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/aio.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>
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 (*iov_fn_t
)(struct kiocb
*, const struct iovec
*,
27 unsigned long, loff_t
);
28 typedef ssize_t (*iter_fn_t
)(struct kiocb
*, struct iov_iter
*);
30 const struct file_operations generic_ro_fops
= {
31 .llseek
= generic_file_llseek
,
32 .read
= new_sync_read
,
33 .read_iter
= generic_file_read_iter
,
34 .mmap
= generic_file_readonly_mmap
,
35 .splice_read
= generic_file_splice_read
,
38 EXPORT_SYMBOL(generic_ro_fops
);
40 static inline int unsigned_offsets(struct file
*file
)
42 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
46 * vfs_setpos - update the file offset for lseek
47 * @file: file structure in question
48 * @offset: file offset to seek to
49 * @maxsize: maximum file size
51 * This is a low-level filesystem helper for updating the file offset to
52 * the value specified by @offset if the given offset is valid and it is
53 * not equal to the current file offset.
55 * Return the specified offset on success and -EINVAL on invalid offset.
57 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
59 if (offset
< 0 && !unsigned_offsets(file
))
64 if (offset
!= file
->f_pos
) {
70 EXPORT_SYMBOL(vfs_setpos
);
73 * generic_file_llseek_size - generic llseek implementation for regular files
74 * @file: file structure to seek on
75 * @offset: file offset to seek to
76 * @whence: type of seek
77 * @size: max size of this file in file system
78 * @eof: offset used for SEEK_END position
80 * This is a variant of generic_file_llseek that allows passing in a custom
81 * maximum file size and a custom EOF position, for e.g. hashed directories
84 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
85 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
86 * read/writes behave like SEEK_SET against seeks.
89 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
90 loff_t maxsize
, loff_t eof
)
98 * Here we special-case the lseek(fd, 0, SEEK_CUR)
99 * position-querying operation. Avoid rewriting the "same"
100 * f_pos value back to the file because a concurrent read(),
101 * write() or lseek() might have altered it
106 * f_lock protects against read/modify/write race with other
107 * SEEK_CURs. Note that parallel writes and reads behave
110 spin_lock(&file
->f_lock
);
111 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
112 spin_unlock(&file
->f_lock
);
116 * In the generic case the entire file is data, so as long as
117 * offset isn't at the end of the file then the offset is data.
124 * There is a virtual hole at the end of the file, so as long as
125 * offset isn't i_size or larger, return i_size.
133 return vfs_setpos(file
, offset
, maxsize
);
135 EXPORT_SYMBOL(generic_file_llseek_size
);
138 * generic_file_llseek - generic llseek implementation for regular files
139 * @file: file structure to seek on
140 * @offset: file offset to seek to
141 * @whence: type of seek
143 * This is a generic implemenation of ->llseek useable for all normal local
144 * filesystems. It just updates the file offset to the value specified by
145 * @offset and @whence.
147 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
149 struct inode
*inode
= file
->f_mapping
->host
;
151 return generic_file_llseek_size(file
, offset
, whence
,
152 inode
->i_sb
->s_maxbytes
,
155 EXPORT_SYMBOL(generic_file_llseek
);
158 * fixed_size_llseek - llseek implementation for fixed-sized devices
159 * @file: file structure to seek on
160 * @offset: file offset to seek to
161 * @whence: type of seek
162 * @size: size of the file
165 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
168 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
169 return generic_file_llseek_size(file
, offset
, whence
,
175 EXPORT_SYMBOL(fixed_size_llseek
);
178 * noop_llseek - No Operation Performed llseek implementation
179 * @file: file structure to seek on
180 * @offset: file offset to seek to
181 * @whence: type of seek
183 * This is an implementation of ->llseek useable for the rare special case when
184 * userspace expects the seek to succeed but the (device) file is actually not
185 * able to perform the seek. In this case you use noop_llseek() instead of
186 * falling back to the default implementation of ->llseek.
188 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
192 EXPORT_SYMBOL(noop_llseek
);
194 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
198 EXPORT_SYMBOL(no_llseek
);
200 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
202 struct inode
*inode
= file_inode(file
);
205 mutex_lock(&inode
->i_mutex
);
208 offset
+= i_size_read(inode
);
212 retval
= file
->f_pos
;
215 offset
+= file
->f_pos
;
219 * In the generic case the entire file is data, so as
220 * long as offset isn't at the end of the file then the
223 if (offset
>= inode
->i_size
) {
230 * There is a virtual hole at the end of the file, so
231 * as long as offset isn't i_size or larger, return
234 if (offset
>= inode
->i_size
) {
238 offset
= inode
->i_size
;
242 if (offset
>= 0 || unsigned_offsets(file
)) {
243 if (offset
!= file
->f_pos
) {
244 file
->f_pos
= offset
;
250 mutex_unlock(&inode
->i_mutex
);
253 EXPORT_SYMBOL(default_llseek
);
255 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
257 loff_t (*fn
)(struct file
*, loff_t
, int);
260 if (file
->f_mode
& FMODE_LSEEK
) {
261 if (file
->f_op
->llseek
)
262 fn
= file
->f_op
->llseek
;
264 return fn(file
, offset
, whence
);
266 EXPORT_SYMBOL(vfs_llseek
);
268 static inline struct fd
fdget_pos(int fd
)
270 return __to_fd(__fdget_pos(fd
));
273 static inline void fdput_pos(struct fd f
)
275 if (f
.flags
& FDPUT_POS_UNLOCK
)
276 mutex_unlock(&f
.file
->f_pos_lock
);
280 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
283 struct fd f
= fdget_pos(fd
);
288 if (whence
<= SEEK_MAX
) {
289 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
291 if (res
!= (loff_t
)retval
)
292 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
299 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
301 return sys_lseek(fd
, offset
, whence
);
305 #ifdef __ARCH_WANT_SYS_LLSEEK
306 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
307 unsigned long, offset_low
, loff_t __user
*, result
,
308 unsigned int, whence
)
311 struct fd f
= fdget_pos(fd
);
318 if (whence
> SEEK_MAX
)
321 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
324 retval
= (int)offset
;
327 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
337 * rw_verify_area doesn't like huge counts. We limit
338 * them to something that fits in "int" so that others
339 * won't have to do range checks all the time.
341 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
345 int retval
= -EINVAL
;
347 inode
= file_inode(file
);
348 if (unlikely((ssize_t
) count
< 0))
351 if (unlikely(pos
< 0)) {
352 if (!unsigned_offsets(file
))
354 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
356 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
357 if (!unsigned_offsets(file
))
361 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
362 retval
= locks_mandatory_area(
363 read_write
== READ
? FLOCK_VERIFY_READ
: FLOCK_VERIFY_WRITE
,
364 inode
, file
, pos
, count
);
368 retval
= security_file_permission(file
,
369 read_write
== READ
? MAY_READ
: MAY_WRITE
);
372 return count
> MAX_RW_COUNT
? MAX_RW_COUNT
: count
;
375 ssize_t
do_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
377 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
381 init_sync_kiocb(&kiocb
, filp
);
382 kiocb
.ki_pos
= *ppos
;
383 kiocb
.ki_nbytes
= len
;
385 ret
= filp
->f_op
->aio_read(&kiocb
, &iov
, 1, kiocb
.ki_pos
);
386 if (-EIOCBQUEUED
== ret
)
387 ret
= wait_on_sync_kiocb(&kiocb
);
388 *ppos
= kiocb
.ki_pos
;
392 EXPORT_SYMBOL(do_sync_read
);
394 ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
396 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
398 struct iov_iter iter
;
401 init_sync_kiocb(&kiocb
, filp
);
402 kiocb
.ki_pos
= *ppos
;
403 kiocb
.ki_nbytes
= len
;
404 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
406 ret
= filp
->f_op
->read_iter(&kiocb
, &iter
);
407 if (-EIOCBQUEUED
== ret
)
408 ret
= wait_on_sync_kiocb(&kiocb
);
409 *ppos
= kiocb
.ki_pos
;
413 EXPORT_SYMBOL(new_sync_read
);
415 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
420 if (file
->f_op
->read
)
421 ret
= file
->f_op
->read(file
, buf
, count
, pos
);
422 else if (file
->f_op
->aio_read
)
423 ret
= do_sync_read(file
, buf
, count
, pos
);
424 else if (file
->f_op
->read_iter
)
425 ret
= new_sync_read(file
, buf
, count
, pos
);
432 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
436 if (!(file
->f_mode
& FMODE_READ
))
438 if (!(file
->f_mode
& FMODE_CAN_READ
))
440 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
443 ret
= rw_verify_area(READ
, file
, pos
, count
);
446 ret
= __vfs_read(file
, buf
, count
, pos
);
448 fsnotify_access(file
);
449 add_rchar(current
, ret
);
457 EXPORT_SYMBOL(vfs_read
);
459 ssize_t
do_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
461 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
465 init_sync_kiocb(&kiocb
, filp
);
466 kiocb
.ki_pos
= *ppos
;
467 kiocb
.ki_nbytes
= len
;
469 ret
= filp
->f_op
->aio_write(&kiocb
, &iov
, 1, kiocb
.ki_pos
);
470 if (-EIOCBQUEUED
== ret
)
471 ret
= wait_on_sync_kiocb(&kiocb
);
472 *ppos
= kiocb
.ki_pos
;
476 EXPORT_SYMBOL(do_sync_write
);
478 ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
480 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
482 struct iov_iter iter
;
485 init_sync_kiocb(&kiocb
, filp
);
486 kiocb
.ki_pos
= *ppos
;
487 kiocb
.ki_nbytes
= len
;
488 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
490 ret
= filp
->f_op
->write_iter(&kiocb
, &iter
);
491 if (-EIOCBQUEUED
== ret
)
492 ret
= wait_on_sync_kiocb(&kiocb
);
493 *ppos
= kiocb
.ki_pos
;
497 EXPORT_SYMBOL(new_sync_write
);
499 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
502 const char __user
*p
;
505 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
510 p
= (__force
const char __user
*)buf
;
511 if (count
> MAX_RW_COUNT
)
512 count
= MAX_RW_COUNT
;
513 if (file
->f_op
->write
)
514 ret
= file
->f_op
->write(file
, p
, count
, pos
);
515 else if (file
->f_op
->aio_write
)
516 ret
= do_sync_write(file
, p
, count
, pos
);
518 ret
= new_sync_write(file
, p
, count
, pos
);
521 fsnotify_modify(file
);
522 add_wchar(current
, ret
);
528 EXPORT_SYMBOL(__kernel_write
);
530 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
534 if (!(file
->f_mode
& FMODE_WRITE
))
536 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
538 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
541 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
544 file_start_write(file
);
545 if (file
->f_op
->write
)
546 ret
= file
->f_op
->write(file
, buf
, count
, pos
);
547 else if (file
->f_op
->aio_write
)
548 ret
= do_sync_write(file
, buf
, count
, pos
);
550 ret
= new_sync_write(file
, buf
, count
, pos
);
552 fsnotify_modify(file
);
553 add_wchar(current
, ret
);
556 file_end_write(file
);
562 EXPORT_SYMBOL(vfs_write
);
564 static inline loff_t
file_pos_read(struct file
*file
)
569 static inline void file_pos_write(struct file
*file
, loff_t pos
)
574 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
576 struct fd f
= fdget_pos(fd
);
577 ssize_t ret
= -EBADF
;
580 loff_t pos
= file_pos_read(f
.file
);
581 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
583 file_pos_write(f
.file
, pos
);
589 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
592 struct fd f
= fdget_pos(fd
);
593 ssize_t ret
= -EBADF
;
596 loff_t pos
= file_pos_read(f
.file
);
597 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
599 file_pos_write(f
.file
, pos
);
606 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
607 size_t, count
, loff_t
, pos
)
610 ssize_t ret
= -EBADF
;
618 if (f
.file
->f_mode
& FMODE_PREAD
)
619 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
626 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
627 size_t, count
, loff_t
, pos
)
630 ssize_t ret
= -EBADF
;
638 if (f
.file
->f_mode
& FMODE_PWRITE
)
639 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
647 * Reduce an iovec's length in-place. Return the resulting number of segments
649 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
651 unsigned long seg
= 0;
654 while (seg
< nr_segs
) {
656 if (len
+ iov
->iov_len
>= to
) {
657 iov
->iov_len
= to
- len
;
665 EXPORT_SYMBOL(iov_shorten
);
667 static ssize_t
do_iter_readv_writev(struct file
*filp
, int rw
, const struct iovec
*iov
,
668 unsigned long nr_segs
, size_t len
, loff_t
*ppos
, iter_fn_t fn
)
671 struct iov_iter iter
;
674 init_sync_kiocb(&kiocb
, filp
);
675 kiocb
.ki_pos
= *ppos
;
676 kiocb
.ki_nbytes
= len
;
678 iov_iter_init(&iter
, rw
, iov
, nr_segs
, len
);
679 ret
= fn(&kiocb
, &iter
);
680 if (ret
== -EIOCBQUEUED
)
681 ret
= wait_on_sync_kiocb(&kiocb
);
682 *ppos
= kiocb
.ki_pos
;
686 static ssize_t
do_sync_readv_writev(struct file
*filp
, const struct iovec
*iov
,
687 unsigned long nr_segs
, size_t len
, loff_t
*ppos
, iov_fn_t fn
)
692 init_sync_kiocb(&kiocb
, filp
);
693 kiocb
.ki_pos
= *ppos
;
694 kiocb
.ki_nbytes
= len
;
696 ret
= fn(&kiocb
, iov
, nr_segs
, kiocb
.ki_pos
);
697 if (ret
== -EIOCBQUEUED
)
698 ret
= wait_on_sync_kiocb(&kiocb
);
699 *ppos
= kiocb
.ki_pos
;
703 /* Do it by hand, with file-ops */
704 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iovec
*iov
,
705 unsigned long nr_segs
, loff_t
*ppos
, io_fn_t fn
)
707 struct iovec
*vector
= iov
;
710 while (nr_segs
> 0) {
715 base
= vector
->iov_base
;
716 len
= vector
->iov_len
;
720 nr
= fn(filp
, base
, len
, ppos
);
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
;
825 ret
= rw_copy_check_uvector(type
, uvector
, nr_segs
,
826 ARRAY_SIZE(iovstack
), iovstack
, &iov
);
831 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
837 fn
= file
->f_op
->read
;
838 fnv
= file
->f_op
->aio_read
;
839 iter_fn
= file
->f_op
->read_iter
;
841 fn
= (io_fn_t
)file
->f_op
->write
;
842 fnv
= file
->f_op
->aio_write
;
843 iter_fn
= file
->f_op
->write_iter
;
844 file_start_write(file
);
848 ret
= do_iter_readv_writev(file
, type
, iov
, nr_segs
, tot_len
,
851 ret
= do_sync_readv_writev(file
, iov
, nr_segs
, tot_len
,
854 ret
= do_loop_readv_writev(file
, iov
, nr_segs
, pos
, fn
);
857 file_end_write(file
);
862 if ((ret
+ (type
== READ
)) > 0) {
864 fsnotify_access(file
);
866 fsnotify_modify(file
);
871 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
872 unsigned long vlen
, loff_t
*pos
)
874 if (!(file
->f_mode
& FMODE_READ
))
876 if (!(file
->f_mode
& FMODE_CAN_READ
))
879 return do_readv_writev(READ
, file
, vec
, vlen
, pos
);
882 EXPORT_SYMBOL(vfs_readv
);
884 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
885 unsigned long vlen
, loff_t
*pos
)
887 if (!(file
->f_mode
& FMODE_WRITE
))
889 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
892 return do_readv_writev(WRITE
, file
, vec
, vlen
, pos
);
895 EXPORT_SYMBOL(vfs_writev
);
897 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
900 struct fd f
= fdget_pos(fd
);
901 ssize_t ret
= -EBADF
;
904 loff_t pos
= file_pos_read(f
.file
);
905 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
);
907 file_pos_write(f
.file
, pos
);
912 add_rchar(current
, ret
);
917 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
920 struct fd f
= fdget_pos(fd
);
921 ssize_t ret
= -EBADF
;
924 loff_t pos
= file_pos_read(f
.file
);
925 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
);
927 file_pos_write(f
.file
, pos
);
932 add_wchar(current
, ret
);
937 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
939 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
940 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
943 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
944 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
946 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
948 ssize_t ret
= -EBADF
;
956 if (f
.file
->f_mode
& FMODE_PREAD
)
957 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
);
962 add_rchar(current
, ret
);
967 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
968 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
970 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
972 ssize_t ret
= -EBADF
;
980 if (f
.file
->f_mode
& FMODE_PWRITE
)
981 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
);
986 add_wchar(current
, ret
);
993 static ssize_t
compat_do_readv_writev(int type
, struct file
*file
,
994 const struct compat_iovec __user
*uvector
,
995 unsigned long nr_segs
, loff_t
*pos
)
997 compat_ssize_t tot_len
;
998 struct iovec iovstack
[UIO_FASTIOV
];
999 struct iovec
*iov
= iovstack
;
1005 ret
= compat_rw_copy_check_uvector(type
, uvector
, nr_segs
,
1006 UIO_FASTIOV
, iovstack
, &iov
);
1011 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
1017 fn
= file
->f_op
->read
;
1018 fnv
= file
->f_op
->aio_read
;
1019 iter_fn
= file
->f_op
->read_iter
;
1021 fn
= (io_fn_t
)file
->f_op
->write
;
1022 fnv
= file
->f_op
->aio_write
;
1023 iter_fn
= file
->f_op
->write_iter
;
1024 file_start_write(file
);
1028 ret
= do_iter_readv_writev(file
, type
, iov
, nr_segs
, tot_len
,
1031 ret
= do_sync_readv_writev(file
, iov
, nr_segs
, tot_len
,
1034 ret
= do_loop_readv_writev(file
, iov
, nr_segs
, pos
, fn
);
1037 file_end_write(file
);
1040 if (iov
!= iovstack
)
1042 if ((ret
+ (type
== READ
)) > 0) {
1044 fsnotify_access(file
);
1046 fsnotify_modify(file
);
1051 static size_t compat_readv(struct file
*file
,
1052 const struct compat_iovec __user
*vec
,
1053 unsigned long vlen
, loff_t
*pos
)
1055 ssize_t ret
= -EBADF
;
1057 if (!(file
->f_mode
& FMODE_READ
))
1061 if (!(file
->f_mode
& FMODE_CAN_READ
))
1064 ret
= compat_do_readv_writev(READ
, file
, vec
, vlen
, pos
);
1068 add_rchar(current
, ret
);
1073 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1074 const struct compat_iovec __user
*,vec
,
1075 compat_ulong_t
, vlen
)
1077 struct fd f
= fdget_pos(fd
);
1083 pos
= f
.file
->f_pos
;
1084 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
);
1086 f
.file
->f_pos
= pos
;
1091 static long __compat_sys_preadv64(unsigned long fd
,
1092 const struct compat_iovec __user
*vec
,
1093 unsigned long vlen
, loff_t pos
)
1104 if (f
.file
->f_mode
& FMODE_PREAD
)
1105 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
);
1110 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1111 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1112 const struct compat_iovec __user
*,vec
,
1113 unsigned long, vlen
, loff_t
, pos
)
1115 return __compat_sys_preadv64(fd
, vec
, vlen
, pos
);
1119 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1120 const struct compat_iovec __user
*,vec
,
1121 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1123 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1125 return __compat_sys_preadv64(fd
, vec
, vlen
, pos
);
1128 static size_t compat_writev(struct file
*file
,
1129 const struct compat_iovec __user
*vec
,
1130 unsigned long vlen
, loff_t
*pos
)
1132 ssize_t ret
= -EBADF
;
1134 if (!(file
->f_mode
& FMODE_WRITE
))
1138 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1141 ret
= compat_do_readv_writev(WRITE
, file
, vec
, vlen
, pos
);
1145 add_wchar(current
, ret
);
1150 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1151 const struct compat_iovec __user
*, vec
,
1152 compat_ulong_t
, vlen
)
1154 struct fd f
= fdget_pos(fd
);
1160 pos
= f
.file
->f_pos
;
1161 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
);
1163 f
.file
->f_pos
= pos
;
1168 static long __compat_sys_pwritev64(unsigned long fd
,
1169 const struct compat_iovec __user
*vec
,
1170 unsigned long vlen
, loff_t pos
)
1181 if (f
.file
->f_mode
& FMODE_PWRITE
)
1182 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
);
1187 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1188 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1189 const struct compat_iovec __user
*,vec
,
1190 unsigned long, vlen
, loff_t
, pos
)
1192 return __compat_sys_pwritev64(fd
, vec
, vlen
, pos
);
1196 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1197 const struct compat_iovec __user
*,vec
,
1198 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1200 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1202 return __compat_sys_pwritev64(fd
, vec
, vlen
, pos
);
1206 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1207 size_t count
, loff_t max
)
1210 struct inode
*in_inode
, *out_inode
;
1217 * Get input file, and verify that it is ok..
1223 if (!(in
.file
->f_mode
& FMODE_READ
))
1227 pos
= in
.file
->f_pos
;
1230 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1233 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1239 * Get output file, and verify that it is ok..
1242 out
= fdget(out_fd
);
1245 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1248 in_inode
= file_inode(in
.file
);
1249 out_inode
= file_inode(out
.file
);
1250 out_pos
= out
.file
->f_pos
;
1251 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1257 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1259 if (unlikely(pos
+ count
> max
)) {
1260 retval
= -EOVERFLOW
;
1269 * We need to debate whether we can enable this or not. The
1270 * man page documents EAGAIN return for the output at least,
1271 * and the application is arguably buggy if it doesn't expect
1272 * EAGAIN on a non-blocking file descriptor.
1274 if (in
.file
->f_flags
& O_NONBLOCK
)
1275 fl
= SPLICE_F_NONBLOCK
;
1277 file_start_write(out
.file
);
1278 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1279 file_end_write(out
.file
);
1282 add_rchar(current
, retval
);
1283 add_wchar(current
, retval
);
1284 fsnotify_access(in
.file
);
1285 fsnotify_modify(out
.file
);
1286 out
.file
->f_pos
= out_pos
;
1290 in
.file
->f_pos
= pos
;
1296 retval
= -EOVERFLOW
;
1306 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1313 if (unlikely(get_user(off
, offset
)))
1316 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1317 if (unlikely(put_user(pos
, offset
)))
1322 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1325 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1331 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1333 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1334 if (unlikely(put_user(pos
, offset
)))
1339 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1342 #ifdef CONFIG_COMPAT
1343 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1344 compat_off_t __user
*, offset
, compat_size_t
, count
)
1351 if (unlikely(get_user(off
, offset
)))
1354 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1355 if (unlikely(put_user(pos
, offset
)))
1360 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1363 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1364 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1370 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1372 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1373 if (unlikely(put_user(pos
, offset
)))
1378 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);