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
;
155 /* The perms bits whose corresponding bytes are already locked in
157 uint64_t locked_perm
;
158 uint64_t locked_shared_perm
;
164 bool has_write_zeroes
:1;
165 bool discard_zeroes
:1;
166 bool use_linux_aio
:1;
167 bool page_cache_inconsistent
:1;
169 bool needs_alignment
;
170 bool check_cache_dropped
;
175 typedef struct BDRVRawReopenState
{
178 bool check_cache_dropped
;
179 } BDRVRawReopenState
;
181 static int fd_open(BlockDriverState
*bs
);
182 static int64_t raw_getlength(BlockDriverState
*bs
);
184 typedef struct RawPosixAIOData
{
185 BlockDriverState
*bs
;
188 struct iovec
*aio_iov
;
193 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
202 PreallocMode prealloc
;
208 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
209 static int cdrom_reopen(BlockDriverState
*bs
);
212 #if defined(__NetBSD__)
213 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
215 static char namebuf
[PATH_MAX
];
216 const char *dp
, *fname
;
220 dp
= strrchr(fname
, '/');
221 if (lstat(fname
, &sb
) < 0) {
222 error_setg_errno(errp
, errno
, "%s: stat failed", fname
);
226 if (!S_ISBLK(sb
.st_mode
)) {
231 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
233 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
234 (int)(dp
- fname
), fname
, dp
+ 1);
237 warn_report("%s is a block device, using %s", fname
, *filename
);
242 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
249 * Get logical block size via ioctl. On success store it in @sector_size_p.
251 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
253 unsigned int sector_size
;
254 bool success
= false;
258 static const unsigned long ioctl_list
[] = {
262 #ifdef DKIOCGETBLOCKSIZE
265 #ifdef DIOCGSECTORSIZE
270 /* Try a few ioctls to get the right size */
271 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
272 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
273 *sector_size_p
= sector_size
;
278 return success
? 0 : -errno
;
282 * Get physical block size of @fd.
283 * On success, store it in @blk_size and return 0.
284 * On failure, return -errno.
286 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
289 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
298 /* Check if read is allowed with given memory buffer and length.
300 * This function is used to check O_DIRECT memory buffer and request alignment.
302 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
304 ssize_t ret
= pread(fd
, buf
, len
, 0);
311 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
312 * other errors (e.g. real I/O error), which could happen on a failed
313 * drive, since we only care about probing alignment.
315 if (errno
!= EINVAL
) {
323 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
325 BDRVRawState
*s
= bs
->opaque
;
327 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
329 /* For SCSI generic devices the alignment is not really used.
330 With buffered I/O, we don't have any restrictions. */
331 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
332 bs
->bl
.request_alignment
= 1;
337 bs
->bl
.request_alignment
= 0;
339 /* Let's try to use the logical blocksize for the alignment. */
340 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
341 bs
->bl
.request_alignment
= 0;
346 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
347 bs
->bl
.request_alignment
= da
.d_miniosz
;
348 /* The kernel returns wrong information for d_mem */
349 /* s->buf_align = da.d_mem; */
354 /* If we could not get the sizes so far, we can only guess them */
357 buf
= qemu_memalign(max_align
, 2 * max_align
);
358 for (align
= 512; align
<= max_align
; align
<<= 1) {
359 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
360 s
->buf_align
= align
;
367 if (!bs
->bl
.request_alignment
) {
369 buf
= qemu_memalign(s
->buf_align
, max_align
);
370 for (align
= 512; align
<= max_align
; align
<<= 1) {
371 if (raw_is_io_aligned(fd
, buf
, align
)) {
372 bs
->bl
.request_alignment
= align
;
379 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
380 error_setg(errp
, "Could not find working O_DIRECT alignment");
381 error_append_hint(errp
, "Try cache.direct=off\n");
385 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
387 assert(open_flags
!= NULL
);
389 *open_flags
|= O_BINARY
;
390 *open_flags
&= ~O_ACCMODE
;
391 if (bdrv_flags
& BDRV_O_RDWR
) {
392 *open_flags
|= O_RDWR
;
394 *open_flags
|= O_RDONLY
;
397 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
398 * and O_DIRECT for no caching. */
399 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
400 *open_flags
|= O_DIRECT
;
404 static void raw_parse_filename(const char *filename
, QDict
*options
,
407 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
410 static QemuOptsList raw_runtime_opts
= {
412 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
416 .type
= QEMU_OPT_STRING
,
417 .help
= "File name of the image",
421 .type
= QEMU_OPT_STRING
,
422 .help
= "host AIO implementation (threads, native)",
426 .type
= QEMU_OPT_STRING
,
427 .help
= "file locking mode (on/off/auto, default: auto)",
430 .name
= "pr-manager",
431 .type
= QEMU_OPT_STRING
,
432 .help
= "id of persistent reservation manager object (default: none)",
435 .name
= "x-check-cache-dropped",
436 .type
= QEMU_OPT_BOOL
,
437 .help
= "check that page cache was dropped on live migration (default: off)"
439 { /* end of list */ }
443 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
444 int bdrv_flags
, int open_flags
,
445 bool device
, Error
**errp
)
447 BDRVRawState
*s
= bs
->opaque
;
449 Error
*local_err
= NULL
;
450 const char *filename
= NULL
;
452 BlockdevAioOptions aio
, aio_default
;
457 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
458 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
460 error_propagate(errp
, local_err
);
465 filename
= qemu_opt_get(opts
, "filename");
467 ret
= raw_normalize_devicepath(&filename
, errp
);
472 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
473 ? BLOCKDEV_AIO_OPTIONS_NATIVE
474 : BLOCKDEV_AIO_OPTIONS_THREADS
;
475 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
476 qemu_opt_get(opts
, "aio"),
477 aio_default
, &local_err
);
479 error_propagate(errp
, local_err
);
483 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
485 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
486 qemu_opt_get(opts
, "locking"),
487 ON_OFF_AUTO_AUTO
, &local_err
);
489 error_propagate(errp
, local_err
);
496 if (!qemu_has_ofd_lock()) {
497 warn_report("File lock requested but OFD locking syscall is "
498 "unavailable, falling back to POSIX file locks");
499 error_printf("Due to the implementation, locks can be lost "
503 case ON_OFF_AUTO_OFF
:
506 case ON_OFF_AUTO_AUTO
:
507 s
->use_lock
= qemu_has_ofd_lock();
513 str
= qemu_opt_get(opts
, "pr-manager");
515 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
517 error_propagate(errp
, local_err
);
523 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
526 s
->open_flags
= open_flags
;
527 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
530 fd
= qemu_open(filename
, s
->open_flags
, 0644);
531 ret
= fd
< 0 ? -errno
: 0;
533 if (ret
== -EACCES
|| ret
== -EROFS
) {
534 /* Try to degrade to read-only, but if it doesn't work, still use the
535 * normal error message. */
536 if (bdrv_apply_auto_read_only(bs
, NULL
, NULL
) == 0) {
537 bdrv_flags
&= ~BDRV_O_RDWR
;
538 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
539 assert(!(s
->open_flags
& O_CREAT
));
540 fd
= qemu_open(filename
, s
->open_flags
);
541 ret
= fd
< 0 ? -errno
: 0;
546 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
556 fd
= qemu_open(filename
, s
->open_flags
);
559 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
567 s
->shared_perm
= BLK_PERM_ALL
;
569 #ifdef CONFIG_LINUX_AIO
570 /* Currently Linux does AIO only for files opened with O_DIRECT */
571 if (s
->use_linux_aio
) {
572 if (!(s
->open_flags
& O_DIRECT
)) {
573 error_setg(errp
, "aio=native was specified, but it requires "
574 "cache.direct=on, which was not specified.");
578 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
579 error_prepend(errp
, "Unable to use native AIO: ");
584 if (s
->use_linux_aio
) {
585 error_setg(errp
, "aio=native was specified, but is not supported "
590 #endif /* !defined(CONFIG_LINUX_AIO) */
592 s
->has_discard
= true;
593 s
->has_write_zeroes
= true;
594 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
595 s
->needs_alignment
= true;
598 if (fstat(s
->fd
, &st
) < 0) {
600 error_setg_errno(errp
, errno
, "Could not stat file");
605 if (S_ISBLK(st
.st_mode
)) {
606 warn_report("Opening a block device as a file using the '%s' "
607 "driver is deprecated", bs
->drv
->format_name
);
608 } else if (S_ISCHR(st
.st_mode
)) {
609 warn_report("Opening a character device as a file using the '%s' "
610 "driver is deprecated", bs
->drv
->format_name
);
611 } else if (!S_ISREG(st
.st_mode
)) {
612 error_setg(errp
, "A regular file was expected by the '%s' driver, "
613 "but something else was given", bs
->drv
->format_name
);
617 s
->discard_zeroes
= true;
618 s
->has_fallocate
= true;
621 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
622 error_setg(errp
, "'%s' driver expects either "
623 "a character or block device", bs
->drv
->format_name
);
629 if (S_ISBLK(st
.st_mode
)) {
630 #ifdef BLKDISCARDZEROES
632 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
633 s
->discard_zeroes
= true;
637 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
638 * not rely on the contents of discarded blocks unless using O_DIRECT.
639 * Same for BLKZEROOUT.
641 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
642 s
->discard_zeroes
= false;
643 s
->has_write_zeroes
= false;
648 if (S_ISCHR(st
.st_mode
)) {
650 * The file is a char device (disk), which on FreeBSD isn't behind
651 * a pager, so force all requests to be aligned. This is needed
652 * so QEMU makes sure all IO operations on the device are aligned
653 * to sector size, or else FreeBSD will reject them with EINVAL.
655 s
->needs_alignment
= true;
660 if (platform_test_xfs_fd(s
->fd
)) {
665 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
668 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
675 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
678 BDRVRawState
*s
= bs
->opaque
;
680 s
->type
= FTYPE_FILE
;
681 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
690 #define PERM_FOREACH(i) \
691 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
693 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
694 * file; if @unlock == true, also unlock the unneeded bytes.
695 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
697 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
698 uint64_t perm_lock_bits
,
699 uint64_t shared_perm_lock_bits
,
700 bool unlock
, Error
**errp
)
704 uint64_t locked_perm
, locked_shared_perm
;
707 locked_perm
= s
->locked_perm
;
708 locked_shared_perm
= s
->locked_shared_perm
;
711 * We don't have the previous bits, just lock/unlock for each of the
715 locked_perm
= BLK_PERM_ALL
;
716 locked_shared_perm
= BLK_PERM_ALL
;
719 locked_shared_perm
= 0;
724 int off
= RAW_LOCK_PERM_BASE
+ i
;
725 uint64_t bit
= (1ULL << i
);
726 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
727 ret
= qemu_lock_fd(fd
, off
, 1, false);
729 error_setg(errp
, "Failed to lock byte %d", off
);
732 s
->locked_perm
|= bit
;
734 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
735 ret
= qemu_unlock_fd(fd
, off
, 1);
737 error_setg(errp
, "Failed to unlock byte %d", off
);
740 s
->locked_perm
&= ~bit
;
745 int off
= RAW_LOCK_SHARED_BASE
+ i
;
746 uint64_t bit
= (1ULL << i
);
747 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
748 ret
= qemu_lock_fd(fd
, off
, 1, false);
750 error_setg(errp
, "Failed to lock byte %d", off
);
753 s
->locked_shared_perm
|= bit
;
755 } else if (unlock
&& (locked_shared_perm
& bit
) &&
756 !(shared_perm_lock_bits
& bit
)) {
757 ret
= qemu_unlock_fd(fd
, off
, 1);
759 error_setg(errp
, "Failed to unlock byte %d", off
);
762 s
->locked_shared_perm
&= ~bit
;
769 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
770 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
777 int off
= RAW_LOCK_SHARED_BASE
+ i
;
778 uint64_t p
= 1ULL << i
;
780 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
782 char *perm_name
= bdrv_perm_names(p
);
784 "Failed to get \"%s\" lock",
792 int off
= RAW_LOCK_PERM_BASE
+ i
;
793 uint64_t p
= 1ULL << i
;
794 if (!(shared_perm
& p
)) {
795 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
797 char *perm_name
= bdrv_perm_names(p
);
799 "Failed to get shared \"%s\" lock",
809 static int raw_handle_perm_lock(BlockDriverState
*bs
,
811 uint64_t new_perm
, uint64_t new_shared
,
814 BDRVRawState
*s
= bs
->opaque
;
816 Error
*local_err
= NULL
;
822 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
826 assert(s
->lock_fd
> 0);
830 ret
= raw_apply_lock_bytes(s
, s
->lock_fd
, s
->perm
| new_perm
,
831 ~s
->shared_perm
| ~new_shared
,
834 ret
= raw_check_lock_bytes(s
->lock_fd
, new_perm
, new_shared
, errp
);
838 error_append_hint(errp
,
839 "Is another process using the image [%s]?\n",
843 /* fall through to unlock bytes. */
845 raw_apply_lock_bytes(s
, s
->lock_fd
, s
->perm
, ~s
->shared_perm
,
848 /* Theoretically the above call only unlocks bytes and it cannot
849 * fail. Something weird happened, report it.
851 warn_report_err(local_err
);
855 raw_apply_lock_bytes(s
, s
->lock_fd
, new_perm
, ~new_shared
,
858 /* Theoretically the above call only unlocks bytes and it cannot
859 * fail. Something weird happened, report it.
861 warn_report_err(local_err
);
868 static int raw_reopen_prepare(BDRVReopenState
*state
,
869 BlockReopenQueue
*queue
, Error
**errp
)
872 BDRVRawReopenState
*rs
;
875 Error
*local_err
= NULL
;
877 assert(state
!= NULL
);
878 assert(state
->bs
!= NULL
);
880 s
= state
->bs
->opaque
;
882 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
886 /* Handle options changes */
887 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
888 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
890 error_propagate(errp
, local_err
);
895 rs
->check_cache_dropped
=
896 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
898 /* This driver's reopen function doesn't currently allow changing
899 * other options, so let's put them back in the original QDict and
900 * bdrv_reopen_prepare() will detect changes and complain. */
901 qemu_opts_to_qdict(opts
, state
->options
);
903 if (s
->type
== FTYPE_CD
) {
904 rs
->open_flags
|= O_NONBLOCK
;
907 raw_parse_flags(state
->flags
, &rs
->open_flags
);
909 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
911 fcntl_flags
|= O_NOATIME
;
915 /* Not all operating systems have O_ASYNC, and those that don't
916 * will not let us track the state into rs->open_flags (typically
917 * you achieve the same effect with an ioctl, for example I_SETSIG
918 * on Solaris). But we do not use O_ASYNC, so that's fine.
920 assert((s
->open_flags
& O_ASYNC
) == 0);
923 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
924 /* dup the original fd */
925 rs
->fd
= qemu_dup(s
->fd
);
927 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
935 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
937 const char *normalized_filename
= state
->bs
->filename
;
938 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
940 assert(!(rs
->open_flags
& O_CREAT
));
941 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
943 error_setg_errno(errp
, errno
, "Could not reopen file");
949 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
950 * alignment with the new fd. */
952 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
956 error_propagate(errp
, local_err
);
966 static void raw_reopen_commit(BDRVReopenState
*state
)
968 BDRVRawReopenState
*rs
= state
->opaque
;
969 BDRVRawState
*s
= state
->bs
->opaque
;
971 s
->check_cache_dropped
= rs
->check_cache_dropped
;
972 s
->open_flags
= rs
->open_flags
;
977 g_free(state
->opaque
);
978 state
->opaque
= NULL
;
982 static void raw_reopen_abort(BDRVReopenState
*state
)
984 BDRVRawReopenState
*rs
= state
->opaque
;
986 /* nothing to do if NULL, we didn't get far enough */
995 g_free(state
->opaque
);
996 state
->opaque
= NULL
;
999 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
1003 short max_sectors
= 0;
1004 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1006 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1007 return max_sectors
<< BDRV_SECTOR_BITS
;
1016 static int hdev_get_max_segments(const struct stat
*st
)
1026 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1027 major(st
->st_rdev
), minor(st
->st_rdev
));
1028 fd
= open(sysfspath
, O_RDONLY
);
1034 ret
= read(fd
, buf
, sizeof(buf
) - 1);
1035 } while (ret
== -1 && errno
== EINTR
);
1039 } else if (ret
== 0) {
1044 /* The file is ended with '\n', pass 'end' to accept that. */
1045 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1046 if (ret
== 0 && end
&& *end
== '\n') {
1061 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1063 BDRVRawState
*s
= bs
->opaque
;
1066 if (!fstat(s
->fd
, &st
)) {
1067 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1068 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1069 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1070 bs
->bl
.max_transfer
= pow2floor(ret
);
1072 ret
= hdev_get_max_segments(&st
);
1074 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1075 ret
* getpagesize());
1080 raw_probe_alignment(bs
, s
->fd
, errp
);
1081 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1082 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1085 static int check_for_dasd(int fd
)
1088 struct dasd_information2_t info
= {0};
1090 return ioctl(fd
, BIODASDINFO2
, &info
);
1097 * Try to get @bs's logical and physical block size.
1098 * On success, store them in @bsz and return zero.
1099 * On failure, return negative errno.
1101 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1103 BDRVRawState
*s
= bs
->opaque
;
1106 /* If DASD, get blocksizes */
1107 if (check_for_dasd(s
->fd
) < 0) {
1110 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1114 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1118 * Try to get @bs's geometry: cyls, heads, sectors.
1119 * On success, store them in @geo and return 0.
1120 * On failure return -errno.
1121 * (Allows block driver to assign default geometry values that guest sees)
1124 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1126 BDRVRawState
*s
= bs
->opaque
;
1127 struct hd_geometry ioctl_geo
= {0};
1129 /* If DASD, get its geometry */
1130 if (check_for_dasd(s
->fd
) < 0) {
1133 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1136 /* HDIO_GETGEO may return success even though geo contains zeros
1137 (e.g. certain multipath setups) */
1138 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1141 /* Do not return a geometry for partition */
1142 if (ioctl_geo
.start
!= 0) {
1145 geo
->heads
= ioctl_geo
.heads
;
1146 geo
->sectors
= ioctl_geo
.sectors
;
1147 geo
->cylinders
= ioctl_geo
.cylinders
;
1151 #else /* __linux__ */
1152 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1158 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1162 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1170 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1172 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1175 if (s
->page_cache_inconsistent
) {
1179 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1181 /* There is no clear definition of the semantics of a failing fsync(),
1182 * so we may have to assume the worst. The sad truth is that this
1183 * assumption is correct for Linux. Some pages are now probably marked
1184 * clean in the page cache even though they are inconsistent with the
1185 * on-disk contents. The next fdatasync() call would succeed, but no
1186 * further writeback attempt will be made. We can't get back to a state
1187 * in which we know what is on disk (we would have to rewrite
1188 * everything that was touched since the last fdatasync() at least), so
1189 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1190 * really defined, I have little hope that other OSes are doing better.
1192 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1194 if ((s
->open_flags
& O_DIRECT
) == 0) {
1195 s
->page_cache_inconsistent
= true;
1202 #ifdef CONFIG_PREADV
1204 static bool preadv_present
= true;
1207 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1209 return preadv(fd
, iov
, nr_iov
, offset
);
1213 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1215 return pwritev(fd
, iov
, nr_iov
, offset
);
1220 static bool preadv_present
= false;
1223 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1229 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1236 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1241 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1242 len
= qemu_pwritev(aiocb
->aio_fildes
,
1247 len
= qemu_preadv(aiocb
->aio_fildes
,
1251 } while (len
== -1 && errno
== EINTR
);
1260 * Read/writes the data to/from a given linear buffer.
1262 * Returns the number of bytes handles or -errno in case of an error. Short
1263 * reads are only returned if the end of the file is reached.
1265 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1270 while (offset
< aiocb
->aio_nbytes
) {
1271 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1272 len
= pwrite(aiocb
->aio_fildes
,
1273 (const char *)buf
+ offset
,
1274 aiocb
->aio_nbytes
- offset
,
1275 aiocb
->aio_offset
+ offset
);
1277 len
= pread(aiocb
->aio_fildes
,
1279 aiocb
->aio_nbytes
- offset
,
1280 aiocb
->aio_offset
+ offset
);
1282 if (len
== -1 && errno
== EINTR
) {
1284 } else if (len
== -1 && errno
== EINVAL
&&
1285 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1286 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1288 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1289 * after a short read. Assume that O_DIRECT short reads only occur
1290 * at EOF. Therefore this is a short read, not an I/O error.
1293 } else if (len
== -1) {
1296 } else if (len
== 0) {
1305 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1310 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1312 * If there is just a single buffer, and it is properly aligned
1313 * we can just use plain pread/pwrite without any problems.
1315 if (aiocb
->aio_niov
== 1) {
1316 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1319 * We have more than one iovec, and all are properly aligned.
1321 * Try preadv/pwritev first and fall back to linearizing the
1322 * buffer if it's not supported.
1324 if (preadv_present
) {
1325 nbytes
= handle_aiocb_rw_vector(aiocb
);
1326 if (nbytes
== aiocb
->aio_nbytes
||
1327 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1330 preadv_present
= false;
1334 * XXX(hch): short read/write. no easy way to handle the reminder
1335 * using these interfaces. For now retry using plain
1341 * Ok, we have to do it the hard way, copy all segments into
1342 * a single aligned buffer.
1344 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1349 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1353 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1354 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1355 p
+= aiocb
->aio_iov
[i
].iov_len
;
1357 assert(p
- buf
== aiocb
->aio_nbytes
);
1360 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1361 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1363 size_t count
= aiocb
->aio_nbytes
, copy
;
1366 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1368 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1369 copy
= aiocb
->aio_iov
[i
].iov_len
;
1371 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1372 assert(count
>= copy
);
1384 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1386 struct xfs_flock64 fl
;
1389 memset(&fl
, 0, sizeof(fl
));
1390 fl
.l_whence
= SEEK_SET
;
1391 fl
.l_start
= offset
;
1394 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1396 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1403 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1405 struct xfs_flock64 fl
;
1408 memset(&fl
, 0, sizeof(fl
));
1409 fl
.l_whence
= SEEK_SET
;
1410 fl
.l_start
= offset
;
1413 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1415 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1423 static int translate_err(int err
)
1425 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1432 #ifdef CONFIG_FALLOCATE
1433 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1436 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1439 } while (errno
== EINTR
);
1440 return translate_err(-errno
);
1444 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1447 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1449 if (!s
->has_write_zeroes
) {
1455 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1456 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1459 } while (errno
== EINTR
);
1461 ret
= translate_err(-errno
);
1464 if (ret
== -ENOTSUP
) {
1465 s
->has_write_zeroes
= false;
1470 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1472 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1473 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1475 #ifdef CONFIG_FALLOCATE
1479 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1480 return handle_aiocb_write_zeroes_block(aiocb
);
1485 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1489 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1490 if (s
->has_write_zeroes
) {
1491 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1492 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1493 if (ret
== 0 || ret
!= -ENOTSUP
) {
1496 s
->has_write_zeroes
= false;
1500 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1501 if (s
->has_discard
&& s
->has_fallocate
) {
1502 int ret
= do_fallocate(s
->fd
,
1503 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1504 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1506 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1507 if (ret
== 0 || ret
!= -ENOTSUP
) {
1510 s
->has_fallocate
= false;
1511 } else if (ret
!= -ENOTSUP
) {
1514 s
->has_discard
= false;
1519 #ifdef CONFIG_FALLOCATE
1520 /* Last resort: we are trying to extend the file with zeroed data. This
1521 * can be done via fallocate(fd, 0) */
1522 len
= bdrv_getlength(aiocb
->bs
);
1523 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1524 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1525 if (ret
== 0 || ret
!= -ENOTSUP
) {
1528 s
->has_fallocate
= false;
1535 static ssize_t
handle_aiocb_write_zeroes_unmap(RawPosixAIOData
*aiocb
)
1537 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1540 /* First try to write zeros and unmap at the same time */
1542 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1543 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1544 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1545 if (ret
!= -ENOTSUP
) {
1552 /* xfs_discard() guarantees that the discarded area reads as all-zero
1553 * afterwards, so we can use it here. */
1554 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1558 /* If we couldn't manage to unmap while guaranteed that the area reads as
1559 * all-zero afterwards, just write zeroes without unmapping */
1560 ret
= handle_aiocb_write_zeroes(aiocb
);
1564 #ifndef HAVE_COPY_FILE_RANGE
1565 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1566 off_t
*out_off
, size_t len
, unsigned int flags
)
1568 #ifdef __NR_copy_file_range
1569 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1570 out_off
, len
, flags
);
1578 static ssize_t
handle_aiocb_copy_range(RawPosixAIOData
*aiocb
)
1580 uint64_t bytes
= aiocb
->aio_nbytes
;
1581 off_t in_off
= aiocb
->aio_offset
;
1582 off_t out_off
= aiocb
->aio_offset2
;
1585 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1586 aiocb
->aio_fd2
, &out_off
,
1588 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1589 aiocb
->aio_fd2
, out_off
, bytes
, 0, ret
);
1591 /* No progress (e.g. when beyond EOF), let the caller fall back to
1610 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1612 int ret
= -EOPNOTSUPP
;
1613 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1615 if (!s
->has_discard
) {
1619 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1622 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1623 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1626 } while (errno
== EINTR
);
1633 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1637 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1638 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1639 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1643 ret
= translate_err(ret
);
1644 if (ret
== -ENOTSUP
) {
1645 s
->has_discard
= false;
1650 static int handle_aiocb_truncate(RawPosixAIOData
*aiocb
)
1653 int64_t current_length
= 0;
1656 int fd
= aiocb
->aio_fildes
;
1657 int64_t offset
= aiocb
->aio_offset
;
1658 Error
**errp
= aiocb
->errp
;
1660 if (fstat(fd
, &st
) < 0) {
1662 error_setg_errno(errp
, -result
, "Could not stat file");
1666 current_length
= st
.st_size
;
1667 if (current_length
> offset
&& aiocb
->prealloc
!= PREALLOC_MODE_OFF
) {
1668 error_setg(errp
, "Cannot use preallocation for shrinking files");
1672 switch (aiocb
->prealloc
) {
1673 #ifdef CONFIG_POSIX_FALLOCATE
1674 case PREALLOC_MODE_FALLOC
:
1676 * Truncating before posix_fallocate() makes it about twice slower on
1677 * file systems that do not support fallocate(), trying to check if a
1678 * block is allocated before allocating it, so don't do that here.
1680 if (offset
!= current_length
) {
1681 result
= -posix_fallocate(fd
, current_length
,
1682 offset
- current_length
);
1684 /* posix_fallocate() doesn't set errno. */
1685 error_setg_errno(errp
, -result
,
1686 "Could not preallocate new data");
1693 case PREALLOC_MODE_FULL
:
1695 int64_t num
= 0, left
= offset
- current_length
;
1699 * Knowing the final size from the beginning could allow the file
1700 * system driver to do less allocations and possibly avoid
1701 * fragmentation of the file.
1703 if (ftruncate(fd
, offset
) != 0) {
1705 error_setg_errno(errp
, -result
, "Could not resize file");
1709 buf
= g_malloc0(65536);
1711 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1712 if (seek_result
< 0) {
1714 error_setg_errno(errp
, -result
,
1715 "Failed to seek to the old end of file");
1720 num
= MIN(left
, 65536);
1721 result
= write(fd
, buf
, num
);
1723 if (errno
== EINTR
) {
1727 error_setg_errno(errp
, -result
,
1728 "Could not write zeros for preallocation");
1737 error_setg_errno(errp
, -result
,
1738 "Could not flush file to disk");
1744 case PREALLOC_MODE_OFF
:
1745 if (ftruncate(fd
, offset
) != 0) {
1747 error_setg_errno(errp
, -result
, "Could not resize file");
1752 error_setg(errp
, "Unsupported preallocation mode: %s",
1753 PreallocMode_str(aiocb
->prealloc
));
1759 if (ftruncate(fd
, current_length
) < 0) {
1760 error_report("Failed to restore old file length: %s",
1769 static int aio_worker(void *arg
)
1771 RawPosixAIOData
*aiocb
= arg
;
1774 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1776 ret
= handle_aiocb_rw(aiocb
);
1777 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1778 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1779 0, aiocb
->aio_nbytes
- ret
);
1781 ret
= aiocb
->aio_nbytes
;
1783 if (ret
== aiocb
->aio_nbytes
) {
1785 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1789 case QEMU_AIO_WRITE
:
1790 ret
= handle_aiocb_rw(aiocb
);
1791 if (ret
== aiocb
->aio_nbytes
) {
1793 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1797 case QEMU_AIO_FLUSH
:
1798 ret
= handle_aiocb_flush(aiocb
);
1800 case QEMU_AIO_IOCTL
:
1801 ret
= handle_aiocb_ioctl(aiocb
);
1803 case QEMU_AIO_DISCARD
:
1804 ret
= handle_aiocb_discard(aiocb
);
1806 case QEMU_AIO_WRITE_ZEROES
:
1807 ret
= handle_aiocb_write_zeroes(aiocb
);
1809 case QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_DISCARD
:
1810 ret
= handle_aiocb_write_zeroes_unmap(aiocb
);
1812 case QEMU_AIO_COPY_RANGE
:
1813 ret
= handle_aiocb_copy_range(aiocb
);
1815 case QEMU_AIO_TRUNCATE
:
1816 ret
= handle_aiocb_truncate(aiocb
);
1819 error_report("invalid aio request (0x%x)", aiocb
->aio_type
);
1828 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1829 int64_t offset
, int fd2
, int64_t offset2
,
1831 int bytes
, int type
)
1833 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1837 acb
->aio_type
= type
;
1838 acb
->aio_fildes
= fd
;
1840 acb
->aio_offset2
= offset2
;
1842 acb
->aio_nbytes
= bytes
;
1843 acb
->aio_offset
= offset
;
1846 acb
->aio_iov
= qiov
->iov
;
1847 acb
->aio_niov
= qiov
->niov
;
1848 assert(qiov
->size
== bytes
);
1851 trace_file_paio_submit_co(offset
, bytes
, type
);
1852 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1853 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1856 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1857 int64_t offset
, QEMUIOVector
*qiov
,
1858 int bytes
, int type
)
1860 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1863 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1864 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1866 BDRVRawState
*s
= bs
->opaque
;
1868 if (fd_open(bs
) < 0)
1872 * Check if the underlying device requires requests to be aligned,
1873 * and if the request we are trying to submit is aligned or not.
1874 * If this is the case tell the low-level driver that it needs
1875 * to copy the buffer.
1877 if (s
->needs_alignment
) {
1878 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1879 type
|= QEMU_AIO_MISALIGNED
;
1880 #ifdef CONFIG_LINUX_AIO
1881 } else if (s
->use_linux_aio
) {
1882 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1883 assert(qiov
->size
== bytes
);
1884 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1889 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1892 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1893 uint64_t bytes
, QEMUIOVector
*qiov
,
1896 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1899 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1900 uint64_t bytes
, QEMUIOVector
*qiov
,
1904 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1907 static void raw_aio_plug(BlockDriverState
*bs
)
1909 #ifdef CONFIG_LINUX_AIO
1910 BDRVRawState
*s
= bs
->opaque
;
1911 if (s
->use_linux_aio
) {
1912 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1913 laio_io_plug(bs
, aio
);
1918 static void raw_aio_unplug(BlockDriverState
*bs
)
1920 #ifdef CONFIG_LINUX_AIO
1921 BDRVRawState
*s
= bs
->opaque
;
1922 if (s
->use_linux_aio
) {
1923 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1924 laio_io_unplug(bs
, aio
);
1929 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1931 BDRVRawState
*s
= bs
->opaque
;
1939 return paio_submit_co(bs
, s
->fd
, 0, NULL
, 0, QEMU_AIO_FLUSH
);
1942 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1943 AioContext
*new_context
)
1945 #ifdef CONFIG_LINUX_AIO
1946 BDRVRawState
*s
= bs
->opaque
;
1947 if (s
->use_linux_aio
) {
1949 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1950 error_reportf_err(local_err
, "Unable to use native AIO, "
1951 "falling back to thread pool: ");
1952 s
->use_linux_aio
= false;
1958 static void raw_close(BlockDriverState
*bs
)
1960 BDRVRawState
*s
= bs
->opaque
;
1966 if (s
->lock_fd
>= 0) {
1967 qemu_close(s
->lock_fd
);
1973 * Truncates the given regular file @fd to @offset and, when growing, fills the
1974 * new space according to @prealloc.
1976 * Returns: 0 on success, -errno on failure.
1978 static int coroutine_fn
1979 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1980 PreallocMode prealloc
, Error
**errp
)
1982 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1985 *acb
= (RawPosixAIOData
) {
1988 .aio_type
= QEMU_AIO_TRUNCATE
,
1989 .aio_offset
= offset
,
1990 .prealloc
= prealloc
,
1994 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1995 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1996 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1999 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2000 PreallocMode prealloc
, Error
**errp
)
2002 BDRVRawState
*s
= bs
->opaque
;
2006 if (fstat(s
->fd
, &st
)) {
2008 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2012 if (S_ISREG(st
.st_mode
)) {
2013 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2016 if (prealloc
!= PREALLOC_MODE_OFF
) {
2017 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2018 "non-regular file", PreallocMode_str(prealloc
));
2022 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2023 if (offset
> raw_getlength(bs
)) {
2024 error_setg(errp
, "Cannot grow device files");
2028 error_setg(errp
, "Resizing this file is not supported");
2036 static int64_t raw_getlength(BlockDriverState
*bs
)
2038 BDRVRawState
*s
= bs
->opaque
;
2044 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2045 struct disklabel dl
;
2047 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2049 return (uint64_t)dl
.d_secsize
*
2050 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2054 #elif defined(__NetBSD__)
2055 static int64_t raw_getlength(BlockDriverState
*bs
)
2057 BDRVRawState
*s
= bs
->opaque
;
2063 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2064 struct dkwedge_info dkw
;
2066 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2067 return dkw
.dkw_size
* 512;
2069 struct disklabel dl
;
2071 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2073 return (uint64_t)dl
.d_secsize
*
2074 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2079 #elif defined(__sun__)
2080 static int64_t raw_getlength(BlockDriverState
*bs
)
2082 BDRVRawState
*s
= bs
->opaque
;
2083 struct dk_minfo minfo
;
2093 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2095 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2097 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2101 * There are reports that lseek on some devices fails, but
2102 * irc discussion said that contingency on contingency was overkill.
2104 size
= lseek(s
->fd
, 0, SEEK_END
);
2110 #elif defined(CONFIG_BSD)
2111 static int64_t raw_getlength(BlockDriverState
*bs
)
2113 BDRVRawState
*s
= bs
->opaque
;
2117 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2126 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2129 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2130 #ifdef DIOCGMEDIASIZE
2131 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2132 #elif defined(DIOCGPART)
2135 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2136 size
= pi
.media_size
;
2142 #if defined(__APPLE__) && defined(__MACH__)
2144 uint64_t sectors
= 0;
2145 uint32_t sector_size
= 0;
2147 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2148 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2149 size
= sectors
* sector_size
;
2151 size
= lseek(fd
, 0LL, SEEK_END
);
2158 size
= lseek(fd
, 0LL, SEEK_END
);
2163 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2166 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2167 if (size
== 2048LL * (unsigned)-1)
2169 /* XXX no disc? maybe we need to reopen... */
2170 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2177 size
= lseek(fd
, 0, SEEK_END
);
2185 static int64_t raw_getlength(BlockDriverState
*bs
)
2187 BDRVRawState
*s
= bs
->opaque
;
2196 size
= lseek(s
->fd
, 0, SEEK_END
);
2204 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2207 BDRVRawState
*s
= bs
->opaque
;
2209 if (fstat(s
->fd
, &st
) < 0) {
2212 return (int64_t)st
.st_blocks
* 512;
2215 static int coroutine_fn
2216 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2218 BlockdevCreateOptionsFile
*file_opts
;
2219 Error
*local_err
= NULL
;
2221 uint64_t perm
, shared
;
2224 /* Validate options and set default values */
2225 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2226 file_opts
= &options
->u
.file
;
2228 if (!file_opts
->has_nocow
) {
2229 file_opts
->nocow
= false;
2231 if (!file_opts
->has_preallocation
) {
2232 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2236 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2239 error_setg_errno(errp
, -result
, "Could not create file");
2243 /* Take permissions: We want to discard everything, so we need
2244 * BLK_PERM_WRITE; and truncation to the desired size requires
2246 * On the other hand, we cannot share the RESIZE permission
2247 * because we promise that after this function, the file has the
2248 * size given in the options. If someone else were to resize it
2249 * concurrently, we could not guarantee that.
2250 * Note that after this function, we can no longer guarantee that
2251 * the file is not touched by a third party, so it may be resized
2253 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2254 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2256 /* Step one: Take locks */
2257 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2262 /* Step two: Check that nobody else has taken conflicting locks */
2263 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2265 error_append_hint(errp
,
2266 "Is another process using the image [%s]?\n",
2267 file_opts
->filename
);
2271 /* Clear the file by truncating it to 0 */
2272 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2277 if (file_opts
->nocow
) {
2279 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2280 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2281 * will be ignored since any failure of this operation should not
2282 * block the left work.
2285 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2286 attr
|= FS_NOCOW_FL
;
2287 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2292 /* Resize and potentially preallocate the file to the desired
2294 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2295 file_opts
->preallocation
, errp
);
2301 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2303 /* The above call should not fail, and if it does, that does
2304 * not mean the whole creation operation has failed. So
2305 * report it the user for their convenience, but do not report
2306 * it to the caller. */
2307 warn_report_err(local_err
);
2311 if (qemu_close(fd
) != 0 && result
== 0) {
2313 error_setg_errno(errp
, -result
, "Could not close the new file");
2319 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2322 BlockdevCreateOptions options
;
2323 int64_t total_size
= 0;
2325 PreallocMode prealloc
;
2327 Error
*local_err
= NULL
;
2329 /* Skip file: protocol prefix */
2330 strstart(filename
, "file:", &filename
);
2332 /* Read out options */
2333 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2335 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2336 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2337 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2338 PREALLOC_MODE_OFF
, &local_err
);
2341 error_propagate(errp
, local_err
);
2345 options
= (BlockdevCreateOptions
) {
2346 .driver
= BLOCKDEV_DRIVER_FILE
,
2348 .filename
= (char *) filename
,
2350 .has_preallocation
= true,
2351 .preallocation
= prealloc
,
2356 return raw_co_create(&options
, errp
);
2360 * Find allocation range in @bs around offset @start.
2361 * May change underlying file descriptor's file offset.
2362 * If @start is not in a hole, store @start in @data, and the
2363 * beginning of the next hole in @hole, and return 0.
2364 * If @start is in a non-trailing hole, store @start in @hole and the
2365 * beginning of the next non-hole in @data, and return 0.
2366 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2367 * If we can't find out, return a negative errno other than -ENXIO.
2369 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2370 off_t
*data
, off_t
*hole
)
2372 #if defined SEEK_HOLE && defined SEEK_DATA
2373 BDRVRawState
*s
= bs
->opaque
;
2378 * D1. offs == start: start is in data
2379 * D2. offs > start: start is in a hole, next data at offs
2380 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2381 * or start is beyond EOF
2382 * If the latter happens, the file has been truncated behind
2383 * our back since we opened it. All bets are off then.
2384 * Treating like a trailing hole is simplest.
2385 * D4. offs < 0, errno != ENXIO: we learned nothing
2387 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2389 return -errno
; /* D3 or D4 */
2393 /* This is not a valid return by lseek(). We are safe to just return
2394 * -EIO in this case, and we'll treat it like D4. */
2399 /* D2: in hole, next data at offs */
2405 /* D1: in data, end not yet known */
2409 * H1. offs == start: start is in a hole
2410 * If this happens here, a hole has been dug behind our back
2411 * since the previous lseek().
2412 * H2. offs > start: either start is in data, next hole at offs,
2413 * or start is in trailing hole, EOF at offs
2414 * Linux treats trailing holes like any other hole: offs ==
2415 * start. Solaris seeks to EOF instead: offs > start (blech).
2416 * If that happens here, a hole has been dug behind our back
2417 * since the previous lseek().
2418 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2419 * If this happens, the file has been truncated behind our
2420 * back since we opened it. Treat it like a trailing hole.
2421 * H4. offs < 0, errno != ENXIO: we learned nothing
2422 * Pretend we know nothing at all, i.e. "forget" about D1.
2424 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2426 return -errno
; /* D1 and (H3 or H4) */
2430 /* This is not a valid return by lseek(). We are safe to just return
2431 * -EIO in this case, and we'll treat it like H4. */
2437 * D1 and H2: either in data, next hole at offs, or it was in
2438 * data but is now in a trailing hole. In the latter case,
2439 * all bets are off. Treating it as if it there was data all
2440 * the way to EOF is safe, so simply do that.
2455 * Returns the allocation status of the specified offset.
2457 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2459 * 'pnum' is set to the number of bytes (including and immediately following
2460 * the specified offset) that are known to be in the same
2461 * allocated/unallocated state.
2463 * 'bytes' is the max value 'pnum' should be set to.
2465 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2468 int64_t bytes
, int64_t *pnum
,
2470 BlockDriverState
**file
)
2472 off_t data
= 0, hole
= 0;
2484 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2487 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2488 if (ret
== -ENXIO
) {
2491 ret
= BDRV_BLOCK_ZERO
;
2492 } else if (ret
< 0) {
2493 /* No info available, so pretend there are no holes */
2495 ret
= BDRV_BLOCK_DATA
;
2496 } else if (data
== offset
) {
2497 /* On a data extent, compute bytes to the end of the extent,
2498 * possibly including a partial sector at EOF. */
2499 *pnum
= MIN(bytes
, hole
- offset
);
2500 ret
= BDRV_BLOCK_DATA
;
2502 /* On a hole, compute bytes to the beginning of the next extent. */
2503 assert(hole
== offset
);
2504 *pnum
= MIN(bytes
, data
- offset
);
2505 ret
= BDRV_BLOCK_ZERO
;
2509 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2512 #if defined(__linux__)
2513 /* Verify that the file is not in the page cache */
2514 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2516 const size_t window_size
= 128 * 1024 * 1024;
2517 BDRVRawState
*s
= bs
->opaque
;
2518 void *window
= NULL
;
2525 /* mincore(2) page status information requires 1 byte per page */
2526 page_size
= sysconf(_SC_PAGESIZE
);
2527 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2529 end
= raw_getlength(bs
);
2531 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2538 /* Unmap previous window if size has changed */
2539 new_length
= MIN(end
- offset
, window_size
);
2540 if (new_length
!= length
) {
2541 munmap(window
, length
);
2546 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2548 if (new_window
== MAP_FAILED
) {
2549 error_setg_errno(errp
, errno
, "mmap failed");
2553 window
= new_window
;
2554 length
= new_length
;
2556 ret
= mincore(window
, length
, vec
);
2558 error_setg_errno(errp
, errno
, "mincore failed");
2562 vec_end
= DIV_ROUND_UP(length
, page_size
);
2563 for (i
= 0; i
< vec_end
; i
++) {
2565 error_setg(errp
, "page cache still in use!");
2572 munmap(window
, length
);
2577 #endif /* __linux__ */
2579 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2582 BDRVRawState
*s
= bs
->opaque
;
2587 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2591 if (s
->open_flags
& O_DIRECT
) {
2592 return; /* No host kernel page cache */
2595 #if defined(__linux__)
2596 /* This sets the scene for the next syscall... */
2597 ret
= bdrv_co_flush(bs
);
2599 error_setg_errno(errp
, -ret
, "flush failed");
2603 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2604 * process. These limitations are okay because we just fsynced the file,
2605 * we don't use mmap, and the file should not be in use by other processes.
2607 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2608 if (ret
!= 0) { /* the return value is a positive errno */
2609 error_setg_errno(errp
, ret
, "fadvise failed");
2613 if (s
->check_cache_dropped
) {
2614 check_cache_dropped(bs
, errp
);
2616 #else /* __linux__ */
2617 /* Do nothing. Live migration to a remote host with cache.direct=off is
2618 * unsupported on other host operating systems. Cache consistency issues
2619 * may occur but no error is reported here, partly because that's the
2620 * historical behavior and partly because it's hard to differentiate valid
2621 * configurations that should not cause errors.
2623 #endif /* !__linux__ */
2626 static coroutine_fn
int
2627 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2629 BDRVRawState
*s
= bs
->opaque
;
2631 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, QEMU_AIO_DISCARD
);
2634 static int coroutine_fn
raw_co_pwrite_zeroes(
2635 BlockDriverState
*bs
, int64_t offset
,
2636 int bytes
, BdrvRequestFlags flags
)
2638 BDRVRawState
*s
= bs
->opaque
;
2639 int operation
= QEMU_AIO_WRITE_ZEROES
;
2641 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2642 operation
|= QEMU_AIO_DISCARD
;
2645 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, operation
);
2648 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2650 BDRVRawState
*s
= bs
->opaque
;
2652 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2656 static QemuOptsList raw_create_opts
= {
2657 .name
= "raw-create-opts",
2658 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2661 .name
= BLOCK_OPT_SIZE
,
2662 .type
= QEMU_OPT_SIZE
,
2663 .help
= "Virtual disk size"
2666 .name
= BLOCK_OPT_NOCOW
,
2667 .type
= QEMU_OPT_BOOL
,
2668 .help
= "Turn off copy-on-write (valid only on btrfs)"
2671 .name
= BLOCK_OPT_PREALLOC
,
2672 .type
= QEMU_OPT_STRING
,
2673 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2675 { /* end of list */ }
2679 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2682 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2685 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2687 BDRVRawState
*s
= bs
->opaque
;
2688 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2690 s
->shared_perm
= shared
;
2693 static void raw_abort_perm_update(BlockDriverState
*bs
)
2695 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2698 static int coroutine_fn
raw_co_copy_range_from(
2699 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2700 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2701 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2703 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2704 read_flags
, write_flags
);
2707 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2709 uint64_t src_offset
,
2711 uint64_t dst_offset
,
2713 BdrvRequestFlags read_flags
,
2714 BdrvRequestFlags write_flags
)
2716 BDRVRawState
*s
= bs
->opaque
;
2717 BDRVRawState
*src_s
;
2719 assert(dst
->bs
== bs
);
2720 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2724 src_s
= src
->bs
->opaque
;
2725 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2728 return paio_submit_co_full(bs
, src_s
->fd
, src_offset
, s
->fd
, dst_offset
,
2729 NULL
, bytes
, QEMU_AIO_COPY_RANGE
);
2732 BlockDriver bdrv_file
= {
2733 .format_name
= "file",
2734 .protocol_name
= "file",
2735 .instance_size
= sizeof(BDRVRawState
),
2736 .bdrv_needs_filename
= true,
2737 .bdrv_probe
= NULL
, /* no probe for protocols */
2738 .bdrv_parse_filename
= raw_parse_filename
,
2739 .bdrv_file_open
= raw_open
,
2740 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2741 .bdrv_reopen_commit
= raw_reopen_commit
,
2742 .bdrv_reopen_abort
= raw_reopen_abort
,
2743 .bdrv_close
= raw_close
,
2744 .bdrv_co_create
= raw_co_create
,
2745 .bdrv_co_create_opts
= raw_co_create_opts
,
2746 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2747 .bdrv_co_block_status
= raw_co_block_status
,
2748 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2749 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2751 .bdrv_co_preadv
= raw_co_preadv
,
2752 .bdrv_co_pwritev
= raw_co_pwritev
,
2753 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2754 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2755 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2756 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2757 .bdrv_refresh_limits
= raw_refresh_limits
,
2758 .bdrv_io_plug
= raw_aio_plug
,
2759 .bdrv_io_unplug
= raw_aio_unplug
,
2760 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2762 .bdrv_co_truncate
= raw_co_truncate
,
2763 .bdrv_getlength
= raw_getlength
,
2764 .bdrv_get_info
= raw_get_info
,
2765 .bdrv_get_allocated_file_size
2766 = raw_get_allocated_file_size
,
2767 .bdrv_check_perm
= raw_check_perm
,
2768 .bdrv_set_perm
= raw_set_perm
,
2769 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2770 .create_opts
= &raw_create_opts
,
2773 /***********************************************/
2776 #if defined(__APPLE__) && defined(__MACH__)
2777 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2778 CFIndex maxPathSize
, int flags
);
2779 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2781 kern_return_t kernResult
= KERN_FAILURE
;
2782 mach_port_t masterPort
;
2783 CFMutableDictionaryRef classesToMatch
;
2784 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2785 char *mediaType
= NULL
;
2787 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2788 if ( KERN_SUCCESS
!= kernResult
) {
2789 printf( "IOMasterPort returned %d\n", kernResult
);
2793 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2794 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2795 if (classesToMatch
== NULL
) {
2796 error_report("IOServiceMatching returned NULL for %s",
2797 matching_array
[index
]);
2800 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2802 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2804 if (kernResult
!= KERN_SUCCESS
) {
2805 error_report("Note: IOServiceGetMatchingServices returned %d",
2810 /* If a match was found, leave the loop */
2811 if (*mediaIterator
!= 0) {
2812 DPRINTF("Matching using %s\n", matching_array
[index
]);
2813 mediaType
= g_strdup(matching_array
[index
]);
2820 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2821 CFIndex maxPathSize
, int flags
)
2823 io_object_t nextMedia
;
2824 kern_return_t kernResult
= KERN_FAILURE
;
2826 nextMedia
= IOIteratorNext( mediaIterator
);
2829 CFTypeRef bsdPathAsCFString
;
2830 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2831 if ( bsdPathAsCFString
) {
2832 size_t devPathLength
;
2833 strcpy( bsdPath
, _PATH_DEV
);
2834 if (flags
& BDRV_O_NOCACHE
) {
2835 strcat(bsdPath
, "r");
2837 devPathLength
= strlen( bsdPath
);
2838 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2839 kernResult
= KERN_SUCCESS
;
2841 CFRelease( bsdPathAsCFString
);
2843 IOObjectRelease( nextMedia
);
2849 /* Sets up a real cdrom for use in QEMU */
2850 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2852 int index
, num_of_test_partitions
= 2, fd
;
2853 char test_partition
[MAXPATHLEN
];
2854 bool partition_found
= false;
2856 /* look for a working partition */
2857 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2858 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2860 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2862 partition_found
= true;
2868 /* if a working partition on the device was not found */
2869 if (partition_found
== false) {
2870 error_setg(errp
, "Failed to find a working partition on disc");
2872 DPRINTF("Using %s as optical disc\n", test_partition
);
2873 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2875 return partition_found
;
2878 /* Prints directions on mounting and unmounting a device */
2879 static void print_unmounting_directions(const char *file_name
)
2881 error_report("If device %s is mounted on the desktop, unmount"
2882 " it first before using it in QEMU", file_name
);
2883 error_report("Command to unmount device: diskutil unmountDisk %s",
2885 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2888 #endif /* defined(__APPLE__) && defined(__MACH__) */
2890 static int hdev_probe_device(const char *filename
)
2894 /* allow a dedicated CD-ROM driver to match with a higher priority */
2895 if (strstart(filename
, "/dev/cdrom", NULL
))
2898 if (stat(filename
, &st
) >= 0 &&
2899 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2906 static int check_hdev_writable(BDRVRawState
*s
)
2908 #if defined(BLKROGET)
2909 /* Linux block devices can be configured "read-only" using blockdev(8).
2910 * This is independent of device node permissions and therefore open(2)
2911 * with O_RDWR succeeds. Actual writes fail with EPERM.
2913 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2914 * check for read-only block devices so that Linux block devices behave
2920 if (fstat(s
->fd
, &st
)) {
2924 if (!S_ISBLK(st
.st_mode
)) {
2928 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2935 #endif /* defined(BLKROGET) */
2939 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2942 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2945 static bool hdev_is_sg(BlockDriverState
*bs
)
2948 #if defined(__linux__)
2950 BDRVRawState
*s
= bs
->opaque
;
2952 struct sg_scsi_id scsiid
;
2956 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2960 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2965 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2967 DPRINTF("SG device found: type=%d, version=%d\n",
2968 scsiid
.scsi_type
, sg_version
);
2977 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2980 BDRVRawState
*s
= bs
->opaque
;
2981 Error
*local_err
= NULL
;
2984 #if defined(__APPLE__) && defined(__MACH__)
2986 * Caution: while qdict_get_str() is fine, getting non-string types
2987 * would require more care. When @options come from -blockdev or
2988 * blockdev_add, its members are typed according to the QAPI
2989 * schema, but when they come from -drive, they're all QString.
2991 const char *filename
= qdict_get_str(options
, "filename");
2992 char bsd_path
[MAXPATHLEN
] = "";
2993 bool error_occurred
= false;
2995 /* If using a real cdrom */
2996 if (strcmp(filename
, "/dev/cdrom") == 0) {
2997 char *mediaType
= NULL
;
2998 kern_return_t ret_val
;
2999 io_iterator_t mediaIterator
= 0;
3001 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3002 if (mediaType
== NULL
) {
3003 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3005 error_occurred
= true;
3006 goto hdev_open_Mac_error
;
3009 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3010 if (ret_val
!= KERN_SUCCESS
) {
3011 error_setg(errp
, "Could not get BSD path for optical drive");
3012 error_occurred
= true;
3013 goto hdev_open_Mac_error
;
3016 /* If a real optical drive was not found */
3017 if (bsd_path
[0] == '\0') {
3018 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3019 error_occurred
= true;
3020 goto hdev_open_Mac_error
;
3023 /* If using a cdrom disc and finding a partition on the disc failed */
3024 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3025 setup_cdrom(bsd_path
, errp
) == false) {
3026 print_unmounting_directions(bsd_path
);
3027 error_occurred
= true;
3028 goto hdev_open_Mac_error
;
3031 qdict_put_str(options
, "filename", bsd_path
);
3033 hdev_open_Mac_error
:
3035 if (mediaIterator
) {
3036 IOObjectRelease(mediaIterator
);
3038 if (error_occurred
) {
3042 #endif /* defined(__APPLE__) && defined(__MACH__) */
3044 s
->type
= FTYPE_FILE
;
3046 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3048 error_propagate(errp
, local_err
);
3049 #if defined(__APPLE__) && defined(__MACH__)
3051 filename
= bsd_path
;
3053 /* if a physical device experienced an error while being opened */
3054 if (strncmp(filename
, "/dev/", 5) == 0) {
3055 print_unmounting_directions(filename
);
3057 #endif /* defined(__APPLE__) && defined(__MACH__) */
3061 /* Since this does ioctl the device must be already opened */
3062 bs
->sg
= hdev_is_sg(bs
);
3064 if (flags
& BDRV_O_RDWR
) {
3065 ret
= check_hdev_writable(s
);
3068 error_setg_errno(errp
, -ret
, "The device is not writable");
3076 #if defined(__linux__)
3078 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
3079 unsigned long int req
, void *buf
,
3080 BlockCompletionFunc
*cb
, void *opaque
)
3082 BDRVRawState
*s
= bs
->opaque
;
3083 RawPosixAIOData
*acb
;
3086 if (fd_open(bs
) < 0)
3089 if (req
== SG_IO
&& s
->pr_mgr
) {
3090 struct sg_io_hdr
*io_hdr
= buf
;
3091 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3092 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3093 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3094 s
->fd
, io_hdr
, cb
, opaque
);
3098 acb
= g_new(RawPosixAIOData
, 1);
3100 acb
->aio_type
= QEMU_AIO_IOCTL
;
3101 acb
->aio_fildes
= s
->fd
;
3102 acb
->aio_offset
= 0;
3103 acb
->aio_ioctl_buf
= buf
;
3104 acb
->aio_ioctl_cmd
= req
;
3105 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3106 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
3110 static int fd_open(BlockDriverState
*bs
)
3112 BDRVRawState
*s
= bs
->opaque
;
3114 /* this is just to ensure s->fd is sane (its called by io ops) */
3120 static coroutine_fn
int
3121 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3123 BDRVRawState
*s
= bs
->opaque
;
3130 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3131 QEMU_AIO_DISCARD
| QEMU_AIO_BLKDEV
);
3134 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3135 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3137 BDRVRawState
*s
= bs
->opaque
;
3138 int operation
= QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_BLKDEV
;
3146 if (flags
& BDRV_REQ_MAY_UNMAP
) {
3147 operation
|= QEMU_AIO_DISCARD
;
3150 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, operation
);
3153 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3158 struct stat stat_buf
;
3159 int64_t total_size
= 0;
3162 /* This function is used by both protocol block drivers and therefore either
3163 * of these prefixes may be given.
3164 * The return value has to be stored somewhere, otherwise this is an error
3165 * due to -Werror=unused-value. */
3167 strstart(filename
, "host_device:", &filename
) ||
3168 strstart(filename
, "host_cdrom:" , &filename
);
3172 ret
= raw_normalize_devicepath(&filename
, errp
);
3177 /* Read out options */
3178 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3181 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3184 error_setg_errno(errp
, -ret
, "Could not open device");
3188 if (fstat(fd
, &stat_buf
) < 0) {
3190 error_setg_errno(errp
, -ret
, "Could not stat device");
3191 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3193 "The given file is neither a block nor a character device");
3195 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3196 error_setg(errp
, "Device is too small");
3200 if (!ret
&& total_size
) {
3201 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3202 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3203 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3206 ret
= qemu_write_full(fd
, buf
, zero_size
);
3207 ret
= ret
== zero_size
? 0 : -errno
;
3214 static BlockDriver bdrv_host_device
= {
3215 .format_name
= "host_device",
3216 .protocol_name
= "host_device",
3217 .instance_size
= sizeof(BDRVRawState
),
3218 .bdrv_needs_filename
= true,
3219 .bdrv_probe_device
= hdev_probe_device
,
3220 .bdrv_parse_filename
= hdev_parse_filename
,
3221 .bdrv_file_open
= hdev_open
,
3222 .bdrv_close
= raw_close
,
3223 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3224 .bdrv_reopen_commit
= raw_reopen_commit
,
3225 .bdrv_reopen_abort
= raw_reopen_abort
,
3226 .bdrv_co_create_opts
= hdev_co_create_opts
,
3227 .create_opts
= &raw_create_opts
,
3228 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3229 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3231 .bdrv_co_preadv
= raw_co_preadv
,
3232 .bdrv_co_pwritev
= raw_co_pwritev
,
3233 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3234 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3235 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3236 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3237 .bdrv_refresh_limits
= raw_refresh_limits
,
3238 .bdrv_io_plug
= raw_aio_plug
,
3239 .bdrv_io_unplug
= raw_aio_unplug
,
3240 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3242 .bdrv_co_truncate
= raw_co_truncate
,
3243 .bdrv_getlength
= raw_getlength
,
3244 .bdrv_get_info
= raw_get_info
,
3245 .bdrv_get_allocated_file_size
3246 = raw_get_allocated_file_size
,
3247 .bdrv_check_perm
= raw_check_perm
,
3248 .bdrv_set_perm
= raw_set_perm
,
3249 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3250 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3251 .bdrv_probe_geometry
= hdev_probe_geometry
,
3253 /* generic scsi device */
3255 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3259 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3260 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3263 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3268 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3271 BDRVRawState
*s
= bs
->opaque
;
3275 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3276 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3279 static int cdrom_probe_device(const char *filename
)
3285 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3289 ret
= fstat(fd
, &st
);
3290 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3294 /* Attempt to detect via a CDROM specific ioctl */
3295 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3305 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3307 BDRVRawState
*s
= bs
->opaque
;
3310 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3311 return ret
== CDS_DISC_OK
;
3314 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3316 BDRVRawState
*s
= bs
->opaque
;
3319 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3320 perror("CDROMEJECT");
3322 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3323 perror("CDROMEJECT");
3327 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3329 BDRVRawState
*s
= bs
->opaque
;
3331 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3333 * Note: an error can happen if the distribution automatically
3336 /* perror("CDROM_LOCKDOOR"); */
3340 static BlockDriver bdrv_host_cdrom
= {
3341 .format_name
= "host_cdrom",
3342 .protocol_name
= "host_cdrom",
3343 .instance_size
= sizeof(BDRVRawState
),
3344 .bdrv_needs_filename
= true,
3345 .bdrv_probe_device
= cdrom_probe_device
,
3346 .bdrv_parse_filename
= cdrom_parse_filename
,
3347 .bdrv_file_open
= cdrom_open
,
3348 .bdrv_close
= raw_close
,
3349 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3350 .bdrv_reopen_commit
= raw_reopen_commit
,
3351 .bdrv_reopen_abort
= raw_reopen_abort
,
3352 .bdrv_co_create_opts
= hdev_co_create_opts
,
3353 .create_opts
= &raw_create_opts
,
3354 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3357 .bdrv_co_preadv
= raw_co_preadv
,
3358 .bdrv_co_pwritev
= raw_co_pwritev
,
3359 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3360 .bdrv_refresh_limits
= raw_refresh_limits
,
3361 .bdrv_io_plug
= raw_aio_plug
,
3362 .bdrv_io_unplug
= raw_aio_unplug
,
3363 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3365 .bdrv_co_truncate
= raw_co_truncate
,
3366 .bdrv_getlength
= raw_getlength
,
3367 .has_variable_length
= true,
3368 .bdrv_get_allocated_file_size
3369 = raw_get_allocated_file_size
,
3371 /* removable device support */
3372 .bdrv_is_inserted
= cdrom_is_inserted
,
3373 .bdrv_eject
= cdrom_eject
,
3374 .bdrv_lock_medium
= cdrom_lock_medium
,
3376 /* generic scsi device */
3377 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3379 #endif /* __linux__ */
3381 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3382 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3385 BDRVRawState
*s
= bs
->opaque
;
3386 Error
*local_err
= NULL
;
3391 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3393 error_propagate(errp
, local_err
);
3397 /* make sure the door isn't locked at this time */
3398 ioctl(s
->fd
, CDIOCALLOW
);
3402 static int cdrom_probe_device(const char *filename
)
3404 if (strstart(filename
, "/dev/cd", NULL
) ||
3405 strstart(filename
, "/dev/acd", NULL
))
3410 static int cdrom_reopen(BlockDriverState
*bs
)
3412 BDRVRawState
*s
= bs
->opaque
;
3416 * Force reread of possibly changed/newly loaded disc,
3417 * FreeBSD seems to not notice sometimes...
3421 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3428 /* make sure the door isn't locked at this time */
3429 ioctl(s
->fd
, CDIOCALLOW
);
3433 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3435 return raw_getlength(bs
) > 0;
3438 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3440 BDRVRawState
*s
= bs
->opaque
;
3445 (void) ioctl(s
->fd
, CDIOCALLOW
);
3448 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3449 perror("CDIOCEJECT");
3451 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3452 perror("CDIOCCLOSE");
3458 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3460 BDRVRawState
*s
= bs
->opaque
;
3464 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3466 * Note: an error can happen if the distribution automatically
3469 /* perror("CDROM_LOCKDOOR"); */
3473 static BlockDriver bdrv_host_cdrom
= {
3474 .format_name
= "host_cdrom",
3475 .protocol_name
= "host_cdrom",
3476 .instance_size
= sizeof(BDRVRawState
),
3477 .bdrv_needs_filename
= true,
3478 .bdrv_probe_device
= cdrom_probe_device
,
3479 .bdrv_parse_filename
= cdrom_parse_filename
,
3480 .bdrv_file_open
= cdrom_open
,
3481 .bdrv_close
= raw_close
,
3482 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3483 .bdrv_reopen_commit
= raw_reopen_commit
,
3484 .bdrv_reopen_abort
= raw_reopen_abort
,
3485 .bdrv_co_create_opts
= hdev_co_create_opts
,
3486 .create_opts
= &raw_create_opts
,
3488 .bdrv_co_preadv
= raw_co_preadv
,
3489 .bdrv_co_pwritev
= raw_co_pwritev
,
3490 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3491 .bdrv_refresh_limits
= raw_refresh_limits
,
3492 .bdrv_io_plug
= raw_aio_plug
,
3493 .bdrv_io_unplug
= raw_aio_unplug
,
3494 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3496 .bdrv_co_truncate
= raw_co_truncate
,
3497 .bdrv_getlength
= raw_getlength
,
3498 .has_variable_length
= true,
3499 .bdrv_get_allocated_file_size
3500 = raw_get_allocated_file_size
,
3502 /* removable device support */
3503 .bdrv_is_inserted
= cdrom_is_inserted
,
3504 .bdrv_eject
= cdrom_eject
,
3505 .bdrv_lock_medium
= cdrom_lock_medium
,
3507 #endif /* __FreeBSD__ */
3509 static void bdrv_file_init(void)
3512 * Register all the drivers. Note that order is important, the driver
3513 * registered last will get probed first.
3515 bdrv_register(&bdrv_file
);
3516 bdrv_register(&bdrv_host_device
);
3518 bdrv_register(&bdrv_host_cdrom
);
3520 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3521 bdrv_register(&bdrv_host_cdrom
);
3525 block_init(bdrv_file_init
);