Re: [xmpp] [POSH] What's the point of using JWKs in POSH?

[Thijs Alkemade <[e-mail]>]

[Resending this message from the right address. Sorry for the duplicate copy, Matt.]

On 5 jun. 2014, at 00:46, Matt Miller <[e-mail]> wrote:

> Signed PGP part
> On 6/4/14, 4:17 PM, Matt Miller wrote:
>> [ Forwarding to the [e-mail] mailing list on behalf of Thjis
>> Alkemade ]
>> 
>> Hello,
>> 
>> Today, I've spent some time on trying to implement POSH-checking
>> for xmpp.net. My implementation aimed to do two things: doing the
>> validation as described and showing someone how they could set up
>> their .well-known file by converting their X509 certificates to
>> JSON Web Keys.
>> 
>> The latter part was a lot more work than the former and made me
>> wonder why it is defined the way it is.
>> 
>> From draft-ietf-xmpp-posh:
>> 
>> Each included JWK object MUST possess the following information:
>> 
>> o  The "kty" field set to the appropriate key type used for TLS
>> connections (e.g., "RSA" for a certificate using an RSA key).
>> 
>> o  The required public parameters for the key type (e.g., "n" and
>> "e" for a certificate using an RSA key).
>> 
>> o  The "x5t" field set to the certificate thumbprint, as described
>> in section 3.6 of [JOSE-JWK].
>> 
>> Yet the data that is required in the first and second bullet is
>> never used. It doesn't specify if and how clients should verify
>> it. Verification only uses the x5t field and optionally x5c.
>> 
>> There are good arguments for "pinning" just the public key.
>> draft-ietf-websec-key-pinning only uses the SPKI field, DANE can
>> use either the full cert or its SPKI field (and optionally hashed).
>> But the way it is specified here won't allow that: the x5t field
>> always needs to be present and clients should verify it.
>> 
>> So the public parameters of the key are useless here, but they make
>> a key >10x as large is they have to be. Generating them is also not
>> as easy: most certificate viewers show a SHA1 fingerprint and it's
>> really easy to do with the openssl cli tool, but extracting n and e
>> and base64-encoding them is a lot more work. I wouldn't even know
>> what to do for ECDSA keys.
>> 
>> Are there any interoperability reasons for using JWKs that I'm not
>> aware of? Couldn't it just use a list of SHA1 hashes?
>> 
>> Best regards, Thijs
> 
> As I stated in the previous venue ([e-mail]), us authors were
> originally working to support various other use-cases, such as
> browserid.  However, no one is arguing to actually support those other
> use-cases, so the desire to use JWKs is much less.
> 
> My co-author and I discussed this today, and think what would be best
> is to switch from using a JWK-set to (roughly) your suggestion of a
> list of hashes.  It would allow us to stay with a single syntax for
> both the "by-reference" and "by-value" documents, as well as provide a
> simple point of extension (if that is ever necessary).
> 
> An example:
> 
> {
>    "fingerprints": [
>        {
>            "sha-1": "ij39Ctarv+LwSw45qoqaZl7venM=",
>            "sha-256": "WhEr4Lpv2L5pv769aRj9rrm4G6MNNCfQlre23Gol/eA="
>        },
>        {
>            "sha-1": "JWow1EHNSbNyRfhQchi22bjurr0=",
>            "sha-256": "K52a2gXfrjchMLYwv16QyOtv5bkKRE6rnR30hY3JM8k="
>        }
>    ],
>    "expires": 604800
> }
> 
> Each "fingerprint" is a JSON object, where the key is the hash
> algorithm and the value is the base64 encoding of hashing the
> DER-encoded certificate with the given algorithm.  I do think that
> algorithm agility is necessary, which means something more than a
> simple array in my opinion.  Generating this should be very simple; I
> could kludge this together on the command-line pretty quickly
> 
> If the WG is ok with this, we can get a new revision of
> draft-ietf-xmpp-posh out relatively soon (by next week).
> 

This looks good to me!

Thijs