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
, Error
**errp
)
443 BDRVRawState
*s
= bs
->opaque
;
445 Error
*local_err
= NULL
;
446 const char *filename
= NULL
;
448 BlockdevAioOptions aio
, aio_default
;
453 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
454 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
456 error_propagate(errp
, local_err
);
461 filename
= qemu_opt_get(opts
, "filename");
463 ret
= raw_normalize_devicepath(&filename
);
465 error_setg_errno(errp
, -ret
, "Could not normalize device path");
469 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
470 ? BLOCKDEV_AIO_OPTIONS_NATIVE
471 : BLOCKDEV_AIO_OPTIONS_THREADS
;
472 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
473 qemu_opt_get(opts
, "aio"),
474 aio_default
, &local_err
);
476 error_propagate(errp
, local_err
);
480 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
482 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
483 qemu_opt_get(opts
, "locking"),
484 ON_OFF_AUTO_AUTO
, &local_err
);
486 error_propagate(errp
, local_err
);
493 if (!qemu_has_ofd_lock()) {
495 "File lock requested but OFD locking syscall is "
496 "unavailable, falling back to POSIX file locks.\n"
497 "Due to the implementation, locks can be lost "
501 case ON_OFF_AUTO_OFF
:
504 case ON_OFF_AUTO_AUTO
:
505 s
->use_lock
= qemu_has_ofd_lock();
511 str
= qemu_opt_get(opts
, "pr-manager");
513 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
515 error_propagate(errp
, local_err
);
521 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
524 s
->open_flags
= open_flags
;
525 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
528 fd
= qemu_open(filename
, s
->open_flags
, 0644);
531 error_setg_errno(errp
, errno
, "Could not open '%s'", filename
);
541 fd
= qemu_open(filename
, s
->open_flags
);
544 error_setg_errno(errp
, errno
, "Could not open '%s' for locking",
552 s
->shared_perm
= BLK_PERM_ALL
;
554 #ifdef CONFIG_LINUX_AIO
555 /* Currently Linux does AIO only for files opened with O_DIRECT */
556 if (s
->use_linux_aio
) {
557 if (!(s
->open_flags
& O_DIRECT
)) {
558 error_setg(errp
, "aio=native was specified, but it requires "
559 "cache.direct=on, which was not specified.");
563 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
564 error_prepend(errp
, "Unable to use native AIO: ");
569 if (s
->use_linux_aio
) {
570 error_setg(errp
, "aio=native was specified, but is not supported "
575 #endif /* !defined(CONFIG_LINUX_AIO) */
577 s
->has_discard
= true;
578 s
->has_write_zeroes
= true;
579 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
580 s
->needs_alignment
= true;
583 if (fstat(s
->fd
, &st
) < 0) {
585 error_setg_errno(errp
, errno
, "Could not stat file");
588 if (S_ISREG(st
.st_mode
)) {
589 s
->discard_zeroes
= true;
590 s
->has_fallocate
= true;
592 if (S_ISBLK(st
.st_mode
)) {
593 #ifdef BLKDISCARDZEROES
595 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
596 s
->discard_zeroes
= true;
600 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
601 * not rely on the contents of discarded blocks unless using O_DIRECT.
602 * Same for BLKZEROOUT.
604 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
605 s
->discard_zeroes
= false;
606 s
->has_write_zeroes
= false;
611 if (S_ISCHR(st
.st_mode
)) {
613 * The file is a char device (disk), which on FreeBSD isn't behind
614 * a pager, so force all requests to be aligned. This is needed
615 * so QEMU makes sure all IO operations on the device are aligned
616 * to sector size, or else FreeBSD will reject them with EINVAL.
618 s
->needs_alignment
= true;
623 if (platform_test_xfs_fd(s
->fd
)) {
628 bs
->supported_zero_flags
= s
->discard_zeroes
? BDRV_REQ_MAY_UNMAP
: 0;
631 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
638 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
641 BDRVRawState
*s
= bs
->opaque
;
643 s
->type
= FTYPE_FILE
;
644 return raw_open_common(bs
, options
, flags
, 0, errp
);
653 #define PERM_FOREACH(i) \
654 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
656 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
657 * file; if @unlock == true, also unlock the unneeded bytes.
658 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
660 static int raw_apply_lock_bytes(int fd
,
661 uint64_t perm_lock_bits
,
662 uint64_t shared_perm_lock_bits
,
663 bool unlock
, Error
**errp
)
669 int off
= RAW_LOCK_PERM_BASE
+ i
;
670 if (perm_lock_bits
& (1ULL << i
)) {
671 ret
= qemu_lock_fd(fd
, off
, 1, false);
673 error_setg(errp
, "Failed to lock byte %d", off
);
677 ret
= qemu_unlock_fd(fd
, off
, 1);
679 error_setg(errp
, "Failed to unlock byte %d", off
);
685 int off
= RAW_LOCK_SHARED_BASE
+ i
;
686 if (shared_perm_lock_bits
& (1ULL << i
)) {
687 ret
= qemu_lock_fd(fd
, off
, 1, false);
689 error_setg(errp
, "Failed to lock byte %d", off
);
693 ret
= qemu_unlock_fd(fd
, off
, 1);
695 error_setg(errp
, "Failed to unlock byte %d", off
);
703 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
704 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
711 int off
= RAW_LOCK_SHARED_BASE
+ i
;
712 uint64_t p
= 1ULL << i
;
714 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
716 char *perm_name
= bdrv_perm_names(p
);
718 "Failed to get \"%s\" lock",
721 error_append_hint(errp
,
722 "Is another process using the image?\n");
728 int off
= RAW_LOCK_PERM_BASE
+ i
;
729 uint64_t p
= 1ULL << i
;
730 if (!(shared_perm
& p
)) {
731 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
733 char *perm_name
= bdrv_perm_names(p
);
735 "Failed to get shared \"%s\" lock",
738 error_append_hint(errp
,
739 "Is another process using the image?\n");
747 static int raw_handle_perm_lock(BlockDriverState
*bs
,
749 uint64_t new_perm
, uint64_t new_shared
,
752 BDRVRawState
*s
= bs
->opaque
;
754 Error
*local_err
= NULL
;
760 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
764 assert(s
->lock_fd
> 0);
768 ret
= raw_apply_lock_bytes(s
->lock_fd
, s
->perm
| new_perm
,
769 ~s
->shared_perm
| ~new_shared
,
772 ret
= raw_check_lock_bytes(s
->lock_fd
, new_perm
, new_shared
, errp
);
778 /* fall through to unlock bytes. */
780 raw_apply_lock_bytes(s
->lock_fd
, s
->perm
, ~s
->shared_perm
,
783 /* Theoretically the above call only unlocks bytes and it cannot
784 * fail. Something weird happened, report it.
786 error_report_err(local_err
);
790 raw_apply_lock_bytes(s
->lock_fd
, new_perm
, ~new_shared
,
793 /* Theoretically the above call only unlocks bytes and it cannot
794 * fail. Something weird happened, report it.
796 error_report_err(local_err
);
803 static int raw_reopen_prepare(BDRVReopenState
*state
,
804 BlockReopenQueue
*queue
, Error
**errp
)
807 BDRVRawReopenState
*rs
;
810 Error
*local_err
= NULL
;
812 assert(state
!= NULL
);
813 assert(state
->bs
!= NULL
);
815 s
= state
->bs
->opaque
;
817 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
821 /* Handle options changes */
822 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
823 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
825 error_propagate(errp
, local_err
);
830 rs
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
831 s
->check_cache_dropped
);
833 if (s
->type
== FTYPE_CD
) {
834 rs
->open_flags
|= O_NONBLOCK
;
837 raw_parse_flags(state
->flags
, &rs
->open_flags
);
839 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
841 fcntl_flags
|= O_NOATIME
;
845 /* Not all operating systems have O_ASYNC, and those that don't
846 * will not let us track the state into rs->open_flags (typically
847 * you achieve the same effect with an ioctl, for example I_SETSIG
848 * on Solaris). But we do not use O_ASYNC, so that's fine.
850 assert((s
->open_flags
& O_ASYNC
) == 0);
853 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
854 /* dup the original fd */
855 rs
->fd
= qemu_dup(s
->fd
);
857 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
865 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
867 const char *normalized_filename
= state
->bs
->filename
;
868 ret
= raw_normalize_devicepath(&normalized_filename
);
870 error_setg_errno(errp
, -ret
, "Could not normalize device path");
872 assert(!(rs
->open_flags
& O_CREAT
));
873 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
875 error_setg_errno(errp
, errno
, "Could not reopen file");
881 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
882 * alignment with the new fd. */
884 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
888 error_propagate(errp
, local_err
);
898 static void raw_reopen_commit(BDRVReopenState
*state
)
900 BDRVRawReopenState
*rs
= state
->opaque
;
901 BDRVRawState
*s
= state
->bs
->opaque
;
903 s
->check_cache_dropped
= rs
->check_cache_dropped
;
904 s
->open_flags
= rs
->open_flags
;
909 g_free(state
->opaque
);
910 state
->opaque
= NULL
;
914 static void raw_reopen_abort(BDRVReopenState
*state
)
916 BDRVRawReopenState
*rs
= state
->opaque
;
918 /* nothing to do if NULL, we didn't get far enough */
927 g_free(state
->opaque
);
928 state
->opaque
= NULL
;
931 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
935 short max_sectors
= 0;
936 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
938 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
939 return max_sectors
<< BDRV_SECTOR_BITS
;
948 static int hdev_get_max_segments(const struct stat
*st
)
958 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
959 major(st
->st_rdev
), minor(st
->st_rdev
));
960 fd
= open(sysfspath
, O_RDONLY
);
966 ret
= read(fd
, buf
, sizeof(buf
) - 1);
967 } while (ret
== -1 && errno
== EINTR
);
971 } else if (ret
== 0) {
976 /* The file is ended with '\n', pass 'end' to accept that. */
977 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
978 if (ret
== 0 && end
&& *end
== '\n') {
993 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
995 BDRVRawState
*s
= bs
->opaque
;
998 if (!fstat(s
->fd
, &st
)) {
999 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1000 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1001 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1002 bs
->bl
.max_transfer
= pow2floor(ret
);
1004 ret
= hdev_get_max_segments(&st
);
1006 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1007 ret
* getpagesize());
1012 raw_probe_alignment(bs
, s
->fd
, errp
);
1013 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1014 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1017 static int check_for_dasd(int fd
)
1020 struct dasd_information2_t info
= {0};
1022 return ioctl(fd
, BIODASDINFO2
, &info
);
1029 * Try to get @bs's logical and physical block size.
1030 * On success, store them in @bsz and return zero.
1031 * On failure, return negative errno.
1033 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1035 BDRVRawState
*s
= bs
->opaque
;
1038 /* If DASD, get blocksizes */
1039 if (check_for_dasd(s
->fd
) < 0) {
1042 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1046 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1050 * Try to get @bs's geometry: cyls, heads, sectors.
1051 * On success, store them in @geo and return 0.
1052 * On failure return -errno.
1053 * (Allows block driver to assign default geometry values that guest sees)
1056 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1058 BDRVRawState
*s
= bs
->opaque
;
1059 struct hd_geometry ioctl_geo
= {0};
1061 /* If DASD, get its geometry */
1062 if (check_for_dasd(s
->fd
) < 0) {
1065 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1068 /* HDIO_GETGEO may return success even though geo contains zeros
1069 (e.g. certain multipath setups) */
1070 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1073 /* Do not return a geometry for partition */
1074 if (ioctl_geo
.start
!= 0) {
1077 geo
->heads
= ioctl_geo
.heads
;
1078 geo
->sectors
= ioctl_geo
.sectors
;
1079 geo
->cylinders
= ioctl_geo
.cylinders
;
1083 #else /* __linux__ */
1084 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1090 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1094 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->aio_ioctl_cmd
, aiocb
->aio_ioctl_buf
);
1102 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1104 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1107 if (s
->page_cache_inconsistent
) {
1111 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1113 /* There is no clear definition of the semantics of a failing fsync(),
1114 * so we may have to assume the worst. The sad truth is that this
1115 * assumption is correct for Linux. Some pages are now probably marked
1116 * clean in the page cache even though they are inconsistent with the
1117 * on-disk contents. The next fdatasync() call would succeed, but no
1118 * further writeback attempt will be made. We can't get back to a state
1119 * in which we know what is on disk (we would have to rewrite
1120 * everything that was touched since the last fdatasync() at least), so
1121 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1122 * really defined, I have little hope that other OSes are doing better.
1124 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1126 if ((s
->open_flags
& O_DIRECT
) == 0) {
1127 s
->page_cache_inconsistent
= true;
1134 #ifdef CONFIG_PREADV
1136 static bool preadv_present
= true;
1139 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1141 return preadv(fd
, iov
, nr_iov
, offset
);
1145 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1147 return pwritev(fd
, iov
, nr_iov
, offset
);
1152 static bool preadv_present
= false;
1155 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1161 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1168 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1173 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1174 len
= qemu_pwritev(aiocb
->aio_fildes
,
1179 len
= qemu_preadv(aiocb
->aio_fildes
,
1183 } while (len
== -1 && errno
== EINTR
);
1192 * Read/writes the data to/from a given linear buffer.
1194 * Returns the number of bytes handles or -errno in case of an error. Short
1195 * reads are only returned if the end of the file is reached.
1197 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1202 while (offset
< aiocb
->aio_nbytes
) {
1203 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1204 len
= pwrite(aiocb
->aio_fildes
,
1205 (const char *)buf
+ offset
,
1206 aiocb
->aio_nbytes
- offset
,
1207 aiocb
->aio_offset
+ offset
);
1209 len
= pread(aiocb
->aio_fildes
,
1211 aiocb
->aio_nbytes
- offset
,
1212 aiocb
->aio_offset
+ offset
);
1214 if (len
== -1 && errno
== EINTR
) {
1216 } else if (len
== -1 && errno
== EINVAL
&&
1217 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1218 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1220 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1221 * after a short read. Assume that O_DIRECT short reads only occur
1222 * at EOF. Therefore this is a short read, not an I/O error.
1225 } else if (len
== -1) {
1228 } else if (len
== 0) {
1237 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1242 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1244 * If there is just a single buffer, and it is properly aligned
1245 * we can just use plain pread/pwrite without any problems.
1247 if (aiocb
->aio_niov
== 1) {
1248 return handle_aiocb_rw_linear(aiocb
, aiocb
->aio_iov
->iov_base
);
1251 * We have more than one iovec, and all are properly aligned.
1253 * Try preadv/pwritev first and fall back to linearizing the
1254 * buffer if it's not supported.
1256 if (preadv_present
) {
1257 nbytes
= handle_aiocb_rw_vector(aiocb
);
1258 if (nbytes
== aiocb
->aio_nbytes
||
1259 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1262 preadv_present
= false;
1266 * XXX(hch): short read/write. no easy way to handle the reminder
1267 * using these interfaces. For now retry using plain
1273 * Ok, we have to do it the hard way, copy all segments into
1274 * a single aligned buffer.
1276 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1281 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1285 for (i
= 0; i
< aiocb
->aio_niov
; ++i
) {
1286 memcpy(p
, aiocb
->aio_iov
[i
].iov_base
, aiocb
->aio_iov
[i
].iov_len
);
1287 p
+= aiocb
->aio_iov
[i
].iov_len
;
1289 assert(p
- buf
== aiocb
->aio_nbytes
);
1292 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1293 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1295 size_t count
= aiocb
->aio_nbytes
, copy
;
1298 for (i
= 0; i
< aiocb
->aio_niov
&& count
; ++i
) {
1300 if (copy
> aiocb
->aio_iov
[i
].iov_len
) {
1301 copy
= aiocb
->aio_iov
[i
].iov_len
;
1303 memcpy(aiocb
->aio_iov
[i
].iov_base
, p
, copy
);
1304 assert(count
>= copy
);
1316 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1318 struct xfs_flock64 fl
;
1321 memset(&fl
, 0, sizeof(fl
));
1322 fl
.l_whence
= SEEK_SET
;
1323 fl
.l_start
= offset
;
1326 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1328 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1335 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1337 struct xfs_flock64 fl
;
1340 memset(&fl
, 0, sizeof(fl
));
1341 fl
.l_whence
= SEEK_SET
;
1342 fl
.l_start
= offset
;
1345 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1347 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1355 static int translate_err(int err
)
1357 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1364 #ifdef CONFIG_FALLOCATE
1365 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1368 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1371 } while (errno
== EINTR
);
1372 return translate_err(-errno
);
1376 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1379 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1381 if (!s
->has_write_zeroes
) {
1387 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1388 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1391 } while (errno
== EINTR
);
1393 ret
= translate_err(-errno
);
1396 if (ret
== -ENOTSUP
) {
1397 s
->has_write_zeroes
= false;
1402 static ssize_t
handle_aiocb_write_zeroes(RawPosixAIOData
*aiocb
)
1404 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1405 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1407 #ifdef CONFIG_FALLOCATE
1411 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1412 return handle_aiocb_write_zeroes_block(aiocb
);
1417 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1421 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1422 if (s
->has_write_zeroes
) {
1423 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1424 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1425 if (ret
== 0 || ret
!= -ENOTSUP
) {
1428 s
->has_write_zeroes
= false;
1432 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1433 if (s
->has_discard
&& s
->has_fallocate
) {
1434 int ret
= do_fallocate(s
->fd
,
1435 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1436 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1438 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1439 if (ret
== 0 || ret
!= -ENOTSUP
) {
1442 s
->has_fallocate
= false;
1443 } else if (ret
!= -ENOTSUP
) {
1446 s
->has_discard
= false;
1451 #ifdef CONFIG_FALLOCATE
1452 /* Last resort: we are trying to extend the file with zeroed data. This
1453 * can be done via fallocate(fd, 0) */
1454 len
= bdrv_getlength(aiocb
->bs
);
1455 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1456 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1457 if (ret
== 0 || ret
!= -ENOTSUP
) {
1460 s
->has_fallocate
= false;
1467 #ifndef HAVE_COPY_FILE_RANGE
1468 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1469 off_t
*out_off
, size_t len
, unsigned int flags
)
1471 #ifdef __NR_copy_file_range
1472 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1473 out_off
, len
, flags
);
1481 static ssize_t
handle_aiocb_copy_range(RawPosixAIOData
*aiocb
)
1483 uint64_t bytes
= aiocb
->aio_nbytes
;
1484 off_t in_off
= aiocb
->aio_offset
;
1485 off_t out_off
= aiocb
->aio_offset2
;
1488 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1489 aiocb
->aio_fd2
, &out_off
,
1492 /* No progress (e.g. when beyond EOF), let the caller fall back to
1511 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1513 int ret
= -EOPNOTSUPP
;
1514 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1516 if (!s
->has_discard
) {
1520 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1523 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1524 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1527 } while (errno
== EINTR
);
1534 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1538 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1539 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1540 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1544 ret
= translate_err(ret
);
1545 if (ret
== -ENOTSUP
) {
1546 s
->has_discard
= false;
1551 static int handle_aiocb_truncate(RawPosixAIOData
*aiocb
)
1554 int64_t current_length
= 0;
1557 int fd
= aiocb
->aio_fildes
;
1558 int64_t offset
= aiocb
->aio_offset
;
1559 Error
**errp
= aiocb
->errp
;
1561 if (fstat(fd
, &st
) < 0) {
1563 error_setg_errno(errp
, -result
, "Could not stat file");
1567 current_length
= st
.st_size
;
1568 if (current_length
> offset
&& aiocb
->prealloc
!= PREALLOC_MODE_OFF
) {
1569 error_setg(errp
, "Cannot use preallocation for shrinking files");
1573 switch (aiocb
->prealloc
) {
1574 #ifdef CONFIG_POSIX_FALLOCATE
1575 case PREALLOC_MODE_FALLOC
:
1577 * Truncating before posix_fallocate() makes it about twice slower on
1578 * file systems that do not support fallocate(), trying to check if a
1579 * block is allocated before allocating it, so don't do that here.
1581 if (offset
!= current_length
) {
1582 result
= -posix_fallocate(fd
, current_length
,
1583 offset
- current_length
);
1585 /* posix_fallocate() doesn't set errno. */
1586 error_setg_errno(errp
, -result
,
1587 "Could not preallocate new data");
1594 case PREALLOC_MODE_FULL
:
1596 int64_t num
= 0, left
= offset
- current_length
;
1600 * Knowing the final size from the beginning could allow the file
1601 * system driver to do less allocations and possibly avoid
1602 * fragmentation of the file.
1604 if (ftruncate(fd
, offset
) != 0) {
1606 error_setg_errno(errp
, -result
, "Could not resize file");
1610 buf
= g_malloc0(65536);
1612 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1613 if (seek_result
< 0) {
1615 error_setg_errno(errp
, -result
,
1616 "Failed to seek to the old end of file");
1621 num
= MIN(left
, 65536);
1622 result
= write(fd
, buf
, num
);
1625 error_setg_errno(errp
, -result
,
1626 "Could not write zeros for preallocation");
1635 error_setg_errno(errp
, -result
,
1636 "Could not flush file to disk");
1642 case PREALLOC_MODE_OFF
:
1643 if (ftruncate(fd
, offset
) != 0) {
1645 error_setg_errno(errp
, -result
, "Could not resize file");
1650 error_setg(errp
, "Unsupported preallocation mode: %s",
1651 PreallocMode_str(aiocb
->prealloc
));
1657 if (ftruncate(fd
, current_length
) < 0) {
1658 error_report("Failed to restore old file length: %s",
1667 static int aio_worker(void *arg
)
1669 RawPosixAIOData
*aiocb
= arg
;
1672 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1674 ret
= handle_aiocb_rw(aiocb
);
1675 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1676 iov_memset(aiocb
->aio_iov
, aiocb
->aio_niov
, ret
,
1677 0, aiocb
->aio_nbytes
- ret
);
1679 ret
= aiocb
->aio_nbytes
;
1681 if (ret
== aiocb
->aio_nbytes
) {
1683 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1687 case QEMU_AIO_WRITE
:
1688 ret
= handle_aiocb_rw(aiocb
);
1689 if (ret
== aiocb
->aio_nbytes
) {
1691 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1695 case QEMU_AIO_FLUSH
:
1696 ret
= handle_aiocb_flush(aiocb
);
1698 case QEMU_AIO_IOCTL
:
1699 ret
= handle_aiocb_ioctl(aiocb
);
1701 case QEMU_AIO_DISCARD
:
1702 ret
= handle_aiocb_discard(aiocb
);
1704 case QEMU_AIO_WRITE_ZEROES
:
1705 ret
= handle_aiocb_write_zeroes(aiocb
);
1707 case QEMU_AIO_COPY_RANGE
:
1708 ret
= handle_aiocb_copy_range(aiocb
);
1710 case QEMU_AIO_TRUNCATE
:
1711 ret
= handle_aiocb_truncate(aiocb
);
1714 fprintf(stderr
, "invalid aio request (0x%x)\n", aiocb
->aio_type
);
1723 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1724 int64_t offset
, int fd2
, int64_t offset2
,
1726 int bytes
, int type
)
1728 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1732 acb
->aio_type
= type
;
1733 acb
->aio_fildes
= fd
;
1735 acb
->aio_offset2
= offset2
;
1737 acb
->aio_nbytes
= bytes
;
1738 acb
->aio_offset
= offset
;
1741 acb
->aio_iov
= qiov
->iov
;
1742 acb
->aio_niov
= qiov
->niov
;
1743 assert(qiov
->size
== bytes
);
1746 trace_paio_submit_co(offset
, bytes
, type
);
1747 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1748 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1751 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1752 int64_t offset
, QEMUIOVector
*qiov
,
1753 int bytes
, int type
)
1755 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1758 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1759 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1761 BDRVRawState
*s
= bs
->opaque
;
1763 if (fd_open(bs
) < 0)
1767 * Check if the underlying device requires requests to be aligned,
1768 * and if the request we are trying to submit is aligned or not.
1769 * If this is the case tell the low-level driver that it needs
1770 * to copy the buffer.
1772 if (s
->needs_alignment
) {
1773 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1774 type
|= QEMU_AIO_MISALIGNED
;
1775 #ifdef CONFIG_LINUX_AIO
1776 } else if (s
->use_linux_aio
) {
1777 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1778 assert(qiov
->size
== bytes
);
1779 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1784 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1787 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1788 uint64_t bytes
, QEMUIOVector
*qiov
,
1791 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1794 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1795 uint64_t bytes
, QEMUIOVector
*qiov
,
1799 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1802 static void raw_aio_plug(BlockDriverState
*bs
)
1804 #ifdef CONFIG_LINUX_AIO
1805 BDRVRawState
*s
= bs
->opaque
;
1806 if (s
->use_linux_aio
) {
1807 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1808 laio_io_plug(bs
, aio
);
1813 static void raw_aio_unplug(BlockDriverState
*bs
)
1815 #ifdef CONFIG_LINUX_AIO
1816 BDRVRawState
*s
= bs
->opaque
;
1817 if (s
->use_linux_aio
) {
1818 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1819 laio_io_unplug(bs
, aio
);
1824 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1826 BDRVRawState
*s
= bs
->opaque
;
1834 return paio_submit_co(bs
, s
->fd
, 0, NULL
, 0, QEMU_AIO_FLUSH
);
1837 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1838 AioContext
*new_context
)
1840 #ifdef CONFIG_LINUX_AIO
1841 BDRVRawState
*s
= bs
->opaque
;
1842 if (s
->use_linux_aio
) {
1844 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1845 error_reportf_err(local_err
, "Unable to use native AIO, "
1846 "falling back to thread pool: ");
1847 s
->use_linux_aio
= false;
1853 static void raw_close(BlockDriverState
*bs
)
1855 BDRVRawState
*s
= bs
->opaque
;
1861 if (s
->lock_fd
>= 0) {
1862 qemu_close(s
->lock_fd
);
1868 * Truncates the given regular file @fd to @offset and, when growing, fills the
1869 * new space according to @prealloc.
1871 * Returns: 0 on success, -errno on failure.
1873 static int coroutine_fn
1874 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1875 PreallocMode prealloc
, Error
**errp
)
1877 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1880 *acb
= (RawPosixAIOData
) {
1883 .aio_type
= QEMU_AIO_TRUNCATE
,
1884 .aio_offset
= offset
,
1885 .prealloc
= prealloc
,
1889 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1890 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1891 return thread_pool_submit_co(pool
, aio_worker
, acb
);
1894 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1895 PreallocMode prealloc
, Error
**errp
)
1897 BDRVRawState
*s
= bs
->opaque
;
1901 if (fstat(s
->fd
, &st
)) {
1903 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1907 if (S_ISREG(st
.st_mode
)) {
1908 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
1911 if (prealloc
!= PREALLOC_MODE_OFF
) {
1912 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1913 "non-regular file", PreallocMode_str(prealloc
));
1917 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1918 if (offset
> raw_getlength(bs
)) {
1919 error_setg(errp
, "Cannot grow device files");
1923 error_setg(errp
, "Resizing this file is not supported");
1931 static int64_t raw_getlength(BlockDriverState
*bs
)
1933 BDRVRawState
*s
= bs
->opaque
;
1939 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1940 struct disklabel dl
;
1942 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1944 return (uint64_t)dl
.d_secsize
*
1945 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1949 #elif defined(__NetBSD__)
1950 static int64_t raw_getlength(BlockDriverState
*bs
)
1952 BDRVRawState
*s
= bs
->opaque
;
1958 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1959 struct dkwedge_info dkw
;
1961 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
1962 return dkw
.dkw_size
* 512;
1964 struct disklabel dl
;
1966 if (ioctl(fd
, DIOCGDINFO
, &dl
))
1968 return (uint64_t)dl
.d_secsize
*
1969 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
1974 #elif defined(__sun__)
1975 static int64_t raw_getlength(BlockDriverState
*bs
)
1977 BDRVRawState
*s
= bs
->opaque
;
1978 struct dk_minfo minfo
;
1988 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1990 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
1992 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
1996 * There are reports that lseek on some devices fails, but
1997 * irc discussion said that contingency on contingency was overkill.
1999 size
= lseek(s
->fd
, 0, SEEK_END
);
2005 #elif defined(CONFIG_BSD)
2006 static int64_t raw_getlength(BlockDriverState
*bs
)
2008 BDRVRawState
*s
= bs
->opaque
;
2012 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2021 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2024 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2025 #ifdef DIOCGMEDIASIZE
2026 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2027 #elif defined(DIOCGPART)
2030 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2031 size
= pi
.media_size
;
2037 #if defined(__APPLE__) && defined(__MACH__)
2039 uint64_t sectors
= 0;
2040 uint32_t sector_size
= 0;
2042 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2043 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2044 size
= sectors
* sector_size
;
2046 size
= lseek(fd
, 0LL, SEEK_END
);
2053 size
= lseek(fd
, 0LL, SEEK_END
);
2058 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2061 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2062 if (size
== 2048LL * (unsigned)-1)
2064 /* XXX no disc? maybe we need to reopen... */
2065 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2072 size
= lseek(fd
, 0, SEEK_END
);
2080 static int64_t raw_getlength(BlockDriverState
*bs
)
2082 BDRVRawState
*s
= bs
->opaque
;
2091 size
= lseek(s
->fd
, 0, SEEK_END
);
2099 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2102 BDRVRawState
*s
= bs
->opaque
;
2104 if (fstat(s
->fd
, &st
) < 0) {
2107 return (int64_t)st
.st_blocks
* 512;
2110 static int coroutine_fn
2111 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2113 BlockdevCreateOptionsFile
*file_opts
;
2115 uint64_t perm
, shared
;
2118 /* Validate options and set default values */
2119 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2120 file_opts
= &options
->u
.file
;
2122 if (!file_opts
->has_nocow
) {
2123 file_opts
->nocow
= false;
2125 if (!file_opts
->has_preallocation
) {
2126 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2130 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2133 error_setg_errno(errp
, -result
, "Could not create file");
2137 /* Take permissions: We want to discard everything, so we need
2138 * BLK_PERM_WRITE; and truncation to the desired size requires
2140 * On the other hand, we cannot share the RESIZE permission
2141 * because we promise that after this function, the file has the
2142 * size given in the options. If someone else were to resize it
2143 * concurrently, we could not guarantee that.
2144 * Note that after this function, we can no longer guarantee that
2145 * the file is not touched by a third party, so it may be resized
2147 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2148 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2150 /* Step one: Take locks */
2151 result
= raw_apply_lock_bytes(fd
, perm
, ~shared
, false, errp
);
2156 /* Step two: Check that nobody else has taken conflicting locks */
2157 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2162 /* Clear the file by truncating it to 0 */
2163 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2168 if (file_opts
->nocow
) {
2170 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2171 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2172 * will be ignored since any failure of this operation should not
2173 * block the left work.
2176 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2177 attr
|= FS_NOCOW_FL
;
2178 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2183 /* Resize and potentially preallocate the file to the desired
2185 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2186 file_opts
->preallocation
, errp
);
2192 if (qemu_close(fd
) != 0 && result
== 0) {
2194 error_setg_errno(errp
, -result
, "Could not close the new file");
2200 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2203 BlockdevCreateOptions options
;
2204 int64_t total_size
= 0;
2206 PreallocMode prealloc
;
2208 Error
*local_err
= NULL
;
2210 /* Skip file: protocol prefix */
2211 strstart(filename
, "file:", &filename
);
2213 /* Read out options */
2214 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2216 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2217 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2218 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2219 PREALLOC_MODE_OFF
, &local_err
);
2222 error_propagate(errp
, local_err
);
2226 options
= (BlockdevCreateOptions
) {
2227 .driver
= BLOCKDEV_DRIVER_FILE
,
2229 .filename
= (char *) filename
,
2231 .has_preallocation
= true,
2232 .preallocation
= prealloc
,
2237 return raw_co_create(&options
, errp
);
2241 * Find allocation range in @bs around offset @start.
2242 * May change underlying file descriptor's file offset.
2243 * If @start is not in a hole, store @start in @data, and the
2244 * beginning of the next hole in @hole, and return 0.
2245 * If @start is in a non-trailing hole, store @start in @hole and the
2246 * beginning of the next non-hole in @data, and return 0.
2247 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2248 * If we can't find out, return a negative errno other than -ENXIO.
2250 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2251 off_t
*data
, off_t
*hole
)
2253 #if defined SEEK_HOLE && defined SEEK_DATA
2254 BDRVRawState
*s
= bs
->opaque
;
2259 * D1. offs == start: start is in data
2260 * D2. offs > start: start is in a hole, next data at offs
2261 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2262 * or start is beyond EOF
2263 * If the latter happens, the file has been truncated behind
2264 * our back since we opened it. All bets are off then.
2265 * Treating like a trailing hole is simplest.
2266 * D4. offs < 0, errno != ENXIO: we learned nothing
2268 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2270 return -errno
; /* D3 or D4 */
2274 /* This is not a valid return by lseek(). We are safe to just return
2275 * -EIO in this case, and we'll treat it like D4. */
2280 /* D2: in hole, next data at offs */
2286 /* D1: in data, end not yet known */
2290 * H1. offs == start: start is in a hole
2291 * If this happens here, a hole has been dug behind our back
2292 * since the previous lseek().
2293 * H2. offs > start: either start is in data, next hole at offs,
2294 * or start is in trailing hole, EOF at offs
2295 * Linux treats trailing holes like any other hole: offs ==
2296 * start. Solaris seeks to EOF instead: offs > start (blech).
2297 * If that happens here, a hole has been dug behind our back
2298 * since the previous lseek().
2299 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2300 * If this happens, the file has been truncated behind our
2301 * back since we opened it. Treat it like a trailing hole.
2302 * H4. offs < 0, errno != ENXIO: we learned nothing
2303 * Pretend we know nothing at all, i.e. "forget" about D1.
2305 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2307 return -errno
; /* D1 and (H3 or H4) */
2311 /* This is not a valid return by lseek(). We are safe to just return
2312 * -EIO in this case, and we'll treat it like H4. */
2318 * D1 and H2: either in data, next hole at offs, or it was in
2319 * data but is now in a trailing hole. In the latter case,
2320 * all bets are off. Treating it as if it there was data all
2321 * the way to EOF is safe, so simply do that.
2336 * Returns the allocation status of the specified offset.
2338 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2340 * 'pnum' is set to the number of bytes (including and immediately following
2341 * the specified offset) that are known to be in the same
2342 * allocated/unallocated state.
2344 * 'bytes' is the max value 'pnum' should be set to.
2346 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2349 int64_t bytes
, int64_t *pnum
,
2351 BlockDriverState
**file
)
2353 off_t data
= 0, hole
= 0;
2365 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2368 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2369 if (ret
== -ENXIO
) {
2372 ret
= BDRV_BLOCK_ZERO
;
2373 } else if (ret
< 0) {
2374 /* No info available, so pretend there are no holes */
2376 ret
= BDRV_BLOCK_DATA
;
2377 } else if (data
== offset
) {
2378 /* On a data extent, compute bytes to the end of the extent,
2379 * possibly including a partial sector at EOF. */
2380 *pnum
= MIN(bytes
, hole
- offset
);
2381 ret
= BDRV_BLOCK_DATA
;
2383 /* On a hole, compute bytes to the beginning of the next extent. */
2384 assert(hole
== offset
);
2385 *pnum
= MIN(bytes
, data
- offset
);
2386 ret
= BDRV_BLOCK_ZERO
;
2390 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2393 #if defined(__linux__)
2394 /* Verify that the file is not in the page cache */
2395 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2397 const size_t window_size
= 128 * 1024 * 1024;
2398 BDRVRawState
*s
= bs
->opaque
;
2399 void *window
= NULL
;
2406 /* mincore(2) page status information requires 1 byte per page */
2407 page_size
= sysconf(_SC_PAGESIZE
);
2408 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2410 end
= raw_getlength(bs
);
2412 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2419 /* Unmap previous window if size has changed */
2420 new_length
= MIN(end
- offset
, window_size
);
2421 if (new_length
!= length
) {
2422 munmap(window
, length
);
2427 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2429 if (new_window
== MAP_FAILED
) {
2430 error_setg_errno(errp
, errno
, "mmap failed");
2434 window
= new_window
;
2435 length
= new_length
;
2437 ret
= mincore(window
, length
, vec
);
2439 error_setg_errno(errp
, errno
, "mincore failed");
2443 vec_end
= DIV_ROUND_UP(length
, page_size
);
2444 for (i
= 0; i
< vec_end
; i
++) {
2446 error_setg(errp
, "page cache still in use!");
2453 munmap(window
, length
);
2458 #endif /* __linux__ */
2460 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2463 BDRVRawState
*s
= bs
->opaque
;
2468 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2472 if (s
->open_flags
& O_DIRECT
) {
2473 return; /* No host kernel page cache */
2476 #if defined(__linux__)
2477 /* This sets the scene for the next syscall... */
2478 ret
= bdrv_co_flush(bs
);
2480 error_setg_errno(errp
, -ret
, "flush failed");
2484 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2485 * process. These limitations are okay because we just fsynced the file,
2486 * we don't use mmap, and the file should not be in use by other processes.
2488 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2489 if (ret
!= 0) { /* the return value is a positive errno */
2490 error_setg_errno(errp
, ret
, "fadvise failed");
2494 if (s
->check_cache_dropped
) {
2495 check_cache_dropped(bs
, errp
);
2497 #else /* __linux__ */
2498 /* Do nothing. Live migration to a remote host with cache.direct=off is
2499 * unsupported on other host operating systems. Cache consistency issues
2500 * may occur but no error is reported here, partly because that's the
2501 * historical behavior and partly because it's hard to differentiate valid
2502 * configurations that should not cause errors.
2504 #endif /* !__linux__ */
2507 static coroutine_fn
int
2508 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2510 BDRVRawState
*s
= bs
->opaque
;
2512 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, QEMU_AIO_DISCARD
);
2515 static int coroutine_fn
raw_co_pwrite_zeroes(
2516 BlockDriverState
*bs
, int64_t offset
,
2517 int bytes
, BdrvRequestFlags flags
)
2519 BDRVRawState
*s
= bs
->opaque
;
2521 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
2522 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2523 QEMU_AIO_WRITE_ZEROES
);
2524 } else if (s
->discard_zeroes
) {
2525 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
2531 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2533 BDRVRawState
*s
= bs
->opaque
;
2535 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2539 static QemuOptsList raw_create_opts
= {
2540 .name
= "raw-create-opts",
2541 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2544 .name
= BLOCK_OPT_SIZE
,
2545 .type
= QEMU_OPT_SIZE
,
2546 .help
= "Virtual disk size"
2549 .name
= BLOCK_OPT_NOCOW
,
2550 .type
= QEMU_OPT_BOOL
,
2551 .help
= "Turn off copy-on-write (valid only on btrfs)"
2554 .name
= BLOCK_OPT_PREALLOC
,
2555 .type
= QEMU_OPT_STRING
,
2556 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2558 { /* end of list */ }
2562 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2565 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2568 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2570 BDRVRawState
*s
= bs
->opaque
;
2571 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2573 s
->shared_perm
= shared
;
2576 static void raw_abort_perm_update(BlockDriverState
*bs
)
2578 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2581 static int coroutine_fn
raw_co_copy_range_from(BlockDriverState
*bs
,
2582 BdrvChild
*src
, uint64_t src_offset
,
2583 BdrvChild
*dst
, uint64_t dst_offset
,
2584 uint64_t bytes
, BdrvRequestFlags flags
)
2586 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
, flags
);
2589 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2590 BdrvChild
*src
, uint64_t src_offset
,
2591 BdrvChild
*dst
, uint64_t dst_offset
,
2592 uint64_t bytes
, BdrvRequestFlags flags
)
2594 BDRVRawState
*s
= bs
->opaque
;
2595 BDRVRawState
*src_s
;
2597 assert(dst
->bs
== bs
);
2598 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2602 src_s
= src
->bs
->opaque
;
2603 if (fd_open(bs
) < 0 || fd_open(bs
) < 0) {
2606 return paio_submit_co_full(bs
, src_s
->fd
, src_offset
, s
->fd
, dst_offset
,
2607 NULL
, bytes
, QEMU_AIO_COPY_RANGE
);
2610 BlockDriver bdrv_file
= {
2611 .format_name
= "file",
2612 .protocol_name
= "file",
2613 .instance_size
= sizeof(BDRVRawState
),
2614 .bdrv_needs_filename
= true,
2615 .bdrv_probe
= NULL
, /* no probe for protocols */
2616 .bdrv_parse_filename
= raw_parse_filename
,
2617 .bdrv_file_open
= raw_open
,
2618 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2619 .bdrv_reopen_commit
= raw_reopen_commit
,
2620 .bdrv_reopen_abort
= raw_reopen_abort
,
2621 .bdrv_close
= raw_close
,
2622 .bdrv_co_create
= raw_co_create
,
2623 .bdrv_co_create_opts
= raw_co_create_opts
,
2624 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2625 .bdrv_co_block_status
= raw_co_block_status
,
2626 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2627 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2629 .bdrv_co_preadv
= raw_co_preadv
,
2630 .bdrv_co_pwritev
= raw_co_pwritev
,
2631 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2632 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2633 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2634 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2635 .bdrv_refresh_limits
= raw_refresh_limits
,
2636 .bdrv_io_plug
= raw_aio_plug
,
2637 .bdrv_io_unplug
= raw_aio_unplug
,
2638 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2640 .bdrv_co_truncate
= raw_co_truncate
,
2641 .bdrv_getlength
= raw_getlength
,
2642 .bdrv_get_info
= raw_get_info
,
2643 .bdrv_get_allocated_file_size
2644 = raw_get_allocated_file_size
,
2645 .bdrv_check_perm
= raw_check_perm
,
2646 .bdrv_set_perm
= raw_set_perm
,
2647 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2648 .create_opts
= &raw_create_opts
,
2651 /***********************************************/
2654 #if defined(__APPLE__) && defined(__MACH__)
2655 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2656 CFIndex maxPathSize
, int flags
);
2657 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2659 kern_return_t kernResult
= KERN_FAILURE
;
2660 mach_port_t masterPort
;
2661 CFMutableDictionaryRef classesToMatch
;
2662 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2663 char *mediaType
= NULL
;
2665 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2666 if ( KERN_SUCCESS
!= kernResult
) {
2667 printf( "IOMasterPort returned %d\n", kernResult
);
2671 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2672 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2673 if (classesToMatch
== NULL
) {
2674 error_report("IOServiceMatching returned NULL for %s",
2675 matching_array
[index
]);
2678 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2680 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2682 if (kernResult
!= KERN_SUCCESS
) {
2683 error_report("Note: IOServiceGetMatchingServices returned %d",
2688 /* If a match was found, leave the loop */
2689 if (*mediaIterator
!= 0) {
2690 DPRINTF("Matching using %s\n", matching_array
[index
]);
2691 mediaType
= g_strdup(matching_array
[index
]);
2698 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2699 CFIndex maxPathSize
, int flags
)
2701 io_object_t nextMedia
;
2702 kern_return_t kernResult
= KERN_FAILURE
;
2704 nextMedia
= IOIteratorNext( mediaIterator
);
2707 CFTypeRef bsdPathAsCFString
;
2708 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2709 if ( bsdPathAsCFString
) {
2710 size_t devPathLength
;
2711 strcpy( bsdPath
, _PATH_DEV
);
2712 if (flags
& BDRV_O_NOCACHE
) {
2713 strcat(bsdPath
, "r");
2715 devPathLength
= strlen( bsdPath
);
2716 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2717 kernResult
= KERN_SUCCESS
;
2719 CFRelease( bsdPathAsCFString
);
2721 IOObjectRelease( nextMedia
);
2727 /* Sets up a real cdrom for use in QEMU */
2728 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2730 int index
, num_of_test_partitions
= 2, fd
;
2731 char test_partition
[MAXPATHLEN
];
2732 bool partition_found
= false;
2734 /* look for a working partition */
2735 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2736 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2738 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2740 partition_found
= true;
2746 /* if a working partition on the device was not found */
2747 if (partition_found
== false) {
2748 error_setg(errp
, "Failed to find a working partition on disc");
2750 DPRINTF("Using %s as optical disc\n", test_partition
);
2751 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2753 return partition_found
;
2756 /* Prints directions on mounting and unmounting a device */
2757 static void print_unmounting_directions(const char *file_name
)
2759 error_report("If device %s is mounted on the desktop, unmount"
2760 " it first before using it in QEMU", file_name
);
2761 error_report("Command to unmount device: diskutil unmountDisk %s",
2763 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2766 #endif /* defined(__APPLE__) && defined(__MACH__) */
2768 static int hdev_probe_device(const char *filename
)
2772 /* allow a dedicated CD-ROM driver to match with a higher priority */
2773 if (strstart(filename
, "/dev/cdrom", NULL
))
2776 if (stat(filename
, &st
) >= 0 &&
2777 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2784 static int check_hdev_writable(BDRVRawState
*s
)
2786 #if defined(BLKROGET)
2787 /* Linux block devices can be configured "read-only" using blockdev(8).
2788 * This is independent of device node permissions and therefore open(2)
2789 * with O_RDWR succeeds. Actual writes fail with EPERM.
2791 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2792 * check for read-only block devices so that Linux block devices behave
2798 if (fstat(s
->fd
, &st
)) {
2802 if (!S_ISBLK(st
.st_mode
)) {
2806 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2813 #endif /* defined(BLKROGET) */
2817 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2820 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2823 static bool hdev_is_sg(BlockDriverState
*bs
)
2826 #if defined(__linux__)
2828 BDRVRawState
*s
= bs
->opaque
;
2830 struct sg_scsi_id scsiid
;
2834 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2838 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2843 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
2845 DPRINTF("SG device found: type=%d, version=%d\n",
2846 scsiid
.scsi_type
, sg_version
);
2855 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
2858 BDRVRawState
*s
= bs
->opaque
;
2859 Error
*local_err
= NULL
;
2862 #if defined(__APPLE__) && defined(__MACH__)
2864 * Caution: while qdict_get_str() is fine, getting non-string types
2865 * would require more care. When @options come from -blockdev or
2866 * blockdev_add, its members are typed according to the QAPI
2867 * schema, but when they come from -drive, they're all QString.
2869 const char *filename
= qdict_get_str(options
, "filename");
2870 char bsd_path
[MAXPATHLEN
] = "";
2871 bool error_occurred
= false;
2873 /* If using a real cdrom */
2874 if (strcmp(filename
, "/dev/cdrom") == 0) {
2875 char *mediaType
= NULL
;
2876 kern_return_t ret_val
;
2877 io_iterator_t mediaIterator
= 0;
2879 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
2880 if (mediaType
== NULL
) {
2881 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
2883 error_occurred
= true;
2884 goto hdev_open_Mac_error
;
2887 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
2888 if (ret_val
!= KERN_SUCCESS
) {
2889 error_setg(errp
, "Could not get BSD path for optical drive");
2890 error_occurred
= true;
2891 goto hdev_open_Mac_error
;
2894 /* If a real optical drive was not found */
2895 if (bsd_path
[0] == '\0') {
2896 error_setg(errp
, "Failed to obtain bsd path for optical drive");
2897 error_occurred
= true;
2898 goto hdev_open_Mac_error
;
2901 /* If using a cdrom disc and finding a partition on the disc failed */
2902 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
2903 setup_cdrom(bsd_path
, errp
) == false) {
2904 print_unmounting_directions(bsd_path
);
2905 error_occurred
= true;
2906 goto hdev_open_Mac_error
;
2909 qdict_put_str(options
, "filename", bsd_path
);
2911 hdev_open_Mac_error
:
2913 if (mediaIterator
) {
2914 IOObjectRelease(mediaIterator
);
2916 if (error_occurred
) {
2920 #endif /* defined(__APPLE__) && defined(__MACH__) */
2922 s
->type
= FTYPE_FILE
;
2924 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
2926 error_propagate(errp
, local_err
);
2927 #if defined(__APPLE__) && defined(__MACH__)
2929 filename
= bsd_path
;
2931 /* if a physical device experienced an error while being opened */
2932 if (strncmp(filename
, "/dev/", 5) == 0) {
2933 print_unmounting_directions(filename
);
2935 #endif /* defined(__APPLE__) && defined(__MACH__) */
2939 /* Since this does ioctl the device must be already opened */
2940 bs
->sg
= hdev_is_sg(bs
);
2942 if (flags
& BDRV_O_RDWR
) {
2943 ret
= check_hdev_writable(s
);
2946 error_setg_errno(errp
, -ret
, "The device is not writable");
2954 #if defined(__linux__)
2956 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
2957 unsigned long int req
, void *buf
,
2958 BlockCompletionFunc
*cb
, void *opaque
)
2960 BDRVRawState
*s
= bs
->opaque
;
2961 RawPosixAIOData
*acb
;
2964 if (fd_open(bs
) < 0)
2967 if (req
== SG_IO
&& s
->pr_mgr
) {
2968 struct sg_io_hdr
*io_hdr
= buf
;
2969 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
2970 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
2971 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
2972 s
->fd
, io_hdr
, cb
, opaque
);
2976 acb
= g_new(RawPosixAIOData
, 1);
2978 acb
->aio_type
= QEMU_AIO_IOCTL
;
2979 acb
->aio_fildes
= s
->fd
;
2980 acb
->aio_offset
= 0;
2981 acb
->aio_ioctl_buf
= buf
;
2982 acb
->aio_ioctl_cmd
= req
;
2983 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
2984 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
2988 static int fd_open(BlockDriverState
*bs
)
2990 BDRVRawState
*s
= bs
->opaque
;
2992 /* this is just to ensure s->fd is sane (its called by io ops) */
2998 static coroutine_fn
int
2999 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3001 BDRVRawState
*s
= bs
->opaque
;
3008 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3009 QEMU_AIO_DISCARD
| QEMU_AIO_BLKDEV
);
3012 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3013 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3015 BDRVRawState
*s
= bs
->opaque
;
3022 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
3023 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3024 QEMU_AIO_WRITE_ZEROES
|QEMU_AIO_BLKDEV
);
3025 } else if (s
->discard_zeroes
) {
3026 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3027 QEMU_AIO_DISCARD
|QEMU_AIO_BLKDEV
);
3032 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3037 struct stat stat_buf
;
3038 int64_t total_size
= 0;
3041 /* This function is used by both protocol block drivers and therefore either
3042 * of these prefixes may be given.
3043 * The return value has to be stored somewhere, otherwise this is an error
3044 * due to -Werror=unused-value. */
3046 strstart(filename
, "host_device:", &filename
) ||
3047 strstart(filename
, "host_cdrom:" , &filename
);
3051 ret
= raw_normalize_devicepath(&filename
);
3053 error_setg_errno(errp
, -ret
, "Could not normalize device path");
3057 /* Read out options */
3058 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3061 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3064 error_setg_errno(errp
, -ret
, "Could not open device");
3068 if (fstat(fd
, &stat_buf
) < 0) {
3070 error_setg_errno(errp
, -ret
, "Could not stat device");
3071 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3073 "The given file is neither a block nor a character device");
3075 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3076 error_setg(errp
, "Device is too small");
3080 if (!ret
&& total_size
) {
3081 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3082 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3083 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3086 ret
= qemu_write_full(fd
, buf
, zero_size
);
3087 ret
= ret
== zero_size
? 0 : -errno
;
3094 static BlockDriver bdrv_host_device
= {
3095 .format_name
= "host_device",
3096 .protocol_name
= "host_device",
3097 .instance_size
= sizeof(BDRVRawState
),
3098 .bdrv_needs_filename
= true,
3099 .bdrv_probe_device
= hdev_probe_device
,
3100 .bdrv_parse_filename
= hdev_parse_filename
,
3101 .bdrv_file_open
= hdev_open
,
3102 .bdrv_close
= raw_close
,
3103 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3104 .bdrv_reopen_commit
= raw_reopen_commit
,
3105 .bdrv_reopen_abort
= raw_reopen_abort
,
3106 .bdrv_co_create_opts
= hdev_co_create_opts
,
3107 .create_opts
= &raw_create_opts
,
3108 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3109 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3111 .bdrv_co_preadv
= raw_co_preadv
,
3112 .bdrv_co_pwritev
= raw_co_pwritev
,
3113 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3114 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3115 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3116 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3117 .bdrv_refresh_limits
= raw_refresh_limits
,
3118 .bdrv_io_plug
= raw_aio_plug
,
3119 .bdrv_io_unplug
= raw_aio_unplug
,
3121 .bdrv_co_truncate
= raw_co_truncate
,
3122 .bdrv_getlength
= raw_getlength
,
3123 .bdrv_get_info
= raw_get_info
,
3124 .bdrv_get_allocated_file_size
3125 = raw_get_allocated_file_size
,
3126 .bdrv_check_perm
= raw_check_perm
,
3127 .bdrv_set_perm
= raw_set_perm
,
3128 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3129 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3130 .bdrv_probe_geometry
= hdev_probe_geometry
,
3132 /* generic scsi device */
3134 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3138 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3139 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3142 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3147 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3150 BDRVRawState
*s
= bs
->opaque
;
3154 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3155 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, errp
);
3158 static int cdrom_probe_device(const char *filename
)
3164 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3168 ret
= fstat(fd
, &st
);
3169 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3173 /* Attempt to detect via a CDROM specific ioctl */
3174 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3184 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3186 BDRVRawState
*s
= bs
->opaque
;
3189 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3190 return ret
== CDS_DISC_OK
;
3193 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3195 BDRVRawState
*s
= bs
->opaque
;
3198 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3199 perror("CDROMEJECT");
3201 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3202 perror("CDROMEJECT");
3206 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3208 BDRVRawState
*s
= bs
->opaque
;
3210 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3212 * Note: an error can happen if the distribution automatically
3215 /* perror("CDROM_LOCKDOOR"); */
3219 static BlockDriver bdrv_host_cdrom
= {
3220 .format_name
= "host_cdrom",
3221 .protocol_name
= "host_cdrom",
3222 .instance_size
= sizeof(BDRVRawState
),
3223 .bdrv_needs_filename
= true,
3224 .bdrv_probe_device
= cdrom_probe_device
,
3225 .bdrv_parse_filename
= cdrom_parse_filename
,
3226 .bdrv_file_open
= cdrom_open
,
3227 .bdrv_close
= raw_close
,
3228 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3229 .bdrv_reopen_commit
= raw_reopen_commit
,
3230 .bdrv_reopen_abort
= raw_reopen_abort
,
3231 .bdrv_co_create_opts
= hdev_co_create_opts
,
3232 .create_opts
= &raw_create_opts
,
3233 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3236 .bdrv_co_preadv
= raw_co_preadv
,
3237 .bdrv_co_pwritev
= raw_co_pwritev
,
3238 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3239 .bdrv_refresh_limits
= raw_refresh_limits
,
3240 .bdrv_io_plug
= raw_aio_plug
,
3241 .bdrv_io_unplug
= raw_aio_unplug
,
3243 .bdrv_co_truncate
= raw_co_truncate
,
3244 .bdrv_getlength
= raw_getlength
,
3245 .has_variable_length
= true,
3246 .bdrv_get_allocated_file_size
3247 = raw_get_allocated_file_size
,
3249 /* removable device support */
3250 .bdrv_is_inserted
= cdrom_is_inserted
,
3251 .bdrv_eject
= cdrom_eject
,
3252 .bdrv_lock_medium
= cdrom_lock_medium
,
3254 /* generic scsi device */
3255 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3257 #endif /* __linux__ */
3259 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3260 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3263 BDRVRawState
*s
= bs
->opaque
;
3264 Error
*local_err
= NULL
;
3269 ret
= raw_open_common(bs
, options
, flags
, 0, &local_err
);
3271 error_propagate(errp
, local_err
);
3275 /* make sure the door isn't locked at this time */
3276 ioctl(s
->fd
, CDIOCALLOW
);
3280 static int cdrom_probe_device(const char *filename
)
3282 if (strstart(filename
, "/dev/cd", NULL
) ||
3283 strstart(filename
, "/dev/acd", NULL
))
3288 static int cdrom_reopen(BlockDriverState
*bs
)
3290 BDRVRawState
*s
= bs
->opaque
;
3294 * Force reread of possibly changed/newly loaded disc,
3295 * FreeBSD seems to not notice sometimes...
3299 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3306 /* make sure the door isn't locked at this time */
3307 ioctl(s
->fd
, CDIOCALLOW
);
3311 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3313 return raw_getlength(bs
) > 0;
3316 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3318 BDRVRawState
*s
= bs
->opaque
;
3323 (void) ioctl(s
->fd
, CDIOCALLOW
);
3326 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3327 perror("CDIOCEJECT");
3329 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3330 perror("CDIOCCLOSE");
3336 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3338 BDRVRawState
*s
= bs
->opaque
;
3342 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3344 * Note: an error can happen if the distribution automatically
3347 /* perror("CDROM_LOCKDOOR"); */
3351 static BlockDriver bdrv_host_cdrom
= {
3352 .format_name
= "host_cdrom",
3353 .protocol_name
= "host_cdrom",
3354 .instance_size
= sizeof(BDRVRawState
),
3355 .bdrv_needs_filename
= true,
3356 .bdrv_probe_device
= cdrom_probe_device
,
3357 .bdrv_parse_filename
= cdrom_parse_filename
,
3358 .bdrv_file_open
= cdrom_open
,
3359 .bdrv_close
= raw_close
,
3360 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3361 .bdrv_reopen_commit
= raw_reopen_commit
,
3362 .bdrv_reopen_abort
= raw_reopen_abort
,
3363 .bdrv_co_create_opts
= hdev_co_create_opts
,
3364 .create_opts
= &raw_create_opts
,
3366 .bdrv_co_preadv
= raw_co_preadv
,
3367 .bdrv_co_pwritev
= raw_co_pwritev
,
3368 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3369 .bdrv_refresh_limits
= raw_refresh_limits
,
3370 .bdrv_io_plug
= raw_aio_plug
,
3371 .bdrv_io_unplug
= raw_aio_unplug
,
3373 .bdrv_co_truncate
= raw_co_truncate
,
3374 .bdrv_getlength
= raw_getlength
,
3375 .has_variable_length
= true,
3376 .bdrv_get_allocated_file_size
3377 = raw_get_allocated_file_size
,
3379 /* removable device support */
3380 .bdrv_is_inserted
= cdrom_is_inserted
,
3381 .bdrv_eject
= cdrom_eject
,
3382 .bdrv_lock_medium
= cdrom_lock_medium
,
3384 #endif /* __FreeBSD__ */
3386 static void bdrv_file_init(void)
3389 * Register all the drivers. Note that order is important, the driver
3390 * registered last will get probed first.
3392 bdrv_register(&bdrv_file
);
3393 bdrv_register(&bdrv_host_device
);
3395 bdrv_register(&bdrv_host_cdrom
);
3397 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3398 bdrv_register(&bdrv_host_cdrom
);
3402 block_init(bdrv_file_init
);