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,
528 if (flags & BDRV_O_SNAPSHOT) {
529 BlockDriverState *bs1;
532 BlockDriver *bdrv_qcow2;
533 QEMUOptionParameter *options;
534 char tmp_filename[PATH_MAX];
535 char backing_filename[PATH_MAX];
537 /* if snapshot, we create a temporary backing file and open it
538 instead of opening 'filename' directly */
540 /* if there is a backing file, use it */
542 ret = bdrv_open(bs1, filename, 0, drv);
547 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
549 if (bs1->drv && bs1->drv->protocol_name)
554 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
556 /* Real path is meaningless for protocols */
558 snprintf(backing_filename, sizeof(backing_filename),
560 else if (!realpath(filename, backing_filename))
563 bdrv_qcow2 = bdrv_find_format("qcow2");
564 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
566 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
567 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
569 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
573 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
574 free_option_parameters(options);
579 filename = tmp_filename;
581 bs->is_temporary = 1;
584 /* Find the right image format driver */
586 ret = find_image_format(filename, &drv);
591 goto unlink_and_fail;
595 ret = bdrv_open_common(bs, filename, flags, drv);
597 goto unlink_and_fail;
602 /* If there is a backing file, use it */
603 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
604 char backing_filename[PATH_MAX];
606 BlockDriver *back_drv = NULL;
608 bs->backing_hd = bdrv_new("");
609 path_combine(backing_filename, sizeof(backing_filename),
610 filename, bs->backing_file);
611 if (bs->backing_format[0] != '\0')
612 back_drv = bdrv_find_format(bs->backing_format);
614 /* backing files always opened read-only */
616 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
618 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
623 if (bs->is_temporary) {
624 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
626 /* base image inherits from "parent" */
627 bs->backing_hd->keep_read_only = bs->keep_read_only;
631 if (!bdrv_key_required(bs)) {
632 /* call the change callback */
633 bs->media_changed = 1;
635 bs->change_cb(bs->change_opaque);
641 if (bs->is_temporary) {
647 void bdrv_close(BlockDriverState *bs)
650 if (bs == bs_snapshots) {
653 if (bs->backing_hd) {
654 bdrv_delete(bs->backing_hd);
655 bs->backing_hd = NULL;
657 bs->drv->bdrv_close(bs);
658 qemu_free(bs->opaque);
660 if (bs->is_temporary) {
661 unlink(bs->filename);
667 if (bs->file != NULL) {
668 bdrv_close(bs->file);
671 /* call the change callback */
672 bs->media_changed = 1;
674 bs->change_cb(bs->change_opaque);
678 void bdrv_close_all(void)
680 BlockDriverState *bs;
682 QTAILQ_FOREACH(bs, &bdrv_states, list) {
687 void bdrv_delete(BlockDriverState *bs)
691 /* remove from list, if necessary */
692 if (bs->device_name[0] != '\0') {
693 QTAILQ_REMOVE(&bdrv_states, bs, list);
697 if (bs->file != NULL) {
698 bdrv_delete(bs->file);
701 assert(bs != bs_snapshots);
705 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
714 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
716 assert(bs->peer == qdev);
720 DeviceState *bdrv_get_attached(BlockDriverState *bs)
726 * Run consistency checks on an image
728 * Returns 0 if the check could be completed (it doesn't mean that the image is
729 * free of errors) or -errno when an internal error occured. The results of the
730 * check are stored in res.
732 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
734 if (bs->drv->bdrv_check == NULL) {
738 memset(res, 0, sizeof(*res));
739 return bs->drv->bdrv_check(bs, res);
742 #define COMMIT_BUF_SECTORS 2048
744 /* commit COW file into the raw image */
745 int bdrv_commit(BlockDriverState *bs)
747 BlockDriver *drv = bs->drv;
748 int64_t sector, total_sectors;
749 int n, ro, open_flags;
750 int ret = 0, rw_ret = 0;
753 BlockDriverState *bs_rw, *bs_ro;
758 if (!bs->backing_hd) {
762 if (bs->backing_hd->keep_read_only) {
766 ro = bs->backing_hd->read_only;
767 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
768 open_flags = bs->backing_hd->open_flags;
772 bdrv_delete(bs->backing_hd);
773 bs->backing_hd = NULL;
774 bs_rw = bdrv_new("");
775 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, drv);
778 /* try to re-open read-only */
779 bs_ro = bdrv_new("");
780 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
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, drv);
834 /* drive not functional anymore */
838 bs->backing_hd = bs_ro;
839 bs->backing_hd->keep_read_only = 0;
845 void bdrv_commit_all(void)
847 BlockDriverState *bs;
849 QTAILQ_FOREACH(bs, &bdrv_states, list) {
857 * -EINVAL - backing format specified, but no file
858 * -ENOSPC - can't update the backing file because no space is left in the
860 * -ENOTSUP - format driver doesn't support changing the backing file
862 int bdrv_change_backing_file(BlockDriverState *bs,
863 const char *backing_file, const char *backing_fmt)
865 BlockDriver *drv = bs->drv;
867 if (drv->bdrv_change_backing_file != NULL) {
868 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
874 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
879 if (!bdrv_is_inserted(bs))
885 len = bdrv_getlength(bs);
890 if ((offset > len) || (len - offset < size))
896 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
899 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
900 nb_sectors * BDRV_SECTOR_SIZE);
903 /* return < 0 if error. See bdrv_write() for the return codes */
904 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
905 uint8_t *buf, int nb_sectors)
907 BlockDriver *drv = bs->drv;
911 if (bdrv_check_request(bs, sector_num, nb_sectors))
914 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
917 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
918 int nb_sectors, int dirty)
921 unsigned long val, idx, bit;
923 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
924 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
926 for (; start <= end; start++) {
927 idx = start / (sizeof(unsigned long) * 8);
928 bit = start % (sizeof(unsigned long) * 8);
929 val = bs->dirty_bitmap[idx];
931 if (!(val & (1 << bit))) {
936 if (val & (1 << bit)) {
941 bs->dirty_bitmap[idx] = val;
945 /* Return < 0 if error. Important errors are:
946 -EIO generic I/O error (may happen for all errors)
947 -ENOMEDIUM No media inserted.
948 -EINVAL Invalid sector number or nb_sectors
949 -EACCES Trying to write a read-only device
951 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
952 const uint8_t *buf, int nb_sectors)
954 BlockDriver *drv = bs->drv;
959 if (bdrv_check_request(bs, sector_num, nb_sectors))
962 if (bs->dirty_bitmap) {
963 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
966 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
967 bs->wr_highest_sector = sector_num + nb_sectors - 1;
970 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
973 int bdrv_pread(BlockDriverState *bs, int64_t offset,
974 void *buf, int count1)
976 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
977 int len, nb_sectors, count;
982 /* first read to align to sector start */
983 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
986 sector_num = offset >> BDRV_SECTOR_BITS;
988 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
990 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
998 /* read the sectors "in place" */
999 nb_sectors = count >> BDRV_SECTOR_BITS;
1000 if (nb_sectors > 0) {
1001 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1003 sector_num += nb_sectors;
1004 len = nb_sectors << BDRV_SECTOR_BITS;
1009 /* add data from the last sector */
1011 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1013 memcpy(buf, tmp_buf, count);
1018 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1019 const void *buf, int count1)
1021 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1022 int len, nb_sectors, count;
1027 /* first write to align to sector start */
1028 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1031 sector_num = offset >> BDRV_SECTOR_BITS;
1033 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1035 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1036 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1045 /* write the sectors "in place" */
1046 nb_sectors = count >> BDRV_SECTOR_BITS;
1047 if (nb_sectors > 0) {
1048 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1050 sector_num += nb_sectors;
1051 len = nb_sectors << BDRV_SECTOR_BITS;
1056 /* add data from the last sector */
1058 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1060 memcpy(tmp_buf, buf, count);
1061 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1068 * Writes to the file and ensures that no writes are reordered across this
1069 * request (acts as a barrier)
1071 * Returns 0 on success, -errno in error cases.
1073 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1074 const void *buf, int count)
1078 ret = bdrv_pwrite(bs, offset, buf, count);
1083 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1084 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1092 * Writes to the file and ensures that no writes are reordered across this
1093 * request (acts as a barrier)
1095 * Returns 0 on success, -errno in error cases.
1097 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1098 const uint8_t *buf, int nb_sectors)
1100 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1101 buf, BDRV_SECTOR_SIZE * nb_sectors);
1105 * Truncate file to 'offset' bytes (needed only for file protocols)
1107 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1109 BlockDriver *drv = bs->drv;
1113 if (!drv->bdrv_truncate)
1117 ret = drv->bdrv_truncate(bs, offset);
1119 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1125 * Length of a file in bytes. Return < 0 if error or unknown.
1127 int64_t bdrv_getlength(BlockDriverState *bs)
1129 BlockDriver *drv = bs->drv;
1133 /* Fixed size devices use the total_sectors value for speed instead of
1134 issuing a length query (like lseek) on each call. Also, legacy block
1135 drivers don't provide a bdrv_getlength function and must use
1137 if (!bs->growable || !drv->bdrv_getlength) {
1138 return bs->total_sectors * BDRV_SECTOR_SIZE;
1140 return drv->bdrv_getlength(bs);
1143 /* return 0 as number of sectors if no device present or error */
1144 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1147 length = bdrv_getlength(bs);
1151 length = length >> BDRV_SECTOR_BITS;
1152 *nb_sectors_ptr = length;
1156 uint8_t boot_ind; /* 0x80 - active */
1157 uint8_t head; /* starting head */
1158 uint8_t sector; /* starting sector */
1159 uint8_t cyl; /* starting cylinder */
1160 uint8_t sys_ind; /* What partition type */
1161 uint8_t end_head; /* end head */
1162 uint8_t end_sector; /* end sector */
1163 uint8_t end_cyl; /* end cylinder */
1164 uint32_t start_sect; /* starting sector counting from 0 */
1165 uint32_t nr_sects; /* nr of sectors in partition */
1166 } __attribute__((packed));
1168 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1169 static int guess_disk_lchs(BlockDriverState *bs,
1170 int *pcylinders, int *pheads, int *psectors)
1172 uint8_t buf[BDRV_SECTOR_SIZE];
1173 int ret, i, heads, sectors, cylinders;
1174 struct partition *p;
1176 uint64_t nb_sectors;
1178 bdrv_get_geometry(bs, &nb_sectors);
1180 ret = bdrv_read(bs, 0, buf, 1);
1183 /* test msdos magic */
1184 if (buf[510] != 0x55 || buf[511] != 0xaa)
1186 for(i = 0; i < 4; i++) {
1187 p = ((struct partition *)(buf + 0x1be)) + i;
1188 nr_sects = le32_to_cpu(p->nr_sects);
1189 if (nr_sects && p->end_head) {
1190 /* We make the assumption that the partition terminates on
1191 a cylinder boundary */
1192 heads = p->end_head + 1;
1193 sectors = p->end_sector & 63;
1196 cylinders = nb_sectors / (heads * sectors);
1197 if (cylinders < 1 || cylinders > 16383)
1200 *psectors = sectors;
1201 *pcylinders = cylinders;
1203 printf("guessed geometry: LCHS=%d %d %d\n",
1204 cylinders, heads, sectors);
1212 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1214 int translation, lba_detected = 0;
1215 int cylinders, heads, secs;
1216 uint64_t nb_sectors;
1218 /* if a geometry hint is available, use it */
1219 bdrv_get_geometry(bs, &nb_sectors);
1220 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1221 translation = bdrv_get_translation_hint(bs);
1222 if (cylinders != 0) {
1227 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1229 /* if heads > 16, it means that a BIOS LBA
1230 translation was active, so the default
1231 hardware geometry is OK */
1233 goto default_geometry;
1238 /* disable any translation to be in sync with
1239 the logical geometry */
1240 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1241 bdrv_set_translation_hint(bs,
1242 BIOS_ATA_TRANSLATION_NONE);
1247 /* if no geometry, use a standard physical disk geometry */
1248 cylinders = nb_sectors / (16 * 63);
1250 if (cylinders > 16383)
1252 else if (cylinders < 2)
1257 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1258 if ((*pcyls * *pheads) <= 131072) {
1259 bdrv_set_translation_hint(bs,
1260 BIOS_ATA_TRANSLATION_LARGE);
1262 bdrv_set_translation_hint(bs,
1263 BIOS_ATA_TRANSLATION_LBA);
1267 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1271 void bdrv_set_geometry_hint(BlockDriverState *bs,
1272 int cyls, int heads, int secs)
1279 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1282 bs->removable = ((type == BDRV_TYPE_CDROM ||
1283 type == BDRV_TYPE_FLOPPY));
1286 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1288 bs->translation = translation;
1291 void bdrv_get_geometry_hint(BlockDriverState *bs,
1292 int *pcyls, int *pheads, int *psecs)
1295 *pheads = bs->heads;
1299 int bdrv_get_type_hint(BlockDriverState *bs)
1304 int bdrv_get_translation_hint(BlockDriverState *bs)
1306 return bs->translation;
1309 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1310 BlockErrorAction on_write_error)
1312 bs->on_read_error = on_read_error;
1313 bs->on_write_error = on_write_error;
1316 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1318 return is_read ? bs->on_read_error : bs->on_write_error;
1321 void bdrv_set_removable(BlockDriverState *bs, int removable)
1323 bs->removable = removable;
1324 if (removable && bs == bs_snapshots) {
1325 bs_snapshots = NULL;
1329 int bdrv_is_removable(BlockDriverState *bs)
1331 return bs->removable;
1334 int bdrv_is_read_only(BlockDriverState *bs)
1336 return bs->read_only;
1339 int bdrv_is_sg(BlockDriverState *bs)
1344 int bdrv_enable_write_cache(BlockDriverState *bs)
1346 return bs->enable_write_cache;
1349 /* XXX: no longer used */
1350 void bdrv_set_change_cb(BlockDriverState *bs,
1351 void (*change_cb)(void *opaque), void *opaque)
1353 bs->change_cb = change_cb;
1354 bs->change_opaque = opaque;
1357 int bdrv_is_encrypted(BlockDriverState *bs)
1359 if (bs->backing_hd && bs->backing_hd->encrypted)
1361 return bs->encrypted;
1364 int bdrv_key_required(BlockDriverState *bs)
1366 BlockDriverState *backing_hd = bs->backing_hd;
1368 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1370 return (bs->encrypted && !bs->valid_key);
1373 int bdrv_set_key(BlockDriverState *bs, const char *key)
1376 if (bs->backing_hd && bs->backing_hd->encrypted) {
1377 ret = bdrv_set_key(bs->backing_hd, key);
1383 if (!bs->encrypted) {
1385 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1388 ret = bs->drv->bdrv_set_key(bs, key);
1391 } else if (!bs->valid_key) {
1393 /* call the change callback now, we skipped it on open */
1394 bs->media_changed = 1;
1396 bs->change_cb(bs->change_opaque);
1401 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1406 pstrcpy(buf, buf_size, bs->drv->format_name);
1410 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1415 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1416 it(opaque, drv->format_name);
1420 BlockDriverState *bdrv_find(const char *name)
1422 BlockDriverState *bs;
1424 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1425 if (!strcmp(name, bs->device_name)) {
1432 BlockDriverState *bdrv_next(BlockDriverState *bs)
1435 return QTAILQ_FIRST(&bdrv_states);
1437 return QTAILQ_NEXT(bs, list);
1440 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1442 BlockDriverState *bs;
1444 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1449 const char *bdrv_get_device_name(BlockDriverState *bs)
1451 return bs->device_name;
1454 void bdrv_flush(BlockDriverState *bs)
1456 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1460 if (bs->drv && bs->drv->bdrv_flush)
1461 bs->drv->bdrv_flush(bs);
1464 void bdrv_flush_all(void)
1466 BlockDriverState *bs;
1468 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1469 if (bs->drv && !bdrv_is_read_only(bs) &&
1470 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1476 int bdrv_has_zero_init(BlockDriverState *bs)
1480 if (bs->drv->bdrv_has_zero_init) {
1481 return bs->drv->bdrv_has_zero_init(bs);
1488 * Returns true iff the specified sector is present in the disk image. Drivers
1489 * not implementing the functionality are assumed to not support backing files,
1490 * hence all their sectors are reported as allocated.
1492 * 'pnum' is set to the number of sectors (including and immediately following
1493 * the specified sector) that are known to be in the same
1494 * allocated/unallocated state.
1496 * 'nb_sectors' is the max value 'pnum' should be set to.
1498 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1502 if (!bs->drv->bdrv_is_allocated) {
1503 if (sector_num >= bs->total_sectors) {
1507 n = bs->total_sectors - sector_num;
1508 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1511 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1514 void bdrv_mon_event(const BlockDriverState *bdrv,
1515 BlockMonEventAction action, int is_read)
1518 const char *action_str;
1521 case BDRV_ACTION_REPORT:
1522 action_str = "report";
1524 case BDRV_ACTION_IGNORE:
1525 action_str = "ignore";
1527 case BDRV_ACTION_STOP:
1528 action_str = "stop";
1534 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1537 is_read ? "read" : "write");
1538 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1540 qobject_decref(data);
1543 static void bdrv_print_dict(QObject *obj, void *opaque)
1546 Monitor *mon = opaque;
1548 bs_dict = qobject_to_qdict(obj);
1550 monitor_printf(mon, "%s: type=%s removable=%d",
1551 qdict_get_str(bs_dict, "device"),
1552 qdict_get_str(bs_dict, "type"),
1553 qdict_get_bool(bs_dict, "removable"));
1555 if (qdict_get_bool(bs_dict, "removable")) {
1556 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1559 if (qdict_haskey(bs_dict, "inserted")) {
1560 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1562 monitor_printf(mon, " file=");
1563 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1564 if (qdict_haskey(qdict, "backing_file")) {
1565 monitor_printf(mon, " backing_file=");
1566 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1568 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1569 qdict_get_bool(qdict, "ro"),
1570 qdict_get_str(qdict, "drv"),
1571 qdict_get_bool(qdict, "encrypted"));
1573 monitor_printf(mon, " [not inserted]");
1576 monitor_printf(mon, "\n");
1579 void bdrv_info_print(Monitor *mon, const QObject *data)
1581 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1584 void bdrv_info(Monitor *mon, QObject **ret_data)
1587 BlockDriverState *bs;
1589 bs_list = qlist_new();
1591 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1593 const char *type = "unknown";
1599 case BDRV_TYPE_CDROM:
1602 case BDRV_TYPE_FLOPPY:
1607 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1608 "'removable': %i, 'locked': %i }",
1609 bs->device_name, type, bs->removable,
1614 QDict *bs_dict = qobject_to_qdict(bs_obj);
1616 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1617 "'encrypted': %i }",
1618 bs->filename, bs->read_only,
1619 bs->drv->format_name,
1620 bdrv_is_encrypted(bs));
1621 if (bs->backing_file[0] != '\0') {
1622 QDict *qdict = qobject_to_qdict(obj);
1623 qdict_put(qdict, "backing_file",
1624 qstring_from_str(bs->backing_file));
1627 qdict_put_obj(bs_dict, "inserted", obj);
1629 qlist_append_obj(bs_list, bs_obj);
1632 *ret_data = QOBJECT(bs_list);
1635 static void bdrv_stats_iter(QObject *data, void *opaque)
1638 Monitor *mon = opaque;
1640 qdict = qobject_to_qdict(data);
1641 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1643 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1644 monitor_printf(mon, " rd_bytes=%" PRId64
1645 " wr_bytes=%" PRId64
1646 " rd_operations=%" PRId64
1647 " wr_operations=%" PRId64
1649 qdict_get_int(qdict, "rd_bytes"),
1650 qdict_get_int(qdict, "wr_bytes"),
1651 qdict_get_int(qdict, "rd_operations"),
1652 qdict_get_int(qdict, "wr_operations"));
1655 void bdrv_stats_print(Monitor *mon, const QObject *data)
1657 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1660 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1665 res = qobject_from_jsonf("{ 'stats': {"
1666 "'rd_bytes': %" PRId64 ","
1667 "'wr_bytes': %" PRId64 ","
1668 "'rd_operations': %" PRId64 ","
1669 "'wr_operations': %" PRId64 ","
1670 "'wr_highest_offset': %" PRId64
1672 bs->rd_bytes, bs->wr_bytes,
1673 bs->rd_ops, bs->wr_ops,
1674 bs->wr_highest_sector *
1675 (uint64_t)BDRV_SECTOR_SIZE);
1676 dict = qobject_to_qdict(res);
1678 if (*bs->device_name) {
1679 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1683 QObject *parent = bdrv_info_stats_bs(bs->file);
1684 qdict_put_obj(dict, "parent", parent);
1690 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1694 BlockDriverState *bs;
1696 devices = qlist_new();
1698 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1699 obj = bdrv_info_stats_bs(bs);
1700 qlist_append_obj(devices, obj);
1703 *ret_data = QOBJECT(devices);
1706 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1708 if (bs->backing_hd && bs->backing_hd->encrypted)
1709 return bs->backing_file;
1710 else if (bs->encrypted)
1711 return bs->filename;
1716 void bdrv_get_backing_filename(BlockDriverState *bs,
1717 char *filename, int filename_size)
1719 if (!bs->backing_file) {
1720 pstrcpy(filename, filename_size, "");
1722 pstrcpy(filename, filename_size, bs->backing_file);
1726 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1727 const uint8_t *buf, int nb_sectors)
1729 BlockDriver *drv = bs->drv;
1732 if (!drv->bdrv_write_compressed)
1734 if (bdrv_check_request(bs, sector_num, nb_sectors))
1737 if (bs->dirty_bitmap) {
1738 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1741 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1744 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1746 BlockDriver *drv = bs->drv;
1749 if (!drv->bdrv_get_info)
1751 memset(bdi, 0, sizeof(*bdi));
1752 return drv->bdrv_get_info(bs, bdi);
1755 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1756 int64_t pos, int size)
1758 BlockDriver *drv = bs->drv;
1761 if (drv->bdrv_save_vmstate)
1762 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1764 return bdrv_save_vmstate(bs->file, buf, pos, size);
1768 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1769 int64_t pos, int size)
1771 BlockDriver *drv = bs->drv;
1774 if (drv->bdrv_load_vmstate)
1775 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1777 return bdrv_load_vmstate(bs->file, buf, pos, size);
1781 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1783 BlockDriver *drv = bs->drv;
1785 if (!drv || !drv->bdrv_debug_event) {
1789 return drv->bdrv_debug_event(bs, event);
1793 /**************************************************************/
1794 /* handling of snapshots */
1796 int bdrv_can_snapshot(BlockDriverState *bs)
1798 BlockDriver *drv = bs->drv;
1799 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1803 if (!drv->bdrv_snapshot_create) {
1804 if (bs->file != NULL) {
1805 return bdrv_can_snapshot(bs->file);
1813 int bdrv_is_snapshot(BlockDriverState *bs)
1815 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1818 BlockDriverState *bdrv_snapshots(void)
1820 BlockDriverState *bs;
1823 return bs_snapshots;
1827 while ((bs = bdrv_next(bs))) {
1828 if (bdrv_can_snapshot(bs)) {
1836 int bdrv_snapshot_create(BlockDriverState *bs,
1837 QEMUSnapshotInfo *sn_info)
1839 BlockDriver *drv = bs->drv;
1842 if (drv->bdrv_snapshot_create)
1843 return drv->bdrv_snapshot_create(bs, sn_info);
1845 return bdrv_snapshot_create(bs->file, sn_info);
1849 int bdrv_snapshot_goto(BlockDriverState *bs,
1850 const char *snapshot_id)
1852 BlockDriver *drv = bs->drv;
1857 if (drv->bdrv_snapshot_goto)
1858 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1861 drv->bdrv_close(bs);
1862 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1863 open_ret = drv->bdrv_open(bs, bs->open_flags);
1865 bdrv_delete(bs->file);
1875 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1877 BlockDriver *drv = bs->drv;
1880 if (drv->bdrv_snapshot_delete)
1881 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1883 return bdrv_snapshot_delete(bs->file, snapshot_id);
1887 int bdrv_snapshot_list(BlockDriverState *bs,
1888 QEMUSnapshotInfo **psn_info)
1890 BlockDriver *drv = bs->drv;
1893 if (drv->bdrv_snapshot_list)
1894 return drv->bdrv_snapshot_list(bs, psn_info);
1896 return bdrv_snapshot_list(bs->file, psn_info);
1900 #define NB_SUFFIXES 4
1902 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1904 static const char suffixes[NB_SUFFIXES] = "KMGT";
1909 snprintf(buf, buf_size, "%" PRId64, size);
1912 for(i = 0; i < NB_SUFFIXES; i++) {
1913 if (size < (10 * base)) {
1914 snprintf(buf, buf_size, "%0.1f%c",
1915 (double)size / base,
1918 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1919 snprintf(buf, buf_size, "%" PRId64 "%c",
1920 ((size + (base >> 1)) / base),
1930 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1932 char buf1[128], date_buf[128], clock_buf[128];
1942 snprintf(buf, buf_size,
1943 "%-10s%-20s%7s%20s%15s",
1944 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1948 ptm = localtime(&ti);
1949 strftime(date_buf, sizeof(date_buf),
1950 "%Y-%m-%d %H:%M:%S", ptm);
1952 localtime_r(&ti, &tm);
1953 strftime(date_buf, sizeof(date_buf),
1954 "%Y-%m-%d %H:%M:%S", &tm);
1956 secs = sn->vm_clock_nsec / 1000000000;
1957 snprintf(clock_buf, sizeof(clock_buf),
1958 "%02d:%02d:%02d.%03d",
1960 (int)((secs / 60) % 60),
1962 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1963 snprintf(buf, buf_size,
1964 "%-10s%-20s%7s%20s%15s",
1965 sn->id_str, sn->name,
1966 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1974 /**************************************************************/
1977 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1978 QEMUIOVector *qiov, int nb_sectors,
1979 BlockDriverCompletionFunc *cb, void *opaque)
1981 BlockDriver *drv = bs->drv;
1982 BlockDriverAIOCB *ret;
1986 if (bdrv_check_request(bs, sector_num, nb_sectors))
1989 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1993 /* Update stats even though technically transfer has not happened. */
1994 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2001 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2002 QEMUIOVector *qiov, int nb_sectors,
2003 BlockDriverCompletionFunc *cb, void *opaque)
2005 BlockDriver *drv = bs->drv;
2006 BlockDriverAIOCB *ret;
2012 if (bdrv_check_request(bs, sector_num, nb_sectors))
2015 if (bs->dirty_bitmap) {
2016 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2019 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2023 /* Update stats even though technically transfer has not happened. */
2024 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2026 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2027 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2035 typedef struct MultiwriteCB {
2040 BlockDriverCompletionFunc *cb;
2042 QEMUIOVector *free_qiov;
2047 static void multiwrite_user_cb(MultiwriteCB *mcb)
2051 for (i = 0; i < mcb->num_callbacks; i++) {
2052 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2053 if (mcb->callbacks[i].free_qiov) {
2054 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2056 qemu_free(mcb->callbacks[i].free_qiov);
2057 qemu_vfree(mcb->callbacks[i].free_buf);
2061 static void multiwrite_cb(void *opaque, int ret)
2063 MultiwriteCB *mcb = opaque;
2065 if (ret < 0 && !mcb->error) {
2069 mcb->num_requests--;
2070 if (mcb->num_requests == 0) {
2071 multiwrite_user_cb(mcb);
2076 static int multiwrite_req_compare(const void *a, const void *b)
2078 const BlockRequest *req1 = a, *req2 = b;
2081 * Note that we can't simply subtract req2->sector from req1->sector
2082 * here as that could overflow the return value.
2084 if (req1->sector > req2->sector) {
2086 } else if (req1->sector < req2->sector) {
2094 * Takes a bunch of requests and tries to merge them. Returns the number of
2095 * requests that remain after merging.
2097 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2098 int num_reqs, MultiwriteCB *mcb)
2102 // Sort requests by start sector
2103 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2105 // Check if adjacent requests touch the same clusters. If so, combine them,
2106 // filling up gaps with zero sectors.
2108 for (i = 1; i < num_reqs; i++) {
2110 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2112 // This handles the cases that are valid for all block drivers, namely
2113 // exactly sequential writes and overlapping writes.
2114 if (reqs[i].sector <= oldreq_last) {
2118 // The block driver may decide that it makes sense to combine requests
2119 // even if there is a gap of some sectors between them. In this case,
2120 // the gap is filled with zeros (therefore only applicable for yet
2121 // unused space in format like qcow2).
2122 if (!merge && bs->drv->bdrv_merge_requests) {
2123 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2126 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2132 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2133 qemu_iovec_init(qiov,
2134 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2136 // Add the first request to the merged one. If the requests are
2137 // overlapping, drop the last sectors of the first request.
2138 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2139 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2141 // We might need to add some zeros between the two requests
2142 if (reqs[i].sector > oldreq_last) {
2143 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2144 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2145 memset(buf, 0, zero_bytes);
2146 qemu_iovec_add(qiov, buf, zero_bytes);
2147 mcb->callbacks[i].free_buf = buf;
2150 // Add the second request
2151 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2153 reqs[outidx].nb_sectors = qiov->size >> 9;
2154 reqs[outidx].qiov = qiov;
2156 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2159 reqs[outidx].sector = reqs[i].sector;
2160 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2161 reqs[outidx].qiov = reqs[i].qiov;
2169 * Submit multiple AIO write requests at once.
2171 * On success, the function returns 0 and all requests in the reqs array have
2172 * been submitted. In error case this function returns -1, and any of the
2173 * requests may or may not be submitted yet. In particular, this means that the
2174 * callback will be called for some of the requests, for others it won't. The
2175 * caller must check the error field of the BlockRequest to wait for the right
2176 * callbacks (if error != 0, no callback will be called).
2178 * The implementation may modify the contents of the reqs array, e.g. to merge
2179 * requests. However, the fields opaque and error are left unmodified as they
2180 * are used to signal failure for a single request to the caller.
2182 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2184 BlockDriverAIOCB *acb;
2188 if (num_reqs == 0) {
2192 // Create MultiwriteCB structure
2193 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2194 mcb->num_requests = 0;
2195 mcb->num_callbacks = num_reqs;
2197 for (i = 0; i < num_reqs; i++) {
2198 mcb->callbacks[i].cb = reqs[i].cb;
2199 mcb->callbacks[i].opaque = reqs[i].opaque;
2202 // Check for mergable requests
2203 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2206 * Run the aio requests. As soon as one request can't be submitted
2207 * successfully, fail all requests that are not yet submitted (we must
2208 * return failure for all requests anyway)
2210 * num_requests cannot be set to the right value immediately: If
2211 * bdrv_aio_writev fails for some request, num_requests would be too high
2212 * and therefore multiwrite_cb() would never recognize the multiwrite
2213 * request as completed. We also cannot use the loop variable i to set it
2214 * when the first request fails because the callback may already have been
2215 * called for previously submitted requests. Thus, num_requests must be
2216 * incremented for each request that is submitted.
2218 * The problem that callbacks may be called early also means that we need
2219 * to take care that num_requests doesn't become 0 before all requests are
2220 * submitted - multiwrite_cb() would consider the multiwrite request
2221 * completed. A dummy request that is "completed" by a manual call to
2222 * multiwrite_cb() takes care of this.
2224 mcb->num_requests = 1;
2226 for (i = 0; i < num_reqs; i++) {
2227 mcb->num_requests++;
2228 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2229 reqs[i].nb_sectors, multiwrite_cb, mcb);
2232 // We can only fail the whole thing if no request has been
2233 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2234 // complete and report the error in the callback.
2238 multiwrite_cb(mcb, -EIO);
2244 /* Complete the dummy request */
2245 multiwrite_cb(mcb, 0);
2250 for (i = 0; i < mcb->num_callbacks; i++) {
2251 reqs[i].error = -EIO;
2257 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2258 BlockDriverCompletionFunc *cb, void *opaque)
2260 BlockDriver *drv = bs->drv;
2262 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2263 return bdrv_aio_noop_em(bs, cb, opaque);
2268 return drv->bdrv_aio_flush(bs, cb, opaque);
2271 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2273 acb->pool->cancel(acb);
2277 /**************************************************************/
2278 /* async block device emulation */
2280 typedef struct BlockDriverAIOCBSync {
2281 BlockDriverAIOCB common;
2284 /* vector translation state */
2288 } BlockDriverAIOCBSync;
2290 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2292 BlockDriverAIOCBSync *acb =
2293 container_of(blockacb, BlockDriverAIOCBSync, common);
2294 qemu_bh_delete(acb->bh);
2296 qemu_aio_release(acb);
2299 static AIOPool bdrv_em_aio_pool = {
2300 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2301 .cancel = bdrv_aio_cancel_em,
2304 static void bdrv_aio_bh_cb(void *opaque)
2306 BlockDriverAIOCBSync *acb = opaque;
2309 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2310 qemu_vfree(acb->bounce);
2311 acb->common.cb(acb->common.opaque, acb->ret);
2312 qemu_bh_delete(acb->bh);
2314 qemu_aio_release(acb);
2317 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2321 BlockDriverCompletionFunc *cb,
2326 BlockDriverAIOCBSync *acb;
2328 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2329 acb->is_write = is_write;
2331 acb->bounce = qemu_blockalign(bs, qiov->size);
2334 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2337 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2338 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2340 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2343 qemu_bh_schedule(acb->bh);
2345 return &acb->common;
2348 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2349 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2350 BlockDriverCompletionFunc *cb, void *opaque)
2352 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2355 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2356 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2357 BlockDriverCompletionFunc *cb, void *opaque)
2359 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2362 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2363 BlockDriverCompletionFunc *cb, void *opaque)
2365 BlockDriverAIOCBSync *acb;
2367 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2368 acb->is_write = 1; /* don't bounce in the completion hadler */
2374 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2377 qemu_bh_schedule(acb->bh);
2378 return &acb->common;
2381 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2382 BlockDriverCompletionFunc *cb, void *opaque)
2384 BlockDriverAIOCBSync *acb;
2386 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2387 acb->is_write = 1; /* don't bounce in the completion handler */
2393 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2396 qemu_bh_schedule(acb->bh);
2397 return &acb->common;
2400 /**************************************************************/
2401 /* sync block device emulation */
2403 static void bdrv_rw_em_cb(void *opaque, int ret)
2405 *(int *)opaque = ret;
2408 #define NOT_DONE 0x7fffffff
2410 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2411 uint8_t *buf, int nb_sectors)
2414 BlockDriverAIOCB *acb;
2418 async_context_push();
2420 async_ret = NOT_DONE;
2421 iov.iov_base = (void *)buf;
2422 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2423 qemu_iovec_init_external(&qiov, &iov, 1);
2424 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2425 bdrv_rw_em_cb, &async_ret);
2431 while (async_ret == NOT_DONE) {
2437 async_context_pop();
2441 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2442 const uint8_t *buf, int nb_sectors)
2445 BlockDriverAIOCB *acb;
2449 async_context_push();
2451 async_ret = NOT_DONE;
2452 iov.iov_base = (void *)buf;
2453 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2454 qemu_iovec_init_external(&qiov, &iov, 1);
2455 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2456 bdrv_rw_em_cb, &async_ret);
2461 while (async_ret == NOT_DONE) {
2466 async_context_pop();
2470 void bdrv_init(void)
2472 module_call_init(MODULE_INIT_BLOCK);
2475 void bdrv_init_with_whitelist(void)
2477 use_bdrv_whitelist = 1;
2481 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2482 BlockDriverCompletionFunc *cb, void *opaque)
2484 BlockDriverAIOCB *acb;
2486 if (pool->free_aiocb) {
2487 acb = pool->free_aiocb;
2488 pool->free_aiocb = acb->next;
2490 acb = qemu_mallocz(pool->aiocb_size);
2495 acb->opaque = opaque;
2499 void qemu_aio_release(void *p)
2501 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2502 AIOPool *pool = acb->pool;
2503 acb->next = pool->free_aiocb;
2504 pool->free_aiocb = acb;
2507 /**************************************************************/
2508 /* removable device support */
2511 * Return TRUE if the media is present
2513 int bdrv_is_inserted(BlockDriverState *bs)
2515 BlockDriver *drv = bs->drv;
2519 if (!drv->bdrv_is_inserted)
2520 return !bs->tray_open;
2521 ret = drv->bdrv_is_inserted(bs);
2526 * Return TRUE if the media changed since the last call to this
2527 * function. It is currently only used for floppy disks
2529 int bdrv_media_changed(BlockDriverState *bs)
2531 BlockDriver *drv = bs->drv;
2534 if (!drv || !drv->bdrv_media_changed)
2537 ret = drv->bdrv_media_changed(bs);
2538 if (ret == -ENOTSUP)
2539 ret = bs->media_changed;
2540 bs->media_changed = 0;
2545 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2547 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2549 BlockDriver *drv = bs->drv;
2556 if (!drv || !drv->bdrv_eject) {
2559 ret = drv->bdrv_eject(bs, eject_flag);
2561 if (ret == -ENOTSUP) {
2565 bs->tray_open = eject_flag;
2571 int bdrv_is_locked(BlockDriverState *bs)
2577 * Lock or unlock the media (if it is locked, the user won't be able
2578 * to eject it manually).
2580 void bdrv_set_locked(BlockDriverState *bs, int locked)
2582 BlockDriver *drv = bs->drv;
2584 bs->locked = locked;
2585 if (drv && drv->bdrv_set_locked) {
2586 drv->bdrv_set_locked(bs, locked);
2590 /* needed for generic scsi interface */
2592 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2594 BlockDriver *drv = bs->drv;
2596 if (drv && drv->bdrv_ioctl)
2597 return drv->bdrv_ioctl(bs, req, buf);
2601 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2602 unsigned long int req, void *buf,
2603 BlockDriverCompletionFunc *cb, void *opaque)
2605 BlockDriver *drv = bs->drv;
2607 if (drv && drv->bdrv_aio_ioctl)
2608 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2614 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2616 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2619 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2621 int64_t bitmap_size;
2623 bs->dirty_count = 0;
2625 if (!bs->dirty_bitmap) {
2626 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2627 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2628 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2630 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2633 if (bs->dirty_bitmap) {
2634 qemu_free(bs->dirty_bitmap);
2635 bs->dirty_bitmap = NULL;
2640 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2642 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2644 if (bs->dirty_bitmap &&
2645 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2646 return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2647 (1 << (chunk % (sizeof(unsigned long) * 8)));
2653 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2656 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2659 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2661 return bs->dirty_count;