[AFS3-std] Re: rxgk token: encrypted blob or not

[Andrew Deason]

On Tue, 6 Nov 2012 20:47:43 -0500 (EST)
Benjamin Kaduk <kaduk@MIT.EDU> wrote:

> > says:
> >
> >      If the token is an encrypted blob, it should be encrypted using
> >      the key usage RXGK_SERVER_ENC_TOKEN.
> >
> > should that be a SHOULD ?
> 
> I don't think so.  If we needed 2119-language, I think it would be a
> MUST.  But I'm not sure that we need 2119 language.  We don't use it
> when talking about the other key usages, if I remember correctly.

I wouldn't think so, since the key usage would be up to the application
to define; the whole token is opaque outside of application-specific
usage, so I don't think you'd need to _require_ the use of a specific
usage key. That is, you're not violating rxgk by using something else,
since the only time something would 'break' based on token content
construction would be application-specific stuff. So, you'd be violating
the e.g. rxgk-afs spec if you used the wrong key usage, not the rxgk
spec.

I thought this was more of a guidance to application-specific
specifications to just say "hey, we have a key usage value allocated
specifically for this purpose, so use this one if you can". But if so,
that also doesn't sound like a 2119-language-appropriate place, since
we're not telling implementors what to do. So yes, not using
2119-language makes sense.

-- 
Andrew Deason
adeason@sinenomine.net