2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
22 #include <asm/uaccess.h>
23 #include <asm/unistd.h>
25 typedef ssize_t (*io_fn_t
)(struct file
*, char __user
*, size_t, loff_t
*);
26 typedef ssize_t (*iter_fn_t
)(struct kiocb
*, struct iov_iter
*);
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 int 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.
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.
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 static inline struct fd
fdget_pos(int fd
)
306 return __to_fd(__fdget_pos(fd
));
309 static inline void fdput_pos(struct fd f
)
311 if (f
.flags
& FDPUT_POS_UNLOCK
)
312 mutex_unlock(&f
.file
->f_pos_lock
);
316 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
319 struct fd f
= fdget_pos(fd
);
324 if (whence
<= SEEK_MAX
) {
325 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
327 if (res
!= (loff_t
)retval
)
328 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
335 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
337 return sys_lseek(fd
, offset
, whence
);
341 #ifdef __ARCH_WANT_SYS_LLSEEK
342 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
343 unsigned long, offset_low
, loff_t __user
*, result
,
344 unsigned int, whence
)
347 struct fd f
= fdget_pos(fd
);
354 if (whence
> SEEK_MAX
)
357 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
360 retval
= (int)offset
;
363 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
372 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
377 if (!file
->f_op
->read_iter
)
380 init_sync_kiocb(&kiocb
, file
);
381 kiocb
.ki_pos
= *ppos
;
384 ret
= file
->f_op
->read_iter(&kiocb
, iter
);
385 BUG_ON(ret
== -EIOCBQUEUED
);
387 *ppos
= kiocb
.ki_pos
;
390 EXPORT_SYMBOL(vfs_iter_read
);
392 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
397 if (!file
->f_op
->write_iter
)
400 init_sync_kiocb(&kiocb
, file
);
401 kiocb
.ki_pos
= *ppos
;
404 ret
= file
->f_op
->write_iter(&kiocb
, iter
);
405 BUG_ON(ret
== -EIOCBQUEUED
);
407 *ppos
= kiocb
.ki_pos
;
410 EXPORT_SYMBOL(vfs_iter_write
);
413 * rw_verify_area doesn't like huge counts. We limit
414 * them to something that fits in "int" so that others
415 * won't have to do range checks all the time.
417 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
421 int retval
= -EINVAL
;
423 inode
= file_inode(file
);
424 if (unlikely((ssize_t
) count
< 0))
427 if (unlikely(pos
< 0)) {
428 if (!unsigned_offsets(file
))
430 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
432 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
433 if (!unsigned_offsets(file
))
437 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
438 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
439 read_write
== READ
? F_RDLCK
: F_WRLCK
);
443 retval
= security_file_permission(file
,
444 read_write
== READ
? MAY_READ
: MAY_WRITE
);
447 return count
> MAX_RW_COUNT
? MAX_RW_COUNT
: count
;
450 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
452 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
454 struct iov_iter iter
;
457 init_sync_kiocb(&kiocb
, filp
);
458 kiocb
.ki_pos
= *ppos
;
459 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
461 ret
= filp
->f_op
->read_iter(&kiocb
, &iter
);
462 BUG_ON(ret
== -EIOCBQUEUED
);
463 *ppos
= kiocb
.ki_pos
;
467 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
470 if (file
->f_op
->read
)
471 return file
->f_op
->read(file
, buf
, count
, pos
);
472 else if (file
->f_op
->read_iter
)
473 return new_sync_read(file
, buf
, count
, pos
);
477 EXPORT_SYMBOL(__vfs_read
);
479 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
483 if (!(file
->f_mode
& FMODE_READ
))
485 if (!(file
->f_mode
& FMODE_CAN_READ
))
487 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
490 ret
= rw_verify_area(READ
, file
, pos
, count
);
493 ret
= __vfs_read(file
, buf
, count
, pos
);
495 fsnotify_access(file
);
496 add_rchar(current
, ret
);
504 EXPORT_SYMBOL(vfs_read
);
506 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
508 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
510 struct iov_iter iter
;
513 init_sync_kiocb(&kiocb
, filp
);
514 kiocb
.ki_pos
= *ppos
;
515 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
517 ret
= filp
->f_op
->write_iter(&kiocb
, &iter
);
518 BUG_ON(ret
== -EIOCBQUEUED
);
520 *ppos
= kiocb
.ki_pos
;
524 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
527 if (file
->f_op
->write
)
528 return file
->f_op
->write(file
, p
, count
, pos
);
529 else if (file
->f_op
->write_iter
)
530 return new_sync_write(file
, p
, count
, pos
);
534 EXPORT_SYMBOL(__vfs_write
);
536 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
539 const char __user
*p
;
542 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
547 p
= (__force
const char __user
*)buf
;
548 if (count
> MAX_RW_COUNT
)
549 count
= MAX_RW_COUNT
;
550 ret
= __vfs_write(file
, p
, count
, pos
);
553 fsnotify_modify(file
);
554 add_wchar(current
, ret
);
560 EXPORT_SYMBOL(__kernel_write
);
562 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
566 if (!(file
->f_mode
& FMODE_WRITE
))
568 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
570 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
573 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
576 file_start_write(file
);
577 ret
= __vfs_write(file
, buf
, count
, pos
);
579 fsnotify_modify(file
);
580 add_wchar(current
, ret
);
583 file_end_write(file
);
589 EXPORT_SYMBOL(vfs_write
);
591 static inline loff_t
file_pos_read(struct file
*file
)
596 static inline void file_pos_write(struct file
*file
, loff_t pos
)
601 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
603 struct fd f
= fdget_pos(fd
);
604 ssize_t ret
= -EBADF
;
607 loff_t pos
= file_pos_read(f
.file
);
608 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
610 file_pos_write(f
.file
, pos
);
616 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
619 struct fd f
= fdget_pos(fd
);
620 ssize_t ret
= -EBADF
;
623 loff_t pos
= file_pos_read(f
.file
);
624 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
626 file_pos_write(f
.file
, pos
);
633 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
634 size_t, count
, loff_t
, pos
)
637 ssize_t ret
= -EBADF
;
645 if (f
.file
->f_mode
& FMODE_PREAD
)
646 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
653 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
654 size_t, count
, loff_t
, pos
)
657 ssize_t ret
= -EBADF
;
665 if (f
.file
->f_mode
& FMODE_PWRITE
)
666 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
674 * Reduce an iovec's length in-place. Return the resulting number of segments
676 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
678 unsigned long seg
= 0;
681 while (seg
< nr_segs
) {
683 if (len
+ iov
->iov_len
>= to
) {
684 iov
->iov_len
= to
- len
;
692 EXPORT_SYMBOL(iov_shorten
);
694 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
695 loff_t
*ppos
, iter_fn_t fn
)
700 init_sync_kiocb(&kiocb
, filp
);
701 kiocb
.ki_pos
= *ppos
;
703 ret
= fn(&kiocb
, iter
);
704 BUG_ON(ret
== -EIOCBQUEUED
);
705 *ppos
= kiocb
.ki_pos
;
709 /* Do it by hand, with file-ops */
710 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
711 loff_t
*ppos
, io_fn_t fn
)
715 while (iov_iter_count(iter
)) {
716 struct iovec iovec
= iov_iter_iovec(iter
);
719 nr
= fn(filp
, iovec
.iov_base
, iovec
.iov_len
, ppos
);
727 if (nr
!= iovec
.iov_len
)
729 iov_iter_advance(iter
, nr
);
735 /* A write operation does a read from user space and vice versa */
736 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
738 ssize_t
rw_copy_check_uvector(int type
, const struct iovec __user
* uvector
,
739 unsigned long nr_segs
, unsigned long fast_segs
,
740 struct iovec
*fast_pointer
,
741 struct iovec
**ret_pointer
)
745 struct iovec
*iov
= fast_pointer
;
748 * SuS says "The readv() function *may* fail if the iovcnt argument
749 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
750 * traditionally returned zero for zero segments, so...
758 * First get the "struct iovec" from user memory and
759 * verify all the pointers
761 if (nr_segs
> UIO_MAXIOV
) {
765 if (nr_segs
> fast_segs
) {
766 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
772 if (copy_from_user(iov
, uvector
, nr_segs
*sizeof(*uvector
))) {
778 * According to the Single Unix Specification we should return EINVAL
779 * if an element length is < 0 when cast to ssize_t or if the
780 * total length would overflow the ssize_t return value of the
783 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
787 for (seg
= 0; seg
< nr_segs
; seg
++) {
788 void __user
*buf
= iov
[seg
].iov_base
;
789 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
791 /* see if we we're about to use an invalid len or if
792 * it's about to overflow ssize_t */
798 && unlikely(!access_ok(vrfy_dir(type
), buf
, len
))) {
802 if (len
> MAX_RW_COUNT
- ret
) {
803 len
= MAX_RW_COUNT
- ret
;
804 iov
[seg
].iov_len
= len
;
813 static ssize_t
do_readv_writev(int type
, struct file
*file
,
814 const struct iovec __user
* uvector
,
815 unsigned long nr_segs
, loff_t
*pos
)
818 struct iovec iovstack
[UIO_FASTIOV
];
819 struct iovec
*iov
= iovstack
;
820 struct iov_iter iter
;
825 ret
= import_iovec(type
, uvector
, nr_segs
,
826 ARRAY_SIZE(iovstack
), &iov
, &iter
);
830 tot_len
= iov_iter_count(&iter
);
833 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
838 fn
= file
->f_op
->read
;
839 iter_fn
= file
->f_op
->read_iter
;
841 fn
= (io_fn_t
)file
->f_op
->write
;
842 iter_fn
= file
->f_op
->write_iter
;
843 file_start_write(file
);
847 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
);
849 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
);
852 file_end_write(file
);
856 if ((ret
+ (type
== READ
)) > 0) {
858 fsnotify_access(file
);
860 fsnotify_modify(file
);
865 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
866 unsigned long vlen
, loff_t
*pos
)
868 if (!(file
->f_mode
& FMODE_READ
))
870 if (!(file
->f_mode
& FMODE_CAN_READ
))
873 return do_readv_writev(READ
, file
, vec
, vlen
, pos
);
876 EXPORT_SYMBOL(vfs_readv
);
878 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
879 unsigned long vlen
, loff_t
*pos
)
881 if (!(file
->f_mode
& FMODE_WRITE
))
883 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
886 return do_readv_writev(WRITE
, file
, vec
, vlen
, pos
);
889 EXPORT_SYMBOL(vfs_writev
);
891 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
894 struct fd f
= fdget_pos(fd
);
895 ssize_t ret
= -EBADF
;
898 loff_t pos
= file_pos_read(f
.file
);
899 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
);
901 file_pos_write(f
.file
, pos
);
906 add_rchar(current
, ret
);
911 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
914 struct fd f
= fdget_pos(fd
);
915 ssize_t ret
= -EBADF
;
918 loff_t pos
= file_pos_read(f
.file
);
919 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
);
921 file_pos_write(f
.file
, pos
);
926 add_wchar(current
, ret
);
931 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
933 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
934 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
937 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
938 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
940 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
942 ssize_t ret
= -EBADF
;
950 if (f
.file
->f_mode
& FMODE_PREAD
)
951 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
);
956 add_rchar(current
, ret
);
961 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
962 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
964 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
966 ssize_t ret
= -EBADF
;
974 if (f
.file
->f_mode
& FMODE_PWRITE
)
975 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
);
980 add_wchar(current
, ret
);
987 static ssize_t
compat_do_readv_writev(int type
, struct file
*file
,
988 const struct compat_iovec __user
*uvector
,
989 unsigned long nr_segs
, loff_t
*pos
)
991 compat_ssize_t tot_len
;
992 struct iovec iovstack
[UIO_FASTIOV
];
993 struct iovec
*iov
= iovstack
;
994 struct iov_iter iter
;
999 ret
= compat_import_iovec(type
, uvector
, nr_segs
,
1000 UIO_FASTIOV
, &iov
, &iter
);
1004 tot_len
= iov_iter_count(&iter
);
1007 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
1012 fn
= file
->f_op
->read
;
1013 iter_fn
= file
->f_op
->read_iter
;
1015 fn
= (io_fn_t
)file
->f_op
->write
;
1016 iter_fn
= file
->f_op
->write_iter
;
1017 file_start_write(file
);
1021 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
);
1023 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
);
1026 file_end_write(file
);
1030 if ((ret
+ (type
== READ
)) > 0) {
1032 fsnotify_access(file
);
1034 fsnotify_modify(file
);
1039 static size_t compat_readv(struct file
*file
,
1040 const struct compat_iovec __user
*vec
,
1041 unsigned long vlen
, loff_t
*pos
)
1043 ssize_t ret
= -EBADF
;
1045 if (!(file
->f_mode
& FMODE_READ
))
1049 if (!(file
->f_mode
& FMODE_CAN_READ
))
1052 ret
= compat_do_readv_writev(READ
, file
, vec
, vlen
, pos
);
1056 add_rchar(current
, ret
);
1061 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1062 const struct compat_iovec __user
*,vec
,
1063 compat_ulong_t
, vlen
)
1065 struct fd f
= fdget_pos(fd
);
1071 pos
= f
.file
->f_pos
;
1072 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
);
1074 f
.file
->f_pos
= pos
;
1079 static long __compat_sys_preadv64(unsigned long fd
,
1080 const struct compat_iovec __user
*vec
,
1081 unsigned long vlen
, loff_t pos
)
1092 if (f
.file
->f_mode
& FMODE_PREAD
)
1093 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
);
1098 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1099 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1100 const struct compat_iovec __user
*,vec
,
1101 unsigned long, vlen
, loff_t
, pos
)
1103 return __compat_sys_preadv64(fd
, vec
, vlen
, pos
);
1107 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1108 const struct compat_iovec __user
*,vec
,
1109 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1111 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1113 return __compat_sys_preadv64(fd
, vec
, vlen
, pos
);
1116 static size_t compat_writev(struct file
*file
,
1117 const struct compat_iovec __user
*vec
,
1118 unsigned long vlen
, loff_t
*pos
)
1120 ssize_t ret
= -EBADF
;
1122 if (!(file
->f_mode
& FMODE_WRITE
))
1126 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1129 ret
= compat_do_readv_writev(WRITE
, file
, vec
, vlen
, pos
);
1133 add_wchar(current
, ret
);
1138 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1139 const struct compat_iovec __user
*, vec
,
1140 compat_ulong_t
, vlen
)
1142 struct fd f
= fdget_pos(fd
);
1148 pos
= f
.file
->f_pos
;
1149 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
);
1151 f
.file
->f_pos
= pos
;
1156 static long __compat_sys_pwritev64(unsigned long fd
,
1157 const struct compat_iovec __user
*vec
,
1158 unsigned long vlen
, loff_t pos
)
1169 if (f
.file
->f_mode
& FMODE_PWRITE
)
1170 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
);
1175 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1176 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1177 const struct compat_iovec __user
*,vec
,
1178 unsigned long, vlen
, loff_t
, pos
)
1180 return __compat_sys_pwritev64(fd
, vec
, vlen
, pos
);
1184 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1185 const struct compat_iovec __user
*,vec
,
1186 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1188 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1190 return __compat_sys_pwritev64(fd
, vec
, vlen
, pos
);
1194 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1195 size_t count
, loff_t max
)
1198 struct inode
*in_inode
, *out_inode
;
1205 * Get input file, and verify that it is ok..
1211 if (!(in
.file
->f_mode
& FMODE_READ
))
1215 pos
= in
.file
->f_pos
;
1218 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1221 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1227 * Get output file, and verify that it is ok..
1230 out
= fdget(out_fd
);
1233 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1236 in_inode
= file_inode(in
.file
);
1237 out_inode
= file_inode(out
.file
);
1238 out_pos
= out
.file
->f_pos
;
1239 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1245 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1247 if (unlikely(pos
+ count
> max
)) {
1248 retval
= -EOVERFLOW
;
1257 * We need to debate whether we can enable this or not. The
1258 * man page documents EAGAIN return for the output at least,
1259 * and the application is arguably buggy if it doesn't expect
1260 * EAGAIN on a non-blocking file descriptor.
1262 if (in
.file
->f_flags
& O_NONBLOCK
)
1263 fl
= SPLICE_F_NONBLOCK
;
1265 file_start_write(out
.file
);
1266 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1267 file_end_write(out
.file
);
1270 add_rchar(current
, retval
);
1271 add_wchar(current
, retval
);
1272 fsnotify_access(in
.file
);
1273 fsnotify_modify(out
.file
);
1274 out
.file
->f_pos
= out_pos
;
1278 in
.file
->f_pos
= pos
;
1284 retval
= -EOVERFLOW
;
1294 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1301 if (unlikely(get_user(off
, offset
)))
1304 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1305 if (unlikely(put_user(pos
, offset
)))
1310 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1313 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1319 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1321 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1322 if (unlikely(put_user(pos
, offset
)))
1327 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1330 #ifdef CONFIG_COMPAT
1331 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1332 compat_off_t __user
*, offset
, compat_size_t
, count
)
1339 if (unlikely(get_user(off
, offset
)))
1342 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1343 if (unlikely(put_user(pos
, offset
)))
1348 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1351 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1352 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1358 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1360 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1361 if (unlikely(put_user(pos
, offset
)))
1366 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1371 * copy_file_range() differs from regular file read and write in that it
1372 * specifically allows return partial success. When it does so is up to
1373 * the copy_file_range method.
1375 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1376 struct file
*file_out
, loff_t pos_out
,
1377 size_t len
, unsigned int flags
)
1379 struct inode
*inode_in
= file_inode(file_in
);
1380 struct inode
*inode_out
= file_inode(file_out
);
1386 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1387 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1389 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1393 if (!(file_in
->f_mode
& FMODE_READ
) ||
1394 !(file_out
->f_mode
& FMODE_WRITE
) ||
1395 (file_out
->f_flags
& O_APPEND
))
1398 /* this could be relaxed once a method supports cross-fs copies */
1399 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1405 ret
= mnt_want_write_file(file_out
);
1410 if (file_out
->f_op
->copy_file_range
)
1411 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1412 pos_out
, len
, flags
);
1413 if (ret
== -EOPNOTSUPP
)
1414 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1415 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1418 fsnotify_access(file_in
);
1419 add_rchar(current
, ret
);
1420 fsnotify_modify(file_out
);
1421 add_wchar(current
, ret
);
1426 mnt_drop_write_file(file_out
);
1430 EXPORT_SYMBOL(vfs_copy_file_range
);
1432 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1433 int, fd_out
, loff_t __user
*, off_out
,
1434 size_t, len
, unsigned int, flags
)
1440 ssize_t ret
= -EBADF
;
1442 f_in
= fdget(fd_in
);
1446 f_out
= fdget(fd_out
);
1452 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1455 pos_in
= f_in
.file
->f_pos
;
1459 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1462 pos_out
= f_out
.file
->f_pos
;
1465 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1472 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1475 f_in
.file
->f_pos
= pos_in
;
1479 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1482 f_out
.file
->f_pos
= pos_out
;
1494 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1496 struct inode
*inode
= file_inode(file
);
1498 if (unlikely(pos
< 0))
1501 if (unlikely((loff_t
) (pos
+ len
) < 0))
1504 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1505 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1508 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1509 write
? F_WRLCK
: F_RDLCK
);
1514 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1517 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1518 struct file
*file_out
, loff_t pos_out
, u64 len
)
1520 struct inode
*inode_in
= file_inode(file_in
);
1521 struct inode
*inode_out
= file_inode(file_out
);
1524 if (inode_in
->i_sb
!= inode_out
->i_sb
||
1525 file_in
->f_path
.mnt
!= file_out
->f_path
.mnt
)
1528 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1530 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1533 if (!(file_in
->f_mode
& FMODE_READ
) ||
1534 !(file_out
->f_mode
& FMODE_WRITE
) ||
1535 (file_out
->f_flags
& O_APPEND
) ||
1536 !file_in
->f_op
->clone_file_range
)
1539 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1543 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1547 if (pos_in
+ len
> i_size_read(inode_in
))
1550 ret
= mnt_want_write_file(file_out
);
1554 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1555 file_out
, pos_out
, len
);
1557 fsnotify_access(file_in
);
1558 fsnotify_modify(file_out
);
1561 mnt_drop_write_file(file_out
);
1564 EXPORT_SYMBOL(vfs_clone_file_range
);
1566 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1568 struct file_dedupe_range_info
*info
;
1569 struct inode
*src
= file_inode(file
);
1574 bool is_admin
= capable(CAP_SYS_ADMIN
);
1575 u16 count
= same
->dest_count
;
1576 struct file
*dst_file
;
1580 if (!(file
->f_mode
& FMODE_READ
))
1583 if (same
->reserved1
|| same
->reserved2
)
1586 off
= same
->src_offset
;
1587 len
= same
->src_length
;
1590 if (S_ISDIR(src
->i_mode
))
1594 if (!S_ISREG(src
->i_mode
))
1597 ret
= clone_verify_area(file
, off
, len
, false);
1602 /* pre-format output fields to sane values */
1603 for (i
= 0; i
< count
; i
++) {
1604 same
->info
[i
].bytes_deduped
= 0ULL;
1605 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
1608 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
1610 struct fd dst_fd
= fdget(info
->dest_fd
);
1612 dst_file
= dst_fd
.file
;
1614 info
->status
= -EBADF
;
1617 dst
= file_inode(dst_file
);
1619 ret
= mnt_want_write_file(dst_file
);
1625 dst_off
= info
->dest_offset
;
1626 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
1633 if (info
->reserved
) {
1634 info
->status
= -EINVAL
;
1635 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
1636 info
->status
= -EINVAL
;
1637 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
1638 info
->status
= -EXDEV
;
1639 } else if (S_ISDIR(dst
->i_mode
)) {
1640 info
->status
= -EISDIR
;
1641 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
1642 info
->status
= -EINVAL
;
1644 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
1647 if (deduped
== -EBADE
)
1648 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
1649 else if (deduped
< 0)
1650 info
->status
= deduped
;
1652 info
->bytes_deduped
+= deduped
;
1656 mnt_drop_write_file(dst_file
);
1660 if (fatal_signal_pending(current
))
1667 EXPORT_SYMBOL(vfs_dedupe_file_range
);