2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
31 #include "qemu/option.h"
32 #include "qemu/units.h"
33 #include "qemu/memalign.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__)
45 #include <sys/ioctl.h>
46 #if defined(HAVE_HOST_BLOCK_DEVICE)
48 #include <sys/param.h>
49 #include <sys/mount.h>
50 #include <IOKit/IOKitLib.h>
51 #include <IOKit/IOBSD.h>
52 #include <IOKit/storage/IOMediaBSDClient.h>
53 #include <IOKit/storage/IOMedia.h>
54 #include <IOKit/storage/IOCDMedia.h>
55 //#include <IOKit/storage/IOCDTypes.h>
56 #include <IOKit/storage/IODVDMedia.h>
57 #include <CoreFoundation/CoreFoundation.h>
58 #endif /* defined(HAVE_HOST_BLOCK_DEVICE) */
62 #define _POSIX_PTHREAD_SEMANTICS 1
66 #include <sys/ioctl.h>
67 #include <sys/param.h>
68 #include <sys/syscall.h>
70 #include <linux/cdrom.h>
73 #include <linux/hdreg.h>
74 #include <linux/magic.h>
80 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
83 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
84 #include <linux/falloc.h>
86 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
92 #include <sys/ioctl.h>
93 #include <sys/disklabel.h>
98 #include <sys/ioctl.h>
99 #include <sys/disklabel.h>
100 #include <sys/dkio.h>
101 #include <sys/disk.h>
105 #include <sys/ioctl.h>
106 #include <sys/diskslice.h>
109 /* OS X does not have O_DSYNC */
112 #define O_DSYNC O_SYNC
113 #elif defined(O_FSYNC)
114 #define O_DSYNC O_FSYNC
118 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
120 #define O_DIRECT O_DSYNC
126 #define MAX_BLOCKSIZE 4096
128 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
129 * leaving a few more bytes for its future use. */
130 #define RAW_LOCK_PERM_BASE 100
131 #define RAW_LOCK_SHARED_BASE 200
133 typedef struct BDRVRawState
{
140 /* The current permissions. */
142 uint64_t shared_perm
;
144 /* The perms bits whose corresponding bytes are already locked in
146 uint64_t locked_perm
;
147 uint64_t locked_shared_perm
;
149 uint64_t aio_max_batch
;
152 int perm_change_flags
;
153 BDRVReopenState
*reopen_state
;
156 bool has_write_zeroes
:1;
157 bool use_linux_aio
:1;
158 bool use_linux_io_uring
:1;
159 int page_cache_inconsistent
; /* errno from fdatasync failure */
161 bool needs_alignment
;
162 bool force_alignment
;
164 bool check_cache_dropped
;
166 uint64_t discard_nb_ok
;
167 uint64_t discard_nb_failed
;
168 uint64_t discard_bytes_ok
;
174 typedef struct BDRVRawReopenState
{
177 bool check_cache_dropped
;
178 } BDRVRawReopenState
;
180 static int fd_open(BlockDriverState
*bs
)
182 BDRVRawState
*s
= bs
->opaque
;
184 /* this is just to ensure s->fd is sane (its called by io ops) */
191 static int64_t raw_getlength(BlockDriverState
*bs
);
193 typedef struct RawPosixAIOData
{
194 BlockDriverState
*bs
;
215 PreallocMode prealloc
;
221 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
222 static int cdrom_reopen(BlockDriverState
*bs
);
226 * Elide EAGAIN and EACCES details when failing to lock, as this
227 * indicates that the specified file region is already locked by
228 * another process, which is considered a common scenario.
230 #define raw_lock_error_setg_errno(errp, err, fmt, ...) \
232 if ((err) == EAGAIN || (err) == EACCES) { \
233 error_setg((errp), (fmt), ## __VA_ARGS__); \
235 error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \
239 #if defined(__NetBSD__)
240 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
242 static char namebuf
[PATH_MAX
];
243 const char *dp
, *fname
;
247 dp
= strrchr(fname
, '/');
248 if (lstat(fname
, &sb
) < 0) {
249 error_setg_file_open(errp
, errno
, fname
);
253 if (!S_ISBLK(sb
.st_mode
)) {
258 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
260 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
261 (int)(dp
- fname
), fname
, dp
+ 1);
264 warn_report("%s is a block device, using %s", fname
, *filename
);
269 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
276 * Get logical block size via ioctl. On success store it in @sector_size_p.
278 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
280 unsigned int sector_size
;
281 bool success
= false;
285 static const unsigned long ioctl_list
[] = {
289 #ifdef DKIOCGETBLOCKSIZE
292 #ifdef DIOCGSECTORSIZE
297 /* Try a few ioctls to get the right size */
298 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
299 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
300 *sector_size_p
= sector_size
;
305 return success
? 0 : -errno
;
309 * Get physical block size of @fd.
310 * On success, store it in @blk_size and return 0.
311 * On failure, return -errno.
313 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
316 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
326 * Returns true if no alignment restrictions are necessary even for files
327 * opened with O_DIRECT.
329 * raw_probe_alignment() probes the required alignment and assume that 1 means
330 * the probing failed, so it falls back to a safe default of 4k. This can be
331 * avoided if we know that byte alignment is okay for the file.
333 static bool dio_byte_aligned(int fd
)
339 ret
= fstatfs(fd
, &buf
);
340 if (ret
== 0 && buf
.f_type
== NFS_SUPER_MAGIC
) {
347 static bool raw_needs_alignment(BlockDriverState
*bs
)
349 BDRVRawState
*s
= bs
->opaque
;
351 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0 && !dio_byte_aligned(s
->fd
)) {
355 return s
->force_alignment
;
358 /* Check if read is allowed with given memory buffer and length.
360 * This function is used to check O_DIRECT memory buffer and request alignment.
362 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
364 ssize_t ret
= pread(fd
, buf
, len
, 0);
371 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
372 * other errors (e.g. real I/O error), which could happen on a failed
373 * drive, since we only care about probing alignment.
375 if (errno
!= EINVAL
) {
383 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
385 BDRVRawState
*s
= bs
->opaque
;
387 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size());
388 size_t alignments
[] = {1, 512, 1024, 2048, 4096};
390 /* For SCSI generic devices the alignment is not really used.
391 With buffered I/O, we don't have any restrictions. */
392 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
393 bs
->bl
.request_alignment
= 1;
398 bs
->bl
.request_alignment
= 0;
400 /* Let's try to use the logical blocksize for the alignment. */
401 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
402 bs
->bl
.request_alignment
= 0;
407 * The XFS ioctl definitions are shipped in extra packages that might
408 * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl
409 * here, we simply use our own definition instead:
416 if (ioctl(fd
, _IOR('X', 30, struct xfs_dioattr
), &da
) >= 0) {
417 bs
->bl
.request_alignment
= da
.d_miniosz
;
418 /* The kernel returns wrong information for d_mem */
419 /* s->buf_align = da.d_mem; */
424 * If we could not get the sizes so far, we can only guess them. First try
425 * to detect request alignment, since it is more likely to succeed. Then
426 * try to detect buf_align, which cannot be detected in some cases (e.g.
427 * Gluster). If buf_align cannot be detected, we fallback to the value of
431 if (!bs
->bl
.request_alignment
) {
434 buf
= qemu_memalign(max_align
, max_align
);
435 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
436 align
= alignments
[i
];
437 if (raw_is_io_aligned(fd
, buf
, align
)) {
438 /* Fallback to safe value. */
439 bs
->bl
.request_alignment
= (align
!= 1) ? align
: max_align
;
449 buf
= qemu_memalign(max_align
, 2 * max_align
);
450 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
451 align
= alignments
[i
];
452 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
453 /* Fallback to request_alignment. */
454 s
->buf_align
= (align
!= 1) ? align
: bs
->bl
.request_alignment
;
461 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
462 error_setg(errp
, "Could not find working O_DIRECT alignment");
463 error_append_hint(errp
, "Try cache.direct=off\n");
467 static int check_hdev_writable(int fd
)
469 #if defined(BLKROGET)
470 /* Linux block devices can be configured "read-only" using blockdev(8).
471 * This is independent of device node permissions and therefore open(2)
472 * with O_RDWR succeeds. Actual writes fail with EPERM.
474 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
475 * check for read-only block devices so that Linux block devices behave
481 if (fstat(fd
, &st
)) {
485 if (!S_ISBLK(st
.st_mode
)) {
489 if (ioctl(fd
, BLKROGET
, &readonly
) < 0) {
496 #endif /* defined(BLKROGET) */
500 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
502 bool read_write
= false;
503 assert(open_flags
!= NULL
);
505 *open_flags
|= O_BINARY
;
506 *open_flags
&= ~O_ACCMODE
;
508 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
509 read_write
= has_writers
;
510 } else if (bdrv_flags
& BDRV_O_RDWR
) {
515 *open_flags
|= O_RDWR
;
517 *open_flags
|= O_RDONLY
;
520 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
521 * and O_DIRECT for no caching. */
522 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
523 *open_flags
|= O_DIRECT
;
527 static void raw_parse_filename(const char *filename
, QDict
*options
,
530 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
533 static QemuOptsList raw_runtime_opts
= {
535 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
539 .type
= QEMU_OPT_STRING
,
540 .help
= "File name of the image",
544 .type
= QEMU_OPT_STRING
,
545 .help
= "host AIO implementation (threads, native, io_uring)",
548 .name
= "aio-max-batch",
549 .type
= QEMU_OPT_NUMBER
,
550 .help
= "AIO max batch size (0 = auto handled by AIO backend, default: 0)",
554 .type
= QEMU_OPT_STRING
,
555 .help
= "file locking mode (on/off/auto, default: auto)",
558 .name
= "pr-manager",
559 .type
= QEMU_OPT_STRING
,
560 .help
= "id of persistent reservation manager object (default: none)",
562 #if defined(__linux__)
564 .name
= "drop-cache",
565 .type
= QEMU_OPT_BOOL
,
566 .help
= "invalidate page cache during live migration (default: on)",
570 .name
= "x-check-cache-dropped",
571 .type
= QEMU_OPT_BOOL
,
572 .help
= "check that page cache was dropped on live migration (default: off)"
574 { /* end of list */ }
578 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
580 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
581 int bdrv_flags
, int open_flags
,
582 bool device
, Error
**errp
)
584 BDRVRawState
*s
= bs
->opaque
;
586 Error
*local_err
= NULL
;
587 const char *filename
= NULL
;
589 BlockdevAioOptions aio
, aio_default
;
594 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
595 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
600 filename
= qemu_opt_get(opts
, "filename");
602 ret
= raw_normalize_devicepath(&filename
, errp
);
607 if (bdrv_flags
& BDRV_O_NATIVE_AIO
) {
608 aio_default
= BLOCKDEV_AIO_OPTIONS_NATIVE
;
609 #ifdef CONFIG_LINUX_IO_URING
610 } else if (bdrv_flags
& BDRV_O_IO_URING
) {
611 aio_default
= BLOCKDEV_AIO_OPTIONS_IO_URING
;
614 aio_default
= BLOCKDEV_AIO_OPTIONS_THREADS
;
617 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
618 qemu_opt_get(opts
, "aio"),
619 aio_default
, &local_err
);
621 error_propagate(errp
, local_err
);
626 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
627 #ifdef CONFIG_LINUX_IO_URING
628 s
->use_linux_io_uring
= (aio
== BLOCKDEV_AIO_OPTIONS_IO_URING
);
631 s
->aio_max_batch
= qemu_opt_get_number(opts
, "aio-max-batch", 0);
633 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
634 qemu_opt_get(opts
, "locking"),
635 ON_OFF_AUTO_AUTO
, &local_err
);
637 error_propagate(errp
, local_err
);
644 if (!qemu_has_ofd_lock()) {
645 warn_report("File lock requested but OFD locking syscall is "
646 "unavailable, falling back to POSIX file locks");
647 error_printf("Due to the implementation, locks can be lost "
651 case ON_OFF_AUTO_OFF
:
654 case ON_OFF_AUTO_AUTO
:
655 s
->use_lock
= qemu_has_ofd_lock();
661 str
= qemu_opt_get(opts
, "pr-manager");
663 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
665 error_propagate(errp
, local_err
);
671 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
672 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
675 s
->open_flags
= open_flags
;
676 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
679 fd
= qemu_open(filename
, s
->open_flags
, errp
);
680 ret
= fd
< 0 ? -errno
: 0;
690 /* Check s->open_flags rather than bdrv_flags due to auto-read-only */
691 if (s
->open_flags
& O_RDWR
) {
692 ret
= check_hdev_writable(s
->fd
);
694 error_setg_errno(errp
, -ret
, "The device is not writable");
700 s
->shared_perm
= BLK_PERM_ALL
;
702 #ifdef CONFIG_LINUX_AIO
703 /* Currently Linux does AIO only for files opened with O_DIRECT */
704 if (s
->use_linux_aio
) {
705 if (!(s
->open_flags
& O_DIRECT
)) {
706 error_setg(errp
, "aio=native was specified, but it requires "
707 "cache.direct=on, which was not specified.");
711 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
712 error_prepend(errp
, "Unable to use native AIO: ");
717 if (s
->use_linux_aio
) {
718 error_setg(errp
, "aio=native was specified, but is not supported "
723 #endif /* !defined(CONFIG_LINUX_AIO) */
725 #ifdef CONFIG_LINUX_IO_URING
726 if (s
->use_linux_io_uring
) {
727 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs
), errp
)) {
728 error_prepend(errp
, "Unable to use io_uring: ");
733 if (s
->use_linux_io_uring
) {
734 error_setg(errp
, "aio=io_uring was specified, but is not supported "
739 #endif /* !defined(CONFIG_LINUX_IO_URING) */
741 s
->has_discard
= true;
742 s
->has_write_zeroes
= true;
744 if (fstat(s
->fd
, &st
) < 0) {
746 error_setg_errno(errp
, errno
, "Could not stat file");
751 if (!S_ISREG(st
.st_mode
)) {
752 error_setg(errp
, "'%s' driver requires '%s' to be a regular file",
753 bs
->drv
->format_name
, bs
->filename
);
757 s
->has_fallocate
= true;
760 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
761 error_setg(errp
, "'%s' driver requires '%s' to be either "
762 "a character or block device",
763 bs
->drv
->format_name
, bs
->filename
);
769 if (S_ISBLK(st
.st_mode
)) {
771 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
772 * not rely on the contents of discarded blocks unless using O_DIRECT.
773 * Same for BLKZEROOUT.
775 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
776 s
->has_write_zeroes
= false;
781 if (S_ISCHR(st
.st_mode
)) {
783 * The file is a char device (disk), which on FreeBSD isn't behind
784 * a pager, so force all requests to be aligned. This is needed
785 * so QEMU makes sure all IO operations on the device are aligned
786 * to sector size, or else FreeBSD will reject them with EINVAL.
788 s
->force_alignment
= true;
791 s
->needs_alignment
= raw_needs_alignment(bs
);
793 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
794 if (S_ISREG(st
.st_mode
)) {
795 /* When extending regular files, we get zeros from the OS */
796 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
800 if (ret
< 0 && s
->fd
!= -1) {
803 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
810 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
813 BDRVRawState
*s
= bs
->opaque
;
815 s
->type
= FTYPE_FILE
;
816 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
825 #define PERM_FOREACH(i) \
826 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
828 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
829 * file; if @unlock == true, also unlock the unneeded bytes.
830 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
832 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
833 uint64_t perm_lock_bits
,
834 uint64_t shared_perm_lock_bits
,
835 bool unlock
, Error
**errp
)
839 uint64_t locked_perm
, locked_shared_perm
;
842 locked_perm
= s
->locked_perm
;
843 locked_shared_perm
= s
->locked_shared_perm
;
846 * We don't have the previous bits, just lock/unlock for each of the
850 locked_perm
= BLK_PERM_ALL
;
851 locked_shared_perm
= BLK_PERM_ALL
;
854 locked_shared_perm
= 0;
859 int off
= RAW_LOCK_PERM_BASE
+ i
;
860 uint64_t bit
= (1ULL << i
);
861 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
862 ret
= qemu_lock_fd(fd
, off
, 1, false);
864 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
868 s
->locked_perm
|= bit
;
870 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
871 ret
= qemu_unlock_fd(fd
, off
, 1);
873 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
876 s
->locked_perm
&= ~bit
;
881 int off
= RAW_LOCK_SHARED_BASE
+ i
;
882 uint64_t bit
= (1ULL << i
);
883 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
884 ret
= qemu_lock_fd(fd
, off
, 1, false);
886 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
890 s
->locked_shared_perm
|= bit
;
892 } else if (unlock
&& (locked_shared_perm
& bit
) &&
893 !(shared_perm_lock_bits
& bit
)) {
894 ret
= qemu_unlock_fd(fd
, off
, 1);
896 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
899 s
->locked_shared_perm
&= ~bit
;
906 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
907 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
914 int off
= RAW_LOCK_SHARED_BASE
+ i
;
915 uint64_t p
= 1ULL << i
;
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 \"%s\" lock",
930 int off
= RAW_LOCK_PERM_BASE
+ i
;
931 uint64_t p
= 1ULL << i
;
932 if (!(shared_perm
& p
)) {
933 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
935 char *perm_name
= bdrv_perm_names(p
);
937 raw_lock_error_setg_errno(errp
, -ret
,
938 "Failed to get shared \"%s\" lock",
948 static int raw_handle_perm_lock(BlockDriverState
*bs
,
950 uint64_t new_perm
, uint64_t new_shared
,
953 BDRVRawState
*s
= bs
->opaque
;
955 Error
*local_err
= NULL
;
961 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
967 if ((s
->perm
| new_perm
) == s
->perm
&&
968 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
971 * We are going to unlock bytes, it should not fail. If it fail due
972 * to some fs-dependent permission-unrelated reasons (which occurs
973 * sometimes on NFS and leads to abort in bdrv_replace_child) we
974 * can't prevent such errors by any check here. And we ignore them
975 * anyway in ABORT and COMMIT.
979 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
980 ~s
->shared_perm
| ~new_shared
,
983 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
987 error_append_hint(errp
,
988 "Is another process using the image [%s]?\n",
991 /* fall through to unlock bytes. */
993 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
996 /* Theoretically the above call only unlocks bytes and it cannot
997 * fail. Something weird happened, report it.
999 warn_report_err(local_err
);
1003 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
1006 /* Theoretically the above call only unlocks bytes and it cannot
1007 * fail. Something weird happened, report it.
1009 warn_report_err(local_err
);
1016 /* Sets a specific flag */
1017 static int fcntl_setfl(int fd
, int flag
)
1021 flags
= fcntl(fd
, F_GETFL
);
1025 if (fcntl(fd
, F_SETFL
, flags
| flag
) == -1) {
1031 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
1032 int *open_flags
, uint64_t perm
, bool force_dup
,
1035 BDRVRawState
*s
= bs
->opaque
;
1038 bool has_writers
= perm
&
1039 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
1040 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
1042 fcntl_flags
|= O_NOATIME
;
1046 if (s
->type
== FTYPE_CD
) {
1047 *open_flags
|= O_NONBLOCK
;
1050 raw_parse_flags(flags
, open_flags
, has_writers
);
1053 /* Not all operating systems have O_ASYNC, and those that don't
1054 * will not let us track the state into rs->open_flags (typically
1055 * you achieve the same effect with an ioctl, for example I_SETSIG
1056 * on Solaris). But we do not use O_ASYNC, so that's fine.
1058 assert((s
->open_flags
& O_ASYNC
) == 0);
1061 if (!force_dup
&& *open_flags
== s
->open_flags
) {
1062 /* We're lucky, the existing fd is fine */
1066 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
1067 /* dup the original fd */
1068 fd
= qemu_dup(s
->fd
);
1070 ret
= fcntl_setfl(fd
, *open_flags
);
1078 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
1080 const char *normalized_filename
= bs
->filename
;
1081 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
1083 fd
= qemu_open(normalized_filename
, *open_flags
, errp
);
1090 if (fd
!= -1 && (*open_flags
& O_RDWR
)) {
1091 ret
= check_hdev_writable(fd
);
1094 error_setg_errno(errp
, -ret
, "The device is not writable");
1102 static int raw_reopen_prepare(BDRVReopenState
*state
,
1103 BlockReopenQueue
*queue
, Error
**errp
)
1106 BDRVRawReopenState
*rs
;
1110 assert(state
!= NULL
);
1111 assert(state
->bs
!= NULL
);
1113 s
= state
->bs
->opaque
;
1115 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
1118 /* Handle options changes */
1119 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
1120 if (!qemu_opts_absorb_qdict(opts
, state
->options
, errp
)) {
1125 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
1126 rs
->check_cache_dropped
=
1127 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
1129 /* This driver's reopen function doesn't currently allow changing
1130 * other options, so let's put them back in the original QDict and
1131 * bdrv_reopen_prepare() will detect changes and complain. */
1132 qemu_opts_to_qdict(opts
, state
->options
);
1135 * As part of reopen prepare we also want to create new fd by
1136 * raw_reconfigure_getfd(). But it wants updated "perm", when in
1137 * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to
1138 * permission update. Happily, permission update is always a part (a seprate
1139 * stage) of bdrv_reopen_multiple() so we can rely on this fact and
1140 * reconfigure fd in raw_check_perm().
1143 s
->reopen_state
= state
;
1147 qemu_opts_del(opts
);
1151 static void raw_reopen_commit(BDRVReopenState
*state
)
1153 BDRVRawReopenState
*rs
= state
->opaque
;
1154 BDRVRawState
*s
= state
->bs
->opaque
;
1156 s
->drop_cache
= rs
->drop_cache
;
1157 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1158 s
->open_flags
= rs
->open_flags
;
1159 g_free(state
->opaque
);
1160 state
->opaque
= NULL
;
1162 assert(s
->reopen_state
== state
);
1163 s
->reopen_state
= NULL
;
1167 static void raw_reopen_abort(BDRVReopenState
*state
)
1169 BDRVRawReopenState
*rs
= state
->opaque
;
1170 BDRVRawState
*s
= state
->bs
->opaque
;
1172 /* nothing to do if NULL, we didn't get far enough */
1177 g_free(state
->opaque
);
1178 state
->opaque
= NULL
;
1180 assert(s
->reopen_state
== state
);
1181 s
->reopen_state
= NULL
;
1184 static int hdev_get_max_hw_transfer(int fd
, struct stat
*st
)
1187 if (S_ISBLK(st
->st_mode
)) {
1188 unsigned short max_sectors
= 0;
1189 if (ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1190 return max_sectors
* 512;
1194 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1204 static int hdev_get_max_segments(int fd
, struct stat
*st
)
1209 char *sysfspath
= NULL
;
1214 if (S_ISCHR(st
->st_mode
)) {
1215 if (ioctl(fd
, SG_GET_SG_TABLESIZE
, &ret
) == 0) {
1221 if (!S_ISBLK(st
->st_mode
)) {
1225 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1226 major(st
->st_rdev
), minor(st
->st_rdev
));
1227 sysfd
= open(sysfspath
, O_RDONLY
);
1233 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1234 } while (ret
== -1 && errno
== EINTR
);
1238 } else if (ret
== 0) {
1243 /* The file is ended with '\n', pass 'end' to accept that. */
1244 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1245 if (ret
== 0 && end
&& *end
== '\n') {
1260 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1262 BDRVRawState
*s
= bs
->opaque
;
1265 s
->needs_alignment
= raw_needs_alignment(bs
);
1266 raw_probe_alignment(bs
, s
->fd
, errp
);
1268 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1269 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size());
1272 * Maximum transfers are best effort, so it is okay to ignore any
1273 * errors. That said, based on the man page errors in fstat would be
1274 * very much unexpected; the only possible case seems to be ENOMEM.
1276 if (fstat(s
->fd
, &st
)) {
1280 #if defined(__APPLE__) && (__MACH__)
1283 if (!fstatfs(s
->fd
, &buf
)) {
1284 bs
->bl
.opt_transfer
= buf
.f_iosize
;
1285 bs
->bl
.pdiscard_alignment
= buf
.f_bsize
;
1289 if (bdrv_is_sg(bs
) || S_ISBLK(st
.st_mode
)) {
1290 int ret
= hdev_get_max_hw_transfer(s
->fd
, &st
);
1292 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1293 bs
->bl
.max_hw_transfer
= ret
;
1296 ret
= hdev_get_max_segments(s
->fd
, &st
);
1298 bs
->bl
.max_hw_iov
= ret
;
1303 static int check_for_dasd(int fd
)
1306 struct dasd_information2_t info
= {0};
1308 return ioctl(fd
, BIODASDINFO2
, &info
);
1315 * Try to get @bs's logical and physical block size.
1316 * On success, store them in @bsz and return zero.
1317 * On failure, return negative errno.
1319 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1321 BDRVRawState
*s
= bs
->opaque
;
1324 /* If DASD, get blocksizes */
1325 if (check_for_dasd(s
->fd
) < 0) {
1328 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1332 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1336 * Try to get @bs's geometry: cyls, heads, sectors.
1337 * On success, store them in @geo and return 0.
1338 * On failure return -errno.
1339 * (Allows block driver to assign default geometry values that guest sees)
1342 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1344 BDRVRawState
*s
= bs
->opaque
;
1345 struct hd_geometry ioctl_geo
= {0};
1347 /* If DASD, get its geometry */
1348 if (check_for_dasd(s
->fd
) < 0) {
1351 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1354 /* HDIO_GETGEO may return success even though geo contains zeros
1355 (e.g. certain multipath setups) */
1356 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1359 /* Do not return a geometry for partition */
1360 if (ioctl_geo
.start
!= 0) {
1363 geo
->heads
= ioctl_geo
.heads
;
1364 geo
->sectors
= ioctl_geo
.sectors
;
1365 geo
->cylinders
= ioctl_geo
.cylinders
;
1369 #else /* __linux__ */
1370 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1376 #if defined(__linux__)
1377 static int handle_aiocb_ioctl(void *opaque
)
1379 RawPosixAIOData
*aiocb
= opaque
;
1383 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1384 } while (ret
== -1 && errno
== EINTR
);
1393 static int handle_aiocb_flush(void *opaque
)
1395 RawPosixAIOData
*aiocb
= opaque
;
1396 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1399 if (s
->page_cache_inconsistent
) {
1400 return -s
->page_cache_inconsistent
;
1403 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1405 trace_file_flush_fdatasync_failed(errno
);
1407 /* There is no clear definition of the semantics of a failing fsync(),
1408 * so we may have to assume the worst. The sad truth is that this
1409 * assumption is correct for Linux. Some pages are now probably marked
1410 * clean in the page cache even though they are inconsistent with the
1411 * on-disk contents. The next fdatasync() call would succeed, but no
1412 * further writeback attempt will be made. We can't get back to a state
1413 * in which we know what is on disk (we would have to rewrite
1414 * everything that was touched since the last fdatasync() at least), so
1415 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1416 * really defined, I have little hope that other OSes are doing better.
1418 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1420 if ((s
->open_flags
& O_DIRECT
) == 0) {
1421 s
->page_cache_inconsistent
= errno
;
1428 #ifdef CONFIG_PREADV
1430 static bool preadv_present
= true;
1433 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1435 return preadv(fd
, iov
, nr_iov
, offset
);
1439 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1441 return pwritev(fd
, iov
, nr_iov
, offset
);
1446 static bool preadv_present
= false;
1449 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1455 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1462 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1467 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1468 len
= qemu_pwritev(aiocb
->aio_fildes
,
1473 len
= qemu_preadv(aiocb
->aio_fildes
,
1477 } while (len
== -1 && errno
== EINTR
);
1486 * Read/writes the data to/from a given linear buffer.
1488 * Returns the number of bytes handles or -errno in case of an error. Short
1489 * reads are only returned if the end of the file is reached.
1491 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1496 while (offset
< aiocb
->aio_nbytes
) {
1497 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1498 len
= pwrite(aiocb
->aio_fildes
,
1499 (const char *)buf
+ offset
,
1500 aiocb
->aio_nbytes
- offset
,
1501 aiocb
->aio_offset
+ offset
);
1503 len
= pread(aiocb
->aio_fildes
,
1505 aiocb
->aio_nbytes
- offset
,
1506 aiocb
->aio_offset
+ offset
);
1508 if (len
== -1 && errno
== EINTR
) {
1510 } else if (len
== -1 && errno
== EINVAL
&&
1511 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1512 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1514 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1515 * after a short read. Assume that O_DIRECT short reads only occur
1516 * at EOF. Therefore this is a short read, not an I/O error.
1519 } else if (len
== -1) {
1522 } else if (len
== 0) {
1531 static int handle_aiocb_rw(void *opaque
)
1533 RawPosixAIOData
*aiocb
= opaque
;
1537 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1539 * If there is just a single buffer, and it is properly aligned
1540 * we can just use plain pread/pwrite without any problems.
1542 if (aiocb
->io
.niov
== 1) {
1543 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1547 * We have more than one iovec, and all are properly aligned.
1549 * Try preadv/pwritev first and fall back to linearizing the
1550 * buffer if it's not supported.
1552 if (preadv_present
) {
1553 nbytes
= handle_aiocb_rw_vector(aiocb
);
1554 if (nbytes
== aiocb
->aio_nbytes
||
1555 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1558 preadv_present
= false;
1562 * XXX(hch): short read/write. no easy way to handle the reminder
1563 * using these interfaces. For now retry using plain
1569 * Ok, we have to do it the hard way, copy all segments into
1570 * a single aligned buffer.
1572 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1578 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1582 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1583 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1584 p
+= aiocb
->io
.iov
[i
].iov_len
;
1586 assert(p
- buf
== aiocb
->aio_nbytes
);
1589 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1590 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1592 size_t count
= aiocb
->aio_nbytes
, copy
;
1595 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1597 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1598 copy
= aiocb
->io
.iov
[i
].iov_len
;
1600 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1601 assert(count
>= copy
);
1610 if (nbytes
== aiocb
->aio_nbytes
) {
1612 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1613 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1616 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1617 0, aiocb
->aio_nbytes
- nbytes
);
1626 #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD)
1627 static int translate_err(int err
)
1629 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1637 #ifdef CONFIG_FALLOCATE
1638 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1641 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1644 } while (errno
== EINTR
);
1645 return translate_err(-errno
);
1649 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1652 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1654 if (!s
->has_write_zeroes
) {
1659 /* The BLKZEROOUT implementation in the kernel doesn't set
1660 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1662 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1664 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1665 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1668 } while (errno
== EINTR
);
1670 ret
= translate_err(-errno
);
1671 if (ret
== -ENOTSUP
) {
1672 s
->has_write_zeroes
= false;
1680 static int handle_aiocb_write_zeroes(void *opaque
)
1682 RawPosixAIOData
*aiocb
= opaque
;
1683 #ifdef CONFIG_FALLOCATE
1684 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1688 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1689 return handle_aiocb_write_zeroes_block(aiocb
);
1692 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1693 if (s
->has_write_zeroes
) {
1694 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1695 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1696 if (ret
== -ENOTSUP
) {
1697 s
->has_write_zeroes
= false;
1698 } else if (ret
== 0 || ret
!= -EINVAL
) {
1702 * Note: Some file systems do not like unaligned byte ranges, and
1703 * return EINVAL in such a case, though they should not do it according
1704 * to the man-page of fallocate(). Thus we simply ignore this return
1705 * value and try the other fallbacks instead.
1710 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1711 if (s
->has_discard
&& s
->has_fallocate
) {
1712 int ret
= do_fallocate(s
->fd
,
1713 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1714 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1716 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1717 if (ret
== 0 || ret
!= -ENOTSUP
) {
1720 s
->has_fallocate
= false;
1721 } else if (ret
== -EINVAL
) {
1723 * Some file systems like older versions of GPFS do not like un-
1724 * aligned byte ranges, and return EINVAL in such a case, though
1725 * they should not do it according to the man-page of fallocate().
1726 * Warn about the bad filesystem and try the final fallback instead.
1728 warn_report_once("Your file system is misbehaving: "
1729 "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. "
1730 "Please report this bug to your file system "
1732 } else if (ret
!= -ENOTSUP
) {
1735 s
->has_discard
= false;
1740 #ifdef CONFIG_FALLOCATE
1741 /* Last resort: we are trying to extend the file with zeroed data. This
1742 * can be done via fallocate(fd, 0) */
1743 len
= bdrv_getlength(aiocb
->bs
);
1744 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1745 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1746 if (ret
== 0 || ret
!= -ENOTSUP
) {
1749 s
->has_fallocate
= false;
1756 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1758 RawPosixAIOData
*aiocb
= opaque
;
1759 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1761 /* First try to write zeros and unmap at the same time */
1763 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1764 int ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1765 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1776 /* If we couldn't manage to unmap while guaranteed that the area reads as
1777 * all-zero afterwards, just write zeroes without unmapping */
1778 return handle_aiocb_write_zeroes(aiocb
);
1781 #ifndef HAVE_COPY_FILE_RANGE
1782 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1783 off_t
*out_off
, size_t len
, unsigned int flags
)
1785 #ifdef __NR_copy_file_range
1786 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1787 out_off
, len
, flags
);
1795 static int handle_aiocb_copy_range(void *opaque
)
1797 RawPosixAIOData
*aiocb
= opaque
;
1798 uint64_t bytes
= aiocb
->aio_nbytes
;
1799 off_t in_off
= aiocb
->aio_offset
;
1800 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1803 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1804 aiocb
->copy_range
.aio_fd2
, &out_off
,
1806 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1807 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1810 /* No progress (e.g. when beyond EOF), let the caller fall back to
1829 static int handle_aiocb_discard(void *opaque
)
1831 RawPosixAIOData
*aiocb
= opaque
;
1833 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1835 if (!s
->has_discard
) {
1839 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1842 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1843 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1846 } while (errno
== EINTR
);
1848 ret
= translate_err(-errno
);
1851 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1852 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1853 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1854 ret
= translate_err(ret
);
1855 #elif defined(__APPLE__) && (__MACH__)
1856 fpunchhole_t fpunchhole
;
1857 fpunchhole
.fp_flags
= 0;
1858 fpunchhole
.reserved
= 0;
1859 fpunchhole
.fp_offset
= aiocb
->aio_offset
;
1860 fpunchhole
.fp_length
= aiocb
->aio_nbytes
;
1861 if (fcntl(s
->fd
, F_PUNCHHOLE
, &fpunchhole
) == -1) {
1862 ret
= errno
== ENODEV
? -ENOTSUP
: -errno
;
1869 if (ret
== -ENOTSUP
) {
1870 s
->has_discard
= false;
1876 * Help alignment probing by allocating the first block.
1878 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1879 * reading succeeds regardless of request length. In this case we fallback to
1880 * safe alignment which is not optimal. Allocating the first block avoids this
1883 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1884 * request alignment, so we use safe values.
1886 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1887 * caller may ignore failures.
1889 static int allocate_first_block(int fd
, size_t max_size
)
1891 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1894 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size());
1899 buf
= qemu_memalign(max_align
, write_size
);
1900 memset(buf
, 0, write_size
);
1903 n
= pwrite(fd
, buf
, write_size
, 0);
1904 } while (n
== -1 && errno
== EINTR
);
1906 ret
= (n
== -1) ? -errno
: 0;
1912 static int handle_aiocb_truncate(void *opaque
)
1914 RawPosixAIOData
*aiocb
= opaque
;
1916 int64_t current_length
= 0;
1919 int fd
= aiocb
->aio_fildes
;
1920 int64_t offset
= aiocb
->aio_offset
;
1921 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1922 Error
**errp
= aiocb
->truncate
.errp
;
1924 if (fstat(fd
, &st
) < 0) {
1926 error_setg_errno(errp
, -result
, "Could not stat file");
1930 current_length
= st
.st_size
;
1931 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1932 error_setg(errp
, "Cannot use preallocation for shrinking files");
1937 #ifdef CONFIG_POSIX_FALLOCATE
1938 case PREALLOC_MODE_FALLOC
:
1940 * Truncating before posix_fallocate() makes it about twice slower on
1941 * file systems that do not support fallocate(), trying to check if a
1942 * block is allocated before allocating it, so don't do that here.
1944 if (offset
!= current_length
) {
1945 result
= -posix_fallocate(fd
, current_length
,
1946 offset
- current_length
);
1948 /* posix_fallocate() doesn't set errno. */
1949 error_setg_errno(errp
, -result
,
1950 "Could not preallocate new data");
1951 } else if (current_length
== 0) {
1953 * posix_fallocate() uses fallocate() if the filesystem
1954 * supports it, or fallback to manually writing zeroes. If
1955 * fallocate() was used, unaligned reads from the fallocated
1956 * area in raw_probe_alignment() will succeed, hence we need to
1957 * allocate the first block.
1959 * Optimize future alignment probing; ignore failures.
1961 allocate_first_block(fd
, offset
);
1968 case PREALLOC_MODE_FULL
:
1970 int64_t num
= 0, left
= offset
- current_length
;
1974 * Knowing the final size from the beginning could allow the file
1975 * system driver to do less allocations and possibly avoid
1976 * fragmentation of the file.
1978 if (ftruncate(fd
, offset
) != 0) {
1980 error_setg_errno(errp
, -result
, "Could not resize file");
1984 buf
= g_malloc0(65536);
1986 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1987 if (seek_result
< 0) {
1989 error_setg_errno(errp
, -result
,
1990 "Failed to seek to the old end of file");
1995 num
= MIN(left
, 65536);
1996 result
= write(fd
, buf
, num
);
1998 if (errno
== EINTR
) {
2002 error_setg_errno(errp
, -result
,
2003 "Could not write zeros for preallocation");
2012 error_setg_errno(errp
, -result
,
2013 "Could not flush file to disk");
2019 case PREALLOC_MODE_OFF
:
2020 if (ftruncate(fd
, offset
) != 0) {
2022 error_setg_errno(errp
, -result
, "Could not resize file");
2023 } else if (current_length
== 0 && offset
> current_length
) {
2024 /* Optimize future alignment probing; ignore failures. */
2025 allocate_first_block(fd
, offset
);
2030 error_setg(errp
, "Unsupported preallocation mode: %s",
2031 PreallocMode_str(prealloc
));
2037 if (ftruncate(fd
, current_length
) < 0) {
2038 error_report("Failed to restore old file length: %s",
2047 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
2048 ThreadPoolFunc func
, void *arg
)
2050 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
2051 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
2052 return thread_pool_submit_co(pool
, func
, arg
);
2056 * Check if all memory in this vector is sector aligned.
2058 static bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2061 size_t alignment
= bdrv_min_mem_align(bs
);
2062 size_t len
= bs
->bl
.request_alignment
;
2065 for (i
= 0; i
< qiov
->niov
; i
++) {
2066 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2069 if (qiov
->iov
[i
].iov_len
% len
) {
2077 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
2078 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
2080 BDRVRawState
*s
= bs
->opaque
;
2081 RawPosixAIOData acb
;
2083 if (fd_open(bs
) < 0)
2087 * When using O_DIRECT, the request must be aligned to be able to use
2088 * either libaio or io_uring interface. If not fail back to regular thread
2089 * pool read/write code which emulates this for us if we
2090 * set QEMU_AIO_MISALIGNED.
2092 if (s
->needs_alignment
&& !bdrv_qiov_is_aligned(bs
, qiov
)) {
2093 type
|= QEMU_AIO_MISALIGNED
;
2094 #ifdef CONFIG_LINUX_IO_URING
2095 } else if (s
->use_linux_io_uring
) {
2096 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2097 assert(qiov
->size
== bytes
);
2098 return luring_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
2100 #ifdef CONFIG_LINUX_AIO
2101 } else if (s
->use_linux_aio
) {
2102 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
2103 assert(qiov
->size
== bytes
);
2104 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
,
2109 acb
= (RawPosixAIOData
) {
2111 .aio_fildes
= s
->fd
,
2113 .aio_offset
= offset
,
2114 .aio_nbytes
= bytes
,
2121 assert(qiov
->size
== bytes
);
2122 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
2125 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, int64_t offset
,
2126 int64_t bytes
, QEMUIOVector
*qiov
,
2127 BdrvRequestFlags flags
)
2129 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
2132 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, int64_t offset
,
2133 int64_t bytes
, QEMUIOVector
*qiov
,
2134 BdrvRequestFlags flags
)
2137 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
2140 static void raw_aio_plug(BlockDriverState
*bs
)
2142 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2143 #ifdef CONFIG_LINUX_AIO
2144 if (s
->use_linux_aio
) {
2145 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
2146 laio_io_plug(bs
, aio
);
2149 #ifdef CONFIG_LINUX_IO_URING
2150 if (s
->use_linux_io_uring
) {
2151 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2152 luring_io_plug(bs
, aio
);
2157 static void raw_aio_unplug(BlockDriverState
*bs
)
2159 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2160 #ifdef CONFIG_LINUX_AIO
2161 if (s
->use_linux_aio
) {
2162 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
2163 laio_io_unplug(bs
, aio
, s
->aio_max_batch
);
2166 #ifdef CONFIG_LINUX_IO_URING
2167 if (s
->use_linux_io_uring
) {
2168 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2169 luring_io_unplug(bs
, aio
);
2174 static int coroutine_fn
raw_co_flush_to_disk(BlockDriverState
*bs
)
2176 BDRVRawState
*s
= bs
->opaque
;
2177 RawPosixAIOData acb
;
2185 acb
= (RawPosixAIOData
) {
2187 .aio_fildes
= s
->fd
,
2188 .aio_type
= QEMU_AIO_FLUSH
,
2191 #ifdef CONFIG_LINUX_IO_URING
2192 if (s
->use_linux_io_uring
) {
2193 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2194 return luring_co_submit(bs
, aio
, s
->fd
, 0, NULL
, QEMU_AIO_FLUSH
);
2197 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
2200 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
2201 AioContext
*new_context
)
2203 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2204 #ifdef CONFIG_LINUX_AIO
2205 if (s
->use_linux_aio
) {
2206 Error
*local_err
= NULL
;
2207 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
2208 error_reportf_err(local_err
, "Unable to use native AIO, "
2209 "falling back to thread pool: ");
2210 s
->use_linux_aio
= false;
2214 #ifdef CONFIG_LINUX_IO_URING
2215 if (s
->use_linux_io_uring
) {
2216 Error
*local_err
= NULL
;
2217 if (!aio_setup_linux_io_uring(new_context
, &local_err
)) {
2218 error_reportf_err(local_err
, "Unable to use linux io_uring, "
2219 "falling back to thread pool: ");
2220 s
->use_linux_io_uring
= false;
2226 static void raw_close(BlockDriverState
*bs
)
2228 BDRVRawState
*s
= bs
->opaque
;
2237 * Truncates the given regular file @fd to @offset and, when growing, fills the
2238 * new space according to @prealloc.
2240 * Returns: 0 on success, -errno on failure.
2242 static int coroutine_fn
2243 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2244 PreallocMode prealloc
, Error
**errp
)
2246 RawPosixAIOData acb
;
2248 acb
= (RawPosixAIOData
) {
2251 .aio_type
= QEMU_AIO_TRUNCATE
,
2252 .aio_offset
= offset
,
2254 .prealloc
= prealloc
,
2259 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2262 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2263 bool exact
, PreallocMode prealloc
,
2264 BdrvRequestFlags flags
, Error
**errp
)
2266 BDRVRawState
*s
= bs
->opaque
;
2270 if (fstat(s
->fd
, &st
)) {
2272 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2276 if (S_ISREG(st
.st_mode
)) {
2277 /* Always resizes to the exact @offset */
2278 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2281 if (prealloc
!= PREALLOC_MODE_OFF
) {
2282 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2283 "non-regular file", PreallocMode_str(prealloc
));
2287 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2288 int64_t cur_length
= raw_getlength(bs
);
2290 if (offset
!= cur_length
&& exact
) {
2291 error_setg(errp
, "Cannot resize device files");
2293 } else if (offset
> cur_length
) {
2294 error_setg(errp
, "Cannot grow device files");
2298 error_setg(errp
, "Resizing this file is not supported");
2306 static int64_t raw_getlength(BlockDriverState
*bs
)
2308 BDRVRawState
*s
= bs
->opaque
;
2314 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2315 struct disklabel dl
;
2317 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2319 return (uint64_t)dl
.d_secsize
*
2320 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2324 #elif defined(__NetBSD__)
2325 static int64_t raw_getlength(BlockDriverState
*bs
)
2327 BDRVRawState
*s
= bs
->opaque
;
2333 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2334 struct dkwedge_info dkw
;
2336 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2337 return dkw
.dkw_size
* 512;
2339 struct disklabel dl
;
2341 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2343 return (uint64_t)dl
.d_secsize
*
2344 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2349 #elif defined(__sun__)
2350 static int64_t raw_getlength(BlockDriverState
*bs
)
2352 BDRVRawState
*s
= bs
->opaque
;
2353 struct dk_minfo minfo
;
2363 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2365 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2367 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2371 * There are reports that lseek on some devices fails, but
2372 * irc discussion said that contingency on contingency was overkill.
2374 size
= lseek(s
->fd
, 0, SEEK_END
);
2380 #elif defined(CONFIG_BSD)
2381 static int64_t raw_getlength(BlockDriverState
*bs
)
2383 BDRVRawState
*s
= bs
->opaque
;
2387 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2396 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2399 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2401 #ifdef DIOCGMEDIASIZE
2402 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
)) {
2409 if (ioctl(fd
, DIOCGPART
, &pi
) == 0) {
2410 size
= pi
.media_size
;
2414 #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE)
2416 uint64_t sectors
= 0;
2417 uint32_t sector_size
= 0;
2419 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2420 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2421 size
= sectors
* sector_size
;
2426 size
= lseek(fd
, 0LL, SEEK_END
);
2431 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2434 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2435 if (size
== 2048LL * (unsigned)-1)
2437 /* XXX no disc? maybe we need to reopen... */
2438 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2445 size
= lseek(fd
, 0, SEEK_END
);
2453 static int64_t raw_getlength(BlockDriverState
*bs
)
2455 BDRVRawState
*s
= bs
->opaque
;
2464 size
= lseek(s
->fd
, 0, SEEK_END
);
2472 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2475 BDRVRawState
*s
= bs
->opaque
;
2477 if (fstat(s
->fd
, &st
) < 0) {
2480 return (int64_t)st
.st_blocks
* 512;
2483 static int coroutine_fn
2484 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2486 BlockdevCreateOptionsFile
*file_opts
;
2487 Error
*local_err
= NULL
;
2489 uint64_t perm
, shared
;
2492 /* Validate options and set default values */
2493 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2494 file_opts
= &options
->u
.file
;
2496 if (!file_opts
->has_nocow
) {
2497 file_opts
->nocow
= false;
2499 if (!file_opts
->has_preallocation
) {
2500 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2502 if (!file_opts
->has_extent_size_hint
) {
2503 file_opts
->extent_size_hint
= 1 * MiB
;
2505 if (file_opts
->extent_size_hint
> UINT32_MAX
) {
2507 error_setg(errp
, "Extent size hint is too large");
2512 fd
= qemu_create(file_opts
->filename
, O_RDWR
| O_BINARY
, 0644, errp
);
2518 /* Take permissions: We want to discard everything, so we need
2519 * BLK_PERM_WRITE; and truncation to the desired size requires
2521 * On the other hand, we cannot share the RESIZE permission
2522 * because we promise that after this function, the file has the
2523 * size given in the options. If someone else were to resize it
2524 * concurrently, we could not guarantee that.
2525 * Note that after this function, we can no longer guarantee that
2526 * the file is not touched by a third party, so it may be resized
2528 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2529 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2531 /* Step one: Take locks */
2532 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2537 /* Step two: Check that nobody else has taken conflicting locks */
2538 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2540 error_append_hint(errp
,
2541 "Is another process using the image [%s]?\n",
2542 file_opts
->filename
);
2546 /* Clear the file by truncating it to 0 */
2547 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2552 if (file_opts
->nocow
) {
2554 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2555 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2556 * will be ignored since any failure of this operation should not
2557 * block the left work.
2560 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2561 attr
|= FS_NOCOW_FL
;
2562 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2566 #ifdef FS_IOC_FSSETXATTR
2568 * Try to set the extent size hint. Failure is not fatal, and a warning is
2569 * only printed if the option was explicitly specified.
2572 struct fsxattr attr
;
2573 result
= ioctl(fd
, FS_IOC_FSGETXATTR
, &attr
);
2575 attr
.fsx_xflags
|= FS_XFLAG_EXTSIZE
;
2576 attr
.fsx_extsize
= file_opts
->extent_size_hint
;
2577 result
= ioctl(fd
, FS_IOC_FSSETXATTR
, &attr
);
2579 if (result
< 0 && file_opts
->has_extent_size_hint
&&
2580 file_opts
->extent_size_hint
)
2582 warn_report("Failed to set extent size hint: %s",
2588 /* Resize and potentially preallocate the file to the desired
2590 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2591 file_opts
->preallocation
, errp
);
2597 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2599 /* The above call should not fail, and if it does, that does
2600 * not mean the whole creation operation has failed. So
2601 * report it the user for their convenience, but do not report
2602 * it to the caller. */
2603 warn_report_err(local_err
);
2607 if (qemu_close(fd
) != 0 && result
== 0) {
2609 error_setg_errno(errp
, -result
, "Could not close the new file");
2615 static int coroutine_fn
raw_co_create_opts(BlockDriver
*drv
,
2616 const char *filename
,
2620 BlockdevCreateOptions options
;
2621 int64_t total_size
= 0;
2622 int64_t extent_size_hint
= 0;
2623 bool has_extent_size_hint
= false;
2625 PreallocMode prealloc
;
2627 Error
*local_err
= NULL
;
2629 /* Skip file: protocol prefix */
2630 strstart(filename
, "file:", &filename
);
2632 /* Read out options */
2633 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2635 if (qemu_opt_get(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
)) {
2636 has_extent_size_hint
= true;
2638 qemu_opt_get_size_del(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
, -1);
2640 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2641 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2642 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2643 PREALLOC_MODE_OFF
, &local_err
);
2646 error_propagate(errp
, local_err
);
2650 options
= (BlockdevCreateOptions
) {
2651 .driver
= BLOCKDEV_DRIVER_FILE
,
2653 .filename
= (char *) filename
,
2655 .has_preallocation
= true,
2656 .preallocation
= prealloc
,
2659 .has_extent_size_hint
= has_extent_size_hint
,
2660 .extent_size_hint
= extent_size_hint
,
2663 return raw_co_create(&options
, errp
);
2666 static int coroutine_fn
raw_co_delete_file(BlockDriverState
*bs
,
2672 if (!(stat(bs
->filename
, &st
) == 0) || !S_ISREG(st
.st_mode
)) {
2673 error_setg_errno(errp
, ENOENT
, "%s is not a regular file",
2678 ret
= unlink(bs
->filename
);
2681 error_setg_errno(errp
, -ret
, "Error when deleting file %s",
2689 * Find allocation range in @bs around offset @start.
2690 * May change underlying file descriptor's file offset.
2691 * If @start is not in a hole, store @start in @data, and the
2692 * beginning of the next hole in @hole, and return 0.
2693 * If @start is in a non-trailing hole, store @start in @hole and the
2694 * beginning of the next non-hole in @data, and return 0.
2695 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2696 * If we can't find out, return a negative errno other than -ENXIO.
2698 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2699 off_t
*data
, off_t
*hole
)
2701 #if defined SEEK_HOLE && defined SEEK_DATA
2702 BDRVRawState
*s
= bs
->opaque
;
2707 * D1. offs == start: start is in data
2708 * D2. offs > start: start is in a hole, next data at offs
2709 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2710 * or start is beyond EOF
2711 * If the latter happens, the file has been truncated behind
2712 * our back since we opened it. All bets are off then.
2713 * Treating like a trailing hole is simplest.
2714 * D4. offs < 0, errno != ENXIO: we learned nothing
2716 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2718 return -errno
; /* D3 or D4 */
2722 /* This is not a valid return by lseek(). We are safe to just return
2723 * -EIO in this case, and we'll treat it like D4. */
2728 /* D2: in hole, next data at offs */
2734 /* D1: in data, end not yet known */
2738 * H1. offs == start: start is in a hole
2739 * If this happens here, a hole has been dug behind our back
2740 * since the previous lseek().
2741 * H2. offs > start: either start is in data, next hole at offs,
2742 * or start is in trailing hole, EOF at offs
2743 * Linux treats trailing holes like any other hole: offs ==
2744 * start. Solaris seeks to EOF instead: offs > start (blech).
2745 * If that happens here, a hole has been dug behind our back
2746 * since the previous lseek().
2747 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2748 * If this happens, the file has been truncated behind our
2749 * back since we opened it. Treat it like a trailing hole.
2750 * H4. offs < 0, errno != ENXIO: we learned nothing
2751 * Pretend we know nothing at all, i.e. "forget" about D1.
2753 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2755 return -errno
; /* D1 and (H3 or H4) */
2759 /* This is not a valid return by lseek(). We are safe to just return
2760 * -EIO in this case, and we'll treat it like H4. */
2766 * D1 and H2: either in data, next hole at offs, or it was in
2767 * data but is now in a trailing hole. In the latter case,
2768 * all bets are off. Treating it as if it there was data all
2769 * the way to EOF is safe, so simply do that.
2784 * Returns the allocation status of the specified offset.
2786 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2788 * 'pnum' is set to the number of bytes (including and immediately following
2789 * the specified offset) that are known to be in the same
2790 * allocated/unallocated state.
2792 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may
2795 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2798 int64_t bytes
, int64_t *pnum
,
2800 BlockDriverState
**file
)
2802 off_t data
= 0, hole
= 0;
2805 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2816 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2819 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2820 if (ret
== -ENXIO
) {
2823 ret
= BDRV_BLOCK_ZERO
;
2824 } else if (ret
< 0) {
2825 /* No info available, so pretend there are no holes */
2827 ret
= BDRV_BLOCK_DATA
;
2828 } else if (data
== offset
) {
2829 /* On a data extent, compute bytes to the end of the extent,
2830 * possibly including a partial sector at EOF. */
2831 *pnum
= hole
- offset
;
2834 * We are not allowed to return partial sectors, though, so
2835 * round up if necessary.
2837 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2838 int64_t file_length
= raw_getlength(bs
);
2839 if (file_length
> 0) {
2840 /* Ignore errors, this is just a safeguard */
2841 assert(hole
== file_length
);
2843 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2846 ret
= BDRV_BLOCK_DATA
;
2848 /* On a hole, compute bytes to the beginning of the next extent. */
2849 assert(hole
== offset
);
2850 *pnum
= data
- offset
;
2851 ret
= BDRV_BLOCK_ZERO
;
2855 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2858 #if defined(__linux__)
2859 /* Verify that the file is not in the page cache */
2860 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2862 const size_t window_size
= 128 * 1024 * 1024;
2863 BDRVRawState
*s
= bs
->opaque
;
2864 void *window
= NULL
;
2871 /* mincore(2) page status information requires 1 byte per page */
2872 page_size
= sysconf(_SC_PAGESIZE
);
2873 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2875 end
= raw_getlength(bs
);
2877 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2884 /* Unmap previous window if size has changed */
2885 new_length
= MIN(end
- offset
, window_size
);
2886 if (new_length
!= length
) {
2887 munmap(window
, length
);
2892 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2894 if (new_window
== MAP_FAILED
) {
2895 error_setg_errno(errp
, errno
, "mmap failed");
2899 window
= new_window
;
2900 length
= new_length
;
2902 ret
= mincore(window
, length
, vec
);
2904 error_setg_errno(errp
, errno
, "mincore failed");
2908 vec_end
= DIV_ROUND_UP(length
, page_size
);
2909 for (i
= 0; i
< vec_end
; i
++) {
2915 error_setg(errp
, "page cache still in use!");
2921 munmap(window
, length
);
2926 #endif /* __linux__ */
2928 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2931 BDRVRawState
*s
= bs
->opaque
;
2936 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2940 if (!s
->drop_cache
) {
2944 if (s
->open_flags
& O_DIRECT
) {
2945 return; /* No host kernel page cache */
2948 #if defined(__linux__)
2949 /* This sets the scene for the next syscall... */
2950 ret
= bdrv_co_flush(bs
);
2952 error_setg_errno(errp
, -ret
, "flush failed");
2956 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2957 * process. These limitations are okay because we just fsynced the file,
2958 * we don't use mmap, and the file should not be in use by other processes.
2960 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2961 if (ret
!= 0) { /* the return value is a positive errno */
2962 error_setg_errno(errp
, ret
, "fadvise failed");
2966 if (s
->check_cache_dropped
) {
2967 check_cache_dropped(bs
, errp
);
2969 #else /* __linux__ */
2970 /* Do nothing. Live migration to a remote host with cache.direct=off is
2971 * unsupported on other host operating systems. Cache consistency issues
2972 * may occur but no error is reported here, partly because that's the
2973 * historical behavior and partly because it's hard to differentiate valid
2974 * configurations that should not cause errors.
2976 #endif /* !__linux__ */
2979 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
2982 s
->stats
.discard_nb_failed
++;
2984 s
->stats
.discard_nb_ok
++;
2985 s
->stats
.discard_bytes_ok
+= nbytes
;
2989 static coroutine_fn
int
2990 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2993 BDRVRawState
*s
= bs
->opaque
;
2994 RawPosixAIOData acb
;
2997 acb
= (RawPosixAIOData
) {
2999 .aio_fildes
= s
->fd
,
3000 .aio_type
= QEMU_AIO_DISCARD
,
3001 .aio_offset
= offset
,
3002 .aio_nbytes
= bytes
,
3006 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
3009 ret
= raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
3010 raw_account_discard(s
, bytes
, ret
);
3014 static coroutine_fn
int
3015 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3017 return raw_do_pdiscard(bs
, offset
, bytes
, false);
3020 static int coroutine_fn
3021 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3022 BdrvRequestFlags flags
, bool blkdev
)
3024 BDRVRawState
*s
= bs
->opaque
;
3025 RawPosixAIOData acb
;
3026 ThreadPoolFunc
*handler
;
3028 #ifdef CONFIG_FALLOCATE
3029 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3030 BdrvTrackedRequest
*req
;
3033 * This is a workaround for a bug in the Linux XFS driver,
3034 * where writes submitted through the AIO interface will be
3035 * discarded if they happen beyond a concurrently running
3036 * fallocate() that increases the file length (i.e., both the
3037 * write and the fallocate() happen beyond the EOF).
3039 * To work around it, we extend the tracked request for this
3040 * zero write until INT64_MAX (effectively infinity), and mark
3041 * it as serializing.
3043 * We have to enable this workaround for all filesystems and
3044 * AIO modes (not just XFS with aio=native), because for
3045 * remote filesystems we do not know the host configuration.
3048 req
= bdrv_co_get_self_request(bs
);
3050 assert(req
->type
== BDRV_TRACKED_WRITE
);
3051 assert(req
->offset
<= offset
);
3052 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
3054 req
->bytes
= BDRV_MAX_LENGTH
- req
->offset
;
3056 bdrv_check_request(req
->offset
, req
->bytes
, &error_abort
);
3058 bdrv_make_request_serialising(req
, bs
->bl
.request_alignment
);
3062 acb
= (RawPosixAIOData
) {
3064 .aio_fildes
= s
->fd
,
3065 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
3066 .aio_offset
= offset
,
3067 .aio_nbytes
= bytes
,
3071 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
3073 if (flags
& BDRV_REQ_NO_FALLBACK
) {
3074 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
3077 if (flags
& BDRV_REQ_MAY_UNMAP
) {
3078 acb
.aio_type
|= QEMU_AIO_DISCARD
;
3079 handler
= handle_aiocb_write_zeroes_unmap
;
3081 handler
= handle_aiocb_write_zeroes
;
3084 return raw_thread_pool_submit(bs
, handler
, &acb
);
3087 static int coroutine_fn
raw_co_pwrite_zeroes(
3088 BlockDriverState
*bs
, int64_t offset
,
3089 int64_t bytes
, BdrvRequestFlags flags
)
3091 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
3094 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3099 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
3101 BDRVRawState
*s
= bs
->opaque
;
3102 return (BlockStatsSpecificFile
) {
3103 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
3104 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
3105 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
3109 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
3111 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3113 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
3114 stats
->u
.file
= get_blockstats_specific_file(bs
);
3119 #if defined(HAVE_HOST_BLOCK_DEVICE)
3120 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
3122 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3124 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
3125 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
3129 #endif /* HAVE_HOST_BLOCK_DEVICE */
3131 static QemuOptsList raw_create_opts
= {
3132 .name
= "raw-create-opts",
3133 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
3136 .name
= BLOCK_OPT_SIZE
,
3137 .type
= QEMU_OPT_SIZE
,
3138 .help
= "Virtual disk size"
3141 .name
= BLOCK_OPT_NOCOW
,
3142 .type
= QEMU_OPT_BOOL
,
3143 .help
= "Turn off copy-on-write (valid only on btrfs)"
3146 .name
= BLOCK_OPT_PREALLOC
,
3147 .type
= QEMU_OPT_STRING
,
3148 .help
= "Preallocation mode (allowed values: off"
3149 #ifdef CONFIG_POSIX_FALLOCATE
3155 .name
= BLOCK_OPT_EXTENT_SIZE_HINT
,
3156 .type
= QEMU_OPT_SIZE
,
3157 .help
= "Extent size hint for the image file, 0 to disable"
3159 { /* end of list */ }
3163 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
3166 BDRVRawState
*s
= bs
->opaque
;
3167 int input_flags
= s
->reopen_state
? s
->reopen_state
->flags
: bs
->open_flags
;
3171 /* We may need a new fd if auto-read-only switches the mode */
3172 ret
= raw_reconfigure_getfd(bs
, input_flags
, &open_flags
, perm
,
3176 } else if (ret
!= s
->fd
) {
3177 Error
*local_err
= NULL
;
3180 * Fail already check_perm() if we can't get a working O_DIRECT
3181 * alignment with the new fd.
3183 raw_probe_alignment(bs
, ret
, &local_err
);
3185 error_propagate(errp
, local_err
);
3189 s
->perm_change_fd
= ret
;
3190 s
->perm_change_flags
= open_flags
;
3193 /* Prepare permissions on old fd to avoid conflicts between old and new,
3194 * but keep everything locked that new will need. */
3195 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
3200 /* Copy locks to the new fd */
3201 if (s
->perm_change_fd
&& s
->use_lock
) {
3202 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
3205 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3212 if (s
->perm_change_fd
) {
3213 qemu_close(s
->perm_change_fd
);
3215 s
->perm_change_fd
= 0;
3219 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
3221 BDRVRawState
*s
= bs
->opaque
;
3223 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3224 * called after .bdrv_reopen_commit) */
3225 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
3227 s
->fd
= s
->perm_change_fd
;
3228 s
->open_flags
= s
->perm_change_flags
;
3230 s
->perm_change_fd
= 0;
3232 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
3234 s
->shared_perm
= shared
;
3237 static void raw_abort_perm_update(BlockDriverState
*bs
)
3239 BDRVRawState
*s
= bs
->opaque
;
3241 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3242 * the file descriptor. */
3243 if (s
->perm_change_fd
) {
3244 qemu_close(s
->perm_change_fd
);
3246 s
->perm_change_fd
= 0;
3248 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3251 static int coroutine_fn
raw_co_copy_range_from(
3252 BlockDriverState
*bs
, BdrvChild
*src
, int64_t src_offset
,
3253 BdrvChild
*dst
, int64_t dst_offset
, int64_t bytes
,
3254 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
3256 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3257 read_flags
, write_flags
);
3260 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
3266 BdrvRequestFlags read_flags
,
3267 BdrvRequestFlags write_flags
)
3269 RawPosixAIOData acb
;
3270 BDRVRawState
*s
= bs
->opaque
;
3271 BDRVRawState
*src_s
;
3273 assert(dst
->bs
== bs
);
3274 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
3278 src_s
= src
->bs
->opaque
;
3279 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
3283 acb
= (RawPosixAIOData
) {
3285 .aio_type
= QEMU_AIO_COPY_RANGE
,
3286 .aio_fildes
= src_s
->fd
,
3287 .aio_offset
= src_offset
,
3288 .aio_nbytes
= bytes
,
3291 .aio_offset2
= dst_offset
,
3295 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
3298 BlockDriver bdrv_file
= {
3299 .format_name
= "file",
3300 .protocol_name
= "file",
3301 .instance_size
= sizeof(BDRVRawState
),
3302 .bdrv_needs_filename
= true,
3303 .bdrv_probe
= NULL
, /* no probe for protocols */
3304 .bdrv_parse_filename
= raw_parse_filename
,
3305 .bdrv_file_open
= raw_open
,
3306 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3307 .bdrv_reopen_commit
= raw_reopen_commit
,
3308 .bdrv_reopen_abort
= raw_reopen_abort
,
3309 .bdrv_close
= raw_close
,
3310 .bdrv_co_create
= raw_co_create
,
3311 .bdrv_co_create_opts
= raw_co_create_opts
,
3312 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3313 .bdrv_co_block_status
= raw_co_block_status
,
3314 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3315 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3316 .bdrv_co_delete_file
= raw_co_delete_file
,
3318 .bdrv_co_preadv
= raw_co_preadv
,
3319 .bdrv_co_pwritev
= raw_co_pwritev
,
3320 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3321 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3322 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3323 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3324 .bdrv_refresh_limits
= raw_refresh_limits
,
3325 .bdrv_io_plug
= raw_aio_plug
,
3326 .bdrv_io_unplug
= raw_aio_unplug
,
3327 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3329 .bdrv_co_truncate
= raw_co_truncate
,
3330 .bdrv_getlength
= raw_getlength
,
3331 .bdrv_get_info
= raw_get_info
,
3332 .bdrv_get_allocated_file_size
3333 = raw_get_allocated_file_size
,
3334 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3335 .bdrv_check_perm
= raw_check_perm
,
3336 .bdrv_set_perm
= raw_set_perm
,
3337 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3338 .create_opts
= &raw_create_opts
,
3339 .mutable_opts
= mutable_opts
,
3342 /***********************************************/
3345 #if defined(HAVE_HOST_BLOCK_DEVICE)
3347 #if defined(__APPLE__) && defined(__MACH__)
3348 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3349 CFIndex maxPathSize
, int flags
);
3351 #if !defined(MAC_OS_VERSION_12_0) \
3352 || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0)
3353 #define IOMainPort IOMasterPort
3356 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3358 kern_return_t kernResult
= KERN_FAILURE
;
3359 mach_port_t mainPort
;
3360 CFMutableDictionaryRef classesToMatch
;
3361 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3362 char *mediaType
= NULL
;
3364 kernResult
= IOMainPort(MACH_PORT_NULL
, &mainPort
);
3365 if ( KERN_SUCCESS
!= kernResult
) {
3366 printf("IOMainPort returned %d\n", kernResult
);
3370 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3371 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3372 if (classesToMatch
== NULL
) {
3373 error_report("IOServiceMatching returned NULL for %s",
3374 matching_array
[index
]);
3377 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3379 kernResult
= IOServiceGetMatchingServices(mainPort
, classesToMatch
,
3381 if (kernResult
!= KERN_SUCCESS
) {
3382 error_report("Note: IOServiceGetMatchingServices returned %d",
3387 /* If a match was found, leave the loop */
3388 if (*mediaIterator
!= 0) {
3389 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3390 mediaType
= g_strdup(matching_array
[index
]);
3397 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3398 CFIndex maxPathSize
, int flags
)
3400 io_object_t nextMedia
;
3401 kern_return_t kernResult
= KERN_FAILURE
;
3403 nextMedia
= IOIteratorNext( mediaIterator
);
3406 CFTypeRef bsdPathAsCFString
;
3407 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3408 if ( bsdPathAsCFString
) {
3409 size_t devPathLength
;
3410 strcpy( bsdPath
, _PATH_DEV
);
3411 if (flags
& BDRV_O_NOCACHE
) {
3412 strcat(bsdPath
, "r");
3414 devPathLength
= strlen( bsdPath
);
3415 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3416 kernResult
= KERN_SUCCESS
;
3418 CFRelease( bsdPathAsCFString
);
3420 IOObjectRelease( nextMedia
);
3426 /* Sets up a real cdrom for use in QEMU */
3427 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3429 int index
, num_of_test_partitions
= 2, fd
;
3430 char test_partition
[MAXPATHLEN
];
3431 bool partition_found
= false;
3433 /* look for a working partition */
3434 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3435 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3437 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
, NULL
);
3439 partition_found
= true;
3445 /* if a working partition on the device was not found */
3446 if (partition_found
== false) {
3447 error_setg(errp
, "Failed to find a working partition on disc");
3449 trace_file_setup_cdrom(test_partition
);
3450 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3452 return partition_found
;
3455 /* Prints directions on mounting and unmounting a device */
3456 static void print_unmounting_directions(const char *file_name
)
3458 error_report("If device %s is mounted on the desktop, unmount"
3459 " it first before using it in QEMU", file_name
);
3460 error_report("Command to unmount device: diskutil unmountDisk %s",
3462 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3465 #endif /* defined(__APPLE__) && defined(__MACH__) */
3467 static int hdev_probe_device(const char *filename
)
3471 /* allow a dedicated CD-ROM driver to match with a higher priority */
3472 if (strstart(filename
, "/dev/cdrom", NULL
))
3475 if (stat(filename
, &st
) >= 0 &&
3476 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3483 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3486 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3489 static bool hdev_is_sg(BlockDriverState
*bs
)
3492 #if defined(__linux__)
3494 BDRVRawState
*s
= bs
->opaque
;
3496 struct sg_scsi_id scsiid
;
3500 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3504 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3509 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3511 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3520 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3523 BDRVRawState
*s
= bs
->opaque
;
3526 #if defined(__APPLE__) && defined(__MACH__)
3528 * Caution: while qdict_get_str() is fine, getting non-string types
3529 * would require more care. When @options come from -blockdev or
3530 * blockdev_add, its members are typed according to the QAPI
3531 * schema, but when they come from -drive, they're all QString.
3533 const char *filename
= qdict_get_str(options
, "filename");
3534 char bsd_path
[MAXPATHLEN
] = "";
3535 bool error_occurred
= false;
3537 /* If using a real cdrom */
3538 if (strcmp(filename
, "/dev/cdrom") == 0) {
3539 char *mediaType
= NULL
;
3540 kern_return_t ret_val
;
3541 io_iterator_t mediaIterator
= 0;
3543 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3544 if (mediaType
== NULL
) {
3545 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3547 error_occurred
= true;
3548 goto hdev_open_Mac_error
;
3551 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3552 if (ret_val
!= KERN_SUCCESS
) {
3553 error_setg(errp
, "Could not get BSD path for optical drive");
3554 error_occurred
= true;
3555 goto hdev_open_Mac_error
;
3558 /* If a real optical drive was not found */
3559 if (bsd_path
[0] == '\0') {
3560 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3561 error_occurred
= true;
3562 goto hdev_open_Mac_error
;
3565 /* If using a cdrom disc and finding a partition on the disc failed */
3566 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3567 setup_cdrom(bsd_path
, errp
) == false) {
3568 print_unmounting_directions(bsd_path
);
3569 error_occurred
= true;
3570 goto hdev_open_Mac_error
;
3573 qdict_put_str(options
, "filename", bsd_path
);
3575 hdev_open_Mac_error
:
3577 if (mediaIterator
) {
3578 IOObjectRelease(mediaIterator
);
3580 if (error_occurred
) {
3584 #endif /* defined(__APPLE__) && defined(__MACH__) */
3586 s
->type
= FTYPE_FILE
;
3588 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
3590 #if defined(__APPLE__) && defined(__MACH__)
3592 filename
= bsd_path
;
3594 /* if a physical device experienced an error while being opened */
3595 if (strncmp(filename
, "/dev/", 5) == 0) {
3596 print_unmounting_directions(filename
);
3598 #endif /* defined(__APPLE__) && defined(__MACH__) */
3602 /* Since this does ioctl the device must be already opened */
3603 bs
->sg
= hdev_is_sg(bs
);
3608 #if defined(__linux__)
3609 static int coroutine_fn
3610 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3612 BDRVRawState
*s
= bs
->opaque
;
3613 RawPosixAIOData acb
;
3621 if (req
== SG_IO
&& s
->pr_mgr
) {
3622 struct sg_io_hdr
*io_hdr
= buf
;
3623 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3624 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3625 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3630 acb
= (RawPosixAIOData
) {
3632 .aio_type
= QEMU_AIO_IOCTL
,
3633 .aio_fildes
= s
->fd
,
3641 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3645 static coroutine_fn
int
3646 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3648 BDRVRawState
*s
= bs
->opaque
;
3653 raw_account_discard(s
, bytes
, ret
);
3656 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3659 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3660 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
)
3669 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3672 static BlockDriver bdrv_host_device
= {
3673 .format_name
= "host_device",
3674 .protocol_name
= "host_device",
3675 .instance_size
= sizeof(BDRVRawState
),
3676 .bdrv_needs_filename
= true,
3677 .bdrv_probe_device
= hdev_probe_device
,
3678 .bdrv_parse_filename
= hdev_parse_filename
,
3679 .bdrv_file_open
= hdev_open
,
3680 .bdrv_close
= raw_close
,
3681 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3682 .bdrv_reopen_commit
= raw_reopen_commit
,
3683 .bdrv_reopen_abort
= raw_reopen_abort
,
3684 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3685 .create_opts
= &bdrv_create_opts_simple
,
3686 .mutable_opts
= mutable_opts
,
3687 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3688 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3690 .bdrv_co_preadv
= raw_co_preadv
,
3691 .bdrv_co_pwritev
= raw_co_pwritev
,
3692 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3693 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3694 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3695 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3696 .bdrv_refresh_limits
= raw_refresh_limits
,
3697 .bdrv_io_plug
= raw_aio_plug
,
3698 .bdrv_io_unplug
= raw_aio_unplug
,
3699 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3701 .bdrv_co_truncate
= raw_co_truncate
,
3702 .bdrv_getlength
= raw_getlength
,
3703 .bdrv_get_info
= raw_get_info
,
3704 .bdrv_get_allocated_file_size
3705 = raw_get_allocated_file_size
,
3706 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
3707 .bdrv_check_perm
= raw_check_perm
,
3708 .bdrv_set_perm
= raw_set_perm
,
3709 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3710 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3711 .bdrv_probe_geometry
= hdev_probe_geometry
,
3713 /* generic scsi device */
3715 .bdrv_co_ioctl
= hdev_co_ioctl
,
3719 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3720 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3723 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3728 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3731 BDRVRawState
*s
= bs
->opaque
;
3735 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3736 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3739 static int cdrom_probe_device(const char *filename
)
3745 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
, NULL
);
3749 ret
= fstat(fd
, &st
);
3750 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3754 /* Attempt to detect via a CDROM specific ioctl */
3755 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3765 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3767 BDRVRawState
*s
= bs
->opaque
;
3770 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3771 return ret
== CDS_DISC_OK
;
3774 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3776 BDRVRawState
*s
= bs
->opaque
;
3779 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3780 perror("CDROMEJECT");
3782 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3783 perror("CDROMEJECT");
3787 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3789 BDRVRawState
*s
= bs
->opaque
;
3791 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3793 * Note: an error can happen if the distribution automatically
3796 /* perror("CDROM_LOCKDOOR"); */
3800 static BlockDriver bdrv_host_cdrom
= {
3801 .format_name
= "host_cdrom",
3802 .protocol_name
= "host_cdrom",
3803 .instance_size
= sizeof(BDRVRawState
),
3804 .bdrv_needs_filename
= true,
3805 .bdrv_probe_device
= cdrom_probe_device
,
3806 .bdrv_parse_filename
= cdrom_parse_filename
,
3807 .bdrv_file_open
= cdrom_open
,
3808 .bdrv_close
= raw_close
,
3809 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3810 .bdrv_reopen_commit
= raw_reopen_commit
,
3811 .bdrv_reopen_abort
= raw_reopen_abort
,
3812 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3813 .create_opts
= &bdrv_create_opts_simple
,
3814 .mutable_opts
= mutable_opts
,
3815 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3817 .bdrv_co_preadv
= raw_co_preadv
,
3818 .bdrv_co_pwritev
= raw_co_pwritev
,
3819 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3820 .bdrv_refresh_limits
= raw_refresh_limits
,
3821 .bdrv_io_plug
= raw_aio_plug
,
3822 .bdrv_io_unplug
= raw_aio_unplug
,
3823 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3825 .bdrv_co_truncate
= raw_co_truncate
,
3826 .bdrv_getlength
= raw_getlength
,
3827 .has_variable_length
= true,
3828 .bdrv_get_allocated_file_size
3829 = raw_get_allocated_file_size
,
3831 /* removable device support */
3832 .bdrv_is_inserted
= cdrom_is_inserted
,
3833 .bdrv_eject
= cdrom_eject
,
3834 .bdrv_lock_medium
= cdrom_lock_medium
,
3836 /* generic scsi device */
3837 .bdrv_co_ioctl
= hdev_co_ioctl
,
3839 #endif /* __linux__ */
3841 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3842 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3845 BDRVRawState
*s
= bs
->opaque
;
3850 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
3855 /* make sure the door isn't locked at this time */
3856 ioctl(s
->fd
, CDIOCALLOW
);
3860 static int cdrom_probe_device(const char *filename
)
3862 if (strstart(filename
, "/dev/cd", NULL
) ||
3863 strstart(filename
, "/dev/acd", NULL
))
3868 static int cdrom_reopen(BlockDriverState
*bs
)
3870 BDRVRawState
*s
= bs
->opaque
;
3874 * Force reread of possibly changed/newly loaded disc,
3875 * FreeBSD seems to not notice sometimes...
3879 fd
= qemu_open(bs
->filename
, s
->open_flags
, NULL
);
3886 /* make sure the door isn't locked at this time */
3887 ioctl(s
->fd
, CDIOCALLOW
);
3891 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3893 return raw_getlength(bs
) > 0;
3896 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3898 BDRVRawState
*s
= bs
->opaque
;
3903 (void) ioctl(s
->fd
, CDIOCALLOW
);
3906 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3907 perror("CDIOCEJECT");
3909 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3910 perror("CDIOCCLOSE");
3916 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3918 BDRVRawState
*s
= bs
->opaque
;
3922 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3924 * Note: an error can happen if the distribution automatically
3927 /* perror("CDROM_LOCKDOOR"); */
3931 static BlockDriver bdrv_host_cdrom
= {
3932 .format_name
= "host_cdrom",
3933 .protocol_name
= "host_cdrom",
3934 .instance_size
= sizeof(BDRVRawState
),
3935 .bdrv_needs_filename
= true,
3936 .bdrv_probe_device
= cdrom_probe_device
,
3937 .bdrv_parse_filename
= cdrom_parse_filename
,
3938 .bdrv_file_open
= cdrom_open
,
3939 .bdrv_close
= raw_close
,
3940 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3941 .bdrv_reopen_commit
= raw_reopen_commit
,
3942 .bdrv_reopen_abort
= raw_reopen_abort
,
3943 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3944 .create_opts
= &bdrv_create_opts_simple
,
3945 .mutable_opts
= mutable_opts
,
3947 .bdrv_co_preadv
= raw_co_preadv
,
3948 .bdrv_co_pwritev
= raw_co_pwritev
,
3949 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3950 .bdrv_refresh_limits
= raw_refresh_limits
,
3951 .bdrv_io_plug
= raw_aio_plug
,
3952 .bdrv_io_unplug
= raw_aio_unplug
,
3953 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3955 .bdrv_co_truncate
= raw_co_truncate
,
3956 .bdrv_getlength
= raw_getlength
,
3957 .has_variable_length
= true,
3958 .bdrv_get_allocated_file_size
3959 = raw_get_allocated_file_size
,
3961 /* removable device support */
3962 .bdrv_is_inserted
= cdrom_is_inserted
,
3963 .bdrv_eject
= cdrom_eject
,
3964 .bdrv_lock_medium
= cdrom_lock_medium
,
3966 #endif /* __FreeBSD__ */
3968 #endif /* HAVE_HOST_BLOCK_DEVICE */
3970 static void bdrv_file_init(void)
3973 * Register all the drivers. Note that order is important, the driver
3974 * registered last will get probed first.
3976 bdrv_register(&bdrv_file
);
3977 #if defined(HAVE_HOST_BLOCK_DEVICE)
3978 bdrv_register(&bdrv_host_device
);
3980 bdrv_register(&bdrv_host_cdrom
);
3982 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3983 bdrv_register(&bdrv_host_cdrom
);
3985 #endif /* HAVE_HOST_BLOCK_DEVICE */
3988 block_init(bdrv_file_init
);