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);
28 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
30 const struct file_operations generic_ro_fops = {
31 .llseek = generic_file_llseek,
32 .read = new_sync_read,
33 .read_iter = generic_file_read_iter,
34 .mmap = generic_file_readonly_mmap,
35 .splice_read = generic_file_splice_read,
38 EXPORT_SYMBOL(generic_ro_fops);
40 static inline int unsigned_offsets(struct file *file)
42 return file->f_mode & FMODE_UNSIGNED_OFFSET;
46 * vfs_setpos - update the file offset for lseek
47 * @file: file structure in question
48 * @offset: file offset to seek to
49 * @maxsize: maximum file size
51 * This is a low-level filesystem helper for updating the file offset to
52 * the value specified by @offset if the given offset is valid and it is
53 * not equal to the current file offset.
55 * Return the specified offset on success and -EINVAL on invalid offset.
57 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
59 if (offset < 0 && !unsigned_offsets(file))
64 if (offset != file->f_pos) {
70 EXPORT_SYMBOL(vfs_setpos);
73 * generic_file_llseek_size - generic llseek implementation for regular files
74 * @file: file structure to seek on
75 * @offset: file offset to seek to
76 * @whence: type of seek
77 * @size: max size of this file in file system
78 * @eof: offset used for SEEK_END position
80 * This is a variant of generic_file_llseek that allows passing in a custom
81 * maximum file size and a custom EOF position, for e.g. hashed directories
84 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
85 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
86 * read/writes behave like SEEK_SET against seeks.
89 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
90 loff_t maxsize, loff_t eof)
98 * Here we special-case the lseek(fd, 0, SEEK_CUR)
99 * position-querying operation. Avoid rewriting the "same"
100 * f_pos value back to the file because a concurrent read(),
101 * write() or lseek() might have altered it
106 * f_lock protects against read/modify/write race with other
107 * SEEK_CURs. Note that parallel writes and reads behave
110 spin_lock(&file->f_lock);
111 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
112 spin_unlock(&file->f_lock);
116 * In the generic case the entire file is data, so as long as
117 * offset isn't at the end of the file then the offset is data.
124 * There is a virtual hole at the end of the file, so as long as
125 * offset isn't i_size or larger, return i_size.
133 return vfs_setpos(file, offset, maxsize);
135 EXPORT_SYMBOL(generic_file_llseek_size);
138 * generic_file_llseek - generic llseek implementation for regular files
139 * @file: file structure to seek on
140 * @offset: file offset to seek to
141 * @whence: type of seek
143 * This is a generic implemenation of ->llseek useable for all normal local
144 * filesystems. It just updates the file offset to the value specified by
145 * @offset and @whence.
147 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
149 struct inode *inode = file->f_mapping->host;
151 return generic_file_llseek_size(file, offset, whence,
152 inode->i_sb->s_maxbytes,
155 EXPORT_SYMBOL(generic_file_llseek);
158 * fixed_size_llseek - llseek implementation for fixed-sized devices
159 * @file: file structure to seek on
160 * @offset: file offset to seek to
161 * @whence: type of seek
162 * @size: size of the file
165 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
168 case SEEK_SET: case SEEK_CUR: case SEEK_END:
169 return generic_file_llseek_size(file, offset, whence,
175 EXPORT_SYMBOL(fixed_size_llseek);
178 * noop_llseek - No Operation Performed llseek implementation
179 * @file: file structure to seek on
180 * @offset: file offset to seek to
181 * @whence: type of seek
183 * This is an implementation of ->llseek useable for the rare special case when
184 * userspace expects the seek to succeed but the (device) file is actually not
185 * able to perform the seek. In this case you use noop_llseek() instead of
186 * falling back to the default implementation of ->llseek.
188 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
192 EXPORT_SYMBOL(noop_llseek);
194 loff_t no_llseek(struct file *file, loff_t offset, int whence)
198 EXPORT_SYMBOL(no_llseek);
200 loff_t default_llseek(struct file *file, loff_t offset, int whence)
202 struct inode *inode = file_inode(file);
205 mutex_lock(&inode->i_mutex);
208 offset += i_size_read(inode);
212 retval = file->f_pos;
215 offset += file->f_pos;
219 * In the generic case the entire file is data, so as
220 * long as offset isn't at the end of the file then the
223 if (offset >= inode->i_size) {
230 * There is a virtual hole at the end of the file, so
231 * as long as offset isn't i_size or larger, return
234 if (offset >= inode->i_size) {
238 offset = inode->i_size;
242 if (offset >= 0 || unsigned_offsets(file)) {
243 if (offset != file->f_pos) {
244 file->f_pos = offset;
250 mutex_unlock(&inode->i_mutex);
253 EXPORT_SYMBOL(default_llseek);
255 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
257 loff_t (*fn)(struct file *, loff_t, int);
260 if (file->f_mode & FMODE_LSEEK) {
261 if (file->f_op->llseek)
262 fn = file->f_op->llseek;
264 return fn(file, offset, whence);
266 EXPORT_SYMBOL(vfs_llseek);
268 static inline struct fd fdget_pos(int fd)
270 return __to_fd(__fdget_pos(fd));
273 static inline void fdput_pos(struct fd f)
275 if (f.flags & FDPUT_POS_UNLOCK)
276 mutex_unlock(&f.file->f_pos_lock);
280 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
283 struct fd f = fdget_pos(fd);
288 if (whence <= SEEK_MAX) {
289 loff_t res = vfs_llseek(f.file, offset, whence);
291 if (res != (loff_t)retval)
292 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
299 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
301 return sys_lseek(fd, offset, whence);
305 #ifdef __ARCH_WANT_SYS_LLSEEK
306 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
307 unsigned long, offset_low, loff_t __user *, result,
308 unsigned int, whence)
311 struct fd f = fdget_pos(fd);
318 if (whence > SEEK_MAX)
321 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
324 retval = (int)offset;
327 if (!copy_to_user(result, &offset, sizeof(offset)))
337 * rw_verify_area doesn't like huge counts. We limit
338 * them to something that fits in "int" so that others
339 * won't have to do range checks all the time.
341 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
345 int retval = -EINVAL;
347 inode = file_inode(file);
348 if (unlikely((ssize_t) count < 0))
351 if (unlikely(pos < 0)) {
352 if (!unsigned_offsets(file))
354 if (count >= -pos) /* both values are in 0..LLONG_MAX */
356 } else if (unlikely((loff_t) (pos + count) < 0)) {
357 if (!unsigned_offsets(file))
361 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
362 retval = locks_mandatory_area(
363 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
364 inode, file, pos, count);
368 retval = security_file_permission(file,
369 read_write == READ ? MAY_READ : MAY_WRITE);
372 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
375 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
377 struct iovec iov = { .iov_base = buf, .iov_len = len };
381 init_sync_kiocb(&kiocb, filp);
382 kiocb.ki_pos = *ppos;
383 kiocb.ki_nbytes = len;
385 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
386 if (-EIOCBQUEUED == ret)
387 ret = wait_on_sync_kiocb(&kiocb);
388 *ppos = kiocb.ki_pos;
392 EXPORT_SYMBOL(do_sync_read);
394 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
396 struct iovec iov = { .iov_base = buf, .iov_len = len };
398 struct iov_iter iter;
401 init_sync_kiocb(&kiocb, filp);
402 kiocb.ki_pos = *ppos;
403 kiocb.ki_nbytes = len;
404 iov_iter_init(&iter, READ, &iov, 1, len);
406 ret = filp->f_op->read_iter(&kiocb, &iter);
407 if (-EIOCBQUEUED == ret)
408 ret = wait_on_sync_kiocb(&kiocb);
409 *ppos = kiocb.ki_pos;
413 EXPORT_SYMBOL(new_sync_read);
415 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
419 if (!(file->f_mode & FMODE_READ))
421 if (!(file->f_mode & FMODE_CAN_READ))
423 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
426 ret = rw_verify_area(READ, file, pos, count);
429 if (file->f_op->read)
430 ret = file->f_op->read(file, buf, count, pos);
431 else if (file->f_op->aio_read)
432 ret = do_sync_read(file, buf, count, pos);
434 ret = new_sync_read(file, buf, count, pos);
436 fsnotify_access(file);
437 add_rchar(current, ret);
445 EXPORT_SYMBOL(vfs_read);
447 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
449 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
453 init_sync_kiocb(&kiocb, filp);
454 kiocb.ki_pos = *ppos;
455 kiocb.ki_nbytes = len;
457 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
458 if (-EIOCBQUEUED == ret)
459 ret = wait_on_sync_kiocb(&kiocb);
460 *ppos = kiocb.ki_pos;
464 EXPORT_SYMBOL(do_sync_write);
466 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
468 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
470 struct iov_iter iter;
473 init_sync_kiocb(&kiocb, filp);
474 kiocb.ki_pos = *ppos;
475 kiocb.ki_nbytes = len;
476 iov_iter_init(&iter, WRITE, &iov, 1, len);
478 ret = filp->f_op->write_iter(&kiocb, &iter);
479 if (-EIOCBQUEUED == ret)
480 ret = wait_on_sync_kiocb(&kiocb);
481 *ppos = kiocb.ki_pos;
485 EXPORT_SYMBOL(new_sync_write);
487 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
490 const char __user *p;
493 if (!(file->f_mode & FMODE_CAN_WRITE))
498 p = (__force const char __user *)buf;
499 if (count > MAX_RW_COUNT)
500 count = MAX_RW_COUNT;
501 if (file->f_op->write)
502 ret = file->f_op->write(file, p, count, pos);
503 else if (file->f_op->aio_write)
504 ret = do_sync_write(file, p, count, pos);
506 ret = new_sync_write(file, p, count, pos);
509 fsnotify_modify(file);
510 add_wchar(current, ret);
516 EXPORT_SYMBOL(__kernel_write);
518 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
522 if (!(file->f_mode & FMODE_WRITE))
524 if (!(file->f_mode & FMODE_CAN_WRITE))
526 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
529 ret = rw_verify_area(WRITE, file, pos, count);
532 file_start_write(file);
533 if (file->f_op->write)
534 ret = file->f_op->write(file, buf, count, pos);
535 else if (file->f_op->aio_write)
536 ret = do_sync_write(file, buf, count, pos);
538 ret = new_sync_write(file, buf, count, pos);
540 fsnotify_modify(file);
541 add_wchar(current, ret);
544 file_end_write(file);
550 EXPORT_SYMBOL(vfs_write);
552 static inline loff_t file_pos_read(struct file *file)
557 static inline void file_pos_write(struct file *file, loff_t pos)
562 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
564 struct fd f = fdget_pos(fd);
565 ssize_t ret = -EBADF;
568 loff_t pos = file_pos_read(f.file);
569 ret = vfs_read(f.file, buf, count, &pos);
571 file_pos_write(f.file, pos);
577 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
580 struct fd f = fdget_pos(fd);
581 ssize_t ret = -EBADF;
584 loff_t pos = file_pos_read(f.file);
585 ret = vfs_write(f.file, buf, count, &pos);
587 file_pos_write(f.file, pos);
594 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
595 size_t, count, loff_t, pos)
598 ssize_t ret = -EBADF;
606 if (f.file->f_mode & FMODE_PREAD)
607 ret = vfs_read(f.file, buf, count, &pos);
614 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
615 size_t, count, loff_t, pos)
618 ssize_t ret = -EBADF;
626 if (f.file->f_mode & FMODE_PWRITE)
627 ret = vfs_write(f.file, buf, count, &pos);
635 * Reduce an iovec's length in-place. Return the resulting number of segments
637 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
639 unsigned long seg = 0;
642 while (seg < nr_segs) {
644 if (len + iov->iov_len >= to) {
645 iov->iov_len = to - len;
653 EXPORT_SYMBOL(iov_shorten);
655 static ssize_t do_iter_readv_writev(struct file *filp, int rw, const struct iovec *iov,
656 unsigned long nr_segs, size_t len, loff_t *ppos, iter_fn_t fn)
659 struct iov_iter iter;
662 init_sync_kiocb(&kiocb, filp);
663 kiocb.ki_pos = *ppos;
664 kiocb.ki_nbytes = len;
666 iov_iter_init(&iter, rw, iov, nr_segs, len);
667 ret = fn(&kiocb, &iter);
668 if (ret == -EIOCBQUEUED)
669 ret = wait_on_sync_kiocb(&kiocb);
670 *ppos = kiocb.ki_pos;
674 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
675 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
680 init_sync_kiocb(&kiocb, filp);
681 kiocb.ki_pos = *ppos;
682 kiocb.ki_nbytes = len;
684 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
685 if (ret == -EIOCBQUEUED)
686 ret = wait_on_sync_kiocb(&kiocb);
687 *ppos = kiocb.ki_pos;
691 /* Do it by hand, with file-ops */
692 static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
693 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
695 struct iovec *vector = iov;
698 while (nr_segs > 0) {
703 base = vector->iov_base;
704 len = vector->iov_len;
708 nr = fn(filp, base, len, ppos);
723 /* A write operation does a read from user space and vice versa */
724 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
726 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
727 unsigned long nr_segs, unsigned long fast_segs,
728 struct iovec *fast_pointer,
729 struct iovec **ret_pointer)
733 struct iovec *iov = fast_pointer;
736 * SuS says "The readv() function *may* fail if the iovcnt argument
737 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
738 * traditionally returned zero for zero segments, so...
746 * First get the "struct iovec" from user memory and
747 * verify all the pointers
749 if (nr_segs > UIO_MAXIOV) {
753 if (nr_segs > fast_segs) {
754 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
760 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
766 * According to the Single Unix Specification we should return EINVAL
767 * if an element length is < 0 when cast to ssize_t or if the
768 * total length would overflow the ssize_t return value of the
771 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
775 for (seg = 0; seg < nr_segs; seg++) {
776 void __user *buf = iov[seg].iov_base;
777 ssize_t len = (ssize_t)iov[seg].iov_len;
779 /* see if we we're about to use an invalid len or if
780 * it's about to overflow ssize_t */
786 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
790 if (len > MAX_RW_COUNT - ret) {
791 len = MAX_RW_COUNT - ret;
792 iov[seg].iov_len = len;
801 static ssize_t do_readv_writev(int type, struct file *file,
802 const struct iovec __user * uvector,
803 unsigned long nr_segs, loff_t *pos)
806 struct iovec iovstack[UIO_FASTIOV];
807 struct iovec *iov = iovstack;
813 ret = rw_copy_check_uvector(type, uvector, nr_segs,
814 ARRAY_SIZE(iovstack), iovstack, &iov);
819 ret = rw_verify_area(type, file, pos, tot_len);
825 fn = file->f_op->read;
826 fnv = file->f_op->aio_read;
827 iter_fn = file->f_op->read_iter;
829 fn = (io_fn_t)file->f_op->write;
830 fnv = file->f_op->aio_write;
831 iter_fn = file->f_op->write_iter;
832 file_start_write(file);
836 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
839 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
842 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
845 file_end_write(file);
850 if ((ret + (type == READ)) > 0) {
852 fsnotify_access(file);
854 fsnotify_modify(file);
859 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
860 unsigned long vlen, loff_t *pos)
862 if (!(file->f_mode & FMODE_READ))
864 if (!(file->f_mode & FMODE_CAN_READ))
867 return do_readv_writev(READ, file, vec, vlen, pos);
870 EXPORT_SYMBOL(vfs_readv);
872 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
873 unsigned long vlen, loff_t *pos)
875 if (!(file->f_mode & FMODE_WRITE))
877 if (!(file->f_mode & FMODE_CAN_WRITE))
880 return do_readv_writev(WRITE, file, vec, vlen, pos);
883 EXPORT_SYMBOL(vfs_writev);
885 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
888 struct fd f = fdget_pos(fd);
889 ssize_t ret = -EBADF;
892 loff_t pos = file_pos_read(f.file);
893 ret = vfs_readv(f.file, vec, vlen, &pos);
895 file_pos_write(f.file, pos);
900 add_rchar(current, ret);
905 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
908 struct fd f = fdget_pos(fd);
909 ssize_t ret = -EBADF;
912 loff_t pos = file_pos_read(f.file);
913 ret = vfs_writev(f.file, vec, vlen, &pos);
915 file_pos_write(f.file, pos);
920 add_wchar(current, ret);
925 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
927 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
928 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
931 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
932 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
934 loff_t pos = pos_from_hilo(pos_h, pos_l);
936 ssize_t ret = -EBADF;
944 if (f.file->f_mode & FMODE_PREAD)
945 ret = vfs_readv(f.file, vec, vlen, &pos);
950 add_rchar(current, ret);
955 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
956 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
958 loff_t pos = pos_from_hilo(pos_h, pos_l);
960 ssize_t ret = -EBADF;
968 if (f.file->f_mode & FMODE_PWRITE)
969 ret = vfs_writev(f.file, vec, vlen, &pos);
974 add_wchar(current, ret);
981 static ssize_t compat_do_readv_writev(int type, struct file *file,
982 const struct compat_iovec __user *uvector,
983 unsigned long nr_segs, loff_t *pos)
985 compat_ssize_t tot_len;
986 struct iovec iovstack[UIO_FASTIOV];
987 struct iovec *iov = iovstack;
993 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
994 UIO_FASTIOV, iovstack, &iov);
999 ret = rw_verify_area(type, file, pos, tot_len);
1005 fn = file->f_op->read;
1006 fnv = file->f_op->aio_read;
1007 iter_fn = file->f_op->read_iter;
1009 fn = (io_fn_t)file->f_op->write;
1010 fnv = file->f_op->aio_write;
1011 iter_fn = file->f_op->write_iter;
1012 file_start_write(file);
1016 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
1019 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1022 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1025 file_end_write(file);
1028 if (iov != iovstack)
1030 if ((ret + (type == READ)) > 0) {
1032 fsnotify_access(file);
1034 fsnotify_modify(file);
1039 static size_t compat_readv(struct file *file,
1040 const struct compat_iovec __user *vec,
1041 unsigned long vlen, loff_t *pos)
1043 ssize_t ret = -EBADF;
1045 if (!(file->f_mode & FMODE_READ))
1049 if (!(file->f_mode & FMODE_CAN_READ))
1052 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1056 add_rchar(current, ret);
1061 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1062 const struct compat_iovec __user *,vec,
1063 compat_ulong_t, vlen)
1065 struct fd f = fdget_pos(fd);
1071 pos = f.file->f_pos;
1072 ret = compat_readv(f.file, vec, vlen, &pos);
1074 f.file->f_pos = pos;
1079 static long __compat_sys_preadv64(unsigned long fd,
1080 const struct compat_iovec __user *vec,
1081 unsigned long vlen, loff_t pos)
1092 if (f.file->f_mode & FMODE_PREAD)
1093 ret = compat_readv(f.file, vec, vlen, &pos);
1098 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1099 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1100 const struct compat_iovec __user *,vec,
1101 unsigned long, vlen, loff_t, pos)
1103 return __compat_sys_preadv64(fd, vec, vlen, pos);
1107 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1108 const struct compat_iovec __user *,vec,
1109 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1111 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1113 return __compat_sys_preadv64(fd, vec, vlen, pos);
1116 static size_t compat_writev(struct file *file,
1117 const struct compat_iovec __user *vec,
1118 unsigned long vlen, loff_t *pos)
1120 ssize_t ret = -EBADF;
1122 if (!(file->f_mode & FMODE_WRITE))
1126 if (!(file->f_mode & FMODE_CAN_WRITE))
1129 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1133 add_wchar(current, ret);
1138 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1139 const struct compat_iovec __user *, vec,
1140 compat_ulong_t, vlen)
1142 struct fd f = fdget_pos(fd);
1148 pos = f.file->f_pos;
1149 ret = compat_writev(f.file, vec, vlen, &pos);
1151 f.file->f_pos = pos;
1156 static long __compat_sys_pwritev64(unsigned long fd,
1157 const struct compat_iovec __user *vec,
1158 unsigned long vlen, loff_t pos)
1169 if (f.file->f_mode & FMODE_PWRITE)
1170 ret = compat_writev(f.file, vec, vlen, &pos);
1175 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1176 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1177 const struct compat_iovec __user *,vec,
1178 unsigned long, vlen, loff_t, pos)
1180 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1184 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1185 const struct compat_iovec __user *,vec,
1186 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1188 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1190 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1194 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1195 size_t count, loff_t max)
1198 struct inode *in_inode, *out_inode;
1205 * Get input file, and verify that it is ok..
1211 if (!(in.file->f_mode & FMODE_READ))
1215 pos = in.file->f_pos;
1218 if (!(in.file->f_mode & FMODE_PREAD))
1221 retval = rw_verify_area(READ, in.file, &pos, count);
1227 * Get output file, and verify that it is ok..
1230 out = fdget(out_fd);
1233 if (!(out.file->f_mode & FMODE_WRITE))
1236 in_inode = file_inode(in.file);
1237 out_inode = file_inode(out.file);
1238 out_pos = out.file->f_pos;
1239 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1245 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1247 if (unlikely(pos + count > max)) {
1248 retval = -EOVERFLOW;
1257 * We need to debate whether we can enable this or not. The
1258 * man page documents EAGAIN return for the output at least,
1259 * and the application is arguably buggy if it doesn't expect
1260 * EAGAIN on a non-blocking file descriptor.
1262 if (in.file->f_flags & O_NONBLOCK)
1263 fl = SPLICE_F_NONBLOCK;
1265 file_start_write(out.file);
1266 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1267 file_end_write(out.file);
1270 add_rchar(current, retval);
1271 add_wchar(current, retval);
1272 fsnotify_access(in.file);
1273 fsnotify_modify(out.file);
1274 out.file->f_pos = out_pos;
1278 in.file->f_pos = pos;
1284 retval = -EOVERFLOW;
1294 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1301 if (unlikely(get_user(off, offset)))
1304 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1305 if (unlikely(put_user(pos, offset)))
1310 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1313 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1319 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1321 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1322 if (unlikely(put_user(pos, offset)))
1327 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1330 #ifdef CONFIG_COMPAT
1331 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1332 compat_off_t __user *, offset, compat_size_t, count)
1339 if (unlikely(get_user(off, offset)))
1342 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1343 if (unlikely(put_user(pos, offset)))
1348 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1351 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1352 compat_loff_t __user *, offset, compat_size_t, count)
1358 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1360 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1361 if (unlikely(put_user(pos, offset)))
1366 return do_sendfile(out_fd, in_fd, NULL, count, 0);