Proactive caching holds the promise of improving network throughput and lowering transmission costs in content-centric networks. However, how to encode data in cache-aided networks is still an open problem. In this paper, a two-phase caching system is investigated, in which users are equipped with buffers and user requests are characterized by a probability measure. Based on content division, a unified framework for caching in arbitrary networks is presented. From the perspective of linear space, optimization problems are formulated to obtain the optimal linear coded joint placement and delivery (LCJPD) policy. A low-complexity algorithm is also proposed to give a sub-optimal LCJPD policy. Simulation results demonstrate that caching in networks can significantly reduce the transmission cost.