1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/read_write.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
28 const struct file_operations generic_ro_fops
= {
29 .llseek
= generic_file_llseek
,
30 .read_iter
= generic_file_read_iter
,
31 .mmap
= generic_file_readonly_mmap
,
32 .splice_read
= generic_file_splice_read
,
35 EXPORT_SYMBOL(generic_ro_fops
);
37 static inline bool unsigned_offsets(struct file
*file
)
39 return file
->f_mode
& FMODE_UNSIGNED_OFFSET
;
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
52 * Return the specified offset on success and -EINVAL on invalid offset.
54 loff_t
vfs_setpos(struct file
*file
, loff_t offset
, loff_t maxsize
)
56 if (offset
< 0 && !unsigned_offsets(file
))
61 if (offset
!= file
->f_pos
) {
67 EXPORT_SYMBOL(vfs_setpos
);
70 * generic_file_llseek_size - generic llseek implementation for regular files
71 * @file: file structure to seek on
72 * @offset: file offset to seek to
73 * @whence: type of seek
74 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
77 * This is a variant of generic_file_llseek that allows passing in a custom
78 * maximum file size and a custom EOF position, for e.g. hashed directories
81 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
86 generic_file_llseek_size(struct file
*file
, loff_t offset
, int whence
,
87 loff_t maxsize
, loff_t eof
)
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
107 spin_lock(&file
->f_lock
);
108 offset
= vfs_setpos(file
, file
->f_pos
+ offset
, maxsize
);
109 spin_unlock(&file
->f_lock
);
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
116 if ((unsigned long long)offset
>= eof
)
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
124 if ((unsigned long long)offset
>= eof
)
130 return vfs_setpos(file
, offset
, maxsize
);
132 EXPORT_SYMBOL(generic_file_llseek_size
);
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
138 * @whence: type of seek
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
142 * @offset and @whence.
144 loff_t
generic_file_llseek(struct file
*file
, loff_t offset
, int whence
)
146 struct inode
*inode
= file
->f_mapping
->host
;
148 return generic_file_llseek_size(file
, offset
, whence
,
149 inode
->i_sb
->s_maxbytes
,
152 EXPORT_SYMBOL(generic_file_llseek
);
155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
162 loff_t
fixed_size_llseek(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
165 case SEEK_SET
: case SEEK_CUR
: case SEEK_END
:
166 return generic_file_llseek_size(file
, offset
, whence
,
172 EXPORT_SYMBOL(fixed_size_llseek
);
175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
181 loff_t
no_seek_end_llseek(struct file
*file
, loff_t offset
, int whence
)
184 case SEEK_SET
: case SEEK_CUR
:
185 return generic_file_llseek_size(file
, offset
, whence
,
191 EXPORT_SYMBOL(no_seek_end_llseek
);
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
201 loff_t
no_seek_end_llseek_size(struct file
*file
, loff_t offset
, int whence
, loff_t size
)
204 case SEEK_SET
: case SEEK_CUR
:
205 return generic_file_llseek_size(file
, offset
, whence
,
211 EXPORT_SYMBOL(no_seek_end_llseek_size
);
214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
217 * @whence: type of seek
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
224 loff_t
noop_llseek(struct file
*file
, loff_t offset
, int whence
)
228 EXPORT_SYMBOL(noop_llseek
);
230 loff_t
no_llseek(struct file
*file
, loff_t offset
, int whence
)
234 EXPORT_SYMBOL(no_llseek
);
236 loff_t
default_llseek(struct file
*file
, loff_t offset
, int whence
)
238 struct inode
*inode
= file_inode(file
);
244 offset
+= i_size_read(inode
);
248 retval
= file
->f_pos
;
251 offset
+= file
->f_pos
;
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
259 if (offset
>= inode
->i_size
) {
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
270 if (offset
>= inode
->i_size
) {
274 offset
= inode
->i_size
;
278 if (offset
>= 0 || unsigned_offsets(file
)) {
279 if (offset
!= file
->f_pos
) {
280 file
->f_pos
= offset
;
289 EXPORT_SYMBOL(default_llseek
);
291 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
293 loff_t (*fn
)(struct file
*, loff_t
, int);
296 if (file
->f_mode
& FMODE_LSEEK
) {
297 if (file
->f_op
->llseek
)
298 fn
= file
->f_op
->llseek
;
300 return fn(file
, offset
, whence
);
302 EXPORT_SYMBOL(vfs_llseek
);
304 static off_t
ksys_lseek(unsigned int fd
, off_t offset
, unsigned int whence
)
307 struct fd f
= fdget_pos(fd
);
312 if (whence
<= SEEK_MAX
) {
313 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
315 if (res
!= (loff_t
)retval
)
316 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
322 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
324 return ksys_lseek(fd
, offset
, whence
);
328 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
330 return ksys_lseek(fd
, offset
, whence
);
334 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
335 defined(__ARCH_WANT_SYS_LLSEEK)
336 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
337 unsigned long, offset_low
, loff_t __user
*, result
,
338 unsigned int, whence
)
341 struct fd f
= fdget_pos(fd
);
348 if (whence
> SEEK_MAX
)
351 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
354 retval
= (int)offset
;
357 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
366 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
368 if (unlikely((ssize_t
) count
< 0))
372 * ranged mandatory locking does not apply to streams - it makes sense
373 * only for files where position has a meaning.
378 if (unlikely(pos
< 0)) {
379 if (!unsigned_offsets(file
))
381 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
383 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
384 if (!unsigned_offsets(file
))
389 return security_file_permission(file
,
390 read_write
== READ
? MAY_READ
: MAY_WRITE
);
393 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
395 struct iovec iov
= { .iov_base
= buf
, .iov_len
= len
};
397 struct iov_iter iter
;
400 init_sync_kiocb(&kiocb
, filp
);
401 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
402 iov_iter_init(&iter
, READ
, &iov
, 1, len
);
404 ret
= call_read_iter(filp
, &kiocb
, &iter
);
405 BUG_ON(ret
== -EIOCBQUEUED
);
407 *ppos
= kiocb
.ki_pos
;
411 static int warn_unsupported(struct file
*file
, const char *op
)
414 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
415 op
, file
, current
->pid
, current
->comm
);
419 ssize_t
__kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
423 .iov_len
= min_t(size_t, count
, MAX_RW_COUNT
),
426 struct iov_iter iter
;
429 if (WARN_ON_ONCE(!(file
->f_mode
& FMODE_READ
)))
431 if (!(file
->f_mode
& FMODE_CAN_READ
))
434 * Also fail if ->read_iter and ->read are both wired up as that
435 * implies very convoluted semantics.
437 if (unlikely(!file
->f_op
->read_iter
|| file
->f_op
->read
))
438 return warn_unsupported(file
, "read");
440 init_sync_kiocb(&kiocb
, file
);
441 kiocb
.ki_pos
= pos
? *pos
: 0;
442 iov_iter_kvec(&iter
, READ
, &iov
, 1, iov
.iov_len
);
443 ret
= file
->f_op
->read_iter(&kiocb
, &iter
);
447 fsnotify_access(file
);
448 add_rchar(current
, ret
);
454 ssize_t
kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
458 ret
= rw_verify_area(READ
, file
, pos
, count
);
461 return __kernel_read(file
, buf
, count
, pos
);
463 EXPORT_SYMBOL(kernel_read
);
465 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
469 if (!(file
->f_mode
& FMODE_READ
))
471 if (!(file
->f_mode
& FMODE_CAN_READ
))
473 if (unlikely(!access_ok(buf
, count
)))
476 ret
= rw_verify_area(READ
, file
, pos
, count
);
479 if (count
> MAX_RW_COUNT
)
480 count
= MAX_RW_COUNT
;
482 if (file
->f_op
->read
)
483 ret
= file
->f_op
->read(file
, buf
, count
, pos
);
484 else if (file
->f_op
->read_iter
)
485 ret
= new_sync_read(file
, buf
, count
, pos
);
489 fsnotify_access(file
);
490 add_rchar(current
, ret
);
495 EXPORT_SYMBOL_GPL(vfs_read
);
497 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
499 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
501 struct iov_iter iter
;
504 init_sync_kiocb(&kiocb
, filp
);
505 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
506 iov_iter_init(&iter
, WRITE
, &iov
, 1, len
);
508 ret
= call_write_iter(filp
, &kiocb
, &iter
);
509 BUG_ON(ret
== -EIOCBQUEUED
);
511 *ppos
= kiocb
.ki_pos
;
515 /* caller is responsible for file_start_write/file_end_write */
516 ssize_t
__kernel_write(struct file
*file
, const void *buf
, size_t count
, loff_t
*pos
)
519 .iov_base
= (void *)buf
,
520 .iov_len
= min_t(size_t, count
, MAX_RW_COUNT
),
523 struct iov_iter iter
;
526 if (WARN_ON_ONCE(!(file
->f_mode
& FMODE_WRITE
)))
528 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
531 * Also fail if ->write_iter and ->write are both wired up as that
532 * implies very convoluted semantics.
534 if (unlikely(!file
->f_op
->write_iter
|| file
->f_op
->write
))
535 return warn_unsupported(file
, "write");
537 init_sync_kiocb(&kiocb
, file
);
538 kiocb
.ki_pos
= pos
? *pos
: 0;
539 iov_iter_kvec(&iter
, WRITE
, &iov
, 1, iov
.iov_len
);
540 ret
= file
->f_op
->write_iter(&kiocb
, &iter
);
544 fsnotify_modify(file
);
545 add_wchar(current
, ret
);
551 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
552 * but autofs is one of the few internal kernel users that actually
553 * wants this _and_ can be built as a module. So we need to export
554 * this symbol for autofs, even though it really isn't appropriate
555 * for any other kernel modules.
557 EXPORT_SYMBOL_GPL(__kernel_write
);
559 ssize_t
kernel_write(struct file
*file
, const void *buf
, size_t count
,
564 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
568 file_start_write(file
);
569 ret
= __kernel_write(file
, buf
, count
, pos
);
570 file_end_write(file
);
573 EXPORT_SYMBOL(kernel_write
);
575 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
579 if (!(file
->f_mode
& FMODE_WRITE
))
581 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
583 if (unlikely(!access_ok(buf
, count
)))
586 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
589 if (count
> MAX_RW_COUNT
)
590 count
= MAX_RW_COUNT
;
591 file_start_write(file
);
592 if (file
->f_op
->write
)
593 ret
= file
->f_op
->write(file
, buf
, count
, pos
);
594 else if (file
->f_op
->write_iter
)
595 ret
= new_sync_write(file
, buf
, count
, pos
);
599 fsnotify_modify(file
);
600 add_wchar(current
, ret
);
603 file_end_write(file
);
606 EXPORT_SYMBOL_GPL(vfs_write
);
608 /* file_ppos returns &file->f_pos or NULL if file is stream */
609 static inline loff_t
*file_ppos(struct file
*file
)
611 return file
->f_mode
& FMODE_STREAM
? NULL
: &file
->f_pos
;
614 ssize_t
ksys_read(unsigned int fd
, char __user
*buf
, size_t count
)
616 struct fd f
= fdget_pos(fd
);
617 ssize_t ret
= -EBADF
;
620 loff_t pos
, *ppos
= file_ppos(f
.file
);
625 ret
= vfs_read(f
.file
, buf
, count
, ppos
);
626 if (ret
>= 0 && ppos
)
633 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
635 return ksys_read(fd
, buf
, count
);
638 ssize_t
ksys_write(unsigned int fd
, const char __user
*buf
, size_t count
)
640 struct fd f
= fdget_pos(fd
);
641 ssize_t ret
= -EBADF
;
644 loff_t pos
, *ppos
= file_ppos(f
.file
);
649 ret
= vfs_write(f
.file
, buf
, count
, ppos
);
650 if (ret
>= 0 && ppos
)
658 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
661 return ksys_write(fd
, buf
, count
);
664 ssize_t
ksys_pread64(unsigned int fd
, char __user
*buf
, size_t count
,
668 ssize_t ret
= -EBADF
;
676 if (f
.file
->f_mode
& FMODE_PREAD
)
677 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
684 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
685 size_t, count
, loff_t
, pos
)
687 return ksys_pread64(fd
, buf
, count
, pos
);
690 ssize_t
ksys_pwrite64(unsigned int fd
, const char __user
*buf
,
691 size_t count
, loff_t pos
)
694 ssize_t ret
= -EBADF
;
702 if (f
.file
->f_mode
& FMODE_PWRITE
)
703 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
710 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
711 size_t, count
, loff_t
, pos
)
713 return ksys_pwrite64(fd
, buf
, count
, pos
);
716 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
717 loff_t
*ppos
, int type
, rwf_t flags
)
722 init_sync_kiocb(&kiocb
, filp
);
723 ret
= kiocb_set_rw_flags(&kiocb
, flags
);
726 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
729 ret
= call_read_iter(filp
, &kiocb
, iter
);
731 ret
= call_write_iter(filp
, &kiocb
, iter
);
732 BUG_ON(ret
== -EIOCBQUEUED
);
734 *ppos
= kiocb
.ki_pos
;
738 /* Do it by hand, with file-ops */
739 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
740 loff_t
*ppos
, int type
, rwf_t flags
)
744 if (flags
& ~RWF_HIPRI
)
747 while (iov_iter_count(iter
)) {
748 struct iovec iovec
= iov_iter_iovec(iter
);
752 nr
= filp
->f_op
->read(filp
, iovec
.iov_base
,
753 iovec
.iov_len
, ppos
);
755 nr
= filp
->f_op
->write(filp
, iovec
.iov_base
,
756 iovec
.iov_len
, ppos
);
765 if (nr
!= iovec
.iov_len
)
767 iov_iter_advance(iter
, nr
);
773 static ssize_t
do_iter_read(struct file
*file
, struct iov_iter
*iter
,
774 loff_t
*pos
, rwf_t flags
)
779 if (!(file
->f_mode
& FMODE_READ
))
781 if (!(file
->f_mode
& FMODE_CAN_READ
))
784 tot_len
= iov_iter_count(iter
);
787 ret
= rw_verify_area(READ
, file
, pos
, tot_len
);
791 if (file
->f_op
->read_iter
)
792 ret
= do_iter_readv_writev(file
, iter
, pos
, READ
, flags
);
794 ret
= do_loop_readv_writev(file
, iter
, pos
, READ
, flags
);
797 fsnotify_access(file
);
801 ssize_t
vfs_iocb_iter_read(struct file
*file
, struct kiocb
*iocb
,
802 struct iov_iter
*iter
)
807 if (!file
->f_op
->read_iter
)
809 if (!(file
->f_mode
& FMODE_READ
))
811 if (!(file
->f_mode
& FMODE_CAN_READ
))
814 tot_len
= iov_iter_count(iter
);
817 ret
= rw_verify_area(READ
, file
, &iocb
->ki_pos
, tot_len
);
821 ret
= call_read_iter(file
, iocb
, iter
);
824 fsnotify_access(file
);
827 EXPORT_SYMBOL(vfs_iocb_iter_read
);
829 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
832 if (!file
->f_op
->read_iter
)
834 return do_iter_read(file
, iter
, ppos
, flags
);
836 EXPORT_SYMBOL(vfs_iter_read
);
838 static ssize_t
do_iter_write(struct file
*file
, struct iov_iter
*iter
,
839 loff_t
*pos
, rwf_t flags
)
844 if (!(file
->f_mode
& FMODE_WRITE
))
846 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
849 tot_len
= iov_iter_count(iter
);
852 ret
= rw_verify_area(WRITE
, file
, pos
, tot_len
);
856 if (file
->f_op
->write_iter
)
857 ret
= do_iter_readv_writev(file
, iter
, pos
, WRITE
, flags
);
859 ret
= do_loop_readv_writev(file
, iter
, pos
, WRITE
, flags
);
861 fsnotify_modify(file
);
865 ssize_t
vfs_iocb_iter_write(struct file
*file
, struct kiocb
*iocb
,
866 struct iov_iter
*iter
)
871 if (!file
->f_op
->write_iter
)
873 if (!(file
->f_mode
& FMODE_WRITE
))
875 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
878 tot_len
= iov_iter_count(iter
);
881 ret
= rw_verify_area(WRITE
, file
, &iocb
->ki_pos
, tot_len
);
885 ret
= call_write_iter(file
, iocb
, iter
);
887 fsnotify_modify(file
);
891 EXPORT_SYMBOL(vfs_iocb_iter_write
);
893 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
896 if (!file
->f_op
->write_iter
)
898 return do_iter_write(file
, iter
, ppos
, flags
);
900 EXPORT_SYMBOL(vfs_iter_write
);
902 static ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
903 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
905 struct iovec iovstack
[UIO_FASTIOV
];
906 struct iovec
*iov
= iovstack
;
907 struct iov_iter iter
;
910 ret
= import_iovec(READ
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
912 ret
= do_iter_read(file
, &iter
, pos
, flags
);
919 static ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
920 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
922 struct iovec iovstack
[UIO_FASTIOV
];
923 struct iovec
*iov
= iovstack
;
924 struct iov_iter iter
;
927 ret
= import_iovec(WRITE
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
, &iter
);
929 file_start_write(file
);
930 ret
= do_iter_write(file
, &iter
, pos
, flags
);
931 file_end_write(file
);
937 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
938 unsigned long vlen
, rwf_t flags
)
940 struct fd f
= fdget_pos(fd
);
941 ssize_t ret
= -EBADF
;
944 loff_t pos
, *ppos
= file_ppos(f
.file
);
949 ret
= vfs_readv(f
.file
, vec
, vlen
, ppos
, flags
);
950 if (ret
>= 0 && ppos
)
956 add_rchar(current
, ret
);
961 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
962 unsigned long vlen
, rwf_t flags
)
964 struct fd f
= fdget_pos(fd
);
965 ssize_t ret
= -EBADF
;
968 loff_t pos
, *ppos
= file_ppos(f
.file
);
973 ret
= vfs_writev(f
.file
, vec
, vlen
, ppos
, flags
);
974 if (ret
>= 0 && ppos
)
980 add_wchar(current
, ret
);
985 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
987 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
988 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
991 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
992 unsigned long vlen
, loff_t pos
, rwf_t flags
)
995 ssize_t ret
= -EBADF
;
1003 if (f
.file
->f_mode
& FMODE_PREAD
)
1004 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1009 add_rchar(current
, ret
);
1014 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1015 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1018 ssize_t ret
= -EBADF
;
1026 if (f
.file
->f_mode
& FMODE_PWRITE
)
1027 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1032 add_wchar(current
, ret
);
1037 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1038 unsigned long, vlen
)
1040 return do_readv(fd
, vec
, vlen
, 0);
1043 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1044 unsigned long, vlen
)
1046 return do_writev(fd
, vec
, vlen
, 0);
1049 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1050 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1052 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1054 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1057 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1058 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1061 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1064 return do_readv(fd
, vec
, vlen
, flags
);
1066 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1069 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1070 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1072 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1074 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1077 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1078 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1081 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1084 return do_writev(fd
, vec
, vlen
, flags
);
1086 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1090 * Various compat syscalls. Note that they all pretend to take a native
1091 * iovec - import_iovec will properly treat those as compat_iovecs based on
1092 * in_compat_syscall().
1094 #ifdef CONFIG_COMPAT
1095 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1096 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1097 const struct iovec __user
*, vec
,
1098 unsigned long, vlen
, loff_t
, pos
)
1100 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1104 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1105 const struct iovec __user
*, vec
,
1106 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1108 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1110 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1113 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1114 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1115 const struct iovec __user
*, vec
,
1116 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1119 return do_readv(fd
, vec
, vlen
, flags
);
1120 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1124 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1125 const struct iovec __user
*, vec
,
1126 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1129 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1132 return do_readv(fd
, vec
, vlen
, flags
);
1133 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1136 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1137 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1138 const struct iovec __user
*, vec
,
1139 unsigned long, vlen
, loff_t
, pos
)
1141 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1145 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1146 const struct iovec __user
*,vec
,
1147 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1149 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1151 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1154 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1155 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1156 const struct iovec __user
*, vec
,
1157 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1160 return do_writev(fd
, vec
, vlen
, flags
);
1161 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1165 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1166 const struct iovec __user
*,vec
,
1167 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, rwf_t
, flags
)
1169 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1172 return do_writev(fd
, vec
, vlen
, flags
);
1173 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1175 #endif /* CONFIG_COMPAT */
1177 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1178 size_t count
, loff_t max
)
1181 struct inode
*in_inode
, *out_inode
;
1182 struct pipe_inode_info
*opipe
;
1189 * Get input file, and verify that it is ok..
1195 if (!(in
.file
->f_mode
& FMODE_READ
))
1199 pos
= in
.file
->f_pos
;
1202 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1205 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1208 if (count
> MAX_RW_COUNT
)
1209 count
= MAX_RW_COUNT
;
1212 * Get output file, and verify that it is ok..
1215 out
= fdget(out_fd
);
1218 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1220 in_inode
= file_inode(in
.file
);
1221 out_inode
= file_inode(out
.file
);
1222 out_pos
= out
.file
->f_pos
;
1225 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1227 if (unlikely(pos
+ count
> max
)) {
1228 retval
= -EOVERFLOW
;
1237 * We need to debate whether we can enable this or not. The
1238 * man page documents EAGAIN return for the output at least,
1239 * and the application is arguably buggy if it doesn't expect
1240 * EAGAIN on a non-blocking file descriptor.
1242 if (in
.file
->f_flags
& O_NONBLOCK
)
1243 fl
= SPLICE_F_NONBLOCK
;
1245 opipe
= get_pipe_info(out
.file
, true);
1247 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1250 file_start_write(out
.file
);
1251 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
,
1253 file_end_write(out
.file
);
1255 retval
= splice_file_to_pipe(in
.file
, opipe
, &pos
, count
, fl
);
1259 add_rchar(current
, retval
);
1260 add_wchar(current
, retval
);
1261 fsnotify_access(in
.file
);
1262 fsnotify_modify(out
.file
);
1263 out
.file
->f_pos
= out_pos
;
1267 in
.file
->f_pos
= pos
;
1273 retval
= -EOVERFLOW
;
1283 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1290 if (unlikely(get_user(off
, offset
)))
1293 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1294 if (unlikely(put_user(pos
, offset
)))
1299 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1302 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1308 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1310 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1311 if (unlikely(put_user(pos
, offset
)))
1316 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1319 #ifdef CONFIG_COMPAT
1320 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1321 compat_off_t __user
*, offset
, compat_size_t
, count
)
1328 if (unlikely(get_user(off
, offset
)))
1331 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1332 if (unlikely(put_user(pos
, offset
)))
1337 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1340 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1341 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1347 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1349 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1350 if (unlikely(put_user(pos
, offset
)))
1355 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1360 * generic_copy_file_range - copy data between two files
1361 * @file_in: file structure to read from
1362 * @pos_in: file offset to read from
1363 * @file_out: file structure to write data to
1364 * @pos_out: file offset to write data to
1365 * @len: amount of data to copy
1366 * @flags: copy flags
1368 * This is a generic filesystem helper to copy data from one file to another.
1369 * It has no constraints on the source or destination file owners - the files
1370 * can belong to different superblocks and different filesystem types. Short
1371 * copies are allowed.
1373 * This should be called from the @file_out filesystem, as per the
1374 * ->copy_file_range() method.
1376 * Returns the number of bytes copied or a negative error indicating the
1380 ssize_t
generic_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1381 struct file
*file_out
, loff_t pos_out
,
1382 size_t len
, unsigned int flags
)
1384 return do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1385 len
> MAX_RW_COUNT
? MAX_RW_COUNT
: len
, 0);
1387 EXPORT_SYMBOL(generic_copy_file_range
);
1389 static ssize_t
do_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1390 struct file
*file_out
, loff_t pos_out
,
1391 size_t len
, unsigned int flags
)
1394 * Although we now allow filesystems to handle cross sb copy, passing
1395 * a file of the wrong filesystem type to filesystem driver can result
1396 * in an attempt to dereference the wrong type of ->private_data, so
1397 * avoid doing that until we really have a good reason. NFS defines
1398 * several different file_system_type structures, but they all end up
1399 * using the same ->copy_file_range() function pointer.
1401 if (file_out
->f_op
->copy_file_range
&&
1402 file_out
->f_op
->copy_file_range
== file_in
->f_op
->copy_file_range
)
1403 return file_out
->f_op
->copy_file_range(file_in
, pos_in
,
1407 return generic_copy_file_range(file_in
, pos_in
, file_out
, pos_out
, len
,
1412 * Performs necessary checks before doing a file copy
1414 * Can adjust amount of bytes to copy via @req_count argument.
1415 * Returns appropriate error code that caller should return or
1416 * zero in case the copy should be allowed.
1418 static int generic_copy_file_checks(struct file
*file_in
, loff_t pos_in
,
1419 struct file
*file_out
, loff_t pos_out
,
1420 size_t *req_count
, unsigned int flags
)
1422 struct inode
*inode_in
= file_inode(file_in
);
1423 struct inode
*inode_out
= file_inode(file_out
);
1424 uint64_t count
= *req_count
;
1428 ret
= generic_file_rw_checks(file_in
, file_out
);
1432 /* Don't touch certain kinds of inodes */
1433 if (IS_IMMUTABLE(inode_out
))
1436 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
1439 /* Ensure offsets don't wrap. */
1440 if (pos_in
+ count
< pos_in
|| pos_out
+ count
< pos_out
)
1443 /* Shorten the copy to EOF */
1444 size_in
= i_size_read(inode_in
);
1445 if (pos_in
>= size_in
)
1448 count
= min(count
, size_in
- (uint64_t)pos_in
);
1450 ret
= generic_write_check_limits(file_out
, pos_out
, &count
);
1454 /* Don't allow overlapped copying within the same file. */
1455 if (inode_in
== inode_out
&&
1456 pos_out
+ count
> pos_in
&&
1457 pos_out
< pos_in
+ count
)
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
)
1478 ret
= generic_copy_file_checks(file_in
, pos_in
, file_out
, pos_out
, &len
,
1483 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1487 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1494 file_start_write(file_out
);
1497 * Try cloning first, this is supported by more file systems, and
1498 * more efficient if both clone and copy are supported (e.g. NFS).
1500 if (file_in
->f_op
->remap_file_range
&&
1501 file_inode(file_in
)->i_sb
== file_inode(file_out
)->i_sb
) {
1504 cloned
= file_in
->f_op
->remap_file_range(file_in
, pos_in
,
1506 min_t(loff_t
, MAX_RW_COUNT
, len
),
1507 REMAP_FILE_CAN_SHORTEN
);
1514 ret
= do_copy_file_range(file_in
, pos_in
, file_out
, pos_out
, len
,
1516 WARN_ON_ONCE(ret
== -EOPNOTSUPP
);
1519 fsnotify_access(file_in
);
1520 add_rchar(current
, ret
);
1521 fsnotify_modify(file_out
);
1522 add_wchar(current
, ret
);
1528 file_end_write(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
;
1597 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1598 * LFS limits. If pos is under the limit it becomes a short access. If it
1599 * exceeds the limit we return -EFBIG.
1601 int generic_write_check_limits(struct file
*file
, loff_t pos
, loff_t
*count
)
1603 struct inode
*inode
= file
->f_mapping
->host
;
1604 loff_t max_size
= inode
->i_sb
->s_maxbytes
;
1605 loff_t limit
= rlimit(RLIMIT_FSIZE
);
1607 if (limit
!= RLIM_INFINITY
) {
1609 send_sig(SIGXFSZ
, current
, 0);
1612 *count
= min(*count
, limit
- pos
);
1615 if (!(file
->f_flags
& O_LARGEFILE
))
1616 max_size
= MAX_NON_LFS
;
1618 if (unlikely(pos
>= max_size
))
1621 *count
= min(*count
, max_size
- pos
);
1627 * Performs necessary checks before doing a write
1629 * Can adjust writing position or amount of bytes to write.
1630 * Returns appropriate error code that caller should return or
1631 * zero in case that write should be allowed.
1633 ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
1635 struct file
*file
= iocb
->ki_filp
;
1636 struct inode
*inode
= file
->f_mapping
->host
;
1640 if (IS_SWAPFILE(inode
))
1643 if (!iov_iter_count(from
))
1646 /* FIXME: this is for backwards compatibility with 2.4 */
1647 if (iocb
->ki_flags
& IOCB_APPEND
)
1648 iocb
->ki_pos
= i_size_read(inode
);
1650 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
1653 count
= iov_iter_count(from
);
1654 ret
= generic_write_check_limits(file
, iocb
->ki_pos
, &count
);
1658 iov_iter_truncate(from
, count
);
1659 return iov_iter_count(from
);
1661 EXPORT_SYMBOL(generic_write_checks
);
1664 * Performs common checks before doing a file copy/clone
1665 * from @file_in to @file_out.
1667 int generic_file_rw_checks(struct file
*file_in
, struct file
*file_out
)
1669 struct inode
*inode_in
= file_inode(file_in
);
1670 struct inode
*inode_out
= file_inode(file_out
);
1672 /* Don't copy dirs, pipes, sockets... */
1673 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1675 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1678 if (!(file_in
->f_mode
& FMODE_READ
) ||
1679 !(file_out
->f_mode
& FMODE_WRITE
) ||
1680 (file_out
->f_flags
& O_APPEND
))