;
2482 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2483 BlockDriverCompletionFunc *cb, void *opaque)
2485 BlockDriverAIOCB *acb;
2487 if (pool->free_aiocb) {
2488 acb = pool->free_aiocb;
2489 pool->free_aiocb = acb->next;
2491 acb = qemu_mallocz(pool->aiocb_size);
2496 acb->opaque = opaque;
2500 void qemu_aio_release(void *p)
2502 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2503 AIOPool *pool = acb->pool;
2504 acb->next = pool->free_aiocb;
2505 pool->free_aiocb = acb;
2508 /**************************************************************/
2509 /* removable device support */
2512 * Return TRUE if the media is present
2514 int bdrv_is_inserted(BlockDriverState *bs)
2516 BlockDriver *drv = bs->drv;
2520 if (!drv->bdrv_is_inserted)
2521 return !bs->tray_open;
2522 ret = drv->bdrv_is_inserted(bs);
2527 * Return TRUE if the media changed since the last call to this
2528 * function. It is currently only used for floppy disks
2530 int bdrv_media_changed(BlockDriverState *bs)
2532 BlockDriver *drv = bs->drv;
2535 if (!drv || !drv->bdrv_media_changed)
2538 ret = drv->bdrv_media_changed(bs);
2539 if (ret == -ENOTSUP)
2540 ret = bs->media_changed;
2541 bs->media_changed = 0;
2546 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2548 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2550 BlockDriver *drv = bs->drv;
2557 if (!drv || !drv->bdrv_eject) {
2560 ret = drv->bdrv_eject(bs, eject_flag);
2562 if (ret == -ENOTSUP) {
2566 bs->tray_open = eject_flag;
2572 int bdrv_is_locked(BlockDriverState *bs)
2578 * Lock or unlock the media (if it is locked, the user won't be able
2579 * to eject it manually).
2581 void bdrv_set_locked(BlockDriverState *bs, int locked)
2583 BlockDriver *drv = bs->drv;
2585 bs->locked = locked;
2586 if (drv && drv->bdrv_set_locked) {
2587 drv->bdrv_set_locked(bs, locked);
2591 /* needed for generic scsi interface */
2593 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2595 BlockDriver *drv = bs->drv;
2597 if (drv && drv->bdrv_ioctl)
2598 return drv->bdrv_ioctl(bs, req, buf);
2602 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2603 unsigned long int req, void *buf,
2604 BlockDriverCompletionFunc *cb, void *opaque)
2606 BlockDriver *drv = bs->drv;
2608 if (drv && drv->bdrv_aio_ioctl)
2609 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2615 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2617 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2620 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2622 int64_t bitmap_size;
2624 bs->dirty_count = 0;
2626 if (!bs->dirty_bitmap) {
2627 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2628 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2629 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2631 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2634 if (bs->dirty_bitmap) {
2635 qemu_free(bs->dirty_bitmap);
2636 bs->dirty_bitmap = NULL;
2641 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2643 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2645 if (bs->dirty_bitmap &&
2646 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2647 return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2648 (1 << (chunk % (sizeof(unsigned long) * 8)));
2654 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2657 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2660 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2662 return bs->dirty_count;