I'm strictly against the use of what has been known as "relativistic" mass in the early days of special relativity. Then it was even worse: They introduced also two kinds of relativstic mass, called transverse and longitudinal mass. All this makes a clear subject, namely the special relativity very confusing, and as soon as you go even further to general relativity it's impossible to get anything clear with such non-covariant concepts.
In modern language thus mass refers to the invariant mass of a system. This clarifies a lot of confusing issues of the early days, particularly when it comes to really non-trivial concepts like the renormalization of mass of charged bodies/particles in classical and quantum electrodynamics or even more complicated quantum field theories as the complete standard model, etc.
That said, the photon mass in free space is, within the standard model, 0. There is, however, no first principle telling us that it must be 0, because you can give even a "naive" mass to a renormalizable Abelian gauge-field theory without spoiling gauge invariance and renormalizability (Stückelberg formalism for Abelian massive gauge fields). Thus, we have to consider the masslessness of the photon a pretty precisely measured empirical input into the standard model.
Of course, there is not only free space but also macroscopic bodies, and there the whole issue becomes even more involved. There you also have contributions to the mass from the interaction of the photons with the medium. The funny thing in the context of the photon mass here are superconductors. Effectively superconductors can be described as a "Higgsed QED", i.e., the electromagnetic gauge symmetry is "spontaneously broken" in the sense of the Higgs mechanism. This has been deduced by Anderson some time before Higgs's famous paper on the Higgs mechanism in electroweak theory! Thus, in a superconductor photons acquire a mass through this Anderson-Higgs mechanism.