Massive machine type communications (mMTC) have been regarded as promising applica-tions in the future. One main feature of mMTC is short packet communication. Different from tradi-tional long packet communication, short packet com-munication suffers from transmission rate degradation and a significant error rate is introduced. In this case, traditional resource allocation scheme for mMTC is no longer applicable. In this paper, we explore resource allocation for cellular-based mMTC in the finite block-length regime. First, to mitigate the load of the base station (BS), we establish a framework for cellular-based mMTC, where MTCGs reuse the resources of cellular users (CUs), aggregate the packets generated by MTCDs, and forward them to the BS. Next, we adopt short packet theory to obtain the minimum re-quired blocklength of a packet that transmits a certain amount of information. Then, by modeling the process of MTCGs-assisted communication as a queuing pro-cess, we derive the closed-form expression of the aver-age delay of all MTCDs. Guided by this, we propose a joint power allocation and spectrum sharing scheme to minimize the average delay. Finally, the simulation results verify the correctness of the theoretical results and show that the proposed scheme can reduce the av-erage delay efficiently.