Reducing the network overhead generated during the search is important in the study of unstructured P2P network. Flooding and random walks are simple and easily implemented. However, a large number of redundant messages generated in the search process are the main reason of producing excessive network overhead. An effective limited search mechanism RFSA (restricted forward search algorithm) is proposed. The search path and redundant search path are defined. As the query messages reaching the node are received by introducing the local messages index caching mechanism, the repeat messages forwarding are eliminated. Using the real-time search path information carried in the search process, the neighbor notes that do not appear in the search path are selected to forward the query messages. Theoretically, the number of messages and network overhead generated by the RFSA. In the simulation, comparative analysis of the limited search mechanism and non-limited mechanism flooding and random walk algorithm is done in the network overhead, query hit rate, search coverage rate, and the number of redundant messages, etc. The results show that this method reduces the generation of a great number of redundant messages, and cuts down the network overload.