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>
21 #include <asm/uaccess.h>
22 #include <asm/unistd.h>
24 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
25 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
27 const struct file_operations generic_ro_fops = {
28 .llseek = generic_file_llseek,
29 .read_iter = generic_file_read_iter,
30 .mmap = generic_file_readonly_mmap,
31 .splice_read = generic_file_splice_read,
34 EXPORT_SYMBOL(generic_ro_fops);
36 static inline int unsigned_offsets(struct file *file)
38 return file->f_mode & FMODE_UNSIGNED_OFFSET;
42 * vfs_setpos - update the file offset for lseek
43 * @file: file structure in question
44 * @offset: file offset to seek to
45 * @maxsize: maximum file size
47 * This is a low-level filesystem helper for updating the file offset to
48 * the value specified by @offset if the given offset is valid and it is
49 * not equal to the current file offset.
51 * Return the specified offset on success and -EINVAL on invalid offset.
53 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
55 if (offset < 0 && !unsigned_offsets(file))
60 if (offset != file->f_pos) {
66 EXPORT_SYMBOL(vfs_setpos);
69 * generic_file_llseek_size - generic llseek implementation for regular files
70 * @file: file structure to seek on
71 * @offset: file offset to seek to
72 * @whence: type of seek
73 * @size: max size of this file in file system
74 * @eof: offset used for SEEK_END position
76 * This is a variant of generic_file_llseek that allows passing in a custom
77 * maximum file size and a custom EOF position, for e.g. hashed directories
80 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
81 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
82 * read/writes behave like SEEK_SET against seeks.
85 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
86 loff_t maxsize, loff_t eof)
94 * Here we special-case the lseek(fd, 0, SEEK_CUR)
95 * position-querying operation. Avoid rewriting the "same"
96 * f_pos value back to the file because a concurrent read(),
97 * write() or lseek() might have altered it
102 * f_lock protects against read/modify/write race with other
103 * SEEK_CURs. Note that parallel writes and reads behave
106 spin_lock(&file->f_lock);
107 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
108 spin_unlock(&file->f_lock);
112 * In the generic case the entire file is data, so as long as
113 * offset isn't at the end of the file then the offset is data.
115 if ((unsigned long long)offset >= eof)
120 * There is a virtual hole at the end of the file, so as long as
121 * offset isn't i_size or larger, return i_size.
123 if ((unsigned long long)offset >= eof)
129 return vfs_setpos(file, offset, maxsize);
131 EXPORT_SYMBOL(generic_file_llseek_size);
134 * generic_file_llseek - generic llseek implementation for regular files
135 * @file: file structure to seek on
136 * @offset: file offset to seek to
137 * @whence: type of seek
139 * This is a generic implemenation of ->llseek useable for all normal local
140 * filesystems. It just updates the file offset to the value specified by
141 * @offset and @whence.
143 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
145 struct inode *inode = file->f_mapping->host;
147 return generic_file_llseek_size(file, offset, whence,
148 inode->i_sb->s_maxbytes,
151 EXPORT_SYMBOL(generic_file_llseek);
154 * fixed_size_llseek - llseek implementation for fixed-sized devices
155 * @file: file structure to seek on
156 * @offset: file offset to seek to
157 * @whence: type of seek
158 * @size: size of the file
161 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
164 case SEEK_SET: case SEEK_CUR: case SEEK_END:
165 return generic_file_llseek_size(file, offset, whence,
171 EXPORT_SYMBOL(fixed_size_llseek);
174 * noop_llseek - No Operation Performed llseek implementation
175 * @file: file structure to seek on
176 * @offset: file offset to seek to
177 * @whence: type of seek
179 * This is an implementation of ->llseek useable for the rare special case when
180 * userspace expects the seek to succeed but the (device) file is actually not
181 * able to perform the seek. In this case you use noop_llseek() instead of
182 * falling back to the default implementation of ->llseek.
184 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
188 EXPORT_SYMBOL(noop_llseek);
190 loff_t no_llseek(struct file *file, loff_t offset, int whence)
194 EXPORT_SYMBOL(no_llseek);
196 loff_t default_llseek(struct file *file, loff_t offset, int whence)
198 struct inode *inode = file_inode(file);
201 mutex_lock(&inode->i_mutex);
204 offset += i_size_read(inode);
208 retval = file->f_pos;
211 offset += file->f_pos;
215 * In the generic case the entire file is data, so as
216 * long as offset isn't at the end of the file then the
219 if (offset >= inode->i_size) {
226 * There is a virtual hole at the end of the file, so
227 * as long as offset isn't i_size or larger, return
230 if (offset >= inode->i_size) {
234 offset = inode->i_size;
238 if (offset >= 0 || unsigned_offsets(file)) {
239 if (offset != file->f_pos) {
240 file->f_pos = offset;
246 mutex_unlock(&inode->i_mutex);
249 EXPORT_SYMBOL(default_llseek);
251 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
253 loff_t (*fn)(struct file *, loff_t, int);
256 if (file->f_mode & FMODE_LSEEK) {
257 if (file->f_op->llseek)
258 fn = file->f_op->llseek;
260 return fn(file, offset, whence);
262 EXPORT_SYMBOL(vfs_llseek);
264 static inline struct fd fdget_pos(int fd)
266 return __to_fd(__fdget_pos(fd));
269 static inline void fdput_pos(struct fd f)
271 if (f.flags & FDPUT_POS_UNLOCK)
272 mutex_unlock(&f.file->f_pos_lock);
276 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
279 struct fd f = fdget_pos(fd);
284 if (whence <= SEEK_MAX) {
285 loff_t res = vfs_llseek(f.file, offset, whence);
287 if (res != (loff_t)retval)
288 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
295 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
297 return sys_lseek(fd, offset, whence);
301 #ifdef __ARCH_WANT_SYS_LLSEEK
302 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
303 unsigned long, offset_low, loff_t __user *, result,
304 unsigned int, whence)
307 struct fd f = fdget_pos(fd);
314 if (whence > SEEK_MAX)
317 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
320 retval = (int)offset;
323 if (!copy_to_user(result, &offset, sizeof(offset)))
332 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
337 if (!file->f_op->read_iter)
340 init_sync_kiocb(&kiocb, file);
341 kiocb.ki_pos = *ppos;
344 ret = file->f_op->read_iter(&kiocb, iter);
345 BUG_ON(ret == -EIOCBQUEUED);
347 *ppos = kiocb.ki_pos;
350 EXPORT_SYMBOL(vfs_iter_read);
352 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
357 if (!file->f_op->write_iter)
360 init_sync_kiocb(&kiocb, file);
361 kiocb.ki_pos = *ppos;
364 ret = file->f_op->write_iter(&kiocb, iter);
365 BUG_ON(ret == -EIOCBQUEUED);
367 *ppos = kiocb.ki_pos;
368 fsnotify_modify(file);
372 EXPORT_SYMBOL(vfs_iter_write);
375 * rw_verify_area doesn't like huge counts. We limit
376 * them to something that fits in "int" so that others
377 * won't have to do range checks all the time.
379 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
383 int retval = -EINVAL;
385 inode = file_inode(file);
386 if (unlikely((ssize_t) count < 0))
389 if (unlikely(pos < 0)) {
390 if (!unsigned_offsets(file))
392 if (count >= -pos) /* both values are in 0..LLONG_MAX */
394 } else if (unlikely((loff_t) (pos + count) < 0)) {
395 if (!unsigned_offsets(file))
399 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
400 retval = locks_mandatory_area(
401 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
402 inode, file, pos, count);
406 retval = security_file_permission(file,
407 read_write == READ ? MAY_READ : MAY_WRITE);
410 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
413 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
415 struct iovec iov = { .iov_base = buf, .iov_len = len };
417 struct iov_iter iter;
420 init_sync_kiocb(&kiocb, filp);
421 kiocb.ki_pos = *ppos;
422 iov_iter_init(&iter, READ, &iov, 1, len);
424 ret = filp->f_op->read_iter(&kiocb, &iter);
425 BUG_ON(ret == -EIOCBQUEUED);
426 *ppos = kiocb.ki_pos;
430 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
433 if (file->f_op->read)
434 return file->f_op->read(file, buf, count, pos);
435 else if (file->f_op->read_iter)
436 return new_sync_read(file, buf, count, pos);
440 EXPORT_SYMBOL(__vfs_read);
442 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
446 if (!(file->f_mode & FMODE_READ))
448 if (!(file->f_mode & FMODE_CAN_READ))
450 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
453 ret = rw_verify_area(READ, file, pos, count);
456 ret = __vfs_read(file, buf, count, pos);
458 fsnotify_access(file);
459 add_rchar(current, ret);
467 EXPORT_SYMBOL(vfs_read);
469 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
471 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
473 struct iov_iter iter;
476 init_sync_kiocb(&kiocb, filp);
477 kiocb.ki_pos = *ppos;
478 iov_iter_init(&iter, WRITE, &iov, 1, len);
480 ret = filp->f_op->write_iter(&kiocb, &iter);
481 BUG_ON(ret == -EIOCBQUEUED);
483 *ppos = kiocb.ki_pos;
487 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
490 if (file->f_op->write)
491 return file->f_op->write(file, p, count, pos);
492 else if (file->f_op->write_iter)
493 return new_sync_write(file, p, count, pos);
497 EXPORT_SYMBOL(__vfs_write);
499 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
502 const char __user *p;
505 if (!(file->f_mode & FMODE_CAN_WRITE))
510 p = (__force const char __user *)buf;
511 if (count > MAX_RW_COUNT)
512 count = MAX_RW_COUNT;
513 ret = __vfs_write(file, p, count, pos);
516 fsnotify_modify(file);
517 add_wchar(current, ret);
523 EXPORT_SYMBOL(__kernel_write);
525 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
529 if (!(file->f_mode & FMODE_WRITE))
531 if (!(file->f_mode & FMODE_CAN_WRITE))
533 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
536 ret = rw_verify_area(WRITE, file, pos, count);
539 file_start_write(file);
540 ret = __vfs_write(file, buf, count, pos);
542 fsnotify_modify(file);
543 add_wchar(current, ret);
546 file_end_write(file);
552 EXPORT_SYMBOL(vfs_write);
554 static inline loff_t file_pos_read(struct file *file)
559 static inline void file_pos_write(struct file *file, loff_t pos)
564 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
566 struct fd f = fdget_pos(fd);
567 ssize_t ret = -EBADF;
570 loff_t pos = file_pos_read(f.file);
571 ret = vfs_read(f.file, buf, count, &pos);
573 file_pos_write(f.file, pos);
579 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
582 struct fd f = fdget_pos(fd);
583 ssize_t ret = -EBADF;
586 loff_t pos = file_pos_read(f.file);
587 ret = vfs_write(f.file, buf, count, &pos);
589 file_pos_write(f.file, pos);
596 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
597 size_t, count, loff_t, pos)
600 ssize_t ret = -EBADF;
608 if (f.file->f_mode & FMODE_PREAD)
609 ret = vfs_read(f.file, buf, count, &pos);
616 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
617 size_t, count, loff_t, pos)
620 ssize_t ret = -EBADF;
628 if (f.file->f_mode & FMODE_PWRITE)
629 ret = vfs_write(f.file, buf, count, &pos);
637 * Reduce an iovec's length in-place. Return the resulting number of segments
639 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
641 unsigned long seg = 0;
644 while (seg < nr_segs) {
646 if (len + iov->iov_len >= to) {
647 iov->iov_len = to - len;
655 EXPORT_SYMBOL(iov_shorten);
657 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
658 loff_t *ppos, iter_fn_t fn)
663 init_sync_kiocb(&kiocb, filp);
664 kiocb.ki_pos = *ppos;
666 ret = fn(&kiocb, iter);
667 BUG_ON(ret == -EIOCBQUEUED);
668 *ppos = kiocb.ki_pos;
672 /* Do it by hand, with file-ops */
673 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
674 loff_t *ppos, io_fn_t fn)
678 while (iov_iter_count(iter)) {
679 struct iovec iovec = iov_iter_iovec(iter);
682 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
690 if (nr != iovec.iov_len)
692 iov_iter_advance(iter, nr);
698 /* A write operation does a read from user space and vice versa */
699 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
701 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
702 unsigned long nr_segs, unsigned long fast_segs,
703 struct iovec *fast_pointer,
704 struct iovec **ret_pointer)
708 struct iovec *iov = fast_pointer;
711 * SuS says "The readv() function *may* fail if the iovcnt argument
712 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
713 * traditionally returned zero for zero segments, so...
721 * First get the "struct iovec" from user memory and
722 * verify all the pointers
724 if (nr_segs > UIO_MAXIOV) {
728 if (nr_segs > fast_segs) {
729 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
735 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
741 * According to the Single Unix Specification we should return EINVAL
742 * if an element length is < 0 when cast to ssize_t or if the
743 * total length would overflow the ssize_t return value of the
746 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
750 for (seg = 0; seg < nr_segs; seg++) {
751 void __user *buf = iov[seg].iov_base;
752 ssize_t len = (ssize_t)iov[seg].iov_len;
754 /* see if we we're about to use an invalid len or if
755 * it's about to overflow ssize_t */
761 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
765 if (len > MAX_RW_COUNT - ret) {
766 len = MAX_RW_COUNT - ret;
767 iov[seg].iov_len = len;
776 static ssize_t do_readv_writev(int type, struct file *file,
777 const struct iovec __user * uvector,
778 unsigned long nr_segs, loff_t *pos)
781 struct iovec iovstack[UIO_FASTIOV];
782 struct iovec *iov = iovstack;
783 struct iov_iter iter;
788 ret = import_iovec(type, uvector, nr_segs,
789 ARRAY_SIZE(iovstack), &iov, &iter);
793 tot_len = iov_iter_count(&iter);
796 ret = rw_verify_area(type, file, pos, tot_len);
801 fn = file->f_op->read;
802 iter_fn = file->f_op->read_iter;
804 fn = (io_fn_t)file->f_op->write;
805 iter_fn = file->f_op->write_iter;
806 file_start_write(file);
810 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
812 ret = do_loop_readv_writev(file, &iter, pos, fn);
815 file_end_write(file);
819 if ((ret + (type == READ)) > 0) {
821 fsnotify_access(file);
823 fsnotify_modify(file);
828 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
829 unsigned long vlen, loff_t *pos)
831 if (!(file->f_mode & FMODE_READ))
833 if (!(file->f_mode & FMODE_CAN_READ))
836 return do_readv_writev(READ, file, vec, vlen, pos);
839 EXPORT_SYMBOL(vfs_readv);
841 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
842 unsigned long vlen, loff_t *pos)
844 if (!(file->f_mode & FMODE_WRITE))
846 if (!(file->f_mode & FMODE_CAN_WRITE))
849 return do_readv_writev(WRITE, file, vec, vlen, pos);
852 EXPORT_SYMBOL(vfs_writev);
854 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
857 struct fd f = fdget_pos(fd);
858 ssize_t ret = -EBADF;
861 loff_t pos = file_pos_read(f.file);
862 ret = vfs_readv(f.file, vec, vlen, &pos);
864 file_pos_write(f.file, pos);
869 add_rchar(current, ret);
874 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
877 struct fd f = fdget_pos(fd);
878 ssize_t ret = -EBADF;
881 loff_t pos = file_pos_read(f.file);
882 ret = vfs_writev(f.file, vec, vlen, &pos);
884 file_pos_write(f.file, pos);
889 add_wchar(current, ret);
894 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
896 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
897 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
900 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
901 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
903 loff_t pos = pos_from_hilo(pos_h, pos_l);
905 ssize_t ret = -EBADF;
913 if (f.file->f_mode & FMODE_PREAD)
914 ret = vfs_readv(f.file, vec, vlen, &pos);
919 add_rchar(current, ret);
924 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
925 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
927 loff_t pos = pos_from_hilo(pos_h, pos_l);
929 ssize_t ret = -EBADF;
937 if (f.file->f_mode & FMODE_PWRITE)
938 ret = vfs_writev(f.file, vec, vlen, &pos);
943 add_wchar(current, ret);
950 static ssize_t compat_do_readv_writev(int type, struct file *file,
951 const struct compat_iovec __user *uvector,
952 unsigned long nr_segs, loff_t *pos)
954 compat_ssize_t tot_len;
955 struct iovec iovstack[UIO_FASTIOV];
956 struct iovec *iov = iovstack;
957 struct iov_iter iter;
962 ret = compat_import_iovec(type, uvector, nr_segs,
963 UIO_FASTIOV, &iov, &iter);
967 tot_len = iov_iter_count(&iter);
970 ret = rw_verify_area(type, file, pos, tot_len);
975 fn = file->f_op->read;
976 iter_fn = file->f_op->read_iter;
978 fn = (io_fn_t)file->f_op->write;
979 iter_fn = file->f_op->write_iter;
980 file_start_write(file);
984 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
986 ret = do_loop_readv_writev(file, &iter, pos, fn);
989 file_end_write(file);
993 if ((ret + (type == READ)) > 0) {
995 fsnotify_access(file);
997 fsnotify_modify(file);
1002 static size_t compat_readv(struct file *file,
1003 const struct compat_iovec __user *vec,
1004 unsigned long vlen, loff_t *pos)
1006 ssize_t ret = -EBADF;
1008 if (!(file->f_mode & FMODE_READ))
1012 if (!(file->f_mode & FMODE_CAN_READ))
1015 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1019 add_rchar(current, ret);
1024 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1025 const struct compat_iovec __user *,vec,
1026 compat_ulong_t, vlen)
1028 struct fd f = fdget_pos(fd);
1034 pos = f.file->f_pos;
1035 ret = compat_readv(f.file, vec, vlen, &pos);
1037 f.file->f_pos = pos;
1042 static long __compat_sys_preadv64(unsigned long fd,
1043 const struct compat_iovec __user *vec,
1044 unsigned long vlen, loff_t pos)
1055 if (f.file->f_mode & FMODE_PREAD)
1056 ret = compat_readv(f.file, vec, vlen, &pos);
1061 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1062 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1063 const struct compat_iovec __user *,vec,
1064 unsigned long, vlen, loff_t, pos)
1066 return __compat_sys_preadv64(fd, vec, vlen, pos);
1070 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1071 const struct compat_iovec __user *,vec,
1072 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1074 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1076 return __compat_sys_preadv64(fd, vec, vlen, pos);
1079 static size_t compat_writev(struct file *file,
1080 const struct compat_iovec __user *vec,
1081 unsigned long vlen, loff_t *pos)
1083 ssize_t ret = -EBADF;
1085 if (!(file->f_mode & FMODE_WRITE))
1089 if (!(file->f_mode & FMODE_CAN_WRITE))
1092 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1096 add_wchar(current, ret);
1101 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1102 const struct compat_iovec __user *, vec,
1103 compat_ulong_t, vlen)
1105 struct fd f = fdget_pos(fd);
1111 pos = f.file->f_pos;
1112 ret = compat_writev(f.file, vec, vlen, &pos);
1114 f.file->f_pos = pos;
1119 static long __compat_sys_pwritev64(unsigned long fd,
1120 const struct compat_iovec __user *vec,
1121 unsigned long vlen, loff_t pos)
1132 if (f.file->f_mode & FMODE_PWRITE)
1133 ret = compat_writev(f.file, vec, vlen, &pos);
1138 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1139 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1140 const struct compat_iovec __user *,vec,
1141 unsigned long, vlen, loff_t, pos)
1143 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1147 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1148 const struct compat_iovec __user *,vec,
1149 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1151 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1153 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1157 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1158 size_t count, loff_t max)
1161 struct inode *in_inode, *out_inode;
1168 * Get input file, and verify that it is ok..
1174 if (!(in.file->f_mode & FMODE_READ))
1178 pos = in.file->f_pos;
1181 if (!(in.file->f_mode & FMODE_PREAD))
1184 retval = rw_verify_area(READ, in.file, &pos, count);
1190 * Get output file, and verify that it is ok..
1193 out = fdget(out_fd);
1196 if (!(out.file->f_mode & FMODE_WRITE))
1199 in_inode = file_inode(in.file);
1200 out_inode = file_inode(out.file);
1201 out_pos = out.file->f_pos;
1202 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1208 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1210 if (unlikely(pos + count > max)) {
1211 retval = -EOVERFLOW;
1220 * We need to debate whether we can enable this or not. The
1221 * man page documents EAGAIN return for the output at least,
1222 * and the application is arguably buggy if it doesn't expect
1223 * EAGAIN on a non-blocking file descriptor.
1225 if (in.file->f_flags & O_NONBLOCK)
1226 fl = SPLICE_F_NONBLOCK;
1228 file_start_write(out.file);
1229 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1230 file_end_write(out.file);
1233 add_rchar(current, retval);
1234 add_wchar(current, retval);
1235 fsnotify_access(in.file);
1236 fsnotify_modify(out.file);
1237 out.file->f_pos = out_pos;
1241 in.file->f_pos = pos;
1247 retval = -EOVERFLOW;
1257 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1264 if (unlikely(get_user(off, offset)))
1267 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1268 if (unlikely(put_user(pos, offset)))
1273 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1276 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1282 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1284 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1285 if (unlikely(put_user(pos, offset)))
1290 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1293 #ifdef CONFIG_COMPAT
1294 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1295 compat_off_t __user *, offset, compat_size_t, count)
1302 if (unlikely(get_user(off, offset)))
1305 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1306 if (unlikely(put_user(pos, offset)))
1311 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1314 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1315 compat_loff_t __user *, offset, compat_size_t, count)
1321 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1323 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1324 if (unlikely(put_user(pos, offset)))
1329 return do_sendfile(out_fd, in_fd, NULL, count, 0);