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)))
336 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
341 if (!file->f_op->read_iter)
344 init_sync_kiocb(&kiocb, file);
345 kiocb.ki_pos = *ppos;
348 ret = file->f_op->read_iter(&kiocb, iter);
349 BUG_ON(ret == -EIOCBQUEUED);
351 *ppos = kiocb.ki_pos;
354 EXPORT_SYMBOL(vfs_iter_read);
356 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
361 if (!file->f_op->write_iter)
364 init_sync_kiocb(&kiocb, file);
365 kiocb.ki_pos = *ppos;
368 ret = file->f_op->write_iter(&kiocb, iter);
369 BUG_ON(ret == -EIOCBQUEUED);
371 *ppos = kiocb.ki_pos;
374 EXPORT_SYMBOL(vfs_iter_write);
377 * rw_verify_area doesn't like huge counts. We limit
378 * them to something that fits in "int" so that others
379 * won't have to do range checks all the time.
381 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
385 int retval = -EINVAL;
387 inode = file_inode(file);
388 if (unlikely((ssize_t) count < 0))
391 if (unlikely(pos < 0)) {
392 if (!unsigned_offsets(file))
394 if (count >= -pos) /* both values are in 0..LLONG_MAX */
396 } else if (unlikely((loff_t) (pos + count) < 0)) {
397 if (!unsigned_offsets(file))
401 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
402 retval = locks_mandatory_area(
403 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
404 inode, file, pos, count);
408 retval = security_file_permission(file,
409 read_write == READ ? MAY_READ : MAY_WRITE);
412 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
415 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
417 struct iovec iov = { .iov_base = buf, .iov_len = len };
421 init_sync_kiocb(&kiocb, filp);
422 kiocb.ki_pos = *ppos;
424 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
425 BUG_ON(ret == -EIOCBQUEUED);
426 *ppos = kiocb.ki_pos;
430 EXPORT_SYMBOL(do_sync_read);
432 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
434 struct iovec iov = { .iov_base = buf, .iov_len = len };
436 struct iov_iter iter;
439 init_sync_kiocb(&kiocb, filp);
440 kiocb.ki_pos = *ppos;
441 iov_iter_init(&iter, READ, &iov, 1, len);
443 ret = filp->f_op->read_iter(&kiocb, &iter);
444 BUG_ON(ret == -EIOCBQUEUED);
445 *ppos = kiocb.ki_pos;
449 EXPORT_SYMBOL(new_sync_read);
451 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
456 if (file->f_op->read)
457 ret = file->f_op->read(file, buf, count, pos);
458 else if (file->f_op->aio_read)
459 ret = do_sync_read(file, buf, count, pos);
460 else if (file->f_op->read_iter)
461 ret = new_sync_read(file, buf, count, pos);
468 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
472 if (!(file->f_mode & FMODE_READ))
474 if (!(file->f_mode & FMODE_CAN_READ))
476 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
479 ret = rw_verify_area(READ, file, pos, count);
482 ret = __vfs_read(file, buf, count, pos);
484 fsnotify_access(file);
485 add_rchar(current, ret);
493 EXPORT_SYMBOL(vfs_read);
495 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
497 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
501 init_sync_kiocb(&kiocb, filp);
502 kiocb.ki_pos = *ppos;
504 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
505 BUG_ON(ret == -EIOCBQUEUED);
506 *ppos = kiocb.ki_pos;
510 EXPORT_SYMBOL(do_sync_write);
512 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
514 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
516 struct iov_iter iter;
519 init_sync_kiocb(&kiocb, filp);
520 kiocb.ki_pos = *ppos;
521 iov_iter_init(&iter, WRITE, &iov, 1, len);
523 ret = filp->f_op->write_iter(&kiocb, &iter);
524 BUG_ON(ret == -EIOCBQUEUED);
525 *ppos = kiocb.ki_pos;
529 EXPORT_SYMBOL(new_sync_write);
531 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
534 const char __user *p;
537 if (!(file->f_mode & FMODE_CAN_WRITE))
542 p = (__force const char __user *)buf;
543 if (count > MAX_RW_COUNT)
544 count = MAX_RW_COUNT;
545 if (file->f_op->write)
546 ret = file->f_op->write(file, p, count, pos);
547 else if (file->f_op->aio_write)
548 ret = do_sync_write(file, p, count, pos);
550 ret = new_sync_write(file, p, count, pos);
553 fsnotify_modify(file);
554 add_wchar(current, ret);
560 EXPORT_SYMBOL(__kernel_write);
562 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
566 if (!(file->f_mode & FMODE_WRITE))
568 if (!(file->f_mode & FMODE_CAN_WRITE))
570 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
573 ret = rw_verify_area(WRITE, file, pos, count);
576 file_start_write(file);
577 if (file->f_op->write)
578 ret = file->f_op->write(file, buf, count, pos);
579 else if (file->f_op->aio_write)
580 ret = do_sync_write(file, buf, count, pos);
582 ret = new_sync_write(file, buf, count, pos);
584 fsnotify_modify(file);
585 add_wchar(current, ret);
588 file_end_write(file);
594 EXPORT_SYMBOL(vfs_write);
596 static inline loff_t file_pos_read(struct file *file)
601 static inline void file_pos_write(struct file *file, loff_t pos)
606 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
608 struct fd f = fdget_pos(fd);
609 ssize_t ret = -EBADF;
612 loff_t pos = file_pos_read(f.file);
613 ret = vfs_read(f.file, buf, count, &pos);
615 file_pos_write(f.file, pos);
621 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
624 struct fd f = fdget_pos(fd);
625 ssize_t ret = -EBADF;
628 loff_t pos = file_pos_read(f.file);
629 ret = vfs_write(f.file, buf, count, &pos);
631 file_pos_write(f.file, pos);
638 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
639 size_t, count, loff_t, pos)
642 ssize_t ret = -EBADF;
650 if (f.file->f_mode & FMODE_PREAD)
651 ret = vfs_read(f.file, buf, count, &pos);
658 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
659 size_t, count, loff_t, pos)
662 ssize_t ret = -EBADF;
670 if (f.file->f_mode & FMODE_PWRITE)
671 ret = vfs_write(f.file, buf, count, &pos);
679 * Reduce an iovec's length in-place. Return the resulting number of segments
681 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
683 unsigned long seg = 0;
686 while (seg < nr_segs) {
688 if (len + iov->iov_len >= to) {
689 iov->iov_len = to - len;
697 EXPORT_SYMBOL(iov_shorten);
699 static ssize_t do_iter_readv_writev(struct file *filp, int rw, const struct iovec *iov,
700 unsigned long nr_segs, size_t len, loff_t *ppos, iter_fn_t fn)
703 struct iov_iter iter;
706 init_sync_kiocb(&kiocb, filp);
707 kiocb.ki_pos = *ppos;
709 iov_iter_init(&iter, rw, iov, nr_segs, len);
710 ret = fn(&kiocb, &iter);
711 BUG_ON(ret == -EIOCBQUEUED);
712 *ppos = kiocb.ki_pos;
716 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
717 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
722 init_sync_kiocb(&kiocb, filp);
723 kiocb.ki_pos = *ppos;
725 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
726 BUG_ON(ret == -EIOCBQUEUED);
727 *ppos = kiocb.ki_pos;
731 /* Do it by hand, with file-ops */
732 static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
733 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
735 struct iovec *vector = iov;
738 while (nr_segs > 0) {
743 base = vector->iov_base;
744 len = vector->iov_len;
748 nr = fn(filp, base, len, ppos);
763 /* A write operation does a read from user space and vice versa */
764 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
766 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
767 unsigned long nr_segs, unsigned long fast_segs,
768 struct iovec *fast_pointer,
769 struct iovec **ret_pointer)
773 struct iovec *iov = fast_pointer;
776 * SuS says "The readv() function *may* fail if the iovcnt argument
777 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
778 * traditionally returned zero for zero segments, so...
786 * First get the "struct iovec" from user memory and
787 * verify all the pointers
789 if (nr_segs > UIO_MAXIOV) {
793 if (nr_segs > fast_segs) {
794 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
800 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
806 * According to the Single Unix Specification we should return EINVAL
807 * if an element length is < 0 when cast to ssize_t or if the
808 * total length would overflow the ssize_t return value of the
811 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
815 for (seg = 0; seg < nr_segs; seg++) {
816 void __user *buf = iov[seg].iov_base;
817 ssize_t len = (ssize_t)iov[seg].iov_len;
819 /* see if we we're about to use an invalid len or if
820 * it's about to overflow ssize_t */
826 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
830 if (len > MAX_RW_COUNT - ret) {
831 len = MAX_RW_COUNT - ret;
832 iov[seg].iov_len = len;
841 static ssize_t do_readv_writev(int type, struct file *file,
842 const struct iovec __user * uvector,
843 unsigned long nr_segs, loff_t *pos)
846 struct iovec iovstack[UIO_FASTIOV];
847 struct iovec *iov = iovstack;
853 ret = rw_copy_check_uvector(type, uvector, nr_segs,
854 ARRAY_SIZE(iovstack), iovstack, &iov);
859 ret = rw_verify_area(type, file, pos, tot_len);
865 fn = file->f_op->read;
866 fnv = file->f_op->aio_read;
867 iter_fn = file->f_op->read_iter;
869 fn = (io_fn_t)file->f_op->write;
870 fnv = file->f_op->aio_write;
871 iter_fn = file->f_op->write_iter;
872 file_start_write(file);
876 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
879 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
882 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
885 file_end_write(file);
890 if ((ret + (type == READ)) > 0) {
892 fsnotify_access(file);
894 fsnotify_modify(file);
899 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
900 unsigned long vlen, loff_t *pos)
902 if (!(file->f_mode & FMODE_READ))
904 if (!(file->f_mode & FMODE_CAN_READ))
907 return do_readv_writev(READ, file, vec, vlen, pos);
910 EXPORT_SYMBOL(vfs_readv);
912 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
913 unsigned long vlen, loff_t *pos)
915 if (!(file->f_mode & FMODE_WRITE))
917 if (!(file->f_mode & FMODE_CAN_WRITE))
920 return do_readv_writev(WRITE, file, vec, vlen, pos);
923 EXPORT_SYMBOL(vfs_writev);
925 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
928 struct fd f = fdget_pos(fd);
929 ssize_t ret = -EBADF;
932 loff_t pos = file_pos_read(f.file);
933 ret = vfs_readv(f.file, vec, vlen, &pos);
935 file_pos_write(f.file, pos);
940 add_rchar(current, ret);
945 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
948 struct fd f = fdget_pos(fd);
949 ssize_t ret = -EBADF;
952 loff_t pos = file_pos_read(f.file);
953 ret = vfs_writev(f.file, vec, vlen, &pos);
955 file_pos_write(f.file, pos);
960 add_wchar(current, ret);
965 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
967 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
968 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
971 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
972 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
974 loff_t pos = pos_from_hilo(pos_h, pos_l);
976 ssize_t ret = -EBADF;
984 if (f.file->f_mode & FMODE_PREAD)
985 ret = vfs_readv(f.file, vec, vlen, &pos);
990 add_rchar(current, ret);
995 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
996 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
998 loff_t pos = pos_from_hilo(pos_h, pos_l);
1000 ssize_t ret = -EBADF;
1008 if (f.file->f_mode & FMODE_PWRITE)
1009 ret = vfs_writev(f.file, vec, vlen, &pos);
1014 add_wchar(current, ret);
1019 #ifdef CONFIG_COMPAT
1021 static ssize_t compat_do_readv_writev(int type, struct file *file,
1022 const struct compat_iovec __user *uvector,
1023 unsigned long nr_segs, loff_t *pos)
1025 compat_ssize_t tot_len;
1026 struct iovec iovstack[UIO_FASTIOV];
1027 struct iovec *iov = iovstack;
1033 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1034 UIO_FASTIOV, iovstack, &iov);
1039 ret = rw_verify_area(type, file, pos, tot_len);
1045 fn = file->f_op->read;
1046 fnv = file->f_op->aio_read;
1047 iter_fn = file->f_op->read_iter;
1049 fn = (io_fn_t)file->f_op->write;
1050 fnv = file->f_op->aio_write;
1051 iter_fn = file->f_op->write_iter;
1052 file_start_write(file);
1056 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
1059 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1062 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1065 file_end_write(file);
1068 if (iov != iovstack)
1070 if ((ret + (type == READ)) > 0) {
1072 fsnotify_access(file);
1074 fsnotify_modify(file);
1079 static size_t compat_readv(struct file *file,
1080 const struct compat_iovec __user *vec,
1081 unsigned long vlen, loff_t *pos)
1083 ssize_t ret = -EBADF;
1085 if (!(file->f_mode & FMODE_READ))
1089 if (!(file->f_mode & FMODE_CAN_READ))
1092 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1096 add_rchar(current, ret);
1101 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1102 const struct compat_iovec __user *,vec,
1103 compat_ulong_t, vlen)
1105 struct fd f = fdget_pos(fd);
1111 pos = f.file->f_pos;
1112 ret = compat_readv(f.file, vec, vlen, &pos);
1114 f.file->f_pos = pos;
1119 static long __compat_sys_preadv64(unsigned long fd,
1120 const struct compat_iovec __user *vec,
1121 unsigned long vlen, loff_t pos)
1132 if (f.file->f_mode & FMODE_PREAD)
1133 ret = compat_readv(f.file, vec, vlen, &pos);
1138 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1139 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1140 const struct compat_iovec __user *,vec,
1141 unsigned long, vlen, loff_t, pos)
1143 return __compat_sys_preadv64(fd, vec, vlen, pos);
1147 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1148 const struct compat_iovec __user *,vec,
1149 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1151 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1153 return __compat_sys_preadv64(fd, vec, vlen, pos);
1156 static size_t compat_writev(struct file *file,
1157 const struct compat_iovec __user *vec,
1158 unsigned long vlen, loff_t *pos)
1160 ssize_t ret = -EBADF;
1162 if (!(file->f_mode & FMODE_WRITE))
1166 if (!(file->f_mode & FMODE_CAN_WRITE))
1169 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1173 add_wchar(current, ret);
1178 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1179 const struct compat_iovec __user *, vec,
1180 compat_ulong_t, vlen)
1182 struct fd f = fdget_pos(fd);
1188 pos = f.file->f_pos;
1189 ret = compat_writev(f.file, vec, vlen, &pos);
1191 f.file->f_pos = pos;
1196 static long __compat_sys_pwritev64(unsigned long fd,
1197 const struct compat_iovec __user *vec,
1198 unsigned long vlen, loff_t pos)
1209 if (f.file->f_mode & FMODE_PWRITE)
1210 ret = compat_writev(f.file, vec, vlen, &pos);
1215 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1216 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1217 const struct compat_iovec __user *,vec,
1218 unsigned long, vlen, loff_t, pos)
1220 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1224 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1225 const struct compat_iovec __user *,vec,
1226 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1228 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1230 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1234 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1235 size_t count, loff_t max)
1238 struct inode *in_inode, *out_inode;
1245 * Get input file, and verify that it is ok..
1251 if (!(in.file->f_mode & FMODE_READ))
1255 pos = in.file->f_pos;
1258 if (!(in.file->f_mode & FMODE_PREAD))
1261 retval = rw_verify_area(READ, in.file, &pos, count);
1267 * Get output file, and verify that it is ok..
1270 out = fdget(out_fd);
1273 if (!(out.file->f_mode & FMODE_WRITE))
1276 in_inode = file_inode(in.file);
1277 out_inode = file_inode(out.file);
1278 out_pos = out.file->f_pos;
1279 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1285 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1287 if (unlikely(pos + count > max)) {
1288 retval = -EOVERFLOW;
1297 * We need to debate whether we can enable this or not. The
1298 * man page documents EAGAIN return for the output at least,
1299 * and the application is arguably buggy if it doesn't expect
1300 * EAGAIN on a non-blocking file descriptor.
1302 if (in.file->f_flags & O_NONBLOCK)
1303 fl = SPLICE_F_NONBLOCK;
1305 file_start_write(out.file);
1306 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1307 file_end_write(out.file);
1310 add_rchar(current, retval);
1311 add_wchar(current, retval);
1312 fsnotify_access(in.file);
1313 fsnotify_modify(out.file);
1314 out.file->f_pos = out_pos;
1318 in.file->f_pos = pos;
1324 retval = -EOVERFLOW;
1334 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1341 if (unlikely(get_user(off, offset)))
1344 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1345 if (unlikely(put_user(pos, offset)))
1350 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1353 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1359 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1361 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1362 if (unlikely(put_user(pos, offset)))
1367 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1370 #ifdef CONFIG_COMPAT
1371 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1372 compat_off_t __user *, offset, compat_size_t, count)
1379 if (unlikely(get_user(off, offset)))
1382 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1383 if (unlikely(put_user(pos, offset)))
1388 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1391 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1392 compat_loff_t __user *, offset, compat_size_t, count)
1398 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1400 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1401 if (unlikely(put_user(pos, offset)))
1406 return do_sendfile(out_fd, in_fd, NULL, count, 0);