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 /* If non-zero, use only whitelisted block drivers */
67 static int use_bdrv_whitelist
;
69 int path_is_absolute(const char *path
)
73 /* specific case for names like: "\\.\d:" */
74 if (*path
== '/' || *path
== '\\')
77 p
= strchr(path
, ':');
83 return (*p
== '/' || *p
== '\\');
89 /* if filename is absolute, just copy it to dest. Otherwise, build a
90 path to it by considering it is relative to base_path. URL are
92 void path_combine(char *dest
, int dest_size
,
93 const char *base_path
,
101 if (path_is_absolute(filename
)) {
102 pstrcpy(dest
, dest_size
, filename
);
104 p
= strchr(base_path
, ':');
109 p1
= strrchr(base_path
, '/');
113 p2
= strrchr(base_path
, '\\');
125 if (len
> dest_size
- 1)
127 memcpy(dest
, base_path
, len
);
129 pstrcat(dest
, dest_size
, filename
);
133 void bdrv_register(BlockDriver
*bdrv
)
135 if (!bdrv
->bdrv_aio_readv
) {
136 /* add AIO emulation layer */
137 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
138 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
139 } else if (!bdrv
->bdrv_read
) {
140 /* add synchronous IO emulation layer */
141 bdrv
->bdrv_read
= bdrv_read_em
;
142 bdrv
->bdrv_write
= bdrv_write_em
;
145 if (!bdrv
->bdrv_aio_flush
)
146 bdrv
->bdrv_aio_flush
= bdrv_aio_flush_em
;
148 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
151 /* create a new block device (by default it is empty) */
152 BlockDriverState
*bdrv_new(const char *device_name
)
154 BlockDriverState
*bs
;
156 bs
= qemu_mallocz(sizeof(BlockDriverState
));
157 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
158 if (device_name
[0] != '\0') {
159 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
164 BlockDriver
*bdrv_find_format(const char *format_name
)
167 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
168 if (!strcmp(drv1
->format_name
, format_name
)) {
175 static int bdrv_is_whitelisted(BlockDriver
*drv
)
177 static const char *whitelist
[] = {
178 CONFIG_BDRV_WHITELIST
183 return 1; /* no whitelist, anything goes */
185 for (p
= whitelist
; *p
; p
++) {
186 if (!strcmp(drv
->format_name
, *p
)) {
193 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
195 BlockDriver
*drv
= bdrv_find_format(format_name
);
196 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
199 int bdrv_create(BlockDriver
*drv
, const char* filename
,
200 QEMUOptionParameter
*options
)
202 if (!drv
->bdrv_create
)
205 return drv
->bdrv_create(filename
, options
);
208 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
212 drv
= bdrv_find_protocol(filename
);
214 drv
= bdrv_find_format("file");
217 return bdrv_create(drv
, filename
, options
);
221 void get_tmp_filename(char *filename
, int size
)
223 char temp_dir
[MAX_PATH
];
225 GetTempPath(MAX_PATH
, temp_dir
);
226 GetTempFileName(temp_dir
, "qem", 0, filename
);
229 void get_tmp_filename(char *filename
, int size
)
233 /* XXX: race condition possible */
234 tmpdir
= getenv("TMPDIR");
237 snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
);
238 fd
= mkstemp(filename
);
244 static int is_windows_drive_prefix(const char *filename
)
246 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
247 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
251 int is_windows_drive(const char *filename
)
253 if (is_windows_drive_prefix(filename
) &&
256 if (strstart(filename
, "\\\\.\\", NULL
) ||
257 strstart(filename
, "//./", NULL
))
264 * Detect host devices. By convention, /dev/cdrom[N] is always
265 * recognized as a host CDROM.
267 static BlockDriver
*find_hdev_driver(const char *filename
)
269 int score_max
= 0, score
;
270 BlockDriver
*drv
= NULL
, *d
;
272 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
273 if (d
->bdrv_probe_device
) {
274 score
= d
->bdrv_probe_device(filename
);
275 if (score
> score_max
) {
285 BlockDriver
*bdrv_find_protocol(const char *filename
)
293 /* TODO Drivers without bdrv_file_open must be specified explicitly */
296 is_drive
= is_windows_drive(filename
) ||
297 is_windows_drive_prefix(filename
);
301 p
= strchr(filename
, ':');
302 if (!p
|| is_drive
) {
303 drv1
= find_hdev_driver(filename
);
305 drv1
= bdrv_find_format("file");
310 if (len
> sizeof(protocol
) - 1)
311 len
= sizeof(protocol
) - 1;
312 memcpy(protocol
, filename
, len
);
313 protocol
[len
] = '\0';
314 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
315 if (drv1
->protocol_name
&&
316 !strcmp(drv1
->protocol_name
, protocol
)) {
323 static BlockDriver
*find_image_format(const char *filename
)
325 int ret
, score
, score_max
;
326 BlockDriver
*drv1
, *drv
;
328 BlockDriverState
*bs
;
330 ret
= bdrv_file_open(&bs
, filename
, 0);
334 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
335 if (bs
->sg
|| !bdrv_is_inserted(bs
)) {
337 return bdrv_find_format("raw");
340 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
348 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
349 if (drv1
->bdrv_probe
) {
350 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
351 if (score
> score_max
) {
361 * Set the current 'total_sectors' value
363 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
365 BlockDriver
*drv
= bs
->drv
;
367 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
371 /* query actual device if possible, otherwise just trust the hint */
372 if (drv
->bdrv_getlength
) {
373 int64_t length
= drv
->bdrv_getlength(bs
);
377 hint
= length
>> BDRV_SECTOR_BITS
;
380 bs
->total_sectors
= hint
;
385 * Common part for opening disk images and files
387 static int bdrv_open_common(BlockDriverState
*bs
, const char *filename
,
388 int flags
, BlockDriver
*drv
)
395 bs
->total_sectors
= 0;
396 bs
->is_temporary
= 0;
399 bs
->open_flags
= flags
;
400 /* buffer_alignment defaulted to 512, drivers can change this value */
401 bs
->buffer_alignment
= 512;
403 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
405 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
410 bs
->opaque
= qemu_mallocz(drv
->instance_size
);
413 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
414 * write cache to the guest. We do need the fdatasync to flush
415 * out transactions for block allocations, and we maybe have a
416 * volatile write cache in our backing device to deal with.
418 if (flags
& (BDRV_O_CACHE_WB
|BDRV_O_NOCACHE
))
419 bs
->enable_write_cache
= 1;
422 * Clear flags that are internal to the block layer before opening the
425 open_flags
= flags
& ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
428 * Snapshots should be writeable.
430 if (bs
->is_temporary
) {
431 open_flags
|= BDRV_O_RDWR
;
434 /* Open the image, either directly or using a protocol */
435 if (drv
->bdrv_file_open
) {
436 ret
= drv
->bdrv_file_open(bs
, filename
, open_flags
);
438 ret
= bdrv_file_open(&bs
->file
, filename
, open_flags
);
440 ret
= drv
->bdrv_open(bs
, open_flags
);
448 bs
->keep_read_only
= bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
450 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
456 if (bs
->is_temporary
) {
464 bdrv_delete(bs
->file
);
467 qemu_free(bs
->opaque
);
474 * Opens a file using a protocol (file, host_device, nbd, ...)
476 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
, int flags
)
478 BlockDriverState
*bs
;
482 drv
= bdrv_find_protocol(filename
);
488 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
499 * Opens a disk image (raw, qcow2, vmdk, ...)
501 int bdrv_open(BlockDriverState
*bs
, const char *filename
, int flags
,
506 if (flags
& BDRV_O_SNAPSHOT
) {
507 BlockDriverState
*bs1
;
510 BlockDriver
*bdrv_qcow2
;
511 QEMUOptionParameter
*options
;
512 char tmp_filename
[PATH_MAX
];
513 char backing_filename
[PATH_MAX
];
515 /* if snapshot, we create a temporary backing file and open it
516 instead of opening 'filename' directly */
518 /* if there is a backing file, use it */
520 ret
= bdrv_open(bs1
, filename
, 0, drv
);
525 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
527 if (bs1
->drv
&& bs1
->drv
->protocol_name
)
532 get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
534 /* Real path is meaningless for protocols */
536 snprintf(backing_filename
, sizeof(backing_filename
),
538 else if (!realpath(filename
, backing_filename
))
541 bdrv_qcow2
= bdrv_find_format("qcow2");
542 options
= parse_option_parameters("", bdrv_qcow2
->create_options
, NULL
);
544 set_option_parameter_int(options
, BLOCK_OPT_SIZE
, total_size
);
545 set_option_parameter(options
, BLOCK_OPT_BACKING_FILE
, backing_filename
);
547 set_option_parameter(options
, BLOCK_OPT_BACKING_FMT
,
551 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, options
);
552 free_option_parameters(options
);
557 filename
= tmp_filename
;
559 bs
->is_temporary
= 1;
562 /* Find the right image format driver */
564 drv
= find_image_format(filename
);
569 goto unlink_and_fail
;
573 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
575 goto unlink_and_fail
;
578 /* If there is a backing file, use it */
579 if ((flags
& BDRV_O_NO_BACKING
) == 0 && bs
->backing_file
[0] != '\0') {
580 char backing_filename
[PATH_MAX
];
582 BlockDriver
*back_drv
= NULL
;
584 bs
->backing_hd
= bdrv_new("");
585 path_combine(backing_filename
, sizeof(backing_filename
),
586 filename
, bs
->backing_file
);
587 if (bs
->backing_format
[0] != '\0')
588 back_drv
= bdrv_find_format(bs
->backing_format
);
590 /* backing files always opened read-only */
592 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
594 ret
= bdrv_open(bs
->backing_hd
, backing_filename
, back_flags
, back_drv
);
599 if (bs
->is_temporary
) {
600 bs
->backing_hd
->keep_read_only
= !(flags
& BDRV_O_RDWR
);
602 /* base image inherits from "parent" */
603 bs
->backing_hd
->keep_read_only
= bs
->keep_read_only
;
607 if (!bdrv_key_required(bs
)) {
608 /* call the change callback */
609 bs
->media_changed
= 1;
611 bs
->change_cb(bs
->change_opaque
);
617 if (bs
->is_temporary
) {
623 void bdrv_close(BlockDriverState
*bs
)
626 if (bs
->backing_hd
) {
627 bdrv_delete(bs
->backing_hd
);
628 bs
->backing_hd
= NULL
;
630 bs
->drv
->bdrv_close(bs
);
631 qemu_free(bs
->opaque
);
633 if (bs
->is_temporary
) {
634 unlink(bs
->filename
);
640 if (bs
->file
!= NULL
) {
641 bdrv_close(bs
->file
);
644 /* call the change callback */
645 bs
->media_changed
= 1;
647 bs
->change_cb(bs
->change_opaque
);
651 void bdrv_close_all(void)
653 BlockDriverState
*bs
;
655 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
660 void bdrv_delete(BlockDriverState
*bs
)
662 /* remove from list, if necessary */
663 if (bs
->device_name
[0] != '\0') {
664 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
668 if (bs
->file
!= NULL
) {
669 bdrv_delete(bs
->file
);
676 * Run consistency checks on an image
678 * Returns the number of errors or -errno when an internal error occurs
680 int bdrv_check(BlockDriverState
*bs
)
682 if (bs
->drv
->bdrv_check
== NULL
) {
686 return bs
->drv
->bdrv_check(bs
);
689 /* commit COW file into the raw image */
690 int bdrv_commit(BlockDriverState
*bs
)
692 BlockDriver
*drv
= bs
->drv
;
693 int64_t i
, total_sectors
;
694 int n
, j
, ro
, open_flags
;
695 int ret
= 0, rw_ret
= 0;
696 unsigned char sector
[BDRV_SECTOR_SIZE
];
698 BlockDriverState
*bs_rw
, *bs_ro
;
703 if (!bs
->backing_hd
) {
707 if (bs
->backing_hd
->keep_read_only
) {
711 ro
= bs
->backing_hd
->read_only
;
712 strncpy(filename
, bs
->backing_hd
->filename
, sizeof(filename
));
713 open_flags
= bs
->backing_hd
->open_flags
;
717 bdrv_delete(bs
->backing_hd
);
718 bs
->backing_hd
= NULL
;
719 bs_rw
= bdrv_new("");
720 rw_ret
= bdrv_open(bs_rw
, filename
, open_flags
| BDRV_O_RDWR
, drv
);
723 /* try to re-open read-only */
724 bs_ro
= bdrv_new("");
725 ret
= bdrv_open(bs_ro
, filename
, open_flags
& ~BDRV_O_RDWR
, drv
);
728 /* drive not functional anymore */
732 bs
->backing_hd
= bs_ro
;
735 bs
->backing_hd
= bs_rw
;
738 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
739 for (i
= 0; i
< total_sectors
;) {
740 if (drv
->bdrv_is_allocated(bs
, i
, 65536, &n
)) {
741 for(j
= 0; j
< n
; j
++) {
742 if (bdrv_read(bs
, i
, sector
, 1) != 0) {
747 if (bdrv_write(bs
->backing_hd
, i
, sector
, 1) != 0) {
758 if (drv
->bdrv_make_empty
) {
759 ret
= drv
->bdrv_make_empty(bs
);
764 * Make sure all data we wrote to the backing device is actually
768 bdrv_flush(bs
->backing_hd
);
774 bdrv_delete(bs
->backing_hd
);
775 bs
->backing_hd
= NULL
;
776 bs_ro
= bdrv_new("");
777 ret
= bdrv_open(bs_ro
, filename
, open_flags
& ~BDRV_O_RDWR
, drv
);
780 /* drive not functional anymore */
784 bs
->backing_hd
= bs_ro
;
785 bs
->backing_hd
->keep_read_only
= 0;
791 void bdrv_commit_all(void)
793 BlockDriverState
*bs
;
795 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
803 * -EINVAL - backing format specified, but no file
804 * -ENOSPC - can't update the backing file because no space is left in the
806 * -ENOTSUP - format driver doesn't support changing the backing file
808 int bdrv_change_backing_file(BlockDriverState
*bs
,
809 const char *backing_file
, const char *backing_fmt
)
811 BlockDriver
*drv
= bs
->drv
;
813 if (drv
->bdrv_change_backing_file
!= NULL
) {
814 return drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
820 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
825 if (!bdrv_is_inserted(bs
))
831 len
= bdrv_getlength(bs
);
836 if ((offset
> len
) || (len
- offset
< size
))
842 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
845 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
846 nb_sectors
* BDRV_SECTOR_SIZE
);
849 /* return < 0 if error. See bdrv_write() for the return codes */
850 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
851 uint8_t *buf
, int nb_sectors
)
853 BlockDriver
*drv
= bs
->drv
;
857 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
860 return drv
->bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
863 static void set_dirty_bitmap(BlockDriverState
*bs
, int64_t sector_num
,
864 int nb_sectors
, int dirty
)
867 unsigned long val
, idx
, bit
;
869 start
= sector_num
/ BDRV_SECTORS_PER_DIRTY_CHUNK
;
870 end
= (sector_num
+ nb_sectors
- 1) / BDRV_SECTORS_PER_DIRTY_CHUNK
;
872 for (; start
<= end
; start
++) {
873 idx
= start
/ (sizeof(unsigned long) * 8);
874 bit
= start
% (sizeof(unsigned long) * 8);
875 val
= bs
->dirty_bitmap
[idx
];
877 if (!(val
& (1 << bit
))) {
882 if (val
& (1 << bit
)) {
887 bs
->dirty_bitmap
[idx
] = val
;
891 /* Return < 0 if error. Important errors are:
892 -EIO generic I/O error (may happen for all errors)
893 -ENOMEDIUM No media inserted.
894 -EINVAL Invalid sector number or nb_sectors
895 -EACCES Trying to write a read-only device
897 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
898 const uint8_t *buf
, int nb_sectors
)
900 BlockDriver
*drv
= bs
->drv
;
905 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
908 if (bs
->dirty_bitmap
) {
909 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
912 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
913 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
916 return drv
->bdrv_write(bs
, sector_num
, buf
, nb_sectors
);
919 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
920 void *buf
, int count1
)
922 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
923 int len
, nb_sectors
, count
;
928 /* first read to align to sector start */
929 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
932 sector_num
= offset
>> BDRV_SECTOR_BITS
;
934 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
936 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
944 /* read the sectors "in place" */
945 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
946 if (nb_sectors
> 0) {
947 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
949 sector_num
+= nb_sectors
;
950 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
955 /* add data from the last sector */
957 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
959 memcpy(buf
, tmp_buf
, count
);
964 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
965 const void *buf
, int count1
)
967 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
968 int len
, nb_sectors
, count
;
973 /* first write to align to sector start */
974 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
977 sector_num
= offset
>> BDRV_SECTOR_BITS
;
979 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
981 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
982 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
991 /* write the sectors "in place" */
992 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
993 if (nb_sectors
> 0) {
994 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
996 sector_num
+= nb_sectors
;
997 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
1002 /* add data from the last sector */
1004 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1006 memcpy(tmp_buf
, buf
, count
);
1007 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
1014 * Truncate file to 'offset' bytes (needed only for file protocols)
1016 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
1018 BlockDriver
*drv
= bs
->drv
;
1022 if (!drv
->bdrv_truncate
)
1026 ret
= drv
->bdrv_truncate(bs
, offset
);
1028 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
1034 * Length of a file in bytes. Return < 0 if error or unknown.
1036 int64_t bdrv_getlength(BlockDriverState
*bs
)
1038 BlockDriver
*drv
= bs
->drv
;
1042 /* Fixed size devices use the total_sectors value for speed instead of
1043 issuing a length query (like lseek) on each call. Also, legacy block
1044 drivers don't provide a bdrv_getlength function and must use
1046 if (!bs
->growable
|| !drv
->bdrv_getlength
) {
1047 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
1049 return drv
->bdrv_getlength(bs
);
1052 /* return 0 as number of sectors if no device present or error */
1053 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
1056 length
= bdrv_getlength(bs
);
1060 length
= length
>> BDRV_SECTOR_BITS
;
1061 *nb_sectors_ptr
= length
;
1065 uint8_t boot_ind
; /* 0x80 - active */
1066 uint8_t head
; /* starting head */
1067 uint8_t sector
; /* starting sector */
1068 uint8_t cyl
; /* starting cylinder */
1069 uint8_t sys_ind
; /* What partition type */
1070 uint8_t end_head
; /* end head */
1071 uint8_t end_sector
; /* end sector */
1072 uint8_t end_cyl
; /* end cylinder */
1073 uint32_t start_sect
; /* starting sector counting from 0 */
1074 uint32_t nr_sects
; /* nr of sectors in partition */
1075 } __attribute__((packed
));
1077 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1078 static int guess_disk_lchs(BlockDriverState
*bs
,
1079 int *pcylinders
, int *pheads
, int *psectors
)
1081 uint8_t buf
[BDRV_SECTOR_SIZE
];
1082 int ret
, i
, heads
, sectors
, cylinders
;
1083 struct partition
*p
;
1085 uint64_t nb_sectors
;
1087 bdrv_get_geometry(bs
, &nb_sectors
);
1089 ret
= bdrv_read(bs
, 0, buf
, 1);
1092 /* test msdos magic */
1093 if (buf
[510] != 0x55 || buf
[511] != 0xaa)
1095 for(i
= 0; i
< 4; i
++) {
1096 p
= ((struct partition
*)(buf
+ 0x1be)) + i
;
1097 nr_sects
= le32_to_cpu(p
->nr_sects
);
1098 if (nr_sects
&& p
->end_head
) {
1099 /* We make the assumption that the partition terminates on
1100 a cylinder boundary */
1101 heads
= p
->end_head
+ 1;
1102 sectors
= p
->end_sector
& 63;
1105 cylinders
= nb_sectors
/ (heads
* sectors
);
1106 if (cylinders
< 1 || cylinders
> 16383)
1109 *psectors
= sectors
;
1110 *pcylinders
= cylinders
;
1112 printf("guessed geometry: LCHS=%d %d %d\n",
1113 cylinders
, heads
, sectors
);
1121 void bdrv_guess_geometry(BlockDriverState
*bs
, int *pcyls
, int *pheads
, int *psecs
)
1123 int translation
, lba_detected
= 0;
1124 int cylinders
, heads
, secs
;
1125 uint64_t nb_sectors
;
1127 /* if a geometry hint is available, use it */
1128 bdrv_get_geometry(bs
, &nb_sectors
);
1129 bdrv_get_geometry_hint(bs
, &cylinders
, &heads
, &secs
);
1130 translation
= bdrv_get_translation_hint(bs
);
1131 if (cylinders
!= 0) {
1136 if (guess_disk_lchs(bs
, &cylinders
, &heads
, &secs
) == 0) {
1138 /* if heads > 16, it means that a BIOS LBA
1139 translation was active, so the default
1140 hardware geometry is OK */
1142 goto default_geometry
;
1147 /* disable any translation to be in sync with
1148 the logical geometry */
1149 if (translation
== BIOS_ATA_TRANSLATION_AUTO
) {
1150 bdrv_set_translation_hint(bs
,
1151 BIOS_ATA_TRANSLATION_NONE
);
1156 /* if no geometry, use a standard physical disk geometry */
1157 cylinders
= nb_sectors
/ (16 * 63);
1159 if (cylinders
> 16383)
1161 else if (cylinders
< 2)
1166 if ((lba_detected
== 1) && (translation
== BIOS_ATA_TRANSLATION_AUTO
)) {
1167 if ((*pcyls
* *pheads
) <= 131072) {
1168 bdrv_set_translation_hint(bs
,
1169 BIOS_ATA_TRANSLATION_LARGE
);
1171 bdrv_set_translation_hint(bs
,
1172 BIOS_ATA_TRANSLATION_LBA
);
1176 bdrv_set_geometry_hint(bs
, *pcyls
, *pheads
, *psecs
);
1180 void bdrv_set_geometry_hint(BlockDriverState
*bs
,
1181 int cyls
, int heads
, int secs
)
1188 void bdrv_set_type_hint(BlockDriverState
*bs
, int type
)
1191 bs
->removable
= ((type
== BDRV_TYPE_CDROM
||
1192 type
== BDRV_TYPE_FLOPPY
));
1195 void bdrv_set_translation_hint(BlockDriverState
*bs
, int translation
)
1197 bs
->translation
= translation
;
1200 void bdrv_get_geometry_hint(BlockDriverState
*bs
,
1201 int *pcyls
, int *pheads
, int *psecs
)
1204 *pheads
= bs
->heads
;
1208 int bdrv_get_type_hint(BlockDriverState
*bs
)
1213 int bdrv_get_translation_hint(BlockDriverState
*bs
)
1215 return bs
->translation
;
1218 void bdrv_set_on_error(BlockDriverState
*bs
, BlockErrorAction on_read_error
,
1219 BlockErrorAction on_write_error
)
1221 bs
->on_read_error
= on_read_error
;
1222 bs
->on_write_error
= on_write_error
;
1225 BlockErrorAction
bdrv_get_on_error(BlockDriverState
*bs
, int is_read
)
1227 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
1230 int bdrv_is_removable(BlockDriverState
*bs
)
1232 return bs
->removable
;
1235 int bdrv_is_read_only(BlockDriverState
*bs
)
1237 return bs
->read_only
;
1240 int bdrv_is_sg(BlockDriverState
*bs
)
1245 int bdrv_enable_write_cache(BlockDriverState
*bs
)
1247 return bs
->enable_write_cache
;
1250 /* XXX: no longer used */
1251 void bdrv_set_change_cb(BlockDriverState
*bs
,
1252 void (*change_cb
)(void *opaque
), void *opaque
)
1254 bs
->change_cb
= change_cb
;
1255 bs
->change_opaque
= opaque
;
1258 int bdrv_is_encrypted(BlockDriverState
*bs
)
1260 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
1262 return bs
->encrypted
;
1265 int bdrv_key_required(BlockDriverState
*bs
)
1267 BlockDriverState
*backing_hd
= bs
->backing_hd
;
1269 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
1271 return (bs
->encrypted
&& !bs
->valid_key
);
1274 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
1277 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
1278 ret
= bdrv_set_key(bs
->backing_hd
, key
);
1284 if (!bs
->encrypted
) {
1286 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
1289 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
1292 } else if (!bs
->valid_key
) {
1294 /* call the change callback now, we skipped it on open */
1295 bs
->media_changed
= 1;
1297 bs
->change_cb(bs
->change_opaque
);
1302 void bdrv_get_format(BlockDriverState
*bs
, char *buf
, int buf_size
)
1307 pstrcpy(buf
, buf_size
, bs
->drv
->format_name
);
1311 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
1316 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
1317 it(opaque
, drv
->format_name
);
1321 BlockDriverState
*bdrv_find(const char *name
)
1323 BlockDriverState
*bs
;
1325 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1326 if (!strcmp(name
, bs
->device_name
)) {
1333 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
1336 return QTAILQ_FIRST(&bdrv_states
);
1338 return QTAILQ_NEXT(bs
, list
);
1341 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
1343 BlockDriverState
*bs
;
1345 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1350 const char *bdrv_get_device_name(BlockDriverState
*bs
)
1352 return bs
->device_name
;
1355 void bdrv_flush(BlockDriverState
*bs
)
1357 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
1361 if (bs
->drv
&& bs
->drv
->bdrv_flush
)
1362 bs
->drv
->bdrv_flush(bs
);
1365 void bdrv_flush_all(void)
1367 BlockDriverState
*bs
;
1369 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1370 if (bs
->drv
&& !bdrv_is_read_only(bs
) &&
1371 (!bdrv_is_removable(bs
) || bdrv_is_inserted(bs
))) {
1377 int bdrv_has_zero_init(BlockDriverState
*bs
)
1381 if (bs
->drv
->no_zero_init
) {
1383 } else if (bs
->file
) {
1384 return bdrv_has_zero_init(bs
->file
);
1391 * Returns true iff the specified sector is present in the disk image. Drivers
1392 * not implementing the functionality are assumed to not support backing files,
1393 * hence all their sectors are reported as allocated.
1395 * 'pnum' is set to the number of sectors (including and immediately following
1396 * the specified sector) that are known to be in the same
1397 * allocated/unallocated state.
1399 * 'nb_sectors' is the max value 'pnum' should be set to.
1401 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
1405 if (!bs
->drv
->bdrv_is_allocated
) {
1406 if (sector_num
>= bs
->total_sectors
) {
1410 n
= bs
->total_sectors
- sector_num
;
1411 *pnum
= (n
< nb_sectors
) ? (n
) : (nb_sectors
);
1414 return bs
->drv
->bdrv_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
1417 void bdrv_mon_event(const BlockDriverState
*bdrv
,
1418 BlockMonEventAction action
, int is_read
)
1421 const char *action_str
;
1424 case BDRV_ACTION_REPORT
:
1425 action_str
= "report";
1427 case BDRV_ACTION_IGNORE
:
1428 action_str
= "ignore";
1430 case BDRV_ACTION_STOP
:
1431 action_str
= "stop";
1437 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1440 is_read
? "read" : "write");
1441 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR
, data
);
1443 qobject_decref(data
);
1446 static void bdrv_print_dict(QObject
*obj
, void *opaque
)
1449 Monitor
*mon
= opaque
;
1451 bs_dict
= qobject_to_qdict(obj
);
1453 monitor_printf(mon
, "%s: type=%s removable=%d",
1454 qdict_get_str(bs_dict
, "device"),
1455 qdict_get_str(bs_dict
, "type"),
1456 qdict_get_bool(bs_dict
, "removable"));
1458 if (qdict_get_bool(bs_dict
, "removable")) {
1459 monitor_printf(mon
, " locked=%d", qdict_get_bool(bs_dict
, "locked"));
1462 if (qdict_haskey(bs_dict
, "inserted")) {
1463 QDict
*qdict
= qobject_to_qdict(qdict_get(bs_dict
, "inserted"));
1465 monitor_printf(mon
, " file=");
1466 monitor_print_filename(mon
, qdict_get_str(qdict
, "file"));
1467 if (qdict_haskey(qdict
, "backing_file")) {
1468 monitor_printf(mon
, " backing_file=");
1469 monitor_print_filename(mon
, qdict_get_str(qdict
, "backing_file"));
1471 monitor_printf(mon
, " ro=%d drv=%s encrypted=%d",
1472 qdict_get_bool(qdict
, "ro"),
1473 qdict_get_str(qdict
, "drv"),
1474 qdict_get_bool(qdict
, "encrypted"));
1476 monitor_printf(mon
, " [not inserted]");
1479 monitor_printf(mon
, "\n");
1482 void bdrv_info_print(Monitor
*mon
, const QObject
*data
)
1484 qlist_iter(qobject_to_qlist(data
), bdrv_print_dict
, mon
);
1487 void bdrv_info(Monitor
*mon
, QObject
**ret_data
)
1490 BlockDriverState
*bs
;
1492 bs_list
= qlist_new();
1494 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1496 const char *type
= "unknown";
1502 case BDRV_TYPE_CDROM
:
1505 case BDRV_TYPE_FLOPPY
:
1510 bs_obj
= qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1511 "'removable': %i, 'locked': %i }",
1512 bs
->device_name
, type
, bs
->removable
,
1517 QDict
*bs_dict
= qobject_to_qdict(bs_obj
);
1519 obj
= qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1520 "'encrypted': %i }",
1521 bs
->filename
, bs
->read_only
,
1522 bs
->drv
->format_name
,
1523 bdrv_is_encrypted(bs
));
1524 if (bs
->backing_file
[0] != '\0') {
1525 QDict
*qdict
= qobject_to_qdict(obj
);
1526 qdict_put(qdict
, "backing_file",
1527 qstring_from_str(bs
->backing_file
));
1530 qdict_put_obj(bs_dict
, "inserted", obj
);
1532 qlist_append_obj(bs_list
, bs_obj
);
1535 *ret_data
= QOBJECT(bs_list
);
1538 static void bdrv_stats_iter(QObject
*data
, void *opaque
)
1541 Monitor
*mon
= opaque
;
1543 qdict
= qobject_to_qdict(data
);
1544 monitor_printf(mon
, "%s:", qdict_get_str(qdict
, "device"));
1546 qdict
= qobject_to_qdict(qdict_get(qdict
, "stats"));
1547 monitor_printf(mon
, " rd_bytes=%" PRId64
1548 " wr_bytes=%" PRId64
1549 " rd_operations=%" PRId64
1550 " wr_operations=%" PRId64
1552 qdict_get_int(qdict
, "rd_bytes"),
1553 qdict_get_int(qdict
, "wr_bytes"),
1554 qdict_get_int(qdict
, "rd_operations"),
1555 qdict_get_int(qdict
, "wr_operations"));
1558 void bdrv_stats_print(Monitor
*mon
, const QObject
*data
)
1560 qlist_iter(qobject_to_qlist(data
), bdrv_stats_iter
, mon
);
1563 static QObject
* bdrv_info_stats_bs(BlockDriverState
*bs
)
1568 res
= qobject_from_jsonf("{ 'stats': {"
1569 "'rd_bytes': %" PRId64
","
1570 "'wr_bytes': %" PRId64
","
1571 "'rd_operations': %" PRId64
","
1572 "'wr_operations': %" PRId64
","
1573 "'wr_highest_offset': %" PRId64
1575 bs
->rd_bytes
, bs
->wr_bytes
,
1576 bs
->rd_ops
, bs
->wr_ops
,
1577 bs
->wr_highest_sector
* (long)BDRV_SECTOR_SIZE
);
1578 dict
= qobject_to_qdict(res
);
1580 if (*bs
->device_name
) {
1581 qdict_put(dict
, "device", qstring_from_str(bs
->device_name
));
1585 QObject
*parent
= bdrv_info_stats_bs(bs
->file
);
1586 qdict_put_obj(dict
, "parent", parent
);
1592 void bdrv_info_stats(Monitor
*mon
, QObject
**ret_data
)
1596 BlockDriverState
*bs
;
1598 devices
= qlist_new();
1600 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1601 obj
= bdrv_info_stats_bs(bs
);
1602 qlist_append_obj(devices
, obj
);
1605 *ret_data
= QOBJECT(devices
);
1608 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
1610 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
1611 return bs
->backing_file
;
1612 else if (bs
->encrypted
)
1613 return bs
->filename
;
1618 void bdrv_get_backing_filename(BlockDriverState
*bs
,
1619 char *filename
, int filename_size
)
1621 if (!bs
->backing_file
) {
1622 pstrcpy(filename
, filename_size
, "");
1624 pstrcpy(filename
, filename_size
, bs
->backing_file
);
1628 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1629 const uint8_t *buf
, int nb_sectors
)
1631 BlockDriver
*drv
= bs
->drv
;
1634 if (!drv
->bdrv_write_compressed
)
1636 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1639 if (bs
->dirty_bitmap
) {
1640 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1643 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1646 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1648 BlockDriver
*drv
= bs
->drv
;
1651 if (!drv
->bdrv_get_info
)
1653 memset(bdi
, 0, sizeof(*bdi
));
1654 return drv
->bdrv_get_info(bs
, bdi
);
1657 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1658 int64_t pos
, int size
)
1660 BlockDriver
*drv
= bs
->drv
;
1663 if (drv
->bdrv_save_vmstate
)
1664 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
1666 return bdrv_save_vmstate(bs
->file
, buf
, pos
, size
);
1670 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1671 int64_t pos
, int size
)
1673 BlockDriver
*drv
= bs
->drv
;
1676 if (drv
->bdrv_load_vmstate
)
1677 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1679 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
1683 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
1685 BlockDriver
*drv
= bs
->drv
;
1687 if (!drv
|| !drv
->bdrv_debug_event
) {
1691 return drv
->bdrv_debug_event(bs
, event
);
1695 /**************************************************************/
1696 /* handling of snapshots */
1698 int bdrv_can_snapshot(BlockDriverState
*bs
)
1700 BlockDriver
*drv
= bs
->drv
;
1701 if (!drv
|| bdrv_is_removable(bs
) || bdrv_is_read_only(bs
)) {
1705 if (!drv
->bdrv_snapshot_create
) {
1706 if (bs
->file
!= NULL
) {
1707 return bdrv_can_snapshot(bs
->file
);
1715 int bdrv_snapshot_create(BlockDriverState
*bs
,
1716 QEMUSnapshotInfo
*sn_info
)
1718 BlockDriver
*drv
= bs
->drv
;
1721 if (drv
->bdrv_snapshot_create
)
1722 return drv
->bdrv_snapshot_create(bs
, sn_info
);
1724 return bdrv_snapshot_create(bs
->file
, sn_info
);
1728 int bdrv_snapshot_goto(BlockDriverState
*bs
,
1729 const char *snapshot_id
)
1731 BlockDriver
*drv
= bs
->drv
;
1736 if (drv
->bdrv_snapshot_goto
)
1737 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
1740 drv
->bdrv_close(bs
);
1741 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
1742 open_ret
= drv
->bdrv_open(bs
, bs
->open_flags
);
1744 bdrv_delete(bs
->file
);
1754 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
1756 BlockDriver
*drv
= bs
->drv
;
1759 if (drv
->bdrv_snapshot_delete
)
1760 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
1762 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
1766 int bdrv_snapshot_list(BlockDriverState
*bs
,
1767 QEMUSnapshotInfo
**psn_info
)
1769 BlockDriver
*drv
= bs
->drv
;
1772 if (drv
->bdrv_snapshot_list
)
1773 return drv
->bdrv_snapshot_list(bs
, psn_info
);
1775 return bdrv_snapshot_list(bs
->file
, psn_info
);
1779 #define NB_SUFFIXES 4
1781 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
1783 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
1788 snprintf(buf
, buf_size
, "%" PRId64
, size
);
1791 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
1792 if (size
< (10 * base
)) {
1793 snprintf(buf
, buf_size
, "%0.1f%c",
1794 (double)size
/ base
,
1797 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
1798 snprintf(buf
, buf_size
, "%" PRId64
"%c",
1799 ((size
+ (base
>> 1)) / base
),
1809 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
1811 char buf1
[128], date_buf
[128], clock_buf
[128];
1821 snprintf(buf
, buf_size
,
1822 "%-10s%-20s%7s%20s%15s",
1823 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1827 ptm
= localtime(&ti
);
1828 strftime(date_buf
, sizeof(date_buf
),
1829 "%Y-%m-%d %H:%M:%S", ptm
);
1831 localtime_r(&ti
, &tm
);
1832 strftime(date_buf
, sizeof(date_buf
),
1833 "%Y-%m-%d %H:%M:%S", &tm
);
1835 secs
= sn
->vm_clock_nsec
/ 1000000000;
1836 snprintf(clock_buf
, sizeof(clock_buf
),
1837 "%02d:%02d:%02d.%03d",
1839 (int)((secs
/ 60) % 60),
1841 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
1842 snprintf(buf
, buf_size
,
1843 "%-10s%-20s%7s%20s%15s",
1844 sn
->id_str
, sn
->name
,
1845 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
1853 /**************************************************************/
1856 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1857 QEMUIOVector
*qiov
, int nb_sectors
,
1858 BlockDriverCompletionFunc
*cb
, void *opaque
)
1860 BlockDriver
*drv
= bs
->drv
;
1861 BlockDriverAIOCB
*ret
;
1865 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1868 ret
= drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
1872 /* Update stats even though technically transfer has not happened. */
1873 bs
->rd_bytes
+= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
1880 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1881 QEMUIOVector
*qiov
, int nb_sectors
,
1882 BlockDriverCompletionFunc
*cb
, void *opaque
)
1884 BlockDriver
*drv
= bs
->drv
;
1885 BlockDriverAIOCB
*ret
;
1891 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1894 if (bs
->dirty_bitmap
) {
1895 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1898 ret
= drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
1902 /* Update stats even though technically transfer has not happened. */
1903 bs
->wr_bytes
+= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
1905 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
1906 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
1914 typedef struct MultiwriteCB
{
1919 BlockDriverCompletionFunc
*cb
;
1921 QEMUIOVector
*free_qiov
;
1926 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
1930 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
1931 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
1932 if (mcb
->callbacks
[i
].free_qiov
) {
1933 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
1935 qemu_free(mcb
->callbacks
[i
].free_qiov
);
1936 qemu_vfree(mcb
->callbacks
[i
].free_buf
);
1940 static void multiwrite_cb(void *opaque
, int ret
)
1942 MultiwriteCB
*mcb
= opaque
;
1944 if (ret
< 0 && !mcb
->error
) {
1946 multiwrite_user_cb(mcb
);
1949 mcb
->num_requests
--;
1950 if (mcb
->num_requests
== 0) {
1951 if (mcb
->error
== 0) {
1952 multiwrite_user_cb(mcb
);
1958 static int multiwrite_req_compare(const void *a
, const void *b
)
1960 const BlockRequest
*req1
= a
, *req2
= b
;
1963 * Note that we can't simply subtract req2->sector from req1->sector
1964 * here as that could overflow the return value.
1966 if (req1
->sector
> req2
->sector
) {
1968 } else if (req1
->sector
< req2
->sector
) {
1976 * Takes a bunch of requests and tries to merge them. Returns the number of
1977 * requests that remain after merging.
1979 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
1980 int num_reqs
, MultiwriteCB
*mcb
)
1984 // Sort requests by start sector
1985 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
1987 // Check if adjacent requests touch the same clusters. If so, combine them,
1988 // filling up gaps with zero sectors.
1990 for (i
= 1; i
< num_reqs
; i
++) {
1992 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
1994 // This handles the cases that are valid for all block drivers, namely
1995 // exactly sequential writes and overlapping writes.
1996 if (reqs
[i
].sector
<= oldreq_last
) {
2000 // The block driver may decide that it makes sense to combine requests
2001 // even if there is a gap of some sectors between them. In this case,
2002 // the gap is filled with zeros (therefore only applicable for yet
2003 // unused space in format like qcow2).
2004 if (!merge
&& bs
->drv
->bdrv_merge_requests
) {
2005 merge
= bs
->drv
->bdrv_merge_requests(bs
, &reqs
[outidx
], &reqs
[i
]);
2008 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
2014 QEMUIOVector
*qiov
= qemu_mallocz(sizeof(*qiov
));
2015 qemu_iovec_init(qiov
,
2016 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
2018 // Add the first request to the merged one. If the requests are
2019 // overlapping, drop the last sectors of the first request.
2020 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
2021 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, size
);
2023 // We might need to add some zeros between the two requests
2024 if (reqs
[i
].sector
> oldreq_last
) {
2025 size_t zero_bytes
= (reqs
[i
].sector
- oldreq_last
) << 9;
2026 uint8_t *buf
= qemu_blockalign(bs
, zero_bytes
);
2027 memset(buf
, 0, zero_bytes
);
2028 qemu_iovec_add(qiov
, buf
, zero_bytes
);
2029 mcb
->callbacks
[i
].free_buf
= buf
;
2032 // Add the second request
2033 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, reqs
[i
].qiov
->size
);
2035 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
2036 reqs
[outidx
].qiov
= qiov
;
2038 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
2041 reqs
[outidx
].sector
= reqs
[i
].sector
;
2042 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
2043 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
2051 * Submit multiple AIO write requests at once.
2053 * On success, the function returns 0 and all requests in the reqs array have
2054 * been submitted. In error case this function returns -1, and any of the
2055 * requests may or may not be submitted yet. In particular, this means that the
2056 * callback will be called for some of the requests, for others it won't. The
2057 * caller must check the error field of the BlockRequest to wait for the right
2058 * callbacks (if error != 0, no callback will be called).
2060 * The implementation may modify the contents of the reqs array, e.g. to merge
2061 * requests. However, the fields opaque and error are left unmodified as they
2062 * are used to signal failure for a single request to the caller.
2064 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
2066 BlockDriverAIOCB
*acb
;
2070 if (num_reqs
== 0) {
2074 // Create MultiwriteCB structure
2075 mcb
= qemu_mallocz(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
2076 mcb
->num_requests
= 0;
2077 mcb
->num_callbacks
= num_reqs
;
2079 for (i
= 0; i
< num_reqs
; i
++) {
2080 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
2081 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
2084 // Check for mergable requests
2085 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
2087 // Run the aio requests
2088 for (i
= 0; i
< num_reqs
; i
++) {
2089 acb
= bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
2090 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
2093 // We can only fail the whole thing if no request has been
2094 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2095 // complete and report the error in the callback.
2096 if (mcb
->num_requests
== 0) {
2097 reqs
[i
].error
= -EIO
;
2100 mcb
->num_requests
++;
2101 multiwrite_cb(mcb
, -EIO
);
2105 mcb
->num_requests
++;
2116 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2117 BlockDriverCompletionFunc
*cb
, void *opaque
)
2119 BlockDriver
*drv
= bs
->drv
;
2121 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2122 return bdrv_aio_noop_em(bs
, cb
, opaque
);
2127 return drv
->bdrv_aio_flush(bs
, cb
, opaque
);
2130 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
2132 acb
->pool
->cancel(acb
);
2136 /**************************************************************/
2137 /* async block device emulation */
2139 typedef struct BlockDriverAIOCBSync
{
2140 BlockDriverAIOCB common
;
2143 /* vector translation state */
2147 } BlockDriverAIOCBSync
;
2149 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
2151 BlockDriverAIOCBSync
*acb
=
2152 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
2153 qemu_bh_delete(acb
->bh
);
2155 qemu_aio_release(acb
);
2158 static AIOPool bdrv_em_aio_pool
= {
2159 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
2160 .cancel
= bdrv_aio_cancel_em
,
2163 static void bdrv_aio_bh_cb(void *opaque
)
2165 BlockDriverAIOCBSync
*acb
= opaque
;
2168 qemu_iovec_from_buffer(acb
->qiov
, acb
->bounce
, acb
->qiov
->size
);
2169 qemu_vfree(acb
->bounce
);
2170 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
2171 qemu_bh_delete(acb
->bh
);
2173 qemu_aio_release(acb
);
2176 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
2180 BlockDriverCompletionFunc
*cb
,
2185 BlockDriverAIOCBSync
*acb
;
2187 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2188 acb
->is_write
= is_write
;
2190 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
2193 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2196 qemu_iovec_to_buffer(acb
->qiov
, acb
->bounce
);
2197 acb
->ret
= bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2199 acb
->ret
= bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2202 qemu_bh_schedule(acb
->bh
);
2204 return &acb
->common
;
2207 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2208 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2209 BlockDriverCompletionFunc
*cb
, void *opaque
)
2211 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2214 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2215 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2216 BlockDriverCompletionFunc
*cb
, void *opaque
)
2218 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2221 static BlockDriverAIOCB
*bdrv_aio_flush_em(BlockDriverState
*bs
,
2222 BlockDriverCompletionFunc
*cb
, void *opaque
)
2224 BlockDriverAIOCBSync
*acb
;
2226 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2227 acb
->is_write
= 1; /* don't bounce in the completion hadler */
2233 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2236 qemu_bh_schedule(acb
->bh
);
2237 return &acb
->common
;
2240 static BlockDriverAIOCB
*bdrv_aio_noop_em(BlockDriverState
*bs
,
2241 BlockDriverCompletionFunc
*cb
, void *opaque
)
2243 BlockDriverAIOCBSync
*acb
;
2245 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2246 acb
->is_write
= 1; /* don't bounce in the completion handler */
2252 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2255 qemu_bh_schedule(acb
->bh
);
2256 return &acb
->common
;
2259 /**************************************************************/
2260 /* sync block device emulation */
2262 static void bdrv_rw_em_cb(void *opaque
, int ret
)
2264 *(int *)opaque
= ret
;
2267 #define NOT_DONE 0x7fffffff
2269 static int bdrv_read_em(BlockDriverState
*bs
, int64_t sector_num
,
2270 uint8_t *buf
, int nb_sectors
)
2273 BlockDriverAIOCB
*acb
;
2277 async_context_push();
2279 async_ret
= NOT_DONE
;
2280 iov
.iov_base
= (void *)buf
;
2281 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2282 qemu_iovec_init_external(&qiov
, &iov
, 1);
2283 acb
= bdrv_aio_readv(bs
, sector_num
, &qiov
, nb_sectors
,
2284 bdrv_rw_em_cb
, &async_ret
);
2290 while (async_ret
== NOT_DONE
) {
2296 async_context_pop();
2300 static int bdrv_write_em(BlockDriverState
*bs
, int64_t sector_num
,
2301 const uint8_t *buf
, int nb_sectors
)
2304 BlockDriverAIOCB
*acb
;
2308 async_context_push();
2310 async_ret
= NOT_DONE
;
2311 iov
.iov_base
= (void *)buf
;
2312 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2313 qemu_iovec_init_external(&qiov
, &iov
, 1);
2314 acb
= bdrv_aio_writev(bs
, sector_num
, &qiov
, nb_sectors
,
2315 bdrv_rw_em_cb
, &async_ret
);
2320 while (async_ret
== NOT_DONE
) {
2325 async_context_pop();
2329 void bdrv_init(void)
2331 module_call_init(MODULE_INIT_BLOCK
);
2334 void bdrv_init_with_whitelist(void)
2336 use_bdrv_whitelist
= 1;
2340 void *qemu_aio_get(AIOPool
*pool
, BlockDriverState
*bs
,
2341 BlockDriverCompletionFunc
*cb
, void *opaque
)
2343 BlockDriverAIOCB
*acb
;
2345 if (pool
->free_aiocb
) {
2346 acb
= pool
->free_aiocb
;
2347 pool
->free_aiocb
= acb
->next
;
2349 acb
= qemu_mallocz(pool
->aiocb_size
);
2354 acb
->opaque
= opaque
;
2358 void qemu_aio_release(void *p
)
2360 BlockDriverAIOCB
*acb
= (BlockDriverAIOCB
*)p
;
2361 AIOPool
*pool
= acb
->pool
;
2362 acb
->next
= pool
->free_aiocb
;
2363 pool
->free_aiocb
= acb
;
2366 /**************************************************************/
2367 /* removable device support */
2370 * Return TRUE if the media is present
2372 int bdrv_is_inserted(BlockDriverState
*bs
)
2374 BlockDriver
*drv
= bs
->drv
;
2378 if (!drv
->bdrv_is_inserted
)
2380 ret
= drv
->bdrv_is_inserted(bs
);
2385 * Return TRUE if the media changed since the last call to this
2386 * function. It is currently only used for floppy disks
2388 int bdrv_media_changed(BlockDriverState
*bs
)
2390 BlockDriver
*drv
= bs
->drv
;
2393 if (!drv
|| !drv
->bdrv_media_changed
)
2396 ret
= drv
->bdrv_media_changed(bs
);
2397 if (ret
== -ENOTSUP
)
2398 ret
= bs
->media_changed
;
2399 bs
->media_changed
= 0;
2404 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2406 int bdrv_eject(BlockDriverState
*bs
, int eject_flag
)
2408 BlockDriver
*drv
= bs
->drv
;
2415 if (!drv
|| !drv
->bdrv_eject
) {
2418 ret
= drv
->bdrv_eject(bs
, eject_flag
);
2420 if (ret
== -ENOTSUP
) {
2429 int bdrv_is_locked(BlockDriverState
*bs
)
2435 * Lock or unlock the media (if it is locked, the user won't be able
2436 * to eject it manually).
2438 void bdrv_set_locked(BlockDriverState
*bs
, int locked
)
2440 BlockDriver
*drv
= bs
->drv
;
2442 bs
->locked
= locked
;
2443 if (drv
&& drv
->bdrv_set_locked
) {
2444 drv
->bdrv_set_locked(bs
, locked
);
2448 /* needed for generic scsi interface */
2450 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2452 BlockDriver
*drv
= bs
->drv
;
2454 if (drv
&& drv
->bdrv_ioctl
)
2455 return drv
->bdrv_ioctl(bs
, req
, buf
);
2459 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2460 unsigned long int req
, void *buf
,
2461 BlockDriverCompletionFunc
*cb
, void *opaque
)
2463 BlockDriver
*drv
= bs
->drv
;
2465 if (drv
&& drv
->bdrv_aio_ioctl
)
2466 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
2472 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2474 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
2477 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int enable
)
2479 int64_t bitmap_size
;
2481 bs
->dirty_count
= 0;
2483 if (!bs
->dirty_bitmap
) {
2484 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
) +
2485 BDRV_SECTORS_PER_DIRTY_CHUNK
* 8 - 1;
2486 bitmap_size
/= BDRV_SECTORS_PER_DIRTY_CHUNK
* 8;
2488 bs
->dirty_bitmap
= qemu_mallocz(bitmap_size
);
2491 if (bs
->dirty_bitmap
) {
2492 qemu_free(bs
->dirty_bitmap
);
2493 bs
->dirty_bitmap
= NULL
;
2498 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
2500 int64_t chunk
= sector
/ (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK
;
2502 if (bs
->dirty_bitmap
&&
2503 (sector
<< BDRV_SECTOR_BITS
) < bdrv_getlength(bs
)) {
2504 return bs
->dirty_bitmap
[chunk
/ (sizeof(unsigned long) * 8)] &
2505 (1 << (chunk
% (sizeof(unsigned long) * 8)));
2511 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
2514 set_dirty_bitmap(bs
, cur_sector
, nr_sectors
, 0);
2517 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
2519 return bs
->dirty_count
;