2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/sched/xacct.h>
10 #include <linux/fcntl.h>
11 #include <linux/file.h>
12 #include <linux/uio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
20 #include <linux/mount.h>
24 #include <linux/uaccess.h>
25 #include <asm/unistd.h>
27 const struct file_operations generic_ro_fops
= {
28 .llseek
= generic_file_llseek
,
29 .read_iter
= generic_file_read_iter
,
30 .mmap
= generic_file_readonly_mmap
,
31 .splice_read
= generic_file_splice_read
,
34 EXPORT_SYMBOL(generic_ro_fops
);
36 static inline int unsigned_offsets(struct file
*file
)
38 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
42 * vfs_setpos - update the file offset for lseek
43 * @file: file structure in question
44 * @offset: file offset to seek to
45 * @maxsize: maximum file size
47 * This is a low-level filesystem helper for updating the file offset to
48 * the value specified by @offset if the given offset is valid and it is
49 * not equal to the current file offset.
51 * Return the specified offset on success and -EINVAL on invalid offset.
53 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
55 if (offset
< 0 && !unsigned_offsets(file
))
60 if (offset
!= file
->f_pos
) {
66 EXPORT_SYMBOL(vfs_setpos
);
69 * generic_file_llseek_size - generic llseek implementation for regular files
70 * @file: file structure to seek on
71 * @offset: file offset to seek to
72 * @whence: type of seek
73 * @size: max size of this file in file system
74 * @eof: offset used for SEEK_END position
76 * This is a variant of generic_file_llseek that allows passing in a custom
77 * maximum file size and a custom EOF position, for e.g. hashed directories
80 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
81 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
82 * read/writes behave like SEEK_SET against seeks.
85 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
86 loff_t maxsize
, loff_t eof
)
94 * Here we special-case the lseek(fd, 0, SEEK_CUR)
95 * position-querying operation. Avoid rewriting the "same"
96 * f_pos value back to the file because a concurrent read(),
97 * write() or lseek() might have altered it
102 * f_lock protects against read/modify/write race with other
103 * SEEK_CURs. Note that parallel writes and reads behave
106 spin_lock(&file
->f_lock
);
107 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
108 spin_unlock(&file
->f_lock
);
112 * In the generic case the entire file is data, so as long as
113 * offset isn't at the end of the file then the offset is data.
120 * There is a virtual hole at the end of the file, so as long as
121 * offset isn't i_size or larger, return i_size.
129 return vfs_setpos(file
, offset
, maxsize
);
131 EXPORT_SYMBOL(generic_file_llseek_size
);
134 * generic_file_llseek - generic llseek implementation for regular files
135 * @file: file structure to seek on
136 * @offset: file offset to seek to
137 * @whence: type of seek
139 * This is a generic implemenation of ->llseek useable for all normal local
140 * filesystems. It just updates the file offset to the value specified by
141 * @offset and @whence.
143 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
145 struct inode
*inode
= file
->f_mapping
->host
;
147 return generic_file_llseek_size(file
, offset
, whence
,
148 inode
->i_sb
->s_maxbytes
,
151 EXPORT_SYMBOL(generic_file_llseek
);
154 * fixed_size_llseek - llseek implementation for fixed-sized devices
155 * @file: file structure to seek on
156 * @offset: file offset to seek to
157 * @whence: type of seek
158 * @size: size of the file
161 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
164 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
165 return generic_file_llseek_size(file
, offset
, whence
,
171 EXPORT_SYMBOL(fixed_size_llseek
);
174 * no_seek_end_llseek - llseek implementation for fixed-sized devices
175 * @file: file structure to seek on
176 * @offset: file offset to seek to
177 * @whence: type of seek
180 loff_t
no_seek_end_llseek(struct file
*file
, loff_t offset
, int whence
)
183 case SEEK_SET
: case SEEK_CUR
:
184 return generic_file_llseek_size(file
, offset
, whence
,
190 EXPORT_SYMBOL(no_seek_end_llseek
);
193 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
194 * @file: file structure to seek on
195 * @offset: file offset to seek to
196 * @whence: type of seek
197 * @size: maximal offset allowed
200 loff_t
no_seek_end_llseek_size(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
203 case SEEK_SET
: case SEEK_CUR
:
204 return generic_file_llseek_size(file
, offset
, whence
,
210 EXPORT_SYMBOL(no_seek_end_llseek_size
);
213 * noop_llseek - No Operation Performed llseek implementation
214 * @file: file structure to seek on
215 * @offset: file offset to seek to
216 * @whence: type of seek
218 * This is an implementation of ->llseek useable for the rare special case when
219 * userspace expects the seek to succeed but the (device) file is actually not
220 * able to perform the seek. In this case you use noop_llseek() instead of
221 * falling back to the default implementation of ->llseek.
223 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
227 EXPORT_SYMBOL(noop_llseek
);
229 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
233 EXPORT_SYMBOL(no_llseek
);
235 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
237 struct inode
*inode
= file_inode(file
);
243 offset
+= i_size_read(inode
);
247 retval
= file
->f_pos
;
250 offset
+= file
->f_pos
;
254 * In the generic case the entire file is data, so as
255 * long as offset isn't at the end of the file then the
258 if (offset
>= inode
->i_size
) {
265 * There is a virtual hole at the end of the file, so
266 * as long as offset isn't i_size or larger, return
269 if (offset
>= inode
->i_size
) {
273 offset
= inode
->i_size
;
277 if (offset
>= 0 || unsigned_offsets(file
)) {
278 if (offset
!= file
->f_pos
) {
279 file
->f_pos
= offset
;
288 EXPORT_SYMBOL(default_llseek
);
290 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
292 loff_t (*fn
)(struct file
*, loff_t
, int);
295 if (file
->f_mode
& FMODE_LSEEK
) {
296 if (file
->f_op
->llseek
)
297 fn
= file
->f_op
->llseek
;
299 return fn(file
, offset
, whence
);
301 EXPORT_SYMBOL(vfs_llseek
);
303 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
306 struct fd f
= fdget_pos(fd
);
311 if (whence
<= SEEK_MAX
) {
312 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
314 if (res
!= (loff_t
)retval
)
315 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
322 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
324 return sys_lseek(fd
, offset
, whence
);
328 #ifdef __ARCH_WANT_SYS_LLSEEK
329 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
330 unsigned long, offset_low
, loff_t __user
*, result
,
331 unsigned int, whence
)
334 struct fd f
= fdget_pos(fd
);
341 if (whence
> SEEK_MAX
)
344 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
347 retval
= (int)offset
;
350 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
359 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
364 if (!file
->f_op
->read_iter
)
367 init_sync_kiocb(&kiocb
, file
);
368 kiocb
.ki_pos
= *ppos
;
371 ret
= call_read_iter(file
, &kiocb
, iter
);
372 BUG_ON(ret
== -EIOCBQUEUED
);
374 *ppos
= kiocb
.ki_pos
;
377 EXPORT_SYMBOL(vfs_iter_read
);
379 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
384 if (!file
->f_op
->write_iter
)
387 init_sync_kiocb(&kiocb
, file
);
388 kiocb
.ki_pos
= *ppos
;
391 ret
= call_write_iter(file
, &kiocb
, iter
);
392 BUG_ON(ret
== -EIOCBQUEUED
);
394 *ppos
= kiocb
.ki_pos
;
397 EXPORT_SYMBOL(vfs_iter_write
);
399 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
403 int retval
= -EINVAL
;
405 inode
= file_inode(file
);
406 if (unlikely((ssize_t
) count
< 0))
409 if (unlikely(pos
< 0)) {
410 if (!unsigned_offsets(file
))
412 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
414 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
415 if (!unsigned_offsets(file
))
419 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
420 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
421 read_write
== READ
? F_RDLCK
: F_WRLCK
);
425 return security_file_permission(file
,
426 read_write
== READ
? MAY_READ
: MAY_WRITE
);
429 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
431 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
433 struct iov_iter iter
;
436 init_sync_kiocb(&kiocb
, filp
);
437 kiocb
.ki_pos
= *ppos
;
438 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
440 ret
= call_read_iter(filp
, &kiocb
, &iter
);
441 BUG_ON(ret
== -EIOCBQUEUED
);
442 *ppos
= kiocb
.ki_pos
;
446 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
449 if (file
->f_op
->read
)
450 return file
->f_op
->read(file
, buf
, count
, pos
);
451 else if (file
->f_op
->read_iter
)
452 return new_sync_read(file
, buf
, count
, pos
);
456 EXPORT_SYMBOL(__vfs_read
);
458 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
462 if (!(file
->f_mode
& FMODE_READ
))
464 if (!(file
->f_mode
& FMODE_CAN_READ
))
466 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
469 ret
= rw_verify_area(READ
, file
, pos
, count
);
471 if (count
> MAX_RW_COUNT
)
472 count
= MAX_RW_COUNT
;
473 ret
= __vfs_read(file
, buf
, count
, pos
);
475 fsnotify_access(file
);
476 add_rchar(current
, ret
);
484 EXPORT_SYMBOL(vfs_read
);
486 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
488 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
490 struct iov_iter iter
;
493 init_sync_kiocb(&kiocb
, filp
);
494 kiocb
.ki_pos
= *ppos
;
495 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
497 ret
= call_write_iter(filp
, &kiocb
, &iter
);
498 BUG_ON(ret
== -EIOCBQUEUED
);
500 *ppos
= kiocb
.ki_pos
;
504 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
507 if (file
->f_op
->write
)
508 return file
->f_op
->write(file
, p
, count
, pos
);
509 else if (file
->f_op
->write_iter
)
510 return new_sync_write(file
, p
, count
, pos
);
514 EXPORT_SYMBOL(__vfs_write
);
516 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
519 const char __user
*p
;
522 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
527 p
= (__force
const char __user
*)buf
;
528 if (count
> MAX_RW_COUNT
)
529 count
= MAX_RW_COUNT
;
530 ret
= __vfs_write(file
, p
, count
, pos
);
533 fsnotify_modify(file
);
534 add_wchar(current
, ret
);
540 EXPORT_SYMBOL(__kernel_write
);
542 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
546 if (!(file
->f_mode
& FMODE_WRITE
))
548 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
550 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
553 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
555 if (count
> MAX_RW_COUNT
)
556 count
= MAX_RW_COUNT
;
557 file_start_write(file
);
558 ret
= __vfs_write(file
, buf
, count
, pos
);
560 fsnotify_modify(file
);
561 add_wchar(current
, ret
);
564 file_end_write(file
);
570 EXPORT_SYMBOL(vfs_write
);
572 static inline loff_t
file_pos_read(struct file
*file
)
577 static inline void file_pos_write(struct file
*file
, loff_t pos
)
582 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
584 struct fd f
= fdget_pos(fd
);
585 ssize_t ret
= -EBADF
;
588 loff_t pos
= file_pos_read(f
.file
);
589 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
591 file_pos_write(f
.file
, pos
);
597 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
600 struct fd f
= fdget_pos(fd
);
601 ssize_t ret
= -EBADF
;
604 loff_t pos
= file_pos_read(f
.file
);
605 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
607 file_pos_write(f
.file
, pos
);
614 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
615 size_t, count
, loff_t
, pos
)
618 ssize_t ret
= -EBADF
;
626 if (f
.file
->f_mode
& FMODE_PREAD
)
627 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
634 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
635 size_t, count
, loff_t
, pos
)
638 ssize_t ret
= -EBADF
;
646 if (f
.file
->f_mode
& FMODE_PWRITE
)
647 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
655 * Reduce an iovec's length in-place. Return the resulting number of segments
657 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
659 unsigned long seg
= 0;
662 while (seg
< nr_segs
) {
664 if (len
+ iov
->iov_len
>= to
) {
665 iov
->iov_len
= to
- len
;
673 EXPORT_SYMBOL(iov_shorten
);
675 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
676 loff_t
*ppos
, int type
, int flags
)
681 if (flags
& ~(RWF_HIPRI
| RWF_DSYNC
| RWF_SYNC
))
684 init_sync_kiocb(&kiocb
, filp
);
685 if (flags
& RWF_HIPRI
)
686 kiocb
.ki_flags
|= IOCB_HIPRI
;
687 if (flags
& RWF_DSYNC
)
688 kiocb
.ki_flags
|= IOCB_DSYNC
;
689 if (flags
& RWF_SYNC
)
690 kiocb
.ki_flags
|= (IOCB_DSYNC
| IOCB_SYNC
);
691 kiocb
.ki_pos
= *ppos
;
694 ret
= call_read_iter(filp
, &kiocb
, iter
);
696 ret
= call_write_iter(filp
, &kiocb
, iter
);
697 BUG_ON(ret
== -EIOCBQUEUED
);
698 *ppos
= kiocb
.ki_pos
;
702 /* Do it by hand, with file-ops */
703 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
704 loff_t
*ppos
, int type
, int flags
)
708 if (flags
& ~RWF_HIPRI
)
711 while (iov_iter_count(iter
)) {
712 struct iovec iovec
= iov_iter_iovec(iter
);
716 nr
= filp
->f_op
->read(filp
, iovec
.iov_base
,
717 iovec
.iov_len
, ppos
);
719 nr
= filp
->f_op
->write(filp
, iovec
.iov_base
,
720 iovec
.iov_len
, ppos
);
729 if (nr
!= iovec
.iov_len
)
731 iov_iter_advance(iter
, nr
);
737 /* A write operation does a read from user space and vice versa */
738 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
741 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
742 * into the kernel and check that it is valid.
744 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
745 * @uvector: Pointer to the userspace array.
746 * @nr_segs: Number of elements in userspace array.
747 * @fast_segs: Number of elements in @fast_pointer.
748 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
749 * @ret_pointer: (output parameter) Pointer to a variable that will point to
750 * either @fast_pointer, a newly allocated kernel array, or NULL,
751 * depending on which array was used.
753 * This function copies an array of &struct iovec of @nr_segs from
754 * userspace into the kernel and checks that each element is valid (e.g.
755 * it does not point to a kernel address or cause overflow by being too
758 * As an optimization, the caller may provide a pointer to a small
759 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
760 * (the size of this array, or 0 if unused, should be given in @fast_segs).
762 * @ret_pointer will always point to the array that was used, so the
763 * caller must take care not to call kfree() on it e.g. in case the
764 * @fast_pointer array was used and it was allocated on the stack.
766 * Return: The total number of bytes covered by the iovec array on success
767 * or a negative error code on error.
769 ssize_t
rw_copy_check_uvector(int type
, const struct iovec __user
* uvector
,
770 unsigned long nr_segs
, unsigned long fast_segs
,
771 struct iovec
*fast_pointer
,
772 struct iovec
**ret_pointer
)
776 struct iovec
*iov
= fast_pointer
;
779 * SuS says "The readv() function *may* fail if the iovcnt argument
780 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
781 * traditionally returned zero for zero segments, so...
789 * First get the "struct iovec" from user memory and
790 * verify all the pointers
792 if (nr_segs
> UIO_MAXIOV
) {
796 if (nr_segs
> fast_segs
) {
797 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
803 if (copy_from_user(iov
, uvector
, nr_segs
*sizeof(*uvector
))) {
809 * According to the Single Unix Specification we should return EINVAL
810 * if an element length is < 0 when cast to ssize_t or if the
811 * total length would overflow the ssize_t return value of the
814 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
818 for (seg
= 0; seg
< nr_segs
; seg
++) {
819 void __user
*buf
= iov
[seg
].iov_base
;
820 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
822 /* see if we we're about to use an invalid len or if
823 * it's about to overflow ssize_t */
829 && unlikely(!access_ok(vrfy_dir(type
), buf
, len
))) {
833 if (len
> MAX_RW_COUNT
- ret
) {
834 len
= MAX_RW_COUNT
- ret
;
835 iov
[seg
].iov_len
= len
;
845 ssize_t
compat_rw_copy_check_uvector(int type
,
846 const struct compat_iovec __user
*uvector
, unsigned long nr_segs
,
847 unsigned long fast_segs
, struct iovec
*fast_pointer
,
848 struct iovec
**ret_pointer
)
850 compat_ssize_t tot_len
;
851 struct iovec
*iov
= *ret_pointer
= fast_pointer
;
856 * SuS says "The readv() function *may* fail if the iovcnt argument
857 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
858 * traditionally returned zero for zero segments, so...
864 if (nr_segs
> UIO_MAXIOV
)
866 if (nr_segs
> fast_segs
) {
868 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
875 if (!access_ok(VERIFY_READ
, uvector
, nr_segs
*sizeof(*uvector
)))
879 * Single unix specification:
880 * We should -EINVAL if an element length is not >= 0 and fitting an
883 * In Linux, the total length is limited to MAX_RW_COUNT, there is
884 * no overflow possibility.
888 for (seg
= 0; seg
< nr_segs
; seg
++) {
892 if (__get_user(len
, &uvector
->iov_len
) ||
893 __get_user(buf
, &uvector
->iov_base
)) {
897 if (len
< 0) /* size_t not fitting in compat_ssize_t .. */
900 !access_ok(vrfy_dir(type
), compat_ptr(buf
), len
)) {
904 if (len
> MAX_RW_COUNT
- tot_len
)
905 len
= MAX_RW_COUNT
- tot_len
;
907 iov
->iov_base
= compat_ptr(buf
);
908 iov
->iov_len
= (compat_size_t
) len
;
919 static ssize_t
do_iter_read(struct file
*file
, struct iov_iter
*iter
,
920 loff_t
*pos
, int flags
)
925 tot_len
= iov_iter_count(iter
);
928 ret
= rw_verify_area(READ
, file
, pos
, tot_len
);
932 if (file
->f_op
->read_iter
)
933 ret
= do_iter_readv_writev(file
, iter
, pos
, READ
, flags
);
935 ret
= do_loop_readv_writev(file
, iter
, pos
, READ
, flags
);
938 fsnotify_access(file
);
942 static ssize_t
do_iter_write(struct file
*file
, struct iov_iter
*iter
,
943 loff_t
*pos
, int flags
)
948 tot_len
= iov_iter_count(iter
);
951 ret
= rw_verify_area(WRITE
, file
, pos
, tot_len
);
955 file_start_write(file
);
956 if (file
->f_op
->write_iter
)
957 ret
= do_iter_readv_writev(file
, iter
, pos
, WRITE
, flags
);
959 ret
= do_loop_readv_writev(file
, iter
, pos
, WRITE
, flags
);
960 file_end_write(file
);
962 fsnotify_modify(file
);
966 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
967 unsigned long vlen
, loff_t
*pos
, int flags
)
969 struct iovec iovstack
[UIO_FASTIOV
];
970 struct iovec
*iov
= iovstack
;
971 struct iov_iter iter
;
974 if (!(file
->f_mode
& FMODE_READ
))
976 if (!(file
->f_mode
& FMODE_CAN_READ
))
979 ret
= import_iovec(READ
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
983 ret
= do_iter_read(file
, &iter
, pos
, flags
);
987 EXPORT_SYMBOL(vfs_readv
);
989 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
990 unsigned long vlen
, loff_t
*pos
, int flags
)
992 struct iovec iovstack
[UIO_FASTIOV
];
993 struct iovec
*iov
= iovstack
;
994 struct iov_iter iter
;
997 if (!(file
->f_mode
& FMODE_WRITE
))
999 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1002 ret
= import_iovec(WRITE
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
1006 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1010 EXPORT_SYMBOL(vfs_writev
);
1012 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
1013 unsigned long vlen
, int flags
)
1015 struct fd f
= fdget_pos(fd
);
1016 ssize_t ret
= -EBADF
;
1019 loff_t pos
= file_pos_read(f
.file
);
1020 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1022 file_pos_write(f
.file
, pos
);
1027 add_rchar(current
, ret
);
1032 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
1033 unsigned long vlen
, int flags
)
1035 struct fd f
= fdget_pos(fd
);
1036 ssize_t ret
= -EBADF
;
1039 loff_t pos
= file_pos_read(f
.file
);
1040 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1042 file_pos_write(f
.file
, pos
);
1047 add_wchar(current
, ret
);
1052 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
1054 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1055 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
1058 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
1059 unsigned long vlen
, loff_t pos
, int flags
)
1062 ssize_t ret
= -EBADF
;
1070 if (f
.file
->f_mode
& FMODE_PREAD
)
1071 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1076 add_rchar(current
, ret
);
1081 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1082 unsigned long vlen
, loff_t pos
, int flags
)
1085 ssize_t ret
= -EBADF
;
1093 if (f
.file
->f_mode
& FMODE_PWRITE
)
1094 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1099 add_wchar(current
, ret
);
1104 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1105 unsigned long, vlen
)
1107 return do_readv(fd
, vec
, vlen
, 0);
1110 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1111 unsigned long, vlen
)
1113 return do_writev(fd
, vec
, vlen
, 0);
1116 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1117 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1119 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1121 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1124 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1125 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1128 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1131 return do_readv(fd
, vec
, vlen
, flags
);
1133 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1136 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1137 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1139 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1141 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1144 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1145 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1148 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1151 return do_writev(fd
, vec
, vlen
, flags
);
1153 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1156 #ifdef CONFIG_COMPAT
1157 static size_t compat_readv(struct file
*file
,
1158 const struct compat_iovec __user
*vec
,
1159 unsigned long vlen
, loff_t
*pos
, int flags
)
1161 struct iovec iovstack
[UIO_FASTIOV
];
1162 struct iovec
*iov
= iovstack
;
1163 struct iov_iter iter
;
1164 ssize_t ret
= -EBADF
;
1166 if (!(file
->f_mode
& FMODE_READ
))
1170 if (!(file
->f_mode
& FMODE_CAN_READ
))
1173 ret
= compat_import_iovec(READ
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1176 ret
= do_iter_read(file
, &iter
, pos
, flags
);
1180 add_rchar(current
, ret
);
1185 static size_t do_compat_readv(compat_ulong_t fd
,
1186 const struct compat_iovec __user
*vec
,
1187 compat_ulong_t vlen
, int flags
)
1189 struct fd f
= fdget_pos(fd
);
1195 pos
= f
.file
->f_pos
;
1196 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1198 f
.file
->f_pos
= pos
;
1204 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1205 const struct compat_iovec __user
*,vec
,
1206 compat_ulong_t
, vlen
)
1208 return do_compat_readv(fd
, vec
, vlen
, 0);
1211 static long do_compat_preadv64(unsigned long fd
,
1212 const struct compat_iovec __user
*vec
,
1213 unsigned long vlen
, loff_t pos
, int flags
)
1224 if (f
.file
->f_mode
& FMODE_PREAD
)
1225 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1230 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1231 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1232 const struct compat_iovec __user
*,vec
,
1233 unsigned long, vlen
, loff_t
, pos
)
1235 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1239 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1240 const struct compat_iovec __user
*,vec
,
1241 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1243 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1245 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1248 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1249 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1250 const struct compat_iovec __user
*,vec
,
1251 unsigned long, vlen
, loff_t
, pos
, int, flags
)
1253 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1257 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1258 const struct compat_iovec __user
*,vec
,
1259 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1262 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1265 return do_compat_readv(fd
, vec
, vlen
, flags
);
1267 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1270 static size_t compat_writev(struct file
*file
,
1271 const struct compat_iovec __user
*vec
,
1272 unsigned long vlen
, loff_t
*pos
, int flags
)
1274 struct iovec iovstack
[UIO_FASTIOV
];
1275 struct iovec
*iov
= iovstack
;
1276 struct iov_iter iter
;
1277 ssize_t ret
= -EBADF
;
1279 if (!(file
->f_mode
& FMODE_WRITE
))
1283 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1286 ret
= compat_import_iovec(WRITE
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1289 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1293 add_wchar(current
, ret
);
1298 static size_t do_compat_writev(compat_ulong_t fd
,
1299 const struct compat_iovec __user
* vec
,
1300 compat_ulong_t vlen
, int flags
)
1302 struct fd f
= fdget_pos(fd
);
1308 pos
= f
.file
->f_pos
;
1309 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1311 f
.file
->f_pos
= pos
;
1316 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1317 const struct compat_iovec __user
*, vec
,
1318 compat_ulong_t
, vlen
)
1320 return do_compat_writev(fd
, vec
, vlen
, 0);
1323 static long do_compat_pwritev64(unsigned long fd
,
1324 const struct compat_iovec __user
*vec
,
1325 unsigned long vlen
, loff_t pos
, int flags
)
1336 if (f
.file
->f_mode
& FMODE_PWRITE
)
1337 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1342 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1343 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1344 const struct compat_iovec __user
*,vec
,
1345 unsigned long, vlen
, loff_t
, pos
)
1347 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1351 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1352 const struct compat_iovec __user
*,vec
,
1353 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1355 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1357 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1360 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1361 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1362 const struct compat_iovec __user
*,vec
,
1363 unsigned long, vlen
, loff_t
, pos
, int, flags
)
1365 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1369 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1370 const struct compat_iovec __user
*,vec
,
1371 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, int, flags
)
1373 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1376 return do_compat_writev(fd
, vec
, vlen
, flags
);
1378 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1383 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1384 size_t count
, loff_t max
)
1387 struct inode
*in_inode
, *out_inode
;
1394 * Get input file, and verify that it is ok..
1400 if (!(in
.file
->f_mode
& FMODE_READ
))
1404 pos
= in
.file
->f_pos
;
1407 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1410 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1413 if (count
> MAX_RW_COUNT
)
1414 count
= MAX_RW_COUNT
;
1417 * Get output file, and verify that it is ok..
1420 out
= fdget(out_fd
);
1423 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1426 in_inode
= file_inode(in
.file
);
1427 out_inode
= file_inode(out
.file
);
1428 out_pos
= out
.file
->f_pos
;
1429 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1434 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1436 if (unlikely(pos
+ count
> max
)) {
1437 retval
= -EOVERFLOW
;
1446 * We need to debate whether we can enable this or not. The
1447 * man page documents EAGAIN return for the output at least,
1448 * and the application is arguably buggy if it doesn't expect
1449 * EAGAIN on a non-blocking file descriptor.
1451 if (in
.file
->f_flags
& O_NONBLOCK
)
1452 fl
= SPLICE_F_NONBLOCK
;
1454 file_start_write(out
.file
);
1455 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1456 file_end_write(out
.file
);
1459 add_rchar(current
, retval
);
1460 add_wchar(current
, retval
);
1461 fsnotify_access(in
.file
);
1462 fsnotify_modify(out
.file
);
1463 out
.file
->f_pos
= out_pos
;
1467 in
.file
->f_pos
= pos
;
1473 retval
= -EOVERFLOW
;
1483 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1490 if (unlikely(get_user(off
, offset
)))
1493 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1494 if (unlikely(put_user(pos
, offset
)))
1499 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1502 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1508 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1510 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1511 if (unlikely(put_user(pos
, offset
)))
1516 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1519 #ifdef CONFIG_COMPAT
1520 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1521 compat_off_t __user
*, offset
, compat_size_t
, count
)
1528 if (unlikely(get_user(off
, offset
)))
1531 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1532 if (unlikely(put_user(pos
, offset
)))
1537 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1540 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1541 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1547 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1549 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1550 if (unlikely(put_user(pos
, offset
)))
1555 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1560 * copy_file_range() differs from regular file read and write in that it
1561 * specifically allows return partial success. When it does so is up to
1562 * the copy_file_range method.
1564 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1565 struct file
*file_out
, loff_t pos_out
,
1566 size_t len
, unsigned int flags
)
1568 struct inode
*inode_in
= file_inode(file_in
);
1569 struct inode
*inode_out
= file_inode(file_out
);
1575 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1577 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1580 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1584 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1588 if (!(file_in
->f_mode
& FMODE_READ
) ||
1589 !(file_out
->f_mode
& FMODE_WRITE
) ||
1590 (file_out
->f_flags
& O_APPEND
))
1593 /* this could be relaxed once a method supports cross-fs copies */
1594 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1600 file_start_write(file_out
);
1603 * Try cloning first, this is supported by more file systems, and
1604 * more efficient if both clone and copy are supported (e.g. NFS).
1606 if (file_in
->f_op
->clone_file_range
) {
1607 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1608 file_out
, pos_out
, len
);
1615 if (file_out
->f_op
->copy_file_range
) {
1616 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1617 pos_out
, len
, flags
);
1618 if (ret
!= -EOPNOTSUPP
)
1622 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1623 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1627 fsnotify_access(file_in
);
1628 add_rchar(current
, ret
);
1629 fsnotify_modify(file_out
);
1630 add_wchar(current
, ret
);
1636 file_end_write(file_out
);
1640 EXPORT_SYMBOL(vfs_copy_file_range
);
1642 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1643 int, fd_out
, loff_t __user
*, off_out
,
1644 size_t, len
, unsigned int, flags
)
1650 ssize_t ret
= -EBADF
;
1652 f_in
= fdget(fd_in
);
1656 f_out
= fdget(fd_out
);
1662 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1665 pos_in
= f_in
.file
->f_pos
;
1669 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1672 pos_out
= f_out
.file
->f_pos
;
1675 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1682 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1685 f_in
.file
->f_pos
= pos_in
;
1689 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1692 f_out
.file
->f_pos
= pos_out
;
1704 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1706 struct inode
*inode
= file_inode(file
);
1708 if (unlikely(pos
< 0))
1711 if (unlikely((loff_t
) (pos
+ len
) < 0))
1714 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1715 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1718 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1719 write
? F_WRLCK
: F_RDLCK
);
1724 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1728 * Check that the two inodes are eligible for cloning, the ranges make
1729 * sense, and then flush all dirty data. Caller must ensure that the
1730 * inodes have been locked against any other modifications.
1732 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1733 * the usual negative error code.
1735 int vfs_clone_file_prep_inodes(struct inode
*inode_in
, loff_t pos_in
,
1736 struct inode
*inode_out
, loff_t pos_out
,
1737 u64
*len
, bool is_dedupe
)
1739 loff_t bs
= inode_out
->i_sb
->s_blocksize
;
1742 bool same_inode
= (inode_in
== inode_out
);
1745 /* Don't touch certain kinds of inodes */
1746 if (IS_IMMUTABLE(inode_out
))
1749 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
1752 /* Don't reflink dirs, pipes, sockets... */
1753 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1755 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1758 /* Are we going all the way to the end? */
1759 isize
= i_size_read(inode_in
);
1763 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1765 if (is_dedupe
|| pos_in
== isize
)
1769 *len
= isize
- pos_in
;
1772 /* Ensure offsets don't wrap and the input is inside i_size */
1773 if (pos_in
+ *len
< pos_in
|| pos_out
+ *len
< pos_out
||
1774 pos_in
+ *len
> isize
)
1777 /* Don't allow dedupe past EOF in the dest file */
1781 disize
= i_size_read(inode_out
);
1782 if (pos_out
>= disize
|| pos_out
+ *len
> disize
)
1786 /* If we're linking to EOF, continue to the block boundary. */
1787 if (pos_in
+ *len
== isize
)
1788 blen
= ALIGN(isize
, bs
) - pos_in
;
1792 /* Only reflink if we're aligned to block boundaries */
1793 if (!IS_ALIGNED(pos_in
, bs
) || !IS_ALIGNED(pos_in
+ blen
, bs
) ||
1794 !IS_ALIGNED(pos_out
, bs
) || !IS_ALIGNED(pos_out
+ blen
, bs
))
1797 /* Don't allow overlapped reflink within the same file */
1799 if (pos_out
+ blen
> pos_in
&& pos_out
< pos_in
+ blen
)
1803 /* Wait for the completion of any pending IOs on both files */
1804 inode_dio_wait(inode_in
);
1806 inode_dio_wait(inode_out
);
1808 ret
= filemap_write_and_wait_range(inode_in
->i_mapping
,
1809 pos_in
, pos_in
+ *len
- 1);
1813 ret
= filemap_write_and_wait_range(inode_out
->i_mapping
,
1814 pos_out
, pos_out
+ *len
- 1);
1819 * Check that the extents are the same.
1822 bool is_same
= false;
1824 ret
= vfs_dedupe_file_range_compare(inode_in
, pos_in
,
1825 inode_out
, pos_out
, *len
, &is_same
);
1834 EXPORT_SYMBOL(vfs_clone_file_prep_inodes
);
1836 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1837 struct file
*file_out
, loff_t pos_out
, u64 len
)
1839 struct inode
*inode_in
= file_inode(file_in
);
1840 struct inode
*inode_out
= file_inode(file_out
);
1843 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1845 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1849 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1850 * the same mount. Practically, they only need to be on the same file
1853 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1856 if (!(file_in
->f_mode
& FMODE_READ
) ||
1857 !(file_out
->f_mode
& FMODE_WRITE
) ||
1858 (file_out
->f_flags
& O_APPEND
))
1861 if (!file_in
->f_op
->clone_file_range
)
1864 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1868 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1872 if (pos_in
+ len
> i_size_read(inode_in
))
1875 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1876 file_out
, pos_out
, len
);
1878 fsnotify_access(file_in
);
1879 fsnotify_modify(file_out
);
1884 EXPORT_SYMBOL(vfs_clone_file_range
);
1887 * Read a page's worth of file data into the page cache. Return the page
1890 static struct page
*vfs_dedupe_get_page(struct inode
*inode
, loff_t offset
)
1892 struct address_space
*mapping
;
1896 n
= offset
>> PAGE_SHIFT
;
1897 mapping
= inode
->i_mapping
;
1898 page
= read_mapping_page(mapping
, n
, NULL
);
1901 if (!PageUptodate(page
)) {
1903 return ERR_PTR(-EIO
);
1910 * Compare extents of two files to see if they are the same.
1911 * Caller must have locked both inodes to prevent write races.
1913 int vfs_dedupe_file_range_compare(struct inode
*src
, loff_t srcoff
,
1914 struct inode
*dest
, loff_t destoff
,
1915 loff_t len
, bool *is_same
)
1921 struct page
*src_page
;
1922 struct page
*dest_page
;
1930 src_poff
= srcoff
& (PAGE_SIZE
- 1);
1931 dest_poff
= destoff
& (PAGE_SIZE
- 1);
1932 cmp_len
= min(PAGE_SIZE
- src_poff
,
1933 PAGE_SIZE
- dest_poff
);
1934 cmp_len
= min(cmp_len
, len
);
1938 src_page
= vfs_dedupe_get_page(src
, srcoff
);
1939 if (IS_ERR(src_page
)) {
1940 error
= PTR_ERR(src_page
);
1943 dest_page
= vfs_dedupe_get_page(dest
, destoff
);
1944 if (IS_ERR(dest_page
)) {
1945 error
= PTR_ERR(dest_page
);
1946 unlock_page(src_page
);
1950 src_addr
= kmap_atomic(src_page
);
1951 dest_addr
= kmap_atomic(dest_page
);
1953 flush_dcache_page(src_page
);
1954 flush_dcache_page(dest_page
);
1956 if (memcmp(src_addr
+ src_poff
, dest_addr
+ dest_poff
, cmp_len
))
1959 kunmap_atomic(dest_addr
);
1960 kunmap_atomic(src_addr
);
1961 unlock_page(dest_page
);
1962 unlock_page(src_page
);
1963 put_page(dest_page
);
1980 EXPORT_SYMBOL(vfs_dedupe_file_range_compare
);
1982 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1984 struct file_dedupe_range_info
*info
;
1985 struct inode
*src
= file_inode(file
);
1990 bool is_admin
= capable(CAP_SYS_ADMIN
);
1991 u16 count
= same
->dest_count
;
1992 struct file
*dst_file
;
1996 if (!(file
->f_mode
& FMODE_READ
))
1999 if (same
->reserved1
|| same
->reserved2
)
2002 off
= same
->src_offset
;
2003 len
= same
->src_length
;
2006 if (S_ISDIR(src
->i_mode
))
2010 if (!S_ISREG(src
->i_mode
))
2013 ret
= clone_verify_area(file
, off
, len
, false);
2018 if (off
+ len
> i_size_read(src
))
2021 /* pre-format output fields to sane values */
2022 for (i
= 0; i
< count
; i
++) {
2023 same
->info
[i
].bytes_deduped
= 0ULL;
2024 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
2027 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
2029 struct fd dst_fd
= fdget(info
->dest_fd
);
2031 dst_file
= dst_fd
.file
;
2033 info
->status
= -EBADF
;
2036 dst
= file_inode(dst_file
);
2038 ret
= mnt_want_write_file(dst_file
);
2044 dst_off
= info
->dest_offset
;
2045 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
2052 if (info
->reserved
) {
2053 info
->status
= -EINVAL
;
2054 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
2055 info
->status
= -EINVAL
;
2056 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
2057 info
->status
= -EXDEV
;
2058 } else if (S_ISDIR(dst
->i_mode
)) {
2059 info
->status
= -EISDIR
;
2060 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
2061 info
->status
= -EINVAL
;
2063 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
2066 if (deduped
== -EBADE
)
2067 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
2068 else if (deduped
< 0)
2069 info
->status
= deduped
;
2071 info
->bytes_deduped
+= deduped
;
2075 mnt_drop_write_file(dst_file
);
2079 if (fatal_signal_pending(current
))
2086 EXPORT_SYMBOL(vfs_dedupe_file_range
);