A fundamental challenge in millimeter-wave (mmWave) communication is the susceptibility to blocking objects. One way to alleviate this problem is the use of reconfigurable intelligent surfaces (RIS). Nevertheless, due to the large number of passive reflecting elements on RIS, channel estimation turns out to be a challenging task. In this paper, we address the channel estimation for RIS-aided mmWave communication systems based on a localization method. The proposed idea consists of exploiting the sparsity of the mmWave channel and the topology of the RIS. In particular, we first propose the concept of reflecting unit set (RUS) to improve the flexibility of RIS. We then propose a novel coplanar maximum likelihood-based (CML) 3D positioning method based on the RUS, and derive the Cramer-Rao lower bound (CRLB) for the positioning method. Furthermore, we develop an efficient positioning-based channel estimation scheme with low computational complexity. Compared to state-of-the-art methods, our proposed method requires less time-frequency resources in channel acquisition as the complexity is independent to the total size of the RIS but depends on the size of the RUSs, which is only a small portion of the RIS. Large performance gains are confirmed in simulations, which proves the effectiveness of the proposed method.