Hi Neil, thanks! This does sound very interesting. Just to clarify, you would document this in a separate doc extending JOSE? We could then mention it from the JWT AT profile, whihc would remain lightweight and implementation independent. thanks V.
On Tue, Mar 26, 2019 at 3:11 AM Neil Madden <neil.mad...@forgerock.com> wrote: > There was a brief discussion at OSW about signing vs encryption for > JWT-based access tokens. I think it was Brian Campbell that pointed out > that you often want authenticated encryption rather than signing, and I > agree with this. > > Currently JOSE only supports authenticated encryption for symmetric > cryptography, which means that the AS and RS must agree a shared secret key > beforehand. (Or else the RS uses the token introspection endpoint and > doesn’t decrypt the token itself). Symmetric cryptography is less than > ideal when the AS and RS are in separate trust boundaries (e.g., different > companies). > > There are actually ways to do *public key* authenticated encryption, but > JOSE doesn’t support them yet. I wrote a 3-part blog series about this > recently [1], but I’ll summarise the tl;dr version here. I think for > JWT-based access tokens that contain identity assertions, this is probably > what you want in most cases as it provides both confidentiality and > authenticity without needing a bulky nested signed-then-encrypted JWT. Is > this something people would be interested in, if I propose a draft? > > Details: > > The basic idea is to introduce one or more variants on ECDH-ES that > provide sender authentication. The most straightforward is to use ECDH-SS — > i.e., Diffie-Hellman between two static key pairs, one for the sender and > one for the recipient, with no ephemeral keys. This provides authenticated > encryption so long as the content encryption method is authenticated (which > they all are in JOSE). But it has a number of security downsides, which I > describe in more detail in the blog. (It might be useful in some IoT > scenarios though). > > The better variant is to instead do ECDH-ESSS. That is, we generate a > random ephemeral key pair and do an agreement with the recipient’s static > public key, just like in ECDH-ES, but then we do another key agreement > between the sender’s static private key and the recipient’s static private > key. We then concatenate the two shared secrets and feed them into > ConcatKDF just like you would for ECDH-ES. This is what NIST SP.800-56A [2] > calls the “one-pass unified model” (section 6.2.1.2). If you squint a bit > then it is also very similar to the “K” one-way pattern in the Noise > protocol framework [3]. > > To spell it out, the process for encrypting a JWE with this new scheme is > as follows: > > Sender has long-term “static” key pair: ssk,_spk (ssk = sender secret key, > etc) > Recipient has long-term static key pair: rsk, rpk > > 1. Sender generates a random ephemeral key pair: esk, epk > 2. Calculate Ze := ecdh(esk, rpk) — just like in ECDH-ES > 3. Calculate Zs := ecdh(ssk, rpk) > 4. Let Z = Ze || Zs where || is concatenation > 5. Run Z through ConcatKDF with PartyUInfo/PartyVInfo just as in ECDH-ES > 6. Encrypt the message using the chosen “enc” encryption method using the > key derived in step 5. > > On its own ECDH-ESSS has some decent security properties (including > authenticated encryption), but it is especially good when you want to > exchange lots of messages with the recipient. If the recipient replies to > your initial message using ECDH-ESSS but using the ephemeral public key you > sent in the first message as if it was your static public key, then what > you get is an interactive handshake very similar to the Noise KK pattern > [4] (squinting quite hard at this point). Both parties can then use the > derived key from step 5 of the second message as a shared session key and > send “direct” encrypted JWEs to each other from that point on. This > provides very strong security properties listed later in the Noise spec, > including forward secrecy and both sender and recipient authentication with > resistance to key compromise impersonation. So beyond its usefulness for > Access Token JWTs, this scheme is a really versatile building block that > you can use for lots of advanced use-cases (e.g., PoP schemes). > > There are even nice formal models of the Noise handshake patterns in > ProVerif [5], but I think they are only valid for the specific details of > how Noise performs key derivation and transcript hashing so they wouldn’t > directly apply to a JOSE version. > > [1] > https://neilmadden.blog/2018/11/14/public-key-authenticated-encryption-and-why-you-want-it-part-i/ > [2] > https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-56Ar3.pdf > [3] https://noiseprotocol.org/noise.html#one-way-handshake-patterns > [4] > https://noiseprotocol.org/noise.html#interactive-handshake-patterns-fundamental > > [5] https://noiseexplorer.com > > Cheers, > > Neil > _______________________________________________ > OAuth mailing list > OAuth@ietf.org > https://www.ietf.org/mailman/listinfo/oauth >
_______________________________________________ OAuth mailing list OAuth@ietf.org https://www.ietf.org/mailman/listinfo/oauth
