In recent years, network accountability and forensic analysis have become increasingly important, as a means of performing network diagnostics, identifying malicious nodes, enforcing trust management policies, and imposing diverse billing over the Internet. This has lead to a series of work to provide better network support for accountability, and efficient mechanisms to trace packets and information flows through the Internet. In this paper, we make the following contributions. First, we show that network accountability and forensic analysis can be posed generally as data provenance computations and queries over distributed streams. In particular, one can utilize provenance-aware declarative networks with appropriate security extensions to provide a flexible declarative framework for specifying, analyzing and auditing networks. Second, we propose a taxonomy of data provenance along multiple axes, and show that they map naturally to different use cases in networks. Third, we suggest techniques to efficiently compute and store network provenance, and provide an initial performance evaluation on the P2 declarative networking system with modifications to support provenance and authenticated communication.