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 off_t
ksys_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 */
322 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
324 return ksys_lseek(fd
, offset
, whence
);
328 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
330 return ksys_lseek(fd
, offset
, whence
);
334 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
335 defined(__ARCH_WANT_SYS_LLSEEK)
336 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
337 unsigned long, offset_low
, loff_t __user
*, result
,
338 unsigned int, whence
)
341 struct fd f
= fdget_pos(fd
);
348 if (whence
> SEEK_MAX
)
351 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
354 retval
= (int)offset
;
357 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
366 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
369 int retval
= -EINVAL
;
371 inode
= file_inode(file
);
372 if (unlikely((ssize_t
) count
< 0))
376 * ranged mandatory locking does not apply to streams - it makes sense
377 * only for files where position has a meaning.
382 if (unlikely(pos
< 0)) {
383 if (!unsigned_offsets(file
))
385 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
387 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
388 if (!unsigned_offsets(file
))
392 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
393 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
394 read_write
== READ
? F_RDLCK
: F_WRLCK
);
400 return security_file_permission(file
,
401 read_write
== READ
? MAY_READ
: MAY_WRITE
);
404 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
406 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
408 struct iov_iter iter
;
411 init_sync_kiocb(&kiocb
, filp
);
412 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
413 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
415 ret
= call_read_iter(filp
, &kiocb
, &iter
);
416 BUG_ON(ret
== -EIOCBQUEUED
);
418 *ppos
= kiocb
.ki_pos
;
422 ssize_t
__kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
424 mm_segment_t old_fs
= get_fs();
427 if (WARN_ON_ONCE(!(file
->f_mode
& FMODE_READ
)))
429 if (!(file
->f_mode
& FMODE_CAN_READ
))
432 if (count
> MAX_RW_COUNT
)
433 count
= MAX_RW_COUNT
;
435 if (file
->f_op
->read
)
436 ret
= file
->f_op
->read(file
, (void __user
*)buf
, count
, pos
);
437 else if (file
->f_op
->read_iter
)
438 ret
= new_sync_read(file
, (void __user
*)buf
, count
, pos
);
443 fsnotify_access(file
);
444 add_rchar(current
, ret
);
450 ssize_t
kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
454 ret
= rw_verify_area(READ
, file
, pos
, count
);
457 return __kernel_read(file
, buf
, count
, pos
);
459 EXPORT_SYMBOL(kernel_read
);
461 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
465 if (!(file
->f_mode
& FMODE_READ
))
467 if (!(file
->f_mode
& FMODE_CAN_READ
))
469 if (unlikely(!access_ok(buf
, count
)))
472 ret
= rw_verify_area(READ
, file
, pos
, count
);
475 if (count
> MAX_RW_COUNT
)
476 count
= MAX_RW_COUNT
;
478 if (file
->f_op
->read
)
479 ret
= file
->f_op
->read(file
, buf
, count
, pos
);
480 else if (file
->f_op
->read_iter
)
481 ret
= new_sync_read(file
, buf
, count
, pos
);
485 fsnotify_access(file
);
486 add_rchar(current
, ret
);
492 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
494 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
496 struct iov_iter iter
;
499 init_sync_kiocb(&kiocb
, filp
);
500 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
501 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
503 ret
= call_write_iter(filp
, &kiocb
, &iter
);
504 BUG_ON(ret
== -EIOCBQUEUED
);
506 *ppos
= kiocb
.ki_pos
;
510 /* caller is responsible for file_start_write/file_end_write */
511 ssize_t
__kernel_write(struct file
*file
, const void *buf
, size_t count
, loff_t
*pos
)
514 const char __user
*p
;
517 if (WARN_ON_ONCE(!(file
->f_mode
& FMODE_WRITE
)))
519 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
524 p
= (__force
const char __user
*)buf
;
525 if (count
> MAX_RW_COUNT
)
526 count
= MAX_RW_COUNT
;
527 if (file
->f_op
->write
)
528 ret
= file
->f_op
->write(file
, p
, count
, pos
);
529 else if (file
->f_op
->write_iter
)
530 ret
= new_sync_write(file
, p
, count
, pos
);
535 fsnotify_modify(file
);
536 add_wchar(current
, ret
);
542 ssize_t
kernel_write(struct file
*file
, const void *buf
, size_t count
,
547 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
551 file_start_write(file
);
552 ret
= __kernel_write(file
, buf
, count
, pos
);
553 file_end_write(file
);
556 EXPORT_SYMBOL(kernel_write
);
558 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
562 if (!(file
->f_mode
& FMODE_WRITE
))
564 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
566 if (unlikely(!access_ok(buf
, count
)))
569 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
572 if (count
> MAX_RW_COUNT
)
573 count
= MAX_RW_COUNT
;
574 file_start_write(file
);
575 if (file
->f_op
->write
)
576 ret
= file
->f_op
->write(file
, buf
, count
, pos
);
577 else if (file
->f_op
->write_iter
)
578 ret
= new_sync_write(file
, buf
, count
, pos
);
582 fsnotify_modify(file
);
583 add_wchar(current
, ret
);
586 file_end_write(file
);
590 /* file_ppos returns &file->f_pos or NULL if file is stream */
591 static inline loff_t
*file_ppos(struct file
*file
)
593 return file
->f_mode
& FMODE_STREAM
? NULL
: &file
->f_pos
;
596 ssize_t
ksys_read(unsigned int fd
, char __user
*buf
, size_t count
)
598 struct fd f
= fdget_pos(fd
);
599 ssize_t ret
= -EBADF
;
602 loff_t pos
, *ppos
= file_ppos(f
.file
);
607 ret
= vfs_read(f
.file
, buf
, count
, ppos
);
608 if (ret
>= 0 && ppos
)
615 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
617 return ksys_read(fd
, buf
, count
);
620 ssize_t
ksys_write(unsigned int fd
, const char __user
*buf
, size_t count
)
622 struct fd f
= fdget_pos(fd
);
623 ssize_t ret
= -EBADF
;
626 loff_t pos
, *ppos
= file_ppos(f
.file
);
631 ret
= vfs_write(f
.file
, buf
, count
, ppos
);
632 if (ret
>= 0 && ppos
)
640 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
643 return ksys_write(fd
, buf
, count
);
646 ssize_t
ksys_pread64(unsigned int fd
, char __user
*buf
, size_t count
,
650 ssize_t ret
= -EBADF
;
658 if (f
.file
->f_mode
& FMODE_PREAD
)
659 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
666 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
667 size_t, count
, loff_t
, pos
)
669 return ksys_pread64(fd
, buf
, count
, pos
);
672 ssize_t
ksys_pwrite64(unsigned int fd
, const char __user
*buf
,
673 size_t count
, loff_t pos
)
676 ssize_t ret
= -EBADF
;
684 if (f
.file
->f_mode
& FMODE_PWRITE
)
685 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
692 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
693 size_t, count
, loff_t
, pos
)
695 return ksys_pwrite64(fd
, buf
, count
, pos
);
698 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
699 loff_t
*ppos
, int type
, rwf_t flags
)
704 init_sync_kiocb(&kiocb
, filp
);
705 ret
= kiocb_set_rw_flags(&kiocb
, flags
);
708 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
711 ret
= call_read_iter(filp
, &kiocb
, iter
);
713 ret
= call_write_iter(filp
, &kiocb
, iter
);
714 BUG_ON(ret
== -EIOCBQUEUED
);
716 *ppos
= kiocb
.ki_pos
;
720 /* Do it by hand, with file-ops */
721 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
722 loff_t
*ppos
, int type
, rwf_t flags
)
726 if (flags
& ~RWF_HIPRI
)
729 while (iov_iter_count(iter
)) {
730 struct iovec iovec
= iov_iter_iovec(iter
);
734 nr
= filp
->f_op
->read(filp
, iovec
.iov_base
,
735 iovec
.iov_len
, ppos
);
737 nr
= filp
->f_op
->write(filp
, iovec
.iov_base
,
738 iovec
.iov_len
, ppos
);
747 if (nr
!= iovec
.iov_len
)
749 iov_iter_advance(iter
, nr
);
756 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
757 * into the kernel and check that it is valid.
759 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
760 * @uvector: Pointer to the userspace array.
761 * @nr_segs: Number of elements in userspace array.
762 * @fast_segs: Number of elements in @fast_pointer.
763 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
764 * @ret_pointer: (output parameter) Pointer to a variable that will point to
765 * either @fast_pointer, a newly allocated kernel array, or NULL,
766 * depending on which array was used.
768 * This function copies an array of &struct iovec of @nr_segs from
769 * userspace into the kernel and checks that each element is valid (e.g.
770 * it does not point to a kernel address or cause overflow by being too
773 * As an optimization, the caller may provide a pointer to a small
774 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
775 * (the size of this array, or 0 if unused, should be given in @fast_segs).
777 * @ret_pointer will always point to the array that was used, so the
778 * caller must take care not to call kfree() on it e.g. in case the
779 * @fast_pointer array was used and it was allocated on the stack.
781 * Return: The total number of bytes covered by the iovec array on success
782 * or a negative error code on error.
784 ssize_t
rw_copy_check_uvector(int type
, const struct iovec __user
* uvector
,
785 unsigned long nr_segs
, unsigned long fast_segs
,
786 struct iovec
*fast_pointer
,
787 struct iovec
**ret_pointer
)
791 struct iovec
*iov
= fast_pointer
;
794 * SuS says "The readv() function *may* fail if the iovcnt argument
795 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
796 * traditionally returned zero for zero segments, so...
804 * First get the "struct iovec" from user memory and
805 * verify all the pointers
807 if (nr_segs
> UIO_MAXIOV
) {
811 if (nr_segs
> fast_segs
) {
812 iov
= kmalloc_array(nr_segs
, sizeof(struct iovec
), GFP_KERNEL
);
818 if (copy_from_user(iov
, uvector
, nr_segs
*sizeof(*uvector
))) {
824 * According to the Single Unix Specification we should return EINVAL
825 * if an element length is < 0 when cast to ssize_t or if the
826 * total length would overflow the ssize_t return value of the
829 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
833 for (seg
= 0; seg
< nr_segs
; seg
++) {
834 void __user
*buf
= iov
[seg
].iov_base
;
835 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
837 /* see if we we're about to use an invalid len or if
838 * it's about to overflow ssize_t */
844 && unlikely(!access_ok(buf
, len
))) {
848 if (len
> MAX_RW_COUNT
- ret
) {
849 len
= MAX_RW_COUNT
- ret
;
850 iov
[seg
].iov_len
= len
;
860 ssize_t
compat_rw_copy_check_uvector(int type
,
861 const struct compat_iovec __user
*uvector
, unsigned long nr_segs
,
862 unsigned long fast_segs
, struct iovec
*fast_pointer
,
863 struct iovec
**ret_pointer
)
865 compat_ssize_t tot_len
;
866 struct iovec
*iov
= *ret_pointer
= fast_pointer
;
871 * SuS says "The readv() function *may* fail if the iovcnt argument
872 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
873 * traditionally returned zero for zero segments, so...
879 if (nr_segs
> UIO_MAXIOV
)
881 if (nr_segs
> fast_segs
) {
883 iov
= kmalloc_array(nr_segs
, sizeof(struct iovec
), GFP_KERNEL
);
890 if (!access_ok(uvector
, nr_segs
*sizeof(*uvector
)))
894 * Single unix specification:
895 * We should -EINVAL if an element length is not >= 0 and fitting an
898 * In Linux, the total length is limited to MAX_RW_COUNT, there is
899 * no overflow possibility.
903 for (seg
= 0; seg
< nr_segs
; seg
++) {
907 if (__get_user(len
, &uvector
->iov_len
) ||
908 __get_user(buf
, &uvector
->iov_base
)) {
912 if (len
< 0) /* size_t not fitting in compat_ssize_t .. */
915 !access_ok(compat_ptr(buf
), len
)) {
919 if (len
> MAX_RW_COUNT
- tot_len
)
920 len
= MAX_RW_COUNT
- tot_len
;
922 iov
->iov_base
= compat_ptr(buf
);
923 iov
->iov_len
= (compat_size_t
) len
;
934 static ssize_t
do_iter_read(struct file
*file
, struct iov_iter
*iter
,
935 loff_t
*pos
, rwf_t flags
)
940 if (!(file
->f_mode
& FMODE_READ
))
942 if (!(file
->f_mode
& FMODE_CAN_READ
))
945 tot_len
= iov_iter_count(iter
);
948 ret
= rw_verify_area(READ
, file
, pos
, tot_len
);
952 if (file
->f_op
->read_iter
)
953 ret
= do_iter_readv_writev(file
, iter
, pos
, READ
, flags
);
955 ret
= do_loop_readv_writev(file
, iter
, pos
, READ
, flags
);
958 fsnotify_access(file
);
962 ssize_t
vfs_iocb_iter_read(struct file
*file
, struct kiocb
*iocb
,
963 struct iov_iter
*iter
)
968 if (!file
->f_op
->read_iter
)
970 if (!(file
->f_mode
& FMODE_READ
))
972 if (!(file
->f_mode
& FMODE_CAN_READ
))
975 tot_len
= iov_iter_count(iter
);
978 ret
= rw_verify_area(READ
, file
, &iocb
->ki_pos
, tot_len
);
982 ret
= call_read_iter(file
, iocb
, iter
);
985 fsnotify_access(file
);
988 EXPORT_SYMBOL(vfs_iocb_iter_read
);
990 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
993 if (!file
->f_op
->read_iter
)
995 return do_iter_read(file
, iter
, ppos
, flags
);
997 EXPORT_SYMBOL(vfs_iter_read
);
999 static ssize_t
do_iter_write(struct file
*file
, struct iov_iter
*iter
,
1000 loff_t
*pos
, rwf_t flags
)
1005 if (!(file
->f_mode
& FMODE_WRITE
))
1007 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1010 tot_len
= iov_iter_count(iter
);
1013 ret
= rw_verify_area(WRITE
, file
, pos
, tot_len
);
1017 if (file
->f_op
->write_iter
)
1018 ret
= do_iter_readv_writev(file
, iter
, pos
, WRITE
, flags
);
1020 ret
= do_loop_readv_writev(file
, iter
, pos
, WRITE
, flags
);
1022 fsnotify_modify(file
);
1026 ssize_t
vfs_iocb_iter_write(struct file
*file
, struct kiocb
*iocb
,
1027 struct iov_iter
*iter
)
1032 if (!file
->f_op
->write_iter
)
1034 if (!(file
->f_mode
& FMODE_WRITE
))
1036 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1039 tot_len
= iov_iter_count(iter
);
1042 ret
= rw_verify_area(WRITE
, file
, &iocb
->ki_pos
, tot_len
);
1046 ret
= call_write_iter(file
, iocb
, iter
);
1048 fsnotify_modify(file
);
1052 EXPORT_SYMBOL(vfs_iocb_iter_write
);
1054 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
1057 if (!file
->f_op
->write_iter
)
1059 return do_iter_write(file
, iter
, ppos
, flags
);
1061 EXPORT_SYMBOL(vfs_iter_write
);
1063 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
1064 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
1066 struct iovec iovstack
[UIO_FASTIOV
];
1067 struct iovec
*iov
= iovstack
;
1068 struct iov_iter iter
;
1071 ret
= import_iovec(READ
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
1073 ret
= do_iter_read(file
, &iter
, pos
, flags
);
1080 static ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
1081 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
1083 struct iovec iovstack
[UIO_FASTIOV
];
1084 struct iovec
*iov
= iovstack
;
1085 struct iov_iter iter
;
1088 ret
= import_iovec(WRITE
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
1090 file_start_write(file
);
1091 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1092 file_end_write(file
);
1098 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
1099 unsigned long vlen
, rwf_t flags
)
1101 struct fd f
= fdget_pos(fd
);
1102 ssize_t ret
= -EBADF
;
1105 loff_t pos
, *ppos
= file_ppos(f
.file
);
1110 ret
= vfs_readv(f
.file
, vec
, vlen
, ppos
, flags
);
1111 if (ret
>= 0 && ppos
)
1112 f
.file
->f_pos
= pos
;
1117 add_rchar(current
, ret
);
1122 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
1123 unsigned long vlen
, rwf_t flags
)
1125 struct fd f
= fdget_pos(fd
);
1126 ssize_t ret
= -EBADF
;
1129 loff_t pos
, *ppos
= file_ppos(f
.file
);
1134 ret
= vfs_writev(f
.file
, vec
, vlen
, ppos
, flags
);
1135 if (ret
>= 0 && ppos
)
1136 f
.file
->f_pos
= pos
;
1141 add_wchar(current
, ret
);
1146 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
1148 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1149 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
1152 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
1153 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1156 ssize_t ret
= -EBADF
;
1164 if (f
.file
->f_mode
& FMODE_PREAD
)
1165 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1170 add_rchar(current
, ret
);
1175 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1176 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1179 ssize_t ret
= -EBADF
;
1187 if (f
.file
->f_mode
& FMODE_PWRITE
)
1188 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1193 add_wchar(current
, ret
);
1198 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1199 unsigned long, vlen
)
1201 return do_readv(fd
, vec
, vlen
, 0);
1204 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1205 unsigned long, vlen
)
1207 return do_writev(fd
, vec
, vlen
, 0);
1210 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1211 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1213 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1215 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1218 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1219 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1222 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1225 return do_readv(fd
, vec
, vlen
, flags
);
1227 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1230 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1231 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1233 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1235 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1238 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1239 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1242 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1245 return do_writev(fd
, vec
, vlen
, flags
);
1247 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1250 #ifdef CONFIG_COMPAT
1251 static size_t compat_readv(struct file
*file
,
1252 const struct compat_iovec __user
*vec
,
1253 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
1255 struct iovec iovstack
[UIO_FASTIOV
];
1256 struct iovec
*iov
= iovstack
;
1257 struct iov_iter iter
;
1260 ret
= compat_import_iovec(READ
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1262 ret
= do_iter_read(file
, &iter
, pos
, flags
);
1266 add_rchar(current
, ret
);
1271 static size_t do_compat_readv(compat_ulong_t fd
,
1272 const struct compat_iovec __user
*vec
,
1273 compat_ulong_t vlen
, rwf_t flags
)
1275 struct fd f
= fdget_pos(fd
);
1281 pos
= f
.file
->f_pos
;
1282 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1284 f
.file
->f_pos
= pos
;
1290 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1291 const struct compat_iovec __user
*,vec
,
1292 compat_ulong_t
, vlen
)
1294 return do_compat_readv(fd
, vec
, vlen
, 0);
1297 static long do_compat_preadv64(unsigned long fd
,
1298 const struct compat_iovec __user
*vec
,
1299 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1310 if (f
.file
->f_mode
& FMODE_PREAD
)
1311 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1316 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1317 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1318 const struct compat_iovec __user
*,vec
,
1319 unsigned long, vlen
, loff_t
, pos
)
1321 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1325 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1326 const struct compat_iovec __user
*,vec
,
1327 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1329 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1331 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1334 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1335 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1336 const struct compat_iovec __user
*,vec
,
1337 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1340 return do_compat_readv(fd
, vec
, vlen
, flags
);
1342 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1346 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1347 const struct compat_iovec __user
*,vec
,
1348 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1351 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1354 return do_compat_readv(fd
, vec
, vlen
, flags
);
1356 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1359 static size_t compat_writev(struct file
*file
,
1360 const struct compat_iovec __user
*vec
,
1361 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
1363 struct iovec iovstack
[UIO_FASTIOV
];
1364 struct iovec
*iov
= iovstack
;
1365 struct iov_iter iter
;
1368 ret
= compat_import_iovec(WRITE
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1370 file_start_write(file
);
1371 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1372 file_end_write(file
);
1376 add_wchar(current
, ret
);
1381 static size_t do_compat_writev(compat_ulong_t fd
,
1382 const struct compat_iovec __user
* vec
,
1383 compat_ulong_t vlen
, rwf_t flags
)
1385 struct fd f
= fdget_pos(fd
);
1391 pos
= f
.file
->f_pos
;
1392 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1394 f
.file
->f_pos
= pos
;
1399 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1400 const struct compat_iovec __user
*, vec
,
1401 compat_ulong_t
, vlen
)
1403 return do_compat_writev(fd
, vec
, vlen
, 0);
1406 static long do_compat_pwritev64(unsigned long fd
,
1407 const struct compat_iovec __user
*vec
,
1408 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1419 if (f
.file
->f_mode
& FMODE_PWRITE
)
1420 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1425 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1426 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1427 const struct compat_iovec __user
*,vec
,
1428 unsigned long, vlen
, loff_t
, pos
)
1430 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1434 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1435 const struct compat_iovec __user
*,vec
,
1436 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1438 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1440 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1443 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1444 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1445 const struct compat_iovec __user
*,vec
,
1446 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1449 return do_compat_writev(fd
, vec
, vlen
, flags
);
1451 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1455 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1456 const struct compat_iovec __user
*,vec
,
1457 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, rwf_t
, flags
)
1459 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1462 return do_compat_writev(fd
, vec
, vlen
, flags
);
1464 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1469 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1470 size_t count
, loff_t max
)
1473 struct inode
*in_inode
, *out_inode
;
1480 * Get input file, and verify that it is ok..
1486 if (!(in
.file
->f_mode
& FMODE_READ
))
1490 pos
= in
.file
->f_pos
;
1493 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1496 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1499 if (count
> MAX_RW_COUNT
)
1500 count
= MAX_RW_COUNT
;
1503 * Get output file, and verify that it is ok..
1506 out
= fdget(out_fd
);
1509 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1511 in_inode
= file_inode(in
.file
);
1512 out_inode
= file_inode(out
.file
);
1513 out_pos
= out
.file
->f_pos
;
1514 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1519 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1521 if (unlikely(pos
+ count
> max
)) {
1522 retval
= -EOVERFLOW
;
1531 * We need to debate whether we can enable this or not. The
1532 * man page documents EAGAIN return for the output at least,
1533 * and the application is arguably buggy if it doesn't expect
1534 * EAGAIN on a non-blocking file descriptor.
1536 if (in
.file
->f_flags
& O_NONBLOCK
)
1537 fl
= SPLICE_F_NONBLOCK
;
1539 file_start_write(out
.file
);
1540 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1541 file_end_write(out
.file
);
1544 add_rchar(current
, retval
);
1545 add_wchar(current
, retval
);
1546 fsnotify_access(in
.file
);
1547 fsnotify_modify(out
.file
);
1548 out
.file
->f_pos
= out_pos
;
1552 in
.file
->f_pos
= pos
;
1558 retval
= -EOVERFLOW
;
1568 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1575 if (unlikely(get_user(off
, offset
)))
1578 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1579 if (unlikely(put_user(pos
, offset
)))
1584 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1587 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1593 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1595 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1596 if (unlikely(put_user(pos
, offset
)))
1601 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1604 #ifdef CONFIG_COMPAT
1605 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1606 compat_off_t __user
*, offset
, compat_size_t
, count
)
1613 if (unlikely(get_user(off
, offset
)))
1616 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1617 if (unlikely(put_user(pos
, offset
)))
1622 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1625 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1626 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1632 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1634 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1635 if (unlikely(put_user(pos
, offset
)))
1640 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1645 * generic_copy_file_range - copy data between two files
1646 * @file_in: file structure to read from
1647 * @pos_in: file offset to read from
1648 * @file_out: file structure to write data to
1649 * @pos_out: file offset to write data to
1650 * @len: amount of data to copy
1651 * @flags: copy flags
1653 * This is a generic filesystem helper to copy data from one file to another.
1654 * It has no constraints on the source or destination file owners - the files
1655 * can belong to different superblocks and different filesystem types. Short
1656 * copies are allowed.
1658 * This should be called from the @file_out filesystem, as per the
1659 * ->copy_file_range() method.
1661 * Returns the number of bytes copied or a negative error indicating the
1665 ssize_t
generic_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1666 struct file
*file_out
, loff_t pos_out
,
1667 size_t len
, unsigned int flags
)
1669 return do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1670 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1672 EXPORT_SYMBOL(generic_copy_file_range
);
1674 static ssize_t
do_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1675 struct file
*file_out
, loff_t pos_out
,
1676 size_t len
, unsigned int flags
)
1679 * Although we now allow filesystems to handle cross sb copy, passing
1680 * a file of the wrong filesystem type to filesystem driver can result
1681 * in an attempt to dereference the wrong type of ->private_data, so
1682 * avoid doing that until we really have a good reason. NFS defines
1683 * several different file_system_type structures, but they all end up
1684 * using the same ->copy_file_range() function pointer.
1686 if (file_out
->f_op
->copy_file_range
&&
1687 file_out
->f_op
->copy_file_range
== file_in
->f_op
->copy_file_range
)
1688 return file_out
->f_op
->copy_file_range(file_in
, pos_in
,
1692 return generic_copy_file_range(file_in
, pos_in
, file_out
, pos_out
, len
,
1697 * copy_file_range() differs from regular file read and write in that it
1698 * specifically allows return partial success. When it does so is up to
1699 * the copy_file_range method.
1701 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1702 struct file
*file_out
, loff_t pos_out
,
1703 size_t len
, unsigned int flags
)
1710 ret
= generic_copy_file_checks(file_in
, pos_in
, file_out
, pos_out
, &len
,
1715 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1719 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1726 file_start_write(file_out
);
1729 * Try cloning first, this is supported by more file systems, and
1730 * more efficient if both clone and copy are supported (e.g. NFS).
1732 if (file_in
->f_op
->remap_file_range
&&
1733 file_inode(file_in
)->i_sb
== file_inode(file_out
)->i_sb
) {
1736 cloned
= file_in
->f_op
->remap_file_range(file_in
, pos_in
,
1738 min_t(loff_t
, MAX_RW_COUNT
, len
),
1739 REMAP_FILE_CAN_SHORTEN
);
1746 ret
= do_copy_file_range(file_in
, pos_in
, file_out
, pos_out
, len
,
1748 WARN_ON_ONCE(ret
== -EOPNOTSUPP
);
1751 fsnotify_access(file_in
);
1752 add_rchar(current
, ret
);
1753 fsnotify_modify(file_out
);
1754 add_wchar(current
, ret
);
1760 file_end_write(file_out
);
1764 EXPORT_SYMBOL(vfs_copy_file_range
);
1766 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1767 int, fd_out
, loff_t __user
*, off_out
,
1768 size_t, len
, unsigned int, flags
)
1774 ssize_t ret
= -EBADF
;
1776 f_in
= fdget(fd_in
);
1780 f_out
= fdget(fd_out
);
1786 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1789 pos_in
= f_in
.file
->f_pos
;
1793 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1796 pos_out
= f_out
.file
->f_pos
;
1799 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1806 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1809 f_in
.file
->f_pos
= pos_in
;
1813 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1816 f_out
.file
->f_pos
= pos_out
;
1828 static int remap_verify_area(struct file
*file
, loff_t pos
, loff_t len
,
1831 struct inode
*inode
= file_inode(file
);
1833 if (unlikely(pos
< 0 || len
< 0))
1836 if (unlikely((loff_t
) (pos
+ len
) < 0))
1839 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1840 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1843 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1844 write
? F_WRLCK
: F_RDLCK
);
1849 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1852 * Ensure that we don't remap a partial EOF block in the middle of something
1853 * else. Assume that the offsets have already been checked for block
1856 * For clone we only link a partial EOF block above or at the destination file's
1857 * EOF. For deduplication we accept a partial EOF block only if it ends at the
1858 * destination file's EOF (can not link it into the middle of a file).
1860 * Shorten the request if possible.
1862 static int generic_remap_check_len(struct inode
*inode_in
,
1863 struct inode
*inode_out
,
1866 unsigned int remap_flags
)
1868 u64 blkmask
= i_blocksize(inode_in
) - 1;
1869 loff_t new_len
= *len
;
1871 if ((*len
& blkmask
) == 0)
1874 if (pos_out
+ *len
< i_size_read(inode_out
))
1875 new_len
&= ~blkmask
;
1877 if (new_len
== *len
)
1880 if (remap_flags
& REMAP_FILE_CAN_SHORTEN
) {
1885 return (remap_flags
& REMAP_FILE_DEDUP
) ? -EBADE
: -EINVAL
;
1888 /* Read a page's worth of file data into the page cache. */
1889 static struct page
*vfs_dedupe_get_page(struct inode
*inode
, loff_t offset
)
1893 page
= read_mapping_page(inode
->i_mapping
, offset
>> PAGE_SHIFT
, NULL
);
1896 if (!PageUptodate(page
)) {
1898 return ERR_PTR(-EIO
);
1904 * Lock two pages, ensuring that we lock in offset order if the pages are from
1907 static void vfs_lock_two_pages(struct page
*page1
, struct page
*page2
)
1909 /* Always lock in order of increasing index. */
1910 if (page1
->index
> page2
->index
)
1918 /* Unlock two pages, being careful not to unlock the same page twice. */
1919 static void vfs_unlock_two_pages(struct page
*page1
, struct page
*page2
)
1927 * Compare extents of two files to see if they are the same.
1928 * Caller must have locked both inodes to prevent write races.
1930 static int vfs_dedupe_file_range_compare(struct inode
*src
, loff_t srcoff
,
1931 struct inode
*dest
, loff_t destoff
,
1932 loff_t len
, bool *is_same
)
1938 struct page
*src_page
;
1939 struct page
*dest_page
;
1947 src_poff
= srcoff
& (PAGE_SIZE
- 1);
1948 dest_poff
= destoff
& (PAGE_SIZE
- 1);
1949 cmp_len
= min(PAGE_SIZE
- src_poff
,
1950 PAGE_SIZE
- dest_poff
);
1951 cmp_len
= min(cmp_len
, len
);
1955 src_page
= vfs_dedupe_get_page(src
, srcoff
);
1956 if (IS_ERR(src_page
)) {
1957 error
= PTR_ERR(src_page
);
1960 dest_page
= vfs_dedupe_get_page(dest
, destoff
);
1961 if (IS_ERR(dest_page
)) {
1962 error
= PTR_ERR(dest_page
);
1967 vfs_lock_two_pages(src_page
, dest_page
);
1970 * Now that we've locked both pages, make sure they're still
1971 * mapped to the file data we're interested in. If not,
1972 * someone is invalidating pages on us and we lose.
1974 if (!PageUptodate(src_page
) || !PageUptodate(dest_page
) ||
1975 src_page
->mapping
!= src
->i_mapping
||
1976 dest_page
->mapping
!= dest
->i_mapping
) {
1981 src_addr
= kmap_atomic(src_page
);
1982 dest_addr
= kmap_atomic(dest_page
);
1984 flush_dcache_page(src_page
);
1985 flush_dcache_page(dest_page
);
1987 if (memcmp(src_addr
+ src_poff
, dest_addr
+ dest_poff
, cmp_len
))
1990 kunmap_atomic(dest_addr
);
1991 kunmap_atomic(src_addr
);
1993 vfs_unlock_two_pages(src_page
, dest_page
);
1994 put_page(dest_page
);
2013 * Check that the two inodes are eligible for cloning, the ranges make
2014 * sense, and then flush all dirty data. Caller must ensure that the
2015 * inodes have been locked against any other modifications.
2017 * If there's an error, then the usual negative error code is returned.
2018 * Otherwise returns 0 with *len set to the request length.
2020 int generic_remap_file_range_prep(struct file
*file_in
, loff_t pos_in
,
2021 struct file
*file_out
, loff_t pos_out
,
2022 loff_t
*len
, unsigned int remap_flags
)
2024 struct inode
*inode_in
= file_inode(file_in
);
2025 struct inode
*inode_out
= file_inode(file_out
);
2026 bool same_inode
= (inode_in
== inode_out
);
2029 /* Don't touch certain kinds of inodes */
2030 if (IS_IMMUTABLE(inode_out
))
2033 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
2036 /* Don't reflink dirs, pipes, sockets... */
2037 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
2039 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
2042 /* Zero length dedupe exits immediately; reflink goes to EOF. */
2044 loff_t isize
= i_size_read(inode_in
);
2046 if ((remap_flags
& REMAP_FILE_DEDUP
) || pos_in
== isize
)
2050 *len
= isize
- pos_in
;
2055 /* Check that we don't violate system file offset limits. */
2056 ret
= generic_remap_checks(file_in
, pos_in
, file_out
, pos_out
, len
,
2061 /* Wait for the completion of any pending IOs on both files */
2062 inode_dio_wait(inode_in
);
2064 inode_dio_wait(inode_out
);
2066 ret
= filemap_write_and_wait_range(inode_in
->i_mapping
,
2067 pos_in
, pos_in
+ *len
- 1);
2071 ret
= filemap_write_and_wait_range(inode_out
->i_mapping
,
2072 pos_out
, pos_out
+ *len
- 1);
2077 * Check that the extents are the same.
2079 if (remap_flags
& REMAP_FILE_DEDUP
) {
2080 bool is_same
= false;
2082 ret
= vfs_dedupe_file_range_compare(inode_in
, pos_in
,
2083 inode_out
, pos_out
, *len
, &is_same
);
2090 ret
= generic_remap_check_len(inode_in
, inode_out
, pos_out
, len
,
2095 /* If can't alter the file contents, we're done. */
2096 if (!(remap_flags
& REMAP_FILE_DEDUP
))
2097 ret
= file_modified(file_out
);
2101 EXPORT_SYMBOL(generic_remap_file_range_prep
);
2103 loff_t
do_clone_file_range(struct file
*file_in
, loff_t pos_in
,
2104 struct file
*file_out
, loff_t pos_out
,
2105 loff_t len
, unsigned int remap_flags
)
2109 WARN_ON_ONCE(remap_flags
& REMAP_FILE_DEDUP
);
2112 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
2113 * the same mount. Practically, they only need to be on the same file
2116 if (file_inode(file_in
)->i_sb
!= file_inode(file_out
)->i_sb
)
2119 ret
= generic_file_rw_checks(file_in
, file_out
);
2123 if (!file_in
->f_op
->remap_file_range
)
2126 ret
= remap_verify_area(file_in
, pos_in
, len
, false);
2130 ret
= remap_verify_area(file_out
, pos_out
, len
, true);
2134 ret
= file_in
->f_op
->remap_file_range(file_in
, pos_in
,
2135 file_out
, pos_out
, len
, remap_flags
);
2139 fsnotify_access(file_in
);
2140 fsnotify_modify(file_out
);
2143 EXPORT_SYMBOL(do_clone_file_range
);
2145 loff_t
vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
2146 struct file
*file_out
, loff_t pos_out
,
2147 loff_t len
, unsigned int remap_flags
)
2151 file_start_write(file_out
);
2152 ret
= do_clone_file_range(file_in
, pos_in
, file_out
, pos_out
, len
,
2154 file_end_write(file_out
);
2158 EXPORT_SYMBOL(vfs_clone_file_range
);
2160 /* Check whether we are allowed to dedupe the destination file */
2161 static bool allow_file_dedupe(struct file
*file
)
2163 if (capable(CAP_SYS_ADMIN
))
2165 if (file
->f_mode
& FMODE_WRITE
)
2167 if (uid_eq(current_fsuid(), file_inode(file
)->i_uid
))
2169 if (!inode_permission(file_inode(file
), MAY_WRITE
))
2174 loff_t
vfs_dedupe_file_range_one(struct file
*src_file
, loff_t src_pos
,
2175 struct file
*dst_file
, loff_t dst_pos
,
2176 loff_t len
, unsigned int remap_flags
)
2180 WARN_ON_ONCE(remap_flags
& ~(REMAP_FILE_DEDUP
|
2181 REMAP_FILE_CAN_SHORTEN
));
2183 ret
= mnt_want_write_file(dst_file
);
2187 ret
= remap_verify_area(dst_file
, dst_pos
, len
, true);
2189 goto out_drop_write
;
2192 if (!allow_file_dedupe(dst_file
))
2193 goto out_drop_write
;
2196 if (src_file
->f_path
.mnt
!= dst_file
->f_path
.mnt
)
2197 goto out_drop_write
;
2200 if (S_ISDIR(file_inode(dst_file
)->i_mode
))
2201 goto out_drop_write
;
2204 if (!dst_file
->f_op
->remap_file_range
)
2205 goto out_drop_write
;
2209 goto out_drop_write
;
2212 ret
= dst_file
->f_op
->remap_file_range(src_file
, src_pos
, dst_file
,
2213 dst_pos
, len
, remap_flags
| REMAP_FILE_DEDUP
);
2215 mnt_drop_write_file(dst_file
);
2219 EXPORT_SYMBOL(vfs_dedupe_file_range_one
);
2221 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
2223 struct file_dedupe_range_info
*info
;
2224 struct inode
*src
= file_inode(file
);
2229 u16 count
= same
->dest_count
;
2232 if (!(file
->f_mode
& FMODE_READ
))
2235 if (same
->reserved1
|| same
->reserved2
)
2238 off
= same
->src_offset
;
2239 len
= same
->src_length
;
2241 if (S_ISDIR(src
->i_mode
))
2244 if (!S_ISREG(src
->i_mode
))
2247 if (!file
->f_op
->remap_file_range
)
2250 ret
= remap_verify_area(file
, off
, len
, false);
2255 if (off
+ len
> i_size_read(src
))
2258 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
2259 len
= min_t(u64
, len
, 1 << 30);
2261 /* pre-format output fields to sane values */
2262 for (i
= 0; i
< count
; i
++) {
2263 same
->info
[i
].bytes_deduped
= 0ULL;
2264 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
2267 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
2268 struct fd dst_fd
= fdget(info
->dest_fd
);
2269 struct file
*dst_file
= dst_fd
.file
;
2272 info
->status
= -EBADF
;
2276 if (info
->reserved
) {
2277 info
->status
= -EINVAL
;
2281 deduped
= vfs_dedupe_file_range_one(file
, off
, dst_file
,
2282 info
->dest_offset
, len
,
2283 REMAP_FILE_CAN_SHORTEN
);
2284 if (deduped
== -EBADE
)
2285 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
2286 else if (deduped
< 0)
2287 info
->status
= deduped
;
2289 info
->bytes_deduped
= len
;
2294 if (fatal_signal_pending(current
))
2299 EXPORT_SYMBOL(vfs_dedupe_file_range
);