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
{
150 /* The current permissions. */
152 uint64_t shared_perm
;
154 /* The perms bits whose corresponding bytes are already locked in
156 uint64_t locked_perm
;
157 uint64_t locked_shared_perm
;
163 bool has_write_zeroes
:1;
164 bool discard_zeroes
:1;
165 bool use_linux_aio
:1;
166 bool page_cache_inconsistent
:1;
168 bool needs_alignment
;
169 bool check_cache_dropped
;
174 typedef struct BDRVRawReopenState
{
177 bool check_cache_dropped
;
178 } BDRVRawReopenState
;
180 static int fd_open(BlockDriverState
*bs
);
181 static int64_t raw_getlength(BlockDriverState
*bs
);
183 typedef struct RawPosixAIOData
{
184 BlockDriverState
*bs
;
205 PreallocMode prealloc
;
211 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
212 static int cdrom_reopen(BlockDriverState
*bs
);
215 #if defined(__NetBSD__)
216 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
218 static char namebuf
[PATH_MAX
];
219 const char *dp
, *fname
;
223 dp
= strrchr(fname
, '/');
224 if (lstat(fname
, &sb
) < 0) {
225 error_setg_errno(errp
, errno
, "%s: stat failed", fname
);
229 if (!S_ISBLK(sb
.st_mode
)) {
234 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
236 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
237 (int)(dp
- fname
), fname
, dp
+ 1);
240 warn_report("%s is a block device, using %s", fname
, *filename
);
245 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
252 * Get logical block size via ioctl. On success store it in @sector_size_p.
254 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
256 unsigned int sector_size
;
257 bool success
= false;
261 static const unsigned long ioctl_list
[] = {
265 #ifdef DKIOCGETBLOCKSIZE
268 #ifdef DIOCGSECTORSIZE
273 /* Try a few ioctls to get the right size */
274 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
275 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
276 *sector_size_p
= sector_size
;
281 return success
? 0 : -errno
;
285 * Get physical block size of @fd.
286 * On success, store it in @blk_size and return 0.
287 * On failure, return -errno.
289 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
292 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
301 /* Check if read is allowed with given memory buffer and length.
303 * This function is used to check O_DIRECT memory buffer and request alignment.
305 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
307 ssize_t ret
= pread(fd
, buf
, len
, 0);
314 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
315 * other errors (e.g. real I/O error), which could happen on a failed
316 * drive, since we only care about probing alignment.
318 if (errno
!= EINVAL
) {
326 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
328 BDRVRawState
*s
= bs
->opaque
;
330 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
332 /* For SCSI generic devices the alignment is not really used.
333 With buffered I/O, we don't have any restrictions. */
334 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
335 bs
->bl
.request_alignment
= 1;
340 bs
->bl
.request_alignment
= 0;
342 /* Let's try to use the logical blocksize for the alignment. */
343 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
344 bs
->bl
.request_alignment
= 0;
349 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
350 bs
->bl
.request_alignment
= da
.d_miniosz
;
351 /* The kernel returns wrong information for d_mem */
352 /* s->buf_align = da.d_mem; */
357 /* If we could not get the sizes so far, we can only guess them */
360 buf
= qemu_memalign(max_align
, 2 * max_align
);
361 for (align
= 512; align
<= max_align
; align
<<= 1) {
362 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
363 s
->buf_align
= align
;
370 if (!bs
->bl
.request_alignment
) {
372 buf
= qemu_memalign(s
->buf_align
, max_align
);
373 for (align
= 512; align
<= max_align
; align
<<= 1) {
374 if (raw_is_io_aligned(fd
, buf
, align
)) {
375 bs
->bl
.request_alignment
= align
;
382 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
383 error_setg(errp
, "Could not find working O_DIRECT alignment");
384 error_append_hint(errp
, "Try cache.direct=off\n");
388 static void raw_parse_flags(int bdrv_flags
, int *open_flags
)
390 assert(open_flags
!= NULL
);
392 *open_flags
|= O_BINARY
;
393 *open_flags
&= ~O_ACCMODE
;
394 if (bdrv_flags
& BDRV_O_RDWR
) {
395 *open_flags
|= O_RDWR
;
397 *open_flags
|= O_RDONLY
;
400 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
401 * and O_DIRECT for no caching. */
402 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
403 *open_flags
|= O_DIRECT
;
407 static void raw_parse_filename(const char *filename
, QDict
*options
,
410 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
413 static QemuOptsList raw_runtime_opts
= {
415 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
419 .type
= QEMU_OPT_STRING
,
420 .help
= "File name of the image",
424 .type
= QEMU_OPT_STRING
,
425 .help
= "host AIO implementation (threads, native)",
429 .type
= QEMU_OPT_STRING
,
430 .help
= "file locking mode (on/off/auto, default: auto)",
433 .name
= "pr-manager",
434 .type
= QEMU_OPT_STRING
,
435 .help
= "id of persistent reservation manager object (default: none)",
438 .name
= "x-check-cache-dropped",
439 .type
= QEMU_OPT_BOOL
,
440 .help
= "check that page cache was dropped on live migration (default: off)"
442 { /* end of list */ }
446 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
447 int bdrv_flags
, int open_flags
,
448 bool device
, Error
**errp
)
450 BDRVRawState
*s
= bs
->opaque
;
452 Error
*local_err
= NULL
;
453 const char *filename
= NULL
;
455 BlockdevAioOptions aio
, aio_default
;
460 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
461 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
463 error_propagate(errp
, local_err
);
468 filename
= qemu_opt_get(opts
, "filename");
470 ret
= raw_normalize_devicepath(&filename
, errp
);
475 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
476 ? BLOCKDEV_AIO_OPTIONS_NATIVE
477 : BLOCKDEV_AIO_OPTIONS_THREADS
;
478 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
479 qemu_opt_get(opts
, "aio"),
480 aio_default
, &local_err
);
482 error_propagate(errp
, local_err
);
486 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
488 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
489 qemu_opt_get(opts
, "locking"),
490 ON_OFF_AUTO_AUTO
, &local_err
);
492 error_propagate(errp
, local_err
);
499 if (!qemu_has_ofd_lock()) {
500 warn_report("File lock requested but OFD locking syscall is "
501 "unavailable, falling back to POSIX file locks");
502 error_printf("Due to the implementation, locks can be lost "
506 case ON_OFF_AUTO_OFF
:
509 case ON_OFF_AUTO_AUTO
:
510 s
->use_lock
= qemu_has_ofd_lock();
516 str
= qemu_opt_get(opts
, "pr-manager");
518 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
520 error_propagate(errp
, local_err
);
526 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
529 s
->open_flags
= open_flags
;
530 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
533 fd
= qemu_open(filename
, s
->open_flags
, 0644);
534 ret
= fd
< 0 ? -errno
: 0;
536 if (ret
== -EACCES
|| ret
== -EROFS
) {
537 /* Try to degrade to read-only, but if it doesn't work, still use the
538 * normal error message. */
539 if (bdrv_apply_auto_read_only(bs
, NULL
, NULL
) == 0) {
540 bdrv_flags
&= ~BDRV_O_RDWR
;
541 raw_parse_flags(bdrv_flags
, &s
->open_flags
);
542 assert(!(s
->open_flags
& O_CREAT
));
543 fd
= qemu_open(filename
, s
->open_flags
);
544 ret
= fd
< 0 ? -errno
: 0;
549 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
558 s
->shared_perm
= BLK_PERM_ALL
;
560 #ifdef CONFIG_LINUX_AIO
561 /* Currently Linux does AIO only for files opened with O_DIRECT */
562 if (s
->use_linux_aio
) {
563 if (!(s
->open_flags
& O_DIRECT
)) {
564 error_setg(errp
, "aio=native was specified, but it requires "
565 "cache.direct=on, which was not specified.");
569 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
570 error_prepend(errp
, "Unable to use native AIO: ");
575 if (s
->use_linux_aio
) {
576 error_setg(errp
, "aio=native was specified, but is not supported "
581 #endif /* !defined(CONFIG_LINUX_AIO) */
583 s
->has_discard
= true;
584 s
->has_write_zeroes
= true;
585 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
586 s
->needs_alignment
= true;
589 if (fstat(s
->fd
, &st
) < 0) {
591 error_setg_errno(errp
, errno
, "Could not stat file");
596 if (S_ISBLK(st
.st_mode
)) {
597 warn_report("Opening a block device as a file using the '%s' "
598 "driver is deprecated", bs
->drv
->format_name
);
599 } else if (S_ISCHR(st
.st_mode
)) {
600 warn_report("Opening a character device as a file using the '%s' "
601 "driver is deprecated", bs
->drv
->format_name
);
602 } else if (!S_ISREG(st
.st_mode
)) {
603 error_setg(errp
, "A regular file was expected by the '%s' driver, "
604 "but something else was given", bs
->drv
->format_name
);
608 s
->discard_zeroes
= true;
609 s
->has_fallocate
= true;
612 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
613 error_setg(errp
, "'%s' driver expects either "
614 "a character or block device", bs
->drv
->format_name
);
620 if (S_ISBLK(st
.st_mode
)) {
621 #ifdef BLKDISCARDZEROES
623 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
624 s
->discard_zeroes
= true;
628 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
629 * not rely on the contents of discarded blocks unless using O_DIRECT.
630 * Same for BLKZEROOUT.
632 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
633 s
->discard_zeroes
= false;
634 s
->has_write_zeroes
= false;
639 if (S_ISCHR(st
.st_mode
)) {
641 * The file is a char device (disk), which on FreeBSD isn't behind
642 * a pager, so force all requests to be aligned. This is needed
643 * so QEMU makes sure all IO operations on the device are aligned
644 * to sector size, or else FreeBSD will reject them with EINVAL.
646 s
->needs_alignment
= true;
651 if (platform_test_xfs_fd(s
->fd
)) {
656 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
659 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
666 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
669 BDRVRawState
*s
= bs
->opaque
;
671 s
->type
= FTYPE_FILE
;
672 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
681 #define PERM_FOREACH(i) \
682 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
684 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
685 * file; if @unlock == true, also unlock the unneeded bytes.
686 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
688 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
689 uint64_t perm_lock_bits
,
690 uint64_t shared_perm_lock_bits
,
691 bool unlock
, Error
**errp
)
695 uint64_t locked_perm
, locked_shared_perm
;
698 locked_perm
= s
->locked_perm
;
699 locked_shared_perm
= s
->locked_shared_perm
;
702 * We don't have the previous bits, just lock/unlock for each of the
706 locked_perm
= BLK_PERM_ALL
;
707 locked_shared_perm
= BLK_PERM_ALL
;
710 locked_shared_perm
= 0;
715 int off
= RAW_LOCK_PERM_BASE
+ i
;
716 uint64_t bit
= (1ULL << i
);
717 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
718 ret
= qemu_lock_fd(fd
, off
, 1, false);
720 error_setg(errp
, "Failed to lock byte %d", off
);
723 s
->locked_perm
|= bit
;
725 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
726 ret
= qemu_unlock_fd(fd
, off
, 1);
728 error_setg(errp
, "Failed to unlock byte %d", off
);
731 s
->locked_perm
&= ~bit
;
736 int off
= RAW_LOCK_SHARED_BASE
+ i
;
737 uint64_t bit
= (1ULL << i
);
738 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
739 ret
= qemu_lock_fd(fd
, off
, 1, false);
741 error_setg(errp
, "Failed to lock byte %d", off
);
744 s
->locked_shared_perm
|= bit
;
746 } else if (unlock
&& (locked_shared_perm
& bit
) &&
747 !(shared_perm_lock_bits
& bit
)) {
748 ret
= qemu_unlock_fd(fd
, off
, 1);
750 error_setg(errp
, "Failed to unlock byte %d", off
);
753 s
->locked_shared_perm
&= ~bit
;
760 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
761 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
768 int off
= RAW_LOCK_SHARED_BASE
+ i
;
769 uint64_t p
= 1ULL << i
;
771 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
773 char *perm_name
= bdrv_perm_names(p
);
775 "Failed to get \"%s\" lock",
783 int off
= RAW_LOCK_PERM_BASE
+ i
;
784 uint64_t p
= 1ULL << i
;
785 if (!(shared_perm
& p
)) {
786 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
788 char *perm_name
= bdrv_perm_names(p
);
790 "Failed to get shared \"%s\" lock",
800 static int raw_handle_perm_lock(BlockDriverState
*bs
,
802 uint64_t new_perm
, uint64_t new_shared
,
805 BDRVRawState
*s
= bs
->opaque
;
807 Error
*local_err
= NULL
;
813 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
819 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
820 ~s
->shared_perm
| ~new_shared
,
823 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
827 error_append_hint(errp
,
828 "Is another process using the image [%s]?\n",
832 /* fall through to unlock bytes. */
834 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
837 /* Theoretically the above call only unlocks bytes and it cannot
838 * fail. Something weird happened, report it.
840 warn_report_err(local_err
);
844 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
847 /* Theoretically the above call only unlocks bytes and it cannot
848 * fail. Something weird happened, report it.
850 warn_report_err(local_err
);
857 static int raw_reopen_prepare(BDRVReopenState
*state
,
858 BlockReopenQueue
*queue
, Error
**errp
)
861 BDRVRawReopenState
*rs
;
864 Error
*local_err
= NULL
;
866 assert(state
!= NULL
);
867 assert(state
->bs
!= NULL
);
869 s
= state
->bs
->opaque
;
871 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
875 /* Handle options changes */
876 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
877 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
879 error_propagate(errp
, local_err
);
884 rs
->check_cache_dropped
=
885 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
887 /* This driver's reopen function doesn't currently allow changing
888 * other options, so let's put them back in the original QDict and
889 * bdrv_reopen_prepare() will detect changes and complain. */
890 qemu_opts_to_qdict(opts
, state
->options
);
892 if (s
->type
== FTYPE_CD
) {
893 rs
->open_flags
|= O_NONBLOCK
;
896 raw_parse_flags(state
->flags
, &rs
->open_flags
);
898 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
900 fcntl_flags
|= O_NOATIME
;
904 /* Not all operating systems have O_ASYNC, and those that don't
905 * will not let us track the state into rs->open_flags (typically
906 * you achieve the same effect with an ioctl, for example I_SETSIG
907 * on Solaris). But we do not use O_ASYNC, so that's fine.
909 assert((s
->open_flags
& O_ASYNC
) == 0);
912 if ((rs
->open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
913 /* dup the original fd */
914 rs
->fd
= qemu_dup(s
->fd
);
916 ret
= fcntl_setfl(rs
->fd
, rs
->open_flags
);
924 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
926 const char *normalized_filename
= state
->bs
->filename
;
927 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
929 assert(!(rs
->open_flags
& O_CREAT
));
930 rs
->fd
= qemu_open(normalized_filename
, rs
->open_flags
);
932 error_setg_errno(errp
, errno
, "Could not reopen file");
938 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
939 * alignment with the new fd. */
941 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
945 error_propagate(errp
, local_err
);
955 static void raw_reopen_commit(BDRVReopenState
*state
)
957 BDRVRawReopenState
*rs
= state
->opaque
;
958 BDRVRawState
*s
= state
->bs
->opaque
;
959 Error
*local_err
= NULL
;
961 s
->check_cache_dropped
= rs
->check_cache_dropped
;
962 s
->open_flags
= rs
->open_flags
;
964 /* Copy locks to the new fd before closing the old one. */
965 raw_apply_lock_bytes(NULL
, rs
->fd
, s
->locked_perm
,
966 s
->locked_shared_perm
, false, &local_err
);
968 /* shouldn't fail in a sane host, but report it just in case. */
969 error_report_err(local_err
);
974 g_free(state
->opaque
);
975 state
->opaque
= NULL
;
979 static void raw_reopen_abort(BDRVReopenState
*state
)
981 BDRVRawReopenState
*rs
= state
->opaque
;
983 /* nothing to do if NULL, we didn't get far enough */
992 g_free(state
->opaque
);
993 state
->opaque
= NULL
;
996 static int hdev_get_max_transfer_length(BlockDriverState
*bs
, int fd
)
1000 short max_sectors
= 0;
1001 if (bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1003 } else if (!bs
->sg
&& ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1004 return max_sectors
<< BDRV_SECTOR_BITS
;
1013 static int hdev_get_max_segments(const struct stat
*st
)
1023 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1024 major(st
->st_rdev
), minor(st
->st_rdev
));
1025 fd
= open(sysfspath
, O_RDONLY
);
1031 ret
= read(fd
, buf
, sizeof(buf
) - 1);
1032 } while (ret
== -1 && errno
== EINTR
);
1036 } else if (ret
== 0) {
1041 /* The file is ended with '\n', pass 'end' to accept that. */
1042 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1043 if (ret
== 0 && end
&& *end
== '\n') {
1058 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1060 BDRVRawState
*s
= bs
->opaque
;
1063 if (!fstat(s
->fd
, &st
)) {
1064 if (S_ISBLK(st
.st_mode
) || S_ISCHR(st
.st_mode
)) {
1065 int ret
= hdev_get_max_transfer_length(bs
, s
->fd
);
1066 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1067 bs
->bl
.max_transfer
= pow2floor(ret
);
1069 ret
= hdev_get_max_segments(&st
);
1071 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1072 ret
* getpagesize());
1077 raw_probe_alignment(bs
, s
->fd
, errp
);
1078 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1079 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1082 static int check_for_dasd(int fd
)
1085 struct dasd_information2_t info
= {0};
1087 return ioctl(fd
, BIODASDINFO2
, &info
);
1094 * Try to get @bs's logical and physical block size.
1095 * On success, store them in @bsz and return zero.
1096 * On failure, return negative errno.
1098 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1100 BDRVRawState
*s
= bs
->opaque
;
1103 /* If DASD, get blocksizes */
1104 if (check_for_dasd(s
->fd
) < 0) {
1107 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1111 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1115 * Try to get @bs's geometry: cyls, heads, sectors.
1116 * On success, store them in @geo and return 0.
1117 * On failure return -errno.
1118 * (Allows block driver to assign default geometry values that guest sees)
1121 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1123 BDRVRawState
*s
= bs
->opaque
;
1124 struct hd_geometry ioctl_geo
= {0};
1126 /* If DASD, get its geometry */
1127 if (check_for_dasd(s
->fd
) < 0) {
1130 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1133 /* HDIO_GETGEO may return success even though geo contains zeros
1134 (e.g. certain multipath setups) */
1135 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1138 /* Do not return a geometry for partition */
1139 if (ioctl_geo
.start
!= 0) {
1142 geo
->heads
= ioctl_geo
.heads
;
1143 geo
->sectors
= ioctl_geo
.sectors
;
1144 geo
->cylinders
= ioctl_geo
.cylinders
;
1148 #else /* __linux__ */
1149 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1155 static ssize_t
handle_aiocb_ioctl(RawPosixAIOData
*aiocb
)
1159 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1167 static ssize_t
handle_aiocb_flush(RawPosixAIOData
*aiocb
)
1169 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1172 if (s
->page_cache_inconsistent
) {
1176 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1178 /* There is no clear definition of the semantics of a failing fsync(),
1179 * so we may have to assume the worst. The sad truth is that this
1180 * assumption is correct for Linux. Some pages are now probably marked
1181 * clean in the page cache even though they are inconsistent with the
1182 * on-disk contents. The next fdatasync() call would succeed, but no
1183 * further writeback attempt will be made. We can't get back to a state
1184 * in which we know what is on disk (we would have to rewrite
1185 * everything that was touched since the last fdatasync() at least), so
1186 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1187 * really defined, I have little hope that other OSes are doing better.
1189 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1191 if ((s
->open_flags
& O_DIRECT
) == 0) {
1192 s
->page_cache_inconsistent
= true;
1199 #ifdef CONFIG_PREADV
1201 static bool preadv_present
= true;
1204 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1206 return preadv(fd
, iov
, nr_iov
, offset
);
1210 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1212 return pwritev(fd
, iov
, nr_iov
, offset
);
1217 static bool preadv_present
= false;
1220 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1226 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1233 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1238 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1239 len
= qemu_pwritev(aiocb
->aio_fildes
,
1244 len
= qemu_preadv(aiocb
->aio_fildes
,
1248 } while (len
== -1 && errno
== EINTR
);
1257 * Read/writes the data to/from a given linear buffer.
1259 * Returns the number of bytes handles or -errno in case of an error. Short
1260 * reads are only returned if the end of the file is reached.
1262 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1267 while (offset
< aiocb
->aio_nbytes
) {
1268 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1269 len
= pwrite(aiocb
->aio_fildes
,
1270 (const char *)buf
+ offset
,
1271 aiocb
->aio_nbytes
- offset
,
1272 aiocb
->aio_offset
+ offset
);
1274 len
= pread(aiocb
->aio_fildes
,
1276 aiocb
->aio_nbytes
- offset
,
1277 aiocb
->aio_offset
+ offset
);
1279 if (len
== -1 && errno
== EINTR
) {
1281 } else if (len
== -1 && errno
== EINVAL
&&
1282 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1283 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1285 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1286 * after a short read. Assume that O_DIRECT short reads only occur
1287 * at EOF. Therefore this is a short read, not an I/O error.
1290 } else if (len
== -1) {
1293 } else if (len
== 0) {
1302 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1307 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1309 * If there is just a single buffer, and it is properly aligned
1310 * we can just use plain pread/pwrite without any problems.
1312 if (aiocb
->io
.niov
== 1) {
1313 return handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1316 * We have more than one iovec, and all are properly aligned.
1318 * Try preadv/pwritev first and fall back to linearizing the
1319 * buffer if it's not supported.
1321 if (preadv_present
) {
1322 nbytes
= handle_aiocb_rw_vector(aiocb
);
1323 if (nbytes
== aiocb
->aio_nbytes
||
1324 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1327 preadv_present
= false;
1331 * XXX(hch): short read/write. no easy way to handle the reminder
1332 * using these interfaces. For now retry using plain
1338 * Ok, we have to do it the hard way, copy all segments into
1339 * a single aligned buffer.
1341 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1346 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1350 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1351 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1352 p
+= aiocb
->io
.iov
[i
].iov_len
;
1354 assert(p
- buf
== aiocb
->aio_nbytes
);
1357 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1358 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1360 size_t count
= aiocb
->aio_nbytes
, copy
;
1363 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1365 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1366 copy
= aiocb
->io
.iov
[i
].iov_len
;
1368 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1369 assert(count
>= copy
);
1381 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1383 struct xfs_flock64 fl
;
1386 memset(&fl
, 0, sizeof(fl
));
1387 fl
.l_whence
= SEEK_SET
;
1388 fl
.l_start
= offset
;
1391 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1393 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1400 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1402 struct xfs_flock64 fl
;
1405 memset(&fl
, 0, sizeof(fl
));
1406 fl
.l_whence
= SEEK_SET
;
1407 fl
.l_start
= offset
;
1410 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1412 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1420 static int translate_err(int err
)
1422 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1429 #ifdef CONFIG_FALLOCATE
1430 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1433 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1436 } while (errno
== EINTR
);
1437 return translate_err(-errno
);
1441 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1444 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1446 if (!s
->has_write_zeroes
) {
1452 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1453 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1456 } while (errno
== EINTR
);
1458 ret
= translate_err(-errno
);
1461 if (ret
== -ENOTSUP
) {
1462 s
->has_write_zeroes
= false;
1467 static int handle_aiocb_write_zeroes(void *opaque
)
1469 RawPosixAIOData
*aiocb
= opaque
;
1470 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1471 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1473 #ifdef CONFIG_FALLOCATE
1477 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1478 return handle_aiocb_write_zeroes_block(aiocb
);
1483 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1487 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1488 if (s
->has_write_zeroes
) {
1489 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1490 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1491 if (ret
== 0 || ret
!= -ENOTSUP
) {
1494 s
->has_write_zeroes
= false;
1498 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1499 if (s
->has_discard
&& s
->has_fallocate
) {
1500 int ret
= do_fallocate(s
->fd
,
1501 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1502 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1504 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1505 if (ret
== 0 || ret
!= -ENOTSUP
) {
1508 s
->has_fallocate
= false;
1509 } else if (ret
!= -ENOTSUP
) {
1512 s
->has_discard
= false;
1517 #ifdef CONFIG_FALLOCATE
1518 /* Last resort: we are trying to extend the file with zeroed data. This
1519 * can be done via fallocate(fd, 0) */
1520 len
= bdrv_getlength(aiocb
->bs
);
1521 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1522 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1523 if (ret
== 0 || ret
!= -ENOTSUP
) {
1526 s
->has_fallocate
= false;
1533 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1535 RawPosixAIOData
*aiocb
= opaque
;
1536 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1539 /* First try to write zeros and unmap at the same time */
1541 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1542 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1543 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1544 if (ret
!= -ENOTSUP
) {
1551 /* xfs_discard() guarantees that the discarded area reads as all-zero
1552 * afterwards, so we can use it here. */
1553 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1557 /* If we couldn't manage to unmap while guaranteed that the area reads as
1558 * all-zero afterwards, just write zeroes without unmapping */
1559 ret
= handle_aiocb_write_zeroes(aiocb
);
1563 #ifndef HAVE_COPY_FILE_RANGE
1564 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1565 off_t
*out_off
, size_t len
, unsigned int flags
)
1567 #ifdef __NR_copy_file_range
1568 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1569 out_off
, len
, flags
);
1577 static int handle_aiocb_copy_range(void *opaque
)
1579 RawPosixAIOData
*aiocb
= opaque
;
1580 uint64_t bytes
= aiocb
->aio_nbytes
;
1581 off_t in_off
= aiocb
->aio_offset
;
1582 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1585 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1586 aiocb
->copy_range
.aio_fd2
, &out_off
,
1588 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1589 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1592 /* No progress (e.g. when beyond EOF), let the caller fall back to
1611 static ssize_t
handle_aiocb_discard(RawPosixAIOData
*aiocb
)
1613 int ret
= -EOPNOTSUPP
;
1614 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1616 if (!s
->has_discard
) {
1620 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1623 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1624 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1627 } while (errno
== EINTR
);
1634 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1638 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1639 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1640 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1644 ret
= translate_err(ret
);
1645 if (ret
== -ENOTSUP
) {
1646 s
->has_discard
= false;
1651 static int handle_aiocb_truncate(void *opaque
)
1653 RawPosixAIOData
*aiocb
= opaque
;
1655 int64_t current_length
= 0;
1658 int fd
= aiocb
->aio_fildes
;
1659 int64_t offset
= aiocb
->aio_offset
;
1660 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1661 Error
**errp
= aiocb
->truncate
.errp
;
1663 if (fstat(fd
, &st
) < 0) {
1665 error_setg_errno(errp
, -result
, "Could not stat file");
1669 current_length
= st
.st_size
;
1670 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1671 error_setg(errp
, "Cannot use preallocation for shrinking files");
1676 #ifdef CONFIG_POSIX_FALLOCATE
1677 case PREALLOC_MODE_FALLOC
:
1679 * Truncating before posix_fallocate() makes it about twice slower on
1680 * file systems that do not support fallocate(), trying to check if a
1681 * block is allocated before allocating it, so don't do that here.
1683 if (offset
!= current_length
) {
1684 result
= -posix_fallocate(fd
, current_length
,
1685 offset
- current_length
);
1687 /* posix_fallocate() doesn't set errno. */
1688 error_setg_errno(errp
, -result
,
1689 "Could not preallocate new data");
1696 case PREALLOC_MODE_FULL
:
1698 int64_t num
= 0, left
= offset
- current_length
;
1702 * Knowing the final size from the beginning could allow the file
1703 * system driver to do less allocations and possibly avoid
1704 * fragmentation of the file.
1706 if (ftruncate(fd
, offset
) != 0) {
1708 error_setg_errno(errp
, -result
, "Could not resize file");
1712 buf
= g_malloc0(65536);
1714 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1715 if (seek_result
< 0) {
1717 error_setg_errno(errp
, -result
,
1718 "Failed to seek to the old end of file");
1723 num
= MIN(left
, 65536);
1724 result
= write(fd
, buf
, num
);
1726 if (errno
== EINTR
) {
1730 error_setg_errno(errp
, -result
,
1731 "Could not write zeros for preallocation");
1740 error_setg_errno(errp
, -result
,
1741 "Could not flush file to disk");
1747 case PREALLOC_MODE_OFF
:
1748 if (ftruncate(fd
, offset
) != 0) {
1750 error_setg_errno(errp
, -result
, "Could not resize file");
1755 error_setg(errp
, "Unsupported preallocation mode: %s",
1756 PreallocMode_str(prealloc
));
1762 if (ftruncate(fd
, current_length
) < 0) {
1763 error_report("Failed to restore old file length: %s",
1772 static int aio_worker(void *arg
)
1774 RawPosixAIOData
*aiocb
= arg
;
1777 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1779 ret
= handle_aiocb_rw(aiocb
);
1780 if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1781 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, ret
,
1782 0, aiocb
->aio_nbytes
- ret
);
1784 ret
= aiocb
->aio_nbytes
;
1786 if (ret
== aiocb
->aio_nbytes
) {
1788 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1792 case QEMU_AIO_WRITE
:
1793 ret
= handle_aiocb_rw(aiocb
);
1794 if (ret
== aiocb
->aio_nbytes
) {
1796 } else if (ret
>= 0 && ret
< aiocb
->aio_nbytes
) {
1800 case QEMU_AIO_FLUSH
:
1801 ret
= handle_aiocb_flush(aiocb
);
1803 case QEMU_AIO_IOCTL
:
1804 ret
= handle_aiocb_ioctl(aiocb
);
1806 case QEMU_AIO_DISCARD
:
1807 ret
= handle_aiocb_discard(aiocb
);
1809 case QEMU_AIO_WRITE_ZEROES
:
1810 case QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_DISCARD
:
1811 case QEMU_AIO_COPY_RANGE
:
1812 case QEMU_AIO_TRUNCATE
:
1813 g_assert_not_reached();
1815 error_report("invalid aio request (0x%x)", aiocb
->aio_type
);
1824 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1825 ThreadPoolFunc func
, void *arg
)
1827 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1828 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1829 return thread_pool_submit_co(pool
, func
, arg
);
1832 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1833 int64_t offset
, int fd2
, int64_t offset2
,
1835 int bytes
, int type
)
1837 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1840 acb
->aio_type
= type
;
1841 acb
->aio_fildes
= fd
;
1843 acb
->aio_nbytes
= bytes
;
1844 acb
->aio_offset
= offset
;
1847 acb
->io
.iov
= qiov
->iov
;
1848 acb
->io
.niov
= qiov
->niov
;
1849 assert(qiov
->size
== bytes
);
1851 acb
->copy_range
.aio_fd2
= fd2
;
1852 acb
->copy_range
.aio_offset2
= offset2
;
1855 trace_file_paio_submit_co(offset
, bytes
, type
);
1856 return raw_thread_pool_submit(bs
, aio_worker
, acb
);
1859 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1860 int64_t offset
, QEMUIOVector
*qiov
,
1861 int bytes
, int type
)
1863 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1866 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1867 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1869 BDRVRawState
*s
= bs
->opaque
;
1871 if (fd_open(bs
) < 0)
1875 * Check if the underlying device requires requests to be aligned,
1876 * and if the request we are trying to submit is aligned or not.
1877 * If this is the case tell the low-level driver that it needs
1878 * to copy the buffer.
1880 if (s
->needs_alignment
) {
1881 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1882 type
|= QEMU_AIO_MISALIGNED
;
1883 #ifdef CONFIG_LINUX_AIO
1884 } else if (s
->use_linux_aio
) {
1885 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1886 assert(qiov
->size
== bytes
);
1887 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1892 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1895 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1896 uint64_t bytes
, QEMUIOVector
*qiov
,
1899 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1902 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1903 uint64_t bytes
, QEMUIOVector
*qiov
,
1907 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1910 static void raw_aio_plug(BlockDriverState
*bs
)
1912 #ifdef CONFIG_LINUX_AIO
1913 BDRVRawState
*s
= bs
->opaque
;
1914 if (s
->use_linux_aio
) {
1915 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1916 laio_io_plug(bs
, aio
);
1921 static void raw_aio_unplug(BlockDriverState
*bs
)
1923 #ifdef CONFIG_LINUX_AIO
1924 BDRVRawState
*s
= bs
->opaque
;
1925 if (s
->use_linux_aio
) {
1926 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1927 laio_io_unplug(bs
, aio
);
1932 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1934 BDRVRawState
*s
= bs
->opaque
;
1942 return paio_submit_co(bs
, s
->fd
, 0, NULL
, 0, QEMU_AIO_FLUSH
);
1945 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1946 AioContext
*new_context
)
1948 #ifdef CONFIG_LINUX_AIO
1949 BDRVRawState
*s
= bs
->opaque
;
1950 if (s
->use_linux_aio
) {
1952 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1953 error_reportf_err(local_err
, "Unable to use native AIO, "
1954 "falling back to thread pool: ");
1955 s
->use_linux_aio
= false;
1961 static void raw_close(BlockDriverState
*bs
)
1963 BDRVRawState
*s
= bs
->opaque
;
1972 * Truncates the given regular file @fd to @offset and, when growing, fills the
1973 * new space according to @prealloc.
1975 * Returns: 0 on success, -errno on failure.
1977 static int coroutine_fn
1978 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1979 PreallocMode prealloc
, Error
**errp
)
1981 RawPosixAIOData acb
;
1983 acb
= (RawPosixAIOData
) {
1986 .aio_type
= QEMU_AIO_TRUNCATE
,
1987 .aio_offset
= offset
,
1989 .prealloc
= prealloc
,
1994 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
1997 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1998 PreallocMode prealloc
, Error
**errp
)
2000 BDRVRawState
*s
= bs
->opaque
;
2004 if (fstat(s
->fd
, &st
)) {
2006 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2010 if (S_ISREG(st
.st_mode
)) {
2011 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2014 if (prealloc
!= PREALLOC_MODE_OFF
) {
2015 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2016 "non-regular file", PreallocMode_str(prealloc
));
2020 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2021 if (offset
> raw_getlength(bs
)) {
2022 error_setg(errp
, "Cannot grow device files");
2026 error_setg(errp
, "Resizing this file is not supported");
2034 static int64_t raw_getlength(BlockDriverState
*bs
)
2036 BDRVRawState
*s
= bs
->opaque
;
2042 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
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
;
2052 #elif defined(__NetBSD__)
2053 static int64_t raw_getlength(BlockDriverState
*bs
)
2055 BDRVRawState
*s
= bs
->opaque
;
2061 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2062 struct dkwedge_info dkw
;
2064 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2065 return dkw
.dkw_size
* 512;
2067 struct disklabel dl
;
2069 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2071 return (uint64_t)dl
.d_secsize
*
2072 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2077 #elif defined(__sun__)
2078 static int64_t raw_getlength(BlockDriverState
*bs
)
2080 BDRVRawState
*s
= bs
->opaque
;
2081 struct dk_minfo minfo
;
2091 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2093 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2095 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2099 * There are reports that lseek on some devices fails, but
2100 * irc discussion said that contingency on contingency was overkill.
2102 size
= lseek(s
->fd
, 0, SEEK_END
);
2108 #elif defined(CONFIG_BSD)
2109 static int64_t raw_getlength(BlockDriverState
*bs
)
2111 BDRVRawState
*s
= bs
->opaque
;
2115 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2124 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2127 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2128 #ifdef DIOCGMEDIASIZE
2129 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2130 #elif defined(DIOCGPART)
2133 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2134 size
= pi
.media_size
;
2140 #if defined(__APPLE__) && defined(__MACH__)
2142 uint64_t sectors
= 0;
2143 uint32_t sector_size
= 0;
2145 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2146 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2147 size
= sectors
* sector_size
;
2149 size
= lseek(fd
, 0LL, SEEK_END
);
2156 size
= lseek(fd
, 0LL, SEEK_END
);
2161 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2164 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2165 if (size
== 2048LL * (unsigned)-1)
2167 /* XXX no disc? maybe we need to reopen... */
2168 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2175 size
= lseek(fd
, 0, SEEK_END
);
2183 static int64_t raw_getlength(BlockDriverState
*bs
)
2185 BDRVRawState
*s
= bs
->opaque
;
2194 size
= lseek(s
->fd
, 0, SEEK_END
);
2202 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2205 BDRVRawState
*s
= bs
->opaque
;
2207 if (fstat(s
->fd
, &st
) < 0) {
2210 return (int64_t)st
.st_blocks
* 512;
2213 static int coroutine_fn
2214 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2216 BlockdevCreateOptionsFile
*file_opts
;
2217 Error
*local_err
= NULL
;
2219 uint64_t perm
, shared
;
2222 /* Validate options and set default values */
2223 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2224 file_opts
= &options
->u
.file
;
2226 if (!file_opts
->has_nocow
) {
2227 file_opts
->nocow
= false;
2229 if (!file_opts
->has_preallocation
) {
2230 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2234 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2237 error_setg_errno(errp
, -result
, "Could not create file");
2241 /* Take permissions: We want to discard everything, so we need
2242 * BLK_PERM_WRITE; and truncation to the desired size requires
2244 * On the other hand, we cannot share the RESIZE permission
2245 * because we promise that after this function, the file has the
2246 * size given in the options. If someone else were to resize it
2247 * concurrently, we could not guarantee that.
2248 * Note that after this function, we can no longer guarantee that
2249 * the file is not touched by a third party, so it may be resized
2251 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2252 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2254 /* Step one: Take locks */
2255 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2260 /* Step two: Check that nobody else has taken conflicting locks */
2261 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2263 error_append_hint(errp
,
2264 "Is another process using the image [%s]?\n",
2265 file_opts
->filename
);
2269 /* Clear the file by truncating it to 0 */
2270 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2275 if (file_opts
->nocow
) {
2277 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2278 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2279 * will be ignored since any failure of this operation should not
2280 * block the left work.
2283 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2284 attr
|= FS_NOCOW_FL
;
2285 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2290 /* Resize and potentially preallocate the file to the desired
2292 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2293 file_opts
->preallocation
, errp
);
2299 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2301 /* The above call should not fail, and if it does, that does
2302 * not mean the whole creation operation has failed. So
2303 * report it the user for their convenience, but do not report
2304 * it to the caller. */
2305 warn_report_err(local_err
);
2309 if (qemu_close(fd
) != 0 && result
== 0) {
2311 error_setg_errno(errp
, -result
, "Could not close the new file");
2317 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2320 BlockdevCreateOptions options
;
2321 int64_t total_size
= 0;
2323 PreallocMode prealloc
;
2325 Error
*local_err
= NULL
;
2327 /* Skip file: protocol prefix */
2328 strstart(filename
, "file:", &filename
);
2330 /* Read out options */
2331 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2333 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2334 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2335 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2336 PREALLOC_MODE_OFF
, &local_err
);
2339 error_propagate(errp
, local_err
);
2343 options
= (BlockdevCreateOptions
) {
2344 .driver
= BLOCKDEV_DRIVER_FILE
,
2346 .filename
= (char *) filename
,
2348 .has_preallocation
= true,
2349 .preallocation
= prealloc
,
2354 return raw_co_create(&options
, errp
);
2358 * Find allocation range in @bs around offset @start.
2359 * May change underlying file descriptor's file offset.
2360 * If @start is not in a hole, store @start in @data, and the
2361 * beginning of the next hole in @hole, and return 0.
2362 * If @start is in a non-trailing hole, store @start in @hole and the
2363 * beginning of the next non-hole in @data, and return 0.
2364 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2365 * If we can't find out, return a negative errno other than -ENXIO.
2367 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2368 off_t
*data
, off_t
*hole
)
2370 #if defined SEEK_HOLE && defined SEEK_DATA
2371 BDRVRawState
*s
= bs
->opaque
;
2376 * D1. offs == start: start is in data
2377 * D2. offs > start: start is in a hole, next data at offs
2378 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2379 * or start is beyond EOF
2380 * If the latter happens, the file has been truncated behind
2381 * our back since we opened it. All bets are off then.
2382 * Treating like a trailing hole is simplest.
2383 * D4. offs < 0, errno != ENXIO: we learned nothing
2385 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2387 return -errno
; /* D3 or D4 */
2391 /* This is not a valid return by lseek(). We are safe to just return
2392 * -EIO in this case, and we'll treat it like D4. */
2397 /* D2: in hole, next data at offs */
2403 /* D1: in data, end not yet known */
2407 * H1. offs == start: start is in a hole
2408 * If this happens here, a hole has been dug behind our back
2409 * since the previous lseek().
2410 * H2. offs > start: either start is in data, next hole at offs,
2411 * or start is in trailing hole, EOF at offs
2412 * Linux treats trailing holes like any other hole: offs ==
2413 * start. Solaris seeks to EOF instead: offs > start (blech).
2414 * If that happens here, a hole has been dug behind our back
2415 * since the previous lseek().
2416 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2417 * If this happens, the file has been truncated behind our
2418 * back since we opened it. Treat it like a trailing hole.
2419 * H4. offs < 0, errno != ENXIO: we learned nothing
2420 * Pretend we know nothing at all, i.e. "forget" about D1.
2422 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2424 return -errno
; /* D1 and (H3 or H4) */
2428 /* This is not a valid return by lseek(). We are safe to just return
2429 * -EIO in this case, and we'll treat it like H4. */
2435 * D1 and H2: either in data, next hole at offs, or it was in
2436 * data but is now in a trailing hole. In the latter case,
2437 * all bets are off. Treating it as if it there was data all
2438 * the way to EOF is safe, so simply do that.
2453 * Returns the allocation status of the specified offset.
2455 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2457 * 'pnum' is set to the number of bytes (including and immediately following
2458 * the specified offset) that are known to be in the same
2459 * allocated/unallocated state.
2461 * 'bytes' is the max value 'pnum' should be set to.
2463 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2466 int64_t bytes
, int64_t *pnum
,
2468 BlockDriverState
**file
)
2470 off_t data
= 0, hole
= 0;
2482 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2485 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2486 if (ret
== -ENXIO
) {
2489 ret
= BDRV_BLOCK_ZERO
;
2490 } else if (ret
< 0) {
2491 /* No info available, so pretend there are no holes */
2493 ret
= BDRV_BLOCK_DATA
;
2494 } else if (data
== offset
) {
2495 /* On a data extent, compute bytes to the end of the extent,
2496 * possibly including a partial sector at EOF. */
2497 *pnum
= MIN(bytes
, hole
- offset
);
2498 ret
= BDRV_BLOCK_DATA
;
2500 /* On a hole, compute bytes to the beginning of the next extent. */
2501 assert(hole
== offset
);
2502 *pnum
= MIN(bytes
, data
- offset
);
2503 ret
= BDRV_BLOCK_ZERO
;
2507 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2510 #if defined(__linux__)
2511 /* Verify that the file is not in the page cache */
2512 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2514 const size_t window_size
= 128 * 1024 * 1024;
2515 BDRVRawState
*s
= bs
->opaque
;
2516 void *window
= NULL
;
2523 /* mincore(2) page status information requires 1 byte per page */
2524 page_size
= sysconf(_SC_PAGESIZE
);
2525 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2527 end
= raw_getlength(bs
);
2529 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2536 /* Unmap previous window if size has changed */
2537 new_length
= MIN(end
- offset
, window_size
);
2538 if (new_length
!= length
) {
2539 munmap(window
, length
);
2544 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2546 if (new_window
== MAP_FAILED
) {
2547 error_setg_errno(errp
, errno
, "mmap failed");
2551 window
= new_window
;
2552 length
= new_length
;
2554 ret
= mincore(window
, length
, vec
);
2556 error_setg_errno(errp
, errno
, "mincore failed");
2560 vec_end
= DIV_ROUND_UP(length
, page_size
);
2561 for (i
= 0; i
< vec_end
; i
++) {
2563 error_setg(errp
, "page cache still in use!");
2570 munmap(window
, length
);
2575 #endif /* __linux__ */
2577 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2580 BDRVRawState
*s
= bs
->opaque
;
2585 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2589 if (s
->open_flags
& O_DIRECT
) {
2590 return; /* No host kernel page cache */
2593 #if defined(__linux__)
2594 /* This sets the scene for the next syscall... */
2595 ret
= bdrv_co_flush(bs
);
2597 error_setg_errno(errp
, -ret
, "flush failed");
2601 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2602 * process. These limitations are okay because we just fsynced the file,
2603 * we don't use mmap, and the file should not be in use by other processes.
2605 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2606 if (ret
!= 0) { /* the return value is a positive errno */
2607 error_setg_errno(errp
, ret
, "fadvise failed");
2611 if (s
->check_cache_dropped
) {
2612 check_cache_dropped(bs
, errp
);
2614 #else /* __linux__ */
2615 /* Do nothing. Live migration to a remote host with cache.direct=off is
2616 * unsupported on other host operating systems. Cache consistency issues
2617 * may occur but no error is reported here, partly because that's the
2618 * historical behavior and partly because it's hard to differentiate valid
2619 * configurations that should not cause errors.
2621 #endif /* !__linux__ */
2624 static coroutine_fn
int
2625 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2627 BDRVRawState
*s
= bs
->opaque
;
2629 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
, QEMU_AIO_DISCARD
);
2632 static int coroutine_fn
2633 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2634 BdrvRequestFlags flags
, bool blkdev
)
2636 BDRVRawState
*s
= bs
->opaque
;
2637 RawPosixAIOData acb
;
2638 ThreadPoolFunc
*handler
;
2640 acb
= (RawPosixAIOData
) {
2642 .aio_fildes
= s
->fd
,
2643 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2644 .aio_offset
= offset
,
2645 .aio_nbytes
= bytes
,
2649 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2652 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2653 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2654 handler
= handle_aiocb_write_zeroes_unmap
;
2656 handler
= handle_aiocb_write_zeroes
;
2659 return raw_thread_pool_submit(bs
, handler
, &acb
);
2662 static int coroutine_fn
raw_co_pwrite_zeroes(
2663 BlockDriverState
*bs
, int64_t offset
,
2664 int bytes
, BdrvRequestFlags flags
)
2666 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2669 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2671 BDRVRawState
*s
= bs
->opaque
;
2673 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2677 static QemuOptsList raw_create_opts
= {
2678 .name
= "raw-create-opts",
2679 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2682 .name
= BLOCK_OPT_SIZE
,
2683 .type
= QEMU_OPT_SIZE
,
2684 .help
= "Virtual disk size"
2687 .name
= BLOCK_OPT_NOCOW
,
2688 .type
= QEMU_OPT_BOOL
,
2689 .help
= "Turn off copy-on-write (valid only on btrfs)"
2692 .name
= BLOCK_OPT_PREALLOC
,
2693 .type
= QEMU_OPT_STRING
,
2694 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2696 { /* end of list */ }
2700 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2703 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2706 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2708 BDRVRawState
*s
= bs
->opaque
;
2709 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2711 s
->shared_perm
= shared
;
2714 static void raw_abort_perm_update(BlockDriverState
*bs
)
2716 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2719 static int coroutine_fn
raw_co_copy_range_from(
2720 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2721 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2722 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2724 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2725 read_flags
, write_flags
);
2728 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2730 uint64_t src_offset
,
2732 uint64_t dst_offset
,
2734 BdrvRequestFlags read_flags
,
2735 BdrvRequestFlags write_flags
)
2737 RawPosixAIOData acb
;
2738 BDRVRawState
*s
= bs
->opaque
;
2739 BDRVRawState
*src_s
;
2741 assert(dst
->bs
== bs
);
2742 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2746 src_s
= src
->bs
->opaque
;
2747 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2751 acb
= (RawPosixAIOData
) {
2753 .aio_type
= QEMU_AIO_COPY_RANGE
,
2754 .aio_fildes
= src_s
->fd
,
2755 .aio_offset
= src_offset
,
2756 .aio_nbytes
= bytes
,
2759 .aio_offset2
= dst_offset
,
2763 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2766 BlockDriver bdrv_file
= {
2767 .format_name
= "file",
2768 .protocol_name
= "file",
2769 .instance_size
= sizeof(BDRVRawState
),
2770 .bdrv_needs_filename
= true,
2771 .bdrv_probe
= NULL
, /* no probe for protocols */
2772 .bdrv_parse_filename
= raw_parse_filename
,
2773 .bdrv_file_open
= raw_open
,
2774 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2775 .bdrv_reopen_commit
= raw_reopen_commit
,
2776 .bdrv_reopen_abort
= raw_reopen_abort
,
2777 .bdrv_close
= raw_close
,
2778 .bdrv_co_create
= raw_co_create
,
2779 .bdrv_co_create_opts
= raw_co_create_opts
,
2780 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2781 .bdrv_co_block_status
= raw_co_block_status
,
2782 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2783 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2785 .bdrv_co_preadv
= raw_co_preadv
,
2786 .bdrv_co_pwritev
= raw_co_pwritev
,
2787 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2788 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2789 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2790 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2791 .bdrv_refresh_limits
= raw_refresh_limits
,
2792 .bdrv_io_plug
= raw_aio_plug
,
2793 .bdrv_io_unplug
= raw_aio_unplug
,
2794 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2796 .bdrv_co_truncate
= raw_co_truncate
,
2797 .bdrv_getlength
= raw_getlength
,
2798 .bdrv_get_info
= raw_get_info
,
2799 .bdrv_get_allocated_file_size
2800 = raw_get_allocated_file_size
,
2801 .bdrv_check_perm
= raw_check_perm
,
2802 .bdrv_set_perm
= raw_set_perm
,
2803 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2804 .create_opts
= &raw_create_opts
,
2807 /***********************************************/
2810 #if defined(__APPLE__) && defined(__MACH__)
2811 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2812 CFIndex maxPathSize
, int flags
);
2813 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2815 kern_return_t kernResult
= KERN_FAILURE
;
2816 mach_port_t masterPort
;
2817 CFMutableDictionaryRef classesToMatch
;
2818 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2819 char *mediaType
= NULL
;
2821 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2822 if ( KERN_SUCCESS
!= kernResult
) {
2823 printf( "IOMasterPort returned %d\n", kernResult
);
2827 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2828 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2829 if (classesToMatch
== NULL
) {
2830 error_report("IOServiceMatching returned NULL for %s",
2831 matching_array
[index
]);
2834 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2836 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2838 if (kernResult
!= KERN_SUCCESS
) {
2839 error_report("Note: IOServiceGetMatchingServices returned %d",
2844 /* If a match was found, leave the loop */
2845 if (*mediaIterator
!= 0) {
2846 DPRINTF("Matching using %s\n", matching_array
[index
]);
2847 mediaType
= g_strdup(matching_array
[index
]);
2854 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2855 CFIndex maxPathSize
, int flags
)
2857 io_object_t nextMedia
;
2858 kern_return_t kernResult
= KERN_FAILURE
;
2860 nextMedia
= IOIteratorNext( mediaIterator
);
2863 CFTypeRef bsdPathAsCFString
;
2864 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2865 if ( bsdPathAsCFString
) {
2866 size_t devPathLength
;
2867 strcpy( bsdPath
, _PATH_DEV
);
2868 if (flags
& BDRV_O_NOCACHE
) {
2869 strcat(bsdPath
, "r");
2871 devPathLength
= strlen( bsdPath
);
2872 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2873 kernResult
= KERN_SUCCESS
;
2875 CFRelease( bsdPathAsCFString
);
2877 IOObjectRelease( nextMedia
);
2883 /* Sets up a real cdrom for use in QEMU */
2884 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2886 int index
, num_of_test_partitions
= 2, fd
;
2887 char test_partition
[MAXPATHLEN
];
2888 bool partition_found
= false;
2890 /* look for a working partition */
2891 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2892 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2894 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2896 partition_found
= true;
2902 /* if a working partition on the device was not found */
2903 if (partition_found
== false) {
2904 error_setg(errp
, "Failed to find a working partition on disc");
2906 DPRINTF("Using %s as optical disc\n", test_partition
);
2907 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2909 return partition_found
;
2912 /* Prints directions on mounting and unmounting a device */
2913 static void print_unmounting_directions(const char *file_name
)
2915 error_report("If device %s is mounted on the desktop, unmount"
2916 " it first before using it in QEMU", file_name
);
2917 error_report("Command to unmount device: diskutil unmountDisk %s",
2919 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2922 #endif /* defined(__APPLE__) && defined(__MACH__) */
2924 static int hdev_probe_device(const char *filename
)
2928 /* allow a dedicated CD-ROM driver to match with a higher priority */
2929 if (strstart(filename
, "/dev/cdrom", NULL
))
2932 if (stat(filename
, &st
) >= 0 &&
2933 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2940 static int check_hdev_writable(BDRVRawState
*s
)
2942 #if defined(BLKROGET)
2943 /* Linux block devices can be configured "read-only" using blockdev(8).
2944 * This is independent of device node permissions and therefore open(2)
2945 * with O_RDWR succeeds. Actual writes fail with EPERM.
2947 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2948 * check for read-only block devices so that Linux block devices behave
2954 if (fstat(s
->fd
, &st
)) {
2958 if (!S_ISBLK(st
.st_mode
)) {
2962 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2969 #endif /* defined(BLKROGET) */
2973 static void hdev_parse_filename(const char *filename
, QDict
*options
,
2976 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
2979 static bool hdev_is_sg(BlockDriverState
*bs
)
2982 #if defined(__linux__)
2984 BDRVRawState
*s
= bs
->opaque
;
2986 struct sg_scsi_id scsiid
;
2990 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
2994 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
2999 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3001 DPRINTF("SG device found: type=%d, version=%d\n",
3002 scsiid
.scsi_type
, sg_version
);
3011 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3014 BDRVRawState
*s
= bs
->opaque
;
3015 Error
*local_err
= NULL
;
3018 #if defined(__APPLE__) && defined(__MACH__)
3020 * Caution: while qdict_get_str() is fine, getting non-string types
3021 * would require more care. When @options come from -blockdev or
3022 * blockdev_add, its members are typed according to the QAPI
3023 * schema, but when they come from -drive, they're all QString.
3025 const char *filename
= qdict_get_str(options
, "filename");
3026 char bsd_path
[MAXPATHLEN
] = "";
3027 bool error_occurred
= false;
3029 /* If using a real cdrom */
3030 if (strcmp(filename
, "/dev/cdrom") == 0) {
3031 char *mediaType
= NULL
;
3032 kern_return_t ret_val
;
3033 io_iterator_t mediaIterator
= 0;
3035 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3036 if (mediaType
== NULL
) {
3037 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3039 error_occurred
= true;
3040 goto hdev_open_Mac_error
;
3043 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3044 if (ret_val
!= KERN_SUCCESS
) {
3045 error_setg(errp
, "Could not get BSD path for optical drive");
3046 error_occurred
= true;
3047 goto hdev_open_Mac_error
;
3050 /* If a real optical drive was not found */
3051 if (bsd_path
[0] == '\0') {
3052 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3053 error_occurred
= true;
3054 goto hdev_open_Mac_error
;
3057 /* If using a cdrom disc and finding a partition on the disc failed */
3058 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3059 setup_cdrom(bsd_path
, errp
) == false) {
3060 print_unmounting_directions(bsd_path
);
3061 error_occurred
= true;
3062 goto hdev_open_Mac_error
;
3065 qdict_put_str(options
, "filename", bsd_path
);
3067 hdev_open_Mac_error
:
3069 if (mediaIterator
) {
3070 IOObjectRelease(mediaIterator
);
3072 if (error_occurred
) {
3076 #endif /* defined(__APPLE__) && defined(__MACH__) */
3078 s
->type
= FTYPE_FILE
;
3080 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3082 error_propagate(errp
, local_err
);
3083 #if defined(__APPLE__) && defined(__MACH__)
3085 filename
= bsd_path
;
3087 /* if a physical device experienced an error while being opened */
3088 if (strncmp(filename
, "/dev/", 5) == 0) {
3089 print_unmounting_directions(filename
);
3091 #endif /* defined(__APPLE__) && defined(__MACH__) */
3095 /* Since this does ioctl the device must be already opened */
3096 bs
->sg
= hdev_is_sg(bs
);
3098 if (flags
& BDRV_O_RDWR
) {
3099 ret
= check_hdev_writable(s
);
3102 error_setg_errno(errp
, -ret
, "The device is not writable");
3110 #if defined(__linux__)
3112 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
3113 unsigned long int req
, void *buf
,
3114 BlockCompletionFunc
*cb
, void *opaque
)
3116 BDRVRawState
*s
= bs
->opaque
;
3117 RawPosixAIOData
*acb
;
3120 if (fd_open(bs
) < 0)
3123 if (req
== SG_IO
&& s
->pr_mgr
) {
3124 struct sg_io_hdr
*io_hdr
= buf
;
3125 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3126 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3127 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3128 s
->fd
, io_hdr
, cb
, opaque
);
3132 acb
= g_new(RawPosixAIOData
, 1);
3134 acb
->aio_type
= QEMU_AIO_IOCTL
;
3135 acb
->aio_fildes
= s
->fd
;
3136 acb
->aio_offset
= 0;
3137 acb
->ioctl
.buf
= buf
;
3138 acb
->ioctl
.cmd
= req
;
3139 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3140 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
3144 static int fd_open(BlockDriverState
*bs
)
3146 BDRVRawState
*s
= bs
->opaque
;
3148 /* this is just to ensure s->fd is sane (its called by io ops) */
3154 static coroutine_fn
int
3155 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3157 BDRVRawState
*s
= bs
->opaque
;
3164 return paio_submit_co(bs
, s
->fd
, offset
, NULL
, bytes
,
3165 QEMU_AIO_DISCARD
| QEMU_AIO_BLKDEV
);
3168 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3169 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3178 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3181 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3186 struct stat stat_buf
;
3187 int64_t total_size
= 0;
3190 /* This function is used by both protocol block drivers and therefore either
3191 * of these prefixes may be given.
3192 * The return value has to be stored somewhere, otherwise this is an error
3193 * due to -Werror=unused-value. */
3195 strstart(filename
, "host_device:", &filename
) ||
3196 strstart(filename
, "host_cdrom:" , &filename
);
3200 ret
= raw_normalize_devicepath(&filename
, errp
);
3205 /* Read out options */
3206 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3209 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3212 error_setg_errno(errp
, -ret
, "Could not open device");
3216 if (fstat(fd
, &stat_buf
) < 0) {
3218 error_setg_errno(errp
, -ret
, "Could not stat device");
3219 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3221 "The given file is neither a block nor a character device");
3223 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3224 error_setg(errp
, "Device is too small");
3228 if (!ret
&& total_size
) {
3229 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3230 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3231 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3234 ret
= qemu_write_full(fd
, buf
, zero_size
);
3235 ret
= ret
== zero_size
? 0 : -errno
;
3242 static BlockDriver bdrv_host_device
= {
3243 .format_name
= "host_device",
3244 .protocol_name
= "host_device",
3245 .instance_size
= sizeof(BDRVRawState
),
3246 .bdrv_needs_filename
= true,
3247 .bdrv_probe_device
= hdev_probe_device
,
3248 .bdrv_parse_filename
= hdev_parse_filename
,
3249 .bdrv_file_open
= hdev_open
,
3250 .bdrv_close
= raw_close
,
3251 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3252 .bdrv_reopen_commit
= raw_reopen_commit
,
3253 .bdrv_reopen_abort
= raw_reopen_abort
,
3254 .bdrv_co_create_opts
= hdev_co_create_opts
,
3255 .create_opts
= &raw_create_opts
,
3256 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3257 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3259 .bdrv_co_preadv
= raw_co_preadv
,
3260 .bdrv_co_pwritev
= raw_co_pwritev
,
3261 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3262 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3263 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3264 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3265 .bdrv_refresh_limits
= raw_refresh_limits
,
3266 .bdrv_io_plug
= raw_aio_plug
,
3267 .bdrv_io_unplug
= raw_aio_unplug
,
3268 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3270 .bdrv_co_truncate
= raw_co_truncate
,
3271 .bdrv_getlength
= raw_getlength
,
3272 .bdrv_get_info
= raw_get_info
,
3273 .bdrv_get_allocated_file_size
3274 = raw_get_allocated_file_size
,
3275 .bdrv_check_perm
= raw_check_perm
,
3276 .bdrv_set_perm
= raw_set_perm
,
3277 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3278 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3279 .bdrv_probe_geometry
= hdev_probe_geometry
,
3281 /* generic scsi device */
3283 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3287 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3288 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3291 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3296 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3299 BDRVRawState
*s
= bs
->opaque
;
3303 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3304 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3307 static int cdrom_probe_device(const char *filename
)
3313 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3317 ret
= fstat(fd
, &st
);
3318 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3322 /* Attempt to detect via a CDROM specific ioctl */
3323 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3333 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3335 BDRVRawState
*s
= bs
->opaque
;
3338 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3339 return ret
== CDS_DISC_OK
;
3342 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3344 BDRVRawState
*s
= bs
->opaque
;
3347 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3348 perror("CDROMEJECT");
3350 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3351 perror("CDROMEJECT");
3355 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3357 BDRVRawState
*s
= bs
->opaque
;
3359 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3361 * Note: an error can happen if the distribution automatically
3364 /* perror("CDROM_LOCKDOOR"); */
3368 static BlockDriver bdrv_host_cdrom
= {
3369 .format_name
= "host_cdrom",
3370 .protocol_name
= "host_cdrom",
3371 .instance_size
= sizeof(BDRVRawState
),
3372 .bdrv_needs_filename
= true,
3373 .bdrv_probe_device
= cdrom_probe_device
,
3374 .bdrv_parse_filename
= cdrom_parse_filename
,
3375 .bdrv_file_open
= cdrom_open
,
3376 .bdrv_close
= raw_close
,
3377 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3378 .bdrv_reopen_commit
= raw_reopen_commit
,
3379 .bdrv_reopen_abort
= raw_reopen_abort
,
3380 .bdrv_co_create_opts
= hdev_co_create_opts
,
3381 .create_opts
= &raw_create_opts
,
3382 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3385 .bdrv_co_preadv
= raw_co_preadv
,
3386 .bdrv_co_pwritev
= raw_co_pwritev
,
3387 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3388 .bdrv_refresh_limits
= raw_refresh_limits
,
3389 .bdrv_io_plug
= raw_aio_plug
,
3390 .bdrv_io_unplug
= raw_aio_unplug
,
3391 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3393 .bdrv_co_truncate
= raw_co_truncate
,
3394 .bdrv_getlength
= raw_getlength
,
3395 .has_variable_length
= true,
3396 .bdrv_get_allocated_file_size
3397 = raw_get_allocated_file_size
,
3399 /* removable device support */
3400 .bdrv_is_inserted
= cdrom_is_inserted
,
3401 .bdrv_eject
= cdrom_eject
,
3402 .bdrv_lock_medium
= cdrom_lock_medium
,
3404 /* generic scsi device */
3405 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3407 #endif /* __linux__ */
3409 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3410 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3413 BDRVRawState
*s
= bs
->opaque
;
3414 Error
*local_err
= NULL
;
3419 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3421 error_propagate(errp
, local_err
);
3425 /* make sure the door isn't locked at this time */
3426 ioctl(s
->fd
, CDIOCALLOW
);
3430 static int cdrom_probe_device(const char *filename
)
3432 if (strstart(filename
, "/dev/cd", NULL
) ||
3433 strstart(filename
, "/dev/acd", NULL
))
3438 static int cdrom_reopen(BlockDriverState
*bs
)
3440 BDRVRawState
*s
= bs
->opaque
;
3444 * Force reread of possibly changed/newly loaded disc,
3445 * FreeBSD seems to not notice sometimes...
3449 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3456 /* make sure the door isn't locked at this time */
3457 ioctl(s
->fd
, CDIOCALLOW
);
3461 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3463 return raw_getlength(bs
) > 0;
3466 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3468 BDRVRawState
*s
= bs
->opaque
;
3473 (void) ioctl(s
->fd
, CDIOCALLOW
);
3476 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3477 perror("CDIOCEJECT");
3479 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3480 perror("CDIOCCLOSE");
3486 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3488 BDRVRawState
*s
= bs
->opaque
;
3492 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3494 * Note: an error can happen if the distribution automatically
3497 /* perror("CDROM_LOCKDOOR"); */
3501 static BlockDriver bdrv_host_cdrom
= {
3502 .format_name
= "host_cdrom",
3503 .protocol_name
= "host_cdrom",
3504 .instance_size
= sizeof(BDRVRawState
),
3505 .bdrv_needs_filename
= true,
3506 .bdrv_probe_device
= cdrom_probe_device
,
3507 .bdrv_parse_filename
= cdrom_parse_filename
,
3508 .bdrv_file_open
= cdrom_open
,
3509 .bdrv_close
= raw_close
,
3510 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3511 .bdrv_reopen_commit
= raw_reopen_commit
,
3512 .bdrv_reopen_abort
= raw_reopen_abort
,
3513 .bdrv_co_create_opts
= hdev_co_create_opts
,
3514 .create_opts
= &raw_create_opts
,
3516 .bdrv_co_preadv
= raw_co_preadv
,
3517 .bdrv_co_pwritev
= raw_co_pwritev
,
3518 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3519 .bdrv_refresh_limits
= raw_refresh_limits
,
3520 .bdrv_io_plug
= raw_aio_plug
,
3521 .bdrv_io_unplug
= raw_aio_unplug
,
3522 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3524 .bdrv_co_truncate
= raw_co_truncate
,
3525 .bdrv_getlength
= raw_getlength
,
3526 .has_variable_length
= true,
3527 .bdrv_get_allocated_file_size
3528 = raw_get_allocated_file_size
,
3530 /* removable device support */
3531 .bdrv_is_inserted
= cdrom_is_inserted
,
3532 .bdrv_eject
= cdrom_eject
,
3533 .bdrv_lock_medium
= cdrom_lock_medium
,
3535 #endif /* __FreeBSD__ */
3537 static void bdrv_file_init(void)
3540 * Register all the drivers. Note that order is important, the driver
3541 * registered last will get probed first.
3543 bdrv_register(&bdrv_file
);
3544 bdrv_register(&bdrv_host_device
);
3546 bdrv_register(&bdrv_host_cdrom
);
3548 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3549 bdrv_register(&bdrv_host_cdrom
);
3553 block_init(bdrv_file_init
);