Pinching-Antenna Systems (PASS) have attracted considerable attention, since they are capable of beneficially reconfiguring the wireless environment by establishing powerful line-of-sight paths. In this paper, a Bayesian localization and tracking framework is developed for uplink PASS, which explicitly exploits propagation delay information enabled by orthogonal frequency division multiplexing (OFDM). The delay-dispersed received signals are modeled in the frequency domain, and a Bayesian localization tracking algorithm is proposed by jointly leveraging variational inference and belief propagation to extract path delays, infer user locations, and exploit the temporal correlation across consecutive time slots. Numerical results demonstrate that the proposed PASS-based tracking scheme achieves a user tracking performance comparable to that of a conventional multi-base-station system, while requiring significantly fewer RF chains. Moreover, it consistently outperforms conventional baselines when the number of RF chains is aligned.