Considering bandwidth-hungry and low latency requirement of vehicular communications applications, we propose a novel dynamic reinforcement learning-based slicing framework and optimization solutions for efficient resource provisioning in virtualized network for D2D-based vehicle-to-vehicle (V2V) communication. The aim is to balance resource utilization and quality of service (QoS) satisfaction levels for multiple slices. The slicing framework is designed as a three-stage layered framework. In the first stage, we propose dynamic deep reinforcement learning-based virtual resource allocation scheme to allocate distinct resources to slices. In the second stage, we aggregate the D2D resource portion of the slice resource for D2D-based V2V communication. In the third stage, due to the computational complexity and signaling overhead of the physical resource allocation, we transform the problem into a convex optimization problem and solve with an alternating direction method of multipliers-based distributed algorithm. Performance results are provided in terms of resource utilization, QoS satisfaction and throughput to show the benefit of integrating resource slices dedicated to supporting interslice D2D-based V2V communication in vehicular network.