2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
28 #include "block_int.h"
30 #include "qemu-objects.h"
31 #include "qemu-coroutine.h"
34 #include <sys/types.h>
36 #include <sys/ioctl.h>
37 #include <sys/queue.h>
47 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
);
48 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
49 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
50 BlockDriverCompletionFunc
*cb
, void *opaque
);
51 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
52 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
53 BlockDriverCompletionFunc
*cb
, void *opaque
);
54 static BlockDriverAIOCB
*bdrv_aio_flush_em(BlockDriverState
*bs
,
55 BlockDriverCompletionFunc
*cb
, void *opaque
);
56 static BlockDriverAIOCB
*bdrv_aio_noop_em(BlockDriverState
*bs
,
57 BlockDriverCompletionFunc
*cb
, void *opaque
);
58 static int bdrv_read_em(BlockDriverState
*bs
, int64_t sector_num
,
59 uint8_t *buf
, int nb_sectors
);
60 static int bdrv_write_em(BlockDriverState
*bs
, int64_t sector_num
,
61 const uint8_t *buf
, int nb_sectors
);
62 static BlockDriverAIOCB
*bdrv_co_aio_readv_em(BlockDriverState
*bs
,
63 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
64 BlockDriverCompletionFunc
*cb
, void *opaque
);
65 static BlockDriverAIOCB
*bdrv_co_aio_writev_em(BlockDriverState
*bs
,
66 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
67 BlockDriverCompletionFunc
*cb
, void *opaque
);
68 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
69 int64_t sector_num
, int nb_sectors
,
71 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
72 int64_t sector_num
, int nb_sectors
,
74 static int coroutine_fn
bdrv_co_flush_em(BlockDriverState
*bs
);
76 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
77 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
79 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
80 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
82 /* The device to use for VM snapshots */
83 static BlockDriverState
*bs_snapshots
;
85 /* If non-zero, use only whitelisted block drivers */
86 static int use_bdrv_whitelist
;
89 static int is_windows_drive_prefix(const char *filename
)
91 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
92 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
96 int is_windows_drive(const char *filename
)
98 if (is_windows_drive_prefix(filename
) &&
101 if (strstart(filename
, "\\\\.\\", NULL
) ||
102 strstart(filename
, "//./", NULL
))
108 /* check if the path starts with "<protocol>:" */
109 static int path_has_protocol(const char *path
)
112 if (is_windows_drive(path
) ||
113 is_windows_drive_prefix(path
)) {
118 return strchr(path
, ':') != NULL
;
121 int path_is_absolute(const char *path
)
125 /* specific case for names like: "\\.\d:" */
126 if (*path
== '/' || *path
== '\\')
129 p
= strchr(path
, ':');
135 return (*p
== '/' || *p
== '\\');
141 /* if filename is absolute, just copy it to dest. Otherwise, build a
142 path to it by considering it is relative to base_path. URL are
144 void path_combine(char *dest
, int dest_size
,
145 const char *base_path
,
146 const char *filename
)
153 if (path_is_absolute(filename
)) {
154 pstrcpy(dest
, dest_size
, filename
);
156 p
= strchr(base_path
, ':');
161 p1
= strrchr(base_path
, '/');
165 p2
= strrchr(base_path
, '\\');
177 if (len
> dest_size
- 1)
179 memcpy(dest
, base_path
, len
);
181 pstrcat(dest
, dest_size
, filename
);
185 void bdrv_register(BlockDriver
*bdrv
)
187 if (bdrv
->bdrv_co_readv
) {
188 /* Emulate AIO by coroutines, and sync by AIO */
189 bdrv
->bdrv_aio_readv
= bdrv_co_aio_readv_em
;
190 bdrv
->bdrv_aio_writev
= bdrv_co_aio_writev_em
;
191 bdrv
->bdrv_read
= bdrv_read_em
;
192 bdrv
->bdrv_write
= bdrv_write_em
;
194 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
195 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
197 if (!bdrv
->bdrv_aio_readv
) {
198 /* add AIO emulation layer */
199 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
200 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
201 } else if (!bdrv
->bdrv_read
) {
202 /* add synchronous IO emulation layer */
203 bdrv
->bdrv_read
= bdrv_read_em
;
204 bdrv
->bdrv_write
= bdrv_write_em
;
208 if (!bdrv
->bdrv_aio_flush
)
209 bdrv
->bdrv_aio_flush
= bdrv_aio_flush_em
;
211 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
214 /* create a new block device (by default it is empty) */
215 BlockDriverState
*bdrv_new(const char *device_name
)
217 BlockDriverState
*bs
;
219 bs
= g_malloc0(sizeof(BlockDriverState
));
220 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
221 if (device_name
[0] != '\0') {
222 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
227 BlockDriver
*bdrv_find_format(const char *format_name
)
230 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
231 if (!strcmp(drv1
->format_name
, format_name
)) {
238 static int bdrv_is_whitelisted(BlockDriver
*drv
)
240 static const char *whitelist
[] = {
241 CONFIG_BDRV_WHITELIST
246 return 1; /* no whitelist, anything goes */
248 for (p
= whitelist
; *p
; p
++) {
249 if (!strcmp(drv
->format_name
, *p
)) {
256 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
258 BlockDriver
*drv
= bdrv_find_format(format_name
);
259 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
262 int bdrv_create(BlockDriver
*drv
, const char* filename
,
263 QEMUOptionParameter
*options
)
265 if (!drv
->bdrv_create
)
268 return drv
->bdrv_create(filename
, options
);
271 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
275 drv
= bdrv_find_protocol(filename
);
280 return bdrv_create(drv
, filename
, options
);
284 void get_tmp_filename(char *filename
, int size
)
286 char temp_dir
[MAX_PATH
];
288 GetTempPath(MAX_PATH
, temp_dir
);
289 GetTempFileName(temp_dir
, "qem", 0, filename
);
292 void get_tmp_filename(char *filename
, int size
)
296 /* XXX: race condition possible */
297 tmpdir
= getenv("TMPDIR");
300 snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
);
301 fd
= mkstemp(filename
);
307 * Detect host devices. By convention, /dev/cdrom[N] is always
308 * recognized as a host CDROM.
310 static BlockDriver
*find_hdev_driver(const char *filename
)
312 int score_max
= 0, score
;
313 BlockDriver
*drv
= NULL
, *d
;
315 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
316 if (d
->bdrv_probe_device
) {
317 score
= d
->bdrv_probe_device(filename
);
318 if (score
> score_max
) {
328 BlockDriver
*bdrv_find_protocol(const char *filename
)
335 /* TODO Drivers without bdrv_file_open must be specified explicitly */
338 * XXX(hch): we really should not let host device detection
339 * override an explicit protocol specification, but moving this
340 * later breaks access to device names with colons in them.
341 * Thanks to the brain-dead persistent naming schemes on udev-
342 * based Linux systems those actually are quite common.
344 drv1
= find_hdev_driver(filename
);
349 if (!path_has_protocol(filename
)) {
350 return bdrv_find_format("file");
352 p
= strchr(filename
, ':');
355 if (len
> sizeof(protocol
) - 1)
356 len
= sizeof(protocol
) - 1;
357 memcpy(protocol
, filename
, len
);
358 protocol
[len
] = '\0';
359 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
360 if (drv1
->protocol_name
&&
361 !strcmp(drv1
->protocol_name
, protocol
)) {
368 static int find_image_format(const char *filename
, BlockDriver
**pdrv
)
370 int ret
, score
, score_max
;
371 BlockDriver
*drv1
, *drv
;
373 BlockDriverState
*bs
;
375 ret
= bdrv_file_open(&bs
, filename
, 0);
381 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
382 if (bs
->sg
|| !bdrv_is_inserted(bs
)) {
384 drv
= bdrv_find_format("raw");
392 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
401 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
402 if (drv1
->bdrv_probe
) {
403 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
404 if (score
> score_max
) {
418 * Set the current 'total_sectors' value
420 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
422 BlockDriver
*drv
= bs
->drv
;
424 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
428 /* query actual device if possible, otherwise just trust the hint */
429 if (drv
->bdrv_getlength
) {
430 int64_t length
= drv
->bdrv_getlength(bs
);
434 hint
= length
>> BDRV_SECTOR_BITS
;
437 bs
->total_sectors
= hint
;
442 * Set open flags for a given cache mode
444 * Return 0 on success, -1 if the cache mode was invalid.
446 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
448 *flags
&= ~BDRV_O_CACHE_MASK
;
450 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
451 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
452 } else if (!strcmp(mode
, "directsync")) {
453 *flags
|= BDRV_O_NOCACHE
;
454 } else if (!strcmp(mode
, "writeback")) {
455 *flags
|= BDRV_O_CACHE_WB
;
456 } else if (!strcmp(mode
, "unsafe")) {
457 *flags
|= BDRV_O_CACHE_WB
;
458 *flags
|= BDRV_O_NO_FLUSH
;
459 } else if (!strcmp(mode
, "writethrough")) {
460 /* this is the default */
469 * Common part for opening disk images and files
471 static int bdrv_open_common(BlockDriverState
*bs
, const char *filename
,
472 int flags
, BlockDriver
*drv
)
479 bs
->total_sectors
= 0;
482 bs
->open_flags
= flags
;
483 /* buffer_alignment defaulted to 512, drivers can change this value */
484 bs
->buffer_alignment
= 512;
486 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
488 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
493 bs
->opaque
= g_malloc0(drv
->instance_size
);
495 if (flags
& BDRV_O_CACHE_WB
)
496 bs
->enable_write_cache
= 1;
499 * Clear flags that are internal to the block layer before opening the
502 open_flags
= flags
& ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
505 * Snapshots should be writable.
507 if (bs
->is_temporary
) {
508 open_flags
|= BDRV_O_RDWR
;
511 /* Open the image, either directly or using a protocol */
512 if (drv
->bdrv_file_open
) {
513 ret
= drv
->bdrv_file_open(bs
, filename
, open_flags
);
515 ret
= bdrv_file_open(&bs
->file
, filename
, open_flags
);
517 ret
= drv
->bdrv_open(bs
, open_flags
);
525 bs
->keep_read_only
= bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
527 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
533 if (bs
->is_temporary
) {
541 bdrv_delete(bs
->file
);
551 * Opens a file using a protocol (file, host_device, nbd, ...)
553 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
, int flags
)
555 BlockDriverState
*bs
;
559 drv
= bdrv_find_protocol(filename
);
565 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
576 * Opens a disk image (raw, qcow2, vmdk, ...)
578 int bdrv_open(BlockDriverState
*bs
, const char *filename
, int flags
,
583 if (flags
& BDRV_O_SNAPSHOT
) {
584 BlockDriverState
*bs1
;
587 BlockDriver
*bdrv_qcow2
;
588 QEMUOptionParameter
*options
;
589 char tmp_filename
[PATH_MAX
];
590 char backing_filename
[PATH_MAX
];
592 /* if snapshot, we create a temporary backing file and open it
593 instead of opening 'filename' directly */
595 /* if there is a backing file, use it */
597 ret
= bdrv_open(bs1
, filename
, 0, drv
);
602 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
604 if (bs1
->drv
&& bs1
->drv
->protocol_name
)
609 get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
611 /* Real path is meaningless for protocols */
613 snprintf(backing_filename
, sizeof(backing_filename
),
615 else if (!realpath(filename
, backing_filename
))
618 bdrv_qcow2
= bdrv_find_format("qcow2");
619 options
= parse_option_parameters("", bdrv_qcow2
->create_options
, NULL
);
621 set_option_parameter_int(options
, BLOCK_OPT_SIZE
, total_size
);
622 set_option_parameter(options
, BLOCK_OPT_BACKING_FILE
, backing_filename
);
624 set_option_parameter(options
, BLOCK_OPT_BACKING_FMT
,
628 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, options
);
629 free_option_parameters(options
);
634 filename
= tmp_filename
;
636 bs
->is_temporary
= 1;
639 /* Find the right image format driver */
641 ret
= find_image_format(filename
, &drv
);
645 goto unlink_and_fail
;
649 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
651 goto unlink_and_fail
;
654 /* If there is a backing file, use it */
655 if ((flags
& BDRV_O_NO_BACKING
) == 0 && bs
->backing_file
[0] != '\0') {
656 char backing_filename
[PATH_MAX
];
658 BlockDriver
*back_drv
= NULL
;
660 bs
->backing_hd
= bdrv_new("");
662 if (path_has_protocol(bs
->backing_file
)) {
663 pstrcpy(backing_filename
, sizeof(backing_filename
),
666 path_combine(backing_filename
, sizeof(backing_filename
),
667 filename
, bs
->backing_file
);
670 if (bs
->backing_format
[0] != '\0') {
671 back_drv
= bdrv_find_format(bs
->backing_format
);
674 /* backing files always opened read-only */
676 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
678 ret
= bdrv_open(bs
->backing_hd
, backing_filename
, back_flags
, back_drv
);
683 if (bs
->is_temporary
) {
684 bs
->backing_hd
->keep_read_only
= !(flags
& BDRV_O_RDWR
);
686 /* base image inherits from "parent" */
687 bs
->backing_hd
->keep_read_only
= bs
->keep_read_only
;
691 if (!bdrv_key_required(bs
)) {
692 bs
->media_changed
= 1;
693 bdrv_dev_change_media_cb(bs
);
699 if (bs
->is_temporary
) {
705 void bdrv_close(BlockDriverState
*bs
)
708 if (bs
== bs_snapshots
) {
711 if (bs
->backing_hd
) {
712 bdrv_delete(bs
->backing_hd
);
713 bs
->backing_hd
= NULL
;
715 bs
->drv
->bdrv_close(bs
);
718 if (bs
->is_temporary
) {
719 unlink(bs
->filename
);
725 if (bs
->file
!= NULL
) {
726 bdrv_close(bs
->file
);
729 bs
->media_changed
= 1;
730 bdrv_dev_change_media_cb(bs
);
734 void bdrv_close_all(void)
736 BlockDriverState
*bs
;
738 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
743 /* make a BlockDriverState anonymous by removing from bdrv_state list.
744 Also, NULL terminate the device_name to prevent double remove */
745 void bdrv_make_anon(BlockDriverState
*bs
)
747 if (bs
->device_name
[0] != '\0') {
748 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
750 bs
->device_name
[0] = '\0';
753 void bdrv_delete(BlockDriverState
*bs
)
757 /* remove from list, if necessary */
761 if (bs
->file
!= NULL
) {
762 bdrv_delete(bs
->file
);
765 assert(bs
!= bs_snapshots
);
769 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
770 /* TODO change to DeviceState *dev when all users are qdevified */
779 /* TODO qdevified devices don't use this, remove when devices are qdevified */
780 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
782 if (bdrv_attach_dev(bs
, dev
) < 0) {
787 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
788 /* TODO change to DeviceState *dev when all users are qdevified */
790 assert(bs
->dev
== dev
);
793 bs
->dev_opaque
= NULL
;
796 /* TODO change to return DeviceState * when all users are qdevified */
797 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
802 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
806 bs
->dev_opaque
= opaque
;
809 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
)
811 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
812 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
);
816 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
818 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
819 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
824 * Run consistency checks on an image
826 * Returns 0 if the check could be completed (it doesn't mean that the image is
827 * free of errors) or -errno when an internal error occurred. The results of the
828 * check are stored in res.
830 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
)
832 if (bs
->drv
->bdrv_check
== NULL
) {
836 memset(res
, 0, sizeof(*res
));
837 return bs
->drv
->bdrv_check(bs
, res
);
840 #define COMMIT_BUF_SECTORS 2048
842 /* commit COW file into the raw image */
843 int bdrv_commit(BlockDriverState
*bs
)
845 BlockDriver
*drv
= bs
->drv
;
846 BlockDriver
*backing_drv
;
847 int64_t sector
, total_sectors
;
848 int n
, ro
, open_flags
;
849 int ret
= 0, rw_ret
= 0;
852 BlockDriverState
*bs_rw
, *bs_ro
;
857 if (!bs
->backing_hd
) {
861 if (bs
->backing_hd
->keep_read_only
) {
865 backing_drv
= bs
->backing_hd
->drv
;
866 ro
= bs
->backing_hd
->read_only
;
867 strncpy(filename
, bs
->backing_hd
->filename
, sizeof(filename
));
868 open_flags
= bs
->backing_hd
->open_flags
;
872 bdrv_delete(bs
->backing_hd
);
873 bs
->backing_hd
= NULL
;
874 bs_rw
= bdrv_new("");
875 rw_ret
= bdrv_open(bs_rw
, filename
, open_flags
| BDRV_O_RDWR
,
879 /* try to re-open read-only */
880 bs_ro
= bdrv_new("");
881 ret
= bdrv_open(bs_ro
, filename
, open_flags
& ~BDRV_O_RDWR
,
885 /* drive not functional anymore */
889 bs
->backing_hd
= bs_ro
;
892 bs
->backing_hd
= bs_rw
;
895 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
896 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
898 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
899 if (drv
->bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
901 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
906 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
913 if (drv
->bdrv_make_empty
) {
914 ret
= drv
->bdrv_make_empty(bs
);
919 * Make sure all data we wrote to the backing device is actually
923 bdrv_flush(bs
->backing_hd
);
930 bdrv_delete(bs
->backing_hd
);
931 bs
->backing_hd
= NULL
;
932 bs_ro
= bdrv_new("");
933 ret
= bdrv_open(bs_ro
, filename
, open_flags
& ~BDRV_O_RDWR
,
937 /* drive not functional anymore */
941 bs
->backing_hd
= bs_ro
;
942 bs
->backing_hd
->keep_read_only
= 0;
948 void bdrv_commit_all(void)
950 BlockDriverState
*bs
;
952 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
960 * -EINVAL - backing format specified, but no file
961 * -ENOSPC - can't update the backing file because no space is left in the
963 * -ENOTSUP - format driver doesn't support changing the backing file
965 int bdrv_change_backing_file(BlockDriverState
*bs
,
966 const char *backing_file
, const char *backing_fmt
)
968 BlockDriver
*drv
= bs
->drv
;
970 if (drv
->bdrv_change_backing_file
!= NULL
) {
971 return drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
977 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
982 if (!bdrv_is_inserted(bs
))
988 len
= bdrv_getlength(bs
);
993 if ((offset
> len
) || (len
- offset
< size
))
999 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
1002 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
1003 nb_sectors
* BDRV_SECTOR_SIZE
);
1006 static inline bool bdrv_has_async_rw(BlockDriver
*drv
)
1008 return drv
->bdrv_co_readv
!= bdrv_co_readv_em
1009 || drv
->bdrv_aio_readv
!= bdrv_aio_readv_em
;
1012 static inline bool bdrv_has_async_flush(BlockDriver
*drv
)
1014 return drv
->bdrv_aio_flush
!= bdrv_aio_flush_em
;
1017 /* return < 0 if error. See bdrv_write() for the return codes */
1018 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
1019 uint8_t *buf
, int nb_sectors
)
1021 BlockDriver
*drv
= bs
->drv
;
1026 if (bdrv_has_async_rw(drv
) && qemu_in_coroutine()) {
1028 struct iovec iov
= {
1029 .iov_base
= (void *)buf
,
1030 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1033 qemu_iovec_init_external(&qiov
, &iov
, 1);
1034 return bdrv_co_readv(bs
, sector_num
, nb_sectors
, &qiov
);
1037 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1040 return drv
->bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
1043 static void set_dirty_bitmap(BlockDriverState
*bs
, int64_t sector_num
,
1044 int nb_sectors
, int dirty
)
1047 unsigned long val
, idx
, bit
;
1049 start
= sector_num
/ BDRV_SECTORS_PER_DIRTY_CHUNK
;
1050 end
= (sector_num
+ nb_sectors
- 1) / BDRV_SECTORS_PER_DIRTY_CHUNK
;
1052 for (; start
<= end
; start
++) {
1053 idx
= start
/ (sizeof(unsigned long) * 8);
1054 bit
= start
% (sizeof(unsigned long) * 8);
1055 val
= bs
->dirty_bitmap
[idx
];
1057 if (!(val
& (1UL << bit
))) {
1062 if (val
& (1UL << bit
)) {
1064 val
&= ~(1UL << bit
);
1067 bs
->dirty_bitmap
[idx
] = val
;
1071 /* Return < 0 if error. Important errors are:
1072 -EIO generic I/O error (may happen for all errors)
1073 -ENOMEDIUM No media inserted.
1074 -EINVAL Invalid sector number or nb_sectors
1075 -EACCES Trying to write a read-only device
1077 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
1078 const uint8_t *buf
, int nb_sectors
)
1080 BlockDriver
*drv
= bs
->drv
;
1085 if (bdrv_has_async_rw(drv
) && qemu_in_coroutine()) {
1087 struct iovec iov
= {
1088 .iov_base
= (void *)buf
,
1089 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1092 qemu_iovec_init_external(&qiov
, &iov
, 1);
1093 return bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
1098 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1101 if (bs
->dirty_bitmap
) {
1102 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1105 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
1106 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
1109 return drv
->bdrv_write(bs
, sector_num
, buf
, nb_sectors
);
1112 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
1113 void *buf
, int count1
)
1115 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
1116 int len
, nb_sectors
, count
;
1121 /* first read to align to sector start */
1122 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
1125 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1127 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1129 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
1137 /* read the sectors "in place" */
1138 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
1139 if (nb_sectors
> 0) {
1140 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
1142 sector_num
+= nb_sectors
;
1143 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
1148 /* add data from the last sector */
1150 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1152 memcpy(buf
, tmp_buf
, count
);
1157 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
1158 const void *buf
, int count1
)
1160 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
1161 int len
, nb_sectors
, count
;
1166 /* first write to align to sector start */
1167 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
1170 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1172 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1174 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
1175 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
1184 /* write the sectors "in place" */
1185 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
1186 if (nb_sectors
> 0) {
1187 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
1189 sector_num
+= nb_sectors
;
1190 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
1195 /* add data from the last sector */
1197 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1199 memcpy(tmp_buf
, buf
, count
);
1200 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
1207 * Writes to the file and ensures that no writes are reordered across this
1208 * request (acts as a barrier)
1210 * Returns 0 on success, -errno in error cases.
1212 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
1213 const void *buf
, int count
)
1217 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
1222 /* No flush needed for cache modes that use O_DSYNC */
1223 if ((bs
->open_flags
& BDRV_O_CACHE_WB
) != 0) {
1230 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1231 int nb_sectors
, QEMUIOVector
*qiov
)
1233 BlockDriver
*drv
= bs
->drv
;
1235 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1240 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
1244 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
1247 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1248 int nb_sectors
, QEMUIOVector
*qiov
)
1250 BlockDriver
*drv
= bs
->drv
;
1252 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1257 if (bs
->read_only
) {
1260 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
1264 if (bs
->dirty_bitmap
) {
1265 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1268 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
1269 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
1272 return drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
1276 * Truncate file to 'offset' bytes (needed only for file protocols)
1278 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
1280 BlockDriver
*drv
= bs
->drv
;
1284 if (!drv
->bdrv_truncate
)
1288 if (bdrv_in_use(bs
))
1290 ret
= drv
->bdrv_truncate(bs
, offset
);
1292 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
1293 bdrv_dev_resize_cb(bs
);
1299 * Length of a allocated file in bytes. Sparse files are counted by actual
1300 * allocated space. Return < 0 if error or unknown.
1302 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
1304 BlockDriver
*drv
= bs
->drv
;
1308 if (drv
->bdrv_get_allocated_file_size
) {
1309 return drv
->bdrv_get_allocated_file_size(bs
);
1312 return bdrv_get_allocated_file_size(bs
->file
);
1318 * Length of a file in bytes. Return < 0 if error or unknown.
1320 int64_t bdrv_getlength(BlockDriverState
*bs
)
1322 BlockDriver
*drv
= bs
->drv
;
1326 if (bs
->growable
|| bs
->removable
) {
1327 if (drv
->bdrv_getlength
) {
1328 return drv
->bdrv_getlength(bs
);
1331 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
1334 /* return 0 as number of sectors if no device present or error */
1335 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
1338 length
= bdrv_getlength(bs
);
1342 length
= length
>> BDRV_SECTOR_BITS
;
1343 *nb_sectors_ptr
= length
;
1347 uint8_t boot_ind
; /* 0x80 - active */
1348 uint8_t head
; /* starting head */
1349 uint8_t sector
; /* starting sector */
1350 uint8_t cyl
; /* starting cylinder */
1351 uint8_t sys_ind
; /* What partition type */
1352 uint8_t end_head
; /* end head */
1353 uint8_t end_sector
; /* end sector */
1354 uint8_t end_cyl
; /* end cylinder */
1355 uint32_t start_sect
; /* starting sector counting from 0 */
1356 uint32_t nr_sects
; /* nr of sectors in partition */
1359 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1360 static int guess_disk_lchs(BlockDriverState
*bs
,
1361 int *pcylinders
, int *pheads
, int *psectors
)
1363 uint8_t buf
[BDRV_SECTOR_SIZE
];
1364 int ret
, i
, heads
, sectors
, cylinders
;
1365 struct partition
*p
;
1367 uint64_t nb_sectors
;
1369 bdrv_get_geometry(bs
, &nb_sectors
);
1371 ret
= bdrv_read(bs
, 0, buf
, 1);
1374 /* test msdos magic */
1375 if (buf
[510] != 0x55 || buf
[511] != 0xaa)
1377 for(i
= 0; i
< 4; i
++) {
1378 p
= ((struct partition
*)(buf
+ 0x1be)) + i
;
1379 nr_sects
= le32_to_cpu(p
->nr_sects
);
1380 if (nr_sects
&& p
->end_head
) {
1381 /* We make the assumption that the partition terminates on
1382 a cylinder boundary */
1383 heads
= p
->end_head
+ 1;
1384 sectors
= p
->end_sector
& 63;
1387 cylinders
= nb_sectors
/ (heads
* sectors
);
1388 if (cylinders
< 1 || cylinders
> 16383)
1391 *psectors
= sectors
;
1392 *pcylinders
= cylinders
;
1394 printf("guessed geometry: LCHS=%d %d %d\n",
1395 cylinders
, heads
, sectors
);
1403 void bdrv_guess_geometry(BlockDriverState
*bs
, int *pcyls
, int *pheads
, int *psecs
)
1405 int translation
, lba_detected
= 0;
1406 int cylinders
, heads
, secs
;
1407 uint64_t nb_sectors
;
1409 /* if a geometry hint is available, use it */
1410 bdrv_get_geometry(bs
, &nb_sectors
);
1411 bdrv_get_geometry_hint(bs
, &cylinders
, &heads
, &secs
);
1412 translation
= bdrv_get_translation_hint(bs
);
1413 if (cylinders
!= 0) {
1418 if (guess_disk_lchs(bs
, &cylinders
, &heads
, &secs
) == 0) {
1420 /* if heads > 16, it means that a BIOS LBA
1421 translation was active, so the default
1422 hardware geometry is OK */
1424 goto default_geometry
;
1429 /* disable any translation to be in sync with
1430 the logical geometry */
1431 if (translation
== BIOS_ATA_TRANSLATION_AUTO
) {
1432 bdrv_set_translation_hint(bs
,
1433 BIOS_ATA_TRANSLATION_NONE
);
1438 /* if no geometry, use a standard physical disk geometry */
1439 cylinders
= nb_sectors
/ (16 * 63);
1441 if (cylinders
> 16383)
1443 else if (cylinders
< 2)
1448 if ((lba_detected
== 1) && (translation
== BIOS_ATA_TRANSLATION_AUTO
)) {
1449 if ((*pcyls
* *pheads
) <= 131072) {
1450 bdrv_set_translation_hint(bs
,
1451 BIOS_ATA_TRANSLATION_LARGE
);
1453 bdrv_set_translation_hint(bs
,
1454 BIOS_ATA_TRANSLATION_LBA
);
1458 bdrv_set_geometry_hint(bs
, *pcyls
, *pheads
, *psecs
);
1462 void bdrv_set_geometry_hint(BlockDriverState
*bs
,
1463 int cyls
, int heads
, int secs
)
1470 void bdrv_set_translation_hint(BlockDriverState
*bs
, int translation
)
1472 bs
->translation
= translation
;
1475 void bdrv_get_geometry_hint(BlockDriverState
*bs
,
1476 int *pcyls
, int *pheads
, int *psecs
)
1479 *pheads
= bs
->heads
;
1483 /* Recognize floppy formats */
1484 typedef struct FDFormat
{
1491 static const FDFormat fd_formats
[] = {
1492 /* First entry is default format */
1493 /* 1.44 MB 3"1/2 floppy disks */
1494 { FDRIVE_DRV_144
, 18, 80, 1, },
1495 { FDRIVE_DRV_144
, 20, 80, 1, },
1496 { FDRIVE_DRV_144
, 21, 80, 1, },
1497 { FDRIVE_DRV_144
, 21, 82, 1, },
1498 { FDRIVE_DRV_144
, 21, 83, 1, },
1499 { FDRIVE_DRV_144
, 22, 80, 1, },
1500 { FDRIVE_DRV_144
, 23, 80, 1, },
1501 { FDRIVE_DRV_144
, 24, 80, 1, },
1502 /* 2.88 MB 3"1/2 floppy disks */
1503 { FDRIVE_DRV_288
, 36, 80, 1, },
1504 { FDRIVE_DRV_288
, 39, 80, 1, },
1505 { FDRIVE_DRV_288
, 40, 80, 1, },
1506 { FDRIVE_DRV_288
, 44, 80, 1, },
1507 { FDRIVE_DRV_288
, 48, 80, 1, },
1508 /* 720 kB 3"1/2 floppy disks */
1509 { FDRIVE_DRV_144
, 9, 80, 1, },
1510 { FDRIVE_DRV_144
, 10, 80, 1, },
1511 { FDRIVE_DRV_144
, 10, 82, 1, },
1512 { FDRIVE_DRV_144
, 10, 83, 1, },
1513 { FDRIVE_DRV_144
, 13, 80, 1, },
1514 { FDRIVE_DRV_144
, 14, 80, 1, },
1515 /* 1.2 MB 5"1/4 floppy disks */
1516 { FDRIVE_DRV_120
, 15, 80, 1, },
1517 { FDRIVE_DRV_120
, 18, 80, 1, },
1518 { FDRIVE_DRV_120
, 18, 82, 1, },
1519 { FDRIVE_DRV_120
, 18, 83, 1, },
1520 { FDRIVE_DRV_120
, 20, 80, 1, },
1521 /* 720 kB 5"1/4 floppy disks */
1522 { FDRIVE_DRV_120
, 9, 80, 1, },
1523 { FDRIVE_DRV_120
, 11, 80, 1, },
1524 /* 360 kB 5"1/4 floppy disks */
1525 { FDRIVE_DRV_120
, 9, 40, 1, },
1526 { FDRIVE_DRV_120
, 9, 40, 0, },
1527 { FDRIVE_DRV_120
, 10, 41, 1, },
1528 { FDRIVE_DRV_120
, 10, 42, 1, },
1529 /* 320 kB 5"1/4 floppy disks */
1530 { FDRIVE_DRV_120
, 8, 40, 1, },
1531 { FDRIVE_DRV_120
, 8, 40, 0, },
1532 /* 360 kB must match 5"1/4 better than 3"1/2... */
1533 { FDRIVE_DRV_144
, 9, 80, 0, },
1535 { FDRIVE_DRV_NONE
, -1, -1, 0, },
1538 void bdrv_get_floppy_geometry_hint(BlockDriverState
*bs
, int *nb_heads
,
1539 int *max_track
, int *last_sect
,
1540 FDriveType drive_in
, FDriveType
*drive
)
1542 const FDFormat
*parse
;
1543 uint64_t nb_sectors
, size
;
1544 int i
, first_match
, match
;
1546 bdrv_get_geometry_hint(bs
, nb_heads
, max_track
, last_sect
);
1547 if (*nb_heads
!= 0 && *max_track
!= 0 && *last_sect
!= 0) {
1548 /* User defined disk */
1550 bdrv_get_geometry(bs
, &nb_sectors
);
1553 for (i
= 0; ; i
++) {
1554 parse
= &fd_formats
[i
];
1555 if (parse
->drive
== FDRIVE_DRV_NONE
) {
1558 if (drive_in
== parse
->drive
||
1559 drive_in
== FDRIVE_DRV_NONE
) {
1560 size
= (parse
->max_head
+ 1) * parse
->max_track
*
1562 if (nb_sectors
== size
) {
1566 if (first_match
== -1) {
1572 if (first_match
== -1) {
1575 match
= first_match
;
1577 parse
= &fd_formats
[match
];
1579 *nb_heads
= parse
->max_head
+ 1;
1580 *max_track
= parse
->max_track
;
1581 *last_sect
= parse
->last_sect
;
1582 *drive
= parse
->drive
;
1586 int bdrv_get_translation_hint(BlockDriverState
*bs
)
1588 return bs
->translation
;
1591 void bdrv_set_on_error(BlockDriverState
*bs
, BlockErrorAction on_read_error
,
1592 BlockErrorAction on_write_error
)
1594 bs
->on_read_error
= on_read_error
;
1595 bs
->on_write_error
= on_write_error
;
1598 BlockErrorAction
bdrv_get_on_error(BlockDriverState
*bs
, int is_read
)
1600 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
1603 void bdrv_set_removable(BlockDriverState
*bs
, int removable
)
1605 bs
->removable
= removable
;
1606 if (removable
&& bs
== bs_snapshots
) {
1607 bs_snapshots
= NULL
;
1611 int bdrv_is_removable(BlockDriverState
*bs
)
1613 return bs
->removable
;
1616 int bdrv_is_read_only(BlockDriverState
*bs
)
1618 return bs
->read_only
;
1621 int bdrv_is_sg(BlockDriverState
*bs
)
1626 int bdrv_enable_write_cache(BlockDriverState
*bs
)
1628 return bs
->enable_write_cache
;
1631 int bdrv_is_encrypted(BlockDriverState
*bs
)
1633 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
1635 return bs
->encrypted
;
1638 int bdrv_key_required(BlockDriverState
*bs
)
1640 BlockDriverState
*backing_hd
= bs
->backing_hd
;
1642 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
1644 return (bs
->encrypted
&& !bs
->valid_key
);
1647 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
1650 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
1651 ret
= bdrv_set_key(bs
->backing_hd
, key
);
1657 if (!bs
->encrypted
) {
1659 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
1662 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
1665 } else if (!bs
->valid_key
) {
1667 /* call the change callback now, we skipped it on open */
1668 bs
->media_changed
= 1;
1669 bdrv_dev_change_media_cb(bs
);
1674 void bdrv_get_format(BlockDriverState
*bs
, char *buf
, int buf_size
)
1679 pstrcpy(buf
, buf_size
, bs
->drv
->format_name
);
1683 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
1688 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
1689 it(opaque
, drv
->format_name
);
1693 BlockDriverState
*bdrv_find(const char *name
)
1695 BlockDriverState
*bs
;
1697 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1698 if (!strcmp(name
, bs
->device_name
)) {
1705 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
1708 return QTAILQ_FIRST(&bdrv_states
);
1710 return QTAILQ_NEXT(bs
, list
);
1713 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
1715 BlockDriverState
*bs
;
1717 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1722 const char *bdrv_get_device_name(BlockDriverState
*bs
)
1724 return bs
->device_name
;
1727 int bdrv_flush(BlockDriverState
*bs
)
1729 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
1733 if (bs
->drv
&& bdrv_has_async_flush(bs
->drv
) && qemu_in_coroutine()) {
1734 return bdrv_co_flush_em(bs
);
1737 if (bs
->drv
&& bs
->drv
->bdrv_flush
) {
1738 return bs
->drv
->bdrv_flush(bs
);
1742 * Some block drivers always operate in either writethrough or unsafe mode
1743 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1744 * the server works (because the behaviour is hardcoded or depends on
1745 * server-side configuration), so we can't ensure that everything is safe
1746 * on disk. Returning an error doesn't work because that would break guests
1747 * even if the server operates in writethrough mode.
1749 * Let's hope the user knows what he's doing.
1754 void bdrv_flush_all(void)
1756 BlockDriverState
*bs
;
1758 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1759 if (bs
->drv
&& !bdrv_is_read_only(bs
) &&
1760 (!bdrv_is_removable(bs
) || bdrv_is_inserted(bs
))) {
1766 int bdrv_has_zero_init(BlockDriverState
*bs
)
1770 if (bs
->drv
->bdrv_has_zero_init
) {
1771 return bs
->drv
->bdrv_has_zero_init(bs
);
1777 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
1782 if (!bs
->drv
->bdrv_discard
) {
1785 return bs
->drv
->bdrv_discard(bs
, sector_num
, nb_sectors
);
1789 * Returns true iff the specified sector is present in the disk image. Drivers
1790 * not implementing the functionality are assumed to not support backing files,
1791 * hence all their sectors are reported as allocated.
1793 * 'pnum' is set to the number of sectors (including and immediately following
1794 * the specified sector) that are known to be in the same
1795 * allocated/unallocated state.
1797 * 'nb_sectors' is the max value 'pnum' should be set to.
1799 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
1803 if (!bs
->drv
->bdrv_is_allocated
) {
1804 if (sector_num
>= bs
->total_sectors
) {
1808 n
= bs
->total_sectors
- sector_num
;
1809 *pnum
= (n
< nb_sectors
) ? (n
) : (nb_sectors
);
1812 return bs
->drv
->bdrv_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
1815 void bdrv_mon_event(const BlockDriverState
*bdrv
,
1816 BlockMonEventAction action
, int is_read
)
1819 const char *action_str
;
1822 case BDRV_ACTION_REPORT
:
1823 action_str
= "report";
1825 case BDRV_ACTION_IGNORE
:
1826 action_str
= "ignore";
1828 case BDRV_ACTION_STOP
:
1829 action_str
= "stop";
1835 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1838 is_read
? "read" : "write");
1839 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR
, data
);
1841 qobject_decref(data
);
1844 static void bdrv_print_dict(QObject
*obj
, void *opaque
)
1847 Monitor
*mon
= opaque
;
1849 bs_dict
= qobject_to_qdict(obj
);
1851 monitor_printf(mon
, "%s: removable=%d",
1852 qdict_get_str(bs_dict
, "device"),
1853 qdict_get_bool(bs_dict
, "removable"));
1855 if (qdict_get_bool(bs_dict
, "removable")) {
1856 monitor_printf(mon
, " locked=%d", qdict_get_bool(bs_dict
, "locked"));
1859 if (qdict_haskey(bs_dict
, "inserted")) {
1860 QDict
*qdict
= qobject_to_qdict(qdict_get(bs_dict
, "inserted"));
1862 monitor_printf(mon
, " file=");
1863 monitor_print_filename(mon
, qdict_get_str(qdict
, "file"));
1864 if (qdict_haskey(qdict
, "backing_file")) {
1865 monitor_printf(mon
, " backing_file=");
1866 monitor_print_filename(mon
, qdict_get_str(qdict
, "backing_file"));
1868 monitor_printf(mon
, " ro=%d drv=%s encrypted=%d",
1869 qdict_get_bool(qdict
, "ro"),
1870 qdict_get_str(qdict
, "drv"),
1871 qdict_get_bool(qdict
, "encrypted"));
1873 monitor_printf(mon
, " [not inserted]");
1876 monitor_printf(mon
, "\n");
1879 void bdrv_info_print(Monitor
*mon
, const QObject
*data
)
1881 qlist_iter(qobject_to_qlist(data
), bdrv_print_dict
, mon
);
1884 void bdrv_info(Monitor
*mon
, QObject
**ret_data
)
1887 BlockDriverState
*bs
;
1889 bs_list
= qlist_new();
1891 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1894 bs_obj
= qobject_from_jsonf("{ 'device': %s, 'type': 'unknown', "
1895 "'removable': %i, 'locked': %i }",
1896 bs
->device_name
, bs
->removable
,
1901 QDict
*bs_dict
= qobject_to_qdict(bs_obj
);
1903 obj
= qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1904 "'encrypted': %i }",
1905 bs
->filename
, bs
->read_only
,
1906 bs
->drv
->format_name
,
1907 bdrv_is_encrypted(bs
));
1908 if (bs
->backing_file
[0] != '\0') {
1909 QDict
*qdict
= qobject_to_qdict(obj
);
1910 qdict_put(qdict
, "backing_file",
1911 qstring_from_str(bs
->backing_file
));
1914 qdict_put_obj(bs_dict
, "inserted", obj
);
1916 qlist_append_obj(bs_list
, bs_obj
);
1919 *ret_data
= QOBJECT(bs_list
);
1922 static void bdrv_stats_iter(QObject
*data
, void *opaque
)
1925 Monitor
*mon
= opaque
;
1927 qdict
= qobject_to_qdict(data
);
1928 monitor_printf(mon
, "%s:", qdict_get_str(qdict
, "device"));
1930 qdict
= qobject_to_qdict(qdict_get(qdict
, "stats"));
1931 monitor_printf(mon
, " rd_bytes=%" PRId64
1932 " wr_bytes=%" PRId64
1933 " rd_operations=%" PRId64
1934 " wr_operations=%" PRId64
1935 " flush_operations=%" PRId64
1936 " wr_total_time_ns=%" PRId64
1937 " rd_total_time_ns=%" PRId64
1938 " flush_total_time_ns=%" PRId64
1940 qdict_get_int(qdict
, "rd_bytes"),
1941 qdict_get_int(qdict
, "wr_bytes"),
1942 qdict_get_int(qdict
, "rd_operations"),
1943 qdict_get_int(qdict
, "wr_operations"),
1944 qdict_get_int(qdict
, "flush_operations"),
1945 qdict_get_int(qdict
, "wr_total_time_ns"),
1946 qdict_get_int(qdict
, "rd_total_time_ns"),
1947 qdict_get_int(qdict
, "flush_total_time_ns"));
1950 void bdrv_stats_print(Monitor
*mon
, const QObject
*data
)
1952 qlist_iter(qobject_to_qlist(data
), bdrv_stats_iter
, mon
);
1955 static QObject
* bdrv_info_stats_bs(BlockDriverState
*bs
)
1960 res
= qobject_from_jsonf("{ 'stats': {"
1961 "'rd_bytes': %" PRId64
","
1962 "'wr_bytes': %" PRId64
","
1963 "'rd_operations': %" PRId64
","
1964 "'wr_operations': %" PRId64
","
1965 "'wr_highest_offset': %" PRId64
","
1966 "'flush_operations': %" PRId64
","
1967 "'wr_total_time_ns': %" PRId64
","
1968 "'rd_total_time_ns': %" PRId64
","
1969 "'flush_total_time_ns': %" PRId64
1971 bs
->nr_bytes
[BDRV_ACCT_READ
],
1972 bs
->nr_bytes
[BDRV_ACCT_WRITE
],
1973 bs
->nr_ops
[BDRV_ACCT_READ
],
1974 bs
->nr_ops
[BDRV_ACCT_WRITE
],
1975 bs
->wr_highest_sector
*
1976 (uint64_t)BDRV_SECTOR_SIZE
,
1977 bs
->nr_ops
[BDRV_ACCT_FLUSH
],
1978 bs
->total_time_ns
[BDRV_ACCT_WRITE
],
1979 bs
->total_time_ns
[BDRV_ACCT_READ
],
1980 bs
->total_time_ns
[BDRV_ACCT_FLUSH
]);
1981 dict
= qobject_to_qdict(res
);
1983 if (*bs
->device_name
) {
1984 qdict_put(dict
, "device", qstring_from_str(bs
->device_name
));
1988 QObject
*parent
= bdrv_info_stats_bs(bs
->file
);
1989 qdict_put_obj(dict
, "parent", parent
);
1995 void bdrv_info_stats(Monitor
*mon
, QObject
**ret_data
)
1999 BlockDriverState
*bs
;
2001 devices
= qlist_new();
2003 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2004 obj
= bdrv_info_stats_bs(bs
);
2005 qlist_append_obj(devices
, obj
);
2008 *ret_data
= QOBJECT(devices
);
2011 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
2013 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2014 return bs
->backing_file
;
2015 else if (bs
->encrypted
)
2016 return bs
->filename
;
2021 void bdrv_get_backing_filename(BlockDriverState
*bs
,
2022 char *filename
, int filename_size
)
2024 if (!bs
->backing_file
) {
2025 pstrcpy(filename
, filename_size
, "");
2027 pstrcpy(filename
, filename_size
, bs
->backing_file
);
2031 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
2032 const uint8_t *buf
, int nb_sectors
)
2034 BlockDriver
*drv
= bs
->drv
;
2037 if (!drv
->bdrv_write_compressed
)
2039 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2042 if (bs
->dirty_bitmap
) {
2043 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
2046 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
2049 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2051 BlockDriver
*drv
= bs
->drv
;
2054 if (!drv
->bdrv_get_info
)
2056 memset(bdi
, 0, sizeof(*bdi
));
2057 return drv
->bdrv_get_info(bs
, bdi
);
2060 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2061 int64_t pos
, int size
)
2063 BlockDriver
*drv
= bs
->drv
;
2066 if (drv
->bdrv_save_vmstate
)
2067 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
2069 return bdrv_save_vmstate(bs
->file
, buf
, pos
, size
);
2073 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2074 int64_t pos
, int size
)
2076 BlockDriver
*drv
= bs
->drv
;
2079 if (drv
->bdrv_load_vmstate
)
2080 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
2082 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
2086 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
2088 BlockDriver
*drv
= bs
->drv
;
2090 if (!drv
|| !drv
->bdrv_debug_event
) {
2094 return drv
->bdrv_debug_event(bs
, event
);
2098 /**************************************************************/
2099 /* handling of snapshots */
2101 int bdrv_can_snapshot(BlockDriverState
*bs
)
2103 BlockDriver
*drv
= bs
->drv
;
2104 if (!drv
|| bdrv_is_removable(bs
) || bdrv_is_read_only(bs
)) {
2108 if (!drv
->bdrv_snapshot_create
) {
2109 if (bs
->file
!= NULL
) {
2110 return bdrv_can_snapshot(bs
->file
);
2118 int bdrv_is_snapshot(BlockDriverState
*bs
)
2120 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
2123 BlockDriverState
*bdrv_snapshots(void)
2125 BlockDriverState
*bs
;
2128 return bs_snapshots
;
2132 while ((bs
= bdrv_next(bs
))) {
2133 if (bdrv_can_snapshot(bs
)) {
2141 int bdrv_snapshot_create(BlockDriverState
*bs
,
2142 QEMUSnapshotInfo
*sn_info
)
2144 BlockDriver
*drv
= bs
->drv
;
2147 if (drv
->bdrv_snapshot_create
)
2148 return drv
->bdrv_snapshot_create(bs
, sn_info
);
2150 return bdrv_snapshot_create(bs
->file
, sn_info
);
2154 int bdrv_snapshot_goto(BlockDriverState
*bs
,
2155 const char *snapshot_id
)
2157 BlockDriver
*drv
= bs
->drv
;
2162 if (drv
->bdrv_snapshot_goto
)
2163 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
2166 drv
->bdrv_close(bs
);
2167 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
2168 open_ret
= drv
->bdrv_open(bs
, bs
->open_flags
);
2170 bdrv_delete(bs
->file
);
2180 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
2182 BlockDriver
*drv
= bs
->drv
;
2185 if (drv
->bdrv_snapshot_delete
)
2186 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
2188 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
2192 int bdrv_snapshot_list(BlockDriverState
*bs
,
2193 QEMUSnapshotInfo
**psn_info
)
2195 BlockDriver
*drv
= bs
->drv
;
2198 if (drv
->bdrv_snapshot_list
)
2199 return drv
->bdrv_snapshot_list(bs
, psn_info
);
2201 return bdrv_snapshot_list(bs
->file
, psn_info
);
2205 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
2206 const char *snapshot_name
)
2208 BlockDriver
*drv
= bs
->drv
;
2212 if (!bs
->read_only
) {
2215 if (drv
->bdrv_snapshot_load_tmp
) {
2216 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
2221 #define NB_SUFFIXES 4
2223 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
2225 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
2230 snprintf(buf
, buf_size
, "%" PRId64
, size
);
2233 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
2234 if (size
< (10 * base
)) {
2235 snprintf(buf
, buf_size
, "%0.1f%c",
2236 (double)size
/ base
,
2239 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
2240 snprintf(buf
, buf_size
, "%" PRId64
"%c",
2241 ((size
+ (base
>> 1)) / base
),
2251 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
2253 char buf1
[128], date_buf
[128], clock_buf
[128];
2263 snprintf(buf
, buf_size
,
2264 "%-10s%-20s%7s%20s%15s",
2265 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2269 ptm
= localtime(&ti
);
2270 strftime(date_buf
, sizeof(date_buf
),
2271 "%Y-%m-%d %H:%M:%S", ptm
);
2273 localtime_r(&ti
, &tm
);
2274 strftime(date_buf
, sizeof(date_buf
),
2275 "%Y-%m-%d %H:%M:%S", &tm
);
2277 secs
= sn
->vm_clock_nsec
/ 1000000000;
2278 snprintf(clock_buf
, sizeof(clock_buf
),
2279 "%02d:%02d:%02d.%03d",
2281 (int)((secs
/ 60) % 60),
2283 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
2284 snprintf(buf
, buf_size
,
2285 "%-10s%-20s%7s%20s%15s",
2286 sn
->id_str
, sn
->name
,
2287 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
2294 /**************************************************************/
2297 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
2298 QEMUIOVector
*qiov
, int nb_sectors
,
2299 BlockDriverCompletionFunc
*cb
, void *opaque
)
2301 BlockDriver
*drv
= bs
->drv
;
2303 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
2307 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2310 return drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
2314 typedef struct BlockCompleteData
{
2315 BlockDriverCompletionFunc
*cb
;
2317 BlockDriverState
*bs
;
2320 } BlockCompleteData
;
2322 static void block_complete_cb(void *opaque
, int ret
)
2324 BlockCompleteData
*b
= opaque
;
2326 if (b
->bs
->dirty_bitmap
) {
2327 set_dirty_bitmap(b
->bs
, b
->sector_num
, b
->nb_sectors
, 1);
2329 b
->cb(b
->opaque
, ret
);
2333 static BlockCompleteData
*blk_dirty_cb_alloc(BlockDriverState
*bs
,
2336 BlockDriverCompletionFunc
*cb
,
2339 BlockCompleteData
*blkdata
= g_malloc0(sizeof(BlockCompleteData
));
2343 blkdata
->opaque
= opaque
;
2344 blkdata
->sector_num
= sector_num
;
2345 blkdata
->nb_sectors
= nb_sectors
;
2350 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
2351 QEMUIOVector
*qiov
, int nb_sectors
,
2352 BlockDriverCompletionFunc
*cb
, void *opaque
)
2354 BlockDriver
*drv
= bs
->drv
;
2355 BlockDriverAIOCB
*ret
;
2356 BlockCompleteData
*blk_cb_data
;
2358 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
2364 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2367 if (bs
->dirty_bitmap
) {
2368 blk_cb_data
= blk_dirty_cb_alloc(bs
, sector_num
, nb_sectors
, cb
,
2370 cb
= &block_complete_cb
;
2371 opaque
= blk_cb_data
;
2374 ret
= drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
2378 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2379 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2387 typedef struct MultiwriteCB
{
2392 BlockDriverCompletionFunc
*cb
;
2394 QEMUIOVector
*free_qiov
;
2399 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
2403 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
2404 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
2405 if (mcb
->callbacks
[i
].free_qiov
) {
2406 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
2408 g_free(mcb
->callbacks
[i
].free_qiov
);
2409 qemu_vfree(mcb
->callbacks
[i
].free_buf
);
2413 static void multiwrite_cb(void *opaque
, int ret
)
2415 MultiwriteCB
*mcb
= opaque
;
2417 trace_multiwrite_cb(mcb
, ret
);
2419 if (ret
< 0 && !mcb
->error
) {
2423 mcb
->num_requests
--;
2424 if (mcb
->num_requests
== 0) {
2425 multiwrite_user_cb(mcb
);
2430 static int multiwrite_req_compare(const void *a
, const void *b
)
2432 const BlockRequest
*req1
= a
, *req2
= b
;
2435 * Note that we can't simply subtract req2->sector from req1->sector
2436 * here as that could overflow the return value.
2438 if (req1
->sector
> req2
->sector
) {
2440 } else if (req1
->sector
< req2
->sector
) {
2448 * Takes a bunch of requests and tries to merge them. Returns the number of
2449 * requests that remain after merging.
2451 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
2452 int num_reqs
, MultiwriteCB
*mcb
)
2456 // Sort requests by start sector
2457 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
2459 // Check if adjacent requests touch the same clusters. If so, combine them,
2460 // filling up gaps with zero sectors.
2462 for (i
= 1; i
< num_reqs
; i
++) {
2464 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
2466 // This handles the cases that are valid for all block drivers, namely
2467 // exactly sequential writes and overlapping writes.
2468 if (reqs
[i
].sector
<= oldreq_last
) {
2472 // The block driver may decide that it makes sense to combine requests
2473 // even if there is a gap of some sectors between them. In this case,
2474 // the gap is filled with zeros (therefore only applicable for yet
2475 // unused space in format like qcow2).
2476 if (!merge
&& bs
->drv
->bdrv_merge_requests
) {
2477 merge
= bs
->drv
->bdrv_merge_requests(bs
, &reqs
[outidx
], &reqs
[i
]);
2480 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
2486 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
2487 qemu_iovec_init(qiov
,
2488 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
2490 // Add the first request to the merged one. If the requests are
2491 // overlapping, drop the last sectors of the first request.
2492 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
2493 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, size
);
2495 // We might need to add some zeros between the two requests
2496 if (reqs
[i
].sector
> oldreq_last
) {
2497 size_t zero_bytes
= (reqs
[i
].sector
- oldreq_last
) << 9;
2498 uint8_t *buf
= qemu_blockalign(bs
, zero_bytes
);
2499 memset(buf
, 0, zero_bytes
);
2500 qemu_iovec_add(qiov
, buf
, zero_bytes
);
2501 mcb
->callbacks
[i
].free_buf
= buf
;
2504 // Add the second request
2505 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, reqs
[i
].qiov
->size
);
2507 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
2508 reqs
[outidx
].qiov
= qiov
;
2510 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
2513 reqs
[outidx
].sector
= reqs
[i
].sector
;
2514 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
2515 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
2523 * Submit multiple AIO write requests at once.
2525 * On success, the function returns 0 and all requests in the reqs array have
2526 * been submitted. In error case this function returns -1, and any of the
2527 * requests may or may not be submitted yet. In particular, this means that the
2528 * callback will be called for some of the requests, for others it won't. The
2529 * caller must check the error field of the BlockRequest to wait for the right
2530 * callbacks (if error != 0, no callback will be called).
2532 * The implementation may modify the contents of the reqs array, e.g. to merge
2533 * requests. However, the fields opaque and error are left unmodified as they
2534 * are used to signal failure for a single request to the caller.
2536 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
2538 BlockDriverAIOCB
*acb
;
2542 /* don't submit writes if we don't have a medium */
2543 if (bs
->drv
== NULL
) {
2544 for (i
= 0; i
< num_reqs
; i
++) {
2545 reqs
[i
].error
= -ENOMEDIUM
;
2550 if (num_reqs
== 0) {
2554 // Create MultiwriteCB structure
2555 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
2556 mcb
->num_requests
= 0;
2557 mcb
->num_callbacks
= num_reqs
;
2559 for (i
= 0; i
< num_reqs
; i
++) {
2560 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
2561 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
2564 // Check for mergable requests
2565 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
2567 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
2570 * Run the aio requests. As soon as one request can't be submitted
2571 * successfully, fail all requests that are not yet submitted (we must
2572 * return failure for all requests anyway)
2574 * num_requests cannot be set to the right value immediately: If
2575 * bdrv_aio_writev fails for some request, num_requests would be too high
2576 * and therefore multiwrite_cb() would never recognize the multiwrite
2577 * request as completed. We also cannot use the loop variable i to set it
2578 * when the first request fails because the callback may already have been
2579 * called for previously submitted requests. Thus, num_requests must be
2580 * incremented for each request that is submitted.
2582 * The problem that callbacks may be called early also means that we need
2583 * to take care that num_requests doesn't become 0 before all requests are
2584 * submitted - multiwrite_cb() would consider the multiwrite request
2585 * completed. A dummy request that is "completed" by a manual call to
2586 * multiwrite_cb() takes care of this.
2588 mcb
->num_requests
= 1;
2590 // Run the aio requests
2591 for (i
= 0; i
< num_reqs
; i
++) {
2592 mcb
->num_requests
++;
2593 acb
= bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
2594 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
2597 // We can only fail the whole thing if no request has been
2598 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2599 // complete and report the error in the callback.
2601 trace_bdrv_aio_multiwrite_earlyfail(mcb
);
2604 trace_bdrv_aio_multiwrite_latefail(mcb
, i
);
2605 multiwrite_cb(mcb
, -EIO
);
2611 /* Complete the dummy request */
2612 multiwrite_cb(mcb
, 0);
2617 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
2618 reqs
[i
].error
= -EIO
;
2624 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2625 BlockDriverCompletionFunc
*cb
, void *opaque
)
2627 BlockDriver
*drv
= bs
->drv
;
2629 trace_bdrv_aio_flush(bs
, opaque
);
2631 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2632 return bdrv_aio_noop_em(bs
, cb
, opaque
);
2637 return drv
->bdrv_aio_flush(bs
, cb
, opaque
);
2640 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
2642 acb
->pool
->cancel(acb
);
2646 /**************************************************************/
2647 /* async block device emulation */
2649 typedef struct BlockDriverAIOCBSync
{
2650 BlockDriverAIOCB common
;
2653 /* vector translation state */
2657 } BlockDriverAIOCBSync
;
2659 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
2661 BlockDriverAIOCBSync
*acb
=
2662 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
2663 qemu_bh_delete(acb
->bh
);
2665 qemu_aio_release(acb
);
2668 static AIOPool bdrv_em_aio_pool
= {
2669 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
2670 .cancel
= bdrv_aio_cancel_em
,
2673 static void bdrv_aio_bh_cb(void *opaque
)
2675 BlockDriverAIOCBSync
*acb
= opaque
;
2678 qemu_iovec_from_buffer(acb
->qiov
, acb
->bounce
, acb
->qiov
->size
);
2679 qemu_vfree(acb
->bounce
);
2680 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
2681 qemu_bh_delete(acb
->bh
);
2683 qemu_aio_release(acb
);
2686 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
2690 BlockDriverCompletionFunc
*cb
,
2695 BlockDriverAIOCBSync
*acb
;
2697 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2698 acb
->is_write
= is_write
;
2700 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
2703 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2706 qemu_iovec_to_buffer(acb
->qiov
, acb
->bounce
);
2707 acb
->ret
= bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2709 acb
->ret
= bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2712 qemu_bh_schedule(acb
->bh
);
2714 return &acb
->common
;
2717 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2718 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2719 BlockDriverCompletionFunc
*cb
, void *opaque
)
2721 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2724 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2725 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2726 BlockDriverCompletionFunc
*cb
, void *opaque
)
2728 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2732 typedef struct BlockDriverAIOCBCoroutine
{
2733 BlockDriverAIOCB common
;
2737 } BlockDriverAIOCBCoroutine
;
2739 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
2744 static AIOPool bdrv_em_co_aio_pool
= {
2745 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
2746 .cancel
= bdrv_aio_co_cancel_em
,
2749 static void bdrv_co_rw_bh(void *opaque
)
2751 BlockDriverAIOCBCoroutine
*acb
= opaque
;
2753 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2754 qemu_bh_delete(acb
->bh
);
2755 qemu_aio_release(acb
);
2758 static void coroutine_fn
bdrv_co_rw(void *opaque
)
2760 BlockDriverAIOCBCoroutine
*acb
= opaque
;
2761 BlockDriverState
*bs
= acb
->common
.bs
;
2763 if (!acb
->is_write
) {
2764 acb
->req
.error
= bs
->drv
->bdrv_co_readv(bs
, acb
->req
.sector
,
2765 acb
->req
.nb_sectors
, acb
->req
.qiov
);
2767 acb
->req
.error
= bs
->drv
->bdrv_co_writev(bs
, acb
->req
.sector
,
2768 acb
->req
.nb_sectors
, acb
->req
.qiov
);
2771 acb
->bh
= qemu_bh_new(bdrv_co_rw_bh
, acb
);
2772 qemu_bh_schedule(acb
->bh
);
2775 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2779 BlockDriverCompletionFunc
*cb
,
2784 BlockDriverAIOCBCoroutine
*acb
;
2786 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
2787 acb
->req
.sector
= sector_num
;
2788 acb
->req
.nb_sectors
= nb_sectors
;
2789 acb
->req
.qiov
= qiov
;
2790 acb
->is_write
= is_write
;
2792 co
= qemu_coroutine_create(bdrv_co_rw
);
2793 qemu_coroutine_enter(co
, acb
);
2795 return &acb
->common
;
2798 static BlockDriverAIOCB
*bdrv_co_aio_readv_em(BlockDriverState
*bs
,
2799 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2800 BlockDriverCompletionFunc
*cb
, void *opaque
)
2802 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
,
2806 static BlockDriverAIOCB
*bdrv_co_aio_writev_em(BlockDriverState
*bs
,
2807 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2808 BlockDriverCompletionFunc
*cb
, void *opaque
)
2810 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
,
2814 static BlockDriverAIOCB
*bdrv_aio_flush_em(BlockDriverState
*bs
,
2815 BlockDriverCompletionFunc
*cb
, void *opaque
)
2817 BlockDriverAIOCBSync
*acb
;
2819 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2820 acb
->is_write
= 1; /* don't bounce in the completion hadler */
2826 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2829 qemu_bh_schedule(acb
->bh
);
2830 return &acb
->common
;
2833 static BlockDriverAIOCB
*bdrv_aio_noop_em(BlockDriverState
*bs
,
2834 BlockDriverCompletionFunc
*cb
, void *opaque
)
2836 BlockDriverAIOCBSync
*acb
;
2838 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2839 acb
->is_write
= 1; /* don't bounce in the completion handler */
2845 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2848 qemu_bh_schedule(acb
->bh
);
2849 return &acb
->common
;
2852 /**************************************************************/
2853 /* sync block device emulation */
2855 static void bdrv_rw_em_cb(void *opaque
, int ret
)
2857 *(int *)opaque
= ret
;
2860 #define NOT_DONE 0x7fffffff
2862 static int bdrv_read_em(BlockDriverState
*bs
, int64_t sector_num
,
2863 uint8_t *buf
, int nb_sectors
)
2866 BlockDriverAIOCB
*acb
;
2870 async_ret
= NOT_DONE
;
2871 iov
.iov_base
= (void *)buf
;
2872 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2873 qemu_iovec_init_external(&qiov
, &iov
, 1);
2874 acb
= bdrv_aio_readv(bs
, sector_num
, &qiov
, nb_sectors
,
2875 bdrv_rw_em_cb
, &async_ret
);
2881 while (async_ret
== NOT_DONE
) {
2890 static int bdrv_write_em(BlockDriverState
*bs
, int64_t sector_num
,
2891 const uint8_t *buf
, int nb_sectors
)
2894 BlockDriverAIOCB
*acb
;
2898 async_ret
= NOT_DONE
;
2899 iov
.iov_base
= (void *)buf
;
2900 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2901 qemu_iovec_init_external(&qiov
, &iov
, 1);
2902 acb
= bdrv_aio_writev(bs
, sector_num
, &qiov
, nb_sectors
,
2903 bdrv_rw_em_cb
, &async_ret
);
2908 while (async_ret
== NOT_DONE
) {
2916 void bdrv_init(void)
2918 module_call_init(MODULE_INIT_BLOCK
);
2921 void bdrv_init_with_whitelist(void)
2923 use_bdrv_whitelist
= 1;
2927 void *qemu_aio_get(AIOPool
*pool
, BlockDriverState
*bs
,
2928 BlockDriverCompletionFunc
*cb
, void *opaque
)
2930 BlockDriverAIOCB
*acb
;
2932 if (pool
->free_aiocb
) {
2933 acb
= pool
->free_aiocb
;
2934 pool
->free_aiocb
= acb
->next
;
2936 acb
= g_malloc0(pool
->aiocb_size
);
2941 acb
->opaque
= opaque
;
2945 void qemu_aio_release(void *p
)
2947 BlockDriverAIOCB
*acb
= (BlockDriverAIOCB
*)p
;
2948 AIOPool
*pool
= acb
->pool
;
2949 acb
->next
= pool
->free_aiocb
;
2950 pool
->free_aiocb
= acb
;
2953 /**************************************************************/
2954 /* Coroutine block device emulation */
2956 typedef struct CoroutineIOCompletion
{
2957 Coroutine
*coroutine
;
2959 } CoroutineIOCompletion
;
2961 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
2963 CoroutineIOCompletion
*co
= opaque
;
2966 qemu_coroutine_enter(co
->coroutine
, NULL
);
2969 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
2970 int nb_sectors
, QEMUIOVector
*iov
,
2973 CoroutineIOCompletion co
= {
2974 .coroutine
= qemu_coroutine_self(),
2976 BlockDriverAIOCB
*acb
;
2979 acb
= bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
2980 bdrv_co_io_em_complete
, &co
);
2982 acb
= bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
2983 bdrv_co_io_em_complete
, &co
);
2986 trace_bdrv_co_io(is_write
, acb
);
2990 qemu_coroutine_yield();
2995 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
2996 int64_t sector_num
, int nb_sectors
,
2999 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
3002 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
3003 int64_t sector_num
, int nb_sectors
,
3006 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
3009 static int coroutine_fn
bdrv_co_flush_em(BlockDriverState
*bs
)
3011 CoroutineIOCompletion co
= {
3012 .coroutine
= qemu_coroutine_self(),
3014 BlockDriverAIOCB
*acb
;
3016 acb
= bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
3020 qemu_coroutine_yield();
3024 /**************************************************************/
3025 /* removable device support */
3028 * Return TRUE if the media is present
3030 int bdrv_is_inserted(BlockDriverState
*bs
)
3032 BlockDriver
*drv
= bs
->drv
;
3036 if (!drv
->bdrv_is_inserted
)
3037 return !bs
->tray_open
;
3038 ret
= drv
->bdrv_is_inserted(bs
);
3043 * Return TRUE if the media changed since the last call to this
3044 * function. It is currently only used for floppy disks
3046 int bdrv_media_changed(BlockDriverState
*bs
)
3048 BlockDriver
*drv
= bs
->drv
;
3051 if (!drv
|| !drv
->bdrv_media_changed
)
3054 ret
= drv
->bdrv_media_changed(bs
);
3055 if (ret
== -ENOTSUP
)
3056 ret
= bs
->media_changed
;
3057 bs
->media_changed
= 0;
3062 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3064 int bdrv_eject(BlockDriverState
*bs
, int eject_flag
)
3066 BlockDriver
*drv
= bs
->drv
;
3068 if (eject_flag
&& bs
->locked
) {
3072 if (drv
&& drv
->bdrv_eject
) {
3073 drv
->bdrv_eject(bs
, eject_flag
);
3075 bs
->tray_open
= eject_flag
;
3079 int bdrv_is_locked(BlockDriverState
*bs
)
3085 * Lock or unlock the media (if it is locked, the user won't be able
3086 * to eject it manually).
3088 void bdrv_set_locked(BlockDriverState
*bs
, int locked
)
3090 BlockDriver
*drv
= bs
->drv
;
3092 trace_bdrv_set_locked(bs
, locked
);
3094 bs
->locked
= locked
;
3095 if (drv
&& drv
->bdrv_set_locked
) {
3096 drv
->bdrv_set_locked(bs
, locked
);
3100 /* needed for generic scsi interface */
3102 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3104 BlockDriver
*drv
= bs
->drv
;
3106 if (drv
&& drv
->bdrv_ioctl
)
3107 return drv
->bdrv_ioctl(bs
, req
, buf
);
3111 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
3112 unsigned long int req
, void *buf
,
3113 BlockDriverCompletionFunc
*cb
, void *opaque
)
3115 BlockDriver
*drv
= bs
->drv
;
3117 if (drv
&& drv
->bdrv_aio_ioctl
)
3118 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
3124 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
3126 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
3129 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int enable
)
3131 int64_t bitmap_size
;
3133 bs
->dirty_count
= 0;
3135 if (!bs
->dirty_bitmap
) {
3136 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
) +
3137 BDRV_SECTORS_PER_DIRTY_CHUNK
* 8 - 1;
3138 bitmap_size
/= BDRV_SECTORS_PER_DIRTY_CHUNK
* 8;
3140 bs
->dirty_bitmap
= g_malloc0(bitmap_size
);
3143 if (bs
->dirty_bitmap
) {
3144 g_free(bs
->dirty_bitmap
);
3145 bs
->dirty_bitmap
= NULL
;
3150 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
3152 int64_t chunk
= sector
/ (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK
;
3154 if (bs
->dirty_bitmap
&&
3155 (sector
<< BDRV_SECTOR_BITS
) < bdrv_getlength(bs
)) {
3156 return !!(bs
->dirty_bitmap
[chunk
/ (sizeof(unsigned long) * 8)] &
3157 (1UL << (chunk
% (sizeof(unsigned long) * 8))));
3163 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
3166 set_dirty_bitmap(bs
, cur_sector
, nr_sectors
, 0);
3169 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
3171 return bs
->dirty_count
;
3174 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
3176 assert(bs
->in_use
!= in_use
);
3177 bs
->in_use
= in_use
;
3180 int bdrv_in_use(BlockDriverState
*bs
)
3186 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
3187 enum BlockAcctType type
)
3189 assert(type
< BDRV_MAX_IOTYPE
);
3191 cookie
->bytes
= bytes
;
3192 cookie
->start_time_ns
= get_clock();
3193 cookie
->type
= type
;
3197 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
3199 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
3201 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
3202 bs
->nr_ops
[cookie
->type
]++;
3203 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
3206 int bdrv_img_create(const char *filename
, const char *fmt
,
3207 const char *base_filename
, const char *base_fmt
,
3208 char *options
, uint64_t img_size
, int flags
)
3210 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
3211 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
3212 BlockDriverState
*bs
= NULL
;
3213 BlockDriver
*drv
, *proto_drv
;
3214 BlockDriver
*backing_drv
= NULL
;
3217 /* Find driver and parse its options */
3218 drv
= bdrv_find_format(fmt
);
3220 error_report("Unknown file format '%s'", fmt
);
3225 proto_drv
= bdrv_find_protocol(filename
);
3227 error_report("Unknown protocol '%s'", filename
);
3232 create_options
= append_option_parameters(create_options
,
3233 drv
->create_options
);
3234 create_options
= append_option_parameters(create_options
,
3235 proto_drv
->create_options
);
3237 /* Create parameter list with default values */
3238 param
= parse_option_parameters("", create_options
, param
);
3240 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
3242 /* Parse -o options */
3244 param
= parse_option_parameters(options
, create_options
, param
);
3245 if (param
== NULL
) {
3246 error_report("Invalid options for file format '%s'.", fmt
);
3252 if (base_filename
) {
3253 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
3255 error_report("Backing file not supported for file format '%s'",
3263 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
3264 error_report("Backing file format not supported for file "
3265 "format '%s'", fmt
);
3271 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
3272 if (backing_file
&& backing_file
->value
.s
) {
3273 if (!strcmp(filename
, backing_file
->value
.s
)) {
3274 error_report("Error: Trying to create an image with the "
3275 "same filename as the backing file");
3281 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
3282 if (backing_fmt
&& backing_fmt
->value
.s
) {
3283 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
3285 error_report("Unknown backing file format '%s'",
3286 backing_fmt
->value
.s
);
3292 // The size for the image must always be specified, with one exception:
3293 // If we are using a backing file, we can obtain the size from there
3294 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
3295 if (size
&& size
->value
.n
== -1) {
3296 if (backing_file
&& backing_file
->value
.s
) {
3302 ret
= bdrv_open(bs
, backing_file
->value
.s
, flags
, backing_drv
);
3304 error_report("Could not open '%s'", backing_file
->value
.s
);
3307 bdrv_get_geometry(bs
, &size
);
3310 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
3311 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
3313 error_report("Image creation needs a size parameter");
3319 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
3320 print_option_parameters(param
);
3323 ret
= bdrv_create(drv
, filename
, param
);
3326 if (ret
== -ENOTSUP
) {
3327 error_report("Formatting or formatting option not supported for "
3328 "file format '%s'", fmt
);
3329 } else if (ret
== -EFBIG
) {
3330 error_report("The image size is too large for file format '%s'",
3333 error_report("%s: error while creating %s: %s", filename
, fmt
,
3339 free_option_parameters(create_options
);
3340 free_option_parameters(param
);