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 = new_sync_read,
32 .read_iter = generic_file_read_iter,
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->llseek)
261 fn = file->f_op->llseek;
263 return fn(file, offset, whence);
265 EXPORT_SYMBOL(vfs_llseek);
267 static inline struct fd fdget_pos(int fd)
269 return __to_fd(__fdget_pos(fd));
272 static inline void fdput_pos(struct fd f)
274 if (f.flags & FDPUT_POS_UNLOCK)
275 mutex_unlock(&f.file->f_pos_lock);
279 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
282 struct fd f = fdget_pos(fd);
287 if (whence <= SEEK_MAX) {
288 loff_t res = vfs_llseek(f.file, offset, whence);
290 if (res != (loff_t)retval)
291 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
298 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
300 return sys_lseek(fd, offset, whence);
304 #ifdef __ARCH_WANT_SYS_LLSEEK
305 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
306 unsigned long, offset_low, loff_t __user *, result,
307 unsigned int, whence)
310 struct fd f = fdget_pos(fd);
317 if (whence > SEEK_MAX)
320 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
323 retval = (int)offset;
326 if (!copy_to_user(result, &offset, sizeof(offset)))
335 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
340 if (!file->f_op->read_iter)
343 init_sync_kiocb(&kiocb, file);
344 kiocb.ki_pos = *ppos;
347 ret = file->f_op->read_iter(&kiocb, iter);
348 BUG_ON(ret == -EIOCBQUEUED);
350 *ppos = kiocb.ki_pos;
353 EXPORT_SYMBOL(vfs_iter_read);
355 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
360 if (!file->f_op->write_iter)
363 init_sync_kiocb(&kiocb, file);
364 kiocb.ki_pos = *ppos;
367 ret = file->f_op->write_iter(&kiocb, iter);
368 BUG_ON(ret == -EIOCBQUEUED);
370 *ppos = kiocb.ki_pos;
373 EXPORT_SYMBOL(vfs_iter_write);
376 * rw_verify_area doesn't like huge counts. We limit
377 * them to something that fits in "int" so that others
378 * won't have to do range checks all the time.
380 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
384 int retval = -EINVAL;
386 inode = file_inode(file);
387 if (unlikely((ssize_t) count < 0))
390 if (unlikely(pos < 0)) {
391 if (!unsigned_offsets(file))
393 if (count >= -pos) /* both values are in 0..LLONG_MAX */
395 } else if (unlikely((loff_t) (pos + count) < 0)) {
396 if (!unsigned_offsets(file))
400 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
401 retval = locks_mandatory_area(
402 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
403 inode, file, pos, count);
407 retval = security_file_permission(file,
408 read_write == READ ? MAY_READ : MAY_WRITE);
411 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
414 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
416 struct iovec iov = { .iov_base = buf, .iov_len = len };
420 init_sync_kiocb(&kiocb, filp);
421 kiocb.ki_pos = *ppos;
423 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
424 BUG_ON(ret == -EIOCBQUEUED);
425 *ppos = kiocb.ki_pos;
429 EXPORT_SYMBOL(do_sync_read);
431 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
433 struct iovec iov = { .iov_base = buf, .iov_len = len };
435 struct iov_iter iter;
438 init_sync_kiocb(&kiocb, filp);
439 kiocb.ki_pos = *ppos;
440 iov_iter_init(&iter, READ, &iov, 1, len);
442 ret = filp->f_op->read_iter(&kiocb, &iter);
443 BUG_ON(ret == -EIOCBQUEUED);
444 *ppos = kiocb.ki_pos;
448 EXPORT_SYMBOL(new_sync_read);
450 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
455 if (file->f_op->read)
456 ret = file->f_op->read(file, buf, count, pos);
457 else if (file->f_op->aio_read)
458 ret = do_sync_read(file, buf, count, pos);
459 else if (file->f_op->read_iter)
460 ret = new_sync_read(file, buf, count, pos);
467 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
471 if (!(file->f_mode & FMODE_READ))
473 if (!(file->f_mode & FMODE_CAN_READ))
475 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
478 ret = rw_verify_area(READ, file, pos, count);
481 ret = __vfs_read(file, buf, count, pos);
483 fsnotify_access(file);
484 add_rchar(current, ret);
492 EXPORT_SYMBOL(vfs_read);
494 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
496 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
500 init_sync_kiocb(&kiocb, filp);
501 kiocb.ki_pos = *ppos;
503 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
504 BUG_ON(ret == -EIOCBQUEUED);
505 *ppos = kiocb.ki_pos;
509 EXPORT_SYMBOL(do_sync_write);
511 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
513 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
515 struct iov_iter iter;
518 init_sync_kiocb(&kiocb, filp);
519 kiocb.ki_pos = *ppos;
520 iov_iter_init(&iter, WRITE, &iov, 1, len);
522 ret = filp->f_op->write_iter(&kiocb, &iter);
523 BUG_ON(ret == -EIOCBQUEUED);
524 *ppos = kiocb.ki_pos;
528 EXPORT_SYMBOL(new_sync_write);
530 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
533 const char __user *p;
536 if (!(file->f_mode & FMODE_CAN_WRITE))
541 p = (__force const char __user *)buf;
542 if (count > MAX_RW_COUNT)
543 count = MAX_RW_COUNT;
544 if (file->f_op->write)
545 ret = file->f_op->write(file, p, count, pos);
546 else if (file->f_op->aio_write)
547 ret = do_sync_write(file, p, count, pos);
549 ret = new_sync_write(file, p, count, pos);
552 fsnotify_modify(file);
553 add_wchar(current, ret);
559 EXPORT_SYMBOL(__kernel_write);
561 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
565 if (!(file->f_mode & FMODE_WRITE))
567 if (!(file->f_mode & FMODE_CAN_WRITE))
569 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
572 ret = rw_verify_area(WRITE, file, pos, count);
575 file_start_write(file);
576 if (file->f_op->write)
577 ret = file->f_op->write(file, buf, count, pos);
578 else if (file->f_op->aio_write)
579 ret = do_sync_write(file, buf, count, pos);
581 ret = new_sync_write(file, buf, count, pos);
583 fsnotify_modify(file);
584 add_wchar(current, ret);
587 file_end_write(file);
593 EXPORT_SYMBOL(vfs_write);
595 static inline loff_t file_pos_read(struct file *file)
600 static inline void file_pos_write(struct file *file, loff_t pos)
605 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
607 struct fd f = fdget_pos(fd);
608 ssize_t ret = -EBADF;
611 loff_t pos = file_pos_read(f.file);
612 ret = vfs_read(f.file, buf, count, &pos);
614 file_pos_write(f.file, pos);
620 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
623 struct fd f = fdget_pos(fd);
624 ssize_t ret = -EBADF;
627 loff_t pos = file_pos_read(f.file);
628 ret = vfs_write(f.file, buf, count, &pos);
630 file_pos_write(f.file, pos);
637 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
638 size_t, count, loff_t, pos)
641 ssize_t ret = -EBADF;
649 if (f.file->f_mode & FMODE_PREAD)
650 ret = vfs_read(f.file, buf, count, &pos);
657 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
658 size_t, count, loff_t, pos)
661 ssize_t ret = -EBADF;
669 if (f.file->f_mode & FMODE_PWRITE)
670 ret = vfs_write(f.file, buf, count, &pos);
678 * Reduce an iovec's length in-place. Return the resulting number of segments
680 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
682 unsigned long seg = 0;
685 while (seg < nr_segs) {
687 if (len + iov->iov_len >= to) {
688 iov->iov_len = to - len;
696 EXPORT_SYMBOL(iov_shorten);
698 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
699 loff_t *ppos, iter_fn_t fn)
704 init_sync_kiocb(&kiocb, filp);
705 kiocb.ki_pos = *ppos;
707 ret = fn(&kiocb, iter);
708 BUG_ON(ret == -EIOCBQUEUED);
709 *ppos = kiocb.ki_pos;
713 static ssize_t do_sync_readv_writev(struct file *filp, struct iov_iter *iter,
714 loff_t *ppos, iov_fn_t fn)
719 init_sync_kiocb(&kiocb, filp);
720 kiocb.ki_pos = *ppos;
722 ret = fn(&kiocb, iter->iov, iter->nr_segs, kiocb.ki_pos);
723 BUG_ON(ret == -EIOCBQUEUED);
724 *ppos = kiocb.ki_pos;
728 /* Do it by hand, with file-ops */
729 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
730 loff_t *ppos, io_fn_t fn)
734 while (iov_iter_count(iter)) {
735 struct iovec iovec = iov_iter_iovec(iter);
738 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
746 if (nr != iovec.iov_len)
748 iov_iter_advance(iter, nr);
754 /* A write operation does a read from user space and vice versa */
755 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
757 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
758 unsigned long nr_segs, unsigned long fast_segs,
759 struct iovec *fast_pointer,
760 struct iovec **ret_pointer)
764 struct iovec *iov = fast_pointer;
767 * SuS says "The readv() function *may* fail if the iovcnt argument
768 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
769 * traditionally returned zero for zero segments, so...
777 * First get the "struct iovec" from user memory and
778 * verify all the pointers
780 if (nr_segs > UIO_MAXIOV) {
784 if (nr_segs > fast_segs) {
785 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
791 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
797 * According to the Single Unix Specification we should return EINVAL
798 * if an element length is < 0 when cast to ssize_t or if the
799 * total length would overflow the ssize_t return value of the
802 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
806 for (seg = 0; seg < nr_segs; seg++) {
807 void __user *buf = iov[seg].iov_base;
808 ssize_t len = (ssize_t)iov[seg].iov_len;
810 /* see if we we're about to use an invalid len or if
811 * it's about to overflow ssize_t */
817 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
821 if (len > MAX_RW_COUNT - ret) {
822 len = MAX_RW_COUNT - ret;
823 iov[seg].iov_len = len;
832 static ssize_t do_readv_writev(int type, struct file *file,
833 const struct iovec __user * uvector,
834 unsigned long nr_segs, loff_t *pos)
837 struct iovec iovstack[UIO_FASTIOV];
838 struct iovec *iov = iovstack;
839 struct iov_iter iter;
845 ret = import_iovec(type, uvector, nr_segs,
846 ARRAY_SIZE(iovstack), &iov, &iter);
850 tot_len = iov_iter_count(&iter);
853 ret = rw_verify_area(type, file, pos, tot_len);
859 fn = file->f_op->read;
860 fnv = file->f_op->aio_read;
861 iter_fn = file->f_op->read_iter;
863 fn = (io_fn_t)file->f_op->write;
864 fnv = file->f_op->aio_write;
865 iter_fn = file->f_op->write_iter;
866 file_start_write(file);
870 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
872 ret = do_sync_readv_writev(file, &iter, pos, fnv);
874 ret = do_loop_readv_writev(file, &iter, pos, fn);
877 file_end_write(file);
881 if ((ret + (type == READ)) > 0) {
883 fsnotify_access(file);
885 fsnotify_modify(file);
890 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
891 unsigned long vlen, loff_t *pos)
893 if (!(file->f_mode & FMODE_READ))
895 if (!(file->f_mode & FMODE_CAN_READ))
898 return do_readv_writev(READ, file, vec, vlen, pos);
901 EXPORT_SYMBOL(vfs_readv);
903 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
904 unsigned long vlen, loff_t *pos)
906 if (!(file->f_mode & FMODE_WRITE))
908 if (!(file->f_mode & FMODE_CAN_WRITE))
911 return do_readv_writev(WRITE, file, vec, vlen, pos);
914 EXPORT_SYMBOL(vfs_writev);
916 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
919 struct fd f = fdget_pos(fd);
920 ssize_t ret = -EBADF;
923 loff_t pos = file_pos_read(f.file);
924 ret = vfs_readv(f.file, vec, vlen, &pos);
926 file_pos_write(f.file, pos);
931 add_rchar(current, ret);
936 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
939 struct fd f = fdget_pos(fd);
940 ssize_t ret = -EBADF;
943 loff_t pos = file_pos_read(f.file);
944 ret = vfs_writev(f.file, vec, vlen, &pos);
946 file_pos_write(f.file, pos);
951 add_wchar(current, ret);
956 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
958 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
959 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
962 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
963 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
965 loff_t pos = pos_from_hilo(pos_h, pos_l);
967 ssize_t ret = -EBADF;
975 if (f.file->f_mode & FMODE_PREAD)
976 ret = vfs_readv(f.file, vec, vlen, &pos);
981 add_rchar(current, ret);
986 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
987 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
989 loff_t pos = pos_from_hilo(pos_h, pos_l);
991 ssize_t ret = -EBADF;
999 if (f.file->f_mode & FMODE_PWRITE)
1000 ret = vfs_writev(f.file, vec, vlen, &pos);
1005 add_wchar(current, ret);
1010 #ifdef CONFIG_COMPAT
1012 static ssize_t compat_do_readv_writev(int type, struct file *file,
1013 const struct compat_iovec __user *uvector,
1014 unsigned long nr_segs, loff_t *pos)
1016 compat_ssize_t tot_len;
1017 struct iovec iovstack[UIO_FASTIOV];
1018 struct iovec *iov = iovstack;
1019 struct iov_iter iter;
1025 ret = compat_import_iovec(type, uvector, nr_segs,
1026 UIO_FASTIOV, &iov, &iter);
1030 tot_len = iov_iter_count(&iter);
1033 ret = rw_verify_area(type, file, pos, tot_len);
1039 fn = file->f_op->read;
1040 fnv = file->f_op->aio_read;
1041 iter_fn = file->f_op->read_iter;
1043 fn = (io_fn_t)file->f_op->write;
1044 fnv = file->f_op->aio_write;
1045 iter_fn = file->f_op->write_iter;
1046 file_start_write(file);
1050 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
1052 ret = do_sync_readv_writev(file, &iter, pos, fnv);
1054 ret = do_loop_readv_writev(file, &iter, pos, fn);
1057 file_end_write(file);
1061 if ((ret + (type == READ)) > 0) {
1063 fsnotify_access(file);
1065 fsnotify_modify(file);
1070 static size_t compat_readv(struct file *file,
1071 const struct compat_iovec __user *vec,
1072 unsigned long vlen, loff_t *pos)
1074 ssize_t ret = -EBADF;
1076 if (!(file->f_mode & FMODE_READ))
1080 if (!(file->f_mode & FMODE_CAN_READ))
1083 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1087 add_rchar(current, ret);
1092 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1093 const struct compat_iovec __user *,vec,
1094 compat_ulong_t, vlen)
1096 struct fd f = fdget_pos(fd);
1102 pos = f.file->f_pos;
1103 ret = compat_readv(f.file, vec, vlen, &pos);
1105 f.file->f_pos = pos;
1110 static long __compat_sys_preadv64(unsigned long fd,
1111 const struct compat_iovec __user *vec,
1112 unsigned long vlen, loff_t pos)
1123 if (f.file->f_mode & FMODE_PREAD)
1124 ret = compat_readv(f.file, vec, vlen, &pos);
1129 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1130 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1131 const struct compat_iovec __user *,vec,
1132 unsigned long, vlen, loff_t, pos)
1134 return __compat_sys_preadv64(fd, vec, vlen, pos);
1138 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1139 const struct compat_iovec __user *,vec,
1140 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1142 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1144 return __compat_sys_preadv64(fd, vec, vlen, pos);
1147 static size_t compat_writev(struct file *file,
1148 const struct compat_iovec __user *vec,
1149 unsigned long vlen, loff_t *pos)
1151 ssize_t ret = -EBADF;
1153 if (!(file->f_mode & FMODE_WRITE))
1157 if (!(file->f_mode & FMODE_CAN_WRITE))
1160 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1164 add_wchar(current, ret);
1169 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1170 const struct compat_iovec __user *, vec,
1171 compat_ulong_t, vlen)
1173 struct fd f = fdget_pos(fd);
1179 pos = f.file->f_pos;
1180 ret = compat_writev(f.file, vec, vlen, &pos);
1182 f.file->f_pos = pos;
1187 static long __compat_sys_pwritev64(unsigned long fd,
1188 const struct compat_iovec __user *vec,
1189 unsigned long vlen, loff_t pos)
1200 if (f.file->f_mode & FMODE_PWRITE)
1201 ret = compat_writev(f.file, vec, vlen, &pos);
1206 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1207 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1208 const struct compat_iovec __user *,vec,
1209 unsigned long, vlen, loff_t, pos)
1211 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1215 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1216 const struct compat_iovec __user *,vec,
1217 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1219 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1221 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1225 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1226 size_t count, loff_t max)
1229 struct inode *in_inode, *out_inode;
1236 * Get input file, and verify that it is ok..
1242 if (!(in.file->f_mode & FMODE_READ))
1246 pos = in.file->f_pos;
1249 if (!(in.file->f_mode & FMODE_PREAD))
1252 retval = rw_verify_area(READ, in.file, &pos, count);
1258 * Get output file, and verify that it is ok..
1261 out = fdget(out_fd);
1264 if (!(out.file->f_mode & FMODE_WRITE))
1267 in_inode = file_inode(in.file);
1268 out_inode = file_inode(out.file);
1269 out_pos = out.file->f_pos;
1270 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1276 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1278 if (unlikely(pos + count > max)) {
1279 retval = -EOVERFLOW;
1288 * We need to debate whether we can enable this or not. The
1289 * man page documents EAGAIN return for the output at least,
1290 * and the application is arguably buggy if it doesn't expect
1291 * EAGAIN on a non-blocking file descriptor.
1293 if (in.file->f_flags & O_NONBLOCK)
1294 fl = SPLICE_F_NONBLOCK;
1296 file_start_write(out.file);
1297 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1298 file_end_write(out.file);
1301 add_rchar(current, retval);
1302 add_wchar(current, retval);
1303 fsnotify_access(in.file);
1304 fsnotify_modify(out.file);
1305 out.file->f_pos = out_pos;
1309 in.file->f_pos = pos;
1315 retval = -EOVERFLOW;
1325 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1332 if (unlikely(get_user(off, offset)))
1335 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1336 if (unlikely(put_user(pos, offset)))
1341 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1344 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1350 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1352 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1353 if (unlikely(put_user(pos, offset)))
1358 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1361 #ifdef CONFIG_COMPAT
1362 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1363 compat_off_t __user *, offset, compat_size_t, count)
1370 if (unlikely(get_user(off, offset)))
1373 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1374 if (unlikely(put_user(pos, offset)))
1379 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1382 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1383 compat_loff_t __user *, offset, compat_size_t, count)
1389 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1391 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1392 if (unlikely(put_user(pos, offset)))
1397 return do_sendfile(out_fd, in_fd, NULL, count, 0);