This question is probably too old now for any answer to be useful to the original poster.
But anyway, the best methods are correction and inspection: First one must get the problem right; in this case it seems obvious to me that the matrix entry in the upper right corner should be b instead of 1, and the entry in the lower left corner should be c instead of 1.
With that in place one notes, by inspection, that the problem is translation invariant. Hence an orthonormal set of eigenvectors are (discretised) plane waves,
Ψn(km) = exp(ikmn)/√N, with km = 2πm/N and m=0, 1, ..., N-1.
Here it is natural (but not mandatory) to let the vector index n run from zero, n = 0, 1, ..., N-1. The corresponding eigenvalues are
λ(k) = a + b e-ik + c eik.
In the unlikely case that the original problem is correct, considerable progress can still be made, analytically reducing the problem to a non-standard 2x2 matrix eigenvalue problem. I.e., with the eigenvalue parameter entering in a highly nonlinear manner, so for explicit results one must ultimately resort to numerical computations.