Traditional cellular networks have been considered the most promising candidates to support machine to machine (M2M) communication mainly due to their ubiquitous coverage. Optimally designed to support human to human (H2H) communication, an innovative access to radio resources is required to accommodate M2M unique features such as the massive number of machine type devices (MTDs) as well as the limited data transmission session. In this paper, we consider a simultaneous access to the spectrum in an M2M/H2H coexistence scenario. Taking the advantage of the new device to device (D2D) communication paradigm enabled in long term evolution-advanced (LTE-A), we propose to combine M2M and D2D owing to the MTD low transmit power and thus enabling efficiently the resource sharing. First, we formulate the resource sharing problem as a maximization of the sum-rate, problem for which the optimal solution has been proved to be non deterministic polynomial time hard (NP-Hard). We next model the problem as a novel interference-aware bipartite graph to overcome the computational complexity of the optimal solution. To solve this problem, we consider here a two-phase resource allocation approach. In the first phase, H2H users resource assignment is performed in a conventional way. In the second phase, we introduce two alternative algorithms, one centralized and one semi-distributed to perform M2M resource allocation. The computational complexity of both introduced algorithms whose aim is to solve the M2M resource allocation, is of polynomial complexity. Simulation results show that the semi-distributed M2M resource allocation algorithm achieves quite good performance in terms of network aggregate sum-rate with markedly lower communication overhead compared to the centralized one.