2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
31 #include "qemu/option.h"
33 #include "block/thread-pool.h"
35 #include "block/raw-aio.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
39 #include "scsi/pr-manager.h"
40 #include "scsi/constants.h"
42 #if defined(__APPLE__) && (__MACH__)
44 #include <sys/param.h>
45 #include <IOKit/IOKitLib.h>
46 #include <IOKit/IOBSD.h>
47 #include <IOKit/storage/IOMediaBSDClient.h>
48 #include <IOKit/storage/IOMedia.h>
49 #include <IOKit/storage/IOCDMedia.h>
50 //#include <IOKit/storage/IOCDTypes.h>
51 #include <IOKit/storage/IODVDMedia.h>
52 #include <CoreFoundation/CoreFoundation.h>
56 #define _POSIX_PTHREAD_SEMANTICS 1
60 #include <sys/ioctl.h>
61 #include <sys/param.h>
62 #include <sys/syscall.h>
63 #include <linux/cdrom.h>
66 #include <linux/hdreg.h>
72 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
75 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
76 #include <linux/falloc.h>
78 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
84 #include <sys/ioctl.h>
85 #include <sys/disklabel.h>
90 #include <sys/ioctl.h>
91 #include <sys/disklabel.h>
97 #include <sys/ioctl.h>
98 #include <sys/diskslice.h>
105 //#define DEBUG_BLOCK
108 # define DEBUG_BLOCK_PRINT 1
110 # define DEBUG_BLOCK_PRINT 0
112 #define DPRINTF(fmt, ...) \
114 if (DEBUG_BLOCK_PRINT) { \
115 printf(fmt, ## __VA_ARGS__); \
119 /* OS X does not have O_DSYNC */
122 #define O_DSYNC O_SYNC
123 #elif defined(O_FSYNC)
124 #define O_DSYNC O_FSYNC
128 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
130 #define O_DIRECT O_DSYNC
136 #define MAX_BLOCKSIZE 4096
138 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
139 * leaving a few more bytes for its future use. */
140 #define RAW_LOCK_PERM_BASE 100
141 #define RAW_LOCK_SHARED_BASE 200
143 typedef struct BDRVRawState
{
151 /* The current permissions. */
153 uint64_t shared_perm
;
159 bool has_write_zeroes
:1;
160 bool discard_zeroes
:1;
161 bool use_linux_aio
:1;
162 bool page_cache_inconsistent
:1;
164 bool needs_alignment
;
165 bool check_cache_dropped
;
170 typedef struct BDRVRawReopenState
{
173 bool check_cache_dropped
;
174 } BDRVRawReopenState
;
176 static int fd_open(BlockDriverState
*bs
);
177 static int64_t raw_getlength(BlockDriverState
*bs
);
179 typedef struct RawPosixAIOData
{
180 BlockDriverState
*bs
;
183 struct iovec
*aio_iov
;
188 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
195 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
196 static int cdrom_reopen(BlockDriverState
*bs
);
199 #if defined(__NetBSD__)
200 static int raw_normalize_devicepath(const char **filename
)
202 static char namebuf
[PATH_MAX
];
203 const char *dp
, *fname
;
207 dp
= strrchr(fname
, '/');
208 if (lstat(fname
, &sb
) < 0) {
209 fprintf(stderr
, "%s: stat failed: %s\n",
210 fname
, strerror(errno
));
214 if (!S_ISBLK(sb
.st_mode
)) {
219 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
221 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
222 (int)(dp
- fname
), fname
, dp
+ 1);
224 fprintf(stderr
, "%s is a block device", fname
);
226 fprintf(stderr
, ", using %s\n", *filename
);
231 static int raw_normalize_devicepath(const char **filename
)
238 * Get logical block size via ioctl. On success store it in @sector_size_p.
240 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
242 unsigned int sector_size
;
243 bool success
= false;
247 static const unsigned long ioctl_list
[] = {
251 #ifdef DKIOCGETBLOCKSIZE
254 #ifdef DIOCGSECTORSIZE
259 /* Try a few ioctls to get the right size */
260 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
261 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
262 *sector_size_p
= sector_size
;
267 return success
? 0 : -errno
;
271 * Get physical block size of @fd.
272 * On success, store it in @blk_size and return 0.
273 * On failure, return -errno.
275 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
278 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
287 /* Check if read is allowed with given memory buffer and length.
289 * This function is used to check O_DIRECT memory buffer and request alignment.
291 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
293 ssize_t ret
= pread(fd
, buf
, len
, 0);
300 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
301 * other errors (e.g. real I/O error), which could happen on a failed
302 * drive, since we only care about probing alignment.
304 if (errno
!= EINVAL
) {
312 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
314 BDRVRawState
*s
= bs
->opaque
;
316 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
318 /* For SCSI generic devices the alignment is not really used.
319 With buffered I/O, we don't have any restrictions. */
320 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
321 bs
->bl
.request_alignment
= 1;
326 bs
->bl
.request_alignment
= 0;
328 /* Let's try to use the logical blocksize for the alignment. */
329 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
330 bs
->bl
.request_alignment
= 0;
335 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
336 bs
->bl
.request_alignment
= da
.d_miniosz
;
337 /* The kernel returns wrong information for d_mem */
338 /* s->buf_align = da.d_mem; */
343 /* If we could not get the sizes so far, we can only guess them */
346 buf
= qemu_memalign(max_align
, 2 * max_align
);
347 for (align
= 512; align
<= max_align
; align
<<= 1) {
348 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
349 s
->buf_align
= align
;
356 if (!bs
->bl
.request_alignment
) {
358 buf
= qemu_memalign(s
->buf_align
, max_align
);
359 for (align
= 512; align
<= max_align
; align
<<= 1) {
360 if (raw_is_io_aligned(fd
, buf
, align
)) {
361 bs
->bl
.request_alignment
= align
;
368 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
369 error_setg(errp
, "Could not find working O_DIRECT alignment");
370 error_append_hint(errp
, "Try cache.direct=off\n");
374 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
376 assert(open_flags
!= NULL
);
378 *open_flags
|= O_BINARY
;
379 *open_flags
&= ~O_ACCMODE
;
380 if (bdrv_flags
& BDRV_O_RDWR
) {
381 *open_flags
|= O_RDWR
;
383 *open_flags
|= O_RDONLY
;
386 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
387 * and O_DIRECT for no caching. */
388 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
389 *open_flags
|= O_DIRECT
;
393 static void raw_parse_filename(const char *filename
, QDict
*options
,
396 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
399 static QemuOptsList raw_runtime_opts
= {
401 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
405 .type
= QEMU_OPT_STRING
,
406 .help
= "File name of the image",
410 .type
= QEMU_OPT_STRING
,
411 .help
= "host AIO implementation (threads, native)",
415 .type
= QEMU_OPT_STRING
,
416 .help
= "file locking mode (on/off/auto, default: auto)",
419 .name
= "pr-manager",
420 .type
= QEMU_OPT_STRING
,
421 .help
= "id of persistent reservation manager object (default: none)",
424 .name
= "x-check-cache-dropped",
425 .type
= QEMU_OPT_BOOL
,
426 .help
= "check that page cache was dropped on live migration (default: off)"
428 { /* end of list */ }
432 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
433 int bdrv_flags
, int open_flags
, Error
**errp
)
435 BDRVRawState
*s
= bs
->opaque
;
437 Error
*local_err
= NULL
;
438 const char *filename
= NULL
;
440 BlockdevAioOptions aio
, aio_default
;
445 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
446 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
448 error_propagate(errp
, local_err
);
453 filename
= qemu_opt_get(opts
, "filename");
455 ret
= raw_normalize_devicepath(&filename
);
457 error_setg_errno(errp
, -ret
, "Could not normalize device path");
461 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
462 ? BLOCKDEV_AIO_OPTIONS_NATIVE
463 : BLOCKDEV_AIO_OPTIONS_THREADS
;
464 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
465 qemu_opt_get(opts
, "aio"),
466 aio_default
, &local_err
);
468 error_propagate(errp
, local_err
);
472 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
474 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
475 qemu_opt_get(opts
, "locking"),
476 ON_OFF_AUTO_AUTO
, &local_err
);
478 error_propagate(errp
, local_err
);
485 if (!qemu_has_ofd_lock()) {
487 "File lock requested but OFD locking syscall is "
488 "unavailable, falling back to POSIX file locks.\n"
489 "Due to the implementation, locks can be lost "
493 case ON_OFF_AUTO_OFF
:
496 case ON_OFF_AUTO_AUTO
:
497 s
->use_lock
= qemu_has_ofd_lock();
503 str
= qemu_opt_get(opts
, "pr-manager");
505 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
507 error_propagate(errp
, local_err
);
513 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
516 s
->open_flags
= open_flags
;
517 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
520 fd
= qemu_open(filename
, s
->open_flags
, 0644);
523 error_setg_errno(errp
, errno
, "Could not open '%s'", filename
);
533 fd
= qemu_open(filename
, s
->open_flags
);
536 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
544 s
->shared_perm
= BLK_PERM_ALL
;
546 #ifdef CONFIG_LINUX_AIO
547 /* Currently Linux does AIO only for files opened with O_DIRECT */
548 if (s
->use_linux_aio
) {
549 if (!(s
->open_flags
& O_DIRECT
)) {
550 error_setg(errp
, "aio=native was specified, but it requires "
551 "cache.direct=on, which was not specified.");
555 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
556 error_prepend(errp
, "Unable to use native AIO: ");
561 if (s
->use_linux_aio
) {
562 error_setg(errp
, "aio=native was specified, but is not supported "
567 #endif /* !defined(CONFIG_LINUX_AIO) */
569 s
->has_discard
= true;
570 s
->has_write_zeroes
= true;
571 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
572 s
->needs_alignment
= true;
575 if (fstat(s
->fd
, &st
) < 0) {
577 error_setg_errno(errp
, errno
, "Could not stat file");
580 if (S_ISREG(st
.st_mode
)) {
581 s
->discard_zeroes
= true;
582 s
->has_fallocate
= true;
584 if (S_ISBLK(st
.st_mode
)) {
585 #ifdef BLKDISCARDZEROES
587 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
588 s
->discard_zeroes
= true;
592 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
593 * not rely on the contents of discarded blocks unless using O_DIRECT.
594 * Same for BLKZEROOUT.
596 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
597 s
->discard_zeroes
= false;
598 s
->has_write_zeroes
= false;
603 if (S_ISCHR(st
.st_mode
)) {
605 * The file is a char device (disk), which on FreeBSD isn't behind
606 * a pager, so force all requests to be aligned. This is needed
607 * so QEMU makes sure all IO operations on the device are aligned
608 * to sector size, or else FreeBSD will reject them with EINVAL.
610 s
->needs_alignment
= true;
615 if (platform_test_xfs_fd(s
->fd
)) {
620 bs
->supported_zero_flags
= s
->discard_zeroes
? BDRV_REQ_MAY_UNMAP
: 0;
623 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
630 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
633 BDRVRawState
*s
= bs
->opaque
;
635 s
->type
= FTYPE_FILE
;
636 return raw_open_common(bs
, options
, flags
, 0, errp
);
645 #define PERM_FOREACH(i) \
646 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
648 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
649 * file; if @unlock == true, also unlock the unneeded bytes.
650 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
652 static int raw_apply_lock_bytes(int fd
,
653 uint64_t perm_lock_bits
,
654 uint64_t shared_perm_lock_bits
,
655 bool unlock
, Error
**errp
)
661 int off
= RAW_LOCK_PERM_BASE
+ i
;
662 if (perm_lock_bits
& (1ULL << i
)) {
663 ret
= qemu_lock_fd(fd
, off
, 1, false);
665 error_setg(errp
, "Failed to lock byte %d", off
);
669 ret
= qemu_unlock_fd(fd
, off
, 1);
671 error_setg(errp
, "Failed to unlock byte %d", off
);
677 int off
= RAW_LOCK_SHARED_BASE
+ i
;
678 if (shared_perm_lock_bits
& (1ULL << i
)) {
679 ret
= qemu_lock_fd(fd
, off
, 1, false);
681 error_setg(errp
, "Failed to lock byte %d", off
);
685 ret
= qemu_unlock_fd(fd
, off
, 1);
687 error_setg(errp
, "Failed to unlock byte %d", off
);
695 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
696 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
703 int off
= RAW_LOCK_SHARED_BASE
+ i
;
704 uint64_t p
= 1ULL << i
;
706 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
708 char *perm_name
= bdrv_perm_names(p
);
710 "Failed to get \"%s\" lock",
713 error_append_hint(errp
,
714 "Is another process using the image?\n");
720 int off
= RAW_LOCK_PERM_BASE
+ i
;
721 uint64_t p
= 1ULL << i
;
722 if (!(shared_perm
& p
)) {
723 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
725 char *perm_name
= bdrv_perm_names(p
);
727 "Failed to get shared \"%s\" lock",
730 error_append_hint(errp
,
731 "Is another process using the image?\n");
739 static int raw_handle_perm_lock(BlockDriverState
*bs
,
741 uint64_t new_perm
, uint64_t new_shared
,
744 BDRVRawState
*s
= bs
->opaque
;
746 Error
*local_err
= NULL
;
752 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
756 assert(s
->lock_fd
> 0);
760 ret
= raw_apply_lock_bytes(s
->lock_fd
, s
->perm
| new_perm
,
761 ~s
->shared_perm
| ~new_shared
,
764 ret
= raw_check_lock_bytes(s
->lock_fd
, new_perm
, new_shared
, errp
);
770 /* fall through to unlock bytes. */
772 raw_apply_lock_bytes(s
->lock_fd
, s
->perm
, ~s
->shared_perm
,
775 /* Theoretically the above call only unlocks bytes and it cannot
776 * fail. Something weird happened, report it.
778 error_report_err(local_err
);
782 raw_apply_lock_bytes(s
->lock_fd
, new_perm
, ~new_shared
,
785 /* Theoretically the above call only unlocks bytes and it cannot
786 * fail. Something weird happened, report it.
788 error_report_err(local_err
);
795 static int raw_reopen_prepare(BDRVReopenState
*state
,
796 BlockReopenQueue
*queue
, Error
**errp
)
799 BDRVRawReopenState
*rs
;
802 Error
*local_err
= NULL
;
804 assert(state
!= NULL
);
805 assert(state
->bs
!= NULL
);
807 s
= state
->bs
->opaque
;
809 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
813 /* Handle options changes */
814 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
815 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
817 error_propagate(errp
, local_err
);
822 rs
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
823 s
->check_cache_dropped
);
825 if (s
->type
== FTYPE_CD
) {
826 rs
->open_flags
|= O_NONBLOCK
;
829 raw_parse_flags(state
->flags
, &rs
->open_flags
);
831 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
833 fcntl_flags
|= O_NOATIME
;
837 /* Not all operating systems have O_ASYNC, and those that don't
838 * will not let us track the state into rs->open_flags (typically
839 * you achieve the same effect with an ioctl, for example I_SETSIG
840 * on Solaris). But we do not use O_ASYNC, so that's fine.
842 assert((s
->open_flags
& O_ASYNC
) == 0);
845 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
846 /* dup the original fd */
847 rs
->fd
= qemu_dup(s
->fd
);
849 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
857 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
859 const char *normalized_filename
= state
->bs
->filename
;
860 ret
= raw_normalize_devicepath(&normalized_filename
);
862 error_setg_errno(errp
, -ret
, "Could not normalize device path");
864 assert(!(rs
->open_flags
& O_CREAT
));
865 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
867 error_setg_errno(errp
, errno
, "Could not reopen file");
873 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
874 * alignment with the new fd. */
876 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
880 error_propagate(errp
, local_err
);
890 static void raw_reopen_commit(BDRVReopenState
*state
)
892 BDRVRawReopenState
*rs
= state
->opaque
;
893 BDRVRawState
*s
= state
->bs
->opaque
;
895 s
->check_cache_dropped
= rs
->check_cache_dropped
;
896 s
->open_flags
= rs
->open_flags
;
901 g_free(state
->opaque
);
902 state
->opaque
= NULL
;
906 static void raw_reopen_abort(BDRVReopenState
*state
)
908 BDRVRawReopenState
*rs
= state
->opaque
;
910 /* nothing to do if NULL, we didn't get far enough */
919 g_free(state
->opaque
);
920 state
->opaque
= NULL
;
923 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
927 short max_sectors
= 0;
928 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
930 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
931 return max_sectors
<< BDRV_SECTOR_BITS
;
940 static int hdev_get_max_segments(const struct stat
*st
)
950 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
951 major(st
->st_rdev
), minor(st
->st_rdev
));
952 fd
= open(sysfspath
, O_RDONLY
);
958 ret
= read(fd
, buf
, sizeof(buf
) - 1);
959 } while (ret
== -1 && errno
== EINTR
);
963 } else if (ret
== 0) {
968 /* The file is ended with '\n', pass 'end' to accept that. */
969 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
970 if (ret
== 0 && end
&& *end
== '\n') {
985 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
987 BDRVRawState
*s
= bs
->opaque
;
990 if (!fstat(s
->fd
, &st
)) {
991 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
992 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
993 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
994 bs
->bl
.max_transfer
= pow2floor(ret
);
996 ret
= hdev_get_max_segments(&st
);
998 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
999 ret
* getpagesize());
1004 raw_probe_alignment(bs
, s
->fd
, errp
);
1005 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1006 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1009 static int check_for_dasd(int fd
)
1012 struct dasd_information2_t info
= {0};
1014 return ioctl(fd
, BIODASDINFO2
, &info
);
1021 * Try to get @bs's logical and physical block size.
1022 * On success, store them in @bsz and return zero.
1023 * On failure, return negative errno.
1025 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1027 BDRVRawState
*s
= bs
->opaque
;
1030 /* If DASD, get blocksizes */
1031 if (check_for_dasd(s
->fd
) < 0) {
1034 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1038 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1042 * Try to get @bs's geometry: cyls, heads, sectors.
1043 * On success, store them in @geo and return 0.
1044 * On failure return -errno.
1045 * (Allows block driver to assign default geometry values that guest sees)
1048 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1050 BDRVRawState
*s
= bs
->opaque
;
1051 struct hd_geometry ioctl_geo
= {0};
1053 /* If DASD, get its geometry */
1054 if (check_for_dasd(s
->fd
) < 0) {
1057 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1060 /* HDIO_GETGEO may return success even though geo contains zeros
1061 (e.g. certain multipath setups) */
1062 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1065 /* Do not return a geometry for partition */
1066 if (ioctl_geo
.start
!= 0) {
1069 geo
->heads
= ioctl_geo
.heads
;
1070 geo
->sectors
= ioctl_geo
.sectors
;
1071 geo
->cylinders
= ioctl_geo
.cylinders
;
1075 #else /* __linux__ */
1076 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1082 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1086 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1094 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1096 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1099 if (s
->page_cache_inconsistent
) {
1103 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1105 /* There is no clear definition of the semantics of a failing fsync(),
1106 * so we may have to assume the worst. The sad truth is that this
1107 * assumption is correct for Linux. Some pages are now probably marked
1108 * clean in the page cache even though they are inconsistent with the
1109 * on-disk contents. The next fdatasync() call would succeed, but no
1110 * further writeback attempt will be made. We can't get back to a state
1111 * in which we know what is on disk (we would have to rewrite
1112 * everything that was touched since the last fdatasync() at least), so
1113 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1114 * really defined, I have little hope that other OSes are doing better.
1116 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1118 if ((s
->open_flags
& O_DIRECT
) == 0) {
1119 s
->page_cache_inconsistent
= true;
1126 #ifdef CONFIG_PREADV
1128 static bool preadv_present
= true;
1131 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1133 return preadv(fd
, iov
, nr_iov
, offset
);
1137 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1139 return pwritev(fd
, iov
, nr_iov
, offset
);
1144 static bool preadv_present
= false;
1147 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1153 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1160 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1165 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1166 len
= qemu_pwritev(aiocb
->aio_fildes
,
1171 len
= qemu_preadv(aiocb
->aio_fildes
,
1175 } while (len
== -1 && errno
== EINTR
);
1184 * Read/writes the data to/from a given linear buffer.
1186 * Returns the number of bytes handles or -errno in case of an error. Short
1187 * reads are only returned if the end of the file is reached.
1189 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1194 while (offset
< aiocb
->aio_nbytes
) {
1195 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1196 len
= pwrite(aiocb
->aio_fildes
,
1197 (const char *)buf
+ offset
,
1198 aiocb
->aio_nbytes
- offset
,
1199 aiocb
->aio_offset
+ offset
);
1201 len
= pread(aiocb
->aio_fildes
,
1203 aiocb
->aio_nbytes
- offset
,
1204 aiocb
->aio_offset
+ offset
);
1206 if (len
== -1 && errno
== EINTR
) {
1208 } else if (len
== -1 && errno
== EINVAL
&&
1209 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1210 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1212 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1213 * after a short read. Assume that O_DIRECT short reads only occur
1214 * at EOF. Therefore this is a short read, not an I/O error.
1217 } else if (len
== -1) {
1220 } else if (len
== 0) {
1229 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1234 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1236 * If there is just a single buffer, and it is properly aligned
1237 * we can just use plain pread/pwrite without any problems.
1239 if (aiocb
->aio_niov
== 1) {
1240 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1243 * We have more than one iovec, and all are properly aligned.
1245 * Try preadv/pwritev first and fall back to linearizing the
1246 * buffer if it's not supported.
1248 if (preadv_present
) {
1249 nbytes
= handle_aiocb_rw_vector(aiocb
);
1250 if (nbytes
== aiocb
->aio_nbytes
||
1251 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1254 preadv_present
= false;
1258 * XXX(hch): short read/write. no easy way to handle the reminder
1259 * using these interfaces. For now retry using plain
1265 * Ok, we have to do it the hard way, copy all segments into
1266 * a single aligned buffer.
1268 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1273 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1277 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1278 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1279 p
+= aiocb
->aio_iov
[i
].iov_len
;
1281 assert(p
- buf
== aiocb
->aio_nbytes
);
1284 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1285 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1287 size_t count
= aiocb
->aio_nbytes
, copy
;
1290 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1292 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1293 copy
= aiocb
->aio_iov
[i
].iov_len
;
1295 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1296 assert(count
>= copy
);
1308 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1310 struct xfs_flock64 fl
;
1313 memset(&fl
, 0, sizeof(fl
));
1314 fl
.l_whence
= SEEK_SET
;
1315 fl
.l_start
= offset
;
1318 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1320 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1327 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1329 struct xfs_flock64 fl
;
1332 memset(&fl
, 0, sizeof(fl
));
1333 fl
.l_whence
= SEEK_SET
;
1334 fl
.l_start
= offset
;
1337 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1339 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1347 static int translate_err(int err
)
1349 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1356 #ifdef CONFIG_FALLOCATE
1357 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1360 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1363 } while (errno
== EINTR
);
1364 return translate_err(-errno
);
1368 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1371 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1373 if (!s
->has_write_zeroes
) {
1379 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1380 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1383 } while (errno
== EINTR
);
1385 ret
= translate_err(-errno
);
1388 if (ret
== -ENOTSUP
) {
1389 s
->has_write_zeroes
= false;
1394 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1396 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1397 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1399 #ifdef CONFIG_FALLOCATE
1403 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1404 return handle_aiocb_write_zeroes_block(aiocb
);
1409 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1413 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1414 if (s
->has_write_zeroes
) {
1415 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1416 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1417 if (ret
== 0 || ret
!= -ENOTSUP
) {
1420 s
->has_write_zeroes
= false;
1424 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1425 if (s
->has_discard
&& s
->has_fallocate
) {
1426 int ret
= do_fallocate(s
->fd
,
1427 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1428 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1430 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1431 if (ret
== 0 || ret
!= -ENOTSUP
) {
1434 s
->has_fallocate
= false;
1435 } else if (ret
!= -ENOTSUP
) {
1438 s
->has_discard
= false;
1443 #ifdef CONFIG_FALLOCATE
1444 /* Last resort: we are trying to extend the file with zeroed data. This
1445 * can be done via fallocate(fd, 0) */
1446 len
= bdrv_getlength(aiocb
->bs
);
1447 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1448 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1449 if (ret
== 0 || ret
!= -ENOTSUP
) {
1452 s
->has_fallocate
= false;
1459 #ifndef HAVE_COPY_FILE_RANGE
1460 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1461 off_t
*out_off
, size_t len
, unsigned int flags
)
1463 #ifdef __NR_copy_file_range
1464 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1465 out_off
, len
, flags
);
1473 static ssize_t
handle_aiocb_copy_range(RawPosixAIOData
*aiocb
)
1475 uint64_t bytes
= aiocb
->aio_nbytes
;
1476 off_t in_off
= aiocb
->aio_offset
;
1477 off_t out_off
= aiocb
->aio_offset2
;
1480 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1481 aiocb
->aio_fd2
, &out_off
,
1483 if (ret
== -EINTR
) {
1487 if (errno
== ENOSYS
) {
1494 /* No progress (e.g. when beyond EOF), fall back to buffer I/O. */
1502 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1504 int ret
= -EOPNOTSUPP
;
1505 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1507 if (!s
->has_discard
) {
1511 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1514 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1515 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1518 } while (errno
== EINTR
);
1525 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1529 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1530 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1531 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1535 ret
= translate_err(ret
);
1536 if (ret
== -ENOTSUP
) {
1537 s
->has_discard
= false;
1542 static int aio_worker(void *arg
)
1544 RawPosixAIOData
*aiocb
= arg
;
1547 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1549 ret
= handle_aiocb_rw(aiocb
);
1550 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1551 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1552 0, aiocb
->aio_nbytes
- ret
);
1554 ret
= aiocb
->aio_nbytes
;
1556 if (ret
== aiocb
->aio_nbytes
) {
1558 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1562 case QEMU_AIO_WRITE
:
1563 ret
= handle_aiocb_rw(aiocb
);
1564 if (ret
== aiocb
->aio_nbytes
) {
1566 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1570 case QEMU_AIO_FLUSH
:
1571 ret
= handle_aiocb_flush(aiocb
);
1573 case QEMU_AIO_IOCTL
:
1574 ret
= handle_aiocb_ioctl(aiocb
);
1576 case QEMU_AIO_DISCARD
:
1577 ret
= handle_aiocb_discard(aiocb
);
1579 case QEMU_AIO_WRITE_ZEROES
:
1580 ret
= handle_aiocb_write_zeroes(aiocb
);
1582 case QEMU_AIO_COPY_RANGE
:
1583 ret
= handle_aiocb_copy_range(aiocb
);
1586 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1595 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1596 int64_t offset
, int fd2
, int64_t offset2
,
1598 int bytes
, int type
)
1600 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1604 acb
->aio_type
= type
;
1605 acb
->aio_fildes
= fd
;
1607 acb
->aio_offset2
= offset2
;
1609 acb
->aio_nbytes
= bytes
;
1610 acb
->aio_offset
= offset
;
1613 acb
->aio_iov
= qiov
->iov
;
1614 acb
->aio_niov
= qiov
->niov
;
1615 assert(qiov
->size
== bytes
);
1618 trace_paio_submit_co(offset
, bytes
, type
);
1619 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1620 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1623 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1624 int64_t offset
, QEMUIOVector
*qiov
,
1625 int bytes
, int type
)
1627 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1630 static BlockAIOCB
*paio_submit(BlockDriverState
*bs
, int fd
,
1631 int64_t offset
, QEMUIOVector
*qiov
, int bytes
,
1632 BlockCompletionFunc
*cb
, void *opaque
, int type
)
1634 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1638 acb
->aio_type
= type
;
1639 acb
->aio_fildes
= fd
;
1641 acb
->aio_nbytes
= bytes
;
1642 acb
->aio_offset
= offset
;
1645 acb
->aio_iov
= qiov
->iov
;
1646 acb
->aio_niov
= qiov
->niov
;
1647 assert(qiov
->size
== acb
->aio_nbytes
);
1650 trace_paio_submit(acb
, opaque
, offset
, bytes
, type
);
1651 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1652 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
1655 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1656 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1658 BDRVRawState
*s
= bs
->opaque
;
1660 if (fd_open(bs
) < 0)
1664 * Check if the underlying device requires requests to be aligned,
1665 * and if the request we are trying to submit is aligned or not.
1666 * If this is the case tell the low-level driver that it needs
1667 * to copy the buffer.
1669 if (s
->needs_alignment
) {
1670 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1671 type
|= QEMU_AIO_MISALIGNED
;
1672 #ifdef CONFIG_LINUX_AIO
1673 } else if (s
->use_linux_aio
) {
1674 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1675 assert(qiov
->size
== bytes
);
1676 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1681 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1684 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1685 uint64_t bytes
, QEMUIOVector
*qiov
,
1688 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1691 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1692 uint64_t bytes
, QEMUIOVector
*qiov
,
1696 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1699 static void raw_aio_plug(BlockDriverState
*bs
)
1701 #ifdef CONFIG_LINUX_AIO
1702 BDRVRawState
*s
= bs
->opaque
;
1703 if (s
->use_linux_aio
) {
1704 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1705 laio_io_plug(bs
, aio
);
1710 static void raw_aio_unplug(BlockDriverState
*bs
)
1712 #ifdef CONFIG_LINUX_AIO
1713 BDRVRawState
*s
= bs
->opaque
;
1714 if (s
->use_linux_aio
) {
1715 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1716 laio_io_unplug(bs
, aio
);
1721 static BlockAIOCB
*raw_aio_flush(BlockDriverState
*bs
,
1722 BlockCompletionFunc
*cb
, void *opaque
)
1724 BDRVRawState
*s
= bs
->opaque
;
1726 if (fd_open(bs
) < 0)
1729 return paio_submit(bs
, s
->fd
, 0, NULL
, 0, cb
, opaque
, QEMU_AIO_FLUSH
);
1732 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1733 AioContext
*new_context
)
1735 #ifdef CONFIG_LINUX_AIO
1736 BDRVRawState
*s
= bs
->opaque
;
1737 if (s
->use_linux_aio
) {
1739 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1740 error_reportf_err(local_err
, "Unable to use native AIO, "
1741 "falling back to thread pool: ");
1742 s
->use_linux_aio
= false;
1748 static void raw_close(BlockDriverState
*bs
)
1750 BDRVRawState
*s
= bs
->opaque
;
1756 if (s
->lock_fd
>= 0) {
1757 qemu_close(s
->lock_fd
);
1763 * Truncates the given regular file @fd to @offset and, when growing, fills the
1764 * new space according to @prealloc.
1766 * Returns: 0 on success, -errno on failure.
1768 static int raw_regular_truncate(int fd
, int64_t offset
, PreallocMode prealloc
,
1772 int64_t current_length
= 0;
1776 if (fstat(fd
, &st
) < 0) {
1778 error_setg_errno(errp
, -result
, "Could not stat file");
1782 current_length
= st
.st_size
;
1783 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1784 error_setg(errp
, "Cannot use preallocation for shrinking files");
1789 #ifdef CONFIG_POSIX_FALLOCATE
1790 case PREALLOC_MODE_FALLOC
:
1792 * Truncating before posix_fallocate() makes it about twice slower on
1793 * file systems that do not support fallocate(), trying to check if a
1794 * block is allocated before allocating it, so don't do that here.
1796 if (offset
!= current_length
) {
1797 result
= -posix_fallocate(fd
, current_length
, offset
- current_length
);
1799 /* posix_fallocate() doesn't set errno. */
1800 error_setg_errno(errp
, -result
,
1801 "Could not preallocate new data");
1808 case PREALLOC_MODE_FULL
:
1810 int64_t num
= 0, left
= offset
- current_length
;
1814 * Knowing the final size from the beginning could allow the file
1815 * system driver to do less allocations and possibly avoid
1816 * fragmentation of the file.
1818 if (ftruncate(fd
, offset
) != 0) {
1820 error_setg_errno(errp
, -result
, "Could not resize file");
1824 buf
= g_malloc0(65536);
1826 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1827 if (seek_result
< 0) {
1829 error_setg_errno(errp
, -result
,
1830 "Failed to seek to the old end of file");
1835 num
= MIN(left
, 65536);
1836 result
= write(fd
, buf
, num
);
1839 error_setg_errno(errp
, -result
,
1840 "Could not write zeros for preallocation");
1849 error_setg_errno(errp
, -result
,
1850 "Could not flush file to disk");
1856 case PREALLOC_MODE_OFF
:
1857 if (ftruncate(fd
, offset
) != 0) {
1859 error_setg_errno(errp
, -result
, "Could not resize file");
1864 error_setg(errp
, "Unsupported preallocation mode: %s",
1865 PreallocMode_str(prealloc
));
1871 if (ftruncate(fd
, current_length
) < 0) {
1872 error_report("Failed to restore old file length: %s",
1881 static int raw_truncate(BlockDriverState
*bs
, int64_t offset
,
1882 PreallocMode prealloc
, Error
**errp
)
1884 BDRVRawState
*s
= bs
->opaque
;
1888 if (fstat(s
->fd
, &st
)) {
1890 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1894 if (S_ISREG(st
.st_mode
)) {
1895 return raw_regular_truncate(s
->fd
, offset
, prealloc
, errp
);
1898 if (prealloc
!= PREALLOC_MODE_OFF
) {
1899 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1900 "non-regular file", PreallocMode_str(prealloc
));
1904 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1905 if (offset
> raw_getlength(bs
)) {
1906 error_setg(errp
, "Cannot grow device files");
1910 error_setg(errp
, "Resizing this file is not supported");
1918 static int64_t raw_getlength(BlockDriverState
*bs
)
1920 BDRVRawState
*s
= bs
->opaque
;
1926 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1927 struct disklabel dl
;
1929 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1931 return (uint64_t)dl
.d_secsize
*
1932 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1936 #elif defined(__NetBSD__)
1937 static int64_t raw_getlength(BlockDriverState
*bs
)
1939 BDRVRawState
*s
= bs
->opaque
;
1945 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1946 struct dkwedge_info dkw
;
1948 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1949 return dkw
.dkw_size
* 512;
1951 struct disklabel dl
;
1953 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1955 return (uint64_t)dl
.d_secsize
*
1956 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1961 #elif defined(__sun__)
1962 static int64_t raw_getlength(BlockDriverState
*bs
)
1964 BDRVRawState
*s
= bs
->opaque
;
1965 struct dk_minfo minfo
;
1975 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1977 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
1979 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
1983 * There are reports that lseek on some devices fails, but
1984 * irc discussion said that contingency on contingency was overkill.
1986 size
= lseek(s
->fd
, 0, SEEK_END
);
1992 #elif defined(CONFIG_BSD)
1993 static int64_t raw_getlength(BlockDriverState
*bs
)
1995 BDRVRawState
*s
= bs
->opaque
;
1999 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2008 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2011 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2012 #ifdef DIOCGMEDIASIZE
2013 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2014 #elif defined(DIOCGPART)
2017 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2018 size
= pi
.media_size
;
2024 #if defined(__APPLE__) && defined(__MACH__)
2026 uint64_t sectors
= 0;
2027 uint32_t sector_size
= 0;
2029 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2030 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2031 size
= sectors
* sector_size
;
2033 size
= lseek(fd
, 0LL, SEEK_END
);
2040 size
= lseek(fd
, 0LL, SEEK_END
);
2045 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2048 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2049 if (size
== 2048LL * (unsigned)-1)
2051 /* XXX no disc? maybe we need to reopen... */
2052 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2059 size
= lseek(fd
, 0, SEEK_END
);
2067 static int64_t raw_getlength(BlockDriverState
*bs
)
2069 BDRVRawState
*s
= bs
->opaque
;
2078 size
= lseek(s
->fd
, 0, SEEK_END
);
2086 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2089 BDRVRawState
*s
= bs
->opaque
;
2091 if (fstat(s
->fd
, &st
) < 0) {
2094 return (int64_t)st
.st_blocks
* 512;
2097 static int raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2099 BlockdevCreateOptionsFile
*file_opts
;
2104 /* Validate options and set default values */
2105 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2106 file_opts
= &options
->u
.file
;
2108 if (!file_opts
->has_nocow
) {
2109 file_opts
->nocow
= false;
2111 if (!file_opts
->has_preallocation
) {
2112 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2116 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2119 error_setg_errno(errp
, -result
, "Could not create file");
2123 /* Take permissions: We want to discard everything, so we need
2124 * BLK_PERM_WRITE; and truncation to the desired size requires
2126 * On the other hand, we cannot share the RESIZE permission
2127 * because we promise that after this function, the file has the
2128 * size given in the options. If someone else were to resize it
2129 * concurrently, we could not guarantee that.
2130 * Note that after this function, we can no longer guarantee that
2131 * the file is not touched by a third party, so it may be resized
2133 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2134 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2136 /* Step one: Take locks */
2137 result
= raw_apply_lock_bytes(fd
, perm
, shared
, false, errp
);
2142 /* Step two: Check that nobody else has taken conflicting locks */
2143 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2148 /* Clear the file by truncating it to 0 */
2149 result
= raw_regular_truncate(fd
, 0, PREALLOC_MODE_OFF
, errp
);
2154 if (file_opts
->nocow
) {
2156 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2157 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2158 * will be ignored since any failure of this operation should not
2159 * block the left work.
2162 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2163 attr
|= FS_NOCOW_FL
;
2164 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2169 /* Resize and potentially preallocate the file to the desired
2171 result
= raw_regular_truncate(fd
, file_opts
->size
, file_opts
->preallocation
,
2178 if (qemu_close(fd
) != 0 && result
== 0) {
2180 error_setg_errno(errp
, -result
, "Could not close the new file");
2186 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2189 BlockdevCreateOptions options
;
2190 int64_t total_size
= 0;
2192 PreallocMode prealloc
;
2194 Error
*local_err
= NULL
;
2196 /* Skip file: protocol prefix */
2197 strstart(filename
, "file:", &filename
);
2199 /* Read out options */
2200 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2202 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2203 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2204 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2205 PREALLOC_MODE_OFF
, &local_err
);
2208 error_propagate(errp
, local_err
);
2212 options
= (BlockdevCreateOptions
) {
2213 .driver
= BLOCKDEV_DRIVER_FILE
,
2215 .filename
= (char *) filename
,
2217 .has_preallocation
= true,
2218 .preallocation
= prealloc
,
2223 return raw_co_create(&options
, errp
);
2227 * Find allocation range in @bs around offset @start.
2228 * May change underlying file descriptor's file offset.
2229 * If @start is not in a hole, store @start in @data, and the
2230 * beginning of the next hole in @hole, and return 0.
2231 * If @start is in a non-trailing hole, store @start in @hole and the
2232 * beginning of the next non-hole in @data, and return 0.
2233 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2234 * If we can't find out, return a negative errno other than -ENXIO.
2236 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2237 off_t
*data
, off_t
*hole
)
2239 #if defined SEEK_HOLE && defined SEEK_DATA
2240 BDRVRawState
*s
= bs
->opaque
;
2245 * D1. offs == start: start is in data
2246 * D2. offs > start: start is in a hole, next data at offs
2247 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2248 * or start is beyond EOF
2249 * If the latter happens, the file has been truncated behind
2250 * our back since we opened it. All bets are off then.
2251 * Treating like a trailing hole is simplest.
2252 * D4. offs < 0, errno != ENXIO: we learned nothing
2254 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2256 return -errno
; /* D3 or D4 */
2260 /* This is not a valid return by lseek(). We are safe to just return
2261 * -EIO in this case, and we'll treat it like D4. */
2266 /* D2: in hole, next data at offs */
2272 /* D1: in data, end not yet known */
2276 * H1. offs == start: start is in a hole
2277 * If this happens here, a hole has been dug behind our back
2278 * since the previous lseek().
2279 * H2. offs > start: either start is in data, next hole at offs,
2280 * or start is in trailing hole, EOF at offs
2281 * Linux treats trailing holes like any other hole: offs ==
2282 * start. Solaris seeks to EOF instead: offs > start (blech).
2283 * If that happens here, a hole has been dug behind our back
2284 * since the previous lseek().
2285 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2286 * If this happens, the file has been truncated behind our
2287 * back since we opened it. Treat it like a trailing hole.
2288 * H4. offs < 0, errno != ENXIO: we learned nothing
2289 * Pretend we know nothing at all, i.e. "forget" about D1.
2291 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2293 return -errno
; /* D1 and (H3 or H4) */
2297 /* This is not a valid return by lseek(). We are safe to just return
2298 * -EIO in this case, and we'll treat it like H4. */
2304 * D1 and H2: either in data, next hole at offs, or it was in
2305 * data but is now in a trailing hole. In the latter case,
2306 * all bets are off. Treating it as if it there was data all
2307 * the way to EOF is safe, so simply do that.
2322 * Returns the allocation status of the specified offset.
2324 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2326 * 'pnum' is set to the number of bytes (including and immediately following
2327 * the specified offset) that are known to be in the same
2328 * allocated/unallocated state.
2330 * 'bytes' is the max value 'pnum' should be set to.
2332 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2335 int64_t bytes
, int64_t *pnum
,
2337 BlockDriverState
**file
)
2339 off_t data
= 0, hole
= 0;
2351 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2354 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2355 if (ret
== -ENXIO
) {
2358 ret
= BDRV_BLOCK_ZERO
;
2359 } else if (ret
< 0) {
2360 /* No info available, so pretend there are no holes */
2362 ret
= BDRV_BLOCK_DATA
;
2363 } else if (data
== offset
) {
2364 /* On a data extent, compute bytes to the end of the extent,
2365 * possibly including a partial sector at EOF. */
2366 *pnum
= MIN(bytes
, hole
- offset
);
2367 ret
= BDRV_BLOCK_DATA
;
2369 /* On a hole, compute bytes to the beginning of the next extent. */
2370 assert(hole
== offset
);
2371 *pnum
= MIN(bytes
, data
- offset
);
2372 ret
= BDRV_BLOCK_ZERO
;
2376 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2379 #if defined(__linux__)
2380 /* Verify that the file is not in the page cache */
2381 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2383 const size_t window_size
= 128 * 1024 * 1024;
2384 BDRVRawState
*s
= bs
->opaque
;
2385 void *window
= NULL
;
2392 /* mincore(2) page status information requires 1 byte per page */
2393 page_size
= sysconf(_SC_PAGESIZE
);
2394 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2396 end
= raw_getlength(bs
);
2398 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2405 /* Unmap previous window if size has changed */
2406 new_length
= MIN(end
- offset
, window_size
);
2407 if (new_length
!= length
) {
2408 munmap(window
, length
);
2413 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2415 if (new_window
== MAP_FAILED
) {
2416 error_setg_errno(errp
, errno
, "mmap failed");
2420 window
= new_window
;
2421 length
= new_length
;
2423 ret
= mincore(window
, length
, vec
);
2425 error_setg_errno(errp
, errno
, "mincore failed");
2429 vec_end
= DIV_ROUND_UP(length
, page_size
);
2430 for (i
= 0; i
< vec_end
; i
++) {
2432 error_setg(errp
, "page cache still in use!");
2439 munmap(window
, length
);
2444 #endif /* __linux__ */
2446 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2449 BDRVRawState
*s
= bs
->opaque
;
2454 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2458 if (s
->open_flags
& O_DIRECT
) {
2459 return; /* No host kernel page cache */
2462 #if defined(__linux__)
2463 /* This sets the scene for the next syscall... */
2464 ret
= bdrv_co_flush(bs
);
2466 error_setg_errno(errp
, -ret
, "flush failed");
2470 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2471 * process. These limitations are okay because we just fsynced the file,
2472 * we don't use mmap, and the file should not be in use by other processes.
2474 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2475 if (ret
!= 0) { /* the return value is a positive errno */
2476 error_setg_errno(errp
, ret
, "fadvise failed");
2480 if (s
->check_cache_dropped
) {
2481 check_cache_dropped(bs
, errp
);
2483 #else /* __linux__ */
2484 /* Do nothing. Live migration to a remote host with cache.direct=off is
2485 * unsupported on other host operating systems. Cache consistency issues
2486 * may occur but no error is reported here, partly because that's the
2487 * historical behavior and partly because it's hard to differentiate valid
2488 * configurations that should not cause errors.
2490 #endif /* !__linux__ */
2493 static coroutine_fn BlockAIOCB
*raw_aio_pdiscard(BlockDriverState
*bs
,
2494 int64_t offset
, int bytes
,
2495 BlockCompletionFunc
*cb
, void *opaque
)
2497 BDRVRawState
*s
= bs
->opaque
;
2499 return paio_submit(bs
, s
->fd
, offset
, NULL
, bytes
,
2500 cb
, opaque
, QEMU_AIO_DISCARD
);
2503 static int coroutine_fn
raw_co_pwrite_zeroes(
2504 BlockDriverState
*bs
, int64_t offset
,
2505 int bytes
, BdrvRequestFlags flags
)
2507 BDRVRawState
*s
= bs
->opaque
;
2509 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
2510 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2511 QEMU_AIO_WRITE_ZEROES
);
2512 } else if (s
->discard_zeroes
) {
2513 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2519 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2521 BDRVRawState
*s
= bs
->opaque
;
2523 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2527 static QemuOptsList raw_create_opts
= {
2528 .name
= "raw-create-opts",
2529 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2532 .name
= BLOCK_OPT_SIZE
,
2533 .type
= QEMU_OPT_SIZE
,
2534 .help
= "Virtual disk size"
2537 .name
= BLOCK_OPT_NOCOW
,
2538 .type
= QEMU_OPT_BOOL
,
2539 .help
= "Turn off copy-on-write (valid only on btrfs)"
2542 .name
= BLOCK_OPT_PREALLOC
,
2543 .type
= QEMU_OPT_STRING
,
2544 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2546 { /* end of list */ }
2550 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2553 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2556 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2558 BDRVRawState
*s
= bs
->opaque
;
2559 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2561 s
->shared_perm
= shared
;
2564 static void raw_abort_perm_update(BlockDriverState
*bs
)
2566 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2569 static int coroutine_fn
raw_co_copy_range_from(BlockDriverState
*bs
,
2570 BdrvChild
*src
, uint64_t src_offset
,
2571 BdrvChild
*dst
, uint64_t dst_offset
,
2572 uint64_t bytes
, BdrvRequestFlags flags
)
2574 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
, flags
);
2577 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2578 BdrvChild
*src
, uint64_t src_offset
,
2579 BdrvChild
*dst
, uint64_t dst_offset
,
2580 uint64_t bytes
, BdrvRequestFlags flags
)
2582 BDRVRawState
*s
= bs
->opaque
;
2583 BDRVRawState
*src_s
;
2585 assert(dst
->bs
== bs
);
2586 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2590 src_s
= src
->bs
->opaque
;
2591 if (fd_open(bs
) < 0 || fd_open(bs
) < 0) {
2594 return paio_submit_co_full(bs
, src_s
->fd
, src_offset
, s
->fd
, dst_offset
,
2595 NULL
, bytes
, QEMU_AIO_COPY_RANGE
);
2598 BlockDriver bdrv_file
= {
2599 .format_name
= "file",
2600 .protocol_name
= "file",
2601 .instance_size
= sizeof(BDRVRawState
),
2602 .bdrv_needs_filename
= true,
2603 .bdrv_probe
= NULL
, /* no probe for protocols */
2604 .bdrv_parse_filename
= raw_parse_filename
,
2605 .bdrv_file_open
= raw_open
,
2606 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2607 .bdrv_reopen_commit
= raw_reopen_commit
,
2608 .bdrv_reopen_abort
= raw_reopen_abort
,
2609 .bdrv_close
= raw_close
,
2610 .bdrv_co_create
= raw_co_create
,
2611 .bdrv_co_create_opts
= raw_co_create_opts
,
2612 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2613 .bdrv_co_block_status
= raw_co_block_status
,
2614 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2615 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2617 .bdrv_co_preadv
= raw_co_preadv
,
2618 .bdrv_co_pwritev
= raw_co_pwritev
,
2619 .bdrv_aio_flush
= raw_aio_flush
,
2620 .bdrv_aio_pdiscard
= raw_aio_pdiscard
,
2621 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2622 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2623 .bdrv_refresh_limits
= raw_refresh_limits
,
2624 .bdrv_io_plug
= raw_aio_plug
,
2625 .bdrv_io_unplug
= raw_aio_unplug
,
2626 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2628 .bdrv_truncate
= raw_truncate
,
2629 .bdrv_getlength
= raw_getlength
,
2630 .bdrv_get_info
= raw_get_info
,
2631 .bdrv_get_allocated_file_size
2632 = raw_get_allocated_file_size
,
2633 .bdrv_check_perm
= raw_check_perm
,
2634 .bdrv_set_perm
= raw_set_perm
,
2635 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2636 .create_opts
= &raw_create_opts
,
2639 /***********************************************/
2642 #if defined(__APPLE__) && defined(__MACH__)
2643 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2644 CFIndex maxPathSize
, int flags
);
2645 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2647 kern_return_t kernResult
= KERN_FAILURE
;
2648 mach_port_t masterPort
;
2649 CFMutableDictionaryRef classesToMatch
;
2650 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2651 char *mediaType
= NULL
;
2653 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2654 if ( KERN_SUCCESS
!= kernResult
) {
2655 printf( "IOMasterPort returned %d\n", kernResult
);
2659 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2660 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2661 if (classesToMatch
== NULL
) {
2662 error_report("IOServiceMatching returned NULL for %s",
2663 matching_array
[index
]);
2666 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2668 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2670 if (kernResult
!= KERN_SUCCESS
) {
2671 error_report("Note: IOServiceGetMatchingServices returned %d",
2676 /* If a match was found, leave the loop */
2677 if (*mediaIterator
!= 0) {
2678 DPRINTF("Matching using %s\n", matching_array
[index
]);
2679 mediaType
= g_strdup(matching_array
[index
]);
2686 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2687 CFIndex maxPathSize
, int flags
)
2689 io_object_t nextMedia
;
2690 kern_return_t kernResult
= KERN_FAILURE
;
2692 nextMedia
= IOIteratorNext( mediaIterator
);
2695 CFTypeRef bsdPathAsCFString
;
2696 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2697 if ( bsdPathAsCFString
) {
2698 size_t devPathLength
;
2699 strcpy( bsdPath
, _PATH_DEV
);
2700 if (flags
& BDRV_O_NOCACHE
) {
2701 strcat(bsdPath
, "r");
2703 devPathLength
= strlen( bsdPath
);
2704 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2705 kernResult
= KERN_SUCCESS
;
2707 CFRelease( bsdPathAsCFString
);
2709 IOObjectRelease( nextMedia
);
2715 /* Sets up a real cdrom for use in QEMU */
2716 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2718 int index
, num_of_test_partitions
= 2, fd
;
2719 char test_partition
[MAXPATHLEN
];
2720 bool partition_found
= false;
2722 /* look for a working partition */
2723 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2724 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2726 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2728 partition_found
= true;
2734 /* if a working partition on the device was not found */
2735 if (partition_found
== false) {
2736 error_setg(errp
, "Failed to find a working partition on disc");
2738 DPRINTF("Using %s as optical disc\n", test_partition
);
2739 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2741 return partition_found
;
2744 /* Prints directions on mounting and unmounting a device */
2745 static void print_unmounting_directions(const char *file_name
)
2747 error_report("If device %s is mounted on the desktop, unmount"
2748 " it first before using it in QEMU", file_name
);
2749 error_report("Command to unmount device: diskutil unmountDisk %s",
2751 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2754 #endif /* defined(__APPLE__) && defined(__MACH__) */
2756 static int hdev_probe_device(const char *filename
)
2760 /* allow a dedicated CD-ROM driver to match with a higher priority */
2761 if (strstart(filename
, "/dev/cdrom", NULL
))
2764 if (stat(filename
, &st
) >= 0 &&
2765 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2772 static int check_hdev_writable(BDRVRawState
*s
)
2774 #if defined(BLKROGET)
2775 /* Linux block devices can be configured "read-only" using blockdev(8).
2776 * This is independent of device node permissions and therefore open(2)
2777 * with O_RDWR succeeds. Actual writes fail with EPERM.
2779 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2780 * check for read-only block devices so that Linux block devices behave
2786 if (fstat(s
->fd
, &st
)) {
2790 if (!S_ISBLK(st
.st_mode
)) {
2794 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2801 #endif /* defined(BLKROGET) */
2805 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2808 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2811 static bool hdev_is_sg(BlockDriverState
*bs
)
2814 #if defined(__linux__)
2816 BDRVRawState
*s
= bs
->opaque
;
2818 struct sg_scsi_id scsiid
;
2822 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2826 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2831 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2833 DPRINTF("SG device found: type=%d, version=%d\n",
2834 scsiid
.scsi_type
, sg_version
);
2843 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2846 BDRVRawState
*s
= bs
->opaque
;
2847 Error
*local_err
= NULL
;
2850 #if defined(__APPLE__) && defined(__MACH__)
2852 * Caution: while qdict_get_str() is fine, getting non-string types
2853 * would require more care. When @options come from -blockdev or
2854 * blockdev_add, its members are typed according to the QAPI
2855 * schema, but when they come from -drive, they're all QString.
2857 const char *filename
= qdict_get_str(options
, "filename");
2858 char bsd_path
[MAXPATHLEN
] = "";
2859 bool error_occurred
= false;
2861 /* If using a real cdrom */
2862 if (strcmp(filename
, "/dev/cdrom") == 0) {
2863 char *mediaType
= NULL
;
2864 kern_return_t ret_val
;
2865 io_iterator_t mediaIterator
= 0;
2867 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
2868 if (mediaType
== NULL
) {
2869 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
2871 error_occurred
= true;
2872 goto hdev_open_Mac_error
;
2875 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
2876 if (ret_val
!= KERN_SUCCESS
) {
2877 error_setg(errp
, "Could not get BSD path for optical drive");
2878 error_occurred
= true;
2879 goto hdev_open_Mac_error
;
2882 /* If a real optical drive was not found */
2883 if (bsd_path
[0] == '\0') {
2884 error_setg(errp
, "Failed to obtain bsd path for optical drive");
2885 error_occurred
= true;
2886 goto hdev_open_Mac_error
;
2889 /* If using a cdrom disc and finding a partition on the disc failed */
2890 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
2891 setup_cdrom(bsd_path
, errp
) == false) {
2892 print_unmounting_directions(bsd_path
);
2893 error_occurred
= true;
2894 goto hdev_open_Mac_error
;
2897 qdict_put_str(options
, "filename", bsd_path
);
2899 hdev_open_Mac_error
:
2901 if (mediaIterator
) {
2902 IOObjectRelease(mediaIterator
);
2904 if (error_occurred
) {
2908 #endif /* defined(__APPLE__) && defined(__MACH__) */
2910 s
->type
= FTYPE_FILE
;
2912 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2914 error_propagate(errp
, local_err
);
2915 #if defined(__APPLE__) && defined(__MACH__)
2917 filename
= bsd_path
;
2919 /* if a physical device experienced an error while being opened */
2920 if (strncmp(filename
, "/dev/", 5) == 0) {
2921 print_unmounting_directions(filename
);
2923 #endif /* defined(__APPLE__) && defined(__MACH__) */
2927 /* Since this does ioctl the device must be already opened */
2928 bs
->sg
= hdev_is_sg(bs
);
2930 if (flags
& BDRV_O_RDWR
) {
2931 ret
= check_hdev_writable(s
);
2934 error_setg_errno(errp
, -ret
, "The device is not writable");
2942 #if defined(__linux__)
2944 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
2945 unsigned long int req
, void *buf
,
2946 BlockCompletionFunc
*cb
, void *opaque
)
2948 BDRVRawState
*s
= bs
->opaque
;
2949 RawPosixAIOData
*acb
;
2952 if (fd_open(bs
) < 0)
2955 if (req
== SG_IO
&& s
->pr_mgr
) {
2956 struct sg_io_hdr
*io_hdr
= buf
;
2957 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
2958 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
2959 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
2960 s
->fd
, io_hdr
, cb
, opaque
);
2964 acb
= g_new(RawPosixAIOData
, 1);
2966 acb
->aio_type
= QEMU_AIO_IOCTL
;
2967 acb
->aio_fildes
= s
->fd
;
2968 acb
->aio_offset
= 0;
2969 acb
->aio_ioctl_buf
= buf
;
2970 acb
->aio_ioctl_cmd
= req
;
2971 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
2972 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
2976 static int fd_open(BlockDriverState
*bs
)
2978 BDRVRawState
*s
= bs
->opaque
;
2980 /* this is just to ensure s->fd is sane (its called by io ops) */
2986 static coroutine_fn BlockAIOCB
*hdev_aio_pdiscard(BlockDriverState
*bs
,
2987 int64_t offset
, int bytes
,
2988 BlockCompletionFunc
*cb
, void *opaque
)
2990 BDRVRawState
*s
= bs
->opaque
;
2992 if (fd_open(bs
) < 0) {
2995 return paio_submit(bs
, s
->fd
, offset
, NULL
, bytes
,
2996 cb
, opaque
, QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
2999 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3000 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3002 BDRVRawState
*s
= bs
->opaque
;
3009 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
3010 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3011 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
3012 } else if (s
->discard_zeroes
) {
3013 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3014 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
3019 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3024 struct stat stat_buf
;
3025 int64_t total_size
= 0;
3028 /* This function is used by both protocol block drivers and therefore either
3029 * of these prefixes may be given.
3030 * The return value has to be stored somewhere, otherwise this is an error
3031 * due to -Werror=unused-value. */
3033 strstart(filename
, "host_device:", &filename
) ||
3034 strstart(filename
, "host_cdrom:" , &filename
);
3038 ret
= raw_normalize_devicepath(&filename
);
3040 error_setg_errno(errp
, -ret
, "Could not normalize device path");
3044 /* Read out options */
3045 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3048 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3051 error_setg_errno(errp
, -ret
, "Could not open device");
3055 if (fstat(fd
, &stat_buf
) < 0) {
3057 error_setg_errno(errp
, -ret
, "Could not stat device");
3058 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3060 "The given file is neither a block nor a character device");
3062 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3063 error_setg(errp
, "Device is too small");
3067 if (!ret
&& total_size
) {
3068 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3069 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3070 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3073 ret
= qemu_write_full(fd
, buf
, zero_size
);
3074 ret
= ret
== zero_size
? 0 : -errno
;
3081 static BlockDriver bdrv_host_device
= {
3082 .format_name
= "host_device",
3083 .protocol_name
= "host_device",
3084 .instance_size
= sizeof(BDRVRawState
),
3085 .bdrv_needs_filename
= true,
3086 .bdrv_probe_device
= hdev_probe_device
,
3087 .bdrv_parse_filename
= hdev_parse_filename
,
3088 .bdrv_file_open
= hdev_open
,
3089 .bdrv_close
= raw_close
,
3090 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3091 .bdrv_reopen_commit
= raw_reopen_commit
,
3092 .bdrv_reopen_abort
= raw_reopen_abort
,
3093 .bdrv_co_create_opts
= hdev_co_create_opts
,
3094 .create_opts
= &raw_create_opts
,
3095 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3096 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3098 .bdrv_co_preadv
= raw_co_preadv
,
3099 .bdrv_co_pwritev
= raw_co_pwritev
,
3100 .bdrv_aio_flush
= raw_aio_flush
,
3101 .bdrv_aio_pdiscard
= hdev_aio_pdiscard
,
3102 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3103 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3104 .bdrv_refresh_limits
= raw_refresh_limits
,
3105 .bdrv_io_plug
= raw_aio_plug
,
3106 .bdrv_io_unplug
= raw_aio_unplug
,
3108 .bdrv_truncate
= raw_truncate
,
3109 .bdrv_getlength
= raw_getlength
,
3110 .bdrv_get_info
= raw_get_info
,
3111 .bdrv_get_allocated_file_size
3112 = raw_get_allocated_file_size
,
3113 .bdrv_check_perm
= raw_check_perm
,
3114 .bdrv_set_perm
= raw_set_perm
,
3115 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3116 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3117 .bdrv_probe_geometry
= hdev_probe_geometry
,
3119 /* generic scsi device */
3121 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3125 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3126 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3129 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3134 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3137 BDRVRawState
*s
= bs
->opaque
;
3141 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3142 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, errp
);
3145 static int cdrom_probe_device(const char *filename
)
3151 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3155 ret
= fstat(fd
, &st
);
3156 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3160 /* Attempt to detect via a CDROM specific ioctl */
3161 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3171 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3173 BDRVRawState
*s
= bs
->opaque
;
3176 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3177 return ret
== CDS_DISC_OK
;
3180 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3182 BDRVRawState
*s
= bs
->opaque
;
3185 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3186 perror("CDROMEJECT");
3188 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3189 perror("CDROMEJECT");
3193 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3195 BDRVRawState
*s
= bs
->opaque
;
3197 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3199 * Note: an error can happen if the distribution automatically
3202 /* perror("CDROM_LOCKDOOR"); */
3206 static BlockDriver bdrv_host_cdrom
= {
3207 .format_name
= "host_cdrom",
3208 .protocol_name
= "host_cdrom",
3209 .instance_size
= sizeof(BDRVRawState
),
3210 .bdrv_needs_filename
= true,
3211 .bdrv_probe_device
= cdrom_probe_device
,
3212 .bdrv_parse_filename
= cdrom_parse_filename
,
3213 .bdrv_file_open
= cdrom_open
,
3214 .bdrv_close
= raw_close
,
3215 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3216 .bdrv_reopen_commit
= raw_reopen_commit
,
3217 .bdrv_reopen_abort
= raw_reopen_abort
,
3218 .bdrv_co_create_opts
= hdev_co_create_opts
,
3219 .create_opts
= &raw_create_opts
,
3220 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3223 .bdrv_co_preadv
= raw_co_preadv
,
3224 .bdrv_co_pwritev
= raw_co_pwritev
,
3225 .bdrv_aio_flush
= raw_aio_flush
,
3226 .bdrv_refresh_limits
= raw_refresh_limits
,
3227 .bdrv_io_plug
= raw_aio_plug
,
3228 .bdrv_io_unplug
= raw_aio_unplug
,
3230 .bdrv_truncate
= raw_truncate
,
3231 .bdrv_getlength
= raw_getlength
,
3232 .has_variable_length
= true,
3233 .bdrv_get_allocated_file_size
3234 = raw_get_allocated_file_size
,
3236 /* removable device support */
3237 .bdrv_is_inserted
= cdrom_is_inserted
,
3238 .bdrv_eject
= cdrom_eject
,
3239 .bdrv_lock_medium
= cdrom_lock_medium
,
3241 /* generic scsi device */
3242 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3244 #endif /* __linux__ */
3246 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3247 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3250 BDRVRawState
*s
= bs
->opaque
;
3251 Error
*local_err
= NULL
;
3256 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
3258 error_propagate(errp
, local_err
);
3262 /* make sure the door isn't locked at this time */
3263 ioctl(s
->fd
, CDIOCALLOW
);
3267 static int cdrom_probe_device(const char *filename
)
3269 if (strstart(filename
, "/dev/cd", NULL
) ||
3270 strstart(filename
, "/dev/acd", NULL
))
3275 static int cdrom_reopen(BlockDriverState
*bs
)
3277 BDRVRawState
*s
= bs
->opaque
;
3281 * Force reread of possibly changed/newly loaded disc,
3282 * FreeBSD seems to not notice sometimes...
3286 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3293 /* make sure the door isn't locked at this time */
3294 ioctl(s
->fd
, CDIOCALLOW
);
3298 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3300 return raw_getlength(bs
) > 0;
3303 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3305 BDRVRawState
*s
= bs
->opaque
;
3310 (void) ioctl(s
->fd
, CDIOCALLOW
);
3313 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3314 perror("CDIOCEJECT");
3316 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3317 perror("CDIOCCLOSE");
3323 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3325 BDRVRawState
*s
= bs
->opaque
;
3329 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3331 * Note: an error can happen if the distribution automatically
3334 /* perror("CDROM_LOCKDOOR"); */
3338 static BlockDriver bdrv_host_cdrom
= {
3339 .format_name
= "host_cdrom",
3340 .protocol_name
= "host_cdrom",
3341 .instance_size
= sizeof(BDRVRawState
),
3342 .bdrv_needs_filename
= true,
3343 .bdrv_probe_device
= cdrom_probe_device
,
3344 .bdrv_parse_filename
= cdrom_parse_filename
,
3345 .bdrv_file_open
= cdrom_open
,
3346 .bdrv_close
= raw_close
,
3347 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3348 .bdrv_reopen_commit
= raw_reopen_commit
,
3349 .bdrv_reopen_abort
= raw_reopen_abort
,
3350 .bdrv_co_create_opts
= hdev_co_create_opts
,
3351 .create_opts
= &raw_create_opts
,
3353 .bdrv_co_preadv
= raw_co_preadv
,
3354 .bdrv_co_pwritev
= raw_co_pwritev
,
3355 .bdrv_aio_flush
= raw_aio_flush
,
3356 .bdrv_refresh_limits
= raw_refresh_limits
,
3357 .bdrv_io_plug
= raw_aio_plug
,
3358 .bdrv_io_unplug
= raw_aio_unplug
,
3360 .bdrv_truncate
= raw_truncate
,
3361 .bdrv_getlength
= raw_getlength
,
3362 .has_variable_length
= true,
3363 .bdrv_get_allocated_file_size
3364 = raw_get_allocated_file_size
,
3366 /* removable device support */
3367 .bdrv_is_inserted
= cdrom_is_inserted
,
3368 .bdrv_eject
= cdrom_eject
,
3369 .bdrv_lock_medium
= cdrom_lock_medium
,
3371 #endif /* __FreeBSD__ */
3373 static void bdrv_file_init(void)
3376 * Register all the drivers. Note that order is important, the driver
3377 * registered last will get probed first.
3379 bdrv_register(&bdrv_file
);
3380 bdrv_register(&bdrv_host_device
);
3382 bdrv_register(&bdrv_host_cdrom
);
3384 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3385 bdrv_register(&bdrv_host_cdrom
);
3389 block_init(bdrv_file_init
);