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_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
802 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
806 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
807 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
808 p
+= aiocb
->aio_iov
[i
].iov_len
;
810 assert(p
- buf
== aiocb
->aio_nbytes
);
813 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
814 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
816 size_t count
= aiocb
->aio_nbytes
, copy
;
819 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
821 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
822 copy
= aiocb
->aio_iov
[i
].iov_len
;
824 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
825 assert(count
>= copy
);
837 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
839 struct xfs_flock64 fl
;
841 memset(&fl
, 0, sizeof(fl
));
842 fl
.l_whence
= SEEK_SET
;
846 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
847 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno
));
854 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
856 struct xfs_flock64 fl
;
858 memset(&fl
, 0, sizeof(fl
));
859 fl
.l_whence
= SEEK_SET
;
863 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
864 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno
));
872 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
874 int ret
= -EOPNOTSUPP
;
875 BDRVRawState
*s
= aiocb
->bs
->opaque
;
877 if (s
->has_write_zeroes
== 0) {
881 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
884 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
885 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
888 } while (errno
== EINTR
);
895 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
900 if (ret
== -ENODEV
|| ret
== -ENOSYS
|| ret
== -EOPNOTSUPP
||
902 s
->has_write_zeroes
= false;
908 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
910 int ret
= -EOPNOTSUPP
;
911 BDRVRawState
*s
= aiocb
->bs
->opaque
;
913 if (!s
->has_discard
) {
917 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
920 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
921 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
924 } while (errno
== EINTR
);
931 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
935 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
937 if (fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
938 aiocb
->aio_offset
, aiocb
->aio_nbytes
) == 0) {
941 } while (errno
== EINTR
);
947 if (ret
== -ENODEV
|| ret
== -ENOSYS
|| ret
== -EOPNOTSUPP
||
949 s
->has_discard
= false;
955 static int aio_worker(void *arg
)
957 RawPosixAIOData
*aiocb
= arg
;
960 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
962 ret
= handle_aiocb_rw(aiocb
);
963 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
&& aiocb
->bs
->growable
) {
964 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
965 0, aiocb
->aio_nbytes
- ret
);
967 ret
= aiocb
->aio_nbytes
;
969 if (ret
== aiocb
->aio_nbytes
) {
971 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
976 ret
= handle_aiocb_rw(aiocb
);
977 if (ret
== aiocb
->aio_nbytes
) {
979 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
984 ret
= handle_aiocb_flush(aiocb
);
987 ret
= handle_aiocb_ioctl(aiocb
);
989 case QEMU_AIO_DISCARD
:
990 ret
= handle_aiocb_discard(aiocb
);
992 case QEMU_AIO_WRITE_ZEROES
:
993 ret
= handle_aiocb_write_zeroes(aiocb
);
996 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1001 g_slice_free(RawPosixAIOData
, aiocb
);
1005 static int paio_submit_co(BlockDriverState
*bs
, int fd
,
1006 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1009 RawPosixAIOData
*acb
= g_slice_new(RawPosixAIOData
);
1013 acb
->aio_type
= type
;
1014 acb
->aio_fildes
= fd
;
1016 acb
->aio_nbytes
= nb_sectors
* BDRV_SECTOR_SIZE
;
1017 acb
->aio_offset
= sector_num
* BDRV_SECTOR_SIZE
;
1020 acb
->aio_iov
= qiov
->iov
;
1021 acb
->aio_niov
= qiov
->niov
;
1022 assert(qiov
->size
== acb
->aio_nbytes
);
1025 trace_paio_submit_co(sector_num
, nb_sectors
, type
);
1026 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1027 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1030 static BlockDriverAIOCB
*paio_submit(BlockDriverState
*bs
, int fd
,
1031 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1032 BlockDriverCompletionFunc
*cb
, void *opaque
, int type
)
1034 RawPosixAIOData
*acb
= g_slice_new(RawPosixAIOData
);
1038 acb
->aio_type
= type
;
1039 acb
->aio_fildes
= fd
;
1041 acb
->aio_nbytes
= nb_sectors
* BDRV_SECTOR_SIZE
;
1042 acb
->aio_offset
= sector_num
* BDRV_SECTOR_SIZE
;
1045 acb
->aio_iov
= qiov
->iov
;
1046 acb
->aio_niov
= qiov
->niov
;
1047 assert(qiov
->size
== acb
->aio_nbytes
);
1050 trace_paio_submit(acb
, opaque
, sector_num
, nb_sectors
, type
);
1051 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1052 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1055 static BlockDriverAIOCB
*raw_aio_submit(BlockDriverState
*bs
,
1056 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1057 BlockDriverCompletionFunc
*cb
, void *opaque
, int type
)
1059 BDRVRawState
*s
= bs
->opaque
;
1061 if (fd_open(bs
) < 0)
1065 * If O_DIRECT is used the buffer needs to be aligned on a sector
1066 * boundary. Check if this is the case or tell the low-level
1067 * driver that it needs to copy the buffer.
1069 if ((bs
->open_flags
& BDRV_O_NOCACHE
)) {
1070 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1071 type
|= QEMU_AIO_MISALIGNED
;
1072 #ifdef CONFIG_LINUX_AIO
1073 } else if (s
->use_aio
) {
1074 return laio_submit(bs
, s
->aio_ctx
, s
->fd
, sector_num
, qiov
,
1075 nb_sectors
, cb
, opaque
, type
);
1080 return paio_submit(bs
, s
->fd
, sector_num
, qiov
, nb_sectors
,
1084 static void raw_aio_plug(BlockDriverState
*bs
)
1086 #ifdef CONFIG_LINUX_AIO
1087 BDRVRawState
*s
= bs
->opaque
;
1089 laio_io_plug(bs
, s
->aio_ctx
);
1094 static void raw_aio_unplug(BlockDriverState
*bs
)
1096 #ifdef CONFIG_LINUX_AIO
1097 BDRVRawState
*s
= bs
->opaque
;
1099 laio_io_unplug(bs
, s
->aio_ctx
, true);
1104 static void raw_aio_flush_io_queue(BlockDriverState
*bs
)
1106 #ifdef CONFIG_LINUX_AIO
1107 BDRVRawState
*s
= bs
->opaque
;
1109 laio_io_unplug(bs
, s
->aio_ctx
, false);
1114 static BlockDriverAIOCB
*raw_aio_readv(BlockDriverState
*bs
,
1115 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1116 BlockDriverCompletionFunc
*cb
, void *opaque
)
1118 return raw_aio_submit(bs
, sector_num
, qiov
, nb_sectors
,
1119 cb
, opaque
, QEMU_AIO_READ
);
1122 static BlockDriverAIOCB
*raw_aio_writev(BlockDriverState
*bs
,
1123 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
1124 BlockDriverCompletionFunc
*cb
, void *opaque
)
1126 return raw_aio_submit(bs
, sector_num
, qiov
, nb_sectors
,
1127 cb
, opaque
, QEMU_AIO_WRITE
);
1130 static BlockDriverAIOCB
*raw_aio_flush(BlockDriverState
*bs
,
1131 BlockDriverCompletionFunc
*cb
, void *opaque
)
1133 BDRVRawState
*s
= bs
->opaque
;
1135 if (fd_open(bs
) < 0)
1138 return paio_submit(bs
, s
->fd
, 0, NULL
, 0, cb
, opaque
, QEMU_AIO_FLUSH
);
1141 static void raw_close(BlockDriverState
*bs
)
1143 BDRVRawState
*s
= bs
->opaque
;
1145 raw_detach_aio_context(bs
);
1147 #ifdef CONFIG_LINUX_AIO
1149 laio_cleanup(s
->aio_ctx
);
1158 static int raw_truncate(BlockDriverState
*bs
, int64_t offset
)
1160 BDRVRawState
*s
= bs
->opaque
;
1163 if (fstat(s
->fd
, &st
)) {
1167 if (S_ISREG(st
.st_mode
)) {
1168 if (ftruncate(s
->fd
, offset
) < 0) {
1171 } else if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1172 if (offset
> raw_getlength(bs
)) {
1183 static int64_t raw_getlength(BlockDriverState
*bs
)
1185 BDRVRawState
*s
= bs
->opaque
;
1191 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1192 struct disklabel dl
;
1194 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1196 return (uint64_t)dl
.d_secsize
*
1197 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1201 #elif defined(__NetBSD__)
1202 static int64_t raw_getlength(BlockDriverState
*bs
)
1204 BDRVRawState
*s
= bs
->opaque
;
1210 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1211 struct dkwedge_info dkw
;
1213 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1214 return dkw
.dkw_size
* 512;
1216 struct disklabel dl
;
1218 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1220 return (uint64_t)dl
.d_secsize
*
1221 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1226 #elif defined(__sun__)
1227 static int64_t raw_getlength(BlockDriverState
*bs
)
1229 BDRVRawState
*s
= bs
->opaque
;
1230 struct dk_minfo minfo
;
1240 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1242 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
1244 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
1248 * There are reports that lseek on some devices fails, but
1249 * irc discussion said that contingency on contingency was overkill.
1251 size
= lseek(s
->fd
, 0, SEEK_END
);
1257 #elif defined(CONFIG_BSD)
1258 static int64_t raw_getlength(BlockDriverState
*bs
)
1260 BDRVRawState
*s
= bs
->opaque
;
1264 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1273 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1276 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
1277 #ifdef DIOCGMEDIASIZE
1278 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
1279 #elif defined(DIOCGPART)
1282 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
1283 size
= pi
.media_size
;
1289 #if defined(__APPLE__) && defined(__MACH__)
1292 size
= lseek(fd
, 0LL, SEEK_END
);
1297 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1300 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1301 if (size
== 2048LL * (unsigned)-1)
1303 /* XXX no disc? maybe we need to reopen... */
1304 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
1311 size
= lseek(fd
, 0, SEEK_END
);
1319 static int64_t raw_getlength(BlockDriverState
*bs
)
1321 BDRVRawState
*s
= bs
->opaque
;
1330 size
= lseek(s
->fd
, 0, SEEK_END
);
1338 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
1341 BDRVRawState
*s
= bs
->opaque
;
1343 if (fstat(s
->fd
, &st
) < 0) {
1346 return (int64_t)st
.st_blocks
* 512;
1349 static int raw_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1353 int64_t total_size
= 0;
1356 strstart(filename
, "file:", &filename
);
1358 /* Read out options */
1360 qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0) / BDRV_SECTOR_SIZE
;
1361 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
1363 fd
= qemu_open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
,
1367 error_setg_errno(errp
, -result
, "Could not create file");
1371 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1372 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1373 * will be ignored since any failure of this operation should not
1374 * block the left work.
1377 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
1378 attr
|= FS_NOCOW_FL
;
1379 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
1384 if (ftruncate(fd
, total_size
* BDRV_SECTOR_SIZE
) != 0) {
1386 error_setg_errno(errp
, -result
, "Could not resize file");
1388 if (qemu_close(fd
) != 0) {
1390 error_setg_errno(errp
, -result
, "Could not close the new file");
1396 static int64_t try_fiemap(BlockDriverState
*bs
, off_t start
, off_t
*data
,
1397 off_t
*hole
, int nb_sectors
, int *pnum
)
1399 #ifdef CONFIG_FIEMAP
1400 BDRVRawState
*s
= bs
->opaque
;
1401 int64_t ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1404 struct fiemap_extent fe
;
1407 if (s
->skip_fiemap
) {
1411 f
.fm
.fm_start
= start
;
1412 f
.fm
.fm_length
= (int64_t)nb_sectors
* BDRV_SECTOR_SIZE
;
1414 f
.fm
.fm_extent_count
= 1;
1415 f
.fm
.fm_reserved
= 0;
1416 if (ioctl(s
->fd
, FS_IOC_FIEMAP
, &f
) == -1) {
1417 s
->skip_fiemap
= true;
1421 if (f
.fm
.fm_mapped_extents
== 0) {
1422 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1423 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1425 off_t length
= lseek(s
->fd
, 0, SEEK_END
);
1426 *hole
= f
.fm
.fm_start
;
1427 *data
= MIN(f
.fm
.fm_start
+ f
.fm
.fm_length
, length
);
1429 *data
= f
.fe
.fe_logical
;
1430 *hole
= f
.fe
.fe_logical
+ f
.fe
.fe_length
;
1431 if (f
.fe
.fe_flags
& FIEMAP_EXTENT_UNWRITTEN
) {
1432 ret
|= BDRV_BLOCK_ZERO
;
1442 static int64_t try_seek_hole(BlockDriverState
*bs
, off_t start
, off_t
*data
,
1443 off_t
*hole
, int *pnum
)
1445 #if defined SEEK_HOLE && defined SEEK_DATA
1446 BDRVRawState
*s
= bs
->opaque
;
1448 *hole
= lseek(s
->fd
, start
, SEEK_HOLE
);
1450 /* -ENXIO indicates that sector_num was past the end of the file.
1451 * There is a virtual hole there. */
1452 assert(errno
!= -ENXIO
);
1457 if (*hole
> start
) {
1460 /* On a hole. We need another syscall to find its end. */
1461 *data
= lseek(s
->fd
, start
, SEEK_DATA
);
1463 *data
= lseek(s
->fd
, 0, SEEK_END
);
1467 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1474 * Returns true iff the specified sector is present in the disk image. Drivers
1475 * not implementing the functionality are assumed to not support backing files,
1476 * hence all their sectors are reported as allocated.
1478 * If 'sector_num' is beyond the end of the disk image the return value is 0
1479 * and 'pnum' is set to 0.
1481 * 'pnum' is set to the number of sectors (including and immediately following
1482 * the specified sector) that are known to be in the same
1483 * allocated/unallocated state.
1485 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1486 * beyond the end of the disk image it will be clamped.
1488 static int64_t coroutine_fn
raw_co_get_block_status(BlockDriverState
*bs
,
1490 int nb_sectors
, int *pnum
)
1492 off_t start
, data
= 0, hole
= 0;
1500 start
= sector_num
* BDRV_SECTOR_SIZE
;
1502 ret
= try_fiemap(bs
, start
, &data
, &hole
, nb_sectors
, pnum
);
1504 ret
= try_seek_hole(bs
, start
, &data
, &hole
, pnum
);
1506 /* Assume everything is allocated. */
1508 hole
= start
+ nb_sectors
* BDRV_SECTOR_SIZE
;
1509 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| start
;
1513 if (data
<= start
) {
1514 /* On a data extent, compute sectors to the end of the extent. */
1515 *pnum
= MIN(nb_sectors
, (hole
- start
) / BDRV_SECTOR_SIZE
);
1517 /* On a hole, compute sectors to the beginning of the next extent. */
1518 *pnum
= MIN(nb_sectors
, (data
- start
) / BDRV_SECTOR_SIZE
);
1519 ret
&= ~BDRV_BLOCK_DATA
;
1520 ret
|= BDRV_BLOCK_ZERO
;
1526 static coroutine_fn BlockDriverAIOCB
*raw_aio_discard(BlockDriverState
*bs
,
1527 int64_t sector_num
, int nb_sectors
,
1528 BlockDriverCompletionFunc
*cb
, void *opaque
)
1530 BDRVRawState
*s
= bs
->opaque
;
1532 return paio_submit(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1533 cb
, opaque
, QEMU_AIO_DISCARD
);
1536 static int coroutine_fn
raw_co_write_zeroes(
1537 BlockDriverState
*bs
, int64_t sector_num
,
1538 int nb_sectors
, BdrvRequestFlags flags
)
1540 BDRVRawState
*s
= bs
->opaque
;
1542 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1543 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1544 QEMU_AIO_WRITE_ZEROES
);
1545 } else if (s
->discard_zeroes
) {
1546 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1552 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1554 BDRVRawState
*s
= bs
->opaque
;
1556 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
1557 bdi
->can_write_zeroes_with_unmap
= s
->discard_zeroes
;
1561 static QemuOptsList raw_create_opts
= {
1562 .name
= "raw-create-opts",
1563 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
1566 .name
= BLOCK_OPT_SIZE
,
1567 .type
= QEMU_OPT_SIZE
,
1568 .help
= "Virtual disk size"
1571 .name
= BLOCK_OPT_NOCOW
,
1572 .type
= QEMU_OPT_BOOL
,
1573 .help
= "Turn off copy-on-write (valid only on btrfs)"
1575 { /* end of list */ }
1579 static BlockDriver bdrv_file
= {
1580 .format_name
= "file",
1581 .protocol_name
= "file",
1582 .instance_size
= sizeof(BDRVRawState
),
1583 .bdrv_needs_filename
= true,
1584 .bdrv_probe
= NULL
, /* no probe for protocols */
1585 .bdrv_parse_filename
= raw_parse_filename
,
1586 .bdrv_file_open
= raw_open
,
1587 .bdrv_reopen_prepare
= raw_reopen_prepare
,
1588 .bdrv_reopen_commit
= raw_reopen_commit
,
1589 .bdrv_reopen_abort
= raw_reopen_abort
,
1590 .bdrv_close
= raw_close
,
1591 .bdrv_create
= raw_create
,
1592 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1593 .bdrv_co_get_block_status
= raw_co_get_block_status
,
1594 .bdrv_co_write_zeroes
= raw_co_write_zeroes
,
1596 .bdrv_aio_readv
= raw_aio_readv
,
1597 .bdrv_aio_writev
= raw_aio_writev
,
1598 .bdrv_aio_flush
= raw_aio_flush
,
1599 .bdrv_aio_discard
= raw_aio_discard
,
1600 .bdrv_refresh_limits
= raw_refresh_limits
,
1601 .bdrv_io_plug
= raw_aio_plug
,
1602 .bdrv_io_unplug
= raw_aio_unplug
,
1603 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
1605 .bdrv_truncate
= raw_truncate
,
1606 .bdrv_getlength
= raw_getlength
,
1607 .bdrv_get_info
= raw_get_info
,
1608 .bdrv_get_allocated_file_size
1609 = raw_get_allocated_file_size
,
1611 .bdrv_detach_aio_context
= raw_detach_aio_context
,
1612 .bdrv_attach_aio_context
= raw_attach_aio_context
,
1614 .create_opts
= &raw_create_opts
,
1617 /***********************************************/
1620 #if defined(__APPLE__) && defined(__MACH__)
1621 static kern_return_t
FindEjectableCDMedia( io_iterator_t
*mediaIterator
);
1622 static kern_return_t
GetBSDPath( io_iterator_t mediaIterator
, char *bsdPath
, CFIndex maxPathSize
);
1624 kern_return_t
FindEjectableCDMedia( io_iterator_t
*mediaIterator
)
1626 kern_return_t kernResult
;
1627 mach_port_t masterPort
;
1628 CFMutableDictionaryRef classesToMatch
;
1630 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
1631 if ( KERN_SUCCESS
!= kernResult
) {
1632 printf( "IOMasterPort returned %d\n", kernResult
);
1635 classesToMatch
= IOServiceMatching( kIOCDMediaClass
);
1636 if ( classesToMatch
== NULL
) {
1637 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1639 CFDictionarySetValue( classesToMatch
, CFSTR( kIOMediaEjectableKey
), kCFBooleanTrue
);
1641 kernResult
= IOServiceGetMatchingServices( masterPort
, classesToMatch
, mediaIterator
);
1642 if ( KERN_SUCCESS
!= kernResult
)
1644 printf( "IOServiceGetMatchingServices returned %d\n", kernResult
);
1650 kern_return_t
GetBSDPath( io_iterator_t mediaIterator
, char *bsdPath
, CFIndex maxPathSize
)
1652 io_object_t nextMedia
;
1653 kern_return_t kernResult
= KERN_FAILURE
;
1655 nextMedia
= IOIteratorNext( mediaIterator
);
1658 CFTypeRef bsdPathAsCFString
;
1659 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
1660 if ( bsdPathAsCFString
) {
1661 size_t devPathLength
;
1662 strcpy( bsdPath
, _PATH_DEV
);
1663 strcat( bsdPath
, "r" );
1664 devPathLength
= strlen( bsdPath
);
1665 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
1666 kernResult
= KERN_SUCCESS
;
1668 CFRelease( bsdPathAsCFString
);
1670 IOObjectRelease( nextMedia
);
1678 static int hdev_probe_device(const char *filename
)
1682 /* allow a dedicated CD-ROM driver to match with a higher priority */
1683 if (strstart(filename
, "/dev/cdrom", NULL
))
1686 if (stat(filename
, &st
) >= 0 &&
1687 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
1694 static int check_hdev_writable(BDRVRawState
*s
)
1696 #if defined(BLKROGET)
1697 /* Linux block devices can be configured "read-only" using blockdev(8).
1698 * This is independent of device node permissions and therefore open(2)
1699 * with O_RDWR succeeds. Actual writes fail with EPERM.
1701 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1702 * check for read-only block devices so that Linux block devices behave
1708 if (fstat(s
->fd
, &st
)) {
1712 if (!S_ISBLK(st
.st_mode
)) {
1716 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
1723 #endif /* defined(BLKROGET) */
1727 static void hdev_parse_filename(const char *filename
, QDict
*options
,
1730 /* The prefix is optional, just as for "file". */
1731 strstart(filename
, "host_device:", &filename
);
1733 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
1736 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1739 BDRVRawState
*s
= bs
->opaque
;
1740 Error
*local_err
= NULL
;
1742 const char *filename
= qdict_get_str(options
, "filename");
1744 #if defined(__APPLE__) && defined(__MACH__)
1745 if (strstart(filename
, "/dev/cdrom", NULL
)) {
1746 kern_return_t kernResult
;
1747 io_iterator_t mediaIterator
;
1748 char bsdPath
[ MAXPATHLEN
];
1751 kernResult
= FindEjectableCDMedia( &mediaIterator
);
1752 kernResult
= GetBSDPath( mediaIterator
, bsdPath
, sizeof( bsdPath
) );
1754 if ( bsdPath
[ 0 ] != '\0' ) {
1755 strcat(bsdPath
,"s0");
1756 /* some CDs don't have a partition 0 */
1757 fd
= qemu_open(bsdPath
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
1759 bsdPath
[strlen(bsdPath
)-1] = '1';
1764 qdict_put(options
, "filename", qstring_from_str(filename
));
1767 if ( mediaIterator
)
1768 IOObjectRelease( mediaIterator
);
1772 s
->type
= FTYPE_FILE
;
1773 #if defined(__linux__)
1775 char resolved_path
[ MAXPATHLEN
], *temp
;
1777 temp
= realpath(filename
, resolved_path
);
1778 if (temp
&& strstart(temp
, "/dev/sg", NULL
)) {
1784 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
1787 error_propagate(errp
, local_err
);
1792 if (flags
& BDRV_O_RDWR
) {
1793 ret
= check_hdev_writable(s
);
1796 error_setg_errno(errp
, -ret
, "The device is not writable");
1804 #if defined(__linux__)
1805 /* Note: we do not have a reliable method to detect if the floppy is
1806 present. The current method is to try to open the floppy at every
1807 I/O and to keep it opened during a few hundreds of ms. */
1808 static int fd_open(BlockDriverState
*bs
)
1810 BDRVRawState
*s
= bs
->opaque
;
1811 int last_media_present
;
1813 if (s
->type
!= FTYPE_FD
)
1815 last_media_present
= (s
->fd
>= 0);
1817 (get_clock() - s
->fd_open_time
) >= FD_OPEN_TIMEOUT
) {
1821 printf("Floppy closed\n");
1825 if (s
->fd_got_error
&&
1826 (get_clock() - s
->fd_error_time
) < FD_OPEN_TIMEOUT
) {
1828 printf("No floppy (open delayed)\n");
1832 s
->fd
= qemu_open(bs
->filename
, s
->open_flags
& ~O_NONBLOCK
);
1834 s
->fd_error_time
= get_clock();
1835 s
->fd_got_error
= 1;
1836 if (last_media_present
)
1837 s
->fd_media_changed
= 1;
1839 printf("No floppy\n");
1844 printf("Floppy opened\n");
1847 if (!last_media_present
)
1848 s
->fd_media_changed
= 1;
1849 s
->fd_open_time
= get_clock();
1850 s
->fd_got_error
= 0;
1854 static int hdev_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
1856 BDRVRawState
*s
= bs
->opaque
;
1858 return ioctl(s
->fd
, req
, buf
);
1861 static BlockDriverAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
1862 unsigned long int req
, void *buf
,
1863 BlockDriverCompletionFunc
*cb
, void *opaque
)
1865 BDRVRawState
*s
= bs
->opaque
;
1866 RawPosixAIOData
*acb
;
1869 if (fd_open(bs
) < 0)
1872 acb
= g_slice_new(RawPosixAIOData
);
1874 acb
->aio_type
= QEMU_AIO_IOCTL
;
1875 acb
->aio_fildes
= s
->fd
;
1876 acb
->aio_offset
= 0;
1877 acb
->aio_ioctl_buf
= buf
;
1878 acb
->aio_ioctl_cmd
= req
;
1879 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1880 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1883 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1884 static int fd_open(BlockDriverState
*bs
)
1886 BDRVRawState
*s
= bs
->opaque
;
1888 /* this is just to ensure s->fd is sane (its called by io ops) */
1893 #else /* !linux && !FreeBSD */
1895 static int fd_open(BlockDriverState
*bs
)
1900 #endif /* !linux && !FreeBSD */
1902 static coroutine_fn BlockDriverAIOCB
*hdev_aio_discard(BlockDriverState
*bs
,
1903 int64_t sector_num
, int nb_sectors
,
1904 BlockDriverCompletionFunc
*cb
, void *opaque
)
1906 BDRVRawState
*s
= bs
->opaque
;
1908 if (fd_open(bs
) < 0) {
1911 return paio_submit(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1912 cb
, opaque
, QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
1915 static coroutine_fn
int hdev_co_write_zeroes(BlockDriverState
*bs
,
1916 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1918 BDRVRawState
*s
= bs
->opaque
;
1925 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1926 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1927 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
1928 } else if (s
->discard_zeroes
) {
1929 return paio_submit_co(bs
, s
->fd
, sector_num
, NULL
, nb_sectors
,
1930 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
1935 static int hdev_create(const char *filename
, QemuOpts
*opts
,
1940 struct stat stat_buf
;
1941 int64_t total_size
= 0;
1944 /* This function is used by all three protocol block drivers and therefore
1945 * any of these three prefixes may be given.
1946 * The return value has to be stored somewhere, otherwise this is an error
1947 * due to -Werror=unused-value. */
1949 strstart(filename
, "host_device:", &filename
) ||
1950 strstart(filename
, "host_cdrom:" , &filename
) ||
1951 strstart(filename
, "host_floppy:", &filename
);
1955 /* Read out options */
1957 qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0) / BDRV_SECTOR_SIZE
;
1959 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
1962 error_setg_errno(errp
, -ret
, "Could not open device");
1966 if (fstat(fd
, &stat_buf
) < 0) {
1968 error_setg_errno(errp
, -ret
, "Could not stat device");
1969 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
1971 "The given file is neither a block nor a character device");
1973 } else if (lseek(fd
, 0, SEEK_END
) < total_size
* BDRV_SECTOR_SIZE
) {
1974 error_setg(errp
, "Device is too small");
1982 static BlockDriver bdrv_host_device
= {
1983 .format_name
= "host_device",
1984 .protocol_name
= "host_device",
1985 .instance_size
= sizeof(BDRVRawState
),
1986 .bdrv_needs_filename
= true,
1987 .bdrv_probe_device
= hdev_probe_device
,
1988 .bdrv_parse_filename
= hdev_parse_filename
,
1989 .bdrv_file_open
= hdev_open
,
1990 .bdrv_close
= raw_close
,
1991 .bdrv_reopen_prepare
= raw_reopen_prepare
,
1992 .bdrv_reopen_commit
= raw_reopen_commit
,
1993 .bdrv_reopen_abort
= raw_reopen_abort
,
1994 .bdrv_create
= hdev_create
,
1995 .create_opts
= &raw_create_opts
,
1996 .bdrv_co_write_zeroes
= hdev_co_write_zeroes
,
1998 .bdrv_aio_readv
= raw_aio_readv
,
1999 .bdrv_aio_writev
= raw_aio_writev
,
2000 .bdrv_aio_flush
= raw_aio_flush
,
2001 .bdrv_aio_discard
= hdev_aio_discard
,
2002 .bdrv_refresh_limits
= raw_refresh_limits
,
2003 .bdrv_io_plug
= raw_aio_plug
,
2004 .bdrv_io_unplug
= raw_aio_unplug
,
2005 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2007 .bdrv_truncate
= raw_truncate
,
2008 .bdrv_getlength
= raw_getlength
,
2009 .bdrv_get_info
= raw_get_info
,
2010 .bdrv_get_allocated_file_size
2011 = raw_get_allocated_file_size
,
2013 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2014 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2016 /* generic scsi device */
2018 .bdrv_ioctl
= hdev_ioctl
,
2019 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2024 static void floppy_parse_filename(const char *filename
, QDict
*options
,
2027 /* The prefix is optional, just as for "file". */
2028 strstart(filename
, "host_floppy:", &filename
);
2030 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
2033 static int floppy_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2036 BDRVRawState
*s
= bs
->opaque
;
2037 Error
*local_err
= NULL
;
2042 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2043 ret
= raw_open_common(bs
, options
, flags
, O_NONBLOCK
, &local_err
);
2046 error_propagate(errp
, local_err
);
2051 /* close fd so that we can reopen it as needed */
2054 s
->fd_media_changed
= 1;
2059 static int floppy_probe_device(const char *filename
)
2063 struct floppy_struct fdparam
;
2066 if (strstart(filename
, "/dev/fd", NULL
) &&
2067 !strstart(filename
, "/dev/fdset/", NULL
)) {
2071 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
2075 ret
= fstat(fd
, &st
);
2076 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2080 /* Attempt to detect via a floppy specific ioctl */
2081 ret
= ioctl(fd
, FDGETPRM
, &fdparam
);
2092 static int floppy_is_inserted(BlockDriverState
*bs
)
2094 return fd_open(bs
) >= 0;
2097 static int floppy_media_changed(BlockDriverState
*bs
)
2099 BDRVRawState
*s
= bs
->opaque
;
2103 * XXX: we do not have a true media changed indication.
2104 * It does not work if the floppy is changed without trying to read it.
2107 ret
= s
->fd_media_changed
;
2108 s
->fd_media_changed
= 0;
2110 printf("Floppy changed=%d\n", ret
);
2115 static void floppy_eject(BlockDriverState
*bs
, bool eject_flag
)
2117 BDRVRawState
*s
= bs
->opaque
;
2124 fd
= qemu_open(bs
->filename
, s
->open_flags
| O_NONBLOCK
);
2126 if (ioctl(fd
, FDEJECT
, 0) < 0)
2132 static BlockDriver bdrv_host_floppy
= {
2133 .format_name
= "host_floppy",
2134 .protocol_name
= "host_floppy",
2135 .instance_size
= sizeof(BDRVRawState
),
2136 .bdrv_needs_filename
= true,
2137 .bdrv_probe_device
= floppy_probe_device
,
2138 .bdrv_parse_filename
= floppy_parse_filename
,
2139 .bdrv_file_open
= floppy_open
,
2140 .bdrv_close
= raw_close
,
2141 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2142 .bdrv_reopen_commit
= raw_reopen_commit
,
2143 .bdrv_reopen_abort
= raw_reopen_abort
,
2144 .bdrv_create
= hdev_create
,
2145 .create_opts
= &raw_create_opts
,
2147 .bdrv_aio_readv
= raw_aio_readv
,
2148 .bdrv_aio_writev
= raw_aio_writev
,
2149 .bdrv_aio_flush
= raw_aio_flush
,
2150 .bdrv_refresh_limits
= raw_refresh_limits
,
2151 .bdrv_io_plug
= raw_aio_plug
,
2152 .bdrv_io_unplug
= raw_aio_unplug
,
2153 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2155 .bdrv_truncate
= raw_truncate
,
2156 .bdrv_getlength
= raw_getlength
,
2157 .has_variable_length
= true,
2158 .bdrv_get_allocated_file_size
2159 = raw_get_allocated_file_size
,
2161 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2162 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2164 /* removable device support */
2165 .bdrv_is_inserted
= floppy_is_inserted
,
2166 .bdrv_media_changed
= floppy_media_changed
,
2167 .bdrv_eject
= floppy_eject
,
2171 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2172 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
2175 /* The prefix is optional, just as for "file". */
2176 strstart(filename
, "host_cdrom:", &filename
);
2178 qdict_put_obj(options
, "filename", QOBJECT(qstring_from_str(filename
)));
2183 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2186 BDRVRawState
*s
= bs
->opaque
;
2187 Error
*local_err
= NULL
;
2192 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2193 ret
= raw_open_common(bs
, options
, flags
, O_NONBLOCK
, &local_err
);
2195 error_propagate(errp
, local_err
);
2200 static int cdrom_probe_device(const char *filename
)
2206 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
2210 ret
= fstat(fd
, &st
);
2211 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
2215 /* Attempt to detect via a CDROM specific ioctl */
2216 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2226 static int cdrom_is_inserted(BlockDriverState
*bs
)
2228 BDRVRawState
*s
= bs
->opaque
;
2231 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
2232 if (ret
== CDS_DISC_OK
)
2237 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2239 BDRVRawState
*s
= bs
->opaque
;
2242 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
2243 perror("CDROMEJECT");
2245 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
2246 perror("CDROMEJECT");
2250 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2252 BDRVRawState
*s
= bs
->opaque
;
2254 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
2256 * Note: an error can happen if the distribution automatically
2259 /* perror("CDROM_LOCKDOOR"); */
2263 static BlockDriver bdrv_host_cdrom
= {
2264 .format_name
= "host_cdrom",
2265 .protocol_name
= "host_cdrom",
2266 .instance_size
= sizeof(BDRVRawState
),
2267 .bdrv_needs_filename
= true,
2268 .bdrv_probe_device
= cdrom_probe_device
,
2269 .bdrv_parse_filename
= cdrom_parse_filename
,
2270 .bdrv_file_open
= cdrom_open
,
2271 .bdrv_close
= raw_close
,
2272 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2273 .bdrv_reopen_commit
= raw_reopen_commit
,
2274 .bdrv_reopen_abort
= raw_reopen_abort
,
2275 .bdrv_create
= hdev_create
,
2276 .create_opts
= &raw_create_opts
,
2278 .bdrv_aio_readv
= raw_aio_readv
,
2279 .bdrv_aio_writev
= raw_aio_writev
,
2280 .bdrv_aio_flush
= raw_aio_flush
,
2281 .bdrv_refresh_limits
= raw_refresh_limits
,
2282 .bdrv_io_plug
= raw_aio_plug
,
2283 .bdrv_io_unplug
= raw_aio_unplug
,
2284 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2286 .bdrv_truncate
= raw_truncate
,
2287 .bdrv_getlength
= raw_getlength
,
2288 .has_variable_length
= true,
2289 .bdrv_get_allocated_file_size
2290 = raw_get_allocated_file_size
,
2292 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2293 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2295 /* removable device support */
2296 .bdrv_is_inserted
= cdrom_is_inserted
,
2297 .bdrv_eject
= cdrom_eject
,
2298 .bdrv_lock_medium
= cdrom_lock_medium
,
2300 /* generic scsi device */
2301 .bdrv_ioctl
= hdev_ioctl
,
2302 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
2304 #endif /* __linux__ */
2306 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2307 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2310 BDRVRawState
*s
= bs
->opaque
;
2311 Error
*local_err
= NULL
;
2316 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2319 error_propagate(errp
, local_err
);
2324 /* make sure the door isn't locked at this time */
2325 ioctl(s
->fd
, CDIOCALLOW
);
2329 static int cdrom_probe_device(const char *filename
)
2331 if (strstart(filename
, "/dev/cd", NULL
) ||
2332 strstart(filename
, "/dev/acd", NULL
))
2337 static int cdrom_reopen(BlockDriverState
*bs
)
2339 BDRVRawState
*s
= bs
->opaque
;
2343 * Force reread of possibly changed/newly loaded disc,
2344 * FreeBSD seems to not notice sometimes...
2348 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
2355 /* make sure the door isn't locked at this time */
2356 ioctl(s
->fd
, CDIOCALLOW
);
2360 static int cdrom_is_inserted(BlockDriverState
*bs
)
2362 return raw_getlength(bs
) > 0;
2365 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
2367 BDRVRawState
*s
= bs
->opaque
;
2372 (void) ioctl(s
->fd
, CDIOCALLOW
);
2375 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
2376 perror("CDIOCEJECT");
2378 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
2379 perror("CDIOCCLOSE");
2385 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
2387 BDRVRawState
*s
= bs
->opaque
;
2391 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
2393 * Note: an error can happen if the distribution automatically
2396 /* perror("CDROM_LOCKDOOR"); */
2400 static BlockDriver bdrv_host_cdrom
= {
2401 .format_name
= "host_cdrom",
2402 .protocol_name
= "host_cdrom",
2403 .instance_size
= sizeof(BDRVRawState
),
2404 .bdrv_needs_filename
= true,
2405 .bdrv_probe_device
= cdrom_probe_device
,
2406 .bdrv_parse_filename
= cdrom_parse_filename
,
2407 .bdrv_file_open
= cdrom_open
,
2408 .bdrv_close
= raw_close
,
2409 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2410 .bdrv_reopen_commit
= raw_reopen_commit
,
2411 .bdrv_reopen_abort
= raw_reopen_abort
,
2412 .bdrv_create
= hdev_create
,
2413 .create_opts
= &raw_create_opts
,
2415 .bdrv_aio_readv
= raw_aio_readv
,
2416 .bdrv_aio_writev
= raw_aio_writev
,
2417 .bdrv_aio_flush
= raw_aio_flush
,
2418 .bdrv_refresh_limits
= raw_refresh_limits
,
2419 .bdrv_io_plug
= raw_aio_plug
,
2420 .bdrv_io_unplug
= raw_aio_unplug
,
2421 .bdrv_flush_io_queue
= raw_aio_flush_io_queue
,
2423 .bdrv_truncate
= raw_truncate
,
2424 .bdrv_getlength
= raw_getlength
,
2425 .has_variable_length
= true,
2426 .bdrv_get_allocated_file_size
2427 = raw_get_allocated_file_size
,
2429 .bdrv_detach_aio_context
= raw_detach_aio_context
,
2430 .bdrv_attach_aio_context
= raw_attach_aio_context
,
2432 /* removable device support */
2433 .bdrv_is_inserted
= cdrom_is_inserted
,
2434 .bdrv_eject
= cdrom_eject
,
2435 .bdrv_lock_medium
= cdrom_lock_medium
,
2437 #endif /* __FreeBSD__ */
2439 static void bdrv_file_init(void)
2442 * Register all the drivers. Note that order is important, the driver
2443 * registered last will get probed first.
2445 bdrv_register(&bdrv_file
);
2446 bdrv_register(&bdrv_host_device
);
2448 bdrv_register(&bdrv_host_floppy
);
2449 bdrv_register(&bdrv_host_cdrom
);
2451 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2452 bdrv_register(&bdrv_host_cdrom
);
2456 block_init(bdrv_file_init
);