RPL has received universal acceptance in IPv6 routing of the Internet of things (IoT). However, for large-scale multi-hops networks, the RPL routing model is faced with the problem that some IoT nodes heavily consume routing table storage. Besides, the flattened address architecture in IoT subnet makes this problem more prominent. In this paper, TFAD (tree forwarding model with address automatically distributed), a light-weight and tree-based forwarding model, is proposed to support automatic IPv6 address assignment. TFAD constructs a forwarding-level-tree for all the IoT nodes that make the IPv6 addresses of nodes aggregate highly in each sub-tree. In TFAD, each node only needs to maintain a few forwarding entries, the number of which is equivalent to the number of its direct son-nodes. Moreover, the backup mechanism of parent node in TFAD is designed. This mechanism supports the network topology reconstitution based on the whole sub-tree, achieving fast route-recovery from network failure. The experiments based on real sensor nodes prove that TFAD model possesses not only high performance on routing table storage but also rapidity on routing table learning and routing recovery from failure.