Power distribution grids are currently challenged by observability issues due to limited metering infrastructure. On the other hand, smart meter data, including local voltage magnitudes and power injections, are collected at grid nodes with renewable generation and demand response programs. A power flow-based approach using these data is put forth here to infer the unknown power injections at non-metered grid nodes. Exploiting the control capabilities of smart inverters and the relative time-invariance of conventional loads, the idea is to solve the non-linear power flow equations jointly over two system realizations. An intuitive condition pertaining to the graph of the underlying grid is shown to be necessary and sufficient for the local identifiability of this task. The derived graph theoretic criterion can be checked efficiently and is numerically verified under realistic scenarios.