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 int bdrv_read_em(BlockDriverState
*bs
, int64_t sector_num
,
54 uint8_t *buf
, int nb_sectors
);
55 static int bdrv_write_em(BlockDriverState
*bs
, int64_t sector_num
,
56 const uint8_t *buf
, int nb_sectors
);
58 BlockDriverState
*bdrv_first
;
60 static BlockDriver
*first_drv
;
62 /* If non-zero, use only whitelisted block drivers */
63 static int use_bdrv_whitelist
;
65 int path_is_absolute(const char *path
)
69 /* specific case for names like: "\\.\d:" */
70 if (*path
== '/' || *path
== '\\')
73 p
= strchr(path
, ':');
79 return (*p
== '/' || *p
== '\\');
85 /* if filename is absolute, just copy it to dest. Otherwise, build a
86 path to it by considering it is relative to base_path. URL are
88 void path_combine(char *dest
, int dest_size
,
89 const char *base_path
,
97 if (path_is_absolute(filename
)) {
98 pstrcpy(dest
, dest_size
, filename
);
100 p
= strchr(base_path
, ':');
105 p1
= strrchr(base_path
, '/');
109 p2
= strrchr(base_path
, '\\');
121 if (len
> dest_size
- 1)
123 memcpy(dest
, base_path
, len
);
125 pstrcat(dest
, dest_size
, filename
);
129 void bdrv_register(BlockDriver
*bdrv
)
131 if (!bdrv
->bdrv_aio_readv
) {
132 /* add AIO emulation layer */
133 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
134 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
135 } else if (!bdrv
->bdrv_read
) {
136 /* add synchronous IO emulation layer */
137 bdrv
->bdrv_read
= bdrv_read_em
;
138 bdrv
->bdrv_write
= bdrv_write_em
;
141 if (!bdrv
->bdrv_aio_flush
)
142 bdrv
->bdrv_aio_flush
= bdrv_aio_flush_em
;
144 bdrv
->next
= first_drv
;
148 /* create a new block device (by default it is empty) */
149 BlockDriverState
*bdrv_new(const char *device_name
)
151 BlockDriverState
**pbs
, *bs
;
153 bs
= qemu_mallocz(sizeof(BlockDriverState
));
154 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
155 if (device_name
[0] != '\0') {
156 /* insert at the end */
165 BlockDriver
*bdrv_find_format(const char *format_name
)
168 for(drv1
= first_drv
; drv1
!= NULL
; drv1
= drv1
->next
) {
169 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
);
209 void get_tmp_filename(char *filename
, int size
)
211 char temp_dir
[MAX_PATH
];
213 GetTempPath(MAX_PATH
, temp_dir
);
214 GetTempFileName(temp_dir
, "qem", 0, filename
);
217 void get_tmp_filename(char *filename
, int size
)
221 /* XXX: race condition possible */
222 tmpdir
= getenv("TMPDIR");
225 snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
);
226 fd
= mkstemp(filename
);
232 static int is_windows_drive_prefix(const char *filename
)
234 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
235 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
239 int is_windows_drive(const char *filename
)
241 if (is_windows_drive_prefix(filename
) &&
244 if (strstart(filename
, "\\\\.\\", NULL
) ||
245 strstart(filename
, "//./", NULL
))
251 static BlockDriver
*find_protocol(const char *filename
)
259 if (is_windows_drive(filename
) ||
260 is_windows_drive_prefix(filename
))
261 return bdrv_find_format("raw");
263 p
= strchr(filename
, ':');
265 return bdrv_find_format("raw");
267 if (len
> sizeof(protocol
) - 1)
268 len
= sizeof(protocol
) - 1;
269 memcpy(protocol
, filename
, len
);
270 protocol
[len
] = '\0';
271 for(drv1
= first_drv
; drv1
!= NULL
; drv1
= drv1
->next
) {
272 if (drv1
->protocol_name
&&
273 !strcmp(drv1
->protocol_name
, protocol
))
280 * Detect host devices. By convention, /dev/cdrom[N] is always
281 * recognized as a host CDROM.
283 static BlockDriver
*find_hdev_driver(const char *filename
)
285 int score_max
= 0, score
;
286 BlockDriver
*drv
= NULL
, *d
;
288 for (d
= first_drv
; d
; d
= d
->next
) {
289 if (d
->bdrv_probe_device
) {
290 score
= d
->bdrv_probe_device(filename
);
291 if (score
> score_max
) {
301 static BlockDriver
*find_image_format(const char *filename
)
303 int ret
, score
, score_max
;
304 BlockDriver
*drv1
, *drv
;
306 BlockDriverState
*bs
;
308 drv
= find_protocol(filename
);
309 /* no need to test disk image formats for vvfat */
310 if (drv
&& strcmp(drv
->format_name
, "vvfat") == 0)
313 ret
= bdrv_file_open(&bs
, filename
, BDRV_O_RDONLY
);
316 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
323 for(drv1
= first_drv
; drv1
!= NULL
; drv1
= drv1
->next
) {
324 if (drv1
->bdrv_probe
) {
325 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
326 if (score
> score_max
) {
335 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
, int flags
)
337 BlockDriverState
*bs
;
341 ret
= bdrv_open2(bs
, filename
, flags
| BDRV_O_FILE
, NULL
);
351 int bdrv_open(BlockDriverState
*bs
, const char *filename
, int flags
)
353 return bdrv_open2(bs
, filename
, flags
, NULL
);
356 int bdrv_open2(BlockDriverState
*bs
, const char *filename
, int flags
,
359 int ret
, open_flags
, try_rw
;
360 char tmp_filename
[PATH_MAX
];
361 char backing_filename
[PATH_MAX
];
363 bs
->is_temporary
= 0;
366 /* buffer_alignment defaulted to 512, drivers can change this value */
367 bs
->buffer_alignment
= 512;
369 if (flags
& BDRV_O_SNAPSHOT
) {
370 BlockDriverState
*bs1
;
373 BlockDriver
*bdrv_qcow2
;
374 QEMUOptionParameter
*options
;
376 /* if snapshot, we create a temporary backing file and open it
377 instead of opening 'filename' directly */
379 /* if there is a backing file, use it */
381 ret
= bdrv_open2(bs1
, filename
, 0, drv
);
386 total_size
= bdrv_getlength(bs1
) >> BDRV_SECTOR_BITS
;
388 if (bs1
->drv
&& bs1
->drv
->protocol_name
)
393 get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
395 /* Real path is meaningless for protocols */
397 snprintf(backing_filename
, sizeof(backing_filename
),
400 realpath(filename
, backing_filename
);
402 bdrv_qcow2
= bdrv_find_format("qcow2");
403 options
= parse_option_parameters("", bdrv_qcow2
->create_options
, NULL
);
405 set_option_parameter_int(options
, BLOCK_OPT_SIZE
, total_size
* 512);
406 set_option_parameter(options
, BLOCK_OPT_BACKING_FILE
, backing_filename
);
408 set_option_parameter(options
, BLOCK_OPT_BACKING_FMT
,
412 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, options
);
417 filename
= tmp_filename
;
419 bs
->is_temporary
= 1;
422 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
423 if (flags
& BDRV_O_FILE
) {
424 drv
= find_protocol(filename
);
426 drv
= find_hdev_driver(filename
);
428 drv
= find_image_format(filename
);
433 goto unlink_and_fail
;
436 bs
->opaque
= qemu_mallocz(drv
->instance_size
);
439 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
440 * write cache to the guest. We do need the fdatasync to flush
441 * out transactions for block allocations, and we maybe have a
442 * volatile write cache in our backing device to deal with.
444 if (flags
& (BDRV_O_CACHE_WB
|BDRV_O_NOCACHE
))
445 bs
->enable_write_cache
= 1;
447 /* Note: for compatibility, we open disk image files as RDWR, and
448 RDONLY as fallback */
449 try_rw
= !bs
->read_only
|| bs
->is_temporary
;
450 if (!(flags
& BDRV_O_FILE
))
451 open_flags
= (try_rw
? BDRV_O_RDWR
: 0) |
452 (flags
& (BDRV_O_CACHE_MASK
|BDRV_O_NATIVE_AIO
));
454 open_flags
= flags
& ~(BDRV_O_FILE
| BDRV_O_SNAPSHOT
);
456 bs
->open_flags
= open_flags
;
457 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
))
460 ret
= drv
->bdrv_open(bs
, filename
, open_flags
);
461 if ((ret
== -EACCES
|| ret
== -EPERM
) && !(flags
& BDRV_O_FILE
)) {
462 ret
= drv
->bdrv_open(bs
, filename
, open_flags
& ~BDRV_O_RDWR
);
466 qemu_free(bs
->opaque
);
470 if (bs
->is_temporary
)
474 if (drv
->bdrv_getlength
) {
475 bs
->total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
478 if (bs
->is_temporary
) {
482 if (bs
->backing_file
[0] != '\0') {
483 /* if there is a backing file, use it */
484 BlockDriver
*back_drv
= NULL
;
485 bs
->backing_hd
= bdrv_new("");
486 /* pass on read_only property to the backing_hd */
487 bs
->backing_hd
->read_only
= bs
->read_only
;
488 path_combine(backing_filename
, sizeof(backing_filename
),
489 filename
, bs
->backing_file
);
490 if (bs
->backing_format
[0] != '\0')
491 back_drv
= bdrv_find_format(bs
->backing_format
);
492 ret
= bdrv_open2(bs
->backing_hd
, backing_filename
, open_flags
,
500 if (!bdrv_key_required(bs
)) {
501 /* call the change callback */
502 bs
->media_changed
= 1;
504 bs
->change_cb(bs
->change_opaque
);
509 void bdrv_close(BlockDriverState
*bs
)
513 bdrv_delete(bs
->backing_hd
);
514 bs
->drv
->bdrv_close(bs
);
515 qemu_free(bs
->opaque
);
517 if (bs
->is_temporary
) {
518 unlink(bs
->filename
);
524 /* call the change callback */
525 bs
->media_changed
= 1;
527 bs
->change_cb(bs
->change_opaque
);
531 void bdrv_delete(BlockDriverState
*bs
)
533 BlockDriverState
**pbs
;
536 while (*pbs
!= bs
&& *pbs
!= NULL
)
546 * Run consistency checks on an image
548 * Returns the number of errors or -errno when an internal error occurs
550 int bdrv_check(BlockDriverState
*bs
)
552 if (bs
->drv
->bdrv_check
== NULL
) {
556 return bs
->drv
->bdrv_check(bs
);
559 /* commit COW file into the raw image */
560 int bdrv_commit(BlockDriverState
*bs
)
562 BlockDriver
*drv
= bs
->drv
;
563 int64_t i
, total_sectors
;
565 unsigned char sector
[512];
574 if (!bs
->backing_hd
) {
578 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
579 for (i
= 0; i
< total_sectors
;) {
580 if (drv
->bdrv_is_allocated(bs
, i
, 65536, &n
)) {
581 for(j
= 0; j
< n
; j
++) {
582 if (bdrv_read(bs
, i
, sector
, 1) != 0) {
586 if (bdrv_write(bs
->backing_hd
, i
, sector
, 1) != 0) {
596 if (drv
->bdrv_make_empty
)
597 return drv
->bdrv_make_empty(bs
);
602 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
607 if (!bdrv_is_inserted(bs
))
613 len
= bdrv_getlength(bs
);
618 if ((offset
> len
) || (len
- offset
< size
))
624 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
627 return bdrv_check_byte_request(bs
, sector_num
* 512, nb_sectors
* 512);
630 /* return < 0 if error. See bdrv_write() for the return codes */
631 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
632 uint8_t *buf
, int nb_sectors
)
634 BlockDriver
*drv
= bs
->drv
;
638 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
641 return drv
->bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
644 static void set_dirty_bitmap(BlockDriverState
*bs
, int64_t sector_num
,
645 int nb_sectors
, int dirty
)
648 unsigned long val
, idx
, bit
;
650 start
= sector_num
/ BDRV_SECTORS_PER_DIRTY_CHUNK
;
651 end
= (sector_num
+ nb_sectors
- 1) / BDRV_SECTORS_PER_DIRTY_CHUNK
;
653 for (; start
<= end
; start
++) {
654 idx
= start
/ (sizeof(unsigned long) * 8);
655 bit
= start
% (sizeof(unsigned long) * 8);
656 val
= bs
->dirty_bitmap
[idx
];
662 bs
->dirty_bitmap
[idx
] = val
;
666 /* Return < 0 if error. Important errors are:
667 -EIO generic I/O error (may happen for all errors)
668 -ENOMEDIUM No media inserted.
669 -EINVAL Invalid sector number or nb_sectors
670 -EACCES Trying to write a read-only device
672 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
673 const uint8_t *buf
, int nb_sectors
)
675 BlockDriver
*drv
= bs
->drv
;
680 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
683 if (bs
->dirty_bitmap
) {
684 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
687 return drv
->bdrv_write(bs
, sector_num
, buf
, nb_sectors
);
690 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
691 void *buf
, int count1
)
693 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
694 int len
, nb_sectors
, count
;
699 /* first read to align to sector start */
700 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
703 sector_num
= offset
>> BDRV_SECTOR_BITS
;
705 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
707 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
715 /* read the sectors "in place" */
716 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
717 if (nb_sectors
> 0) {
718 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
720 sector_num
+= nb_sectors
;
721 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
726 /* add data from the last sector */
728 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
730 memcpy(buf
, tmp_buf
, count
);
735 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
736 const void *buf
, int count1
)
738 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
739 int len
, nb_sectors
, count
;
744 /* first write to align to sector start */
745 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
748 sector_num
= offset
>> BDRV_SECTOR_BITS
;
750 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
752 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
753 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
762 /* write the sectors "in place" */
763 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
764 if (nb_sectors
> 0) {
765 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
767 sector_num
+= nb_sectors
;
768 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
773 /* add data from the last sector */
775 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
777 memcpy(tmp_buf
, buf
, count
);
778 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
785 * Writes to the file and ensures that no writes are reordered across this
786 * request (acts as a barrier)
788 * Returns 0 on success, -errno in error cases.
790 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
791 const void *buf
, int count
)
795 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
800 /* No flush needed for cache=writethrough, it uses O_DSYNC */
801 if ((bs
->open_flags
& BDRV_O_CACHE_MASK
) != 0) {
809 * Writes to the file and ensures that no writes are reordered across this
810 * request (acts as a barrier)
812 * Returns 0 on success, -errno in error cases.
814 int bdrv_write_sync(BlockDriverState
*bs
, int64_t sector_num
,
815 const uint8_t *buf
, int nb_sectors
)
817 return bdrv_pwrite_sync(bs
, BDRV_SECTOR_SIZE
* sector_num
,
818 buf
, BDRV_SECTOR_SIZE
* nb_sectors
);
822 * Truncate file to 'offset' bytes (needed only for file protocols)
824 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
826 BlockDriver
*drv
= bs
->drv
;
829 if (!drv
->bdrv_truncate
)
833 return drv
->bdrv_truncate(bs
, offset
);
837 * Length of a file in bytes. Return < 0 if error or unknown.
839 int64_t bdrv_getlength(BlockDriverState
*bs
)
841 BlockDriver
*drv
= bs
->drv
;
844 if (!drv
->bdrv_getlength
) {
846 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
848 return drv
->bdrv_getlength(bs
);
851 /* return 0 as number of sectors if no device present or error */
852 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
855 length
= bdrv_getlength(bs
);
859 length
= length
>> BDRV_SECTOR_BITS
;
860 *nb_sectors_ptr
= length
;
864 uint8_t boot_ind
; /* 0x80 - active */
865 uint8_t head
; /* starting head */
866 uint8_t sector
; /* starting sector */
867 uint8_t cyl
; /* starting cylinder */
868 uint8_t sys_ind
; /* What partition type */
869 uint8_t end_head
; /* end head */
870 uint8_t end_sector
; /* end sector */
871 uint8_t end_cyl
; /* end cylinder */
872 uint32_t start_sect
; /* starting sector counting from 0 */
873 uint32_t nr_sects
; /* nr of sectors in partition */
874 } __attribute__((packed
));
876 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
877 static int guess_disk_lchs(BlockDriverState
*bs
,
878 int *pcylinders
, int *pheads
, int *psectors
)
881 int ret
, i
, heads
, sectors
, cylinders
;
886 bdrv_get_geometry(bs
, &nb_sectors
);
888 ret
= bdrv_read(bs
, 0, buf
, 1);
891 /* test msdos magic */
892 if (buf
[510] != 0x55 || buf
[511] != 0xaa)
894 for(i
= 0; i
< 4; i
++) {
895 p
= ((struct partition
*)(buf
+ 0x1be)) + i
;
896 nr_sects
= le32_to_cpu(p
->nr_sects
);
897 if (nr_sects
&& p
->end_head
) {
898 /* We make the assumption that the partition terminates on
899 a cylinder boundary */
900 heads
= p
->end_head
+ 1;
901 sectors
= p
->end_sector
& 63;
904 cylinders
= nb_sectors
/ (heads
* sectors
);
905 if (cylinders
< 1 || cylinders
> 16383)
909 *pcylinders
= cylinders
;
911 printf("guessed geometry: LCHS=%d %d %d\n",
912 cylinders
, heads
, sectors
);
920 void bdrv_guess_geometry(BlockDriverState
*bs
, int *pcyls
, int *pheads
, int *psecs
)
922 int translation
, lba_detected
= 0;
923 int cylinders
, heads
, secs
;
926 /* if a geometry hint is available, use it */
927 bdrv_get_geometry(bs
, &nb_sectors
);
928 bdrv_get_geometry_hint(bs
, &cylinders
, &heads
, &secs
);
929 translation
= bdrv_get_translation_hint(bs
);
930 if (cylinders
!= 0) {
935 if (guess_disk_lchs(bs
, &cylinders
, &heads
, &secs
) == 0) {
937 /* if heads > 16, it means that a BIOS LBA
938 translation was active, so the default
939 hardware geometry is OK */
941 goto default_geometry
;
946 /* disable any translation to be in sync with
947 the logical geometry */
948 if (translation
== BIOS_ATA_TRANSLATION_AUTO
) {
949 bdrv_set_translation_hint(bs
,
950 BIOS_ATA_TRANSLATION_NONE
);
955 /* if no geometry, use a standard physical disk geometry */
956 cylinders
= nb_sectors
/ (16 * 63);
958 if (cylinders
> 16383)
960 else if (cylinders
< 2)
965 if ((lba_detected
== 1) && (translation
== BIOS_ATA_TRANSLATION_AUTO
)) {
966 if ((*pcyls
* *pheads
) <= 131072) {
967 bdrv_set_translation_hint(bs
,
968 BIOS_ATA_TRANSLATION_LARGE
);
970 bdrv_set_translation_hint(bs
,
971 BIOS_ATA_TRANSLATION_LBA
);
975 bdrv_set_geometry_hint(bs
, *pcyls
, *pheads
, *psecs
);
979 void bdrv_set_geometry_hint(BlockDriverState
*bs
,
980 int cyls
, int heads
, int secs
)
987 void bdrv_set_type_hint(BlockDriverState
*bs
, int type
)
990 bs
->removable
= ((type
== BDRV_TYPE_CDROM
||
991 type
== BDRV_TYPE_FLOPPY
));
994 void bdrv_set_translation_hint(BlockDriverState
*bs
, int translation
)
996 bs
->translation
= translation
;
999 void bdrv_get_geometry_hint(BlockDriverState
*bs
,
1000 int *pcyls
, int *pheads
, int *psecs
)
1003 *pheads
= bs
->heads
;
1007 int bdrv_get_type_hint(BlockDriverState
*bs
)
1012 int bdrv_get_translation_hint(BlockDriverState
*bs
)
1014 return bs
->translation
;
1017 int bdrv_is_removable(BlockDriverState
*bs
)
1019 return bs
->removable
;
1022 int bdrv_is_read_only(BlockDriverState
*bs
)
1024 return bs
->read_only
;
1027 int bdrv_set_read_only(BlockDriverState
*bs
, int read_only
)
1029 int ret
= bs
->read_only
;
1030 bs
->read_only
= read_only
;
1034 int bdrv_is_sg(BlockDriverState
*bs
)
1039 int bdrv_enable_write_cache(BlockDriverState
*bs
)
1041 return bs
->enable_write_cache
;
1044 /* XXX: no longer used */
1045 void bdrv_set_change_cb(BlockDriverState
*bs
,
1046 void (*change_cb
)(void *opaque
), void *opaque
)
1048 bs
->change_cb
= change_cb
;
1049 bs
->change_opaque
= opaque
;
1052 int bdrv_is_encrypted(BlockDriverState
*bs
)
1054 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
1056 return bs
->encrypted
;
1059 int bdrv_key_required(BlockDriverState
*bs
)
1061 BlockDriverState
*backing_hd
= bs
->backing_hd
;
1063 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
1065 return (bs
->encrypted
&& !bs
->valid_key
);
1068 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
1071 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
1072 ret
= bdrv_set_key(bs
->backing_hd
, key
);
1078 if (!bs
->encrypted
|| !bs
->drv
|| !bs
->drv
->bdrv_set_key
)
1080 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
1083 } else if (!bs
->valid_key
) {
1085 /* call the change callback now, we skipped it on open */
1086 bs
->media_changed
= 1;
1088 bs
->change_cb(bs
->change_opaque
);
1093 void bdrv_get_format(BlockDriverState
*bs
, char *buf
, int buf_size
)
1098 pstrcpy(buf
, buf_size
, bs
->drv
->format_name
);
1102 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
1107 for (drv
= first_drv
; drv
!= NULL
; drv
= drv
->next
) {
1108 it(opaque
, drv
->format_name
);
1112 BlockDriverState
*bdrv_find(const char *name
)
1114 BlockDriverState
*bs
;
1116 for (bs
= bdrv_first
; bs
!= NULL
; bs
= bs
->next
) {
1117 if (!strcmp(name
, bs
->device_name
))
1123 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
1125 BlockDriverState
*bs
;
1127 for (bs
= bdrv_first
; bs
!= NULL
; bs
= bs
->next
) {
1132 const char *bdrv_get_device_name(BlockDriverState
*bs
)
1134 return bs
->device_name
;
1137 void bdrv_flush(BlockDriverState
*bs
)
1141 if (bs
->drv
->bdrv_flush
)
1142 bs
->drv
->bdrv_flush(bs
);
1144 bdrv_flush(bs
->backing_hd
);
1147 void bdrv_flush_all(void)
1149 BlockDriverState
*bs
;
1151 for (bs
= bdrv_first
; bs
!= NULL
; bs
= bs
->next
)
1152 if (bs
->drv
&& !bdrv_is_read_only(bs
) &&
1153 (!bdrv_is_removable(bs
) || bdrv_is_inserted(bs
)))
1158 * Returns true iff the specified sector is present in the disk image. Drivers
1159 * not implementing the functionality are assumed to not support backing files,
1160 * hence all their sectors are reported as allocated.
1162 * 'pnum' is set to the number of sectors (including and immediately following
1163 * the specified sector) that are known to be in the same
1164 * allocated/unallocated state.
1166 * 'nb_sectors' is the max value 'pnum' should be set to.
1168 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
1172 if (!bs
->drv
->bdrv_is_allocated
) {
1173 if (sector_num
>= bs
->total_sectors
) {
1177 n
= bs
->total_sectors
- sector_num
;
1178 *pnum
= (n
< nb_sectors
) ? (n
) : (nb_sectors
);
1181 return bs
->drv
->bdrv_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
1184 static void bdrv_print_dict(QObject
*obj
, void *opaque
)
1187 Monitor
*mon
= opaque
;
1189 bs_dict
= qobject_to_qdict(obj
);
1191 monitor_printf(mon
, "%s: type=%s removable=%d",
1192 qdict_get_str(bs_dict
, "device"),
1193 qdict_get_str(bs_dict
, "type"),
1194 qdict_get_bool(bs_dict
, "removable"));
1196 if (qdict_get_bool(bs_dict
, "removable")) {
1197 monitor_printf(mon
, " locked=%d", qdict_get_bool(bs_dict
, "locked"));
1200 if (qdict_haskey(bs_dict
, "inserted")) {
1201 QDict
*qdict
= qobject_to_qdict(qdict_get(bs_dict
, "inserted"));
1203 monitor_printf(mon
, " file=");
1204 monitor_print_filename(mon
, qdict_get_str(qdict
, "file"));
1205 if (qdict_haskey(qdict
, "backing_file")) {
1206 monitor_printf(mon
, " backing_file=");
1207 monitor_print_filename(mon
, qdict_get_str(qdict
, "backing_file"));
1209 monitor_printf(mon
, " ro=%d drv=%s encrypted=%d",
1210 qdict_get_bool(qdict
, "ro"),
1211 qdict_get_str(qdict
, "drv"),
1212 qdict_get_bool(qdict
, "encrypted"));
1214 monitor_printf(mon
, " [not inserted]");
1217 monitor_printf(mon
, "\n");
1220 void bdrv_info_print(Monitor
*mon
, const QObject
*data
)
1222 qlist_iter(qobject_to_qlist(data
), bdrv_print_dict
, mon
);
1226 * bdrv_info(): Block devices information
1228 * Each block device information is stored in a QDict and the
1229 * returned QObject is a QList of all devices.
1231 * The QDict contains the following:
1233 * - "device": device name
1234 * - "type": device type
1235 * - "removable": true if the device is removable, false otherwise
1236 * - "locked": true if the device is locked, false otherwise
1237 * - "inserted": only present if the device is inserted, it is a QDict
1238 * containing the following:
1239 * - "file": device file name
1240 * - "ro": true if read-only, false otherwise
1241 * - "drv": driver format name
1242 * - "backing_file": backing file name if one is used
1243 * - "encrypted": true if encrypted, false otherwise
1247 * [ { "device": "ide0-hd0", "type": "hd", "removable": false, "locked": false,
1248 * "inserted": { "file": "/tmp/foobar", "ro": false, "drv": "qcow2" } },
1249 * { "device": "floppy0", "type": "floppy", "removable": true,
1250 * "locked": false } ]
1252 void bdrv_info(Monitor
*mon
, QObject
**ret_data
)
1255 BlockDriverState
*bs
;
1257 bs_list
= qlist_new();
1259 for (bs
= bdrv_first
; bs
!= NULL
; bs
= bs
->next
) {
1261 const char *type
= "unknown";
1267 case BDRV_TYPE_CDROM
:
1270 case BDRV_TYPE_FLOPPY
:
1275 bs_obj
= qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1276 "'removable': %i, 'locked': %i }",
1277 bs
->device_name
, type
, bs
->removable
,
1279 assert(bs_obj
!= NULL
);
1283 QDict
*bs_dict
= qobject_to_qdict(bs_obj
);
1285 obj
= qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1286 "'encrypted': %i }",
1287 bs
->filename
, bs
->read_only
,
1288 bs
->drv
->format_name
,
1289 bdrv_is_encrypted(bs
));
1290 assert(obj
!= NULL
);
1291 if (bs
->backing_file
[0] != '\0') {
1292 QDict
*qdict
= qobject_to_qdict(obj
);
1293 qdict_put(qdict
, "backing_file",
1294 qstring_from_str(bs
->backing_file
));
1297 qdict_put_obj(bs_dict
, "inserted", obj
);
1299 qlist_append_obj(bs_list
, bs_obj
);
1302 *ret_data
= QOBJECT(bs_list
);
1305 static void bdrv_stats_iter(QObject
*data
, void *opaque
)
1308 Monitor
*mon
= opaque
;
1310 qdict
= qobject_to_qdict(data
);
1311 monitor_printf(mon
, "%s:", qdict_get_str(qdict
, "device"));
1313 qdict
= qobject_to_qdict(qdict_get(qdict
, "stats"));
1314 monitor_printf(mon
, " rd_bytes=%" PRId64
1315 " wr_bytes=%" PRId64
1316 " rd_operations=%" PRId64
1317 " wr_operations=%" PRId64
1319 qdict_get_int(qdict
, "rd_bytes"),
1320 qdict_get_int(qdict
, "wr_bytes"),
1321 qdict_get_int(qdict
, "rd_operations"),
1322 qdict_get_int(qdict
, "wr_operations"));
1325 void bdrv_stats_print(Monitor
*mon
, const QObject
*data
)
1327 qlist_iter(qobject_to_qlist(data
), bdrv_stats_iter
, mon
);
1331 * bdrv_info_stats(): show block device statistics
1333 * Each device statistic information is stored in a QDict and
1334 * the returned QObject is a QList of all devices.
1336 * The QDict contains the following:
1338 * - "device": device name
1339 * - "stats": A QDict with the statistics information, it contains:
1340 * - "rd_bytes": bytes read
1341 * - "wr_bytes": bytes written
1342 * - "rd_operations": read operations
1343 * - "wr_operations": write operations
1347 * [ { "device": "ide0-hd0",
1348 * "stats": { "rd_bytes": 512,
1350 * "rd_operations": 1,
1351 * "wr_operations": 0 } },
1352 * { "device": "ide1-cd0",
1353 * "stats": { "rd_bytes": 0,
1355 * "rd_operations": 0,
1356 * "wr_operations": 0 } } ]
1358 void bdrv_info_stats(Monitor
*mon
, QObject
**ret_data
)
1362 BlockDriverState
*bs
;
1364 devices
= qlist_new();
1366 for (bs
= bdrv_first
; bs
!= NULL
; bs
= bs
->next
) {
1367 obj
= qobject_from_jsonf("{ 'device': %s, 'stats': {"
1368 "'rd_bytes': %" PRId64
","
1369 "'wr_bytes': %" PRId64
","
1370 "'rd_operations': %" PRId64
","
1371 "'wr_operations': %" PRId64
1374 bs
->rd_bytes
, bs
->wr_bytes
,
1375 bs
->rd_ops
, bs
->wr_ops
);
1376 assert(obj
!= NULL
);
1377 qlist_append_obj(devices
, obj
);
1380 *ret_data
= QOBJECT(devices
);
1383 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
1385 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
1386 return bs
->backing_file
;
1387 else if (bs
->encrypted
)
1388 return bs
->filename
;
1393 void bdrv_get_backing_filename(BlockDriverState
*bs
,
1394 char *filename
, int filename_size
)
1396 if (!bs
->backing_hd
) {
1397 pstrcpy(filename
, filename_size
, "");
1399 pstrcpy(filename
, filename_size
, bs
->backing_file
);
1403 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1404 const uint8_t *buf
, int nb_sectors
)
1406 BlockDriver
*drv
= bs
->drv
;
1409 if (!drv
->bdrv_write_compressed
)
1411 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1414 if (bs
->dirty_bitmap
) {
1415 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1418 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1421 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1423 BlockDriver
*drv
= bs
->drv
;
1426 if (!drv
->bdrv_get_info
)
1428 memset(bdi
, 0, sizeof(*bdi
));
1429 return drv
->bdrv_get_info(bs
, bdi
);
1432 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1433 int64_t pos
, int size
)
1435 BlockDriver
*drv
= bs
->drv
;
1438 if (!drv
->bdrv_save_vmstate
)
1440 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
1443 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1444 int64_t pos
, int size
)
1446 BlockDriver
*drv
= bs
->drv
;
1449 if (!drv
->bdrv_load_vmstate
)
1451 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1454 /**************************************************************/
1455 /* handling of snapshots */
1457 int bdrv_snapshot_create(BlockDriverState
*bs
,
1458 QEMUSnapshotInfo
*sn_info
)
1460 BlockDriver
*drv
= bs
->drv
;
1463 if (!drv
->bdrv_snapshot_create
)
1465 return drv
->bdrv_snapshot_create(bs
, sn_info
);
1468 int bdrv_snapshot_goto(BlockDriverState
*bs
,
1469 const char *snapshot_id
)
1471 BlockDriver
*drv
= bs
->drv
;
1474 if (!drv
->bdrv_snapshot_goto
)
1476 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
1479 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
1481 BlockDriver
*drv
= bs
->drv
;
1484 if (!drv
->bdrv_snapshot_delete
)
1486 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
1489 int bdrv_snapshot_list(BlockDriverState
*bs
,
1490 QEMUSnapshotInfo
**psn_info
)
1492 BlockDriver
*drv
= bs
->drv
;
1495 if (!drv
->bdrv_snapshot_list
)
1497 return drv
->bdrv_snapshot_list(bs
, psn_info
);
1500 #define NB_SUFFIXES 4
1502 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
1504 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
1509 snprintf(buf
, buf_size
, "%" PRId64
, size
);
1512 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
1513 if (size
< (10 * base
)) {
1514 snprintf(buf
, buf_size
, "%0.1f%c",
1515 (double)size
/ base
,
1518 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
1519 snprintf(buf
, buf_size
, "%" PRId64
"%c",
1520 ((size
+ (base
>> 1)) / base
),
1530 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
1532 char buf1
[128], date_buf
[128], clock_buf
[128];
1542 snprintf(buf
, buf_size
,
1543 "%-10s%-20s%7s%20s%15s",
1544 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1548 ptm
= localtime(&ti
);
1549 strftime(date_buf
, sizeof(date_buf
),
1550 "%Y-%m-%d %H:%M:%S", ptm
);
1552 localtime_r(&ti
, &tm
);
1553 strftime(date_buf
, sizeof(date_buf
),
1554 "%Y-%m-%d %H:%M:%S", &tm
);
1556 secs
= sn
->vm_clock_nsec
/ 1000000000;
1557 snprintf(clock_buf
, sizeof(clock_buf
),
1558 "%02d:%02d:%02d.%03d",
1560 (int)((secs
/ 60) % 60),
1562 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
1563 snprintf(buf
, buf_size
,
1564 "%-10s%-20s%7s%20s%15s",
1565 sn
->id_str
, sn
->name
,
1566 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
1574 /**************************************************************/
1577 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1578 QEMUIOVector
*qiov
, int nb_sectors
,
1579 BlockDriverCompletionFunc
*cb
, void *opaque
)
1581 BlockDriver
*drv
= bs
->drv
;
1582 BlockDriverAIOCB
*ret
;
1586 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1589 ret
= drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
1593 /* Update stats even though technically transfer has not happened. */
1594 bs
->rd_bytes
+= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
1601 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1602 QEMUIOVector
*qiov
, int nb_sectors
,
1603 BlockDriverCompletionFunc
*cb
, void *opaque
)
1605 BlockDriver
*drv
= bs
->drv
;
1606 BlockDriverAIOCB
*ret
;
1612 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1615 if (bs
->dirty_bitmap
) {
1616 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1619 ret
= drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
1623 /* Update stats even though technically transfer has not happened. */
1624 bs
->wr_bytes
+= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
1632 typedef struct MultiwriteCB
{
1637 BlockDriverCompletionFunc
*cb
;
1639 QEMUIOVector
*free_qiov
;
1644 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
1648 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
1649 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
1650 if (mcb
->callbacks
[i
].free_qiov
) {
1651 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
1653 qemu_free(mcb
->callbacks
[i
].free_qiov
);
1654 qemu_vfree(mcb
->callbacks
[i
].free_buf
);
1658 static void multiwrite_cb(void *opaque
, int ret
)
1660 MultiwriteCB
*mcb
= opaque
;
1662 if (ret
< 0 && !mcb
->error
) {
1664 multiwrite_user_cb(mcb
);
1667 mcb
->num_requests
--;
1668 if (mcb
->num_requests
== 0) {
1669 if (mcb
->error
== 0) {
1670 multiwrite_user_cb(mcb
);
1676 static int multiwrite_req_compare(const void *a
, const void *b
)
1678 const BlockRequest
*req1
= a
, *req2
= b
;
1681 * Note that we can't simply subtract req2->sector from req1->sector
1682 * here as that could overflow the return value.
1684 if (req1
->sector
> req2
->sector
) {
1686 } else if (req1
->sector
< req2
->sector
) {
1694 * Takes a bunch of requests and tries to merge them. Returns the number of
1695 * requests that remain after merging.
1697 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
1698 int num_reqs
, MultiwriteCB
*mcb
)
1702 // Sort requests by start sector
1703 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
1705 // Check if adjacent requests touch the same clusters. If so, combine them,
1706 // filling up gaps with zero sectors.
1708 for (i
= 1; i
< num_reqs
; i
++) {
1710 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
1712 // This handles the cases that are valid for all block drivers, namely
1713 // exactly sequential writes and overlapping writes.
1714 if (reqs
[i
].sector
<= oldreq_last
) {
1718 // The block driver may decide that it makes sense to combine requests
1719 // even if there is a gap of some sectors between them. In this case,
1720 // the gap is filled with zeros (therefore only applicable for yet
1721 // unused space in format like qcow2).
1722 if (!merge
&& bs
->drv
->bdrv_merge_requests
) {
1723 merge
= bs
->drv
->bdrv_merge_requests(bs
, &reqs
[outidx
], &reqs
[i
]);
1726 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
1732 QEMUIOVector
*qiov
= qemu_mallocz(sizeof(*qiov
));
1733 qemu_iovec_init(qiov
,
1734 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
1736 // Add the first request to the merged one. If the requests are
1737 // overlapping, drop the last sectors of the first request.
1738 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
1739 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, size
);
1741 // We might need to add some zeros between the two requests
1742 if (reqs
[i
].sector
> oldreq_last
) {
1743 size_t zero_bytes
= (reqs
[i
].sector
- oldreq_last
) << 9;
1744 uint8_t *buf
= qemu_blockalign(bs
, zero_bytes
);
1745 memset(buf
, 0, zero_bytes
);
1746 qemu_iovec_add(qiov
, buf
, zero_bytes
);
1747 mcb
->callbacks
[i
].free_buf
= buf
;
1750 // Add the second request
1751 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, reqs
[i
].qiov
->size
);
1753 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
1754 reqs
[outidx
].qiov
= qiov
;
1756 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
1759 reqs
[outidx
].sector
= reqs
[i
].sector
;
1760 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
1761 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
1769 * Submit multiple AIO write requests at once.
1771 * On success, the function returns 0 and all requests in the reqs array have
1772 * been submitted. In error case this function returns -1, and any of the
1773 * requests may or may not be submitted yet. In particular, this means that the
1774 * callback will be called for some of the requests, for others it won't. The
1775 * caller must check the error field of the BlockRequest to wait for the right
1776 * callbacks (if error != 0, no callback will be called).
1778 * The implementation may modify the contents of the reqs array, e.g. to merge
1779 * requests. However, the fields opaque and error are left unmodified as they
1780 * are used to signal failure for a single request to the caller.
1782 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
1784 BlockDriverAIOCB
*acb
;
1788 if (num_reqs
== 0) {
1792 // Create MultiwriteCB structure
1793 mcb
= qemu_mallocz(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
1794 mcb
->num_requests
= 0;
1795 mcb
->num_callbacks
= num_reqs
;
1797 for (i
= 0; i
< num_reqs
; i
++) {
1798 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
1799 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
1802 // Check for mergable requests
1803 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
1805 // Run the aio requests
1806 for (i
= 0; i
< num_reqs
; i
++) {
1807 acb
= bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
1808 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
1811 // We can only fail the whole thing if no request has been
1812 // submitted yet. Otherwise we'll wait for the submitted AIOs to
1813 // complete and report the error in the callback.
1814 if (mcb
->num_requests
== 0) {
1815 reqs
[i
].error
= -EIO
;
1818 mcb
->num_requests
++;
1819 multiwrite_cb(mcb
, -EIO
);
1823 mcb
->num_requests
++;
1834 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
1835 BlockDriverCompletionFunc
*cb
, void *opaque
)
1837 BlockDriver
*drv
= bs
->drv
;
1843 * Note that unlike bdrv_flush the driver is reponsible for flushing a
1844 * backing image if it exists.
1846 return drv
->bdrv_aio_flush(bs
, cb
, opaque
);
1849 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
1851 acb
->pool
->cancel(acb
);
1855 /**************************************************************/
1856 /* async block device emulation */
1858 typedef struct BlockDriverAIOCBSync
{
1859 BlockDriverAIOCB common
;
1862 /* vector translation state */
1866 } BlockDriverAIOCBSync
;
1868 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
1870 BlockDriverAIOCBSync
*acb
= (BlockDriverAIOCBSync
*)blockacb
;
1871 qemu_bh_delete(acb
->bh
);
1873 qemu_aio_release(acb
);
1876 static AIOPool bdrv_em_aio_pool
= {
1877 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
1878 .cancel
= bdrv_aio_cancel_em
,
1881 static void bdrv_aio_bh_cb(void *opaque
)
1883 BlockDriverAIOCBSync
*acb
= opaque
;
1886 qemu_iovec_from_buffer(acb
->qiov
, acb
->bounce
, acb
->qiov
->size
);
1887 qemu_vfree(acb
->bounce
);
1888 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1889 qemu_bh_delete(acb
->bh
);
1891 qemu_aio_release(acb
);
1894 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
1898 BlockDriverCompletionFunc
*cb
,
1903 BlockDriverAIOCBSync
*acb
;
1905 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
1906 acb
->is_write
= is_write
;
1908 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
1911 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
1914 qemu_iovec_to_buffer(acb
->qiov
, acb
->bounce
);
1915 acb
->ret
= bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
1917 acb
->ret
= bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
1920 qemu_bh_schedule(acb
->bh
);
1922 return &acb
->common
;
1925 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
1926 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1927 BlockDriverCompletionFunc
*cb
, void *opaque
)
1929 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
1932 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
1933 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1934 BlockDriverCompletionFunc
*cb
, void *opaque
)
1936 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
1939 static BlockDriverAIOCB
*bdrv_aio_flush_em(BlockDriverState
*bs
,
1940 BlockDriverCompletionFunc
*cb
, void *opaque
)
1942 BlockDriverAIOCBSync
*acb
;
1944 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
1945 acb
->is_write
= 1; /* don't bounce in the completion hadler */
1951 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
1954 qemu_bh_schedule(acb
->bh
);
1955 return &acb
->common
;
1958 /**************************************************************/
1959 /* sync block device emulation */
1961 static void bdrv_rw_em_cb(void *opaque
, int ret
)
1963 *(int *)opaque
= ret
;
1966 #define NOT_DONE 0x7fffffff
1968 static int bdrv_read_em(BlockDriverState
*bs
, int64_t sector_num
,
1969 uint8_t *buf
, int nb_sectors
)
1972 BlockDriverAIOCB
*acb
;
1976 async_context_push();
1978 async_ret
= NOT_DONE
;
1979 iov
.iov_base
= (void *)buf
;
1980 iov
.iov_len
= nb_sectors
* 512;
1981 qemu_iovec_init_external(&qiov
, &iov
, 1);
1982 acb
= bdrv_aio_readv(bs
, sector_num
, &qiov
, nb_sectors
,
1983 bdrv_rw_em_cb
, &async_ret
);
1989 while (async_ret
== NOT_DONE
) {
1995 async_context_pop();
1999 static int bdrv_write_em(BlockDriverState
*bs
, int64_t sector_num
,
2000 const uint8_t *buf
, int nb_sectors
)
2003 BlockDriverAIOCB
*acb
;
2007 async_context_push();
2009 async_ret
= NOT_DONE
;
2010 iov
.iov_base
= (void *)buf
;
2011 iov
.iov_len
= nb_sectors
* 512;
2012 qemu_iovec_init_external(&qiov
, &iov
, 1);
2013 acb
= bdrv_aio_writev(bs
, sector_num
, &qiov
, nb_sectors
,
2014 bdrv_rw_em_cb
, &async_ret
);
2019 while (async_ret
== NOT_DONE
) {
2024 async_context_pop();
2028 void bdrv_init(void)
2030 module_call_init(MODULE_INIT_BLOCK
);
2033 void bdrv_init_with_whitelist(void)
2035 use_bdrv_whitelist
= 1;
2039 void *qemu_aio_get(AIOPool
*pool
, BlockDriverState
*bs
,
2040 BlockDriverCompletionFunc
*cb
, void *opaque
)
2042 BlockDriverAIOCB
*acb
;
2044 if (pool
->free_aiocb
) {
2045 acb
= pool
->free_aiocb
;
2046 pool
->free_aiocb
= acb
->next
;
2048 acb
= qemu_mallocz(pool
->aiocb_size
);
2053 acb
->opaque
= opaque
;
2057 void qemu_aio_release(void *p
)
2059 BlockDriverAIOCB
*acb
= (BlockDriverAIOCB
*)p
;
2060 AIOPool
*pool
= acb
->pool
;
2061 acb
->next
= pool
->free_aiocb
;
2062 pool
->free_aiocb
= acb
;
2065 /**************************************************************/
2066 /* removable device support */
2069 * Return TRUE if the media is present
2071 int bdrv_is_inserted(BlockDriverState
*bs
)
2073 BlockDriver
*drv
= bs
->drv
;
2077 if (!drv
->bdrv_is_inserted
)
2079 ret
= drv
->bdrv_is_inserted(bs
);
2084 * Return TRUE if the media changed since the last call to this
2085 * function. It is currently only used for floppy disks
2087 int bdrv_media_changed(BlockDriverState
*bs
)
2089 BlockDriver
*drv
= bs
->drv
;
2092 if (!drv
|| !drv
->bdrv_media_changed
)
2095 ret
= drv
->bdrv_media_changed(bs
);
2096 if (ret
== -ENOTSUP
)
2097 ret
= bs
->media_changed
;
2098 bs
->media_changed
= 0;
2103 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2105 int bdrv_eject(BlockDriverState
*bs
, int eject_flag
)
2107 BlockDriver
*drv
= bs
->drv
;
2114 if (!drv
|| !drv
->bdrv_eject
) {
2117 ret
= drv
->bdrv_eject(bs
, eject_flag
);
2119 if (ret
== -ENOTSUP
) {
2128 int bdrv_is_locked(BlockDriverState
*bs
)
2134 * Lock or unlock the media (if it is locked, the user won't be able
2135 * to eject it manually).
2137 void bdrv_set_locked(BlockDriverState
*bs
, int locked
)
2139 BlockDriver
*drv
= bs
->drv
;
2141 bs
->locked
= locked
;
2142 if (drv
&& drv
->bdrv_set_locked
) {
2143 drv
->bdrv_set_locked(bs
, locked
);
2147 /* needed for generic scsi interface */
2149 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2151 BlockDriver
*drv
= bs
->drv
;
2153 if (drv
&& drv
->bdrv_ioctl
)
2154 return drv
->bdrv_ioctl(bs
, req
, buf
);
2158 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2159 unsigned long int req
, void *buf
,
2160 BlockDriverCompletionFunc
*cb
, void *opaque
)
2162 BlockDriver
*drv
= bs
->drv
;
2164 if (drv
&& drv
->bdrv_aio_ioctl
)
2165 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
2171 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2173 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
2176 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int enable
)
2178 int64_t bitmap_size
;
2181 if (!bs
->dirty_bitmap
) {
2182 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
) +
2183 BDRV_SECTORS_PER_DIRTY_CHUNK
* 8 - 1;
2184 bitmap_size
/= BDRV_SECTORS_PER_DIRTY_CHUNK
* 8;
2186 bs
->dirty_bitmap
= qemu_mallocz(bitmap_size
);
2189 if (bs
->dirty_bitmap
) {
2190 qemu_free(bs
->dirty_bitmap
);
2191 bs
->dirty_bitmap
= NULL
;
2196 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
2198 int64_t chunk
= sector
/ (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK
;
2200 if (bs
->dirty_bitmap
&&
2201 (sector
<< BDRV_SECTOR_BITS
) < bdrv_getlength(bs
)) {
2202 return bs
->dirty_bitmap
[chunk
/ (sizeof(unsigned long) * 8)] &
2203 (1 << (chunk
% (sizeof(unsigned long) * 8)));
2209 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
2212 set_dirty_bitmap(bs
, cur_sector
, nr_sectors
, 0);