Online Social Networks are ubiquitous, bringing not only numerous new possibilities but also big threats and challenges. Privacy is one of them. Most social networks today offer a limited set of (static) privacy settings, not being able to express dynamic policies. For instance, users might decide to protect their location during the night, or share information with difference audiences depending on their current position. In this paper we introduce TFPPF, a formal framework to express, and reason about, dynamic (and recurrent) privacy policies that are activated or deactivated by context (events) or time. Besides a formal policy language (TPPL), the framework includes a knowledge-based logic extended with (linear) temporal operators and a learning modality (TKBL). Policies, and formulae in the logic, are interpreted over (timed) traces representing the evolution of the social network. We prove that checking privacy policy conformance, and the model-checking problem for TKBL, are both decidable.