A typical cloud-radio access network (C-RAN) architecture in fifth-generation (5G) network is composed of one radio cloud center (RCC), moderate amount of radio aggregation units (RAU) and massive remote radio heads (RRH). This paper proposes a distributed medium access control (MAC) scheduling scheme for C-RAN, where RCC and RAU are connected with non-ideal fronthaul. In this scheme, a full scheduling operation and a limited scheduling operation are executed at RCC and RAU, respectively. The full scheduling operation at RCC is based on the overall network information and hence could generate a global-optimized scheduling result with ideal fronthaul. The limited scheduling operation at RAU is based on the latest hybrid automatic repeat request (HARQ) feedback information and channel state information (CSI), and could provide real-time supplement when the fronthaul transfer between RCC and RAU is delayed or lost. The integration of these two scheduling results are also provided to further exert their individual advantages. Simulation results demonstrate that the proposed distributed MAC scheduling scheme could effectively mitigate the performance loss due to non-ideal fronthaul in terms of cell sum throughput and cell edge throughput, and therefore benefits in promoting the feasibility of C-RAN in 5G network architecture evolution.