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 */
197 PreallocMode prealloc
;
203 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
204 static int cdrom_reopen(BlockDriverState
*bs
);
207 #if defined(__NetBSD__)
208 static int raw_normalize_devicepath(const char **filename
)
210 static char namebuf
[PATH_MAX
];
211 const char *dp
, *fname
;
215 dp
= strrchr(fname
, '/');
216 if (lstat(fname
, &sb
) < 0) {
217 fprintf(stderr
, "%s: stat failed: %s\n",
218 fname
, strerror(errno
));
222 if (!S_ISBLK(sb
.st_mode
)) {
227 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
229 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
230 (int)(dp
- fname
), fname
, dp
+ 1);
232 fprintf(stderr
, "%s is a block device", fname
);
234 fprintf(stderr
, ", using %s\n", *filename
);
239 static int raw_normalize_devicepath(const char **filename
)
246 * Get logical block size via ioctl. On success store it in @sector_size_p.
248 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
250 unsigned int sector_size
;
251 bool success
= false;
255 static const unsigned long ioctl_list
[] = {
259 #ifdef DKIOCGETBLOCKSIZE
262 #ifdef DIOCGSECTORSIZE
267 /* Try a few ioctls to get the right size */
268 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
269 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
270 *sector_size_p
= sector_size
;
275 return success
? 0 : -errno
;
279 * Get physical block size of @fd.
280 * On success, store it in @blk_size and return 0.
281 * On failure, return -errno.
283 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
286 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
295 /* Check if read is allowed with given memory buffer and length.
297 * This function is used to check O_DIRECT memory buffer and request alignment.
299 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
301 ssize_t ret
= pread(fd
, buf
, len
, 0);
308 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
309 * other errors (e.g. real I/O error), which could happen on a failed
310 * drive, since we only care about probing alignment.
312 if (errno
!= EINVAL
) {
320 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
322 BDRVRawState
*s
= bs
->opaque
;
324 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
326 /* For SCSI generic devices the alignment is not really used.
327 With buffered I/O, we don't have any restrictions. */
328 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
329 bs
->bl
.request_alignment
= 1;
334 bs
->bl
.request_alignment
= 0;
336 /* Let's try to use the logical blocksize for the alignment. */
337 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
338 bs
->bl
.request_alignment
= 0;
343 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
344 bs
->bl
.request_alignment
= da
.d_miniosz
;
345 /* The kernel returns wrong information for d_mem */
346 /* s->buf_align = da.d_mem; */
351 /* If we could not get the sizes so far, we can only guess them */
354 buf
= qemu_memalign(max_align
, 2 * max_align
);
355 for (align
= 512; align
<= max_align
; align
<<= 1) {
356 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
357 s
->buf_align
= align
;
364 if (!bs
->bl
.request_alignment
) {
366 buf
= qemu_memalign(s
->buf_align
, max_align
);
367 for (align
= 512; align
<= max_align
; align
<<= 1) {
368 if (raw_is_io_aligned(fd
, buf
, align
)) {
369 bs
->bl
.request_alignment
= align
;
376 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
377 error_setg(errp
, "Could not find working O_DIRECT alignment");
378 error_append_hint(errp
, "Try cache.direct=off\n");
382 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
384 assert(open_flags
!= NULL
);
386 *open_flags
|= O_BINARY
;
387 *open_flags
&= ~O_ACCMODE
;
388 if (bdrv_flags
& BDRV_O_RDWR
) {
389 *open_flags
|= O_RDWR
;
391 *open_flags
|= O_RDONLY
;
394 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
395 * and O_DIRECT for no caching. */
396 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
397 *open_flags
|= O_DIRECT
;
401 static void raw_parse_filename(const char *filename
, QDict
*options
,
404 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
407 static QemuOptsList raw_runtime_opts
= {
409 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
413 .type
= QEMU_OPT_STRING
,
414 .help
= "File name of the image",
418 .type
= QEMU_OPT_STRING
,
419 .help
= "host AIO implementation (threads, native)",
423 .type
= QEMU_OPT_STRING
,
424 .help
= "file locking mode (on/off/auto, default: auto)",
427 .name
= "pr-manager",
428 .type
= QEMU_OPT_STRING
,
429 .help
= "id of persistent reservation manager object (default: none)",
432 .name
= "x-check-cache-dropped",
433 .type
= QEMU_OPT_BOOL
,
434 .help
= "check that page cache was dropped on live migration (default: off)"
436 { /* end of list */ }
440 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
441 int bdrv_flags
, int open_flags
,
442 bool device
, Error
**errp
)
444 BDRVRawState
*s
= bs
->opaque
;
446 Error
*local_err
= NULL
;
447 const char *filename
= NULL
;
449 BlockdevAioOptions aio
, aio_default
;
454 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
455 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
457 error_propagate(errp
, local_err
);
462 filename
= qemu_opt_get(opts
, "filename");
464 ret
= raw_normalize_devicepath(&filename
);
466 error_setg_errno(errp
, -ret
, "Could not normalize device path");
470 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
471 ? BLOCKDEV_AIO_OPTIONS_NATIVE
472 : BLOCKDEV_AIO_OPTIONS_THREADS
;
473 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
474 qemu_opt_get(opts
, "aio"),
475 aio_default
, &local_err
);
477 error_propagate(errp
, local_err
);
481 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
483 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
484 qemu_opt_get(opts
, "locking"),
485 ON_OFF_AUTO_AUTO
, &local_err
);
487 error_propagate(errp
, local_err
);
494 if (!qemu_has_ofd_lock()) {
496 "File lock requested but OFD locking syscall is "
497 "unavailable, falling back to POSIX file locks.\n"
498 "Due to the implementation, locks can be lost "
502 case ON_OFF_AUTO_OFF
:
505 case ON_OFF_AUTO_AUTO
:
506 s
->use_lock
= qemu_has_ofd_lock();
512 str
= qemu_opt_get(opts
, "pr-manager");
514 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
516 error_propagate(errp
, local_err
);
522 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
525 s
->open_flags
= open_flags
;
526 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
529 fd
= qemu_open(filename
, s
->open_flags
, 0644);
532 error_setg_errno(errp
, errno
, "Could not open '%s'", filename
);
542 fd
= qemu_open(filename
, s
->open_flags
);
545 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
553 s
->shared_perm
= BLK_PERM_ALL
;
555 #ifdef CONFIG_LINUX_AIO
556 /* Currently Linux does AIO only for files opened with O_DIRECT */
557 if (s
->use_linux_aio
) {
558 if (!(s
->open_flags
& O_DIRECT
)) {
559 error_setg(errp
, "aio=native was specified, but it requires "
560 "cache.direct=on, which was not specified.");
564 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
565 error_prepend(errp
, "Unable to use native AIO: ");
570 if (s
->use_linux_aio
) {
571 error_setg(errp
, "aio=native was specified, but is not supported "
576 #endif /* !defined(CONFIG_LINUX_AIO) */
578 s
->has_discard
= true;
579 s
->has_write_zeroes
= true;
580 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
581 s
->needs_alignment
= true;
584 if (fstat(s
->fd
, &st
) < 0) {
586 error_setg_errno(errp
, errno
, "Could not stat file");
591 if (S_ISBLK(st
.st_mode
)) {
592 warn_report("Opening a block device as a file using the '%s' "
593 "driver is deprecated", bs
->drv
->format_name
);
594 } else if (S_ISCHR(st
.st_mode
)) {
595 warn_report("Opening a character device as a file using the '%s' "
596 "driver is deprecated", bs
->drv
->format_name
);
597 } else if (!S_ISREG(st
.st_mode
)) {
598 error_setg(errp
, "A regular file was expected by the '%s' driver, "
599 "but something else was given", bs
->drv
->format_name
);
603 s
->discard_zeroes
= true;
604 s
->has_fallocate
= true;
607 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
608 error_setg(errp
, "'%s' driver expects either "
609 "a character or block device", bs
->drv
->format_name
);
615 if (S_ISBLK(st
.st_mode
)) {
616 #ifdef BLKDISCARDZEROES
618 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
619 s
->discard_zeroes
= true;
623 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
624 * not rely on the contents of discarded blocks unless using O_DIRECT.
625 * Same for BLKZEROOUT.
627 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
628 s
->discard_zeroes
= false;
629 s
->has_write_zeroes
= false;
634 if (S_ISCHR(st
.st_mode
)) {
636 * The file is a char device (disk), which on FreeBSD isn't behind
637 * a pager, so force all requests to be aligned. This is needed
638 * so QEMU makes sure all IO operations on the device are aligned
639 * to sector size, or else FreeBSD will reject them with EINVAL.
641 s
->needs_alignment
= true;
646 if (platform_test_xfs_fd(s
->fd
)) {
651 bs
->supported_zero_flags
= s
->discard_zeroes
? BDRV_REQ_MAY_UNMAP
: 0;
654 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
661 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
664 BDRVRawState
*s
= bs
->opaque
;
666 s
->type
= FTYPE_FILE
;
667 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
676 #define PERM_FOREACH(i) \
677 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
679 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
680 * file; if @unlock == true, also unlock the unneeded bytes.
681 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
683 static int raw_apply_lock_bytes(int fd
,
684 uint64_t perm_lock_bits
,
685 uint64_t shared_perm_lock_bits
,
686 bool unlock
, Error
**errp
)
692 int off
= RAW_LOCK_PERM_BASE
+ i
;
693 if (perm_lock_bits
& (1ULL << i
)) {
694 ret
= qemu_lock_fd(fd
, off
, 1, false);
696 error_setg(errp
, "Failed to lock byte %d", off
);
700 ret
= qemu_unlock_fd(fd
, off
, 1);
702 error_setg(errp
, "Failed to unlock byte %d", off
);
708 int off
= RAW_LOCK_SHARED_BASE
+ i
;
709 if (shared_perm_lock_bits
& (1ULL << i
)) {
710 ret
= qemu_lock_fd(fd
, off
, 1, false);
712 error_setg(errp
, "Failed to lock byte %d", off
);
716 ret
= qemu_unlock_fd(fd
, off
, 1);
718 error_setg(errp
, "Failed to unlock byte %d", off
);
726 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
727 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
734 int off
= RAW_LOCK_SHARED_BASE
+ i
;
735 uint64_t p
= 1ULL << i
;
737 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
739 char *perm_name
= bdrv_perm_names(p
);
741 "Failed to get \"%s\" lock",
744 error_append_hint(errp
,
745 "Is another process using the image?\n");
751 int off
= RAW_LOCK_PERM_BASE
+ i
;
752 uint64_t p
= 1ULL << i
;
753 if (!(shared_perm
& p
)) {
754 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
756 char *perm_name
= bdrv_perm_names(p
);
758 "Failed to get shared \"%s\" lock",
761 error_append_hint(errp
,
762 "Is another process using the image?\n");
770 static int raw_handle_perm_lock(BlockDriverState
*bs
,
772 uint64_t new_perm
, uint64_t new_shared
,
775 BDRVRawState
*s
= bs
->opaque
;
777 Error
*local_err
= NULL
;
783 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
787 assert(s
->lock_fd
> 0);
791 ret
= raw_apply_lock_bytes(s
->lock_fd
, s
->perm
| new_perm
,
792 ~s
->shared_perm
| ~new_shared
,
795 ret
= raw_check_lock_bytes(s
->lock_fd
, new_perm
, new_shared
, errp
);
801 /* fall through to unlock bytes. */
803 raw_apply_lock_bytes(s
->lock_fd
, s
->perm
, ~s
->shared_perm
,
806 /* Theoretically the above call only unlocks bytes and it cannot
807 * fail. Something weird happened, report it.
809 error_report_err(local_err
);
813 raw_apply_lock_bytes(s
->lock_fd
, new_perm
, ~new_shared
,
816 /* Theoretically the above call only unlocks bytes and it cannot
817 * fail. Something weird happened, report it.
819 error_report_err(local_err
);
826 static int raw_reopen_prepare(BDRVReopenState
*state
,
827 BlockReopenQueue
*queue
, Error
**errp
)
830 BDRVRawReopenState
*rs
;
833 Error
*local_err
= NULL
;
835 assert(state
!= NULL
);
836 assert(state
->bs
!= NULL
);
838 s
= state
->bs
->opaque
;
840 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
844 /* Handle options changes */
845 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
846 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
848 error_propagate(errp
, local_err
);
853 rs
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
854 s
->check_cache_dropped
);
856 if (s
->type
== FTYPE_CD
) {
857 rs
->open_flags
|= O_NONBLOCK
;
860 raw_parse_flags(state
->flags
, &rs
->open_flags
);
862 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
864 fcntl_flags
|= O_NOATIME
;
868 /* Not all operating systems have O_ASYNC, and those that don't
869 * will not let us track the state into rs->open_flags (typically
870 * you achieve the same effect with an ioctl, for example I_SETSIG
871 * on Solaris). But we do not use O_ASYNC, so that's fine.
873 assert((s
->open_flags
& O_ASYNC
) == 0);
876 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
877 /* dup the original fd */
878 rs
->fd
= qemu_dup(s
->fd
);
880 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
888 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
890 const char *normalized_filename
= state
->bs
->filename
;
891 ret
= raw_normalize_devicepath(&normalized_filename
);
893 error_setg_errno(errp
, -ret
, "Could not normalize device path");
895 assert(!(rs
->open_flags
& O_CREAT
));
896 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
898 error_setg_errno(errp
, errno
, "Could not reopen file");
904 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
905 * alignment with the new fd. */
907 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
911 error_propagate(errp
, local_err
);
921 static void raw_reopen_commit(BDRVReopenState
*state
)
923 BDRVRawReopenState
*rs
= state
->opaque
;
924 BDRVRawState
*s
= state
->bs
->opaque
;
926 s
->check_cache_dropped
= rs
->check_cache_dropped
;
927 s
->open_flags
= rs
->open_flags
;
932 g_free(state
->opaque
);
933 state
->opaque
= NULL
;
937 static void raw_reopen_abort(BDRVReopenState
*state
)
939 BDRVRawReopenState
*rs
= state
->opaque
;
941 /* nothing to do if NULL, we didn't get far enough */
950 g_free(state
->opaque
);
951 state
->opaque
= NULL
;
954 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
958 short max_sectors
= 0;
959 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
961 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
962 return max_sectors
<< BDRV_SECTOR_BITS
;
971 static int hdev_get_max_segments(const struct stat
*st
)
981 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
982 major(st
->st_rdev
), minor(st
->st_rdev
));
983 fd
= open(sysfspath
, O_RDONLY
);
989 ret
= read(fd
, buf
, sizeof(buf
) - 1);
990 } while (ret
== -1 && errno
== EINTR
);
994 } else if (ret
== 0) {
999 /* The file is ended with '\n', pass 'end' to accept that. */
1000 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1001 if (ret
== 0 && end
&& *end
== '\n') {
1016 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1018 BDRVRawState
*s
= bs
->opaque
;
1021 if (!fstat(s
->fd
, &st
)) {
1022 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1023 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1024 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1025 bs
->bl
.max_transfer
= pow2floor(ret
);
1027 ret
= hdev_get_max_segments(&st
);
1029 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1030 ret
* getpagesize());
1035 raw_probe_alignment(bs
, s
->fd
, errp
);
1036 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1037 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1040 static int check_for_dasd(int fd
)
1043 struct dasd_information2_t info
= {0};
1045 return ioctl(fd
, BIODASDINFO2
, &info
);
1052 * Try to get @bs's logical and physical block size.
1053 * On success, store them in @bsz and return zero.
1054 * On failure, return negative errno.
1056 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1058 BDRVRawState
*s
= bs
->opaque
;
1061 /* If DASD, get blocksizes */
1062 if (check_for_dasd(s
->fd
) < 0) {
1065 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1069 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1073 * Try to get @bs's geometry: cyls, heads, sectors.
1074 * On success, store them in @geo and return 0.
1075 * On failure return -errno.
1076 * (Allows block driver to assign default geometry values that guest sees)
1079 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1081 BDRVRawState
*s
= bs
->opaque
;
1082 struct hd_geometry ioctl_geo
= {0};
1084 /* If DASD, get its geometry */
1085 if (check_for_dasd(s
->fd
) < 0) {
1088 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1091 /* HDIO_GETGEO may return success even though geo contains zeros
1092 (e.g. certain multipath setups) */
1093 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1096 /* Do not return a geometry for partition */
1097 if (ioctl_geo
.start
!= 0) {
1100 geo
->heads
= ioctl_geo
.heads
;
1101 geo
->sectors
= ioctl_geo
.sectors
;
1102 geo
->cylinders
= ioctl_geo
.cylinders
;
1106 #else /* __linux__ */
1107 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1113 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1117 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1125 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1127 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1130 if (s
->page_cache_inconsistent
) {
1134 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1136 /* There is no clear definition of the semantics of a failing fsync(),
1137 * so we may have to assume the worst. The sad truth is that this
1138 * assumption is correct for Linux. Some pages are now probably marked
1139 * clean in the page cache even though they are inconsistent with the
1140 * on-disk contents. The next fdatasync() call would succeed, but no
1141 * further writeback attempt will be made. We can't get back to a state
1142 * in which we know what is on disk (we would have to rewrite
1143 * everything that was touched since the last fdatasync() at least), so
1144 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1145 * really defined, I have little hope that other OSes are doing better.
1147 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1149 if ((s
->open_flags
& O_DIRECT
) == 0) {
1150 s
->page_cache_inconsistent
= true;
1157 #ifdef CONFIG_PREADV
1159 static bool preadv_present
= true;
1162 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1164 return preadv(fd
, iov
, nr_iov
, offset
);
1168 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1170 return pwritev(fd
, iov
, nr_iov
, offset
);
1175 static bool preadv_present
= false;
1178 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1184 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1191 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1196 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1197 len
= qemu_pwritev(aiocb
->aio_fildes
,
1202 len
= qemu_preadv(aiocb
->aio_fildes
,
1206 } while (len
== -1 && errno
== EINTR
);
1215 * Read/writes the data to/from a given linear buffer.
1217 * Returns the number of bytes handles or -errno in case of an error. Short
1218 * reads are only returned if the end of the file is reached.
1220 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1225 while (offset
< aiocb
->aio_nbytes
) {
1226 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1227 len
= pwrite(aiocb
->aio_fildes
,
1228 (const char *)buf
+ offset
,
1229 aiocb
->aio_nbytes
- offset
,
1230 aiocb
->aio_offset
+ offset
);
1232 len
= pread(aiocb
->aio_fildes
,
1234 aiocb
->aio_nbytes
- offset
,
1235 aiocb
->aio_offset
+ offset
);
1237 if (len
== -1 && errno
== EINTR
) {
1239 } else if (len
== -1 && errno
== EINVAL
&&
1240 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1241 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1243 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1244 * after a short read. Assume that O_DIRECT short reads only occur
1245 * at EOF. Therefore this is a short read, not an I/O error.
1248 } else if (len
== -1) {
1251 } else if (len
== 0) {
1260 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1265 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1267 * If there is just a single buffer, and it is properly aligned
1268 * we can just use plain pread/pwrite without any problems.
1270 if (aiocb
->aio_niov
== 1) {
1271 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1274 * We have more than one iovec, and all are properly aligned.
1276 * Try preadv/pwritev first and fall back to linearizing the
1277 * buffer if it's not supported.
1279 if (preadv_present
) {
1280 nbytes
= handle_aiocb_rw_vector(aiocb
);
1281 if (nbytes
== aiocb
->aio_nbytes
||
1282 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1285 preadv_present
= false;
1289 * XXX(hch): short read/write. no easy way to handle the reminder
1290 * using these interfaces. For now retry using plain
1296 * Ok, we have to do it the hard way, copy all segments into
1297 * a single aligned buffer.
1299 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1304 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1308 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1309 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1310 p
+= aiocb
->aio_iov
[i
].iov_len
;
1312 assert(p
- buf
== aiocb
->aio_nbytes
);
1315 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1316 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1318 size_t count
= aiocb
->aio_nbytes
, copy
;
1321 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1323 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1324 copy
= aiocb
->aio_iov
[i
].iov_len
;
1326 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1327 assert(count
>= copy
);
1339 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1341 struct xfs_flock64 fl
;
1344 memset(&fl
, 0, sizeof(fl
));
1345 fl
.l_whence
= SEEK_SET
;
1346 fl
.l_start
= offset
;
1349 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1351 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1358 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1360 struct xfs_flock64 fl
;
1363 memset(&fl
, 0, sizeof(fl
));
1364 fl
.l_whence
= SEEK_SET
;
1365 fl
.l_start
= offset
;
1368 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1370 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1378 static int translate_err(int err
)
1380 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1387 #ifdef CONFIG_FALLOCATE
1388 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1391 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1394 } while (errno
== EINTR
);
1395 return translate_err(-errno
);
1399 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1402 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1404 if (!s
->has_write_zeroes
) {
1410 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1411 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1414 } while (errno
== EINTR
);
1416 ret
= translate_err(-errno
);
1419 if (ret
== -ENOTSUP
) {
1420 s
->has_write_zeroes
= false;
1425 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1427 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1428 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1430 #ifdef CONFIG_FALLOCATE
1434 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1435 return handle_aiocb_write_zeroes_block(aiocb
);
1440 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1444 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1445 if (s
->has_write_zeroes
) {
1446 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1447 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1448 if (ret
== 0 || ret
!= -ENOTSUP
) {
1451 s
->has_write_zeroes
= false;
1455 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1456 if (s
->has_discard
&& s
->has_fallocate
) {
1457 int ret
= do_fallocate(s
->fd
,
1458 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1459 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1461 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1462 if (ret
== 0 || ret
!= -ENOTSUP
) {
1465 s
->has_fallocate
= false;
1466 } else if (ret
!= -ENOTSUP
) {
1469 s
->has_discard
= false;
1474 #ifdef CONFIG_FALLOCATE
1475 /* Last resort: we are trying to extend the file with zeroed data. This
1476 * can be done via fallocate(fd, 0) */
1477 len
= bdrv_getlength(aiocb
->bs
);
1478 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1479 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1480 if (ret
== 0 || ret
!= -ENOTSUP
) {
1483 s
->has_fallocate
= false;
1490 #ifndef HAVE_COPY_FILE_RANGE
1491 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1492 off_t
*out_off
, size_t len
, unsigned int flags
)
1494 #ifdef __NR_copy_file_range
1495 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1496 out_off
, len
, flags
);
1504 static ssize_t
handle_aiocb_copy_range(RawPosixAIOData
*aiocb
)
1506 uint64_t bytes
= aiocb
->aio_nbytes
;
1507 off_t in_off
= aiocb
->aio_offset
;
1508 off_t out_off
= aiocb
->aio_offset2
;
1511 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1512 aiocb
->aio_fd2
, &out_off
,
1514 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1515 aiocb
->aio_fd2
, out_off
, bytes
, 0, ret
);
1517 /* No progress (e.g. when beyond EOF), let the caller fall back to
1536 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1538 int ret
= -EOPNOTSUPP
;
1539 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1541 if (!s
->has_discard
) {
1545 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1548 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1549 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1552 } while (errno
== EINTR
);
1559 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1563 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1564 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1565 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1569 ret
= translate_err(ret
);
1570 if (ret
== -ENOTSUP
) {
1571 s
->has_discard
= false;
1576 static int handle_aiocb_truncate(RawPosixAIOData
*aiocb
)
1579 int64_t current_length
= 0;
1582 int fd
= aiocb
->aio_fildes
;
1583 int64_t offset
= aiocb
->aio_offset
;
1584 Error
**errp
= aiocb
->errp
;
1586 if (fstat(fd
, &st
) < 0) {
1588 error_setg_errno(errp
, -result
, "Could not stat file");
1592 current_length
= st
.st_size
;
1593 if (current_length
> offset
&& aiocb
->prealloc
!= PREALLOC_MODE_OFF
) {
1594 error_setg(errp
, "Cannot use preallocation for shrinking files");
1598 switch (aiocb
->prealloc
) {
1599 #ifdef CONFIG_POSIX_FALLOCATE
1600 case PREALLOC_MODE_FALLOC
:
1602 * Truncating before posix_fallocate() makes it about twice slower on
1603 * file systems that do not support fallocate(), trying to check if a
1604 * block is allocated before allocating it, so don't do that here.
1606 if (offset
!= current_length
) {
1607 result
= -posix_fallocate(fd
, current_length
,
1608 offset
- current_length
);
1610 /* posix_fallocate() doesn't set errno. */
1611 error_setg_errno(errp
, -result
,
1612 "Could not preallocate new data");
1619 case PREALLOC_MODE_FULL
:
1621 int64_t num
= 0, left
= offset
- current_length
;
1625 * Knowing the final size from the beginning could allow the file
1626 * system driver to do less allocations and possibly avoid
1627 * fragmentation of the file.
1629 if (ftruncate(fd
, offset
) != 0) {
1631 error_setg_errno(errp
, -result
, "Could not resize file");
1635 buf
= g_malloc0(65536);
1637 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1638 if (seek_result
< 0) {
1640 error_setg_errno(errp
, -result
,
1641 "Failed to seek to the old end of file");
1646 num
= MIN(left
, 65536);
1647 result
= write(fd
, buf
, num
);
1649 if (errno
== EINTR
) {
1653 error_setg_errno(errp
, -result
,
1654 "Could not write zeros for preallocation");
1663 error_setg_errno(errp
, -result
,
1664 "Could not flush file to disk");
1670 case PREALLOC_MODE_OFF
:
1671 if (ftruncate(fd
, offset
) != 0) {
1673 error_setg_errno(errp
, -result
, "Could not resize file");
1678 error_setg(errp
, "Unsupported preallocation mode: %s",
1679 PreallocMode_str(aiocb
->prealloc
));
1685 if (ftruncate(fd
, current_length
) < 0) {
1686 error_report("Failed to restore old file length: %s",
1695 static int aio_worker(void *arg
)
1697 RawPosixAIOData
*aiocb
= arg
;
1700 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1702 ret
= handle_aiocb_rw(aiocb
);
1703 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1704 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1705 0, aiocb
->aio_nbytes
- ret
);
1707 ret
= aiocb
->aio_nbytes
;
1709 if (ret
== aiocb
->aio_nbytes
) {
1711 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1715 case QEMU_AIO_WRITE
:
1716 ret
= handle_aiocb_rw(aiocb
);
1717 if (ret
== aiocb
->aio_nbytes
) {
1719 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1723 case QEMU_AIO_FLUSH
:
1724 ret
= handle_aiocb_flush(aiocb
);
1726 case QEMU_AIO_IOCTL
:
1727 ret
= handle_aiocb_ioctl(aiocb
);
1729 case QEMU_AIO_DISCARD
:
1730 ret
= handle_aiocb_discard(aiocb
);
1732 case QEMU_AIO_WRITE_ZEROES
:
1733 ret
= handle_aiocb_write_zeroes(aiocb
);
1735 case QEMU_AIO_COPY_RANGE
:
1736 ret
= handle_aiocb_copy_range(aiocb
);
1738 case QEMU_AIO_TRUNCATE
:
1739 ret
= handle_aiocb_truncate(aiocb
);
1742 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1751 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1752 int64_t offset
, int fd2
, int64_t offset2
,
1754 int bytes
, int type
)
1756 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1760 acb
->aio_type
= type
;
1761 acb
->aio_fildes
= fd
;
1763 acb
->aio_offset2
= offset2
;
1765 acb
->aio_nbytes
= bytes
;
1766 acb
->aio_offset
= offset
;
1769 acb
->aio_iov
= qiov
->iov
;
1770 acb
->aio_niov
= qiov
->niov
;
1771 assert(qiov
->size
== bytes
);
1774 trace_file_paio_submit_co(offset
, bytes
, type
);
1775 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1776 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1779 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1780 int64_t offset
, QEMUIOVector
*qiov
,
1781 int bytes
, int type
)
1783 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1786 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1787 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1789 BDRVRawState
*s
= bs
->opaque
;
1791 if (fd_open(bs
) < 0)
1795 * Check if the underlying device requires requests to be aligned,
1796 * and if the request we are trying to submit is aligned or not.
1797 * If this is the case tell the low-level driver that it needs
1798 * to copy the buffer.
1800 if (s
->needs_alignment
) {
1801 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1802 type
|= QEMU_AIO_MISALIGNED
;
1803 #ifdef CONFIG_LINUX_AIO
1804 } else if (s
->use_linux_aio
) {
1805 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1806 assert(qiov
->size
== bytes
);
1807 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1812 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1815 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1816 uint64_t bytes
, QEMUIOVector
*qiov
,
1819 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1822 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1823 uint64_t bytes
, QEMUIOVector
*qiov
,
1827 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1830 static void raw_aio_plug(BlockDriverState
*bs
)
1832 #ifdef CONFIG_LINUX_AIO
1833 BDRVRawState
*s
= bs
->opaque
;
1834 if (s
->use_linux_aio
) {
1835 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1836 laio_io_plug(bs
, aio
);
1841 static void raw_aio_unplug(BlockDriverState
*bs
)
1843 #ifdef CONFIG_LINUX_AIO
1844 BDRVRawState
*s
= bs
->opaque
;
1845 if (s
->use_linux_aio
) {
1846 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1847 laio_io_unplug(bs
, aio
);
1852 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1854 BDRVRawState
*s
= bs
->opaque
;
1862 return paio_submit_co(bs
, s
->fd
, 0, NULL
, 0, QEMU_AIO_FLUSH
);
1865 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1866 AioContext
*new_context
)
1868 #ifdef CONFIG_LINUX_AIO
1869 BDRVRawState
*s
= bs
->opaque
;
1870 if (s
->use_linux_aio
) {
1872 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1873 error_reportf_err(local_err
, "Unable to use native AIO, "
1874 "falling back to thread pool: ");
1875 s
->use_linux_aio
= false;
1881 static void raw_close(BlockDriverState
*bs
)
1883 BDRVRawState
*s
= bs
->opaque
;
1889 if (s
->lock_fd
>= 0) {
1890 qemu_close(s
->lock_fd
);
1896 * Truncates the given regular file @fd to @offset and, when growing, fills the
1897 * new space according to @prealloc.
1899 * Returns: 0 on success, -errno on failure.
1901 static int coroutine_fn
1902 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1903 PreallocMode prealloc
, Error
**errp
)
1905 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1908 *acb
= (RawPosixAIOData
) {
1911 .aio_type
= QEMU_AIO_TRUNCATE
,
1912 .aio_offset
= offset
,
1913 .prealloc
= prealloc
,
1917 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1918 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1919 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1922 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1923 PreallocMode prealloc
, Error
**errp
)
1925 BDRVRawState
*s
= bs
->opaque
;
1929 if (fstat(s
->fd
, &st
)) {
1931 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1935 if (S_ISREG(st
.st_mode
)) {
1936 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
1939 if (prealloc
!= PREALLOC_MODE_OFF
) {
1940 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1941 "non-regular file", PreallocMode_str(prealloc
));
1945 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1946 if (offset
> raw_getlength(bs
)) {
1947 error_setg(errp
, "Cannot grow device files");
1951 error_setg(errp
, "Resizing this file is not supported");
1959 static int64_t raw_getlength(BlockDriverState
*bs
)
1961 BDRVRawState
*s
= bs
->opaque
;
1967 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1968 struct disklabel dl
;
1970 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1972 return (uint64_t)dl
.d_secsize
*
1973 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1977 #elif defined(__NetBSD__)
1978 static int64_t raw_getlength(BlockDriverState
*bs
)
1980 BDRVRawState
*s
= bs
->opaque
;
1986 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1987 struct dkwedge_info dkw
;
1989 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1990 return dkw
.dkw_size
* 512;
1992 struct disklabel dl
;
1994 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1996 return (uint64_t)dl
.d_secsize
*
1997 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2002 #elif defined(__sun__)
2003 static int64_t raw_getlength(BlockDriverState
*bs
)
2005 BDRVRawState
*s
= bs
->opaque
;
2006 struct dk_minfo minfo
;
2016 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2018 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2020 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2024 * There are reports that lseek on some devices fails, but
2025 * irc discussion said that contingency on contingency was overkill.
2027 size
= lseek(s
->fd
, 0, SEEK_END
);
2033 #elif defined(CONFIG_BSD)
2034 static int64_t raw_getlength(BlockDriverState
*bs
)
2036 BDRVRawState
*s
= bs
->opaque
;
2040 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2049 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2052 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2053 #ifdef DIOCGMEDIASIZE
2054 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2055 #elif defined(DIOCGPART)
2058 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2059 size
= pi
.media_size
;
2065 #if defined(__APPLE__) && defined(__MACH__)
2067 uint64_t sectors
= 0;
2068 uint32_t sector_size
= 0;
2070 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2071 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2072 size
= sectors
* sector_size
;
2074 size
= lseek(fd
, 0LL, SEEK_END
);
2081 size
= lseek(fd
, 0LL, SEEK_END
);
2086 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2089 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2090 if (size
== 2048LL * (unsigned)-1)
2092 /* XXX no disc? maybe we need to reopen... */
2093 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2100 size
= lseek(fd
, 0, SEEK_END
);
2108 static int64_t raw_getlength(BlockDriverState
*bs
)
2110 BDRVRawState
*s
= bs
->opaque
;
2119 size
= lseek(s
->fd
, 0, SEEK_END
);
2127 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2130 BDRVRawState
*s
= bs
->opaque
;
2132 if (fstat(s
->fd
, &st
) < 0) {
2135 return (int64_t)st
.st_blocks
* 512;
2138 static int coroutine_fn
2139 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2141 BlockdevCreateOptionsFile
*file_opts
;
2142 Error
*local_err
= NULL
;
2144 uint64_t perm
, shared
;
2147 /* Validate options and set default values */
2148 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2149 file_opts
= &options
->u
.file
;
2151 if (!file_opts
->has_nocow
) {
2152 file_opts
->nocow
= false;
2154 if (!file_opts
->has_preallocation
) {
2155 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2159 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2162 error_setg_errno(errp
, -result
, "Could not create file");
2166 /* Take permissions: We want to discard everything, so we need
2167 * BLK_PERM_WRITE; and truncation to the desired size requires
2169 * On the other hand, we cannot share the RESIZE permission
2170 * because we promise that after this function, the file has the
2171 * size given in the options. If someone else were to resize it
2172 * concurrently, we could not guarantee that.
2173 * Note that after this function, we can no longer guarantee that
2174 * the file is not touched by a third party, so it may be resized
2176 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2177 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2179 /* Step one: Take locks */
2180 result
= raw_apply_lock_bytes(fd
, perm
, ~shared
, false, errp
);
2185 /* Step two: Check that nobody else has taken conflicting locks */
2186 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2191 /* Clear the file by truncating it to 0 */
2192 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2197 if (file_opts
->nocow
) {
2199 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2200 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2201 * will be ignored since any failure of this operation should not
2202 * block the left work.
2205 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2206 attr
|= FS_NOCOW_FL
;
2207 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2212 /* Resize and potentially preallocate the file to the desired
2214 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2215 file_opts
->preallocation
, errp
);
2221 raw_apply_lock_bytes(fd
, 0, 0, true, &local_err
);
2223 /* The above call should not fail, and if it does, that does
2224 * not mean the whole creation operation has failed. So
2225 * report it the user for their convenience, but do not report
2226 * it to the caller. */
2227 error_report_err(local_err
);
2231 if (qemu_close(fd
) != 0 && result
== 0) {
2233 error_setg_errno(errp
, -result
, "Could not close the new file");
2239 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2242 BlockdevCreateOptions options
;
2243 int64_t total_size
= 0;
2245 PreallocMode prealloc
;
2247 Error
*local_err
= NULL
;
2249 /* Skip file: protocol prefix */
2250 strstart(filename
, "file:", &filename
);
2252 /* Read out options */
2253 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2255 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2256 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2257 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2258 PREALLOC_MODE_OFF
, &local_err
);
2261 error_propagate(errp
, local_err
);
2265 options
= (BlockdevCreateOptions
) {
2266 .driver
= BLOCKDEV_DRIVER_FILE
,
2268 .filename
= (char *) filename
,
2270 .has_preallocation
= true,
2271 .preallocation
= prealloc
,
2276 return raw_co_create(&options
, errp
);
2280 * Find allocation range in @bs around offset @start.
2281 * May change underlying file descriptor's file offset.
2282 * If @start is not in a hole, store @start in @data, and the
2283 * beginning of the next hole in @hole, and return 0.
2284 * If @start is in a non-trailing hole, store @start in @hole and the
2285 * beginning of the next non-hole in @data, and return 0.
2286 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2287 * If we can't find out, return a negative errno other than -ENXIO.
2289 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2290 off_t
*data
, off_t
*hole
)
2292 #if defined SEEK_HOLE && defined SEEK_DATA
2293 BDRVRawState
*s
= bs
->opaque
;
2298 * D1. offs == start: start is in data
2299 * D2. offs > start: start is in a hole, next data at offs
2300 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2301 * or start is beyond EOF
2302 * If the latter happens, the file has been truncated behind
2303 * our back since we opened it. All bets are off then.
2304 * Treating like a trailing hole is simplest.
2305 * D4. offs < 0, errno != ENXIO: we learned nothing
2307 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2309 return -errno
; /* D3 or D4 */
2313 /* This is not a valid return by lseek(). We are safe to just return
2314 * -EIO in this case, and we'll treat it like D4. */
2319 /* D2: in hole, next data at offs */
2325 /* D1: in data, end not yet known */
2329 * H1. offs == start: start is in a hole
2330 * If this happens here, a hole has been dug behind our back
2331 * since the previous lseek().
2332 * H2. offs > start: either start is in data, next hole at offs,
2333 * or start is in trailing hole, EOF at offs
2334 * Linux treats trailing holes like any other hole: offs ==
2335 * start. Solaris seeks to EOF instead: offs > start (blech).
2336 * If that happens here, a hole has been dug behind our back
2337 * since the previous lseek().
2338 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2339 * If this happens, the file has been truncated behind our
2340 * back since we opened it. Treat it like a trailing hole.
2341 * H4. offs < 0, errno != ENXIO: we learned nothing
2342 * Pretend we know nothing at all, i.e. "forget" about D1.
2344 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2346 return -errno
; /* D1 and (H3 or H4) */
2350 /* This is not a valid return by lseek(). We are safe to just return
2351 * -EIO in this case, and we'll treat it like H4. */
2357 * D1 and H2: either in data, next hole at offs, or it was in
2358 * data but is now in a trailing hole. In the latter case,
2359 * all bets are off. Treating it as if it there was data all
2360 * the way to EOF is safe, so simply do that.
2375 * Returns the allocation status of the specified offset.
2377 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2379 * 'pnum' is set to the number of bytes (including and immediately following
2380 * the specified offset) that are known to be in the same
2381 * allocated/unallocated state.
2383 * 'bytes' is the max value 'pnum' should be set to.
2385 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2388 int64_t bytes
, int64_t *pnum
,
2390 BlockDriverState
**file
)
2392 off_t data
= 0, hole
= 0;
2404 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2407 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2408 if (ret
== -ENXIO
) {
2411 ret
= BDRV_BLOCK_ZERO
;
2412 } else if (ret
< 0) {
2413 /* No info available, so pretend there are no holes */
2415 ret
= BDRV_BLOCK_DATA
;
2416 } else if (data
== offset
) {
2417 /* On a data extent, compute bytes to the end of the extent,
2418 * possibly including a partial sector at EOF. */
2419 *pnum
= MIN(bytes
, hole
- offset
);
2420 ret
= BDRV_BLOCK_DATA
;
2422 /* On a hole, compute bytes to the beginning of the next extent. */
2423 assert(hole
== offset
);
2424 *pnum
= MIN(bytes
, data
- offset
);
2425 ret
= BDRV_BLOCK_ZERO
;
2429 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2432 #if defined(__linux__)
2433 /* Verify that the file is not in the page cache */
2434 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2436 const size_t window_size
= 128 * 1024 * 1024;
2437 BDRVRawState
*s
= bs
->opaque
;
2438 void *window
= NULL
;
2445 /* mincore(2) page status information requires 1 byte per page */
2446 page_size
= sysconf(_SC_PAGESIZE
);
2447 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2449 end
= raw_getlength(bs
);
2451 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2458 /* Unmap previous window if size has changed */
2459 new_length
= MIN(end
- offset
, window_size
);
2460 if (new_length
!= length
) {
2461 munmap(window
, length
);
2466 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2468 if (new_window
== MAP_FAILED
) {
2469 error_setg_errno(errp
, errno
, "mmap failed");
2473 window
= new_window
;
2474 length
= new_length
;
2476 ret
= mincore(window
, length
, vec
);
2478 error_setg_errno(errp
, errno
, "mincore failed");
2482 vec_end
= DIV_ROUND_UP(length
, page_size
);
2483 for (i
= 0; i
< vec_end
; i
++) {
2485 error_setg(errp
, "page cache still in use!");
2492 munmap(window
, length
);
2497 #endif /* __linux__ */
2499 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2502 BDRVRawState
*s
= bs
->opaque
;
2507 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2511 if (s
->open_flags
& O_DIRECT
) {
2512 return; /* No host kernel page cache */
2515 #if defined(__linux__)
2516 /* This sets the scene for the next syscall... */
2517 ret
= bdrv_co_flush(bs
);
2519 error_setg_errno(errp
, -ret
, "flush failed");
2523 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2524 * process. These limitations are okay because we just fsynced the file,
2525 * we don't use mmap, and the file should not be in use by other processes.
2527 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2528 if (ret
!= 0) { /* the return value is a positive errno */
2529 error_setg_errno(errp
, ret
, "fadvise failed");
2533 if (s
->check_cache_dropped
) {
2534 check_cache_dropped(bs
, errp
);
2536 #else /* __linux__ */
2537 /* Do nothing. Live migration to a remote host with cache.direct=off is
2538 * unsupported on other host operating systems. Cache consistency issues
2539 * may occur but no error is reported here, partly because that's the
2540 * historical behavior and partly because it's hard to differentiate valid
2541 * configurations that should not cause errors.
2543 #endif /* !__linux__ */
2546 static coroutine_fn
int
2547 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2549 BDRVRawState
*s
= bs
->opaque
;
2551 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, QEMU_AIO_DISCARD
);
2554 static int coroutine_fn
raw_co_pwrite_zeroes(
2555 BlockDriverState
*bs
, int64_t offset
,
2556 int bytes
, BdrvRequestFlags flags
)
2558 BDRVRawState
*s
= bs
->opaque
;
2560 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
2561 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2562 QEMU_AIO_WRITE_ZEROES
);
2563 } else if (s
->discard_zeroes
) {
2564 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2570 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2572 BDRVRawState
*s
= bs
->opaque
;
2574 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2578 static QemuOptsList raw_create_opts
= {
2579 .name
= "raw-create-opts",
2580 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2583 .name
= BLOCK_OPT_SIZE
,
2584 .type
= QEMU_OPT_SIZE
,
2585 .help
= "Virtual disk size"
2588 .name
= BLOCK_OPT_NOCOW
,
2589 .type
= QEMU_OPT_BOOL
,
2590 .help
= "Turn off copy-on-write (valid only on btrfs)"
2593 .name
= BLOCK_OPT_PREALLOC
,
2594 .type
= QEMU_OPT_STRING
,
2595 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2597 { /* end of list */ }
2601 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2604 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2607 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2609 BDRVRawState
*s
= bs
->opaque
;
2610 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2612 s
->shared_perm
= shared
;
2615 static void raw_abort_perm_update(BlockDriverState
*bs
)
2617 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2620 static int coroutine_fn
raw_co_copy_range_from(
2621 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2622 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2623 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2625 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2626 read_flags
, write_flags
);
2629 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2631 uint64_t src_offset
,
2633 uint64_t dst_offset
,
2635 BdrvRequestFlags read_flags
,
2636 BdrvRequestFlags write_flags
)
2638 BDRVRawState
*s
= bs
->opaque
;
2639 BDRVRawState
*src_s
;
2641 assert(dst
->bs
== bs
);
2642 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2646 src_s
= src
->bs
->opaque
;
2647 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2650 return paio_submit_co_full(bs
, src_s
->fd
, src_offset
, s
->fd
, dst_offset
,
2651 NULL
, bytes
, QEMU_AIO_COPY_RANGE
);
2654 BlockDriver bdrv_file
= {
2655 .format_name
= "file",
2656 .protocol_name
= "file",
2657 .instance_size
= sizeof(BDRVRawState
),
2658 .bdrv_needs_filename
= true,
2659 .bdrv_probe
= NULL
, /* no probe for protocols */
2660 .bdrv_parse_filename
= raw_parse_filename
,
2661 .bdrv_file_open
= raw_open
,
2662 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2663 .bdrv_reopen_commit
= raw_reopen_commit
,
2664 .bdrv_reopen_abort
= raw_reopen_abort
,
2665 .bdrv_close
= raw_close
,
2666 .bdrv_co_create
= raw_co_create
,
2667 .bdrv_co_create_opts
= raw_co_create_opts
,
2668 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2669 .bdrv_co_block_status
= raw_co_block_status
,
2670 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2671 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2673 .bdrv_co_preadv
= raw_co_preadv
,
2674 .bdrv_co_pwritev
= raw_co_pwritev
,
2675 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2676 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2677 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2678 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2679 .bdrv_refresh_limits
= raw_refresh_limits
,
2680 .bdrv_io_plug
= raw_aio_plug
,
2681 .bdrv_io_unplug
= raw_aio_unplug
,
2682 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2684 .bdrv_co_truncate
= raw_co_truncate
,
2685 .bdrv_getlength
= raw_getlength
,
2686 .bdrv_get_info
= raw_get_info
,
2687 .bdrv_get_allocated_file_size
2688 = raw_get_allocated_file_size
,
2689 .bdrv_check_perm
= raw_check_perm
,
2690 .bdrv_set_perm
= raw_set_perm
,
2691 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2692 .create_opts
= &raw_create_opts
,
2695 /***********************************************/
2698 #if defined(__APPLE__) && defined(__MACH__)
2699 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2700 CFIndex maxPathSize
, int flags
);
2701 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2703 kern_return_t kernResult
= KERN_FAILURE
;
2704 mach_port_t masterPort
;
2705 CFMutableDictionaryRef classesToMatch
;
2706 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2707 char *mediaType
= NULL
;
2709 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2710 if ( KERN_SUCCESS
!= kernResult
) {
2711 printf( "IOMasterPort returned %d\n", kernResult
);
2715 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2716 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2717 if (classesToMatch
== NULL
) {
2718 error_report("IOServiceMatching returned NULL for %s",
2719 matching_array
[index
]);
2722 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2724 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2726 if (kernResult
!= KERN_SUCCESS
) {
2727 error_report("Note: IOServiceGetMatchingServices returned %d",
2732 /* If a match was found, leave the loop */
2733 if (*mediaIterator
!= 0) {
2734 DPRINTF("Matching using %s\n", matching_array
[index
]);
2735 mediaType
= g_strdup(matching_array
[index
]);
2742 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2743 CFIndex maxPathSize
, int flags
)
2745 io_object_t nextMedia
;
2746 kern_return_t kernResult
= KERN_FAILURE
;
2748 nextMedia
= IOIteratorNext( mediaIterator
);
2751 CFTypeRef bsdPathAsCFString
;
2752 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2753 if ( bsdPathAsCFString
) {
2754 size_t devPathLength
;
2755 strcpy( bsdPath
, _PATH_DEV
);
2756 if (flags
& BDRV_O_NOCACHE
) {
2757 strcat(bsdPath
, "r");
2759 devPathLength
= strlen( bsdPath
);
2760 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2761 kernResult
= KERN_SUCCESS
;
2763 CFRelease( bsdPathAsCFString
);
2765 IOObjectRelease( nextMedia
);
2771 /* Sets up a real cdrom for use in QEMU */
2772 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2774 int index
, num_of_test_partitions
= 2, fd
;
2775 char test_partition
[MAXPATHLEN
];
2776 bool partition_found
= false;
2778 /* look for a working partition */
2779 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2780 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2782 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2784 partition_found
= true;
2790 /* if a working partition on the device was not found */
2791 if (partition_found
== false) {
2792 error_setg(errp
, "Failed to find a working partition on disc");
2794 DPRINTF("Using %s as optical disc\n", test_partition
);
2795 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2797 return partition_found
;
2800 /* Prints directions on mounting and unmounting a device */
2801 static void print_unmounting_directions(const char *file_name
)
2803 error_report("If device %s is mounted on the desktop, unmount"
2804 " it first before using it in QEMU", file_name
);
2805 error_report("Command to unmount device: diskutil unmountDisk %s",
2807 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2810 #endif /* defined(__APPLE__) && defined(__MACH__) */
2812 static int hdev_probe_device(const char *filename
)
2816 /* allow a dedicated CD-ROM driver to match with a higher priority */
2817 if (strstart(filename
, "/dev/cdrom", NULL
))
2820 if (stat(filename
, &st
) >= 0 &&
2821 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2828 static int check_hdev_writable(BDRVRawState
*s
)
2830 #if defined(BLKROGET)
2831 /* Linux block devices can be configured "read-only" using blockdev(8).
2832 * This is independent of device node permissions and therefore open(2)
2833 * with O_RDWR succeeds. Actual writes fail with EPERM.
2835 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2836 * check for read-only block devices so that Linux block devices behave
2842 if (fstat(s
->fd
, &st
)) {
2846 if (!S_ISBLK(st
.st_mode
)) {
2850 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2857 #endif /* defined(BLKROGET) */
2861 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2864 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2867 static bool hdev_is_sg(BlockDriverState
*bs
)
2870 #if defined(__linux__)
2872 BDRVRawState
*s
= bs
->opaque
;
2874 struct sg_scsi_id scsiid
;
2878 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2882 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2887 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2889 DPRINTF("SG device found: type=%d, version=%d\n",
2890 scsiid
.scsi_type
, sg_version
);
2899 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2902 BDRVRawState
*s
= bs
->opaque
;
2903 Error
*local_err
= NULL
;
2906 #if defined(__APPLE__) && defined(__MACH__)
2908 * Caution: while qdict_get_str() is fine, getting non-string types
2909 * would require more care. When @options come from -blockdev or
2910 * blockdev_add, its members are typed according to the QAPI
2911 * schema, but when they come from -drive, they're all QString.
2913 const char *filename
= qdict_get_str(options
, "filename");
2914 char bsd_path
[MAXPATHLEN
] = "";
2915 bool error_occurred
= false;
2917 /* If using a real cdrom */
2918 if (strcmp(filename
, "/dev/cdrom") == 0) {
2919 char *mediaType
= NULL
;
2920 kern_return_t ret_val
;
2921 io_iterator_t mediaIterator
= 0;
2923 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
2924 if (mediaType
== NULL
) {
2925 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
2927 error_occurred
= true;
2928 goto hdev_open_Mac_error
;
2931 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
2932 if (ret_val
!= KERN_SUCCESS
) {
2933 error_setg(errp
, "Could not get BSD path for optical drive");
2934 error_occurred
= true;
2935 goto hdev_open_Mac_error
;
2938 /* If a real optical drive was not found */
2939 if (bsd_path
[0] == '\0') {
2940 error_setg(errp
, "Failed to obtain bsd path for optical drive");
2941 error_occurred
= true;
2942 goto hdev_open_Mac_error
;
2945 /* If using a cdrom disc and finding a partition on the disc failed */
2946 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
2947 setup_cdrom(bsd_path
, errp
) == false) {
2948 print_unmounting_directions(bsd_path
);
2949 error_occurred
= true;
2950 goto hdev_open_Mac_error
;
2953 qdict_put_str(options
, "filename", bsd_path
);
2955 hdev_open_Mac_error
:
2957 if (mediaIterator
) {
2958 IOObjectRelease(mediaIterator
);
2960 if (error_occurred
) {
2964 #endif /* defined(__APPLE__) && defined(__MACH__) */
2966 s
->type
= FTYPE_FILE
;
2968 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
2970 error_propagate(errp
, local_err
);
2971 #if defined(__APPLE__) && defined(__MACH__)
2973 filename
= bsd_path
;
2975 /* if a physical device experienced an error while being opened */
2976 if (strncmp(filename
, "/dev/", 5) == 0) {
2977 print_unmounting_directions(filename
);
2979 #endif /* defined(__APPLE__) && defined(__MACH__) */
2983 /* Since this does ioctl the device must be already opened */
2984 bs
->sg
= hdev_is_sg(bs
);
2986 if (flags
& BDRV_O_RDWR
) {
2987 ret
= check_hdev_writable(s
);
2990 error_setg_errno(errp
, -ret
, "The device is not writable");
2998 #if defined(__linux__)
3000 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
3001 unsigned long int req
, void *buf
,
3002 BlockCompletionFunc
*cb
, void *opaque
)
3004 BDRVRawState
*s
= bs
->opaque
;
3005 RawPosixAIOData
*acb
;
3008 if (fd_open(bs
) < 0)
3011 if (req
== SG_IO
&& s
->pr_mgr
) {
3012 struct sg_io_hdr
*io_hdr
= buf
;
3013 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3014 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3015 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3016 s
->fd
, io_hdr
, cb
, opaque
);
3020 acb
= g_new(RawPosixAIOData
, 1);
3022 acb
->aio_type
= QEMU_AIO_IOCTL
;
3023 acb
->aio_fildes
= s
->fd
;
3024 acb
->aio_offset
= 0;
3025 acb
->aio_ioctl_buf
= buf
;
3026 acb
->aio_ioctl_cmd
= req
;
3027 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3028 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
3032 static int fd_open(BlockDriverState
*bs
)
3034 BDRVRawState
*s
= bs
->opaque
;
3036 /* this is just to ensure s->fd is sane (its called by io ops) */
3042 static coroutine_fn
int
3043 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3045 BDRVRawState
*s
= bs
->opaque
;
3052 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3053 QEMU_AIO_DISCARD
| QEMU_AIO_BLKDEV
);
3056 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3057 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3059 BDRVRawState
*s
= bs
->opaque
;
3066 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
3067 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3068 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
3069 } else if (s
->discard_zeroes
) {
3070 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3071 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
3076 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3081 struct stat stat_buf
;
3082 int64_t total_size
= 0;
3085 /* This function is used by both protocol block drivers and therefore either
3086 * of these prefixes may be given.
3087 * The return value has to be stored somewhere, otherwise this is an error
3088 * due to -Werror=unused-value. */
3090 strstart(filename
, "host_device:", &filename
) ||
3091 strstart(filename
, "host_cdrom:" , &filename
);
3095 ret
= raw_normalize_devicepath(&filename
);
3097 error_setg_errno(errp
, -ret
, "Could not normalize device path");
3101 /* Read out options */
3102 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3105 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3108 error_setg_errno(errp
, -ret
, "Could not open device");
3112 if (fstat(fd
, &stat_buf
) < 0) {
3114 error_setg_errno(errp
, -ret
, "Could not stat device");
3115 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3117 "The given file is neither a block nor a character device");
3119 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3120 error_setg(errp
, "Device is too small");
3124 if (!ret
&& total_size
) {
3125 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3126 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3127 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3130 ret
= qemu_write_full(fd
, buf
, zero_size
);
3131 ret
= ret
== zero_size
? 0 : -errno
;
3138 static BlockDriver bdrv_host_device
= {
3139 .format_name
= "host_device",
3140 .protocol_name
= "host_device",
3141 .instance_size
= sizeof(BDRVRawState
),
3142 .bdrv_needs_filename
= true,
3143 .bdrv_probe_device
= hdev_probe_device
,
3144 .bdrv_parse_filename
= hdev_parse_filename
,
3145 .bdrv_file_open
= hdev_open
,
3146 .bdrv_close
= raw_close
,
3147 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3148 .bdrv_reopen_commit
= raw_reopen_commit
,
3149 .bdrv_reopen_abort
= raw_reopen_abort
,
3150 .bdrv_co_create_opts
= hdev_co_create_opts
,
3151 .create_opts
= &raw_create_opts
,
3152 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3153 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3155 .bdrv_co_preadv
= raw_co_preadv
,
3156 .bdrv_co_pwritev
= raw_co_pwritev
,
3157 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3158 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3159 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3160 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3161 .bdrv_refresh_limits
= raw_refresh_limits
,
3162 .bdrv_io_plug
= raw_aio_plug
,
3163 .bdrv_io_unplug
= raw_aio_unplug
,
3164 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3166 .bdrv_co_truncate
= raw_co_truncate
,
3167 .bdrv_getlength
= raw_getlength
,
3168 .bdrv_get_info
= raw_get_info
,
3169 .bdrv_get_allocated_file_size
3170 = raw_get_allocated_file_size
,
3171 .bdrv_check_perm
= raw_check_perm
,
3172 .bdrv_set_perm
= raw_set_perm
,
3173 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3174 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3175 .bdrv_probe_geometry
= hdev_probe_geometry
,
3177 /* generic scsi device */
3179 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3183 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3184 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3187 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3192 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3195 BDRVRawState
*s
= bs
->opaque
;
3199 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3200 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3203 static int cdrom_probe_device(const char *filename
)
3209 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3213 ret
= fstat(fd
, &st
);
3214 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3218 /* Attempt to detect via a CDROM specific ioctl */
3219 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3229 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3231 BDRVRawState
*s
= bs
->opaque
;
3234 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3235 return ret
== CDS_DISC_OK
;
3238 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3240 BDRVRawState
*s
= bs
->opaque
;
3243 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3244 perror("CDROMEJECT");
3246 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3247 perror("CDROMEJECT");
3251 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3253 BDRVRawState
*s
= bs
->opaque
;
3255 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3257 * Note: an error can happen if the distribution automatically
3260 /* perror("CDROM_LOCKDOOR"); */
3264 static BlockDriver bdrv_host_cdrom
= {
3265 .format_name
= "host_cdrom",
3266 .protocol_name
= "host_cdrom",
3267 .instance_size
= sizeof(BDRVRawState
),
3268 .bdrv_needs_filename
= true,
3269 .bdrv_probe_device
= cdrom_probe_device
,
3270 .bdrv_parse_filename
= cdrom_parse_filename
,
3271 .bdrv_file_open
= cdrom_open
,
3272 .bdrv_close
= raw_close
,
3273 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3274 .bdrv_reopen_commit
= raw_reopen_commit
,
3275 .bdrv_reopen_abort
= raw_reopen_abort
,
3276 .bdrv_co_create_opts
= hdev_co_create_opts
,
3277 .create_opts
= &raw_create_opts
,
3278 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3281 .bdrv_co_preadv
= raw_co_preadv
,
3282 .bdrv_co_pwritev
= raw_co_pwritev
,
3283 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3284 .bdrv_refresh_limits
= raw_refresh_limits
,
3285 .bdrv_io_plug
= raw_aio_plug
,
3286 .bdrv_io_unplug
= raw_aio_unplug
,
3287 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3289 .bdrv_co_truncate
= raw_co_truncate
,
3290 .bdrv_getlength
= raw_getlength
,
3291 .has_variable_length
= true,
3292 .bdrv_get_allocated_file_size
3293 = raw_get_allocated_file_size
,
3295 /* removable device support */
3296 .bdrv_is_inserted
= cdrom_is_inserted
,
3297 .bdrv_eject
= cdrom_eject
,
3298 .bdrv_lock_medium
= cdrom_lock_medium
,
3300 /* generic scsi device */
3301 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3303 #endif /* __linux__ */
3305 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3306 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3309 BDRVRawState
*s
= bs
->opaque
;
3310 Error
*local_err
= NULL
;
3315 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3317 error_propagate(errp
, local_err
);
3321 /* make sure the door isn't locked at this time */
3322 ioctl(s
->fd
, CDIOCALLOW
);
3326 static int cdrom_probe_device(const char *filename
)
3328 if (strstart(filename
, "/dev/cd", NULL
) ||
3329 strstart(filename
, "/dev/acd", NULL
))
3334 static int cdrom_reopen(BlockDriverState
*bs
)
3336 BDRVRawState
*s
= bs
->opaque
;
3340 * Force reread of possibly changed/newly loaded disc,
3341 * FreeBSD seems to not notice sometimes...
3345 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3352 /* make sure the door isn't locked at this time */
3353 ioctl(s
->fd
, CDIOCALLOW
);
3357 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3359 return raw_getlength(bs
) > 0;
3362 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3364 BDRVRawState
*s
= bs
->opaque
;
3369 (void) ioctl(s
->fd
, CDIOCALLOW
);
3372 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3373 perror("CDIOCEJECT");
3375 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3376 perror("CDIOCCLOSE");
3382 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3384 BDRVRawState
*s
= bs
->opaque
;
3388 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3390 * Note: an error can happen if the distribution automatically
3393 /* perror("CDROM_LOCKDOOR"); */
3397 static BlockDriver bdrv_host_cdrom
= {
3398 .format_name
= "host_cdrom",
3399 .protocol_name
= "host_cdrom",
3400 .instance_size
= sizeof(BDRVRawState
),
3401 .bdrv_needs_filename
= true,
3402 .bdrv_probe_device
= cdrom_probe_device
,
3403 .bdrv_parse_filename
= cdrom_parse_filename
,
3404 .bdrv_file_open
= cdrom_open
,
3405 .bdrv_close
= raw_close
,
3406 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3407 .bdrv_reopen_commit
= raw_reopen_commit
,
3408 .bdrv_reopen_abort
= raw_reopen_abort
,
3409 .bdrv_co_create_opts
= hdev_co_create_opts
,
3410 .create_opts
= &raw_create_opts
,
3412 .bdrv_co_preadv
= raw_co_preadv
,
3413 .bdrv_co_pwritev
= raw_co_pwritev
,
3414 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3415 .bdrv_refresh_limits
= raw_refresh_limits
,
3416 .bdrv_io_plug
= raw_aio_plug
,
3417 .bdrv_io_unplug
= raw_aio_unplug
,
3418 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3420 .bdrv_co_truncate
= raw_co_truncate
,
3421 .bdrv_getlength
= raw_getlength
,
3422 .has_variable_length
= true,
3423 .bdrv_get_allocated_file_size
3424 = raw_get_allocated_file_size
,
3426 /* removable device support */
3427 .bdrv_is_inserted
= cdrom_is_inserted
,
3428 .bdrv_eject
= cdrom_eject
,
3429 .bdrv_lock_medium
= cdrom_lock_medium
,
3431 #endif /* __FreeBSD__ */
3433 static void bdrv_file_init(void)
3436 * Register all the drivers. Note that order is important, the driver
3437 * registered last will get probed first.
3439 bdrv_register(&bdrv_file
);
3440 bdrv_register(&bdrv_host_device
);
3442 bdrv_register(&bdrv_host_cdrom
);
3444 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3445 bdrv_register(&bdrv_host_cdrom
);
3449 block_init(bdrv_file_init
);