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"
33 #include "qemu/units.h"
35 #include "block/thread-pool.h"
37 #include "block/raw-aio.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qstring.h"
41 #include "scsi/pr-manager.h"
42 #include "scsi/constants.h"
44 #if defined(__APPLE__) && (__MACH__)
46 #include <sys/param.h>
47 #include <IOKit/IOKitLib.h>
48 #include <IOKit/IOBSD.h>
49 #include <IOKit/storage/IOMediaBSDClient.h>
50 #include <IOKit/storage/IOMedia.h>
51 #include <IOKit/storage/IOCDMedia.h>
52 //#include <IOKit/storage/IOCDTypes.h>
53 #include <IOKit/storage/IODVDMedia.h>
54 #include <CoreFoundation/CoreFoundation.h>
58 #define _POSIX_PTHREAD_SEMANTICS 1
62 #include <sys/ioctl.h>
63 #include <sys/param.h>
64 #include <sys/syscall.h>
66 #include <linux/cdrom.h>
69 #include <linux/hdreg.h>
70 #include <linux/magic.h>
76 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
79 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
80 #include <linux/falloc.h>
82 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
88 #include <sys/ioctl.h>
89 #include <sys/disklabel.h>
94 #include <sys/ioctl.h>
95 #include <sys/disklabel.h>
101 #include <sys/ioctl.h>
102 #include <sys/diskslice.h>
111 /* OS X does not have O_DSYNC */
114 #define O_DSYNC O_SYNC
115 #elif defined(O_FSYNC)
116 #define O_DSYNC O_FSYNC
120 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
122 #define O_DIRECT O_DSYNC
128 #define MAX_BLOCKSIZE 4096
130 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
131 * leaving a few more bytes for its future use. */
132 #define RAW_LOCK_PERM_BASE 100
133 #define RAW_LOCK_SHARED_BASE 200
135 typedef struct BDRVRawState
{
142 /* The current permissions. */
144 uint64_t shared_perm
;
146 /* The perms bits whose corresponding bytes are already locked in
148 uint64_t locked_perm
;
149 uint64_t locked_shared_perm
;
152 int perm_change_flags
;
153 BDRVReopenState
*reopen_state
;
159 bool has_write_zeroes
:1;
160 bool discard_zeroes
:1;
161 bool use_linux_aio
:1;
162 bool use_linux_io_uring
:1;
163 bool page_cache_inconsistent
:1;
165 bool needs_alignment
;
167 bool check_cache_dropped
;
169 uint64_t discard_nb_ok
;
170 uint64_t discard_nb_failed
;
171 uint64_t discard_bytes_ok
;
177 typedef struct BDRVRawReopenState
{
181 bool check_cache_dropped
;
182 } BDRVRawReopenState
;
184 static int fd_open(BlockDriverState
*bs
);
185 static int64_t raw_getlength(BlockDriverState
*bs
);
187 typedef struct RawPosixAIOData
{
188 BlockDriverState
*bs
;
209 PreallocMode prealloc
;
215 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
216 static int cdrom_reopen(BlockDriverState
*bs
);
220 * Elide EAGAIN and EACCES details when failing to lock, as this
221 * indicates that the specified file region is already locked by
222 * another process, which is considered a common scenario.
224 #define raw_lock_error_setg_errno(errp, err, fmt, ...) \
226 if ((err) == EAGAIN || (err) == EACCES) { \
227 error_setg((errp), (fmt), ## __VA_ARGS__); \
229 error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \
233 #if defined(__NetBSD__)
234 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
236 static char namebuf
[PATH_MAX
];
237 const char *dp
, *fname
;
241 dp
= strrchr(fname
, '/');
242 if (lstat(fname
, &sb
) < 0) {
243 error_setg_file_open(errp
, errno
, fname
);
247 if (!S_ISBLK(sb
.st_mode
)) {
252 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
254 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
255 (int)(dp
- fname
), fname
, dp
+ 1);
258 warn_report("%s is a block device, using %s", fname
, *filename
);
263 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
270 * Get logical block size via ioctl. On success store it in @sector_size_p.
272 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
274 unsigned int sector_size
;
275 bool success
= false;
279 static const unsigned long ioctl_list
[] = {
283 #ifdef DKIOCGETBLOCKSIZE
286 #ifdef DIOCGSECTORSIZE
291 /* Try a few ioctls to get the right size */
292 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
293 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
294 *sector_size_p
= sector_size
;
299 return success
? 0 : -errno
;
303 * Get physical block size of @fd.
304 * On success, store it in @blk_size and return 0.
305 * On failure, return -errno.
307 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
310 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
320 * Returns true if no alignment restrictions are necessary even for files
321 * opened with O_DIRECT.
323 * raw_probe_alignment() probes the required alignment and assume that 1 means
324 * the probing failed, so it falls back to a safe default of 4k. This can be
325 * avoided if we know that byte alignment is okay for the file.
327 static bool dio_byte_aligned(int fd
)
333 ret
= fstatfs(fd
, &buf
);
334 if (ret
== 0 && buf
.f_type
== NFS_SUPER_MAGIC
) {
341 /* Check if read is allowed with given memory buffer and length.
343 * This function is used to check O_DIRECT memory buffer and request alignment.
345 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
347 ssize_t ret
= pread(fd
, buf
, len
, 0);
354 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
355 * other errors (e.g. real I/O error), which could happen on a failed
356 * drive, since we only care about probing alignment.
358 if (errno
!= EINVAL
) {
366 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
368 BDRVRawState
*s
= bs
->opaque
;
370 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size
);
371 size_t alignments
[] = {1, 512, 1024, 2048, 4096};
373 /* For SCSI generic devices the alignment is not really used.
374 With buffered I/O, we don't have any restrictions. */
375 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
376 bs
->bl
.request_alignment
= 1;
381 bs
->bl
.request_alignment
= 0;
383 /* Let's try to use the logical blocksize for the alignment. */
384 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
385 bs
->bl
.request_alignment
= 0;
390 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
391 bs
->bl
.request_alignment
= da
.d_miniosz
;
392 /* The kernel returns wrong information for d_mem */
393 /* s->buf_align = da.d_mem; */
399 * If we could not get the sizes so far, we can only guess them. First try
400 * to detect request alignment, since it is more likely to succeed. Then
401 * try to detect buf_align, which cannot be detected in some cases (e.g.
402 * Gluster). If buf_align cannot be detected, we fallback to the value of
406 if (!bs
->bl
.request_alignment
) {
409 buf
= qemu_memalign(max_align
, max_align
);
410 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
411 align
= alignments
[i
];
412 if (raw_is_io_aligned(fd
, buf
, align
)) {
413 /* Fallback to safe value. */
414 bs
->bl
.request_alignment
= (align
!= 1) ? align
: max_align
;
424 buf
= qemu_memalign(max_align
, 2 * max_align
);
425 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
426 align
= alignments
[i
];
427 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
428 /* Fallback to request_alignment. */
429 s
->buf_align
= (align
!= 1) ? align
: bs
->bl
.request_alignment
;
436 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
437 error_setg(errp
, "Could not find working O_DIRECT alignment");
438 error_append_hint(errp
, "Try cache.direct=off\n");
442 static int check_hdev_writable(int fd
)
444 #if defined(BLKROGET)
445 /* Linux block devices can be configured "read-only" using blockdev(8).
446 * This is independent of device node permissions and therefore open(2)
447 * with O_RDWR succeeds. Actual writes fail with EPERM.
449 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
450 * check for read-only block devices so that Linux block devices behave
456 if (fstat(fd
, &st
)) {
460 if (!S_ISBLK(st
.st_mode
)) {
464 if (ioctl(fd
, BLKROGET
, &readonly
) < 0) {
471 #endif /* defined(BLKROGET) */
475 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
477 bool read_write
= false;
478 assert(open_flags
!= NULL
);
480 *open_flags
|= O_BINARY
;
481 *open_flags
&= ~O_ACCMODE
;
483 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
484 read_write
= has_writers
;
485 } else if (bdrv_flags
& BDRV_O_RDWR
) {
490 *open_flags
|= O_RDWR
;
492 *open_flags
|= O_RDONLY
;
495 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
496 * and O_DIRECT for no caching. */
497 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
498 *open_flags
|= O_DIRECT
;
502 static void raw_parse_filename(const char *filename
, QDict
*options
,
505 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
508 static QemuOptsList raw_runtime_opts
= {
510 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
514 .type
= QEMU_OPT_STRING
,
515 .help
= "File name of the image",
519 .type
= QEMU_OPT_STRING
,
520 .help
= "host AIO implementation (threads, native, io_uring)",
524 .type
= QEMU_OPT_STRING
,
525 .help
= "file locking mode (on/off/auto, default: auto)",
528 .name
= "pr-manager",
529 .type
= QEMU_OPT_STRING
,
530 .help
= "id of persistent reservation manager object (default: none)",
532 #if defined(__linux__)
534 .name
= "drop-cache",
535 .type
= QEMU_OPT_BOOL
,
536 .help
= "invalidate page cache during live migration (default: on)",
540 .name
= "x-check-cache-dropped",
541 .type
= QEMU_OPT_BOOL
,
542 .help
= "check that page cache was dropped on live migration (default: off)"
544 { /* end of list */ }
548 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
550 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
551 int bdrv_flags
, int open_flags
,
552 bool device
, Error
**errp
)
554 BDRVRawState
*s
= bs
->opaque
;
556 Error
*local_err
= NULL
;
557 const char *filename
= NULL
;
559 BlockdevAioOptions aio
, aio_default
;
564 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
565 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
570 filename
= qemu_opt_get(opts
, "filename");
572 ret
= raw_normalize_devicepath(&filename
, errp
);
577 if (bdrv_flags
& BDRV_O_NATIVE_AIO
) {
578 aio_default
= BLOCKDEV_AIO_OPTIONS_NATIVE
;
579 #ifdef CONFIG_LINUX_IO_URING
580 } else if (bdrv_flags
& BDRV_O_IO_URING
) {
581 aio_default
= BLOCKDEV_AIO_OPTIONS_IO_URING
;
584 aio_default
= BLOCKDEV_AIO_OPTIONS_THREADS
;
587 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
588 qemu_opt_get(opts
, "aio"),
589 aio_default
, &local_err
);
591 error_propagate(errp
, local_err
);
596 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
597 #ifdef CONFIG_LINUX_IO_URING
598 s
->use_linux_io_uring
= (aio
== BLOCKDEV_AIO_OPTIONS_IO_URING
);
601 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
602 qemu_opt_get(opts
, "locking"),
603 ON_OFF_AUTO_AUTO
, &local_err
);
605 error_propagate(errp
, local_err
);
612 if (!qemu_has_ofd_lock()) {
613 warn_report("File lock requested but OFD locking syscall is "
614 "unavailable, falling back to POSIX file locks");
615 error_printf("Due to the implementation, locks can be lost "
619 case ON_OFF_AUTO_OFF
:
622 case ON_OFF_AUTO_AUTO
:
623 s
->use_lock
= qemu_has_ofd_lock();
629 str
= qemu_opt_get(opts
, "pr-manager");
631 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
633 error_propagate(errp
, local_err
);
639 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
640 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
643 s
->open_flags
= open_flags
;
644 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
647 fd
= qemu_open(filename
, s
->open_flags
, errp
);
648 ret
= fd
< 0 ? -errno
: 0;
658 /* Check s->open_flags rather than bdrv_flags due to auto-read-only */
659 if (s
->open_flags
& O_RDWR
) {
660 ret
= check_hdev_writable(s
->fd
);
662 error_setg_errno(errp
, -ret
, "The device is not writable");
668 s
->shared_perm
= BLK_PERM_ALL
;
670 #ifdef CONFIG_LINUX_AIO
671 /* Currently Linux does AIO only for files opened with O_DIRECT */
672 if (s
->use_linux_aio
) {
673 if (!(s
->open_flags
& O_DIRECT
)) {
674 error_setg(errp
, "aio=native was specified, but it requires "
675 "cache.direct=on, which was not specified.");
679 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
680 error_prepend(errp
, "Unable to use native AIO: ");
685 if (s
->use_linux_aio
) {
686 error_setg(errp
, "aio=native was specified, but is not supported "
691 #endif /* !defined(CONFIG_LINUX_AIO) */
693 #ifdef CONFIG_LINUX_IO_URING
694 if (s
->use_linux_io_uring
) {
695 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs
), errp
)) {
696 error_prepend(errp
, "Unable to use io_uring: ");
701 if (s
->use_linux_io_uring
) {
702 error_setg(errp
, "aio=io_uring was specified, but is not supported "
707 #endif /* !defined(CONFIG_LINUX_IO_URING) */
709 s
->has_discard
= true;
710 s
->has_write_zeroes
= true;
711 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0 && !dio_byte_aligned(s
->fd
)) {
712 s
->needs_alignment
= true;
715 if (fstat(s
->fd
, &st
) < 0) {
717 error_setg_errno(errp
, errno
, "Could not stat file");
722 if (!S_ISREG(st
.st_mode
)) {
723 error_setg(errp
, "'%s' driver requires '%s' to be a regular file",
724 bs
->drv
->format_name
, bs
->filename
);
728 s
->discard_zeroes
= true;
729 s
->has_fallocate
= true;
732 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
733 error_setg(errp
, "'%s' driver requires '%s' to be either "
734 "a character or block device",
735 bs
->drv
->format_name
, bs
->filename
);
741 if (S_ISBLK(st
.st_mode
)) {
742 #ifdef BLKDISCARDZEROES
744 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
745 s
->discard_zeroes
= true;
749 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
750 * not rely on the contents of discarded blocks unless using O_DIRECT.
751 * Same for BLKZEROOUT.
753 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
754 s
->discard_zeroes
= false;
755 s
->has_write_zeroes
= false;
760 if (S_ISCHR(st
.st_mode
)) {
762 * The file is a char device (disk), which on FreeBSD isn't behind
763 * a pager, so force all requests to be aligned. This is needed
764 * so QEMU makes sure all IO operations on the device are aligned
765 * to sector size, or else FreeBSD will reject them with EINVAL.
767 s
->needs_alignment
= true;
772 if (platform_test_xfs_fd(s
->fd
)) {
777 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
778 if (S_ISREG(st
.st_mode
)) {
779 /* When extending regular files, we get zeros from the OS */
780 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
784 if (ret
< 0 && s
->fd
!= -1) {
787 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
794 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
797 BDRVRawState
*s
= bs
->opaque
;
799 s
->type
= FTYPE_FILE
;
800 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
809 #define PERM_FOREACH(i) \
810 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
812 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
813 * file; if @unlock == true, also unlock the unneeded bytes.
814 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
816 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
817 uint64_t perm_lock_bits
,
818 uint64_t shared_perm_lock_bits
,
819 bool unlock
, Error
**errp
)
823 uint64_t locked_perm
, locked_shared_perm
;
826 locked_perm
= s
->locked_perm
;
827 locked_shared_perm
= s
->locked_shared_perm
;
830 * We don't have the previous bits, just lock/unlock for each of the
834 locked_perm
= BLK_PERM_ALL
;
835 locked_shared_perm
= BLK_PERM_ALL
;
838 locked_shared_perm
= 0;
843 int off
= RAW_LOCK_PERM_BASE
+ i
;
844 uint64_t bit
= (1ULL << i
);
845 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
846 ret
= qemu_lock_fd(fd
, off
, 1, false);
848 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
852 s
->locked_perm
|= bit
;
854 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
855 ret
= qemu_unlock_fd(fd
, off
, 1);
857 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
860 s
->locked_perm
&= ~bit
;
865 int off
= RAW_LOCK_SHARED_BASE
+ i
;
866 uint64_t bit
= (1ULL << i
);
867 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
868 ret
= qemu_lock_fd(fd
, off
, 1, false);
870 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
874 s
->locked_shared_perm
|= bit
;
876 } else if (unlock
&& (locked_shared_perm
& bit
) &&
877 !(shared_perm_lock_bits
& bit
)) {
878 ret
= qemu_unlock_fd(fd
, off
, 1);
880 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
883 s
->locked_shared_perm
&= ~bit
;
890 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
891 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
898 int off
= RAW_LOCK_SHARED_BASE
+ i
;
899 uint64_t p
= 1ULL << i
;
901 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
903 char *perm_name
= bdrv_perm_names(p
);
905 raw_lock_error_setg_errno(errp
, -ret
,
906 "Failed to get \"%s\" lock",
914 int off
= RAW_LOCK_PERM_BASE
+ i
;
915 uint64_t p
= 1ULL << i
;
916 if (!(shared_perm
& p
)) {
917 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
919 char *perm_name
= bdrv_perm_names(p
);
921 raw_lock_error_setg_errno(errp
, -ret
,
922 "Failed to get shared \"%s\" lock",
932 static int raw_handle_perm_lock(BlockDriverState
*bs
,
934 uint64_t new_perm
, uint64_t new_shared
,
937 BDRVRawState
*s
= bs
->opaque
;
939 Error
*local_err
= NULL
;
945 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
951 if ((s
->perm
| new_perm
) == s
->perm
&&
952 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
955 * We are going to unlock bytes, it should not fail. If it fail due
956 * to some fs-dependent permission-unrelated reasons (which occurs
957 * sometimes on NFS and leads to abort in bdrv_replace_child) we
958 * can't prevent such errors by any check here. And we ignore them
959 * anyway in ABORT and COMMIT.
963 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
964 ~s
->shared_perm
| ~new_shared
,
967 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
971 error_append_hint(errp
,
972 "Is another process using the image [%s]?\n",
975 /* fall through to unlock bytes. */
977 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
980 /* Theoretically the above call only unlocks bytes and it cannot
981 * fail. Something weird happened, report it.
983 warn_report_err(local_err
);
987 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
990 /* Theoretically the above call only unlocks bytes and it cannot
991 * fail. Something weird happened, report it.
993 warn_report_err(local_err
);
1000 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
1001 int *open_flags
, uint64_t perm
, bool force_dup
,
1004 BDRVRawState
*s
= bs
->opaque
;
1007 bool has_writers
= perm
&
1008 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
1009 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
1011 fcntl_flags
|= O_NOATIME
;
1015 if (s
->type
== FTYPE_CD
) {
1016 *open_flags
|= O_NONBLOCK
;
1019 raw_parse_flags(flags
, open_flags
, has_writers
);
1022 /* Not all operating systems have O_ASYNC, and those that don't
1023 * will not let us track the state into rs->open_flags (typically
1024 * you achieve the same effect with an ioctl, for example I_SETSIG
1025 * on Solaris). But we do not use O_ASYNC, so that's fine.
1027 assert((s
->open_flags
& O_ASYNC
) == 0);
1030 if (!force_dup
&& *open_flags
== s
->open_flags
) {
1031 /* We're lucky, the existing fd is fine */
1035 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
1036 /* dup the original fd */
1037 fd
= qemu_dup(s
->fd
);
1039 ret
= fcntl_setfl(fd
, *open_flags
);
1047 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
1049 const char *normalized_filename
= bs
->filename
;
1050 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
1052 fd
= qemu_open(normalized_filename
, *open_flags
, errp
);
1059 if (fd
!= -1 && (*open_flags
& O_RDWR
)) {
1060 ret
= check_hdev_writable(fd
);
1063 error_setg_errno(errp
, -ret
, "The device is not writable");
1071 static int raw_reopen_prepare(BDRVReopenState
*state
,
1072 BlockReopenQueue
*queue
, Error
**errp
)
1075 BDRVRawReopenState
*rs
;
1078 Error
*local_err
= NULL
;
1080 assert(state
!= NULL
);
1081 assert(state
->bs
!= NULL
);
1083 s
= state
->bs
->opaque
;
1085 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
1088 /* Handle options changes */
1089 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
1090 if (!qemu_opts_absorb_qdict(opts
, state
->options
, errp
)) {
1095 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
1096 rs
->check_cache_dropped
=
1097 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
1099 /* This driver's reopen function doesn't currently allow changing
1100 * other options, so let's put them back in the original QDict and
1101 * bdrv_reopen_prepare() will detect changes and complain. */
1102 qemu_opts_to_qdict(opts
, state
->options
);
1104 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
1105 state
->perm
, true, &local_err
);
1107 error_propagate(errp
, local_err
);
1112 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
1113 * alignment with the new fd. */
1115 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
1117 error_propagate(errp
, local_err
);
1123 s
->reopen_state
= state
;
1131 qemu_opts_del(opts
);
1135 static void raw_reopen_commit(BDRVReopenState
*state
)
1137 BDRVRawReopenState
*rs
= state
->opaque
;
1138 BDRVRawState
*s
= state
->bs
->opaque
;
1140 s
->drop_cache
= rs
->drop_cache
;
1141 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1142 s
->open_flags
= rs
->open_flags
;
1147 g_free(state
->opaque
);
1148 state
->opaque
= NULL
;
1150 assert(s
->reopen_state
== state
);
1151 s
->reopen_state
= NULL
;
1155 static void raw_reopen_abort(BDRVReopenState
*state
)
1157 BDRVRawReopenState
*rs
= state
->opaque
;
1158 BDRVRawState
*s
= state
->bs
->opaque
;
1160 /* nothing to do if NULL, we didn't get far enough */
1169 g_free(state
->opaque
);
1170 state
->opaque
= NULL
;
1172 assert(s
->reopen_state
== state
);
1173 s
->reopen_state
= NULL
;
1176 static int sg_get_max_transfer_length(int fd
)
1181 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1191 static int sg_get_max_segments(int fd
)
1196 char *sysfspath
= NULL
;
1202 if (fstat(fd
, &st
)) {
1207 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1208 major(st
.st_rdev
), minor(st
.st_rdev
));
1209 sysfd
= open(sysfspath
, O_RDONLY
);
1215 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1216 } while (ret
== -1 && errno
== EINTR
);
1220 } else if (ret
== 0) {
1225 /* The file is ended with '\n', pass 'end' to accept that. */
1226 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1227 if (ret
== 0 && end
&& *end
== '\n') {
1242 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1244 BDRVRawState
*s
= bs
->opaque
;
1247 int ret
= sg_get_max_transfer_length(s
->fd
);
1249 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1250 bs
->bl
.max_transfer
= pow2floor(ret
);
1253 ret
= sg_get_max_segments(s
->fd
);
1255 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1256 ret
* qemu_real_host_page_size
);
1260 raw_probe_alignment(bs
, s
->fd
, errp
);
1261 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1262 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size
);
1265 static int check_for_dasd(int fd
)
1268 struct dasd_information2_t info
= {0};
1270 return ioctl(fd
, BIODASDINFO2
, &info
);
1277 * Try to get @bs's logical and physical block size.
1278 * On success, store them in @bsz and return zero.
1279 * On failure, return negative errno.
1281 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1283 BDRVRawState
*s
= bs
->opaque
;
1286 /* If DASD, get blocksizes */
1287 if (check_for_dasd(s
->fd
) < 0) {
1290 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1294 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1298 * Try to get @bs's geometry: cyls, heads, sectors.
1299 * On success, store them in @geo and return 0.
1300 * On failure return -errno.
1301 * (Allows block driver to assign default geometry values that guest sees)
1304 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1306 BDRVRawState
*s
= bs
->opaque
;
1307 struct hd_geometry ioctl_geo
= {0};
1309 /* If DASD, get its geometry */
1310 if (check_for_dasd(s
->fd
) < 0) {
1313 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1316 /* HDIO_GETGEO may return success even though geo contains zeros
1317 (e.g. certain multipath setups) */
1318 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1321 /* Do not return a geometry for partition */
1322 if (ioctl_geo
.start
!= 0) {
1325 geo
->heads
= ioctl_geo
.heads
;
1326 geo
->sectors
= ioctl_geo
.sectors
;
1327 geo
->cylinders
= ioctl_geo
.cylinders
;
1331 #else /* __linux__ */
1332 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1338 #if defined(__linux__)
1339 static int handle_aiocb_ioctl(void *opaque
)
1341 RawPosixAIOData
*aiocb
= opaque
;
1344 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1353 static int handle_aiocb_flush(void *opaque
)
1355 RawPosixAIOData
*aiocb
= opaque
;
1356 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1359 if (s
->page_cache_inconsistent
) {
1363 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1365 /* There is no clear definition of the semantics of a failing fsync(),
1366 * so we may have to assume the worst. The sad truth is that this
1367 * assumption is correct for Linux. Some pages are now probably marked
1368 * clean in the page cache even though they are inconsistent with the
1369 * on-disk contents. The next fdatasync() call would succeed, but no
1370 * further writeback attempt will be made. We can't get back to a state
1371 * in which we know what is on disk (we would have to rewrite
1372 * everything that was touched since the last fdatasync() at least), so
1373 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1374 * really defined, I have little hope that other OSes are doing better.
1376 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1378 if ((s
->open_flags
& O_DIRECT
) == 0) {
1379 s
->page_cache_inconsistent
= true;
1386 #ifdef CONFIG_PREADV
1388 static bool preadv_present
= true;
1391 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1393 return preadv(fd
, iov
, nr_iov
, offset
);
1397 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1399 return pwritev(fd
, iov
, nr_iov
, offset
);
1404 static bool preadv_present
= false;
1407 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1413 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1420 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1425 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1426 len
= qemu_pwritev(aiocb
->aio_fildes
,
1431 len
= qemu_preadv(aiocb
->aio_fildes
,
1435 } while (len
== -1 && errno
== EINTR
);
1444 * Read/writes the data to/from a given linear buffer.
1446 * Returns the number of bytes handles or -errno in case of an error. Short
1447 * reads are only returned if the end of the file is reached.
1449 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1454 while (offset
< aiocb
->aio_nbytes
) {
1455 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1456 len
= pwrite(aiocb
->aio_fildes
,
1457 (const char *)buf
+ offset
,
1458 aiocb
->aio_nbytes
- offset
,
1459 aiocb
->aio_offset
+ offset
);
1461 len
= pread(aiocb
->aio_fildes
,
1463 aiocb
->aio_nbytes
- offset
,
1464 aiocb
->aio_offset
+ offset
);
1466 if (len
== -1 && errno
== EINTR
) {
1468 } else if (len
== -1 && errno
== EINVAL
&&
1469 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1470 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1472 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1473 * after a short read. Assume that O_DIRECT short reads only occur
1474 * at EOF. Therefore this is a short read, not an I/O error.
1477 } else if (len
== -1) {
1480 } else if (len
== 0) {
1489 static int handle_aiocb_rw(void *opaque
)
1491 RawPosixAIOData
*aiocb
= opaque
;
1495 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1497 * If there is just a single buffer, and it is properly aligned
1498 * we can just use plain pread/pwrite without any problems.
1500 if (aiocb
->io
.niov
== 1) {
1501 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1505 * We have more than one iovec, and all are properly aligned.
1507 * Try preadv/pwritev first and fall back to linearizing the
1508 * buffer if it's not supported.
1510 if (preadv_present
) {
1511 nbytes
= handle_aiocb_rw_vector(aiocb
);
1512 if (nbytes
== aiocb
->aio_nbytes
||
1513 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1516 preadv_present
= false;
1520 * XXX(hch): short read/write. no easy way to handle the reminder
1521 * using these interfaces. For now retry using plain
1527 * Ok, we have to do it the hard way, copy all segments into
1528 * a single aligned buffer.
1530 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1536 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1540 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1541 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1542 p
+= aiocb
->io
.iov
[i
].iov_len
;
1544 assert(p
- buf
== aiocb
->aio_nbytes
);
1547 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1548 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1550 size_t count
= aiocb
->aio_nbytes
, copy
;
1553 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1555 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1556 copy
= aiocb
->io
.iov
[i
].iov_len
;
1558 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1559 assert(count
>= copy
);
1568 if (nbytes
== aiocb
->aio_nbytes
) {
1570 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1571 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1574 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1575 0, aiocb
->aio_nbytes
- nbytes
);
1584 static int translate_err(int err
)
1586 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1593 #ifdef CONFIG_FALLOCATE
1594 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1597 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1600 } while (errno
== EINTR
);
1601 return translate_err(-errno
);
1605 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1608 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1610 if (!s
->has_write_zeroes
) {
1615 /* The BLKZEROOUT implementation in the kernel doesn't set
1616 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1618 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1620 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1621 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1624 } while (errno
== EINTR
);
1626 ret
= translate_err(-errno
);
1627 if (ret
== -ENOTSUP
) {
1628 s
->has_write_zeroes
= false;
1636 static int handle_aiocb_write_zeroes(void *opaque
)
1638 RawPosixAIOData
*aiocb
= opaque
;
1639 #ifdef CONFIG_FALLOCATE
1640 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1644 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1645 return handle_aiocb_write_zeroes_block(aiocb
);
1648 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1649 if (s
->has_write_zeroes
) {
1650 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1651 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1652 if (ret
== -EINVAL
) {
1654 * Allow falling back to pwrite for file systems that
1655 * do not support fallocate() for an unaligned byte range.
1659 if (ret
== 0 || ret
!= -ENOTSUP
) {
1662 s
->has_write_zeroes
= false;
1666 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1667 if (s
->has_discard
&& s
->has_fallocate
) {
1668 int ret
= do_fallocate(s
->fd
,
1669 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1670 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1672 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1673 if (ret
== 0 || ret
!= -ENOTSUP
) {
1676 s
->has_fallocate
= false;
1677 } else if (ret
!= -ENOTSUP
) {
1680 s
->has_discard
= false;
1685 #ifdef CONFIG_FALLOCATE
1686 /* Last resort: we are trying to extend the file with zeroed data. This
1687 * can be done via fallocate(fd, 0) */
1688 len
= bdrv_getlength(aiocb
->bs
);
1689 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1690 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1691 if (ret
== 0 || ret
!= -ENOTSUP
) {
1694 s
->has_fallocate
= false;
1701 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1703 RawPosixAIOData
*aiocb
= opaque
;
1704 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1706 /* First try to write zeros and unmap at the same time */
1708 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1709 int ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1710 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1721 /* If we couldn't manage to unmap while guaranteed that the area reads as
1722 * all-zero afterwards, just write zeroes without unmapping */
1723 return handle_aiocb_write_zeroes(aiocb
);
1726 #ifndef HAVE_COPY_FILE_RANGE
1727 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1728 off_t
*out_off
, size_t len
, unsigned int flags
)
1730 #ifdef __NR_copy_file_range
1731 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1732 out_off
, len
, flags
);
1740 static int handle_aiocb_copy_range(void *opaque
)
1742 RawPosixAIOData
*aiocb
= opaque
;
1743 uint64_t bytes
= aiocb
->aio_nbytes
;
1744 off_t in_off
= aiocb
->aio_offset
;
1745 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1748 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1749 aiocb
->copy_range
.aio_fd2
, &out_off
,
1751 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1752 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1755 /* No progress (e.g. when beyond EOF), let the caller fall back to
1774 static int handle_aiocb_discard(void *opaque
)
1776 RawPosixAIOData
*aiocb
= opaque
;
1777 int ret
= -EOPNOTSUPP
;
1778 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1780 if (!s
->has_discard
) {
1784 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1787 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1788 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1791 } while (errno
== EINTR
);
1796 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1797 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1798 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1802 ret
= translate_err(ret
);
1803 if (ret
== -ENOTSUP
) {
1804 s
->has_discard
= false;
1810 * Help alignment probing by allocating the first block.
1812 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1813 * reading succeeds regardless of request length. In this case we fallback to
1814 * safe alignment which is not optimal. Allocating the first block avoids this
1817 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1818 * request alignment, so we use safe values.
1820 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1821 * caller may ignore failures.
1823 static int allocate_first_block(int fd
, size_t max_size
)
1825 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1828 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size
);
1833 buf
= qemu_memalign(max_align
, write_size
);
1834 memset(buf
, 0, write_size
);
1837 n
= pwrite(fd
, buf
, write_size
, 0);
1838 } while (n
== -1 && errno
== EINTR
);
1840 ret
= (n
== -1) ? -errno
: 0;
1846 static int handle_aiocb_truncate(void *opaque
)
1848 RawPosixAIOData
*aiocb
= opaque
;
1850 int64_t current_length
= 0;
1853 int fd
= aiocb
->aio_fildes
;
1854 int64_t offset
= aiocb
->aio_offset
;
1855 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1856 Error
**errp
= aiocb
->truncate
.errp
;
1858 if (fstat(fd
, &st
) < 0) {
1860 error_setg_errno(errp
, -result
, "Could not stat file");
1864 current_length
= st
.st_size
;
1865 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1866 error_setg(errp
, "Cannot use preallocation for shrinking files");
1871 #ifdef CONFIG_POSIX_FALLOCATE
1872 case PREALLOC_MODE_FALLOC
:
1874 * Truncating before posix_fallocate() makes it about twice slower on
1875 * file systems that do not support fallocate(), trying to check if a
1876 * block is allocated before allocating it, so don't do that here.
1878 if (offset
!= current_length
) {
1879 result
= -posix_fallocate(fd
, current_length
,
1880 offset
- current_length
);
1882 /* posix_fallocate() doesn't set errno. */
1883 error_setg_errno(errp
, -result
,
1884 "Could not preallocate new data");
1885 } else if (current_length
== 0) {
1887 * posix_fallocate() uses fallocate() if the filesystem
1888 * supports it, or fallback to manually writing zeroes. If
1889 * fallocate() was used, unaligned reads from the fallocated
1890 * area in raw_probe_alignment() will succeed, hence we need to
1891 * allocate the first block.
1893 * Optimize future alignment probing; ignore failures.
1895 allocate_first_block(fd
, offset
);
1902 case PREALLOC_MODE_FULL
:
1904 int64_t num
= 0, left
= offset
- current_length
;
1908 * Knowing the final size from the beginning could allow the file
1909 * system driver to do less allocations and possibly avoid
1910 * fragmentation of the file.
1912 if (ftruncate(fd
, offset
) != 0) {
1914 error_setg_errno(errp
, -result
, "Could not resize file");
1918 buf
= g_malloc0(65536);
1920 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1921 if (seek_result
< 0) {
1923 error_setg_errno(errp
, -result
,
1924 "Failed to seek to the old end of file");
1929 num
= MIN(left
, 65536);
1930 result
= write(fd
, buf
, num
);
1932 if (errno
== EINTR
) {
1936 error_setg_errno(errp
, -result
,
1937 "Could not write zeros for preallocation");
1946 error_setg_errno(errp
, -result
,
1947 "Could not flush file to disk");
1953 case PREALLOC_MODE_OFF
:
1954 if (ftruncate(fd
, offset
) != 0) {
1956 error_setg_errno(errp
, -result
, "Could not resize file");
1957 } else if (current_length
== 0 && offset
> current_length
) {
1958 /* Optimize future alignment probing; ignore failures. */
1959 allocate_first_block(fd
, offset
);
1964 error_setg(errp
, "Unsupported preallocation mode: %s",
1965 PreallocMode_str(prealloc
));
1971 if (ftruncate(fd
, current_length
) < 0) {
1972 error_report("Failed to restore old file length: %s",
1981 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1982 ThreadPoolFunc func
, void *arg
)
1984 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1985 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1986 return thread_pool_submit_co(pool
, func
, arg
);
1989 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1990 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1992 BDRVRawState
*s
= bs
->opaque
;
1993 RawPosixAIOData acb
;
1995 if (fd_open(bs
) < 0)
1999 * When using O_DIRECT, the request must be aligned to be able to use
2000 * either libaio or io_uring interface. If not fail back to regular thread
2001 * pool read/write code which emulates this for us if we
2002 * set QEMU_AIO_MISALIGNED.
2004 if (s
->needs_alignment
&& !bdrv_qiov_is_aligned(bs
, qiov
)) {
2005 type
|= QEMU_AIO_MISALIGNED
;
2006 #ifdef CONFIG_LINUX_IO_URING
2007 } else if (s
->use_linux_io_uring
) {
2008 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2009 assert(qiov
->size
== bytes
);
2010 return luring_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
2012 #ifdef CONFIG_LINUX_AIO
2013 } else if (s
->use_linux_aio
) {
2014 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
2015 assert(qiov
->size
== bytes
);
2016 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
2020 acb
= (RawPosixAIOData
) {
2022 .aio_fildes
= s
->fd
,
2024 .aio_offset
= offset
,
2025 .aio_nbytes
= bytes
,
2032 assert(qiov
->size
== bytes
);
2033 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
2036 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
2037 uint64_t bytes
, QEMUIOVector
*qiov
,
2040 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
2043 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
2044 uint64_t bytes
, QEMUIOVector
*qiov
,
2048 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
2051 static void raw_aio_plug(BlockDriverState
*bs
)
2053 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2054 #ifdef CONFIG_LINUX_AIO
2055 if (s
->use_linux_aio
) {
2056 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
2057 laio_io_plug(bs
, aio
);
2060 #ifdef CONFIG_LINUX_IO_URING
2061 if (s
->use_linux_io_uring
) {
2062 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2063 luring_io_plug(bs
, aio
);
2068 static void raw_aio_unplug(BlockDriverState
*bs
)
2070 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2071 #ifdef CONFIG_LINUX_AIO
2072 if (s
->use_linux_aio
) {
2073 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
2074 laio_io_unplug(bs
, aio
);
2077 #ifdef CONFIG_LINUX_IO_URING
2078 if (s
->use_linux_io_uring
) {
2079 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2080 luring_io_unplug(bs
, aio
);
2085 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
2087 BDRVRawState
*s
= bs
->opaque
;
2088 RawPosixAIOData acb
;
2096 acb
= (RawPosixAIOData
) {
2098 .aio_fildes
= s
->fd
,
2099 .aio_type
= QEMU_AIO_FLUSH
,
2102 #ifdef CONFIG_LINUX_IO_URING
2103 if (s
->use_linux_io_uring
) {
2104 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2105 return luring_co_submit(bs
, aio
, s
->fd
, 0, NULL
, QEMU_AIO_FLUSH
);
2108 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
2111 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
2112 AioContext
*new_context
)
2114 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2115 #ifdef CONFIG_LINUX_AIO
2116 if (s
->use_linux_aio
) {
2117 Error
*local_err
= NULL
;
2118 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
2119 error_reportf_err(local_err
, "Unable to use native AIO, "
2120 "falling back to thread pool: ");
2121 s
->use_linux_aio
= false;
2125 #ifdef CONFIG_LINUX_IO_URING
2126 if (s
->use_linux_io_uring
) {
2127 Error
*local_err
= NULL
;
2128 if (!aio_setup_linux_io_uring(new_context
, &local_err
)) {
2129 error_reportf_err(local_err
, "Unable to use linux io_uring, "
2130 "falling back to thread pool: ");
2131 s
->use_linux_io_uring
= false;
2137 static void raw_close(BlockDriverState
*bs
)
2139 BDRVRawState
*s
= bs
->opaque
;
2148 * Truncates the given regular file @fd to @offset and, when growing, fills the
2149 * new space according to @prealloc.
2151 * Returns: 0 on success, -errno on failure.
2153 static int coroutine_fn
2154 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2155 PreallocMode prealloc
, Error
**errp
)
2157 RawPosixAIOData acb
;
2159 acb
= (RawPosixAIOData
) {
2162 .aio_type
= QEMU_AIO_TRUNCATE
,
2163 .aio_offset
= offset
,
2165 .prealloc
= prealloc
,
2170 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2173 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2174 bool exact
, PreallocMode prealloc
,
2175 BdrvRequestFlags flags
, Error
**errp
)
2177 BDRVRawState
*s
= bs
->opaque
;
2181 if (fstat(s
->fd
, &st
)) {
2183 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2187 if (S_ISREG(st
.st_mode
)) {
2188 /* Always resizes to the exact @offset */
2189 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2192 if (prealloc
!= PREALLOC_MODE_OFF
) {
2193 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2194 "non-regular file", PreallocMode_str(prealloc
));
2198 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2199 int64_t cur_length
= raw_getlength(bs
);
2201 if (offset
!= cur_length
&& exact
) {
2202 error_setg(errp
, "Cannot resize device files");
2204 } else if (offset
> cur_length
) {
2205 error_setg(errp
, "Cannot grow device files");
2209 error_setg(errp
, "Resizing this file is not supported");
2217 static int64_t raw_getlength(BlockDriverState
*bs
)
2219 BDRVRawState
*s
= bs
->opaque
;
2225 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2226 struct disklabel dl
;
2228 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2230 return (uint64_t)dl
.d_secsize
*
2231 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2235 #elif defined(__NetBSD__)
2236 static int64_t raw_getlength(BlockDriverState
*bs
)
2238 BDRVRawState
*s
= bs
->opaque
;
2244 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2245 struct dkwedge_info dkw
;
2247 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2248 return dkw
.dkw_size
* 512;
2250 struct disklabel dl
;
2252 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2254 return (uint64_t)dl
.d_secsize
*
2255 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2260 #elif defined(__sun__)
2261 static int64_t raw_getlength(BlockDriverState
*bs
)
2263 BDRVRawState
*s
= bs
->opaque
;
2264 struct dk_minfo minfo
;
2274 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2276 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2278 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2282 * There are reports that lseek on some devices fails, but
2283 * irc discussion said that contingency on contingency was overkill.
2285 size
= lseek(s
->fd
, 0, SEEK_END
);
2291 #elif defined(CONFIG_BSD)
2292 static int64_t raw_getlength(BlockDriverState
*bs
)
2294 BDRVRawState
*s
= bs
->opaque
;
2298 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2307 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2310 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2311 #ifdef DIOCGMEDIASIZE
2312 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2313 #elif defined(DIOCGPART)
2316 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2317 size
= pi
.media_size
;
2323 #if defined(__APPLE__) && defined(__MACH__)
2325 uint64_t sectors
= 0;
2326 uint32_t sector_size
= 0;
2328 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2329 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2330 size
= sectors
* sector_size
;
2332 size
= lseek(fd
, 0LL, SEEK_END
);
2339 size
= lseek(fd
, 0LL, SEEK_END
);
2344 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2347 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2348 if (size
== 2048LL * (unsigned)-1)
2350 /* XXX no disc? maybe we need to reopen... */
2351 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2358 size
= lseek(fd
, 0, SEEK_END
);
2366 static int64_t raw_getlength(BlockDriverState
*bs
)
2368 BDRVRawState
*s
= bs
->opaque
;
2377 size
= lseek(s
->fd
, 0, SEEK_END
);
2385 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2388 BDRVRawState
*s
= bs
->opaque
;
2390 if (fstat(s
->fd
, &st
) < 0) {
2393 return (int64_t)st
.st_blocks
* 512;
2396 static int coroutine_fn
2397 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2399 BlockdevCreateOptionsFile
*file_opts
;
2400 Error
*local_err
= NULL
;
2402 uint64_t perm
, shared
;
2405 /* Validate options and set default values */
2406 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2407 file_opts
= &options
->u
.file
;
2409 if (!file_opts
->has_nocow
) {
2410 file_opts
->nocow
= false;
2412 if (!file_opts
->has_preallocation
) {
2413 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2415 if (!file_opts
->has_extent_size_hint
) {
2416 file_opts
->extent_size_hint
= 1 * MiB
;
2418 if (file_opts
->extent_size_hint
> UINT32_MAX
) {
2420 error_setg(errp
, "Extent size hint is too large");
2425 fd
= qemu_create(file_opts
->filename
, O_RDWR
| O_BINARY
, 0644, errp
);
2431 /* Take permissions: We want to discard everything, so we need
2432 * BLK_PERM_WRITE; and truncation to the desired size requires
2434 * On the other hand, we cannot share the RESIZE permission
2435 * because we promise that after this function, the file has the
2436 * size given in the options. If someone else were to resize it
2437 * concurrently, we could not guarantee that.
2438 * Note that after this function, we can no longer guarantee that
2439 * the file is not touched by a third party, so it may be resized
2441 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2442 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2444 /* Step one: Take locks */
2445 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2450 /* Step two: Check that nobody else has taken conflicting locks */
2451 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2453 error_append_hint(errp
,
2454 "Is another process using the image [%s]?\n",
2455 file_opts
->filename
);
2459 /* Clear the file by truncating it to 0 */
2460 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2465 if (file_opts
->nocow
) {
2467 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2468 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2469 * will be ignored since any failure of this operation should not
2470 * block the left work.
2473 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2474 attr
|= FS_NOCOW_FL
;
2475 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2479 #ifdef FS_IOC_FSSETXATTR
2481 * Try to set the extent size hint. Failure is not fatal, and a warning is
2482 * only printed if the option was explicitly specified.
2485 struct fsxattr attr
;
2486 result
= ioctl(fd
, FS_IOC_FSGETXATTR
, &attr
);
2488 attr
.fsx_xflags
|= FS_XFLAG_EXTSIZE
;
2489 attr
.fsx_extsize
= file_opts
->extent_size_hint
;
2490 result
= ioctl(fd
, FS_IOC_FSSETXATTR
, &attr
);
2492 if (result
< 0 && file_opts
->has_extent_size_hint
&&
2493 file_opts
->extent_size_hint
)
2495 warn_report("Failed to set extent size hint: %s",
2501 /* Resize and potentially preallocate the file to the desired
2503 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2504 file_opts
->preallocation
, errp
);
2510 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2512 /* The above call should not fail, and if it does, that does
2513 * not mean the whole creation operation has failed. So
2514 * report it the user for their convenience, but do not report
2515 * it to the caller. */
2516 warn_report_err(local_err
);
2520 if (qemu_close(fd
) != 0 && result
== 0) {
2522 error_setg_errno(errp
, -result
, "Could not close the new file");
2528 static int coroutine_fn
raw_co_create_opts(BlockDriver
*drv
,
2529 const char *filename
,
2533 BlockdevCreateOptions options
;
2534 int64_t total_size
= 0;
2535 int64_t extent_size_hint
= 0;
2536 bool has_extent_size_hint
= false;
2538 PreallocMode prealloc
;
2540 Error
*local_err
= NULL
;
2542 /* Skip file: protocol prefix */
2543 strstart(filename
, "file:", &filename
);
2545 /* Read out options */
2546 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2548 if (qemu_opt_get(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
)) {
2549 has_extent_size_hint
= true;
2551 qemu_opt_get_size_del(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
, -1);
2553 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2554 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2555 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2556 PREALLOC_MODE_OFF
, &local_err
);
2559 error_propagate(errp
, local_err
);
2563 options
= (BlockdevCreateOptions
) {
2564 .driver
= BLOCKDEV_DRIVER_FILE
,
2566 .filename
= (char *) filename
,
2568 .has_preallocation
= true,
2569 .preallocation
= prealloc
,
2572 .has_extent_size_hint
= has_extent_size_hint
,
2573 .extent_size_hint
= extent_size_hint
,
2576 return raw_co_create(&options
, errp
);
2579 static int coroutine_fn
raw_co_delete_file(BlockDriverState
*bs
,
2585 if (!(stat(bs
->filename
, &st
) == 0) || !S_ISREG(st
.st_mode
)) {
2586 error_setg_errno(errp
, ENOENT
, "%s is not a regular file",
2591 ret
= unlink(bs
->filename
);
2594 error_setg_errno(errp
, -ret
, "Error when deleting file %s",
2602 * Find allocation range in @bs around offset @start.
2603 * May change underlying file descriptor's file offset.
2604 * If @start is not in a hole, store @start in @data, and the
2605 * beginning of the next hole in @hole, and return 0.
2606 * If @start is in a non-trailing hole, store @start in @hole and the
2607 * beginning of the next non-hole in @data, and return 0.
2608 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2609 * If we can't find out, return a negative errno other than -ENXIO.
2611 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2612 off_t
*data
, off_t
*hole
)
2614 #if defined SEEK_HOLE && defined SEEK_DATA
2615 BDRVRawState
*s
= bs
->opaque
;
2620 * D1. offs == start: start is in data
2621 * D2. offs > start: start is in a hole, next data at offs
2622 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2623 * or start is beyond EOF
2624 * If the latter happens, the file has been truncated behind
2625 * our back since we opened it. All bets are off then.
2626 * Treating like a trailing hole is simplest.
2627 * D4. offs < 0, errno != ENXIO: we learned nothing
2629 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2631 return -errno
; /* D3 or D4 */
2635 /* This is not a valid return by lseek(). We are safe to just return
2636 * -EIO in this case, and we'll treat it like D4. */
2641 /* D2: in hole, next data at offs */
2647 /* D1: in data, end not yet known */
2651 * H1. offs == start: start is in a hole
2652 * If this happens here, a hole has been dug behind our back
2653 * since the previous lseek().
2654 * H2. offs > start: either start is in data, next hole at offs,
2655 * or start is in trailing hole, EOF at offs
2656 * Linux treats trailing holes like any other hole: offs ==
2657 * start. Solaris seeks to EOF instead: offs > start (blech).
2658 * If that happens here, a hole has been dug behind our back
2659 * since the previous lseek().
2660 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2661 * If this happens, the file has been truncated behind our
2662 * back since we opened it. Treat it like a trailing hole.
2663 * H4. offs < 0, errno != ENXIO: we learned nothing
2664 * Pretend we know nothing at all, i.e. "forget" about D1.
2666 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2668 return -errno
; /* D1 and (H3 or H4) */
2672 /* This is not a valid return by lseek(). We are safe to just return
2673 * -EIO in this case, and we'll treat it like H4. */
2679 * D1 and H2: either in data, next hole at offs, or it was in
2680 * data but is now in a trailing hole. In the latter case,
2681 * all bets are off. Treating it as if it there was data all
2682 * the way to EOF is safe, so simply do that.
2697 * Returns the allocation status of the specified offset.
2699 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2701 * 'pnum' is set to the number of bytes (including and immediately following
2702 * the specified offset) that are known to be in the same
2703 * allocated/unallocated state.
2705 * 'bytes' is the max value 'pnum' should be set to.
2707 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2710 int64_t bytes
, int64_t *pnum
,
2712 BlockDriverState
**file
)
2714 off_t data
= 0, hole
= 0;
2717 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2728 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2731 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2732 if (ret
== -ENXIO
) {
2735 ret
= BDRV_BLOCK_ZERO
;
2736 } else if (ret
< 0) {
2737 /* No info available, so pretend there are no holes */
2739 ret
= BDRV_BLOCK_DATA
;
2740 } else if (data
== offset
) {
2741 /* On a data extent, compute bytes to the end of the extent,
2742 * possibly including a partial sector at EOF. */
2743 *pnum
= MIN(bytes
, hole
- offset
);
2746 * We are not allowed to return partial sectors, though, so
2747 * round up if necessary.
2749 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2750 int64_t file_length
= raw_getlength(bs
);
2751 if (file_length
> 0) {
2752 /* Ignore errors, this is just a safeguard */
2753 assert(hole
== file_length
);
2755 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2758 ret
= BDRV_BLOCK_DATA
;
2760 /* On a hole, compute bytes to the beginning of the next extent. */
2761 assert(hole
== offset
);
2762 *pnum
= MIN(bytes
, data
- offset
);
2763 ret
= BDRV_BLOCK_ZERO
;
2767 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2770 #if defined(__linux__)
2771 /* Verify that the file is not in the page cache */
2772 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2774 const size_t window_size
= 128 * 1024 * 1024;
2775 BDRVRawState
*s
= bs
->opaque
;
2776 void *window
= NULL
;
2783 /* mincore(2) page status information requires 1 byte per page */
2784 page_size
= sysconf(_SC_PAGESIZE
);
2785 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2787 end
= raw_getlength(bs
);
2789 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2796 /* Unmap previous window if size has changed */
2797 new_length
= MIN(end
- offset
, window_size
);
2798 if (new_length
!= length
) {
2799 munmap(window
, length
);
2804 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2806 if (new_window
== MAP_FAILED
) {
2807 error_setg_errno(errp
, errno
, "mmap failed");
2811 window
= new_window
;
2812 length
= new_length
;
2814 ret
= mincore(window
, length
, vec
);
2816 error_setg_errno(errp
, errno
, "mincore failed");
2820 vec_end
= DIV_ROUND_UP(length
, page_size
);
2821 for (i
= 0; i
< vec_end
; i
++) {
2827 error_setg(errp
, "page cache still in use!");
2833 munmap(window
, length
);
2838 #endif /* __linux__ */
2840 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2843 BDRVRawState
*s
= bs
->opaque
;
2848 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2852 if (!s
->drop_cache
) {
2856 if (s
->open_flags
& O_DIRECT
) {
2857 return; /* No host kernel page cache */
2860 #if defined(__linux__)
2861 /* This sets the scene for the next syscall... */
2862 ret
= bdrv_co_flush(bs
);
2864 error_setg_errno(errp
, -ret
, "flush failed");
2868 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2869 * process. These limitations are okay because we just fsynced the file,
2870 * we don't use mmap, and the file should not be in use by other processes.
2872 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2873 if (ret
!= 0) { /* the return value is a positive errno */
2874 error_setg_errno(errp
, ret
, "fadvise failed");
2878 if (s
->check_cache_dropped
) {
2879 check_cache_dropped(bs
, errp
);
2881 #else /* __linux__ */
2882 /* Do nothing. Live migration to a remote host with cache.direct=off is
2883 * unsupported on other host operating systems. Cache consistency issues
2884 * may occur but no error is reported here, partly because that's the
2885 * historical behavior and partly because it's hard to differentiate valid
2886 * configurations that should not cause errors.
2888 #endif /* !__linux__ */
2891 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
2894 s
->stats
.discard_nb_failed
++;
2896 s
->stats
.discard_nb_ok
++;
2897 s
->stats
.discard_bytes_ok
+= nbytes
;
2901 static coroutine_fn
int
2902 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2904 BDRVRawState
*s
= bs
->opaque
;
2905 RawPosixAIOData acb
;
2908 acb
= (RawPosixAIOData
) {
2910 .aio_fildes
= s
->fd
,
2911 .aio_type
= QEMU_AIO_DISCARD
,
2912 .aio_offset
= offset
,
2913 .aio_nbytes
= bytes
,
2917 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2920 ret
= raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2921 raw_account_discard(s
, bytes
, ret
);
2925 static coroutine_fn
int
2926 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2928 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2931 static int coroutine_fn
2932 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2933 BdrvRequestFlags flags
, bool blkdev
)
2935 BDRVRawState
*s
= bs
->opaque
;
2936 RawPosixAIOData acb
;
2937 ThreadPoolFunc
*handler
;
2939 #ifdef CONFIG_FALLOCATE
2940 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
2941 BdrvTrackedRequest
*req
;
2944 * This is a workaround for a bug in the Linux XFS driver,
2945 * where writes submitted through the AIO interface will be
2946 * discarded if they happen beyond a concurrently running
2947 * fallocate() that increases the file length (i.e., both the
2948 * write and the fallocate() happen beyond the EOF).
2950 * To work around it, we extend the tracked request for this
2951 * zero write until INT64_MAX (effectively infinity), and mark
2952 * it as serializing.
2954 * We have to enable this workaround for all filesystems and
2955 * AIO modes (not just XFS with aio=native), because for
2956 * remote filesystems we do not know the host configuration.
2959 req
= bdrv_co_get_self_request(bs
);
2961 assert(req
->type
== BDRV_TRACKED_WRITE
);
2962 assert(req
->offset
<= offset
);
2963 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
2965 req
->bytes
= BDRV_MAX_LENGTH
- req
->offset
;
2967 bdrv_check_request(req
->offset
, req
->bytes
, &error_abort
);
2969 bdrv_make_request_serialising(req
, bs
->bl
.request_alignment
);
2973 acb
= (RawPosixAIOData
) {
2975 .aio_fildes
= s
->fd
,
2976 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2977 .aio_offset
= offset
,
2978 .aio_nbytes
= bytes
,
2982 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2984 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2985 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2988 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2989 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2990 handler
= handle_aiocb_write_zeroes_unmap
;
2992 handler
= handle_aiocb_write_zeroes
;
2995 return raw_thread_pool_submit(bs
, handler
, &acb
);
2998 static int coroutine_fn
raw_co_pwrite_zeroes(
2999 BlockDriverState
*bs
, int64_t offset
,
3000 int bytes
, BdrvRequestFlags flags
)
3002 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
3005 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3010 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
3012 BDRVRawState
*s
= bs
->opaque
;
3013 return (BlockStatsSpecificFile
) {
3014 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
3015 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
3016 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
3020 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
3022 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3024 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
3025 stats
->u
.file
= get_blockstats_specific_file(bs
);
3030 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
3032 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3034 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
3035 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
3040 static QemuOptsList raw_create_opts
= {
3041 .name
= "raw-create-opts",
3042 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
3045 .name
= BLOCK_OPT_SIZE
,
3046 .type
= QEMU_OPT_SIZE
,
3047 .help
= "Virtual disk size"
3050 .name
= BLOCK_OPT_NOCOW
,
3051 .type
= QEMU_OPT_BOOL
,
3052 .help
= "Turn off copy-on-write (valid only on btrfs)"
3055 .name
= BLOCK_OPT_PREALLOC
,
3056 .type
= QEMU_OPT_STRING
,
3057 .help
= "Preallocation mode (allowed values: off"
3058 #ifdef CONFIG_POSIX_FALLOCATE
3064 .name
= BLOCK_OPT_EXTENT_SIZE_HINT
,
3065 .type
= QEMU_OPT_SIZE
,
3066 .help
= "Extent size hint for the image file, 0 to disable"
3068 { /* end of list */ }
3072 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
3075 BDRVRawState
*s
= bs
->opaque
;
3076 BDRVRawReopenState
*rs
= NULL
;
3080 if (s
->perm_change_fd
) {
3082 * In the context of reopen, this function may be called several times
3083 * (directly and recursively while change permissions of the parent).
3084 * This is even true for children that don't inherit from the original
3085 * reopen node, so s->reopen_state is not set.
3087 * Ignore all but the first call.
3092 if (s
->reopen_state
) {
3093 /* We already have a new file descriptor to set permissions for */
3094 assert(s
->reopen_state
->perm
== perm
);
3095 assert(s
->reopen_state
->shared_perm
== shared
);
3096 rs
= s
->reopen_state
->opaque
;
3097 s
->perm_change_fd
= rs
->fd
;
3098 s
->perm_change_flags
= rs
->open_flags
;
3100 /* We may need a new fd if auto-read-only switches the mode */
3101 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
3105 } else if (ret
!= s
->fd
) {
3106 s
->perm_change_fd
= ret
;
3107 s
->perm_change_flags
= open_flags
;
3111 /* Prepare permissions on old fd to avoid conflicts between old and new,
3112 * but keep everything locked that new will need. */
3113 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
3118 /* Copy locks to the new fd */
3119 if (s
->perm_change_fd
&& s
->use_lock
) {
3120 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
3123 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3130 if (s
->perm_change_fd
&& !s
->reopen_state
) {
3131 qemu_close(s
->perm_change_fd
);
3133 s
->perm_change_fd
= 0;
3137 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
3139 BDRVRawState
*s
= bs
->opaque
;
3141 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3142 * called after .bdrv_reopen_commit) */
3143 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
3145 s
->fd
= s
->perm_change_fd
;
3146 s
->open_flags
= s
->perm_change_flags
;
3148 s
->perm_change_fd
= 0;
3150 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
3152 s
->shared_perm
= shared
;
3155 static void raw_abort_perm_update(BlockDriverState
*bs
)
3157 BDRVRawState
*s
= bs
->opaque
;
3159 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3160 * the file descriptor. */
3161 if (s
->perm_change_fd
&& !s
->reopen_state
) {
3162 qemu_close(s
->perm_change_fd
);
3164 s
->perm_change_fd
= 0;
3166 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3169 static int coroutine_fn
raw_co_copy_range_from(
3170 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
3171 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
3172 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
3174 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3175 read_flags
, write_flags
);
3178 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
3180 uint64_t src_offset
,
3182 uint64_t dst_offset
,
3184 BdrvRequestFlags read_flags
,
3185 BdrvRequestFlags write_flags
)
3187 RawPosixAIOData acb
;
3188 BDRVRawState
*s
= bs
->opaque
;
3189 BDRVRawState
*src_s
;
3191 assert(dst
->bs
== bs
);
3192 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
3196 src_s
= src
->bs
->opaque
;
3197 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
3201 acb
= (RawPosixAIOData
) {
3203 .aio_type
= QEMU_AIO_COPY_RANGE
,
3204 .aio_fildes
= src_s
->fd
,
3205 .aio_offset
= src_offset
,
3206 .aio_nbytes
= bytes
,
3209 .aio_offset2
= dst_offset
,
3213 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
3216 BlockDriver bdrv_file
= {
3217 .format_name
= "file",
3218 .protocol_name
= "file",
3219 .instance_size
= sizeof(BDRVRawState
),
3220 .bdrv_needs_filename
= true,
3221 .bdrv_probe
= NULL
, /* no probe for protocols */
3222 .bdrv_parse_filename
= raw_parse_filename
,
3223 .bdrv_file_open
= raw_open
,
3224 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3225 .bdrv_reopen_commit
= raw_reopen_commit
,
3226 .bdrv_reopen_abort
= raw_reopen_abort
,
3227 .bdrv_close
= raw_close
,
3228 .bdrv_co_create
= raw_co_create
,
3229 .bdrv_co_create_opts
= raw_co_create_opts
,
3230 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3231 .bdrv_co_block_status
= raw_co_block_status
,
3232 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3233 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3234 .bdrv_co_delete_file
= raw_co_delete_file
,
3236 .bdrv_co_preadv
= raw_co_preadv
,
3237 .bdrv_co_pwritev
= raw_co_pwritev
,
3238 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3239 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3240 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3241 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3242 .bdrv_refresh_limits
= raw_refresh_limits
,
3243 .bdrv_io_plug
= raw_aio_plug
,
3244 .bdrv_io_unplug
= raw_aio_unplug
,
3245 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3247 .bdrv_co_truncate
= raw_co_truncate
,
3248 .bdrv_getlength
= raw_getlength
,
3249 .bdrv_get_info
= raw_get_info
,
3250 .bdrv_get_allocated_file_size
3251 = raw_get_allocated_file_size
,
3252 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3253 .bdrv_check_perm
= raw_check_perm
,
3254 .bdrv_set_perm
= raw_set_perm
,
3255 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3256 .create_opts
= &raw_create_opts
,
3257 .mutable_opts
= mutable_opts
,
3260 /***********************************************/
3263 #if defined(__APPLE__) && defined(__MACH__)
3264 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3265 CFIndex maxPathSize
, int flags
);
3266 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3268 kern_return_t kernResult
= KERN_FAILURE
;
3269 mach_port_t masterPort
;
3270 CFMutableDictionaryRef classesToMatch
;
3271 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3272 char *mediaType
= NULL
;
3274 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
3275 if ( KERN_SUCCESS
!= kernResult
) {
3276 printf( "IOMasterPort returned %d\n", kernResult
);
3280 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3281 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3282 if (classesToMatch
== NULL
) {
3283 error_report("IOServiceMatching returned NULL for %s",
3284 matching_array
[index
]);
3287 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3289 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
3291 if (kernResult
!= KERN_SUCCESS
) {
3292 error_report("Note: IOServiceGetMatchingServices returned %d",
3297 /* If a match was found, leave the loop */
3298 if (*mediaIterator
!= 0) {
3299 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3300 mediaType
= g_strdup(matching_array
[index
]);
3307 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3308 CFIndex maxPathSize
, int flags
)
3310 io_object_t nextMedia
;
3311 kern_return_t kernResult
= KERN_FAILURE
;
3313 nextMedia
= IOIteratorNext( mediaIterator
);
3316 CFTypeRef bsdPathAsCFString
;
3317 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3318 if ( bsdPathAsCFString
) {
3319 size_t devPathLength
;
3320 strcpy( bsdPath
, _PATH_DEV
);
3321 if (flags
& BDRV_O_NOCACHE
) {
3322 strcat(bsdPath
, "r");
3324 devPathLength
= strlen( bsdPath
);
3325 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3326 kernResult
= KERN_SUCCESS
;
3328 CFRelease( bsdPathAsCFString
);
3330 IOObjectRelease( nextMedia
);
3336 /* Sets up a real cdrom for use in QEMU */
3337 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3339 int index
, num_of_test_partitions
= 2, fd
;
3340 char test_partition
[MAXPATHLEN
];
3341 bool partition_found
= false;
3343 /* look for a working partition */
3344 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3345 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3347 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
, NULL
);
3349 partition_found
= true;
3355 /* if a working partition on the device was not found */
3356 if (partition_found
== false) {
3357 error_setg(errp
, "Failed to find a working partition on disc");
3359 trace_file_setup_cdrom(test_partition
);
3360 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3362 return partition_found
;
3365 /* Prints directions on mounting and unmounting a device */
3366 static void print_unmounting_directions(const char *file_name
)
3368 error_report("If device %s is mounted on the desktop, unmount"
3369 " it first before using it in QEMU", file_name
);
3370 error_report("Command to unmount device: diskutil unmountDisk %s",
3372 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3375 #endif /* defined(__APPLE__) && defined(__MACH__) */
3377 static int hdev_probe_device(const char *filename
)
3381 /* allow a dedicated CD-ROM driver to match with a higher priority */
3382 if (strstart(filename
, "/dev/cdrom", NULL
))
3385 if (stat(filename
, &st
) >= 0 &&
3386 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3393 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3396 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3399 static bool hdev_is_sg(BlockDriverState
*bs
)
3402 #if defined(__linux__)
3404 BDRVRawState
*s
= bs
->opaque
;
3406 struct sg_scsi_id scsiid
;
3410 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3414 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3419 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3421 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3430 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3433 BDRVRawState
*s
= bs
->opaque
;
3436 #if defined(__APPLE__) && defined(__MACH__)
3438 * Caution: while qdict_get_str() is fine, getting non-string types
3439 * would require more care. When @options come from -blockdev or
3440 * blockdev_add, its members are typed according to the QAPI
3441 * schema, but when they come from -drive, they're all QString.
3443 const char *filename
= qdict_get_str(options
, "filename");
3444 char bsd_path
[MAXPATHLEN
] = "";
3445 bool error_occurred
= false;
3447 /* If using a real cdrom */
3448 if (strcmp(filename
, "/dev/cdrom") == 0) {
3449 char *mediaType
= NULL
;
3450 kern_return_t ret_val
;
3451 io_iterator_t mediaIterator
= 0;
3453 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3454 if (mediaType
== NULL
) {
3455 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3457 error_occurred
= true;
3458 goto hdev_open_Mac_error
;
3461 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3462 if (ret_val
!= KERN_SUCCESS
) {
3463 error_setg(errp
, "Could not get BSD path for optical drive");
3464 error_occurred
= true;
3465 goto hdev_open_Mac_error
;
3468 /* If a real optical drive was not found */
3469 if (bsd_path
[0] == '\0') {
3470 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3471 error_occurred
= true;
3472 goto hdev_open_Mac_error
;
3475 /* If using a cdrom disc and finding a partition on the disc failed */
3476 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3477 setup_cdrom(bsd_path
, errp
) == false) {
3478 print_unmounting_directions(bsd_path
);
3479 error_occurred
= true;
3480 goto hdev_open_Mac_error
;
3483 qdict_put_str(options
, "filename", bsd_path
);
3485 hdev_open_Mac_error
:
3487 if (mediaIterator
) {
3488 IOObjectRelease(mediaIterator
);
3490 if (error_occurred
) {
3494 #endif /* defined(__APPLE__) && defined(__MACH__) */
3496 s
->type
= FTYPE_FILE
;
3498 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
3500 #if defined(__APPLE__) && defined(__MACH__)
3502 filename
= bsd_path
;
3504 /* if a physical device experienced an error while being opened */
3505 if (strncmp(filename
, "/dev/", 5) == 0) {
3506 print_unmounting_directions(filename
);
3508 #endif /* defined(__APPLE__) && defined(__MACH__) */
3512 /* Since this does ioctl the device must be already opened */
3513 bs
->sg
= hdev_is_sg(bs
);
3518 #if defined(__linux__)
3519 static int coroutine_fn
3520 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3522 BDRVRawState
*s
= bs
->opaque
;
3523 RawPosixAIOData acb
;
3531 if (req
== SG_IO
&& s
->pr_mgr
) {
3532 struct sg_io_hdr
*io_hdr
= buf
;
3533 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3534 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3535 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3540 acb
= (RawPosixAIOData
) {
3542 .aio_type
= QEMU_AIO_IOCTL
,
3543 .aio_fildes
= s
->fd
,
3551 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3555 static int fd_open(BlockDriverState
*bs
)
3557 BDRVRawState
*s
= bs
->opaque
;
3559 /* this is just to ensure s->fd is sane (its called by io ops) */
3565 static coroutine_fn
int
3566 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3568 BDRVRawState
*s
= bs
->opaque
;
3573 raw_account_discard(s
, bytes
, ret
);
3576 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3579 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3580 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3589 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3592 static BlockDriver bdrv_host_device
= {
3593 .format_name
= "host_device",
3594 .protocol_name
= "host_device",
3595 .instance_size
= sizeof(BDRVRawState
),
3596 .bdrv_needs_filename
= true,
3597 .bdrv_probe_device
= hdev_probe_device
,
3598 .bdrv_parse_filename
= hdev_parse_filename
,
3599 .bdrv_file_open
= hdev_open
,
3600 .bdrv_close
= raw_close
,
3601 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3602 .bdrv_reopen_commit
= raw_reopen_commit
,
3603 .bdrv_reopen_abort
= raw_reopen_abort
,
3604 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3605 .create_opts
= &bdrv_create_opts_simple
,
3606 .mutable_opts
= mutable_opts
,
3607 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3608 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3610 .bdrv_co_preadv
= raw_co_preadv
,
3611 .bdrv_co_pwritev
= raw_co_pwritev
,
3612 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3613 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3614 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3615 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3616 .bdrv_refresh_limits
= raw_refresh_limits
,
3617 .bdrv_io_plug
= raw_aio_plug
,
3618 .bdrv_io_unplug
= raw_aio_unplug
,
3619 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3621 .bdrv_co_truncate
= raw_co_truncate
,
3622 .bdrv_getlength
= raw_getlength
,
3623 .bdrv_get_info
= raw_get_info
,
3624 .bdrv_get_allocated_file_size
3625 = raw_get_allocated_file_size
,
3626 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
3627 .bdrv_check_perm
= raw_check_perm
,
3628 .bdrv_set_perm
= raw_set_perm
,
3629 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3630 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3631 .bdrv_probe_geometry
= hdev_probe_geometry
,
3633 /* generic scsi device */
3635 .bdrv_co_ioctl
= hdev_co_ioctl
,
3639 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3640 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3643 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3648 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3651 BDRVRawState
*s
= bs
->opaque
;
3655 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3656 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3659 static int cdrom_probe_device(const char *filename
)
3665 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
, NULL
);
3669 ret
= fstat(fd
, &st
);
3670 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3674 /* Attempt to detect via a CDROM specific ioctl */
3675 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3685 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3687 BDRVRawState
*s
= bs
->opaque
;
3690 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3691 return ret
== CDS_DISC_OK
;
3694 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3696 BDRVRawState
*s
= bs
->opaque
;
3699 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3700 perror("CDROMEJECT");
3702 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3703 perror("CDROMEJECT");
3707 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3709 BDRVRawState
*s
= bs
->opaque
;
3711 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3713 * Note: an error can happen if the distribution automatically
3716 /* perror("CDROM_LOCKDOOR"); */
3720 static BlockDriver bdrv_host_cdrom
= {
3721 .format_name
= "host_cdrom",
3722 .protocol_name
= "host_cdrom",
3723 .instance_size
= sizeof(BDRVRawState
),
3724 .bdrv_needs_filename
= true,
3725 .bdrv_probe_device
= cdrom_probe_device
,
3726 .bdrv_parse_filename
= cdrom_parse_filename
,
3727 .bdrv_file_open
= cdrom_open
,
3728 .bdrv_close
= raw_close
,
3729 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3730 .bdrv_reopen_commit
= raw_reopen_commit
,
3731 .bdrv_reopen_abort
= raw_reopen_abort
,
3732 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3733 .create_opts
= &bdrv_create_opts_simple
,
3734 .mutable_opts
= mutable_opts
,
3735 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3737 .bdrv_co_preadv
= raw_co_preadv
,
3738 .bdrv_co_pwritev
= raw_co_pwritev
,
3739 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3740 .bdrv_refresh_limits
= raw_refresh_limits
,
3741 .bdrv_io_plug
= raw_aio_plug
,
3742 .bdrv_io_unplug
= raw_aio_unplug
,
3743 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3745 .bdrv_co_truncate
= raw_co_truncate
,
3746 .bdrv_getlength
= raw_getlength
,
3747 .has_variable_length
= true,
3748 .bdrv_get_allocated_file_size
3749 = raw_get_allocated_file_size
,
3751 /* removable device support */
3752 .bdrv_is_inserted
= cdrom_is_inserted
,
3753 .bdrv_eject
= cdrom_eject
,
3754 .bdrv_lock_medium
= cdrom_lock_medium
,
3756 /* generic scsi device */
3757 .bdrv_co_ioctl
= hdev_co_ioctl
,
3759 #endif /* __linux__ */
3761 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3762 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3765 BDRVRawState
*s
= bs
->opaque
;
3770 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
3775 /* make sure the door isn't locked at this time */
3776 ioctl(s
->fd
, CDIOCALLOW
);
3780 static int cdrom_probe_device(const char *filename
)
3782 if (strstart(filename
, "/dev/cd", NULL
) ||
3783 strstart(filename
, "/dev/acd", NULL
))
3788 static int cdrom_reopen(BlockDriverState
*bs
)
3790 BDRVRawState
*s
= bs
->opaque
;
3794 * Force reread of possibly changed/newly loaded disc,
3795 * FreeBSD seems to not notice sometimes...
3799 fd
= qemu_open(bs
->filename
, s
->open_flags
, NULL
);
3806 /* make sure the door isn't locked at this time */
3807 ioctl(s
->fd
, CDIOCALLOW
);
3811 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3813 return raw_getlength(bs
) > 0;
3816 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3818 BDRVRawState
*s
= bs
->opaque
;
3823 (void) ioctl(s
->fd
, CDIOCALLOW
);
3826 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3827 perror("CDIOCEJECT");
3829 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3830 perror("CDIOCCLOSE");
3836 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3838 BDRVRawState
*s
= bs
->opaque
;
3842 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3844 * Note: an error can happen if the distribution automatically
3847 /* perror("CDROM_LOCKDOOR"); */
3851 static BlockDriver bdrv_host_cdrom
= {
3852 .format_name
= "host_cdrom",
3853 .protocol_name
= "host_cdrom",
3854 .instance_size
= sizeof(BDRVRawState
),
3855 .bdrv_needs_filename
= true,
3856 .bdrv_probe_device
= cdrom_probe_device
,
3857 .bdrv_parse_filename
= cdrom_parse_filename
,
3858 .bdrv_file_open
= cdrom_open
,
3859 .bdrv_close
= raw_close
,
3860 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3861 .bdrv_reopen_commit
= raw_reopen_commit
,
3862 .bdrv_reopen_abort
= raw_reopen_abort
,
3863 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3864 .create_opts
= &bdrv_create_opts_simple
,
3865 .mutable_opts
= mutable_opts
,
3867 .bdrv_co_preadv
= raw_co_preadv
,
3868 .bdrv_co_pwritev
= raw_co_pwritev
,
3869 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3870 .bdrv_refresh_limits
= raw_refresh_limits
,
3871 .bdrv_io_plug
= raw_aio_plug
,
3872 .bdrv_io_unplug
= raw_aio_unplug
,
3873 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3875 .bdrv_co_truncate
= raw_co_truncate
,
3876 .bdrv_getlength
= raw_getlength
,
3877 .has_variable_length
= true,
3878 .bdrv_get_allocated_file_size
3879 = raw_get_allocated_file_size
,
3881 /* removable device support */
3882 .bdrv_is_inserted
= cdrom_is_inserted
,
3883 .bdrv_eject
= cdrom_eject
,
3884 .bdrv_lock_medium
= cdrom_lock_medium
,
3886 #endif /* __FreeBSD__ */
3888 static void bdrv_file_init(void)
3891 * Register all the drivers. Note that order is important, the driver
3892 * registered last will get probed first.
3894 bdrv_register(&bdrv_file
);
3895 bdrv_register(&bdrv_host_device
);
3897 bdrv_register(&bdrv_host_cdrom
);
3899 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3900 bdrv_register(&bdrv_host_cdrom
);
3904 block_init(bdrv_file_init
);