vfs: Add vfs_copy_file_range() support for pagecache copies

This allows us to have an in-kernel copy mechanism that avoids frequent
switches between kernel and user space.  This is especially useful so
NFSD can support server-side copies.

The default (flags=0) means to first attempt copy acceleration, but use
the pagecache if that fails.

Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Padraig Brady <P@draigBrady.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fs/read_write.c

@@ -1354,8 +1354,7 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
 
 	if (!(file_in->f_mode & FMODE_READ) ||
 	    !(file_out->f_mode & FMODE_WRITE) ||
-	    (file_out->f_flags & O_APPEND) ||
-	    !file_out->f_op->copy_file_range)
+	    (file_out->f_flags & O_APPEND))
 		return -EBADF;
 
 	/* this could be relaxed once a method supports cross-fs copies */
@@ -1369,8 +1368,14 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
 	if (ret)
 		return ret;
 
-	ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out, pos_out,
-					      len, flags);
+	ret = -EOPNOTSUPP;
+	if (file_out->f_op->copy_file_range)
+		ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
+						      pos_out, len, flags);
+	if (ret == -EOPNOTSUPP)
+		ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
+				len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
+
 	if (ret > 0) {
 		fsnotify_access(file_in);
 		add_rchar(current, ret);