In this article, we focus on the channel estimation problem for reconfigurable intelligent surface (RIS)-assisted multiuser mmWave MIMO system. The main challenge for channel estimation of this system lies in that all elements of the RIS are passive reflecting antennas incapable of processing signals, which are designed to reflect the incident signals by a certain phase shift. Meanwhile, considering that array blockage is often present at the RIS working in mmWave band, the channel estimation problem becomes more challenging. By introducing the effective channel model derived from the cascaded channel model among the base station (BS), the RIS, and the user equipment, we formulate the original problem to estimate the effective channel and array blockage matrix jointly. For the single-user scenario, we propose the orthogonal matching pursuit-array blockage calibration (OMPABC) algorithm to complete this estimation problem. In the multiuser scenario, with the aid of channel estimation scheme based on feedback, all the channel matrices of UEs and array blockage matrix are estimated at the BS jointly. Furthermore, in the small number of UEs case, we introduce the common support among the different UEs and extend this property to the effective channel. Utilizing the property, we propose the joint OMPABC algorithm to estimate the effective channel of all UEs and array blockage matrix jointly. Finally, in the large number of UEs case, we propose the two-space-scale OMPABC algorithm with a low computational complexity. The simulation results illustrate our proposed algorithms can effectively improve the estimation performance.