ovl: use plain list filler in indexdir and workdir cleanup

Those two cleanup routines are using the helper ovl_dir_read() with the
merge dir filler, which populates an rb tree, that is never used.

The index dir entry names all have a long (42 bytes) constant prefix, so it
is not surprising that perf top has demostrated high CPU usage by rb tree
population during cleanup of a large index dir:

      - 9.53% ovl_fill_merge
         - 78.41% ovl_cache_entry_find_link.constprop.27
            + 72.11% strncmp

Use the plain list filler that does not populate the unneeded rb tree.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fs/overlayfs/readdir.c

@@ -1071,14 +1071,10 @@ static int ovl_workdir_cleanup_recurse(struct ovl_fs *ofs, const struct path *pa
 	int err;
 	struct inode *dir = path->dentry->d_inode;
 	LIST_HEAD(list);
-	struct rb_root root = RB_ROOT;
 	struct ovl_cache_entry *p;
 	struct ovl_readdir_data rdd = {
-		.ctx.actor = ovl_fill_merge,
-		.dentry = NULL,
+		.ctx.actor = ovl_fill_plain,
 		.list = &list,
-		.root = &root,
-		.is_lowest = false,
 	};
 	bool incompat = false;
 
@@ -1159,14 +1155,10 @@ int ovl_indexdir_cleanup(struct ovl_fs *ofs)
 	struct inode *dir = indexdir->d_inode;
 	struct path path = { .mnt = ovl_upper_mnt(ofs), .dentry = indexdir };
 	LIST_HEAD(list);
-	struct rb_root root = RB_ROOT;
 	struct ovl_cache_entry *p;
 	struct ovl_readdir_data rdd = {
-		.ctx.actor = ovl_fill_merge,
-		.dentry = NULL,
+		.ctx.actor = ovl_fill_plain,
 		.list = &list,
-		.root = &root,
-		.is_lowest = false,
 	};
 
 	err = ovl_dir_read(&path, &rdd);