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
= filemap_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 * @maxsize: 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
default_llseek(struct file
*file
, loff_t offset
, int whence
)
232 struct inode
*inode
= file_inode(file
);
238 offset
+= i_size_read(inode
);
242 retval
= file
->f_pos
;
245 offset
+= file
->f_pos
;
249 * In the generic case the entire file is data, so as
250 * long as offset isn't at the end of the file then the
253 if (offset
>= inode
->i_size
) {
260 * There is a virtual hole at the end of the file, so
261 * as long as offset isn't i_size or larger, return
264 if (offset
>= inode
->i_size
) {
268 offset
= inode
->i_size
;
272 if (offset
>= 0 || unsigned_offsets(file
)) {
273 if (offset
!= file
->f_pos
) {
274 file
->f_pos
= offset
;
283 EXPORT_SYMBOL(default_llseek
);
285 loff_t
vfs_llseek(struct file
*file
, loff_t offset
, int whence
)
287 if (!(file
->f_mode
& FMODE_LSEEK
))
289 return file
->f_op
->llseek(file
, offset
, whence
);
291 EXPORT_SYMBOL(vfs_llseek
);
293 static off_t
ksys_lseek(unsigned int fd
, off_t offset
, unsigned int whence
)
296 struct fd f
= fdget_pos(fd
);
301 if (whence
<= SEEK_MAX
) {
302 loff_t res
= vfs_llseek(f
.file
, offset
, whence
);
304 if (res
!= (loff_t
)retval
)
305 retval
= -EOVERFLOW
; /* LFS: should only happen on 32 bit platforms */
311 SYSCALL_DEFINE3(lseek
, unsigned int, fd
, off_t
, offset
, unsigned int, whence
)
313 return ksys_lseek(fd
, offset
, whence
);
317 COMPAT_SYSCALL_DEFINE3(lseek
, unsigned int, fd
, compat_off_t
, offset
, unsigned int, whence
)
319 return ksys_lseek(fd
, offset
, whence
);
323 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
324 defined(__ARCH_WANT_SYS_LLSEEK)
325 SYSCALL_DEFINE5(llseek
, unsigned int, fd
, unsigned long, offset_high
,
326 unsigned long, offset_low
, loff_t __user
*, result
,
327 unsigned int, whence
)
330 struct fd f
= fdget_pos(fd
);
337 if (whence
> SEEK_MAX
)
340 offset
= vfs_llseek(f
.file
, ((loff_t
) offset_high
<< 32) | offset_low
,
343 retval
= (int)offset
;
346 if (!copy_to_user(result
, &offset
, sizeof(offset
)))
355 int rw_verify_area(int read_write
, struct file
*file
, const loff_t
*ppos
, size_t count
)
357 int mask
= read_write
== READ
? MAY_READ
: MAY_WRITE
;
360 if (unlikely((ssize_t
) count
< 0))
366 if (unlikely(pos
< 0)) {
367 if (!unsigned_offsets(file
))
369 if (count
>= -pos
) /* both values are in 0..LLONG_MAX */
371 } else if (unlikely((loff_t
) (pos
+ count
) < 0)) {
372 if (!unsigned_offsets(file
))
377 ret
= security_file_permission(file
, mask
);
381 return fsnotify_file_area_perm(file
, mask
, ppos
, count
);
383 EXPORT_SYMBOL(rw_verify_area
);
385 static ssize_t
new_sync_read(struct file
*filp
, char __user
*buf
, size_t len
, loff_t
*ppos
)
388 struct iov_iter iter
;
391 init_sync_kiocb(&kiocb
, filp
);
392 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
393 iov_iter_ubuf(&iter
, ITER_DEST
, buf
, len
);
395 ret
= call_read_iter(filp
, &kiocb
, &iter
);
396 BUG_ON(ret
== -EIOCBQUEUED
);
398 *ppos
= kiocb
.ki_pos
;
402 static int warn_unsupported(struct file
*file
, const char *op
)
405 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
406 op
, file
, current
->pid
, current
->comm
);
410 ssize_t
__kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
414 .iov_len
= min_t(size_t, count
, MAX_RW_COUNT
),
417 struct iov_iter iter
;
420 if (WARN_ON_ONCE(!(file
->f_mode
& FMODE_READ
)))
422 if (!(file
->f_mode
& FMODE_CAN_READ
))
425 * Also fail if ->read_iter and ->read are both wired up as that
426 * implies very convoluted semantics.
428 if (unlikely(!file
->f_op
->read_iter
|| file
->f_op
->read
))
429 return warn_unsupported(file
, "read");
431 init_sync_kiocb(&kiocb
, file
);
432 kiocb
.ki_pos
= pos
? *pos
: 0;
433 iov_iter_kvec(&iter
, ITER_DEST
, &iov
, 1, iov
.iov_len
);
434 ret
= file
->f_op
->read_iter(&kiocb
, &iter
);
438 fsnotify_access(file
);
439 add_rchar(current
, ret
);
445 ssize_t
kernel_read(struct file
*file
, void *buf
, size_t count
, loff_t
*pos
)
449 ret
= rw_verify_area(READ
, file
, pos
, count
);
452 return __kernel_read(file
, buf
, count
, pos
);
454 EXPORT_SYMBOL(kernel_read
);
456 ssize_t
vfs_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*pos
)
460 if (!(file
->f_mode
& FMODE_READ
))
462 if (!(file
->f_mode
& FMODE_CAN_READ
))
464 if (unlikely(!access_ok(buf
, count
)))
467 ret
= rw_verify_area(READ
, file
, pos
, count
);
470 if (count
> MAX_RW_COUNT
)
471 count
= MAX_RW_COUNT
;
473 if (file
->f_op
->read
)
474 ret
= file
->f_op
->read(file
, buf
, count
, pos
);
475 else if (file
->f_op
->read_iter
)
476 ret
= new_sync_read(file
, buf
, count
, pos
);
480 fsnotify_access(file
);
481 add_rchar(current
, ret
);
487 static ssize_t
new_sync_write(struct file
*filp
, const char __user
*buf
, size_t len
, loff_t
*ppos
)
490 struct iov_iter iter
;
493 init_sync_kiocb(&kiocb
, filp
);
494 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
495 iov_iter_ubuf(&iter
, ITER_SOURCE
, (void __user
*)buf
, len
);
497 ret
= call_write_iter(filp
, &kiocb
, &iter
);
498 BUG_ON(ret
== -EIOCBQUEUED
);
500 *ppos
= kiocb
.ki_pos
;
504 /* caller is responsible for file_start_write/file_end_write */
505 ssize_t
__kernel_write_iter(struct file
*file
, struct iov_iter
*from
, loff_t
*pos
)
510 if (WARN_ON_ONCE(!(file
->f_mode
& FMODE_WRITE
)))
512 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
515 * Also fail if ->write_iter and ->write are both wired up as that
516 * implies very convoluted semantics.
518 if (unlikely(!file
->f_op
->write_iter
|| file
->f_op
->write
))
519 return warn_unsupported(file
, "write");
521 init_sync_kiocb(&kiocb
, file
);
522 kiocb
.ki_pos
= pos
? *pos
: 0;
523 ret
= file
->f_op
->write_iter(&kiocb
, from
);
527 fsnotify_modify(file
);
528 add_wchar(current
, ret
);
534 /* caller is responsible for file_start_write/file_end_write */
535 ssize_t
__kernel_write(struct file
*file
, const void *buf
, size_t count
, loff_t
*pos
)
538 .iov_base
= (void *)buf
,
539 .iov_len
= min_t(size_t, count
, MAX_RW_COUNT
),
541 struct iov_iter iter
;
542 iov_iter_kvec(&iter
, ITER_SOURCE
, &iov
, 1, iov
.iov_len
);
543 return __kernel_write_iter(file
, &iter
, pos
);
546 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
547 * but autofs is one of the few internal kernel users that actually
548 * wants this _and_ can be built as a module. So we need to export
549 * this symbol for autofs, even though it really isn't appropriate
550 * for any other kernel modules.
552 EXPORT_SYMBOL_GPL(__kernel_write
);
554 ssize_t
kernel_write(struct file
*file
, const void *buf
, size_t count
,
559 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
563 file_start_write(file
);
564 ret
= __kernel_write(file
, buf
, count
, pos
);
565 file_end_write(file
);
568 EXPORT_SYMBOL(kernel_write
);
570 ssize_t
vfs_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*pos
)
574 if (!(file
->f_mode
& FMODE_WRITE
))
576 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
578 if (unlikely(!access_ok(buf
, count
)))
581 ret
= rw_verify_area(WRITE
, file
, pos
, count
);
584 if (count
> MAX_RW_COUNT
)
585 count
= MAX_RW_COUNT
;
586 file_start_write(file
);
587 if (file
->f_op
->write
)
588 ret
= file
->f_op
->write(file
, buf
, count
, pos
);
589 else if (file
->f_op
->write_iter
)
590 ret
= new_sync_write(file
, buf
, count
, pos
);
594 fsnotify_modify(file
);
595 add_wchar(current
, ret
);
598 file_end_write(file
);
602 /* file_ppos returns &file->f_pos or NULL if file is stream */
603 static inline loff_t
*file_ppos(struct file
*file
)
605 return file
->f_mode
& FMODE_STREAM
? NULL
: &file
->f_pos
;
608 ssize_t
ksys_read(unsigned int fd
, char __user
*buf
, size_t count
)
610 struct fd f
= fdget_pos(fd
);
611 ssize_t ret
= -EBADF
;
614 loff_t pos
, *ppos
= file_ppos(f
.file
);
619 ret
= vfs_read(f
.file
, buf
, count
, ppos
);
620 if (ret
>= 0 && ppos
)
627 SYSCALL_DEFINE3(read
, unsigned int, fd
, char __user
*, buf
, size_t, count
)
629 return ksys_read(fd
, buf
, count
);
632 ssize_t
ksys_write(unsigned int fd
, const char __user
*buf
, size_t count
)
634 struct fd f
= fdget_pos(fd
);
635 ssize_t ret
= -EBADF
;
638 loff_t pos
, *ppos
= file_ppos(f
.file
);
643 ret
= vfs_write(f
.file
, buf
, count
, ppos
);
644 if (ret
>= 0 && ppos
)
652 SYSCALL_DEFINE3(write
, unsigned int, fd
, const char __user
*, buf
,
655 return ksys_write(fd
, buf
, count
);
658 ssize_t
ksys_pread64(unsigned int fd
, char __user
*buf
, size_t count
,
662 ssize_t ret
= -EBADF
;
670 if (f
.file
->f_mode
& FMODE_PREAD
)
671 ret
= vfs_read(f
.file
, buf
, count
, &pos
);
678 SYSCALL_DEFINE4(pread64
, unsigned int, fd
, char __user
*, buf
,
679 size_t, count
, loff_t
, pos
)
681 return ksys_pread64(fd
, buf
, count
, pos
);
684 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
685 COMPAT_SYSCALL_DEFINE5(pread64
, unsigned int, fd
, char __user
*, buf
,
686 size_t, count
, compat_arg_u64_dual(pos
))
688 return ksys_pread64(fd
, buf
, count
, compat_arg_u64_glue(pos
));
692 ssize_t
ksys_pwrite64(unsigned int fd
, const char __user
*buf
,
693 size_t count
, loff_t pos
)
696 ssize_t ret
= -EBADF
;
704 if (f
.file
->f_mode
& FMODE_PWRITE
)
705 ret
= vfs_write(f
.file
, buf
, count
, &pos
);
712 SYSCALL_DEFINE4(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
713 size_t, count
, loff_t
, pos
)
715 return ksys_pwrite64(fd
, buf
, count
, pos
);
718 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
719 COMPAT_SYSCALL_DEFINE5(pwrite64
, unsigned int, fd
, const char __user
*, buf
,
720 size_t, count
, compat_arg_u64_dual(pos
))
722 return ksys_pwrite64(fd
, buf
, count
, compat_arg_u64_glue(pos
));
726 static ssize_t
do_iter_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
727 loff_t
*ppos
, int type
, rwf_t flags
)
732 init_sync_kiocb(&kiocb
, filp
);
733 ret
= kiocb_set_rw_flags(&kiocb
, flags
);
736 kiocb
.ki_pos
= (ppos
? *ppos
: 0);
739 ret
= call_read_iter(filp
, &kiocb
, iter
);
741 ret
= call_write_iter(filp
, &kiocb
, iter
);
742 BUG_ON(ret
== -EIOCBQUEUED
);
744 *ppos
= kiocb
.ki_pos
;
748 /* Do it by hand, with file-ops */
749 static ssize_t
do_loop_readv_writev(struct file
*filp
, struct iov_iter
*iter
,
750 loff_t
*ppos
, int type
, rwf_t flags
)
754 if (flags
& ~RWF_HIPRI
)
757 while (iov_iter_count(iter
)) {
761 nr
= filp
->f_op
->read(filp
, iter_iov_addr(iter
),
762 iter_iov_len(iter
), ppos
);
764 nr
= filp
->f_op
->write(filp
, iter_iov_addr(iter
),
765 iter_iov_len(iter
), ppos
);
774 if (nr
!= iter_iov_len(iter
))
776 iov_iter_advance(iter
, nr
);
782 ssize_t
vfs_iocb_iter_read(struct file
*file
, struct kiocb
*iocb
,
783 struct iov_iter
*iter
)
788 if (!file
->f_op
->read_iter
)
790 if (!(file
->f_mode
& FMODE_READ
))
792 if (!(file
->f_mode
& FMODE_CAN_READ
))
795 tot_len
= iov_iter_count(iter
);
798 ret
= rw_verify_area(READ
, file
, &iocb
->ki_pos
, tot_len
);
802 ret
= call_read_iter(file
, iocb
, iter
);
805 fsnotify_access(file
);
808 EXPORT_SYMBOL(vfs_iocb_iter_read
);
810 ssize_t
vfs_iter_read(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
816 if (!file
->f_op
->read_iter
)
818 if (!(file
->f_mode
& FMODE_READ
))
820 if (!(file
->f_mode
& FMODE_CAN_READ
))
823 tot_len
= iov_iter_count(iter
);
826 ret
= rw_verify_area(READ
, file
, ppos
, tot_len
);
830 ret
= do_iter_readv_writev(file
, iter
, ppos
, READ
, flags
);
833 fsnotify_access(file
);
836 EXPORT_SYMBOL(vfs_iter_read
);
839 * Caller is responsible for calling kiocb_end_write() on completion
840 * if async iocb was queued.
842 ssize_t
vfs_iocb_iter_write(struct file
*file
, struct kiocb
*iocb
,
843 struct iov_iter
*iter
)
848 if (!file
->f_op
->write_iter
)
850 if (!(file
->f_mode
& FMODE_WRITE
))
852 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
855 tot_len
= iov_iter_count(iter
);
858 ret
= rw_verify_area(WRITE
, file
, &iocb
->ki_pos
, tot_len
);
862 kiocb_start_write(iocb
);
863 ret
= call_write_iter(file
, iocb
, iter
);
864 if (ret
!= -EIOCBQUEUED
)
865 kiocb_end_write(iocb
);
867 fsnotify_modify(file
);
871 EXPORT_SYMBOL(vfs_iocb_iter_write
);
873 ssize_t
vfs_iter_write(struct file
*file
, struct iov_iter
*iter
, loff_t
*ppos
,
879 if (!(file
->f_mode
& FMODE_WRITE
))
881 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
883 if (!file
->f_op
->write_iter
)
886 tot_len
= iov_iter_count(iter
);
890 ret
= rw_verify_area(WRITE
, file
, ppos
, tot_len
);
894 file_start_write(file
);
895 ret
= do_iter_readv_writev(file
, iter
, ppos
, WRITE
, flags
);
897 fsnotify_modify(file
);
898 file_end_write(file
);
902 EXPORT_SYMBOL(vfs_iter_write
);
904 static ssize_t
vfs_readv(struct file
*file
, const struct iovec __user
*vec
,
905 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
907 struct iovec iovstack
[UIO_FASTIOV
];
908 struct iovec
*iov
= iovstack
;
909 struct iov_iter iter
;
913 if (!(file
->f_mode
& FMODE_READ
))
915 if (!(file
->f_mode
& FMODE_CAN_READ
))
918 ret
= import_iovec(ITER_DEST
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
,
923 tot_len
= iov_iter_count(&iter
);
927 ret
= rw_verify_area(READ
, file
, pos
, tot_len
);
931 if (file
->f_op
->read_iter
)
932 ret
= do_iter_readv_writev(file
, &iter
, pos
, READ
, flags
);
934 ret
= do_loop_readv_writev(file
, &iter
, pos
, READ
, flags
);
937 fsnotify_access(file
);
942 static ssize_t
vfs_writev(struct file
*file
, const struct iovec __user
*vec
,
943 unsigned long vlen
, loff_t
*pos
, rwf_t flags
)
945 struct iovec iovstack
[UIO_FASTIOV
];
946 struct iovec
*iov
= iovstack
;
947 struct iov_iter iter
;
951 if (!(file
->f_mode
& FMODE_WRITE
))
953 if (!(file
->f_mode
& FMODE_CAN_WRITE
))
956 ret
= import_iovec(ITER_SOURCE
, vec
, vlen
, ARRAY_SIZE(iovstack
), &iov
,
961 tot_len
= iov_iter_count(&iter
);
965 ret
= rw_verify_area(WRITE
, file
, pos
, tot_len
);
969 file_start_write(file
);
970 if (file
->f_op
->write_iter
)
971 ret
= do_iter_readv_writev(file
, &iter
, pos
, WRITE
, flags
);
973 ret
= do_loop_readv_writev(file
, &iter
, pos
, WRITE
, flags
);
975 fsnotify_modify(file
);
976 file_end_write(file
);
982 static ssize_t
do_readv(unsigned long fd
, const struct iovec __user
*vec
,
983 unsigned long vlen
, rwf_t flags
)
985 struct fd f
= fdget_pos(fd
);
986 ssize_t ret
= -EBADF
;
989 loff_t pos
, *ppos
= file_ppos(f
.file
);
994 ret
= vfs_readv(f
.file
, vec
, vlen
, ppos
, flags
);
995 if (ret
>= 0 && ppos
)
1001 add_rchar(current
, ret
);
1006 static ssize_t
do_writev(unsigned long fd
, const struct iovec __user
*vec
,
1007 unsigned long vlen
, rwf_t flags
)
1009 struct fd f
= fdget_pos(fd
);
1010 ssize_t ret
= -EBADF
;
1013 loff_t pos
, *ppos
= file_ppos(f
.file
);
1018 ret
= vfs_writev(f
.file
, vec
, vlen
, ppos
, flags
);
1019 if (ret
>= 0 && ppos
)
1020 f
.file
->f_pos
= pos
;
1025 add_wchar(current
, ret
);
1030 static inline loff_t
pos_from_hilo(unsigned long high
, unsigned long low
)
1032 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1033 return (((loff_t
)high
<< HALF_LONG_BITS
) << HALF_LONG_BITS
) | low
;
1036 static ssize_t
do_preadv(unsigned long fd
, const struct iovec __user
*vec
,
1037 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1040 ssize_t ret
= -EBADF
;
1048 if (f
.file
->f_mode
& FMODE_PREAD
)
1049 ret
= vfs_readv(f
.file
, vec
, vlen
, &pos
, flags
);
1054 add_rchar(current
, ret
);
1059 static ssize_t
do_pwritev(unsigned long fd
, const struct iovec __user
*vec
,
1060 unsigned long vlen
, loff_t pos
, rwf_t flags
)
1063 ssize_t ret
= -EBADF
;
1071 if (f
.file
->f_mode
& FMODE_PWRITE
)
1072 ret
= vfs_writev(f
.file
, vec
, vlen
, &pos
, flags
);
1077 add_wchar(current
, ret
);
1082 SYSCALL_DEFINE3(readv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1083 unsigned long, vlen
)
1085 return do_readv(fd
, vec
, vlen
, 0);
1088 SYSCALL_DEFINE3(writev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1089 unsigned long, vlen
)
1091 return do_writev(fd
, vec
, vlen
, 0);
1094 SYSCALL_DEFINE5(preadv
, unsigned long, fd
, const struct iovec __user
*, vec
,
1095 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1097 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1099 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1102 SYSCALL_DEFINE6(preadv2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1103 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1106 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1109 return do_readv(fd
, vec
, vlen
, flags
);
1111 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1114 SYSCALL_DEFINE5(pwritev
, unsigned long, fd
, const struct iovec __user
*, vec
,
1115 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
)
1117 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1119 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1122 SYSCALL_DEFINE6(pwritev2
, unsigned long, fd
, const struct iovec __user
*, vec
,
1123 unsigned long, vlen
, unsigned long, pos_l
, unsigned long, pos_h
,
1126 loff_t pos
= pos_from_hilo(pos_h
, pos_l
);
1129 return do_writev(fd
, vec
, vlen
, flags
);
1131 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1135 * Various compat syscalls. Note that they all pretend to take a native
1136 * iovec - import_iovec will properly treat those as compat_iovecs based on
1137 * in_compat_syscall().
1139 #ifdef CONFIG_COMPAT
1140 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1141 COMPAT_SYSCALL_DEFINE4(preadv64
, unsigned long, fd
,
1142 const struct iovec __user
*, vec
,
1143 unsigned long, vlen
, loff_t
, pos
)
1145 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1149 COMPAT_SYSCALL_DEFINE5(preadv
, compat_ulong_t
, fd
,
1150 const struct iovec __user
*, vec
,
1151 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1153 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1155 return do_preadv(fd
, vec
, vlen
, pos
, 0);
1158 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1159 COMPAT_SYSCALL_DEFINE5(preadv64v2
, unsigned long, fd
,
1160 const struct iovec __user
*, vec
,
1161 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1164 return do_readv(fd
, vec
, vlen
, flags
);
1165 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1169 COMPAT_SYSCALL_DEFINE6(preadv2
, compat_ulong_t
, fd
,
1170 const struct iovec __user
*, vec
,
1171 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
,
1174 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1177 return do_readv(fd
, vec
, vlen
, flags
);
1178 return do_preadv(fd
, vec
, vlen
, pos
, flags
);
1181 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1182 COMPAT_SYSCALL_DEFINE4(pwritev64
, unsigned long, fd
,
1183 const struct iovec __user
*, vec
,
1184 unsigned long, vlen
, loff_t
, pos
)
1186 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1190 COMPAT_SYSCALL_DEFINE5(pwritev
, compat_ulong_t
, fd
,
1191 const struct iovec __user
*,vec
,
1192 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
)
1194 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1196 return do_pwritev(fd
, vec
, vlen
, pos
, 0);
1199 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1200 COMPAT_SYSCALL_DEFINE5(pwritev64v2
, unsigned long, fd
,
1201 const struct iovec __user
*, vec
,
1202 unsigned long, vlen
, loff_t
, pos
, rwf_t
, flags
)
1205 return do_writev(fd
, vec
, vlen
, flags
);
1206 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1210 COMPAT_SYSCALL_DEFINE6(pwritev2
, compat_ulong_t
, fd
,
1211 const struct iovec __user
*,vec
,
1212 compat_ulong_t
, vlen
, u32
, pos_low
, u32
, pos_high
, rwf_t
, flags
)
1214 loff_t pos
= ((loff_t
)pos_high
<< 32) | pos_low
;
1217 return do_writev(fd
, vec
, vlen
, flags
);
1218 return do_pwritev(fd
, vec
, vlen
, pos
, flags
);
1220 #endif /* CONFIG_COMPAT */
1222 static ssize_t
do_sendfile(int out_fd
, int in_fd
, loff_t
*ppos
,
1223 size_t count
, loff_t max
)
1226 struct inode
*in_inode
, *out_inode
;
1227 struct pipe_inode_info
*opipe
;
1234 * Get input file, and verify that it is ok..
1240 if (!(in
.file
->f_mode
& FMODE_READ
))
1244 pos
= in
.file
->f_pos
;
1247 if (!(in
.file
->f_mode
& FMODE_PREAD
))
1250 retval
= rw_verify_area(READ
, in
.file
, &pos
, count
);
1253 if (count
> MAX_RW_COUNT
)
1254 count
= MAX_RW_COUNT
;
1257 * Get output file, and verify that it is ok..
1260 out
= fdget(out_fd
);
1263 if (!(out
.file
->f_mode
& FMODE_WRITE
))
1265 in_inode
= file_inode(in
.file
);
1266 out_inode
= file_inode(out
.file
);
1267 out_pos
= out
.file
->f_pos
;
1270 max
= min(in_inode
->i_sb
->s_maxbytes
, out_inode
->i_sb
->s_maxbytes
);
1272 if (unlikely(pos
+ count
> max
)) {
1273 retval
= -EOVERFLOW
;
1282 * We need to debate whether we can enable this or not. The
1283 * man page documents EAGAIN return for the output at least,
1284 * and the application is arguably buggy if it doesn't expect
1285 * EAGAIN on a non-blocking file descriptor.
1287 if (in
.file
->f_flags
& O_NONBLOCK
)
1288 fl
= SPLICE_F_NONBLOCK
;
1290 opipe
= get_pipe_info(out
.file
, true);
1292 retval
= rw_verify_area(WRITE
, out
.file
, &out_pos
, count
);
1295 retval
= do_splice_direct(in
.file
, &pos
, out
.file
, &out_pos
,
1298 if (out
.file
->f_flags
& O_NONBLOCK
)
1299 fl
|= SPLICE_F_NONBLOCK
;
1301 retval
= splice_file_to_pipe(in
.file
, opipe
, &pos
, count
, fl
);
1305 add_rchar(current
, retval
);
1306 add_wchar(current
, retval
);
1307 fsnotify_access(in
.file
);
1308 fsnotify_modify(out
.file
);
1309 out
.file
->f_pos
= out_pos
;
1313 in
.file
->f_pos
= pos
;
1319 retval
= -EOVERFLOW
;
1329 SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
, off_t __user
*, offset
, size_t, count
)
1336 if (unlikely(get_user(off
, offset
)))
1339 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1340 if (unlikely(put_user(pos
, offset
)))
1345 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1348 SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
, loff_t __user
*, offset
, size_t, count
)
1354 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1356 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1357 if (unlikely(put_user(pos
, offset
)))
1362 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1365 #ifdef CONFIG_COMPAT
1366 COMPAT_SYSCALL_DEFINE4(sendfile
, int, out_fd
, int, in_fd
,
1367 compat_off_t __user
*, offset
, compat_size_t
, count
)
1374 if (unlikely(get_user(off
, offset
)))
1377 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, MAX_NON_LFS
);
1378 if (unlikely(put_user(pos
, offset
)))
1383 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1386 COMPAT_SYSCALL_DEFINE4(sendfile64
, int, out_fd
, int, in_fd
,
1387 compat_loff_t __user
*, offset
, compat_size_t
, count
)
1393 if (unlikely(copy_from_user(&pos
, offset
, sizeof(loff_t
))))
1395 ret
= do_sendfile(out_fd
, in_fd
, &pos
, count
, 0);
1396 if (unlikely(put_user(pos
, offset
)))
1401 return do_sendfile(out_fd
, in_fd
, NULL
, count
, 0);
1406 * Performs necessary checks before doing a file copy
1408 * Can adjust amount of bytes to copy via @req_count argument.
1409 * Returns appropriate error code that caller should return or
1410 * zero in case the copy should be allowed.
1412 static int generic_copy_file_checks(struct file
*file_in
, loff_t pos_in
,
1413 struct file
*file_out
, loff_t pos_out
,
1414 size_t *req_count
, unsigned int flags
)
1416 struct inode
*inode_in
= file_inode(file_in
);
1417 struct inode
*inode_out
= file_inode(file_out
);
1418 uint64_t count
= *req_count
;
1422 ret
= generic_file_rw_checks(file_in
, file_out
);
1427 * We allow some filesystems to handle cross sb copy, but passing
1428 * a file of the wrong filesystem type to filesystem driver can result
1429 * in an attempt to dereference the wrong type of ->private_data, so
1430 * avoid doing that until we really have a good reason.
1432 * nfs and cifs define several different file_system_type structures
1433 * and several different sets of file_operations, but they all end up
1434 * using the same ->copy_file_range() function pointer.
1436 if (flags
& COPY_FILE_SPLICE
) {
1437 /* cross sb splice is allowed */
1438 } else if (file_out
->f_op
->copy_file_range
) {
1439 if (file_in
->f_op
->copy_file_range
!=
1440 file_out
->f_op
->copy_file_range
)
1442 } else if (file_inode(file_in
)->i_sb
!= file_inode(file_out
)->i_sb
) {
1446 /* Don't touch certain kinds of inodes */
1447 if (IS_IMMUTABLE(inode_out
))
1450 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
1453 /* Ensure offsets don't wrap. */
1454 if (pos_in
+ count
< pos_in
|| pos_out
+ count
< pos_out
)
1457 /* Shorten the copy to EOF */
1458 size_in
= i_size_read(inode_in
);
1459 if (pos_in
>= size_in
)
1462 count
= min(count
, size_in
- (uint64_t)pos_in
);
1464 ret
= generic_write_check_limits(file_out
, pos_out
, &count
);
1468 /* Don't allow overlapped copying within the same file. */
1469 if (inode_in
== inode_out
&&
1470 pos_out
+ count
> pos_in
&&
1471 pos_out
< pos_in
+ count
)
1479 * copy_file_range() differs from regular file read and write in that it
1480 * specifically allows return partial success. When it does so is up to
1481 * the copy_file_range method.
1483 ssize_t
vfs_copy_file_range(struct file
*file_in
, loff_t pos_in
,
1484 struct file
*file_out
, loff_t pos_out
,
1485 size_t len
, unsigned int flags
)
1488 bool splice
= flags
& COPY_FILE_SPLICE
;
1489 bool samesb
= file_inode(file_in
)->i_sb
== file_inode(file_out
)->i_sb
;
1491 if (flags
& ~COPY_FILE_SPLICE
)
1494 ret
= generic_copy_file_checks(file_in
, pos_in
, file_out
, pos_out
, &len
,
1499 ret
= rw_verify_area(READ
, file_in
, &pos_in
, len
);
1503 ret
= rw_verify_area(WRITE
, file_out
, &pos_out
, len
);
1510 file_start_write(file_out
);
1513 * Cloning is supported by more file systems, so we implement copy on
1514 * same sb using clone, but for filesystems where both clone and copy
1515 * are supported (e.g. nfs,cifs), we only call the copy method.
1517 if (!splice
&& file_out
->f_op
->copy_file_range
) {
1518 ret
= file_out
->f_op
->copy_file_range(file_in
, pos_in
,
1521 } else if (!splice
&& file_in
->f_op
->remap_file_range
&& samesb
) {
1522 ret
= file_in
->f_op
->remap_file_range(file_in
, pos_in
,
1524 min_t(loff_t
, MAX_RW_COUNT
, len
),
1525 REMAP_FILE_CAN_SHORTEN
);
1526 /* fallback to splice */
1529 } else if (samesb
) {
1530 /* Fallback to splice for same sb copy for backward compat */
1534 file_end_write(file_out
);
1540 * We can get here for same sb copy of filesystems that do not implement
1541 * ->copy_file_range() in case filesystem does not support clone or in
1542 * case filesystem supports clone but rejected the clone request (e.g.
1543 * because it was not block aligned).
1545 * In both cases, fall back to kernel copy so we are able to maintain a
1546 * consistent story about which filesystems support copy_file_range()
1547 * and which filesystems do not, that will allow userspace tools to
1548 * make consistent desicions w.r.t using copy_file_range().
1550 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE
1551 * for server-side-copy between any two sb.
1553 * In any case, we call do_splice_direct() and not splice_file_range(),
1554 * without file_start_write() held, to avoid possible deadlocks related
1555 * to splicing from input file, while file_start_write() is held on
1556 * the output file on a different sb.
1558 ret
= do_splice_direct(file_in
, &pos_in
, file_out
, &pos_out
,
1559 min_t(size_t, len
, MAX_RW_COUNT
), 0);
1562 fsnotify_access(file_in
);
1563 add_rchar(current
, ret
);
1564 fsnotify_modify(file_out
);
1565 add_wchar(current
, ret
);
1573 EXPORT_SYMBOL(vfs_copy_file_range
);
1575 SYSCALL_DEFINE6(copy_file_range
, int, fd_in
, loff_t __user
*, off_in
,
1576 int, fd_out
, loff_t __user
*, off_out
,
1577 size_t, len
, unsigned int, flags
)
1583 ssize_t ret
= -EBADF
;
1585 f_in
= fdget(fd_in
);
1589 f_out
= fdget(fd_out
);
1595 if (copy_from_user(&pos_in
, off_in
, sizeof(loff_t
)))
1598 pos_in
= f_in
.file
->f_pos
;
1602 if (copy_from_user(&pos_out
, off_out
, sizeof(loff_t
)))
1605 pos_out
= f_out
.file
->f_pos
;
1612 ret
= vfs_copy_file_range(f_in
.file
, pos_in
, f_out
.file
, pos_out
, len
,
1619 if (copy_to_user(off_in
, &pos_in
, sizeof(loff_t
)))
1622 f_in
.file
->f_pos
= pos_in
;
1626 if (copy_to_user(off_out
, &pos_out
, sizeof(loff_t
)))
1629 f_out
.file
->f_pos
= pos_out
;
1642 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1643 * LFS limits. If pos is under the limit it becomes a short access. If it
1644 * exceeds the limit we return -EFBIG.
1646 int generic_write_check_limits(struct file
*file
, loff_t pos
, loff_t
*count
)
1648 struct inode
*inode
= file
->f_mapping
->host
;
1649 loff_t max_size
= inode
->i_sb
->s_maxbytes
;
1650 loff_t limit
= rlimit(RLIMIT_FSIZE
);
1652 if (limit
!= RLIM_INFINITY
) {
1654 send_sig(SIGXFSZ
, current
, 0);
1657 *count
= min(*count
, limit
- pos
);
1660 if (!(file
->f_flags
& O_LARGEFILE
))
1661 max_size
= MAX_NON_LFS
;
1663 if (unlikely(pos
>= max_size
))
1666 *count
= min(*count
, max_size
- pos
);
1671 /* Like generic_write_checks(), but takes size of write instead of iter. */
1672 int generic_write_checks_count(struct kiocb
*iocb
, loff_t
*count
)
1674 struct file
*file
= iocb
->ki_filp
;
1675 struct inode
*inode
= file
->f_mapping
->host
;
1677 if (IS_SWAPFILE(inode
))
1683 if (iocb
->ki_flags
& IOCB_APPEND
)
1684 iocb
->ki_pos
= i_size_read(inode
);
1686 if ((iocb
->ki_flags
& IOCB_NOWAIT
) &&
1687 !((iocb
->ki_flags
& IOCB_DIRECT
) ||
1688 (file
->f_mode
& FMODE_BUF_WASYNC
)))
1691 return generic_write_check_limits(iocb
->ki_filp
, iocb
->ki_pos
, count
);
1693 EXPORT_SYMBOL(generic_write_checks_count
);
1696 * Performs necessary checks before doing a write
1698 * Can adjust writing position or amount of bytes to write.
1699 * Returns appropriate error code that caller should return or
1700 * zero in case that write should be allowed.
1702 ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
1704 loff_t count
= iov_iter_count(from
);
1707 ret
= generic_write_checks_count(iocb
, &count
);
1711 iov_iter_truncate(from
, count
);
1712 return iov_iter_count(from
);
1714 EXPORT_SYMBOL(generic_write_checks
);
1717 * Performs common checks before doing a file copy/clone
1718 * from @file_in to @file_out.
1720 int generic_file_rw_checks(struct file
*file_in
, struct file
*file_out
)
1722 struct inode
*inode_in
= file_inode(file_in
);
1723 struct inode
*inode_out
= file_inode(file_out
);
1725 /* Don't copy dirs, pipes, sockets... */
1726 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1728 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1731 if (!(file_in
->f_mode
& FMODE_READ
) ||
1732 !(file_out
->f_mode
& FMODE_WRITE
) ||
1733 (file_out
->f_flags
& O_APPEND
))