2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 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
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
31 #include "qemu/option.h"
33 #include "block/thread-pool.h"
35 #include "block/raw-aio.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
39 #include "scsi/pr-manager.h"
40 #include "scsi/constants.h"
42 #if defined(__APPLE__) && (__MACH__)
44 #include <sys/param.h>
45 #include <IOKit/IOKitLib.h>
46 #include <IOKit/IOBSD.h>
47 #include <IOKit/storage/IOMediaBSDClient.h>
48 #include <IOKit/storage/IOMedia.h>
49 #include <IOKit/storage/IOCDMedia.h>
50 //#include <IOKit/storage/IOCDTypes.h>
51 #include <IOKit/storage/IODVDMedia.h>
52 #include <CoreFoundation/CoreFoundation.h>
56 #define _POSIX_PTHREAD_SEMANTICS 1
60 #include <sys/ioctl.h>
61 #include <sys/param.h>
62 #include <sys/syscall.h>
63 #include <linux/cdrom.h>
66 #include <linux/hdreg.h>
72 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
75 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
76 #include <linux/falloc.h>
78 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
84 #include <sys/ioctl.h>
85 #include <sys/disklabel.h>
90 #include <sys/ioctl.h>
91 #include <sys/disklabel.h>
97 #include <sys/ioctl.h>
98 #include <sys/diskslice.h>
107 /* OS X does not have O_DSYNC */
110 #define O_DSYNC O_SYNC
111 #elif defined(O_FSYNC)
112 #define O_DSYNC O_FSYNC
116 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
118 #define O_DIRECT O_DSYNC
124 #define MAX_BLOCKSIZE 4096
126 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
127 * leaving a few more bytes for its future use. */
128 #define RAW_LOCK_PERM_BASE 100
129 #define RAW_LOCK_SHARED_BASE 200
131 typedef struct BDRVRawState
{
138 /* The current permissions. */
140 uint64_t shared_perm
;
142 /* The perms bits whose corresponding bytes are already locked in
144 uint64_t locked_perm
;
145 uint64_t locked_shared_perm
;
147 BDRVReopenState
*reopen_state
;
153 bool has_write_zeroes
:1;
154 bool discard_zeroes
:1;
155 bool use_linux_aio
:1;
156 bool page_cache_inconsistent
:1;
158 bool needs_alignment
;
159 bool check_cache_dropped
;
164 typedef struct BDRVRawReopenState
{
167 bool check_cache_dropped
;
168 } BDRVRawReopenState
;
170 static int fd_open(BlockDriverState
*bs
);
171 static int64_t raw_getlength(BlockDriverState
*bs
);
173 typedef struct RawPosixAIOData
{
174 BlockDriverState
*bs
;
195 PreallocMode prealloc
;
201 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
202 static int cdrom_reopen(BlockDriverState
*bs
);
205 #if defined(__NetBSD__)
206 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
208 static char namebuf
[PATH_MAX
];
209 const char *dp
, *fname
;
213 dp
= strrchr(fname
, '/');
214 if (lstat(fname
, &sb
) < 0) {
215 error_setg_errno(errp
, errno
, "%s: stat failed", fname
);
219 if (!S_ISBLK(sb
.st_mode
)) {
224 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
226 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
227 (int)(dp
- fname
), fname
, dp
+ 1);
230 warn_report("%s is a block device, using %s", fname
, *filename
);
235 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
242 * Get logical block size via ioctl. On success store it in @sector_size_p.
244 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
246 unsigned int sector_size
;
247 bool success
= false;
251 static const unsigned long ioctl_list
[] = {
255 #ifdef DKIOCGETBLOCKSIZE
258 #ifdef DIOCGSECTORSIZE
263 /* Try a few ioctls to get the right size */
264 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
265 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
266 *sector_size_p
= sector_size
;
271 return success
? 0 : -errno
;
275 * Get physical block size of @fd.
276 * On success, store it in @blk_size and return 0.
277 * On failure, return -errno.
279 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
282 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
291 /* Check if read is allowed with given memory buffer and length.
293 * This function is used to check O_DIRECT memory buffer and request alignment.
295 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
297 ssize_t ret
= pread(fd
, buf
, len
, 0);
304 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
305 * other errors (e.g. real I/O error), which could happen on a failed
306 * drive, since we only care about probing alignment.
308 if (errno
!= EINVAL
) {
316 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
318 BDRVRawState
*s
= bs
->opaque
;
320 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
322 /* For SCSI generic devices the alignment is not really used.
323 With buffered I/O, we don't have any restrictions. */
324 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
325 bs
->bl
.request_alignment
= 1;
330 bs
->bl
.request_alignment
= 0;
332 /* Let's try to use the logical blocksize for the alignment. */
333 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
334 bs
->bl
.request_alignment
= 0;
339 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
340 bs
->bl
.request_alignment
= da
.d_miniosz
;
341 /* The kernel returns wrong information for d_mem */
342 /* s->buf_align = da.d_mem; */
347 /* If we could not get the sizes so far, we can only guess them */
350 buf
= qemu_memalign(max_align
, 2 * max_align
);
351 for (align
= 512; align
<= max_align
; align
<<= 1) {
352 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
353 s
->buf_align
= align
;
360 if (!bs
->bl
.request_alignment
) {
362 buf
= qemu_memalign(s
->buf_align
, max_align
);
363 for (align
= 512; align
<= max_align
; align
<<= 1) {
364 if (raw_is_io_aligned(fd
, buf
, align
)) {
365 bs
->bl
.request_alignment
= align
;
372 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
373 error_setg(errp
, "Could not find working O_DIRECT alignment");
374 error_append_hint(errp
, "Try cache.direct=off\n");
378 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
380 assert(open_flags
!= NULL
);
382 *open_flags
|= O_BINARY
;
383 *open_flags
&= ~O_ACCMODE
;
384 if (bdrv_flags
& BDRV_O_RDWR
) {
385 *open_flags
|= O_RDWR
;
387 *open_flags
|= O_RDONLY
;
390 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
391 * and O_DIRECT for no caching. */
392 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
393 *open_flags
|= O_DIRECT
;
397 static void raw_parse_filename(const char *filename
, QDict
*options
,
400 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
403 static QemuOptsList raw_runtime_opts
= {
405 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
409 .type
= QEMU_OPT_STRING
,
410 .help
= "File name of the image",
414 .type
= QEMU_OPT_STRING
,
415 .help
= "host AIO implementation (threads, native)",
419 .type
= QEMU_OPT_STRING
,
420 .help
= "file locking mode (on/off/auto, default: auto)",
423 .name
= "pr-manager",
424 .type
= QEMU_OPT_STRING
,
425 .help
= "id of persistent reservation manager object (default: none)",
428 .name
= "x-check-cache-dropped",
429 .type
= QEMU_OPT_BOOL
,
430 .help
= "check that page cache was dropped on live migration (default: off)"
432 { /* end of list */ }
436 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
437 int bdrv_flags
, int open_flags
,
438 bool device
, Error
**errp
)
440 BDRVRawState
*s
= bs
->opaque
;
442 Error
*local_err
= NULL
;
443 const char *filename
= NULL
;
445 BlockdevAioOptions aio
, aio_default
;
450 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
451 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
453 error_propagate(errp
, local_err
);
458 filename
= qemu_opt_get(opts
, "filename");
460 ret
= raw_normalize_devicepath(&filename
, errp
);
465 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
466 ? BLOCKDEV_AIO_OPTIONS_NATIVE
467 : BLOCKDEV_AIO_OPTIONS_THREADS
;
468 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
469 qemu_opt_get(opts
, "aio"),
470 aio_default
, &local_err
);
472 error_propagate(errp
, local_err
);
476 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
478 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
479 qemu_opt_get(opts
, "locking"),
480 ON_OFF_AUTO_AUTO
, &local_err
);
482 error_propagate(errp
, local_err
);
489 if (!qemu_has_ofd_lock()) {
490 warn_report("File lock requested but OFD locking syscall is "
491 "unavailable, falling back to POSIX file locks");
492 error_printf("Due to the implementation, locks can be lost "
496 case ON_OFF_AUTO_OFF
:
499 case ON_OFF_AUTO_AUTO
:
500 s
->use_lock
= qemu_has_ofd_lock();
506 str
= qemu_opt_get(opts
, "pr-manager");
508 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
510 error_propagate(errp
, local_err
);
516 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
519 s
->open_flags
= open_flags
;
520 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
523 fd
= qemu_open(filename
, s
->open_flags
, 0644);
524 ret
= fd
< 0 ? -errno
: 0;
526 if (ret
== -EACCES
|| ret
== -EROFS
) {
527 /* Try to degrade to read-only, but if it doesn't work, still use the
528 * normal error message. */
529 if (bdrv_apply_auto_read_only(bs
, NULL
, NULL
) == 0) {
530 bdrv_flags
&= ~BDRV_O_RDWR
;
531 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
532 assert(!(s
->open_flags
& O_CREAT
));
533 fd
= qemu_open(filename
, s
->open_flags
);
534 ret
= fd
< 0 ? -errno
: 0;
539 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
548 s
->shared_perm
= BLK_PERM_ALL
;
550 #ifdef CONFIG_LINUX_AIO
551 /* Currently Linux does AIO only for files opened with O_DIRECT */
552 if (s
->use_linux_aio
) {
553 if (!(s
->open_flags
& O_DIRECT
)) {
554 error_setg(errp
, "aio=native was specified, but it requires "
555 "cache.direct=on, which was not specified.");
559 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
560 error_prepend(errp
, "Unable to use native AIO: ");
565 if (s
->use_linux_aio
) {
566 error_setg(errp
, "aio=native was specified, but is not supported "
571 #endif /* !defined(CONFIG_LINUX_AIO) */
573 s
->has_discard
= true;
574 s
->has_write_zeroes
= true;
575 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
576 s
->needs_alignment
= true;
579 if (fstat(s
->fd
, &st
) < 0) {
581 error_setg_errno(errp
, errno
, "Could not stat file");
586 if (S_ISBLK(st
.st_mode
)) {
587 warn_report("Opening a block device as a file using the '%s' "
588 "driver is deprecated", bs
->drv
->format_name
);
589 } else if (S_ISCHR(st
.st_mode
)) {
590 warn_report("Opening a character device as a file using the '%s' "
591 "driver is deprecated", bs
->drv
->format_name
);
592 } else if (!S_ISREG(st
.st_mode
)) {
593 error_setg(errp
, "A regular file was expected by the '%s' driver, "
594 "but something else was given", bs
->drv
->format_name
);
598 s
->discard_zeroes
= true;
599 s
->has_fallocate
= true;
602 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
603 error_setg(errp
, "'%s' driver expects either "
604 "a character or block device", bs
->drv
->format_name
);
610 if (S_ISBLK(st
.st_mode
)) {
611 #ifdef BLKDISCARDZEROES
613 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
614 s
->discard_zeroes
= true;
618 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
619 * not rely on the contents of discarded blocks unless using O_DIRECT.
620 * Same for BLKZEROOUT.
622 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
623 s
->discard_zeroes
= false;
624 s
->has_write_zeroes
= false;
629 if (S_ISCHR(st
.st_mode
)) {
631 * The file is a char device (disk), which on FreeBSD isn't behind
632 * a pager, so force all requests to be aligned. This is needed
633 * so QEMU makes sure all IO operations on the device are aligned
634 * to sector size, or else FreeBSD will reject them with EINVAL.
636 s
->needs_alignment
= true;
641 if (platform_test_xfs_fd(s
->fd
)) {
646 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
649 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
656 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
659 BDRVRawState
*s
= bs
->opaque
;
661 s
->type
= FTYPE_FILE
;
662 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
671 #define PERM_FOREACH(i) \
672 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
674 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
675 * file; if @unlock == true, also unlock the unneeded bytes.
676 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
678 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
679 uint64_t perm_lock_bits
,
680 uint64_t shared_perm_lock_bits
,
681 bool unlock
, Error
**errp
)
685 uint64_t locked_perm
, locked_shared_perm
;
688 locked_perm
= s
->locked_perm
;
689 locked_shared_perm
= s
->locked_shared_perm
;
692 * We don't have the previous bits, just lock/unlock for each of the
696 locked_perm
= BLK_PERM_ALL
;
697 locked_shared_perm
= BLK_PERM_ALL
;
700 locked_shared_perm
= 0;
705 int off
= RAW_LOCK_PERM_BASE
+ i
;
706 uint64_t bit
= (1ULL << i
);
707 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
708 ret
= qemu_lock_fd(fd
, off
, 1, false);
710 error_setg(errp
, "Failed to lock byte %d", off
);
713 s
->locked_perm
|= bit
;
715 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
716 ret
= qemu_unlock_fd(fd
, off
, 1);
718 error_setg(errp
, "Failed to unlock byte %d", off
);
721 s
->locked_perm
&= ~bit
;
726 int off
= RAW_LOCK_SHARED_BASE
+ i
;
727 uint64_t bit
= (1ULL << i
);
728 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
729 ret
= qemu_lock_fd(fd
, off
, 1, false);
731 error_setg(errp
, "Failed to lock byte %d", off
);
734 s
->locked_shared_perm
|= bit
;
736 } else if (unlock
&& (locked_shared_perm
& bit
) &&
737 !(shared_perm_lock_bits
& bit
)) {
738 ret
= qemu_unlock_fd(fd
, off
, 1);
740 error_setg(errp
, "Failed to unlock byte %d", off
);
743 s
->locked_shared_perm
&= ~bit
;
750 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
751 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
758 int off
= RAW_LOCK_SHARED_BASE
+ i
;
759 uint64_t p
= 1ULL << i
;
761 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
763 char *perm_name
= bdrv_perm_names(p
);
765 "Failed to get \"%s\" lock",
773 int off
= RAW_LOCK_PERM_BASE
+ i
;
774 uint64_t p
= 1ULL << i
;
775 if (!(shared_perm
& p
)) {
776 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
778 char *perm_name
= bdrv_perm_names(p
);
780 "Failed to get shared \"%s\" lock",
790 static int raw_handle_perm_lock(BlockDriverState
*bs
,
792 uint64_t new_perm
, uint64_t new_shared
,
795 BDRVRawState
*s
= bs
->opaque
;
797 Error
*local_err
= NULL
;
803 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
809 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
810 ~s
->shared_perm
| ~new_shared
,
813 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
817 error_append_hint(errp
,
818 "Is another process using the image [%s]?\n",
822 /* fall through to unlock bytes. */
824 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
827 /* Theoretically the above call only unlocks bytes and it cannot
828 * fail. Something weird happened, report it.
830 warn_report_err(local_err
);
834 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
837 /* Theoretically the above call only unlocks bytes and it cannot
838 * fail. Something weird happened, report it.
840 warn_report_err(local_err
);
847 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
848 int *open_flags
, Error
**errp
)
850 BDRVRawState
*s
= bs
->opaque
;
853 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
855 fcntl_flags
|= O_NOATIME
;
859 if (s
->type
== FTYPE_CD
) {
860 *open_flags
|= O_NONBLOCK
;
863 raw_parse_flags(flags
, open_flags
);
866 /* Not all operating systems have O_ASYNC, and those that don't
867 * will not let us track the state into rs->open_flags (typically
868 * you achieve the same effect with an ioctl, for example I_SETSIG
869 * on Solaris). But we do not use O_ASYNC, so that's fine.
871 assert((s
->open_flags
& O_ASYNC
) == 0);
874 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
875 /* dup the original fd */
876 fd
= qemu_dup(s
->fd
);
878 ret
= fcntl_setfl(fd
, *open_flags
);
886 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
888 const char *normalized_filename
= bs
->filename
;
889 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
891 assert(!(*open_flags
& O_CREAT
));
892 fd
= qemu_open(normalized_filename
, *open_flags
);
894 error_setg_errno(errp
, errno
, "Could not reopen file");
903 static int raw_reopen_prepare(BDRVReopenState
*state
,
904 BlockReopenQueue
*queue
, Error
**errp
)
907 BDRVRawReopenState
*rs
;
910 Error
*local_err
= NULL
;
912 assert(state
!= NULL
);
913 assert(state
->bs
!= NULL
);
915 s
= state
->bs
->opaque
;
917 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
920 /* Handle options changes */
921 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
922 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
924 error_propagate(errp
, local_err
);
929 rs
->check_cache_dropped
=
930 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
932 /* This driver's reopen function doesn't currently allow changing
933 * other options, so let's put them back in the original QDict and
934 * bdrv_reopen_prepare() will detect changes and complain. */
935 qemu_opts_to_qdict(opts
, state
->options
);
937 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
940 error_propagate(errp
, local_err
);
945 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
946 * alignment with the new fd. */
948 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
952 error_propagate(errp
, local_err
);
957 s
->reopen_state
= state
;
963 static void raw_reopen_commit(BDRVReopenState
*state
)
965 BDRVRawReopenState
*rs
= state
->opaque
;
966 BDRVRawState
*s
= state
->bs
->opaque
;
967 Error
*local_err
= NULL
;
969 s
->check_cache_dropped
= rs
->check_cache_dropped
;
970 s
->open_flags
= rs
->open_flags
;
972 /* Copy locks to the new fd before closing the old one. */
973 raw_apply_lock_bytes(NULL
, rs
->fd
, s
->locked_perm
,
974 s
->locked_shared_perm
, false, &local_err
);
976 /* shouldn't fail in a sane host, but report it just in case. */
977 error_report_err(local_err
);
982 g_free(state
->opaque
);
983 state
->opaque
= NULL
;
985 assert(s
->reopen_state
== state
);
986 s
->reopen_state
= NULL
;
990 static void raw_reopen_abort(BDRVReopenState
*state
)
992 BDRVRawReopenState
*rs
= state
->opaque
;
993 BDRVRawState
*s
= state
->bs
->opaque
;
995 /* nothing to do if NULL, we didn't get far enough */
1004 g_free(state
->opaque
);
1005 state
->opaque
= NULL
;
1007 assert(s
->reopen_state
== state
);
1008 s
->reopen_state
= NULL
;
1011 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
1015 short max_sectors
= 0;
1016 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1018 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1019 return max_sectors
<< BDRV_SECTOR_BITS
;
1028 static int hdev_get_max_segments(const struct stat
*st
)
1038 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1039 major(st
->st_rdev
), minor(st
->st_rdev
));
1040 fd
= open(sysfspath
, O_RDONLY
);
1046 ret
= read(fd
, buf
, sizeof(buf
) - 1);
1047 } while (ret
== -1 && errno
== EINTR
);
1051 } else if (ret
== 0) {
1056 /* The file is ended with '\n', pass 'end' to accept that. */
1057 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1058 if (ret
== 0 && end
&& *end
== '\n') {
1073 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1075 BDRVRawState
*s
= bs
->opaque
;
1078 if (!fstat(s
->fd
, &st
)) {
1079 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1080 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1081 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1082 bs
->bl
.max_transfer
= pow2floor(ret
);
1084 ret
= hdev_get_max_segments(&st
);
1086 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1087 ret
* getpagesize());
1092 raw_probe_alignment(bs
, s
->fd
, errp
);
1093 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1094 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1097 static int check_for_dasd(int fd
)
1100 struct dasd_information2_t info
= {0};
1102 return ioctl(fd
, BIODASDINFO2
, &info
);
1109 * Try to get @bs's logical and physical block size.
1110 * On success, store them in @bsz and return zero.
1111 * On failure, return negative errno.
1113 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1115 BDRVRawState
*s
= bs
->opaque
;
1118 /* If DASD, get blocksizes */
1119 if (check_for_dasd(s
->fd
) < 0) {
1122 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1126 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1130 * Try to get @bs's geometry: cyls, heads, sectors.
1131 * On success, store them in @geo and return 0.
1132 * On failure return -errno.
1133 * (Allows block driver to assign default geometry values that guest sees)
1136 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1138 BDRVRawState
*s
= bs
->opaque
;
1139 struct hd_geometry ioctl_geo
= {0};
1141 /* If DASD, get its geometry */
1142 if (check_for_dasd(s
->fd
) < 0) {
1145 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1148 /* HDIO_GETGEO may return success even though geo contains zeros
1149 (e.g. certain multipath setups) */
1150 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1153 /* Do not return a geometry for partition */
1154 if (ioctl_geo
.start
!= 0) {
1157 geo
->heads
= ioctl_geo
.heads
;
1158 geo
->sectors
= ioctl_geo
.sectors
;
1159 geo
->cylinders
= ioctl_geo
.cylinders
;
1163 #else /* __linux__ */
1164 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1170 #if defined(__linux__)
1171 static int handle_aiocb_ioctl(void *opaque
)
1173 RawPosixAIOData
*aiocb
= opaque
;
1176 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1185 static int handle_aiocb_flush(void *opaque
)
1187 RawPosixAIOData
*aiocb
= opaque
;
1188 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1191 if (s
->page_cache_inconsistent
) {
1195 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1197 /* There is no clear definition of the semantics of a failing fsync(),
1198 * so we may have to assume the worst. The sad truth is that this
1199 * assumption is correct for Linux. Some pages are now probably marked
1200 * clean in the page cache even though they are inconsistent with the
1201 * on-disk contents. The next fdatasync() call would succeed, but no
1202 * further writeback attempt will be made. We can't get back to a state
1203 * in which we know what is on disk (we would have to rewrite
1204 * everything that was touched since the last fdatasync() at least), so
1205 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1206 * really defined, I have little hope that other OSes are doing better.
1208 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1210 if ((s
->open_flags
& O_DIRECT
) == 0) {
1211 s
->page_cache_inconsistent
= true;
1218 #ifdef CONFIG_PREADV
1220 static bool preadv_present
= true;
1223 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1225 return preadv(fd
, iov
, nr_iov
, offset
);
1229 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1231 return pwritev(fd
, iov
, nr_iov
, offset
);
1236 static bool preadv_present
= false;
1239 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1245 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1252 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1257 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1258 len
= qemu_pwritev(aiocb
->aio_fildes
,
1263 len
= qemu_preadv(aiocb
->aio_fildes
,
1267 } while (len
== -1 && errno
== EINTR
);
1276 * Read/writes the data to/from a given linear buffer.
1278 * Returns the number of bytes handles or -errno in case of an error. Short
1279 * reads are only returned if the end of the file is reached.
1281 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1286 while (offset
< aiocb
->aio_nbytes
) {
1287 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1288 len
= pwrite(aiocb
->aio_fildes
,
1289 (const char *)buf
+ offset
,
1290 aiocb
->aio_nbytes
- offset
,
1291 aiocb
->aio_offset
+ offset
);
1293 len
= pread(aiocb
->aio_fildes
,
1295 aiocb
->aio_nbytes
- offset
,
1296 aiocb
->aio_offset
+ offset
);
1298 if (len
== -1 && errno
== EINTR
) {
1300 } else if (len
== -1 && errno
== EINVAL
&&
1301 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1302 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1304 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1305 * after a short read. Assume that O_DIRECT short reads only occur
1306 * at EOF. Therefore this is a short read, not an I/O error.
1309 } else if (len
== -1) {
1312 } else if (len
== 0) {
1321 static int handle_aiocb_rw(void *opaque
)
1323 RawPosixAIOData
*aiocb
= opaque
;
1327 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1329 * If there is just a single buffer, and it is properly aligned
1330 * we can just use plain pread/pwrite without any problems.
1332 if (aiocb
->io
.niov
== 1) {
1333 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1337 * We have more than one iovec, and all are properly aligned.
1339 * Try preadv/pwritev first and fall back to linearizing the
1340 * buffer if it's not supported.
1342 if (preadv_present
) {
1343 nbytes
= handle_aiocb_rw_vector(aiocb
);
1344 if (nbytes
== aiocb
->aio_nbytes
||
1345 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1348 preadv_present
= false;
1352 * XXX(hch): short read/write. no easy way to handle the reminder
1353 * using these interfaces. For now retry using plain
1359 * Ok, we have to do it the hard way, copy all segments into
1360 * a single aligned buffer.
1362 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1368 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1372 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1373 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1374 p
+= aiocb
->io
.iov
[i
].iov_len
;
1376 assert(p
- buf
== aiocb
->aio_nbytes
);
1379 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1380 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1382 size_t count
= aiocb
->aio_nbytes
, copy
;
1385 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1387 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1388 copy
= aiocb
->io
.iov
[i
].iov_len
;
1390 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1391 assert(count
>= copy
);
1400 if (nbytes
== aiocb
->aio_nbytes
) {
1402 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1403 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1406 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1407 0, aiocb
->aio_nbytes
- nbytes
);
1417 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1419 struct xfs_flock64 fl
;
1422 memset(&fl
, 0, sizeof(fl
));
1423 fl
.l_whence
= SEEK_SET
;
1424 fl
.l_start
= offset
;
1427 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1429 trace_file_xfs_write_zeroes(strerror(errno
));
1436 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1438 struct xfs_flock64 fl
;
1441 memset(&fl
, 0, sizeof(fl
));
1442 fl
.l_whence
= SEEK_SET
;
1443 fl
.l_start
= offset
;
1446 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1448 trace_file_xfs_discard(strerror(errno
));
1456 static int translate_err(int err
)
1458 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1465 #ifdef CONFIG_FALLOCATE
1466 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1469 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1472 } while (errno
== EINTR
);
1473 return translate_err(-errno
);
1477 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1480 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1482 if (!s
->has_write_zeroes
) {
1488 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1489 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1492 } while (errno
== EINTR
);
1494 ret
= translate_err(-errno
);
1497 if (ret
== -ENOTSUP
) {
1498 s
->has_write_zeroes
= false;
1503 static int handle_aiocb_write_zeroes(void *opaque
)
1505 RawPosixAIOData
*aiocb
= opaque
;
1506 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1507 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1509 #ifdef CONFIG_FALLOCATE
1513 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1514 return handle_aiocb_write_zeroes_block(aiocb
);
1519 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1523 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1524 if (s
->has_write_zeroes
) {
1525 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1526 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1527 if (ret
== 0 || ret
!= -ENOTSUP
) {
1530 s
->has_write_zeroes
= false;
1534 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1535 if (s
->has_discard
&& s
->has_fallocate
) {
1536 int ret
= do_fallocate(s
->fd
,
1537 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1538 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1540 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1541 if (ret
== 0 || ret
!= -ENOTSUP
) {
1544 s
->has_fallocate
= false;
1545 } else if (ret
!= -ENOTSUP
) {
1548 s
->has_discard
= false;
1553 #ifdef CONFIG_FALLOCATE
1554 /* Last resort: we are trying to extend the file with zeroed data. This
1555 * can be done via fallocate(fd, 0) */
1556 len
= bdrv_getlength(aiocb
->bs
);
1557 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1558 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1559 if (ret
== 0 || ret
!= -ENOTSUP
) {
1562 s
->has_fallocate
= false;
1569 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1571 RawPosixAIOData
*aiocb
= opaque
;
1572 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1575 /* First try to write zeros and unmap at the same time */
1577 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1578 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1579 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1580 if (ret
!= -ENOTSUP
) {
1587 /* xfs_discard() guarantees that the discarded area reads as all-zero
1588 * afterwards, so we can use it here. */
1589 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1593 /* If we couldn't manage to unmap while guaranteed that the area reads as
1594 * all-zero afterwards, just write zeroes without unmapping */
1595 ret
= handle_aiocb_write_zeroes(aiocb
);
1599 #ifndef HAVE_COPY_FILE_RANGE
1600 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1601 off_t
*out_off
, size_t len
, unsigned int flags
)
1603 #ifdef __NR_copy_file_range
1604 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1605 out_off
, len
, flags
);
1613 static int handle_aiocb_copy_range(void *opaque
)
1615 RawPosixAIOData
*aiocb
= opaque
;
1616 uint64_t bytes
= aiocb
->aio_nbytes
;
1617 off_t in_off
= aiocb
->aio_offset
;
1618 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1621 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1622 aiocb
->copy_range
.aio_fd2
, &out_off
,
1624 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1625 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1628 /* No progress (e.g. when beyond EOF), let the caller fall back to
1647 static int handle_aiocb_discard(void *opaque
)
1649 RawPosixAIOData
*aiocb
= opaque
;
1650 int ret
= -EOPNOTSUPP
;
1651 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1653 if (!s
->has_discard
) {
1657 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1660 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1661 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1664 } while (errno
== EINTR
);
1671 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1675 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1676 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1677 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1681 ret
= translate_err(ret
);
1682 if (ret
== -ENOTSUP
) {
1683 s
->has_discard
= false;
1688 static int handle_aiocb_truncate(void *opaque
)
1690 RawPosixAIOData
*aiocb
= opaque
;
1692 int64_t current_length
= 0;
1695 int fd
= aiocb
->aio_fildes
;
1696 int64_t offset
= aiocb
->aio_offset
;
1697 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1698 Error
**errp
= aiocb
->truncate
.errp
;
1700 if (fstat(fd
, &st
) < 0) {
1702 error_setg_errno(errp
, -result
, "Could not stat file");
1706 current_length
= st
.st_size
;
1707 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1708 error_setg(errp
, "Cannot use preallocation for shrinking files");
1713 #ifdef CONFIG_POSIX_FALLOCATE
1714 case PREALLOC_MODE_FALLOC
:
1716 * Truncating before posix_fallocate() makes it about twice slower on
1717 * file systems that do not support fallocate(), trying to check if a
1718 * block is allocated before allocating it, so don't do that here.
1720 if (offset
!= current_length
) {
1721 result
= -posix_fallocate(fd
, current_length
,
1722 offset
- current_length
);
1724 /* posix_fallocate() doesn't set errno. */
1725 error_setg_errno(errp
, -result
,
1726 "Could not preallocate new data");
1733 case PREALLOC_MODE_FULL
:
1735 int64_t num
= 0, left
= offset
- current_length
;
1739 * Knowing the final size from the beginning could allow the file
1740 * system driver to do less allocations and possibly avoid
1741 * fragmentation of the file.
1743 if (ftruncate(fd
, offset
) != 0) {
1745 error_setg_errno(errp
, -result
, "Could not resize file");
1749 buf
= g_malloc0(65536);
1751 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1752 if (seek_result
< 0) {
1754 error_setg_errno(errp
, -result
,
1755 "Failed to seek to the old end of file");
1760 num
= MIN(left
, 65536);
1761 result
= write(fd
, buf
, num
);
1763 if (errno
== EINTR
) {
1767 error_setg_errno(errp
, -result
,
1768 "Could not write zeros for preallocation");
1777 error_setg_errno(errp
, -result
,
1778 "Could not flush file to disk");
1784 case PREALLOC_MODE_OFF
:
1785 if (ftruncate(fd
, offset
) != 0) {
1787 error_setg_errno(errp
, -result
, "Could not resize file");
1792 error_setg(errp
, "Unsupported preallocation mode: %s",
1793 PreallocMode_str(prealloc
));
1799 if (ftruncate(fd
, current_length
) < 0) {
1800 error_report("Failed to restore old file length: %s",
1809 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1810 ThreadPoolFunc func
, void *arg
)
1812 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1813 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1814 return thread_pool_submit_co(pool
, func
, arg
);
1817 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1818 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1820 BDRVRawState
*s
= bs
->opaque
;
1821 RawPosixAIOData acb
;
1823 if (fd_open(bs
) < 0)
1827 * Check if the underlying device requires requests to be aligned,
1828 * and if the request we are trying to submit is aligned or not.
1829 * If this is the case tell the low-level driver that it needs
1830 * to copy the buffer.
1832 if (s
->needs_alignment
) {
1833 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1834 type
|= QEMU_AIO_MISALIGNED
;
1835 #ifdef CONFIG_LINUX_AIO
1836 } else if (s
->use_linux_aio
) {
1837 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1838 assert(qiov
->size
== bytes
);
1839 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1844 acb
= (RawPosixAIOData
) {
1846 .aio_fildes
= s
->fd
,
1848 .aio_offset
= offset
,
1849 .aio_nbytes
= bytes
,
1856 assert(qiov
->size
== bytes
);
1857 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1860 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1861 uint64_t bytes
, QEMUIOVector
*qiov
,
1864 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1867 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1868 uint64_t bytes
, QEMUIOVector
*qiov
,
1872 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1875 static void raw_aio_plug(BlockDriverState
*bs
)
1877 #ifdef CONFIG_LINUX_AIO
1878 BDRVRawState
*s
= bs
->opaque
;
1879 if (s
->use_linux_aio
) {
1880 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1881 laio_io_plug(bs
, aio
);
1886 static void raw_aio_unplug(BlockDriverState
*bs
)
1888 #ifdef CONFIG_LINUX_AIO
1889 BDRVRawState
*s
= bs
->opaque
;
1890 if (s
->use_linux_aio
) {
1891 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1892 laio_io_unplug(bs
, aio
);
1897 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1899 BDRVRawState
*s
= bs
->opaque
;
1900 RawPosixAIOData acb
;
1908 acb
= (RawPosixAIOData
) {
1910 .aio_fildes
= s
->fd
,
1911 .aio_type
= QEMU_AIO_FLUSH
,
1914 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1917 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1918 AioContext
*new_context
)
1920 #ifdef CONFIG_LINUX_AIO
1921 BDRVRawState
*s
= bs
->opaque
;
1922 if (s
->use_linux_aio
) {
1924 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1925 error_reportf_err(local_err
, "Unable to use native AIO, "
1926 "falling back to thread pool: ");
1927 s
->use_linux_aio
= false;
1933 static void raw_close(BlockDriverState
*bs
)
1935 BDRVRawState
*s
= bs
->opaque
;
1944 * Truncates the given regular file @fd to @offset and, when growing, fills the
1945 * new space according to @prealloc.
1947 * Returns: 0 on success, -errno on failure.
1949 static int coroutine_fn
1950 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1951 PreallocMode prealloc
, Error
**errp
)
1953 RawPosixAIOData acb
;
1955 acb
= (RawPosixAIOData
) {
1958 .aio_type
= QEMU_AIO_TRUNCATE
,
1959 .aio_offset
= offset
,
1961 .prealloc
= prealloc
,
1966 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
1969 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1970 PreallocMode prealloc
, Error
**errp
)
1972 BDRVRawState
*s
= bs
->opaque
;
1976 if (fstat(s
->fd
, &st
)) {
1978 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1982 if (S_ISREG(st
.st_mode
)) {
1983 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
1986 if (prealloc
!= PREALLOC_MODE_OFF
) {
1987 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1988 "non-regular file", PreallocMode_str(prealloc
));
1992 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1993 if (offset
> raw_getlength(bs
)) {
1994 error_setg(errp
, "Cannot grow device files");
1998 error_setg(errp
, "Resizing this file is not supported");
2006 static int64_t raw_getlength(BlockDriverState
*bs
)
2008 BDRVRawState
*s
= bs
->opaque
;
2014 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2015 struct disklabel dl
;
2017 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2019 return (uint64_t)dl
.d_secsize
*
2020 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2024 #elif defined(__NetBSD__)
2025 static int64_t raw_getlength(BlockDriverState
*bs
)
2027 BDRVRawState
*s
= bs
->opaque
;
2033 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2034 struct dkwedge_info dkw
;
2036 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2037 return dkw
.dkw_size
* 512;
2039 struct disklabel dl
;
2041 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2043 return (uint64_t)dl
.d_secsize
*
2044 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2049 #elif defined(__sun__)
2050 static int64_t raw_getlength(BlockDriverState
*bs
)
2052 BDRVRawState
*s
= bs
->opaque
;
2053 struct dk_minfo minfo
;
2063 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2065 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2067 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2071 * There are reports that lseek on some devices fails, but
2072 * irc discussion said that contingency on contingency was overkill.
2074 size
= lseek(s
->fd
, 0, SEEK_END
);
2080 #elif defined(CONFIG_BSD)
2081 static int64_t raw_getlength(BlockDriverState
*bs
)
2083 BDRVRawState
*s
= bs
->opaque
;
2087 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2096 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2099 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2100 #ifdef DIOCGMEDIASIZE
2101 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2102 #elif defined(DIOCGPART)
2105 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2106 size
= pi
.media_size
;
2112 #if defined(__APPLE__) && defined(__MACH__)
2114 uint64_t sectors
= 0;
2115 uint32_t sector_size
= 0;
2117 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2118 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2119 size
= sectors
* sector_size
;
2121 size
= lseek(fd
, 0LL, SEEK_END
);
2128 size
= lseek(fd
, 0LL, SEEK_END
);
2133 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2136 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2137 if (size
== 2048LL * (unsigned)-1)
2139 /* XXX no disc? maybe we need to reopen... */
2140 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2147 size
= lseek(fd
, 0, SEEK_END
);
2155 static int64_t raw_getlength(BlockDriverState
*bs
)
2157 BDRVRawState
*s
= bs
->opaque
;
2166 size
= lseek(s
->fd
, 0, SEEK_END
);
2174 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2177 BDRVRawState
*s
= bs
->opaque
;
2179 if (fstat(s
->fd
, &st
) < 0) {
2182 return (int64_t)st
.st_blocks
* 512;
2185 static int coroutine_fn
2186 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2188 BlockdevCreateOptionsFile
*file_opts
;
2189 Error
*local_err
= NULL
;
2191 uint64_t perm
, shared
;
2194 /* Validate options and set default values */
2195 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2196 file_opts
= &options
->u
.file
;
2198 if (!file_opts
->has_nocow
) {
2199 file_opts
->nocow
= false;
2201 if (!file_opts
->has_preallocation
) {
2202 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2206 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2209 error_setg_errno(errp
, -result
, "Could not create file");
2213 /* Take permissions: We want to discard everything, so we need
2214 * BLK_PERM_WRITE; and truncation to the desired size requires
2216 * On the other hand, we cannot share the RESIZE permission
2217 * because we promise that after this function, the file has the
2218 * size given in the options. If someone else were to resize it
2219 * concurrently, we could not guarantee that.
2220 * Note that after this function, we can no longer guarantee that
2221 * the file is not touched by a third party, so it may be resized
2223 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2224 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2226 /* Step one: Take locks */
2227 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2232 /* Step two: Check that nobody else has taken conflicting locks */
2233 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2235 error_append_hint(errp
,
2236 "Is another process using the image [%s]?\n",
2237 file_opts
->filename
);
2241 /* Clear the file by truncating it to 0 */
2242 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2247 if (file_opts
->nocow
) {
2249 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2250 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2251 * will be ignored since any failure of this operation should not
2252 * block the left work.
2255 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2256 attr
|= FS_NOCOW_FL
;
2257 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2262 /* Resize and potentially preallocate the file to the desired
2264 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2265 file_opts
->preallocation
, errp
);
2271 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2273 /* The above call should not fail, and if it does, that does
2274 * not mean the whole creation operation has failed. So
2275 * report it the user for their convenience, but do not report
2276 * it to the caller. */
2277 warn_report_err(local_err
);
2281 if (qemu_close(fd
) != 0 && result
== 0) {
2283 error_setg_errno(errp
, -result
, "Could not close the new file");
2289 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2292 BlockdevCreateOptions options
;
2293 int64_t total_size
= 0;
2295 PreallocMode prealloc
;
2297 Error
*local_err
= NULL
;
2299 /* Skip file: protocol prefix */
2300 strstart(filename
, "file:", &filename
);
2302 /* Read out options */
2303 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2305 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2306 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2307 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2308 PREALLOC_MODE_OFF
, &local_err
);
2311 error_propagate(errp
, local_err
);
2315 options
= (BlockdevCreateOptions
) {
2316 .driver
= BLOCKDEV_DRIVER_FILE
,
2318 .filename
= (char *) filename
,
2320 .has_preallocation
= true,
2321 .preallocation
= prealloc
,
2326 return raw_co_create(&options
, errp
);
2330 * Find allocation range in @bs around offset @start.
2331 * May change underlying file descriptor's file offset.
2332 * If @start is not in a hole, store @start in @data, and the
2333 * beginning of the next hole in @hole, and return 0.
2334 * If @start is in a non-trailing hole, store @start in @hole and the
2335 * beginning of the next non-hole in @data, and return 0.
2336 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2337 * If we can't find out, return a negative errno other than -ENXIO.
2339 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2340 off_t
*data
, off_t
*hole
)
2342 #if defined SEEK_HOLE && defined SEEK_DATA
2343 BDRVRawState
*s
= bs
->opaque
;
2348 * D1. offs == start: start is in data
2349 * D2. offs > start: start is in a hole, next data at offs
2350 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2351 * or start is beyond EOF
2352 * If the latter happens, the file has been truncated behind
2353 * our back since we opened it. All bets are off then.
2354 * Treating like a trailing hole is simplest.
2355 * D4. offs < 0, errno != ENXIO: we learned nothing
2357 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2359 return -errno
; /* D3 or D4 */
2363 /* This is not a valid return by lseek(). We are safe to just return
2364 * -EIO in this case, and we'll treat it like D4. */
2369 /* D2: in hole, next data at offs */
2375 /* D1: in data, end not yet known */
2379 * H1. offs == start: start is in a hole
2380 * If this happens here, a hole has been dug behind our back
2381 * since the previous lseek().
2382 * H2. offs > start: either start is in data, next hole at offs,
2383 * or start is in trailing hole, EOF at offs
2384 * Linux treats trailing holes like any other hole: offs ==
2385 * start. Solaris seeks to EOF instead: offs > start (blech).
2386 * If that happens here, a hole has been dug behind our back
2387 * since the previous lseek().
2388 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2389 * If this happens, the file has been truncated behind our
2390 * back since we opened it. Treat it like a trailing hole.
2391 * H4. offs < 0, errno != ENXIO: we learned nothing
2392 * Pretend we know nothing at all, i.e. "forget" about D1.
2394 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2396 return -errno
; /* D1 and (H3 or H4) */
2400 /* This is not a valid return by lseek(). We are safe to just return
2401 * -EIO in this case, and we'll treat it like H4. */
2407 * D1 and H2: either in data, next hole at offs, or it was in
2408 * data but is now in a trailing hole. In the latter case,
2409 * all bets are off. Treating it as if it there was data all
2410 * the way to EOF is safe, so simply do that.
2425 * Returns the allocation status of the specified offset.
2427 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2429 * 'pnum' is set to the number of bytes (including and immediately following
2430 * the specified offset) that are known to be in the same
2431 * allocated/unallocated state.
2433 * 'bytes' is the max value 'pnum' should be set to.
2435 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2438 int64_t bytes
, int64_t *pnum
,
2440 BlockDriverState
**file
)
2442 off_t data
= 0, hole
= 0;
2454 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2457 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2458 if (ret
== -ENXIO
) {
2461 ret
= BDRV_BLOCK_ZERO
;
2462 } else if (ret
< 0) {
2463 /* No info available, so pretend there are no holes */
2465 ret
= BDRV_BLOCK_DATA
;
2466 } else if (data
== offset
) {
2467 /* On a data extent, compute bytes to the end of the extent,
2468 * possibly including a partial sector at EOF. */
2469 *pnum
= MIN(bytes
, hole
- offset
);
2470 ret
= BDRV_BLOCK_DATA
;
2472 /* On a hole, compute bytes to the beginning of the next extent. */
2473 assert(hole
== offset
);
2474 *pnum
= MIN(bytes
, data
- offset
);
2475 ret
= BDRV_BLOCK_ZERO
;
2479 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2482 #if defined(__linux__)
2483 /* Verify that the file is not in the page cache */
2484 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2486 const size_t window_size
= 128 * 1024 * 1024;
2487 BDRVRawState
*s
= bs
->opaque
;
2488 void *window
= NULL
;
2495 /* mincore(2) page status information requires 1 byte per page */
2496 page_size
= sysconf(_SC_PAGESIZE
);
2497 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2499 end
= raw_getlength(bs
);
2501 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2508 /* Unmap previous window if size has changed */
2509 new_length
= MIN(end
- offset
, window_size
);
2510 if (new_length
!= length
) {
2511 munmap(window
, length
);
2516 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2518 if (new_window
== MAP_FAILED
) {
2519 error_setg_errno(errp
, errno
, "mmap failed");
2523 window
= new_window
;
2524 length
= new_length
;
2526 ret
= mincore(window
, length
, vec
);
2528 error_setg_errno(errp
, errno
, "mincore failed");
2532 vec_end
= DIV_ROUND_UP(length
, page_size
);
2533 for (i
= 0; i
< vec_end
; i
++) {
2535 error_setg(errp
, "page cache still in use!");
2542 munmap(window
, length
);
2547 #endif /* __linux__ */
2549 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2552 BDRVRawState
*s
= bs
->opaque
;
2557 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2561 if (s
->open_flags
& O_DIRECT
) {
2562 return; /* No host kernel page cache */
2565 #if defined(__linux__)
2566 /* This sets the scene for the next syscall... */
2567 ret
= bdrv_co_flush(bs
);
2569 error_setg_errno(errp
, -ret
, "flush failed");
2573 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2574 * process. These limitations are okay because we just fsynced the file,
2575 * we don't use mmap, and the file should not be in use by other processes.
2577 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2578 if (ret
!= 0) { /* the return value is a positive errno */
2579 error_setg_errno(errp
, ret
, "fadvise failed");
2583 if (s
->check_cache_dropped
) {
2584 check_cache_dropped(bs
, errp
);
2586 #else /* __linux__ */
2587 /* Do nothing. Live migration to a remote host with cache.direct=off is
2588 * unsupported on other host operating systems. Cache consistency issues
2589 * may occur but no error is reported here, partly because that's the
2590 * historical behavior and partly because it's hard to differentiate valid
2591 * configurations that should not cause errors.
2593 #endif /* !__linux__ */
2596 static coroutine_fn
int
2597 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2599 BDRVRawState
*s
= bs
->opaque
;
2600 RawPosixAIOData acb
;
2602 acb
= (RawPosixAIOData
) {
2604 .aio_fildes
= s
->fd
,
2605 .aio_type
= QEMU_AIO_DISCARD
,
2606 .aio_offset
= offset
,
2607 .aio_nbytes
= bytes
,
2611 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2614 return raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2617 static coroutine_fn
int
2618 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2620 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2623 static int coroutine_fn
2624 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2625 BdrvRequestFlags flags
, bool blkdev
)
2627 BDRVRawState
*s
= bs
->opaque
;
2628 RawPosixAIOData acb
;
2629 ThreadPoolFunc
*handler
;
2631 acb
= (RawPosixAIOData
) {
2633 .aio_fildes
= s
->fd
,
2634 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2635 .aio_offset
= offset
,
2636 .aio_nbytes
= bytes
,
2640 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2643 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2644 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2645 handler
= handle_aiocb_write_zeroes_unmap
;
2647 handler
= handle_aiocb_write_zeroes
;
2650 return raw_thread_pool_submit(bs
, handler
, &acb
);
2653 static int coroutine_fn
raw_co_pwrite_zeroes(
2654 BlockDriverState
*bs
, int64_t offset
,
2655 int bytes
, BdrvRequestFlags flags
)
2657 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2660 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2662 BDRVRawState
*s
= bs
->opaque
;
2664 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2668 static QemuOptsList raw_create_opts
= {
2669 .name
= "raw-create-opts",
2670 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2673 .name
= BLOCK_OPT_SIZE
,
2674 .type
= QEMU_OPT_SIZE
,
2675 .help
= "Virtual disk size"
2678 .name
= BLOCK_OPT_NOCOW
,
2679 .type
= QEMU_OPT_BOOL
,
2680 .help
= "Turn off copy-on-write (valid only on btrfs)"
2683 .name
= BLOCK_OPT_PREALLOC
,
2684 .type
= QEMU_OPT_STRING
,
2685 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2687 { /* end of list */ }
2691 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2694 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2697 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2699 BDRVRawState
*s
= bs
->opaque
;
2700 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2702 s
->shared_perm
= shared
;
2705 static void raw_abort_perm_update(BlockDriverState
*bs
)
2707 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2710 static int coroutine_fn
raw_co_copy_range_from(
2711 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2712 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2713 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2715 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2716 read_flags
, write_flags
);
2719 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2721 uint64_t src_offset
,
2723 uint64_t dst_offset
,
2725 BdrvRequestFlags read_flags
,
2726 BdrvRequestFlags write_flags
)
2728 RawPosixAIOData acb
;
2729 BDRVRawState
*s
= bs
->opaque
;
2730 BDRVRawState
*src_s
;
2732 assert(dst
->bs
== bs
);
2733 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2737 src_s
= src
->bs
->opaque
;
2738 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2742 acb
= (RawPosixAIOData
) {
2744 .aio_type
= QEMU_AIO_COPY_RANGE
,
2745 .aio_fildes
= src_s
->fd
,
2746 .aio_offset
= src_offset
,
2747 .aio_nbytes
= bytes
,
2750 .aio_offset2
= dst_offset
,
2754 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2757 BlockDriver bdrv_file
= {
2758 .format_name
= "file",
2759 .protocol_name
= "file",
2760 .instance_size
= sizeof(BDRVRawState
),
2761 .bdrv_needs_filename
= true,
2762 .bdrv_probe
= NULL
, /* no probe for protocols */
2763 .bdrv_parse_filename
= raw_parse_filename
,
2764 .bdrv_file_open
= raw_open
,
2765 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2766 .bdrv_reopen_commit
= raw_reopen_commit
,
2767 .bdrv_reopen_abort
= raw_reopen_abort
,
2768 .bdrv_close
= raw_close
,
2769 .bdrv_co_create
= raw_co_create
,
2770 .bdrv_co_create_opts
= raw_co_create_opts
,
2771 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2772 .bdrv_co_block_status
= raw_co_block_status
,
2773 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2774 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2776 .bdrv_co_preadv
= raw_co_preadv
,
2777 .bdrv_co_pwritev
= raw_co_pwritev
,
2778 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2779 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2780 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2781 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2782 .bdrv_refresh_limits
= raw_refresh_limits
,
2783 .bdrv_io_plug
= raw_aio_plug
,
2784 .bdrv_io_unplug
= raw_aio_unplug
,
2785 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2787 .bdrv_co_truncate
= raw_co_truncate
,
2788 .bdrv_getlength
= raw_getlength
,
2789 .bdrv_get_info
= raw_get_info
,
2790 .bdrv_get_allocated_file_size
2791 = raw_get_allocated_file_size
,
2792 .bdrv_check_perm
= raw_check_perm
,
2793 .bdrv_set_perm
= raw_set_perm
,
2794 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2795 .create_opts
= &raw_create_opts
,
2798 /***********************************************/
2801 #if defined(__APPLE__) && defined(__MACH__)
2802 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2803 CFIndex maxPathSize
, int flags
);
2804 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2806 kern_return_t kernResult
= KERN_FAILURE
;
2807 mach_port_t masterPort
;
2808 CFMutableDictionaryRef classesToMatch
;
2809 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2810 char *mediaType
= NULL
;
2812 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2813 if ( KERN_SUCCESS
!= kernResult
) {
2814 printf( "IOMasterPort returned %d\n", kernResult
);
2818 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2819 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2820 if (classesToMatch
== NULL
) {
2821 error_report("IOServiceMatching returned NULL for %s",
2822 matching_array
[index
]);
2825 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2827 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2829 if (kernResult
!= KERN_SUCCESS
) {
2830 error_report("Note: IOServiceGetMatchingServices returned %d",
2835 /* If a match was found, leave the loop */
2836 if (*mediaIterator
!= 0) {
2837 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
2838 mediaType
= g_strdup(matching_array
[index
]);
2845 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2846 CFIndex maxPathSize
, int flags
)
2848 io_object_t nextMedia
;
2849 kern_return_t kernResult
= KERN_FAILURE
;
2851 nextMedia
= IOIteratorNext( mediaIterator
);
2854 CFTypeRef bsdPathAsCFString
;
2855 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2856 if ( bsdPathAsCFString
) {
2857 size_t devPathLength
;
2858 strcpy( bsdPath
, _PATH_DEV
);
2859 if (flags
& BDRV_O_NOCACHE
) {
2860 strcat(bsdPath
, "r");
2862 devPathLength
= strlen( bsdPath
);
2863 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2864 kernResult
= KERN_SUCCESS
;
2866 CFRelease( bsdPathAsCFString
);
2868 IOObjectRelease( nextMedia
);
2874 /* Sets up a real cdrom for use in QEMU */
2875 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2877 int index
, num_of_test_partitions
= 2, fd
;
2878 char test_partition
[MAXPATHLEN
];
2879 bool partition_found
= false;
2881 /* look for a working partition */
2882 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2883 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2885 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2887 partition_found
= true;
2893 /* if a working partition on the device was not found */
2894 if (partition_found
== false) {
2895 error_setg(errp
, "Failed to find a working partition on disc");
2897 trace_file_setup_cdrom(test_partition
);
2898 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2900 return partition_found
;
2903 /* Prints directions on mounting and unmounting a device */
2904 static void print_unmounting_directions(const char *file_name
)
2906 error_report("If device %s is mounted on the desktop, unmount"
2907 " it first before using it in QEMU", file_name
);
2908 error_report("Command to unmount device: diskutil unmountDisk %s",
2910 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2913 #endif /* defined(__APPLE__) && defined(__MACH__) */
2915 static int hdev_probe_device(const char *filename
)
2919 /* allow a dedicated CD-ROM driver to match with a higher priority */
2920 if (strstart(filename
, "/dev/cdrom", NULL
))
2923 if (stat(filename
, &st
) >= 0 &&
2924 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2931 static int check_hdev_writable(BDRVRawState
*s
)
2933 #if defined(BLKROGET)
2934 /* Linux block devices can be configured "read-only" using blockdev(8).
2935 * This is independent of device node permissions and therefore open(2)
2936 * with O_RDWR succeeds. Actual writes fail with EPERM.
2938 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2939 * check for read-only block devices so that Linux block devices behave
2945 if (fstat(s
->fd
, &st
)) {
2949 if (!S_ISBLK(st
.st_mode
)) {
2953 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2960 #endif /* defined(BLKROGET) */
2964 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2967 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2970 static bool hdev_is_sg(BlockDriverState
*bs
)
2973 #if defined(__linux__)
2975 BDRVRawState
*s
= bs
->opaque
;
2977 struct sg_scsi_id scsiid
;
2981 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2985 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2990 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2992 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3001 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3004 BDRVRawState
*s
= bs
->opaque
;
3005 Error
*local_err
= NULL
;
3008 #if defined(__APPLE__) && defined(__MACH__)
3010 * Caution: while qdict_get_str() is fine, getting non-string types
3011 * would require more care. When @options come from -blockdev or
3012 * blockdev_add, its members are typed according to the QAPI
3013 * schema, but when they come from -drive, they're all QString.
3015 const char *filename
= qdict_get_str(options
, "filename");
3016 char bsd_path
[MAXPATHLEN
] = "";
3017 bool error_occurred
= false;
3019 /* If using a real cdrom */
3020 if (strcmp(filename
, "/dev/cdrom") == 0) {
3021 char *mediaType
= NULL
;
3022 kern_return_t ret_val
;
3023 io_iterator_t mediaIterator
= 0;
3025 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3026 if (mediaType
== NULL
) {
3027 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3029 error_occurred
= true;
3030 goto hdev_open_Mac_error
;
3033 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3034 if (ret_val
!= KERN_SUCCESS
) {
3035 error_setg(errp
, "Could not get BSD path for optical drive");
3036 error_occurred
= true;
3037 goto hdev_open_Mac_error
;
3040 /* If a real optical drive was not found */
3041 if (bsd_path
[0] == '\0') {
3042 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3043 error_occurred
= true;
3044 goto hdev_open_Mac_error
;
3047 /* If using a cdrom disc and finding a partition on the disc failed */
3048 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3049 setup_cdrom(bsd_path
, errp
) == false) {
3050 print_unmounting_directions(bsd_path
);
3051 error_occurred
= true;
3052 goto hdev_open_Mac_error
;
3055 qdict_put_str(options
, "filename", bsd_path
);
3057 hdev_open_Mac_error
:
3059 if (mediaIterator
) {
3060 IOObjectRelease(mediaIterator
);
3062 if (error_occurred
) {
3066 #endif /* defined(__APPLE__) && defined(__MACH__) */
3068 s
->type
= FTYPE_FILE
;
3070 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3072 error_propagate(errp
, local_err
);
3073 #if defined(__APPLE__) && defined(__MACH__)
3075 filename
= bsd_path
;
3077 /* if a physical device experienced an error while being opened */
3078 if (strncmp(filename
, "/dev/", 5) == 0) {
3079 print_unmounting_directions(filename
);
3081 #endif /* defined(__APPLE__) && defined(__MACH__) */
3085 /* Since this does ioctl the device must be already opened */
3086 bs
->sg
= hdev_is_sg(bs
);
3088 if (flags
& BDRV_O_RDWR
) {
3089 ret
= check_hdev_writable(s
);
3092 error_setg_errno(errp
, -ret
, "The device is not writable");
3100 #if defined(__linux__)
3101 static int coroutine_fn
3102 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3104 BDRVRawState
*s
= bs
->opaque
;
3105 RawPosixAIOData acb
;
3113 if (req
== SG_IO
&& s
->pr_mgr
) {
3114 struct sg_io_hdr
*io_hdr
= buf
;
3115 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3116 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3117 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3122 acb
= (RawPosixAIOData
) {
3124 .aio_type
= QEMU_AIO_IOCTL
,
3125 .aio_fildes
= s
->fd
,
3133 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3137 static int fd_open(BlockDriverState
*bs
)
3139 BDRVRawState
*s
= bs
->opaque
;
3141 /* this is just to ensure s->fd is sane (its called by io ops) */
3147 static coroutine_fn
int
3148 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3156 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3159 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3160 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3169 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3172 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3177 struct stat stat_buf
;
3178 int64_t total_size
= 0;
3181 /* This function is used by both protocol block drivers and therefore either
3182 * of these prefixes may be given.
3183 * The return value has to be stored somewhere, otherwise this is an error
3184 * due to -Werror=unused-value. */
3186 strstart(filename
, "host_device:", &filename
) ||
3187 strstart(filename
, "host_cdrom:" , &filename
);
3191 ret
= raw_normalize_devicepath(&filename
, errp
);
3196 /* Read out options */
3197 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3200 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3203 error_setg_errno(errp
, -ret
, "Could not open device");
3207 if (fstat(fd
, &stat_buf
) < 0) {
3209 error_setg_errno(errp
, -ret
, "Could not stat device");
3210 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3212 "The given file is neither a block nor a character device");
3214 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3215 error_setg(errp
, "Device is too small");
3219 if (!ret
&& total_size
) {
3220 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3221 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3222 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3225 ret
= qemu_write_full(fd
, buf
, zero_size
);
3226 ret
= ret
== zero_size
? 0 : -errno
;
3233 static BlockDriver bdrv_host_device
= {
3234 .format_name
= "host_device",
3235 .protocol_name
= "host_device",
3236 .instance_size
= sizeof(BDRVRawState
),
3237 .bdrv_needs_filename
= true,
3238 .bdrv_probe_device
= hdev_probe_device
,
3239 .bdrv_parse_filename
= hdev_parse_filename
,
3240 .bdrv_file_open
= hdev_open
,
3241 .bdrv_close
= raw_close
,
3242 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3243 .bdrv_reopen_commit
= raw_reopen_commit
,
3244 .bdrv_reopen_abort
= raw_reopen_abort
,
3245 .bdrv_co_create_opts
= hdev_co_create_opts
,
3246 .create_opts
= &raw_create_opts
,
3247 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3248 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3250 .bdrv_co_preadv
= raw_co_preadv
,
3251 .bdrv_co_pwritev
= raw_co_pwritev
,
3252 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3253 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3254 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3255 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3256 .bdrv_refresh_limits
= raw_refresh_limits
,
3257 .bdrv_io_plug
= raw_aio_plug
,
3258 .bdrv_io_unplug
= raw_aio_unplug
,
3259 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3261 .bdrv_co_truncate
= raw_co_truncate
,
3262 .bdrv_getlength
= raw_getlength
,
3263 .bdrv_get_info
= raw_get_info
,
3264 .bdrv_get_allocated_file_size
3265 = raw_get_allocated_file_size
,
3266 .bdrv_check_perm
= raw_check_perm
,
3267 .bdrv_set_perm
= raw_set_perm
,
3268 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3269 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3270 .bdrv_probe_geometry
= hdev_probe_geometry
,
3272 /* generic scsi device */
3274 .bdrv_co_ioctl
= hdev_co_ioctl
,
3278 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3279 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3282 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3287 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3290 BDRVRawState
*s
= bs
->opaque
;
3294 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3295 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3298 static int cdrom_probe_device(const char *filename
)
3304 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3308 ret
= fstat(fd
, &st
);
3309 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3313 /* Attempt to detect via a CDROM specific ioctl */
3314 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3324 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3326 BDRVRawState
*s
= bs
->opaque
;
3329 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3330 return ret
== CDS_DISC_OK
;
3333 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3335 BDRVRawState
*s
= bs
->opaque
;
3338 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3339 perror("CDROMEJECT");
3341 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3342 perror("CDROMEJECT");
3346 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3348 BDRVRawState
*s
= bs
->opaque
;
3350 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3352 * Note: an error can happen if the distribution automatically
3355 /* perror("CDROM_LOCKDOOR"); */
3359 static BlockDriver bdrv_host_cdrom
= {
3360 .format_name
= "host_cdrom",
3361 .protocol_name
= "host_cdrom",
3362 .instance_size
= sizeof(BDRVRawState
),
3363 .bdrv_needs_filename
= true,
3364 .bdrv_probe_device
= cdrom_probe_device
,
3365 .bdrv_parse_filename
= cdrom_parse_filename
,
3366 .bdrv_file_open
= cdrom_open
,
3367 .bdrv_close
= raw_close
,
3368 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3369 .bdrv_reopen_commit
= raw_reopen_commit
,
3370 .bdrv_reopen_abort
= raw_reopen_abort
,
3371 .bdrv_co_create_opts
= hdev_co_create_opts
,
3372 .create_opts
= &raw_create_opts
,
3373 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3376 .bdrv_co_preadv
= raw_co_preadv
,
3377 .bdrv_co_pwritev
= raw_co_pwritev
,
3378 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3379 .bdrv_refresh_limits
= raw_refresh_limits
,
3380 .bdrv_io_plug
= raw_aio_plug
,
3381 .bdrv_io_unplug
= raw_aio_unplug
,
3382 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3384 .bdrv_co_truncate
= raw_co_truncate
,
3385 .bdrv_getlength
= raw_getlength
,
3386 .has_variable_length
= true,
3387 .bdrv_get_allocated_file_size
3388 = raw_get_allocated_file_size
,
3390 /* removable device support */
3391 .bdrv_is_inserted
= cdrom_is_inserted
,
3392 .bdrv_eject
= cdrom_eject
,
3393 .bdrv_lock_medium
= cdrom_lock_medium
,
3395 /* generic scsi device */
3396 .bdrv_co_ioctl
= hdev_co_ioctl
,
3398 #endif /* __linux__ */
3400 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3401 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3404 BDRVRawState
*s
= bs
->opaque
;
3405 Error
*local_err
= NULL
;
3410 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3412 error_propagate(errp
, local_err
);
3416 /* make sure the door isn't locked at this time */
3417 ioctl(s
->fd
, CDIOCALLOW
);
3421 static int cdrom_probe_device(const char *filename
)
3423 if (strstart(filename
, "/dev/cd", NULL
) ||
3424 strstart(filename
, "/dev/acd", NULL
))
3429 static int cdrom_reopen(BlockDriverState
*bs
)
3431 BDRVRawState
*s
= bs
->opaque
;
3435 * Force reread of possibly changed/newly loaded disc,
3436 * FreeBSD seems to not notice sometimes...
3440 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3447 /* make sure the door isn't locked at this time */
3448 ioctl(s
->fd
, CDIOCALLOW
);
3452 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3454 return raw_getlength(bs
) > 0;
3457 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3459 BDRVRawState
*s
= bs
->opaque
;
3464 (void) ioctl(s
->fd
, CDIOCALLOW
);
3467 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3468 perror("CDIOCEJECT");
3470 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3471 perror("CDIOCCLOSE");
3477 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3479 BDRVRawState
*s
= bs
->opaque
;
3483 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3485 * Note: an error can happen if the distribution automatically
3488 /* perror("CDROM_LOCKDOOR"); */
3492 static BlockDriver bdrv_host_cdrom
= {
3493 .format_name
= "host_cdrom",
3494 .protocol_name
= "host_cdrom",
3495 .instance_size
= sizeof(BDRVRawState
),
3496 .bdrv_needs_filename
= true,
3497 .bdrv_probe_device
= cdrom_probe_device
,
3498 .bdrv_parse_filename
= cdrom_parse_filename
,
3499 .bdrv_file_open
= cdrom_open
,
3500 .bdrv_close
= raw_close
,
3501 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3502 .bdrv_reopen_commit
= raw_reopen_commit
,
3503 .bdrv_reopen_abort
= raw_reopen_abort
,
3504 .bdrv_co_create_opts
= hdev_co_create_opts
,
3505 .create_opts
= &raw_create_opts
,
3507 .bdrv_co_preadv
= raw_co_preadv
,
3508 .bdrv_co_pwritev
= raw_co_pwritev
,
3509 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3510 .bdrv_refresh_limits
= raw_refresh_limits
,
3511 .bdrv_io_plug
= raw_aio_plug
,
3512 .bdrv_io_unplug
= raw_aio_unplug
,
3513 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3515 .bdrv_co_truncate
= raw_co_truncate
,
3516 .bdrv_getlength
= raw_getlength
,
3517 .has_variable_length
= true,
3518 .bdrv_get_allocated_file_size
3519 = raw_get_allocated_file_size
,
3521 /* removable device support */
3522 .bdrv_is_inserted
= cdrom_is_inserted
,
3523 .bdrv_eject
= cdrom_eject
,
3524 .bdrv_lock_medium
= cdrom_lock_medium
,
3526 #endif /* __FreeBSD__ */
3528 static void bdrv_file_init(void)
3531 * Register all the drivers. Note that order is important, the driver
3532 * registered last will get probed first.
3534 bdrv_register(&bdrv_file
);
3535 bdrv_register(&bdrv_host_device
);
3537 bdrv_register(&bdrv_host_cdrom
);
3539 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3540 bdrv_register(&bdrv_host_cdrom
);
3544 block_init(bdrv_file_init
);