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 bdrv_dev_change_media_cb(bs
);
698 if (bs
->is_temporary
) {
704 void bdrv_close(BlockDriverState
*bs
)
707 if (bs
== bs_snapshots
) {
710 if (bs
->backing_hd
) {
711 bdrv_delete(bs
->backing_hd
);
712 bs
->backing_hd
= NULL
;
714 bs
->drv
->bdrv_close(bs
);
717 if (bs
->is_temporary
) {
718 unlink(bs
->filename
);
724 if (bs
->file
!= NULL
) {
725 bdrv_close(bs
->file
);
728 bdrv_dev_change_media_cb(bs
);
732 void bdrv_close_all(void)
734 BlockDriverState
*bs
;
736 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
741 /* make a BlockDriverState anonymous by removing from bdrv_state list.
742 Also, NULL terminate the device_name to prevent double remove */
743 void bdrv_make_anon(BlockDriverState
*bs
)
745 if (bs
->device_name
[0] != '\0') {
746 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
748 bs
->device_name
[0] = '\0';
751 void bdrv_delete(BlockDriverState
*bs
)
755 /* remove from list, if necessary */
759 if (bs
->file
!= NULL
) {
760 bdrv_delete(bs
->file
);
763 assert(bs
!= bs_snapshots
);
767 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
768 /* TODO change to DeviceState *dev when all users are qdevified */
777 /* TODO qdevified devices don't use this, remove when devices are qdevified */
778 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
780 if (bdrv_attach_dev(bs
, dev
) < 0) {
785 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
786 /* TODO change to DeviceState *dev when all users are qdevified */
788 assert(bs
->dev
== dev
);
791 bs
->dev_opaque
= NULL
;
794 /* TODO change to return DeviceState * when all users are qdevified */
795 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
800 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
804 bs
->dev_opaque
= opaque
;
807 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
)
809 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
810 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
);
814 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
816 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
817 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
821 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
823 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
824 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
830 * Run consistency checks on an image
832 * Returns 0 if the check could be completed (it doesn't mean that the image is
833 * free of errors) or -errno when an internal error occurred. The results of the
834 * check are stored in res.
836 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
)
838 if (bs
->drv
->bdrv_check
== NULL
) {
842 memset(res
, 0, sizeof(*res
));
843 return bs
->drv
->bdrv_check(bs
, res
);
846 #define COMMIT_BUF_SECTORS 2048
848 /* commit COW file into the raw image */
849 int bdrv_commit(BlockDriverState
*bs
)
851 BlockDriver
*drv
= bs
->drv
;
852 BlockDriver
*backing_drv
;
853 int64_t sector
, total_sectors
;
854 int n
, ro
, open_flags
;
855 int ret
= 0, rw_ret
= 0;
858 BlockDriverState
*bs_rw
, *bs_ro
;
863 if (!bs
->backing_hd
) {
867 if (bs
->backing_hd
->keep_read_only
) {
871 backing_drv
= bs
->backing_hd
->drv
;
872 ro
= bs
->backing_hd
->read_only
;
873 strncpy(filename
, bs
->backing_hd
->filename
, sizeof(filename
));
874 open_flags
= bs
->backing_hd
->open_flags
;
878 bdrv_delete(bs
->backing_hd
);
879 bs
->backing_hd
= NULL
;
880 bs_rw
= bdrv_new("");
881 rw_ret
= bdrv_open(bs_rw
, filename
, open_flags
| BDRV_O_RDWR
,
885 /* try to re-open read-only */
886 bs_ro
= bdrv_new("");
887 ret
= bdrv_open(bs_ro
, filename
, open_flags
& ~BDRV_O_RDWR
,
891 /* drive not functional anymore */
895 bs
->backing_hd
= bs_ro
;
898 bs
->backing_hd
= bs_rw
;
901 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
902 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
904 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
905 if (drv
->bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
907 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
912 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
919 if (drv
->bdrv_make_empty
) {
920 ret
= drv
->bdrv_make_empty(bs
);
925 * Make sure all data we wrote to the backing device is actually
929 bdrv_flush(bs
->backing_hd
);
936 bdrv_delete(bs
->backing_hd
);
937 bs
->backing_hd
= NULL
;
938 bs_ro
= bdrv_new("");
939 ret
= bdrv_open(bs_ro
, filename
, open_flags
& ~BDRV_O_RDWR
,
943 /* drive not functional anymore */
947 bs
->backing_hd
= bs_ro
;
948 bs
->backing_hd
->keep_read_only
= 0;
954 void bdrv_commit_all(void)
956 BlockDriverState
*bs
;
958 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
966 * -EINVAL - backing format specified, but no file
967 * -ENOSPC - can't update the backing file because no space is left in the
969 * -ENOTSUP - format driver doesn't support changing the backing file
971 int bdrv_change_backing_file(BlockDriverState
*bs
,
972 const char *backing_file
, const char *backing_fmt
)
974 BlockDriver
*drv
= bs
->drv
;
976 if (drv
->bdrv_change_backing_file
!= NULL
) {
977 return drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
983 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
988 if (!bdrv_is_inserted(bs
))
994 len
= bdrv_getlength(bs
);
999 if ((offset
> len
) || (len
- offset
< size
))
1005 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
1008 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
1009 nb_sectors
* BDRV_SECTOR_SIZE
);
1012 static inline bool bdrv_has_async_rw(BlockDriver
*drv
)
1014 return drv
->bdrv_co_readv
!= bdrv_co_readv_em
1015 || drv
->bdrv_aio_readv
!= bdrv_aio_readv_em
;
1018 static inline bool bdrv_has_async_flush(BlockDriver
*drv
)
1020 return drv
->bdrv_aio_flush
!= bdrv_aio_flush_em
;
1023 /* return < 0 if error. See bdrv_write() for the return codes */
1024 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
1025 uint8_t *buf
, int nb_sectors
)
1027 BlockDriver
*drv
= bs
->drv
;
1032 if (bdrv_has_async_rw(drv
) && qemu_in_coroutine()) {
1034 struct iovec iov
= {
1035 .iov_base
= (void *)buf
,
1036 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1039 qemu_iovec_init_external(&qiov
, &iov
, 1);
1040 return bdrv_co_readv(bs
, sector_num
, nb_sectors
, &qiov
);
1043 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1046 return drv
->bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
1049 static void set_dirty_bitmap(BlockDriverState
*bs
, int64_t sector_num
,
1050 int nb_sectors
, int dirty
)
1053 unsigned long val
, idx
, bit
;
1055 start
= sector_num
/ BDRV_SECTORS_PER_DIRTY_CHUNK
;
1056 end
= (sector_num
+ nb_sectors
- 1) / BDRV_SECTORS_PER_DIRTY_CHUNK
;
1058 for (; start
<= end
; start
++) {
1059 idx
= start
/ (sizeof(unsigned long) * 8);
1060 bit
= start
% (sizeof(unsigned long) * 8);
1061 val
= bs
->dirty_bitmap
[idx
];
1063 if (!(val
& (1UL << bit
))) {
1068 if (val
& (1UL << bit
)) {
1070 val
&= ~(1UL << bit
);
1073 bs
->dirty_bitmap
[idx
] = val
;
1077 /* Return < 0 if error. Important errors are:
1078 -EIO generic I/O error (may happen for all errors)
1079 -ENOMEDIUM No media inserted.
1080 -EINVAL Invalid sector number or nb_sectors
1081 -EACCES Trying to write a read-only device
1083 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
1084 const uint8_t *buf
, int nb_sectors
)
1086 BlockDriver
*drv
= bs
->drv
;
1091 if (bdrv_has_async_rw(drv
) && qemu_in_coroutine()) {
1093 struct iovec iov
= {
1094 .iov_base
= (void *)buf
,
1095 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1098 qemu_iovec_init_external(&qiov
, &iov
, 1);
1099 return bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
1104 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
1107 if (bs
->dirty_bitmap
) {
1108 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1111 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
1112 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
1115 return drv
->bdrv_write(bs
, sector_num
, buf
, nb_sectors
);
1118 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
1119 void *buf
, int count1
)
1121 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
1122 int len
, nb_sectors
, count
;
1127 /* first read to align to sector start */
1128 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
1131 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1133 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1135 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
1143 /* read the sectors "in place" */
1144 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
1145 if (nb_sectors
> 0) {
1146 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
1148 sector_num
+= nb_sectors
;
1149 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
1154 /* add data from the last sector */
1156 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1158 memcpy(buf
, tmp_buf
, count
);
1163 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
1164 const void *buf
, int count1
)
1166 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
1167 int len
, nb_sectors
, count
;
1172 /* first write to align to sector start */
1173 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
1176 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1178 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1180 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
1181 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
1190 /* write the sectors "in place" */
1191 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
1192 if (nb_sectors
> 0) {
1193 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
1195 sector_num
+= nb_sectors
;
1196 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
1201 /* add data from the last sector */
1203 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
1205 memcpy(tmp_buf
, buf
, count
);
1206 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
1213 * Writes to the file and ensures that no writes are reordered across this
1214 * request (acts as a barrier)
1216 * Returns 0 on success, -errno in error cases.
1218 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
1219 const void *buf
, int count
)
1223 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
1228 /* No flush needed for cache modes that use O_DSYNC */
1229 if ((bs
->open_flags
& BDRV_O_CACHE_WB
) != 0) {
1236 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1237 int nb_sectors
, QEMUIOVector
*qiov
)
1239 BlockDriver
*drv
= bs
->drv
;
1241 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1246 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
1250 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
1253 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1254 int nb_sectors
, QEMUIOVector
*qiov
)
1256 BlockDriver
*drv
= bs
->drv
;
1258 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1263 if (bs
->read_only
) {
1266 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
1270 if (bs
->dirty_bitmap
) {
1271 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
1274 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
1275 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
1278 return drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
1282 * Truncate file to 'offset' bytes (needed only for file protocols)
1284 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
1286 BlockDriver
*drv
= bs
->drv
;
1290 if (!drv
->bdrv_truncate
)
1294 if (bdrv_in_use(bs
))
1296 ret
= drv
->bdrv_truncate(bs
, offset
);
1298 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
1299 bdrv_dev_resize_cb(bs
);
1305 * Length of a allocated file in bytes. Sparse files are counted by actual
1306 * allocated space. Return < 0 if error or unknown.
1308 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
1310 BlockDriver
*drv
= bs
->drv
;
1314 if (drv
->bdrv_get_allocated_file_size
) {
1315 return drv
->bdrv_get_allocated_file_size(bs
);
1318 return bdrv_get_allocated_file_size(bs
->file
);
1324 * Length of a file in bytes. Return < 0 if error or unknown.
1326 int64_t bdrv_getlength(BlockDriverState
*bs
)
1328 BlockDriver
*drv
= bs
->drv
;
1332 if (bs
->growable
|| bs
->removable
) {
1333 if (drv
->bdrv_getlength
) {
1334 return drv
->bdrv_getlength(bs
);
1337 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
1340 /* return 0 as number of sectors if no device present or error */
1341 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
1344 length
= bdrv_getlength(bs
);
1348 length
= length
>> BDRV_SECTOR_BITS
;
1349 *nb_sectors_ptr
= length
;
1353 uint8_t boot_ind
; /* 0x80 - active */
1354 uint8_t head
; /* starting head */
1355 uint8_t sector
; /* starting sector */
1356 uint8_t cyl
; /* starting cylinder */
1357 uint8_t sys_ind
; /* What partition type */
1358 uint8_t end_head
; /* end head */
1359 uint8_t end_sector
; /* end sector */
1360 uint8_t end_cyl
; /* end cylinder */
1361 uint32_t start_sect
; /* starting sector counting from 0 */
1362 uint32_t nr_sects
; /* nr of sectors in partition */
1365 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1366 static int guess_disk_lchs(BlockDriverState
*bs
,
1367 int *pcylinders
, int *pheads
, int *psectors
)
1369 uint8_t buf
[BDRV_SECTOR_SIZE
];
1370 int ret
, i
, heads
, sectors
, cylinders
;
1371 struct partition
*p
;
1373 uint64_t nb_sectors
;
1375 bdrv_get_geometry(bs
, &nb_sectors
);
1377 ret
= bdrv_read(bs
, 0, buf
, 1);
1380 /* test msdos magic */
1381 if (buf
[510] != 0x55 || buf
[511] != 0xaa)
1383 for(i
= 0; i
< 4; i
++) {
1384 p
= ((struct partition
*)(buf
+ 0x1be)) + i
;
1385 nr_sects
= le32_to_cpu(p
->nr_sects
);
1386 if (nr_sects
&& p
->end_head
) {
1387 /* We make the assumption that the partition terminates on
1388 a cylinder boundary */
1389 heads
= p
->end_head
+ 1;
1390 sectors
= p
->end_sector
& 63;
1393 cylinders
= nb_sectors
/ (heads
* sectors
);
1394 if (cylinders
< 1 || cylinders
> 16383)
1397 *psectors
= sectors
;
1398 *pcylinders
= cylinders
;
1400 printf("guessed geometry: LCHS=%d %d %d\n",
1401 cylinders
, heads
, sectors
);
1409 void bdrv_guess_geometry(BlockDriverState
*bs
, int *pcyls
, int *pheads
, int *psecs
)
1411 int translation
, lba_detected
= 0;
1412 int cylinders
, heads
, secs
;
1413 uint64_t nb_sectors
;
1415 /* if a geometry hint is available, use it */
1416 bdrv_get_geometry(bs
, &nb_sectors
);
1417 bdrv_get_geometry_hint(bs
, &cylinders
, &heads
, &secs
);
1418 translation
= bdrv_get_translation_hint(bs
);
1419 if (cylinders
!= 0) {
1424 if (guess_disk_lchs(bs
, &cylinders
, &heads
, &secs
) == 0) {
1426 /* if heads > 16, it means that a BIOS LBA
1427 translation was active, so the default
1428 hardware geometry is OK */
1430 goto default_geometry
;
1435 /* disable any translation to be in sync with
1436 the logical geometry */
1437 if (translation
== BIOS_ATA_TRANSLATION_AUTO
) {
1438 bdrv_set_translation_hint(bs
,
1439 BIOS_ATA_TRANSLATION_NONE
);
1444 /* if no geometry, use a standard physical disk geometry */
1445 cylinders
= nb_sectors
/ (16 * 63);
1447 if (cylinders
> 16383)
1449 else if (cylinders
< 2)
1454 if ((lba_detected
== 1) && (translation
== BIOS_ATA_TRANSLATION_AUTO
)) {
1455 if ((*pcyls
* *pheads
) <= 131072) {
1456 bdrv_set_translation_hint(bs
,
1457 BIOS_ATA_TRANSLATION_LARGE
);
1459 bdrv_set_translation_hint(bs
,
1460 BIOS_ATA_TRANSLATION_LBA
);
1464 bdrv_set_geometry_hint(bs
, *pcyls
, *pheads
, *psecs
);
1468 void bdrv_set_geometry_hint(BlockDriverState
*bs
,
1469 int cyls
, int heads
, int secs
)
1476 void bdrv_set_translation_hint(BlockDriverState
*bs
, int translation
)
1478 bs
->translation
= translation
;
1481 void bdrv_get_geometry_hint(BlockDriverState
*bs
,
1482 int *pcyls
, int *pheads
, int *psecs
)
1485 *pheads
= bs
->heads
;
1489 /* Recognize floppy formats */
1490 typedef struct FDFormat
{
1497 static const FDFormat fd_formats
[] = {
1498 /* First entry is default format */
1499 /* 1.44 MB 3"1/2 floppy disks */
1500 { FDRIVE_DRV_144
, 18, 80, 1, },
1501 { FDRIVE_DRV_144
, 20, 80, 1, },
1502 { FDRIVE_DRV_144
, 21, 80, 1, },
1503 { FDRIVE_DRV_144
, 21, 82, 1, },
1504 { FDRIVE_DRV_144
, 21, 83, 1, },
1505 { FDRIVE_DRV_144
, 22, 80, 1, },
1506 { FDRIVE_DRV_144
, 23, 80, 1, },
1507 { FDRIVE_DRV_144
, 24, 80, 1, },
1508 /* 2.88 MB 3"1/2 floppy disks */
1509 { FDRIVE_DRV_288
, 36, 80, 1, },
1510 { FDRIVE_DRV_288
, 39, 80, 1, },
1511 { FDRIVE_DRV_288
, 40, 80, 1, },
1512 { FDRIVE_DRV_288
, 44, 80, 1, },
1513 { FDRIVE_DRV_288
, 48, 80, 1, },
1514 /* 720 kB 3"1/2 floppy disks */
1515 { FDRIVE_DRV_144
, 9, 80, 1, },
1516 { FDRIVE_DRV_144
, 10, 80, 1, },
1517 { FDRIVE_DRV_144
, 10, 82, 1, },
1518 { FDRIVE_DRV_144
, 10, 83, 1, },
1519 { FDRIVE_DRV_144
, 13, 80, 1, },
1520 { FDRIVE_DRV_144
, 14, 80, 1, },
1521 /* 1.2 MB 5"1/4 floppy disks */
1522 { FDRIVE_DRV_120
, 15, 80, 1, },
1523 { FDRIVE_DRV_120
, 18, 80, 1, },
1524 { FDRIVE_DRV_120
, 18, 82, 1, },
1525 { FDRIVE_DRV_120
, 18, 83, 1, },
1526 { FDRIVE_DRV_120
, 20, 80, 1, },
1527 /* 720 kB 5"1/4 floppy disks */
1528 { FDRIVE_DRV_120
, 9, 80, 1, },
1529 { FDRIVE_DRV_120
, 11, 80, 1, },
1530 /* 360 kB 5"1/4 floppy disks */
1531 { FDRIVE_DRV_120
, 9, 40, 1, },
1532 { FDRIVE_DRV_120
, 9, 40, 0, },
1533 { FDRIVE_DRV_120
, 10, 41, 1, },
1534 { FDRIVE_DRV_120
, 10, 42, 1, },
1535 /* 320 kB 5"1/4 floppy disks */
1536 { FDRIVE_DRV_120
, 8, 40, 1, },
1537 { FDRIVE_DRV_120
, 8, 40, 0, },
1538 /* 360 kB must match 5"1/4 better than 3"1/2... */
1539 { FDRIVE_DRV_144
, 9, 80, 0, },
1541 { FDRIVE_DRV_NONE
, -1, -1, 0, },
1544 void bdrv_get_floppy_geometry_hint(BlockDriverState
*bs
, int *nb_heads
,
1545 int *max_track
, int *last_sect
,
1546 FDriveType drive_in
, FDriveType
*drive
)
1548 const FDFormat
*parse
;
1549 uint64_t nb_sectors
, size
;
1550 int i
, first_match
, match
;
1552 bdrv_get_geometry_hint(bs
, nb_heads
, max_track
, last_sect
);
1553 if (*nb_heads
!= 0 && *max_track
!= 0 && *last_sect
!= 0) {
1554 /* User defined disk */
1556 bdrv_get_geometry(bs
, &nb_sectors
);
1559 for (i
= 0; ; i
++) {
1560 parse
= &fd_formats
[i
];
1561 if (parse
->drive
== FDRIVE_DRV_NONE
) {
1564 if (drive_in
== parse
->drive
||
1565 drive_in
== FDRIVE_DRV_NONE
) {
1566 size
= (parse
->max_head
+ 1) * parse
->max_track
*
1568 if (nb_sectors
== size
) {
1572 if (first_match
== -1) {
1578 if (first_match
== -1) {
1581 match
= first_match
;
1583 parse
= &fd_formats
[match
];
1585 *nb_heads
= parse
->max_head
+ 1;
1586 *max_track
= parse
->max_track
;
1587 *last_sect
= parse
->last_sect
;
1588 *drive
= parse
->drive
;
1592 int bdrv_get_translation_hint(BlockDriverState
*bs
)
1594 return bs
->translation
;
1597 void bdrv_set_on_error(BlockDriverState
*bs
, BlockErrorAction on_read_error
,
1598 BlockErrorAction on_write_error
)
1600 bs
->on_read_error
= on_read_error
;
1601 bs
->on_write_error
= on_write_error
;
1604 BlockErrorAction
bdrv_get_on_error(BlockDriverState
*bs
, int is_read
)
1606 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
1609 void bdrv_set_removable(BlockDriverState
*bs
, int removable
)
1611 bs
->removable
= removable
;
1612 if (removable
&& bs
== bs_snapshots
) {
1613 bs_snapshots
= NULL
;
1617 int bdrv_is_removable(BlockDriverState
*bs
)
1619 return bs
->removable
;
1622 int bdrv_is_read_only(BlockDriverState
*bs
)
1624 return bs
->read_only
;
1627 int bdrv_is_sg(BlockDriverState
*bs
)
1632 int bdrv_enable_write_cache(BlockDriverState
*bs
)
1634 return bs
->enable_write_cache
;
1637 int bdrv_is_encrypted(BlockDriverState
*bs
)
1639 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
1641 return bs
->encrypted
;
1644 int bdrv_key_required(BlockDriverState
*bs
)
1646 BlockDriverState
*backing_hd
= bs
->backing_hd
;
1648 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
1650 return (bs
->encrypted
&& !bs
->valid_key
);
1653 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
1656 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
1657 ret
= bdrv_set_key(bs
->backing_hd
, key
);
1663 if (!bs
->encrypted
) {
1665 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
1668 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
1671 } else if (!bs
->valid_key
) {
1673 /* call the change callback now, we skipped it on open */
1674 bdrv_dev_change_media_cb(bs
);
1679 void bdrv_get_format(BlockDriverState
*bs
, char *buf
, int buf_size
)
1684 pstrcpy(buf
, buf_size
, bs
->drv
->format_name
);
1688 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
1693 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
1694 it(opaque
, drv
->format_name
);
1698 BlockDriverState
*bdrv_find(const char *name
)
1700 BlockDriverState
*bs
;
1702 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1703 if (!strcmp(name
, bs
->device_name
)) {
1710 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
1713 return QTAILQ_FIRST(&bdrv_states
);
1715 return QTAILQ_NEXT(bs
, list
);
1718 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
1720 BlockDriverState
*bs
;
1722 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1727 const char *bdrv_get_device_name(BlockDriverState
*bs
)
1729 return bs
->device_name
;
1732 int bdrv_flush(BlockDriverState
*bs
)
1734 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
1738 if (bs
->drv
&& bdrv_has_async_flush(bs
->drv
) && qemu_in_coroutine()) {
1739 return bdrv_co_flush_em(bs
);
1742 if (bs
->drv
&& bs
->drv
->bdrv_flush
) {
1743 return bs
->drv
->bdrv_flush(bs
);
1747 * Some block drivers always operate in either writethrough or unsafe mode
1748 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1749 * the server works (because the behaviour is hardcoded or depends on
1750 * server-side configuration), so we can't ensure that everything is safe
1751 * on disk. Returning an error doesn't work because that would break guests
1752 * even if the server operates in writethrough mode.
1754 * Let's hope the user knows what he's doing.
1759 void bdrv_flush_all(void)
1761 BlockDriverState
*bs
;
1763 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1764 if (!bdrv_is_read_only(bs
) && bdrv_is_inserted(bs
)) {
1770 int bdrv_has_zero_init(BlockDriverState
*bs
)
1774 if (bs
->drv
->bdrv_has_zero_init
) {
1775 return bs
->drv
->bdrv_has_zero_init(bs
);
1781 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
1786 if (!bs
->drv
->bdrv_discard
) {
1789 return bs
->drv
->bdrv_discard(bs
, sector_num
, nb_sectors
);
1793 * Returns true iff the specified sector is present in the disk image. Drivers
1794 * not implementing the functionality are assumed to not support backing files,
1795 * hence all their sectors are reported as allocated.
1797 * 'pnum' is set to the number of sectors (including and immediately following
1798 * the specified sector) that are known to be in the same
1799 * allocated/unallocated state.
1801 * 'nb_sectors' is the max value 'pnum' should be set to.
1803 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
1807 if (!bs
->drv
->bdrv_is_allocated
) {
1808 if (sector_num
>= bs
->total_sectors
) {
1812 n
= bs
->total_sectors
- sector_num
;
1813 *pnum
= (n
< nb_sectors
) ? (n
) : (nb_sectors
);
1816 return bs
->drv
->bdrv_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
1819 void bdrv_mon_event(const BlockDriverState
*bdrv
,
1820 BlockMonEventAction action
, int is_read
)
1823 const char *action_str
;
1826 case BDRV_ACTION_REPORT
:
1827 action_str
= "report";
1829 case BDRV_ACTION_IGNORE
:
1830 action_str
= "ignore";
1832 case BDRV_ACTION_STOP
:
1833 action_str
= "stop";
1839 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1842 is_read
? "read" : "write");
1843 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR
, data
);
1845 qobject_decref(data
);
1848 static void bdrv_print_dict(QObject
*obj
, void *opaque
)
1851 Monitor
*mon
= opaque
;
1853 bs_dict
= qobject_to_qdict(obj
);
1855 monitor_printf(mon
, "%s: removable=%d",
1856 qdict_get_str(bs_dict
, "device"),
1857 qdict_get_bool(bs_dict
, "removable"));
1859 if (qdict_get_bool(bs_dict
, "removable")) {
1860 monitor_printf(mon
, " locked=%d", qdict_get_bool(bs_dict
, "locked"));
1863 if (qdict_haskey(bs_dict
, "inserted")) {
1864 QDict
*qdict
= qobject_to_qdict(qdict_get(bs_dict
, "inserted"));
1866 monitor_printf(mon
, " file=");
1867 monitor_print_filename(mon
, qdict_get_str(qdict
, "file"));
1868 if (qdict_haskey(qdict
, "backing_file")) {
1869 monitor_printf(mon
, " backing_file=");
1870 monitor_print_filename(mon
, qdict_get_str(qdict
, "backing_file"));
1872 monitor_printf(mon
, " ro=%d drv=%s encrypted=%d",
1873 qdict_get_bool(qdict
, "ro"),
1874 qdict_get_str(qdict
, "drv"),
1875 qdict_get_bool(qdict
, "encrypted"));
1877 monitor_printf(mon
, " [not inserted]");
1880 monitor_printf(mon
, "\n");
1883 void bdrv_info_print(Monitor
*mon
, const QObject
*data
)
1885 qlist_iter(qobject_to_qlist(data
), bdrv_print_dict
, mon
);
1888 void bdrv_info(Monitor
*mon
, QObject
**ret_data
)
1891 BlockDriverState
*bs
;
1893 bs_list
= qlist_new();
1895 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1898 bs_obj
= qobject_from_jsonf("{ 'device': %s, 'type': 'unknown', "
1899 "'removable': %i, 'locked': %i }",
1900 bs
->device_name
, bs
->removable
,
1901 bdrv_dev_is_medium_locked(bs
));
1905 QDict
*bs_dict
= qobject_to_qdict(bs_obj
);
1907 obj
= qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1908 "'encrypted': %i }",
1909 bs
->filename
, bs
->read_only
,
1910 bs
->drv
->format_name
,
1911 bdrv_is_encrypted(bs
));
1912 if (bs
->backing_file
[0] != '\0') {
1913 QDict
*qdict
= qobject_to_qdict(obj
);
1914 qdict_put(qdict
, "backing_file",
1915 qstring_from_str(bs
->backing_file
));
1918 qdict_put_obj(bs_dict
, "inserted", obj
);
1920 qlist_append_obj(bs_list
, bs_obj
);
1923 *ret_data
= QOBJECT(bs_list
);
1926 static void bdrv_stats_iter(QObject
*data
, void *opaque
)
1929 Monitor
*mon
= opaque
;
1931 qdict
= qobject_to_qdict(data
);
1932 monitor_printf(mon
, "%s:", qdict_get_str(qdict
, "device"));
1934 qdict
= qobject_to_qdict(qdict_get(qdict
, "stats"));
1935 monitor_printf(mon
, " rd_bytes=%" PRId64
1936 " wr_bytes=%" PRId64
1937 " rd_operations=%" PRId64
1938 " wr_operations=%" PRId64
1939 " flush_operations=%" PRId64
1940 " wr_total_time_ns=%" PRId64
1941 " rd_total_time_ns=%" PRId64
1942 " flush_total_time_ns=%" PRId64
1944 qdict_get_int(qdict
, "rd_bytes"),
1945 qdict_get_int(qdict
, "wr_bytes"),
1946 qdict_get_int(qdict
, "rd_operations"),
1947 qdict_get_int(qdict
, "wr_operations"),
1948 qdict_get_int(qdict
, "flush_operations"),
1949 qdict_get_int(qdict
, "wr_total_time_ns"),
1950 qdict_get_int(qdict
, "rd_total_time_ns"),
1951 qdict_get_int(qdict
, "flush_total_time_ns"));
1954 void bdrv_stats_print(Monitor
*mon
, const QObject
*data
)
1956 qlist_iter(qobject_to_qlist(data
), bdrv_stats_iter
, mon
);
1959 static QObject
* bdrv_info_stats_bs(BlockDriverState
*bs
)
1964 res
= qobject_from_jsonf("{ 'stats': {"
1965 "'rd_bytes': %" PRId64
","
1966 "'wr_bytes': %" PRId64
","
1967 "'rd_operations': %" PRId64
","
1968 "'wr_operations': %" PRId64
","
1969 "'wr_highest_offset': %" PRId64
","
1970 "'flush_operations': %" PRId64
","
1971 "'wr_total_time_ns': %" PRId64
","
1972 "'rd_total_time_ns': %" PRId64
","
1973 "'flush_total_time_ns': %" PRId64
1975 bs
->nr_bytes
[BDRV_ACCT_READ
],
1976 bs
->nr_bytes
[BDRV_ACCT_WRITE
],
1977 bs
->nr_ops
[BDRV_ACCT_READ
],
1978 bs
->nr_ops
[BDRV_ACCT_WRITE
],
1979 bs
->wr_highest_sector
*
1980 (uint64_t)BDRV_SECTOR_SIZE
,
1981 bs
->nr_ops
[BDRV_ACCT_FLUSH
],
1982 bs
->total_time_ns
[BDRV_ACCT_WRITE
],
1983 bs
->total_time_ns
[BDRV_ACCT_READ
],
1984 bs
->total_time_ns
[BDRV_ACCT_FLUSH
]);
1985 dict
= qobject_to_qdict(res
);
1987 if (*bs
->device_name
) {
1988 qdict_put(dict
, "device", qstring_from_str(bs
->device_name
));
1992 QObject
*parent
= bdrv_info_stats_bs(bs
->file
);
1993 qdict_put_obj(dict
, "parent", parent
);
1999 void bdrv_info_stats(Monitor
*mon
, QObject
**ret_data
)
2003 BlockDriverState
*bs
;
2005 devices
= qlist_new();
2007 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2008 obj
= bdrv_info_stats_bs(bs
);
2009 qlist_append_obj(devices
, obj
);
2012 *ret_data
= QOBJECT(devices
);
2015 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
2017 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2018 return bs
->backing_file
;
2019 else if (bs
->encrypted
)
2020 return bs
->filename
;
2025 void bdrv_get_backing_filename(BlockDriverState
*bs
,
2026 char *filename
, int filename_size
)
2028 if (!bs
->backing_file
) {
2029 pstrcpy(filename
, filename_size
, "");
2031 pstrcpy(filename
, filename_size
, bs
->backing_file
);
2035 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
2036 const uint8_t *buf
, int nb_sectors
)
2038 BlockDriver
*drv
= bs
->drv
;
2041 if (!drv
->bdrv_write_compressed
)
2043 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2046 if (bs
->dirty_bitmap
) {
2047 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
2050 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
2053 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2055 BlockDriver
*drv
= bs
->drv
;
2058 if (!drv
->bdrv_get_info
)
2060 memset(bdi
, 0, sizeof(*bdi
));
2061 return drv
->bdrv_get_info(bs
, bdi
);
2064 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2065 int64_t pos
, int size
)
2067 BlockDriver
*drv
= bs
->drv
;
2070 if (drv
->bdrv_save_vmstate
)
2071 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
2073 return bdrv_save_vmstate(bs
->file
, buf
, pos
, size
);
2077 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2078 int64_t pos
, int size
)
2080 BlockDriver
*drv
= bs
->drv
;
2083 if (drv
->bdrv_load_vmstate
)
2084 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
2086 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
2090 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
2092 BlockDriver
*drv
= bs
->drv
;
2094 if (!drv
|| !drv
->bdrv_debug_event
) {
2098 return drv
->bdrv_debug_event(bs
, event
);
2102 /**************************************************************/
2103 /* handling of snapshots */
2105 int bdrv_can_snapshot(BlockDriverState
*bs
)
2107 BlockDriver
*drv
= bs
->drv
;
2108 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
2112 if (!drv
->bdrv_snapshot_create
) {
2113 if (bs
->file
!= NULL
) {
2114 return bdrv_can_snapshot(bs
->file
);
2122 int bdrv_is_snapshot(BlockDriverState
*bs
)
2124 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
2127 BlockDriverState
*bdrv_snapshots(void)
2129 BlockDriverState
*bs
;
2132 return bs_snapshots
;
2136 while ((bs
= bdrv_next(bs
))) {
2137 if (bdrv_can_snapshot(bs
)) {
2145 int bdrv_snapshot_create(BlockDriverState
*bs
,
2146 QEMUSnapshotInfo
*sn_info
)
2148 BlockDriver
*drv
= bs
->drv
;
2151 if (drv
->bdrv_snapshot_create
)
2152 return drv
->bdrv_snapshot_create(bs
, sn_info
);
2154 return bdrv_snapshot_create(bs
->file
, sn_info
);
2158 int bdrv_snapshot_goto(BlockDriverState
*bs
,
2159 const char *snapshot_id
)
2161 BlockDriver
*drv
= bs
->drv
;
2166 if (drv
->bdrv_snapshot_goto
)
2167 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
2170 drv
->bdrv_close(bs
);
2171 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
2172 open_ret
= drv
->bdrv_open(bs
, bs
->open_flags
);
2174 bdrv_delete(bs
->file
);
2184 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
2186 BlockDriver
*drv
= bs
->drv
;
2189 if (drv
->bdrv_snapshot_delete
)
2190 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
2192 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
2196 int bdrv_snapshot_list(BlockDriverState
*bs
,
2197 QEMUSnapshotInfo
**psn_info
)
2199 BlockDriver
*drv
= bs
->drv
;
2202 if (drv
->bdrv_snapshot_list
)
2203 return drv
->bdrv_snapshot_list(bs
, psn_info
);
2205 return bdrv_snapshot_list(bs
->file
, psn_info
);
2209 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
2210 const char *snapshot_name
)
2212 BlockDriver
*drv
= bs
->drv
;
2216 if (!bs
->read_only
) {
2219 if (drv
->bdrv_snapshot_load_tmp
) {
2220 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
2225 #define NB_SUFFIXES 4
2227 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
2229 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
2234 snprintf(buf
, buf_size
, "%" PRId64
, size
);
2237 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
2238 if (size
< (10 * base
)) {
2239 snprintf(buf
, buf_size
, "%0.1f%c",
2240 (double)size
/ base
,
2243 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
2244 snprintf(buf
, buf_size
, "%" PRId64
"%c",
2245 ((size
+ (base
>> 1)) / base
),
2255 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
2257 char buf1
[128], date_buf
[128], clock_buf
[128];
2267 snprintf(buf
, buf_size
,
2268 "%-10s%-20s%7s%20s%15s",
2269 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2273 ptm
= localtime(&ti
);
2274 strftime(date_buf
, sizeof(date_buf
),
2275 "%Y-%m-%d %H:%M:%S", ptm
);
2277 localtime_r(&ti
, &tm
);
2278 strftime(date_buf
, sizeof(date_buf
),
2279 "%Y-%m-%d %H:%M:%S", &tm
);
2281 secs
= sn
->vm_clock_nsec
/ 1000000000;
2282 snprintf(clock_buf
, sizeof(clock_buf
),
2283 "%02d:%02d:%02d.%03d",
2285 (int)((secs
/ 60) % 60),
2287 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
2288 snprintf(buf
, buf_size
,
2289 "%-10s%-20s%7s%20s%15s",
2290 sn
->id_str
, sn
->name
,
2291 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
2298 /**************************************************************/
2301 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
2302 QEMUIOVector
*qiov
, int nb_sectors
,
2303 BlockDriverCompletionFunc
*cb
, void *opaque
)
2305 BlockDriver
*drv
= bs
->drv
;
2307 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
2311 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2314 return drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
2318 typedef struct BlockCompleteData
{
2319 BlockDriverCompletionFunc
*cb
;
2321 BlockDriverState
*bs
;
2324 } BlockCompleteData
;
2326 static void block_complete_cb(void *opaque
, int ret
)
2328 BlockCompleteData
*b
= opaque
;
2330 if (b
->bs
->dirty_bitmap
) {
2331 set_dirty_bitmap(b
->bs
, b
->sector_num
, b
->nb_sectors
, 1);
2333 b
->cb(b
->opaque
, ret
);
2337 static BlockCompleteData
*blk_dirty_cb_alloc(BlockDriverState
*bs
,
2340 BlockDriverCompletionFunc
*cb
,
2343 BlockCompleteData
*blkdata
= g_malloc0(sizeof(BlockCompleteData
));
2347 blkdata
->opaque
= opaque
;
2348 blkdata
->sector_num
= sector_num
;
2349 blkdata
->nb_sectors
= nb_sectors
;
2354 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
2355 QEMUIOVector
*qiov
, int nb_sectors
,
2356 BlockDriverCompletionFunc
*cb
, void *opaque
)
2358 BlockDriver
*drv
= bs
->drv
;
2359 BlockDriverAIOCB
*ret
;
2360 BlockCompleteData
*blk_cb_data
;
2362 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
2368 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2371 if (bs
->dirty_bitmap
) {
2372 blk_cb_data
= blk_dirty_cb_alloc(bs
, sector_num
, nb_sectors
, cb
,
2374 cb
= &block_complete_cb
;
2375 opaque
= blk_cb_data
;
2378 ret
= drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
2382 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2383 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2391 typedef struct MultiwriteCB
{
2396 BlockDriverCompletionFunc
*cb
;
2398 QEMUIOVector
*free_qiov
;
2403 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
2407 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
2408 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
2409 if (mcb
->callbacks
[i
].free_qiov
) {
2410 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
2412 g_free(mcb
->callbacks
[i
].free_qiov
);
2413 qemu_vfree(mcb
->callbacks
[i
].free_buf
);
2417 static void multiwrite_cb(void *opaque
, int ret
)
2419 MultiwriteCB
*mcb
= opaque
;
2421 trace_multiwrite_cb(mcb
, ret
);
2423 if (ret
< 0 && !mcb
->error
) {
2427 mcb
->num_requests
--;
2428 if (mcb
->num_requests
== 0) {
2429 multiwrite_user_cb(mcb
);
2434 static int multiwrite_req_compare(const void *a
, const void *b
)
2436 const BlockRequest
*req1
= a
, *req2
= b
;
2439 * Note that we can't simply subtract req2->sector from req1->sector
2440 * here as that could overflow the return value.
2442 if (req1
->sector
> req2
->sector
) {
2444 } else if (req1
->sector
< req2
->sector
) {
2452 * Takes a bunch of requests and tries to merge them. Returns the number of
2453 * requests that remain after merging.
2455 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
2456 int num_reqs
, MultiwriteCB
*mcb
)
2460 // Sort requests by start sector
2461 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
2463 // Check if adjacent requests touch the same clusters. If so, combine them,
2464 // filling up gaps with zero sectors.
2466 for (i
= 1; i
< num_reqs
; i
++) {
2468 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
2470 // This handles the cases that are valid for all block drivers, namely
2471 // exactly sequential writes and overlapping writes.
2472 if (reqs
[i
].sector
<= oldreq_last
) {
2476 // The block driver may decide that it makes sense to combine requests
2477 // even if there is a gap of some sectors between them. In this case,
2478 // the gap is filled with zeros (therefore only applicable for yet
2479 // unused space in format like qcow2).
2480 if (!merge
&& bs
->drv
->bdrv_merge_requests
) {
2481 merge
= bs
->drv
->bdrv_merge_requests(bs
, &reqs
[outidx
], &reqs
[i
]);
2484 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
2490 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
2491 qemu_iovec_init(qiov
,
2492 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
2494 // Add the first request to the merged one. If the requests are
2495 // overlapping, drop the last sectors of the first request.
2496 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
2497 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, size
);
2499 // We might need to add some zeros between the two requests
2500 if (reqs
[i
].sector
> oldreq_last
) {
2501 size_t zero_bytes
= (reqs
[i
].sector
- oldreq_last
) << 9;
2502 uint8_t *buf
= qemu_blockalign(bs
, zero_bytes
);
2503 memset(buf
, 0, zero_bytes
);
2504 qemu_iovec_add(qiov
, buf
, zero_bytes
);
2505 mcb
->callbacks
[i
].free_buf
= buf
;
2508 // Add the second request
2509 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, reqs
[i
].qiov
->size
);
2511 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
2512 reqs
[outidx
].qiov
= qiov
;
2514 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
2517 reqs
[outidx
].sector
= reqs
[i
].sector
;
2518 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
2519 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
2527 * Submit multiple AIO write requests at once.
2529 * On success, the function returns 0 and all requests in the reqs array have
2530 * been submitted. In error case this function returns -1, and any of the
2531 * requests may or may not be submitted yet. In particular, this means that the
2532 * callback will be called for some of the requests, for others it won't. The
2533 * caller must check the error field of the BlockRequest to wait for the right
2534 * callbacks (if error != 0, no callback will be called).
2536 * The implementation may modify the contents of the reqs array, e.g. to merge
2537 * requests. However, the fields opaque and error are left unmodified as they
2538 * are used to signal failure for a single request to the caller.
2540 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
2542 BlockDriverAIOCB
*acb
;
2546 /* don't submit writes if we don't have a medium */
2547 if (bs
->drv
== NULL
) {
2548 for (i
= 0; i
< num_reqs
; i
++) {
2549 reqs
[i
].error
= -ENOMEDIUM
;
2554 if (num_reqs
== 0) {
2558 // Create MultiwriteCB structure
2559 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
2560 mcb
->num_requests
= 0;
2561 mcb
->num_callbacks
= num_reqs
;
2563 for (i
= 0; i
< num_reqs
; i
++) {
2564 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
2565 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
2568 // Check for mergable requests
2569 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
2571 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
2574 * Run the aio requests. As soon as one request can't be submitted
2575 * successfully, fail all requests that are not yet submitted (we must
2576 * return failure for all requests anyway)
2578 * num_requests cannot be set to the right value immediately: If
2579 * bdrv_aio_writev fails for some request, num_requests would be too high
2580 * and therefore multiwrite_cb() would never recognize the multiwrite
2581 * request as completed. We also cannot use the loop variable i to set it
2582 * when the first request fails because the callback may already have been
2583 * called for previously submitted requests. Thus, num_requests must be
2584 * incremented for each request that is submitted.
2586 * The problem that callbacks may be called early also means that we need
2587 * to take care that num_requests doesn't become 0 before all requests are
2588 * submitted - multiwrite_cb() would consider the multiwrite request
2589 * completed. A dummy request that is "completed" by a manual call to
2590 * multiwrite_cb() takes care of this.
2592 mcb
->num_requests
= 1;
2594 // Run the aio requests
2595 for (i
= 0; i
< num_reqs
; i
++) {
2596 mcb
->num_requests
++;
2597 acb
= bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
2598 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
2601 // We can only fail the whole thing if no request has been
2602 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2603 // complete and report the error in the callback.
2605 trace_bdrv_aio_multiwrite_earlyfail(mcb
);
2608 trace_bdrv_aio_multiwrite_latefail(mcb
, i
);
2609 multiwrite_cb(mcb
, -EIO
);
2615 /* Complete the dummy request */
2616 multiwrite_cb(mcb
, 0);
2621 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
2622 reqs
[i
].error
= -EIO
;
2628 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2629 BlockDriverCompletionFunc
*cb
, void *opaque
)
2631 BlockDriver
*drv
= bs
->drv
;
2633 trace_bdrv_aio_flush(bs
, opaque
);
2635 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2636 return bdrv_aio_noop_em(bs
, cb
, opaque
);
2641 return drv
->bdrv_aio_flush(bs
, cb
, opaque
);
2644 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
2646 acb
->pool
->cancel(acb
);
2650 /**************************************************************/
2651 /* async block device emulation */
2653 typedef struct BlockDriverAIOCBSync
{
2654 BlockDriverAIOCB common
;
2657 /* vector translation state */
2661 } BlockDriverAIOCBSync
;
2663 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
2665 BlockDriverAIOCBSync
*acb
=
2666 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
2667 qemu_bh_delete(acb
->bh
);
2669 qemu_aio_release(acb
);
2672 static AIOPool bdrv_em_aio_pool
= {
2673 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
2674 .cancel
= bdrv_aio_cancel_em
,
2677 static void bdrv_aio_bh_cb(void *opaque
)
2679 BlockDriverAIOCBSync
*acb
= opaque
;
2682 qemu_iovec_from_buffer(acb
->qiov
, acb
->bounce
, acb
->qiov
->size
);
2683 qemu_vfree(acb
->bounce
);
2684 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
2685 qemu_bh_delete(acb
->bh
);
2687 qemu_aio_release(acb
);
2690 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
2694 BlockDriverCompletionFunc
*cb
,
2699 BlockDriverAIOCBSync
*acb
;
2701 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2702 acb
->is_write
= is_write
;
2704 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
2707 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2710 qemu_iovec_to_buffer(acb
->qiov
, acb
->bounce
);
2711 acb
->ret
= bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2713 acb
->ret
= bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2716 qemu_bh_schedule(acb
->bh
);
2718 return &acb
->common
;
2721 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2722 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2723 BlockDriverCompletionFunc
*cb
, void *opaque
)
2725 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2728 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2729 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2730 BlockDriverCompletionFunc
*cb
, void *opaque
)
2732 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2736 typedef struct BlockDriverAIOCBCoroutine
{
2737 BlockDriverAIOCB common
;
2741 } BlockDriverAIOCBCoroutine
;
2743 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
2748 static AIOPool bdrv_em_co_aio_pool
= {
2749 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
2750 .cancel
= bdrv_aio_co_cancel_em
,
2753 static void bdrv_co_rw_bh(void *opaque
)
2755 BlockDriverAIOCBCoroutine
*acb
= opaque
;
2757 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2758 qemu_bh_delete(acb
->bh
);
2759 qemu_aio_release(acb
);
2762 static void coroutine_fn
bdrv_co_rw(void *opaque
)
2764 BlockDriverAIOCBCoroutine
*acb
= opaque
;
2765 BlockDriverState
*bs
= acb
->common
.bs
;
2767 if (!acb
->is_write
) {
2768 acb
->req
.error
= bs
->drv
->bdrv_co_readv(bs
, acb
->req
.sector
,
2769 acb
->req
.nb_sectors
, acb
->req
.qiov
);
2771 acb
->req
.error
= bs
->drv
->bdrv_co_writev(bs
, acb
->req
.sector
,
2772 acb
->req
.nb_sectors
, acb
->req
.qiov
);
2775 acb
->bh
= qemu_bh_new(bdrv_co_rw_bh
, acb
);
2776 qemu_bh_schedule(acb
->bh
);
2779 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2783 BlockDriverCompletionFunc
*cb
,
2788 BlockDriverAIOCBCoroutine
*acb
;
2790 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
2791 acb
->req
.sector
= sector_num
;
2792 acb
->req
.nb_sectors
= nb_sectors
;
2793 acb
->req
.qiov
= qiov
;
2794 acb
->is_write
= is_write
;
2796 co
= qemu_coroutine_create(bdrv_co_rw
);
2797 qemu_coroutine_enter(co
, acb
);
2799 return &acb
->common
;
2802 static BlockDriverAIOCB
*bdrv_co_aio_readv_em(BlockDriverState
*bs
,
2803 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2804 BlockDriverCompletionFunc
*cb
, void *opaque
)
2806 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
,
2810 static BlockDriverAIOCB
*bdrv_co_aio_writev_em(BlockDriverState
*bs
,
2811 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2812 BlockDriverCompletionFunc
*cb
, void *opaque
)
2814 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
,
2818 static BlockDriverAIOCB
*bdrv_aio_flush_em(BlockDriverState
*bs
,
2819 BlockDriverCompletionFunc
*cb
, void *opaque
)
2821 BlockDriverAIOCBSync
*acb
;
2823 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2824 acb
->is_write
= 1; /* don't bounce in the completion hadler */
2830 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2833 qemu_bh_schedule(acb
->bh
);
2834 return &acb
->common
;
2837 static BlockDriverAIOCB
*bdrv_aio_noop_em(BlockDriverState
*bs
,
2838 BlockDriverCompletionFunc
*cb
, void *opaque
)
2840 BlockDriverAIOCBSync
*acb
;
2842 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
2843 acb
->is_write
= 1; /* don't bounce in the completion handler */
2849 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
2852 qemu_bh_schedule(acb
->bh
);
2853 return &acb
->common
;
2856 /**************************************************************/
2857 /* sync block device emulation */
2859 static void bdrv_rw_em_cb(void *opaque
, int ret
)
2861 *(int *)opaque
= ret
;
2864 #define NOT_DONE 0x7fffffff
2866 static int bdrv_read_em(BlockDriverState
*bs
, int64_t sector_num
,
2867 uint8_t *buf
, int nb_sectors
)
2870 BlockDriverAIOCB
*acb
;
2874 async_ret
= NOT_DONE
;
2875 iov
.iov_base
= (void *)buf
;
2876 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2877 qemu_iovec_init_external(&qiov
, &iov
, 1);
2878 acb
= bdrv_aio_readv(bs
, sector_num
, &qiov
, nb_sectors
,
2879 bdrv_rw_em_cb
, &async_ret
);
2885 while (async_ret
== NOT_DONE
) {
2894 static int bdrv_write_em(BlockDriverState
*bs
, int64_t sector_num
,
2895 const uint8_t *buf
, int nb_sectors
)
2898 BlockDriverAIOCB
*acb
;
2902 async_ret
= NOT_DONE
;
2903 iov
.iov_base
= (void *)buf
;
2904 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2905 qemu_iovec_init_external(&qiov
, &iov
, 1);
2906 acb
= bdrv_aio_writev(bs
, sector_num
, &qiov
, nb_sectors
,
2907 bdrv_rw_em_cb
, &async_ret
);
2912 while (async_ret
== NOT_DONE
) {
2920 void bdrv_init(void)
2922 module_call_init(MODULE_INIT_BLOCK
);
2925 void bdrv_init_with_whitelist(void)
2927 use_bdrv_whitelist
= 1;
2931 void *qemu_aio_get(AIOPool
*pool
, BlockDriverState
*bs
,
2932 BlockDriverCompletionFunc
*cb
, void *opaque
)
2934 BlockDriverAIOCB
*acb
;
2936 if (pool
->free_aiocb
) {
2937 acb
= pool
->free_aiocb
;
2938 pool
->free_aiocb
= acb
->next
;
2940 acb
= g_malloc0(pool
->aiocb_size
);
2945 acb
->opaque
= opaque
;
2949 void qemu_aio_release(void *p
)
2951 BlockDriverAIOCB
*acb
= (BlockDriverAIOCB
*)p
;
2952 AIOPool
*pool
= acb
->pool
;
2953 acb
->next
= pool
->free_aiocb
;
2954 pool
->free_aiocb
= acb
;
2957 /**************************************************************/
2958 /* Coroutine block device emulation */
2960 typedef struct CoroutineIOCompletion
{
2961 Coroutine
*coroutine
;
2963 } CoroutineIOCompletion
;
2965 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
2967 CoroutineIOCompletion
*co
= opaque
;
2970 qemu_coroutine_enter(co
->coroutine
, NULL
);
2973 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
2974 int nb_sectors
, QEMUIOVector
*iov
,
2977 CoroutineIOCompletion co
= {
2978 .coroutine
= qemu_coroutine_self(),
2980 BlockDriverAIOCB
*acb
;
2983 acb
= bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
2984 bdrv_co_io_em_complete
, &co
);
2986 acb
= bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
2987 bdrv_co_io_em_complete
, &co
);
2990 trace_bdrv_co_io(is_write
, acb
);
2994 qemu_coroutine_yield();
2999 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
3000 int64_t sector_num
, int nb_sectors
,
3003 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
3006 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
3007 int64_t sector_num
, int nb_sectors
,
3010 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
3013 static int coroutine_fn
bdrv_co_flush_em(BlockDriverState
*bs
)
3015 CoroutineIOCompletion co
= {
3016 .coroutine
= qemu_coroutine_self(),
3018 BlockDriverAIOCB
*acb
;
3020 acb
= bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
3024 qemu_coroutine_yield();
3028 /**************************************************************/
3029 /* removable device support */
3032 * Return TRUE if the media is present
3034 int bdrv_is_inserted(BlockDriverState
*bs
)
3036 BlockDriver
*drv
= bs
->drv
;
3040 if (!drv
->bdrv_is_inserted
)
3042 return drv
->bdrv_is_inserted(bs
);
3046 * Return whether the media changed since the last call to this
3047 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3049 int bdrv_media_changed(BlockDriverState
*bs
)
3051 BlockDriver
*drv
= bs
->drv
;
3053 if (drv
&& drv
->bdrv_media_changed
) {
3054 return drv
->bdrv_media_changed(bs
);
3060 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3062 void bdrv_eject(BlockDriverState
*bs
, int eject_flag
)
3064 BlockDriver
*drv
= bs
->drv
;
3066 if (drv
&& drv
->bdrv_eject
) {
3067 drv
->bdrv_eject(bs
, eject_flag
);
3072 * Lock or unlock the media (if it is locked, the user won't be able
3073 * to eject it manually).
3075 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
3077 BlockDriver
*drv
= bs
->drv
;
3079 trace_bdrv_lock_medium(bs
, locked
);
3081 if (drv
&& drv
->bdrv_lock_medium
) {
3082 drv
->bdrv_lock_medium(bs
, locked
);
3086 /* needed for generic scsi interface */
3088 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3090 BlockDriver
*drv
= bs
->drv
;
3092 if (drv
&& drv
->bdrv_ioctl
)
3093 return drv
->bdrv_ioctl(bs
, req
, buf
);
3097 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
3098 unsigned long int req
, void *buf
,
3099 BlockDriverCompletionFunc
*cb
, void *opaque
)
3101 BlockDriver
*drv
= bs
->drv
;
3103 if (drv
&& drv
->bdrv_aio_ioctl
)
3104 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
3110 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
3112 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
3115 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int enable
)
3117 int64_t bitmap_size
;
3119 bs
->dirty_count
= 0;
3121 if (!bs
->dirty_bitmap
) {
3122 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
) +
3123 BDRV_SECTORS_PER_DIRTY_CHUNK
* 8 - 1;
3124 bitmap_size
/= BDRV_SECTORS_PER_DIRTY_CHUNK
* 8;
3126 bs
->dirty_bitmap
= g_malloc0(bitmap_size
);
3129 if (bs
->dirty_bitmap
) {
3130 g_free(bs
->dirty_bitmap
);
3131 bs
->dirty_bitmap
= NULL
;
3136 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
3138 int64_t chunk
= sector
/ (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK
;
3140 if (bs
->dirty_bitmap
&&
3141 (sector
<< BDRV_SECTOR_BITS
) < bdrv_getlength(bs
)) {
3142 return !!(bs
->dirty_bitmap
[chunk
/ (sizeof(unsigned long) * 8)] &
3143 (1UL << (chunk
% (sizeof(unsigned long) * 8))));
3149 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
3152 set_dirty_bitmap(bs
, cur_sector
, nr_sectors
, 0);
3155 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
3157 return bs
->dirty_count
;
3160 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
3162 assert(bs
->in_use
!= in_use
);
3163 bs
->in_use
= in_use
;
3166 int bdrv_in_use(BlockDriverState
*bs
)
3172 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
3173 enum BlockAcctType type
)
3175 assert(type
< BDRV_MAX_IOTYPE
);
3177 cookie
->bytes
= bytes
;
3178 cookie
->start_time_ns
= get_clock();
3179 cookie
->type
= type
;
3183 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
3185 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
3187 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
3188 bs
->nr_ops
[cookie
->type
]++;
3189 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
3192 int bdrv_img_create(const char *filename
, const char *fmt
,
3193 const char *base_filename
, const char *base_fmt
,
3194 char *options
, uint64_t img_size
, int flags
)
3196 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
3197 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
3198 BlockDriverState
*bs
= NULL
;
3199 BlockDriver
*drv
, *proto_drv
;
3200 BlockDriver
*backing_drv
= NULL
;
3203 /* Find driver and parse its options */
3204 drv
= bdrv_find_format(fmt
);
3206 error_report("Unknown file format '%s'", fmt
);
3211 proto_drv
= bdrv_find_protocol(filename
);
3213 error_report("Unknown protocol '%s'", filename
);
3218 create_options
= append_option_parameters(create_options
,
3219 drv
->create_options
);
3220 create_options
= append_option_parameters(create_options
,
3221 proto_drv
->create_options
);
3223 /* Create parameter list with default values */
3224 param
= parse_option_parameters("", create_options
, param
);
3226 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
3228 /* Parse -o options */
3230 param
= parse_option_parameters(options
, create_options
, param
);
3231 if (param
== NULL
) {
3232 error_report("Invalid options for file format '%s'.", fmt
);
3238 if (base_filename
) {
3239 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
3241 error_report("Backing file not supported for file format '%s'",
3249 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
3250 error_report("Backing file format not supported for file "
3251 "format '%s'", fmt
);
3257 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
3258 if (backing_file
&& backing_file
->value
.s
) {
3259 if (!strcmp(filename
, backing_file
->value
.s
)) {
3260 error_report("Error: Trying to create an image with the "
3261 "same filename as the backing file");
3267 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
3268 if (backing_fmt
&& backing_fmt
->value
.s
) {
3269 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
3271 error_report("Unknown backing file format '%s'",
3272 backing_fmt
->value
.s
);
3278 // The size for the image must always be specified, with one exception:
3279 // If we are using a backing file, we can obtain the size from there
3280 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
3281 if (size
&& size
->value
.n
== -1) {
3282 if (backing_file
&& backing_file
->value
.s
) {
3288 ret
= bdrv_open(bs
, backing_file
->value
.s
, flags
, backing_drv
);
3290 error_report("Could not open '%s'", backing_file
->value
.s
);
3293 bdrv_get_geometry(bs
, &size
);
3296 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
3297 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
3299 error_report("Image creation needs a size parameter");
3305 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
3306 print_option_parameters(param
);
3309 ret
= bdrv_create(drv
, filename
, param
);
3312 if (ret
== -ENOTSUP
) {
3313 error_report("Formatting or formatting option not supported for "
3314 "file format '%s'", fmt
);
3315 } else if (ret
== -EFBIG
) {
3316 error_report("The image size is too large for file format '%s'",
3319 error_report("%s: error while creating %s: %s", filename
, fmt
,
3325 free_option_parameters(create_options
);
3326 free_option_parameters(param
);