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 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
308 struct fd f
= fdget_pos(fd
);
313 if (whence
<= SEEK_MAX
) {
314 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
316 if (res
!= (loff_t
)retval
)
317 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
324 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
326 return sys_lseek(fd
, offset
, whence
);
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
332 unsigned long, offset_low
, loff_t __user
*, result
,
333 unsigned int, whence
)
336 struct fd f
= fdget_pos(fd
);
343 if (whence
> SEEK_MAX
)
346 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
349 retval
= (int)offset
;
352 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
361 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
366 if (!file
->f_op
->read_iter
)
369 init_sync_kiocb(&kiocb
, file
);
370 kiocb
.ki_pos
= *ppos
;
373 ret
= file
->f_op
->read_iter(&kiocb
, iter
);
374 BUG_ON(ret
== -EIOCBQUEUED
);
376 *ppos
= kiocb
.ki_pos
;
379 EXPORT_SYMBOL(vfs_iter_read
);
381 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
)
386 if (!file
->f_op
->write_iter
)
389 init_sync_kiocb(&kiocb
, file
);
390 kiocb
.ki_pos
= *ppos
;
393 ret
= file
->f_op
->write_iter(&kiocb
, iter
);
394 BUG_ON(ret
== -EIOCBQUEUED
);
396 *ppos
= kiocb
.ki_pos
;
399 EXPORT_SYMBOL(vfs_iter_write
);
402 * rw_verify_area doesn't like huge counts. We limit
403 * them to something that fits in "int" so that others
404 * won't have to do range checks all the time.
406 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
410 int retval
= -EINVAL
;
412 inode
= file_inode(file
);
413 if (unlikely((ssize_t
) count
< 0))
416 if (unlikely(pos
< 0)) {
417 if (!unsigned_offsets(file
))
419 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
421 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
422 if (!unsigned_offsets(file
))
426 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
427 retval
= locks_mandatory_area(inode
, file
, pos
, pos
+ count
- 1,
428 read_write
== READ
? F_RDLCK
: F_WRLCK
);
432 retval
= security_file_permission(file
,
433 read_write
== READ
? MAY_READ
: MAY_WRITE
);
436 return count
> MAX_RW_COUNT
? MAX_RW_COUNT
: count
;
439 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
441 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
443 struct iov_iter iter
;
446 init_sync_kiocb(&kiocb
, filp
);
447 kiocb
.ki_pos
= *ppos
;
448 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
450 ret
= filp
->f_op
->read_iter(&kiocb
, &iter
);
451 BUG_ON(ret
== -EIOCBQUEUED
);
452 *ppos
= kiocb
.ki_pos
;
456 ssize_t
__vfs_read(struct file
*file
, char __user
*buf
, size_t count
,
459 if (file
->f_op
->read
)
460 return file
->f_op
->read(file
, buf
, count
, pos
);
461 else if (file
->f_op
->read_iter
)
462 return new_sync_read(file
, buf
, count
, pos
);
466 EXPORT_SYMBOL(__vfs_read
);
468 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
472 if (!(file
->f_mode
& FMODE_READ
))
474 if (!(file
->f_mode
& FMODE_CAN_READ
))
476 if (unlikely(!access_ok(VERIFY_WRITE
, buf
, count
)))
479 ret
= rw_verify_area(READ
, file
, pos
, count
);
482 ret
= __vfs_read(file
, buf
, count
, pos
);
484 fsnotify_access(file
);
485 add_rchar(current
, ret
);
493 EXPORT_SYMBOL(vfs_read
);
495 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
497 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
499 struct iov_iter iter
;
502 init_sync_kiocb(&kiocb
, filp
);
503 kiocb
.ki_pos
= *ppos
;
504 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
506 ret
= filp
->f_op
->write_iter(&kiocb
, &iter
);
507 BUG_ON(ret
== -EIOCBQUEUED
);
509 *ppos
= kiocb
.ki_pos
;
513 ssize_t
__vfs_write(struct file
*file
, const char __user
*p
, size_t count
,
516 if (file
->f_op
->write
)
517 return file
->f_op
->write(file
, p
, count
, pos
);
518 else if (file
->f_op
->write_iter
)
519 return new_sync_write(file
, p
, count
, pos
);
523 EXPORT_SYMBOL(__vfs_write
);
525 ssize_t
__kernel_write(struct file
*file
, const char *buf
, size_t count
, loff_t
*pos
)
528 const char __user
*p
;
531 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
536 p
= (__force
const char __user
*)buf
;
537 if (count
> MAX_RW_COUNT
)
538 count
= MAX_RW_COUNT
;
539 ret
= __vfs_write(file
, p
, count
, pos
);
542 fsnotify_modify(file
);
543 add_wchar(current
, ret
);
549 EXPORT_SYMBOL(__kernel_write
);
551 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
555 if (!(file
->f_mode
& FMODE_WRITE
))
557 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
559 if (unlikely(!access_ok(VERIFY_READ
, buf
, count
)))
562 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
565 file_start_write(file
);
566 ret
= __vfs_write(file
, buf
, count
, pos
);
568 fsnotify_modify(file
);
569 add_wchar(current
, ret
);
572 file_end_write(file
);
578 EXPORT_SYMBOL(vfs_write
);
580 static inline loff_t
file_pos_read(struct file
*file
)
585 static inline void file_pos_write(struct file
*file
, loff_t pos
)
590 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
592 struct fd f
= fdget_pos(fd
);
593 ssize_t ret
= -EBADF
;
596 loff_t pos
= file_pos_read(f
.file
);
597 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
599 file_pos_write(f
.file
, pos
);
605 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
608 struct fd f
= fdget_pos(fd
);
609 ssize_t ret
= -EBADF
;
612 loff_t pos
= file_pos_read(f
.file
);
613 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
615 file_pos_write(f
.file
, pos
);
622 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
623 size_t, count
, loff_t
, pos
)
626 ssize_t ret
= -EBADF
;
634 if (f
.file
->f_mode
& FMODE_PREAD
)
635 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
642 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
643 size_t, count
, loff_t
, pos
)
646 ssize_t ret
= -EBADF
;
654 if (f
.file
->f_mode
& FMODE_PWRITE
)
655 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
663 * Reduce an iovec's length in-place. Return the resulting number of segments
665 unsigned long iov_shorten(struct iovec
*iov
, unsigned long nr_segs
, size_t to
)
667 unsigned long seg
= 0;
670 while (seg
< nr_segs
) {
672 if (len
+ iov
->iov_len
>= to
) {
673 iov
->iov_len
= to
- len
;
681 EXPORT_SYMBOL(iov_shorten
);
683 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
684 loff_t
*ppos
, iter_fn_t fn
, int flags
)
689 if (flags
& ~(RWF_HIPRI
| RWF_DSYNC
| RWF_SYNC
))
692 init_sync_kiocb(&kiocb
, filp
);
693 if (flags
& RWF_HIPRI
)
694 kiocb
.ki_flags
|= IOCB_HIPRI
;
695 if (flags
& RWF_DSYNC
)
696 kiocb
.ki_flags
|= IOCB_DSYNC
;
697 if (flags
& RWF_SYNC
)
698 kiocb
.ki_flags
|= (IOCB_DSYNC
| IOCB_SYNC
);
699 kiocb
.ki_pos
= *ppos
;
701 ret
= fn(&kiocb
, iter
);
702 BUG_ON(ret
== -EIOCBQUEUED
);
703 *ppos
= kiocb
.ki_pos
;
707 /* Do it by hand, with file-ops */
708 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
709 loff_t
*ppos
, io_fn_t fn
, int flags
)
713 if (flags
& ~RWF_HIPRI
)
716 while (iov_iter_count(iter
)) {
717 struct iovec iovec
= iov_iter_iovec(iter
);
720 nr
= fn(filp
, iovec
.iov_base
, iovec
.iov_len
, ppos
);
728 if (nr
!= iovec
.iov_len
)
730 iov_iter_advance(iter
, nr
);
736 /* A write operation does a read from user space and vice versa */
737 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
739 ssize_t
rw_copy_check_uvector(int type
, const struct iovec __user
* uvector
,
740 unsigned long nr_segs
, unsigned long fast_segs
,
741 struct iovec
*fast_pointer
,
742 struct iovec
**ret_pointer
)
746 struct iovec
*iov
= fast_pointer
;
749 * SuS says "The readv() function *may* fail if the iovcnt argument
750 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
751 * traditionally returned zero for zero segments, so...
759 * First get the "struct iovec" from user memory and
760 * verify all the pointers
762 if (nr_segs
> UIO_MAXIOV
) {
766 if (nr_segs
> fast_segs
) {
767 iov
= kmalloc(nr_segs
*sizeof(struct iovec
), GFP_KERNEL
);
773 if (copy_from_user(iov
, uvector
, nr_segs
*sizeof(*uvector
))) {
779 * According to the Single Unix Specification we should return EINVAL
780 * if an element length is < 0 when cast to ssize_t or if the
781 * total length would overflow the ssize_t return value of the
784 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
788 for (seg
= 0; seg
< nr_segs
; seg
++) {
789 void __user
*buf
= iov
[seg
].iov_base
;
790 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
792 /* see if we we're about to use an invalid len or if
793 * it's about to overflow ssize_t */
799 && unlikely(!access_ok(vrfy_dir(type
), buf
, len
))) {
803 if (len
> MAX_RW_COUNT
- ret
) {
804 len
= MAX_RW_COUNT
- ret
;
805 iov
[seg
].iov_len
= len
;
814 static ssize_t
do_readv_writev(int type
, struct file
*file
,
815 const struct iovec __user
* uvector
,
816 unsigned long nr_segs
, loff_t
*pos
,
820 struct iovec iovstack
[UIO_FASTIOV
];
821 struct iovec
*iov
= iovstack
;
822 struct iov_iter iter
;
827 ret
= import_iovec(type
, uvector
, nr_segs
,
828 ARRAY_SIZE(iovstack
), &iov
, &iter
);
832 tot_len
= iov_iter_count(&iter
);
835 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
840 fn
= file
->f_op
->read
;
841 iter_fn
= file
->f_op
->read_iter
;
843 fn
= (io_fn_t
)file
->f_op
->write
;
844 iter_fn
= file
->f_op
->write_iter
;
845 file_start_write(file
);
849 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
, flags
);
851 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
, flags
);
854 file_end_write(file
);
858 if ((ret
+ (type
== READ
)) > 0) {
860 fsnotify_access(file
);
862 fsnotify_modify(file
);
867 ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
868 unsigned long vlen
, loff_t
*pos
, int flags
)
870 if (!(file
->f_mode
& FMODE_READ
))
872 if (!(file
->f_mode
& FMODE_CAN_READ
))
875 return do_readv_writev(READ
, file
, vec
, vlen
, pos
, flags
);
878 EXPORT_SYMBOL(vfs_readv
);
880 ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
881 unsigned long vlen
, loff_t
*pos
, int flags
)
883 if (!(file
->f_mode
& FMODE_WRITE
))
885 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
888 return do_readv_writev(WRITE
, file
, vec
, vlen
, pos
, flags
);
891 EXPORT_SYMBOL(vfs_writev
);
893 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
894 unsigned long vlen
, int flags
)
896 struct fd f
= fdget_pos(fd
);
897 ssize_t ret
= -EBADF
;
900 loff_t pos
= file_pos_read(f
.file
);
901 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
903 file_pos_write(f
.file
, pos
);
908 add_rchar(current
, ret
);
913 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
914 unsigned long vlen
, int flags
)
916 struct fd f
= fdget_pos(fd
);
917 ssize_t ret
= -EBADF
;
920 loff_t pos
= file_pos_read(f
.file
);
921 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
923 file_pos_write(f
.file
, pos
);
928 add_wchar(current
, ret
);
933 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
935 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
936 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
939 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
940 unsigned long vlen
, loff_t pos
, int flags
)
943 ssize_t ret
= -EBADF
;
951 if (f
.file
->f_mode
& FMODE_PREAD
)
952 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
957 add_rchar(current
, ret
);
962 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
963 unsigned long vlen
, loff_t pos
, int flags
)
966 ssize_t ret
= -EBADF
;
974 if (f
.file
->f_mode
& FMODE_PWRITE
)
975 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
980 add_wchar(current
, ret
);
985 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
988 return do_readv(fd
, vec
, vlen
, 0);
991 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
994 return do_writev(fd
, vec
, vlen
, 0);
997 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
998 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1000 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1002 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1005 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1006 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1009 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1012 return do_readv(fd
, vec
, vlen
, flags
);
1014 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1017 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1018 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1020 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1022 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1025 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1026 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1029 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1032 return do_writev(fd
, vec
, vlen
, flags
);
1034 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1037 #ifdef CONFIG_COMPAT
1039 static ssize_t
compat_do_readv_writev(int type
, struct file
*file
,
1040 const struct compat_iovec __user
*uvector
,
1041 unsigned long nr_segs
, loff_t
*pos
,
1044 compat_ssize_t tot_len
;
1045 struct iovec iovstack
[UIO_FASTIOV
];
1046 struct iovec
*iov
= iovstack
;
1047 struct iov_iter iter
;
1052 ret
= compat_import_iovec(type
, uvector
, nr_segs
,
1053 UIO_FASTIOV
, &iov
, &iter
);
1057 tot_len
= iov_iter_count(&iter
);
1060 ret
= rw_verify_area(type
, file
, pos
, tot_len
);
1065 fn
= file
->f_op
->read
;
1066 iter_fn
= file
->f_op
->read_iter
;
1068 fn
= (io_fn_t
)file
->f_op
->write
;
1069 iter_fn
= file
->f_op
->write_iter
;
1070 file_start_write(file
);
1074 ret
= do_iter_readv_writev(file
, &iter
, pos
, iter_fn
, flags
);
1076 ret
= do_loop_readv_writev(file
, &iter
, pos
, fn
, flags
);
1079 file_end_write(file
);
1083 if ((ret
+ (type
== READ
)) > 0) {
1085 fsnotify_access(file
);
1087 fsnotify_modify(file
);
1092 static size_t compat_readv(struct file
*file
,
1093 const struct compat_iovec __user
*vec
,
1094 unsigned long vlen
, loff_t
*pos
, int flags
)
1096 ssize_t ret
= -EBADF
;
1098 if (!(file
->f_mode
& FMODE_READ
))
1102 if (!(file
->f_mode
& FMODE_CAN_READ
))
1105 ret
= compat_do_readv_writev(READ
, file
, vec
, vlen
, pos
, flags
);
1109 add_rchar(current
, ret
);
1114 static size_t do_compat_readv(compat_ulong_t fd
,
1115 const struct compat_iovec __user
*vec
,
1116 compat_ulong_t vlen
, int flags
)
1118 struct fd f
= fdget_pos(fd
);
1124 pos
= f
.file
->f_pos
;
1125 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1127 f
.file
->f_pos
= pos
;
1133 COMPAT_SYSCALL_DEFINE3(readv
, compat_ulong_t
, fd
,
1134 const struct compat_iovec __user
*,vec
,
1135 compat_ulong_t
, vlen
)
1137 return do_compat_readv(fd
, vec
, vlen
, 0);
1140 static long do_compat_preadv64(unsigned long fd
,
1141 const struct compat_iovec __user
*vec
,
1142 unsigned long vlen
, loff_t pos
, int flags
)
1153 if (f
.file
->f_mode
& FMODE_PREAD
)
1154 ret
= compat_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1159 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1160 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1161 const struct compat_iovec __user
*,vec
,
1162 unsigned long, vlen
, loff_t
, pos
)
1164 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1168 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1169 const struct compat_iovec __user
*,vec
,
1170 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1172 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1174 return do_compat_preadv64(fd
, vec
, vlen
, pos
, 0);
1177 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1178 const struct compat_iovec __user
*,vec
,
1179 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1182 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1185 return do_compat_readv(fd
, vec
, vlen
, flags
);
1187 return do_compat_preadv64(fd
, vec
, vlen
, pos
, flags
);
1190 static size_t compat_writev(struct file
*file
,
1191 const struct compat_iovec __user
*vec
,
1192 unsigned long vlen
, loff_t
*pos
, int flags
)
1194 ssize_t ret
= -EBADF
;
1196 if (!(file
->f_mode
& FMODE_WRITE
))
1200 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
1203 ret
= compat_do_readv_writev(WRITE
, file
, vec
, vlen
, pos
, 0);
1207 add_wchar(current
, ret
);
1212 static size_t do_compat_writev(compat_ulong_t fd
,
1213 const struct compat_iovec __user
* vec
,
1214 compat_ulong_t vlen
, int flags
)
1216 struct fd f
= fdget_pos(fd
);
1222 pos
= f
.file
->f_pos
;
1223 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1225 f
.file
->f_pos
= pos
;
1230 COMPAT_SYSCALL_DEFINE3(writev
, compat_ulong_t
, fd
,
1231 const struct compat_iovec __user
*, vec
,
1232 compat_ulong_t
, vlen
)
1234 return do_compat_writev(fd
, vec
, vlen
, 0);
1237 static long do_compat_pwritev64(unsigned long fd
,
1238 const struct compat_iovec __user
*vec
,
1239 unsigned long vlen
, loff_t pos
, int flags
)
1250 if (f
.file
->f_mode
& FMODE_PWRITE
)
1251 ret
= compat_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1256 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1257 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1258 const struct compat_iovec __user
*,vec
,
1259 unsigned long, vlen
, loff_t
, pos
)
1261 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1265 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1266 const struct compat_iovec __user
*,vec
,
1267 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1269 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1271 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, 0);
1274 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1275 const struct compat_iovec __user
*,vec
,
1276 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, int, flags
)
1278 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1281 return do_compat_writev(fd
, vec
, vlen
, flags
);
1283 return do_compat_pwritev64(fd
, vec
, vlen
, pos
, flags
);
1288 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1289 size_t count
, loff_t max
)
1292 struct inode
*in_inode
, *out_inode
;
1299 * Get input file, and verify that it is ok..
1305 if (!(in
.file
->f_mode
& FMODE_READ
))
1309 pos
= in
.file
->f_pos
;
1312 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1315 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1321 * Get output file, and verify that it is ok..
1324 out
= fdget(out_fd
);
1327 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1330 in_inode
= file_inode(in
.file
);
1331 out_inode
= file_inode(out
.file
);
1332 out_pos
= out
.file
->f_pos
;
1333 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1339 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1341 if (unlikely(pos
+ count
> max
)) {
1342 retval
= -EOVERFLOW
;
1351 * We need to debate whether we can enable this or not. The
1352 * man page documents EAGAIN return for the output at least,
1353 * and the application is arguably buggy if it doesn't expect
1354 * EAGAIN on a non-blocking file descriptor.
1356 if (in
.file
->f_flags
& O_NONBLOCK
)
1357 fl
= SPLICE_F_NONBLOCK
;
1359 file_start_write(out
.file
);
1360 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
, count
, fl
);
1361 file_end_write(out
.file
);
1364 add_rchar(current
, retval
);
1365 add_wchar(current
, retval
);
1366 fsnotify_access(in
.file
);
1367 fsnotify_modify(out
.file
);
1368 out
.file
->f_pos
= out_pos
;
1372 in
.file
->f_pos
= pos
;
1378 retval
= -EOVERFLOW
;
1388 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1395 if (unlikely(get_user(off
, offset
)))
1398 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1399 if (unlikely(put_user(pos
, offset
)))
1404 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1407 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1413 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1415 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1416 if (unlikely(put_user(pos
, offset
)))
1421 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1424 #ifdef CONFIG_COMPAT
1425 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1426 compat_off_t __user
*, offset
, compat_size_t
, count
)
1433 if (unlikely(get_user(off
, offset
)))
1436 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1437 if (unlikely(put_user(pos
, offset
)))
1442 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1445 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1446 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1452 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1454 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1455 if (unlikely(put_user(pos
, offset
)))
1460 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1465 * copy_file_range() differs from regular file read and write in that it
1466 * specifically allows return partial success. When it does so is up to
1467 * the copy_file_range method.
1469 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1470 struct file
*file_out
, loff_t pos_out
,
1471 size_t len
, unsigned int flags
)
1473 struct inode
*inode_in
= file_inode(file_in
);
1474 struct inode
*inode_out
= file_inode(file_out
);
1480 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1481 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1483 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1487 if (!(file_in
->f_mode
& FMODE_READ
) ||
1488 !(file_out
->f_mode
& FMODE_WRITE
) ||
1489 (file_out
->f_flags
& O_APPEND
))
1492 /* this could be relaxed once a method supports cross-fs copies */
1493 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1499 ret
= mnt_want_write_file(file_out
);
1504 if (file_out
->f_op
->copy_file_range
)
1505 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
, file_out
,
1506 pos_out
, len
, flags
);
1507 if (ret
== -EOPNOTSUPP
)
1508 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1509 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1512 fsnotify_access(file_in
);
1513 add_rchar(current
, ret
);
1514 fsnotify_modify(file_out
);
1515 add_wchar(current
, ret
);
1520 mnt_drop_write_file(file_out
);
1524 EXPORT_SYMBOL(vfs_copy_file_range
);
1526 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1527 int, fd_out
, loff_t __user
*, off_out
,
1528 size_t, len
, unsigned int, flags
)
1534 ssize_t ret
= -EBADF
;
1536 f_in
= fdget(fd_in
);
1540 f_out
= fdget(fd_out
);
1546 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1549 pos_in
= f_in
.file
->f_pos
;
1553 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1556 pos_out
= f_out
.file
->f_pos
;
1559 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1566 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1569 f_in
.file
->f_pos
= pos_in
;
1573 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1576 f_out
.file
->f_pos
= pos_out
;
1588 static int clone_verify_area(struct file
*file
, loff_t pos
, u64 len
, bool write
)
1590 struct inode
*inode
= file_inode(file
);
1592 if (unlikely(pos
< 0))
1595 if (unlikely((loff_t
) (pos
+ len
) < 0))
1598 if (unlikely(inode
->i_flctx
&& mandatory_lock(inode
))) {
1599 loff_t end
= len
? pos
+ len
- 1 : OFFSET_MAX
;
1602 retval
= locks_mandatory_area(inode
, file
, pos
, end
,
1603 write
? F_WRLCK
: F_RDLCK
);
1608 return security_file_permission(file
, write
? MAY_WRITE
: MAY_READ
);
1611 int vfs_clone_file_range(struct file
*file_in
, loff_t pos_in
,
1612 struct file
*file_out
, loff_t pos_out
, u64 len
)
1614 struct inode
*inode_in
= file_inode(file_in
);
1615 struct inode
*inode_out
= file_inode(file_out
);
1618 if (inode_in
->i_sb
!= inode_out
->i_sb
||
1619 file_in
->f_path
.mnt
!= file_out
->f_path
.mnt
)
1622 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1624 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1627 if (!(file_in
->f_mode
& FMODE_READ
) ||
1628 !(file_out
->f_mode
& FMODE_WRITE
) ||
1629 (file_out
->f_flags
& O_APPEND
))
1632 if (!file_in
->f_op
->clone_file_range
)
1635 ret
= clone_verify_area(file_in
, pos_in
, len
, false);
1639 ret
= clone_verify_area(file_out
, pos_out
, len
, true);
1643 if (pos_in
+ len
> i_size_read(inode_in
))
1646 ret
= mnt_want_write_file(file_out
);
1650 ret
= file_in
->f_op
->clone_file_range(file_in
, pos_in
,
1651 file_out
, pos_out
, len
);
1653 fsnotify_access(file_in
);
1654 fsnotify_modify(file_out
);
1657 mnt_drop_write_file(file_out
);
1660 EXPORT_SYMBOL(vfs_clone_file_range
);
1662 int vfs_dedupe_file_range(struct file
*file
, struct file_dedupe_range
*same
)
1664 struct file_dedupe_range_info
*info
;
1665 struct inode
*src
= file_inode(file
);
1670 bool is_admin
= capable(CAP_SYS_ADMIN
);
1671 u16 count
= same
->dest_count
;
1672 struct file
*dst_file
;
1676 if (!(file
->f_mode
& FMODE_READ
))
1679 if (same
->reserved1
|| same
->reserved2
)
1682 off
= same
->src_offset
;
1683 len
= same
->src_length
;
1686 if (S_ISDIR(src
->i_mode
))
1690 if (!S_ISREG(src
->i_mode
))
1693 ret
= clone_verify_area(file
, off
, len
, false);
1698 /* pre-format output fields to sane values */
1699 for (i
= 0; i
< count
; i
++) {
1700 same
->info
[i
].bytes_deduped
= 0ULL;
1701 same
->info
[i
].status
= FILE_DEDUPE_RANGE_SAME
;
1704 for (i
= 0, info
= same
->info
; i
< count
; i
++, info
++) {
1706 struct fd dst_fd
= fdget(info
->dest_fd
);
1708 dst_file
= dst_fd
.file
;
1710 info
->status
= -EBADF
;
1713 dst
= file_inode(dst_file
);
1715 ret
= mnt_want_write_file(dst_file
);
1721 dst_off
= info
->dest_offset
;
1722 ret
= clone_verify_area(dst_file
, dst_off
, len
, true);
1729 if (info
->reserved
) {
1730 info
->status
= -EINVAL
;
1731 } else if (!(is_admin
|| (dst_file
->f_mode
& FMODE_WRITE
))) {
1732 info
->status
= -EINVAL
;
1733 } else if (file
->f_path
.mnt
!= dst_file
->f_path
.mnt
) {
1734 info
->status
= -EXDEV
;
1735 } else if (S_ISDIR(dst
->i_mode
)) {
1736 info
->status
= -EISDIR
;
1737 } else if (dst_file
->f_op
->dedupe_file_range
== NULL
) {
1738 info
->status
= -EINVAL
;
1740 deduped
= dst_file
->f_op
->dedupe_file_range(file
, off
,
1743 if (deduped
== -EBADE
)
1744 info
->status
= FILE_DEDUPE_RANGE_DIFFERS
;
1745 else if (deduped
< 0)
1746 info
->status
= deduped
;
1748 info
->bytes_deduped
+= deduped
;
1752 mnt_drop_write_file(dst_file
);
1756 if (fatal_signal_pending(current
))
1763 EXPORT_SYMBOL(vfs_dedupe_file_range
);