As the explosion of Internet traffic is quickly leading to overloaded cellular network, device-to-device (D2D)-based content sharing is proposed as a method to offload mobile data traffic. The performance of D2D-based content sharing is dramatically affected by the success rate of content fetching from nearby devices and quality of content transmission, which is determined by the geographic distribution of mobile devices, the number of devices having contents in their caches, and the condition of D2D links. Hence, a key problem is how to cache various contents in the limited storage of mobile devices for improving the success rate of content fetching. In this paper, we aim to design a caching policy by considering the joint impact of locality of real-world mobile data traffic and device contact pattern to improve the success rate of content fetching. To do this, we first study the characteristics of network traffic and device contact pattern by analyzing traces from realistic networks. Then, we design a locality-based caching policy and derive the content caching probability and hit ratio through mathematical analysis. Through numerical evaluation and trace-driven simulations, we not only quantify how content popularity, content active lifetime, content size, content bit rate, device storage, transmission rate, and closeness centrality affect the content hit ratio, but also provide comparison on hit ratio and storage cost in different caching policy, which is a strong evidence that the joint impacts from characteristics of content and device are the necessary consideration when to design a caching policy.