2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/sched/xacct.h>
10 #include <linux/fcntl.h>
11 #include <linux/file.h>
12 #include <linux/uio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
20 #include <linux/mount.h>
24 #include <linux/uaccess.h>
25 #include <asm/unistd.h>
27 const struct file_operations generic_ro_fops
= {
28 .llseek
= generic_file_llseek
,
29 .read_iter
= generic_file_read_iter
,
30 .mmap
= generic_file_readonly_mmap
,
31 .splice_read
= generic_file_splice_read
,
34 EXPORT_SYMBOL(generic_ro_fops
);
36 static inline int unsigned_offsets(struct file
*file
)
38 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
42 * vfs_setpos - update the file offset for lseek
43 * @file: file structure in question
44 * @offset: file offset to seek to
45 * @maxsize: maximum file size
47 * This is a low-level filesystem helper for updating the file offset to
48 * the value specified by @offset if the given offset is valid and it is
49 * not equal to the current file offset.
51 * Return the specified offset on success and -EINVAL on invalid offset.
53 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
55 if (offset
< 0 && !unsigned_offsets(file
))
60 if (offset
!= file
->f_pos
) {
66 EXPORT_SYMBOL(vfs_setpos
);
69 * generic_file_llseek_size - generic llseek implementation for regular files
70 * @file: file structure to seek on
71 * @offset: file offset to seek to
72 * @whence: type of seek
73 * @size: max size of this file in file system
74 * @eof: offset used for SEEK_END position
76 * This is a variant of generic_file_llseek that allows passing in a custom
77 * maximum file size and a custom EOF position, for e.g. hashed directories
80 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
81 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
82 * read/writes behave like SEEK_SET against seeks.
85 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
86 loff_t maxsize
, loff_t eof
)
94 * Here we special-case the lseek(fd, 0, SEEK_CUR)
95 * position-querying operation. Avoid rewriting the "same"
96 * f_pos value back to the file because a concurrent read(),
97 * write() or lseek() might have altered it
102 * f_lock protects against read/modify/write race with other
103 * SEEK_CURs. Note that parallel writes and reads behave
106 spin_lock(&file
->f_lock
);
107 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
108 spin_unlock(&file
->f_lock
);
112 * In the generic case the entire file is data, so as long as
113 * offset isn't at the end of the file then the offset is data.
120 * There is a virtual hole at the end of the file, so as long as
121 * offset isn't i_size or larger, return i_size.
129 return vfs_setpos(file
, offset
, maxsize
);
131 EXPORT_SYMBOL(generic_file_llseek_size
);
134 * generic_file_llseek - generic llseek implementation for regular files
135 * @file: file structure to seek on
136 * @offset: file offset to seek to
137 * @whence: type of seek
139 * This is a generic implemenation of ->llseek useable for all normal local
140 * filesystems. It just updates the file offset to the value specified by
141 * @offset and @whence.
143 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
145 struct inode
*inode
= file
->f_mapping
->host
;
147 return generic_file_llseek_size(file
, offset
, whence
,
148 inode
->i_sb
->s_maxbytes
,
151 EXPORT_SYMBOL(generic_file_llseek
);
154 * fixed_size_llseek - llseek implementation for fixed-sized devices
155 * @file: file structure to seek on
156 * @offset: file offset to seek to
157 * @whence: type of seek
158 * @size: size of the file
161 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
164 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
165 return generic_file_llseek_size(file
, offset
, whence
,
171 EXPORT_SYMBOL(fixed_size_llseek
);
174 * no_seek_end_llseek - llseek implementation for fixed-sized devices
175 * @file: file structure to seek on
176 * @offset: file offset to seek to
177 * @whence: type of seek
180 loff_t
no_seek_end_llseek(struct file
*file
, loff_t offset
, int whence
)
183 case SEEK_SET
: case SEEK_CUR
:
184 return generic_file_llseek_size(file
, offset
, whence
,
190 EXPORT_SYMBOL(no_seek_end_llseek
);
193 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
194 * @file: file structure to seek on
195 * @offset: file offset to seek to
196 * @whence: type of seek
197 * @size: maximal offset allowed
200 loff_t
no_seek_end_llseek_size(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
203 case SEEK_SET
: case SEEK_CUR
:
204 return generic_file_llseek_size(file
, offset
, whence
,
210 EXPORT_SYMBOL(no_seek_end_llseek_size
);
213 * noop_llseek - No Operation Performed llseek implementation
214 * @file: file structure to seek on
215 * @offset: file offset to seek to
216 * @whence: type of seek
218 * This is an implementation of ->llseek useable for the rare special case when
219 * userspace expects the seek to succeed but the (device) file is actually not
220 * able to perform the seek. In this case you use noop_llseek() instead of
221 * falling back to the default implementation of ->llseek.
223 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
227 EXPORT_SYMBOL(noop_llseek
);
229 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
233 EXPORT_SYMBOL(no_llseek
);
235 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
237 struct inode
*inode
= file_inode(file
);
243 offset
+= i_size_read(inode
);
247 retval
= file
->f_pos
;
250 offset
+= file
->f_pos
;
254 * In the generic case the entire file is data, so as
255 * long as offset isn't at the end of the file then the
258 if (offset
>= inode
->i_size
) {
265 * There is a virtual hole at the end of the file, so
266 * as long as offset isn't i_size or larger, return
269 if (offset
>= inode
->i_size
) {
273 offset
= inode
->i_size
;
277 if (offset
>= 0 || unsigned_offsets(file
)) {
278 if (offset
!= file
->f_pos
) {
279 file
->f_pos
= offset
;
288 EXPORT_SYMBOL(default_llseek
);
290 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
292 loff_t (*fn
)(struct file
*, loff_t
, int);
295 if (file
->f_mode
& FMODE_LSEEK
) {
296 if (file
->f_op
->llseek
)
297 fn
= file
->f_op
->llseek
;
299 return fn(file
, offset
, whence
);
301 EXPORT_SYMBOL(vfs_llseek
);
303 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
306 struct fd f
= fdget_pos(fd
);
311 if (whence
<= SEEK_MAX
) {
312 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
314 if (res
!= (loff_t
)retval
)
315 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
322 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
324 return sys_lseek(fd
, offset
, whence
);
328 #ifdef __ARCH_WANT_SYS_LLSEEK
329 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
330 unsigned long, offset_low
, loff_t __user
*, result
,
331 unsigned int, whence
)
334 struct fd f
= fdget_pos(fd
);
341 if (whence
> SEEK_MAX
)
344 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
347 retval
= (int)offset
;
350 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
359 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
363 int retval
= -EINVAL
;
365 inode
= file_inode(file
);
366 if (unlikely((ssize_t
) count
< 0))
369 if (unlikely(pos
< 0)) {
370 if (!unsigned_offsets(file
))
372 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
374 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
375 if (!unsigned_offsets(file
))
379 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
380 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
381 read_write
== READ
? F_RDLCK
: F_WRLCK
);
385 return security_file_permission(file
,
386 read_write
== READ
? MAY_READ
: MAY_WRITE
);
389 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
391 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
393 struct iov_iter iter
;
396 init_sync_kiocb(&kiocb
, filp
);
397 kiocb
.ki_pos
= *ppos
;
398 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
400 ret
= call_read_iter(filp
, &kiocb
, &iter
);
401 BUG_ON(ret
== -EIOCBQUEUED
);
402 *ppos
= kiocb
.ki_pos
;
406 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
409 if (file
->f_op
->read
)
410 return file
->f_op
->read(file
, buf
, count
, pos
);
411 else if (file
->f_op
->read_iter
)
412 return new_sync_read(file
, buf
, count
, pos
);
416 EXPORT_SYMBOL(__vfs_read
);
418 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
422 if (!(file
->f_mode
& FMODE_READ
))
424 if (!(file
->f_mode
& FMODE_CAN_READ
))
426 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
429 ret
= rw_verify_area(READ
, file
, pos
, count
);
431 if (count
> MAX_RW_COUNT
)
432 count
= MAX_RW_COUNT
;
433 ret
= __vfs_read(file
, buf
, count
, pos
);
435 fsnotify_access(file
);
436 add_rchar(current
, ret
);
444 EXPORT_SYMBOL(vfs_read
);
446 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
448 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
450 struct iov_iter iter
;
453 init_sync_kiocb(&kiocb
, filp
);
454 kiocb
.ki_pos
= *ppos
;
455 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
457 ret
= call_write_iter(filp
, &kiocb
, &iter
);
458 BUG_ON(ret
== -EIOCBQUEUED
);
460 *ppos
= kiocb
.ki_pos
;
464 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
467 if (file
->f_op
->write
)
468 return file
->f_op
->write(file
, p
, count
, pos
);
469 else if (file
->f_op
->write_iter
)
470 return new_sync_write(file
, p
, count
, pos
);
474 EXPORT_SYMBOL(__vfs_write
);
476 vfs_readf_t
vfs_readf(struct file
*file
)
478 const struct file_operations
*fop
= file
->f_op
;
483 return new_sync_read
;
484 return ERR_PTR(-ENOSYS
);
486 EXPORT_SYMBOL_GPL(vfs_readf
);
488 vfs_writef_t
vfs_writef(struct file
*file
)
490 const struct file_operations
*fop
= file
->f_op
;
495 return new_sync_write
;
496 return ERR_PTR(-ENOSYS
);
498 EXPORT_SYMBOL_GPL(vfs_writef
);
500 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
503 const char __user
*p
;
506 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
511 p
= (__force
const char __user
*)buf
;
512 if (count
> MAX_RW_COUNT
)
513 count
= MAX_RW_COUNT
;
514 ret
= __vfs_write(file
, p
, count
, pos
);
517 fsnotify_modify(file
);
518 add_wchar(current
, ret
);
524 EXPORT_SYMBOL(__kernel_write
);
526 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
530 if (!(file
->f_mode
& FMODE_WRITE
))
532 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
534 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
537 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
539 if (count
> MAX_RW_COUNT
)
540 count
= MAX_RW_COUNT
;
541 file_start_write(file
);
542 ret
= __vfs_write(file
, buf
, count
, pos
);
544 fsnotify_modify(file
);
545 add_wchar(current
, ret
);
548 file_end_write(file
);
554 EXPORT_SYMBOL(vfs_write
);
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
, int 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
, int 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
, int 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
, int 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
, int 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
);
986 EXPORT_SYMBOL(vfs_readv
);
988 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
989 unsigned long vlen
, loff_t
*pos
, int flags
)
991 struct iovec iovstack
[UIO_FASTIOV
];
992 struct iovec
*iov
= iovstack
;
993 struct iov_iter iter
;
996 ret
= import_iovec(WRITE
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
998 file_start_write(file
);
999 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1000 file_end_write(file
);
1005 EXPORT_SYMBOL(vfs_writev
);
1007 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
1008 unsigned long vlen
, int flags
)
1010 struct fd f
= fdget_pos(fd
);
1011 ssize_t ret
= -EBADF
;
1014 loff_t pos
= file_pos_read(f
.file
);
1015 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1017 file_pos_write(f
.file
, pos
);
1022 add_rchar(current
, ret
);
1027 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
1028 unsigned long vlen
, int flags
)
1030 struct fd f
= fdget_pos(fd
);
1031 ssize_t ret
= -EBADF
;
1034 loff_t pos
= file_pos_read(f
.file
);
1035 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1037 file_pos_write(f
.file
, pos
);
1042 add_wchar(current
, ret
);
1047 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
1049 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1050 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
1053 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
1054 unsigned long vlen
, loff_t pos
, int flags
)
1057 ssize_t ret
= -EBADF
;
1065 if (f
.file
->f_mode
& FMODE_PREAD
)
1066 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1071 add_rchar(current
, ret
);
1076 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1077 unsigned long vlen
, loff_t pos
, int flags
)
1080 ssize_t ret
= -EBADF
;
1088 if (f
.file
->f_mode
& FMODE_PWRITE
)
1089 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1094 add_wchar(current
, ret
);
1099 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1100 unsigned long, vlen
)
1102 return do_readv(fd
, vec
, vlen
, 0);
1105 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1106 unsigned long, vlen
)
1108 return do_writev(fd
, vec
, vlen
, 0);
1111 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1112 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1114 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1116 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1119 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1120 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1123 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1126 return do_readv(fd
, vec
, vlen
, flags
);
1128 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1131 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1132 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1134 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1136 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1139 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1140 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1143 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1146 return do_writev(fd
, vec
, vlen
, flags
);
1148 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1151 #ifdef CONFIG_COMPAT
1152 static size_t compat_readv(struct file
*file
,
1153 const struct compat_iovec __user
*vec
,
1154 unsigned long vlen
, loff_t
*pos
, int flags
)
1156 struct iovec iovstack
[UIO_FASTIOV
];
1157 struct iovec
*iov
= iovstack
;
1158 struct iov_iter iter
;
1161 ret
= compat_import_iovec(READ
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1163 ret
= do_iter_read(file
, &iter
, pos
, flags
);
1167 add_rchar(current
, ret
);
1172 static size_t do_compat_readv(compat_ulong_t fd
,
1173 const struct compat_iovec __user
*vec
,
1174 compat_ulong_t vlen
, int flags
)
1176 struct fd f
= fdget_pos(fd
);
1182 pos
= f
.file
->f_pos
;
1183 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1185 f
.file
->f_pos
= pos
;
1191 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1192 const struct compat_iovec __user
*,vec
,
1193 compat_ulong_t
, vlen
)
1195 return do_compat_readv(fd
, vec
, vlen
, 0);
1198 static long do_compat_preadv64(unsigned long fd
,
1199 const struct compat_iovec __user
*vec
,
1200 unsigned long vlen
, loff_t pos
, int flags
)
1211 if (f
.file
->f_mode
& FMODE_PREAD
)
1212 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1217 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1218 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1219 const struct compat_iovec __user
*,vec
,
1220 unsigned long, vlen
, loff_t
, pos
)
1222 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1226 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1227 const struct compat_iovec __user
*,vec
,
1228 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1230 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1232 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1235 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1236 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1237 const struct compat_iovec __user
*,vec
,
1238 unsigned long, vlen
, loff_t
, pos
, int, flags
)
1240 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1244 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1245 const struct compat_iovec __user
*,vec
,
1246 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1249 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1252 return do_compat_readv(fd
, vec
, vlen
, flags
);
1254 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1257 static size_t compat_writev(struct file
*file
,
1258 const struct compat_iovec __user
*vec
,
1259 unsigned long vlen
, loff_t
*pos
, int flags
)
1261 struct iovec iovstack
[UIO_FASTIOV
];
1262 struct iovec
*iov
= iovstack
;
1263 struct iov_iter iter
;
1266 ret
= compat_import_iovec(WRITE
, vec
, vlen
, UIO_FASTIOV
, &iov
, &iter
);
1268 file_start_write(file
);
1269 ret
= do_iter_write(file
, &iter
, pos
, flags
);
1270 file_end_write(file
);
1274 add_wchar(current
, ret
);
1279 static size_t do_compat_writev(compat_ulong_t fd
,
1280 const struct compat_iovec __user
* vec
,
1281 compat_ulong_t vlen
, int flags
)
1283 struct fd f
= fdget_pos(fd
);
1289 pos
= f
.file
->f_pos
;
1290 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1292 f
.file
->f_pos
= pos
;
1297 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1298 const struct compat_iovec __user
*, vec
,
1299 compat_ulong_t
, vlen
)
1301 return do_compat_writev(fd
, vec
, vlen
, 0);
1304 static long do_compat_pwritev64(unsigned long fd
,
1305 const struct compat_iovec __user
*vec
,
1306 unsigned long vlen
, loff_t pos
, int flags
)
1317 if (f
.file
->f_mode
& FMODE_PWRITE
)
1318 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1323 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1324 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1325 const struct compat_iovec __user
*,vec
,
1326 unsigned long, vlen
, loff_t
, pos
)
1328 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1332 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1333 const struct compat_iovec __user
*,vec
,
1334 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1336 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1338 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1341 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1342 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1343 const struct compat_iovec __user
*,vec
,
1344 unsigned long, vlen
, loff_t
, pos
, int, flags
)
1346 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1350 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1351 const struct compat_iovec __user
*,vec
,
1352 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, int, flags
)
1354 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1357 return do_compat_writev(fd
, vec
, vlen
, flags
);
1359 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1364 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1365 size_t count
, loff_t max
)
1368 struct inode
*in_inode
, *out_inode
;
1375 * Get input file, and verify that it is ok..
1381 if (!(in
.file
->f_mode
& FMODE_READ
))
1385 pos
= in
.file
->f_pos
;
1388 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1391 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1394 if (count
> MAX_RW_COUNT
)
1395 count
= MAX_RW_COUNT
;
1398 * Get output file, and verify that it is ok..
1401 out
= fdget(out_fd
);
1404 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1407 in_inode
= file_inode(in
.file
);
1408 out_inode
= file_inode(out
.file
);
1409 out_pos
= out
.file
->f_pos
;
1410 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1415 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1417 if (unlikely(pos
+ count
> max
)) {
1418 retval
= -EOVERFLOW
;
1427 * We need to debate whether we can enable this or not. The
1428 * man page documents EAGAIN return for the output at least,
1429 * and the application is arguably buggy if it doesn't expect
1430 * EAGAIN on a non-blocking file descriptor.
1432 if (in
.file
->f_flags
& O_NONBLOCK
)
1433 fl
= SPLICE_F_NONBLOCK
;
1435 file_start_write(out
.file
);
1436 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1437 file_end_write(out
.file
);
1440 add_rchar(current
, retval
);
1441 add_wchar(current
, retval
);
1442 fsnotify_access(in
.file
);
1443 fsnotify_modify(out
.file
);
1444 out
.file
->f_pos
= out_pos
;
1448 in
.file
->f_pos
= pos
;
1454 retval
= -EOVERFLOW
;
1464 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1471 if (unlikely(get_user(off
, offset
)))
1474 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1475 if (unlikely(put_user(pos
, offset
)))
1480 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1483 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1489 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1491 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1492 if (unlikely(put_user(pos
, offset
)))
1497 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1500 #ifdef CONFIG_COMPAT
1501 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1502 compat_off_t __user
*, offset
, compat_size_t
, count
)
1509 if (unlikely(get_user(off
, offset
)))
1512 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1513 if (unlikely(put_user(pos
, offset
)))
1518 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1521 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1522 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1528 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1530 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1531 if (unlikely(put_user(pos
, offset
)))
1536 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1541 * copy_file_range() differs from regular file read and write in that it
1542 * specifically allows return partial success. When it does so is up to
1543 * the copy_file_range method.
1545 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1546 struct file
*file_out
, loff_t pos_out
,
1547 size_t len
, unsigned int flags
)
1549 struct inode
*inode_in
= file_inode(file_in
);
1550 struct inode
*inode_out
= file_inode(file_out
);
1556 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1558 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1561 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1565 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1569 if (!(file_in
->f_mode
& FMODE_READ
) ||
1570 !(file_out
->f_mode
& FMODE_WRITE
) ||
1571 (file_out
->f_flags
& O_APPEND
))
1574 /* this could be relaxed once a method supports cross-fs copies */
1575 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1581 file_start_write(file_out
);
1584 * Try cloning first, this is supported by more file systems, and
1585 * more efficient if both clone and copy are supported (e.g. NFS).
1587 if (file_in
->f_op
->clone_file_range
) {
1588 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1589 file_out
, pos_out
, len
);
1596 if (file_out
->f_op
->copy_file_range
) {
1597 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1598 pos_out
, len
, flags
);
1599 if (ret
!= -EOPNOTSUPP
)
1603 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1604 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1608 fsnotify_access(file_in
);
1609 add_rchar(current
, ret
);
1610 fsnotify_modify(file_out
);
1611 add_wchar(current
, ret
);
1617 file_end_write(file_out
);
1621 EXPORT_SYMBOL(vfs_copy_file_range
);
1623 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1624 int, fd_out
, loff_t __user
*, off_out
,
1625 size_t, len
, unsigned int, flags
)
1631 ssize_t ret
= -EBADF
;
1633 f_in
= fdget(fd_in
);
1637 f_out
= fdget(fd_out
);
1643 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1646 pos_in
= f_in
.file
->f_pos
;
1650 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1653 pos_out
= f_out
.file
->f_pos
;
1656 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1663 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1666 f_in
.file
->f_pos
= pos_in
;
1670 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1673 f_out
.file
->f_pos
= pos_out
;
1685 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1687 struct inode
*inode
= file_inode(file
);
1689 if (unlikely(pos
< 0))
1692 if (unlikely((loff_t
) (pos
+ len
) < 0))
1695 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1696 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1699 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1700 write
? F_WRLCK
: F_RDLCK
);
1705 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1709 * Check that the two inodes are eligible for cloning, the ranges make
1710 * sense, and then flush all dirty data. Caller must ensure that the
1711 * inodes have been locked against any other modifications.
1713 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1714 * the usual negative error code.
1716 int vfs_clone_file_prep_inodes(struct inode
*inode_in
, loff_t pos_in
,
1717 struct inode
*inode_out
, loff_t pos_out
,
1718 u64
*len
, bool is_dedupe
)
1720 loff_t bs
= inode_out
->i_sb
->s_blocksize
;
1723 bool same_inode
= (inode_in
== inode_out
);
1726 /* Don't touch certain kinds of inodes */
1727 if (IS_IMMUTABLE(inode_out
))
1730 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
1733 /* Don't reflink dirs, pipes, sockets... */
1734 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1736 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1739 /* Are we going all the way to the end? */
1740 isize
= i_size_read(inode_in
);
1744 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1746 if (is_dedupe
|| pos_in
== isize
)
1750 *len
= isize
- pos_in
;
1753 /* Ensure offsets don't wrap and the input is inside i_size */
1754 if (pos_in
+ *len
< pos_in
|| pos_out
+ *len
< pos_out
||
1755 pos_in
+ *len
> isize
)
1758 /* Don't allow dedupe past EOF in the dest file */
1762 disize
= i_size_read(inode_out
);
1763 if (pos_out
>= disize
|| pos_out
+ *len
> disize
)
1767 /* If we're linking to EOF, continue to the block boundary. */
1768 if (pos_in
+ *len
== isize
)
1769 blen
= ALIGN(isize
, bs
) - pos_in
;
1773 /* Only reflink if we're aligned to block boundaries */
1774 if (!IS_ALIGNED(pos_in
, bs
) || !IS_ALIGNED(pos_in
+ blen
, bs
) ||
1775 !IS_ALIGNED(pos_out
, bs
) || !IS_ALIGNED(pos_out
+ blen
, bs
))
1778 /* Don't allow overlapped reflink within the same file */
1780 if (pos_out
+ blen
> pos_in
&& pos_out
< pos_in
+ blen
)
1784 /* Wait for the completion of any pending IOs on both files */
1785 inode_dio_wait(inode_in
);
1787 inode_dio_wait(inode_out
);
1789 ret
= filemap_write_and_wait_range(inode_in
->i_mapping
,
1790 pos_in
, pos_in
+ *len
- 1);
1794 ret
= filemap_write_and_wait_range(inode_out
->i_mapping
,
1795 pos_out
, pos_out
+ *len
- 1);
1800 * Check that the extents are the same.
1803 bool is_same
= false;
1805 ret
= vfs_dedupe_file_range_compare(inode_in
, pos_in
,
1806 inode_out
, pos_out
, *len
, &is_same
);
1815 EXPORT_SYMBOL(vfs_clone_file_prep_inodes
);
1817 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1818 struct file
*file_out
, loff_t pos_out
, u64 len
)
1820 struct inode
*inode_in
= file_inode(file_in
);
1821 struct inode
*inode_out
= file_inode(file_out
);
1824 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1826 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1830 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1831 * the same mount. Practically, they only need to be on the same file
1834 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1837 if (!(file_in
->f_mode
& FMODE_READ
) ||
1838 !(file_out
->f_mode
& FMODE_WRITE
) ||
1839 (file_out
->f_flags
& O_APPEND
))
1842 if (!file_in
->f_op
->clone_file_range
)
1845 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1849 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1853 if (pos_in
+ len
> i_size_read(inode_in
))
1856 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1857 file_out
, pos_out
, len
);
1859 fsnotify_access(file_in
);
1860 fsnotify_modify(file_out
);
1865 EXPORT_SYMBOL(vfs_clone_file_range
);
1868 * Read a page's worth of file data into the page cache. Return the page
1871 static struct page
*vfs_dedupe_get_page(struct inode
*inode
, loff_t offset
)
1873 struct address_space
*mapping
;
1877 n
= offset
>> PAGE_SHIFT
;
1878 mapping
= inode
->i_mapping
;
1879 page
= read_mapping_page(mapping
, n
, NULL
);
1882 if (!PageUptodate(page
)) {
1884 return ERR_PTR(-EIO
);
1891 * Compare extents of two files to see if they are the same.
1892 * Caller must have locked both inodes to prevent write races.
1894 int vfs_dedupe_file_range_compare(struct inode
*src
, loff_t srcoff
,
1895 struct inode
*dest
, loff_t destoff
,
1896 loff_t len
, bool *is_same
)
1902 struct page
*src_page
;
1903 struct page
*dest_page
;
1911 src_poff
= srcoff
& (PAGE_SIZE
- 1);
1912 dest_poff
= destoff
& (PAGE_SIZE
- 1);
1913 cmp_len
= min(PAGE_SIZE
- src_poff
,
1914 PAGE_SIZE
- dest_poff
);
1915 cmp_len
= min(cmp_len
, len
);
1919 src_page
= vfs_dedupe_get_page(src
, srcoff
);
1920 if (IS_ERR(src_page
)) {
1921 error
= PTR_ERR(src_page
);
1924 dest_page
= vfs_dedupe_get_page(dest
, destoff
);
1925 if (IS_ERR(dest_page
)) {
1926 error
= PTR_ERR(dest_page
);
1927 unlock_page(src_page
);
1931 src_addr
= kmap_atomic(src_page
);
1932 dest_addr
= kmap_atomic(dest_page
);
1934 flush_dcache_page(src_page
);
1935 flush_dcache_page(dest_page
);
1937 if (memcmp(src_addr
+ src_poff
, dest_addr
+ dest_poff
, cmp_len
))
1940 kunmap_atomic(dest_addr
);
1941 kunmap_atomic(src_addr
);
1942 unlock_page(dest_page
);
1943 unlock_page(src_page
);
1944 put_page(dest_page
);
1961 EXPORT_SYMBOL(vfs_dedupe_file_range_compare
);
1963 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1965 struct file_dedupe_range_info
*info
;
1966 struct inode
*src
= file_inode(file
);
1971 bool is_admin
= capable(CAP_SYS_ADMIN
);
1972 u16 count
= same
->dest_count
;
1973 struct file
*dst_file
;
1977 if (!(file
->f_mode
& FMODE_READ
))
1980 if (same
->reserved1
|| same
->reserved2
)
1983 off
= same
->src_offset
;
1984 len
= same
->src_length
;
1987 if (S_ISDIR(src
->i_mode
))
1991 if (!S_ISREG(src
->i_mode
))
1994 ret
= clone_verify_area(file
, off
, len
, false);
1999 if (off
+ len
> i_size_read(src
))
2002 /* pre-format output fields to sane values */
2003 for (i
= 0; i
< count
; i
++) {
2004 same
->info
[i
].bytes_deduped
= 0ULL;
2005 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
2008 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
2010 struct fd dst_fd
= fdget(info
->dest_fd
);
2012 dst_file
= dst_fd
.file
;
2014 info
->status
= -EBADF
;
2017 dst
= file_inode(dst_file
);
2019 ret
= mnt_want_write_file(dst_file
);
2025 dst_off
= info
->dest_offset
;
2026 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
2033 if (info
->reserved
) {
2034 info
->status
= -EINVAL
;
2035 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
2036 info
->status
= -EINVAL
;
2037 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
2038 info
->status
= -EXDEV
;
2039 } else if (S_ISDIR(dst
->i_mode
)) {
2040 info
->status
= -EISDIR
;
2041 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
2042 info
->status
= -EINVAL
;
2044 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
2047 if (deduped
== -EBADE
)
2048 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
2049 else if (deduped
< 0)
2050 info
->status
= deduped
;
2052 info
->bytes_deduped
+= deduped
;
2056 mnt_drop_write_file(dst_file
);
2060 if (fatal_signal_pending(current
))
2067 EXPORT_SYMBOL(vfs_dedupe_file_range
);