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/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
23 #include <linux/uaccess.h>
24 #include <asm/unistd.h>
26 typedef ssize_t (*io_fn_t
)(struct file
*, char __user
*, size_t, loff_t
*);
27 typedef ssize_t (*iter_fn_t
)(struct kiocb
*, struct iov_iter
*);
29 const struct file_operations generic_ro_fops
= {
30 .llseek
= generic_file_llseek
,
31 .read_iter
= generic_file_read_iter
,
32 .mmap
= generic_file_readonly_mmap
,
33 .splice_read
= generic_file_splice_read
,
36 EXPORT_SYMBOL(generic_ro_fops
);
38 static inline int unsigned_offsets(struct file
*file
)
40 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
53 * Return the specified offset on success and -EINVAL on invalid offset.
55 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
57 if (offset
< 0 && !unsigned_offsets(file
))
62 if (offset
!= file
->f_pos
) {
68 EXPORT_SYMBOL(vfs_setpos
);
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
87 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
88 loff_t maxsize
, loff_t eof
)
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
108 spin_lock(&file
->f_lock
);
109 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
110 spin_unlock(&file
->f_lock
);
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
131 return vfs_setpos(file
, offset
, maxsize
);
133 EXPORT_SYMBOL(generic_file_llseek_size
);
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
145 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
147 struct inode
*inode
= file
->f_mapping
->host
;
149 return generic_file_llseek_size(file
, offset
, whence
,
150 inode
->i_sb
->s_maxbytes
,
153 EXPORT_SYMBOL(generic_file_llseek
);
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
163 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
166 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
167 return generic_file_llseek_size(file
, offset
, whence
,
173 EXPORT_SYMBOL(fixed_size_llseek
);
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
182 loff_t
no_seek_end_llseek(struct file
*file
, loff_t offset
, int whence
)
185 case SEEK_SET
: case SEEK_CUR
:
186 return generic_file_llseek_size(file
, offset
, whence
,
192 EXPORT_SYMBOL(no_seek_end_llseek
);
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
202 loff_t
no_seek_end_llseek_size(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
205 case SEEK_SET
: case SEEK_CUR
:
206 return generic_file_llseek_size(file
, offset
, whence
,
212 EXPORT_SYMBOL(no_seek_end_llseek_size
);
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
225 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
229 EXPORT_SYMBOL(noop_llseek
);
231 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
235 EXPORT_SYMBOL(no_llseek
);
237 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
239 struct inode
*inode
= file_inode(file
);
245 offset
+= i_size_read(inode
);
249 retval
= file
->f_pos
;
252 offset
+= file
->f_pos
;
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
260 if (offset
>= inode
->i_size
) {
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
271 if (offset
>= inode
->i_size
) {
275 offset
= inode
->i_size
;
279 if (offset
>= 0 || unsigned_offsets(file
)) {
280 if (offset
!= file
->f_pos
) {
281 file
->f_pos
= offset
;
290 EXPORT_SYMBOL(default_llseek
);
292 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
294 loff_t (*fn
)(struct file
*, loff_t
, int);
297 if (file
->f_mode
& FMODE_LSEEK
) {
298 if (file
->f_op
->llseek
)
299 fn
= file
->f_op
->llseek
;
301 return fn(file
, offset
, whence
);
303 EXPORT_SYMBOL(vfs_llseek
);
305 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
308 struct fd f
= fdget_pos(fd
);
313 if (whence
<= SEEK_MAX
) {
314 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
316 if (res
!= (loff_t
)retval
)
317 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
324 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
326 return sys_lseek(fd
, offset
, whence
);
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
332 unsigned long, offset_low
, loff_t __user
*, result
,
333 unsigned int, whence
)
336 struct fd f
= fdget_pos(fd
);
343 if (whence
> SEEK_MAX
)
346 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
349 retval
= (int)offset
;
352 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
361 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
366 if (!file
->f_op
->read_iter
)
369 init_sync_kiocb(&kiocb
, file
);
370 kiocb
.ki_pos
= *ppos
;
373 ret
= file
->f_op
->read_iter(&kiocb
, iter
);
374 BUG_ON(ret
== -EIOCBQUEUED
);
376 *ppos
= kiocb
.ki_pos
;
379 EXPORT_SYMBOL(vfs_iter_read
);
381 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
386 if (!file
->f_op
->write_iter
)
389 init_sync_kiocb(&kiocb
, file
);
390 kiocb
.ki_pos
= *ppos
;
393 ret
= file
->f_op
->write_iter(&kiocb
, iter
);
394 BUG_ON(ret
== -EIOCBQUEUED
);
396 *ppos
= kiocb
.ki_pos
;
399 EXPORT_SYMBOL(vfs_iter_write
);
401 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
405 int retval
= -EINVAL
;
407 inode
= file_inode(file
);
408 if (unlikely((ssize_t
) count
< 0))
411 if (unlikely(pos
< 0)) {
412 if (!unsigned_offsets(file
))
414 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
416 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
417 if (!unsigned_offsets(file
))
421 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
422 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
423 read_write
== READ
? F_RDLCK
: F_WRLCK
);
427 return security_file_permission(file
,
428 read_write
== READ
? MAY_READ
: MAY_WRITE
);
431 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
433 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
435 struct iov_iter iter
;
438 init_sync_kiocb(&kiocb
, filp
);
439 kiocb
.ki_pos
= *ppos
;
440 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
442 ret
= filp
->f_op
->read_iter(&kiocb
, &iter
);
443 BUG_ON(ret
== -EIOCBQUEUED
);
444 *ppos
= kiocb
.ki_pos
;
448 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
451 if (file
->f_op
->read
)
452 return file
->f_op
->read(file
, buf
, count
, pos
);
453 else if (file
->f_op
->read_iter
)
454 return new_sync_read(file
, buf
, count
, pos
);
458 EXPORT_SYMBOL(__vfs_read
);
460 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
464 if (!(file
->f_mode
& FMODE_READ
))
466 if (!(file
->f_mode
& FMODE_CAN_READ
))
468 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
471 ret
= rw_verify_area(READ
, file
, pos
, count
);
473 if (count
> MAX_RW_COUNT
)
474 count
= MAX_RW_COUNT
;
475 ret
= __vfs_read(file
, buf
, count
, pos
);
477 fsnotify_access(file
);
478 add_rchar(current
, ret
);
486 EXPORT_SYMBOL(vfs_read
);
488 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
490 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
492 struct iov_iter iter
;
495 init_sync_kiocb(&kiocb
, filp
);
496 kiocb
.ki_pos
= *ppos
;
497 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
499 ret
= filp
->f_op
->write_iter(&kiocb
, &iter
);
500 BUG_ON(ret
== -EIOCBQUEUED
);
502 *ppos
= kiocb
.ki_pos
;
506 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
509 if (file
->f_op
->write
)
510 return file
->f_op
->write(file
, p
, count
, pos
);
511 else if (file
->f_op
->write_iter
)
512 return new_sync_write(file
, p
, count
, pos
);
516 EXPORT_SYMBOL(__vfs_write
);
518 vfs_readf_t
vfs_readf(struct file
*file
)
520 const struct file_operations
*fop
= file
->f_op
;
525 return new_sync_read
;
526 return ERR_PTR(-ENOSYS
);
528 EXPORT_SYMBOL_GPL(vfs_readf
);
530 vfs_writef_t
vfs_writef(struct file
*file
)
532 const struct file_operations
*fop
= file
->f_op
;
537 return new_sync_write
;
538 return ERR_PTR(-ENOSYS
);
540 EXPORT_SYMBOL_GPL(vfs_writef
);
542 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
545 const char __user
*p
;
548 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
553 p
= (__force
const char __user
*)buf
;
554 if (count
> MAX_RW_COUNT
)
555 count
= MAX_RW_COUNT
;
556 ret
= __vfs_write(file
, p
, count
, pos
);
559 fsnotify_modify(file
);
560 add_wchar(current
, ret
);
566 EXPORT_SYMBOL(__kernel_write
);
568 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
572 if (!(file
->f_mode
& FMODE_WRITE
))
574 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
576 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
579 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
581 if (count
> MAX_RW_COUNT
)
582 count
= MAX_RW_COUNT
;
583 file_start_write(file
);
584 ret
= __vfs_write(file
, buf
, count
, pos
);
586 fsnotify_modify(file
);
587 add_wchar(current
, ret
);
590 file_end_write(file
);
596 EXPORT_SYMBOL(vfs_write
);
598 static inline loff_t
file_pos_read(struct file
*file
)
603 static inline void file_pos_write(struct file
*file
, loff_t pos
)
608 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
610 struct fd f
= fdget_pos(fd
);
611 ssize_t ret
= -EBADF
;
614 loff_t pos
= file_pos_read(f
.file
);
615 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
617 file_pos_write(f
.file
, pos
);
623 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
626 struct fd f
= fdget_pos(fd
);
627 ssize_t ret
= -EBADF
;
630 loff_t pos
= file_pos_read(f
.file
);
631 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
633 file_pos_write(f
.file
, pos
);
640 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
641 size_t, count
, loff_t
, pos
)
644 ssize_t ret
= -EBADF
;
652 if (f
.file
->f_mode
& FMODE_PREAD
)
653 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
660 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
661 size_t, count
, loff_t
, pos
)
664 ssize_t ret
= -EBADF
;
672 if (f
.file
->f_mode
& FMODE_PWRITE
)
673 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
681 * Reduce an iovec's length in-place. Return the resulting number of segments
683 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
685 unsigned long seg
= 0;
688 while (seg
< nr_segs
) {
690 if (len
+ iov
->iov_len
>= to
) {
691 iov
->iov_len
= to
- len
;
699 EXPORT_SYMBOL(iov_shorten
);
701 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
702 loff_t
*ppos
, iter_fn_t fn
, int flags
)
707 if (flags
& ~(RWF_HIPRI
| RWF_DSYNC
| RWF_SYNC
))
710 init_sync_kiocb(&kiocb
, filp
);
711 if (flags
& RWF_HIPRI
)
712 kiocb
.ki_flags
|= IOCB_HIPRI
;
713 if (flags
& RWF_DSYNC
)
714 kiocb
.ki_flags
|= IOCB_DSYNC
;
715 if (flags
& RWF_SYNC
)
716 kiocb
.ki_flags
|= (IOCB_DSYNC
| IOCB_SYNC
);
717 kiocb
.ki_pos
= *ppos
;
719 ret
= fn(&kiocb
, iter
);
720 BUG_ON(ret
== -EIOCBQUEUED
);
721 *ppos
= kiocb
.ki_pos
;
725 /* Do it by hand, with file-ops */
726 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
727 loff_t
*ppos
, io_fn_t fn
, int flags
)
731 if (flags
& ~RWF_HIPRI
)
734 while (iov_iter_count(iter
)) {
735 struct iovec iovec
= iov_iter_iovec(iter
);
738 nr
= fn(filp
, iovec
.iov_base
, iovec
.iov_len
, ppos
);
746 if (nr
!= iovec
.iov_len
)
748 iov_iter_advance(iter
, nr
);
754 /* A write operation does a read from user space and vice versa */
755 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
758 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
759 * into the kernel and check that it is valid.
761 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
762 * @uvector: Pointer to the userspace array.
763 * @nr_segs: Number of elements in userspace array.
764 * @fast_segs: Number of elements in @fast_pointer.
765 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
766 * @ret_pointer: (output parameter) Pointer to a variable that will point to
767 * either @fast_pointer, a newly allocated kernel array, or NULL,
768 * depending on which array was used.
770 * This function copies an array of &struct iovec of @nr_segs from
771 * userspace into the kernel and checks that each element is valid (e.g.
772 * it does not point to a kernel address or cause overflow by being too
775 * As an optimization, the caller may provide a pointer to a small
776 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
777 * (the size of this array, or 0 if unused, should be given in @fast_segs).
779 * @ret_pointer will always point to the array that was used, so the
780 * caller must take care not to call kfree() on it e.g. in case the
781 * @fast_pointer array was used and it was allocated on the stack.
783 * Return: The total number of bytes covered by the iovec array on success
784 * or a negative error code on error.
786 ssize_t
rw_copy_check_uvector(int type
, const struct iovec __user
* uvector
,
787 unsigned long nr_segs
, unsigned long fast_segs
,
788 struct iovec
*fast_pointer
,
789 struct iovec
**ret_pointer
)
793 struct iovec
*iov
= fast_pointer
;
796 * SuS says "The readv() function *may* fail if the iovcnt argument
797 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
798 * traditionally returned zero for zero segments, so...
806 * First get the "struct iovec" from user memory and
807 * verify all the pointers
809 if (nr_segs
> UIO_MAXIOV
) {
813 if (nr_segs
> fast_segs
) {
814 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
820 if (copy_from_user(iov
, uvector
, nr_segs
*sizeof(*uvector
))) {
826 * According to the Single Unix Specification we should return EINVAL
827 * if an element length is < 0 when cast to ssize_t or if the
828 * total length would overflow the ssize_t return value of the
831 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
835 for (seg
= 0; seg
< nr_segs
; seg
++) {
836 void __user
*buf
= iov
[seg
].iov_base
;
837 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
839 /* see if we we're about to use an invalid len or if
840 * it's about to overflow ssize_t */
846 && unlikely(!access_ok(vrfy_dir(type
), buf
, len
))) {
850 if (len
> MAX_RW_COUNT
- ret
) {
851 len
= MAX_RW_COUNT
- ret
;
852 iov
[seg
].iov_len
= len
;
861 static ssize_t
do_readv_writev(int type
, struct file
*file
,
862 const struct iovec __user
* uvector
,
863 unsigned long nr_segs
, loff_t
*pos
,
867 struct iovec iovstack
[UIO_FASTIOV
];
868 struct iovec
*iov
= iovstack
;
869 struct iov_iter iter
;
874 ret
= import_iovec(type
, uvector
, nr_segs
,
875 ARRAY_SIZE(iovstack
), &iov
, &iter
);
879 tot_len
= iov_iter_count(&iter
);
882 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
887 fn
= file
->f_op
->read
;
888 iter_fn
= file
->f_op
->read_iter
;
890 fn
= (io_fn_t
)file
->f_op
->write
;
891 iter_fn
= file
->f_op
->write_iter
;
892 file_start_write(file
);
896 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
, flags
);
898 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
, flags
);
901 file_end_write(file
);
905 if ((ret
+ (type
== READ
)) > 0) {
907 fsnotify_access(file
);
909 fsnotify_modify(file
);
914 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
915 unsigned long vlen
, loff_t
*pos
, int flags
)
917 if (!(file
->f_mode
& FMODE_READ
))
919 if (!(file
->f_mode
& FMODE_CAN_READ
))
922 return do_readv_writev(READ
, file
, vec
, vlen
, pos
, flags
);
925 EXPORT_SYMBOL(vfs_readv
);
927 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
928 unsigned long vlen
, loff_t
*pos
, int flags
)
930 if (!(file
->f_mode
& FMODE_WRITE
))
932 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
935 return do_readv_writev(WRITE
, file
, vec
, vlen
, pos
, flags
);
938 EXPORT_SYMBOL(vfs_writev
);
940 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
941 unsigned long vlen
, int flags
)
943 struct fd f
= fdget_pos(fd
);
944 ssize_t ret
= -EBADF
;
947 loff_t pos
= file_pos_read(f
.file
);
948 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
950 file_pos_write(f
.file
, pos
);
955 add_rchar(current
, ret
);
960 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
961 unsigned long vlen
, int flags
)
963 struct fd f
= fdget_pos(fd
);
964 ssize_t ret
= -EBADF
;
967 loff_t pos
= file_pos_read(f
.file
);
968 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
970 file_pos_write(f
.file
, pos
);
975 add_wchar(current
, ret
);
980 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
982 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
983 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
986 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
987 unsigned long vlen
, loff_t pos
, int flags
)
990 ssize_t ret
= -EBADF
;
998 if (f
.file
->f_mode
& FMODE_PREAD
)
999 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1004 add_rchar(current
, ret
);
1009 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1010 unsigned long vlen
, loff_t pos
, int flags
)
1013 ssize_t ret
= -EBADF
;
1021 if (f
.file
->f_mode
& FMODE_PWRITE
)
1022 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1027 add_wchar(current
, ret
);
1032 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1033 unsigned long, vlen
)
1035 return do_readv(fd
, vec
, vlen
, 0);
1038 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1039 unsigned long, vlen
)
1041 return do_writev(fd
, vec
, vlen
, 0);
1044 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1045 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1047 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1049 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1052 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1053 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1056 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1059 return do_readv(fd
, vec
, vlen
, flags
);
1061 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1064 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1065 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1067 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1069 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1072 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1073 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1076 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1079 return do_writev(fd
, vec
, vlen
, flags
);
1081 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1084 #ifdef CONFIG_COMPAT
1086 static ssize_t
compat_do_readv_writev(int type
, struct file
*file
,
1087 const struct compat_iovec __user
*uvector
,
1088 unsigned long nr_segs
, loff_t
*pos
,
1091 compat_ssize_t tot_len
;
1092 struct iovec iovstack
[UIO_FASTIOV
];
1093 struct iovec
*iov
= iovstack
;
1094 struct iov_iter iter
;
1099 ret
= compat_import_iovec(type
, uvector
, nr_segs
,
1100 UIO_FASTIOV
, &iov
, &iter
);
1104 tot_len
= iov_iter_count(&iter
);
1107 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
1112 fn
= file
->f_op
->read
;
1113 iter_fn
= file
->f_op
->read_iter
;
1115 fn
= (io_fn_t
)file
->f_op
->write
;
1116 iter_fn
= file
->f_op
->write_iter
;
1117 file_start_write(file
);
1121 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
, flags
);
1123 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
, flags
);
1126 file_end_write(file
);
1130 if ((ret
+ (type
== READ
)) > 0) {
1132 fsnotify_access(file
);
1134 fsnotify_modify(file
);
1139 static size_t compat_readv(struct file
*file
,
1140 const struct compat_iovec __user
*vec
,
1141 unsigned long vlen
, loff_t
*pos
, int flags
)
1143 ssize_t ret
= -EBADF
;
1145 if (!(file
->f_mode
& FMODE_READ
))
1149 if (!(file
->f_mode
& FMODE_CAN_READ
))
1152 ret
= compat_do_readv_writev(READ
, file
, vec
, vlen
, pos
, flags
);
1156 add_rchar(current
, ret
);
1161 static size_t do_compat_readv(compat_ulong_t fd
,
1162 const struct compat_iovec __user
*vec
,
1163 compat_ulong_t vlen
, int flags
)
1165 struct fd f
= fdget_pos(fd
);
1171 pos
= f
.file
->f_pos
;
1172 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1174 f
.file
->f_pos
= pos
;
1180 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1181 const struct compat_iovec __user
*,vec
,
1182 compat_ulong_t
, vlen
)
1184 return do_compat_readv(fd
, vec
, vlen
, 0);
1187 static long do_compat_preadv64(unsigned long fd
,
1188 const struct compat_iovec __user
*vec
,
1189 unsigned long vlen
, loff_t pos
, int flags
)
1200 if (f
.file
->f_mode
& FMODE_PREAD
)
1201 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1206 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1207 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1208 const struct compat_iovec __user
*,vec
,
1209 unsigned long, vlen
, loff_t
, pos
)
1211 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1215 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1216 const struct compat_iovec __user
*,vec
,
1217 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1219 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1221 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1224 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1225 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1226 const struct compat_iovec __user
*,vec
,
1227 unsigned long, vlen
, loff_t
, pos
, int, flags
)
1229 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1233 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1234 const struct compat_iovec __user
*,vec
,
1235 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1238 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1241 return do_compat_readv(fd
, vec
, vlen
, flags
);
1243 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1246 static size_t compat_writev(struct file
*file
,
1247 const struct compat_iovec __user
*vec
,
1248 unsigned long vlen
, loff_t
*pos
, int flags
)
1250 ssize_t ret
= -EBADF
;
1252 if (!(file
->f_mode
& FMODE_WRITE
))
1256 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1259 ret
= compat_do_readv_writev(WRITE
, file
, vec
, vlen
, pos
, 0);
1263 add_wchar(current
, ret
);
1268 static size_t do_compat_writev(compat_ulong_t fd
,
1269 const struct compat_iovec __user
* vec
,
1270 compat_ulong_t vlen
, int flags
)
1272 struct fd f
= fdget_pos(fd
);
1278 pos
= f
.file
->f_pos
;
1279 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1281 f
.file
->f_pos
= pos
;
1286 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1287 const struct compat_iovec __user
*, vec
,
1288 compat_ulong_t
, vlen
)
1290 return do_compat_writev(fd
, vec
, vlen
, 0);
1293 static long do_compat_pwritev64(unsigned long fd
,
1294 const struct compat_iovec __user
*vec
,
1295 unsigned long vlen
, loff_t pos
, int flags
)
1306 if (f
.file
->f_mode
& FMODE_PWRITE
)
1307 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1312 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1313 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1314 const struct compat_iovec __user
*,vec
,
1315 unsigned long, vlen
, loff_t
, pos
)
1317 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1321 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1322 const struct compat_iovec __user
*,vec
,
1323 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1325 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1327 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1330 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1331 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1332 const struct compat_iovec __user
*,vec
,
1333 unsigned long, vlen
, loff_t
, pos
, int, flags
)
1335 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1339 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1340 const struct compat_iovec __user
*,vec
,
1341 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, int, flags
)
1343 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1346 return do_compat_writev(fd
, vec
, vlen
, flags
);
1348 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1353 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1354 size_t count
, loff_t max
)
1357 struct inode
*in_inode
, *out_inode
;
1364 * Get input file, and verify that it is ok..
1370 if (!(in
.file
->f_mode
& FMODE_READ
))
1374 pos
= in
.file
->f_pos
;
1377 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1380 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1383 if (count
> MAX_RW_COUNT
)
1384 count
= MAX_RW_COUNT
;
1387 * Get output file, and verify that it is ok..
1390 out
= fdget(out_fd
);
1393 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1396 in_inode
= file_inode(in
.file
);
1397 out_inode
= file_inode(out
.file
);
1398 out_pos
= out
.file
->f_pos
;
1399 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1404 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1406 if (unlikely(pos
+ count
> max
)) {
1407 retval
= -EOVERFLOW
;
1416 * We need to debate whether we can enable this or not. The
1417 * man page documents EAGAIN return for the output at least,
1418 * and the application is arguably buggy if it doesn't expect
1419 * EAGAIN on a non-blocking file descriptor.
1421 if (in
.file
->f_flags
& O_NONBLOCK
)
1422 fl
= SPLICE_F_NONBLOCK
;
1424 file_start_write(out
.file
);
1425 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1426 file_end_write(out
.file
);
1429 add_rchar(current
, retval
);
1430 add_wchar(current
, retval
);
1431 fsnotify_access(in
.file
);
1432 fsnotify_modify(out
.file
);
1433 out
.file
->f_pos
= out_pos
;
1437 in
.file
->f_pos
= pos
;
1443 retval
= -EOVERFLOW
;
1453 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1460 if (unlikely(get_user(off
, offset
)))
1463 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1464 if (unlikely(put_user(pos
, offset
)))
1469 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1472 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1478 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1480 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1481 if (unlikely(put_user(pos
, offset
)))
1486 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1489 #ifdef CONFIG_COMPAT
1490 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1491 compat_off_t __user
*, offset
, compat_size_t
, count
)
1498 if (unlikely(get_user(off
, offset
)))
1501 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1502 if (unlikely(put_user(pos
, offset
)))
1507 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1510 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1511 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1517 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1519 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1520 if (unlikely(put_user(pos
, offset
)))
1525 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1530 * copy_file_range() differs from regular file read and write in that it
1531 * specifically allows return partial success. When it does so is up to
1532 * the copy_file_range method.
1534 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1535 struct file
*file_out
, loff_t pos_out
,
1536 size_t len
, unsigned int flags
)
1538 struct inode
*inode_in
= file_inode(file_in
);
1539 struct inode
*inode_out
= file_inode(file_out
);
1545 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1549 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1553 if (!(file_in
->f_mode
& FMODE_READ
) ||
1554 !(file_out
->f_mode
& FMODE_WRITE
) ||
1555 (file_out
->f_flags
& O_APPEND
))
1558 /* this could be relaxed once a method supports cross-fs copies */
1559 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1565 sb_start_write(inode_out
->i_sb
);
1568 * Try cloning first, this is supported by more file systems, and
1569 * more efficient if both clone and copy are supported (e.g. NFS).
1571 if (file_in
->f_op
->clone_file_range
) {
1572 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1573 file_out
, pos_out
, len
);
1580 if (file_out
->f_op
->copy_file_range
) {
1581 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1582 pos_out
, len
, flags
);
1583 if (ret
!= -EOPNOTSUPP
)
1587 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1588 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1592 fsnotify_access(file_in
);
1593 add_rchar(current
, ret
);
1594 fsnotify_modify(file_out
);
1595 add_wchar(current
, ret
);
1601 sb_end_write(inode_out
->i_sb
);
1605 EXPORT_SYMBOL(vfs_copy_file_range
);
1607 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1608 int, fd_out
, loff_t __user
*, off_out
,
1609 size_t, len
, unsigned int, flags
)
1615 ssize_t ret
= -EBADF
;
1617 f_in
= fdget(fd_in
);
1621 f_out
= fdget(fd_out
);
1627 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1630 pos_in
= f_in
.file
->f_pos
;
1634 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1637 pos_out
= f_out
.file
->f_pos
;
1640 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1647 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1650 f_in
.file
->f_pos
= pos_in
;
1654 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1657 f_out
.file
->f_pos
= pos_out
;
1669 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1671 struct inode
*inode
= file_inode(file
);
1673 if (unlikely(pos
< 0))
1676 if (unlikely((loff_t
) (pos
+ len
) < 0))
1679 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1680 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1683 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1684 write
? F_WRLCK
: F_RDLCK
);
1689 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1693 * Check that the two inodes are eligible for cloning, the ranges make
1694 * sense, and then flush all dirty data. Caller must ensure that the
1695 * inodes have been locked against any other modifications.
1697 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1698 * the usual negative error code.
1700 int vfs_clone_file_prep_inodes(struct inode
*inode_in
, loff_t pos_in
,
1701 struct inode
*inode_out
, loff_t pos_out
,
1702 u64
*len
, bool is_dedupe
)
1704 loff_t bs
= inode_out
->i_sb
->s_blocksize
;
1707 bool same_inode
= (inode_in
== inode_out
);
1710 /* Don't touch certain kinds of inodes */
1711 if (IS_IMMUTABLE(inode_out
))
1714 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
1717 /* Don't reflink dirs, pipes, sockets... */
1718 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1720 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1723 /* Are we going all the way to the end? */
1724 isize
= i_size_read(inode_in
);
1728 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1730 if (is_dedupe
|| pos_in
== isize
)
1734 *len
= isize
- pos_in
;
1737 /* Ensure offsets don't wrap and the input is inside i_size */
1738 if (pos_in
+ *len
< pos_in
|| pos_out
+ *len
< pos_out
||
1739 pos_in
+ *len
> isize
)
1742 /* Don't allow dedupe past EOF in the dest file */
1746 disize
= i_size_read(inode_out
);
1747 if (pos_out
>= disize
|| pos_out
+ *len
> disize
)
1751 /* If we're linking to EOF, continue to the block boundary. */
1752 if (pos_in
+ *len
== isize
)
1753 blen
= ALIGN(isize
, bs
) - pos_in
;
1757 /* Only reflink if we're aligned to block boundaries */
1758 if (!IS_ALIGNED(pos_in
, bs
) || !IS_ALIGNED(pos_in
+ blen
, bs
) ||
1759 !IS_ALIGNED(pos_out
, bs
) || !IS_ALIGNED(pos_out
+ blen
, bs
))
1762 /* Don't allow overlapped reflink within the same file */
1764 if (pos_out
+ blen
> pos_in
&& pos_out
< pos_in
+ blen
)
1768 /* Wait for the completion of any pending IOs on both files */
1769 inode_dio_wait(inode_in
);
1771 inode_dio_wait(inode_out
);
1773 ret
= filemap_write_and_wait_range(inode_in
->i_mapping
,
1774 pos_in
, pos_in
+ *len
- 1);
1778 ret
= filemap_write_and_wait_range(inode_out
->i_mapping
,
1779 pos_out
, pos_out
+ *len
- 1);
1784 * Check that the extents are the same.
1787 bool is_same
= false;
1789 ret
= vfs_dedupe_file_range_compare(inode_in
, pos_in
,
1790 inode_out
, pos_out
, *len
, &is_same
);
1799 EXPORT_SYMBOL(vfs_clone_file_prep_inodes
);
1801 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1802 struct file
*file_out
, loff_t pos_out
, u64 len
)
1804 struct inode
*inode_in
= file_inode(file_in
);
1805 struct inode
*inode_out
= file_inode(file_out
);
1808 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1810 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1814 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1815 * the same mount. Practically, they only need to be on the same file
1818 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1821 if (!(file_in
->f_mode
& FMODE_READ
) ||
1822 !(file_out
->f_mode
& FMODE_WRITE
) ||
1823 (file_out
->f_flags
& O_APPEND
))
1826 if (!file_in
->f_op
->clone_file_range
)
1829 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1833 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1837 if (pos_in
+ len
> i_size_read(inode_in
))
1840 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1841 file_out
, pos_out
, len
);
1843 fsnotify_access(file_in
);
1844 fsnotify_modify(file_out
);
1849 EXPORT_SYMBOL(vfs_clone_file_range
);
1852 * Read a page's worth of file data into the page cache. Return the page
1855 static struct page
*vfs_dedupe_get_page(struct inode
*inode
, loff_t offset
)
1857 struct address_space
*mapping
;
1861 n
= offset
>> PAGE_SHIFT
;
1862 mapping
= inode
->i_mapping
;
1863 page
= read_mapping_page(mapping
, n
, NULL
);
1866 if (!PageUptodate(page
)) {
1868 return ERR_PTR(-EIO
);
1875 * Compare extents of two files to see if they are the same.
1876 * Caller must have locked both inodes to prevent write races.
1878 int vfs_dedupe_file_range_compare(struct inode
*src
, loff_t srcoff
,
1879 struct inode
*dest
, loff_t destoff
,
1880 loff_t len
, bool *is_same
)
1886 struct page
*src_page
;
1887 struct page
*dest_page
;
1895 src_poff
= srcoff
& (PAGE_SIZE
- 1);
1896 dest_poff
= destoff
& (PAGE_SIZE
- 1);
1897 cmp_len
= min(PAGE_SIZE
- src_poff
,
1898 PAGE_SIZE
- dest_poff
);
1899 cmp_len
= min(cmp_len
, len
);
1903 src_page
= vfs_dedupe_get_page(src
, srcoff
);
1904 if (IS_ERR(src_page
)) {
1905 error
= PTR_ERR(src_page
);
1908 dest_page
= vfs_dedupe_get_page(dest
, destoff
);
1909 if (IS_ERR(dest_page
)) {
1910 error
= PTR_ERR(dest_page
);
1911 unlock_page(src_page
);
1915 src_addr
= kmap_atomic(src_page
);
1916 dest_addr
= kmap_atomic(dest_page
);
1918 flush_dcache_page(src_page
);
1919 flush_dcache_page(dest_page
);
1921 if (memcmp(src_addr
+ src_poff
, dest_addr
+ dest_poff
, cmp_len
))
1924 kunmap_atomic(dest_addr
);
1925 kunmap_atomic(src_addr
);
1926 unlock_page(dest_page
);
1927 unlock_page(src_page
);
1928 put_page(dest_page
);
1945 EXPORT_SYMBOL(vfs_dedupe_file_range_compare
);
1947 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1949 struct file_dedupe_range_info
*info
;
1950 struct inode
*src
= file_inode(file
);
1955 bool is_admin
= capable(CAP_SYS_ADMIN
);
1956 u16 count
= same
->dest_count
;
1957 struct file
*dst_file
;
1961 if (!(file
->f_mode
& FMODE_READ
))
1964 if (same
->reserved1
|| same
->reserved2
)
1967 off
= same
->src_offset
;
1968 len
= same
->src_length
;
1971 if (S_ISDIR(src
->i_mode
))
1975 if (!S_ISREG(src
->i_mode
))
1978 ret
= clone_verify_area(file
, off
, len
, false);
1983 if (off
+ len
> i_size_read(src
))
1986 /* pre-format output fields to sane values */
1987 for (i
= 0; i
< count
; i
++) {
1988 same
->info
[i
].bytes_deduped
= 0ULL;
1989 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
1992 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
1994 struct fd dst_fd
= fdget(info
->dest_fd
);
1996 dst_file
= dst_fd
.file
;
1998 info
->status
= -EBADF
;
2001 dst
= file_inode(dst_file
);
2003 ret
= mnt_want_write_file(dst_file
);
2009 dst_off
= info
->dest_offset
;
2010 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
2017 if (info
->reserved
) {
2018 info
->status
= -EINVAL
;
2019 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
2020 info
->status
= -EINVAL
;
2021 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
2022 info
->status
= -EXDEV
;
2023 } else if (S_ISDIR(dst
->i_mode
)) {
2024 info
->status
= -EISDIR
;
2025 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
2026 info
->status
= -EINVAL
;
2028 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
2031 if (deduped
== -EBADE
)
2032 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
2033 else if (deduped
< 0)
2034 info
->status
= deduped
;
2036 info
->bytes_deduped
+= deduped
;
2040 mnt_drop_write_file(dst_file
);
2044 if (fatal_signal_pending(current
))
2051 EXPORT_SYMBOL(vfs_dedupe_file_range
);