Retrosynthesis is a procedure where a molecule is transformed into potential reactants and thus the synthesis routes are identified. We propose a novel generative framework, denoted as $\mathsf{G^2Retro}$, for one-step retrosynthesis prediction. $\mathsf{G^2Retro}$ imitates the reversed logic of synthetic reactions, that is, first predicting the reaction centers to convert the target molecule into fragments named synthons, and then transforming synthons into reactants, following previous semi-template-based methods. In predicting reaction centers, $\mathsf{G^2Retro}$ defines a comprehensive set of reaction center types, and enables diversity in the predicted reactions by considering multiple reaction center candidates. In completing synthons, $\mathsf{G^2Retro}$ deploys a sequence of substructure attachments to transform synthons into reactants, which utilize a holistic view of the most updated structures of the synthons to be completed, as well as all the involved synthon and product structures. Here we show that $\mathsf{G^2Retro}$ is able to better prioritize the most possible reactants in the benchmark dataset than the state-of-the-art methods, and discover novel and highly likely reactions that are not included in the benchmark dataset.