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
, int fd
, Error
**errp
)
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(fd
, BLKSSZGET
, §or_size
) >= 0) {
244 bs
->request_alignment
= sector_size
;
247 #ifdef DKIOCGETBLOCKSIZE
248 if (ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) >= 0) {
249 bs
->request_alignment
= sector_size
;
252 #ifdef DIOCGSECTORSIZE
253 if (ioctl(fd
, DIOCGSECTORSIZE
, §or_size
) >= 0) {
254 bs
->request_alignment
= sector_size
;
260 if (xfsctl(NULL
, 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(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(fd
, buf
, align
, 0) >= 0) {
286 bs
->request_alignment
= align
;
293 if (!s
->buf_align
|| !bs
->request_alignment
) {
294 error_setg(errp
, "Could not find working O_DIRECT alignment. "
295 "Try cache.direct=off.");
299 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
301 assert(open_flags
!= NULL
);
303 *open_flags
|= O_BINARY
;
304 *open_flags
&= ~O_ACCMODE
;
305 if (bdrv_flags
& BDRV_O_RDWR
) {
306 *open_flags
|= O_RDWR
;
308 *open_flags
|= O_RDONLY
;
311 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
312 * and O_DIRECT for no caching. */
313 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
314 *open_flags
|= O_DIRECT
;
318 static void raw_detach_aio_context(BlockDriverState
*bs
)
320 #ifdef CONFIG_LINUX_AIO
321 BDRVRawState
*s
= bs
->opaque
;
324 laio_detach_aio_context(s
->aio_ctx
, bdrv_get_aio_context(bs
));
329 static void raw_attach_aio_context(BlockDriverState
*bs
,
330 AioContext
*new_context
)
332 #ifdef CONFIG_LINUX_AIO
333 BDRVRawState
*s
= bs
->opaque
;
336 laio_attach_aio_context(s
->aio_ctx
, new_context
);
341 #ifdef CONFIG_LINUX_AIO
342 static int raw_set_aio(void **aio_ctx
, int *use_aio
, int bdrv_flags
)
345 assert(aio_ctx
!= NULL
);
346 assert(use_aio
!= NULL
);
348 * Currently Linux do AIO only for files opened with O_DIRECT
349 * specified so check NOCACHE flag too
351 if ((bdrv_flags
& (BDRV_O_NOCACHE
|BDRV_O_NATIVE_AIO
)) ==
352 (BDRV_O_NOCACHE
|BDRV_O_NATIVE_AIO
)) {
354 /* if non-NULL, laio_init() has already been run */
355 if (*aio_ctx
== NULL
) {
356 *aio_ctx
= laio_init();
373 static void raw_parse_filename(const char *filename
, QDict
*options
,
376 /* The filename does not have to be prefixed by the protocol name, since
377 * "file" is the default protocol; therefore, the return value of this
378 * function call can be ignored. */
379 strstart(filename
, "file:", &filename
);
381 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
384 static QemuOptsList raw_runtime_opts
= {
386 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
390 .type
= QEMU_OPT_STRING
,
391 .help
= "File name of the image",
393 { /* end of list */ }
397 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
398 int bdrv_flags
, int open_flags
, Error
**errp
)
400 BDRVRawState
*s
= bs
->opaque
;
402 Error
*local_err
= NULL
;
403 const char *filename
= NULL
;
407 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
408 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
410 error_propagate(errp
, local_err
);
415 filename
= qemu_opt_get(opts
, "filename");
417 ret
= raw_normalize_devicepath(&filename
);
419 error_setg_errno(errp
, -ret
, "Could not normalize device path");
423 s
->open_flags
= open_flags
;
424 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
427 fd
= qemu_open(filename
, s
->open_flags
, 0644);
437 #ifdef CONFIG_LINUX_AIO
438 if (raw_set_aio(&s
->aio_ctx
, &s
->use_aio
, bdrv_flags
)) {
441 error_setg_errno(errp
, -ret
, "Could not set AIO state");
446 s
->has_discard
= true;
447 s
->has_write_zeroes
= true;
449 if (fstat(s
->fd
, &st
) < 0) {
450 error_setg_errno(errp
, errno
, "Could not stat file");
453 if (S_ISREG(st
.st_mode
)) {
454 s
->discard_zeroes
= true;
456 if (S_ISBLK(st
.st_mode
)) {
457 #ifdef BLKDISCARDZEROES
459 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
460 s
->discard_zeroes
= true;
464 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
465 * not rely on the contents of discarded blocks unless using O_DIRECT.
466 * Same for BLKZEROOUT.
468 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
469 s
->discard_zeroes
= false;
470 s
->has_write_zeroes
= false;
476 if (platform_test_xfs_fd(s
->fd
)) {
481 raw_attach_aio_context(bs
, bdrv_get_aio_context(bs
));
485 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
492 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
495 BDRVRawState
*s
= bs
->opaque
;
496 Error
*local_err
= NULL
;
499 s
->type
= FTYPE_FILE
;
500 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
502 error_propagate(errp
, local_err
);
507 static int raw_reopen_prepare(BDRVReopenState
*state
,
508 BlockReopenQueue
*queue
, Error
**errp
)
511 BDRVRawReopenState
*raw_s
;
513 Error
*local_err
= NULL
;
515 assert(state
!= NULL
);
516 assert(state
->bs
!= NULL
);
518 s
= state
->bs
->opaque
;
520 state
->opaque
= g_malloc0(sizeof(BDRVRawReopenState
));
521 raw_s
= state
->opaque
;
523 #ifdef CONFIG_LINUX_AIO
524 raw_s
->use_aio
= s
->use_aio
;
526 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
527 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
528 * won't override aio_ctx if aio_ctx is non-NULL */
529 if (raw_set_aio(&s
->aio_ctx
, &raw_s
->use_aio
, state
->flags
)) {
530 error_setg(errp
, "Could not set AIO state");
535 if (s
->type
== FTYPE_FD
|| s
->type
== FTYPE_CD
) {
536 raw_s
->open_flags
|= O_NONBLOCK
;
539 raw_parse_flags(state
->flags
, &raw_s
->open_flags
);
543 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
545 fcntl_flags
|= O_NOATIME
;
549 /* Not all operating systems have O_ASYNC, and those that don't
550 * will not let us track the state into raw_s->open_flags (typically
551 * you achieve the same effect with an ioctl, for example I_SETSIG
552 * on Solaris). But we do not use O_ASYNC, so that's fine.
554 assert((s
->open_flags
& O_ASYNC
) == 0);
557 if ((raw_s
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
558 /* dup the original fd */
559 /* TODO: use qemu fcntl wrapper */
560 #ifdef F_DUPFD_CLOEXEC
561 raw_s
->fd
= fcntl(s
->fd
, F_DUPFD_CLOEXEC
, 0);
563 raw_s
->fd
= dup(s
->fd
);
564 if (raw_s
->fd
!= -1) {
565 qemu_set_cloexec(raw_s
->fd
);
568 if (raw_s
->fd
>= 0) {
569 ret
= fcntl_setfl(raw_s
->fd
, raw_s
->open_flags
);
571 qemu_close(raw_s
->fd
);
577 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
578 if (raw_s
->fd
== -1) {
579 assert(!(raw_s
->open_flags
& O_CREAT
));
580 raw_s
->fd
= qemu_open(state
->bs
->filename
, raw_s
->open_flags
);
581 if (raw_s
->fd
== -1) {
582 error_setg_errno(errp
, errno
, "Could not reopen file");
587 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
588 * alignment with the new fd. */
589 if (raw_s
->fd
!= -1) {
590 raw_probe_alignment(state
->bs
, raw_s
->fd
, &local_err
);
592 qemu_close(raw_s
->fd
);
594 error_propagate(errp
, local_err
);
602 static void raw_reopen_commit(BDRVReopenState
*state
)
604 BDRVRawReopenState
*raw_s
= state
->opaque
;
605 BDRVRawState
*s
= state
->bs
->opaque
;
607 s
->open_flags
= raw_s
->open_flags
;
611 #ifdef CONFIG_LINUX_AIO
612 s
->use_aio
= raw_s
->use_aio
;
615 g_free(state
->opaque
);
616 state
->opaque
= NULL
;
620 static void raw_reopen_abort(BDRVReopenState
*state
)
622 BDRVRawReopenState
*raw_s
= state
->opaque
;
624 /* nothing to do if NULL, we didn't get far enough */
629 if (raw_s
->fd
>= 0) {
630 qemu_close(raw_s
->fd
);
633 g_free(state
->opaque
);
634 state
->opaque
= NULL
;
637 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
639 BDRVRawState
*s
= bs
->opaque
;
641 raw_probe_alignment(bs
, s
->fd
, errp
);
642 bs
->bl
.opt_mem_alignment
= s
->buf_align
;
645 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
649 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
657 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
661 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
670 static bool preadv_present
= true;
673 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
675 return preadv(fd
, iov
, nr_iov
, offset
);
679 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
681 return pwritev(fd
, iov
, nr_iov
, offset
);
686 static bool preadv_present
= false;
689 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
695 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
702 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
707 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
708 len
= qemu_pwritev(aiocb
->aio_fildes
,
713 len
= qemu_preadv(aiocb
->aio_fildes
,
717 } while (len
== -1 && errno
== EINTR
);
726 * Read/writes the data to/from a given linear buffer.
728 * Returns the number of bytes handles or -errno in case of an error. Short
729 * reads are only returned if the end of the file is reached.
731 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
736 while (offset
< aiocb
->aio_nbytes
) {
737 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
738 len
= pwrite(aiocb
->aio_fildes
,
739 (const char *)buf
+ offset
,
740 aiocb
->aio_nbytes
- offset
,
741 aiocb
->aio_offset
+ offset
);
743 len
= pread(aiocb
->aio_fildes
,
745 aiocb
->aio_nbytes
- offset
,
746 aiocb
->aio_offset
+ offset
);
748 if (len
== -1 && errno
== EINTR
) {
750 } else if (len
== -1) {
753 } else if (len
== 0) {
762 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
767 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
769 * If there is just a single buffer, and it is properly aligned
770 * we can just use plain pread/pwrite without any problems.
772 if (aiocb
->aio_niov
== 1) {
773 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
776 * We have more than one iovec, and all are properly aligned.
778 * Try preadv/pwritev first and fall back to linearizing the
779 * buffer if it's not supported.
781 if (preadv_present
) {
782 nbytes
= handle_aiocb_rw_vector(aiocb
);
783 if (nbytes
== aiocb
->aio_nbytes
||
784 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
787 preadv_present
= false;
791 * XXX(hch): short read/write. no easy way to handle the reminder
792 * using these interfaces. For now retry using plain
798 * Ok, we have to do it the hard way, copy all segments into
799 * a single aligned buffer.
801 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
806 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
810 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
811 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
812 p
+= aiocb
->aio_iov
[i
].iov_len
;
814 assert(p
- buf
== aiocb
->aio_nbytes
);
817 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
818 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
820 size_t count
= aiocb
->aio_nbytes
, copy
;
823 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
825 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
826 copy
= aiocb
->aio_iov
[i
].iov_len
;
828 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
829 assert(count
>= copy
);
841 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
843 struct xfs_flock64 fl
;
845 memset(&fl
, 0, sizeof(fl
));
846 fl
.l_whence
= SEEK_SET
;
850 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
851 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno
));
858 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
860 struct xfs_flock64 fl
;
862 memset(&fl
, 0, sizeof(fl
));
863 fl
.l_whence
= SEEK_SET
;
867 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
868 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno
));
876 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
878 int ret
= -EOPNOTSUPP
;
879 BDRVRawState
*s
= aiocb
->bs
->opaque
;
881 if (s
->has_write_zeroes
== 0) {
885 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
888 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
889 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
892 } while (errno
== EINTR
);
899 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
904 if (ret
== -ENODEV
|| ret
== -ENOSYS
|| ret
== -EOPNOTSUPP
||
906 s
->has_write_zeroes
= false;
912 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
914 int ret
= -EOPNOTSUPP
;
915 BDRVRawState
*s
= aiocb
->bs
->opaque
;
917 if (!s
->has_discard
) {
921 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
924 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
925 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
928 } while (errno
== EINTR
);
935 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
939 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
941 if (fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
942 aiocb
->aio_offset
, aiocb
->aio_nbytes
) == 0) {
945 } while (errno
== EINTR
);
951 if (ret
== -ENODEV
|| ret
== -ENOSYS
|| ret
== -EOPNOTSUPP
||
953 s
->has_discard
= false;
959 static int aio_worker(void *arg
)
961 RawPosixAIOData
*aiocb
= arg
;
964 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
966 ret
= handle_aiocb_rw(aiocb
);
967 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
&& aiocb
->bs
->growable
) {
968 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
969 0, aiocb
->aio_nbytes
- ret
);
971 ret
= aiocb
->aio_nbytes
;
973 if (ret
== aiocb
->aio_nbytes
) {
975 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
980 ret
= handle_aiocb_rw(aiocb
);
981 if (ret
== aiocb
->aio_nbytes
) {
983 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
988 ret
= handle_aiocb_flush(aiocb
);
991 ret
= handle_aiocb_ioctl(aiocb
);
993 case QEMU_AIO_DISCARD
:
994 ret
= handle_aiocb_discard(aiocb
);
996 case QEMU_AIO_WRITE_ZEROES
:
997 ret
= handle_aiocb_write_zeroes(aiocb
);
1000 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1005 g_slice_free(RawPosixAIOData
, aiocb
);
1009 static int paio_submit_co(BlockDriverState
*bs
, int fd
,
1010 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1013 RawPosixAIOData
*acb
= g_slice_new(RawPosixAIOData
);
1017 acb
->aio_type
= type
;
1018 acb
->aio_fildes
= fd
;
1020 acb
->aio_nbytes
= nb_sectors
* BDRV_SECTOR_SIZE
;
1021 acb
->aio_offset
= sector_num
* BDRV_SECTOR_SIZE
;
1024 acb
->aio_iov
= qiov
->iov
;
1025 acb
->aio_niov
= qiov
->niov
;
1026 assert(qiov
->size
== acb
->aio_nbytes
);
1029 trace_paio_submit_co(sector_num
, nb_sectors
, type
);
1030 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1031 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1034 static BlockDriverAIOCB
*paio_submit(BlockDriverState
*bs
, int fd
,
1035 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1036 BlockDriverCompletionFunc
*cb
, void *opaque
, int type
)
1038 RawPosixAIOData
*acb
= g_slice_new(RawPosixAIOData
);
1042 acb
->aio_type
= type
;
1043 acb
->aio_fildes
= fd
;
1045 acb
->aio_nbytes
= nb_sectors
* BDRV_SECTOR_SIZE
;
1046 acb
->aio_offset
= sector_num
* BDRV_SECTOR_SIZE
;
1049 acb
->aio_iov
= qiov
->iov
;
1050 acb
->aio_niov
= qiov
->niov
;
1051 assert(qiov
->size
== acb
->aio_nbytes
);
1054 trace_paio_submit(acb
, opaque
, sector_num
, nb_sectors
, type
);
1055 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1056 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1059 static BlockDriverAIOCB
*raw_aio_submit(BlockDriverState
*bs
,
1060 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1061 BlockDriverCompletionFunc
*cb
, void *opaque
, int type
)
1063 BDRVRawState
*s
= bs
->opaque
;
1065 if (fd_open(bs
) < 0)
1069 * If O_DIRECT is used the buffer needs to be aligned on a sector
1070 * boundary. Check if this is the case or tell the low-level
1071 * driver that it needs to copy the buffer.
1073 if ((bs
->open_flags
& BDRV_O_NOCACHE
)) {
1074 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1075 type
|= QEMU_AIO_MISALIGNED
;
1076 #ifdef CONFIG_LINUX_AIO
1077 } else if (s
->use_aio
) {
1078 return laio_submit(bs
, s
->aio_ctx
, s
->fd
, sector_num
, qiov
,
1079 nb_sectors
, cb
, opaque
, type
);
1084 return paio_submit(bs
, s
->fd
, sector_num
, qiov
, nb_sectors
,
1088 static void raw_aio_plug(BlockDriverState
*bs
)
1090 #ifdef CONFIG_LINUX_AIO
1091 BDRVRawState
*s
= bs
->opaque
;
1093 laio_io_plug(bs
, s
->aio_ctx
);
1098 static void raw_aio_unplug(BlockDriverState
*bs
)
1100 #ifdef CONFIG_LINUX_AIO
1101 BDRVRawState
*s
= bs
->opaque
;
1103 laio_io_unplug(bs
, s
->aio_ctx
, true);
1108 static void raw_aio_flush_io_queue(BlockDriverState
*bs
)
1110 #ifdef CONFIG_LINUX_AIO
1111 BDRVRawState
*s
= bs
->opaque
;
1113 laio_io_unplug(bs
, s
->aio_ctx
, false);
1118 static BlockDriverAIOCB
*raw_aio_readv(BlockDriverState
*bs
,
1119 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1120 BlockDriverCompletionFunc
*cb
, void *opaque
)
1122 return raw_aio_submit(bs
, sector_num
, qiov
, nb_sectors
,
1123 cb
, opaque
, QEMU_AIO_READ
);
1126 static BlockDriverAIOCB
*raw_aio_writev(BlockDriverState
*bs
,
1127 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1128 BlockDriverCompletionFunc
*cb
, void *opaque
)
1130 return raw_aio_submit(bs
, sector_num
, qiov
, nb_sectors
,
1131 cb
, opaque
, QEMU_AIO_WRITE
);
1134 static BlockDriverAIOCB
*raw_aio_flush(BlockDriverState
*bs
,
1135 BlockDriverCompletionFunc
*cb
, void *opaque
)
1137 BDRVRawState
*s
= bs
->opaque
;
1139 if (fd_open(bs
) < 0)
1142 return paio_submit(bs
, s
->fd
, 0, NULL
, 0, cb
, opaque
, QEMU_AIO_FLUSH
);
1145 static void raw_close(BlockDriverState
*bs
)
1147 BDRVRawState
*s
= bs
->opaque
;
1149 raw_detach_aio_context(bs
);
1151 #ifdef CONFIG_LINUX_AIO
1153 laio_cleanup(s
->aio_ctx
);
1162 static int raw_truncate(BlockDriverState
*bs
, int64_t offset
)
1164 BDRVRawState
*s
= bs
->opaque
;
1167 if (fstat(s
->fd
, &st
)) {
1171 if (S_ISREG(st
.st_mode
)) {
1172 if (ftruncate(s
->fd
, offset
) < 0) {
1175 } else if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1176 if (offset
> raw_getlength(bs
)) {
1187 static int64_t raw_getlength(BlockDriverState
*bs
)
1189 BDRVRawState
*s
= bs
->opaque
;
1195 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1196 struct disklabel dl
;
1198 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1200 return (uint64_t)dl
.d_secsize
*
1201 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1205 #elif defined(__NetBSD__)
1206 static int64_t raw_getlength(BlockDriverState
*bs
)
1208 BDRVRawState
*s
= bs
->opaque
;
1214 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1215 struct dkwedge_info dkw
;
1217 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1218 return dkw
.dkw_size
* 512;
1220 struct disklabel dl
;
1222 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1224 return (uint64_t)dl
.d_secsize
*
1225 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1230 #elif defined(__sun__)
1231 static int64_t raw_getlength(BlockDriverState
*bs
)
1233 BDRVRawState
*s
= bs
->opaque
;
1234 struct dk_minfo minfo
;
1244 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1246 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
1248 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
1252 * There are reports that lseek on some devices fails, but
1253 * irc discussion said that contingency on contingency was overkill.
1255 size
= lseek(s
->fd
, 0, SEEK_END
);
1261 #elif defined(CONFIG_BSD)
1262 static int64_t raw_getlength(BlockDriverState
*bs
)
1264 BDRVRawState
*s
= bs
->opaque
;
1268 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1277 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1280 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
1281 #ifdef DIOCGMEDIASIZE
1282 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
1283 #elif defined(DIOCGPART)
1286 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
1287 size
= pi
.media_size
;
1293 #if defined(__APPLE__) && defined(__MACH__)
1296 size
= lseek(fd
, 0LL, SEEK_END
);
1301 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1304 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1305 if (size
== 2048LL * (unsigned)-1)
1307 /* XXX no disc? maybe we need to reopen... */
1308 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
1315 size
= lseek(fd
, 0, SEEK_END
);
1323 static int64_t raw_getlength(BlockDriverState
*bs
)
1325 BDRVRawState
*s
= bs
->opaque
;
1334 size
= lseek(s
->fd
, 0, SEEK_END
);
1342 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
1345 BDRVRawState
*s
= bs
->opaque
;
1347 if (fstat(s
->fd
, &st
) < 0) {
1350 return (int64_t)st
.st_blocks
* 512;
1353 static int raw_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1357 int64_t total_size
= 0;
1360 strstart(filename
, "file:", &filename
);
1362 /* Read out options */
1364 qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0) / BDRV_SECTOR_SIZE
;
1365 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
1367 fd
= qemu_open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
,
1371 error_setg_errno(errp
, -result
, "Could not create file");
1375 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1376 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1377 * will be ignored since any failure of this operation should not
1378 * block the left work.
1381 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
1382 attr
|= FS_NOCOW_FL
;
1383 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
1388 if (ftruncate(fd
, total_size
* BDRV_SECTOR_SIZE
) != 0) {
1390 error_setg_errno(errp
, -result
, "Could not resize file");
1392 if (qemu_close(fd
) != 0) {
1394 error_setg_errno(errp
, -result
, "Could not close the new file");
1400 static int64_t try_fiemap(BlockDriverState
*bs
, off_t start
, off_t
*data
,
1401 off_t
*hole
, int nb_sectors
, int *pnum
)
1403 #ifdef CONFIG_FIEMAP
1404 BDRVRawState
*s
= bs
->opaque
;
1405 int64_t ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1408 struct fiemap_extent fe
;
1411 if (s
->skip_fiemap
) {
1415 f
.fm
.fm_start
= start
;
1416 f
.fm
.fm_length
= (int64_t)nb_sectors
* BDRV_SECTOR_SIZE
;
1418 f
.fm
.fm_extent_count
= 1;
1419 f
.fm
.fm_reserved
= 0;
1420 if (ioctl(s
->fd
, FS_IOC_FIEMAP
, &f
) == -1) {
1421 s
->skip_fiemap
= true;
1425 if (f
.fm
.fm_mapped_extents
== 0) {
1426 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1427 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1429 off_t length
= lseek(s
->fd
, 0, SEEK_END
);
1430 *hole
= f
.fm
.fm_start
;
1431 *data
= MIN(f
.fm
.fm_start
+ f
.fm
.fm_length
, length
);
1433 *data
= f
.fe
.fe_logical
;
1434 *hole
= f
.fe
.fe_logical
+ f
.fe
.fe_length
;
1435 if (f
.fe
.fe_flags
& FIEMAP_EXTENT_UNWRITTEN
) {
1436 ret
|= BDRV_BLOCK_ZERO
;
1446 static int64_t try_seek_hole(BlockDriverState
*bs
, off_t start
, off_t
*data
,
1447 off_t
*hole
, int *pnum
)
1449 #if defined SEEK_HOLE && defined SEEK_DATA
1450 BDRVRawState
*s
= bs
->opaque
;
1452 *hole
= lseek(s
->fd
, start
, SEEK_HOLE
);
1454 /* -ENXIO indicates that sector_num was past the end of the file.
1455 * There is a virtual hole there. */
1456 assert(errno
!= -ENXIO
);
1461 if (*hole
> start
) {
1464 /* On a hole. We need another syscall to find its end. */
1465 *data
= lseek(s
->fd
, start
, SEEK_DATA
);
1467 *data
= lseek(s
->fd
, 0, SEEK_END
);
1471 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1478 * Returns true iff the specified sector is present in the disk image. Drivers
1479 * not implementing the functionality are assumed to not support backing files,
1480 * hence all their sectors are reported as allocated.
1482 * If 'sector_num' is beyond the end of the disk image the return value is 0
1483 * and 'pnum' is set to 0.
1485 * 'pnum' is set to the number of sectors (including and immediately following
1486 * the specified sector) that are known to be in the same
1487 * allocated/unallocated state.
1489 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1490 * beyond the end of the disk image it will be clamped.
1492 static int64_t coroutine_fn
raw_co_get_block_status(BlockDriverState
*bs
,
1494 int nb_sectors
, int *pnum
)
1496 off_t start
, data
= 0, hole
= 0;
1504 start
= sector_num
* BDRV_SECTOR_SIZE
;
1506 ret
= try_fiemap(bs
, start
, &data
, &hole
, nb_sectors
, pnum
);
1508 ret
= try_seek_hole(bs
, start
, &data
, &hole
, pnum
);
1510 /* Assume everything is allocated. */
1512 hole
= start
+ nb_sectors
* BDRV_SECTOR_SIZE
;
1513 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1517 if (data
<= start
) {
1518 /* On a data extent, compute sectors to the end of the extent. */
1519 *pnum
= MIN(nb_sectors
, (hole
- start
) / BDRV_SECTOR_SIZE
);
1521 /* On a hole, compute sectors to the beginning of the next extent. */
1522 *pnum
= MIN(nb_sectors
, (data
- start
) / BDRV_SECTOR_SIZE
);
1523 ret
&= ~BDRV_BLOCK_DATA
;
1524 ret
|= BDRV_BLOCK_ZERO
;
1530 static coroutine_fn BlockDriverAIOCB
*raw_aio_discard(BlockDriverState
*bs
,
1531 int64_t sector_num
, int nb_sectors
,
1532 BlockDriverCompletionFunc
*cb
, void *opaque
)
1534 BDRVRawState
*s
= bs
->opaque
;
1536 return paio_submit(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1537 cb
, opaque
, QEMU_AIO_DISCARD
);
1540 static int coroutine_fn
raw_co_write_zeroes(
1541 BlockDriverState
*bs
, int64_t sector_num
,
1542 int nb_sectors
, BdrvRequestFlags flags
)
1544 BDRVRawState
*s
= bs
->opaque
;
1546 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1547 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1548 QEMU_AIO_WRITE_ZEROES
);
1549 } else if (s
->discard_zeroes
) {
1550 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1556 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1558 BDRVRawState
*s
= bs
->opaque
;
1560 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
1561 bdi
->can_write_zeroes_with_unmap
= s
->discard_zeroes
;
1565 static QemuOptsList raw_create_opts
= {
1566 .name
= "raw-create-opts",
1567 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
1570 .name
= BLOCK_OPT_SIZE
,
1571 .type
= QEMU_OPT_SIZE
,
1572 .help
= "Virtual disk size"
1575 .name
= BLOCK_OPT_NOCOW
,
1576 .type
= QEMU_OPT_BOOL
,
1577 .help
= "Turn off copy-on-write (valid only on btrfs)"
1579 { /* end of list */ }
1583 static BlockDriver bdrv_file
= {
1584 .format_name
= "file",
1585 .protocol_name
= "file",
1586 .instance_size
= sizeof(BDRVRawState
),
1587 .bdrv_needs_filename
= true,
1588 .bdrv_probe
= NULL
, /* no probe for protocols */
1589 .bdrv_parse_filename
= raw_parse_filename
,
1590 .bdrv_file_open
= raw_open
,
1591 .bdrv_reopen_prepare
= raw_reopen_prepare
,
1592 .bdrv_reopen_commit
= raw_reopen_commit
,
1593 .bdrv_reopen_abort
= raw_reopen_abort
,
1594 .bdrv_close
= raw_close
,
1595 .bdrv_create
= raw_create
,
1596 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1597 .bdrv_co_get_block_status
= raw_co_get_block_status
,
1598 .bdrv_co_write_zeroes
= raw_co_write_zeroes
,
1600 .bdrv_aio_readv
= raw_aio_readv
,
1601 .bdrv_aio_writev
= raw_aio_writev
,
1602 .bdrv_aio_flush
= raw_aio_flush
,
1603 .bdrv_aio_discard
= raw_aio_discard
,
1604 .bdrv_refresh_limits
= raw_refresh_limits
,
1605 .bdrv_io_plug
= raw_aio_plug
,
1606 .bdrv_io_unplug
= raw_aio_unplug
,
1607 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
1609 .bdrv_truncate
= raw_truncate
,
1610 .bdrv_getlength
= raw_getlength
,
1611 .bdrv_get_info
= raw_get_info
,
1612 .bdrv_get_allocated_file_size
1613 = raw_get_allocated_file_size
,
1615 .bdrv_detach_aio_context
= raw_detach_aio_context
,
1616 .bdrv_attach_aio_context
= raw_attach_aio_context
,
1618 .create_opts
= &raw_create_opts
,
1621 /***********************************************/
1624 #if defined(__APPLE__) && defined(__MACH__)
1625 static kern_return_t
FindEjectableCDMedia( io_iterator_t
*mediaIterator
);
1626 static kern_return_t
GetBSDPath( io_iterator_t mediaIterator
, char *bsdPath
, CFIndex maxPathSize
);
1628 kern_return_t
FindEjectableCDMedia( io_iterator_t
*mediaIterator
)
1630 kern_return_t kernResult
;
1631 mach_port_t masterPort
;
1632 CFMutableDictionaryRef classesToMatch
;
1634 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
1635 if ( KERN_SUCCESS
!= kernResult
) {
1636 printf( "IOMasterPort returned %d\n", kernResult
);
1639 classesToMatch
= IOServiceMatching( kIOCDMediaClass
);
1640 if ( classesToMatch
== NULL
) {
1641 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1643 CFDictionarySetValue( classesToMatch
, CFSTR( kIOMediaEjectableKey
), kCFBooleanTrue
);
1645 kernResult
= IOServiceGetMatchingServices( masterPort
, classesToMatch
, mediaIterator
);
1646 if ( KERN_SUCCESS
!= kernResult
)
1648 printf( "IOServiceGetMatchingServices returned %d\n", kernResult
);
1654 kern_return_t
GetBSDPath( io_iterator_t mediaIterator
, char *bsdPath
, CFIndex maxPathSize
)
1656 io_object_t nextMedia
;
1657 kern_return_t kernResult
= KERN_FAILURE
;
1659 nextMedia
= IOIteratorNext( mediaIterator
);
1662 CFTypeRef bsdPathAsCFString
;
1663 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
1664 if ( bsdPathAsCFString
) {
1665 size_t devPathLength
;
1666 strcpy( bsdPath
, _PATH_DEV
);
1667 strcat( bsdPath
, "r" );
1668 devPathLength
= strlen( bsdPath
);
1669 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
1670 kernResult
= KERN_SUCCESS
;
1672 CFRelease( bsdPathAsCFString
);
1674 IOObjectRelease( nextMedia
);
1682 static int hdev_probe_device(const char *filename
)
1686 /* allow a dedicated CD-ROM driver to match with a higher priority */
1687 if (strstart(filename
, "/dev/cdrom", NULL
))
1690 if (stat(filename
, &st
) >= 0 &&
1691 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
1698 static int check_hdev_writable(BDRVRawState
*s
)
1700 #if defined(BLKROGET)
1701 /* Linux block devices can be configured "read-only" using blockdev(8).
1702 * This is independent of device node permissions and therefore open(2)
1703 * with O_RDWR succeeds. Actual writes fail with EPERM.
1705 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1706 * check for read-only block devices so that Linux block devices behave
1712 if (fstat(s
->fd
, &st
)) {
1716 if (!S_ISBLK(st
.st_mode
)) {
1720 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
1727 #endif /* defined(BLKROGET) */
1731 static void hdev_parse_filename(const char *filename
, QDict
*options
,
1734 /* The prefix is optional, just as for "file". */
1735 strstart(filename
, "host_device:", &filename
);
1737 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
1740 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1743 BDRVRawState
*s
= bs
->opaque
;
1744 Error
*local_err
= NULL
;
1746 const char *filename
= qdict_get_str(options
, "filename");
1748 #if defined(__APPLE__) && defined(__MACH__)
1749 if (strstart(filename
, "/dev/cdrom", NULL
)) {
1750 kern_return_t kernResult
;
1751 io_iterator_t mediaIterator
;
1752 char bsdPath
[ MAXPATHLEN
];
1755 kernResult
= FindEjectableCDMedia( &mediaIterator
);
1756 kernResult
= GetBSDPath( mediaIterator
, bsdPath
, sizeof( bsdPath
) );
1758 if ( bsdPath
[ 0 ] != '\0' ) {
1759 strcat(bsdPath
,"s0");
1760 /* some CDs don't have a partition 0 */
1761 fd
= qemu_open(bsdPath
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
1763 bsdPath
[strlen(bsdPath
)-1] = '1';
1768 qdict_put(options
, "filename", qstring_from_str(filename
));
1771 if ( mediaIterator
)
1772 IOObjectRelease( mediaIterator
);
1776 s
->type
= FTYPE_FILE
;
1777 #if defined(__linux__)
1779 char resolved_path
[ MAXPATHLEN
], *temp
;
1781 temp
= realpath(filename
, resolved_path
);
1782 if (temp
&& strstart(temp
, "/dev/sg", NULL
)) {
1788 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
1791 error_propagate(errp
, local_err
);
1796 if (flags
& BDRV_O_RDWR
) {
1797 ret
= check_hdev_writable(s
);
1800 error_setg_errno(errp
, -ret
, "The device is not writable");
1808 #if defined(__linux__)
1809 /* Note: we do not have a reliable method to detect if the floppy is
1810 present. The current method is to try to open the floppy at every
1811 I/O and to keep it opened during a few hundreds of ms. */
1812 static int fd_open(BlockDriverState
*bs
)
1814 BDRVRawState
*s
= bs
->opaque
;
1815 int last_media_present
;
1817 if (s
->type
!= FTYPE_FD
)
1819 last_media_present
= (s
->fd
>= 0);
1821 (get_clock() - s
->fd_open_time
) >= FD_OPEN_TIMEOUT
) {
1825 printf("Floppy closed\n");
1829 if (s
->fd_got_error
&&
1830 (get_clock() - s
->fd_error_time
) < FD_OPEN_TIMEOUT
) {
1832 printf("No floppy (open delayed)\n");
1836 s
->fd
= qemu_open(bs
->filename
, s
->open_flags
& ~O_NONBLOCK
);
1838 s
->fd_error_time
= get_clock();
1839 s
->fd_got_error
= 1;
1840 if (last_media_present
)
1841 s
->fd_media_changed
= 1;
1843 printf("No floppy\n");
1848 printf("Floppy opened\n");
1851 if (!last_media_present
)
1852 s
->fd_media_changed
= 1;
1853 s
->fd_open_time
= get_clock();
1854 s
->fd_got_error
= 0;
1858 static int hdev_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
1860 BDRVRawState
*s
= bs
->opaque
;
1862 return ioctl(s
->fd
, req
, buf
);
1865 static BlockDriverAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
1866 unsigned long int req
, void *buf
,
1867 BlockDriverCompletionFunc
*cb
, void *opaque
)
1869 BDRVRawState
*s
= bs
->opaque
;
1870 RawPosixAIOData
*acb
;
1873 if (fd_open(bs
) < 0)
1876 acb
= g_slice_new(RawPosixAIOData
);
1878 acb
->aio_type
= QEMU_AIO_IOCTL
;
1879 acb
->aio_fildes
= s
->fd
;
1880 acb
->aio_offset
= 0;
1881 acb
->aio_ioctl_buf
= buf
;
1882 acb
->aio_ioctl_cmd
= req
;
1883 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1884 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1887 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1888 static int fd_open(BlockDriverState
*bs
)
1890 BDRVRawState
*s
= bs
->opaque
;
1892 /* this is just to ensure s->fd is sane (its called by io ops) */
1897 #else /* !linux && !FreeBSD */
1899 static int fd_open(BlockDriverState
*bs
)
1904 #endif /* !linux && !FreeBSD */
1906 static coroutine_fn BlockDriverAIOCB
*hdev_aio_discard(BlockDriverState
*bs
,
1907 int64_t sector_num
, int nb_sectors
,
1908 BlockDriverCompletionFunc
*cb
, void *opaque
)
1910 BDRVRawState
*s
= bs
->opaque
;
1912 if (fd_open(bs
) < 0) {
1915 return paio_submit(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1916 cb
, opaque
, QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
1919 static coroutine_fn
int hdev_co_write_zeroes(BlockDriverState
*bs
,
1920 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1922 BDRVRawState
*s
= bs
->opaque
;
1929 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1930 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1931 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
1932 } else if (s
->discard_zeroes
) {
1933 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1934 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
1939 static int hdev_create(const char *filename
, QemuOpts
*opts
,
1944 struct stat stat_buf
;
1945 int64_t total_size
= 0;
1948 /* This function is used by all three protocol block drivers and therefore
1949 * any of these three prefixes may be given.
1950 * The return value has to be stored somewhere, otherwise this is an error
1951 * due to -Werror=unused-value. */
1953 strstart(filename
, "host_device:", &filename
) ||
1954 strstart(filename
, "host_cdrom:" , &filename
) ||
1955 strstart(filename
, "host_floppy:", &filename
);
1959 /* Read out options */
1961 qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0) / BDRV_SECTOR_SIZE
;
1963 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
1966 error_setg_errno(errp
, -ret
, "Could not open device");
1970 if (fstat(fd
, &stat_buf
) < 0) {
1972 error_setg_errno(errp
, -ret
, "Could not stat device");
1973 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
1975 "The given file is neither a block nor a character device");
1977 } else if (lseek(fd
, 0, SEEK_END
) < total_size
* BDRV_SECTOR_SIZE
) {
1978 error_setg(errp
, "Device is too small");
1986 static BlockDriver bdrv_host_device
= {
1987 .format_name
= "host_device",
1988 .protocol_name
= "host_device",
1989 .instance_size
= sizeof(BDRVRawState
),
1990 .bdrv_needs_filename
= true,
1991 .bdrv_probe_device
= hdev_probe_device
,
1992 .bdrv_parse_filename
= hdev_parse_filename
,
1993 .bdrv_file_open
= hdev_open
,
1994 .bdrv_close
= raw_close
,
1995 .bdrv_reopen_prepare
= raw_reopen_prepare
,
1996 .bdrv_reopen_commit
= raw_reopen_commit
,
1997 .bdrv_reopen_abort
= raw_reopen_abort
,
1998 .bdrv_create
= hdev_create
,
1999 .create_opts
= &raw_create_opts
,
2000 .bdrv_co_write_zeroes
= hdev_co_write_zeroes
,
2002 .bdrv_aio_readv
= raw_aio_readv
,
2003 .bdrv_aio_writev
= raw_aio_writev
,
2004 .bdrv_aio_flush
= raw_aio_flush
,
2005 .bdrv_aio_discard
= hdev_aio_discard
,
2006 .bdrv_refresh_limits
= raw_refresh_limits
,
2007 .bdrv_io_plug
= raw_aio_plug
,
2008 .bdrv_io_unplug
= raw_aio_unplug
,
2009 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2011 .bdrv_truncate
= raw_truncate
,
2012 .bdrv_getlength
= raw_getlength
,
2013 .bdrv_get_info
= raw_get_info
,
2014 .bdrv_get_allocated_file_size
2015 = raw_get_allocated_file_size
,
2017 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2018 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2020 /* generic scsi device */
2022 .bdrv_ioctl
= hdev_ioctl
,
2023 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2028 static void floppy_parse_filename(const char *filename
, QDict
*options
,
2031 /* The prefix is optional, just as for "file". */
2032 strstart(filename
, "host_floppy:", &filename
);
2034 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
2037 static int floppy_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2040 BDRVRawState
*s
= bs
->opaque
;
2041 Error
*local_err
= NULL
;
2046 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2047 ret
= raw_open_common(bs
, options
, flags
, O_NONBLOCK
, &local_err
);
2050 error_propagate(errp
, local_err
);
2055 /* close fd so that we can reopen it as needed */
2058 s
->fd_media_changed
= 1;
2063 static int floppy_probe_device(const char *filename
)
2067 struct floppy_struct fdparam
;
2070 if (strstart(filename
, "/dev/fd", NULL
) &&
2071 !strstart(filename
, "/dev/fdset/", NULL
)) {
2075 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
2079 ret
= fstat(fd
, &st
);
2080 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2084 /* Attempt to detect via a floppy specific ioctl */
2085 ret
= ioctl(fd
, FDGETPRM
, &fdparam
);
2096 static int floppy_is_inserted(BlockDriverState
*bs
)
2098 return fd_open(bs
) >= 0;
2101 static int floppy_media_changed(BlockDriverState
*bs
)
2103 BDRVRawState
*s
= bs
->opaque
;
2107 * XXX: we do not have a true media changed indication.
2108 * It does not work if the floppy is changed without trying to read it.
2111 ret
= s
->fd_media_changed
;
2112 s
->fd_media_changed
= 0;
2114 printf("Floppy changed=%d\n", ret
);
2119 static void floppy_eject(BlockDriverState
*bs
, bool eject_flag
)
2121 BDRVRawState
*s
= bs
->opaque
;
2128 fd
= qemu_open(bs
->filename
, s
->open_flags
| O_NONBLOCK
);
2130 if (ioctl(fd
, FDEJECT
, 0) < 0)
2136 static BlockDriver bdrv_host_floppy
= {
2137 .format_name
= "host_floppy",
2138 .protocol_name
= "host_floppy",
2139 .instance_size
= sizeof(BDRVRawState
),
2140 .bdrv_needs_filename
= true,
2141 .bdrv_probe_device
= floppy_probe_device
,
2142 .bdrv_parse_filename
= floppy_parse_filename
,
2143 .bdrv_file_open
= floppy_open
,
2144 .bdrv_close
= raw_close
,
2145 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2146 .bdrv_reopen_commit
= raw_reopen_commit
,
2147 .bdrv_reopen_abort
= raw_reopen_abort
,
2148 .bdrv_create
= hdev_create
,
2149 .create_opts
= &raw_create_opts
,
2151 .bdrv_aio_readv
= raw_aio_readv
,
2152 .bdrv_aio_writev
= raw_aio_writev
,
2153 .bdrv_aio_flush
= raw_aio_flush
,
2154 .bdrv_refresh_limits
= raw_refresh_limits
,
2155 .bdrv_io_plug
= raw_aio_plug
,
2156 .bdrv_io_unplug
= raw_aio_unplug
,
2157 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2159 .bdrv_truncate
= raw_truncate
,
2160 .bdrv_getlength
= raw_getlength
,
2161 .has_variable_length
= true,
2162 .bdrv_get_allocated_file_size
2163 = raw_get_allocated_file_size
,
2165 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2166 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2168 /* removable device support */
2169 .bdrv_is_inserted
= floppy_is_inserted
,
2170 .bdrv_media_changed
= floppy_media_changed
,
2171 .bdrv_eject
= floppy_eject
,
2175 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2176 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
2179 /* The prefix is optional, just as for "file". */
2180 strstart(filename
, "host_cdrom:", &filename
);
2182 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
2187 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2190 BDRVRawState
*s
= bs
->opaque
;
2191 Error
*local_err
= NULL
;
2196 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2197 ret
= raw_open_common(bs
, options
, flags
, O_NONBLOCK
, &local_err
);
2199 error_propagate(errp
, local_err
);
2204 static int cdrom_probe_device(const char *filename
)
2210 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
2214 ret
= fstat(fd
, &st
);
2215 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2219 /* Attempt to detect via a CDROM specific ioctl */
2220 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2230 static int cdrom_is_inserted(BlockDriverState
*bs
)
2232 BDRVRawState
*s
= bs
->opaque
;
2235 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2236 if (ret
== CDS_DISC_OK
)
2241 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2243 BDRVRawState
*s
= bs
->opaque
;
2246 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
2247 perror("CDROMEJECT");
2249 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
2250 perror("CDROMEJECT");
2254 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2256 BDRVRawState
*s
= bs
->opaque
;
2258 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
2260 * Note: an error can happen if the distribution automatically
2263 /* perror("CDROM_LOCKDOOR"); */
2267 static BlockDriver bdrv_host_cdrom
= {
2268 .format_name
= "host_cdrom",
2269 .protocol_name
= "host_cdrom",
2270 .instance_size
= sizeof(BDRVRawState
),
2271 .bdrv_needs_filename
= true,
2272 .bdrv_probe_device
= cdrom_probe_device
,
2273 .bdrv_parse_filename
= cdrom_parse_filename
,
2274 .bdrv_file_open
= cdrom_open
,
2275 .bdrv_close
= raw_close
,
2276 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2277 .bdrv_reopen_commit
= raw_reopen_commit
,
2278 .bdrv_reopen_abort
= raw_reopen_abort
,
2279 .bdrv_create
= hdev_create
,
2280 .create_opts
= &raw_create_opts
,
2282 .bdrv_aio_readv
= raw_aio_readv
,
2283 .bdrv_aio_writev
= raw_aio_writev
,
2284 .bdrv_aio_flush
= raw_aio_flush
,
2285 .bdrv_refresh_limits
= raw_refresh_limits
,
2286 .bdrv_io_plug
= raw_aio_plug
,
2287 .bdrv_io_unplug
= raw_aio_unplug
,
2288 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2290 .bdrv_truncate
= raw_truncate
,
2291 .bdrv_getlength
= raw_getlength
,
2292 .has_variable_length
= true,
2293 .bdrv_get_allocated_file_size
2294 = raw_get_allocated_file_size
,
2296 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2297 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2299 /* removable device support */
2300 .bdrv_is_inserted
= cdrom_is_inserted
,
2301 .bdrv_eject
= cdrom_eject
,
2302 .bdrv_lock_medium
= cdrom_lock_medium
,
2304 /* generic scsi device */
2305 .bdrv_ioctl
= hdev_ioctl
,
2306 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2308 #endif /* __linux__ */
2310 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2311 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2314 BDRVRawState
*s
= bs
->opaque
;
2315 Error
*local_err
= NULL
;
2320 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2323 error_propagate(errp
, local_err
);
2328 /* make sure the door isn't locked at this time */
2329 ioctl(s
->fd
, CDIOCALLOW
);
2333 static int cdrom_probe_device(const char *filename
)
2335 if (strstart(filename
, "/dev/cd", NULL
) ||
2336 strstart(filename
, "/dev/acd", NULL
))
2341 static int cdrom_reopen(BlockDriverState
*bs
)
2343 BDRVRawState
*s
= bs
->opaque
;
2347 * Force reread of possibly changed/newly loaded disc,
2348 * FreeBSD seems to not notice sometimes...
2352 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
2359 /* make sure the door isn't locked at this time */
2360 ioctl(s
->fd
, CDIOCALLOW
);
2364 static int cdrom_is_inserted(BlockDriverState
*bs
)
2366 return raw_getlength(bs
) > 0;
2369 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2371 BDRVRawState
*s
= bs
->opaque
;
2376 (void) ioctl(s
->fd
, CDIOCALLOW
);
2379 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
2380 perror("CDIOCEJECT");
2382 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
2383 perror("CDIOCCLOSE");
2389 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2391 BDRVRawState
*s
= bs
->opaque
;
2395 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
2397 * Note: an error can happen if the distribution automatically
2400 /* perror("CDROM_LOCKDOOR"); */
2404 static BlockDriver bdrv_host_cdrom
= {
2405 .format_name
= "host_cdrom",
2406 .protocol_name
= "host_cdrom",
2407 .instance_size
= sizeof(BDRVRawState
),
2408 .bdrv_needs_filename
= true,
2409 .bdrv_probe_device
= cdrom_probe_device
,
2410 .bdrv_parse_filename
= cdrom_parse_filename
,
2411 .bdrv_file_open
= cdrom_open
,
2412 .bdrv_close
= raw_close
,
2413 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2414 .bdrv_reopen_commit
= raw_reopen_commit
,
2415 .bdrv_reopen_abort
= raw_reopen_abort
,
2416 .bdrv_create
= hdev_create
,
2417 .create_opts
= &raw_create_opts
,
2419 .bdrv_aio_readv
= raw_aio_readv
,
2420 .bdrv_aio_writev
= raw_aio_writev
,
2421 .bdrv_aio_flush
= raw_aio_flush
,
2422 .bdrv_refresh_limits
= raw_refresh_limits
,
2423 .bdrv_io_plug
= raw_aio_plug
,
2424 .bdrv_io_unplug
= raw_aio_unplug
,
2425 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2427 .bdrv_truncate
= raw_truncate
,
2428 .bdrv_getlength
= raw_getlength
,
2429 .has_variable_length
= true,
2430 .bdrv_get_allocated_file_size
2431 = raw_get_allocated_file_size
,
2433 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2434 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2436 /* removable device support */
2437 .bdrv_is_inserted
= cdrom_is_inserted
,
2438 .bdrv_eject
= cdrom_eject
,
2439 .bdrv_lock_medium
= cdrom_lock_medium
,
2441 #endif /* __FreeBSD__ */
2443 static void bdrv_file_init(void)
2446 * Register all the drivers. Note that order is important, the driver
2447 * registered last will get probed first.
2449 bdrv_register(&bdrv_file
);
2450 bdrv_register(&bdrv_host_device
);
2452 bdrv_register(&bdrv_host_floppy
);
2453 bdrv_register(&bdrv_host_cdrom
);
2455 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2456 bdrv_register(&bdrv_host_cdrom
);
2460 block_init(bdrv_file_init
);