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 (*iov_fn_t)(struct kiocb *, const struct iovec *,
26 unsigned long, loff_t);
27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
31 .read_iter = generic_file_read_iter,
32 .mmap = generic_file_readonly_mmap,
33 .splice_read = generic_file_splice_read,
36 EXPORT_SYMBOL(generic_ro_fops);
38 static inline int unsigned_offsets(struct file *file)
40 return file->f_mode & FMODE_UNSIGNED_OFFSET;
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
53 * Return the specified offset on success and -EINVAL on invalid offset.
55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
57 if (offset < 0 && !unsigned_offsets(file))
62 if (offset != file->f_pos) {
68 EXPORT_SYMBOL(vfs_setpos);
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
87 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
88 loff_t maxsize, loff_t eof)
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
108 spin_lock(&file->f_lock);
109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
110 spin_unlock(&file->f_lock);
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
131 return vfs_setpos(file, offset, maxsize);
133 EXPORT_SYMBOL(generic_file_llseek_size);
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
147 struct inode *inode = file->f_mapping->host;
149 return generic_file_llseek_size(file, offset, whence,
150 inode->i_sb->s_maxbytes,
153 EXPORT_SYMBOL(generic_file_llseek);
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
173 EXPORT_SYMBOL(fixed_size_llseek);
176 * noop_llseek - No Operation Performed llseek implementation
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
181 * This is an implementation of ->llseek useable for the rare special case when
182 * userspace expects the seek to succeed but the (device) file is actually not
183 * able to perform the seek. In this case you use noop_llseek() instead of
184 * falling back to the default implementation of ->llseek.
186 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
190 EXPORT_SYMBOL(noop_llseek);
192 loff_t no_llseek(struct file *file, loff_t offset, int whence)
196 EXPORT_SYMBOL(no_llseek);
198 loff_t default_llseek(struct file *file, loff_t offset, int whence)
200 struct inode *inode = file_inode(file);
203 mutex_lock(&inode->i_mutex);
206 offset += i_size_read(inode);
210 retval = file->f_pos;
213 offset += file->f_pos;
217 * In the generic case the entire file is data, so as
218 * long as offset isn't at the end of the file then the
221 if (offset >= inode->i_size) {
228 * There is a virtual hole at the end of the file, so
229 * as long as offset isn't i_size or larger, return
232 if (offset >= inode->i_size) {
236 offset = inode->i_size;
240 if (offset >= 0 || unsigned_offsets(file)) {
241 if (offset != file->f_pos) {
242 file->f_pos = offset;
248 mutex_unlock(&inode->i_mutex);
251 EXPORT_SYMBOL(default_llseek);
253 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
255 loff_t (*fn)(struct file *, loff_t, int);
258 if (file->f_mode & FMODE_LSEEK) {
259 if (file->f_op->llseek)
260 fn = file->f_op->llseek;
262 return fn(file, offset, whence);
264 EXPORT_SYMBOL(vfs_llseek);
266 static inline struct fd fdget_pos(int fd)
268 return __to_fd(__fdget_pos(fd));
271 static inline void fdput_pos(struct fd f)
273 if (f.flags & FDPUT_POS_UNLOCK)
274 mutex_unlock(&f.file->f_pos_lock);
278 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
281 struct fd f = fdget_pos(fd);
286 if (whence <= SEEK_MAX) {
287 loff_t res = vfs_llseek(f.file, offset, whence);
289 if (res != (loff_t)retval)
290 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
297 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
299 return sys_lseek(fd, offset, whence);
303 #ifdef __ARCH_WANT_SYS_LLSEEK
304 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
305 unsigned long, offset_low, loff_t __user *, result,
306 unsigned int, whence)
309 struct fd f = fdget_pos(fd);
316 if (whence > SEEK_MAX)
319 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
322 retval = (int)offset;
325 if (!copy_to_user(result, &offset, sizeof(offset)))
334 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
339 if (!file->f_op->read_iter)
342 init_sync_kiocb(&kiocb, file);
343 kiocb.ki_pos = *ppos;
346 ret = file->f_op->read_iter(&kiocb, iter);
347 BUG_ON(ret == -EIOCBQUEUED);
349 *ppos = kiocb.ki_pos;
352 EXPORT_SYMBOL(vfs_iter_read);
354 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
359 if (!file->f_op->write_iter)
362 init_sync_kiocb(&kiocb, file);
363 kiocb.ki_pos = *ppos;
366 ret = file->f_op->write_iter(&kiocb, iter);
367 BUG_ON(ret == -EIOCBQUEUED);
369 *ppos = kiocb.ki_pos;
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);
482 *ppos = kiocb.ki_pos;
486 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
489 if (file->f_op->write)
490 return file->f_op->write(file, p, count, pos);
491 else if (file->f_op->write_iter)
492 return new_sync_write(file, p, count, pos);
496 EXPORT_SYMBOL(__vfs_write);
498 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
501 const char __user *p;
504 if (!(file->f_mode & FMODE_CAN_WRITE))
509 p = (__force const char __user *)buf;
510 if (count > MAX_RW_COUNT)
511 count = MAX_RW_COUNT;
512 ret = __vfs_write(file, p, count, pos);
515 fsnotify_modify(file);
516 add_wchar(current, ret);
522 EXPORT_SYMBOL(__kernel_write);
524 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
528 if (!(file->f_mode & FMODE_WRITE))
530 if (!(file->f_mode & FMODE_CAN_WRITE))
532 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
535 ret = rw_verify_area(WRITE, file, pos, count);
538 file_start_write(file);
539 ret = __vfs_write(file, buf, count, pos);
541 fsnotify_modify(file);
542 add_wchar(current, ret);
545 file_end_write(file);
551 EXPORT_SYMBOL(vfs_write);
553 static inline loff_t file_pos_read(struct file *file)
558 static inline void file_pos_write(struct file *file, loff_t pos)
563 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
565 struct fd f = fdget_pos(fd);
566 ssize_t ret = -EBADF;
569 loff_t pos = file_pos_read(f.file);
570 ret = vfs_read(f.file, buf, count, &pos);
572 file_pos_write(f.file, pos);
578 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
581 struct fd f = fdget_pos(fd);
582 ssize_t ret = -EBADF;
585 loff_t pos = file_pos_read(f.file);
586 ret = vfs_write(f.file, buf, count, &pos);
588 file_pos_write(f.file, pos);
595 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
596 size_t, count, loff_t, pos)
599 ssize_t ret = -EBADF;
607 if (f.file->f_mode & FMODE_PREAD)
608 ret = vfs_read(f.file, buf, count, &pos);
615 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
616 size_t, count, loff_t, pos)
619 ssize_t ret = -EBADF;
627 if (f.file->f_mode & FMODE_PWRITE)
628 ret = vfs_write(f.file, buf, count, &pos);
636 * Reduce an iovec's length in-place. Return the resulting number of segments
638 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
640 unsigned long seg = 0;
643 while (seg < nr_segs) {
645 if (len + iov->iov_len >= to) {
646 iov->iov_len = to - len;
654 EXPORT_SYMBOL(iov_shorten);
656 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
657 loff_t *ppos, iter_fn_t fn)
662 init_sync_kiocb(&kiocb, filp);
663 kiocb.ki_pos = *ppos;
665 ret = fn(&kiocb, iter);
666 BUG_ON(ret == -EIOCBQUEUED);
667 *ppos = kiocb.ki_pos;
671 static ssize_t do_sync_readv_writev(struct file *filp, struct iov_iter *iter,
672 loff_t *ppos, iov_fn_t fn)
677 init_sync_kiocb(&kiocb, filp);
678 kiocb.ki_pos = *ppos;
680 ret = fn(&kiocb, iter->iov, iter->nr_segs, kiocb.ki_pos);
681 BUG_ON(ret == -EIOCBQUEUED);
682 *ppos = kiocb.ki_pos;
686 /* Do it by hand, with file-ops */
687 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
688 loff_t *ppos, io_fn_t fn)
692 while (iov_iter_count(iter)) {
693 struct iovec iovec = iov_iter_iovec(iter);
696 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
704 if (nr != iovec.iov_len)
706 iov_iter_advance(iter, nr);
712 /* A write operation does a read from user space and vice versa */
713 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
715 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
716 unsigned long nr_segs, unsigned long fast_segs,
717 struct iovec *fast_pointer,
718 struct iovec **ret_pointer)
722 struct iovec *iov = fast_pointer;
725 * SuS says "The readv() function *may* fail if the iovcnt argument
726 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
727 * traditionally returned zero for zero segments, so...
735 * First get the "struct iovec" from user memory and
736 * verify all the pointers
738 if (nr_segs > UIO_MAXIOV) {
742 if (nr_segs > fast_segs) {
743 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
749 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
755 * According to the Single Unix Specification we should return EINVAL
756 * if an element length is < 0 when cast to ssize_t or if the
757 * total length would overflow the ssize_t return value of the
760 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
764 for (seg = 0; seg < nr_segs; seg++) {
765 void __user *buf = iov[seg].iov_base;
766 ssize_t len = (ssize_t)iov[seg].iov_len;
768 /* see if we we're about to use an invalid len or if
769 * it's about to overflow ssize_t */
775 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
779 if (len > MAX_RW_COUNT - ret) {
780 len = MAX_RW_COUNT - ret;
781 iov[seg].iov_len = len;
790 static ssize_t do_readv_writev(int type, struct file *file,
791 const struct iovec __user * uvector,
792 unsigned long nr_segs, loff_t *pos)
795 struct iovec iovstack[UIO_FASTIOV];
796 struct iovec *iov = iovstack;
797 struct iov_iter iter;
803 ret = import_iovec(type, uvector, nr_segs,
804 ARRAY_SIZE(iovstack), &iov, &iter);
808 tot_len = iov_iter_count(&iter);
811 ret = rw_verify_area(type, file, pos, tot_len);
817 fn = file->f_op->read;
818 fnv = file->f_op->aio_read;
819 iter_fn = file->f_op->read_iter;
821 fn = (io_fn_t)file->f_op->write;
822 fnv = file->f_op->aio_write;
823 iter_fn = file->f_op->write_iter;
824 file_start_write(file);
828 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
830 ret = do_sync_readv_writev(file, &iter, pos, fnv);
832 ret = do_loop_readv_writev(file, &iter, pos, fn);
835 file_end_write(file);
839 if ((ret + (type == READ)) > 0) {
841 fsnotify_access(file);
843 fsnotify_modify(file);
848 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
849 unsigned long vlen, loff_t *pos)
851 if (!(file->f_mode & FMODE_READ))
853 if (!(file->f_mode & FMODE_CAN_READ))
856 return do_readv_writev(READ, file, vec, vlen, pos);
859 EXPORT_SYMBOL(vfs_readv);
861 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
862 unsigned long vlen, loff_t *pos)
864 if (!(file->f_mode & FMODE_WRITE))
866 if (!(file->f_mode & FMODE_CAN_WRITE))
869 return do_readv_writev(WRITE, file, vec, vlen, pos);
872 EXPORT_SYMBOL(vfs_writev);
874 SYSCALL_DEFINE3(readv, 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_readv(f.file, vec, vlen, &pos);
884 file_pos_write(f.file, pos);
889 add_rchar(current, ret);
894 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
897 struct fd f = fdget_pos(fd);
898 ssize_t ret = -EBADF;
901 loff_t pos = file_pos_read(f.file);
902 ret = vfs_writev(f.file, vec, vlen, &pos);
904 file_pos_write(f.file, pos);
909 add_wchar(current, ret);
914 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
916 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
917 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
920 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
921 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
923 loff_t pos = pos_from_hilo(pos_h, pos_l);
925 ssize_t ret = -EBADF;
933 if (f.file->f_mode & FMODE_PREAD)
934 ret = vfs_readv(f.file, vec, vlen, &pos);
939 add_rchar(current, ret);
944 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
945 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
947 loff_t pos = pos_from_hilo(pos_h, pos_l);
949 ssize_t ret = -EBADF;
957 if (f.file->f_mode & FMODE_PWRITE)
958 ret = vfs_writev(f.file, vec, vlen, &pos);
963 add_wchar(current, ret);
970 static ssize_t compat_do_readv_writev(int type, struct file *file,
971 const struct compat_iovec __user *uvector,
972 unsigned long nr_segs, loff_t *pos)
974 compat_ssize_t tot_len;
975 struct iovec iovstack[UIO_FASTIOV];
976 struct iovec *iov = iovstack;
977 struct iov_iter iter;
983 ret = compat_import_iovec(type, uvector, nr_segs,
984 UIO_FASTIOV, &iov, &iter);
988 tot_len = iov_iter_count(&iter);
991 ret = rw_verify_area(type, file, pos, tot_len);
997 fn = file->f_op->read;
998 fnv = file->f_op->aio_read;
999 iter_fn = file->f_op->read_iter;
1001 fn = (io_fn_t)file->f_op->write;
1002 fnv = file->f_op->aio_write;
1003 iter_fn = file->f_op->write_iter;
1004 file_start_write(file);
1008 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
1010 ret = do_sync_readv_writev(file, &iter, pos, fnv);
1012 ret = do_loop_readv_writev(file, &iter, pos, fn);
1015 file_end_write(file);
1019 if ((ret + (type == READ)) > 0) {
1021 fsnotify_access(file);
1023 fsnotify_modify(file);
1028 static size_t compat_readv(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_READ))
1038 if (!(file->f_mode & FMODE_CAN_READ))
1041 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1045 add_rchar(current, ret);
1050 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1051 const struct compat_iovec __user *,vec,
1052 compat_ulong_t, vlen)
1054 struct fd f = fdget_pos(fd);
1060 pos = f.file->f_pos;
1061 ret = compat_readv(f.file, vec, vlen, &pos);
1063 f.file->f_pos = pos;
1068 static long __compat_sys_preadv64(unsigned long fd,
1069 const struct compat_iovec __user *vec,
1070 unsigned long vlen, loff_t pos)
1081 if (f.file->f_mode & FMODE_PREAD)
1082 ret = compat_readv(f.file, vec, vlen, &pos);
1087 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1088 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1089 const struct compat_iovec __user *,vec,
1090 unsigned long, vlen, loff_t, pos)
1092 return __compat_sys_preadv64(fd, vec, vlen, pos);
1096 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1097 const struct compat_iovec __user *,vec,
1098 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1100 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1102 return __compat_sys_preadv64(fd, vec, vlen, pos);
1105 static size_t compat_writev(struct file *file,
1106 const struct compat_iovec __user *vec,
1107 unsigned long vlen, loff_t *pos)
1109 ssize_t ret = -EBADF;
1111 if (!(file->f_mode & FMODE_WRITE))
1115 if (!(file->f_mode & FMODE_CAN_WRITE))
1118 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1122 add_wchar(current, ret);
1127 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1128 const struct compat_iovec __user *, vec,
1129 compat_ulong_t, vlen)
1131 struct fd f = fdget_pos(fd);
1137 pos = f.file->f_pos;
1138 ret = compat_writev(f.file, vec, vlen, &pos);
1140 f.file->f_pos = pos;
1145 static long __compat_sys_pwritev64(unsigned long fd,
1146 const struct compat_iovec __user *vec,
1147 unsigned long vlen, loff_t pos)
1158 if (f.file->f_mode & FMODE_PWRITE)
1159 ret = compat_writev(f.file, vec, vlen, &pos);
1164 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1165 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1166 const struct compat_iovec __user *,vec,
1167 unsigned long, vlen, loff_t, pos)
1169 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1173 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1174 const struct compat_iovec __user *,vec,
1175 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1177 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1179 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1183 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1184 size_t count, loff_t max)
1187 struct inode *in_inode, *out_inode;
1194 * Get input file, and verify that it is ok..
1200 if (!(in.file->f_mode & FMODE_READ))
1204 pos = in.file->f_pos;
1207 if (!(in.file->f_mode & FMODE_PREAD))
1210 retval = rw_verify_area(READ, in.file, &pos, count);
1216 * Get output file, and verify that it is ok..
1219 out = fdget(out_fd);
1222 if (!(out.file->f_mode & FMODE_WRITE))
1225 in_inode = file_inode(in.file);
1226 out_inode = file_inode(out.file);
1227 out_pos = out.file->f_pos;
1228 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1234 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1236 if (unlikely(pos + count > max)) {
1237 retval = -EOVERFLOW;
1246 * We need to debate whether we can enable this or not. The
1247 * man page documents EAGAIN return for the output at least,
1248 * and the application is arguably buggy if it doesn't expect
1249 * EAGAIN on a non-blocking file descriptor.
1251 if (in.file->f_flags & O_NONBLOCK)
1252 fl = SPLICE_F_NONBLOCK;
1254 file_start_write(out.file);
1255 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1256 file_end_write(out.file);
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);