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/aio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
22 #include <asm/uaccess.h>
23 #include <asm/unistd.h>
25 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
26 typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *,
27 unsigned long, loff_t);
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
32 .aio_read = generic_file_aio_read,
33 .mmap = generic_file_readonly_mmap,
34 .splice_read = generic_file_splice_read,
37 EXPORT_SYMBOL(generic_ro_fops);
39 static inline int unsigned_offsets(struct file *file)
41 return file->f_mode & FMODE_UNSIGNED_OFFSET;
45 * vfs_setpos - update the file offset for lseek
46 * @file: file structure in question
47 * @offset: file offset to seek to
48 * @maxsize: maximum file size
50 * This is a low-level filesystem helper for updating the file offset to
51 * the value specified by @offset if the given offset is valid and it is
52 * not equal to the current file offset.
54 * Return the specified offset on success and -EINVAL on invalid offset.
56 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
58 if (offset < 0 && !unsigned_offsets(file))
63 if (offset != file->f_pos) {
69 EXPORT_SYMBOL(vfs_setpos);
72 * generic_file_llseek_size - generic llseek implementation for regular files
73 * @file: file structure to seek on
74 * @offset: file offset to seek to
75 * @whence: type of seek
76 * @size: max size of this file in file system
77 * @eof: offset used for SEEK_END position
79 * This is a variant of generic_file_llseek that allows passing in a custom
80 * maximum file size and a custom EOF position, for e.g. hashed directories
83 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
84 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
85 * read/writes behave like SEEK_SET against seeks.
88 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
89 loff_t maxsize, loff_t eof)
97 * Here we special-case the lseek(fd, 0, SEEK_CUR)
98 * position-querying operation. Avoid rewriting the "same"
99 * f_pos value back to the file because a concurrent read(),
100 * write() or lseek() might have altered it
105 * f_lock protects against read/modify/write race with other
106 * SEEK_CURs. Note that parallel writes and reads behave
109 spin_lock(&file->f_lock);
110 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
111 spin_unlock(&file->f_lock);
115 * In the generic case the entire file is data, so as long as
116 * offset isn't at the end of the file then the offset is data.
123 * There is a virtual hole at the end of the file, so as long as
124 * offset isn't i_size or larger, return i_size.
132 return vfs_setpos(file, offset, maxsize);
134 EXPORT_SYMBOL(generic_file_llseek_size);
137 * generic_file_llseek - generic llseek implementation for regular files
138 * @file: file structure to seek on
139 * @offset: file offset to seek to
140 * @whence: type of seek
142 * This is a generic implemenation of ->llseek useable for all normal local
143 * filesystems. It just updates the file offset to the value specified by
144 * @offset and @whence.
146 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
148 struct inode *inode = file->f_mapping->host;
150 return generic_file_llseek_size(file, offset, whence,
151 inode->i_sb->s_maxbytes,
154 EXPORT_SYMBOL(generic_file_llseek);
157 * fixed_size_llseek - llseek implementation for fixed-sized devices
158 * @file: file structure to seek on
159 * @offset: file offset to seek to
160 * @whence: type of seek
161 * @size: size of the file
164 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
167 case SEEK_SET: case SEEK_CUR: case SEEK_END:
168 return generic_file_llseek_size(file, offset, whence,
174 EXPORT_SYMBOL(fixed_size_llseek);
177 * noop_llseek - No Operation Performed llseek implementation
178 * @file: file structure to seek on
179 * @offset: file offset to seek to
180 * @whence: type of seek
182 * This is an implementation of ->llseek useable for the rare special case when
183 * userspace expects the seek to succeed but the (device) file is actually not
184 * able to perform the seek. In this case you use noop_llseek() instead of
185 * falling back to the default implementation of ->llseek.
187 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
191 EXPORT_SYMBOL(noop_llseek);
193 loff_t no_llseek(struct file *file, loff_t offset, int whence)
197 EXPORT_SYMBOL(no_llseek);
199 loff_t default_llseek(struct file *file, loff_t offset, int whence)
201 struct inode *inode = file_inode(file);
204 mutex_lock(&inode->i_mutex);
207 offset += i_size_read(inode);
211 retval = file->f_pos;
214 offset += file->f_pos;
218 * In the generic case the entire file is data, so as
219 * long as offset isn't at the end of the file then the
222 if (offset >= inode->i_size) {
229 * There is a virtual hole at the end of the file, so
230 * as long as offset isn't i_size or larger, return
233 if (offset >= inode->i_size) {
237 offset = inode->i_size;
241 if (offset >= 0 || unsigned_offsets(file)) {
242 if (offset != file->f_pos) {
243 file->f_pos = offset;
249 mutex_unlock(&inode->i_mutex);
252 EXPORT_SYMBOL(default_llseek);
254 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
256 loff_t (*fn)(struct file *, loff_t, int);
259 if (file->f_mode & FMODE_LSEEK) {
260 if (file->f_op && file->f_op->llseek)
261 fn = file->f_op->llseek;
263 return fn(file, offset, whence);
265 EXPORT_SYMBOL(vfs_llseek);
267 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
270 struct fd f = fdget(fd);
275 if (whence <= SEEK_MAX) {
276 loff_t res = vfs_llseek(f.file, offset, whence);
278 if (res != (loff_t)retval)
279 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
286 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
288 return sys_lseek(fd, offset, whence);
292 #ifdef __ARCH_WANT_SYS_LLSEEK
293 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
294 unsigned long, offset_low, loff_t __user *, result,
295 unsigned int, whence)
298 struct fd f = fdget(fd);
305 if (whence > SEEK_MAX)
308 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
311 retval = (int)offset;
314 if (!copy_to_user(result, &offset, sizeof(offset)))
324 * rw_verify_area doesn't like huge counts. We limit
325 * them to something that fits in "int" so that others
326 * won't have to do range checks all the time.
328 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
332 int retval = -EINVAL;
334 inode = file_inode(file);
335 if (unlikely((ssize_t) count < 0))
338 if (unlikely(pos < 0)) {
339 if (!unsigned_offsets(file))
341 if (count >= -pos) /* both values are in 0..LLONG_MAX */
343 } else if (unlikely((loff_t) (pos + count) < 0)) {
344 if (!unsigned_offsets(file))
348 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
349 retval = locks_mandatory_area(
350 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
351 inode, file, pos, count);
355 retval = security_file_permission(file,
356 read_write == READ ? MAY_READ : MAY_WRITE);
359 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
362 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
364 struct iovec iov = { .iov_base = buf, .iov_len = len };
368 init_sync_kiocb(&kiocb, filp);
369 kiocb.ki_pos = *ppos;
371 kiocb.ki_nbytes = len;
373 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
374 if (-EIOCBQUEUED == ret)
375 ret = wait_on_sync_kiocb(&kiocb);
376 *ppos = kiocb.ki_pos;
380 EXPORT_SYMBOL(do_sync_read);
382 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
386 if (!(file->f_mode & FMODE_READ))
388 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
390 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
393 ret = rw_verify_area(READ, file, pos, count);
396 if (file->f_op->read)
397 ret = file->f_op->read(file, buf, count, pos);
399 ret = do_sync_read(file, buf, count, pos);
401 fsnotify_access(file);
402 add_rchar(current, ret);
410 EXPORT_SYMBOL(vfs_read);
412 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
414 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
418 init_sync_kiocb(&kiocb, filp);
419 kiocb.ki_pos = *ppos;
421 kiocb.ki_nbytes = len;
423 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
424 if (-EIOCBQUEUED == ret)
425 ret = wait_on_sync_kiocb(&kiocb);
426 *ppos = kiocb.ki_pos;
430 EXPORT_SYMBOL(do_sync_write);
432 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
435 const char __user *p;
438 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
443 p = (__force const char __user *)buf;
444 if (count > MAX_RW_COUNT)
445 count = MAX_RW_COUNT;
446 if (file->f_op->write)
447 ret = file->f_op->write(file, p, count, pos);
449 ret = do_sync_write(file, p, count, pos);
452 fsnotify_modify(file);
453 add_wchar(current, ret);
459 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
463 if (!(file->f_mode & FMODE_WRITE))
465 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
467 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
470 ret = rw_verify_area(WRITE, file, pos, count);
473 file_start_write(file);
474 if (file->f_op->write)
475 ret = file->f_op->write(file, buf, count, pos);
477 ret = do_sync_write(file, buf, count, pos);
479 fsnotify_modify(file);
480 add_wchar(current, ret);
483 file_end_write(file);
489 EXPORT_SYMBOL(vfs_write);
491 static inline loff_t file_pos_read(struct file *file)
496 static inline void file_pos_write(struct file *file, loff_t pos)
501 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
503 struct fd f = fdget(fd);
504 ssize_t ret = -EBADF;
507 loff_t pos = file_pos_read(f.file);
508 ret = vfs_read(f.file, buf, count, &pos);
510 file_pos_write(f.file, pos);
516 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
519 struct fd f = fdget(fd);
520 ssize_t ret = -EBADF;
523 loff_t pos = file_pos_read(f.file);
524 ret = vfs_write(f.file, buf, count, &pos);
526 file_pos_write(f.file, pos);
533 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
534 size_t, count, loff_t, pos)
537 ssize_t ret = -EBADF;
545 if (f.file->f_mode & FMODE_PREAD)
546 ret = vfs_read(f.file, buf, count, &pos);
553 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
554 size_t, count, loff_t, pos)
557 ssize_t ret = -EBADF;
565 if (f.file->f_mode & FMODE_PWRITE)
566 ret = vfs_write(f.file, buf, count, &pos);
574 * Reduce an iovec's length in-place. Return the resulting number of segments
576 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
578 unsigned long seg = 0;
581 while (seg < nr_segs) {
583 if (len + iov->iov_len >= to) {
584 iov->iov_len = to - len;
592 EXPORT_SYMBOL(iov_shorten);
594 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
595 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
600 init_sync_kiocb(&kiocb, filp);
601 kiocb.ki_pos = *ppos;
603 kiocb.ki_nbytes = len;
605 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
606 if (ret == -EIOCBQUEUED)
607 ret = wait_on_sync_kiocb(&kiocb);
608 *ppos = kiocb.ki_pos;
612 /* Do it by hand, with file-ops */
613 static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
614 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
616 struct iovec *vector = iov;
619 while (nr_segs > 0) {
624 base = vector->iov_base;
625 len = vector->iov_len;
629 nr = fn(filp, base, len, ppos);
644 /* A write operation does a read from user space and vice versa */
645 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
647 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
648 unsigned long nr_segs, unsigned long fast_segs,
649 struct iovec *fast_pointer,
650 struct iovec **ret_pointer)
654 struct iovec *iov = fast_pointer;
657 * SuS says "The readv() function *may* fail if the iovcnt argument
658 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
659 * traditionally returned zero for zero segments, so...
667 * First get the "struct iovec" from user memory and
668 * verify all the pointers
670 if (nr_segs > UIO_MAXIOV) {
674 if (nr_segs > fast_segs) {
675 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
681 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
687 * According to the Single Unix Specification we should return EINVAL
688 * if an element length is < 0 when cast to ssize_t or if the
689 * total length would overflow the ssize_t return value of the
692 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
696 for (seg = 0; seg < nr_segs; seg++) {
697 void __user *buf = iov[seg].iov_base;
698 ssize_t len = (ssize_t)iov[seg].iov_len;
700 /* see if we we're about to use an invalid len or if
701 * it's about to overflow ssize_t */
707 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
711 if (len > MAX_RW_COUNT - ret) {
712 len = MAX_RW_COUNT - ret;
713 iov[seg].iov_len = len;
722 static ssize_t do_readv_writev(int type, struct file *file,
723 const struct iovec __user * uvector,
724 unsigned long nr_segs, loff_t *pos)
727 struct iovec iovstack[UIO_FASTIOV];
728 struct iovec *iov = iovstack;
738 ret = rw_copy_check_uvector(type, uvector, nr_segs,
739 ARRAY_SIZE(iovstack), iovstack, &iov);
744 ret = rw_verify_area(type, file, pos, tot_len);
750 fn = file->f_op->read;
751 fnv = file->f_op->aio_read;
753 fn = (io_fn_t)file->f_op->write;
754 fnv = file->f_op->aio_write;
755 file_start_write(file);
759 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
762 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
765 file_end_write(file);
770 if ((ret + (type == READ)) > 0) {
772 fsnotify_access(file);
774 fsnotify_modify(file);
779 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
780 unsigned long vlen, loff_t *pos)
782 if (!(file->f_mode & FMODE_READ))
784 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
787 return do_readv_writev(READ, file, vec, vlen, pos);
790 EXPORT_SYMBOL(vfs_readv);
792 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
793 unsigned long vlen, loff_t *pos)
795 if (!(file->f_mode & FMODE_WRITE))
797 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
800 return do_readv_writev(WRITE, file, vec, vlen, pos);
803 EXPORT_SYMBOL(vfs_writev);
805 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
808 struct fd f = fdget(fd);
809 ssize_t ret = -EBADF;
812 loff_t pos = file_pos_read(f.file);
813 ret = vfs_readv(f.file, vec, vlen, &pos);
815 file_pos_write(f.file, pos);
820 add_rchar(current, ret);
825 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
828 struct fd f = fdget(fd);
829 ssize_t ret = -EBADF;
832 loff_t pos = file_pos_read(f.file);
833 ret = vfs_writev(f.file, vec, vlen, &pos);
835 file_pos_write(f.file, pos);
840 add_wchar(current, ret);
845 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
847 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
848 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
851 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
852 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
854 loff_t pos = pos_from_hilo(pos_h, pos_l);
856 ssize_t ret = -EBADF;
864 if (f.file->f_mode & FMODE_PREAD)
865 ret = vfs_readv(f.file, vec, vlen, &pos);
870 add_rchar(current, ret);
875 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
876 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
878 loff_t pos = pos_from_hilo(pos_h, pos_l);
880 ssize_t ret = -EBADF;
888 if (f.file->f_mode & FMODE_PWRITE)
889 ret = vfs_writev(f.file, vec, vlen, &pos);
894 add_wchar(current, ret);
901 static ssize_t compat_do_readv_writev(int type, struct file *file,
902 const struct compat_iovec __user *uvector,
903 unsigned long nr_segs, loff_t *pos)
905 compat_ssize_t tot_len;
906 struct iovec iovstack[UIO_FASTIOV];
907 struct iovec *iov = iovstack;
917 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
920 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
921 UIO_FASTIOV, iovstack, &iov);
926 ret = rw_verify_area(type, file, pos, tot_len);
932 fn = file->f_op->read;
933 fnv = file->f_op->aio_read;
935 fn = (io_fn_t)file->f_op->write;
936 fnv = file->f_op->aio_write;
937 file_start_write(file);
941 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
944 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
947 file_end_write(file);
952 if ((ret + (type == READ)) > 0) {
954 fsnotify_access(file);
956 fsnotify_modify(file);
961 static size_t compat_readv(struct file *file,
962 const struct compat_iovec __user *vec,
963 unsigned long vlen, loff_t *pos)
965 ssize_t ret = -EBADF;
967 if (!(file->f_mode & FMODE_READ))
971 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
974 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
978 add_rchar(current, ret);
983 COMPAT_SYSCALL_DEFINE3(readv, unsigned long, fd,
984 const struct compat_iovec __user *,vec,
987 struct fd f = fdget(fd);
994 ret = compat_readv(f.file, vec, vlen, &pos);
1001 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1002 const struct compat_iovec __user *,vec,
1003 unsigned long, vlen, loff_t, pos)
1014 if (f.file->f_mode & FMODE_PREAD)
1015 ret = compat_readv(f.file, vec, vlen, &pos);
1020 COMPAT_SYSCALL_DEFINE5(preadv, unsigned long, fd,
1021 const struct compat_iovec __user *,vec,
1022 unsigned long, vlen, u32, pos_low, u32, pos_high)
1024 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1025 return compat_sys_preadv64(fd, vec, vlen, pos);
1028 static size_t compat_writev(struct file *file,
1029 const struct compat_iovec __user *vec,
1030 unsigned long vlen, loff_t *pos)
1032 ssize_t ret = -EBADF;
1034 if (!(file->f_mode & FMODE_WRITE))
1038 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1041 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1045 add_wchar(current, ret);
1050 COMPAT_SYSCALL_DEFINE3(writev, unsigned long, fd,
1051 const struct compat_iovec __user *, vec,
1052 unsigned long, vlen)
1054 struct fd f = fdget(fd);
1060 pos = f.file->f_pos;
1061 ret = compat_writev(f.file, vec, vlen, &pos);
1063 f.file->f_pos = pos;
1068 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1069 const struct compat_iovec __user *,vec,
1070 unsigned long, vlen, loff_t, pos)
1081 if (f.file->f_mode & FMODE_PWRITE)
1082 ret = compat_writev(f.file, vec, vlen, &pos);
1087 COMPAT_SYSCALL_DEFINE5(pwritev, unsigned long, fd,
1088 const struct compat_iovec __user *,vec,
1089 unsigned long, vlen, u32, pos_low, u32, pos_high)
1091 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1092 return compat_sys_pwritev64(fd, vec, vlen, pos);
1096 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1097 size_t count, loff_t max)
1100 struct inode *in_inode, *out_inode;
1107 * Get input file, and verify that it is ok..
1113 if (!(in.file->f_mode & FMODE_READ))
1117 pos = in.file->f_pos;
1120 if (!(in.file->f_mode & FMODE_PREAD))
1123 retval = rw_verify_area(READ, in.file, &pos, count);
1129 * Get output file, and verify that it is ok..
1132 out = fdget(out_fd);
1135 if (!(out.file->f_mode & FMODE_WRITE))
1138 in_inode = file_inode(in.file);
1139 out_inode = file_inode(out.file);
1140 out_pos = out.file->f_pos;
1141 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1147 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1149 if (unlikely(pos + count > max)) {
1150 retval = -EOVERFLOW;
1159 * We need to debate whether we can enable this or not. The
1160 * man page documents EAGAIN return for the output at least,
1161 * and the application is arguably buggy if it doesn't expect
1162 * EAGAIN on a non-blocking file descriptor.
1164 if (in.file->f_flags & O_NONBLOCK)
1165 fl = SPLICE_F_NONBLOCK;
1167 file_start_write(out.file);
1168 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1169 file_end_write(out.file);
1172 add_rchar(current, retval);
1173 add_wchar(current, retval);
1174 fsnotify_access(in.file);
1175 fsnotify_modify(out.file);
1176 out.file->f_pos = out_pos;
1180 in.file->f_pos = pos;
1186 retval = -EOVERFLOW;
1196 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1203 if (unlikely(get_user(off, offset)))
1206 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1207 if (unlikely(put_user(pos, offset)))
1212 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1215 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1221 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1223 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1224 if (unlikely(put_user(pos, offset)))
1229 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1232 #ifdef CONFIG_COMPAT
1233 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1234 compat_off_t __user *, offset, compat_size_t, count)
1241 if (unlikely(get_user(off, offset)))
1244 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1245 if (unlikely(put_user(pos, offset)))
1250 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1253 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1254 compat_loff_t __user *, offset, compat_size_t, count)
1260 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1262 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1263 if (unlikely(put_user(pos, offset)))
1268 return do_sendfile(out_fd, in_fd, NULL, count, 0);