However despite its importance, despite its longevity, and despite the wealth of Dolev-Yao-style security proofs, no reduction based security proof has been published until now. This has two reasons: (1) All widely accepted formal models either deal with two-party protocols, or group key agreement protocols (where all entities have the same role), but not with 3-party protocols where each party has a different role. (2) Kerberos uses timestamps and nonces, and formal security models for timestamps are not well understood up to now.
As a step towards a full security proof of Kerberos, we target problem (1) here: We propose a variant of the Kerberos protocol, where nonces are used instead of timestamps. This requires one additional protocol message, but enables a proof in the standard Bellare-Rogaway (BR) model. The key setup and the roles of the different parties are identical to the original Kerberos protocol.
For our proof, we only require that the authenticated encryption and the message authentication code (MAC) schemes are secure. Under these assumptions we show that the probability that a client or server process oracle accepts maliciously, and the advantage of an adversary trying to distinguish a real Kerberos session key from a random value, are both negligible.
One main idea in the proof is to model the Kerberos server a a public oracle, so that we do not have to consider the security of the connection client--Kerberos. This idea is only applicable to the communication pattern adapted by Kerberos, and not to other 3-party patterns (e.g. EAP protocols).Category / Keywords: cryptographic protocols / Kerberos, Nonce, AKE, authenticated key exchange Date: received 29 Feb 2016 Contact author: joerg schwenk at rub de Available format(s): PDF | BibTeX Citation Version: 20160229:213900 (All versions of this report) Short URL: ia.cr/2016/219 Discussion forum: Show discussion | Start new discussion