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>
23 #include <asm/uaccess.h>
24 #include <asm/unistd.h>
26 typedef ssize_t (*io_fn_t
)(struct file
*, char __user
*, size_t, loff_t
*);
27 typedef ssize_t (*iter_fn_t
)(struct kiocb
*, struct iov_iter
*);
29 const struct file_operations generic_ro_fops
= {
30 .llseek
= generic_file_llseek
,
31 .read_iter
= generic_file_read_iter
,
32 .mmap
= generic_file_readonly_mmap
,
33 .splice_read
= generic_file_splice_read
,
36 EXPORT_SYMBOL(generic_ro_fops
);
38 static inline int unsigned_offsets(struct file
*file
)
40 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
53 * Return the specified offset on success and -EINVAL on invalid offset.
55 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
57 if (offset
< 0 && !unsigned_offsets(file
))
62 if (offset
!= file
->f_pos
) {
68 EXPORT_SYMBOL(vfs_setpos
);
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
87 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
88 loff_t maxsize
, loff_t eof
)
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
108 spin_lock(&file
->f_lock
);
109 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
110 spin_unlock(&file
->f_lock
);
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
131 return vfs_setpos(file
, offset
, maxsize
);
133 EXPORT_SYMBOL(generic_file_llseek_size
);
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
145 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
147 struct inode
*inode
= file
->f_mapping
->host
;
149 return generic_file_llseek_size(file
, offset
, whence
,
150 inode
->i_sb
->s_maxbytes
,
153 EXPORT_SYMBOL(generic_file_llseek
);
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
163 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
166 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
167 return generic_file_llseek_size(file
, offset
, whence
,
173 EXPORT_SYMBOL(fixed_size_llseek
);
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
182 loff_t
no_seek_end_llseek(struct file
*file
, loff_t offset
, int whence
)
185 case SEEK_SET
: case SEEK_CUR
:
186 return generic_file_llseek_size(file
, offset
, whence
,
192 EXPORT_SYMBOL(no_seek_end_llseek
);
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
202 loff_t
no_seek_end_llseek_size(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
205 case SEEK_SET
: case SEEK_CUR
:
206 return generic_file_llseek_size(file
, offset
, whence
,
212 EXPORT_SYMBOL(no_seek_end_llseek_size
);
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
225 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
229 EXPORT_SYMBOL(noop_llseek
);
231 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
235 EXPORT_SYMBOL(no_llseek
);
237 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
239 struct inode
*inode
= file_inode(file
);
245 offset
+= i_size_read(inode
);
249 retval
= file
->f_pos
;
252 offset
+= file
->f_pos
;
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
260 if (offset
>= inode
->i_size
) {
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
271 if (offset
>= inode
->i_size
) {
275 offset
= inode
->i_size
;
279 if (offset
>= 0 || unsigned_offsets(file
)) {
280 if (offset
!= file
->f_pos
) {
281 file
->f_pos
= offset
;
290 EXPORT_SYMBOL(default_llseek
);
292 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
294 loff_t (*fn
)(struct file
*, loff_t
, int);
297 if (file
->f_mode
& FMODE_LSEEK
) {
298 if (file
->f_op
->llseek
)
299 fn
= file
->f_op
->llseek
;
301 return fn(file
, offset
, whence
);
303 EXPORT_SYMBOL(vfs_llseek
);
305 static inline struct fd
fdget_pos(int fd
)
307 return __to_fd(__fdget_pos(fd
));
310 static inline void fdput_pos(struct fd f
)
312 if (f
.flags
& FDPUT_POS_UNLOCK
)
313 mutex_unlock(&f
.file
->f_pos_lock
);
317 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
320 struct fd f
= fdget_pos(fd
);
325 if (whence
<= SEEK_MAX
) {
326 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
328 if (res
!= (loff_t
)retval
)
329 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
336 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
338 return sys_lseek(fd
, offset
, whence
);
342 #ifdef __ARCH_WANT_SYS_LLSEEK
343 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
344 unsigned long, offset_low
, loff_t __user
*, result
,
345 unsigned int, whence
)
348 struct fd f
= fdget_pos(fd
);
355 if (whence
> SEEK_MAX
)
358 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
361 retval
= (int)offset
;
364 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
373 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
378 if (!file
->f_op
->read_iter
)
381 init_sync_kiocb(&kiocb
, file
);
382 kiocb
.ki_pos
= *ppos
;
385 ret
= file
->f_op
->read_iter(&kiocb
, iter
);
386 BUG_ON(ret
== -EIOCBQUEUED
);
388 *ppos
= kiocb
.ki_pos
;
391 EXPORT_SYMBOL(vfs_iter_read
);
393 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
398 if (!file
->f_op
->write_iter
)
401 init_sync_kiocb(&kiocb
, file
);
402 kiocb
.ki_pos
= *ppos
;
405 ret
= file
->f_op
->write_iter(&kiocb
, iter
);
406 BUG_ON(ret
== -EIOCBQUEUED
);
408 *ppos
= kiocb
.ki_pos
;
411 EXPORT_SYMBOL(vfs_iter_write
);
414 * rw_verify_area doesn't like huge counts. We limit
415 * them to something that fits in "int" so that others
416 * won't have to do range checks all the time.
418 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
422 int retval
= -EINVAL
;
424 inode
= file_inode(file
);
425 if (unlikely((ssize_t
) count
< 0))
428 if (unlikely(pos
< 0)) {
429 if (!unsigned_offsets(file
))
431 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
433 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
434 if (!unsigned_offsets(file
))
438 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
439 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
440 read_write
== READ
? F_RDLCK
: F_WRLCK
);
444 retval
= security_file_permission(file
,
445 read_write
== READ
? MAY_READ
: MAY_WRITE
);
448 return count
> MAX_RW_COUNT
? MAX_RW_COUNT
: count
;
451 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
453 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
455 struct iov_iter iter
;
458 init_sync_kiocb(&kiocb
, filp
);
459 kiocb
.ki_pos
= *ppos
;
460 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
462 ret
= filp
->f_op
->read_iter(&kiocb
, &iter
);
463 BUG_ON(ret
== -EIOCBQUEUED
);
464 *ppos
= kiocb
.ki_pos
;
468 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
471 if (file
->f_op
->read
)
472 return file
->f_op
->read(file
, buf
, count
, pos
);
473 else if (file
->f_op
->read_iter
)
474 return new_sync_read(file
, buf
, count
, pos
);
478 EXPORT_SYMBOL(__vfs_read
);
480 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
484 if (!(file
->f_mode
& FMODE_READ
))
486 if (!(file
->f_mode
& FMODE_CAN_READ
))
488 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
491 ret
= rw_verify_area(READ
, file
, pos
, count
);
494 ret
= __vfs_read(file
, buf
, count
, pos
);
496 fsnotify_access(file
);
497 add_rchar(current
, ret
);
505 EXPORT_SYMBOL(vfs_read
);
507 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
509 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
511 struct iov_iter iter
;
514 init_sync_kiocb(&kiocb
, filp
);
515 kiocb
.ki_pos
= *ppos
;
516 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
518 ret
= filp
->f_op
->write_iter(&kiocb
, &iter
);
519 BUG_ON(ret
== -EIOCBQUEUED
);
521 *ppos
= kiocb
.ki_pos
;
525 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
528 if (file
->f_op
->write
)
529 return file
->f_op
->write(file
, p
, count
, pos
);
530 else if (file
->f_op
->write_iter
)
531 return new_sync_write(file
, p
, count
, pos
);
535 EXPORT_SYMBOL(__vfs_write
);
537 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
540 const char __user
*p
;
543 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
548 p
= (__force
const char __user
*)buf
;
549 if (count
> MAX_RW_COUNT
)
550 count
= MAX_RW_COUNT
;
551 ret
= __vfs_write(file
, p
, count
, pos
);
554 fsnotify_modify(file
);
555 add_wchar(current
, ret
);
561 EXPORT_SYMBOL(__kernel_write
);
563 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
567 if (!(file
->f_mode
& FMODE_WRITE
))
569 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
571 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
574 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
577 file_start_write(file
);
578 ret
= __vfs_write(file
, buf
, count
, pos
);
580 fsnotify_modify(file
);
581 add_wchar(current
, ret
);
584 file_end_write(file
);
590 EXPORT_SYMBOL(vfs_write
);
592 static inline loff_t
file_pos_read(struct file
*file
)
597 static inline void file_pos_write(struct file
*file
, loff_t pos
)
602 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
604 struct fd f
= fdget_pos(fd
);
605 ssize_t ret
= -EBADF
;
608 loff_t pos
= file_pos_read(f
.file
);
609 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
611 file_pos_write(f
.file
, pos
);
617 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
620 struct fd f
= fdget_pos(fd
);
621 ssize_t ret
= -EBADF
;
624 loff_t pos
= file_pos_read(f
.file
);
625 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
627 file_pos_write(f
.file
, pos
);
634 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
635 size_t, count
, loff_t
, pos
)
638 ssize_t ret
= -EBADF
;
646 if (f
.file
->f_mode
& FMODE_PREAD
)
647 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
654 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
655 size_t, count
, loff_t
, pos
)
658 ssize_t ret
= -EBADF
;
666 if (f
.file
->f_mode
& FMODE_PWRITE
)
667 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
675 * Reduce an iovec's length in-place. Return the resulting number of segments
677 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
679 unsigned long seg
= 0;
682 while (seg
< nr_segs
) {
684 if (len
+ iov
->iov_len
>= to
) {
685 iov
->iov_len
= to
- len
;
693 EXPORT_SYMBOL(iov_shorten
);
695 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
696 loff_t
*ppos
, iter_fn_t fn
, int flags
)
701 if (flags
& ~RWF_HIPRI
)
704 init_sync_kiocb(&kiocb
, filp
);
705 if (flags
& RWF_HIPRI
)
706 kiocb
.ki_flags
|= IOCB_HIPRI
;
707 kiocb
.ki_pos
= *ppos
;
709 ret
= fn(&kiocb
, iter
);
710 BUG_ON(ret
== -EIOCBQUEUED
);
711 *ppos
= kiocb
.ki_pos
;
715 /* Do it by hand, with file-ops */
716 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
717 loff_t
*ppos
, io_fn_t fn
, int flags
)
721 if (flags
& ~RWF_HIPRI
)
724 while (iov_iter_count(iter
)) {
725 struct iovec iovec
= iov_iter_iovec(iter
);
728 nr
= fn(filp
, iovec
.iov_base
, iovec
.iov_len
, ppos
);
736 if (nr
!= iovec
.iov_len
)
738 iov_iter_advance(iter
, nr
);
744 /* A write operation does a read from user space and vice versa */
745 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
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
;
822 static ssize_t
do_readv_writev(int type
, struct file
*file
,
823 const struct iovec __user
* uvector
,
824 unsigned long nr_segs
, loff_t
*pos
,
828 struct iovec iovstack
[UIO_FASTIOV
];
829 struct iovec
*iov
= iovstack
;
830 struct iov_iter iter
;
835 ret
= import_iovec(type
, uvector
, nr_segs
,
836 ARRAY_SIZE(iovstack
), &iov
, &iter
);
840 tot_len
= iov_iter_count(&iter
);
843 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
848 fn
= file
->f_op
->read
;
849 iter_fn
= file
->f_op
->read_iter
;
851 fn
= (io_fn_t
)file
->f_op
->write
;
852 iter_fn
= file
->f_op
->write_iter
;
853 file_start_write(file
);
857 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
, flags
);
859 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
, flags
);
862 file_end_write(file
);
866 if ((ret
+ (type
== READ
)) > 0) {
868 fsnotify_access(file
);
870 fsnotify_modify(file
);
875 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
876 unsigned long vlen
, loff_t
*pos
, int flags
)
878 if (!(file
->f_mode
& FMODE_READ
))
880 if (!(file
->f_mode
& FMODE_CAN_READ
))
883 return do_readv_writev(READ
, file
, vec
, vlen
, pos
, flags
);
886 EXPORT_SYMBOL(vfs_readv
);
888 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
889 unsigned long vlen
, loff_t
*pos
, int flags
)
891 if (!(file
->f_mode
& FMODE_WRITE
))
893 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
896 return do_readv_writev(WRITE
, file
, vec
, vlen
, pos
, flags
);
899 EXPORT_SYMBOL(vfs_writev
);
901 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
902 unsigned long vlen
, int flags
)
904 struct fd f
= fdget_pos(fd
);
905 ssize_t ret
= -EBADF
;
908 loff_t pos
= file_pos_read(f
.file
);
909 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
911 file_pos_write(f
.file
, pos
);
916 add_rchar(current
, ret
);
921 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
922 unsigned long vlen
, int flags
)
924 struct fd f
= fdget_pos(fd
);
925 ssize_t ret
= -EBADF
;
928 loff_t pos
= file_pos_read(f
.file
);
929 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
931 file_pos_write(f
.file
, pos
);
936 add_wchar(current
, ret
);
941 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
943 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
944 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
947 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
948 unsigned long vlen
, loff_t pos
, int flags
)
951 ssize_t ret
= -EBADF
;
959 if (f
.file
->f_mode
& FMODE_PREAD
)
960 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
965 add_rchar(current
, ret
);
970 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
971 unsigned long vlen
, loff_t pos
, int flags
)
974 ssize_t ret
= -EBADF
;
982 if (f
.file
->f_mode
& FMODE_PWRITE
)
983 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
988 add_wchar(current
, ret
);
993 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
996 return do_readv(fd
, vec
, vlen
, 0);
999 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1000 unsigned long, vlen
)
1002 return do_writev(fd
, vec
, vlen
, 0);
1005 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1006 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1008 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1010 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1013 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1014 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1017 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1020 return do_readv(fd
, vec
, vlen
, flags
);
1022 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1025 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1026 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1028 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1030 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1033 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1034 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1037 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1040 return do_writev(fd
, vec
, vlen
, flags
);
1042 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1045 #ifdef CONFIG_COMPAT
1047 static ssize_t
compat_do_readv_writev(int type
, struct file
*file
,
1048 const struct compat_iovec __user
*uvector
,
1049 unsigned long nr_segs
, loff_t
*pos
,
1052 compat_ssize_t tot_len
;
1053 struct iovec iovstack
[UIO_FASTIOV
];
1054 struct iovec
*iov
= iovstack
;
1055 struct iov_iter iter
;
1060 ret
= compat_import_iovec(type
, uvector
, nr_segs
,
1061 UIO_FASTIOV
, &iov
, &iter
);
1065 tot_len
= iov_iter_count(&iter
);
1068 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
1073 fn
= file
->f_op
->read
;
1074 iter_fn
= file
->f_op
->read_iter
;
1076 fn
= (io_fn_t
)file
->f_op
->write
;
1077 iter_fn
= file
->f_op
->write_iter
;
1078 file_start_write(file
);
1082 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
, flags
);
1084 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
, flags
);
1087 file_end_write(file
);
1091 if ((ret
+ (type
== READ
)) > 0) {
1093 fsnotify_access(file
);
1095 fsnotify_modify(file
);
1100 static size_t compat_readv(struct file
*file
,
1101 const struct compat_iovec __user
*vec
,
1102 unsigned long vlen
, loff_t
*pos
, int flags
)
1104 ssize_t ret
= -EBADF
;
1106 if (!(file
->f_mode
& FMODE_READ
))
1110 if (!(file
->f_mode
& FMODE_CAN_READ
))
1113 ret
= compat_do_readv_writev(READ
, file
, vec
, vlen
, pos
, flags
);
1117 add_rchar(current
, ret
);
1122 static size_t do_compat_readv(compat_ulong_t fd
,
1123 const struct compat_iovec __user
*vec
,
1124 compat_ulong_t vlen
, int flags
)
1126 struct fd f
= fdget_pos(fd
);
1132 pos
= f
.file
->f_pos
;
1133 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1135 f
.file
->f_pos
= pos
;
1141 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1142 const struct compat_iovec __user
*,vec
,
1143 compat_ulong_t
, vlen
)
1145 return do_compat_readv(fd
, vec
, vlen
, 0);
1148 static long do_compat_preadv64(unsigned long fd
,
1149 const struct compat_iovec __user
*vec
,
1150 unsigned long vlen
, loff_t pos
, int flags
)
1161 if (f
.file
->f_mode
& FMODE_PREAD
)
1162 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1167 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1168 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1169 const struct compat_iovec __user
*,vec
,
1170 unsigned long, vlen
, loff_t
, pos
)
1172 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1176 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1177 const struct compat_iovec __user
*,vec
,
1178 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1180 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1182 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1185 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1186 const struct compat_iovec __user
*,vec
,
1187 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1190 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1193 return do_compat_readv(fd
, vec
, vlen
, flags
);
1195 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1198 static size_t compat_writev(struct file
*file
,
1199 const struct compat_iovec __user
*vec
,
1200 unsigned long vlen
, loff_t
*pos
, int flags
)
1202 ssize_t ret
= -EBADF
;
1204 if (!(file
->f_mode
& FMODE_WRITE
))
1208 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1211 ret
= compat_do_readv_writev(WRITE
, file
, vec
, vlen
, pos
, 0);
1215 add_wchar(current
, ret
);
1220 static size_t do_compat_writev(compat_ulong_t fd
,
1221 const struct compat_iovec __user
* vec
,
1222 compat_ulong_t vlen
, int flags
)
1224 struct fd f
= fdget_pos(fd
);
1230 pos
= f
.file
->f_pos
;
1231 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1233 f
.file
->f_pos
= pos
;
1238 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1239 const struct compat_iovec __user
*, vec
,
1240 compat_ulong_t
, vlen
)
1242 return do_compat_writev(fd
, vec
, vlen
, 0);
1245 static long do_compat_pwritev64(unsigned long fd
,
1246 const struct compat_iovec __user
*vec
,
1247 unsigned long vlen
, loff_t pos
, int flags
)
1258 if (f
.file
->f_mode
& FMODE_PWRITE
)
1259 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1264 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1265 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1266 const struct compat_iovec __user
*,vec
,
1267 unsigned long, vlen
, loff_t
, pos
)
1269 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1273 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1274 const struct compat_iovec __user
*,vec
,
1275 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1277 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1279 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1282 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1283 const struct compat_iovec __user
*,vec
,
1284 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, int, flags
)
1286 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1289 return do_compat_writev(fd
, vec
, vlen
, flags
);
1291 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1296 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1297 size_t count
, loff_t max
)
1300 struct inode
*in_inode
, *out_inode
;
1307 * Get input file, and verify that it is ok..
1313 if (!(in
.file
->f_mode
& FMODE_READ
))
1317 pos
= in
.file
->f_pos
;
1320 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1323 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1329 * Get output file, and verify that it is ok..
1332 out
= fdget(out_fd
);
1335 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1338 in_inode
= file_inode(in
.file
);
1339 out_inode
= file_inode(out
.file
);
1340 out_pos
= out
.file
->f_pos
;
1341 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1347 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1349 if (unlikely(pos
+ count
> max
)) {
1350 retval
= -EOVERFLOW
;
1359 * We need to debate whether we can enable this or not. The
1360 * man page documents EAGAIN return for the output at least,
1361 * and the application is arguably buggy if it doesn't expect
1362 * EAGAIN on a non-blocking file descriptor.
1364 if (in
.file
->f_flags
& O_NONBLOCK
)
1365 fl
= SPLICE_F_NONBLOCK
;
1367 file_start_write(out
.file
);
1368 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1369 file_end_write(out
.file
);
1372 add_rchar(current
, retval
);
1373 add_wchar(current
, retval
);
1374 fsnotify_access(in
.file
);
1375 fsnotify_modify(out
.file
);
1376 out
.file
->f_pos
= out_pos
;
1380 in
.file
->f_pos
= pos
;
1386 retval
= -EOVERFLOW
;
1396 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1403 if (unlikely(get_user(off
, offset
)))
1406 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1407 if (unlikely(put_user(pos
, offset
)))
1412 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1415 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1421 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1423 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1424 if (unlikely(put_user(pos
, offset
)))
1429 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1432 #ifdef CONFIG_COMPAT
1433 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1434 compat_off_t __user
*, offset
, compat_size_t
, count
)
1441 if (unlikely(get_user(off
, offset
)))
1444 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1445 if (unlikely(put_user(pos
, offset
)))
1450 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1453 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1454 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1460 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1462 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1463 if (unlikely(put_user(pos
, offset
)))
1468 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1473 * copy_file_range() differs from regular file read and write in that it
1474 * specifically allows return partial success. When it does so is up to
1475 * the copy_file_range method.
1477 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1478 struct file
*file_out
, loff_t pos_out
,
1479 size_t len
, unsigned int flags
)
1481 struct inode
*inode_in
= file_inode(file_in
);
1482 struct inode
*inode_out
= file_inode(file_out
);
1488 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1489 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1491 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1495 if (!(file_in
->f_mode
& FMODE_READ
) ||
1496 !(file_out
->f_mode
& FMODE_WRITE
) ||
1497 (file_out
->f_flags
& O_APPEND
))
1500 /* this could be relaxed once a method supports cross-fs copies */
1501 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1507 ret
= mnt_want_write_file(file_out
);
1512 if (file_out
->f_op
->copy_file_range
)
1513 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1514 pos_out
, len
, flags
);
1515 if (ret
== -EOPNOTSUPP
)
1516 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1517 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1520 fsnotify_access(file_in
);
1521 add_rchar(current
, ret
);
1522 fsnotify_modify(file_out
);
1523 add_wchar(current
, ret
);
1528 mnt_drop_write_file(file_out
);
1532 EXPORT_SYMBOL(vfs_copy_file_range
);
1534 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1535 int, fd_out
, loff_t __user
*, off_out
,
1536 size_t, len
, unsigned int, flags
)
1542 ssize_t ret
= -EBADF
;
1544 f_in
= fdget(fd_in
);
1548 f_out
= fdget(fd_out
);
1554 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1557 pos_in
= f_in
.file
->f_pos
;
1561 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1564 pos_out
= f_out
.file
->f_pos
;
1567 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1574 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1577 f_in
.file
->f_pos
= pos_in
;
1581 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1584 f_out
.file
->f_pos
= pos_out
;
1596 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1598 struct inode
*inode
= file_inode(file
);
1600 if (unlikely(pos
< 0))
1603 if (unlikely((loff_t
) (pos
+ len
) < 0))
1606 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1607 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1610 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1611 write
? F_WRLCK
: F_RDLCK
);
1616 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1619 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1620 struct file
*file_out
, loff_t pos_out
, u64 len
)
1622 struct inode
*inode_in
= file_inode(file_in
);
1623 struct inode
*inode_out
= file_inode(file_out
);
1626 if (inode_in
->i_sb
!= inode_out
->i_sb
||
1627 file_in
->f_path
.mnt
!= file_out
->f_path
.mnt
)
1630 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1632 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1635 if (!(file_in
->f_mode
& FMODE_READ
) ||
1636 !(file_out
->f_mode
& FMODE_WRITE
) ||
1637 (file_out
->f_flags
& O_APPEND
))
1640 if (!file_in
->f_op
->clone_file_range
)
1643 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1647 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1651 if (pos_in
+ len
> i_size_read(inode_in
))
1654 ret
= mnt_want_write_file(file_out
);
1658 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1659 file_out
, pos_out
, len
);
1661 fsnotify_access(file_in
);
1662 fsnotify_modify(file_out
);
1665 mnt_drop_write_file(file_out
);
1668 EXPORT_SYMBOL(vfs_clone_file_range
);
1670 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1672 struct file_dedupe_range_info
*info
;
1673 struct inode
*src
= file_inode(file
);
1678 bool is_admin
= capable(CAP_SYS_ADMIN
);
1679 u16 count
= same
->dest_count
;
1680 struct file
*dst_file
;
1684 if (!(file
->f_mode
& FMODE_READ
))
1687 if (same
->reserved1
|| same
->reserved2
)
1690 off
= same
->src_offset
;
1691 len
= same
->src_length
;
1694 if (S_ISDIR(src
->i_mode
))
1698 if (!S_ISREG(src
->i_mode
))
1701 ret
= clone_verify_area(file
, off
, len
, false);
1706 /* pre-format output fields to sane values */
1707 for (i
= 0; i
< count
; i
++) {
1708 same
->info
[i
].bytes_deduped
= 0ULL;
1709 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
1712 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
1714 struct fd dst_fd
= fdget(info
->dest_fd
);
1716 dst_file
= dst_fd
.file
;
1718 info
->status
= -EBADF
;
1721 dst
= file_inode(dst_file
);
1723 ret
= mnt_want_write_file(dst_file
);
1729 dst_off
= info
->dest_offset
;
1730 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
1737 if (info
->reserved
) {
1738 info
->status
= -EINVAL
;
1739 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
1740 info
->status
= -EINVAL
;
1741 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
1742 info
->status
= -EXDEV
;
1743 } else if (S_ISDIR(dst
->i_mode
)) {
1744 info
->status
= -EISDIR
;
1745 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
1746 info
->status
= -EINVAL
;
1748 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
1751 if (deduped
== -EBADE
)
1752 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
1753 else if (deduped
< 0)
1754 info
->status
= deduped
;
1756 info
->bytes_deduped
+= deduped
;
1760 mnt_drop_write_file(dst_file
);
1764 if (fatal_signal_pending(current
))
1771 EXPORT_SYMBOL(vfs_dedupe_file_range
);