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-common.h"
25 #include "qemu/timer.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
30 #include "block/thread-pool.h"
34 #if defined(__APPLE__) && (__MACH__)
36 #include <sys/param.h>
37 #include <IOKit/IOKitLib.h>
38 #include <IOKit/IOBSD.h>
39 #include <IOKit/storage/IOMediaBSDClient.h>
40 #include <IOKit/storage/IOMedia.h>
41 #include <IOKit/storage/IOCDMedia.h>
42 //#include <IOKit/storage/IOCDTypes.h>
43 #include <CoreFoundation/CoreFoundation.h>
47 #define _POSIX_PTHREAD_SEMANTICS 1
51 #include <sys/types.h>
53 #include <sys/ioctl.h>
54 #include <sys/param.h>
55 #include <linux/cdrom.h>
59 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
63 #include <linux/fiemap.h>
65 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
66 #include <linux/falloc.h>
68 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
74 #include <sys/ioctl.h>
75 #include <sys/disklabel.h>
80 #include <sys/ioctl.h>
81 #include <sys/disklabel.h>
87 #include <sys/ioctl.h>
88 #include <sys/diskslice.h>
95 //#define DEBUG_FLOPPY
98 #if defined(DEBUG_BLOCK)
99 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
100 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
102 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
105 /* OS X does not have O_DSYNC */
108 #define O_DSYNC O_SYNC
109 #elif defined(O_FSYNC)
110 #define O_DSYNC O_FSYNC
114 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
116 #define O_DIRECT O_DSYNC
123 /* if the FD is not accessed during that time (in ns), we try to
124 reopen it to see if the disk has been changed */
125 #define FD_OPEN_TIMEOUT (1000000000)
127 #define MAX_BLOCKSIZE 4096
129 typedef struct BDRVRawState
{
135 #if defined(__linux__)
136 /* linux floppy specific */
137 int64_t fd_open_time
;
138 int64_t fd_error_time
;
140 int fd_media_changed
;
142 #ifdef CONFIG_LINUX_AIO
150 bool has_write_zeroes
:1;
151 bool discard_zeroes
:1;
157 typedef struct BDRVRawReopenState
{
160 #ifdef CONFIG_LINUX_AIO
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
)
224 static void raw_probe_alignment(BlockDriverState
*bs
)
226 BDRVRawState
*s
= bs
->opaque
;
228 unsigned int sector_size
;
230 /* For /dev/sg devices the alignment is not really used.
231 With buffered I/O, we don't have any restrictions. */
232 if (bs
->sg
|| !(s
->open_flags
& O_DIRECT
)) {
233 bs
->request_alignment
= 1;
238 /* Try a few ioctls to get the right size */
239 bs
->request_alignment
= 0;
243 if (ioctl(s
->fd
, BLKSSZGET
, §or_size
) >= 0) {
244 bs
->request_alignment
= sector_size
;
247 #ifdef DKIOCGETBLOCKSIZE
248 if (ioctl(s
->fd
, DKIOCGETBLOCKSIZE
, §or_size
) >= 0) {
249 bs
->request_alignment
= sector_size
;
252 #ifdef DIOCGSECTORSIZE
253 if (ioctl(s
->fd
, DIOCGSECTORSIZE
, §or_size
) >= 0) {
254 bs
->request_alignment
= sector_size
;
260 if (xfsctl(NULL
, s
->fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
261 bs
->request_alignment
= da
.d_miniosz
;
262 /* The kernel returns wrong information for d_mem */
263 /* s->buf_align = da.d_mem; */
268 /* If we could not get the sizes so far, we can only guess them */
271 buf
= qemu_memalign(MAX_BLOCKSIZE
, 2 * MAX_BLOCKSIZE
);
272 for (align
= 512; align
<= MAX_BLOCKSIZE
; align
<<= 1) {
273 if (pread(s
->fd
, buf
+ align
, MAX_BLOCKSIZE
, 0) >= 0) {
274 s
->buf_align
= align
;
281 if (!bs
->request_alignment
) {
283 buf
= qemu_memalign(s
->buf_align
, MAX_BLOCKSIZE
);
284 for (align
= 512; align
<= MAX_BLOCKSIZE
; align
<<= 1) {
285 if (pread(s
->fd
, buf
, align
, 0) >= 0) {
286 bs
->request_alignment
= align
;
294 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
296 assert(open_flags
!= NULL
);
298 *open_flags
|= O_BINARY
;
299 *open_flags
&= ~O_ACCMODE
;
300 if (bdrv_flags
& BDRV_O_RDWR
) {
301 *open_flags
|= O_RDWR
;
303 *open_flags
|= O_RDONLY
;
306 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
307 * and O_DIRECT for no caching. */
308 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
309 *open_flags
|= O_DIRECT
;
313 static void raw_detach_aio_context(BlockDriverState
*bs
)
315 #ifdef CONFIG_LINUX_AIO
316 BDRVRawState
*s
= bs
->opaque
;
319 laio_detach_aio_context(s
->aio_ctx
, bdrv_get_aio_context(bs
));
324 static void raw_attach_aio_context(BlockDriverState
*bs
,
325 AioContext
*new_context
)
327 #ifdef CONFIG_LINUX_AIO
328 BDRVRawState
*s
= bs
->opaque
;
331 laio_attach_aio_context(s
->aio_ctx
, new_context
);
336 #ifdef CONFIG_LINUX_AIO
337 static int raw_set_aio(void **aio_ctx
, int *use_aio
, int bdrv_flags
)
340 assert(aio_ctx
!= NULL
);
341 assert(use_aio
!= NULL
);
343 * Currently Linux do AIO only for files opened with O_DIRECT
344 * specified so check NOCACHE flag too
346 if ((bdrv_flags
& (BDRV_O_NOCACHE
|BDRV_O_NATIVE_AIO
)) ==
347 (BDRV_O_NOCACHE
|BDRV_O_NATIVE_AIO
)) {
349 /* if non-NULL, laio_init() has already been run */
350 if (*aio_ctx
== NULL
) {
351 *aio_ctx
= laio_init();
368 static void raw_parse_filename(const char *filename
, QDict
*options
,
371 /* The filename does not have to be prefixed by the protocol name, since
372 * "file" is the default protocol; therefore, the return value of this
373 * function call can be ignored. */
374 strstart(filename
, "file:", &filename
);
376 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
379 static QemuOptsList raw_runtime_opts
= {
381 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
385 .type
= QEMU_OPT_STRING
,
386 .help
= "File name of the image",
388 { /* end of list */ }
392 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
393 int bdrv_flags
, int open_flags
, Error
**errp
)
395 BDRVRawState
*s
= bs
->opaque
;
397 Error
*local_err
= NULL
;
398 const char *filename
= NULL
;
402 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
403 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
405 error_propagate(errp
, local_err
);
410 filename
= qemu_opt_get(opts
, "filename");
412 ret
= raw_normalize_devicepath(&filename
);
414 error_setg_errno(errp
, -ret
, "Could not normalize device path");
418 s
->open_flags
= open_flags
;
419 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
422 fd
= qemu_open(filename
, s
->open_flags
, 0644);
432 #ifdef CONFIG_LINUX_AIO
433 if (raw_set_aio(&s
->aio_ctx
, &s
->use_aio
, bdrv_flags
)) {
436 error_setg_errno(errp
, -ret
, "Could not set AIO state");
441 s
->has_discard
= true;
442 s
->has_write_zeroes
= true;
444 if (fstat(s
->fd
, &st
) < 0) {
445 error_setg_errno(errp
, errno
, "Could not stat file");
448 if (S_ISREG(st
.st_mode
)) {
449 s
->discard_zeroes
= true;
451 if (S_ISBLK(st
.st_mode
)) {
452 #ifdef BLKDISCARDZEROES
454 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
455 s
->discard_zeroes
= true;
459 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
460 * not rely on the contents of discarded blocks unless using O_DIRECT.
461 * Same for BLKZEROOUT.
463 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
464 s
->discard_zeroes
= false;
465 s
->has_write_zeroes
= false;
471 if (platform_test_xfs_fd(s
->fd
)) {
476 raw_attach_aio_context(bs
, bdrv_get_aio_context(bs
));
480 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
487 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
490 BDRVRawState
*s
= bs
->opaque
;
491 Error
*local_err
= NULL
;
494 s
->type
= FTYPE_FILE
;
495 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
497 error_propagate(errp
, local_err
);
502 static int raw_reopen_prepare(BDRVReopenState
*state
,
503 BlockReopenQueue
*queue
, Error
**errp
)
506 BDRVRawReopenState
*raw_s
;
509 assert(state
!= NULL
);
510 assert(state
->bs
!= NULL
);
512 s
= state
->bs
->opaque
;
514 state
->opaque
= g_malloc0(sizeof(BDRVRawReopenState
));
515 raw_s
= state
->opaque
;
517 #ifdef CONFIG_LINUX_AIO
518 raw_s
->use_aio
= s
->use_aio
;
520 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
521 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
522 * won't override aio_ctx if aio_ctx is non-NULL */
523 if (raw_set_aio(&s
->aio_ctx
, &raw_s
->use_aio
, state
->flags
)) {
524 error_setg(errp
, "Could not set AIO state");
529 if (s
->type
== FTYPE_FD
|| s
->type
== FTYPE_CD
) {
530 raw_s
->open_flags
|= O_NONBLOCK
;
533 raw_parse_flags(state
->flags
, &raw_s
->open_flags
);
537 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
539 fcntl_flags
|= O_NOATIME
;
543 /* Not all operating systems have O_ASYNC, and those that don't
544 * will not let us track the state into raw_s->open_flags (typically
545 * you achieve the same effect with an ioctl, for example I_SETSIG
546 * on Solaris). But we do not use O_ASYNC, so that's fine.
548 assert((s
->open_flags
& O_ASYNC
) == 0);
551 if ((raw_s
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
552 /* dup the original fd */
553 /* TODO: use qemu fcntl wrapper */
554 #ifdef F_DUPFD_CLOEXEC
555 raw_s
->fd
= fcntl(s
->fd
, F_DUPFD_CLOEXEC
, 0);
557 raw_s
->fd
= dup(s
->fd
);
558 if (raw_s
->fd
!= -1) {
559 qemu_set_cloexec(raw_s
->fd
);
562 if (raw_s
->fd
>= 0) {
563 ret
= fcntl_setfl(raw_s
->fd
, raw_s
->open_flags
);
565 qemu_close(raw_s
->fd
);
571 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
572 if (raw_s
->fd
== -1) {
573 assert(!(raw_s
->open_flags
& O_CREAT
));
574 raw_s
->fd
= qemu_open(state
->bs
->filename
, raw_s
->open_flags
);
575 if (raw_s
->fd
== -1) {
576 error_setg_errno(errp
, errno
, "Could not reopen file");
583 static void raw_reopen_commit(BDRVReopenState
*state
)
585 BDRVRawReopenState
*raw_s
= state
->opaque
;
586 BDRVRawState
*s
= state
->bs
->opaque
;
588 s
->open_flags
= raw_s
->open_flags
;
592 #ifdef CONFIG_LINUX_AIO
593 s
->use_aio
= raw_s
->use_aio
;
596 g_free(state
->opaque
);
597 state
->opaque
= NULL
;
601 static void raw_reopen_abort(BDRVReopenState
*state
)
603 BDRVRawReopenState
*raw_s
= state
->opaque
;
605 /* nothing to do if NULL, we didn't get far enough */
610 if (raw_s
->fd
>= 0) {
611 qemu_close(raw_s
->fd
);
614 g_free(state
->opaque
);
615 state
->opaque
= NULL
;
618 static int raw_refresh_limits(BlockDriverState
*bs
)
620 BDRVRawState
*s
= bs
->opaque
;
622 raw_probe_alignment(bs
);
623 bs
->bl
.opt_mem_alignment
= s
->buf_align
;
628 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
632 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
640 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
644 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
653 static bool preadv_present
= true;
656 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
658 return preadv(fd
, iov
, nr_iov
, offset
);
662 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
664 return pwritev(fd
, iov
, nr_iov
, offset
);
669 static bool preadv_present
= false;
672 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
678 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
685 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
690 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
691 len
= qemu_pwritev(aiocb
->aio_fildes
,
696 len
= qemu_preadv(aiocb
->aio_fildes
,
700 } while (len
== -1 && errno
== EINTR
);
709 * Read/writes the data to/from a given linear buffer.
711 * Returns the number of bytes handles or -errno in case of an error. Short
712 * reads are only returned if the end of the file is reached.
714 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
719 while (offset
< aiocb
->aio_nbytes
) {
720 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
721 len
= pwrite(aiocb
->aio_fildes
,
722 (const char *)buf
+ offset
,
723 aiocb
->aio_nbytes
- offset
,
724 aiocb
->aio_offset
+ offset
);
726 len
= pread(aiocb
->aio_fildes
,
728 aiocb
->aio_nbytes
- offset
,
729 aiocb
->aio_offset
+ offset
);
731 if (len
== -1 && errno
== EINTR
) {
733 } else if (len
== -1) {
736 } else if (len
== 0) {
745 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
750 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
752 * If there is just a single buffer, and it is properly aligned
753 * we can just use plain pread/pwrite without any problems.
755 if (aiocb
->aio_niov
== 1) {
756 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
759 * We have more than one iovec, and all are properly aligned.
761 * Try preadv/pwritev first and fall back to linearizing the
762 * buffer if it's not supported.
764 if (preadv_present
) {
765 nbytes
= handle_aiocb_rw_vector(aiocb
);
766 if (nbytes
== aiocb
->aio_nbytes
||
767 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
770 preadv_present
= false;
774 * XXX(hch): short read/write. no easy way to handle the reminder
775 * using these interfaces. For now retry using plain
781 * Ok, we have to do it the hard way, copy all segments into
782 * a single aligned buffer.
784 buf
= qemu_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
785 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
789 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
790 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
791 p
+= aiocb
->aio_iov
[i
].iov_len
;
795 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
796 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
798 size_t count
= aiocb
->aio_nbytes
, copy
;
801 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
803 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
804 copy
= aiocb
->aio_iov
[i
].iov_len
;
806 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
817 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
819 struct xfs_flock64 fl
;
821 memset(&fl
, 0, sizeof(fl
));
822 fl
.l_whence
= SEEK_SET
;
826 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
827 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno
));
834 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
836 struct xfs_flock64 fl
;
838 memset(&fl
, 0, sizeof(fl
));
839 fl
.l_whence
= SEEK_SET
;
843 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
844 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno
));
852 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
854 int ret
= -EOPNOTSUPP
;
855 BDRVRawState
*s
= aiocb
->bs
->opaque
;
857 if (s
->has_write_zeroes
== 0) {
861 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
864 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
865 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
868 } while (errno
== EINTR
);
875 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
880 if (ret
== -ENODEV
|| ret
== -ENOSYS
|| ret
== -EOPNOTSUPP
||
882 s
->has_write_zeroes
= false;
888 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
890 int ret
= -EOPNOTSUPP
;
891 BDRVRawState
*s
= aiocb
->bs
->opaque
;
893 if (!s
->has_discard
) {
897 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
900 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
901 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
904 } while (errno
== EINTR
);
911 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
915 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
917 if (fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
918 aiocb
->aio_offset
, aiocb
->aio_nbytes
) == 0) {
921 } while (errno
== EINTR
);
927 if (ret
== -ENODEV
|| ret
== -ENOSYS
|| ret
== -EOPNOTSUPP
||
929 s
->has_discard
= false;
935 static int aio_worker(void *arg
)
937 RawPosixAIOData
*aiocb
= arg
;
940 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
942 ret
= handle_aiocb_rw(aiocb
);
943 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
&& aiocb
->bs
->growable
) {
944 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
945 0, aiocb
->aio_nbytes
- ret
);
947 ret
= aiocb
->aio_nbytes
;
949 if (ret
== aiocb
->aio_nbytes
) {
951 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
956 ret
= handle_aiocb_rw(aiocb
);
957 if (ret
== aiocb
->aio_nbytes
) {
959 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
964 ret
= handle_aiocb_flush(aiocb
);
967 ret
= handle_aiocb_ioctl(aiocb
);
969 case QEMU_AIO_DISCARD
:
970 ret
= handle_aiocb_discard(aiocb
);
972 case QEMU_AIO_WRITE_ZEROES
:
973 ret
= handle_aiocb_write_zeroes(aiocb
);
976 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
981 g_slice_free(RawPosixAIOData
, aiocb
);
985 static int paio_submit_co(BlockDriverState
*bs
, int fd
,
986 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
989 RawPosixAIOData
*acb
= g_slice_new(RawPosixAIOData
);
993 acb
->aio_type
= type
;
994 acb
->aio_fildes
= fd
;
997 acb
->aio_iov
= qiov
->iov
;
998 acb
->aio_niov
= qiov
->niov
;
1000 acb
->aio_nbytes
= nb_sectors
* 512;
1001 acb
->aio_offset
= sector_num
* 512;
1003 trace_paio_submit_co(sector_num
, nb_sectors
, type
);
1004 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1005 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1008 static BlockDriverAIOCB
*paio_submit(BlockDriverState
*bs
, int fd
,
1009 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1010 BlockDriverCompletionFunc
*cb
, void *opaque
, int type
)
1012 RawPosixAIOData
*acb
= g_slice_new(RawPosixAIOData
);
1016 acb
->aio_type
= type
;
1017 acb
->aio_fildes
= fd
;
1020 acb
->aio_iov
= qiov
->iov
;
1021 acb
->aio_niov
= qiov
->niov
;
1023 acb
->aio_nbytes
= nb_sectors
* 512;
1024 acb
->aio_offset
= sector_num
* 512;
1026 trace_paio_submit(acb
, opaque
, sector_num
, nb_sectors
, type
);
1027 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1028 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1031 static BlockDriverAIOCB
*raw_aio_submit(BlockDriverState
*bs
,
1032 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1033 BlockDriverCompletionFunc
*cb
, void *opaque
, int type
)
1035 BDRVRawState
*s
= bs
->opaque
;
1037 if (fd_open(bs
) < 0)
1041 * If O_DIRECT is used the buffer needs to be aligned on a sector
1042 * boundary. Check if this is the case or tell the low-level
1043 * driver that it needs to copy the buffer.
1045 if ((bs
->open_flags
& BDRV_O_NOCACHE
)) {
1046 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1047 type
|= QEMU_AIO_MISALIGNED
;
1048 #ifdef CONFIG_LINUX_AIO
1049 } else if (s
->use_aio
) {
1050 return laio_submit(bs
, s
->aio_ctx
, s
->fd
, sector_num
, qiov
,
1051 nb_sectors
, cb
, opaque
, type
);
1056 return paio_submit(bs
, s
->fd
, sector_num
, qiov
, nb_sectors
,
1060 static BlockDriverAIOCB
*raw_aio_readv(BlockDriverState
*bs
,
1061 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1062 BlockDriverCompletionFunc
*cb
, void *opaque
)
1064 return raw_aio_submit(bs
, sector_num
, qiov
, nb_sectors
,
1065 cb
, opaque
, QEMU_AIO_READ
);
1068 static BlockDriverAIOCB
*raw_aio_writev(BlockDriverState
*bs
,
1069 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1070 BlockDriverCompletionFunc
*cb
, void *opaque
)
1072 return raw_aio_submit(bs
, sector_num
, qiov
, nb_sectors
,
1073 cb
, opaque
, QEMU_AIO_WRITE
);
1076 static BlockDriverAIOCB
*raw_aio_flush(BlockDriverState
*bs
,
1077 BlockDriverCompletionFunc
*cb
, void *opaque
)
1079 BDRVRawState
*s
= bs
->opaque
;
1081 if (fd_open(bs
) < 0)
1084 return paio_submit(bs
, s
->fd
, 0, NULL
, 0, cb
, opaque
, QEMU_AIO_FLUSH
);
1087 static void raw_close(BlockDriverState
*bs
)
1089 BDRVRawState
*s
= bs
->opaque
;
1091 raw_detach_aio_context(bs
);
1093 #ifdef CONFIG_LINUX_AIO
1095 laio_cleanup(s
->aio_ctx
);
1104 static int raw_truncate(BlockDriverState
*bs
, int64_t offset
)
1106 BDRVRawState
*s
= bs
->opaque
;
1109 if (fstat(s
->fd
, &st
)) {
1113 if (S_ISREG(st
.st_mode
)) {
1114 if (ftruncate(s
->fd
, offset
) < 0) {
1117 } else if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1118 if (offset
> raw_getlength(bs
)) {
1129 static int64_t raw_getlength(BlockDriverState
*bs
)
1131 BDRVRawState
*s
= bs
->opaque
;
1137 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1138 struct disklabel dl
;
1140 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1142 return (uint64_t)dl
.d_secsize
*
1143 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1147 #elif defined(__NetBSD__)
1148 static int64_t raw_getlength(BlockDriverState
*bs
)
1150 BDRVRawState
*s
= bs
->opaque
;
1156 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1157 struct dkwedge_info dkw
;
1159 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1160 return dkw
.dkw_size
* 512;
1162 struct disklabel dl
;
1164 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1166 return (uint64_t)dl
.d_secsize
*
1167 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1172 #elif defined(__sun__)
1173 static int64_t raw_getlength(BlockDriverState
*bs
)
1175 BDRVRawState
*s
= bs
->opaque
;
1176 struct dk_minfo minfo
;
1185 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1187 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
1189 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
1193 * There are reports that lseek on some devices fails, but
1194 * irc discussion said that contingency on contingency was overkill.
1196 return lseek(s
->fd
, 0, SEEK_END
);
1198 #elif defined(CONFIG_BSD)
1199 static int64_t raw_getlength(BlockDriverState
*bs
)
1201 BDRVRawState
*s
= bs
->opaque
;
1205 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1214 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1217 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
1218 #ifdef DIOCGMEDIASIZE
1219 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
1220 #elif defined(DIOCGPART)
1223 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
1224 size
= pi
.media_size
;
1230 #if defined(__APPLE__) && defined(__MACH__)
1233 size
= lseek(fd
, 0LL, SEEK_END
);
1235 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1238 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1239 if (size
== 2048LL * (unsigned)-1)
1241 /* XXX no disc? maybe we need to reopen... */
1242 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
1249 size
= lseek(fd
, 0, SEEK_END
);
1254 static int64_t raw_getlength(BlockDriverState
*bs
)
1256 BDRVRawState
*s
= bs
->opaque
;
1264 return lseek(s
->fd
, 0, SEEK_END
);
1268 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
1271 BDRVRawState
*s
= bs
->opaque
;
1273 if (fstat(s
->fd
, &st
) < 0) {
1276 return (int64_t)st
.st_blocks
* 512;
1279 static int raw_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1283 int64_t total_size
= 0;
1286 strstart(filename
, "file:", &filename
);
1288 /* Read out options */
1290 qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0) / BDRV_SECTOR_SIZE
;
1291 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
1293 fd
= qemu_open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
,
1297 error_setg_errno(errp
, -result
, "Could not create file");
1301 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1302 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1303 * will be ignored since any failure of this operation should not
1304 * block the left work.
1307 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
1308 attr
|= FS_NOCOW_FL
;
1309 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
1314 if (ftruncate(fd
, total_size
* BDRV_SECTOR_SIZE
) != 0) {
1316 error_setg_errno(errp
, -result
, "Could not resize file");
1318 if (qemu_close(fd
) != 0) {
1320 error_setg_errno(errp
, -result
, "Could not close the new file");
1326 static int64_t try_fiemap(BlockDriverState
*bs
, off_t start
, off_t
*data
,
1327 off_t
*hole
, int nb_sectors
, int *pnum
)
1329 #ifdef CONFIG_FIEMAP
1330 BDRVRawState
*s
= bs
->opaque
;
1331 int64_t ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1334 struct fiemap_extent fe
;
1337 if (s
->skip_fiemap
) {
1341 f
.fm
.fm_start
= start
;
1342 f
.fm
.fm_length
= (int64_t)nb_sectors
* BDRV_SECTOR_SIZE
;
1344 f
.fm
.fm_extent_count
= 1;
1345 f
.fm
.fm_reserved
= 0;
1346 if (ioctl(s
->fd
, FS_IOC_FIEMAP
, &f
) == -1) {
1347 s
->skip_fiemap
= true;
1351 if (f
.fm
.fm_mapped_extents
== 0) {
1352 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1353 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1355 off_t length
= lseek(s
->fd
, 0, SEEK_END
);
1356 *hole
= f
.fm
.fm_start
;
1357 *data
= MIN(f
.fm
.fm_start
+ f
.fm
.fm_length
, length
);
1359 *data
= f
.fe
.fe_logical
;
1360 *hole
= f
.fe
.fe_logical
+ f
.fe
.fe_length
;
1361 if (f
.fe
.fe_flags
& FIEMAP_EXTENT_UNWRITTEN
) {
1362 ret
|= BDRV_BLOCK_ZERO
;
1372 static int64_t try_seek_hole(BlockDriverState
*bs
, off_t start
, off_t
*data
,
1373 off_t
*hole
, int *pnum
)
1375 #if defined SEEK_HOLE && defined SEEK_DATA
1376 BDRVRawState
*s
= bs
->opaque
;
1378 *hole
= lseek(s
->fd
, start
, SEEK_HOLE
);
1380 /* -ENXIO indicates that sector_num was past the end of the file.
1381 * There is a virtual hole there. */
1382 assert(errno
!= -ENXIO
);
1387 if (*hole
> start
) {
1390 /* On a hole. We need another syscall to find its end. */
1391 *data
= lseek(s
->fd
, start
, SEEK_DATA
);
1393 *data
= lseek(s
->fd
, 0, SEEK_END
);
1397 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1404 * Returns true iff the specified sector is present in the disk image. Drivers
1405 * not implementing the functionality are assumed to not support backing files,
1406 * hence all their sectors are reported as allocated.
1408 * If 'sector_num' is beyond the end of the disk image the return value is 0
1409 * and 'pnum' is set to 0.
1411 * 'pnum' is set to the number of sectors (including and immediately following
1412 * the specified sector) that are known to be in the same
1413 * allocated/unallocated state.
1415 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1416 * beyond the end of the disk image it will be clamped.
1418 static int64_t coroutine_fn
raw_co_get_block_status(BlockDriverState
*bs
,
1420 int nb_sectors
, int *pnum
)
1422 off_t start
, data
= 0, hole
= 0;
1430 start
= sector_num
* BDRV_SECTOR_SIZE
;
1432 ret
= try_fiemap(bs
, start
, &data
, &hole
, nb_sectors
, pnum
);
1434 ret
= try_seek_hole(bs
, start
, &data
, &hole
, pnum
);
1436 /* Assume everything is allocated. */
1438 hole
= start
+ nb_sectors
* BDRV_SECTOR_SIZE
;
1439 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1443 if (data
<= start
) {
1444 /* On a data extent, compute sectors to the end of the extent. */
1445 *pnum
= MIN(nb_sectors
, (hole
- start
) / BDRV_SECTOR_SIZE
);
1447 /* On a hole, compute sectors to the beginning of the next extent. */
1448 *pnum
= MIN(nb_sectors
, (data
- start
) / BDRV_SECTOR_SIZE
);
1449 ret
&= ~BDRV_BLOCK_DATA
;
1450 ret
|= BDRV_BLOCK_ZERO
;
1456 static coroutine_fn BlockDriverAIOCB
*raw_aio_discard(BlockDriverState
*bs
,
1457 int64_t sector_num
, int nb_sectors
,
1458 BlockDriverCompletionFunc
*cb
, void *opaque
)
1460 BDRVRawState
*s
= bs
->opaque
;
1462 return paio_submit(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1463 cb
, opaque
, QEMU_AIO_DISCARD
);
1466 static int coroutine_fn
raw_co_write_zeroes(
1467 BlockDriverState
*bs
, int64_t sector_num
,
1468 int nb_sectors
, BdrvRequestFlags flags
)
1470 BDRVRawState
*s
= bs
->opaque
;
1472 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1473 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1474 QEMU_AIO_WRITE_ZEROES
);
1475 } else if (s
->discard_zeroes
) {
1476 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1482 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1484 BDRVRawState
*s
= bs
->opaque
;
1486 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
1487 bdi
->can_write_zeroes_with_unmap
= s
->discard_zeroes
;
1491 static QemuOptsList raw_create_opts
= {
1492 .name
= "raw-create-opts",
1493 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
1496 .name
= BLOCK_OPT_SIZE
,
1497 .type
= QEMU_OPT_SIZE
,
1498 .help
= "Virtual disk size"
1501 .name
= BLOCK_OPT_NOCOW
,
1502 .type
= QEMU_OPT_BOOL
,
1503 .help
= "Turn off copy-on-write (valid only on btrfs)"
1505 { /* end of list */ }
1509 static BlockDriver bdrv_file
= {
1510 .format_name
= "file",
1511 .protocol_name
= "file",
1512 .instance_size
= sizeof(BDRVRawState
),
1513 .bdrv_needs_filename
= true,
1514 .bdrv_probe
= NULL
, /* no probe for protocols */
1515 .bdrv_parse_filename
= raw_parse_filename
,
1516 .bdrv_file_open
= raw_open
,
1517 .bdrv_reopen_prepare
= raw_reopen_prepare
,
1518 .bdrv_reopen_commit
= raw_reopen_commit
,
1519 .bdrv_reopen_abort
= raw_reopen_abort
,
1520 .bdrv_close
= raw_close
,
1521 .bdrv_create
= raw_create
,
1522 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1523 .bdrv_co_get_block_status
= raw_co_get_block_status
,
1524 .bdrv_co_write_zeroes
= raw_co_write_zeroes
,
1526 .bdrv_aio_readv
= raw_aio_readv
,
1527 .bdrv_aio_writev
= raw_aio_writev
,
1528 .bdrv_aio_flush
= raw_aio_flush
,
1529 .bdrv_aio_discard
= raw_aio_discard
,
1530 .bdrv_refresh_limits
= raw_refresh_limits
,
1532 .bdrv_truncate
= raw_truncate
,
1533 .bdrv_getlength
= raw_getlength
,
1534 .bdrv_get_info
= raw_get_info
,
1535 .bdrv_get_allocated_file_size
1536 = raw_get_allocated_file_size
,
1538 .bdrv_detach_aio_context
= raw_detach_aio_context
,
1539 .bdrv_attach_aio_context
= raw_attach_aio_context
,
1541 .create_opts
= &raw_create_opts
,
1544 /***********************************************/
1547 #if defined(__APPLE__) && defined(__MACH__)
1548 static kern_return_t
FindEjectableCDMedia( io_iterator_t
*mediaIterator
);
1549 static kern_return_t
GetBSDPath( io_iterator_t mediaIterator
, char *bsdPath
, CFIndex maxPathSize
);
1551 kern_return_t
FindEjectableCDMedia( io_iterator_t
*mediaIterator
)
1553 kern_return_t kernResult
;
1554 mach_port_t masterPort
;
1555 CFMutableDictionaryRef classesToMatch
;
1557 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
1558 if ( KERN_SUCCESS
!= kernResult
) {
1559 printf( "IOMasterPort returned %d\n", kernResult
);
1562 classesToMatch
= IOServiceMatching( kIOCDMediaClass
);
1563 if ( classesToMatch
== NULL
) {
1564 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1566 CFDictionarySetValue( classesToMatch
, CFSTR( kIOMediaEjectableKey
), kCFBooleanTrue
);
1568 kernResult
= IOServiceGetMatchingServices( masterPort
, classesToMatch
, mediaIterator
);
1569 if ( KERN_SUCCESS
!= kernResult
)
1571 printf( "IOServiceGetMatchingServices returned %d\n", kernResult
);
1577 kern_return_t
GetBSDPath( io_iterator_t mediaIterator
, char *bsdPath
, CFIndex maxPathSize
)
1579 io_object_t nextMedia
;
1580 kern_return_t kernResult
= KERN_FAILURE
;
1582 nextMedia
= IOIteratorNext( mediaIterator
);
1585 CFTypeRef bsdPathAsCFString
;
1586 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
1587 if ( bsdPathAsCFString
) {
1588 size_t devPathLength
;
1589 strcpy( bsdPath
, _PATH_DEV
);
1590 strcat( bsdPath
, "r" );
1591 devPathLength
= strlen( bsdPath
);
1592 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
1593 kernResult
= KERN_SUCCESS
;
1595 CFRelease( bsdPathAsCFString
);
1597 IOObjectRelease( nextMedia
);
1605 static int hdev_probe_device(const char *filename
)
1609 /* allow a dedicated CD-ROM driver to match with a higher priority */
1610 if (strstart(filename
, "/dev/cdrom", NULL
))
1613 if (stat(filename
, &st
) >= 0 &&
1614 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
1621 static int check_hdev_writable(BDRVRawState
*s
)
1623 #if defined(BLKROGET)
1624 /* Linux block devices can be configured "read-only" using blockdev(8).
1625 * This is independent of device node permissions and therefore open(2)
1626 * with O_RDWR succeeds. Actual writes fail with EPERM.
1628 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1629 * check for read-only block devices so that Linux block devices behave
1635 if (fstat(s
->fd
, &st
)) {
1639 if (!S_ISBLK(st
.st_mode
)) {
1643 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
1650 #endif /* defined(BLKROGET) */
1654 static void hdev_parse_filename(const char *filename
, QDict
*options
,
1657 /* The prefix is optional, just as for "file". */
1658 strstart(filename
, "host_device:", &filename
);
1660 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
1663 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1666 BDRVRawState
*s
= bs
->opaque
;
1667 Error
*local_err
= NULL
;
1669 const char *filename
= qdict_get_str(options
, "filename");
1671 #if defined(__APPLE__) && defined(__MACH__)
1672 if (strstart(filename
, "/dev/cdrom", NULL
)) {
1673 kern_return_t kernResult
;
1674 io_iterator_t mediaIterator
;
1675 char bsdPath
[ MAXPATHLEN
];
1678 kernResult
= FindEjectableCDMedia( &mediaIterator
);
1679 kernResult
= GetBSDPath( mediaIterator
, bsdPath
, sizeof( bsdPath
) );
1681 if ( bsdPath
[ 0 ] != '\0' ) {
1682 strcat(bsdPath
,"s0");
1683 /* some CDs don't have a partition 0 */
1684 fd
= qemu_open(bsdPath
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
1686 bsdPath
[strlen(bsdPath
)-1] = '1';
1691 qdict_put(options
, "filename", qstring_from_str(filename
));
1694 if ( mediaIterator
)
1695 IOObjectRelease( mediaIterator
);
1699 s
->type
= FTYPE_FILE
;
1700 #if defined(__linux__)
1702 char resolved_path
[ MAXPATHLEN
], *temp
;
1704 temp
= realpath(filename
, resolved_path
);
1705 if (temp
&& strstart(temp
, "/dev/sg", NULL
)) {
1711 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
1714 error_propagate(errp
, local_err
);
1719 if (flags
& BDRV_O_RDWR
) {
1720 ret
= check_hdev_writable(s
);
1723 error_setg_errno(errp
, -ret
, "The device is not writable");
1731 #if defined(__linux__)
1732 /* Note: we do not have a reliable method to detect if the floppy is
1733 present. The current method is to try to open the floppy at every
1734 I/O and to keep it opened during a few hundreds of ms. */
1735 static int fd_open(BlockDriverState
*bs
)
1737 BDRVRawState
*s
= bs
->opaque
;
1738 int last_media_present
;
1740 if (s
->type
!= FTYPE_FD
)
1742 last_media_present
= (s
->fd
>= 0);
1744 (get_clock() - s
->fd_open_time
) >= FD_OPEN_TIMEOUT
) {
1748 printf("Floppy closed\n");
1752 if (s
->fd_got_error
&&
1753 (get_clock() - s
->fd_error_time
) < FD_OPEN_TIMEOUT
) {
1755 printf("No floppy (open delayed)\n");
1759 s
->fd
= qemu_open(bs
->filename
, s
->open_flags
& ~O_NONBLOCK
);
1761 s
->fd_error_time
= get_clock();
1762 s
->fd_got_error
= 1;
1763 if (last_media_present
)
1764 s
->fd_media_changed
= 1;
1766 printf("No floppy\n");
1771 printf("Floppy opened\n");
1774 if (!last_media_present
)
1775 s
->fd_media_changed
= 1;
1776 s
->fd_open_time
= get_clock();
1777 s
->fd_got_error
= 0;
1781 static int hdev_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
1783 BDRVRawState
*s
= bs
->opaque
;
1785 return ioctl(s
->fd
, req
, buf
);
1788 static BlockDriverAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
1789 unsigned long int req
, void *buf
,
1790 BlockDriverCompletionFunc
*cb
, void *opaque
)
1792 BDRVRawState
*s
= bs
->opaque
;
1793 RawPosixAIOData
*acb
;
1796 if (fd_open(bs
) < 0)
1799 acb
= g_slice_new(RawPosixAIOData
);
1801 acb
->aio_type
= QEMU_AIO_IOCTL
;
1802 acb
->aio_fildes
= s
->fd
;
1803 acb
->aio_offset
= 0;
1804 acb
->aio_ioctl_buf
= buf
;
1805 acb
->aio_ioctl_cmd
= req
;
1806 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1807 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1810 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1811 static int fd_open(BlockDriverState
*bs
)
1813 BDRVRawState
*s
= bs
->opaque
;
1815 /* this is just to ensure s->fd is sane (its called by io ops) */
1820 #else /* !linux && !FreeBSD */
1822 static int fd_open(BlockDriverState
*bs
)
1827 #endif /* !linux && !FreeBSD */
1829 static coroutine_fn BlockDriverAIOCB
*hdev_aio_discard(BlockDriverState
*bs
,
1830 int64_t sector_num
, int nb_sectors
,
1831 BlockDriverCompletionFunc
*cb
, void *opaque
)
1833 BDRVRawState
*s
= bs
->opaque
;
1835 if (fd_open(bs
) < 0) {
1838 return paio_submit(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1839 cb
, opaque
, QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
1842 static coroutine_fn
int hdev_co_write_zeroes(BlockDriverState
*bs
,
1843 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1845 BDRVRawState
*s
= bs
->opaque
;
1852 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1853 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1854 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
1855 } else if (s
->discard_zeroes
) {
1856 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1857 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
1862 static int hdev_create(const char *filename
, QemuOpts
*opts
,
1867 struct stat stat_buf
;
1868 int64_t total_size
= 0;
1871 /* This function is used by all three protocol block drivers and therefore
1872 * any of these three prefixes may be given.
1873 * The return value has to be stored somewhere, otherwise this is an error
1874 * due to -Werror=unused-value. */
1876 strstart(filename
, "host_device:", &filename
) ||
1877 strstart(filename
, "host_cdrom:" , &filename
) ||
1878 strstart(filename
, "host_floppy:", &filename
);
1882 /* Read out options */
1884 qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0) / BDRV_SECTOR_SIZE
;
1886 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
1889 error_setg_errno(errp
, -ret
, "Could not open device");
1893 if (fstat(fd
, &stat_buf
) < 0) {
1895 error_setg_errno(errp
, -ret
, "Could not stat device");
1896 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
1898 "The given file is neither a block nor a character device");
1900 } else if (lseek(fd
, 0, SEEK_END
) < total_size
* BDRV_SECTOR_SIZE
) {
1901 error_setg(errp
, "Device is too small");
1909 static BlockDriver bdrv_host_device
= {
1910 .format_name
= "host_device",
1911 .protocol_name
= "host_device",
1912 .instance_size
= sizeof(BDRVRawState
),
1913 .bdrv_needs_filename
= true,
1914 .bdrv_probe_device
= hdev_probe_device
,
1915 .bdrv_parse_filename
= hdev_parse_filename
,
1916 .bdrv_file_open
= hdev_open
,
1917 .bdrv_close
= raw_close
,
1918 .bdrv_reopen_prepare
= raw_reopen_prepare
,
1919 .bdrv_reopen_commit
= raw_reopen_commit
,
1920 .bdrv_reopen_abort
= raw_reopen_abort
,
1921 .bdrv_create
= hdev_create
,
1922 .create_opts
= &raw_create_opts
,
1923 .bdrv_co_write_zeroes
= hdev_co_write_zeroes
,
1925 .bdrv_aio_readv
= raw_aio_readv
,
1926 .bdrv_aio_writev
= raw_aio_writev
,
1927 .bdrv_aio_flush
= raw_aio_flush
,
1928 .bdrv_aio_discard
= hdev_aio_discard
,
1929 .bdrv_refresh_limits
= raw_refresh_limits
,
1931 .bdrv_truncate
= raw_truncate
,
1932 .bdrv_getlength
= raw_getlength
,
1933 .bdrv_get_info
= raw_get_info
,
1934 .bdrv_get_allocated_file_size
1935 = raw_get_allocated_file_size
,
1937 .bdrv_detach_aio_context
= raw_detach_aio_context
,
1938 .bdrv_attach_aio_context
= raw_attach_aio_context
,
1940 /* generic scsi device */
1942 .bdrv_ioctl
= hdev_ioctl
,
1943 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
1948 static void floppy_parse_filename(const char *filename
, QDict
*options
,
1951 /* The prefix is optional, just as for "file". */
1952 strstart(filename
, "host_floppy:", &filename
);
1954 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
1957 static int floppy_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1960 BDRVRawState
*s
= bs
->opaque
;
1961 Error
*local_err
= NULL
;
1966 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1967 ret
= raw_open_common(bs
, options
, flags
, O_NONBLOCK
, &local_err
);
1970 error_propagate(errp
, local_err
);
1975 /* close fd so that we can reopen it as needed */
1978 s
->fd_media_changed
= 1;
1983 static int floppy_probe_device(const char *filename
)
1987 struct floppy_struct fdparam
;
1990 if (strstart(filename
, "/dev/fd", NULL
) &&
1991 !strstart(filename
, "/dev/fdset/", NULL
)) {
1995 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
1999 ret
= fstat(fd
, &st
);
2000 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2004 /* Attempt to detect via a floppy specific ioctl */
2005 ret
= ioctl(fd
, FDGETPRM
, &fdparam
);
2016 static int floppy_is_inserted(BlockDriverState
*bs
)
2018 return fd_open(bs
) >= 0;
2021 static int floppy_media_changed(BlockDriverState
*bs
)
2023 BDRVRawState
*s
= bs
->opaque
;
2027 * XXX: we do not have a true media changed indication.
2028 * It does not work if the floppy is changed without trying to read it.
2031 ret
= s
->fd_media_changed
;
2032 s
->fd_media_changed
= 0;
2034 printf("Floppy changed=%d\n", ret
);
2039 static void floppy_eject(BlockDriverState
*bs
, bool eject_flag
)
2041 BDRVRawState
*s
= bs
->opaque
;
2048 fd
= qemu_open(bs
->filename
, s
->open_flags
| O_NONBLOCK
);
2050 if (ioctl(fd
, FDEJECT
, 0) < 0)
2056 static BlockDriver bdrv_host_floppy
= {
2057 .format_name
= "host_floppy",
2058 .protocol_name
= "host_floppy",
2059 .instance_size
= sizeof(BDRVRawState
),
2060 .bdrv_needs_filename
= true,
2061 .bdrv_probe_device
= floppy_probe_device
,
2062 .bdrv_parse_filename
= floppy_parse_filename
,
2063 .bdrv_file_open
= floppy_open
,
2064 .bdrv_close
= raw_close
,
2065 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2066 .bdrv_reopen_commit
= raw_reopen_commit
,
2067 .bdrv_reopen_abort
= raw_reopen_abort
,
2068 .bdrv_create
= hdev_create
,
2069 .create_opts
= &raw_create_opts
,
2071 .bdrv_aio_readv
= raw_aio_readv
,
2072 .bdrv_aio_writev
= raw_aio_writev
,
2073 .bdrv_aio_flush
= raw_aio_flush
,
2074 .bdrv_refresh_limits
= raw_refresh_limits
,
2076 .bdrv_truncate
= raw_truncate
,
2077 .bdrv_getlength
= raw_getlength
,
2078 .has_variable_length
= true,
2079 .bdrv_get_allocated_file_size
2080 = raw_get_allocated_file_size
,
2082 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2083 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2085 /* removable device support */
2086 .bdrv_is_inserted
= floppy_is_inserted
,
2087 .bdrv_media_changed
= floppy_media_changed
,
2088 .bdrv_eject
= floppy_eject
,
2092 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2093 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
2096 /* The prefix is optional, just as for "file". */
2097 strstart(filename
, "host_cdrom:", &filename
);
2099 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
2104 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2107 BDRVRawState
*s
= bs
->opaque
;
2108 Error
*local_err
= NULL
;
2113 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2114 ret
= raw_open_common(bs
, options
, flags
, O_NONBLOCK
, &local_err
);
2116 error_propagate(errp
, local_err
);
2121 static int cdrom_probe_device(const char *filename
)
2127 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
2131 ret
= fstat(fd
, &st
);
2132 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2136 /* Attempt to detect via a CDROM specific ioctl */
2137 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2147 static int cdrom_is_inserted(BlockDriverState
*bs
)
2149 BDRVRawState
*s
= bs
->opaque
;
2152 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2153 if (ret
== CDS_DISC_OK
)
2158 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2160 BDRVRawState
*s
= bs
->opaque
;
2163 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
2164 perror("CDROMEJECT");
2166 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
2167 perror("CDROMEJECT");
2171 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2173 BDRVRawState
*s
= bs
->opaque
;
2175 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
2177 * Note: an error can happen if the distribution automatically
2180 /* perror("CDROM_LOCKDOOR"); */
2184 static BlockDriver bdrv_host_cdrom
= {
2185 .format_name
= "host_cdrom",
2186 .protocol_name
= "host_cdrom",
2187 .instance_size
= sizeof(BDRVRawState
),
2188 .bdrv_needs_filename
= true,
2189 .bdrv_probe_device
= cdrom_probe_device
,
2190 .bdrv_parse_filename
= cdrom_parse_filename
,
2191 .bdrv_file_open
= cdrom_open
,
2192 .bdrv_close
= raw_close
,
2193 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2194 .bdrv_reopen_commit
= raw_reopen_commit
,
2195 .bdrv_reopen_abort
= raw_reopen_abort
,
2196 .bdrv_create
= hdev_create
,
2197 .create_opts
= &raw_create_opts
,
2199 .bdrv_aio_readv
= raw_aio_readv
,
2200 .bdrv_aio_writev
= raw_aio_writev
,
2201 .bdrv_aio_flush
= raw_aio_flush
,
2202 .bdrv_refresh_limits
= raw_refresh_limits
,
2204 .bdrv_truncate
= raw_truncate
,
2205 .bdrv_getlength
= raw_getlength
,
2206 .has_variable_length
= true,
2207 .bdrv_get_allocated_file_size
2208 = raw_get_allocated_file_size
,
2210 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2211 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2213 /* removable device support */
2214 .bdrv_is_inserted
= cdrom_is_inserted
,
2215 .bdrv_eject
= cdrom_eject
,
2216 .bdrv_lock_medium
= cdrom_lock_medium
,
2218 /* generic scsi device */
2219 .bdrv_ioctl
= hdev_ioctl
,
2220 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2222 #endif /* __linux__ */
2224 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2225 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2228 BDRVRawState
*s
= bs
->opaque
;
2229 Error
*local_err
= NULL
;
2234 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2237 error_propagate(errp
, local_err
);
2242 /* make sure the door isn't locked at this time */
2243 ioctl(s
->fd
, CDIOCALLOW
);
2247 static int cdrom_probe_device(const char *filename
)
2249 if (strstart(filename
, "/dev/cd", NULL
) ||
2250 strstart(filename
, "/dev/acd", NULL
))
2255 static int cdrom_reopen(BlockDriverState
*bs
)
2257 BDRVRawState
*s
= bs
->opaque
;
2261 * Force reread of possibly changed/newly loaded disc,
2262 * FreeBSD seems to not notice sometimes...
2266 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
2273 /* make sure the door isn't locked at this time */
2274 ioctl(s
->fd
, CDIOCALLOW
);
2278 static int cdrom_is_inserted(BlockDriverState
*bs
)
2280 return raw_getlength(bs
) > 0;
2283 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2285 BDRVRawState
*s
= bs
->opaque
;
2290 (void) ioctl(s
->fd
, CDIOCALLOW
);
2293 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
2294 perror("CDIOCEJECT");
2296 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
2297 perror("CDIOCCLOSE");
2303 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2305 BDRVRawState
*s
= bs
->opaque
;
2309 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
2311 * Note: an error can happen if the distribution automatically
2314 /* perror("CDROM_LOCKDOOR"); */
2318 static BlockDriver bdrv_host_cdrom
= {
2319 .format_name
= "host_cdrom",
2320 .protocol_name
= "host_cdrom",
2321 .instance_size
= sizeof(BDRVRawState
),
2322 .bdrv_needs_filename
= true,
2323 .bdrv_probe_device
= cdrom_probe_device
,
2324 .bdrv_parse_filename
= cdrom_parse_filename
,
2325 .bdrv_file_open
= cdrom_open
,
2326 .bdrv_close
= raw_close
,
2327 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2328 .bdrv_reopen_commit
= raw_reopen_commit
,
2329 .bdrv_reopen_abort
= raw_reopen_abort
,
2330 .bdrv_create
= hdev_create
,
2331 .create_opts
= &raw_create_opts
,
2333 .bdrv_aio_readv
= raw_aio_readv
,
2334 .bdrv_aio_writev
= raw_aio_writev
,
2335 .bdrv_aio_flush
= raw_aio_flush
,
2336 .bdrv_refresh_limits
= raw_refresh_limits
,
2338 .bdrv_truncate
= raw_truncate
,
2339 .bdrv_getlength
= raw_getlength
,
2340 .has_variable_length
= true,
2341 .bdrv_get_allocated_file_size
2342 = raw_get_allocated_file_size
,
2344 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2345 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2347 /* removable device support */
2348 .bdrv_is_inserted
= cdrom_is_inserted
,
2349 .bdrv_eject
= cdrom_eject
,
2350 .bdrv_lock_medium
= cdrom_lock_medium
,
2352 #endif /* __FreeBSD__ */
2354 static void bdrv_file_init(void)
2357 * Register all the drivers. Note that order is important, the driver
2358 * registered last will get probed first.
2360 bdrv_register(&bdrv_file
);
2361 bdrv_register(&bdrv_host_device
);
2363 bdrv_register(&bdrv_host_floppy
);
2364 bdrv_register(&bdrv_host_cdrom
);
2366 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2367 bdrv_register(&bdrv_host_cdrom
);
2371 block_init(bdrv_file_init
);