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 "qemu-common.h"
27 #include "qapi/error.h"
28 #include "qemu/cutils.h"
29 #include "qemu/error-report.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
34 #include "block/thread-pool.h"
36 #include "block/raw-aio.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qstring.h"
40 #include "scsi/pr-manager.h"
41 #include "scsi/constants.h"
43 #if defined(__APPLE__) && (__MACH__)
45 #include <sys/param.h>
46 #include <IOKit/IOKitLib.h>
47 #include <IOKit/IOBSD.h>
48 #include <IOKit/storage/IOMediaBSDClient.h>
49 #include <IOKit/storage/IOMedia.h>
50 #include <IOKit/storage/IOCDMedia.h>
51 //#include <IOKit/storage/IOCDTypes.h>
52 #include <IOKit/storage/IODVDMedia.h>
53 #include <CoreFoundation/CoreFoundation.h>
57 #define _POSIX_PTHREAD_SEMANTICS 1
61 #include <sys/ioctl.h>
62 #include <sys/param.h>
63 #include <sys/syscall.h>
64 #include <linux/cdrom.h>
67 #include <linux/hdreg.h>
73 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
76 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
77 #include <linux/falloc.h>
79 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
85 #include <sys/ioctl.h>
86 #include <sys/disklabel.h>
91 #include <sys/ioctl.h>
92 #include <sys/disklabel.h>
98 #include <sys/ioctl.h>
99 #include <sys/diskslice.h>
108 /* OS X does not have O_DSYNC */
111 #define O_DSYNC O_SYNC
112 #elif defined(O_FSYNC)
113 #define O_DSYNC O_FSYNC
117 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
119 #define O_DIRECT O_DSYNC
125 #define MAX_BLOCKSIZE 4096
127 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
128 * leaving a few more bytes for its future use. */
129 #define RAW_LOCK_PERM_BASE 100
130 #define RAW_LOCK_SHARED_BASE 200
132 typedef struct BDRVRawState
{
139 /* The current permissions. */
141 uint64_t shared_perm
;
143 /* The perms bits whose corresponding bytes are already locked in
145 uint64_t locked_perm
;
146 uint64_t locked_shared_perm
;
149 int perm_change_flags
;
150 BDRVReopenState
*reopen_state
;
156 bool has_write_zeroes
:1;
157 bool discard_zeroes
:1;
158 bool use_linux_aio
:1;
159 bool page_cache_inconsistent
:1;
161 bool needs_alignment
;
163 bool check_cache_dropped
;
168 typedef struct BDRVRawReopenState
{
172 bool check_cache_dropped
;
173 } BDRVRawReopenState
;
175 static int fd_open(BlockDriverState
*bs
);
176 static int64_t raw_getlength(BlockDriverState
*bs
);
178 typedef struct RawPosixAIOData
{
179 BlockDriverState
*bs
;
200 PreallocMode prealloc
;
206 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
207 static int cdrom_reopen(BlockDriverState
*bs
);
210 #if defined(__NetBSD__)
211 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
213 static char namebuf
[PATH_MAX
];
214 const char *dp
, *fname
;
218 dp
= strrchr(fname
, '/');
219 if (lstat(fname
, &sb
) < 0) {
220 error_setg_errno(errp
, errno
, "%s: stat failed", fname
);
224 if (!S_ISBLK(sb
.st_mode
)) {
229 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
231 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
232 (int)(dp
- fname
), fname
, dp
+ 1);
235 warn_report("%s is a block device, using %s", fname
, *filename
);
240 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
247 * Get logical block size via ioctl. On success store it in @sector_size_p.
249 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
251 unsigned int sector_size
;
252 bool success
= false;
256 static const unsigned long ioctl_list
[] = {
260 #ifdef DKIOCGETBLOCKSIZE
263 #ifdef DIOCGSECTORSIZE
268 /* Try a few ioctls to get the right size */
269 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
270 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
271 *sector_size_p
= sector_size
;
276 return success
? 0 : -errno
;
280 * Get physical block size of @fd.
281 * On success, store it in @blk_size and return 0.
282 * On failure, return -errno.
284 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
287 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
296 /* Check if read is allowed with given memory buffer and length.
298 * This function is used to check O_DIRECT memory buffer and request alignment.
300 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
302 ssize_t ret
= pread(fd
, buf
, len
, 0);
309 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
310 * other errors (e.g. real I/O error), which could happen on a failed
311 * drive, since we only care about probing alignment.
313 if (errno
!= EINVAL
) {
321 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
323 BDRVRawState
*s
= bs
->opaque
;
325 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
327 /* For SCSI generic devices the alignment is not really used.
328 With buffered I/O, we don't have any restrictions. */
329 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
330 bs
->bl
.request_alignment
= 1;
335 bs
->bl
.request_alignment
= 0;
337 /* Let's try to use the logical blocksize for the alignment. */
338 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
339 bs
->bl
.request_alignment
= 0;
344 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
345 bs
->bl
.request_alignment
= da
.d_miniosz
;
346 /* The kernel returns wrong information for d_mem */
347 /* s->buf_align = da.d_mem; */
352 /* If we could not get the sizes so far, we can only guess them */
355 buf
= qemu_memalign(max_align
, 2 * max_align
);
356 for (align
= 512; align
<= max_align
; align
<<= 1) {
357 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
358 s
->buf_align
= align
;
365 if (!bs
->bl
.request_alignment
) {
367 buf
= qemu_memalign(s
->buf_align
, max_align
);
368 for (align
= 512; align
<= max_align
; align
<<= 1) {
369 if (raw_is_io_aligned(fd
, buf
, align
)) {
370 bs
->bl
.request_alignment
= align
;
377 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
378 error_setg(errp
, "Could not find working O_DIRECT alignment");
379 error_append_hint(errp
, "Try cache.direct=off\n");
383 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
385 bool read_write
= false;
386 assert(open_flags
!= NULL
);
388 *open_flags
|= O_BINARY
;
389 *open_flags
&= ~O_ACCMODE
;
391 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
392 read_write
= has_writers
;
393 } else if (bdrv_flags
& BDRV_O_RDWR
) {
398 *open_flags
|= O_RDWR
;
400 *open_flags
|= O_RDONLY
;
403 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
404 * and O_DIRECT for no caching. */
405 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
406 *open_flags
|= O_DIRECT
;
410 static void raw_parse_filename(const char *filename
, QDict
*options
,
413 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
416 static QemuOptsList raw_runtime_opts
= {
418 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
422 .type
= QEMU_OPT_STRING
,
423 .help
= "File name of the image",
427 .type
= QEMU_OPT_STRING
,
428 .help
= "host AIO implementation (threads, native)",
432 .type
= QEMU_OPT_STRING
,
433 .help
= "file locking mode (on/off/auto, default: auto)",
436 .name
= "pr-manager",
437 .type
= QEMU_OPT_STRING
,
438 .help
= "id of persistent reservation manager object (default: none)",
440 #if defined(__linux__)
442 .name
= "drop-cache",
443 .type
= QEMU_OPT_BOOL
,
444 .help
= "invalidate page cache during live migration (default: on)",
448 .name
= "x-check-cache-dropped",
449 .type
= QEMU_OPT_BOOL
,
450 .help
= "check that page cache was dropped on live migration (default: off)"
452 { /* end of list */ }
456 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
458 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
459 int bdrv_flags
, int open_flags
,
460 bool device
, Error
**errp
)
462 BDRVRawState
*s
= bs
->opaque
;
464 Error
*local_err
= NULL
;
465 const char *filename
= NULL
;
467 BlockdevAioOptions aio
, aio_default
;
472 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
473 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
475 error_propagate(errp
, local_err
);
480 filename
= qemu_opt_get(opts
, "filename");
482 ret
= raw_normalize_devicepath(&filename
, errp
);
487 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
488 ? BLOCKDEV_AIO_OPTIONS_NATIVE
489 : BLOCKDEV_AIO_OPTIONS_THREADS
;
490 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
491 qemu_opt_get(opts
, "aio"),
492 aio_default
, &local_err
);
494 error_propagate(errp
, local_err
);
498 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
500 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
501 qemu_opt_get(opts
, "locking"),
502 ON_OFF_AUTO_AUTO
, &local_err
);
504 error_propagate(errp
, local_err
);
511 if (!qemu_has_ofd_lock()) {
512 warn_report("File lock requested but OFD locking syscall is "
513 "unavailable, falling back to POSIX file locks");
514 error_printf("Due to the implementation, locks can be lost "
518 case ON_OFF_AUTO_OFF
:
521 case ON_OFF_AUTO_AUTO
:
522 s
->use_lock
= qemu_has_ofd_lock();
528 str
= qemu_opt_get(opts
, "pr-manager");
530 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
532 error_propagate(errp
, local_err
);
538 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
539 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
542 s
->open_flags
= open_flags
;
543 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
546 fd
= qemu_open(filename
, s
->open_flags
, 0644);
547 ret
= fd
< 0 ? -errno
: 0;
550 error_setg_errno(errp
, -ret
, "Could not open '%s'", filename
);
559 s
->shared_perm
= BLK_PERM_ALL
;
561 #ifdef CONFIG_LINUX_AIO
562 /* Currently Linux does AIO only for files opened with O_DIRECT */
563 if (s
->use_linux_aio
) {
564 if (!(s
->open_flags
& O_DIRECT
)) {
565 error_setg(errp
, "aio=native was specified, but it requires "
566 "cache.direct=on, which was not specified.");
570 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
571 error_prepend(errp
, "Unable to use native AIO: ");
576 if (s
->use_linux_aio
) {
577 error_setg(errp
, "aio=native was specified, but is not supported "
582 #endif /* !defined(CONFIG_LINUX_AIO) */
584 s
->has_discard
= true;
585 s
->has_write_zeroes
= true;
586 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
587 s
->needs_alignment
= true;
590 if (fstat(s
->fd
, &st
) < 0) {
592 error_setg_errno(errp
, errno
, "Could not stat file");
597 if (S_ISBLK(st
.st_mode
)) {
598 warn_report("Opening a block device as a file using the '%s' "
599 "driver is deprecated", bs
->drv
->format_name
);
600 } else if (S_ISCHR(st
.st_mode
)) {
601 warn_report("Opening a character device as a file using the '%s' "
602 "driver is deprecated", bs
->drv
->format_name
);
603 } else if (!S_ISREG(st
.st_mode
)) {
604 error_setg(errp
, "A regular file was expected by the '%s' driver, "
605 "but something else was given", bs
->drv
->format_name
);
609 s
->discard_zeroes
= true;
610 s
->has_fallocate
= true;
613 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
614 error_setg(errp
, "'%s' driver expects either "
615 "a character or block device", bs
->drv
->format_name
);
621 if (S_ISBLK(st
.st_mode
)) {
622 #ifdef BLKDISCARDZEROES
624 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
625 s
->discard_zeroes
= true;
629 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
630 * not rely on the contents of discarded blocks unless using O_DIRECT.
631 * Same for BLKZEROOUT.
633 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
634 s
->discard_zeroes
= false;
635 s
->has_write_zeroes
= false;
640 if (S_ISCHR(st
.st_mode
)) {
642 * The file is a char device (disk), which on FreeBSD isn't behind
643 * a pager, so force all requests to be aligned. This is needed
644 * so QEMU makes sure all IO operations on the device are aligned
645 * to sector size, or else FreeBSD will reject them with EINVAL.
647 s
->needs_alignment
= true;
652 if (platform_test_xfs_fd(s
->fd
)) {
657 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
660 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
667 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
670 BDRVRawState
*s
= bs
->opaque
;
672 s
->type
= FTYPE_FILE
;
673 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
682 #define PERM_FOREACH(i) \
683 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
685 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
686 * file; if @unlock == true, also unlock the unneeded bytes.
687 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
689 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
690 uint64_t perm_lock_bits
,
691 uint64_t shared_perm_lock_bits
,
692 bool unlock
, Error
**errp
)
696 uint64_t locked_perm
, locked_shared_perm
;
699 locked_perm
= s
->locked_perm
;
700 locked_shared_perm
= s
->locked_shared_perm
;
703 * We don't have the previous bits, just lock/unlock for each of the
707 locked_perm
= BLK_PERM_ALL
;
708 locked_shared_perm
= BLK_PERM_ALL
;
711 locked_shared_perm
= 0;
716 int off
= RAW_LOCK_PERM_BASE
+ i
;
717 uint64_t bit
= (1ULL << i
);
718 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
719 ret
= qemu_lock_fd(fd
, off
, 1, false);
721 error_setg(errp
, "Failed to lock byte %d", off
);
724 s
->locked_perm
|= bit
;
726 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
727 ret
= qemu_unlock_fd(fd
, off
, 1);
729 error_setg(errp
, "Failed to unlock byte %d", off
);
732 s
->locked_perm
&= ~bit
;
737 int off
= RAW_LOCK_SHARED_BASE
+ i
;
738 uint64_t bit
= (1ULL << i
);
739 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
740 ret
= qemu_lock_fd(fd
, off
, 1, false);
742 error_setg(errp
, "Failed to lock byte %d", off
);
745 s
->locked_shared_perm
|= bit
;
747 } else if (unlock
&& (locked_shared_perm
& bit
) &&
748 !(shared_perm_lock_bits
& bit
)) {
749 ret
= qemu_unlock_fd(fd
, off
, 1);
751 error_setg(errp
, "Failed to unlock byte %d", off
);
754 s
->locked_shared_perm
&= ~bit
;
761 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
762 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
769 int off
= RAW_LOCK_SHARED_BASE
+ i
;
770 uint64_t p
= 1ULL << i
;
772 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
774 char *perm_name
= bdrv_perm_names(p
);
776 "Failed to get \"%s\" lock",
784 int off
= RAW_LOCK_PERM_BASE
+ i
;
785 uint64_t p
= 1ULL << i
;
786 if (!(shared_perm
& p
)) {
787 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
789 char *perm_name
= bdrv_perm_names(p
);
791 "Failed to get shared \"%s\" lock",
801 static int raw_handle_perm_lock(BlockDriverState
*bs
,
803 uint64_t new_perm
, uint64_t new_shared
,
806 BDRVRawState
*s
= bs
->opaque
;
808 Error
*local_err
= NULL
;
814 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
820 if ((s
->perm
| new_perm
) == s
->perm
&&
821 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
824 * We are going to unlock bytes, it should not fail. If it fail due
825 * to some fs-dependent permission-unrelated reasons (which occurs
826 * sometimes on NFS and leads to abort in bdrv_replace_child) we
827 * can't prevent such errors by any check here. And we ignore them
828 * anyway in ABORT and COMMIT.
832 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
833 ~s
->shared_perm
| ~new_shared
,
836 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
840 error_append_hint(errp
,
841 "Is another process using the image [%s]?\n",
845 /* fall through to unlock bytes. */
847 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
850 /* Theoretically the above call only unlocks bytes and it cannot
851 * fail. Something weird happened, report it.
853 warn_report_err(local_err
);
857 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
860 /* Theoretically the above call only unlocks bytes and it cannot
861 * fail. Something weird happened, report it.
863 warn_report_err(local_err
);
870 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
871 int *open_flags
, uint64_t perm
, bool force_dup
,
874 BDRVRawState
*s
= bs
->opaque
;
877 bool has_writers
= perm
&
878 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
879 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
881 fcntl_flags
|= O_NOATIME
;
885 if (s
->type
== FTYPE_CD
) {
886 *open_flags
|= O_NONBLOCK
;
889 raw_parse_flags(flags
, open_flags
, has_writers
);
892 /* Not all operating systems have O_ASYNC, and those that don't
893 * will not let us track the state into rs->open_flags (typically
894 * you achieve the same effect with an ioctl, for example I_SETSIG
895 * on Solaris). But we do not use O_ASYNC, so that's fine.
897 assert((s
->open_flags
& O_ASYNC
) == 0);
900 if (!force_dup
&& *open_flags
== s
->open_flags
) {
901 /* We're lucky, the existing fd is fine */
905 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
906 /* dup the original fd */
907 fd
= qemu_dup(s
->fd
);
909 ret
= fcntl_setfl(fd
, *open_flags
);
917 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
919 const char *normalized_filename
= bs
->filename
;
920 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
922 assert(!(*open_flags
& O_CREAT
));
923 fd
= qemu_open(normalized_filename
, *open_flags
);
925 error_setg_errno(errp
, errno
, "Could not reopen file");
934 static int raw_reopen_prepare(BDRVReopenState
*state
,
935 BlockReopenQueue
*queue
, Error
**errp
)
938 BDRVRawReopenState
*rs
;
941 Error
*local_err
= NULL
;
943 assert(state
!= NULL
);
944 assert(state
->bs
!= NULL
);
946 s
= state
->bs
->opaque
;
948 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
951 /* Handle options changes */
952 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
953 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
955 error_propagate(errp
, local_err
);
960 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
961 rs
->check_cache_dropped
=
962 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
964 /* This driver's reopen function doesn't currently allow changing
965 * other options, so let's put them back in the original QDict and
966 * bdrv_reopen_prepare() will detect changes and complain. */
967 qemu_opts_to_qdict(opts
, state
->options
);
969 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
970 state
->perm
, true, &local_err
);
972 error_propagate(errp
, local_err
);
977 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
978 * alignment with the new fd. */
980 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
982 error_propagate(errp
, local_err
);
988 s
->reopen_state
= state
;
1000 static void raw_reopen_commit(BDRVReopenState
*state
)
1002 BDRVRawReopenState
*rs
= state
->opaque
;
1003 BDRVRawState
*s
= state
->bs
->opaque
;
1005 s
->drop_cache
= rs
->drop_cache
;
1006 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1007 s
->open_flags
= rs
->open_flags
;
1012 g_free(state
->opaque
);
1013 state
->opaque
= NULL
;
1015 assert(s
->reopen_state
== state
);
1016 s
->reopen_state
= NULL
;
1020 static void raw_reopen_abort(BDRVReopenState
*state
)
1022 BDRVRawReopenState
*rs
= state
->opaque
;
1023 BDRVRawState
*s
= state
->bs
->opaque
;
1025 /* nothing to do if NULL, we didn't get far enough */
1034 g_free(state
->opaque
);
1035 state
->opaque
= NULL
;
1037 assert(s
->reopen_state
== state
);
1038 s
->reopen_state
= NULL
;
1041 static int sg_get_max_transfer_length(int fd
)
1046 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1056 static int sg_get_max_segments(int fd
)
1061 char *sysfspath
= NULL
;
1067 if (fstat(fd
, &st
)) {
1072 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1073 major(st
.st_rdev
), minor(st
.st_rdev
));
1074 sysfd
= open(sysfspath
, O_RDONLY
);
1080 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1081 } while (ret
== -1 && errno
== EINTR
);
1085 } else if (ret
== 0) {
1090 /* The file is ended with '\n', pass 'end' to accept that. */
1091 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1092 if (ret
== 0 && end
&& *end
== '\n') {
1107 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1109 BDRVRawState
*s
= bs
->opaque
;
1112 int ret
= sg_get_max_transfer_length(s
->fd
);
1114 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1115 bs
->bl
.max_transfer
= pow2floor(ret
);
1118 ret
= sg_get_max_segments(s
->fd
);
1120 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
, ret
* getpagesize());
1124 raw_probe_alignment(bs
, s
->fd
, errp
);
1125 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1126 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1129 static int check_for_dasd(int fd
)
1132 struct dasd_information2_t info
= {0};
1134 return ioctl(fd
, BIODASDINFO2
, &info
);
1141 * Try to get @bs's logical and physical block size.
1142 * On success, store them in @bsz and return zero.
1143 * On failure, return negative errno.
1145 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1147 BDRVRawState
*s
= bs
->opaque
;
1150 /* If DASD, get blocksizes */
1151 if (check_for_dasd(s
->fd
) < 0) {
1154 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1158 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1162 * Try to get @bs's geometry: cyls, heads, sectors.
1163 * On success, store them in @geo and return 0.
1164 * On failure return -errno.
1165 * (Allows block driver to assign default geometry values that guest sees)
1168 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1170 BDRVRawState
*s
= bs
->opaque
;
1171 struct hd_geometry ioctl_geo
= {0};
1173 /* If DASD, get its geometry */
1174 if (check_for_dasd(s
->fd
) < 0) {
1177 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1180 /* HDIO_GETGEO may return success even though geo contains zeros
1181 (e.g. certain multipath setups) */
1182 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1185 /* Do not return a geometry for partition */
1186 if (ioctl_geo
.start
!= 0) {
1189 geo
->heads
= ioctl_geo
.heads
;
1190 geo
->sectors
= ioctl_geo
.sectors
;
1191 geo
->cylinders
= ioctl_geo
.cylinders
;
1195 #else /* __linux__ */
1196 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1202 #if defined(__linux__)
1203 static int handle_aiocb_ioctl(void *opaque
)
1205 RawPosixAIOData
*aiocb
= opaque
;
1208 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1217 static int handle_aiocb_flush(void *opaque
)
1219 RawPosixAIOData
*aiocb
= opaque
;
1220 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1223 if (s
->page_cache_inconsistent
) {
1227 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1229 /* There is no clear definition of the semantics of a failing fsync(),
1230 * so we may have to assume the worst. The sad truth is that this
1231 * assumption is correct for Linux. Some pages are now probably marked
1232 * clean in the page cache even though they are inconsistent with the
1233 * on-disk contents. The next fdatasync() call would succeed, but no
1234 * further writeback attempt will be made. We can't get back to a state
1235 * in which we know what is on disk (we would have to rewrite
1236 * everything that was touched since the last fdatasync() at least), so
1237 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1238 * really defined, I have little hope that other OSes are doing better.
1240 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1242 if ((s
->open_flags
& O_DIRECT
) == 0) {
1243 s
->page_cache_inconsistent
= true;
1250 #ifdef CONFIG_PREADV
1252 static bool preadv_present
= true;
1255 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1257 return preadv(fd
, iov
, nr_iov
, offset
);
1261 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1263 return pwritev(fd
, iov
, nr_iov
, offset
);
1268 static bool preadv_present
= false;
1271 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1277 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1284 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1289 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1290 len
= qemu_pwritev(aiocb
->aio_fildes
,
1295 len
= qemu_preadv(aiocb
->aio_fildes
,
1299 } while (len
== -1 && errno
== EINTR
);
1308 * Read/writes the data to/from a given linear buffer.
1310 * Returns the number of bytes handles or -errno in case of an error. Short
1311 * reads are only returned if the end of the file is reached.
1313 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1318 while (offset
< aiocb
->aio_nbytes
) {
1319 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1320 len
= pwrite(aiocb
->aio_fildes
,
1321 (const char *)buf
+ offset
,
1322 aiocb
->aio_nbytes
- offset
,
1323 aiocb
->aio_offset
+ offset
);
1325 len
= pread(aiocb
->aio_fildes
,
1327 aiocb
->aio_nbytes
- offset
,
1328 aiocb
->aio_offset
+ offset
);
1330 if (len
== -1 && errno
== EINTR
) {
1332 } else if (len
== -1 && errno
== EINVAL
&&
1333 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1334 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1336 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1337 * after a short read. Assume that O_DIRECT short reads only occur
1338 * at EOF. Therefore this is a short read, not an I/O error.
1341 } else if (len
== -1) {
1344 } else if (len
== 0) {
1353 static int handle_aiocb_rw(void *opaque
)
1355 RawPosixAIOData
*aiocb
= opaque
;
1359 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1361 * If there is just a single buffer, and it is properly aligned
1362 * we can just use plain pread/pwrite without any problems.
1364 if (aiocb
->io
.niov
== 1) {
1365 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1369 * We have more than one iovec, and all are properly aligned.
1371 * Try preadv/pwritev first and fall back to linearizing the
1372 * buffer if it's not supported.
1374 if (preadv_present
) {
1375 nbytes
= handle_aiocb_rw_vector(aiocb
);
1376 if (nbytes
== aiocb
->aio_nbytes
||
1377 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1380 preadv_present
= false;
1384 * XXX(hch): short read/write. no easy way to handle the reminder
1385 * using these interfaces. For now retry using plain
1391 * Ok, we have to do it the hard way, copy all segments into
1392 * a single aligned buffer.
1394 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1400 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1404 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1405 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1406 p
+= aiocb
->io
.iov
[i
].iov_len
;
1408 assert(p
- buf
== aiocb
->aio_nbytes
);
1411 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1412 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1414 size_t count
= aiocb
->aio_nbytes
, copy
;
1417 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1419 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1420 copy
= aiocb
->io
.iov
[i
].iov_len
;
1422 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1423 assert(count
>= copy
);
1432 if (nbytes
== aiocb
->aio_nbytes
) {
1434 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1435 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1438 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1439 0, aiocb
->aio_nbytes
- nbytes
);
1449 static int xfs_write_zeroes(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1452 struct xfs_flock64 fl
;
1455 len
= lseek(s
->fd
, 0, SEEK_END
);
1460 if (offset
+ bytes
> len
) {
1461 /* XFS_IOC_ZERO_RANGE does not increase the file length */
1462 if (ftruncate(s
->fd
, offset
+ bytes
) < 0) {
1467 memset(&fl
, 0, sizeof(fl
));
1468 fl
.l_whence
= SEEK_SET
;
1469 fl
.l_start
= offset
;
1472 if (xfsctl(NULL
, s
->fd
, XFS_IOC_ZERO_RANGE
, &fl
) < 0) {
1474 trace_file_xfs_write_zeroes(strerror(errno
));
1481 static int xfs_discard(BDRVRawState
*s
, int64_t offset
, uint64_t bytes
)
1483 struct xfs_flock64 fl
;
1486 memset(&fl
, 0, sizeof(fl
));
1487 fl
.l_whence
= SEEK_SET
;
1488 fl
.l_start
= offset
;
1491 if (xfsctl(NULL
, s
->fd
, XFS_IOC_UNRESVSP64
, &fl
) < 0) {
1493 trace_file_xfs_discard(strerror(errno
));
1501 static int translate_err(int err
)
1503 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1510 #ifdef CONFIG_FALLOCATE
1511 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1514 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1517 } while (errno
== EINTR
);
1518 return translate_err(-errno
);
1522 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1525 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1527 if (!s
->has_write_zeroes
) {
1532 /* The BLKZEROOUT implementation in the kernel doesn't set
1533 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1535 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1537 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1538 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1541 } while (errno
== EINTR
);
1543 ret
= translate_err(-errno
);
1547 if (ret
== -ENOTSUP
) {
1548 s
->has_write_zeroes
= false;
1553 static int handle_aiocb_write_zeroes(void *opaque
)
1555 RawPosixAIOData
*aiocb
= opaque
;
1556 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1557 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1559 #ifdef CONFIG_FALLOCATE
1563 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1564 return handle_aiocb_write_zeroes_block(aiocb
);
1569 return xfs_write_zeroes(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1573 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1574 if (s
->has_write_zeroes
) {
1575 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1576 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1577 if (ret
== 0 || ret
!= -ENOTSUP
) {
1580 s
->has_write_zeroes
= false;
1584 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1585 if (s
->has_discard
&& s
->has_fallocate
) {
1586 int ret
= do_fallocate(s
->fd
,
1587 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1588 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1590 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1591 if (ret
== 0 || ret
!= -ENOTSUP
) {
1594 s
->has_fallocate
= false;
1595 } else if (ret
!= -ENOTSUP
) {
1598 s
->has_discard
= false;
1603 #ifdef CONFIG_FALLOCATE
1604 /* Last resort: we are trying to extend the file with zeroed data. This
1605 * can be done via fallocate(fd, 0) */
1606 len
= bdrv_getlength(aiocb
->bs
);
1607 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1608 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1609 if (ret
== 0 || ret
!= -ENOTSUP
) {
1612 s
->has_fallocate
= false;
1619 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1621 RawPosixAIOData
*aiocb
= opaque
;
1622 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1625 /* First try to write zeros and unmap at the same time */
1627 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1628 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1629 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1630 if (ret
!= -ENOTSUP
) {
1637 /* xfs_discard() guarantees that the discarded area reads as all-zero
1638 * afterwards, so we can use it here. */
1639 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1643 /* If we couldn't manage to unmap while guaranteed that the area reads as
1644 * all-zero afterwards, just write zeroes without unmapping */
1645 ret
= handle_aiocb_write_zeroes(aiocb
);
1649 #ifndef HAVE_COPY_FILE_RANGE
1651 /* Avoid conflict with system header file. */
1652 #define copy_file_range(a, b, c, d, e, f) local_copy_file_range(a, b, c, d, e, f)
1654 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1655 off_t
*out_off
, size_t len
, unsigned int flags
)
1657 #ifdef __NR_copy_file_range
1658 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1659 out_off
, len
, flags
);
1667 static int handle_aiocb_copy_range(void *opaque
)
1669 RawPosixAIOData
*aiocb
= opaque
;
1670 uint64_t bytes
= aiocb
->aio_nbytes
;
1671 off_t in_off
= aiocb
->aio_offset
;
1672 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1675 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1676 aiocb
->copy_range
.aio_fd2
, &out_off
,
1678 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1679 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1682 /* No progress (e.g. when beyond EOF), let the caller fall back to
1701 static int handle_aiocb_discard(void *opaque
)
1703 RawPosixAIOData
*aiocb
= opaque
;
1704 int ret
= -EOPNOTSUPP
;
1705 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1707 if (!s
->has_discard
) {
1711 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1714 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1715 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1718 } while (errno
== EINTR
);
1725 return xfs_discard(s
, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1729 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1730 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1731 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1735 ret
= translate_err(ret
);
1736 if (ret
== -ENOTSUP
) {
1737 s
->has_discard
= false;
1742 static int handle_aiocb_truncate(void *opaque
)
1744 RawPosixAIOData
*aiocb
= opaque
;
1746 int64_t current_length
= 0;
1749 int fd
= aiocb
->aio_fildes
;
1750 int64_t offset
= aiocb
->aio_offset
;
1751 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1752 Error
**errp
= aiocb
->truncate
.errp
;
1754 if (fstat(fd
, &st
) < 0) {
1756 error_setg_errno(errp
, -result
, "Could not stat file");
1760 current_length
= st
.st_size
;
1761 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1762 error_setg(errp
, "Cannot use preallocation for shrinking files");
1767 #ifdef CONFIG_POSIX_FALLOCATE
1768 case PREALLOC_MODE_FALLOC
:
1770 * Truncating before posix_fallocate() makes it about twice slower on
1771 * file systems that do not support fallocate(), trying to check if a
1772 * block is allocated before allocating it, so don't do that here.
1774 if (offset
!= current_length
) {
1775 result
= -posix_fallocate(fd
, current_length
,
1776 offset
- current_length
);
1778 /* posix_fallocate() doesn't set errno. */
1779 error_setg_errno(errp
, -result
,
1780 "Could not preallocate new data");
1787 case PREALLOC_MODE_FULL
:
1789 int64_t num
= 0, left
= offset
- current_length
;
1793 * Knowing the final size from the beginning could allow the file
1794 * system driver to do less allocations and possibly avoid
1795 * fragmentation of the file.
1797 if (ftruncate(fd
, offset
) != 0) {
1799 error_setg_errno(errp
, -result
, "Could not resize file");
1803 buf
= g_malloc0(65536);
1805 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1806 if (seek_result
< 0) {
1808 error_setg_errno(errp
, -result
,
1809 "Failed to seek to the old end of file");
1814 num
= MIN(left
, 65536);
1815 result
= write(fd
, buf
, num
);
1817 if (errno
== EINTR
) {
1821 error_setg_errno(errp
, -result
,
1822 "Could not write zeros for preallocation");
1831 error_setg_errno(errp
, -result
,
1832 "Could not flush file to disk");
1838 case PREALLOC_MODE_OFF
:
1839 if (ftruncate(fd
, offset
) != 0) {
1841 error_setg_errno(errp
, -result
, "Could not resize file");
1846 error_setg(errp
, "Unsupported preallocation mode: %s",
1847 PreallocMode_str(prealloc
));
1853 if (ftruncate(fd
, current_length
) < 0) {
1854 error_report("Failed to restore old file length: %s",
1863 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1864 ThreadPoolFunc func
, void *arg
)
1866 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1867 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1868 return thread_pool_submit_co(pool
, func
, arg
);
1871 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1872 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1874 BDRVRawState
*s
= bs
->opaque
;
1875 RawPosixAIOData acb
;
1877 if (fd_open(bs
) < 0)
1881 * Check if the underlying device requires requests to be aligned,
1882 * and if the request we are trying to submit is aligned or not.
1883 * If this is the case tell the low-level driver that it needs
1884 * to copy the buffer.
1886 if (s
->needs_alignment
) {
1887 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1888 type
|= QEMU_AIO_MISALIGNED
;
1889 #ifdef CONFIG_LINUX_AIO
1890 } else if (s
->use_linux_aio
) {
1891 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1892 assert(qiov
->size
== bytes
);
1893 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1898 acb
= (RawPosixAIOData
) {
1900 .aio_fildes
= s
->fd
,
1902 .aio_offset
= offset
,
1903 .aio_nbytes
= bytes
,
1910 assert(qiov
->size
== bytes
);
1911 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1914 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1915 uint64_t bytes
, QEMUIOVector
*qiov
,
1918 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1921 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1922 uint64_t bytes
, QEMUIOVector
*qiov
,
1926 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1929 static void raw_aio_plug(BlockDriverState
*bs
)
1931 #ifdef CONFIG_LINUX_AIO
1932 BDRVRawState
*s
= bs
->opaque
;
1933 if (s
->use_linux_aio
) {
1934 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1935 laio_io_plug(bs
, aio
);
1940 static void raw_aio_unplug(BlockDriverState
*bs
)
1942 #ifdef CONFIG_LINUX_AIO
1943 BDRVRawState
*s
= bs
->opaque
;
1944 if (s
->use_linux_aio
) {
1945 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1946 laio_io_unplug(bs
, aio
);
1951 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1953 BDRVRawState
*s
= bs
->opaque
;
1954 RawPosixAIOData acb
;
1962 acb
= (RawPosixAIOData
) {
1964 .aio_fildes
= s
->fd
,
1965 .aio_type
= QEMU_AIO_FLUSH
,
1968 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1971 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1972 AioContext
*new_context
)
1974 #ifdef CONFIG_LINUX_AIO
1975 BDRVRawState
*s
= bs
->opaque
;
1976 if (s
->use_linux_aio
) {
1978 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1979 error_reportf_err(local_err
, "Unable to use native AIO, "
1980 "falling back to thread pool: ");
1981 s
->use_linux_aio
= false;
1987 static void raw_close(BlockDriverState
*bs
)
1989 BDRVRawState
*s
= bs
->opaque
;
1998 * Truncates the given regular file @fd to @offset and, when growing, fills the
1999 * new space according to @prealloc.
2001 * Returns: 0 on success, -errno on failure.
2003 static int coroutine_fn
2004 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2005 PreallocMode prealloc
, Error
**errp
)
2007 RawPosixAIOData acb
;
2009 acb
= (RawPosixAIOData
) {
2012 .aio_type
= QEMU_AIO_TRUNCATE
,
2013 .aio_offset
= offset
,
2015 .prealloc
= prealloc
,
2020 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2023 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2024 PreallocMode prealloc
, Error
**errp
)
2026 BDRVRawState
*s
= bs
->opaque
;
2030 if (fstat(s
->fd
, &st
)) {
2032 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2036 if (S_ISREG(st
.st_mode
)) {
2037 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2040 if (prealloc
!= PREALLOC_MODE_OFF
) {
2041 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2042 "non-regular file", PreallocMode_str(prealloc
));
2046 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2047 if (offset
> raw_getlength(bs
)) {
2048 error_setg(errp
, "Cannot grow device files");
2052 error_setg(errp
, "Resizing this file is not supported");
2060 static int64_t raw_getlength(BlockDriverState
*bs
)
2062 BDRVRawState
*s
= bs
->opaque
;
2068 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2069 struct disklabel dl
;
2071 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2073 return (uint64_t)dl
.d_secsize
*
2074 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2078 #elif defined(__NetBSD__)
2079 static int64_t raw_getlength(BlockDriverState
*bs
)
2081 BDRVRawState
*s
= bs
->opaque
;
2087 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2088 struct dkwedge_info dkw
;
2090 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2091 return dkw
.dkw_size
* 512;
2093 struct disklabel dl
;
2095 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2097 return (uint64_t)dl
.d_secsize
*
2098 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2103 #elif defined(__sun__)
2104 static int64_t raw_getlength(BlockDriverState
*bs
)
2106 BDRVRawState
*s
= bs
->opaque
;
2107 struct dk_minfo minfo
;
2117 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2119 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2121 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2125 * There are reports that lseek on some devices fails, but
2126 * irc discussion said that contingency on contingency was overkill.
2128 size
= lseek(s
->fd
, 0, SEEK_END
);
2134 #elif defined(CONFIG_BSD)
2135 static int64_t raw_getlength(BlockDriverState
*bs
)
2137 BDRVRawState
*s
= bs
->opaque
;
2141 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2150 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2153 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2154 #ifdef DIOCGMEDIASIZE
2155 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2156 #elif defined(DIOCGPART)
2159 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2160 size
= pi
.media_size
;
2166 #if defined(__APPLE__) && defined(__MACH__)
2168 uint64_t sectors
= 0;
2169 uint32_t sector_size
= 0;
2171 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2172 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2173 size
= sectors
* sector_size
;
2175 size
= lseek(fd
, 0LL, SEEK_END
);
2182 size
= lseek(fd
, 0LL, SEEK_END
);
2187 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2190 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2191 if (size
== 2048LL * (unsigned)-1)
2193 /* XXX no disc? maybe we need to reopen... */
2194 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2201 size
= lseek(fd
, 0, SEEK_END
);
2209 static int64_t raw_getlength(BlockDriverState
*bs
)
2211 BDRVRawState
*s
= bs
->opaque
;
2220 size
= lseek(s
->fd
, 0, SEEK_END
);
2228 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2231 BDRVRawState
*s
= bs
->opaque
;
2233 if (fstat(s
->fd
, &st
) < 0) {
2236 return (int64_t)st
.st_blocks
* 512;
2239 static int coroutine_fn
2240 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2242 BlockdevCreateOptionsFile
*file_opts
;
2243 Error
*local_err
= NULL
;
2245 uint64_t perm
, shared
;
2248 /* Validate options and set default values */
2249 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2250 file_opts
= &options
->u
.file
;
2252 if (!file_opts
->has_nocow
) {
2253 file_opts
->nocow
= false;
2255 if (!file_opts
->has_preallocation
) {
2256 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2260 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2263 error_setg_errno(errp
, -result
, "Could not create file");
2267 /* Take permissions: We want to discard everything, so we need
2268 * BLK_PERM_WRITE; and truncation to the desired size requires
2270 * On the other hand, we cannot share the RESIZE permission
2271 * because we promise that after this function, the file has the
2272 * size given in the options. If someone else were to resize it
2273 * concurrently, we could not guarantee that.
2274 * Note that after this function, we can no longer guarantee that
2275 * the file is not touched by a third party, so it may be resized
2277 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2278 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2280 /* Step one: Take locks */
2281 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2286 /* Step two: Check that nobody else has taken conflicting locks */
2287 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2289 error_append_hint(errp
,
2290 "Is another process using the image [%s]?\n",
2291 file_opts
->filename
);
2295 /* Clear the file by truncating it to 0 */
2296 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2301 if (file_opts
->nocow
) {
2303 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2304 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2305 * will be ignored since any failure of this operation should not
2306 * block the left work.
2309 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2310 attr
|= FS_NOCOW_FL
;
2311 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2316 /* Resize and potentially preallocate the file to the desired
2318 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2319 file_opts
->preallocation
, errp
);
2325 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2327 /* The above call should not fail, and if it does, that does
2328 * not mean the whole creation operation has failed. So
2329 * report it the user for their convenience, but do not report
2330 * it to the caller. */
2331 warn_report_err(local_err
);
2335 if (qemu_close(fd
) != 0 && result
== 0) {
2337 error_setg_errno(errp
, -result
, "Could not close the new file");
2343 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2346 BlockdevCreateOptions options
;
2347 int64_t total_size
= 0;
2349 PreallocMode prealloc
;
2351 Error
*local_err
= NULL
;
2353 /* Skip file: protocol prefix */
2354 strstart(filename
, "file:", &filename
);
2356 /* Read out options */
2357 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2359 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2360 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2361 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2362 PREALLOC_MODE_OFF
, &local_err
);
2365 error_propagate(errp
, local_err
);
2369 options
= (BlockdevCreateOptions
) {
2370 .driver
= BLOCKDEV_DRIVER_FILE
,
2372 .filename
= (char *) filename
,
2374 .has_preallocation
= true,
2375 .preallocation
= prealloc
,
2380 return raw_co_create(&options
, errp
);
2384 * Find allocation range in @bs around offset @start.
2385 * May change underlying file descriptor's file offset.
2386 * If @start is not in a hole, store @start in @data, and the
2387 * beginning of the next hole in @hole, and return 0.
2388 * If @start is in a non-trailing hole, store @start in @hole and the
2389 * beginning of the next non-hole in @data, and return 0.
2390 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2391 * If we can't find out, return a negative errno other than -ENXIO.
2393 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2394 off_t
*data
, off_t
*hole
)
2396 #if defined SEEK_HOLE && defined SEEK_DATA
2397 BDRVRawState
*s
= bs
->opaque
;
2402 * D1. offs == start: start is in data
2403 * D2. offs > start: start is in a hole, next data at offs
2404 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2405 * or start is beyond EOF
2406 * If the latter happens, the file has been truncated behind
2407 * our back since we opened it. All bets are off then.
2408 * Treating like a trailing hole is simplest.
2409 * D4. offs < 0, errno != ENXIO: we learned nothing
2411 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2413 return -errno
; /* D3 or D4 */
2417 /* This is not a valid return by lseek(). We are safe to just return
2418 * -EIO in this case, and we'll treat it like D4. */
2423 /* D2: in hole, next data at offs */
2429 /* D1: in data, end not yet known */
2433 * H1. offs == start: start is in a hole
2434 * If this happens here, a hole has been dug behind our back
2435 * since the previous lseek().
2436 * H2. offs > start: either start is in data, next hole at offs,
2437 * or start is in trailing hole, EOF at offs
2438 * Linux treats trailing holes like any other hole: offs ==
2439 * start. Solaris seeks to EOF instead: offs > start (blech).
2440 * If that happens here, a hole has been dug behind our back
2441 * since the previous lseek().
2442 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2443 * If this happens, the file has been truncated behind our
2444 * back since we opened it. Treat it like a trailing hole.
2445 * H4. offs < 0, errno != ENXIO: we learned nothing
2446 * Pretend we know nothing at all, i.e. "forget" about D1.
2448 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2450 return -errno
; /* D1 and (H3 or H4) */
2454 /* This is not a valid return by lseek(). We are safe to just return
2455 * -EIO in this case, and we'll treat it like H4. */
2461 * D1 and H2: either in data, next hole at offs, or it was in
2462 * data but is now in a trailing hole. In the latter case,
2463 * all bets are off. Treating it as if it there was data all
2464 * the way to EOF is safe, so simply do that.
2479 * Returns the allocation status of the specified offset.
2481 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2483 * 'pnum' is set to the number of bytes (including and immediately following
2484 * the specified offset) that are known to be in the same
2485 * allocated/unallocated state.
2487 * 'bytes' is the max value 'pnum' should be set to.
2489 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2492 int64_t bytes
, int64_t *pnum
,
2494 BlockDriverState
**file
)
2496 off_t data
= 0, hole
= 0;
2499 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2510 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2513 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2514 if (ret
== -ENXIO
) {
2517 ret
= BDRV_BLOCK_ZERO
;
2518 } else if (ret
< 0) {
2519 /* No info available, so pretend there are no holes */
2521 ret
= BDRV_BLOCK_DATA
;
2522 } else if (data
== offset
) {
2523 /* On a data extent, compute bytes to the end of the extent,
2524 * possibly including a partial sector at EOF. */
2525 *pnum
= MIN(bytes
, hole
- offset
);
2528 * We are not allowed to return partial sectors, though, so
2529 * round up if necessary.
2531 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2532 int64_t file_length
= raw_getlength(bs
);
2533 if (file_length
> 0) {
2534 /* Ignore errors, this is just a safeguard */
2535 assert(hole
== file_length
);
2537 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2540 ret
= BDRV_BLOCK_DATA
;
2542 /* On a hole, compute bytes to the beginning of the next extent. */
2543 assert(hole
== offset
);
2544 *pnum
= MIN(bytes
, data
- offset
);
2545 ret
= BDRV_BLOCK_ZERO
;
2549 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2552 #if defined(__linux__)
2553 /* Verify that the file is not in the page cache */
2554 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2556 const size_t window_size
= 128 * 1024 * 1024;
2557 BDRVRawState
*s
= bs
->opaque
;
2558 void *window
= NULL
;
2565 /* mincore(2) page status information requires 1 byte per page */
2566 page_size
= sysconf(_SC_PAGESIZE
);
2567 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2569 end
= raw_getlength(bs
);
2571 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2578 /* Unmap previous window if size has changed */
2579 new_length
= MIN(end
- offset
, window_size
);
2580 if (new_length
!= length
) {
2581 munmap(window
, length
);
2586 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2588 if (new_window
== MAP_FAILED
) {
2589 error_setg_errno(errp
, errno
, "mmap failed");
2593 window
= new_window
;
2594 length
= new_length
;
2596 ret
= mincore(window
, length
, vec
);
2598 error_setg_errno(errp
, errno
, "mincore failed");
2602 vec_end
= DIV_ROUND_UP(length
, page_size
);
2603 for (i
= 0; i
< vec_end
; i
++) {
2605 error_setg(errp
, "page cache still in use!");
2612 munmap(window
, length
);
2617 #endif /* __linux__ */
2619 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2622 BDRVRawState
*s
= bs
->opaque
;
2627 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2631 if (!s
->drop_cache
) {
2635 if (s
->open_flags
& O_DIRECT
) {
2636 return; /* No host kernel page cache */
2639 #if defined(__linux__)
2640 /* This sets the scene for the next syscall... */
2641 ret
= bdrv_co_flush(bs
);
2643 error_setg_errno(errp
, -ret
, "flush failed");
2647 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2648 * process. These limitations are okay because we just fsynced the file,
2649 * we don't use mmap, and the file should not be in use by other processes.
2651 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2652 if (ret
!= 0) { /* the return value is a positive errno */
2653 error_setg_errno(errp
, ret
, "fadvise failed");
2657 if (s
->check_cache_dropped
) {
2658 check_cache_dropped(bs
, errp
);
2660 #else /* __linux__ */
2661 /* Do nothing. Live migration to a remote host with cache.direct=off is
2662 * unsupported on other host operating systems. Cache consistency issues
2663 * may occur but no error is reported here, partly because that's the
2664 * historical behavior and partly because it's hard to differentiate valid
2665 * configurations that should not cause errors.
2667 #endif /* !__linux__ */
2670 static coroutine_fn
int
2671 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2673 BDRVRawState
*s
= bs
->opaque
;
2674 RawPosixAIOData acb
;
2676 acb
= (RawPosixAIOData
) {
2678 .aio_fildes
= s
->fd
,
2679 .aio_type
= QEMU_AIO_DISCARD
,
2680 .aio_offset
= offset
,
2681 .aio_nbytes
= bytes
,
2685 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2688 return raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2691 static coroutine_fn
int
2692 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2694 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2697 static int coroutine_fn
2698 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2699 BdrvRequestFlags flags
, bool blkdev
)
2701 BDRVRawState
*s
= bs
->opaque
;
2702 RawPosixAIOData acb
;
2703 ThreadPoolFunc
*handler
;
2705 acb
= (RawPosixAIOData
) {
2707 .aio_fildes
= s
->fd
,
2708 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2709 .aio_offset
= offset
,
2710 .aio_nbytes
= bytes
,
2714 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2716 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2717 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2720 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2721 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2722 handler
= handle_aiocb_write_zeroes_unmap
;
2724 handler
= handle_aiocb_write_zeroes
;
2727 return raw_thread_pool_submit(bs
, handler
, &acb
);
2730 static int coroutine_fn
raw_co_pwrite_zeroes(
2731 BlockDriverState
*bs
, int64_t offset
,
2732 int bytes
, BdrvRequestFlags flags
)
2734 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2737 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2739 BDRVRawState
*s
= bs
->opaque
;
2741 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2745 static QemuOptsList raw_create_opts
= {
2746 .name
= "raw-create-opts",
2747 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2750 .name
= BLOCK_OPT_SIZE
,
2751 .type
= QEMU_OPT_SIZE
,
2752 .help
= "Virtual disk size"
2755 .name
= BLOCK_OPT_NOCOW
,
2756 .type
= QEMU_OPT_BOOL
,
2757 .help
= "Turn off copy-on-write (valid only on btrfs)"
2760 .name
= BLOCK_OPT_PREALLOC
,
2761 .type
= QEMU_OPT_STRING
,
2762 .help
= "Preallocation mode (allowed values: off"
2763 #ifdef CONFIG_POSIX_FALLOCATE
2768 { /* end of list */ }
2772 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2775 BDRVRawState
*s
= bs
->opaque
;
2776 BDRVRawReopenState
*rs
= NULL
;
2780 if (s
->perm_change_fd
) {
2782 * In the context of reopen, this function may be called several times
2783 * (directly and recursively while change permissions of the parent).
2784 * This is even true for children that don't inherit from the original
2785 * reopen node, so s->reopen_state is not set.
2787 * Ignore all but the first call.
2792 if (s
->reopen_state
) {
2793 /* We already have a new file descriptor to set permissions for */
2794 assert(s
->reopen_state
->perm
== perm
);
2795 assert(s
->reopen_state
->shared_perm
== shared
);
2796 rs
= s
->reopen_state
->opaque
;
2797 s
->perm_change_fd
= rs
->fd
;
2798 s
->perm_change_flags
= rs
->open_flags
;
2800 /* We may need a new fd if auto-read-only switches the mode */
2801 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2805 } else if (ret
!= s
->fd
) {
2806 s
->perm_change_fd
= ret
;
2807 s
->perm_change_flags
= open_flags
;
2811 /* Prepare permissions on old fd to avoid conflicts between old and new,
2812 * but keep everything locked that new will need. */
2813 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2818 /* Copy locks to the new fd */
2819 if (s
->perm_change_fd
) {
2820 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2823 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2830 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2831 qemu_close(s
->perm_change_fd
);
2833 s
->perm_change_fd
= 0;
2837 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2839 BDRVRawState
*s
= bs
->opaque
;
2841 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2842 * called after .bdrv_reopen_commit) */
2843 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
2845 s
->fd
= s
->perm_change_fd
;
2846 s
->open_flags
= s
->perm_change_flags
;
2848 s
->perm_change_fd
= 0;
2850 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2852 s
->shared_perm
= shared
;
2855 static void raw_abort_perm_update(BlockDriverState
*bs
)
2857 BDRVRawState
*s
= bs
->opaque
;
2859 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
2860 * the file descriptor. */
2861 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2862 qemu_close(s
->perm_change_fd
);
2864 s
->perm_change_fd
= 0;
2866 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2869 static int coroutine_fn
raw_co_copy_range_from(
2870 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2871 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2872 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2874 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2875 read_flags
, write_flags
);
2878 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2880 uint64_t src_offset
,
2882 uint64_t dst_offset
,
2884 BdrvRequestFlags read_flags
,
2885 BdrvRequestFlags write_flags
)
2887 RawPosixAIOData acb
;
2888 BDRVRawState
*s
= bs
->opaque
;
2889 BDRVRawState
*src_s
;
2891 assert(dst
->bs
== bs
);
2892 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2896 src_s
= src
->bs
->opaque
;
2897 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2901 acb
= (RawPosixAIOData
) {
2903 .aio_type
= QEMU_AIO_COPY_RANGE
,
2904 .aio_fildes
= src_s
->fd
,
2905 .aio_offset
= src_offset
,
2906 .aio_nbytes
= bytes
,
2909 .aio_offset2
= dst_offset
,
2913 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2916 BlockDriver bdrv_file
= {
2917 .format_name
= "file",
2918 .protocol_name
= "file",
2919 .instance_size
= sizeof(BDRVRawState
),
2920 .bdrv_needs_filename
= true,
2921 .bdrv_probe
= NULL
, /* no probe for protocols */
2922 .bdrv_parse_filename
= raw_parse_filename
,
2923 .bdrv_file_open
= raw_open
,
2924 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2925 .bdrv_reopen_commit
= raw_reopen_commit
,
2926 .bdrv_reopen_abort
= raw_reopen_abort
,
2927 .bdrv_close
= raw_close
,
2928 .bdrv_co_create
= raw_co_create
,
2929 .bdrv_co_create_opts
= raw_co_create_opts
,
2930 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2931 .bdrv_co_block_status
= raw_co_block_status
,
2932 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2933 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2935 .bdrv_co_preadv
= raw_co_preadv
,
2936 .bdrv_co_pwritev
= raw_co_pwritev
,
2937 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2938 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2939 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2940 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2941 .bdrv_refresh_limits
= raw_refresh_limits
,
2942 .bdrv_io_plug
= raw_aio_plug
,
2943 .bdrv_io_unplug
= raw_aio_unplug
,
2944 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2946 .bdrv_co_truncate
= raw_co_truncate
,
2947 .bdrv_getlength
= raw_getlength
,
2948 .bdrv_get_info
= raw_get_info
,
2949 .bdrv_get_allocated_file_size
2950 = raw_get_allocated_file_size
,
2951 .bdrv_check_perm
= raw_check_perm
,
2952 .bdrv_set_perm
= raw_set_perm
,
2953 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2954 .create_opts
= &raw_create_opts
,
2955 .mutable_opts
= mutable_opts
,
2958 /***********************************************/
2961 #if defined(__APPLE__) && defined(__MACH__)
2962 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2963 CFIndex maxPathSize
, int flags
);
2964 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2966 kern_return_t kernResult
= KERN_FAILURE
;
2967 mach_port_t masterPort
;
2968 CFMutableDictionaryRef classesToMatch
;
2969 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2970 char *mediaType
= NULL
;
2972 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2973 if ( KERN_SUCCESS
!= kernResult
) {
2974 printf( "IOMasterPort returned %d\n", kernResult
);
2978 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2979 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2980 if (classesToMatch
== NULL
) {
2981 error_report("IOServiceMatching returned NULL for %s",
2982 matching_array
[index
]);
2985 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2987 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2989 if (kernResult
!= KERN_SUCCESS
) {
2990 error_report("Note: IOServiceGetMatchingServices returned %d",
2995 /* If a match was found, leave the loop */
2996 if (*mediaIterator
!= 0) {
2997 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
2998 mediaType
= g_strdup(matching_array
[index
]);
3005 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3006 CFIndex maxPathSize
, int flags
)
3008 io_object_t nextMedia
;
3009 kern_return_t kernResult
= KERN_FAILURE
;
3011 nextMedia
= IOIteratorNext( mediaIterator
);
3014 CFTypeRef bsdPathAsCFString
;
3015 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3016 if ( bsdPathAsCFString
) {
3017 size_t devPathLength
;
3018 strcpy( bsdPath
, _PATH_DEV
);
3019 if (flags
& BDRV_O_NOCACHE
) {
3020 strcat(bsdPath
, "r");
3022 devPathLength
= strlen( bsdPath
);
3023 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3024 kernResult
= KERN_SUCCESS
;
3026 CFRelease( bsdPathAsCFString
);
3028 IOObjectRelease( nextMedia
);
3034 /* Sets up a real cdrom for use in QEMU */
3035 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3037 int index
, num_of_test_partitions
= 2, fd
;
3038 char test_partition
[MAXPATHLEN
];
3039 bool partition_found
= false;
3041 /* look for a working partition */
3042 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3043 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3045 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
3047 partition_found
= true;
3053 /* if a working partition on the device was not found */
3054 if (partition_found
== false) {
3055 error_setg(errp
, "Failed to find a working partition on disc");
3057 trace_file_setup_cdrom(test_partition
);
3058 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3060 return partition_found
;
3063 /* Prints directions on mounting and unmounting a device */
3064 static void print_unmounting_directions(const char *file_name
)
3066 error_report("If device %s is mounted on the desktop, unmount"
3067 " it first before using it in QEMU", file_name
);
3068 error_report("Command to unmount device: diskutil unmountDisk %s",
3070 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3073 #endif /* defined(__APPLE__) && defined(__MACH__) */
3075 static int hdev_probe_device(const char *filename
)
3079 /* allow a dedicated CD-ROM driver to match with a higher priority */
3080 if (strstart(filename
, "/dev/cdrom", NULL
))
3083 if (stat(filename
, &st
) >= 0 &&
3084 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3091 static int check_hdev_writable(BDRVRawState
*s
)
3093 #if defined(BLKROGET)
3094 /* Linux block devices can be configured "read-only" using blockdev(8).
3095 * This is independent of device node permissions and therefore open(2)
3096 * with O_RDWR succeeds. Actual writes fail with EPERM.
3098 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3099 * check for read-only block devices so that Linux block devices behave
3105 if (fstat(s
->fd
, &st
)) {
3109 if (!S_ISBLK(st
.st_mode
)) {
3113 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3120 #endif /* defined(BLKROGET) */
3124 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3127 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3130 static bool hdev_is_sg(BlockDriverState
*bs
)
3133 #if defined(__linux__)
3135 BDRVRawState
*s
= bs
->opaque
;
3137 struct sg_scsi_id scsiid
;
3141 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3145 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3150 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3152 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3161 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3164 BDRVRawState
*s
= bs
->opaque
;
3165 Error
*local_err
= NULL
;
3168 #if defined(__APPLE__) && defined(__MACH__)
3170 * Caution: while qdict_get_str() is fine, getting non-string types
3171 * would require more care. When @options come from -blockdev or
3172 * blockdev_add, its members are typed according to the QAPI
3173 * schema, but when they come from -drive, they're all QString.
3175 const char *filename
= qdict_get_str(options
, "filename");
3176 char bsd_path
[MAXPATHLEN
] = "";
3177 bool error_occurred
= false;
3179 /* If using a real cdrom */
3180 if (strcmp(filename
, "/dev/cdrom") == 0) {
3181 char *mediaType
= NULL
;
3182 kern_return_t ret_val
;
3183 io_iterator_t mediaIterator
= 0;
3185 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3186 if (mediaType
== NULL
) {
3187 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3189 error_occurred
= true;
3190 goto hdev_open_Mac_error
;
3193 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3194 if (ret_val
!= KERN_SUCCESS
) {
3195 error_setg(errp
, "Could not get BSD path for optical drive");
3196 error_occurred
= true;
3197 goto hdev_open_Mac_error
;
3200 /* If a real optical drive was not found */
3201 if (bsd_path
[0] == '\0') {
3202 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3203 error_occurred
= true;
3204 goto hdev_open_Mac_error
;
3207 /* If using a cdrom disc and finding a partition on the disc failed */
3208 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3209 setup_cdrom(bsd_path
, errp
) == false) {
3210 print_unmounting_directions(bsd_path
);
3211 error_occurred
= true;
3212 goto hdev_open_Mac_error
;
3215 qdict_put_str(options
, "filename", bsd_path
);
3217 hdev_open_Mac_error
:
3219 if (mediaIterator
) {
3220 IOObjectRelease(mediaIterator
);
3222 if (error_occurred
) {
3226 #endif /* defined(__APPLE__) && defined(__MACH__) */
3228 s
->type
= FTYPE_FILE
;
3230 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3232 error_propagate(errp
, local_err
);
3233 #if defined(__APPLE__) && defined(__MACH__)
3235 filename
= bsd_path
;
3237 /* if a physical device experienced an error while being opened */
3238 if (strncmp(filename
, "/dev/", 5) == 0) {
3239 print_unmounting_directions(filename
);
3241 #endif /* defined(__APPLE__) && defined(__MACH__) */
3245 /* Since this does ioctl the device must be already opened */
3246 bs
->sg
= hdev_is_sg(bs
);
3248 if (flags
& BDRV_O_RDWR
) {
3249 ret
= check_hdev_writable(s
);
3252 error_setg_errno(errp
, -ret
, "The device is not writable");
3260 #if defined(__linux__)
3261 static int coroutine_fn
3262 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3264 BDRVRawState
*s
= bs
->opaque
;
3265 RawPosixAIOData acb
;
3273 if (req
== SG_IO
&& s
->pr_mgr
) {
3274 struct sg_io_hdr
*io_hdr
= buf
;
3275 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3276 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3277 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3282 acb
= (RawPosixAIOData
) {
3284 .aio_type
= QEMU_AIO_IOCTL
,
3285 .aio_fildes
= s
->fd
,
3293 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3297 static int fd_open(BlockDriverState
*bs
)
3299 BDRVRawState
*s
= bs
->opaque
;
3301 /* this is just to ensure s->fd is sane (its called by io ops) */
3307 static coroutine_fn
int
3308 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3316 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3319 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3320 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3329 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3332 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3337 struct stat stat_buf
;
3338 int64_t total_size
= 0;
3341 /* This function is used by both protocol block drivers and therefore either
3342 * of these prefixes may be given.
3343 * The return value has to be stored somewhere, otherwise this is an error
3344 * due to -Werror=unused-value. */
3346 strstart(filename
, "host_device:", &filename
) ||
3347 strstart(filename
, "host_cdrom:" , &filename
);
3351 ret
= raw_normalize_devicepath(&filename
, errp
);
3356 /* Read out options */
3357 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3360 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3363 error_setg_errno(errp
, -ret
, "Could not open device");
3367 if (fstat(fd
, &stat_buf
) < 0) {
3369 error_setg_errno(errp
, -ret
, "Could not stat device");
3370 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3372 "The given file is neither a block nor a character device");
3374 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3375 error_setg(errp
, "Device is too small");
3379 if (!ret
&& total_size
) {
3380 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3381 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3382 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3385 ret
= qemu_write_full(fd
, buf
, zero_size
);
3386 ret
= ret
== zero_size
? 0 : -errno
;
3393 static BlockDriver bdrv_host_device
= {
3394 .format_name
= "host_device",
3395 .protocol_name
= "host_device",
3396 .instance_size
= sizeof(BDRVRawState
),
3397 .bdrv_needs_filename
= true,
3398 .bdrv_probe_device
= hdev_probe_device
,
3399 .bdrv_parse_filename
= hdev_parse_filename
,
3400 .bdrv_file_open
= hdev_open
,
3401 .bdrv_close
= raw_close
,
3402 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3403 .bdrv_reopen_commit
= raw_reopen_commit
,
3404 .bdrv_reopen_abort
= raw_reopen_abort
,
3405 .bdrv_co_create_opts
= hdev_co_create_opts
,
3406 .create_opts
= &raw_create_opts
,
3407 .mutable_opts
= mutable_opts
,
3408 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3409 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3411 .bdrv_co_preadv
= raw_co_preadv
,
3412 .bdrv_co_pwritev
= raw_co_pwritev
,
3413 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3414 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3415 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3416 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3417 .bdrv_refresh_limits
= raw_refresh_limits
,
3418 .bdrv_io_plug
= raw_aio_plug
,
3419 .bdrv_io_unplug
= raw_aio_unplug
,
3420 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3422 .bdrv_co_truncate
= raw_co_truncate
,
3423 .bdrv_getlength
= raw_getlength
,
3424 .bdrv_get_info
= raw_get_info
,
3425 .bdrv_get_allocated_file_size
3426 = raw_get_allocated_file_size
,
3427 .bdrv_check_perm
= raw_check_perm
,
3428 .bdrv_set_perm
= raw_set_perm
,
3429 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3430 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3431 .bdrv_probe_geometry
= hdev_probe_geometry
,
3433 /* generic scsi device */
3435 .bdrv_co_ioctl
= hdev_co_ioctl
,
3439 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3440 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3443 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3448 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3451 BDRVRawState
*s
= bs
->opaque
;
3455 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3456 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3459 static int cdrom_probe_device(const char *filename
)
3465 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3469 ret
= fstat(fd
, &st
);
3470 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3474 /* Attempt to detect via a CDROM specific ioctl */
3475 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3485 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3487 BDRVRawState
*s
= bs
->opaque
;
3490 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3491 return ret
== CDS_DISC_OK
;
3494 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3496 BDRVRawState
*s
= bs
->opaque
;
3499 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3500 perror("CDROMEJECT");
3502 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3503 perror("CDROMEJECT");
3507 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3509 BDRVRawState
*s
= bs
->opaque
;
3511 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3513 * Note: an error can happen if the distribution automatically
3516 /* perror("CDROM_LOCKDOOR"); */
3520 static BlockDriver bdrv_host_cdrom
= {
3521 .format_name
= "host_cdrom",
3522 .protocol_name
= "host_cdrom",
3523 .instance_size
= sizeof(BDRVRawState
),
3524 .bdrv_needs_filename
= true,
3525 .bdrv_probe_device
= cdrom_probe_device
,
3526 .bdrv_parse_filename
= cdrom_parse_filename
,
3527 .bdrv_file_open
= cdrom_open
,
3528 .bdrv_close
= raw_close
,
3529 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3530 .bdrv_reopen_commit
= raw_reopen_commit
,
3531 .bdrv_reopen_abort
= raw_reopen_abort
,
3532 .bdrv_co_create_opts
= hdev_co_create_opts
,
3533 .create_opts
= &raw_create_opts
,
3534 .mutable_opts
= mutable_opts
,
3535 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
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 /* generic scsi device */
3558 .bdrv_co_ioctl
= hdev_co_ioctl
,
3560 #endif /* __linux__ */
3562 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3563 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3566 BDRVRawState
*s
= bs
->opaque
;
3567 Error
*local_err
= NULL
;
3572 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3574 error_propagate(errp
, local_err
);
3578 /* make sure the door isn't locked at this time */
3579 ioctl(s
->fd
, CDIOCALLOW
);
3583 static int cdrom_probe_device(const char *filename
)
3585 if (strstart(filename
, "/dev/cd", NULL
) ||
3586 strstart(filename
, "/dev/acd", NULL
))
3591 static int cdrom_reopen(BlockDriverState
*bs
)
3593 BDRVRawState
*s
= bs
->opaque
;
3597 * Force reread of possibly changed/newly loaded disc,
3598 * FreeBSD seems to not notice sometimes...
3602 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3609 /* make sure the door isn't locked at this time */
3610 ioctl(s
->fd
, CDIOCALLOW
);
3614 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3616 return raw_getlength(bs
) > 0;
3619 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3621 BDRVRawState
*s
= bs
->opaque
;
3626 (void) ioctl(s
->fd
, CDIOCALLOW
);
3629 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3630 perror("CDIOCEJECT");
3632 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3633 perror("CDIOCCLOSE");
3639 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3641 BDRVRawState
*s
= bs
->opaque
;
3645 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3647 * Note: an error can happen if the distribution automatically
3650 /* perror("CDROM_LOCKDOOR"); */
3654 static BlockDriver bdrv_host_cdrom
= {
3655 .format_name
= "host_cdrom",
3656 .protocol_name
= "host_cdrom",
3657 .instance_size
= sizeof(BDRVRawState
),
3658 .bdrv_needs_filename
= true,
3659 .bdrv_probe_device
= cdrom_probe_device
,
3660 .bdrv_parse_filename
= cdrom_parse_filename
,
3661 .bdrv_file_open
= cdrom_open
,
3662 .bdrv_close
= raw_close
,
3663 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3664 .bdrv_reopen_commit
= raw_reopen_commit
,
3665 .bdrv_reopen_abort
= raw_reopen_abort
,
3666 .bdrv_co_create_opts
= hdev_co_create_opts
,
3667 .create_opts
= &raw_create_opts
,
3668 .mutable_opts
= mutable_opts
,
3670 .bdrv_co_preadv
= raw_co_preadv
,
3671 .bdrv_co_pwritev
= raw_co_pwritev
,
3672 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3673 .bdrv_refresh_limits
= raw_refresh_limits
,
3674 .bdrv_io_plug
= raw_aio_plug
,
3675 .bdrv_io_unplug
= raw_aio_unplug
,
3676 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3678 .bdrv_co_truncate
= raw_co_truncate
,
3679 .bdrv_getlength
= raw_getlength
,
3680 .has_variable_length
= true,
3681 .bdrv_get_allocated_file_size
3682 = raw_get_allocated_file_size
,
3684 /* removable device support */
3685 .bdrv_is_inserted
= cdrom_is_inserted
,
3686 .bdrv_eject
= cdrom_eject
,
3687 .bdrv_lock_medium
= cdrom_lock_medium
,
3689 #endif /* __FreeBSD__ */
3691 static void bdrv_file_init(void)
3694 * Register all the drivers. Note that order is important, the driver
3695 * registered last will get probed first.
3697 bdrv_register(&bdrv_file
);
3698 bdrv_register(&bdrv_host_device
);
3700 bdrv_register(&bdrv_host_cdrom
);
3702 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3703 bdrv_register(&bdrv_host_cdrom
);
3707 block_init(bdrv_file_init
);