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>
107 /* OS X does not have O_DSYNC */
110 #define O_DSYNC O_SYNC
111 #elif defined(O_FSYNC)
112 #define O_DSYNC O_FSYNC
116 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
118 #define O_DIRECT O_DSYNC
124 #define MAX_BLOCKSIZE 4096
126 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
127 * leaving a few more bytes for its future use. */
128 #define RAW_LOCK_PERM_BASE 100
129 #define RAW_LOCK_SHARED_BASE 200
131 typedef struct BDRVRawState
{
138 /* The current permissions. */
140 uint64_t shared_perm
;
142 /* The perms bits whose corresponding bytes are already locked in
144 uint64_t locked_perm
;
145 uint64_t locked_shared_perm
;
148 BDRVReopenState
*reopen_state
;
154 bool has_write_zeroes
:1;
155 bool discard_zeroes
:1;
156 bool use_linux_aio
:1;
157 bool page_cache_inconsistent
:1;
159 bool needs_alignment
;
160 bool check_cache_dropped
;
165 typedef struct BDRVRawReopenState
{
168 bool check_cache_dropped
;
169 } BDRVRawReopenState
;
171 static int fd_open(BlockDriverState
*bs
);
172 static int64_t raw_getlength(BlockDriverState
*bs
);
174 typedef struct RawPosixAIOData
{
175 BlockDriverState
*bs
;
196 PreallocMode prealloc
;
202 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
203 static int cdrom_reopen(BlockDriverState
*bs
);
206 #if defined(__NetBSD__)
207 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
209 static char namebuf
[PATH_MAX
];
210 const char *dp
, *fname
;
214 dp
= strrchr(fname
, '/');
215 if (lstat(fname
, &sb
) < 0) {
216 error_setg_errno(errp
, errno
, "%s: stat failed", fname
);
220 if (!S_ISBLK(sb
.st_mode
)) {
225 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
227 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
228 (int)(dp
- fname
), fname
, dp
+ 1);
231 warn_report("%s is a block device, using %s", fname
, *filename
);
236 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
243 * Get logical block size via ioctl. On success store it in @sector_size_p.
245 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
247 unsigned int sector_size
;
248 bool success
= false;
252 static const unsigned long ioctl_list
[] = {
256 #ifdef DKIOCGETBLOCKSIZE
259 #ifdef DIOCGSECTORSIZE
264 /* Try a few ioctls to get the right size */
265 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
266 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
267 *sector_size_p
= sector_size
;
272 return success
? 0 : -errno
;
276 * Get physical block size of @fd.
277 * On success, store it in @blk_size and return 0.
278 * On failure, return -errno.
280 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
283 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
292 /* Check if read is allowed with given memory buffer and length.
294 * This function is used to check O_DIRECT memory buffer and request alignment.
296 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
298 ssize_t ret
= pread(fd
, buf
, len
, 0);
305 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
306 * other errors (e.g. real I/O error), which could happen on a failed
307 * drive, since we only care about probing alignment.
309 if (errno
!= EINVAL
) {
317 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
319 BDRVRawState
*s
= bs
->opaque
;
321 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
323 /* For SCSI generic devices the alignment is not really used.
324 With buffered I/O, we don't have any restrictions. */
325 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
326 bs
->bl
.request_alignment
= 1;
331 bs
->bl
.request_alignment
= 0;
333 /* Let's try to use the logical blocksize for the alignment. */
334 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
335 bs
->bl
.request_alignment
= 0;
340 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
341 bs
->bl
.request_alignment
= da
.d_miniosz
;
342 /* The kernel returns wrong information for d_mem */
343 /* s->buf_align = da.d_mem; */
348 /* If we could not get the sizes so far, we can only guess them */
351 buf
= qemu_memalign(max_align
, 2 * max_align
);
352 for (align
= 512; align
<= max_align
; align
<<= 1) {
353 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
354 s
->buf_align
= align
;
361 if (!bs
->bl
.request_alignment
) {
363 buf
= qemu_memalign(s
->buf_align
, max_align
);
364 for (align
= 512; align
<= max_align
; align
<<= 1) {
365 if (raw_is_io_aligned(fd
, buf
, align
)) {
366 bs
->bl
.request_alignment
= align
;
373 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
374 error_setg(errp
, "Could not find working O_DIRECT alignment");
375 error_append_hint(errp
, "Try cache.direct=off\n");
379 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
381 bool read_write
= false;
382 assert(open_flags
!= NULL
);
384 *open_flags
|= O_BINARY
;
385 *open_flags
&= ~O_ACCMODE
;
387 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
388 read_write
= has_writers
;
389 } else if (bdrv_flags
& BDRV_O_RDWR
) {
394 *open_flags
|= O_RDWR
;
396 *open_flags
|= O_RDONLY
;
399 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
400 * and O_DIRECT for no caching. */
401 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
402 *open_flags
|= O_DIRECT
;
406 static void raw_parse_filename(const char *filename
, QDict
*options
,
409 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
412 static QemuOptsList raw_runtime_opts
= {
414 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
418 .type
= QEMU_OPT_STRING
,
419 .help
= "File name of the image",
423 .type
= QEMU_OPT_STRING
,
424 .help
= "host AIO implementation (threads, native)",
428 .type
= QEMU_OPT_STRING
,
429 .help
= "file locking mode (on/off/auto, default: auto)",
432 .name
= "pr-manager",
433 .type
= QEMU_OPT_STRING
,
434 .help
= "id of persistent reservation manager object (default: none)",
437 .name
= "x-check-cache-dropped",
438 .type
= QEMU_OPT_BOOL
,
439 .help
= "check that page cache was dropped on live migration (default: off)"
441 { /* end of list */ }
445 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
447 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
448 int bdrv_flags
, int open_flags
,
449 bool device
, Error
**errp
)
451 BDRVRawState
*s
= bs
->opaque
;
453 Error
*local_err
= NULL
;
454 const char *filename
= NULL
;
456 BlockdevAioOptions aio
, aio_default
;
461 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
462 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
464 error_propagate(errp
, local_err
);
469 filename
= qemu_opt_get(opts
, "filename");
471 ret
= raw_normalize_devicepath(&filename
, errp
);
476 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
477 ? BLOCKDEV_AIO_OPTIONS_NATIVE
478 : BLOCKDEV_AIO_OPTIONS_THREADS
;
479 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
480 qemu_opt_get(opts
, "aio"),
481 aio_default
, &local_err
);
483 error_propagate(errp
, local_err
);
487 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
489 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
490 qemu_opt_get(opts
, "locking"),
491 ON_OFF_AUTO_AUTO
, &local_err
);
493 error_propagate(errp
, local_err
);
500 if (!qemu_has_ofd_lock()) {
501 warn_report("File lock requested but OFD locking syscall is "
502 "unavailable, falling back to POSIX file locks");
503 error_printf("Due to the implementation, locks can be lost "
507 case ON_OFF_AUTO_OFF
:
510 case ON_OFF_AUTO_AUTO
:
511 s
->use_lock
= qemu_has_ofd_lock();
517 str
= qemu_opt_get(opts
, "pr-manager");
519 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
521 error_propagate(errp
, local_err
);
527 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
530 s
->open_flags
= open_flags
;
531 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
534 fd
= qemu_open(filename
, s
->open_flags
, 0644);
535 ret
= fd
< 0 ? -errno
: 0;
538 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
547 s
->shared_perm
= BLK_PERM_ALL
;
549 #ifdef CONFIG_LINUX_AIO
550 /* Currently Linux does AIO only for files opened with O_DIRECT */
551 if (s
->use_linux_aio
) {
552 if (!(s
->open_flags
& O_DIRECT
)) {
553 error_setg(errp
, "aio=native was specified, but it requires "
554 "cache.direct=on, which was not specified.");
558 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
559 error_prepend(errp
, "Unable to use native AIO: ");
564 if (s
->use_linux_aio
) {
565 error_setg(errp
, "aio=native was specified, but is not supported "
570 #endif /* !defined(CONFIG_LINUX_AIO) */
572 s
->has_discard
= true;
573 s
->has_write_zeroes
= true;
574 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
575 s
->needs_alignment
= true;
578 if (fstat(s
->fd
, &st
) < 0) {
580 error_setg_errno(errp
, errno
, "Could not stat file");
585 if (S_ISBLK(st
.st_mode
)) {
586 warn_report("Opening a block device as a file using the '%s' "
587 "driver is deprecated", bs
->drv
->format_name
);
588 } else if (S_ISCHR(st
.st_mode
)) {
589 warn_report("Opening a character device as a file using the '%s' "
590 "driver is deprecated", bs
->drv
->format_name
);
591 } else if (!S_ISREG(st
.st_mode
)) {
592 error_setg(errp
, "A regular file was expected by the '%s' driver, "
593 "but something else was given", bs
->drv
->format_name
);
597 s
->discard_zeroes
= true;
598 s
->has_fallocate
= true;
601 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
602 error_setg(errp
, "'%s' driver expects either "
603 "a character or block device", bs
->drv
->format_name
);
609 if (S_ISBLK(st
.st_mode
)) {
610 #ifdef BLKDISCARDZEROES
612 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
613 s
->discard_zeroes
= true;
617 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
618 * not rely on the contents of discarded blocks unless using O_DIRECT.
619 * Same for BLKZEROOUT.
621 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
622 s
->discard_zeroes
= false;
623 s
->has_write_zeroes
= false;
628 if (S_ISCHR(st
.st_mode
)) {
630 * The file is a char device (disk), which on FreeBSD isn't behind
631 * a pager, so force all requests to be aligned. This is needed
632 * so QEMU makes sure all IO operations on the device are aligned
633 * to sector size, or else FreeBSD will reject them with EINVAL.
635 s
->needs_alignment
= true;
640 if (platform_test_xfs_fd(s
->fd
)) {
645 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
648 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
655 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
658 BDRVRawState
*s
= bs
->opaque
;
660 s
->type
= FTYPE_FILE
;
661 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
670 #define PERM_FOREACH(i) \
671 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
673 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
674 * file; if @unlock == true, also unlock the unneeded bytes.
675 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
677 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
678 uint64_t perm_lock_bits
,
679 uint64_t shared_perm_lock_bits
,
680 bool unlock
, Error
**errp
)
684 uint64_t locked_perm
, locked_shared_perm
;
687 locked_perm
= s
->locked_perm
;
688 locked_shared_perm
= s
->locked_shared_perm
;
691 * We don't have the previous bits, just lock/unlock for each of the
695 locked_perm
= BLK_PERM_ALL
;
696 locked_shared_perm
= BLK_PERM_ALL
;
699 locked_shared_perm
= 0;
704 int off
= RAW_LOCK_PERM_BASE
+ i
;
705 uint64_t bit
= (1ULL << i
);
706 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
707 ret
= qemu_lock_fd(fd
, off
, 1, false);
709 error_setg(errp
, "Failed to lock byte %d", off
);
712 s
->locked_perm
|= bit
;
714 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
715 ret
= qemu_unlock_fd(fd
, off
, 1);
717 error_setg(errp
, "Failed to unlock byte %d", off
);
720 s
->locked_perm
&= ~bit
;
725 int off
= RAW_LOCK_SHARED_BASE
+ i
;
726 uint64_t bit
= (1ULL << i
);
727 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
728 ret
= qemu_lock_fd(fd
, off
, 1, false);
730 error_setg(errp
, "Failed to lock byte %d", off
);
733 s
->locked_shared_perm
|= bit
;
735 } else if (unlock
&& (locked_shared_perm
& bit
) &&
736 !(shared_perm_lock_bits
& bit
)) {
737 ret
= qemu_unlock_fd(fd
, off
, 1);
739 error_setg(errp
, "Failed to unlock byte %d", off
);
742 s
->locked_shared_perm
&= ~bit
;
749 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
750 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
757 int off
= RAW_LOCK_SHARED_BASE
+ i
;
758 uint64_t p
= 1ULL << i
;
760 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
762 char *perm_name
= bdrv_perm_names(p
);
764 "Failed to get \"%s\" lock",
772 int off
= RAW_LOCK_PERM_BASE
+ i
;
773 uint64_t p
= 1ULL << i
;
774 if (!(shared_perm
& p
)) {
775 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
777 char *perm_name
= bdrv_perm_names(p
);
779 "Failed to get shared \"%s\" lock",
789 static int raw_handle_perm_lock(BlockDriverState
*bs
,
791 uint64_t new_perm
, uint64_t new_shared
,
794 BDRVRawState
*s
= bs
->opaque
;
796 Error
*local_err
= NULL
;
802 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
808 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
809 ~s
->shared_perm
| ~new_shared
,
812 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
816 error_append_hint(errp
,
817 "Is another process using the image [%s]?\n",
821 /* fall through to unlock bytes. */
823 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
826 /* Theoretically the above call only unlocks bytes and it cannot
827 * fail. Something weird happened, report it.
829 warn_report_err(local_err
);
833 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
836 /* Theoretically the above call only unlocks bytes and it cannot
837 * fail. Something weird happened, report it.
839 warn_report_err(local_err
);
846 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
847 int *open_flags
, uint64_t perm
, bool force_dup
,
850 BDRVRawState
*s
= bs
->opaque
;
853 bool has_writers
= perm
&
854 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
855 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
857 fcntl_flags
|= O_NOATIME
;
861 if (s
->type
== FTYPE_CD
) {
862 *open_flags
|= O_NONBLOCK
;
865 raw_parse_flags(flags
, open_flags
, has_writers
);
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 (!force_dup
&& *open_flags
== s
->open_flags
) {
877 /* We're lucky, the existing fd is fine */
881 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
882 /* dup the original fd */
883 fd
= qemu_dup(s
->fd
);
885 ret
= fcntl_setfl(fd
, *open_flags
);
893 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
895 const char *normalized_filename
= bs
->filename
;
896 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
898 assert(!(*open_flags
& O_CREAT
));
899 fd
= qemu_open(normalized_filename
, *open_flags
);
901 error_setg_errno(errp
, errno
, "Could not reopen file");
910 static int raw_reopen_prepare(BDRVReopenState
*state
,
911 BlockReopenQueue
*queue
, Error
**errp
)
914 BDRVRawReopenState
*rs
;
917 Error
*local_err
= NULL
;
919 assert(state
!= NULL
);
920 assert(state
->bs
!= NULL
);
922 s
= state
->bs
->opaque
;
924 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
927 /* Handle options changes */
928 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
929 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
931 error_propagate(errp
, local_err
);
936 rs
->check_cache_dropped
=
937 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
939 /* This driver's reopen function doesn't currently allow changing
940 * other options, so let's put them back in the original QDict and
941 * bdrv_reopen_prepare() will detect changes and complain. */
942 qemu_opts_to_qdict(opts
, state
->options
);
944 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
945 state
->perm
, true, &local_err
);
947 error_propagate(errp
, local_err
);
952 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
953 * alignment with the new fd. */
955 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
957 error_propagate(errp
, local_err
);
963 s
->reopen_state
= state
;
975 static void raw_reopen_commit(BDRVReopenState
*state
)
977 BDRVRawReopenState
*rs
= state
->opaque
;
978 BDRVRawState
*s
= state
->bs
->opaque
;
980 s
->check_cache_dropped
= rs
->check_cache_dropped
;
981 s
->open_flags
= rs
->open_flags
;
986 g_free(state
->opaque
);
987 state
->opaque
= NULL
;
989 assert(s
->reopen_state
== state
);
990 s
->reopen_state
= NULL
;
994 static void raw_reopen_abort(BDRVReopenState
*state
)
996 BDRVRawReopenState
*rs
= state
->opaque
;
997 BDRVRawState
*s
= state
->bs
->opaque
;
999 /* nothing to do if NULL, we didn't get far enough */
1008 g_free(state
->opaque
);
1009 state
->opaque
= NULL
;
1011 assert(s
->reopen_state
== state
);
1012 s
->reopen_state
= NULL
;
1015 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
1019 short max_sectors
= 0;
1020 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1022 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1023 return max_sectors
<< BDRV_SECTOR_BITS
;
1032 static int hdev_get_max_segments(const struct stat
*st
)
1042 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1043 major(st
->st_rdev
), minor(st
->st_rdev
));
1044 fd
= open(sysfspath
, O_RDONLY
);
1050 ret
= read(fd
, buf
, sizeof(buf
) - 1);
1051 } while (ret
== -1 && errno
== EINTR
);
1055 } else if (ret
== 0) {
1060 /* The file is ended with '\n', pass 'end' to accept that. */
1061 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1062 if (ret
== 0 && end
&& *end
== '\n') {
1077 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1079 BDRVRawState
*s
= bs
->opaque
;
1082 if (!fstat(s
->fd
, &st
)) {
1083 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1084 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1085 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1086 bs
->bl
.max_transfer
= pow2floor(ret
);
1088 ret
= hdev_get_max_segments(&st
);
1090 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1091 ret
* getpagesize());
1096 raw_probe_alignment(bs
, s
->fd
, errp
);
1097 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1098 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1101 static int check_for_dasd(int fd
)
1104 struct dasd_information2_t info
= {0};
1106 return ioctl(fd
, BIODASDINFO2
, &info
);
1113 * Try to get @bs's logical and physical block size.
1114 * On success, store them in @bsz and return zero.
1115 * On failure, return negative errno.
1117 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1119 BDRVRawState
*s
= bs
->opaque
;
1122 /* If DASD, get blocksizes */
1123 if (check_for_dasd(s
->fd
) < 0) {
1126 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1130 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1134 * Try to get @bs's geometry: cyls, heads, sectors.
1135 * On success, store them in @geo and return 0.
1136 * On failure return -errno.
1137 * (Allows block driver to assign default geometry values that guest sees)
1140 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1142 BDRVRawState
*s
= bs
->opaque
;
1143 struct hd_geometry ioctl_geo
= {0};
1145 /* If DASD, get its geometry */
1146 if (check_for_dasd(s
->fd
) < 0) {
1149 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1152 /* HDIO_GETGEO may return success even though geo contains zeros
1153 (e.g. certain multipath setups) */
1154 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1157 /* Do not return a geometry for partition */
1158 if (ioctl_geo
.start
!= 0) {
1161 geo
->heads
= ioctl_geo
.heads
;
1162 geo
->sectors
= ioctl_geo
.sectors
;
1163 geo
->cylinders
= ioctl_geo
.cylinders
;
1167 #else /* __linux__ */
1168 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1174 #if defined(__linux__)
1175 static int handle_aiocb_ioctl(void *opaque
)
1177 RawPosixAIOData
*aiocb
= opaque
;
1180 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1189 static int handle_aiocb_flush(void *opaque
)
1191 RawPosixAIOData
*aiocb
= opaque
;
1192 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1195 if (s
->page_cache_inconsistent
) {
1199 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1201 /* There is no clear definition of the semantics of a failing fsync(),
1202 * so we may have to assume the worst. The sad truth is that this
1203 * assumption is correct for Linux. Some pages are now probably marked
1204 * clean in the page cache even though they are inconsistent with the
1205 * on-disk contents. The next fdatasync() call would succeed, but no
1206 * further writeback attempt will be made. We can't get back to a state
1207 * in which we know what is on disk (we would have to rewrite
1208 * everything that was touched since the last fdatasync() at least), so
1209 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1210 * really defined, I have little hope that other OSes are doing better.
1212 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1214 if ((s
->open_flags
& O_DIRECT
) == 0) {
1215 s
->page_cache_inconsistent
= true;
1222 #ifdef CONFIG_PREADV
1224 static bool preadv_present
= true;
1227 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1229 return preadv(fd
, iov
, nr_iov
, offset
);
1233 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1235 return pwritev(fd
, iov
, nr_iov
, offset
);
1240 static bool preadv_present
= false;
1243 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1249 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1256 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1261 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1262 len
= qemu_pwritev(aiocb
->aio_fildes
,
1267 len
= qemu_preadv(aiocb
->aio_fildes
,
1271 } while (len
== -1 && errno
== EINTR
);
1280 * Read/writes the data to/from a given linear buffer.
1282 * Returns the number of bytes handles or -errno in case of an error. Short
1283 * reads are only returned if the end of the file is reached.
1285 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1290 while (offset
< aiocb
->aio_nbytes
) {
1291 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1292 len
= pwrite(aiocb
->aio_fildes
,
1293 (const char *)buf
+ offset
,
1294 aiocb
->aio_nbytes
- offset
,
1295 aiocb
->aio_offset
+ offset
);
1297 len
= pread(aiocb
->aio_fildes
,
1299 aiocb
->aio_nbytes
- offset
,
1300 aiocb
->aio_offset
+ offset
);
1302 if (len
== -1 && errno
== EINTR
) {
1304 } else if (len
== -1 && errno
== EINVAL
&&
1305 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1306 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1308 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1309 * after a short read. Assume that O_DIRECT short reads only occur
1310 * at EOF. Therefore this is a short read, not an I/O error.
1313 } else if (len
== -1) {
1316 } else if (len
== 0) {
1325 static int handle_aiocb_rw(void *opaque
)
1327 RawPosixAIOData
*aiocb
= opaque
;
1331 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1333 * If there is just a single buffer, and it is properly aligned
1334 * we can just use plain pread/pwrite without any problems.
1336 if (aiocb
->io
.niov
== 1) {
1337 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1341 * We have more than one iovec, and all are properly aligned.
1343 * Try preadv/pwritev first and fall back to linearizing the
1344 * buffer if it's not supported.
1346 if (preadv_present
) {
1347 nbytes
= handle_aiocb_rw_vector(aiocb
);
1348 if (nbytes
== aiocb
->aio_nbytes
||
1349 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1352 preadv_present
= false;
1356 * XXX(hch): short read/write. no easy way to handle the reminder
1357 * using these interfaces. For now retry using plain
1363 * Ok, we have to do it the hard way, copy all segments into
1364 * a single aligned buffer.
1366 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1372 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1376 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1377 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1378 p
+= aiocb
->io
.iov
[i
].iov_len
;
1380 assert(p
- buf
== aiocb
->aio_nbytes
);
1383 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1384 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1386 size_t count
= aiocb
->aio_nbytes
, copy
;
1389 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1391 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1392 copy
= aiocb
->io
.iov
[i
].iov_len
;
1394 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1395 assert(count
>= copy
);
1404 if (nbytes
== aiocb
->aio_nbytes
) {
1406 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1407 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1410 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1411 0, aiocb
->aio_nbytes
- nbytes
);
1421 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1423 struct xfs_flock64 fl
;
1426 memset(&fl
, 0, sizeof(fl
));
1427 fl
.l_whence
= SEEK_SET
;
1428 fl
.l_start
= offset
;
1431 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1433 trace_file_xfs_write_zeroes(strerror(errno
));
1440 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1442 struct xfs_flock64 fl
;
1445 memset(&fl
, 0, sizeof(fl
));
1446 fl
.l_whence
= SEEK_SET
;
1447 fl
.l_start
= offset
;
1450 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1452 trace_file_xfs_discard(strerror(errno
));
1460 static int translate_err(int err
)
1462 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1469 #ifdef CONFIG_FALLOCATE
1470 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1473 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1476 } while (errno
== EINTR
);
1477 return translate_err(-errno
);
1481 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1484 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1486 if (!s
->has_write_zeroes
) {
1492 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1493 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1496 } while (errno
== EINTR
);
1498 ret
= translate_err(-errno
);
1501 if (ret
== -ENOTSUP
) {
1502 s
->has_write_zeroes
= false;
1507 static int handle_aiocb_write_zeroes(void *opaque
)
1509 RawPosixAIOData
*aiocb
= opaque
;
1510 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1511 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1513 #ifdef CONFIG_FALLOCATE
1517 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1518 return handle_aiocb_write_zeroes_block(aiocb
);
1523 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1527 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1528 if (s
->has_write_zeroes
) {
1529 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1530 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1531 if (ret
== 0 || ret
!= -ENOTSUP
) {
1534 s
->has_write_zeroes
= false;
1538 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1539 if (s
->has_discard
&& s
->has_fallocate
) {
1540 int ret
= do_fallocate(s
->fd
,
1541 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1542 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1544 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1545 if (ret
== 0 || ret
!= -ENOTSUP
) {
1548 s
->has_fallocate
= false;
1549 } else if (ret
!= -ENOTSUP
) {
1552 s
->has_discard
= false;
1557 #ifdef CONFIG_FALLOCATE
1558 /* Last resort: we are trying to extend the file with zeroed data. This
1559 * can be done via fallocate(fd, 0) */
1560 len
= bdrv_getlength(aiocb
->bs
);
1561 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1562 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1563 if (ret
== 0 || ret
!= -ENOTSUP
) {
1566 s
->has_fallocate
= false;
1573 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1575 RawPosixAIOData
*aiocb
= opaque
;
1576 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1579 /* First try to write zeros and unmap at the same time */
1581 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1582 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1583 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1584 if (ret
!= -ENOTSUP
) {
1591 /* xfs_discard() guarantees that the discarded area reads as all-zero
1592 * afterwards, so we can use it here. */
1593 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1597 /* If we couldn't manage to unmap while guaranteed that the area reads as
1598 * all-zero afterwards, just write zeroes without unmapping */
1599 ret
= handle_aiocb_write_zeroes(aiocb
);
1603 #ifndef HAVE_COPY_FILE_RANGE
1604 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1605 off_t
*out_off
, size_t len
, unsigned int flags
)
1607 #ifdef __NR_copy_file_range
1608 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1609 out_off
, len
, flags
);
1617 static int handle_aiocb_copy_range(void *opaque
)
1619 RawPosixAIOData
*aiocb
= opaque
;
1620 uint64_t bytes
= aiocb
->aio_nbytes
;
1621 off_t in_off
= aiocb
->aio_offset
;
1622 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1625 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1626 aiocb
->copy_range
.aio_fd2
, &out_off
,
1628 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1629 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1632 /* No progress (e.g. when beyond EOF), let the caller fall back to
1651 static int handle_aiocb_discard(void *opaque
)
1653 RawPosixAIOData
*aiocb
= opaque
;
1654 int ret
= -EOPNOTSUPP
;
1655 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1657 if (!s
->has_discard
) {
1661 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1664 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1665 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1668 } while (errno
== EINTR
);
1675 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1679 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1680 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1681 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1685 ret
= translate_err(ret
);
1686 if (ret
== -ENOTSUP
) {
1687 s
->has_discard
= false;
1692 static int handle_aiocb_truncate(void *opaque
)
1694 RawPosixAIOData
*aiocb
= opaque
;
1696 int64_t current_length
= 0;
1699 int fd
= aiocb
->aio_fildes
;
1700 int64_t offset
= aiocb
->aio_offset
;
1701 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1702 Error
**errp
= aiocb
->truncate
.errp
;
1704 if (fstat(fd
, &st
) < 0) {
1706 error_setg_errno(errp
, -result
, "Could not stat file");
1710 current_length
= st
.st_size
;
1711 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1712 error_setg(errp
, "Cannot use preallocation for shrinking files");
1717 #ifdef CONFIG_POSIX_FALLOCATE
1718 case PREALLOC_MODE_FALLOC
:
1720 * Truncating before posix_fallocate() makes it about twice slower on
1721 * file systems that do not support fallocate(), trying to check if a
1722 * block is allocated before allocating it, so don't do that here.
1724 if (offset
!= current_length
) {
1725 result
= -posix_fallocate(fd
, current_length
,
1726 offset
- current_length
);
1728 /* posix_fallocate() doesn't set errno. */
1729 error_setg_errno(errp
, -result
,
1730 "Could not preallocate new data");
1737 case PREALLOC_MODE_FULL
:
1739 int64_t num
= 0, left
= offset
- current_length
;
1743 * Knowing the final size from the beginning could allow the file
1744 * system driver to do less allocations and possibly avoid
1745 * fragmentation of the file.
1747 if (ftruncate(fd
, offset
) != 0) {
1749 error_setg_errno(errp
, -result
, "Could not resize file");
1753 buf
= g_malloc0(65536);
1755 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1756 if (seek_result
< 0) {
1758 error_setg_errno(errp
, -result
,
1759 "Failed to seek to the old end of file");
1764 num
= MIN(left
, 65536);
1765 result
= write(fd
, buf
, num
);
1767 if (errno
== EINTR
) {
1771 error_setg_errno(errp
, -result
,
1772 "Could not write zeros for preallocation");
1781 error_setg_errno(errp
, -result
,
1782 "Could not flush file to disk");
1788 case PREALLOC_MODE_OFF
:
1789 if (ftruncate(fd
, offset
) != 0) {
1791 error_setg_errno(errp
, -result
, "Could not resize file");
1796 error_setg(errp
, "Unsupported preallocation mode: %s",
1797 PreallocMode_str(prealloc
));
1803 if (ftruncate(fd
, current_length
) < 0) {
1804 error_report("Failed to restore old file length: %s",
1813 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1814 ThreadPoolFunc func
, void *arg
)
1816 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1817 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1818 return thread_pool_submit_co(pool
, func
, arg
);
1821 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1822 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1824 BDRVRawState
*s
= bs
->opaque
;
1825 RawPosixAIOData acb
;
1827 if (fd_open(bs
) < 0)
1831 * Check if the underlying device requires requests to be aligned,
1832 * and if the request we are trying to submit is aligned or not.
1833 * If this is the case tell the low-level driver that it needs
1834 * to copy the buffer.
1836 if (s
->needs_alignment
) {
1837 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1838 type
|= QEMU_AIO_MISALIGNED
;
1839 #ifdef CONFIG_LINUX_AIO
1840 } else if (s
->use_linux_aio
) {
1841 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1842 assert(qiov
->size
== bytes
);
1843 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1848 acb
= (RawPosixAIOData
) {
1850 .aio_fildes
= s
->fd
,
1852 .aio_offset
= offset
,
1853 .aio_nbytes
= bytes
,
1860 assert(qiov
->size
== bytes
);
1861 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1864 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1865 uint64_t bytes
, QEMUIOVector
*qiov
,
1868 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1871 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1872 uint64_t bytes
, QEMUIOVector
*qiov
,
1876 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1879 static void raw_aio_plug(BlockDriverState
*bs
)
1881 #ifdef CONFIG_LINUX_AIO
1882 BDRVRawState
*s
= bs
->opaque
;
1883 if (s
->use_linux_aio
) {
1884 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1885 laio_io_plug(bs
, aio
);
1890 static void raw_aio_unplug(BlockDriverState
*bs
)
1892 #ifdef CONFIG_LINUX_AIO
1893 BDRVRawState
*s
= bs
->opaque
;
1894 if (s
->use_linux_aio
) {
1895 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1896 laio_io_unplug(bs
, aio
);
1901 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1903 BDRVRawState
*s
= bs
->opaque
;
1904 RawPosixAIOData acb
;
1912 acb
= (RawPosixAIOData
) {
1914 .aio_fildes
= s
->fd
,
1915 .aio_type
= QEMU_AIO_FLUSH
,
1918 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1921 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1922 AioContext
*new_context
)
1924 #ifdef CONFIG_LINUX_AIO
1925 BDRVRawState
*s
= bs
->opaque
;
1926 if (s
->use_linux_aio
) {
1928 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1929 error_reportf_err(local_err
, "Unable to use native AIO, "
1930 "falling back to thread pool: ");
1931 s
->use_linux_aio
= false;
1937 static void raw_close(BlockDriverState
*bs
)
1939 BDRVRawState
*s
= bs
->opaque
;
1948 * Truncates the given regular file @fd to @offset and, when growing, fills the
1949 * new space according to @prealloc.
1951 * Returns: 0 on success, -errno on failure.
1953 static int coroutine_fn
1954 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1955 PreallocMode prealloc
, Error
**errp
)
1957 RawPosixAIOData acb
;
1959 acb
= (RawPosixAIOData
) {
1962 .aio_type
= QEMU_AIO_TRUNCATE
,
1963 .aio_offset
= offset
,
1965 .prealloc
= prealloc
,
1970 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
1973 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1974 PreallocMode prealloc
, Error
**errp
)
1976 BDRVRawState
*s
= bs
->opaque
;
1980 if (fstat(s
->fd
, &st
)) {
1982 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
1986 if (S_ISREG(st
.st_mode
)) {
1987 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
1990 if (prealloc
!= PREALLOC_MODE_OFF
) {
1991 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
1992 "non-regular file", PreallocMode_str(prealloc
));
1996 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
1997 if (offset
> raw_getlength(bs
)) {
1998 error_setg(errp
, "Cannot grow device files");
2002 error_setg(errp
, "Resizing this file is not supported");
2010 static int64_t raw_getlength(BlockDriverState
*bs
)
2012 BDRVRawState
*s
= bs
->opaque
;
2018 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2019 struct disklabel dl
;
2021 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2023 return (uint64_t)dl
.d_secsize
*
2024 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2028 #elif defined(__NetBSD__)
2029 static int64_t raw_getlength(BlockDriverState
*bs
)
2031 BDRVRawState
*s
= bs
->opaque
;
2037 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2038 struct dkwedge_info dkw
;
2040 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2041 return dkw
.dkw_size
* 512;
2043 struct disklabel dl
;
2045 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2047 return (uint64_t)dl
.d_secsize
*
2048 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2053 #elif defined(__sun__)
2054 static int64_t raw_getlength(BlockDriverState
*bs
)
2056 BDRVRawState
*s
= bs
->opaque
;
2057 struct dk_minfo minfo
;
2067 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2069 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2071 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2075 * There are reports that lseek on some devices fails, but
2076 * irc discussion said that contingency on contingency was overkill.
2078 size
= lseek(s
->fd
, 0, SEEK_END
);
2084 #elif defined(CONFIG_BSD)
2085 static int64_t raw_getlength(BlockDriverState
*bs
)
2087 BDRVRawState
*s
= bs
->opaque
;
2091 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2100 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2103 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2104 #ifdef DIOCGMEDIASIZE
2105 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2106 #elif defined(DIOCGPART)
2109 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2110 size
= pi
.media_size
;
2116 #if defined(__APPLE__) && defined(__MACH__)
2118 uint64_t sectors
= 0;
2119 uint32_t sector_size
= 0;
2121 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2122 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2123 size
= sectors
* sector_size
;
2125 size
= lseek(fd
, 0LL, SEEK_END
);
2132 size
= lseek(fd
, 0LL, SEEK_END
);
2137 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2140 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2141 if (size
== 2048LL * (unsigned)-1)
2143 /* XXX no disc? maybe we need to reopen... */
2144 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2151 size
= lseek(fd
, 0, SEEK_END
);
2159 static int64_t raw_getlength(BlockDriverState
*bs
)
2161 BDRVRawState
*s
= bs
->opaque
;
2170 size
= lseek(s
->fd
, 0, SEEK_END
);
2178 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2181 BDRVRawState
*s
= bs
->opaque
;
2183 if (fstat(s
->fd
, &st
) < 0) {
2186 return (int64_t)st
.st_blocks
* 512;
2189 static int coroutine_fn
2190 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2192 BlockdevCreateOptionsFile
*file_opts
;
2193 Error
*local_err
= NULL
;
2195 uint64_t perm
, shared
;
2198 /* Validate options and set default values */
2199 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2200 file_opts
= &options
->u
.file
;
2202 if (!file_opts
->has_nocow
) {
2203 file_opts
->nocow
= false;
2205 if (!file_opts
->has_preallocation
) {
2206 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2210 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2213 error_setg_errno(errp
, -result
, "Could not create file");
2217 /* Take permissions: We want to discard everything, so we need
2218 * BLK_PERM_WRITE; and truncation to the desired size requires
2220 * On the other hand, we cannot share the RESIZE permission
2221 * because we promise that after this function, the file has the
2222 * size given in the options. If someone else were to resize it
2223 * concurrently, we could not guarantee that.
2224 * Note that after this function, we can no longer guarantee that
2225 * the file is not touched by a third party, so it may be resized
2227 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2228 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2230 /* Step one: Take locks */
2231 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2236 /* Step two: Check that nobody else has taken conflicting locks */
2237 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2239 error_append_hint(errp
,
2240 "Is another process using the image [%s]?\n",
2241 file_opts
->filename
);
2245 /* Clear the file by truncating it to 0 */
2246 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2251 if (file_opts
->nocow
) {
2253 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2254 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2255 * will be ignored since any failure of this operation should not
2256 * block the left work.
2259 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2260 attr
|= FS_NOCOW_FL
;
2261 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2266 /* Resize and potentially preallocate the file to the desired
2268 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2269 file_opts
->preallocation
, errp
);
2275 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2277 /* The above call should not fail, and if it does, that does
2278 * not mean the whole creation operation has failed. So
2279 * report it the user for their convenience, but do not report
2280 * it to the caller. */
2281 warn_report_err(local_err
);
2285 if (qemu_close(fd
) != 0 && result
== 0) {
2287 error_setg_errno(errp
, -result
, "Could not close the new file");
2293 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2296 BlockdevCreateOptions options
;
2297 int64_t total_size
= 0;
2299 PreallocMode prealloc
;
2301 Error
*local_err
= NULL
;
2303 /* Skip file: protocol prefix */
2304 strstart(filename
, "file:", &filename
);
2306 /* Read out options */
2307 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2309 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2310 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2311 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2312 PREALLOC_MODE_OFF
, &local_err
);
2315 error_propagate(errp
, local_err
);
2319 options
= (BlockdevCreateOptions
) {
2320 .driver
= BLOCKDEV_DRIVER_FILE
,
2322 .filename
= (char *) filename
,
2324 .has_preallocation
= true,
2325 .preallocation
= prealloc
,
2330 return raw_co_create(&options
, errp
);
2334 * Find allocation range in @bs around offset @start.
2335 * May change underlying file descriptor's file offset.
2336 * If @start is not in a hole, store @start in @data, and the
2337 * beginning of the next hole in @hole, and return 0.
2338 * If @start is in a non-trailing hole, store @start in @hole and the
2339 * beginning of the next non-hole in @data, and return 0.
2340 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2341 * If we can't find out, return a negative errno other than -ENXIO.
2343 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2344 off_t
*data
, off_t
*hole
)
2346 #if defined SEEK_HOLE && defined SEEK_DATA
2347 BDRVRawState
*s
= bs
->opaque
;
2352 * D1. offs == start: start is in data
2353 * D2. offs > start: start is in a hole, next data at offs
2354 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2355 * or start is beyond EOF
2356 * If the latter happens, the file has been truncated behind
2357 * our back since we opened it. All bets are off then.
2358 * Treating like a trailing hole is simplest.
2359 * D4. offs < 0, errno != ENXIO: we learned nothing
2361 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2363 return -errno
; /* D3 or D4 */
2367 /* This is not a valid return by lseek(). We are safe to just return
2368 * -EIO in this case, and we'll treat it like D4. */
2373 /* D2: in hole, next data at offs */
2379 /* D1: in data, end not yet known */
2383 * H1. offs == start: start is in a hole
2384 * If this happens here, a hole has been dug behind our back
2385 * since the previous lseek().
2386 * H2. offs > start: either start is in data, next hole at offs,
2387 * or start is in trailing hole, EOF at offs
2388 * Linux treats trailing holes like any other hole: offs ==
2389 * start. Solaris seeks to EOF instead: offs > start (blech).
2390 * If that happens here, a hole has been dug behind our back
2391 * since the previous lseek().
2392 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2393 * If this happens, the file has been truncated behind our
2394 * back since we opened it. Treat it like a trailing hole.
2395 * H4. offs < 0, errno != ENXIO: we learned nothing
2396 * Pretend we know nothing at all, i.e. "forget" about D1.
2398 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2400 return -errno
; /* D1 and (H3 or H4) */
2404 /* This is not a valid return by lseek(). We are safe to just return
2405 * -EIO in this case, and we'll treat it like H4. */
2411 * D1 and H2: either in data, next hole at offs, or it was in
2412 * data but is now in a trailing hole. In the latter case,
2413 * all bets are off. Treating it as if it there was data all
2414 * the way to EOF is safe, so simply do that.
2429 * Returns the allocation status of the specified offset.
2431 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2433 * 'pnum' is set to the number of bytes (including and immediately following
2434 * the specified offset) that are known to be in the same
2435 * allocated/unallocated state.
2437 * 'bytes' is the max value 'pnum' should be set to.
2439 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2442 int64_t bytes
, int64_t *pnum
,
2444 BlockDriverState
**file
)
2446 off_t data
= 0, hole
= 0;
2458 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2461 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2462 if (ret
== -ENXIO
) {
2465 ret
= BDRV_BLOCK_ZERO
;
2466 } else if (ret
< 0) {
2467 /* No info available, so pretend there are no holes */
2469 ret
= BDRV_BLOCK_DATA
;
2470 } else if (data
== offset
) {
2471 /* On a data extent, compute bytes to the end of the extent,
2472 * possibly including a partial sector at EOF. */
2473 *pnum
= MIN(bytes
, hole
- offset
);
2474 ret
= BDRV_BLOCK_DATA
;
2476 /* On a hole, compute bytes to the beginning of the next extent. */
2477 assert(hole
== offset
);
2478 *pnum
= MIN(bytes
, data
- offset
);
2479 ret
= BDRV_BLOCK_ZERO
;
2483 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2486 #if defined(__linux__)
2487 /* Verify that the file is not in the page cache */
2488 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2490 const size_t window_size
= 128 * 1024 * 1024;
2491 BDRVRawState
*s
= bs
->opaque
;
2492 void *window
= NULL
;
2499 /* mincore(2) page status information requires 1 byte per page */
2500 page_size
= sysconf(_SC_PAGESIZE
);
2501 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2503 end
= raw_getlength(bs
);
2505 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2512 /* Unmap previous window if size has changed */
2513 new_length
= MIN(end
- offset
, window_size
);
2514 if (new_length
!= length
) {
2515 munmap(window
, length
);
2520 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2522 if (new_window
== MAP_FAILED
) {
2523 error_setg_errno(errp
, errno
, "mmap failed");
2527 window
= new_window
;
2528 length
= new_length
;
2530 ret
= mincore(window
, length
, vec
);
2532 error_setg_errno(errp
, errno
, "mincore failed");
2536 vec_end
= DIV_ROUND_UP(length
, page_size
);
2537 for (i
= 0; i
< vec_end
; i
++) {
2539 error_setg(errp
, "page cache still in use!");
2546 munmap(window
, length
);
2551 #endif /* __linux__ */
2553 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2556 BDRVRawState
*s
= bs
->opaque
;
2561 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2565 if (s
->open_flags
& O_DIRECT
) {
2566 return; /* No host kernel page cache */
2569 #if defined(__linux__)
2570 /* This sets the scene for the next syscall... */
2571 ret
= bdrv_co_flush(bs
);
2573 error_setg_errno(errp
, -ret
, "flush failed");
2577 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2578 * process. These limitations are okay because we just fsynced the file,
2579 * we don't use mmap, and the file should not be in use by other processes.
2581 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2582 if (ret
!= 0) { /* the return value is a positive errno */
2583 error_setg_errno(errp
, ret
, "fadvise failed");
2587 if (s
->check_cache_dropped
) {
2588 check_cache_dropped(bs
, errp
);
2590 #else /* __linux__ */
2591 /* Do nothing. Live migration to a remote host with cache.direct=off is
2592 * unsupported on other host operating systems. Cache consistency issues
2593 * may occur but no error is reported here, partly because that's the
2594 * historical behavior and partly because it's hard to differentiate valid
2595 * configurations that should not cause errors.
2597 #endif /* !__linux__ */
2600 static coroutine_fn
int
2601 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2603 BDRVRawState
*s
= bs
->opaque
;
2604 RawPosixAIOData acb
;
2606 acb
= (RawPosixAIOData
) {
2608 .aio_fildes
= s
->fd
,
2609 .aio_type
= QEMU_AIO_DISCARD
,
2610 .aio_offset
= offset
,
2611 .aio_nbytes
= bytes
,
2615 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2618 return raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2621 static coroutine_fn
int
2622 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2624 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2627 static int coroutine_fn
2628 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2629 BdrvRequestFlags flags
, bool blkdev
)
2631 BDRVRawState
*s
= bs
->opaque
;
2632 RawPosixAIOData acb
;
2633 ThreadPoolFunc
*handler
;
2635 acb
= (RawPosixAIOData
) {
2637 .aio_fildes
= s
->fd
,
2638 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2639 .aio_offset
= offset
,
2640 .aio_nbytes
= bytes
,
2644 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2647 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2648 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2649 handler
= handle_aiocb_write_zeroes_unmap
;
2651 handler
= handle_aiocb_write_zeroes
;
2654 return raw_thread_pool_submit(bs
, handler
, &acb
);
2657 static int coroutine_fn
raw_co_pwrite_zeroes(
2658 BlockDriverState
*bs
, int64_t offset
,
2659 int bytes
, BdrvRequestFlags flags
)
2661 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2664 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2666 BDRVRawState
*s
= bs
->opaque
;
2668 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2672 static QemuOptsList raw_create_opts
= {
2673 .name
= "raw-create-opts",
2674 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2677 .name
= BLOCK_OPT_SIZE
,
2678 .type
= QEMU_OPT_SIZE
,
2679 .help
= "Virtual disk size"
2682 .name
= BLOCK_OPT_NOCOW
,
2683 .type
= QEMU_OPT_BOOL
,
2684 .help
= "Turn off copy-on-write (valid only on btrfs)"
2687 .name
= BLOCK_OPT_PREALLOC
,
2688 .type
= QEMU_OPT_STRING
,
2689 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2691 { /* end of list */ }
2695 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2698 BDRVRawState
*s
= bs
->opaque
;
2699 BDRVRawReopenState
*rs
= NULL
;
2703 if (s
->perm_change_fd
) {
2705 * In the context of reopen, this function may be called several times
2706 * (directly and recursively while change permissions of the parent).
2707 * This is even true for children that don't inherit from the original
2708 * reopen node, so s->reopen_state is not set.
2710 * Ignore all but the first call.
2715 if (s
->reopen_state
) {
2716 /* We already have a new file descriptor to set permissions for */
2717 assert(s
->reopen_state
->perm
== perm
);
2718 assert(s
->reopen_state
->shared_perm
== shared
);
2719 rs
= s
->reopen_state
->opaque
;
2720 s
->perm_change_fd
= rs
->fd
;
2722 /* We may need a new fd if auto-read-only switches the mode */
2723 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2727 } else if (ret
!= s
->fd
) {
2728 s
->perm_change_fd
= ret
;
2732 /* Prepare permissions on old fd to avoid conflicts between old and new,
2733 * but keep everything locked that new will need. */
2734 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2739 /* Copy locks to the new fd */
2740 if (s
->perm_change_fd
) {
2741 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2744 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2751 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2752 qemu_close(s
->perm_change_fd
);
2754 s
->perm_change_fd
= 0;
2758 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2760 BDRVRawState
*s
= bs
->opaque
;
2762 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2763 * called after .bdrv_reopen_commit) */
2764 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
2766 s
->fd
= s
->perm_change_fd
;
2768 s
->perm_change_fd
= 0;
2770 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2772 s
->shared_perm
= shared
;
2775 static void raw_abort_perm_update(BlockDriverState
*bs
)
2777 BDRVRawState
*s
= bs
->opaque
;
2779 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
2780 * the file descriptor. */
2781 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2782 qemu_close(s
->perm_change_fd
);
2784 s
->perm_change_fd
= 0;
2786 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2789 static int coroutine_fn
raw_co_copy_range_from(
2790 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2791 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2792 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2794 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2795 read_flags
, write_flags
);
2798 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2800 uint64_t src_offset
,
2802 uint64_t dst_offset
,
2804 BdrvRequestFlags read_flags
,
2805 BdrvRequestFlags write_flags
)
2807 RawPosixAIOData acb
;
2808 BDRVRawState
*s
= bs
->opaque
;
2809 BDRVRawState
*src_s
;
2811 assert(dst
->bs
== bs
);
2812 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2816 src_s
= src
->bs
->opaque
;
2817 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2821 acb
= (RawPosixAIOData
) {
2823 .aio_type
= QEMU_AIO_COPY_RANGE
,
2824 .aio_fildes
= src_s
->fd
,
2825 .aio_offset
= src_offset
,
2826 .aio_nbytes
= bytes
,
2829 .aio_offset2
= dst_offset
,
2833 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2836 BlockDriver bdrv_file
= {
2837 .format_name
= "file",
2838 .protocol_name
= "file",
2839 .instance_size
= sizeof(BDRVRawState
),
2840 .bdrv_needs_filename
= true,
2841 .bdrv_probe
= NULL
, /* no probe for protocols */
2842 .bdrv_parse_filename
= raw_parse_filename
,
2843 .bdrv_file_open
= raw_open
,
2844 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2845 .bdrv_reopen_commit
= raw_reopen_commit
,
2846 .bdrv_reopen_abort
= raw_reopen_abort
,
2847 .bdrv_close
= raw_close
,
2848 .bdrv_co_create
= raw_co_create
,
2849 .bdrv_co_create_opts
= raw_co_create_opts
,
2850 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2851 .bdrv_co_block_status
= raw_co_block_status
,
2852 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2853 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2855 .bdrv_co_preadv
= raw_co_preadv
,
2856 .bdrv_co_pwritev
= raw_co_pwritev
,
2857 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2858 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2859 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2860 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2861 .bdrv_refresh_limits
= raw_refresh_limits
,
2862 .bdrv_io_plug
= raw_aio_plug
,
2863 .bdrv_io_unplug
= raw_aio_unplug
,
2864 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2866 .bdrv_co_truncate
= raw_co_truncate
,
2867 .bdrv_getlength
= raw_getlength
,
2868 .bdrv_get_info
= raw_get_info
,
2869 .bdrv_get_allocated_file_size
2870 = raw_get_allocated_file_size
,
2871 .bdrv_check_perm
= raw_check_perm
,
2872 .bdrv_set_perm
= raw_set_perm
,
2873 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2874 .create_opts
= &raw_create_opts
,
2875 .mutable_opts
= mutable_opts
,
2878 /***********************************************/
2881 #if defined(__APPLE__) && defined(__MACH__)
2882 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2883 CFIndex maxPathSize
, int flags
);
2884 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2886 kern_return_t kernResult
= KERN_FAILURE
;
2887 mach_port_t masterPort
;
2888 CFMutableDictionaryRef classesToMatch
;
2889 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2890 char *mediaType
= NULL
;
2892 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2893 if ( KERN_SUCCESS
!= kernResult
) {
2894 printf( "IOMasterPort returned %d\n", kernResult
);
2898 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2899 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2900 if (classesToMatch
== NULL
) {
2901 error_report("IOServiceMatching returned NULL for %s",
2902 matching_array
[index
]);
2905 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2907 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2909 if (kernResult
!= KERN_SUCCESS
) {
2910 error_report("Note: IOServiceGetMatchingServices returned %d",
2915 /* If a match was found, leave the loop */
2916 if (*mediaIterator
!= 0) {
2917 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
2918 mediaType
= g_strdup(matching_array
[index
]);
2925 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2926 CFIndex maxPathSize
, int flags
)
2928 io_object_t nextMedia
;
2929 kern_return_t kernResult
= KERN_FAILURE
;
2931 nextMedia
= IOIteratorNext( mediaIterator
);
2934 CFTypeRef bsdPathAsCFString
;
2935 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2936 if ( bsdPathAsCFString
) {
2937 size_t devPathLength
;
2938 strcpy( bsdPath
, _PATH_DEV
);
2939 if (flags
& BDRV_O_NOCACHE
) {
2940 strcat(bsdPath
, "r");
2942 devPathLength
= strlen( bsdPath
);
2943 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2944 kernResult
= KERN_SUCCESS
;
2946 CFRelease( bsdPathAsCFString
);
2948 IOObjectRelease( nextMedia
);
2954 /* Sets up a real cdrom for use in QEMU */
2955 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2957 int index
, num_of_test_partitions
= 2, fd
;
2958 char test_partition
[MAXPATHLEN
];
2959 bool partition_found
= false;
2961 /* look for a working partition */
2962 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2963 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2965 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2967 partition_found
= true;
2973 /* if a working partition on the device was not found */
2974 if (partition_found
== false) {
2975 error_setg(errp
, "Failed to find a working partition on disc");
2977 trace_file_setup_cdrom(test_partition
);
2978 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2980 return partition_found
;
2983 /* Prints directions on mounting and unmounting a device */
2984 static void print_unmounting_directions(const char *file_name
)
2986 error_report("If device %s is mounted on the desktop, unmount"
2987 " it first before using it in QEMU", file_name
);
2988 error_report("Command to unmount device: diskutil unmountDisk %s",
2990 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2993 #endif /* defined(__APPLE__) && defined(__MACH__) */
2995 static int hdev_probe_device(const char *filename
)
2999 /* allow a dedicated CD-ROM driver to match with a higher priority */
3000 if (strstart(filename
, "/dev/cdrom", NULL
))
3003 if (stat(filename
, &st
) >= 0 &&
3004 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3011 static int check_hdev_writable(BDRVRawState
*s
)
3013 #if defined(BLKROGET)
3014 /* Linux block devices can be configured "read-only" using blockdev(8).
3015 * This is independent of device node permissions and therefore open(2)
3016 * with O_RDWR succeeds. Actual writes fail with EPERM.
3018 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3019 * check for read-only block devices so that Linux block devices behave
3025 if (fstat(s
->fd
, &st
)) {
3029 if (!S_ISBLK(st
.st_mode
)) {
3033 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3040 #endif /* defined(BLKROGET) */
3044 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3047 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3050 static bool hdev_is_sg(BlockDriverState
*bs
)
3053 #if defined(__linux__)
3055 BDRVRawState
*s
= bs
->opaque
;
3057 struct sg_scsi_id scsiid
;
3061 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3065 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3070 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3072 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3081 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3084 BDRVRawState
*s
= bs
->opaque
;
3085 Error
*local_err
= NULL
;
3088 #if defined(__APPLE__) && defined(__MACH__)
3090 * Caution: while qdict_get_str() is fine, getting non-string types
3091 * would require more care. When @options come from -blockdev or
3092 * blockdev_add, its members are typed according to the QAPI
3093 * schema, but when they come from -drive, they're all QString.
3095 const char *filename
= qdict_get_str(options
, "filename");
3096 char bsd_path
[MAXPATHLEN
] = "";
3097 bool error_occurred
= false;
3099 /* If using a real cdrom */
3100 if (strcmp(filename
, "/dev/cdrom") == 0) {
3101 char *mediaType
= NULL
;
3102 kern_return_t ret_val
;
3103 io_iterator_t mediaIterator
= 0;
3105 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3106 if (mediaType
== NULL
) {
3107 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3109 error_occurred
= true;
3110 goto hdev_open_Mac_error
;
3113 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3114 if (ret_val
!= KERN_SUCCESS
) {
3115 error_setg(errp
, "Could not get BSD path for optical drive");
3116 error_occurred
= true;
3117 goto hdev_open_Mac_error
;
3120 /* If a real optical drive was not found */
3121 if (bsd_path
[0] == '\0') {
3122 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3123 error_occurred
= true;
3124 goto hdev_open_Mac_error
;
3127 /* If using a cdrom disc and finding a partition on the disc failed */
3128 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3129 setup_cdrom(bsd_path
, errp
) == false) {
3130 print_unmounting_directions(bsd_path
);
3131 error_occurred
= true;
3132 goto hdev_open_Mac_error
;
3135 qdict_put_str(options
, "filename", bsd_path
);
3137 hdev_open_Mac_error
:
3139 if (mediaIterator
) {
3140 IOObjectRelease(mediaIterator
);
3142 if (error_occurred
) {
3146 #endif /* defined(__APPLE__) && defined(__MACH__) */
3148 s
->type
= FTYPE_FILE
;
3150 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3152 error_propagate(errp
, local_err
);
3153 #if defined(__APPLE__) && defined(__MACH__)
3155 filename
= bsd_path
;
3157 /* if a physical device experienced an error while being opened */
3158 if (strncmp(filename
, "/dev/", 5) == 0) {
3159 print_unmounting_directions(filename
);
3161 #endif /* defined(__APPLE__) && defined(__MACH__) */
3165 /* Since this does ioctl the device must be already opened */
3166 bs
->sg
= hdev_is_sg(bs
);
3168 if (flags
& BDRV_O_RDWR
) {
3169 ret
= check_hdev_writable(s
);
3172 error_setg_errno(errp
, -ret
, "The device is not writable");
3180 #if defined(__linux__)
3181 static int coroutine_fn
3182 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3184 BDRVRawState
*s
= bs
->opaque
;
3185 RawPosixAIOData acb
;
3193 if (req
== SG_IO
&& s
->pr_mgr
) {
3194 struct sg_io_hdr
*io_hdr
= buf
;
3195 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3196 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3197 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3202 acb
= (RawPosixAIOData
) {
3204 .aio_type
= QEMU_AIO_IOCTL
,
3205 .aio_fildes
= s
->fd
,
3213 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3217 static int fd_open(BlockDriverState
*bs
)
3219 BDRVRawState
*s
= bs
->opaque
;
3221 /* this is just to ensure s->fd is sane (its called by io ops) */
3227 static coroutine_fn
int
3228 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3236 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3239 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3240 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3249 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3252 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3257 struct stat stat_buf
;
3258 int64_t total_size
= 0;
3261 /* This function is used by both protocol block drivers and therefore either
3262 * of these prefixes may be given.
3263 * The return value has to be stored somewhere, otherwise this is an error
3264 * due to -Werror=unused-value. */
3266 strstart(filename
, "host_device:", &filename
) ||
3267 strstart(filename
, "host_cdrom:" , &filename
);
3271 ret
= raw_normalize_devicepath(&filename
, errp
);
3276 /* Read out options */
3277 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3280 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3283 error_setg_errno(errp
, -ret
, "Could not open device");
3287 if (fstat(fd
, &stat_buf
) < 0) {
3289 error_setg_errno(errp
, -ret
, "Could not stat device");
3290 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3292 "The given file is neither a block nor a character device");
3294 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3295 error_setg(errp
, "Device is too small");
3299 if (!ret
&& total_size
) {
3300 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3301 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3302 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3305 ret
= qemu_write_full(fd
, buf
, zero_size
);
3306 ret
= ret
== zero_size
? 0 : -errno
;
3313 static BlockDriver bdrv_host_device
= {
3314 .format_name
= "host_device",
3315 .protocol_name
= "host_device",
3316 .instance_size
= sizeof(BDRVRawState
),
3317 .bdrv_needs_filename
= true,
3318 .bdrv_probe_device
= hdev_probe_device
,
3319 .bdrv_parse_filename
= hdev_parse_filename
,
3320 .bdrv_file_open
= hdev_open
,
3321 .bdrv_close
= raw_close
,
3322 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3323 .bdrv_reopen_commit
= raw_reopen_commit
,
3324 .bdrv_reopen_abort
= raw_reopen_abort
,
3325 .bdrv_co_create_opts
= hdev_co_create_opts
,
3326 .create_opts
= &raw_create_opts
,
3327 .mutable_opts
= mutable_opts
,
3328 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3329 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3331 .bdrv_co_preadv
= raw_co_preadv
,
3332 .bdrv_co_pwritev
= raw_co_pwritev
,
3333 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3334 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3335 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3336 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3337 .bdrv_refresh_limits
= raw_refresh_limits
,
3338 .bdrv_io_plug
= raw_aio_plug
,
3339 .bdrv_io_unplug
= raw_aio_unplug
,
3340 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3342 .bdrv_co_truncate
= raw_co_truncate
,
3343 .bdrv_getlength
= raw_getlength
,
3344 .bdrv_get_info
= raw_get_info
,
3345 .bdrv_get_allocated_file_size
3346 = raw_get_allocated_file_size
,
3347 .bdrv_check_perm
= raw_check_perm
,
3348 .bdrv_set_perm
= raw_set_perm
,
3349 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3350 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3351 .bdrv_probe_geometry
= hdev_probe_geometry
,
3353 /* generic scsi device */
3355 .bdrv_co_ioctl
= hdev_co_ioctl
,
3359 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3360 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3363 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3368 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3371 BDRVRawState
*s
= bs
->opaque
;
3375 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3376 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3379 static int cdrom_probe_device(const char *filename
)
3385 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3389 ret
= fstat(fd
, &st
);
3390 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3394 /* Attempt to detect via a CDROM specific ioctl */
3395 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3405 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3407 BDRVRawState
*s
= bs
->opaque
;
3410 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3411 return ret
== CDS_DISC_OK
;
3414 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3416 BDRVRawState
*s
= bs
->opaque
;
3419 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3420 perror("CDROMEJECT");
3422 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3423 perror("CDROMEJECT");
3427 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3429 BDRVRawState
*s
= bs
->opaque
;
3431 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3433 * Note: an error can happen if the distribution automatically
3436 /* perror("CDROM_LOCKDOOR"); */
3440 static BlockDriver bdrv_host_cdrom
= {
3441 .format_name
= "host_cdrom",
3442 .protocol_name
= "host_cdrom",
3443 .instance_size
= sizeof(BDRVRawState
),
3444 .bdrv_needs_filename
= true,
3445 .bdrv_probe_device
= cdrom_probe_device
,
3446 .bdrv_parse_filename
= cdrom_parse_filename
,
3447 .bdrv_file_open
= cdrom_open
,
3448 .bdrv_close
= raw_close
,
3449 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3450 .bdrv_reopen_commit
= raw_reopen_commit
,
3451 .bdrv_reopen_abort
= raw_reopen_abort
,
3452 .bdrv_co_create_opts
= hdev_co_create_opts
,
3453 .create_opts
= &raw_create_opts
,
3454 .mutable_opts
= mutable_opts
,
3455 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3458 .bdrv_co_preadv
= raw_co_preadv
,
3459 .bdrv_co_pwritev
= raw_co_pwritev
,
3460 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3461 .bdrv_refresh_limits
= raw_refresh_limits
,
3462 .bdrv_io_plug
= raw_aio_plug
,
3463 .bdrv_io_unplug
= raw_aio_unplug
,
3464 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3466 .bdrv_co_truncate
= raw_co_truncate
,
3467 .bdrv_getlength
= raw_getlength
,
3468 .has_variable_length
= true,
3469 .bdrv_get_allocated_file_size
3470 = raw_get_allocated_file_size
,
3472 /* removable device support */
3473 .bdrv_is_inserted
= cdrom_is_inserted
,
3474 .bdrv_eject
= cdrom_eject
,
3475 .bdrv_lock_medium
= cdrom_lock_medium
,
3477 /* generic scsi device */
3478 .bdrv_co_ioctl
= hdev_co_ioctl
,
3480 #endif /* __linux__ */
3482 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3483 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3486 BDRVRawState
*s
= bs
->opaque
;
3487 Error
*local_err
= NULL
;
3492 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3494 error_propagate(errp
, local_err
);
3498 /* make sure the door isn't locked at this time */
3499 ioctl(s
->fd
, CDIOCALLOW
);
3503 static int cdrom_probe_device(const char *filename
)
3505 if (strstart(filename
, "/dev/cd", NULL
) ||
3506 strstart(filename
, "/dev/acd", NULL
))
3511 static int cdrom_reopen(BlockDriverState
*bs
)
3513 BDRVRawState
*s
= bs
->opaque
;
3517 * Force reread of possibly changed/newly loaded disc,
3518 * FreeBSD seems to not notice sometimes...
3522 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3529 /* make sure the door isn't locked at this time */
3530 ioctl(s
->fd
, CDIOCALLOW
);
3534 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3536 return raw_getlength(bs
) > 0;
3539 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3541 BDRVRawState
*s
= bs
->opaque
;
3546 (void) ioctl(s
->fd
, CDIOCALLOW
);
3549 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3550 perror("CDIOCEJECT");
3552 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3553 perror("CDIOCCLOSE");
3559 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3561 BDRVRawState
*s
= bs
->opaque
;
3565 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3567 * Note: an error can happen if the distribution automatically
3570 /* perror("CDROM_LOCKDOOR"); */
3574 static BlockDriver bdrv_host_cdrom
= {
3575 .format_name
= "host_cdrom",
3576 .protocol_name
= "host_cdrom",
3577 .instance_size
= sizeof(BDRVRawState
),
3578 .bdrv_needs_filename
= true,
3579 .bdrv_probe_device
= cdrom_probe_device
,
3580 .bdrv_parse_filename
= cdrom_parse_filename
,
3581 .bdrv_file_open
= cdrom_open
,
3582 .bdrv_close
= raw_close
,
3583 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3584 .bdrv_reopen_commit
= raw_reopen_commit
,
3585 .bdrv_reopen_abort
= raw_reopen_abort
,
3586 .bdrv_co_create_opts
= hdev_co_create_opts
,
3587 .create_opts
= &raw_create_opts
,
3588 .mutable_opts
= mutable_opts
,
3590 .bdrv_co_preadv
= raw_co_preadv
,
3591 .bdrv_co_pwritev
= raw_co_pwritev
,
3592 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3593 .bdrv_refresh_limits
= raw_refresh_limits
,
3594 .bdrv_io_plug
= raw_aio_plug
,
3595 .bdrv_io_unplug
= raw_aio_unplug
,
3596 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3598 .bdrv_co_truncate
= raw_co_truncate
,
3599 .bdrv_getlength
= raw_getlength
,
3600 .has_variable_length
= true,
3601 .bdrv_get_allocated_file_size
3602 = raw_get_allocated_file_size
,
3604 /* removable device support */
3605 .bdrv_is_inserted
= cdrom_is_inserted
,
3606 .bdrv_eject
= cdrom_eject
,
3607 .bdrv_lock_medium
= cdrom_lock_medium
,
3609 #endif /* __FreeBSD__ */
3611 static void bdrv_file_init(void)
3614 * Register all the drivers. Note that order is important, the driver
3615 * registered last will get probed first.
3617 bdrv_register(&bdrv_file
);
3618 bdrv_register(&bdrv_host_device
);
3620 bdrv_register(&bdrv_host_cdrom
);
3622 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3623 bdrv_register(&bdrv_host_cdrom
);
3627 block_init(bdrv_file_init
);