Decentralised Online Social Networks (DOSN) are evolving as a promising approach to mitigate design-inherent privacy flaws of logically centralised services such as Facebook, Google+ or Twitter. A common approach to build a DOSN is to use a peer-to-peer architecture. While the absence of a single point of data aggregation strikes the most powerful attacker from the list of adversaries, the decentralisation also removes some privacy protection afforded by the central party's intermediation of all communication. As content storage, access right management, retrieval and other administrative tasks of the service become the obligation of the users, it is non-trivial to hide the metadata of objects and information flows, even when the content itself is encrypted. Such metadata is, deliberately or as a side effect, hidden by the provider in a centralised system. In this work, we aim to identify the dangers arising or made more severe from decentralisation, and show how inferences from metadata might invade users' privacy. Furthermore, we discuss general techniques to mitigate or solve the identified issues.