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 int handle_aiocb_flush(void *opaque
)
1169 RawPosixAIOData
*aiocb
= opaque
;
1170 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1173 if (s
->page_cache_inconsistent
) {
1177 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1179 /* There is no clear definition of the semantics of a failing fsync(),
1180 * so we may have to assume the worst. The sad truth is that this
1181 * assumption is correct for Linux. Some pages are now probably marked
1182 * clean in the page cache even though they are inconsistent with the
1183 * on-disk contents. The next fdatasync() call would succeed, but no
1184 * further writeback attempt will be made. We can't get back to a state
1185 * in which we know what is on disk (we would have to rewrite
1186 * everything that was touched since the last fdatasync() at least), so
1187 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1188 * really defined, I have little hope that other OSes are doing better.
1190 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1192 if ((s
->open_flags
& O_DIRECT
) == 0) {
1193 s
->page_cache_inconsistent
= true;
1200 #ifdef CONFIG_PREADV
1202 static bool preadv_present
= true;
1205 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1207 return preadv(fd
, iov
, nr_iov
, offset
);
1211 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1213 return pwritev(fd
, iov
, nr_iov
, offset
);
1218 static bool preadv_present
= false;
1221 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1227 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1234 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1239 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1240 len
= qemu_pwritev(aiocb
->aio_fildes
,
1245 len
= qemu_preadv(aiocb
->aio_fildes
,
1249 } while (len
== -1 && errno
== EINTR
);
1258 * Read/writes the data to/from a given linear buffer.
1260 * Returns the number of bytes handles or -errno in case of an error. Short
1261 * reads are only returned if the end of the file is reached.
1263 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1268 while (offset
< aiocb
->aio_nbytes
) {
1269 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1270 len
= pwrite(aiocb
->aio_fildes
,
1271 (const char *)buf
+ offset
,
1272 aiocb
->aio_nbytes
- offset
,
1273 aiocb
->aio_offset
+ offset
);
1275 len
= pread(aiocb
->aio_fildes
,
1277 aiocb
->aio_nbytes
- offset
,
1278 aiocb
->aio_offset
+ offset
);
1280 if (len
== -1 && errno
== EINTR
) {
1282 } else if (len
== -1 && errno
== EINVAL
&&
1283 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1284 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1286 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1287 * after a short read. Assume that O_DIRECT short reads only occur
1288 * at EOF. Therefore this is a short read, not an I/O error.
1291 } else if (len
== -1) {
1294 } else if (len
== 0) {
1303 static ssize_t
handle_aiocb_rw(RawPosixAIOData
*aiocb
)
1308 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1310 * If there is just a single buffer, and it is properly aligned
1311 * we can just use plain pread/pwrite without any problems.
1313 if (aiocb
->io
.niov
== 1) {
1314 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1318 * We have more than one iovec, and all are properly aligned.
1320 * Try preadv/pwritev first and fall back to linearizing the
1321 * buffer if it's not supported.
1323 if (preadv_present
) {
1324 nbytes
= handle_aiocb_rw_vector(aiocb
);
1325 if (nbytes
== aiocb
->aio_nbytes
||
1326 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1329 preadv_present
= false;
1333 * XXX(hch): short read/write. no easy way to handle the reminder
1334 * using these interfaces. For now retry using plain
1340 * Ok, we have to do it the hard way, copy all segments into
1341 * a single aligned buffer.
1343 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1349 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1353 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1354 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1355 p
+= aiocb
->io
.iov
[i
].iov_len
;
1357 assert(p
- buf
== aiocb
->aio_nbytes
);
1360 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1361 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1363 size_t count
= aiocb
->aio_nbytes
, copy
;
1366 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1368 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1369 copy
= aiocb
->io
.iov
[i
].iov_len
;
1371 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1372 assert(count
>= copy
);
1381 if (nbytes
== aiocb
->aio_nbytes
) {
1383 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1384 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1387 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1388 0, aiocb
->aio_nbytes
- nbytes
);
1398 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1400 struct xfs_flock64 fl
;
1403 memset(&fl
, 0, sizeof(fl
));
1404 fl
.l_whence
= SEEK_SET
;
1405 fl
.l_start
= offset
;
1408 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1410 DPRINTF("cannot write zero range (%s)\n", strerror(errno
));
1417 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1419 struct xfs_flock64 fl
;
1422 memset(&fl
, 0, sizeof(fl
));
1423 fl
.l_whence
= SEEK_SET
;
1424 fl
.l_start
= offset
;
1427 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1429 DPRINTF("cannot punch hole (%s)\n", strerror(errno
));
1437 static int translate_err(int err
)
1439 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1446 #ifdef CONFIG_FALLOCATE
1447 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1450 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1453 } while (errno
== EINTR
);
1454 return translate_err(-errno
);
1458 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1461 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1463 if (!s
->has_write_zeroes
) {
1469 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1470 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1473 } while (errno
== EINTR
);
1475 ret
= translate_err(-errno
);
1478 if (ret
== -ENOTSUP
) {
1479 s
->has_write_zeroes
= false;
1484 static int handle_aiocb_write_zeroes(void *opaque
)
1486 RawPosixAIOData
*aiocb
= opaque
;
1487 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1488 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1490 #ifdef CONFIG_FALLOCATE
1494 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1495 return handle_aiocb_write_zeroes_block(aiocb
);
1500 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1504 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1505 if (s
->has_write_zeroes
) {
1506 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1507 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1508 if (ret
== 0 || ret
!= -ENOTSUP
) {
1511 s
->has_write_zeroes
= false;
1515 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1516 if (s
->has_discard
&& s
->has_fallocate
) {
1517 int ret
= do_fallocate(s
->fd
,
1518 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1519 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1521 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1522 if (ret
== 0 || ret
!= -ENOTSUP
) {
1525 s
->has_fallocate
= false;
1526 } else if (ret
!= -ENOTSUP
) {
1529 s
->has_discard
= false;
1534 #ifdef CONFIG_FALLOCATE
1535 /* Last resort: we are trying to extend the file with zeroed data. This
1536 * can be done via fallocate(fd, 0) */
1537 len
= bdrv_getlength(aiocb
->bs
);
1538 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1539 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1540 if (ret
== 0 || ret
!= -ENOTSUP
) {
1543 s
->has_fallocate
= false;
1550 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1552 RawPosixAIOData
*aiocb
= opaque
;
1553 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1556 /* First try to write zeros and unmap at the same time */
1558 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1559 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1560 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1561 if (ret
!= -ENOTSUP
) {
1568 /* xfs_discard() guarantees that the discarded area reads as all-zero
1569 * afterwards, so we can use it here. */
1570 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1574 /* If we couldn't manage to unmap while guaranteed that the area reads as
1575 * all-zero afterwards, just write zeroes without unmapping */
1576 ret
= handle_aiocb_write_zeroes(aiocb
);
1580 #ifndef HAVE_COPY_FILE_RANGE
1581 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1582 off_t
*out_off
, size_t len
, unsigned int flags
)
1584 #ifdef __NR_copy_file_range
1585 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1586 out_off
, len
, flags
);
1594 static int handle_aiocb_copy_range(void *opaque
)
1596 RawPosixAIOData
*aiocb
= opaque
;
1597 uint64_t bytes
= aiocb
->aio_nbytes
;
1598 off_t in_off
= aiocb
->aio_offset
;
1599 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1602 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1603 aiocb
->copy_range
.aio_fd2
, &out_off
,
1605 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1606 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1609 /* No progress (e.g. when beyond EOF), let the caller fall back to
1628 static int handle_aiocb_discard(void *opaque
)
1630 RawPosixAIOData
*aiocb
= opaque
;
1631 int ret
= -EOPNOTSUPP
;
1632 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1634 if (!s
->has_discard
) {
1638 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1641 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1642 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1645 } while (errno
== EINTR
);
1652 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1656 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1657 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1658 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1662 ret
= translate_err(ret
);
1663 if (ret
== -ENOTSUP
) {
1664 s
->has_discard
= false;
1669 static int handle_aiocb_truncate(void *opaque
)
1671 RawPosixAIOData
*aiocb
= opaque
;
1673 int64_t current_length
= 0;
1676 int fd
= aiocb
->aio_fildes
;
1677 int64_t offset
= aiocb
->aio_offset
;
1678 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1679 Error
**errp
= aiocb
->truncate
.errp
;
1681 if (fstat(fd
, &st
) < 0) {
1683 error_setg_errno(errp
, -result
, "Could not stat file");
1687 current_length
= st
.st_size
;
1688 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1689 error_setg(errp
, "Cannot use preallocation for shrinking files");
1694 #ifdef CONFIG_POSIX_FALLOCATE
1695 case PREALLOC_MODE_FALLOC
:
1697 * Truncating before posix_fallocate() makes it about twice slower on
1698 * file systems that do not support fallocate(), trying to check if a
1699 * block is allocated before allocating it, so don't do that here.
1701 if (offset
!= current_length
) {
1702 result
= -posix_fallocate(fd
, current_length
,
1703 offset
- current_length
);
1705 /* posix_fallocate() doesn't set errno. */
1706 error_setg_errno(errp
, -result
,
1707 "Could not preallocate new data");
1714 case PREALLOC_MODE_FULL
:
1716 int64_t num
= 0, left
= offset
- current_length
;
1720 * Knowing the final size from the beginning could allow the file
1721 * system driver to do less allocations and possibly avoid
1722 * fragmentation of the file.
1724 if (ftruncate(fd
, offset
) != 0) {
1726 error_setg_errno(errp
, -result
, "Could not resize file");
1730 buf
= g_malloc0(65536);
1732 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1733 if (seek_result
< 0) {
1735 error_setg_errno(errp
, -result
,
1736 "Failed to seek to the old end of file");
1741 num
= MIN(left
, 65536);
1742 result
= write(fd
, buf
, num
);
1744 if (errno
== EINTR
) {
1748 error_setg_errno(errp
, -result
,
1749 "Could not write zeros for preallocation");
1758 error_setg_errno(errp
, -result
,
1759 "Could not flush file to disk");
1765 case PREALLOC_MODE_OFF
:
1766 if (ftruncate(fd
, offset
) != 0) {
1768 error_setg_errno(errp
, -result
, "Could not resize file");
1773 error_setg(errp
, "Unsupported preallocation mode: %s",
1774 PreallocMode_str(prealloc
));
1780 if (ftruncate(fd
, current_length
) < 0) {
1781 error_report("Failed to restore old file length: %s",
1790 static int aio_worker(void *arg
)
1792 RawPosixAIOData
*aiocb
= arg
;
1795 switch (aiocb
->aio_type
& QEMU_AIO_TYPE_MASK
) {
1797 ret
= handle_aiocb_rw(aiocb
);
1799 case QEMU_AIO_WRITE
:
1800 ret
= handle_aiocb_rw(aiocb
);
1802 case QEMU_AIO_IOCTL
:
1803 ret
= handle_aiocb_ioctl(aiocb
);
1805 case QEMU_AIO_FLUSH
:
1806 case QEMU_AIO_DISCARD
:
1807 case QEMU_AIO_WRITE_ZEROES
:
1808 case QEMU_AIO_WRITE_ZEROES
| QEMU_AIO_DISCARD
:
1809 case QEMU_AIO_COPY_RANGE
:
1810 case QEMU_AIO_TRUNCATE
:
1811 g_assert_not_reached();
1813 error_report("invalid aio request (0x%x)", aiocb
->aio_type
);
1822 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1823 ThreadPoolFunc func
, void *arg
)
1825 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1826 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1827 return thread_pool_submit_co(pool
, func
, arg
);
1830 static int paio_submit_co_full(BlockDriverState
*bs
, int fd
,
1831 int64_t offset
, int fd2
, int64_t offset2
,
1833 int bytes
, int type
)
1835 RawPosixAIOData
*acb
= g_new(RawPosixAIOData
, 1);
1838 acb
->aio_type
= type
;
1839 acb
->aio_fildes
= fd
;
1841 acb
->aio_nbytes
= bytes
;
1842 acb
->aio_offset
= offset
;
1845 acb
->io
.iov
= qiov
->iov
;
1846 acb
->io
.niov
= qiov
->niov
;
1847 assert(qiov
->size
== bytes
);
1849 acb
->copy_range
.aio_fd2
= fd2
;
1850 acb
->copy_range
.aio_offset2
= offset2
;
1853 trace_file_paio_submit_co(offset
, bytes
, type
);
1854 return raw_thread_pool_submit(bs
, aio_worker
, acb
);
1857 static inline int paio_submit_co(BlockDriverState
*bs
, int fd
,
1858 int64_t offset
, QEMUIOVector
*qiov
,
1859 int bytes
, int type
)
1861 return paio_submit_co_full(bs
, fd
, offset
, -1, 0, qiov
, bytes
, type
);
1864 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1865 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1867 BDRVRawState
*s
= bs
->opaque
;
1869 if (fd_open(bs
) < 0)
1873 * Check if the underlying device requires requests to be aligned,
1874 * and if the request we are trying to submit is aligned or not.
1875 * If this is the case tell the low-level driver that it needs
1876 * to copy the buffer.
1878 if (s
->needs_alignment
) {
1879 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1880 type
|= QEMU_AIO_MISALIGNED
;
1881 #ifdef CONFIG_LINUX_AIO
1882 } else if (s
->use_linux_aio
) {
1883 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1884 assert(qiov
->size
== bytes
);
1885 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1890 return paio_submit_co(bs
, s
->fd
, offset
, qiov
, bytes
, type
);
1893 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1894 uint64_t bytes
, QEMUIOVector
*qiov
,
1897 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1900 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1901 uint64_t bytes
, QEMUIOVector
*qiov
,
1905 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1908 static void raw_aio_plug(BlockDriverState
*bs
)
1910 #ifdef CONFIG_LINUX_AIO
1911 BDRVRawState
*s
= bs
->opaque
;
1912 if (s
->use_linux_aio
) {
1913 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1914 laio_io_plug(bs
, aio
);
1919 static void raw_aio_unplug(BlockDriverState
*bs
)
1921 #ifdef CONFIG_LINUX_AIO
1922 BDRVRawState
*s
= bs
->opaque
;
1923 if (s
->use_linux_aio
) {
1924 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1925 laio_io_unplug(bs
, aio
);
1930 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1932 BDRVRawState
*s
= bs
->opaque
;
1933 RawPosixAIOData acb
;
1941 acb
= (RawPosixAIOData
) {
1943 .aio_fildes
= s
->fd
,
1944 .aio_type
= QEMU_AIO_FLUSH
,
1947 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1950 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1951 AioContext
*new_context
)
1953 #ifdef CONFIG_LINUX_AIO
1954 BDRVRawState
*s
= bs
->opaque
;
1955 if (s
->use_linux_aio
) {
1957 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1958 error_reportf_err(local_err
, "Unable to use native AIO, "
1959 "falling back to thread pool: ");
1960 s
->use_linux_aio
= false;
1966 static void raw_close(BlockDriverState
*bs
)
1968 BDRVRawState
*s
= bs
->opaque
;
1977 * Truncates the given regular file @fd to @offset and, when growing, fills the
1978 * new space according to @prealloc.
1980 * Returns: 0 on success, -errno on failure.
1982 static int coroutine_fn
1983 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1984 PreallocMode prealloc
, Error
**errp
)
1986 RawPosixAIOData acb
;
1988 acb
= (RawPosixAIOData
) {
1991 .aio_type
= QEMU_AIO_TRUNCATE
,
1992 .aio_offset
= offset
,
1994 .prealloc
= prealloc
,
1999 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2002 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2003 PreallocMode prealloc
, Error
**errp
)
2005 BDRVRawState
*s
= bs
->opaque
;
2009 if (fstat(s
->fd
, &st
)) {
2011 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2015 if (S_ISREG(st
.st_mode
)) {
2016 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2019 if (prealloc
!= PREALLOC_MODE_OFF
) {
2020 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2021 "non-regular file", PreallocMode_str(prealloc
));
2025 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2026 if (offset
> raw_getlength(bs
)) {
2027 error_setg(errp
, "Cannot grow device files");
2031 error_setg(errp
, "Resizing this file is not supported");
2039 static int64_t raw_getlength(BlockDriverState
*bs
)
2041 BDRVRawState
*s
= bs
->opaque
;
2047 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2048 struct disklabel dl
;
2050 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2052 return (uint64_t)dl
.d_secsize
*
2053 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2057 #elif defined(__NetBSD__)
2058 static int64_t raw_getlength(BlockDriverState
*bs
)
2060 BDRVRawState
*s
= bs
->opaque
;
2066 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2067 struct dkwedge_info dkw
;
2069 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2070 return dkw
.dkw_size
* 512;
2072 struct disklabel dl
;
2074 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2076 return (uint64_t)dl
.d_secsize
*
2077 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2082 #elif defined(__sun__)
2083 static int64_t raw_getlength(BlockDriverState
*bs
)
2085 BDRVRawState
*s
= bs
->opaque
;
2086 struct dk_minfo minfo
;
2096 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2098 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2100 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2104 * There are reports that lseek on some devices fails, but
2105 * irc discussion said that contingency on contingency was overkill.
2107 size
= lseek(s
->fd
, 0, SEEK_END
);
2113 #elif defined(CONFIG_BSD)
2114 static int64_t raw_getlength(BlockDriverState
*bs
)
2116 BDRVRawState
*s
= bs
->opaque
;
2120 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2129 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2132 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2133 #ifdef DIOCGMEDIASIZE
2134 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2135 #elif defined(DIOCGPART)
2138 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2139 size
= pi
.media_size
;
2145 #if defined(__APPLE__) && defined(__MACH__)
2147 uint64_t sectors
= 0;
2148 uint32_t sector_size
= 0;
2150 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2151 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2152 size
= sectors
* sector_size
;
2154 size
= lseek(fd
, 0LL, SEEK_END
);
2161 size
= lseek(fd
, 0LL, SEEK_END
);
2166 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2169 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2170 if (size
== 2048LL * (unsigned)-1)
2172 /* XXX no disc? maybe we need to reopen... */
2173 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2180 size
= lseek(fd
, 0, SEEK_END
);
2188 static int64_t raw_getlength(BlockDriverState
*bs
)
2190 BDRVRawState
*s
= bs
->opaque
;
2199 size
= lseek(s
->fd
, 0, SEEK_END
);
2207 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2210 BDRVRawState
*s
= bs
->opaque
;
2212 if (fstat(s
->fd
, &st
) < 0) {
2215 return (int64_t)st
.st_blocks
* 512;
2218 static int coroutine_fn
2219 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2221 BlockdevCreateOptionsFile
*file_opts
;
2222 Error
*local_err
= NULL
;
2224 uint64_t perm
, shared
;
2227 /* Validate options and set default values */
2228 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2229 file_opts
= &options
->u
.file
;
2231 if (!file_opts
->has_nocow
) {
2232 file_opts
->nocow
= false;
2234 if (!file_opts
->has_preallocation
) {
2235 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2239 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2242 error_setg_errno(errp
, -result
, "Could not create file");
2246 /* Take permissions: We want to discard everything, so we need
2247 * BLK_PERM_WRITE; and truncation to the desired size requires
2249 * On the other hand, we cannot share the RESIZE permission
2250 * because we promise that after this function, the file has the
2251 * size given in the options. If someone else were to resize it
2252 * concurrently, we could not guarantee that.
2253 * Note that after this function, we can no longer guarantee that
2254 * the file is not touched by a third party, so it may be resized
2256 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2257 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2259 /* Step one: Take locks */
2260 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2265 /* Step two: Check that nobody else has taken conflicting locks */
2266 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2268 error_append_hint(errp
,
2269 "Is another process using the image [%s]?\n",
2270 file_opts
->filename
);
2274 /* Clear the file by truncating it to 0 */
2275 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2280 if (file_opts
->nocow
) {
2282 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2283 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2284 * will be ignored since any failure of this operation should not
2285 * block the left work.
2288 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2289 attr
|= FS_NOCOW_FL
;
2290 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2295 /* Resize and potentially preallocate the file to the desired
2297 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2298 file_opts
->preallocation
, errp
);
2304 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2306 /* The above call should not fail, and if it does, that does
2307 * not mean the whole creation operation has failed. So
2308 * report it the user for their convenience, but do not report
2309 * it to the caller. */
2310 warn_report_err(local_err
);
2314 if (qemu_close(fd
) != 0 && result
== 0) {
2316 error_setg_errno(errp
, -result
, "Could not close the new file");
2322 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2325 BlockdevCreateOptions options
;
2326 int64_t total_size
= 0;
2328 PreallocMode prealloc
;
2330 Error
*local_err
= NULL
;
2332 /* Skip file: protocol prefix */
2333 strstart(filename
, "file:", &filename
);
2335 /* Read out options */
2336 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2338 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2339 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2340 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2341 PREALLOC_MODE_OFF
, &local_err
);
2344 error_propagate(errp
, local_err
);
2348 options
= (BlockdevCreateOptions
) {
2349 .driver
= BLOCKDEV_DRIVER_FILE
,
2351 .filename
= (char *) filename
,
2353 .has_preallocation
= true,
2354 .preallocation
= prealloc
,
2359 return raw_co_create(&options
, errp
);
2363 * Find allocation range in @bs around offset @start.
2364 * May change underlying file descriptor's file offset.
2365 * If @start is not in a hole, store @start in @data, and the
2366 * beginning of the next hole in @hole, and return 0.
2367 * If @start is in a non-trailing hole, store @start in @hole and the
2368 * beginning of the next non-hole in @data, and return 0.
2369 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2370 * If we can't find out, return a negative errno other than -ENXIO.
2372 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2373 off_t
*data
, off_t
*hole
)
2375 #if defined SEEK_HOLE && defined SEEK_DATA
2376 BDRVRawState
*s
= bs
->opaque
;
2381 * D1. offs == start: start is in data
2382 * D2. offs > start: start is in a hole, next data at offs
2383 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2384 * or start is beyond EOF
2385 * If the latter happens, the file has been truncated behind
2386 * our back since we opened it. All bets are off then.
2387 * Treating like a trailing hole is simplest.
2388 * D4. offs < 0, errno != ENXIO: we learned nothing
2390 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2392 return -errno
; /* D3 or D4 */
2396 /* This is not a valid return by lseek(). We are safe to just return
2397 * -EIO in this case, and we'll treat it like D4. */
2402 /* D2: in hole, next data at offs */
2408 /* D1: in data, end not yet known */
2412 * H1. offs == start: start is in a hole
2413 * If this happens here, a hole has been dug behind our back
2414 * since the previous lseek().
2415 * H2. offs > start: either start is in data, next hole at offs,
2416 * or start is in trailing hole, EOF at offs
2417 * Linux treats trailing holes like any other hole: offs ==
2418 * start. Solaris seeks to EOF instead: offs > start (blech).
2419 * If that happens here, a hole has been dug behind our back
2420 * since the previous lseek().
2421 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2422 * If this happens, the file has been truncated behind our
2423 * back since we opened it. Treat it like a trailing hole.
2424 * H4. offs < 0, errno != ENXIO: we learned nothing
2425 * Pretend we know nothing at all, i.e. "forget" about D1.
2427 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2429 return -errno
; /* D1 and (H3 or H4) */
2433 /* This is not a valid return by lseek(). We are safe to just return
2434 * -EIO in this case, and we'll treat it like H4. */
2440 * D1 and H2: either in data, next hole at offs, or it was in
2441 * data but is now in a trailing hole. In the latter case,
2442 * all bets are off. Treating it as if it there was data all
2443 * the way to EOF is safe, so simply do that.
2458 * Returns the allocation status of the specified offset.
2460 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2462 * 'pnum' is set to the number of bytes (including and immediately following
2463 * the specified offset) that are known to be in the same
2464 * allocated/unallocated state.
2466 * 'bytes' is the max value 'pnum' should be set to.
2468 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2471 int64_t bytes
, int64_t *pnum
,
2473 BlockDriverState
**file
)
2475 off_t data
= 0, hole
= 0;
2487 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2490 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2491 if (ret
== -ENXIO
) {
2494 ret
= BDRV_BLOCK_ZERO
;
2495 } else if (ret
< 0) {
2496 /* No info available, so pretend there are no holes */
2498 ret
= BDRV_BLOCK_DATA
;
2499 } else if (data
== offset
) {
2500 /* On a data extent, compute bytes to the end of the extent,
2501 * possibly including a partial sector at EOF. */
2502 *pnum
= MIN(bytes
, hole
- offset
);
2503 ret
= BDRV_BLOCK_DATA
;
2505 /* On a hole, compute bytes to the beginning of the next extent. */
2506 assert(hole
== offset
);
2507 *pnum
= MIN(bytes
, data
- offset
);
2508 ret
= BDRV_BLOCK_ZERO
;
2512 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2515 #if defined(__linux__)
2516 /* Verify that the file is not in the page cache */
2517 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2519 const size_t window_size
= 128 * 1024 * 1024;
2520 BDRVRawState
*s
= bs
->opaque
;
2521 void *window
= NULL
;
2528 /* mincore(2) page status information requires 1 byte per page */
2529 page_size
= sysconf(_SC_PAGESIZE
);
2530 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2532 end
= raw_getlength(bs
);
2534 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2541 /* Unmap previous window if size has changed */
2542 new_length
= MIN(end
- offset
, window_size
);
2543 if (new_length
!= length
) {
2544 munmap(window
, length
);
2549 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2551 if (new_window
== MAP_FAILED
) {
2552 error_setg_errno(errp
, errno
, "mmap failed");
2556 window
= new_window
;
2557 length
= new_length
;
2559 ret
= mincore(window
, length
, vec
);
2561 error_setg_errno(errp
, errno
, "mincore failed");
2565 vec_end
= DIV_ROUND_UP(length
, page_size
);
2566 for (i
= 0; i
< vec_end
; i
++) {
2568 error_setg(errp
, "page cache still in use!");
2575 munmap(window
, length
);
2580 #endif /* __linux__ */
2582 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2585 BDRVRawState
*s
= bs
->opaque
;
2590 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2594 if (s
->open_flags
& O_DIRECT
) {
2595 return; /* No host kernel page cache */
2598 #if defined(__linux__)
2599 /* This sets the scene for the next syscall... */
2600 ret
= bdrv_co_flush(bs
);
2602 error_setg_errno(errp
, -ret
, "flush failed");
2606 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2607 * process. These limitations are okay because we just fsynced the file,
2608 * we don't use mmap, and the file should not be in use by other processes.
2610 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2611 if (ret
!= 0) { /* the return value is a positive errno */
2612 error_setg_errno(errp
, ret
, "fadvise failed");
2616 if (s
->check_cache_dropped
) {
2617 check_cache_dropped(bs
, errp
);
2619 #else /* __linux__ */
2620 /* Do nothing. Live migration to a remote host with cache.direct=off is
2621 * unsupported on other host operating systems. Cache consistency issues
2622 * may occur but no error is reported here, partly because that's the
2623 * historical behavior and partly because it's hard to differentiate valid
2624 * configurations that should not cause errors.
2626 #endif /* !__linux__ */
2629 static coroutine_fn
int
2630 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2632 BDRVRawState
*s
= bs
->opaque
;
2633 RawPosixAIOData acb
;
2635 acb
= (RawPosixAIOData
) {
2637 .aio_fildes
= s
->fd
,
2638 .aio_type
= QEMU_AIO_DISCARD
,
2639 .aio_offset
= offset
,
2640 .aio_nbytes
= bytes
,
2644 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2647 return raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2650 static coroutine_fn
int
2651 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2653 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2656 static int coroutine_fn
2657 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2658 BdrvRequestFlags flags
, bool blkdev
)
2660 BDRVRawState
*s
= bs
->opaque
;
2661 RawPosixAIOData acb
;
2662 ThreadPoolFunc
*handler
;
2664 acb
= (RawPosixAIOData
) {
2666 .aio_fildes
= s
->fd
,
2667 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2668 .aio_offset
= offset
,
2669 .aio_nbytes
= bytes
,
2673 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2676 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2677 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2678 handler
= handle_aiocb_write_zeroes_unmap
;
2680 handler
= handle_aiocb_write_zeroes
;
2683 return raw_thread_pool_submit(bs
, handler
, &acb
);
2686 static int coroutine_fn
raw_co_pwrite_zeroes(
2687 BlockDriverState
*bs
, int64_t offset
,
2688 int bytes
, BdrvRequestFlags flags
)
2690 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2693 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2695 BDRVRawState
*s
= bs
->opaque
;
2697 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2701 static QemuOptsList raw_create_opts
= {
2702 .name
= "raw-create-opts",
2703 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2706 .name
= BLOCK_OPT_SIZE
,
2707 .type
= QEMU_OPT_SIZE
,
2708 .help
= "Virtual disk size"
2711 .name
= BLOCK_OPT_NOCOW
,
2712 .type
= QEMU_OPT_BOOL
,
2713 .help
= "Turn off copy-on-write (valid only on btrfs)"
2716 .name
= BLOCK_OPT_PREALLOC
,
2717 .type
= QEMU_OPT_STRING
,
2718 .help
= "Preallocation mode (allowed values: off, falloc, full)"
2720 { /* end of list */ }
2724 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2727 return raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2730 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2732 BDRVRawState
*s
= bs
->opaque
;
2733 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2735 s
->shared_perm
= shared
;
2738 static void raw_abort_perm_update(BlockDriverState
*bs
)
2740 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2743 static int coroutine_fn
raw_co_copy_range_from(
2744 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2745 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2746 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2748 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2749 read_flags
, write_flags
);
2752 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2754 uint64_t src_offset
,
2756 uint64_t dst_offset
,
2758 BdrvRequestFlags read_flags
,
2759 BdrvRequestFlags write_flags
)
2761 RawPosixAIOData acb
;
2762 BDRVRawState
*s
= bs
->opaque
;
2763 BDRVRawState
*src_s
;
2765 assert(dst
->bs
== bs
);
2766 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2770 src_s
= src
->bs
->opaque
;
2771 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2775 acb
= (RawPosixAIOData
) {
2777 .aio_type
= QEMU_AIO_COPY_RANGE
,
2778 .aio_fildes
= src_s
->fd
,
2779 .aio_offset
= src_offset
,
2780 .aio_nbytes
= bytes
,
2783 .aio_offset2
= dst_offset
,
2787 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2790 BlockDriver bdrv_file
= {
2791 .format_name
= "file",
2792 .protocol_name
= "file",
2793 .instance_size
= sizeof(BDRVRawState
),
2794 .bdrv_needs_filename
= true,
2795 .bdrv_probe
= NULL
, /* no probe for protocols */
2796 .bdrv_parse_filename
= raw_parse_filename
,
2797 .bdrv_file_open
= raw_open
,
2798 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2799 .bdrv_reopen_commit
= raw_reopen_commit
,
2800 .bdrv_reopen_abort
= raw_reopen_abort
,
2801 .bdrv_close
= raw_close
,
2802 .bdrv_co_create
= raw_co_create
,
2803 .bdrv_co_create_opts
= raw_co_create_opts
,
2804 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2805 .bdrv_co_block_status
= raw_co_block_status
,
2806 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2807 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2809 .bdrv_co_preadv
= raw_co_preadv
,
2810 .bdrv_co_pwritev
= raw_co_pwritev
,
2811 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2812 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2813 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2814 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2815 .bdrv_refresh_limits
= raw_refresh_limits
,
2816 .bdrv_io_plug
= raw_aio_plug
,
2817 .bdrv_io_unplug
= raw_aio_unplug
,
2818 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2820 .bdrv_co_truncate
= raw_co_truncate
,
2821 .bdrv_getlength
= raw_getlength
,
2822 .bdrv_get_info
= raw_get_info
,
2823 .bdrv_get_allocated_file_size
2824 = raw_get_allocated_file_size
,
2825 .bdrv_check_perm
= raw_check_perm
,
2826 .bdrv_set_perm
= raw_set_perm
,
2827 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2828 .create_opts
= &raw_create_opts
,
2831 /***********************************************/
2834 #if defined(__APPLE__) && defined(__MACH__)
2835 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2836 CFIndex maxPathSize
, int flags
);
2837 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2839 kern_return_t kernResult
= KERN_FAILURE
;
2840 mach_port_t masterPort
;
2841 CFMutableDictionaryRef classesToMatch
;
2842 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2843 char *mediaType
= NULL
;
2845 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2846 if ( KERN_SUCCESS
!= kernResult
) {
2847 printf( "IOMasterPort returned %d\n", kernResult
);
2851 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2852 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2853 if (classesToMatch
== NULL
) {
2854 error_report("IOServiceMatching returned NULL for %s",
2855 matching_array
[index
]);
2858 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2860 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2862 if (kernResult
!= KERN_SUCCESS
) {
2863 error_report("Note: IOServiceGetMatchingServices returned %d",
2868 /* If a match was found, leave the loop */
2869 if (*mediaIterator
!= 0) {
2870 DPRINTF("Matching using %s\n", matching_array
[index
]);
2871 mediaType
= g_strdup(matching_array
[index
]);
2878 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2879 CFIndex maxPathSize
, int flags
)
2881 io_object_t nextMedia
;
2882 kern_return_t kernResult
= KERN_FAILURE
;
2884 nextMedia
= IOIteratorNext( mediaIterator
);
2887 CFTypeRef bsdPathAsCFString
;
2888 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
2889 if ( bsdPathAsCFString
) {
2890 size_t devPathLength
;
2891 strcpy( bsdPath
, _PATH_DEV
);
2892 if (flags
& BDRV_O_NOCACHE
) {
2893 strcat(bsdPath
, "r");
2895 devPathLength
= strlen( bsdPath
);
2896 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
2897 kernResult
= KERN_SUCCESS
;
2899 CFRelease( bsdPathAsCFString
);
2901 IOObjectRelease( nextMedia
);
2907 /* Sets up a real cdrom for use in QEMU */
2908 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
2910 int index
, num_of_test_partitions
= 2, fd
;
2911 char test_partition
[MAXPATHLEN
];
2912 bool partition_found
= false;
2914 /* look for a working partition */
2915 for (index
= 0; index
< num_of_test_partitions
; index
++) {
2916 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
2918 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
2920 partition_found
= true;
2926 /* if a working partition on the device was not found */
2927 if (partition_found
== false) {
2928 error_setg(errp
, "Failed to find a working partition on disc");
2930 DPRINTF("Using %s as optical disc\n", test_partition
);
2931 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
2933 return partition_found
;
2936 /* Prints directions on mounting and unmounting a device */
2937 static void print_unmounting_directions(const char *file_name
)
2939 error_report("If device %s is mounted on the desktop, unmount"
2940 " it first before using it in QEMU", file_name
);
2941 error_report("Command to unmount device: diskutil unmountDisk %s",
2943 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
2946 #endif /* defined(__APPLE__) && defined(__MACH__) */
2948 static int hdev_probe_device(const char *filename
)
2952 /* allow a dedicated CD-ROM driver to match with a higher priority */
2953 if (strstart(filename
, "/dev/cdrom", NULL
))
2956 if (stat(filename
, &st
) >= 0 &&
2957 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
2964 static int check_hdev_writable(BDRVRawState
*s
)
2966 #if defined(BLKROGET)
2967 /* Linux block devices can be configured "read-only" using blockdev(8).
2968 * This is independent of device node permissions and therefore open(2)
2969 * with O_RDWR succeeds. Actual writes fail with EPERM.
2971 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2972 * check for read-only block devices so that Linux block devices behave
2978 if (fstat(s
->fd
, &st
)) {
2982 if (!S_ISBLK(st
.st_mode
)) {
2986 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
2993 #endif /* defined(BLKROGET) */
2997 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3000 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3003 static bool hdev_is_sg(BlockDriverState
*bs
)
3006 #if defined(__linux__)
3008 BDRVRawState
*s
= bs
->opaque
;
3010 struct sg_scsi_id scsiid
;
3014 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3018 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3023 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3025 DPRINTF("SG device found: type=%d, version=%d\n",
3026 scsiid
.scsi_type
, sg_version
);
3035 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3038 BDRVRawState
*s
= bs
->opaque
;
3039 Error
*local_err
= NULL
;
3042 #if defined(__APPLE__) && defined(__MACH__)
3044 * Caution: while qdict_get_str() is fine, getting non-string types
3045 * would require more care. When @options come from -blockdev or
3046 * blockdev_add, its members are typed according to the QAPI
3047 * schema, but when they come from -drive, they're all QString.
3049 const char *filename
= qdict_get_str(options
, "filename");
3050 char bsd_path
[MAXPATHLEN
] = "";
3051 bool error_occurred
= false;
3053 /* If using a real cdrom */
3054 if (strcmp(filename
, "/dev/cdrom") == 0) {
3055 char *mediaType
= NULL
;
3056 kern_return_t ret_val
;
3057 io_iterator_t mediaIterator
= 0;
3059 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3060 if (mediaType
== NULL
) {
3061 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3063 error_occurred
= true;
3064 goto hdev_open_Mac_error
;
3067 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3068 if (ret_val
!= KERN_SUCCESS
) {
3069 error_setg(errp
, "Could not get BSD path for optical drive");
3070 error_occurred
= true;
3071 goto hdev_open_Mac_error
;
3074 /* If a real optical drive was not found */
3075 if (bsd_path
[0] == '\0') {
3076 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3077 error_occurred
= true;
3078 goto hdev_open_Mac_error
;
3081 /* If using a cdrom disc and finding a partition on the disc failed */
3082 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3083 setup_cdrom(bsd_path
, errp
) == false) {
3084 print_unmounting_directions(bsd_path
);
3085 error_occurred
= true;
3086 goto hdev_open_Mac_error
;
3089 qdict_put_str(options
, "filename", bsd_path
);
3091 hdev_open_Mac_error
:
3093 if (mediaIterator
) {
3094 IOObjectRelease(mediaIterator
);
3096 if (error_occurred
) {
3100 #endif /* defined(__APPLE__) && defined(__MACH__) */
3102 s
->type
= FTYPE_FILE
;
3104 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3106 error_propagate(errp
, local_err
);
3107 #if defined(__APPLE__) && defined(__MACH__)
3109 filename
= bsd_path
;
3111 /* if a physical device experienced an error while being opened */
3112 if (strncmp(filename
, "/dev/", 5) == 0) {
3113 print_unmounting_directions(filename
);
3115 #endif /* defined(__APPLE__) && defined(__MACH__) */
3119 /* Since this does ioctl the device must be already opened */
3120 bs
->sg
= hdev_is_sg(bs
);
3122 if (flags
& BDRV_O_RDWR
) {
3123 ret
= check_hdev_writable(s
);
3126 error_setg_errno(errp
, -ret
, "The device is not writable");
3134 #if defined(__linux__)
3136 static BlockAIOCB
*hdev_aio_ioctl(BlockDriverState
*bs
,
3137 unsigned long int req
, void *buf
,
3138 BlockCompletionFunc
*cb
, void *opaque
)
3140 BDRVRawState
*s
= bs
->opaque
;
3141 RawPosixAIOData
*acb
;
3144 if (fd_open(bs
) < 0)
3147 if (req
== SG_IO
&& s
->pr_mgr
) {
3148 struct sg_io_hdr
*io_hdr
= buf
;
3149 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3150 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3151 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3152 s
->fd
, io_hdr
, cb
, opaque
);
3156 acb
= g_new(RawPosixAIOData
, 1);
3158 acb
->aio_type
= QEMU_AIO_IOCTL
;
3159 acb
->aio_fildes
= s
->fd
;
3160 acb
->aio_offset
= 0;
3161 acb
->ioctl
.buf
= buf
;
3162 acb
->ioctl
.cmd
= req
;
3163 pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3164 return thread_pool_submit_aio(pool
, aio_worker
, acb
, cb
, opaque
);
3168 static int fd_open(BlockDriverState
*bs
)
3170 BDRVRawState
*s
= bs
->opaque
;
3172 /* this is just to ensure s->fd is sane (its called by io ops) */
3178 static coroutine_fn
int
3179 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3187 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3190 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3191 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3200 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3203 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3208 struct stat stat_buf
;
3209 int64_t total_size
= 0;
3212 /* This function is used by both protocol block drivers and therefore either
3213 * of these prefixes may be given.
3214 * The return value has to be stored somewhere, otherwise this is an error
3215 * due to -Werror=unused-value. */
3217 strstart(filename
, "host_device:", &filename
) ||
3218 strstart(filename
, "host_cdrom:" , &filename
);
3222 ret
= raw_normalize_devicepath(&filename
, errp
);
3227 /* Read out options */
3228 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3231 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3234 error_setg_errno(errp
, -ret
, "Could not open device");
3238 if (fstat(fd
, &stat_buf
) < 0) {
3240 error_setg_errno(errp
, -ret
, "Could not stat device");
3241 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3243 "The given file is neither a block nor a character device");
3245 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3246 error_setg(errp
, "Device is too small");
3250 if (!ret
&& total_size
) {
3251 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3252 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3253 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3256 ret
= qemu_write_full(fd
, buf
, zero_size
);
3257 ret
= ret
== zero_size
? 0 : -errno
;
3264 static BlockDriver bdrv_host_device
= {
3265 .format_name
= "host_device",
3266 .protocol_name
= "host_device",
3267 .instance_size
= sizeof(BDRVRawState
),
3268 .bdrv_needs_filename
= true,
3269 .bdrv_probe_device
= hdev_probe_device
,
3270 .bdrv_parse_filename
= hdev_parse_filename
,
3271 .bdrv_file_open
= hdev_open
,
3272 .bdrv_close
= raw_close
,
3273 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3274 .bdrv_reopen_commit
= raw_reopen_commit
,
3275 .bdrv_reopen_abort
= raw_reopen_abort
,
3276 .bdrv_co_create_opts
= hdev_co_create_opts
,
3277 .create_opts
= &raw_create_opts
,
3278 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3279 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3281 .bdrv_co_preadv
= raw_co_preadv
,
3282 .bdrv_co_pwritev
= raw_co_pwritev
,
3283 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3284 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3285 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3286 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3287 .bdrv_refresh_limits
= raw_refresh_limits
,
3288 .bdrv_io_plug
= raw_aio_plug
,
3289 .bdrv_io_unplug
= raw_aio_unplug
,
3290 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3292 .bdrv_co_truncate
= raw_co_truncate
,
3293 .bdrv_getlength
= raw_getlength
,
3294 .bdrv_get_info
= raw_get_info
,
3295 .bdrv_get_allocated_file_size
3296 = raw_get_allocated_file_size
,
3297 .bdrv_check_perm
= raw_check_perm
,
3298 .bdrv_set_perm
= raw_set_perm
,
3299 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3300 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3301 .bdrv_probe_geometry
= hdev_probe_geometry
,
3303 /* generic scsi device */
3305 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3309 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3310 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3313 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3318 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3321 BDRVRawState
*s
= bs
->opaque
;
3325 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3326 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3329 static int cdrom_probe_device(const char *filename
)
3335 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3339 ret
= fstat(fd
, &st
);
3340 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3344 /* Attempt to detect via a CDROM specific ioctl */
3345 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3355 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3357 BDRVRawState
*s
= bs
->opaque
;
3360 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3361 return ret
== CDS_DISC_OK
;
3364 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3366 BDRVRawState
*s
= bs
->opaque
;
3369 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3370 perror("CDROMEJECT");
3372 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3373 perror("CDROMEJECT");
3377 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3379 BDRVRawState
*s
= bs
->opaque
;
3381 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3383 * Note: an error can happen if the distribution automatically
3386 /* perror("CDROM_LOCKDOOR"); */
3390 static BlockDriver bdrv_host_cdrom
= {
3391 .format_name
= "host_cdrom",
3392 .protocol_name
= "host_cdrom",
3393 .instance_size
= sizeof(BDRVRawState
),
3394 .bdrv_needs_filename
= true,
3395 .bdrv_probe_device
= cdrom_probe_device
,
3396 .bdrv_parse_filename
= cdrom_parse_filename
,
3397 .bdrv_file_open
= cdrom_open
,
3398 .bdrv_close
= raw_close
,
3399 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3400 .bdrv_reopen_commit
= raw_reopen_commit
,
3401 .bdrv_reopen_abort
= raw_reopen_abort
,
3402 .bdrv_co_create_opts
= hdev_co_create_opts
,
3403 .create_opts
= &raw_create_opts
,
3404 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3407 .bdrv_co_preadv
= raw_co_preadv
,
3408 .bdrv_co_pwritev
= raw_co_pwritev
,
3409 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3410 .bdrv_refresh_limits
= raw_refresh_limits
,
3411 .bdrv_io_plug
= raw_aio_plug
,
3412 .bdrv_io_unplug
= raw_aio_unplug
,
3413 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3415 .bdrv_co_truncate
= raw_co_truncate
,
3416 .bdrv_getlength
= raw_getlength
,
3417 .has_variable_length
= true,
3418 .bdrv_get_allocated_file_size
3419 = raw_get_allocated_file_size
,
3421 /* removable device support */
3422 .bdrv_is_inserted
= cdrom_is_inserted
,
3423 .bdrv_eject
= cdrom_eject
,
3424 .bdrv_lock_medium
= cdrom_lock_medium
,
3426 /* generic scsi device */
3427 .bdrv_aio_ioctl
= hdev_aio_ioctl
,
3429 #endif /* __linux__ */
3431 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3432 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3435 BDRVRawState
*s
= bs
->opaque
;
3436 Error
*local_err
= NULL
;
3441 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3443 error_propagate(errp
, local_err
);
3447 /* make sure the door isn't locked at this time */
3448 ioctl(s
->fd
, CDIOCALLOW
);
3452 static int cdrom_probe_device(const char *filename
)
3454 if (strstart(filename
, "/dev/cd", NULL
) ||
3455 strstart(filename
, "/dev/acd", NULL
))
3460 static int cdrom_reopen(BlockDriverState
*bs
)
3462 BDRVRawState
*s
= bs
->opaque
;
3466 * Force reread of possibly changed/newly loaded disc,
3467 * FreeBSD seems to not notice sometimes...
3471 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3478 /* make sure the door isn't locked at this time */
3479 ioctl(s
->fd
, CDIOCALLOW
);
3483 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3485 return raw_getlength(bs
) > 0;
3488 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3490 BDRVRawState
*s
= bs
->opaque
;
3495 (void) ioctl(s
->fd
, CDIOCALLOW
);
3498 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3499 perror("CDIOCEJECT");
3501 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3502 perror("CDIOCCLOSE");
3508 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3510 BDRVRawState
*s
= bs
->opaque
;
3514 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3516 * Note: an error can happen if the distribution automatically
3519 /* perror("CDROM_LOCKDOOR"); */
3523 static BlockDriver bdrv_host_cdrom
= {
3524 .format_name
= "host_cdrom",
3525 .protocol_name
= "host_cdrom",
3526 .instance_size
= sizeof(BDRVRawState
),
3527 .bdrv_needs_filename
= true,
3528 .bdrv_probe_device
= cdrom_probe_device
,
3529 .bdrv_parse_filename
= cdrom_parse_filename
,
3530 .bdrv_file_open
= cdrom_open
,
3531 .bdrv_close
= raw_close
,
3532 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3533 .bdrv_reopen_commit
= raw_reopen_commit
,
3534 .bdrv_reopen_abort
= raw_reopen_abort
,
3535 .bdrv_co_create_opts
= hdev_co_create_opts
,
3536 .create_opts
= &raw_create_opts
,
3538 .bdrv_co_preadv
= raw_co_preadv
,
3539 .bdrv_co_pwritev
= raw_co_pwritev
,
3540 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3541 .bdrv_refresh_limits
= raw_refresh_limits
,
3542 .bdrv_io_plug
= raw_aio_plug
,
3543 .bdrv_io_unplug
= raw_aio_unplug
,
3544 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3546 .bdrv_co_truncate
= raw_co_truncate
,
3547 .bdrv_getlength
= raw_getlength
,
3548 .has_variable_length
= true,
3549 .bdrv_get_allocated_file_size
3550 = raw_get_allocated_file_size
,
3552 /* removable device support */
3553 .bdrv_is_inserted
= cdrom_is_inserted
,
3554 .bdrv_eject
= cdrom_eject
,
3555 .bdrv_lock_medium
= cdrom_lock_medium
,
3557 #endif /* __FreeBSD__ */
3559 static void bdrv_file_init(void)
3562 * Register all the drivers. Note that order is important, the driver
3563 * registered last will get probed first.
3565 bdrv_register(&bdrv_file
);
3566 bdrv_register(&bdrv_host_device
);
3568 bdrv_register(&bdrv_host_cdrom
);
3570 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3571 bdrv_register(&bdrv_host_cdrom
);
3575 block_init(bdrv_file_init
);