This paper investigates the joint channel estimation and localization problem for integrated sensing and communication (ISAC) systems assisted by a reconfigurable intelligent surface (RIS). The key technical challenges involve addressing the computational complexity of high-dimensional sparse recovery while effectively exploiting geometric relationships in parameter estimation. To begin with, we formulate the joint estimation problem as a geometrically-constrained sparse recovery task based on maximum a posteriori estimation principles, incorporating physical feasibility constraints. Then, we propose the geometrically-constrained separable multidimensional orthogonal matching pursuit (GSMOMP) algorithm, which iteratively executes three key procedures: (1) coarse candidate selection via separable projections to avoid high-dimensional tensor construction; (2) geometric feasibility filtering to eliminate physically implausible propagation paths; and (3) refined atom selection with joint parameter updates that concurrently optimize both channel estimates and location parameters. The convergence is ensured by analytically demonstrating the monotonic decrease of the residual error. Finally, numerical results verify the effectiveness of the proposed algorithm, which achieves substantial computational efficiency improvement compared to conventional MOMP while maintaining comparable localization and angular estimation accuracy.