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
24 #include "qemu/osdep.h"
25 #include "qapi/error.h"
26 #include "qemu/cutils.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "qemu/module.h"
31 #include "block/thread-pool.h"
33 #include "block/raw-aio.h"
34 #include "qapi/qmp/qstring.h"
36 #if defined(__APPLE__) && (__MACH__)
38 #include <sys/param.h>
39 #include <IOKit/IOKitLib.h>
40 #include <IOKit/IOBSD.h>
41 #include <IOKit/storage/IOMediaBSDClient.h>
42 #include <IOKit/storage/IOMedia.h>
43 #include <IOKit/storage/IOCDMedia.h>
44 //#include <IOKit/storage/IOCDTypes.h>
45 #include <IOKit/storage/IODVDMedia.h>
46 #include <CoreFoundation/CoreFoundation.h>
50 #define _POSIX_PTHREAD_SEMANTICS 1
54 #include <sys/ioctl.h>
55 #include <sys/param.h>
56 #include <linux/cdrom.h>
59 #include <linux/hdreg.h>
65 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
68 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
69 #include <linux/falloc.h>
71 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
77 #include <sys/ioctl.h>
78 #include <sys/disklabel.h>
83 #include <sys/ioctl.h>
84 #include <sys/disklabel.h>
90 #include <sys/ioctl.h>
91 #include <sys/diskslice.h>
101 # define DEBUG_BLOCK_PRINT 1
103 # define DEBUG_BLOCK_PRINT 0
105 #define DPRINTF(fmt, ...) \
107 if (DEBUG_BLOCK_PRINT) { \
108 printf(fmt, ## __VA_ARGS__); \
112 /* OS X does not have O_DSYNC */
115 #define O_DSYNC O_SYNC
116 #elif defined(O_FSYNC)
117 #define O_DSYNC O_FSYNC
121 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
123 #define O_DIRECT O_DSYNC
129 #define MAX_BLOCKSIZE 4096
131 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
132 * leaving a few more bytes for its future use. */
133 #define RAW_LOCK_PERM_BASE 100
134 #define RAW_LOCK_SHARED_BASE 200
136 typedef struct BDRVRawState
{
144 /* The current permissions. */
146 uint64_t shared_perm
;
152 bool has_write_zeroes
:1;
153 bool discard_zeroes
:1;
154 bool use_linux_aio
:1;
155 bool page_cache_inconsistent
:1;
157 bool needs_alignment
;
160 typedef struct BDRVRawReopenState
{
163 } BDRVRawReopenState
;
165 static int fd_open(BlockDriverState
*bs
);
166 static int64_t raw_getlength(BlockDriverState
*bs
);
168 typedef struct RawPosixAIOData
{
169 BlockDriverState
*bs
;
172 struct iovec
*aio_iov
;
177 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
182 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
183 static int cdrom_reopen(BlockDriverState
*bs
);
186 #if defined(__NetBSD__)
187 static int raw_normalize_devicepath(const char **filename
)
189 static char namebuf
[PATH_MAX
];
190 const char *dp
, *fname
;
194 dp
= strrchr(fname
, '/');
195 if (lstat(fname
, &sb
) < 0) {
196 fprintf(stderr
, "%s: stat failed: %s\n",
197 fname
, strerror(errno
));
201 if (!S_ISBLK(sb
.st_mode
)) {
206 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
208 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
209 (int)(dp
- fname
), fname
, dp
+ 1);
211 fprintf(stderr
, "%s is a block device", fname
);
213 fprintf(stderr
, ", using %s\n", *filename
);
218 static int raw_normalize_devicepath(const char **filename
)
225 * Get logical block size via ioctl. On success store it in @sector_size_p.
227 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
229 unsigned int sector_size
;
230 bool success
= false;
234 static const unsigned long ioctl_list
[] = {
238 #ifdef DKIOCGETBLOCKSIZE
241 #ifdef DIOCGSECTORSIZE
246 /* Try a few ioctls to get the right size */
247 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
248 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
249 *sector_size_p
= sector_size
;
254 return success
? 0 : -errno
;
258 * Get physical block size of @fd.
259 * On success, store it in @blk_size and return 0.
260 * On failure, return -errno.
262 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
265 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
274 /* Check if read is allowed with given memory buffer and length.
276 * This function is used to check O_DIRECT memory buffer and request alignment.
278 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
280 ssize_t ret
= pread(fd
, buf
, len
, 0);
287 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
288 * other errors (e.g. real I/O error), which could happen on a failed
289 * drive, since we only care about probing alignment.
291 if (errno
!= EINVAL
) {
299 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
301 BDRVRawState
*s
= bs
->opaque
;
303 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
305 /* For SCSI generic devices the alignment is not really used.
306 With buffered I/O, we don't have any restrictions. */
307 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
308 bs
->bl
.request_alignment
= 1;
313 bs
->bl
.request_alignment
= 0;
315 /* Let's try to use the logical blocksize for the alignment. */
316 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
317 bs
->bl
.request_alignment
= 0;
322 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
323 bs
->bl
.request_alignment
= da
.d_miniosz
;
324 /* The kernel returns wrong information for d_mem */
325 /* s->buf_align = da.d_mem; */
330 /* If we could not get the sizes so far, we can only guess them */
333 buf
= qemu_memalign(max_align
, 2 * max_align
);
334 for (align
= 512; align
<= max_align
; align
<<= 1) {
335 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
336 s
->buf_align
= align
;
343 if (!bs
->bl
.request_alignment
) {
345 buf
= qemu_memalign(s
->buf_align
, max_align
);
346 for (align
= 512; align
<= max_align
; align
<<= 1) {
347 if (raw_is_io_aligned(fd
, buf
, align
)) {
348 bs
->bl
.request_alignment
= align
;
355 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
356 error_setg(errp
, "Could not find working O_DIRECT alignment");
357 error_append_hint(errp
, "Try cache.direct=off\n");
361 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
363 assert(open_flags
!= NULL
);
365 *open_flags
|= O_BINARY
;
366 *open_flags
&= ~O_ACCMODE
;
367 if (bdrv_flags
& BDRV_O_RDWR
) {
368 *open_flags
|= O_RDWR
;
370 *open_flags
|= O_RDONLY
;
373 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
374 * and O_DIRECT for no caching. */
375 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
376 *open_flags
|= O_DIRECT
;
380 static void raw_parse_filename(const char *filename
, QDict
*options
,
383 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
386 static QemuOptsList raw_runtime_opts
= {
388 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
392 .type
= QEMU_OPT_STRING
,
393 .help
= "File name of the image",
397 .type
= QEMU_OPT_STRING
,
398 .help
= "host AIO implementation (threads, native)",
402 .type
= QEMU_OPT_STRING
,
403 .help
= "file locking mode (on/off/auto, default: auto)",
405 { /* end of list */ }
409 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
410 int bdrv_flags
, int open_flags
, Error
**errp
)
412 BDRVRawState
*s
= bs
->opaque
;
414 Error
*local_err
= NULL
;
415 const char *filename
= NULL
;
416 BlockdevAioOptions aio
, aio_default
;
421 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
422 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
424 error_propagate(errp
, local_err
);
429 filename
= qemu_opt_get(opts
, "filename");
431 ret
= raw_normalize_devicepath(&filename
);
433 error_setg_errno(errp
, -ret
, "Could not normalize device path");
437 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
438 ? BLOCKDEV_AIO_OPTIONS_NATIVE
439 : BLOCKDEV_AIO_OPTIONS_THREADS
;
440 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
441 qemu_opt_get(opts
, "aio"),
442 aio_default
, &local_err
);
444 error_propagate(errp
, local_err
);
448 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
450 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
451 qemu_opt_get(opts
, "locking"),
452 ON_OFF_AUTO_AUTO
, &local_err
);
454 error_propagate(errp
, local_err
);
461 if (!qemu_has_ofd_lock()) {
463 "File lock requested but OFD locking syscall is "
464 "unavailable, falling back to POSIX file locks.\n"
465 "Due to the implementation, locks can be lost "
469 case ON_OFF_AUTO_OFF
:
472 case ON_OFF_AUTO_AUTO
:
473 s
->use_lock
= qemu_has_ofd_lock();
479 s
->open_flags
= open_flags
;
480 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
483 fd
= qemu_open(filename
, s
->open_flags
, 0644);
486 error_setg_errno(errp
, errno
, "Could not open '%s'", filename
);
496 fd
= qemu_open(filename
, s
->open_flags
);
499 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
507 s
->shared_perm
= BLK_PERM_ALL
;
509 #ifdef CONFIG_LINUX_AIO
510 /* Currently Linux does AIO only for files opened with O_DIRECT */
511 if (s
->use_linux_aio
&& !(s
->open_flags
& O_DIRECT
)) {
512 error_setg(errp
, "aio=native was specified, but it requires "
513 "cache.direct=on, which was not specified.");
518 if (s
->use_linux_aio
) {
519 error_setg(errp
, "aio=native was specified, but is not supported "
524 #endif /* !defined(CONFIG_LINUX_AIO) */
526 s
->has_discard
= true;
527 s
->has_write_zeroes
= true;
528 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
529 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
530 s
->needs_alignment
= true;
533 if (fstat(s
->fd
, &st
) < 0) {
535 error_setg_errno(errp
, errno
, "Could not stat file");
538 if (S_ISREG(st
.st_mode
)) {
539 s
->discard_zeroes
= true;
540 s
->has_fallocate
= true;
542 if (S_ISBLK(st
.st_mode
)) {
543 #ifdef BLKDISCARDZEROES
545 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
546 s
->discard_zeroes
= true;
550 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
551 * not rely on the contents of discarded blocks unless using O_DIRECT.
552 * Same for BLKZEROOUT.
554 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
555 s
->discard_zeroes
= false;
556 s
->has_write_zeroes
= false;
561 if (S_ISCHR(st
.st_mode
)) {
563 * The file is a char device (disk), which on FreeBSD isn't behind
564 * a pager, so force all requests to be aligned. This is needed
565 * so QEMU makes sure all IO operations on the device are aligned
566 * to sector size, or else FreeBSD will reject them with EINVAL.
568 s
->needs_alignment
= true;
573 if (platform_test_xfs_fd(s
->fd
)) {
580 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
587 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
590 BDRVRawState
*s
= bs
->opaque
;
592 s
->type
= FTYPE_FILE
;
593 return raw_open_common(bs
, options
, flags
, 0, errp
);
602 #define PERM_FOREACH(i) \
603 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
605 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
606 * file; if @unlock == true, also unlock the unneeded bytes.
607 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
609 static int raw_apply_lock_bytes(BDRVRawState
*s
,
610 uint64_t perm_lock_bits
,
611 uint64_t shared_perm_lock_bits
,
612 bool unlock
, Error
**errp
)
618 int off
= RAW_LOCK_PERM_BASE
+ i
;
619 if (perm_lock_bits
& (1ULL << i
)) {
620 ret
= qemu_lock_fd(s
->lock_fd
, off
, 1, false);
622 error_setg(errp
, "Failed to lock byte %d", off
);
626 ret
= qemu_unlock_fd(s
->lock_fd
, off
, 1);
628 error_setg(errp
, "Failed to unlock byte %d", off
);
634 int off
= RAW_LOCK_SHARED_BASE
+ i
;
635 if (shared_perm_lock_bits
& (1ULL << i
)) {
636 ret
= qemu_lock_fd(s
->lock_fd
, off
, 1, false);
638 error_setg(errp
, "Failed to lock byte %d", off
);
642 ret
= qemu_unlock_fd(s
->lock_fd
, off
, 1);
644 error_setg(errp
, "Failed to unlock byte %d", off
);
652 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
653 static int raw_check_lock_bytes(BDRVRawState
*s
,
654 uint64_t perm
, uint64_t shared_perm
,
661 int off
= RAW_LOCK_SHARED_BASE
+ i
;
662 uint64_t p
= 1ULL << i
;
664 ret
= qemu_lock_fd_test(s
->lock_fd
, off
, 1, true);
666 char *perm_name
= bdrv_perm_names(p
);
668 "Failed to get \"%s\" lock",
671 error_append_hint(errp
,
672 "Is another process using the image?\n");
678 int off
= RAW_LOCK_PERM_BASE
+ i
;
679 uint64_t p
= 1ULL << i
;
680 if (!(shared_perm
& p
)) {
681 ret
= qemu_lock_fd_test(s
->lock_fd
, off
, 1, true);
683 char *perm_name
= bdrv_perm_names(p
);
685 "Failed to get shared \"%s\" lock",
688 error_append_hint(errp
,
689 "Is another process using the image?\n");
697 static int raw_handle_perm_lock(BlockDriverState
*bs
,
699 uint64_t new_perm
, uint64_t new_shared
,
702 BDRVRawState
*s
= bs
->opaque
;
704 Error
*local_err
= NULL
;
710 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
714 assert(s
->lock_fd
> 0);
718 ret
= raw_apply_lock_bytes(s
, s
->perm
| new_perm
,
719 ~s
->shared_perm
| ~new_shared
,
722 ret
= raw_check_lock_bytes(s
, new_perm
, new_shared
, errp
);
728 /* fall through to unlock bytes. */
730 raw_apply_lock_bytes(s
, s
->perm
, ~s
->shared_perm
, true, &local_err
);
732 /* Theoretically the above call only unlocks bytes and it cannot
733 * fail. Something weird happened, report it.
735 error_report_err(local_err
);
739 raw_apply_lock_bytes(s
, new_perm
, ~new_shared
, true, &local_err
);
741 /* Theoretically the above call only unlocks bytes and it cannot
742 * fail. Something weird happened, report it.
744 error_report_err(local_err
);
751 static int raw_reopen_prepare(BDRVReopenState
*state
,
752 BlockReopenQueue
*queue
, Error
**errp
)
755 BDRVRawReopenState
*rs
;
757 Error
*local_err
= NULL
;
759 assert(state
!= NULL
);
760 assert(state
->bs
!= NULL
);
762 s
= state
->bs
->opaque
;
764 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
767 if (s
->type
== FTYPE_CD
) {
768 rs
->open_flags
|= O_NONBLOCK
;
771 raw_parse_flags(state
->flags
, &rs
->open_flags
);
775 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
777 fcntl_flags
|= O_NOATIME
;
781 /* Not all operating systems have O_ASYNC, and those that don't
782 * will not let us track the state into rs->open_flags (typically
783 * you achieve the same effect with an ioctl, for example I_SETSIG
784 * on Solaris). But we do not use O_ASYNC, so that's fine.
786 assert((s
->open_flags
& O_ASYNC
) == 0);
789 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
790 /* dup the original fd */
791 rs
->fd
= qemu_dup(s
->fd
);
793 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
801 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
803 const char *normalized_filename
= state
->bs
->filename
;
804 ret
= raw_normalize_devicepath(&normalized_filename
);
806 error_setg_errno(errp
, -ret
, "Could not normalize device path");
808 assert(!(rs
->open_flags
& O_CREAT
));
809 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
811 error_setg_errno(errp
, errno
, "Could not reopen file");
817 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
818 * alignment with the new fd. */
820 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
824 error_propagate(errp
, local_err
);
832 static void raw_reopen_commit(BDRVReopenState
*state
)
834 BDRVRawReopenState
*rs
= state
->opaque
;
835 BDRVRawState
*s
= state
->bs
->opaque
;
837 s
->open_flags
= rs
->open_flags
;
842 g_free(state
->opaque
);
843 state
->opaque
= NULL
;
847 static void raw_reopen_abort(BDRVReopenState
*state
)
849 BDRVRawReopenState
*rs
= state
->opaque
;
851 /* nothing to do if NULL, we didn't get far enough */
860 g_free(state
->opaque
);
861 state
->opaque
= NULL
;
864 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
868 short max_sectors
= 0;
869 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
871 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
872 return max_sectors
<< BDRV_SECTOR_BITS
;
881 static int hdev_get_max_segments(const struct stat
*st
)
891 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
892 major(st
->st_rdev
), minor(st
->st_rdev
));
893 fd
= open(sysfspath
, O_RDONLY
);
899 ret
= read(fd
, buf
, sizeof(buf
) - 1);
900 } while (ret
== -1 && errno
== EINTR
);
904 } else if (ret
== 0) {
909 /* The file is ended with '\n', pass 'end' to accept that. */
910 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
911 if (ret
== 0 && end
&& *end
== '\n') {
926 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
928 BDRVRawState
*s
= bs
->opaque
;
931 if (!fstat(s
->fd
, &st
)) {
932 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
933 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
934 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
935 bs
->bl
.max_transfer
= pow2floor(ret
);
937 ret
= hdev_get_max_segments(&st
);
939 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
940 ret
* getpagesize());
945 raw_probe_alignment(bs
, s
->fd
, errp
);
946 bs
->bl
.min_mem_alignment
= s
->buf_align
;
947 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
950 static int check_for_dasd(int fd
)
953 struct dasd_information2_t info
= {0};
955 return ioctl(fd
, BIODASDINFO2
, &info
);
962 * Try to get @bs's logical and physical block size.
963 * On success, store them in @bsz and return zero.
964 * On failure, return negative errno.
966 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
968 BDRVRawState
*s
= bs
->opaque
;
971 /* If DASD, get blocksizes */
972 if (check_for_dasd(s
->fd
) < 0) {
975 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
979 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
983 * Try to get @bs's geometry: cyls, heads, sectors.
984 * On success, store them in @geo and return 0.
985 * On failure return -errno.
986 * (Allows block driver to assign default geometry values that guest sees)
989 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
991 BDRVRawState
*s
= bs
->opaque
;
992 struct hd_geometry ioctl_geo
= {0};
994 /* If DASD, get its geometry */
995 if (check_for_dasd(s
->fd
) < 0) {
998 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1001 /* HDIO_GETGEO may return success even though geo contains zeros
1002 (e.g. certain multipath setups) */
1003 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1006 /* Do not return a geometry for partition */
1007 if (ioctl_geo
.start
!= 0) {
1010 geo
->heads
= ioctl_geo
.heads
;
1011 geo
->sectors
= ioctl_geo
.sectors
;
1012 geo
->cylinders
= ioctl_geo
.cylinders
;
1016 #else /* __linux__ */
1017 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1023 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1027 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1035 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1037 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1040 if (s
->page_cache_inconsistent
) {
1044 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1046 /* There is no clear definition of the semantics of a failing fsync(),
1047 * so we may have to assume the worst. The sad truth is that this
1048 * assumption is correct for Linux. Some pages are now probably marked
1049 * clean in the page cache even though they are inconsistent with the
1050 * on-disk contents. The next fdatasync() call would succeed, but no
1051 * further writeback attempt will be made. We can't get back to a state
1052 * in which we know what is on disk (we would have to rewrite
1053 * everything that was touched since the last fdatasync() at least), so
1054 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1055 * really defined, I have little hope that other OSes are doing better.
1057 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1059 if ((s
->open_flags
& O_DIRECT
) == 0) {
1060 s
->page_cache_inconsistent
= true;
1067 #ifdef CONFIG_PREADV
1069 static bool preadv_present
= true;
1072 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1074 return preadv(fd
, iov
, nr_iov
, offset
);
1078 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1080 return pwritev(fd
, iov
, nr_iov
, offset
);
1085 static bool preadv_present
= false;
1088 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1094 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1101 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1106 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1107 len
= qemu_pwritev(aiocb
->aio_fildes
,
1112 len
= qemu_preadv(aiocb
->aio_fildes
,
1116 } while (len
== -1 && errno
== EINTR
);
1125 * Read/writes the data to/from a given linear buffer.
1127 * Returns the number of bytes handles or -errno in case of an error. Short
1128 * reads are only returned if the end of the file is reached.
1130 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1135 while (offset
< aiocb
->aio_nbytes
) {
1136 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1137 len
= pwrite(aiocb
->aio_fildes
,
1138 (const char *)buf
+ offset
,
1139 aiocb
->aio_nbytes
- offset
,
1140 aiocb
->aio_offset
+ offset
);
1142 len
= pread(aiocb
->aio_fildes
,
1144 aiocb
->aio_nbytes
- offset
,
1145 aiocb
->aio_offset
+ offset
);
1147 if (len
== -1 && errno
== EINTR
) {
1149 } else if (len
== -1 && errno
== EINVAL
&&
1150 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1151 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1153 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1154 * after a short read. Assume that O_DIRECT short reads only occur
1155 * at EOF. Therefore this is a short read, not an I/O error.
1158 } else if (len
== -1) {
1161 } else if (len
== 0) {
1170 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1175 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1177 * If there is just a single buffer, and it is properly aligned
1178 * we can just use plain pread/pwrite without any problems.
1180 if (aiocb
->aio_niov
== 1) {
1181 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1184 * We have more than one iovec, and all are properly aligned.
1186 * Try preadv/pwritev first and fall back to linearizing the
1187 * buffer if it's not supported.
1189 if (preadv_present
) {
1190 nbytes
= handle_aiocb_rw_vector(aiocb
);
1191 if (nbytes
== aiocb
->aio_nbytes
||
1192 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1195 preadv_present
= false;
1199 * XXX(hch): short read/write. no easy way to handle the reminder
1200 * using these interfaces. For now retry using plain
1206 * Ok, we have to do it the hard way, copy all segments into
1207 * a single aligned buffer.
1209 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1214 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1218 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1219 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1220 p
+= aiocb
->aio_iov
[i
].iov_len
;
1222 assert(p
- buf
== aiocb
->aio_nbytes
);
1225 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1226 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1228 size_t count
= aiocb
->aio_nbytes
, copy
;
1231 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1233 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1234 copy
= aiocb
->aio_iov
[i
].iov_len
;
1236 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1237 assert(count
>= copy
);
1249 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1251 struct xfs_flock64 fl
;
1254 memset(&fl
, 0, sizeof(fl
));
1255 fl
.l_whence
= SEEK_SET
;
1256 fl
.l_start
= offset
;
1259 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1261 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1268 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1270 struct xfs_flock64 fl
;
1273 memset(&fl
, 0, sizeof(fl
));
1274 fl
.l_whence
= SEEK_SET
;
1275 fl
.l_start
= offset
;
1278 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1280 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1288 static int translate_err(int err
)
1290 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1297 #ifdef CONFIG_FALLOCATE
1298 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1301 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1304 } while (errno
== EINTR
);
1305 return translate_err(-errno
);
1309 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1312 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1314 if (!s
->has_write_zeroes
) {
1320 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1321 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1324 } while (errno
== EINTR
);
1326 ret
= translate_err(-errno
);
1329 if (ret
== -ENOTSUP
) {
1330 s
->has_write_zeroes
= false;
1335 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1337 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1338 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1340 #ifdef CONFIG_FALLOCATE
1344 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1345 return handle_aiocb_write_zeroes_block(aiocb
);
1350 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1354 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1355 if (s
->has_write_zeroes
) {
1356 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1357 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1358 if (ret
== 0 || ret
!= -ENOTSUP
) {
1361 s
->has_write_zeroes
= false;
1365 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1366 if (s
->has_discard
&& s
->has_fallocate
) {
1367 int ret
= do_fallocate(s
->fd
,
1368 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1369 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1371 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1372 if (ret
== 0 || ret
!= -ENOTSUP
) {
1375 s
->has_fallocate
= false;
1376 } else if (ret
!= -ENOTSUP
) {
1379 s
->has_discard
= false;
1384 #ifdef CONFIG_FALLOCATE
1385 /* Last resort: we are trying to extend the file with zeroed data. This
1386 * can be done via fallocate(fd, 0) */
1387 len
= bdrv_getlength(aiocb
->bs
);
1388 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1389 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1390 if (ret
== 0 || ret
!= -ENOTSUP
) {
1393 s
->has_fallocate
= false;
1400 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1402 int ret
= -EOPNOTSUPP
;
1403 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1405 if (!s
->has_discard
) {
1409 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1412 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1413 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1416 } while (errno
== EINTR
);
1423 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1427 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1428 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1429 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1433 ret
= translate_err(ret
);
1434 if (ret
== -ENOTSUP
) {
1435 s
->has_discard
= false;
1440 static int aio_worker(void *arg
)
1442 RawPosixAIOData
*aiocb
= arg
;
1445 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1447 ret
= handle_aiocb_rw(aiocb
);
1448 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1449 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1450 0, aiocb
->aio_nbytes
- ret
);
1452 ret
= aiocb
->aio_nbytes
;
1454 if (ret
== aiocb
->aio_nbytes
) {
1456 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1460 case QEMU_AIO_WRITE
:
1461 ret
= handle_aiocb_rw(aiocb
);
1462 if (ret
== aiocb
->aio_nbytes
) {
1464 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1468 case QEMU_AIO_FLUSH
:
1469 ret
= handle_aiocb_flush(aiocb
);
1471 case QEMU_AIO_IOCTL
:
1472 ret
= handle_aiocb_ioctl(aiocb
);
1474 case QEMU_AIO_DISCARD
:
1475 ret
= handle_aiocb_discard(aiocb
);
1477 case QEMU_AIO_WRITE_ZEROES
:
1478 ret
= handle_aiocb_write_zeroes(aiocb
);
1481 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1490 static int paio_submit_co(BlockDriverState
*bs
, int fd
,
1491 int64_t offset
, QEMUIOVector
*qiov
,
1492 int bytes
, int type
)
1494 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1498 acb
->aio_type
= type
;
1499 acb
->aio_fildes
= fd
;
1501 acb
->aio_nbytes
= bytes
;
1502 acb
->aio_offset
= offset
;
1505 acb
->aio_iov
= qiov
->iov
;
1506 acb
->aio_niov
= qiov
->niov
;
1507 assert(qiov
->size
== bytes
);
1510 trace_paio_submit_co(offset
, bytes
, type
);
1511 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1512 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1515 static BlockAIOCB
*paio_submit(BlockDriverState
*bs
, int fd
,
1516 int64_t offset
, QEMUIOVector
*qiov
, int bytes
,
1517 BlockCompletionFunc
*cb
, void *opaque
, int type
)
1519 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1523 acb
->aio_type
= type
;
1524 acb
->aio_fildes
= fd
;
1526 acb
->aio_nbytes
= bytes
;
1527 acb
->aio_offset
= offset
;
1530 acb
->aio_iov
= qiov
->iov
;
1531 acb
->aio_niov
= qiov
->niov
;
1532 assert(qiov
->size
== acb
->aio_nbytes
);
1535 trace_paio_submit(acb
, opaque
, offset
, bytes
, type
);
1536 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1537 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1540 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1541 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1543 BDRVRawState
*s
= bs
->opaque
;
1545 if (fd_open(bs
) < 0)
1549 * Check if the underlying device requires requests to be aligned,
1550 * and if the request we are trying to submit is aligned or not.
1551 * If this is the case tell the low-level driver that it needs
1552 * to copy the buffer.
1554 if (s
->needs_alignment
) {
1555 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1556 type
|= QEMU_AIO_MISALIGNED
;
1557 #ifdef CONFIG_LINUX_AIO
1558 } else if (s
->use_linux_aio
) {
1559 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1560 assert(qiov
->size
== bytes
);
1561 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1566 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1569 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1570 uint64_t bytes
, QEMUIOVector
*qiov
,
1573 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1576 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1577 uint64_t bytes
, QEMUIOVector
*qiov
,
1581 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1584 static void raw_aio_plug(BlockDriverState
*bs
)
1586 #ifdef CONFIG_LINUX_AIO
1587 BDRVRawState
*s
= bs
->opaque
;
1588 if (s
->use_linux_aio
) {
1589 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1590 laio_io_plug(bs
, aio
);
1595 static void raw_aio_unplug(BlockDriverState
*bs
)
1597 #ifdef CONFIG_LINUX_AIO
1598 BDRVRawState
*s
= bs
->opaque
;
1599 if (s
->use_linux_aio
) {
1600 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1601 laio_io_unplug(bs
, aio
);
1606 static BlockAIOCB
*raw_aio_flush(BlockDriverState
*bs
,
1607 BlockCompletionFunc
*cb
, void *opaque
)
1609 BDRVRawState
*s
= bs
->opaque
;
1611 if (fd_open(bs
) < 0)
1614 return paio_submit(bs
, s
->fd
, 0, NULL
, 0, cb
, opaque
, QEMU_AIO_FLUSH
);
1617 static void raw_close(BlockDriverState
*bs
)
1619 BDRVRawState
*s
= bs
->opaque
;
1625 if (s
->lock_fd
>= 0) {
1626 qemu_close(s
->lock_fd
);
1632 * Truncates the given regular file @fd to @offset and, when growing, fills the
1633 * new space according to @prealloc.
1635 * Returns: 0 on success, -errno on failure.
1637 static int raw_regular_truncate(int fd
, int64_t offset
, PreallocMode prealloc
,
1641 int64_t current_length
= 0;
1645 if (fstat(fd
, &st
) < 0) {
1647 error_setg_errno(errp
, -result
, "Could not stat file");
1651 current_length
= st
.st_size
;
1652 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1653 error_setg(errp
, "Cannot use preallocation for shrinking files");
1658 #ifdef CONFIG_POSIX_FALLOCATE
1659 case PREALLOC_MODE_FALLOC
:
1661 * Truncating before posix_fallocate() makes it about twice slower on
1662 * file systems that do not support fallocate(), trying to check if a
1663 * block is allocated before allocating it, so don't do that here.
1665 result
= -posix_fallocate(fd
, current_length
, offset
- current_length
);
1667 /* posix_fallocate() doesn't set errno. */
1668 error_setg_errno(errp
, -result
,
1669 "Could not preallocate new data");
1673 case PREALLOC_MODE_FULL
:
1675 int64_t num
= 0, left
= offset
- current_length
;
1678 * Knowing the final size from the beginning could allow the file
1679 * system driver to do less allocations and possibly avoid
1680 * fragmentation of the file.
1682 if (ftruncate(fd
, offset
) != 0) {
1684 error_setg_errno(errp
, -result
, "Could not resize file");
1688 buf
= g_malloc0(65536);
1690 result
= lseek(fd
, current_length
, SEEK_SET
);
1693 error_setg_errno(errp
, -result
,
1694 "Failed to seek to the old end of file");
1699 num
= MIN(left
, 65536);
1700 result
= write(fd
, buf
, num
);
1703 error_setg_errno(errp
, -result
,
1704 "Could not write zeros for preallocation");
1713 error_setg_errno(errp
, -result
,
1714 "Could not flush file to disk");
1720 case PREALLOC_MODE_OFF
:
1721 if (ftruncate(fd
, offset
) != 0) {
1723 error_setg_errno(errp
, -result
, "Could not resize file");
1728 error_setg(errp
, "Unsupported preallocation mode: %s",
1729 PreallocMode_str(prealloc
));
1735 if (ftruncate(fd
, current_length
) < 0) {
1736 error_report("Failed to restore old file length: %s",
1745 static int raw_truncate(BlockDriverState
*bs
, int64_t offset
,
1746 PreallocMode prealloc
, Error
**errp
)
1748 BDRVRawState
*s
= bs
->opaque
;
1752 if (fstat(s
->fd
, &st
)) {
1754 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1758 if (S_ISREG(st
.st_mode
)) {
1759 return raw_regular_truncate(s
->fd
, offset
, prealloc
, errp
);
1762 if (prealloc
!= PREALLOC_MODE_OFF
) {
1763 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1764 "non-regular file", PreallocMode_str(prealloc
));
1768 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1769 if (offset
> raw_getlength(bs
)) {
1770 error_setg(errp
, "Cannot grow device files");
1774 error_setg(errp
, "Resizing this file is not supported");
1782 static int64_t raw_getlength(BlockDriverState
*bs
)
1784 BDRVRawState
*s
= bs
->opaque
;
1790 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1791 struct disklabel dl
;
1793 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1795 return (uint64_t)dl
.d_secsize
*
1796 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1800 #elif defined(__NetBSD__)
1801 static int64_t raw_getlength(BlockDriverState
*bs
)
1803 BDRVRawState
*s
= bs
->opaque
;
1809 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1810 struct dkwedge_info dkw
;
1812 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1813 return dkw
.dkw_size
* 512;
1815 struct disklabel dl
;
1817 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1819 return (uint64_t)dl
.d_secsize
*
1820 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1825 #elif defined(__sun__)
1826 static int64_t raw_getlength(BlockDriverState
*bs
)
1828 BDRVRawState
*s
= bs
->opaque
;
1829 struct dk_minfo minfo
;
1839 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1841 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
1843 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
1847 * There are reports that lseek on some devices fails, but
1848 * irc discussion said that contingency on contingency was overkill.
1850 size
= lseek(s
->fd
, 0, SEEK_END
);
1856 #elif defined(CONFIG_BSD)
1857 static int64_t raw_getlength(BlockDriverState
*bs
)
1859 BDRVRawState
*s
= bs
->opaque
;
1863 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1872 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1875 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
1876 #ifdef DIOCGMEDIASIZE
1877 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
1878 #elif defined(DIOCGPART)
1881 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
1882 size
= pi
.media_size
;
1888 #if defined(__APPLE__) && defined(__MACH__)
1890 uint64_t sectors
= 0;
1891 uint32_t sector_size
= 0;
1893 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
1894 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
1895 size
= sectors
* sector_size
;
1897 size
= lseek(fd
, 0LL, SEEK_END
);
1904 size
= lseek(fd
, 0LL, SEEK_END
);
1909 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1912 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1913 if (size
== 2048LL * (unsigned)-1)
1915 /* XXX no disc? maybe we need to reopen... */
1916 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
1923 size
= lseek(fd
, 0, SEEK_END
);
1931 static int64_t raw_getlength(BlockDriverState
*bs
)
1933 BDRVRawState
*s
= bs
->opaque
;
1942 size
= lseek(s
->fd
, 0, SEEK_END
);
1950 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
1953 BDRVRawState
*s
= bs
->opaque
;
1955 if (fstat(s
->fd
, &st
) < 0) {
1958 return (int64_t)st
.st_blocks
* 512;
1961 static int raw_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1965 int64_t total_size
= 0;
1967 PreallocMode prealloc
;
1969 Error
*local_err
= NULL
;
1971 strstart(filename
, "file:", &filename
);
1973 /* Read out options */
1974 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1976 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
1977 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
1978 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
1979 PREALLOC_MODE_OFF
, &local_err
);
1982 error_propagate(errp
, local_err
);
1987 fd
= qemu_open(filename
, O_RDWR
| O_CREAT
| O_TRUNC
| O_BINARY
,
1991 error_setg_errno(errp
, -result
, "Could not create file");
1997 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1998 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1999 * will be ignored since any failure of this operation should not
2000 * block the left work.
2003 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2004 attr
|= FS_NOCOW_FL
;
2005 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2010 result
= raw_regular_truncate(fd
, total_size
, prealloc
, errp
);
2016 if (qemu_close(fd
) != 0 && result
== 0) {
2018 error_setg_errno(errp
, -result
, "Could not close the new file");
2025 * Find allocation range in @bs around offset @start.
2026 * May change underlying file descriptor's file offset.
2027 * If @start is not in a hole, store @start in @data, and the
2028 * beginning of the next hole in @hole, and return 0.
2029 * If @start is in a non-trailing hole, store @start in @hole and the
2030 * beginning of the next non-hole in @data, and return 0.
2031 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2032 * If we can't find out, return a negative errno other than -ENXIO.
2034 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2035 off_t
*data
, off_t
*hole
)
2037 #if defined SEEK_HOLE && defined SEEK_DATA
2038 BDRVRawState
*s
= bs
->opaque
;
2043 * D1. offs == start: start is in data
2044 * D2. offs > start: start is in a hole, next data at offs
2045 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2046 * or start is beyond EOF
2047 * If the latter happens, the file has been truncated behind
2048 * our back since we opened it. All bets are off then.
2049 * Treating like a trailing hole is simplest.
2050 * D4. offs < 0, errno != ENXIO: we learned nothing
2052 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2054 return -errno
; /* D3 or D4 */
2056 assert(offs
>= start
);
2059 /* D2: in hole, next data at offs */
2065 /* D1: in data, end not yet known */
2069 * H1. offs == start: start is in a hole
2070 * If this happens here, a hole has been dug behind our back
2071 * since the previous lseek().
2072 * H2. offs > start: either start is in data, next hole at offs,
2073 * or start is in trailing hole, EOF at offs
2074 * Linux treats trailing holes like any other hole: offs ==
2075 * start. Solaris seeks to EOF instead: offs > start (blech).
2076 * If that happens here, a hole has been dug behind our back
2077 * since the previous lseek().
2078 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2079 * If this happens, the file has been truncated behind our
2080 * back since we opened it. Treat it like a trailing hole.
2081 * H4. offs < 0, errno != ENXIO: we learned nothing
2082 * Pretend we know nothing at all, i.e. "forget" about D1.
2084 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2086 return -errno
; /* D1 and (H3 or H4) */
2088 assert(offs
>= start
);
2092 * D1 and H2: either in data, next hole at offs, or it was in
2093 * data but is now in a trailing hole. In the latter case,
2094 * all bets are off. Treating it as if it there was data all
2095 * the way to EOF is safe, so simply do that.
2110 * Returns the allocation status of the specified sectors.
2112 * If 'sector_num' is beyond the end of the disk image the return value is 0
2113 * and 'pnum' is set to 0.
2115 * 'pnum' is set to the number of sectors (including and immediately following
2116 * the specified sector) that are known to be in the same
2117 * allocated/unallocated state.
2119 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2120 * beyond the end of the disk image it will be clamped.
2122 static int64_t coroutine_fn
raw_co_get_block_status(BlockDriverState
*bs
,
2124 int nb_sectors
, int *pnum
,
2125 BlockDriverState
**file
)
2127 off_t start
, data
= 0, hole
= 0;
2136 start
= sector_num
* BDRV_SECTOR_SIZE
;
2137 total_size
= bdrv_getlength(bs
);
2138 if (total_size
< 0) {
2140 } else if (start
>= total_size
) {
2143 } else if (start
+ nb_sectors
* BDRV_SECTOR_SIZE
> total_size
) {
2144 nb_sectors
= DIV_ROUND_UP(total_size
- start
, BDRV_SECTOR_SIZE
);
2147 ret
= find_allocation(bs
, start
, &data
, &hole
);
2148 if (ret
== -ENXIO
) {
2151 ret
= BDRV_BLOCK_ZERO
;
2152 } else if (ret
< 0) {
2153 /* No info available, so pretend there are no holes */
2155 ret
= BDRV_BLOCK_DATA
;
2156 } else if (data
== start
) {
2157 /* On a data extent, compute sectors to the end of the extent,
2158 * possibly including a partial sector at EOF. */
2159 *pnum
= MIN(nb_sectors
, DIV_ROUND_UP(hole
- start
, BDRV_SECTOR_SIZE
));
2160 ret
= BDRV_BLOCK_DATA
;
2162 /* On a hole, compute sectors to the beginning of the next extent. */
2163 assert(hole
== start
);
2164 *pnum
= MIN(nb_sectors
, (data
- start
) / BDRV_SECTOR_SIZE
);
2165 ret
= BDRV_BLOCK_ZERO
;
2168 return ret
| BDRV_BLOCK_OFFSET_VALID
| start
;
2171 static coroutine_fn BlockAIOCB
*raw_aio_pdiscard(BlockDriverState
*bs
,
2172 int64_t offset
, int bytes
,
2173 BlockCompletionFunc
*cb
, void *opaque
)
2175 BDRVRawState
*s
= bs
->opaque
;
2177 return paio_submit(bs
, s
->fd
, offset
, NULL
, bytes
,
2178 cb
, opaque
, QEMU_AIO_DISCARD
);
2181 static int coroutine_fn
raw_co_pwrite_zeroes(
2182 BlockDriverState
*bs
, int64_t offset
,
2183 int bytes
, BdrvRequestFlags flags
)
2185 BDRVRawState
*s
= bs
->opaque
;
2187 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
2188 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2189 QEMU_AIO_WRITE_ZEROES
);
2190 } else if (s
->discard_zeroes
) {
2191 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2197 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2199 BDRVRawState
*s
= bs
->opaque
;
2201 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2202 bdi
->can_write_zeroes_with_unmap
= s
->discard_zeroes
;
2206 static QemuOptsList raw_create_opts
= {
2207 .name
= "raw-create-opts",
2208 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2211 .name
= BLOCK_OPT_SIZE
,
2212 .type
= QEMU_OPT_SIZE
,
2213 .help
= "Virtual disk size"
2216 .name
= BLOCK_OPT_NOCOW
,
2217 .type
= QEMU_OPT_BOOL
,
2218 .help
= "Turn off copy-on-write (valid only on btrfs)"
2221 .name
= BLOCK_OPT_PREALLOC
,
2222 .type
= QEMU_OPT_STRING
,
2223 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2225 { /* end of list */ }
2229 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2232 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2235 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2237 BDRVRawState
*s
= bs
->opaque
;
2238 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2240 s
->shared_perm
= shared
;
2243 static void raw_abort_perm_update(BlockDriverState
*bs
)
2245 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2248 BlockDriver bdrv_file
= {
2249 .format_name
= "file",
2250 .protocol_name
= "file",
2251 .instance_size
= sizeof(BDRVRawState
),
2252 .bdrv_needs_filename
= true,
2253 .bdrv_probe
= NULL
, /* no probe for protocols */
2254 .bdrv_parse_filename
= raw_parse_filename
,
2255 .bdrv_file_open
= raw_open
,
2256 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2257 .bdrv_reopen_commit
= raw_reopen_commit
,
2258 .bdrv_reopen_abort
= raw_reopen_abort
,
2259 .bdrv_close
= raw_close
,
2260 .bdrv_create
= raw_create
,
2261 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2262 .bdrv_co_get_block_status
= raw_co_get_block_status
,
2263 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2265 .bdrv_co_preadv
= raw_co_preadv
,
2266 .bdrv_co_pwritev
= raw_co_pwritev
,
2267 .bdrv_aio_flush
= raw_aio_flush
,
2268 .bdrv_aio_pdiscard
= raw_aio_pdiscard
,
2269 .bdrv_refresh_limits
= raw_refresh_limits
,
2270 .bdrv_io_plug
= raw_aio_plug
,
2271 .bdrv_io_unplug
= raw_aio_unplug
,
2273 .bdrv_truncate
= raw_truncate
,
2274 .bdrv_getlength
= raw_getlength
,
2275 .bdrv_get_info
= raw_get_info
,
2276 .bdrv_get_allocated_file_size
2277 = raw_get_allocated_file_size
,
2278 .bdrv_check_perm
= raw_check_perm
,
2279 .bdrv_set_perm
= raw_set_perm
,
2280 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2281 .create_opts
= &raw_create_opts
,
2284 /***********************************************/
2287 #if defined(__APPLE__) && defined(__MACH__)
2288 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2289 CFIndex maxPathSize
, int flags
);
2290 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2292 kern_return_t kernResult
= KERN_FAILURE
;
2293 mach_port_t masterPort
;
2294 CFMutableDictionaryRef classesToMatch
;
2295 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2296 char *mediaType
= NULL
;
2298 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2299 if ( KERN_SUCCESS
!= kernResult
) {
2300 printf( "IOMasterPort returned %d\n", kernResult
);
2304 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2305 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2306 if (classesToMatch
== NULL
) {
2307 error_report("IOServiceMatching returned NULL for %s",
2308 matching_array
[index
]);
2311 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2313 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2315 if (kernResult
!= KERN_SUCCESS
) {
2316 error_report("Note: IOServiceGetMatchingServices returned %d",
2321 /* If a match was found, leave the loop */
2322 if (*mediaIterator
!= 0) {
2323 DPRINTF("Matching using %s\n", matching_array
[index
]);
2324 mediaType
= g_strdup(matching_array
[index
]);
2331 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2332 CFIndex maxPathSize
, int flags
)
2334 io_object_t nextMedia
;
2335 kern_return_t kernResult
= KERN_FAILURE
;
2337 nextMedia
= IOIteratorNext( mediaIterator
);
2340 CFTypeRef bsdPathAsCFString
;
2341 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2342 if ( bsdPathAsCFString
) {
2343 size_t devPathLength
;
2344 strcpy( bsdPath
, _PATH_DEV
);
2345 if (flags
& BDRV_O_NOCACHE
) {
2346 strcat(bsdPath
, "r");
2348 devPathLength
= strlen( bsdPath
);
2349 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2350 kernResult
= KERN_SUCCESS
;
2352 CFRelease( bsdPathAsCFString
);
2354 IOObjectRelease( nextMedia
);
2360 /* Sets up a real cdrom for use in QEMU */
2361 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2363 int index
, num_of_test_partitions
= 2, fd
;
2364 char test_partition
[MAXPATHLEN
];
2365 bool partition_found
= false;
2367 /* look for a working partition */
2368 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2369 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2371 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2373 partition_found
= true;
2379 /* if a working partition on the device was not found */
2380 if (partition_found
== false) {
2381 error_setg(errp
, "Failed to find a working partition on disc");
2383 DPRINTF("Using %s as optical disc\n", test_partition
);
2384 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2386 return partition_found
;
2389 /* Prints directions on mounting and unmounting a device */
2390 static void print_unmounting_directions(const char *file_name
)
2392 error_report("If device %s is mounted on the desktop, unmount"
2393 " it first before using it in QEMU", file_name
);
2394 error_report("Command to unmount device: diskutil unmountDisk %s",
2396 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2399 #endif /* defined(__APPLE__) && defined(__MACH__) */
2401 static int hdev_probe_device(const char *filename
)
2405 /* allow a dedicated CD-ROM driver to match with a higher priority */
2406 if (strstart(filename
, "/dev/cdrom", NULL
))
2409 if (stat(filename
, &st
) >= 0 &&
2410 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2417 static int check_hdev_writable(BDRVRawState
*s
)
2419 #if defined(BLKROGET)
2420 /* Linux block devices can be configured "read-only" using blockdev(8).
2421 * This is independent of device node permissions and therefore open(2)
2422 * with O_RDWR succeeds. Actual writes fail with EPERM.
2424 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2425 * check for read-only block devices so that Linux block devices behave
2431 if (fstat(s
->fd
, &st
)) {
2435 if (!S_ISBLK(st
.st_mode
)) {
2439 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2446 #endif /* defined(BLKROGET) */
2450 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2453 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2456 static bool hdev_is_sg(BlockDriverState
*bs
)
2459 #if defined(__linux__)
2461 BDRVRawState
*s
= bs
->opaque
;
2463 struct sg_scsi_id scsiid
;
2467 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2471 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2476 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2478 DPRINTF("SG device found: type=%d, version=%d\n",
2479 scsiid
.scsi_type
, sg_version
);
2488 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2491 BDRVRawState
*s
= bs
->opaque
;
2492 Error
*local_err
= NULL
;
2495 #if defined(__APPLE__) && defined(__MACH__)
2497 * Caution: while qdict_get_str() is fine, getting non-string types
2498 * would require more care. When @options come from -blockdev or
2499 * blockdev_add, its members are typed according to the QAPI
2500 * schema, but when they come from -drive, they're all QString.
2502 const char *filename
= qdict_get_str(options
, "filename");
2503 char bsd_path
[MAXPATHLEN
] = "";
2504 bool error_occurred
= false;
2506 /* If using a real cdrom */
2507 if (strcmp(filename
, "/dev/cdrom") == 0) {
2508 char *mediaType
= NULL
;
2509 kern_return_t ret_val
;
2510 io_iterator_t mediaIterator
= 0;
2512 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
2513 if (mediaType
== NULL
) {
2514 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
2516 error_occurred
= true;
2517 goto hdev_open_Mac_error
;
2520 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
2521 if (ret_val
!= KERN_SUCCESS
) {
2522 error_setg(errp
, "Could not get BSD path for optical drive");
2523 error_occurred
= true;
2524 goto hdev_open_Mac_error
;
2527 /* If a real optical drive was not found */
2528 if (bsd_path
[0] == '\0') {
2529 error_setg(errp
, "Failed to obtain bsd path for optical drive");
2530 error_occurred
= true;
2531 goto hdev_open_Mac_error
;
2534 /* If using a cdrom disc and finding a partition on the disc failed */
2535 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
2536 setup_cdrom(bsd_path
, errp
) == false) {
2537 print_unmounting_directions(bsd_path
);
2538 error_occurred
= true;
2539 goto hdev_open_Mac_error
;
2542 qdict_put_str(options
, "filename", bsd_path
);
2544 hdev_open_Mac_error
:
2546 if (mediaIterator
) {
2547 IOObjectRelease(mediaIterator
);
2549 if (error_occurred
) {
2553 #endif /* defined(__APPLE__) && defined(__MACH__) */
2555 s
->type
= FTYPE_FILE
;
2557 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2559 error_propagate(errp
, local_err
);
2560 #if defined(__APPLE__) && defined(__MACH__)
2562 filename
= bsd_path
;
2564 /* if a physical device experienced an error while being opened */
2565 if (strncmp(filename
, "/dev/", 5) == 0) {
2566 print_unmounting_directions(filename
);
2568 #endif /* defined(__APPLE__) && defined(__MACH__) */
2572 /* Since this does ioctl the device must be already opened */
2573 bs
->sg
= hdev_is_sg(bs
);
2575 if (flags
& BDRV_O_RDWR
) {
2576 ret
= check_hdev_writable(s
);
2579 error_setg_errno(errp
, -ret
, "The device is not writable");
2587 #if defined(__linux__)
2589 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
2590 unsigned long int req
, void *buf
,
2591 BlockCompletionFunc
*cb
, void *opaque
)
2593 BDRVRawState
*s
= bs
->opaque
;
2594 RawPosixAIOData
*acb
;
2597 if (fd_open(bs
) < 0)
2600 acb
= g_new(RawPosixAIOData
, 1);
2602 acb
->aio_type
= QEMU_AIO_IOCTL
;
2603 acb
->aio_fildes
= s
->fd
;
2604 acb
->aio_offset
= 0;
2605 acb
->aio_ioctl_buf
= buf
;
2606 acb
->aio_ioctl_cmd
= req
;
2607 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
2608 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
2612 static int fd_open(BlockDriverState
*bs
)
2614 BDRVRawState
*s
= bs
->opaque
;
2616 /* this is just to ensure s->fd is sane (its called by io ops) */
2622 static coroutine_fn BlockAIOCB
*hdev_aio_pdiscard(BlockDriverState
*bs
,
2623 int64_t offset
, int bytes
,
2624 BlockCompletionFunc
*cb
, void *opaque
)
2626 BDRVRawState
*s
= bs
->opaque
;
2628 if (fd_open(bs
) < 0) {
2631 return paio_submit(bs
, s
->fd
, offset
, NULL
, bytes
,
2632 cb
, opaque
, QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
2635 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
2636 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
2638 BDRVRawState
*s
= bs
->opaque
;
2645 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
2646 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2647 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
2648 } else if (s
->discard_zeroes
) {
2649 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2650 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
2655 static int hdev_create(const char *filename
, QemuOpts
*opts
,
2660 struct stat stat_buf
;
2661 int64_t total_size
= 0;
2664 /* This function is used by both protocol block drivers and therefore either
2665 * of these prefixes may be given.
2666 * The return value has to be stored somewhere, otherwise this is an error
2667 * due to -Werror=unused-value. */
2669 strstart(filename
, "host_device:", &filename
) ||
2670 strstart(filename
, "host_cdrom:" , &filename
);
2674 ret
= raw_normalize_devicepath(&filename
);
2676 error_setg_errno(errp
, -ret
, "Could not normalize device path");
2680 /* Read out options */
2681 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2684 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
2687 error_setg_errno(errp
, -ret
, "Could not open device");
2691 if (fstat(fd
, &stat_buf
) < 0) {
2693 error_setg_errno(errp
, -ret
, "Could not stat device");
2694 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
2696 "The given file is neither a block nor a character device");
2698 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
2699 error_setg(errp
, "Device is too small");
2707 static BlockDriver bdrv_host_device
= {
2708 .format_name
= "host_device",
2709 .protocol_name
= "host_device",
2710 .instance_size
= sizeof(BDRVRawState
),
2711 .bdrv_needs_filename
= true,
2712 .bdrv_probe_device
= hdev_probe_device
,
2713 .bdrv_parse_filename
= hdev_parse_filename
,
2714 .bdrv_file_open
= hdev_open
,
2715 .bdrv_close
= raw_close
,
2716 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2717 .bdrv_reopen_commit
= raw_reopen_commit
,
2718 .bdrv_reopen_abort
= raw_reopen_abort
,
2719 .bdrv_create
= hdev_create
,
2720 .create_opts
= &raw_create_opts
,
2721 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
2723 .bdrv_co_preadv
= raw_co_preadv
,
2724 .bdrv_co_pwritev
= raw_co_pwritev
,
2725 .bdrv_aio_flush
= raw_aio_flush
,
2726 .bdrv_aio_pdiscard
= hdev_aio_pdiscard
,
2727 .bdrv_refresh_limits
= raw_refresh_limits
,
2728 .bdrv_io_plug
= raw_aio_plug
,
2729 .bdrv_io_unplug
= raw_aio_unplug
,
2731 .bdrv_truncate
= raw_truncate
,
2732 .bdrv_getlength
= raw_getlength
,
2733 .bdrv_get_info
= raw_get_info
,
2734 .bdrv_get_allocated_file_size
2735 = raw_get_allocated_file_size
,
2736 .bdrv_check_perm
= raw_check_perm
,
2737 .bdrv_set_perm
= raw_set_perm
,
2738 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2739 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
2740 .bdrv_probe_geometry
= hdev_probe_geometry
,
2742 /* generic scsi device */
2744 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2748 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2749 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
2752 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
2757 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2760 BDRVRawState
*s
= bs
->opaque
;
2764 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2765 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, errp
);
2768 static int cdrom_probe_device(const char *filename
)
2774 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
2778 ret
= fstat(fd
, &st
);
2779 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2783 /* Attempt to detect via a CDROM specific ioctl */
2784 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2794 static bool cdrom_is_inserted(BlockDriverState
*bs
)
2796 BDRVRawState
*s
= bs
->opaque
;
2799 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2800 return ret
== CDS_DISC_OK
;
2803 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2805 BDRVRawState
*s
= bs
->opaque
;
2808 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
2809 perror("CDROMEJECT");
2811 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
2812 perror("CDROMEJECT");
2816 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2818 BDRVRawState
*s
= bs
->opaque
;
2820 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
2822 * Note: an error can happen if the distribution automatically
2825 /* perror("CDROM_LOCKDOOR"); */
2829 static BlockDriver bdrv_host_cdrom
= {
2830 .format_name
= "host_cdrom",
2831 .protocol_name
= "host_cdrom",
2832 .instance_size
= sizeof(BDRVRawState
),
2833 .bdrv_needs_filename
= true,
2834 .bdrv_probe_device
= cdrom_probe_device
,
2835 .bdrv_parse_filename
= cdrom_parse_filename
,
2836 .bdrv_file_open
= cdrom_open
,
2837 .bdrv_close
= raw_close
,
2838 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2839 .bdrv_reopen_commit
= raw_reopen_commit
,
2840 .bdrv_reopen_abort
= raw_reopen_abort
,
2841 .bdrv_create
= hdev_create
,
2842 .create_opts
= &raw_create_opts
,
2845 .bdrv_co_preadv
= raw_co_preadv
,
2846 .bdrv_co_pwritev
= raw_co_pwritev
,
2847 .bdrv_aio_flush
= raw_aio_flush
,
2848 .bdrv_refresh_limits
= raw_refresh_limits
,
2849 .bdrv_io_plug
= raw_aio_plug
,
2850 .bdrv_io_unplug
= raw_aio_unplug
,
2852 .bdrv_truncate
= raw_truncate
,
2853 .bdrv_getlength
= raw_getlength
,
2854 .has_variable_length
= true,
2855 .bdrv_get_allocated_file_size
2856 = raw_get_allocated_file_size
,
2858 /* removable device support */
2859 .bdrv_is_inserted
= cdrom_is_inserted
,
2860 .bdrv_eject
= cdrom_eject
,
2861 .bdrv_lock_medium
= cdrom_lock_medium
,
2863 /* generic scsi device */
2864 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2866 #endif /* __linux__ */
2868 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2869 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2872 BDRVRawState
*s
= bs
->opaque
;
2873 Error
*local_err
= NULL
;
2878 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2880 error_propagate(errp
, local_err
);
2884 /* make sure the door isn't locked at this time */
2885 ioctl(s
->fd
, CDIOCALLOW
);
2889 static int cdrom_probe_device(const char *filename
)
2891 if (strstart(filename
, "/dev/cd", NULL
) ||
2892 strstart(filename
, "/dev/acd", NULL
))
2897 static int cdrom_reopen(BlockDriverState
*bs
)
2899 BDRVRawState
*s
= bs
->opaque
;
2903 * Force reread of possibly changed/newly loaded disc,
2904 * FreeBSD seems to not notice sometimes...
2908 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
2915 /* make sure the door isn't locked at this time */
2916 ioctl(s
->fd
, CDIOCALLOW
);
2920 static bool cdrom_is_inserted(BlockDriverState
*bs
)
2922 return raw_getlength(bs
) > 0;
2925 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2927 BDRVRawState
*s
= bs
->opaque
;
2932 (void) ioctl(s
->fd
, CDIOCALLOW
);
2935 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
2936 perror("CDIOCEJECT");
2938 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
2939 perror("CDIOCCLOSE");
2945 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2947 BDRVRawState
*s
= bs
->opaque
;
2951 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
2953 * Note: an error can happen if the distribution automatically
2956 /* perror("CDROM_LOCKDOOR"); */
2960 static BlockDriver bdrv_host_cdrom
= {
2961 .format_name
= "host_cdrom",
2962 .protocol_name
= "host_cdrom",
2963 .instance_size
= sizeof(BDRVRawState
),
2964 .bdrv_needs_filename
= true,
2965 .bdrv_probe_device
= cdrom_probe_device
,
2966 .bdrv_parse_filename
= cdrom_parse_filename
,
2967 .bdrv_file_open
= cdrom_open
,
2968 .bdrv_close
= raw_close
,
2969 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2970 .bdrv_reopen_commit
= raw_reopen_commit
,
2971 .bdrv_reopen_abort
= raw_reopen_abort
,
2972 .bdrv_create
= hdev_create
,
2973 .create_opts
= &raw_create_opts
,
2975 .bdrv_co_preadv
= raw_co_preadv
,
2976 .bdrv_co_pwritev
= raw_co_pwritev
,
2977 .bdrv_aio_flush
= raw_aio_flush
,
2978 .bdrv_refresh_limits
= raw_refresh_limits
,
2979 .bdrv_io_plug
= raw_aio_plug
,
2980 .bdrv_io_unplug
= raw_aio_unplug
,
2982 .bdrv_truncate
= raw_truncate
,
2983 .bdrv_getlength
= raw_getlength
,
2984 .has_variable_length
= true,
2985 .bdrv_get_allocated_file_size
2986 = raw_get_allocated_file_size
,
2988 /* removable device support */
2989 .bdrv_is_inserted
= cdrom_is_inserted
,
2990 .bdrv_eject
= cdrom_eject
,
2991 .bdrv_lock_medium
= cdrom_lock_medium
,
2993 #endif /* __FreeBSD__ */
2995 static void bdrv_file_init(void)
2998 * Register all the drivers. Note that order is important, the driver
2999 * registered last will get probed first.
3001 bdrv_register(&bdrv_file
);
3002 bdrv_register(&bdrv_host_device
);
3004 bdrv_register(&bdrv_host_cdrom
);
3006 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3007 bdrv_register(&bdrv_host_cdrom
);
3011 block_init(bdrv_file_init
);