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>
105 //#define DEBUG_BLOCK
108 # define DEBUG_BLOCK_PRINT 1
110 # define DEBUG_BLOCK_PRINT 0
112 #define DPRINTF(fmt, ...) \
114 if (DEBUG_BLOCK_PRINT) { \
115 printf(fmt, ## __VA_ARGS__); \
119 /* OS X does not have O_DSYNC */
122 #define O_DSYNC O_SYNC
123 #elif defined(O_FSYNC)
124 #define O_DSYNC O_FSYNC
128 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
130 #define O_DIRECT O_DSYNC
136 #define MAX_BLOCKSIZE 4096
138 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
139 * leaving a few more bytes for its future use. */
140 #define RAW_LOCK_PERM_BASE 100
141 #define RAW_LOCK_SHARED_BASE 200
143 typedef struct BDRVRawState
{
151 /* The current permissions. */
153 uint64_t shared_perm
;
159 bool has_write_zeroes
:1;
160 bool discard_zeroes
:1;
161 bool use_linux_aio
:1;
162 bool page_cache_inconsistent
:1;
164 bool needs_alignment
;
165 bool check_cache_dropped
;
170 typedef struct BDRVRawReopenState
{
173 bool check_cache_dropped
;
174 } BDRVRawReopenState
;
176 static int fd_open(BlockDriverState
*bs
);
177 static int64_t raw_getlength(BlockDriverState
*bs
);
179 typedef struct RawPosixAIOData
{
180 BlockDriverState
*bs
;
183 struct iovec
*aio_iov
;
188 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
197 PreallocMode prealloc
;
203 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
204 static int cdrom_reopen(BlockDriverState
*bs
);
207 #if defined(__NetBSD__)
208 static int raw_normalize_devicepath(const char **filename
)
210 static char namebuf
[PATH_MAX
];
211 const char *dp
, *fname
;
215 dp
= strrchr(fname
, '/');
216 if (lstat(fname
, &sb
) < 0) {
217 fprintf(stderr
, "%s: stat failed: %s\n",
218 fname
, strerror(errno
));
222 if (!S_ISBLK(sb
.st_mode
)) {
227 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
229 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
230 (int)(dp
- fname
), fname
, dp
+ 1);
232 fprintf(stderr
, "%s is a block device", fname
);
234 fprintf(stderr
, ", using %s\n", *filename
);
239 static int raw_normalize_devicepath(const char **filename
)
246 * Get logical block size via ioctl. On success store it in @sector_size_p.
248 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
250 unsigned int sector_size
;
251 bool success
= false;
255 static const unsigned long ioctl_list
[] = {
259 #ifdef DKIOCGETBLOCKSIZE
262 #ifdef DIOCGSECTORSIZE
267 /* Try a few ioctls to get the right size */
268 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
269 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
270 *sector_size_p
= sector_size
;
275 return success
? 0 : -errno
;
279 * Get physical block size of @fd.
280 * On success, store it in @blk_size and return 0.
281 * On failure, return -errno.
283 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
286 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
295 /* Check if read is allowed with given memory buffer and length.
297 * This function is used to check O_DIRECT memory buffer and request alignment.
299 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
301 ssize_t ret
= pread(fd
, buf
, len
, 0);
308 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
309 * other errors (e.g. real I/O error), which could happen on a failed
310 * drive, since we only care about probing alignment.
312 if (errno
!= EINVAL
) {
320 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
322 BDRVRawState
*s
= bs
->opaque
;
324 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
326 /* For SCSI generic devices the alignment is not really used.
327 With buffered I/O, we don't have any restrictions. */
328 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
329 bs
->bl
.request_alignment
= 1;
334 bs
->bl
.request_alignment
= 0;
336 /* Let's try to use the logical blocksize for the alignment. */
337 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
338 bs
->bl
.request_alignment
= 0;
343 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
344 bs
->bl
.request_alignment
= da
.d_miniosz
;
345 /* The kernel returns wrong information for d_mem */
346 /* s->buf_align = da.d_mem; */
351 /* If we could not get the sizes so far, we can only guess them */
354 buf
= qemu_memalign(max_align
, 2 * max_align
);
355 for (align
= 512; align
<= max_align
; align
<<= 1) {
356 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
357 s
->buf_align
= align
;
364 if (!bs
->bl
.request_alignment
) {
366 buf
= qemu_memalign(s
->buf_align
, max_align
);
367 for (align
= 512; align
<= max_align
; align
<<= 1) {
368 if (raw_is_io_aligned(fd
, buf
, align
)) {
369 bs
->bl
.request_alignment
= align
;
376 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
377 error_setg(errp
, "Could not find working O_DIRECT alignment");
378 error_append_hint(errp
, "Try cache.direct=off\n");
382 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
384 assert(open_flags
!= NULL
);
386 *open_flags
|= O_BINARY
;
387 *open_flags
&= ~O_ACCMODE
;
388 if (bdrv_flags
& BDRV_O_RDWR
) {
389 *open_flags
|= O_RDWR
;
391 *open_flags
|= O_RDONLY
;
394 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
395 * and O_DIRECT for no caching. */
396 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
397 *open_flags
|= O_DIRECT
;
401 static void raw_parse_filename(const char *filename
, QDict
*options
,
404 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
407 static QemuOptsList raw_runtime_opts
= {
409 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
413 .type
= QEMU_OPT_STRING
,
414 .help
= "File name of the image",
418 .type
= QEMU_OPT_STRING
,
419 .help
= "host AIO implementation (threads, native)",
423 .type
= QEMU_OPT_STRING
,
424 .help
= "file locking mode (on/off/auto, default: auto)",
427 .name
= "pr-manager",
428 .type
= QEMU_OPT_STRING
,
429 .help
= "id of persistent reservation manager object (default: none)",
432 .name
= "x-check-cache-dropped",
433 .type
= QEMU_OPT_BOOL
,
434 .help
= "check that page cache was dropped on live migration (default: off)"
436 { /* end of list */ }
440 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
441 int bdrv_flags
, int open_flags
,
442 bool device
, Error
**errp
)
444 BDRVRawState
*s
= bs
->opaque
;
446 Error
*local_err
= NULL
;
447 const char *filename
= NULL
;
449 BlockdevAioOptions aio
, aio_default
;
454 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
455 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
457 error_propagate(errp
, local_err
);
462 filename
= qemu_opt_get(opts
, "filename");
464 ret
= raw_normalize_devicepath(&filename
);
466 error_setg_errno(errp
, -ret
, "Could not normalize device path");
470 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
471 ? BLOCKDEV_AIO_OPTIONS_NATIVE
472 : BLOCKDEV_AIO_OPTIONS_THREADS
;
473 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
474 qemu_opt_get(opts
, "aio"),
475 aio_default
, &local_err
);
477 error_propagate(errp
, local_err
);
481 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
483 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
484 qemu_opt_get(opts
, "locking"),
485 ON_OFF_AUTO_AUTO
, &local_err
);
487 error_propagate(errp
, local_err
);
494 if (!qemu_has_ofd_lock()) {
496 "File lock requested but OFD locking syscall is "
497 "unavailable, falling back to POSIX file locks.\n"
498 "Due to the implementation, locks can be lost "
502 case ON_OFF_AUTO_OFF
:
505 case ON_OFF_AUTO_AUTO
:
506 s
->use_lock
= qemu_has_ofd_lock();
512 str
= qemu_opt_get(opts
, "pr-manager");
514 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
516 error_propagate(errp
, local_err
);
522 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
525 s
->open_flags
= open_flags
;
526 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
529 fd
= qemu_open(filename
, s
->open_flags
, 0644);
530 ret
= fd
< 0 ? -errno
: 0;
532 if (ret
== -EACCES
|| ret
== -EROFS
) {
533 /* Try to degrade to read-only, but if it doesn't work, still use the
534 * normal error message. */
535 if (bdrv_apply_auto_read_only(bs
, NULL
, NULL
) == 0) {
536 bdrv_flags
&= ~BDRV_O_RDWR
;
537 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
538 assert(!(s
->open_flags
& O_CREAT
));
539 fd
= qemu_open(filename
, s
->open_flags
);
540 ret
= fd
< 0 ? -errno
: 0;
545 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
555 fd
= qemu_open(filename
, s
->open_flags
);
558 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
566 s
->shared_perm
= BLK_PERM_ALL
;
568 #ifdef CONFIG_LINUX_AIO
569 /* Currently Linux does AIO only for files opened with O_DIRECT */
570 if (s
->use_linux_aio
) {
571 if (!(s
->open_flags
& O_DIRECT
)) {
572 error_setg(errp
, "aio=native was specified, but it requires "
573 "cache.direct=on, which was not specified.");
577 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
578 error_prepend(errp
, "Unable to use native AIO: ");
583 if (s
->use_linux_aio
) {
584 error_setg(errp
, "aio=native was specified, but is not supported "
589 #endif /* !defined(CONFIG_LINUX_AIO) */
591 s
->has_discard
= true;
592 s
->has_write_zeroes
= true;
593 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
594 s
->needs_alignment
= true;
597 if (fstat(s
->fd
, &st
) < 0) {
599 error_setg_errno(errp
, errno
, "Could not stat file");
604 if (S_ISBLK(st
.st_mode
)) {
605 warn_report("Opening a block device as a file using the '%s' "
606 "driver is deprecated", bs
->drv
->format_name
);
607 } else if (S_ISCHR(st
.st_mode
)) {
608 warn_report("Opening a character device as a file using the '%s' "
609 "driver is deprecated", bs
->drv
->format_name
);
610 } else if (!S_ISREG(st
.st_mode
)) {
611 error_setg(errp
, "A regular file was expected by the '%s' driver, "
612 "but something else was given", bs
->drv
->format_name
);
616 s
->discard_zeroes
= true;
617 s
->has_fallocate
= true;
620 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
621 error_setg(errp
, "'%s' driver expects either "
622 "a character or block device", bs
->drv
->format_name
);
628 if (S_ISBLK(st
.st_mode
)) {
629 #ifdef BLKDISCARDZEROES
631 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
632 s
->discard_zeroes
= true;
636 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
637 * not rely on the contents of discarded blocks unless using O_DIRECT.
638 * Same for BLKZEROOUT.
640 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
641 s
->discard_zeroes
= false;
642 s
->has_write_zeroes
= false;
647 if (S_ISCHR(st
.st_mode
)) {
649 * The file is a char device (disk), which on FreeBSD isn't behind
650 * a pager, so force all requests to be aligned. This is needed
651 * so QEMU makes sure all IO operations on the device are aligned
652 * to sector size, or else FreeBSD will reject them with EINVAL.
654 s
->needs_alignment
= true;
659 if (platform_test_xfs_fd(s
->fd
)) {
664 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
667 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
674 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
677 BDRVRawState
*s
= bs
->opaque
;
679 s
->type
= FTYPE_FILE
;
680 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
689 #define PERM_FOREACH(i) \
690 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
692 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
693 * file; if @unlock == true, also unlock the unneeded bytes.
694 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
696 static int raw_apply_lock_bytes(int fd
,
697 uint64_t perm_lock_bits
,
698 uint64_t shared_perm_lock_bits
,
699 bool unlock
, Error
**errp
)
705 int off
= RAW_LOCK_PERM_BASE
+ i
;
706 if (perm_lock_bits
& (1ULL << i
)) {
707 ret
= qemu_lock_fd(fd
, off
, 1, false);
709 error_setg(errp
, "Failed to lock byte %d", off
);
713 ret
= qemu_unlock_fd(fd
, off
, 1);
715 error_setg(errp
, "Failed to unlock byte %d", off
);
721 int off
= RAW_LOCK_SHARED_BASE
+ i
;
722 if (shared_perm_lock_bits
& (1ULL << i
)) {
723 ret
= qemu_lock_fd(fd
, off
, 1, false);
725 error_setg(errp
, "Failed to lock byte %d", off
);
729 ret
= qemu_unlock_fd(fd
, off
, 1);
731 error_setg(errp
, "Failed to unlock byte %d", off
);
739 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
740 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
747 int off
= RAW_LOCK_SHARED_BASE
+ i
;
748 uint64_t p
= 1ULL << i
;
750 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
752 char *perm_name
= bdrv_perm_names(p
);
754 "Failed to get \"%s\" lock",
762 int off
= RAW_LOCK_PERM_BASE
+ i
;
763 uint64_t p
= 1ULL << i
;
764 if (!(shared_perm
& p
)) {
765 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
767 char *perm_name
= bdrv_perm_names(p
);
769 "Failed to get shared \"%s\" lock",
779 static int raw_handle_perm_lock(BlockDriverState
*bs
,
781 uint64_t new_perm
, uint64_t new_shared
,
784 BDRVRawState
*s
= bs
->opaque
;
786 Error
*local_err
= NULL
;
792 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
796 assert(s
->lock_fd
> 0);
800 ret
= raw_apply_lock_bytes(s
->lock_fd
, s
->perm
| new_perm
,
801 ~s
->shared_perm
| ~new_shared
,
804 ret
= raw_check_lock_bytes(s
->lock_fd
, new_perm
, new_shared
, errp
);
808 error_append_hint(errp
,
809 "Is another process using the image [%s]?\n",
813 /* fall through to unlock bytes. */
815 raw_apply_lock_bytes(s
->lock_fd
, s
->perm
, ~s
->shared_perm
,
818 /* Theoretically the above call only unlocks bytes and it cannot
819 * fail. Something weird happened, report it.
821 error_report_err(local_err
);
825 raw_apply_lock_bytes(s
->lock_fd
, new_perm
, ~new_shared
,
828 /* Theoretically the above call only unlocks bytes and it cannot
829 * fail. Something weird happened, report it.
831 error_report_err(local_err
);
838 static int raw_reopen_prepare(BDRVReopenState
*state
,
839 BlockReopenQueue
*queue
, Error
**errp
)
842 BDRVRawReopenState
*rs
;
845 Error
*local_err
= NULL
;
847 assert(state
!= NULL
);
848 assert(state
->bs
!= NULL
);
850 s
= state
->bs
->opaque
;
852 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
856 /* Handle options changes */
857 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
858 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
860 error_propagate(errp
, local_err
);
865 rs
->check_cache_dropped
=
866 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
868 /* This driver's reopen function doesn't currently allow changing
869 * other options, so let's put them back in the original QDict and
870 * bdrv_reopen_prepare() will detect changes and complain. */
871 qemu_opts_to_qdict(opts
, state
->options
);
873 if (s
->type
== FTYPE_CD
) {
874 rs
->open_flags
|= O_NONBLOCK
;
877 raw_parse_flags(state
->flags
, &rs
->open_flags
);
879 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
881 fcntl_flags
|= O_NOATIME
;
885 /* Not all operating systems have O_ASYNC, and those that don't
886 * will not let us track the state into rs->open_flags (typically
887 * you achieve the same effect with an ioctl, for example I_SETSIG
888 * on Solaris). But we do not use O_ASYNC, so that's fine.
890 assert((s
->open_flags
& O_ASYNC
) == 0);
893 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
894 /* dup the original fd */
895 rs
->fd
= qemu_dup(s
->fd
);
897 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
905 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
907 const char *normalized_filename
= state
->bs
->filename
;
908 ret
= raw_normalize_devicepath(&normalized_filename
);
910 error_setg_errno(errp
, -ret
, "Could not normalize device path");
912 assert(!(rs
->open_flags
& O_CREAT
));
913 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
915 error_setg_errno(errp
, errno
, "Could not reopen file");
921 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
922 * alignment with the new fd. */
924 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
928 error_propagate(errp
, local_err
);
938 static void raw_reopen_commit(BDRVReopenState
*state
)
940 BDRVRawReopenState
*rs
= state
->opaque
;
941 BDRVRawState
*s
= state
->bs
->opaque
;
943 s
->check_cache_dropped
= rs
->check_cache_dropped
;
944 s
->open_flags
= rs
->open_flags
;
949 g_free(state
->opaque
);
950 state
->opaque
= NULL
;
954 static void raw_reopen_abort(BDRVReopenState
*state
)
956 BDRVRawReopenState
*rs
= state
->opaque
;
958 /* nothing to do if NULL, we didn't get far enough */
967 g_free(state
->opaque
);
968 state
->opaque
= NULL
;
971 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
975 short max_sectors
= 0;
976 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
978 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
979 return max_sectors
<< BDRV_SECTOR_BITS
;
988 static int hdev_get_max_segments(const struct stat
*st
)
998 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
999 major(st
->st_rdev
), minor(st
->st_rdev
));
1000 fd
= open(sysfspath
, O_RDONLY
);
1006 ret
= read(fd
, buf
, sizeof(buf
) - 1);
1007 } while (ret
== -1 && errno
== EINTR
);
1011 } else if (ret
== 0) {
1016 /* The file is ended with '\n', pass 'end' to accept that. */
1017 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1018 if (ret
== 0 && end
&& *end
== '\n') {
1033 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1035 BDRVRawState
*s
= bs
->opaque
;
1038 if (!fstat(s
->fd
, &st
)) {
1039 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1040 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1041 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1042 bs
->bl
.max_transfer
= pow2floor(ret
);
1044 ret
= hdev_get_max_segments(&st
);
1046 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1047 ret
* getpagesize());
1052 raw_probe_alignment(bs
, s
->fd
, errp
);
1053 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1054 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1057 static int check_for_dasd(int fd
)
1060 struct dasd_information2_t info
= {0};
1062 return ioctl(fd
, BIODASDINFO2
, &info
);
1069 * Try to get @bs's logical and physical block size.
1070 * On success, store them in @bsz and return zero.
1071 * On failure, return negative errno.
1073 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1075 BDRVRawState
*s
= bs
->opaque
;
1078 /* If DASD, get blocksizes */
1079 if (check_for_dasd(s
->fd
) < 0) {
1082 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1086 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1090 * Try to get @bs's geometry: cyls, heads, sectors.
1091 * On success, store them in @geo and return 0.
1092 * On failure return -errno.
1093 * (Allows block driver to assign default geometry values that guest sees)
1096 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1098 BDRVRawState
*s
= bs
->opaque
;
1099 struct hd_geometry ioctl_geo
= {0};
1101 /* If DASD, get its geometry */
1102 if (check_for_dasd(s
->fd
) < 0) {
1105 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1108 /* HDIO_GETGEO may return success even though geo contains zeros
1109 (e.g. certain multipath setups) */
1110 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1113 /* Do not return a geometry for partition */
1114 if (ioctl_geo
.start
!= 0) {
1117 geo
->heads
= ioctl_geo
.heads
;
1118 geo
->sectors
= ioctl_geo
.sectors
;
1119 geo
->cylinders
= ioctl_geo
.cylinders
;
1123 #else /* __linux__ */
1124 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1130 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1134 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1142 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1144 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1147 if (s
->page_cache_inconsistent
) {
1151 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1153 /* There is no clear definition of the semantics of a failing fsync(),
1154 * so we may have to assume the worst. The sad truth is that this
1155 * assumption is correct for Linux. Some pages are now probably marked
1156 * clean in the page cache even though they are inconsistent with the
1157 * on-disk contents. The next fdatasync() call would succeed, but no
1158 * further writeback attempt will be made. We can't get back to a state
1159 * in which we know what is on disk (we would have to rewrite
1160 * everything that was touched since the last fdatasync() at least), so
1161 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1162 * really defined, I have little hope that other OSes are doing better.
1164 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1166 if ((s
->open_flags
& O_DIRECT
) == 0) {
1167 s
->page_cache_inconsistent
= true;
1174 #ifdef CONFIG_PREADV
1176 static bool preadv_present
= true;
1179 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1181 return preadv(fd
, iov
, nr_iov
, offset
);
1185 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1187 return pwritev(fd
, iov
, nr_iov
, offset
);
1192 static bool preadv_present
= false;
1195 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1201 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1208 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1213 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1214 len
= qemu_pwritev(aiocb
->aio_fildes
,
1219 len
= qemu_preadv(aiocb
->aio_fildes
,
1223 } while (len
== -1 && errno
== EINTR
);
1232 * Read/writes the data to/from a given linear buffer.
1234 * Returns the number of bytes handles or -errno in case of an error. Short
1235 * reads are only returned if the end of the file is reached.
1237 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1242 while (offset
< aiocb
->aio_nbytes
) {
1243 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1244 len
= pwrite(aiocb
->aio_fildes
,
1245 (const char *)buf
+ offset
,
1246 aiocb
->aio_nbytes
- offset
,
1247 aiocb
->aio_offset
+ offset
);
1249 len
= pread(aiocb
->aio_fildes
,
1251 aiocb
->aio_nbytes
- offset
,
1252 aiocb
->aio_offset
+ offset
);
1254 if (len
== -1 && errno
== EINTR
) {
1256 } else if (len
== -1 && errno
== EINVAL
&&
1257 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1258 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1260 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1261 * after a short read. Assume that O_DIRECT short reads only occur
1262 * at EOF. Therefore this is a short read, not an I/O error.
1265 } else if (len
== -1) {
1268 } else if (len
== 0) {
1277 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1282 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1284 * If there is just a single buffer, and it is properly aligned
1285 * we can just use plain pread/pwrite without any problems.
1287 if (aiocb
->aio_niov
== 1) {
1288 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1291 * We have more than one iovec, and all are properly aligned.
1293 * Try preadv/pwritev first and fall back to linearizing the
1294 * buffer if it's not supported.
1296 if (preadv_present
) {
1297 nbytes
= handle_aiocb_rw_vector(aiocb
);
1298 if (nbytes
== aiocb
->aio_nbytes
||
1299 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1302 preadv_present
= false;
1306 * XXX(hch): short read/write. no easy way to handle the reminder
1307 * using these interfaces. For now retry using plain
1313 * Ok, we have to do it the hard way, copy all segments into
1314 * a single aligned buffer.
1316 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1321 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1325 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1326 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1327 p
+= aiocb
->aio_iov
[i
].iov_len
;
1329 assert(p
- buf
== aiocb
->aio_nbytes
);
1332 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1333 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1335 size_t count
= aiocb
->aio_nbytes
, copy
;
1338 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1340 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1341 copy
= aiocb
->aio_iov
[i
].iov_len
;
1343 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1344 assert(count
>= copy
);
1356 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1358 struct xfs_flock64 fl
;
1361 memset(&fl
, 0, sizeof(fl
));
1362 fl
.l_whence
= SEEK_SET
;
1363 fl
.l_start
= offset
;
1366 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1368 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1375 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1377 struct xfs_flock64 fl
;
1380 memset(&fl
, 0, sizeof(fl
));
1381 fl
.l_whence
= SEEK_SET
;
1382 fl
.l_start
= offset
;
1385 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1387 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1395 static int translate_err(int err
)
1397 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1404 #ifdef CONFIG_FALLOCATE
1405 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1408 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1411 } while (errno
== EINTR
);
1412 return translate_err(-errno
);
1416 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1419 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1421 if (!s
->has_write_zeroes
) {
1427 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1428 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1431 } while (errno
== EINTR
);
1433 ret
= translate_err(-errno
);
1436 if (ret
== -ENOTSUP
) {
1437 s
->has_write_zeroes
= false;
1442 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1444 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1445 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1447 #ifdef CONFIG_FALLOCATE
1451 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1452 return handle_aiocb_write_zeroes_block(aiocb
);
1457 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1461 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1462 if (s
->has_write_zeroes
) {
1463 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1464 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1465 if (ret
== 0 || ret
!= -ENOTSUP
) {
1468 s
->has_write_zeroes
= false;
1472 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1473 if (s
->has_discard
&& s
->has_fallocate
) {
1474 int ret
= do_fallocate(s
->fd
,
1475 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1476 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1478 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1479 if (ret
== 0 || ret
!= -ENOTSUP
) {
1482 s
->has_fallocate
= false;
1483 } else if (ret
!= -ENOTSUP
) {
1486 s
->has_discard
= false;
1491 #ifdef CONFIG_FALLOCATE
1492 /* Last resort: we are trying to extend the file with zeroed data. This
1493 * can be done via fallocate(fd, 0) */
1494 len
= bdrv_getlength(aiocb
->bs
);
1495 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1496 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1497 if (ret
== 0 || ret
!= -ENOTSUP
) {
1500 s
->has_fallocate
= false;
1507 static ssize_t
handle_aiocb_write_zeroes_unmap(RawPosixAIOData
*aiocb
)
1509 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1512 /* First try to write zeros and unmap at the same time */
1514 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1515 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1516 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1517 if (ret
!= -ENOTSUP
) {
1524 /* xfs_discard() guarantees that the discarded area reads as all-zero
1525 * afterwards, so we can use it here. */
1526 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1530 /* If we couldn't manage to unmap while guaranteed that the area reads as
1531 * all-zero afterwards, just write zeroes without unmapping */
1532 ret
= handle_aiocb_write_zeroes(aiocb
);
1536 #ifndef HAVE_COPY_FILE_RANGE
1537 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1538 off_t
*out_off
, size_t len
, unsigned int flags
)
1540 #ifdef __NR_copy_file_range
1541 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1542 out_off
, len
, flags
);
1550 static ssize_t
handle_aiocb_copy_range(RawPosixAIOData
*aiocb
)
1552 uint64_t bytes
= aiocb
->aio_nbytes
;
1553 off_t in_off
= aiocb
->aio_offset
;
1554 off_t out_off
= aiocb
->aio_offset2
;
1557 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1558 aiocb
->aio_fd2
, &out_off
,
1560 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1561 aiocb
->aio_fd2
, out_off
, bytes
, 0, ret
);
1563 /* No progress (e.g. when beyond EOF), let the caller fall back to
1582 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1584 int ret
= -EOPNOTSUPP
;
1585 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1587 if (!s
->has_discard
) {
1591 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1594 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1595 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1598 } while (errno
== EINTR
);
1605 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1609 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1610 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1611 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1615 ret
= translate_err(ret
);
1616 if (ret
== -ENOTSUP
) {
1617 s
->has_discard
= false;
1622 static int handle_aiocb_truncate(RawPosixAIOData
*aiocb
)
1625 int64_t current_length
= 0;
1628 int fd
= aiocb
->aio_fildes
;
1629 int64_t offset
= aiocb
->aio_offset
;
1630 Error
**errp
= aiocb
->errp
;
1632 if (fstat(fd
, &st
) < 0) {
1634 error_setg_errno(errp
, -result
, "Could not stat file");
1638 current_length
= st
.st_size
;
1639 if (current_length
> offset
&& aiocb
->prealloc
!= PREALLOC_MODE_OFF
) {
1640 error_setg(errp
, "Cannot use preallocation for shrinking files");
1644 switch (aiocb
->prealloc
) {
1645 #ifdef CONFIG_POSIX_FALLOCATE
1646 case PREALLOC_MODE_FALLOC
:
1648 * Truncating before posix_fallocate() makes it about twice slower on
1649 * file systems that do not support fallocate(), trying to check if a
1650 * block is allocated before allocating it, so don't do that here.
1652 if (offset
!= current_length
) {
1653 result
= -posix_fallocate(fd
, current_length
,
1654 offset
- current_length
);
1656 /* posix_fallocate() doesn't set errno. */
1657 error_setg_errno(errp
, -result
,
1658 "Could not preallocate new data");
1665 case PREALLOC_MODE_FULL
:
1667 int64_t num
= 0, left
= offset
- current_length
;
1671 * Knowing the final size from the beginning could allow the file
1672 * system driver to do less allocations and possibly avoid
1673 * fragmentation of the file.
1675 if (ftruncate(fd
, offset
) != 0) {
1677 error_setg_errno(errp
, -result
, "Could not resize file");
1681 buf
= g_malloc0(65536);
1683 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1684 if (seek_result
< 0) {
1686 error_setg_errno(errp
, -result
,
1687 "Failed to seek to the old end of file");
1692 num
= MIN(left
, 65536);
1693 result
= write(fd
, buf
, num
);
1695 if (errno
== EINTR
) {
1699 error_setg_errno(errp
, -result
,
1700 "Could not write zeros for preallocation");
1709 error_setg_errno(errp
, -result
,
1710 "Could not flush file to disk");
1716 case PREALLOC_MODE_OFF
:
1717 if (ftruncate(fd
, offset
) != 0) {
1719 error_setg_errno(errp
, -result
, "Could not resize file");
1724 error_setg(errp
, "Unsupported preallocation mode: %s",
1725 PreallocMode_str(aiocb
->prealloc
));
1731 if (ftruncate(fd
, current_length
) < 0) {
1732 error_report("Failed to restore old file length: %s",
1741 static int aio_worker(void *arg
)
1743 RawPosixAIOData
*aiocb
= arg
;
1746 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1748 ret
= handle_aiocb_rw(aiocb
);
1749 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1750 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1751 0, aiocb
->aio_nbytes
- ret
);
1753 ret
= aiocb
->aio_nbytes
;
1755 if (ret
== aiocb
->aio_nbytes
) {
1757 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1761 case QEMU_AIO_WRITE
:
1762 ret
= handle_aiocb_rw(aiocb
);
1763 if (ret
== aiocb
->aio_nbytes
) {
1765 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1769 case QEMU_AIO_FLUSH
:
1770 ret
= handle_aiocb_flush(aiocb
);
1772 case QEMU_AIO_IOCTL
:
1773 ret
= handle_aiocb_ioctl(aiocb
);
1775 case QEMU_AIO_DISCARD
:
1776 ret
= handle_aiocb_discard(aiocb
);
1778 case QEMU_AIO_WRITE_ZEROES
:
1779 ret
= handle_aiocb_write_zeroes(aiocb
);
1781 case QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_DISCARD
:
1782 ret
= handle_aiocb_write_zeroes_unmap(aiocb
);
1784 case QEMU_AIO_COPY_RANGE
:
1785 ret
= handle_aiocb_copy_range(aiocb
);
1787 case QEMU_AIO_TRUNCATE
:
1788 ret
= handle_aiocb_truncate(aiocb
);
1791 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1800 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1801 int64_t offset
, int fd2
, int64_t offset2
,
1803 int bytes
, int type
)
1805 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1809 acb
->aio_type
= type
;
1810 acb
->aio_fildes
= fd
;
1812 acb
->aio_offset2
= offset2
;
1814 acb
->aio_nbytes
= bytes
;
1815 acb
->aio_offset
= offset
;
1818 acb
->aio_iov
= qiov
->iov
;
1819 acb
->aio_niov
= qiov
->niov
;
1820 assert(qiov
->size
== bytes
);
1823 trace_file_paio_submit_co(offset
, bytes
, type
);
1824 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1825 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1828 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1829 int64_t offset
, QEMUIOVector
*qiov
,
1830 int bytes
, int type
)
1832 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1835 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1836 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1838 BDRVRawState
*s
= bs
->opaque
;
1840 if (fd_open(bs
) < 0)
1844 * Check if the underlying device requires requests to be aligned,
1845 * and if the request we are trying to submit is aligned or not.
1846 * If this is the case tell the low-level driver that it needs
1847 * to copy the buffer.
1849 if (s
->needs_alignment
) {
1850 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1851 type
|= QEMU_AIO_MISALIGNED
;
1852 #ifdef CONFIG_LINUX_AIO
1853 } else if (s
->use_linux_aio
) {
1854 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1855 assert(qiov
->size
== bytes
);
1856 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1861 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1864 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1865 uint64_t bytes
, QEMUIOVector
*qiov
,
1868 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1871 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1872 uint64_t bytes
, QEMUIOVector
*qiov
,
1876 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1879 static void raw_aio_plug(BlockDriverState
*bs
)
1881 #ifdef CONFIG_LINUX_AIO
1882 BDRVRawState
*s
= bs
->opaque
;
1883 if (s
->use_linux_aio
) {
1884 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1885 laio_io_plug(bs
, aio
);
1890 static void raw_aio_unplug(BlockDriverState
*bs
)
1892 #ifdef CONFIG_LINUX_AIO
1893 BDRVRawState
*s
= bs
->opaque
;
1894 if (s
->use_linux_aio
) {
1895 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1896 laio_io_unplug(bs
, aio
);
1901 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1903 BDRVRawState
*s
= bs
->opaque
;
1911 return paio_submit_co(bs
, s
->fd
, 0, NULL
, 0, QEMU_AIO_FLUSH
);
1914 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1915 AioContext
*new_context
)
1917 #ifdef CONFIG_LINUX_AIO
1918 BDRVRawState
*s
= bs
->opaque
;
1919 if (s
->use_linux_aio
) {
1921 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1922 error_reportf_err(local_err
, "Unable to use native AIO, "
1923 "falling back to thread pool: ");
1924 s
->use_linux_aio
= false;
1930 static void raw_close(BlockDriverState
*bs
)
1932 BDRVRawState
*s
= bs
->opaque
;
1938 if (s
->lock_fd
>= 0) {
1939 qemu_close(s
->lock_fd
);
1945 * Truncates the given regular file @fd to @offset and, when growing, fills the
1946 * new space according to @prealloc.
1948 * Returns: 0 on success, -errno on failure.
1950 static int coroutine_fn
1951 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1952 PreallocMode prealloc
, Error
**errp
)
1954 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1957 *acb
= (RawPosixAIOData
) {
1960 .aio_type
= QEMU_AIO_TRUNCATE
,
1961 .aio_offset
= offset
,
1962 .prealloc
= prealloc
,
1966 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1967 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1968 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1971 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1972 PreallocMode prealloc
, Error
**errp
)
1974 BDRVRawState
*s
= bs
->opaque
;
1978 if (fstat(s
->fd
, &st
)) {
1980 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1984 if (S_ISREG(st
.st_mode
)) {
1985 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
1988 if (prealloc
!= PREALLOC_MODE_OFF
) {
1989 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1990 "non-regular file", PreallocMode_str(prealloc
));
1994 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1995 if (offset
> raw_getlength(bs
)) {
1996 error_setg(errp
, "Cannot grow device files");
2000 error_setg(errp
, "Resizing this file is not supported");
2008 static int64_t raw_getlength(BlockDriverState
*bs
)
2010 BDRVRawState
*s
= bs
->opaque
;
2016 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2017 struct disklabel dl
;
2019 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2021 return (uint64_t)dl
.d_secsize
*
2022 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2026 #elif defined(__NetBSD__)
2027 static int64_t raw_getlength(BlockDriverState
*bs
)
2029 BDRVRawState
*s
= bs
->opaque
;
2035 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2036 struct dkwedge_info dkw
;
2038 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2039 return dkw
.dkw_size
* 512;
2041 struct disklabel dl
;
2043 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2045 return (uint64_t)dl
.d_secsize
*
2046 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2051 #elif defined(__sun__)
2052 static int64_t raw_getlength(BlockDriverState
*bs
)
2054 BDRVRawState
*s
= bs
->opaque
;
2055 struct dk_minfo minfo
;
2065 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2067 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2069 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2073 * There are reports that lseek on some devices fails, but
2074 * irc discussion said that contingency on contingency was overkill.
2076 size
= lseek(s
->fd
, 0, SEEK_END
);
2082 #elif defined(CONFIG_BSD)
2083 static int64_t raw_getlength(BlockDriverState
*bs
)
2085 BDRVRawState
*s
= bs
->opaque
;
2089 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2098 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2101 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2102 #ifdef DIOCGMEDIASIZE
2103 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2104 #elif defined(DIOCGPART)
2107 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2108 size
= pi
.media_size
;
2114 #if defined(__APPLE__) && defined(__MACH__)
2116 uint64_t sectors
= 0;
2117 uint32_t sector_size
= 0;
2119 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2120 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2121 size
= sectors
* sector_size
;
2123 size
= lseek(fd
, 0LL, SEEK_END
);
2130 size
= lseek(fd
, 0LL, SEEK_END
);
2135 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2138 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2139 if (size
== 2048LL * (unsigned)-1)
2141 /* XXX no disc? maybe we need to reopen... */
2142 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2149 size
= lseek(fd
, 0, SEEK_END
);
2157 static int64_t raw_getlength(BlockDriverState
*bs
)
2159 BDRVRawState
*s
= bs
->opaque
;
2168 size
= lseek(s
->fd
, 0, SEEK_END
);
2176 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2179 BDRVRawState
*s
= bs
->opaque
;
2181 if (fstat(s
->fd
, &st
) < 0) {
2184 return (int64_t)st
.st_blocks
* 512;
2187 static int coroutine_fn
2188 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2190 BlockdevCreateOptionsFile
*file_opts
;
2191 Error
*local_err
= NULL
;
2193 uint64_t perm
, shared
;
2196 /* Validate options and set default values */
2197 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2198 file_opts
= &options
->u
.file
;
2200 if (!file_opts
->has_nocow
) {
2201 file_opts
->nocow
= false;
2203 if (!file_opts
->has_preallocation
) {
2204 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2208 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2211 error_setg_errno(errp
, -result
, "Could not create file");
2215 /* Take permissions: We want to discard everything, so we need
2216 * BLK_PERM_WRITE; and truncation to the desired size requires
2218 * On the other hand, we cannot share the RESIZE permission
2219 * because we promise that after this function, the file has the
2220 * size given in the options. If someone else were to resize it
2221 * concurrently, we could not guarantee that.
2222 * Note that after this function, we can no longer guarantee that
2223 * the file is not touched by a third party, so it may be resized
2225 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2226 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2228 /* Step one: Take locks */
2229 result
= raw_apply_lock_bytes(fd
, perm
, ~shared
, false, errp
);
2234 /* Step two: Check that nobody else has taken conflicting locks */
2235 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2237 error_append_hint(errp
,
2238 "Is another process using the image [%s]?\n",
2239 file_opts
->filename
);
2243 /* Clear the file by truncating it to 0 */
2244 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2249 if (file_opts
->nocow
) {
2251 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2252 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2253 * will be ignored since any failure of this operation should not
2254 * block the left work.
2257 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2258 attr
|= FS_NOCOW_FL
;
2259 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2264 /* Resize and potentially preallocate the file to the desired
2266 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2267 file_opts
->preallocation
, errp
);
2273 raw_apply_lock_bytes(fd
, 0, 0, true, &local_err
);
2275 /* The above call should not fail, and if it does, that does
2276 * not mean the whole creation operation has failed. So
2277 * report it the user for their convenience, but do not report
2278 * it to the caller. */
2279 error_report_err(local_err
);
2283 if (qemu_close(fd
) != 0 && result
== 0) {
2285 error_setg_errno(errp
, -result
, "Could not close the new file");
2291 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2294 BlockdevCreateOptions options
;
2295 int64_t total_size
= 0;
2297 PreallocMode prealloc
;
2299 Error
*local_err
= NULL
;
2301 /* Skip file: protocol prefix */
2302 strstart(filename
, "file:", &filename
);
2304 /* Read out options */
2305 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2307 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2308 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2309 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2310 PREALLOC_MODE_OFF
, &local_err
);
2313 error_propagate(errp
, local_err
);
2317 options
= (BlockdevCreateOptions
) {
2318 .driver
= BLOCKDEV_DRIVER_FILE
,
2320 .filename
= (char *) filename
,
2322 .has_preallocation
= true,
2323 .preallocation
= prealloc
,
2328 return raw_co_create(&options
, errp
);
2332 * Find allocation range in @bs around offset @start.
2333 * May change underlying file descriptor's file offset.
2334 * If @start is not in a hole, store @start in @data, and the
2335 * beginning of the next hole in @hole, and return 0.
2336 * If @start is in a non-trailing hole, store @start in @hole and the
2337 * beginning of the next non-hole in @data, and return 0.
2338 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2339 * If we can't find out, return a negative errno other than -ENXIO.
2341 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2342 off_t
*data
, off_t
*hole
)
2344 #if defined SEEK_HOLE && defined SEEK_DATA
2345 BDRVRawState
*s
= bs
->opaque
;
2350 * D1. offs == start: start is in data
2351 * D2. offs > start: start is in a hole, next data at offs
2352 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2353 * or start is beyond EOF
2354 * If the latter happens, the file has been truncated behind
2355 * our back since we opened it. All bets are off then.
2356 * Treating like a trailing hole is simplest.
2357 * D4. offs < 0, errno != ENXIO: we learned nothing
2359 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2361 return -errno
; /* D3 or D4 */
2365 /* This is not a valid return by lseek(). We are safe to just return
2366 * -EIO in this case, and we'll treat it like D4. */
2371 /* D2: in hole, next data at offs */
2377 /* D1: in data, end not yet known */
2381 * H1. offs == start: start is in a hole
2382 * If this happens here, a hole has been dug behind our back
2383 * since the previous lseek().
2384 * H2. offs > start: either start is in data, next hole at offs,
2385 * or start is in trailing hole, EOF at offs
2386 * Linux treats trailing holes like any other hole: offs ==
2387 * start. Solaris seeks to EOF instead: offs > start (blech).
2388 * If that happens here, a hole has been dug behind our back
2389 * since the previous lseek().
2390 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2391 * If this happens, the file has been truncated behind our
2392 * back since we opened it. Treat it like a trailing hole.
2393 * H4. offs < 0, errno != ENXIO: we learned nothing
2394 * Pretend we know nothing at all, i.e. "forget" about D1.
2396 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2398 return -errno
; /* D1 and (H3 or H4) */
2402 /* This is not a valid return by lseek(). We are safe to just return
2403 * -EIO in this case, and we'll treat it like H4. */
2409 * D1 and H2: either in data, next hole at offs, or it was in
2410 * data but is now in a trailing hole. In the latter case,
2411 * all bets are off. Treating it as if it there was data all
2412 * the way to EOF is safe, so simply do that.
2427 * Returns the allocation status of the specified offset.
2429 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2431 * 'pnum' is set to the number of bytes (including and immediately following
2432 * the specified offset) that are known to be in the same
2433 * allocated/unallocated state.
2435 * 'bytes' is the max value 'pnum' should be set to.
2437 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2440 int64_t bytes
, int64_t *pnum
,
2442 BlockDriverState
**file
)
2444 off_t data
= 0, hole
= 0;
2456 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2459 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2460 if (ret
== -ENXIO
) {
2463 ret
= BDRV_BLOCK_ZERO
;
2464 } else if (ret
< 0) {
2465 /* No info available, so pretend there are no holes */
2467 ret
= BDRV_BLOCK_DATA
;
2468 } else if (data
== offset
) {
2469 /* On a data extent, compute bytes to the end of the extent,
2470 * possibly including a partial sector at EOF. */
2471 *pnum
= MIN(bytes
, hole
- offset
);
2472 ret
= BDRV_BLOCK_DATA
;
2474 /* On a hole, compute bytes to the beginning of the next extent. */
2475 assert(hole
== offset
);
2476 *pnum
= MIN(bytes
, data
- offset
);
2477 ret
= BDRV_BLOCK_ZERO
;
2481 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2484 #if defined(__linux__)
2485 /* Verify that the file is not in the page cache */
2486 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2488 const size_t window_size
= 128 * 1024 * 1024;
2489 BDRVRawState
*s
= bs
->opaque
;
2490 void *window
= NULL
;
2497 /* mincore(2) page status information requires 1 byte per page */
2498 page_size
= sysconf(_SC_PAGESIZE
);
2499 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2501 end
= raw_getlength(bs
);
2503 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2510 /* Unmap previous window if size has changed */
2511 new_length
= MIN(end
- offset
, window_size
);
2512 if (new_length
!= length
) {
2513 munmap(window
, length
);
2518 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2520 if (new_window
== MAP_FAILED
) {
2521 error_setg_errno(errp
, errno
, "mmap failed");
2525 window
= new_window
;
2526 length
= new_length
;
2528 ret
= mincore(window
, length
, vec
);
2530 error_setg_errno(errp
, errno
, "mincore failed");
2534 vec_end
= DIV_ROUND_UP(length
, page_size
);
2535 for (i
= 0; i
< vec_end
; i
++) {
2537 error_setg(errp
, "page cache still in use!");
2544 munmap(window
, length
);
2549 #endif /* __linux__ */
2551 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2554 BDRVRawState
*s
= bs
->opaque
;
2559 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2563 if (s
->open_flags
& O_DIRECT
) {
2564 return; /* No host kernel page cache */
2567 #if defined(__linux__)
2568 /* This sets the scene for the next syscall... */
2569 ret
= bdrv_co_flush(bs
);
2571 error_setg_errno(errp
, -ret
, "flush failed");
2575 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2576 * process. These limitations are okay because we just fsynced the file,
2577 * we don't use mmap, and the file should not be in use by other processes.
2579 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2580 if (ret
!= 0) { /* the return value is a positive errno */
2581 error_setg_errno(errp
, ret
, "fadvise failed");
2585 if (s
->check_cache_dropped
) {
2586 check_cache_dropped(bs
, errp
);
2588 #else /* __linux__ */
2589 /* Do nothing. Live migration to a remote host with cache.direct=off is
2590 * unsupported on other host operating systems. Cache consistency issues
2591 * may occur but no error is reported here, partly because that's the
2592 * historical behavior and partly because it's hard to differentiate valid
2593 * configurations that should not cause errors.
2595 #endif /* !__linux__ */
2598 static coroutine_fn
int
2599 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2601 BDRVRawState
*s
= bs
->opaque
;
2603 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, QEMU_AIO_DISCARD
);
2606 static int coroutine_fn
raw_co_pwrite_zeroes(
2607 BlockDriverState
*bs
, int64_t offset
,
2608 int bytes
, BdrvRequestFlags flags
)
2610 BDRVRawState
*s
= bs
->opaque
;
2611 int operation
= QEMU_AIO_WRITE_ZEROES
;
2613 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2614 operation
|= QEMU_AIO_DISCARD
;
2617 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, operation
);
2620 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2622 BDRVRawState
*s
= bs
->opaque
;
2624 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2628 static QemuOptsList raw_create_opts
= {
2629 .name
= "raw-create-opts",
2630 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2633 .name
= BLOCK_OPT_SIZE
,
2634 .type
= QEMU_OPT_SIZE
,
2635 .help
= "Virtual disk size"
2638 .name
= BLOCK_OPT_NOCOW
,
2639 .type
= QEMU_OPT_BOOL
,
2640 .help
= "Turn off copy-on-write (valid only on btrfs)"
2643 .name
= BLOCK_OPT_PREALLOC
,
2644 .type
= QEMU_OPT_STRING
,
2645 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2647 { /* end of list */ }
2651 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2654 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2657 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2659 BDRVRawState
*s
= bs
->opaque
;
2660 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2662 s
->shared_perm
= shared
;
2665 static void raw_abort_perm_update(BlockDriverState
*bs
)
2667 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2670 static int coroutine_fn
raw_co_copy_range_from(
2671 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2672 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2673 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2675 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2676 read_flags
, write_flags
);
2679 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2681 uint64_t src_offset
,
2683 uint64_t dst_offset
,
2685 BdrvRequestFlags read_flags
,
2686 BdrvRequestFlags write_flags
)
2688 BDRVRawState
*s
= bs
->opaque
;
2689 BDRVRawState
*src_s
;
2691 assert(dst
->bs
== bs
);
2692 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2696 src_s
= src
->bs
->opaque
;
2697 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2700 return paio_submit_co_full(bs
, src_s
->fd
, src_offset
, s
->fd
, dst_offset
,
2701 NULL
, bytes
, QEMU_AIO_COPY_RANGE
);
2704 BlockDriver bdrv_file
= {
2705 .format_name
= "file",
2706 .protocol_name
= "file",
2707 .instance_size
= sizeof(BDRVRawState
),
2708 .bdrv_needs_filename
= true,
2709 .bdrv_probe
= NULL
, /* no probe for protocols */
2710 .bdrv_parse_filename
= raw_parse_filename
,
2711 .bdrv_file_open
= raw_open
,
2712 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2713 .bdrv_reopen_commit
= raw_reopen_commit
,
2714 .bdrv_reopen_abort
= raw_reopen_abort
,
2715 .bdrv_close
= raw_close
,
2716 .bdrv_co_create
= raw_co_create
,
2717 .bdrv_co_create_opts
= raw_co_create_opts
,
2718 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2719 .bdrv_co_block_status
= raw_co_block_status
,
2720 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2721 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2723 .bdrv_co_preadv
= raw_co_preadv
,
2724 .bdrv_co_pwritev
= raw_co_pwritev
,
2725 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2726 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2727 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2728 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2729 .bdrv_refresh_limits
= raw_refresh_limits
,
2730 .bdrv_io_plug
= raw_aio_plug
,
2731 .bdrv_io_unplug
= raw_aio_unplug
,
2732 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2734 .bdrv_co_truncate
= raw_co_truncate
,
2735 .bdrv_getlength
= raw_getlength
,
2736 .bdrv_get_info
= raw_get_info
,
2737 .bdrv_get_allocated_file_size
2738 = raw_get_allocated_file_size
,
2739 .bdrv_check_perm
= raw_check_perm
,
2740 .bdrv_set_perm
= raw_set_perm
,
2741 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2742 .create_opts
= &raw_create_opts
,
2745 /***********************************************/
2748 #if defined(__APPLE__) && defined(__MACH__)
2749 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2750 CFIndex maxPathSize
, int flags
);
2751 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2753 kern_return_t kernResult
= KERN_FAILURE
;
2754 mach_port_t masterPort
;
2755 CFMutableDictionaryRef classesToMatch
;
2756 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2757 char *mediaType
= NULL
;
2759 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2760 if ( KERN_SUCCESS
!= kernResult
) {
2761 printf( "IOMasterPort returned %d\n", kernResult
);
2765 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2766 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2767 if (classesToMatch
== NULL
) {
2768 error_report("IOServiceMatching returned NULL for %s",
2769 matching_array
[index
]);
2772 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2774 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2776 if (kernResult
!= KERN_SUCCESS
) {
2777 error_report("Note: IOServiceGetMatchingServices returned %d",
2782 /* If a match was found, leave the loop */
2783 if (*mediaIterator
!= 0) {
2784 DPRINTF("Matching using %s\n", matching_array
[index
]);
2785 mediaType
= g_strdup(matching_array
[index
]);
2792 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2793 CFIndex maxPathSize
, int flags
)
2795 io_object_t nextMedia
;
2796 kern_return_t kernResult
= KERN_FAILURE
;
2798 nextMedia
= IOIteratorNext( mediaIterator
);
2801 CFTypeRef bsdPathAsCFString
;
2802 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2803 if ( bsdPathAsCFString
) {
2804 size_t devPathLength
;
2805 strcpy( bsdPath
, _PATH_DEV
);
2806 if (flags
& BDRV_O_NOCACHE
) {
2807 strcat(bsdPath
, "r");
2809 devPathLength
= strlen( bsdPath
);
2810 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2811 kernResult
= KERN_SUCCESS
;
2813 CFRelease( bsdPathAsCFString
);
2815 IOObjectRelease( nextMedia
);
2821 /* Sets up a real cdrom for use in QEMU */
2822 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2824 int index
, num_of_test_partitions
= 2, fd
;
2825 char test_partition
[MAXPATHLEN
];
2826 bool partition_found
= false;
2828 /* look for a working partition */
2829 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2830 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2832 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2834 partition_found
= true;
2840 /* if a working partition on the device was not found */
2841 if (partition_found
== false) {
2842 error_setg(errp
, "Failed to find a working partition on disc");
2844 DPRINTF("Using %s as optical disc\n", test_partition
);
2845 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2847 return partition_found
;
2850 /* Prints directions on mounting and unmounting a device */
2851 static void print_unmounting_directions(const char *file_name
)
2853 error_report("If device %s is mounted on the desktop, unmount"
2854 " it first before using it in QEMU", file_name
);
2855 error_report("Command to unmount device: diskutil unmountDisk %s",
2857 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2860 #endif /* defined(__APPLE__) && defined(__MACH__) */
2862 static int hdev_probe_device(const char *filename
)
2866 /* allow a dedicated CD-ROM driver to match with a higher priority */
2867 if (strstart(filename
, "/dev/cdrom", NULL
))
2870 if (stat(filename
, &st
) >= 0 &&
2871 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2878 static int check_hdev_writable(BDRVRawState
*s
)
2880 #if defined(BLKROGET)
2881 /* Linux block devices can be configured "read-only" using blockdev(8).
2882 * This is independent of device node permissions and therefore open(2)
2883 * with O_RDWR succeeds. Actual writes fail with EPERM.
2885 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2886 * check for read-only block devices so that Linux block devices behave
2892 if (fstat(s
->fd
, &st
)) {
2896 if (!S_ISBLK(st
.st_mode
)) {
2900 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2907 #endif /* defined(BLKROGET) */
2911 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2914 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2917 static bool hdev_is_sg(BlockDriverState
*bs
)
2920 #if defined(__linux__)
2922 BDRVRawState
*s
= bs
->opaque
;
2924 struct sg_scsi_id scsiid
;
2928 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2932 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2937 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2939 DPRINTF("SG device found: type=%d, version=%d\n",
2940 scsiid
.scsi_type
, sg_version
);
2949 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2952 BDRVRawState
*s
= bs
->opaque
;
2953 Error
*local_err
= NULL
;
2956 #if defined(__APPLE__) && defined(__MACH__)
2958 * Caution: while qdict_get_str() is fine, getting non-string types
2959 * would require more care. When @options come from -blockdev or
2960 * blockdev_add, its members are typed according to the QAPI
2961 * schema, but when they come from -drive, they're all QString.
2963 const char *filename
= qdict_get_str(options
, "filename");
2964 char bsd_path
[MAXPATHLEN
] = "";
2965 bool error_occurred
= false;
2967 /* If using a real cdrom */
2968 if (strcmp(filename
, "/dev/cdrom") == 0) {
2969 char *mediaType
= NULL
;
2970 kern_return_t ret_val
;
2971 io_iterator_t mediaIterator
= 0;
2973 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
2974 if (mediaType
== NULL
) {
2975 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
2977 error_occurred
= true;
2978 goto hdev_open_Mac_error
;
2981 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
2982 if (ret_val
!= KERN_SUCCESS
) {
2983 error_setg(errp
, "Could not get BSD path for optical drive");
2984 error_occurred
= true;
2985 goto hdev_open_Mac_error
;
2988 /* If a real optical drive was not found */
2989 if (bsd_path
[0] == '\0') {
2990 error_setg(errp
, "Failed to obtain bsd path for optical drive");
2991 error_occurred
= true;
2992 goto hdev_open_Mac_error
;
2995 /* If using a cdrom disc and finding a partition on the disc failed */
2996 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
2997 setup_cdrom(bsd_path
, errp
) == false) {
2998 print_unmounting_directions(bsd_path
);
2999 error_occurred
= true;
3000 goto hdev_open_Mac_error
;
3003 qdict_put_str(options
, "filename", bsd_path
);
3005 hdev_open_Mac_error
:
3007 if (mediaIterator
) {
3008 IOObjectRelease(mediaIterator
);
3010 if (error_occurred
) {
3014 #endif /* defined(__APPLE__) && defined(__MACH__) */
3016 s
->type
= FTYPE_FILE
;
3018 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3020 error_propagate(errp
, local_err
);
3021 #if defined(__APPLE__) && defined(__MACH__)
3023 filename
= bsd_path
;
3025 /* if a physical device experienced an error while being opened */
3026 if (strncmp(filename
, "/dev/", 5) == 0) {
3027 print_unmounting_directions(filename
);
3029 #endif /* defined(__APPLE__) && defined(__MACH__) */
3033 /* Since this does ioctl the device must be already opened */
3034 bs
->sg
= hdev_is_sg(bs
);
3036 if (flags
& BDRV_O_RDWR
) {
3037 ret
= check_hdev_writable(s
);
3040 error_setg_errno(errp
, -ret
, "The device is not writable");
3048 #if defined(__linux__)
3050 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
3051 unsigned long int req
, void *buf
,
3052 BlockCompletionFunc
*cb
, void *opaque
)
3054 BDRVRawState
*s
= bs
->opaque
;
3055 RawPosixAIOData
*acb
;
3058 if (fd_open(bs
) < 0)
3061 if (req
== SG_IO
&& s
->pr_mgr
) {
3062 struct sg_io_hdr
*io_hdr
= buf
;
3063 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3064 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3065 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3066 s
->fd
, io_hdr
, cb
, opaque
);
3070 acb
= g_new(RawPosixAIOData
, 1);
3072 acb
->aio_type
= QEMU_AIO_IOCTL
;
3073 acb
->aio_fildes
= s
->fd
;
3074 acb
->aio_offset
= 0;
3075 acb
->aio_ioctl_buf
= buf
;
3076 acb
->aio_ioctl_cmd
= req
;
3077 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3078 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
3082 static int fd_open(BlockDriverState
*bs
)
3084 BDRVRawState
*s
= bs
->opaque
;
3086 /* this is just to ensure s->fd is sane (its called by io ops) */
3092 static coroutine_fn
int
3093 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3095 BDRVRawState
*s
= bs
->opaque
;
3102 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3103 QEMU_AIO_DISCARD
| QEMU_AIO_BLKDEV
);
3106 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3107 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3109 BDRVRawState
*s
= bs
->opaque
;
3110 int operation
= QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_BLKDEV
;
3118 if (flags
& BDRV_REQ_MAY_UNMAP
) {
3119 operation
|= QEMU_AIO_DISCARD
;
3122 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, operation
);
3125 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3130 struct stat stat_buf
;
3131 int64_t total_size
= 0;
3134 /* This function is used by both protocol block drivers and therefore either
3135 * of these prefixes may be given.
3136 * The return value has to be stored somewhere, otherwise this is an error
3137 * due to -Werror=unused-value. */
3139 strstart(filename
, "host_device:", &filename
) ||
3140 strstart(filename
, "host_cdrom:" , &filename
);
3144 ret
= raw_normalize_devicepath(&filename
);
3146 error_setg_errno(errp
, -ret
, "Could not normalize device path");
3150 /* Read out options */
3151 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3154 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3157 error_setg_errno(errp
, -ret
, "Could not open device");
3161 if (fstat(fd
, &stat_buf
) < 0) {
3163 error_setg_errno(errp
, -ret
, "Could not stat device");
3164 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3166 "The given file is neither a block nor a character device");
3168 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3169 error_setg(errp
, "Device is too small");
3173 if (!ret
&& total_size
) {
3174 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3175 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3176 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3179 ret
= qemu_write_full(fd
, buf
, zero_size
);
3180 ret
= ret
== zero_size
? 0 : -errno
;
3187 static BlockDriver bdrv_host_device
= {
3188 .format_name
= "host_device",
3189 .protocol_name
= "host_device",
3190 .instance_size
= sizeof(BDRVRawState
),
3191 .bdrv_needs_filename
= true,
3192 .bdrv_probe_device
= hdev_probe_device
,
3193 .bdrv_parse_filename
= hdev_parse_filename
,
3194 .bdrv_file_open
= hdev_open
,
3195 .bdrv_close
= raw_close
,
3196 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3197 .bdrv_reopen_commit
= raw_reopen_commit
,
3198 .bdrv_reopen_abort
= raw_reopen_abort
,
3199 .bdrv_co_create_opts
= hdev_co_create_opts
,
3200 .create_opts
= &raw_create_opts
,
3201 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3202 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3204 .bdrv_co_preadv
= raw_co_preadv
,
3205 .bdrv_co_pwritev
= raw_co_pwritev
,
3206 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3207 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3208 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3209 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3210 .bdrv_refresh_limits
= raw_refresh_limits
,
3211 .bdrv_io_plug
= raw_aio_plug
,
3212 .bdrv_io_unplug
= raw_aio_unplug
,
3213 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3215 .bdrv_co_truncate
= raw_co_truncate
,
3216 .bdrv_getlength
= raw_getlength
,
3217 .bdrv_get_info
= raw_get_info
,
3218 .bdrv_get_allocated_file_size
3219 = raw_get_allocated_file_size
,
3220 .bdrv_check_perm
= raw_check_perm
,
3221 .bdrv_set_perm
= raw_set_perm
,
3222 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3223 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3224 .bdrv_probe_geometry
= hdev_probe_geometry
,
3226 /* generic scsi device */
3228 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3232 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3233 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3236 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3241 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3244 BDRVRawState
*s
= bs
->opaque
;
3248 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3249 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3252 static int cdrom_probe_device(const char *filename
)
3258 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3262 ret
= fstat(fd
, &st
);
3263 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3267 /* Attempt to detect via a CDROM specific ioctl */
3268 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3278 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3280 BDRVRawState
*s
= bs
->opaque
;
3283 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3284 return ret
== CDS_DISC_OK
;
3287 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3289 BDRVRawState
*s
= bs
->opaque
;
3292 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3293 perror("CDROMEJECT");
3295 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3296 perror("CDROMEJECT");
3300 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3302 BDRVRawState
*s
= bs
->opaque
;
3304 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3306 * Note: an error can happen if the distribution automatically
3309 /* perror("CDROM_LOCKDOOR"); */
3313 static BlockDriver bdrv_host_cdrom
= {
3314 .format_name
= "host_cdrom",
3315 .protocol_name
= "host_cdrom",
3316 .instance_size
= sizeof(BDRVRawState
),
3317 .bdrv_needs_filename
= true,
3318 .bdrv_probe_device
= cdrom_probe_device
,
3319 .bdrv_parse_filename
= cdrom_parse_filename
,
3320 .bdrv_file_open
= cdrom_open
,
3321 .bdrv_close
= raw_close
,
3322 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3323 .bdrv_reopen_commit
= raw_reopen_commit
,
3324 .bdrv_reopen_abort
= raw_reopen_abort
,
3325 .bdrv_co_create_opts
= hdev_co_create_opts
,
3326 .create_opts
= &raw_create_opts
,
3327 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3330 .bdrv_co_preadv
= raw_co_preadv
,
3331 .bdrv_co_pwritev
= raw_co_pwritev
,
3332 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3333 .bdrv_refresh_limits
= raw_refresh_limits
,
3334 .bdrv_io_plug
= raw_aio_plug
,
3335 .bdrv_io_unplug
= raw_aio_unplug
,
3336 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3338 .bdrv_co_truncate
= raw_co_truncate
,
3339 .bdrv_getlength
= raw_getlength
,
3340 .has_variable_length
= true,
3341 .bdrv_get_allocated_file_size
3342 = raw_get_allocated_file_size
,
3344 /* removable device support */
3345 .bdrv_is_inserted
= cdrom_is_inserted
,
3346 .bdrv_eject
= cdrom_eject
,
3347 .bdrv_lock_medium
= cdrom_lock_medium
,
3349 /* generic scsi device */
3350 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3352 #endif /* __linux__ */
3354 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3355 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3358 BDRVRawState
*s
= bs
->opaque
;
3359 Error
*local_err
= NULL
;
3364 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3366 error_propagate(errp
, local_err
);
3370 /* make sure the door isn't locked at this time */
3371 ioctl(s
->fd
, CDIOCALLOW
);
3375 static int cdrom_probe_device(const char *filename
)
3377 if (strstart(filename
, "/dev/cd", NULL
) ||
3378 strstart(filename
, "/dev/acd", NULL
))
3383 static int cdrom_reopen(BlockDriverState
*bs
)
3385 BDRVRawState
*s
= bs
->opaque
;
3389 * Force reread of possibly changed/newly loaded disc,
3390 * FreeBSD seems to not notice sometimes...
3394 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3401 /* make sure the door isn't locked at this time */
3402 ioctl(s
->fd
, CDIOCALLOW
);
3406 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3408 return raw_getlength(bs
) > 0;
3411 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3413 BDRVRawState
*s
= bs
->opaque
;
3418 (void) ioctl(s
->fd
, CDIOCALLOW
);
3421 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3422 perror("CDIOCEJECT");
3424 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3425 perror("CDIOCCLOSE");
3431 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3433 BDRVRawState
*s
= bs
->opaque
;
3437 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3439 * Note: an error can happen if the distribution automatically
3442 /* perror("CDROM_LOCKDOOR"); */
3446 static BlockDriver bdrv_host_cdrom
= {
3447 .format_name
= "host_cdrom",
3448 .protocol_name
= "host_cdrom",
3449 .instance_size
= sizeof(BDRVRawState
),
3450 .bdrv_needs_filename
= true,
3451 .bdrv_probe_device
= cdrom_probe_device
,
3452 .bdrv_parse_filename
= cdrom_parse_filename
,
3453 .bdrv_file_open
= cdrom_open
,
3454 .bdrv_close
= raw_close
,
3455 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3456 .bdrv_reopen_commit
= raw_reopen_commit
,
3457 .bdrv_reopen_abort
= raw_reopen_abort
,
3458 .bdrv_co_create_opts
= hdev_co_create_opts
,
3459 .create_opts
= &raw_create_opts
,
3461 .bdrv_co_preadv
= raw_co_preadv
,
3462 .bdrv_co_pwritev
= raw_co_pwritev
,
3463 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3464 .bdrv_refresh_limits
= raw_refresh_limits
,
3465 .bdrv_io_plug
= raw_aio_plug
,
3466 .bdrv_io_unplug
= raw_aio_unplug
,
3467 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3469 .bdrv_co_truncate
= raw_co_truncate
,
3470 .bdrv_getlength
= raw_getlength
,
3471 .has_variable_length
= true,
3472 .bdrv_get_allocated_file_size
3473 = raw_get_allocated_file_size
,
3475 /* removable device support */
3476 .bdrv_is_inserted
= cdrom_is_inserted
,
3477 .bdrv_eject
= cdrom_eject
,
3478 .bdrv_lock_medium
= cdrom_lock_medium
,
3480 #endif /* __FreeBSD__ */
3482 static void bdrv_file_init(void)
3485 * Register all the drivers. Note that order is important, the driver
3486 * registered last will get probed first.
3488 bdrv_register(&bdrv_file
);
3489 bdrv_register(&bdrv_host_device
);
3491 bdrv_register(&bdrv_host_cdrom
);
3493 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3494 bdrv_register(&bdrv_host_cdrom
);
3498 block_init(bdrv_file_init
);