1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/read_write.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
28 const struct file_operations generic_ro_fops
= {
29 .llseek
= generic_file_llseek
,
30 .read_iter
= generic_file_read_iter
,
31 .mmap
= generic_file_readonly_mmap
,
32 .splice_read
= generic_file_splice_read
,
35 EXPORT_SYMBOL(generic_ro_fops
);
37 static inline bool unsigned_offsets(struct file
*file
)
39 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
52 * Return the specified offset on success and -EINVAL on invalid offset.
54 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
56 if (offset
< 0 && !unsigned_offsets(file
))
61 if (offset
!= file
->f_pos
) {
67 EXPORT_SYMBOL(vfs_setpos
);
70 * generic_file_llseek_size - generic llseek implementation for regular files
71 * @file: file structure to seek on
72 * @offset: file offset to seek to
73 * @whence: type of seek
74 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
77 * This is a variant of generic_file_llseek that allows passing in a custom
78 * maximum file size and a custom EOF position, for e.g. hashed directories
81 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
86 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
87 loff_t maxsize
, loff_t eof
)
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
107 spin_lock(&file
->f_lock
);
108 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
109 spin_unlock(&file
->f_lock
);
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
116 if ((unsigned long long)offset
>= eof
)
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
124 if ((unsigned long long)offset
>= eof
)
130 return vfs_setpos(file
, offset
, maxsize
);
132 EXPORT_SYMBOL(generic_file_llseek_size
);
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
138 * @whence: type of seek
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
142 * @offset and @whence.
144 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
146 struct inode
*inode
= file
->f_mapping
->host
;
148 return generic_file_llseek_size(file
, offset
, whence
,
149 inode
->i_sb
->s_maxbytes
,
152 EXPORT_SYMBOL(generic_file_llseek
);
155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
162 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
165 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
166 return generic_file_llseek_size(file
, offset
, whence
,
172 EXPORT_SYMBOL(fixed_size_llseek
);
175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
181 loff_t
no_seek_end_llseek(struct file
*file
, loff_t offset
, int whence
)
184 case SEEK_SET
: case SEEK_CUR
:
185 return generic_file_llseek_size(file
, offset
, whence
,
191 EXPORT_SYMBOL(no_seek_end_llseek
);
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
201 loff_t
no_seek_end_llseek_size(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
204 case SEEK_SET
: case SEEK_CUR
:
205 return generic_file_llseek_size(file
, offset
, whence
,
211 EXPORT_SYMBOL(no_seek_end_llseek_size
);
214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
217 * @whence: type of seek
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
224 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
228 EXPORT_SYMBOL(noop_llseek
);
230 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
234 EXPORT_SYMBOL(no_llseek
);
236 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
238 struct inode
*inode
= file_inode(file
);
244 offset
+= i_size_read(inode
);
248 retval
= file
->f_pos
;
251 offset
+= file
->f_pos
;
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
259 if (offset
>= inode
->i_size
) {
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
270 if (offset
>= inode
->i_size
) {
274 offset
= inode
->i_size
;
278 if (offset
>= 0 || unsigned_offsets(file
)) {
279 if (offset
!= file
->f_pos
) {
280 file
->f_pos
= offset
;
289 EXPORT_SYMBOL(default_llseek
);
291 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
293 loff_t (*fn
)(struct file
*, loff_t
, int);
296 if (file
->f_mode
& FMODE_LSEEK
) {
297 if (file
->f_op
->llseek
)
298 fn
= file
->f_op
->llseek
;
300 return fn(file
, offset
, whence
);
302 EXPORT_SYMBOL(vfs_llseek
);
304 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
307 struct fd f
= fdget_pos(fd
);
312 if (whence
<= SEEK_MAX
) {
313 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
315 if (res
!= (loff_t
)retval
)
316 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
323 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
325 return sys_lseek(fd
, offset
, whence
);
329 #ifdef __ARCH_WANT_SYS_LLSEEK
330 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
331 unsigned long, offset_low
, loff_t __user
*, result
,
332 unsigned int, whence
)
335 struct fd f
= fdget_pos(fd
);
342 if (whence
> SEEK_MAX
)
345 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
348 retval
= (int)offset
;
351 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
360 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
364 int retval
= -EINVAL
;
366 inode
= file_inode(file
);
367 if (unlikely((ssize_t
) count
< 0))
370 if (unlikely(pos
< 0)) {
371 if (!unsigned_offsets(file
))
373 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
375 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
376 if (!unsigned_offsets(file
))
380 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
381 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
382 read_write
== READ
? F_RDLCK
: F_WRLCK
);
386 return security_file_permission(file
,
387 read_write
== READ
? MAY_READ
: MAY_WRITE
);
390 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
392 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
394 struct iov_iter iter
;
397 init_sync_kiocb(&kiocb
, filp
);
398 kiocb
.ki_pos
= *ppos
;
399 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
401 ret
= call_read_iter(filp
, &kiocb
, &iter
);
402 BUG_ON(ret
== -EIOCBQUEUED
);
403 *ppos
= kiocb
.ki_pos
;
407 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
410 if (file
->f_op
->read
)
411 return file
->f_op
->read(file
, buf
, count
, pos
);
412 else if (file
->f_op
->read_iter
)
413 return new_sync_read(file
, buf
, count
, pos
);
418 ssize_t
kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
425 /* The cast to a user pointer is valid due to the set_fs() */
426 result
= vfs_read(file
, (void __user
*)buf
, count
, pos
);
430 EXPORT_SYMBOL(kernel_read
);
432 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
436 if (!(file
->f_mode
& FMODE_READ
))
438 if (!(file
->f_mode
& FMODE_CAN_READ
))
440 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
443 ret
= rw_verify_area(READ
, file
, pos
, count
);
445 if (count
> MAX_RW_COUNT
)
446 count
= MAX_RW_COUNT
;
447 ret
= __vfs_read(file
, buf
, count
, pos
);
449 fsnotify_access(file
);
450 add_rchar(current
, ret
);
458 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
460 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
462 struct iov_iter iter
;
465 init_sync_kiocb(&kiocb
, filp
);
466 kiocb
.ki_pos
= *ppos
;
467 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
469 ret
= call_write_iter(filp
, &kiocb
, &iter
);
470 BUG_ON(ret
== -EIOCBQUEUED
);
472 *ppos
= kiocb
.ki_pos
;
476 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
479 if (file
->f_op
->write
)
480 return file
->f_op
->write(file
, p
, count
, pos
);
481 else if (file
->f_op
->write_iter
)
482 return new_sync_write(file
, p
, count
, pos
);
487 ssize_t
__kernel_write(struct file
*file
, const void *buf
, size_t count
, loff_t
*pos
)
490 const char __user
*p
;
493 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
498 p
= (__force
const char __user
*)buf
;
499 if (count
> MAX_RW_COUNT
)
500 count
= MAX_RW_COUNT
;
501 ret
= __vfs_write(file
, p
, count
, pos
);
504 fsnotify_modify(file
);
505 add_wchar(current
, ret
);
510 EXPORT_SYMBOL(__kernel_write
);
512 ssize_t
kernel_write(struct file
*file
, const void *buf
, size_t count
,
520 /* The cast to a user pointer is valid due to the set_fs() */
521 res
= vfs_write(file
, (__force
const char __user
*)buf
, count
, pos
);
526 EXPORT_SYMBOL(kernel_write
);
528 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
532 if (!(file
->f_mode
& FMODE_WRITE
))
534 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
536 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
539 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
541 if (count
> MAX_RW_COUNT
)
542 count
= MAX_RW_COUNT
;
543 file_start_write(file
);
544 ret
= __vfs_write(file
, buf
, count
, pos
);
546 fsnotify_modify(file
);
547 add_wchar(current
, ret
);
550 file_end_write(file
);
556 static inline loff_t
file_pos_read(struct file
*file
)
561 static inline void file_pos_write(struct file
*file
, loff_t pos
)
566 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
568 struct fd f
= fdget_pos(fd
);
569 ssize_t ret
= -EBADF
;
572 loff_t pos
= file_pos_read(f
.file
);
573 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
575 file_pos_write(f
.file
, pos
);
581 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
584 struct fd f
= fdget_pos(fd
);
585 ssize_t ret
= -EBADF
;
588 loff_t pos
= file_pos_read(f
.file
);
589 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
591 file_pos_write(f
.file
, pos
);
598 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
599 size_t, count
, loff_t
, pos
)
602 ssize_t ret
= -EBADF
;
610 if (f
.file
->f_mode
& FMODE_PREAD
)
611 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
618 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
619 size_t, count
, loff_t
, pos
)
622 ssize_t ret
= -EBADF
;
630 if (f
.file
->f_mode
& FMODE_PWRITE
)
631 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
639 * Reduce an iovec's length in-place. Return the resulting number of segments
641 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
643 unsigned long seg
= 0;
646 while (seg
< nr_segs
) {
648 if (len
+ iov
->iov_len
>= to
) {
649 iov
->iov_len
= to
- len
;
657 EXPORT_SYMBOL(iov_shorten
);
659 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
660 loff_t
*ppos
, int type
, rwf_t flags
)
665 init_sync_kiocb(&kiocb
, filp
);
666 ret
= kiocb_set_rw_flags(&kiocb
, flags
);
669 kiocb
.ki_pos
= *ppos
;
672 ret
= call_read_iter(filp
, &kiocb
, iter
);
674 ret
= call_write_iter(filp
, &kiocb
, iter
);
675 BUG_ON(ret
== -EIOCBQUEUED
);
676 *ppos
= kiocb
.ki_pos
;
680 /* Do it by hand, with file-ops */
681 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
682 loff_t
*ppos
, int type
, rwf_t flags
)
686 if (flags
& ~RWF_HIPRI
)
689 while (iov_iter_count(iter
)) {
690 struct iovec iovec
= iov_iter_iovec(iter
);
694 nr
= filp
->f_op
->read(filp
, iovec
.iov_base
,
695 iovec
.iov_len
, ppos
);
697 nr
= filp
->f_op
->write(filp
, iovec
.iov_base
,
698 iovec
.iov_len
, ppos
);
707 if (nr
!= iovec
.iov_len
)
709 iov_iter_advance(iter
, nr
);
715 /* A write operation does a read from user space and vice versa */
716 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
719 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
720 * into the kernel and check that it is valid.
722 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
723 * @uvector: Pointer to the userspace array.
724 * @nr_segs: Number of elements in userspace array.
725 * @fast_segs: Number of elements in @fast_pointer.
726 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
727 * @ret_pointer: (output parameter) Pointer to a variable that will point to
728 * either @fast_pointer, a newly allocated kernel array, or NULL,
729 * depending on which array was used.
731 * This function copies an array of &struct iovec of @nr_segs from
732 * userspace into the kernel and checks that each element is valid (e.g.
733 * it does not point to a kernel address or cause overflow by being too
736 * As an optimization, the caller may provide a pointer to a small
737 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
738 * (the size of this array, or 0 if unused, should be given in @fast_segs).
740 * @ret_pointer will always point to the array that was used, so the
741 * caller must take care not to call kfree() on it e.g. in case the
742 * @fast_pointer array was used and it was allocated on the stack.
744 * Return: The total number of bytes covered by the iovec array on success
745 * or a negative error code on error.
747 ssize_t
rw_copy_check_uvector(int type
, const struct iovec __user
* uvector
,
748 unsigned long nr_segs
, unsigned long fast_segs
,
749 struct iovec
*fast_pointer
,
750 struct iovec
**ret_pointer
)
754 struct iovec
*iov
= fast_pointer
;
757 * SuS says "The readv() function *may* fail if the iovcnt argument
758 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
759 * traditionally returned zero for zero segments, so...
767 * First get the "struct iovec" from user memory and
768 * verify all the pointers
770 if (nr_segs
> UIO_MAXIOV
) {
774 if (nr_segs
> fast_segs
) {
775 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
781 if (copy_from_user(iov
, uvector
, nr_segs
*sizeof(*uvector
))) {
787 * According to the Single Unix Specification we should return EINVAL
788 * if an element length is < 0 when cast to ssize_t or if the
789 * total length would overflow the ssize_t return value of the
792 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
796 for (seg
= 0; seg
< nr_segs
; seg
++) {
797 void __user
*buf
= iov
[seg
].iov_base
;
798 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
800 /* see if we we're about to use an invalid len or if
801 * it's about to overflow ssize_t */
807 && unlikely(!access_ok(vrfy_dir(type
), buf
, len
))) {
811 if (len
> MAX_RW_COUNT
- ret
) {
812 len
= MAX_RW_COUNT
- ret
;
813 iov
[seg
].iov_len
= len
;
823 ssize_t
compat_rw_copy_check_uvector(int type
,
824 const struct compat_iovec __user
*uvector
, unsigned long nr_segs
,
825 unsigned long fast_segs
, struct iovec
*fast_pointer
,
826 struct iovec
**ret_pointer
)
828 compat_ssize_t tot_len
;
829 struct iovec
*iov
= *ret_pointer
= fast_pointer
;
834 * SuS says "The readv() function *may* fail if the iovcnt argument
835 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
836 * traditionally returned zero for zero segments, so...
842 if (nr_segs
> UIO_MAXIOV
)
844 if (nr_segs
> fast_segs
) {
846 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
853 if (!access_ok(VERIFY_READ
, uvector
, nr_segs
*sizeof(*uvector
)))
857 * Single unix specification:
858 * We should -EINVAL if an element length is not >= 0 and fitting an
861 * In Linux, the total length is limited to MAX_RW_COUNT, there is
862 * no overflow possibility.
866 for (seg
= 0; seg
< nr_segs
; seg
++) {
870 if (__get_user(len
, &uvector
->iov_len
) ||
871 __get_user(buf
, &uvector
->iov_base
)) {
875 if (len
< 0) /* size_t not fitting in compat_ssize_t .. */
878 !access_ok(vrfy_dir(type
), compat_ptr(buf
), len
)) {
882 if (len
> MAX_RW_COUNT
- tot_len
)
883 len
= MAX_RW_COUNT
- tot_len
;
885 iov
->iov_base
= compat_ptr(buf
);
886 iov
->iov_len
= (compat_size_t
) len
;
897 static ssize_t
do_iter_read(struct file
*file
, struct iov_iter
*iter
,
898 loff_t
*pos
, rwf_t flags
)
903 if (!(file
->f_mode
& FMODE_READ
))
905 if (!(file
->f_mode
& FMODE_CAN_READ
))
908 tot_len
= iov_iter_count(iter
);
911 ret
= rw_verify_area(READ
, file
, pos
, tot_len
);
915 if (file
->f_op
->read_iter
)
916 ret
= do_iter_readv_writev(file
, iter
, pos
, READ
, flags
);
918 ret
= do_loop_readv_writev(file
, iter
, pos
, READ
, flags
);
921 fsnotify_access(file
);
925 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
928 if (!file
->f_op
->read_iter
)
930 return do_iter_read(file
, iter
, ppos
, flags
);
932 EXPORT_SYMBOL(vfs_iter_read
);
934 static ssize_t
do_iter_write(struct file
*file
, struct iov_iter
*iter
,
935 loff_t
*pos
, rwf_t flags
)
940 if (!(file
->f_mode
& FMODE_WRITE
))
942 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
945 tot_len
= iov_iter_count(iter
);
948 ret
= rw_verify_area(WRITE
, file
, pos
, tot_len
);
952 if (file
->f_op
->write_iter
)
953 ret
= do_iter_readv_writev(file
, iter
, pos
, WRITE
, flags
);
955 ret
= do_loop_readv_writev(file
, iter
, pos
, WRITE
, flags
);
957 fsnotify_modify(file
);
961 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
964 if (!file
->f_op
->write_iter
)
966 return do_iter_write(file
, iter
, ppos
, flags
);
968 EXPORT_SYMBOL(vfs_iter_write
);
970 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
971 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
973 struct iovec iovstack
[UIO_FASTIOV
];
974 struct iovec
*iov
= iovstack
;
975 struct iov_iter iter
;
978 ret
= import_iovec(READ
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
980 ret
= do_iter_read(file
, &iter
, pos
, flags
);
987 static ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
988 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
990 struct iovec iovstack
[UIO_FASTIOV
];
991 struct iovec
*iov
= iovstack
;
992 struct iov_iter iter
;
995 ret
= import_iovec(WRITE
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
997 file_start_write(file
);
998 ret
= do_iter_write(file
, &iter
, pos
, flags
);
999 file_end_write(file
);
1005 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
1006 unsigned long vlen
, rwf_t flags
)
1008 struct fd f
= fdget_pos(fd
);
1009 ssize_t ret
= -EBADF
;
1012 loff_t pos
= file_pos_read(f
.file
);
1013 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1015 file_pos_write(f
.file
, pos
);
1020 add_rchar(current
, ret
);
1025 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
1026 unsigned long vlen
, rwf_t flags
)
1028 struct fd f
= fdget_pos(fd
);
1029 ssize_t ret
= -EBADF
;
1032 loff_t pos
= file_pos_read(f
.file
);
1033 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1035 file_pos_write(f
.file
, pos
);
1040 add_wchar(current
, ret
);
1045 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
1047 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1048 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
1051 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
1052 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1055 ssize_t ret
= -EBADF
;
1063 if (f
.file
->f_mode
& FMODE_PREAD
)
1064 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1069 add_rchar(current
, ret
);
1074 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1075 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1078 ssize_t ret
= -EBADF
;
1086 if (f
.file
->f_mode
& FMODE_PWRITE
)
1087 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1092 add_wchar(current
, ret
);
1097 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1098 unsigned long, vlen
)
1100 return do_readv(fd
, vec
, vlen
, 0);
1103 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1104 unsigned long, vlen
)
1106 return do_writev(fd
, vec
, vlen
, 0);
1109 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1110 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1112 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1114 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1117 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1118 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1121 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1124 return do_readv(fd
, vec
, vlen
, flags
);
1126 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1129 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1130 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1132 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1134 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1137 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1138 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1141 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1144 return do_writev(fd
, vec
, vlen
, flags
);
1146 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1149 #ifdef CONFIG_COMPAT
1150 static size_t compat_readv(struct file
*file
,
1151 const struct compat_iovec __user
*vec
,
1152 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
1154 struct iovec iovstack
[UIO_FASTIOV
];
1155 struct iovec
*iov
= iovstack
;
1156 struct iov_iter iter
;
1159 ret
= compat_import_iovec(READ
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1161 ret
= do_iter_read(file
, &iter
, pos
, flags
);
1165 add_rchar(current
, ret
);
1170 static size_t do_compat_readv(compat_ulong_t fd
,
1171 const struct compat_iovec __user
*vec
,
1172 compat_ulong_t vlen
, rwf_t flags
)
1174 struct fd f
= fdget_pos(fd
);
1180 pos
= f
.file
->f_pos
;
1181 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1183 f
.file
->f_pos
= pos
;
1189 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1190 const struct compat_iovec __user
*,vec
,
1191 compat_ulong_t
, vlen
)
1193 return do_compat_readv(fd
, vec
, vlen
, 0);
1196 static long do_compat_preadv64(unsigned long fd
,
1197 const struct compat_iovec __user
*vec
,
1198 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1209 if (f
.file
->f_mode
& FMODE_PREAD
)
1210 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1215 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1216 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1217 const struct compat_iovec __user
*,vec
,
1218 unsigned long, vlen
, loff_t
, pos
)
1220 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1224 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1225 const struct compat_iovec __user
*,vec
,
1226 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1228 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1230 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1233 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1234 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1235 const struct compat_iovec __user
*,vec
,
1236 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1238 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1242 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1243 const struct compat_iovec __user
*,vec
,
1244 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1247 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1250 return do_compat_readv(fd
, vec
, vlen
, flags
);
1252 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1255 static size_t compat_writev(struct file
*file
,
1256 const struct compat_iovec __user
*vec
,
1257 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
1259 struct iovec iovstack
[UIO_FASTIOV
];
1260 struct iovec
*iov
= iovstack
;
1261 struct iov_iter iter
;
1264 ret
= compat_import_iovec(WRITE
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1266 file_start_write(file
);
1267 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1268 file_end_write(file
);
1272 add_wchar(current
, ret
);
1277 static size_t do_compat_writev(compat_ulong_t fd
,
1278 const struct compat_iovec __user
* vec
,
1279 compat_ulong_t vlen
, rwf_t flags
)
1281 struct fd f
= fdget_pos(fd
);
1287 pos
= f
.file
->f_pos
;
1288 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1290 f
.file
->f_pos
= pos
;
1295 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1296 const struct compat_iovec __user
*, vec
,
1297 compat_ulong_t
, vlen
)
1299 return do_compat_writev(fd
, vec
, vlen
, 0);
1302 static long do_compat_pwritev64(unsigned long fd
,
1303 const struct compat_iovec __user
*vec
,
1304 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1315 if (f
.file
->f_mode
& FMODE_PWRITE
)
1316 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1321 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1322 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1323 const struct compat_iovec __user
*,vec
,
1324 unsigned long, vlen
, loff_t
, pos
)
1326 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1330 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1331 const struct compat_iovec __user
*,vec
,
1332 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1334 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1336 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1339 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1340 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1341 const struct compat_iovec __user
*,vec
,
1342 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1344 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1348 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1349 const struct compat_iovec __user
*,vec
,
1350 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, rwf_t
, flags
)
1352 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1355 return do_compat_writev(fd
, vec
, vlen
, flags
);
1357 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1362 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1363 size_t count
, loff_t max
)
1366 struct inode
*in_inode
, *out_inode
;
1373 * Get input file, and verify that it is ok..
1379 if (!(in
.file
->f_mode
& FMODE_READ
))
1383 pos
= in
.file
->f_pos
;
1386 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1389 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1392 if (count
> MAX_RW_COUNT
)
1393 count
= MAX_RW_COUNT
;
1396 * Get output file, and verify that it is ok..
1399 out
= fdget(out_fd
);
1402 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1405 in_inode
= file_inode(in
.file
);
1406 out_inode
= file_inode(out
.file
);
1407 out_pos
= out
.file
->f_pos
;
1408 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1413 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1415 if (unlikely(pos
+ count
> max
)) {
1416 retval
= -EOVERFLOW
;
1425 * We need to debate whether we can enable this or not. The
1426 * man page documents EAGAIN return for the output at least,
1427 * and the application is arguably buggy if it doesn't expect
1428 * EAGAIN on a non-blocking file descriptor.
1430 if (in
.file
->f_flags
& O_NONBLOCK
)
1431 fl
= SPLICE_F_NONBLOCK
;
1433 file_start_write(out
.file
);
1434 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1435 file_end_write(out
.file
);
1438 add_rchar(current
, retval
);
1439 add_wchar(current
, retval
);
1440 fsnotify_access(in
.file
);
1441 fsnotify_modify(out
.file
);
1442 out
.file
->f_pos
= out_pos
;
1446 in
.file
->f_pos
= pos
;
1452 retval
= -EOVERFLOW
;
1462 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1469 if (unlikely(get_user(off
, offset
)))
1472 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1473 if (unlikely(put_user(pos
, offset
)))
1478 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1481 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1487 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1489 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1490 if (unlikely(put_user(pos
, offset
)))
1495 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1498 #ifdef CONFIG_COMPAT
1499 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1500 compat_off_t __user
*, offset
, compat_size_t
, count
)
1507 if (unlikely(get_user(off
, offset
)))
1510 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1511 if (unlikely(put_user(pos
, offset
)))
1516 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1519 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1520 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1526 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1528 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1529 if (unlikely(put_user(pos
, offset
)))
1534 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1539 * copy_file_range() differs from regular file read and write in that it
1540 * specifically allows return partial success. When it does so is up to
1541 * the copy_file_range method.
1543 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1544 struct file
*file_out
, loff_t pos_out
,
1545 size_t len
, unsigned int flags
)
1547 struct inode
*inode_in
= file_inode(file_in
);
1548 struct inode
*inode_out
= file_inode(file_out
);
1554 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1556 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1559 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1563 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1567 if (!(file_in
->f_mode
& FMODE_READ
) ||
1568 !(file_out
->f_mode
& FMODE_WRITE
) ||
1569 (file_out
->f_flags
& O_APPEND
))
1572 /* this could be relaxed once a method supports cross-fs copies */
1573 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1579 file_start_write(file_out
);
1582 * Try cloning first, this is supported by more file systems, and
1583 * more efficient if both clone and copy are supported (e.g. NFS).
1585 if (file_in
->f_op
->clone_file_range
) {
1586 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1587 file_out
, pos_out
, len
);
1594 if (file_out
->f_op
->copy_file_range
) {
1595 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1596 pos_out
, len
, flags
);
1597 if (ret
!= -EOPNOTSUPP
)
1601 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1602 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1606 fsnotify_access(file_in
);
1607 add_rchar(current
, ret
);
1608 fsnotify_modify(file_out
);
1609 add_wchar(current
, ret
);
1615 file_end_write(file_out
);
1619 EXPORT_SYMBOL(vfs_copy_file_range
);
1621 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1622 int, fd_out
, loff_t __user
*, off_out
,
1623 size_t, len
, unsigned int, flags
)
1629 ssize_t ret
= -EBADF
;
1631 f_in
= fdget(fd_in
);
1635 f_out
= fdget(fd_out
);
1641 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1644 pos_in
= f_in
.file
->f_pos
;
1648 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1651 pos_out
= f_out
.file
->f_pos
;
1654 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1661 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1664 f_in
.file
->f_pos
= pos_in
;
1668 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1671 f_out
.file
->f_pos
= pos_out
;
1683 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1685 struct inode
*inode
= file_inode(file
);
1687 if (unlikely(pos
< 0))
1690 if (unlikely((loff_t
) (pos
+ len
) < 0))
1693 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1694 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1697 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1698 write
? F_WRLCK
: F_RDLCK
);
1703 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1707 * Check that the two inodes are eligible for cloning, the ranges make
1708 * sense, and then flush all dirty data. Caller must ensure that the
1709 * inodes have been locked against any other modifications.
1711 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1712 * the usual negative error code.
1714 int vfs_clone_file_prep_inodes(struct inode
*inode_in
, loff_t pos_in
,
1715 struct inode
*inode_out
, loff_t pos_out
,
1716 u64
*len
, bool is_dedupe
)
1718 loff_t bs
= inode_out
->i_sb
->s_blocksize
;
1721 bool same_inode
= (inode_in
== inode_out
);
1724 /* Don't touch certain kinds of inodes */
1725 if (IS_IMMUTABLE(inode_out
))
1728 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
1731 /* Don't reflink dirs, pipes, sockets... */
1732 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1734 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1737 /* Are we going all the way to the end? */
1738 isize
= i_size_read(inode_in
);
1742 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1744 if (is_dedupe
|| pos_in
== isize
)
1748 *len
= isize
- pos_in
;
1751 /* Ensure offsets don't wrap and the input is inside i_size */
1752 if (pos_in
+ *len
< pos_in
|| pos_out
+ *len
< pos_out
||
1753 pos_in
+ *len
> isize
)
1756 /* Don't allow dedupe past EOF in the dest file */
1760 disize
= i_size_read(inode_out
);
1761 if (pos_out
>= disize
|| pos_out
+ *len
> disize
)
1765 /* If we're linking to EOF, continue to the block boundary. */
1766 if (pos_in
+ *len
== isize
)
1767 blen
= ALIGN(isize
, bs
) - pos_in
;
1771 /* Only reflink if we're aligned to block boundaries */
1772 if (!IS_ALIGNED(pos_in
, bs
) || !IS_ALIGNED(pos_in
+ blen
, bs
) ||
1773 !IS_ALIGNED(pos_out
, bs
) || !IS_ALIGNED(pos_out
+ blen
, bs
))
1776 /* Don't allow overlapped reflink within the same file */
1778 if (pos_out
+ blen
> pos_in
&& pos_out
< pos_in
+ blen
)
1782 /* Wait for the completion of any pending IOs on both files */
1783 inode_dio_wait(inode_in
);
1785 inode_dio_wait(inode_out
);
1787 ret
= filemap_write_and_wait_range(inode_in
->i_mapping
,
1788 pos_in
, pos_in
+ *len
- 1);
1792 ret
= filemap_write_and_wait_range(inode_out
->i_mapping
,
1793 pos_out
, pos_out
+ *len
- 1);
1798 * Check that the extents are the same.
1801 bool is_same
= false;
1803 ret
= vfs_dedupe_file_range_compare(inode_in
, pos_in
,
1804 inode_out
, pos_out
, *len
, &is_same
);
1813 EXPORT_SYMBOL(vfs_clone_file_prep_inodes
);
1815 int do_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1816 struct file
*file_out
, loff_t pos_out
, u64 len
)
1818 struct inode
*inode_in
= file_inode(file_in
);
1819 struct inode
*inode_out
= file_inode(file_out
);
1822 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1824 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1828 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1829 * the same mount. Practically, they only need to be on the same file
1832 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1835 if (!(file_in
->f_mode
& FMODE_READ
) ||
1836 !(file_out
->f_mode
& FMODE_WRITE
) ||
1837 (file_out
->f_flags
& O_APPEND
))
1840 if (!file_in
->f_op
->clone_file_range
)
1843 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1847 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1851 if (pos_in
+ len
> i_size_read(inode_in
))
1854 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1855 file_out
, pos_out
, len
);
1857 fsnotify_access(file_in
);
1858 fsnotify_modify(file_out
);
1863 EXPORT_SYMBOL(do_clone_file_range
);
1865 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1866 struct file
*file_out
, loff_t pos_out
, u64 len
)
1870 file_start_write(file_out
);
1871 ret
= do_clone_file_range(file_in
, pos_in
, file_out
, pos_out
, len
);
1872 file_end_write(file_out
);
1876 EXPORT_SYMBOL(vfs_clone_file_range
);
1879 * Read a page's worth of file data into the page cache. Return the page
1882 static struct page
*vfs_dedupe_get_page(struct inode
*inode
, loff_t offset
)
1884 struct address_space
*mapping
;
1888 n
= offset
>> PAGE_SHIFT
;
1889 mapping
= inode
->i_mapping
;
1890 page
= read_mapping_page(mapping
, n
, NULL
);
1893 if (!PageUptodate(page
)) {
1895 return ERR_PTR(-EIO
);
1902 * Compare extents of two files to see if they are the same.
1903 * Caller must have locked both inodes to prevent write races.
1905 int vfs_dedupe_file_range_compare(struct inode
*src
, loff_t srcoff
,
1906 struct inode
*dest
, loff_t destoff
,
1907 loff_t len
, bool *is_same
)
1913 struct page
*src_page
;
1914 struct page
*dest_page
;
1922 src_poff
= srcoff
& (PAGE_SIZE
- 1);
1923 dest_poff
= destoff
& (PAGE_SIZE
- 1);
1924 cmp_len
= min(PAGE_SIZE
- src_poff
,
1925 PAGE_SIZE
- dest_poff
);
1926 cmp_len
= min(cmp_len
, len
);
1930 src_page
= vfs_dedupe_get_page(src
, srcoff
);
1931 if (IS_ERR(src_page
)) {
1932 error
= PTR_ERR(src_page
);
1935 dest_page
= vfs_dedupe_get_page(dest
, destoff
);
1936 if (IS_ERR(dest_page
)) {
1937 error
= PTR_ERR(dest_page
);
1938 unlock_page(src_page
);
1942 src_addr
= kmap_atomic(src_page
);
1943 dest_addr
= kmap_atomic(dest_page
);
1945 flush_dcache_page(src_page
);
1946 flush_dcache_page(dest_page
);
1948 if (memcmp(src_addr
+ src_poff
, dest_addr
+ dest_poff
, cmp_len
))
1951 kunmap_atomic(dest_addr
);
1952 kunmap_atomic(src_addr
);
1953 unlock_page(dest_page
);
1954 unlock_page(src_page
);
1955 put_page(dest_page
);
1972 EXPORT_SYMBOL(vfs_dedupe_file_range_compare
);
1974 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1976 struct file_dedupe_range_info
*info
;
1977 struct inode
*src
= file_inode(file
);
1982 bool is_admin
= capable(CAP_SYS_ADMIN
);
1983 u16 count
= same
->dest_count
;
1984 struct file
*dst_file
;
1988 if (!(file
->f_mode
& FMODE_READ
))
1991 if (same
->reserved1
|| same
->reserved2
)
1994 off
= same
->src_offset
;
1995 len
= same
->src_length
;
1998 if (S_ISDIR(src
->i_mode
))
2002 if (!S_ISREG(src
->i_mode
))
2005 ret
= clone_verify_area(file
, off
, len
, false);
2010 if (off
+ len
> i_size_read(src
))
2013 /* pre-format output fields to sane values */
2014 for (i
= 0; i
< count
; i
++) {
2015 same
->info
[i
].bytes_deduped
= 0ULL;
2016 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
2019 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
2021 struct fd dst_fd
= fdget(info
->dest_fd
);
2023 dst_file
= dst_fd
.file
;
2025 info
->status
= -EBADF
;
2028 dst
= file_inode(dst_file
);
2030 ret
= mnt_want_write_file(dst_file
);
2036 dst_off
= info
->dest_offset
;
2037 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
2044 if (info
->reserved
) {
2045 info
->status
= -EINVAL
;
2046 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
2047 info
->status
= -EINVAL
;
2048 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
2049 info
->status
= -EXDEV
;
2050 } else if (S_ISDIR(dst
->i_mode
)) {
2051 info
->status
= -EISDIR
;
2052 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
2053 info
->status
= -EINVAL
;
2055 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
2058 if (deduped
== -EBADE
)
2059 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
2060 else if (deduped
< 0)
2061 info
->status
= deduped
;
2063 info
->bytes_deduped
+= deduped
;
2067 mnt_drop_write_file(dst_file
);
2071 if (fatal_signal_pending(current
))
2078 EXPORT_SYMBOL(vfs_dedupe_file_range
);