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
, Error
**errp
)
210 static char namebuf
[PATH_MAX
];
211 const char *dp
, *fname
;
215 dp
= strrchr(fname
, '/');
216 if (lstat(fname
, &sb
) < 0) {
217 error_setg_errno(errp
, errno
, "%s: stat failed", fname
);
221 if (!S_ISBLK(sb
.st_mode
)) {
226 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
228 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
229 (int)(dp
- fname
), fname
, dp
+ 1);
232 warn_report("%s is a block device, using %s", fname
, *filename
);
237 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
244 * Get logical block size via ioctl. On success store it in @sector_size_p.
246 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
248 unsigned int sector_size
;
249 bool success
= false;
253 static const unsigned long ioctl_list
[] = {
257 #ifdef DKIOCGETBLOCKSIZE
260 #ifdef DIOCGSECTORSIZE
265 /* Try a few ioctls to get the right size */
266 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
267 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
268 *sector_size_p
= sector_size
;
273 return success
? 0 : -errno
;
277 * Get physical block size of @fd.
278 * On success, store it in @blk_size and return 0.
279 * On failure, return -errno.
281 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
284 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
293 /* Check if read is allowed with given memory buffer and length.
295 * This function is used to check O_DIRECT memory buffer and request alignment.
297 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
299 ssize_t ret
= pread(fd
, buf
, len
, 0);
306 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
307 * other errors (e.g. real I/O error), which could happen on a failed
308 * drive, since we only care about probing alignment.
310 if (errno
!= EINVAL
) {
318 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
320 BDRVRawState
*s
= bs
->opaque
;
322 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
324 /* For SCSI generic devices the alignment is not really used.
325 With buffered I/O, we don't have any restrictions. */
326 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
327 bs
->bl
.request_alignment
= 1;
332 bs
->bl
.request_alignment
= 0;
334 /* Let's try to use the logical blocksize for the alignment. */
335 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
336 bs
->bl
.request_alignment
= 0;
341 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
342 bs
->bl
.request_alignment
= da
.d_miniosz
;
343 /* The kernel returns wrong information for d_mem */
344 /* s->buf_align = da.d_mem; */
349 /* If we could not get the sizes so far, we can only guess them */
352 buf
= qemu_memalign(max_align
, 2 * max_align
);
353 for (align
= 512; align
<= max_align
; align
<<= 1) {
354 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
355 s
->buf_align
= align
;
362 if (!bs
->bl
.request_alignment
) {
364 buf
= qemu_memalign(s
->buf_align
, max_align
);
365 for (align
= 512; align
<= max_align
; align
<<= 1) {
366 if (raw_is_io_aligned(fd
, buf
, align
)) {
367 bs
->bl
.request_alignment
= align
;
374 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
375 error_setg(errp
, "Could not find working O_DIRECT alignment");
376 error_append_hint(errp
, "Try cache.direct=off\n");
380 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
382 assert(open_flags
!= NULL
);
384 *open_flags
|= O_BINARY
;
385 *open_flags
&= ~O_ACCMODE
;
386 if (bdrv_flags
& BDRV_O_RDWR
) {
387 *open_flags
|= O_RDWR
;
389 *open_flags
|= O_RDONLY
;
392 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
393 * and O_DIRECT for no caching. */
394 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
395 *open_flags
|= O_DIRECT
;
399 static void raw_parse_filename(const char *filename
, QDict
*options
,
402 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
405 static QemuOptsList raw_runtime_opts
= {
407 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
411 .type
= QEMU_OPT_STRING
,
412 .help
= "File name of the image",
416 .type
= QEMU_OPT_STRING
,
417 .help
= "host AIO implementation (threads, native)",
421 .type
= QEMU_OPT_STRING
,
422 .help
= "file locking mode (on/off/auto, default: auto)",
425 .name
= "pr-manager",
426 .type
= QEMU_OPT_STRING
,
427 .help
= "id of persistent reservation manager object (default: none)",
430 .name
= "x-check-cache-dropped",
431 .type
= QEMU_OPT_BOOL
,
432 .help
= "check that page cache was dropped on live migration (default: off)"
434 { /* end of list */ }
438 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
439 int bdrv_flags
, int open_flags
,
440 bool device
, Error
**errp
)
442 BDRVRawState
*s
= bs
->opaque
;
444 Error
*local_err
= NULL
;
445 const char *filename
= NULL
;
447 BlockdevAioOptions aio
, aio_default
;
452 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
453 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
455 error_propagate(errp
, local_err
);
460 filename
= qemu_opt_get(opts
, "filename");
462 ret
= raw_normalize_devicepath(&filename
, errp
);
467 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
468 ? BLOCKDEV_AIO_OPTIONS_NATIVE
469 : BLOCKDEV_AIO_OPTIONS_THREADS
;
470 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
471 qemu_opt_get(opts
, "aio"),
472 aio_default
, &local_err
);
474 error_propagate(errp
, local_err
);
478 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
480 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
481 qemu_opt_get(opts
, "locking"),
482 ON_OFF_AUTO_AUTO
, &local_err
);
484 error_propagate(errp
, local_err
);
491 if (!qemu_has_ofd_lock()) {
492 warn_report("File lock requested but OFD locking syscall is "
493 "unavailable, falling back to POSIX file locks");
494 error_printf("Due to the implementation, locks can be lost "
498 case ON_OFF_AUTO_OFF
:
501 case ON_OFF_AUTO_AUTO
:
502 s
->use_lock
= qemu_has_ofd_lock();
508 str
= qemu_opt_get(opts
, "pr-manager");
510 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
512 error_propagate(errp
, local_err
);
518 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
521 s
->open_flags
= open_flags
;
522 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
525 fd
= qemu_open(filename
, s
->open_flags
, 0644);
526 ret
= fd
< 0 ? -errno
: 0;
528 if (ret
== -EACCES
|| ret
== -EROFS
) {
529 /* Try to degrade to read-only, but if it doesn't work, still use the
530 * normal error message. */
531 if (bdrv_apply_auto_read_only(bs
, NULL
, NULL
) == 0) {
532 bdrv_flags
&= ~BDRV_O_RDWR
;
533 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
534 assert(!(s
->open_flags
& O_CREAT
));
535 fd
= qemu_open(filename
, s
->open_flags
);
536 ret
= fd
< 0 ? -errno
: 0;
541 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
551 fd
= qemu_open(filename
, s
->open_flags
);
554 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
562 s
->shared_perm
= BLK_PERM_ALL
;
564 #ifdef CONFIG_LINUX_AIO
565 /* Currently Linux does AIO only for files opened with O_DIRECT */
566 if (s
->use_linux_aio
) {
567 if (!(s
->open_flags
& O_DIRECT
)) {
568 error_setg(errp
, "aio=native was specified, but it requires "
569 "cache.direct=on, which was not specified.");
573 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
574 error_prepend(errp
, "Unable to use native AIO: ");
579 if (s
->use_linux_aio
) {
580 error_setg(errp
, "aio=native was specified, but is not supported "
585 #endif /* !defined(CONFIG_LINUX_AIO) */
587 s
->has_discard
= true;
588 s
->has_write_zeroes
= true;
589 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
590 s
->needs_alignment
= true;
593 if (fstat(s
->fd
, &st
) < 0) {
595 error_setg_errno(errp
, errno
, "Could not stat file");
600 if (S_ISBLK(st
.st_mode
)) {
601 warn_report("Opening a block device as a file using the '%s' "
602 "driver is deprecated", bs
->drv
->format_name
);
603 } else if (S_ISCHR(st
.st_mode
)) {
604 warn_report("Opening a character device as a file using the '%s' "
605 "driver is deprecated", bs
->drv
->format_name
);
606 } else if (!S_ISREG(st
.st_mode
)) {
607 error_setg(errp
, "A regular file was expected by the '%s' driver, "
608 "but something else was given", bs
->drv
->format_name
);
612 s
->discard_zeroes
= true;
613 s
->has_fallocate
= true;
616 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
617 error_setg(errp
, "'%s' driver expects either "
618 "a character or block device", bs
->drv
->format_name
);
624 if (S_ISBLK(st
.st_mode
)) {
625 #ifdef BLKDISCARDZEROES
627 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
628 s
->discard_zeroes
= true;
632 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
633 * not rely on the contents of discarded blocks unless using O_DIRECT.
634 * Same for BLKZEROOUT.
636 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
637 s
->discard_zeroes
= false;
638 s
->has_write_zeroes
= false;
643 if (S_ISCHR(st
.st_mode
)) {
645 * The file is a char device (disk), which on FreeBSD isn't behind
646 * a pager, so force all requests to be aligned. This is needed
647 * so QEMU makes sure all IO operations on the device are aligned
648 * to sector size, or else FreeBSD will reject them with EINVAL.
650 s
->needs_alignment
= true;
655 if (platform_test_xfs_fd(s
->fd
)) {
660 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
663 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
670 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
673 BDRVRawState
*s
= bs
->opaque
;
675 s
->type
= FTYPE_FILE
;
676 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
685 #define PERM_FOREACH(i) \
686 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
688 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
689 * file; if @unlock == true, also unlock the unneeded bytes.
690 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
692 static int raw_apply_lock_bytes(int fd
,
693 uint64_t perm_lock_bits
,
694 uint64_t shared_perm_lock_bits
,
695 bool unlock
, Error
**errp
)
701 int off
= RAW_LOCK_PERM_BASE
+ i
;
702 if (perm_lock_bits
& (1ULL << i
)) {
703 ret
= qemu_lock_fd(fd
, off
, 1, false);
705 error_setg(errp
, "Failed to lock byte %d", off
);
709 ret
= qemu_unlock_fd(fd
, off
, 1);
711 error_setg(errp
, "Failed to unlock byte %d", off
);
717 int off
= RAW_LOCK_SHARED_BASE
+ i
;
718 if (shared_perm_lock_bits
& (1ULL << i
)) {
719 ret
= qemu_lock_fd(fd
, off
, 1, false);
721 error_setg(errp
, "Failed to lock byte %d", off
);
725 ret
= qemu_unlock_fd(fd
, off
, 1);
727 error_setg(errp
, "Failed to unlock byte %d", off
);
735 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
736 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
743 int off
= RAW_LOCK_SHARED_BASE
+ i
;
744 uint64_t p
= 1ULL << i
;
746 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
748 char *perm_name
= bdrv_perm_names(p
);
750 "Failed to get \"%s\" lock",
758 int off
= RAW_LOCK_PERM_BASE
+ i
;
759 uint64_t p
= 1ULL << i
;
760 if (!(shared_perm
& p
)) {
761 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
763 char *perm_name
= bdrv_perm_names(p
);
765 "Failed to get shared \"%s\" lock",
775 static int raw_handle_perm_lock(BlockDriverState
*bs
,
777 uint64_t new_perm
, uint64_t new_shared
,
780 BDRVRawState
*s
= bs
->opaque
;
782 Error
*local_err
= NULL
;
788 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
792 assert(s
->lock_fd
> 0);
796 ret
= raw_apply_lock_bytes(s
->lock_fd
, s
->perm
| new_perm
,
797 ~s
->shared_perm
| ~new_shared
,
800 ret
= raw_check_lock_bytes(s
->lock_fd
, new_perm
, new_shared
, errp
);
804 error_append_hint(errp
,
805 "Is another process using the image [%s]?\n",
809 /* fall through to unlock bytes. */
811 raw_apply_lock_bytes(s
->lock_fd
, s
->perm
, ~s
->shared_perm
,
814 /* Theoretically the above call only unlocks bytes and it cannot
815 * fail. Something weird happened, report it.
817 warn_report_err(local_err
);
821 raw_apply_lock_bytes(s
->lock_fd
, new_perm
, ~new_shared
,
824 /* Theoretically the above call only unlocks bytes and it cannot
825 * fail. Something weird happened, report it.
827 warn_report_err(local_err
);
834 static int raw_reopen_prepare(BDRVReopenState
*state
,
835 BlockReopenQueue
*queue
, Error
**errp
)
838 BDRVRawReopenState
*rs
;
841 Error
*local_err
= NULL
;
843 assert(state
!= NULL
);
844 assert(state
->bs
!= NULL
);
846 s
= state
->bs
->opaque
;
848 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
852 /* Handle options changes */
853 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
854 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
856 error_propagate(errp
, local_err
);
861 rs
->check_cache_dropped
=
862 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
864 /* This driver's reopen function doesn't currently allow changing
865 * other options, so let's put them back in the original QDict and
866 * bdrv_reopen_prepare() will detect changes and complain. */
867 qemu_opts_to_qdict(opts
, state
->options
);
869 if (s
->type
== FTYPE_CD
) {
870 rs
->open_flags
|= O_NONBLOCK
;
873 raw_parse_flags(state
->flags
, &rs
->open_flags
);
875 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
877 fcntl_flags
|= O_NOATIME
;
881 /* Not all operating systems have O_ASYNC, and those that don't
882 * will not let us track the state into rs->open_flags (typically
883 * you achieve the same effect with an ioctl, for example I_SETSIG
884 * on Solaris). But we do not use O_ASYNC, so that's fine.
886 assert((s
->open_flags
& O_ASYNC
) == 0);
889 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
890 /* dup the original fd */
891 rs
->fd
= qemu_dup(s
->fd
);
893 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
901 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
903 const char *normalized_filename
= state
->bs
->filename
;
904 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
906 assert(!(rs
->open_flags
& O_CREAT
));
907 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
909 error_setg_errno(errp
, errno
, "Could not reopen file");
915 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
916 * alignment with the new fd. */
918 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
922 error_propagate(errp
, local_err
);
932 static void raw_reopen_commit(BDRVReopenState
*state
)
934 BDRVRawReopenState
*rs
= state
->opaque
;
935 BDRVRawState
*s
= state
->bs
->opaque
;
937 s
->check_cache_dropped
= rs
->check_cache_dropped
;
938 s
->open_flags
= rs
->open_flags
;
943 g_free(state
->opaque
);
944 state
->opaque
= NULL
;
948 static void raw_reopen_abort(BDRVReopenState
*state
)
950 BDRVRawReopenState
*rs
= state
->opaque
;
952 /* nothing to do if NULL, we didn't get far enough */
961 g_free(state
->opaque
);
962 state
->opaque
= NULL
;
965 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
969 short max_sectors
= 0;
970 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
972 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
973 return max_sectors
<< BDRV_SECTOR_BITS
;
982 static int hdev_get_max_segments(const struct stat
*st
)
992 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
993 major(st
->st_rdev
), minor(st
->st_rdev
));
994 fd
= open(sysfspath
, O_RDONLY
);
1000 ret
= read(fd
, buf
, sizeof(buf
) - 1);
1001 } while (ret
== -1 && errno
== EINTR
);
1005 } else if (ret
== 0) {
1010 /* The file is ended with '\n', pass 'end' to accept that. */
1011 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1012 if (ret
== 0 && end
&& *end
== '\n') {
1027 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1029 BDRVRawState
*s
= bs
->opaque
;
1032 if (!fstat(s
->fd
, &st
)) {
1033 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1034 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1035 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1036 bs
->bl
.max_transfer
= pow2floor(ret
);
1038 ret
= hdev_get_max_segments(&st
);
1040 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1041 ret
* getpagesize());
1046 raw_probe_alignment(bs
, s
->fd
, errp
);
1047 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1048 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1051 static int check_for_dasd(int fd
)
1054 struct dasd_information2_t info
= {0};
1056 return ioctl(fd
, BIODASDINFO2
, &info
);
1063 * Try to get @bs's logical and physical block size.
1064 * On success, store them in @bsz and return zero.
1065 * On failure, return negative errno.
1067 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1069 BDRVRawState
*s
= bs
->opaque
;
1072 /* If DASD, get blocksizes */
1073 if (check_for_dasd(s
->fd
) < 0) {
1076 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1080 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1084 * Try to get @bs's geometry: cyls, heads, sectors.
1085 * On success, store them in @geo and return 0.
1086 * On failure return -errno.
1087 * (Allows block driver to assign default geometry values that guest sees)
1090 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1092 BDRVRawState
*s
= bs
->opaque
;
1093 struct hd_geometry ioctl_geo
= {0};
1095 /* If DASD, get its geometry */
1096 if (check_for_dasd(s
->fd
) < 0) {
1099 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1102 /* HDIO_GETGEO may return success even though geo contains zeros
1103 (e.g. certain multipath setups) */
1104 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1107 /* Do not return a geometry for partition */
1108 if (ioctl_geo
.start
!= 0) {
1111 geo
->heads
= ioctl_geo
.heads
;
1112 geo
->sectors
= ioctl_geo
.sectors
;
1113 geo
->cylinders
= ioctl_geo
.cylinders
;
1117 #else /* __linux__ */
1118 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1124 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1128 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1136 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1138 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1141 if (s
->page_cache_inconsistent
) {
1145 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1147 /* There is no clear definition of the semantics of a failing fsync(),
1148 * so we may have to assume the worst. The sad truth is that this
1149 * assumption is correct for Linux. Some pages are now probably marked
1150 * clean in the page cache even though they are inconsistent with the
1151 * on-disk contents. The next fdatasync() call would succeed, but no
1152 * further writeback attempt will be made. We can't get back to a state
1153 * in which we know what is on disk (we would have to rewrite
1154 * everything that was touched since the last fdatasync() at least), so
1155 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1156 * really defined, I have little hope that other OSes are doing better.
1158 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1160 if ((s
->open_flags
& O_DIRECT
) == 0) {
1161 s
->page_cache_inconsistent
= true;
1168 #ifdef CONFIG_PREADV
1170 static bool preadv_present
= true;
1173 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1175 return preadv(fd
, iov
, nr_iov
, offset
);
1179 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1181 return pwritev(fd
, iov
, nr_iov
, offset
);
1186 static bool preadv_present
= false;
1189 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1195 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1202 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1207 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1208 len
= qemu_pwritev(aiocb
->aio_fildes
,
1213 len
= qemu_preadv(aiocb
->aio_fildes
,
1217 } while (len
== -1 && errno
== EINTR
);
1226 * Read/writes the data to/from a given linear buffer.
1228 * Returns the number of bytes handles or -errno in case of an error. Short
1229 * reads are only returned if the end of the file is reached.
1231 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1236 while (offset
< aiocb
->aio_nbytes
) {
1237 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1238 len
= pwrite(aiocb
->aio_fildes
,
1239 (const char *)buf
+ offset
,
1240 aiocb
->aio_nbytes
- offset
,
1241 aiocb
->aio_offset
+ offset
);
1243 len
= pread(aiocb
->aio_fildes
,
1245 aiocb
->aio_nbytes
- offset
,
1246 aiocb
->aio_offset
+ offset
);
1248 if (len
== -1 && errno
== EINTR
) {
1250 } else if (len
== -1 && errno
== EINVAL
&&
1251 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1252 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1254 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1255 * after a short read. Assume that O_DIRECT short reads only occur
1256 * at EOF. Therefore this is a short read, not an I/O error.
1259 } else if (len
== -1) {
1262 } else if (len
== 0) {
1271 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1276 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1278 * If there is just a single buffer, and it is properly aligned
1279 * we can just use plain pread/pwrite without any problems.
1281 if (aiocb
->aio_niov
== 1) {
1282 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1285 * We have more than one iovec, and all are properly aligned.
1287 * Try preadv/pwritev first and fall back to linearizing the
1288 * buffer if it's not supported.
1290 if (preadv_present
) {
1291 nbytes
= handle_aiocb_rw_vector(aiocb
);
1292 if (nbytes
== aiocb
->aio_nbytes
||
1293 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1296 preadv_present
= false;
1300 * XXX(hch): short read/write. no easy way to handle the reminder
1301 * using these interfaces. For now retry using plain
1307 * Ok, we have to do it the hard way, copy all segments into
1308 * a single aligned buffer.
1310 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1315 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1319 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1320 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1321 p
+= aiocb
->aio_iov
[i
].iov_len
;
1323 assert(p
- buf
== aiocb
->aio_nbytes
);
1326 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1327 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1329 size_t count
= aiocb
->aio_nbytes
, copy
;
1332 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1334 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1335 copy
= aiocb
->aio_iov
[i
].iov_len
;
1337 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1338 assert(count
>= copy
);
1350 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1352 struct xfs_flock64 fl
;
1355 memset(&fl
, 0, sizeof(fl
));
1356 fl
.l_whence
= SEEK_SET
;
1357 fl
.l_start
= offset
;
1360 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1362 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1369 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1371 struct xfs_flock64 fl
;
1374 memset(&fl
, 0, sizeof(fl
));
1375 fl
.l_whence
= SEEK_SET
;
1376 fl
.l_start
= offset
;
1379 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1381 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1389 static int translate_err(int err
)
1391 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1398 #ifdef CONFIG_FALLOCATE
1399 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1402 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1405 } while (errno
== EINTR
);
1406 return translate_err(-errno
);
1410 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1413 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1415 if (!s
->has_write_zeroes
) {
1421 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1422 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1425 } while (errno
== EINTR
);
1427 ret
= translate_err(-errno
);
1430 if (ret
== -ENOTSUP
) {
1431 s
->has_write_zeroes
= false;
1436 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1438 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1439 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1441 #ifdef CONFIG_FALLOCATE
1445 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1446 return handle_aiocb_write_zeroes_block(aiocb
);
1451 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1455 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1456 if (s
->has_write_zeroes
) {
1457 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1458 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1459 if (ret
== 0 || ret
!= -ENOTSUP
) {
1462 s
->has_write_zeroes
= false;
1466 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1467 if (s
->has_discard
&& s
->has_fallocate
) {
1468 int ret
= do_fallocate(s
->fd
,
1469 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1470 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1472 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1473 if (ret
== 0 || ret
!= -ENOTSUP
) {
1476 s
->has_fallocate
= false;
1477 } else if (ret
!= -ENOTSUP
) {
1480 s
->has_discard
= false;
1485 #ifdef CONFIG_FALLOCATE
1486 /* Last resort: we are trying to extend the file with zeroed data. This
1487 * can be done via fallocate(fd, 0) */
1488 len
= bdrv_getlength(aiocb
->bs
);
1489 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1490 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1491 if (ret
== 0 || ret
!= -ENOTSUP
) {
1494 s
->has_fallocate
= false;
1501 static ssize_t
handle_aiocb_write_zeroes_unmap(RawPosixAIOData
*aiocb
)
1503 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1506 /* First try to write zeros and unmap at the same time */
1508 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1509 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1510 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1511 if (ret
!= -ENOTSUP
) {
1518 /* xfs_discard() guarantees that the discarded area reads as all-zero
1519 * afterwards, so we can use it here. */
1520 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1524 /* If we couldn't manage to unmap while guaranteed that the area reads as
1525 * all-zero afterwards, just write zeroes without unmapping */
1526 ret
= handle_aiocb_write_zeroes(aiocb
);
1530 #ifndef HAVE_COPY_FILE_RANGE
1531 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1532 off_t
*out_off
, size_t len
, unsigned int flags
)
1534 #ifdef __NR_copy_file_range
1535 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1536 out_off
, len
, flags
);
1544 static ssize_t
handle_aiocb_copy_range(RawPosixAIOData
*aiocb
)
1546 uint64_t bytes
= aiocb
->aio_nbytes
;
1547 off_t in_off
= aiocb
->aio_offset
;
1548 off_t out_off
= aiocb
->aio_offset2
;
1551 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1552 aiocb
->aio_fd2
, &out_off
,
1554 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1555 aiocb
->aio_fd2
, out_off
, bytes
, 0, ret
);
1557 /* No progress (e.g. when beyond EOF), let the caller fall back to
1576 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1578 int ret
= -EOPNOTSUPP
;
1579 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1581 if (!s
->has_discard
) {
1585 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1588 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1589 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1592 } while (errno
== EINTR
);
1599 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1603 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1604 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1605 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1609 ret
= translate_err(ret
);
1610 if (ret
== -ENOTSUP
) {
1611 s
->has_discard
= false;
1616 static int handle_aiocb_truncate(RawPosixAIOData
*aiocb
)
1619 int64_t current_length
= 0;
1622 int fd
= aiocb
->aio_fildes
;
1623 int64_t offset
= aiocb
->aio_offset
;
1624 Error
**errp
= aiocb
->errp
;
1626 if (fstat(fd
, &st
) < 0) {
1628 error_setg_errno(errp
, -result
, "Could not stat file");
1632 current_length
= st
.st_size
;
1633 if (current_length
> offset
&& aiocb
->prealloc
!= PREALLOC_MODE_OFF
) {
1634 error_setg(errp
, "Cannot use preallocation for shrinking files");
1638 switch (aiocb
->prealloc
) {
1639 #ifdef CONFIG_POSIX_FALLOCATE
1640 case PREALLOC_MODE_FALLOC
:
1642 * Truncating before posix_fallocate() makes it about twice slower on
1643 * file systems that do not support fallocate(), trying to check if a
1644 * block is allocated before allocating it, so don't do that here.
1646 if (offset
!= current_length
) {
1647 result
= -posix_fallocate(fd
, current_length
,
1648 offset
- current_length
);
1650 /* posix_fallocate() doesn't set errno. */
1651 error_setg_errno(errp
, -result
,
1652 "Could not preallocate new data");
1659 case PREALLOC_MODE_FULL
:
1661 int64_t num
= 0, left
= offset
- current_length
;
1665 * Knowing the final size from the beginning could allow the file
1666 * system driver to do less allocations and possibly avoid
1667 * fragmentation of the file.
1669 if (ftruncate(fd
, offset
) != 0) {
1671 error_setg_errno(errp
, -result
, "Could not resize file");
1675 buf
= g_malloc0(65536);
1677 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1678 if (seek_result
< 0) {
1680 error_setg_errno(errp
, -result
,
1681 "Failed to seek to the old end of file");
1686 num
= MIN(left
, 65536);
1687 result
= write(fd
, buf
, num
);
1689 if (errno
== EINTR
) {
1693 error_setg_errno(errp
, -result
,
1694 "Could not write zeros for preallocation");
1703 error_setg_errno(errp
, -result
,
1704 "Could not flush file to disk");
1710 case PREALLOC_MODE_OFF
:
1711 if (ftruncate(fd
, offset
) != 0) {
1713 error_setg_errno(errp
, -result
, "Could not resize file");
1718 error_setg(errp
, "Unsupported preallocation mode: %s",
1719 PreallocMode_str(aiocb
->prealloc
));
1725 if (ftruncate(fd
, current_length
) < 0) {
1726 error_report("Failed to restore old file length: %s",
1735 static int aio_worker(void *arg
)
1737 RawPosixAIOData
*aiocb
= arg
;
1740 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1742 ret
= handle_aiocb_rw(aiocb
);
1743 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1744 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1745 0, aiocb
->aio_nbytes
- ret
);
1747 ret
= aiocb
->aio_nbytes
;
1749 if (ret
== aiocb
->aio_nbytes
) {
1751 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1755 case QEMU_AIO_WRITE
:
1756 ret
= handle_aiocb_rw(aiocb
);
1757 if (ret
== aiocb
->aio_nbytes
) {
1759 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1763 case QEMU_AIO_FLUSH
:
1764 ret
= handle_aiocb_flush(aiocb
);
1766 case QEMU_AIO_IOCTL
:
1767 ret
= handle_aiocb_ioctl(aiocb
);
1769 case QEMU_AIO_DISCARD
:
1770 ret
= handle_aiocb_discard(aiocb
);
1772 case QEMU_AIO_WRITE_ZEROES
:
1773 ret
= handle_aiocb_write_zeroes(aiocb
);
1775 case QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_DISCARD
:
1776 ret
= handle_aiocb_write_zeroes_unmap(aiocb
);
1778 case QEMU_AIO_COPY_RANGE
:
1779 ret
= handle_aiocb_copy_range(aiocb
);
1781 case QEMU_AIO_TRUNCATE
:
1782 ret
= handle_aiocb_truncate(aiocb
);
1785 error_report("invalid aio request (0x%x)", aiocb
->aio_type
);
1794 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1795 int64_t offset
, int fd2
, int64_t offset2
,
1797 int bytes
, int type
)
1799 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1803 acb
->aio_type
= type
;
1804 acb
->aio_fildes
= fd
;
1806 acb
->aio_offset2
= offset2
;
1808 acb
->aio_nbytes
= bytes
;
1809 acb
->aio_offset
= offset
;
1812 acb
->aio_iov
= qiov
->iov
;
1813 acb
->aio_niov
= qiov
->niov
;
1814 assert(qiov
->size
== bytes
);
1817 trace_file_paio_submit_co(offset
, bytes
, type
);
1818 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1819 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1822 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1823 int64_t offset
, QEMUIOVector
*qiov
,
1824 int bytes
, int type
)
1826 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1829 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1830 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1832 BDRVRawState
*s
= bs
->opaque
;
1834 if (fd_open(bs
) < 0)
1838 * Check if the underlying device requires requests to be aligned,
1839 * and if the request we are trying to submit is aligned or not.
1840 * If this is the case tell the low-level driver that it needs
1841 * to copy the buffer.
1843 if (s
->needs_alignment
) {
1844 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1845 type
|= QEMU_AIO_MISALIGNED
;
1846 #ifdef CONFIG_LINUX_AIO
1847 } else if (s
->use_linux_aio
) {
1848 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1849 assert(qiov
->size
== bytes
);
1850 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1855 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1858 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1859 uint64_t bytes
, QEMUIOVector
*qiov
,
1862 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1865 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1866 uint64_t bytes
, QEMUIOVector
*qiov
,
1870 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1873 static void raw_aio_plug(BlockDriverState
*bs
)
1875 #ifdef CONFIG_LINUX_AIO
1876 BDRVRawState
*s
= bs
->opaque
;
1877 if (s
->use_linux_aio
) {
1878 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1879 laio_io_plug(bs
, aio
);
1884 static void raw_aio_unplug(BlockDriverState
*bs
)
1886 #ifdef CONFIG_LINUX_AIO
1887 BDRVRawState
*s
= bs
->opaque
;
1888 if (s
->use_linux_aio
) {
1889 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1890 laio_io_unplug(bs
, aio
);
1895 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1897 BDRVRawState
*s
= bs
->opaque
;
1905 return paio_submit_co(bs
, s
->fd
, 0, NULL
, 0, QEMU_AIO_FLUSH
);
1908 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1909 AioContext
*new_context
)
1911 #ifdef CONFIG_LINUX_AIO
1912 BDRVRawState
*s
= bs
->opaque
;
1913 if (s
->use_linux_aio
) {
1915 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1916 error_reportf_err(local_err
, "Unable to use native AIO, "
1917 "falling back to thread pool: ");
1918 s
->use_linux_aio
= false;
1924 static void raw_close(BlockDriverState
*bs
)
1926 BDRVRawState
*s
= bs
->opaque
;
1932 if (s
->lock_fd
>= 0) {
1933 qemu_close(s
->lock_fd
);
1939 * Truncates the given regular file @fd to @offset and, when growing, fills the
1940 * new space according to @prealloc.
1942 * Returns: 0 on success, -errno on failure.
1944 static int coroutine_fn
1945 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1946 PreallocMode prealloc
, Error
**errp
)
1948 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1951 *acb
= (RawPosixAIOData
) {
1954 .aio_type
= QEMU_AIO_TRUNCATE
,
1955 .aio_offset
= offset
,
1956 .prealloc
= prealloc
,
1960 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1961 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1962 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1965 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1966 PreallocMode prealloc
, Error
**errp
)
1968 BDRVRawState
*s
= bs
->opaque
;
1972 if (fstat(s
->fd
, &st
)) {
1974 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1978 if (S_ISREG(st
.st_mode
)) {
1979 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
1982 if (prealloc
!= PREALLOC_MODE_OFF
) {
1983 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1984 "non-regular file", PreallocMode_str(prealloc
));
1988 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1989 if (offset
> raw_getlength(bs
)) {
1990 error_setg(errp
, "Cannot grow device files");
1994 error_setg(errp
, "Resizing this file is not supported");
2002 static int64_t raw_getlength(BlockDriverState
*bs
)
2004 BDRVRawState
*s
= bs
->opaque
;
2010 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2011 struct disklabel dl
;
2013 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2015 return (uint64_t)dl
.d_secsize
*
2016 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2020 #elif defined(__NetBSD__)
2021 static int64_t raw_getlength(BlockDriverState
*bs
)
2023 BDRVRawState
*s
= bs
->opaque
;
2029 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2030 struct dkwedge_info dkw
;
2032 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2033 return dkw
.dkw_size
* 512;
2035 struct disklabel dl
;
2037 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2039 return (uint64_t)dl
.d_secsize
*
2040 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2045 #elif defined(__sun__)
2046 static int64_t raw_getlength(BlockDriverState
*bs
)
2048 BDRVRawState
*s
= bs
->opaque
;
2049 struct dk_minfo minfo
;
2059 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2061 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2063 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2067 * There are reports that lseek on some devices fails, but
2068 * irc discussion said that contingency on contingency was overkill.
2070 size
= lseek(s
->fd
, 0, SEEK_END
);
2076 #elif defined(CONFIG_BSD)
2077 static int64_t raw_getlength(BlockDriverState
*bs
)
2079 BDRVRawState
*s
= bs
->opaque
;
2083 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2092 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2095 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2096 #ifdef DIOCGMEDIASIZE
2097 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2098 #elif defined(DIOCGPART)
2101 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2102 size
= pi
.media_size
;
2108 #if defined(__APPLE__) && defined(__MACH__)
2110 uint64_t sectors
= 0;
2111 uint32_t sector_size
= 0;
2113 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2114 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2115 size
= sectors
* sector_size
;
2117 size
= lseek(fd
, 0LL, SEEK_END
);
2124 size
= lseek(fd
, 0LL, SEEK_END
);
2129 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2132 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2133 if (size
== 2048LL * (unsigned)-1)
2135 /* XXX no disc? maybe we need to reopen... */
2136 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2143 size
= lseek(fd
, 0, SEEK_END
);
2151 static int64_t raw_getlength(BlockDriverState
*bs
)
2153 BDRVRawState
*s
= bs
->opaque
;
2162 size
= lseek(s
->fd
, 0, SEEK_END
);
2170 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2173 BDRVRawState
*s
= bs
->opaque
;
2175 if (fstat(s
->fd
, &st
) < 0) {
2178 return (int64_t)st
.st_blocks
* 512;
2181 static int coroutine_fn
2182 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2184 BlockdevCreateOptionsFile
*file_opts
;
2185 Error
*local_err
= NULL
;
2187 uint64_t perm
, shared
;
2190 /* Validate options and set default values */
2191 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2192 file_opts
= &options
->u
.file
;
2194 if (!file_opts
->has_nocow
) {
2195 file_opts
->nocow
= false;
2197 if (!file_opts
->has_preallocation
) {
2198 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2202 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2205 error_setg_errno(errp
, -result
, "Could not create file");
2209 /* Take permissions: We want to discard everything, so we need
2210 * BLK_PERM_WRITE; and truncation to the desired size requires
2212 * On the other hand, we cannot share the RESIZE permission
2213 * because we promise that after this function, the file has the
2214 * size given in the options. If someone else were to resize it
2215 * concurrently, we could not guarantee that.
2216 * Note that after this function, we can no longer guarantee that
2217 * the file is not touched by a third party, so it may be resized
2219 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2220 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2222 /* Step one: Take locks */
2223 result
= raw_apply_lock_bytes(fd
, perm
, ~shared
, false, errp
);
2228 /* Step two: Check that nobody else has taken conflicting locks */
2229 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2231 error_append_hint(errp
,
2232 "Is another process using the image [%s]?\n",
2233 file_opts
->filename
);
2237 /* Clear the file by truncating it to 0 */
2238 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2243 if (file_opts
->nocow
) {
2245 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2246 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2247 * will be ignored since any failure of this operation should not
2248 * block the left work.
2251 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2252 attr
|= FS_NOCOW_FL
;
2253 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2258 /* Resize and potentially preallocate the file to the desired
2260 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2261 file_opts
->preallocation
, errp
);
2267 raw_apply_lock_bytes(fd
, 0, 0, true, &local_err
);
2269 /* The above call should not fail, and if it does, that does
2270 * not mean the whole creation operation has failed. So
2271 * report it the user for their convenience, but do not report
2272 * it to the caller. */
2273 warn_report_err(local_err
);
2277 if (qemu_close(fd
) != 0 && result
== 0) {
2279 error_setg_errno(errp
, -result
, "Could not close the new file");
2285 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2288 BlockdevCreateOptions options
;
2289 int64_t total_size
= 0;
2291 PreallocMode prealloc
;
2293 Error
*local_err
= NULL
;
2295 /* Skip file: protocol prefix */
2296 strstart(filename
, "file:", &filename
);
2298 /* Read out options */
2299 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2301 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2302 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2303 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2304 PREALLOC_MODE_OFF
, &local_err
);
2307 error_propagate(errp
, local_err
);
2311 options
= (BlockdevCreateOptions
) {
2312 .driver
= BLOCKDEV_DRIVER_FILE
,
2314 .filename
= (char *) filename
,
2316 .has_preallocation
= true,
2317 .preallocation
= prealloc
,
2322 return raw_co_create(&options
, errp
);
2326 * Find allocation range in @bs around offset @start.
2327 * May change underlying file descriptor's file offset.
2328 * If @start is not in a hole, store @start in @data, and the
2329 * beginning of the next hole in @hole, and return 0.
2330 * If @start is in a non-trailing hole, store @start in @hole and the
2331 * beginning of the next non-hole in @data, and return 0.
2332 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2333 * If we can't find out, return a negative errno other than -ENXIO.
2335 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2336 off_t
*data
, off_t
*hole
)
2338 #if defined SEEK_HOLE && defined SEEK_DATA
2339 BDRVRawState
*s
= bs
->opaque
;
2344 * D1. offs == start: start is in data
2345 * D2. offs > start: start is in a hole, next data at offs
2346 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2347 * or start is beyond EOF
2348 * If the latter happens, the file has been truncated behind
2349 * our back since we opened it. All bets are off then.
2350 * Treating like a trailing hole is simplest.
2351 * D4. offs < 0, errno != ENXIO: we learned nothing
2353 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2355 return -errno
; /* D3 or D4 */
2359 /* This is not a valid return by lseek(). We are safe to just return
2360 * -EIO in this case, and we'll treat it like D4. */
2365 /* D2: in hole, next data at offs */
2371 /* D1: in data, end not yet known */
2375 * H1. offs == start: start is in a hole
2376 * If this happens here, a hole has been dug behind our back
2377 * since the previous lseek().
2378 * H2. offs > start: either start is in data, next hole at offs,
2379 * or start is in trailing hole, EOF at offs
2380 * Linux treats trailing holes like any other hole: offs ==
2381 * start. Solaris seeks to EOF instead: offs > start (blech).
2382 * If that happens here, a hole has been dug behind our back
2383 * since the previous lseek().
2384 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2385 * If this happens, the file has been truncated behind our
2386 * back since we opened it. Treat it like a trailing hole.
2387 * H4. offs < 0, errno != ENXIO: we learned nothing
2388 * Pretend we know nothing at all, i.e. "forget" about D1.
2390 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2392 return -errno
; /* D1 and (H3 or H4) */
2396 /* This is not a valid return by lseek(). We are safe to just return
2397 * -EIO in this case, and we'll treat it like H4. */
2403 * D1 and H2: either in data, next hole at offs, or it was in
2404 * data but is now in a trailing hole. In the latter case,
2405 * all bets are off. Treating it as if it there was data all
2406 * the way to EOF is safe, so simply do that.
2421 * Returns the allocation status of the specified offset.
2423 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2425 * 'pnum' is set to the number of bytes (including and immediately following
2426 * the specified offset) that are known to be in the same
2427 * allocated/unallocated state.
2429 * 'bytes' is the max value 'pnum' should be set to.
2431 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2434 int64_t bytes
, int64_t *pnum
,
2436 BlockDriverState
**file
)
2438 off_t data
= 0, hole
= 0;
2450 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2453 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2454 if (ret
== -ENXIO
) {
2457 ret
= BDRV_BLOCK_ZERO
;
2458 } else if (ret
< 0) {
2459 /* No info available, so pretend there are no holes */
2461 ret
= BDRV_BLOCK_DATA
;
2462 } else if (data
== offset
) {
2463 /* On a data extent, compute bytes to the end of the extent,
2464 * possibly including a partial sector at EOF. */
2465 *pnum
= MIN(bytes
, hole
- offset
);
2466 ret
= BDRV_BLOCK_DATA
;
2468 /* On a hole, compute bytes to the beginning of the next extent. */
2469 assert(hole
== offset
);
2470 *pnum
= MIN(bytes
, data
- offset
);
2471 ret
= BDRV_BLOCK_ZERO
;
2475 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2478 #if defined(__linux__)
2479 /* Verify that the file is not in the page cache */
2480 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2482 const size_t window_size
= 128 * 1024 * 1024;
2483 BDRVRawState
*s
= bs
->opaque
;
2484 void *window
= NULL
;
2491 /* mincore(2) page status information requires 1 byte per page */
2492 page_size
= sysconf(_SC_PAGESIZE
);
2493 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2495 end
= raw_getlength(bs
);
2497 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2504 /* Unmap previous window if size has changed */
2505 new_length
= MIN(end
- offset
, window_size
);
2506 if (new_length
!= length
) {
2507 munmap(window
, length
);
2512 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2514 if (new_window
== MAP_FAILED
) {
2515 error_setg_errno(errp
, errno
, "mmap failed");
2519 window
= new_window
;
2520 length
= new_length
;
2522 ret
= mincore(window
, length
, vec
);
2524 error_setg_errno(errp
, errno
, "mincore failed");
2528 vec_end
= DIV_ROUND_UP(length
, page_size
);
2529 for (i
= 0; i
< vec_end
; i
++) {
2531 error_setg(errp
, "page cache still in use!");
2538 munmap(window
, length
);
2543 #endif /* __linux__ */
2545 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2548 BDRVRawState
*s
= bs
->opaque
;
2553 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2557 if (s
->open_flags
& O_DIRECT
) {
2558 return; /* No host kernel page cache */
2561 #if defined(__linux__)
2562 /* This sets the scene for the next syscall... */
2563 ret
= bdrv_co_flush(bs
);
2565 error_setg_errno(errp
, -ret
, "flush failed");
2569 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2570 * process. These limitations are okay because we just fsynced the file,
2571 * we don't use mmap, and the file should not be in use by other processes.
2573 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2574 if (ret
!= 0) { /* the return value is a positive errno */
2575 error_setg_errno(errp
, ret
, "fadvise failed");
2579 if (s
->check_cache_dropped
) {
2580 check_cache_dropped(bs
, errp
);
2582 #else /* __linux__ */
2583 /* Do nothing. Live migration to a remote host with cache.direct=off is
2584 * unsupported on other host operating systems. Cache consistency issues
2585 * may occur but no error is reported here, partly because that's the
2586 * historical behavior and partly because it's hard to differentiate valid
2587 * configurations that should not cause errors.
2589 #endif /* !__linux__ */
2592 static coroutine_fn
int
2593 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2595 BDRVRawState
*s
= bs
->opaque
;
2597 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, QEMU_AIO_DISCARD
);
2600 static int coroutine_fn
raw_co_pwrite_zeroes(
2601 BlockDriverState
*bs
, int64_t offset
,
2602 int bytes
, BdrvRequestFlags flags
)
2604 BDRVRawState
*s
= bs
->opaque
;
2605 int operation
= QEMU_AIO_WRITE_ZEROES
;
2607 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2608 operation
|= QEMU_AIO_DISCARD
;
2611 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, operation
);
2614 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2616 BDRVRawState
*s
= bs
->opaque
;
2618 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2622 static QemuOptsList raw_create_opts
= {
2623 .name
= "raw-create-opts",
2624 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2627 .name
= BLOCK_OPT_SIZE
,
2628 .type
= QEMU_OPT_SIZE
,
2629 .help
= "Virtual disk size"
2632 .name
= BLOCK_OPT_NOCOW
,
2633 .type
= QEMU_OPT_BOOL
,
2634 .help
= "Turn off copy-on-write (valid only on btrfs)"
2637 .name
= BLOCK_OPT_PREALLOC
,
2638 .type
= QEMU_OPT_STRING
,
2639 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2641 { /* end of list */ }
2645 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2648 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2651 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2653 BDRVRawState
*s
= bs
->opaque
;
2654 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2656 s
->shared_perm
= shared
;
2659 static void raw_abort_perm_update(BlockDriverState
*bs
)
2661 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2664 static int coroutine_fn
raw_co_copy_range_from(
2665 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2666 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2667 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2669 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2670 read_flags
, write_flags
);
2673 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2675 uint64_t src_offset
,
2677 uint64_t dst_offset
,
2679 BdrvRequestFlags read_flags
,
2680 BdrvRequestFlags write_flags
)
2682 BDRVRawState
*s
= bs
->opaque
;
2683 BDRVRawState
*src_s
;
2685 assert(dst
->bs
== bs
);
2686 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2690 src_s
= src
->bs
->opaque
;
2691 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2694 return paio_submit_co_full(bs
, src_s
->fd
, src_offset
, s
->fd
, dst_offset
,
2695 NULL
, bytes
, QEMU_AIO_COPY_RANGE
);
2698 BlockDriver bdrv_file
= {
2699 .format_name
= "file",
2700 .protocol_name
= "file",
2701 .instance_size
= sizeof(BDRVRawState
),
2702 .bdrv_needs_filename
= true,
2703 .bdrv_probe
= NULL
, /* no probe for protocols */
2704 .bdrv_parse_filename
= raw_parse_filename
,
2705 .bdrv_file_open
= raw_open
,
2706 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2707 .bdrv_reopen_commit
= raw_reopen_commit
,
2708 .bdrv_reopen_abort
= raw_reopen_abort
,
2709 .bdrv_close
= raw_close
,
2710 .bdrv_co_create
= raw_co_create
,
2711 .bdrv_co_create_opts
= raw_co_create_opts
,
2712 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2713 .bdrv_co_block_status
= raw_co_block_status
,
2714 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2715 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2717 .bdrv_co_preadv
= raw_co_preadv
,
2718 .bdrv_co_pwritev
= raw_co_pwritev
,
2719 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2720 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2721 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2722 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2723 .bdrv_refresh_limits
= raw_refresh_limits
,
2724 .bdrv_io_plug
= raw_aio_plug
,
2725 .bdrv_io_unplug
= raw_aio_unplug
,
2726 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2728 .bdrv_co_truncate
= raw_co_truncate
,
2729 .bdrv_getlength
= raw_getlength
,
2730 .bdrv_get_info
= raw_get_info
,
2731 .bdrv_get_allocated_file_size
2732 = raw_get_allocated_file_size
,
2733 .bdrv_check_perm
= raw_check_perm
,
2734 .bdrv_set_perm
= raw_set_perm
,
2735 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2736 .create_opts
= &raw_create_opts
,
2739 /***********************************************/
2742 #if defined(__APPLE__) && defined(__MACH__)
2743 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2744 CFIndex maxPathSize
, int flags
);
2745 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2747 kern_return_t kernResult
= KERN_FAILURE
;
2748 mach_port_t masterPort
;
2749 CFMutableDictionaryRef classesToMatch
;
2750 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2751 char *mediaType
= NULL
;
2753 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2754 if ( KERN_SUCCESS
!= kernResult
) {
2755 printf( "IOMasterPort returned %d\n", kernResult
);
2759 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2760 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2761 if (classesToMatch
== NULL
) {
2762 error_report("IOServiceMatching returned NULL for %s",
2763 matching_array
[index
]);
2766 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2768 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2770 if (kernResult
!= KERN_SUCCESS
) {
2771 error_report("Note: IOServiceGetMatchingServices returned %d",
2776 /* If a match was found, leave the loop */
2777 if (*mediaIterator
!= 0) {
2778 DPRINTF("Matching using %s\n", matching_array
[index
]);
2779 mediaType
= g_strdup(matching_array
[index
]);
2786 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2787 CFIndex maxPathSize
, int flags
)
2789 io_object_t nextMedia
;
2790 kern_return_t kernResult
= KERN_FAILURE
;
2792 nextMedia
= IOIteratorNext( mediaIterator
);
2795 CFTypeRef bsdPathAsCFString
;
2796 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2797 if ( bsdPathAsCFString
) {
2798 size_t devPathLength
;
2799 strcpy( bsdPath
, _PATH_DEV
);
2800 if (flags
& BDRV_O_NOCACHE
) {
2801 strcat(bsdPath
, "r");
2803 devPathLength
= strlen( bsdPath
);
2804 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2805 kernResult
= KERN_SUCCESS
;
2807 CFRelease( bsdPathAsCFString
);
2809 IOObjectRelease( nextMedia
);
2815 /* Sets up a real cdrom for use in QEMU */
2816 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2818 int index
, num_of_test_partitions
= 2, fd
;
2819 char test_partition
[MAXPATHLEN
];
2820 bool partition_found
= false;
2822 /* look for a working partition */
2823 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2824 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2826 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2828 partition_found
= true;
2834 /* if a working partition on the device was not found */
2835 if (partition_found
== false) {
2836 error_setg(errp
, "Failed to find a working partition on disc");
2838 DPRINTF("Using %s as optical disc\n", test_partition
);
2839 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2841 return partition_found
;
2844 /* Prints directions on mounting and unmounting a device */
2845 static void print_unmounting_directions(const char *file_name
)
2847 error_report("If device %s is mounted on the desktop, unmount"
2848 " it first before using it in QEMU", file_name
);
2849 error_report("Command to unmount device: diskutil unmountDisk %s",
2851 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2854 #endif /* defined(__APPLE__) && defined(__MACH__) */
2856 static int hdev_probe_device(const char *filename
)
2860 /* allow a dedicated CD-ROM driver to match with a higher priority */
2861 if (strstart(filename
, "/dev/cdrom", NULL
))
2864 if (stat(filename
, &st
) >= 0 &&
2865 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2872 static int check_hdev_writable(BDRVRawState
*s
)
2874 #if defined(BLKROGET)
2875 /* Linux block devices can be configured "read-only" using blockdev(8).
2876 * This is independent of device node permissions and therefore open(2)
2877 * with O_RDWR succeeds. Actual writes fail with EPERM.
2879 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2880 * check for read-only block devices so that Linux block devices behave
2886 if (fstat(s
->fd
, &st
)) {
2890 if (!S_ISBLK(st
.st_mode
)) {
2894 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2901 #endif /* defined(BLKROGET) */
2905 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2908 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2911 static bool hdev_is_sg(BlockDriverState
*bs
)
2914 #if defined(__linux__)
2916 BDRVRawState
*s
= bs
->opaque
;
2918 struct sg_scsi_id scsiid
;
2922 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2926 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2931 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2933 DPRINTF("SG device found: type=%d, version=%d\n",
2934 scsiid
.scsi_type
, sg_version
);
2943 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2946 BDRVRawState
*s
= bs
->opaque
;
2947 Error
*local_err
= NULL
;
2950 #if defined(__APPLE__) && defined(__MACH__)
2952 * Caution: while qdict_get_str() is fine, getting non-string types
2953 * would require more care. When @options come from -blockdev or
2954 * blockdev_add, its members are typed according to the QAPI
2955 * schema, but when they come from -drive, they're all QString.
2957 const char *filename
= qdict_get_str(options
, "filename");
2958 char bsd_path
[MAXPATHLEN
] = "";
2959 bool error_occurred
= false;
2961 /* If using a real cdrom */
2962 if (strcmp(filename
, "/dev/cdrom") == 0) {
2963 char *mediaType
= NULL
;
2964 kern_return_t ret_val
;
2965 io_iterator_t mediaIterator
= 0;
2967 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
2968 if (mediaType
== NULL
) {
2969 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
2971 error_occurred
= true;
2972 goto hdev_open_Mac_error
;
2975 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
2976 if (ret_val
!= KERN_SUCCESS
) {
2977 error_setg(errp
, "Could not get BSD path for optical drive");
2978 error_occurred
= true;
2979 goto hdev_open_Mac_error
;
2982 /* If a real optical drive was not found */
2983 if (bsd_path
[0] == '\0') {
2984 error_setg(errp
, "Failed to obtain bsd path for optical drive");
2985 error_occurred
= true;
2986 goto hdev_open_Mac_error
;
2989 /* If using a cdrom disc and finding a partition on the disc failed */
2990 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
2991 setup_cdrom(bsd_path
, errp
) == false) {
2992 print_unmounting_directions(bsd_path
);
2993 error_occurred
= true;
2994 goto hdev_open_Mac_error
;
2997 qdict_put_str(options
, "filename", bsd_path
);
2999 hdev_open_Mac_error
:
3001 if (mediaIterator
) {
3002 IOObjectRelease(mediaIterator
);
3004 if (error_occurred
) {
3008 #endif /* defined(__APPLE__) && defined(__MACH__) */
3010 s
->type
= FTYPE_FILE
;
3012 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3014 error_propagate(errp
, local_err
);
3015 #if defined(__APPLE__) && defined(__MACH__)
3017 filename
= bsd_path
;
3019 /* if a physical device experienced an error while being opened */
3020 if (strncmp(filename
, "/dev/", 5) == 0) {
3021 print_unmounting_directions(filename
);
3023 #endif /* defined(__APPLE__) && defined(__MACH__) */
3027 /* Since this does ioctl the device must be already opened */
3028 bs
->sg
= hdev_is_sg(bs
);
3030 if (flags
& BDRV_O_RDWR
) {
3031 ret
= check_hdev_writable(s
);
3034 error_setg_errno(errp
, -ret
, "The device is not writable");
3042 #if defined(__linux__)
3044 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
3045 unsigned long int req
, void *buf
,
3046 BlockCompletionFunc
*cb
, void *opaque
)
3048 BDRVRawState
*s
= bs
->opaque
;
3049 RawPosixAIOData
*acb
;
3052 if (fd_open(bs
) < 0)
3055 if (req
== SG_IO
&& s
->pr_mgr
) {
3056 struct sg_io_hdr
*io_hdr
= buf
;
3057 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3058 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3059 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3060 s
->fd
, io_hdr
, cb
, opaque
);
3064 acb
= g_new(RawPosixAIOData
, 1);
3066 acb
->aio_type
= QEMU_AIO_IOCTL
;
3067 acb
->aio_fildes
= s
->fd
;
3068 acb
->aio_offset
= 0;
3069 acb
->aio_ioctl_buf
= buf
;
3070 acb
->aio_ioctl_cmd
= req
;
3071 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3072 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
3076 static int fd_open(BlockDriverState
*bs
)
3078 BDRVRawState
*s
= bs
->opaque
;
3080 /* this is just to ensure s->fd is sane (its called by io ops) */
3086 static coroutine_fn
int
3087 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3089 BDRVRawState
*s
= bs
->opaque
;
3096 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3097 QEMU_AIO_DISCARD
| QEMU_AIO_BLKDEV
);
3100 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3101 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3103 BDRVRawState
*s
= bs
->opaque
;
3104 int operation
= QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_BLKDEV
;
3112 if (flags
& BDRV_REQ_MAY_UNMAP
) {
3113 operation
|= QEMU_AIO_DISCARD
;
3116 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, operation
);
3119 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3124 struct stat stat_buf
;
3125 int64_t total_size
= 0;
3128 /* This function is used by both protocol block drivers and therefore either
3129 * of these prefixes may be given.
3130 * The return value has to be stored somewhere, otherwise this is an error
3131 * due to -Werror=unused-value. */
3133 strstart(filename
, "host_device:", &filename
) ||
3134 strstart(filename
, "host_cdrom:" , &filename
);
3138 ret
= raw_normalize_devicepath(&filename
, errp
);
3143 /* Read out options */
3144 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3147 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3150 error_setg_errno(errp
, -ret
, "Could not open device");
3154 if (fstat(fd
, &stat_buf
) < 0) {
3156 error_setg_errno(errp
, -ret
, "Could not stat device");
3157 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3159 "The given file is neither a block nor a character device");
3161 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3162 error_setg(errp
, "Device is too small");
3166 if (!ret
&& total_size
) {
3167 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3168 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3169 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3172 ret
= qemu_write_full(fd
, buf
, zero_size
);
3173 ret
= ret
== zero_size
? 0 : -errno
;
3180 static BlockDriver bdrv_host_device
= {
3181 .format_name
= "host_device",
3182 .protocol_name
= "host_device",
3183 .instance_size
= sizeof(BDRVRawState
),
3184 .bdrv_needs_filename
= true,
3185 .bdrv_probe_device
= hdev_probe_device
,
3186 .bdrv_parse_filename
= hdev_parse_filename
,
3187 .bdrv_file_open
= hdev_open
,
3188 .bdrv_close
= raw_close
,
3189 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3190 .bdrv_reopen_commit
= raw_reopen_commit
,
3191 .bdrv_reopen_abort
= raw_reopen_abort
,
3192 .bdrv_co_create_opts
= hdev_co_create_opts
,
3193 .create_opts
= &raw_create_opts
,
3194 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3195 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3197 .bdrv_co_preadv
= raw_co_preadv
,
3198 .bdrv_co_pwritev
= raw_co_pwritev
,
3199 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3200 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3201 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3202 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3203 .bdrv_refresh_limits
= raw_refresh_limits
,
3204 .bdrv_io_plug
= raw_aio_plug
,
3205 .bdrv_io_unplug
= raw_aio_unplug
,
3206 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3208 .bdrv_co_truncate
= raw_co_truncate
,
3209 .bdrv_getlength
= raw_getlength
,
3210 .bdrv_get_info
= raw_get_info
,
3211 .bdrv_get_allocated_file_size
3212 = raw_get_allocated_file_size
,
3213 .bdrv_check_perm
= raw_check_perm
,
3214 .bdrv_set_perm
= raw_set_perm
,
3215 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3216 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3217 .bdrv_probe_geometry
= hdev_probe_geometry
,
3219 /* generic scsi device */
3221 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3225 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3226 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3229 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3234 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3237 BDRVRawState
*s
= bs
->opaque
;
3241 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3242 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3245 static int cdrom_probe_device(const char *filename
)
3251 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3255 ret
= fstat(fd
, &st
);
3256 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3260 /* Attempt to detect via a CDROM specific ioctl */
3261 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3271 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3273 BDRVRawState
*s
= bs
->opaque
;
3276 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3277 return ret
== CDS_DISC_OK
;
3280 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3282 BDRVRawState
*s
= bs
->opaque
;
3285 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3286 perror("CDROMEJECT");
3288 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3289 perror("CDROMEJECT");
3293 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3295 BDRVRawState
*s
= bs
->opaque
;
3297 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3299 * Note: an error can happen if the distribution automatically
3302 /* perror("CDROM_LOCKDOOR"); */
3306 static BlockDriver bdrv_host_cdrom
= {
3307 .format_name
= "host_cdrom",
3308 .protocol_name
= "host_cdrom",
3309 .instance_size
= sizeof(BDRVRawState
),
3310 .bdrv_needs_filename
= true,
3311 .bdrv_probe_device
= cdrom_probe_device
,
3312 .bdrv_parse_filename
= cdrom_parse_filename
,
3313 .bdrv_file_open
= cdrom_open
,
3314 .bdrv_close
= raw_close
,
3315 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3316 .bdrv_reopen_commit
= raw_reopen_commit
,
3317 .bdrv_reopen_abort
= raw_reopen_abort
,
3318 .bdrv_co_create_opts
= hdev_co_create_opts
,
3319 .create_opts
= &raw_create_opts
,
3320 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3323 .bdrv_co_preadv
= raw_co_preadv
,
3324 .bdrv_co_pwritev
= raw_co_pwritev
,
3325 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3326 .bdrv_refresh_limits
= raw_refresh_limits
,
3327 .bdrv_io_plug
= raw_aio_plug
,
3328 .bdrv_io_unplug
= raw_aio_unplug
,
3329 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3331 .bdrv_co_truncate
= raw_co_truncate
,
3332 .bdrv_getlength
= raw_getlength
,
3333 .has_variable_length
= true,
3334 .bdrv_get_allocated_file_size
3335 = raw_get_allocated_file_size
,
3337 /* removable device support */
3338 .bdrv_is_inserted
= cdrom_is_inserted
,
3339 .bdrv_eject
= cdrom_eject
,
3340 .bdrv_lock_medium
= cdrom_lock_medium
,
3342 /* generic scsi device */
3343 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3345 #endif /* __linux__ */
3347 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3348 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3351 BDRVRawState
*s
= bs
->opaque
;
3352 Error
*local_err
= NULL
;
3357 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3359 error_propagate(errp
, local_err
);
3363 /* make sure the door isn't locked at this time */
3364 ioctl(s
->fd
, CDIOCALLOW
);
3368 static int cdrom_probe_device(const char *filename
)
3370 if (strstart(filename
, "/dev/cd", NULL
) ||
3371 strstart(filename
, "/dev/acd", NULL
))
3376 static int cdrom_reopen(BlockDriverState
*bs
)
3378 BDRVRawState
*s
= bs
->opaque
;
3382 * Force reread of possibly changed/newly loaded disc,
3383 * FreeBSD seems to not notice sometimes...
3387 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3394 /* make sure the door isn't locked at this time */
3395 ioctl(s
->fd
, CDIOCALLOW
);
3399 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3401 return raw_getlength(bs
) > 0;
3404 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3406 BDRVRawState
*s
= bs
->opaque
;
3411 (void) ioctl(s
->fd
, CDIOCALLOW
);
3414 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3415 perror("CDIOCEJECT");
3417 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3418 perror("CDIOCCLOSE");
3424 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3426 BDRVRawState
*s
= bs
->opaque
;
3430 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3432 * Note: an error can happen if the distribution automatically
3435 /* perror("CDROM_LOCKDOOR"); */
3439 static BlockDriver bdrv_host_cdrom
= {
3440 .format_name
= "host_cdrom",
3441 .protocol_name
= "host_cdrom",
3442 .instance_size
= sizeof(BDRVRawState
),
3443 .bdrv_needs_filename
= true,
3444 .bdrv_probe_device
= cdrom_probe_device
,
3445 .bdrv_parse_filename
= cdrom_parse_filename
,
3446 .bdrv_file_open
= cdrom_open
,
3447 .bdrv_close
= raw_close
,
3448 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3449 .bdrv_reopen_commit
= raw_reopen_commit
,
3450 .bdrv_reopen_abort
= raw_reopen_abort
,
3451 .bdrv_co_create_opts
= hdev_co_create_opts
,
3452 .create_opts
= &raw_create_opts
,
3454 .bdrv_co_preadv
= raw_co_preadv
,
3455 .bdrv_co_pwritev
= raw_co_pwritev
,
3456 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3457 .bdrv_refresh_limits
= raw_refresh_limits
,
3458 .bdrv_io_plug
= raw_aio_plug
,
3459 .bdrv_io_unplug
= raw_aio_unplug
,
3460 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3462 .bdrv_co_truncate
= raw_co_truncate
,
3463 .bdrv_getlength
= raw_getlength
,
3464 .has_variable_length
= true,
3465 .bdrv_get_allocated_file_size
3466 = raw_get_allocated_file_size
,
3468 /* removable device support */
3469 .bdrv_is_inserted
= cdrom_is_inserted
,
3470 .bdrv_eject
= cdrom_eject
,
3471 .bdrv_lock_medium
= cdrom_lock_medium
,
3473 #endif /* __FreeBSD__ */
3475 static void bdrv_file_init(void)
3478 * Register all the drivers. Note that order is important, the driver
3479 * registered last will get probed first.
3481 bdrv_register(&bdrv_file
);
3482 bdrv_register(&bdrv_host_device
);
3484 bdrv_register(&bdrv_host_cdrom
);
3486 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3487 bdrv_register(&bdrv_host_cdrom
);
3491 block_init(bdrv_file_init
);