Attacking and securing Network Time Protocol
MetadataShow full item record
Network Time Protocol (NTP) is used to synchronize time between computer systems communicating over unreliable, variable-latency, and untrusted network paths. Time is critical for many applications; in particular it is heavily utilized by cryptographic protocols. Despite its importance, the community still lacks visibility into the robustness of the NTP ecosystem itself, the integrity of the timing information transmitted by NTP, and the impact that any error in NTP might have upon the security of other protocols that rely on timing information. In this thesis, we seek to accomplish the following broad goals: 1. Demonstrate that the current design presents a security risk, by showing that network attackers can exploit NTP and then use it to attack other core Internet protocols that rely on time. 2. Improve NTP to make it more robust, and rigorously analyze the security of the improved protocol. 3. Establish formal and precise security requirements that should be satisfied by a network time-synchronization protocol, and prove that these are sufficient for the security of other protocols that rely on time. We take the following approach to achieve our goals incrementally. 1. We begin by (a) scrutinizing NTP's core protocol (RFC 5905) and (b) statically analyzing code of its reference implementation to identify vulnerabilities in protocol design, ambiguities in specifications, and flaws in reference implementations. We then leverage these observations to show several off- and on-path denial-of-service and time-shifting attacks on NTP clients. We then show cache-flushing and cache-sticking attacks on DNS(SEC) that leverage NTP. We quantify the attack surface using Internet measurements, and suggest simple countermeasures that can improve the security of NTP and DNS(SEC). 2. Next we move beyond identifying attacks and leverage ideas from Universal Composability (UC) security framework to develop a cryptographic model for attacks on NTP's datagram protocol. We use this model to prove the security of a new backwards-compatible protocol that correctly synchronizes time in the face of both off- and on-path network attackers. 3. Next, we propose general security notions for network time-synchronization protocols within the UC framework and formulate ideal functionalities that capture a number of prevalent forms of time measurement within existing systems. We show how they can be realized by real-world protocols (including but not limited to NTP), and how they can be used to assert security of time-reliant applications-specifically, cryptographic certificates with revocation and expiration times. Our security framework allows for a clear and modular treatment of the use of time in security-sensitive systems. Our work makes the core NTP protocol and its implementations more robust and secure, thus improving the security of applications and protocols that rely on time.
RightsAttribution-NonCommercial-NoDerivatives 4.0 International