Re: [radext] AD review draft-ietf-radext-tls-psk-08

[Alan DeKok <[e-mail]>]

On Feb 29, 2024, at 3:43 PM, Paul Wouters <paul.wouters=[e-mail]> wrote:
> Thanks for this very clear document, and especially for re-using other RFCs security considerations where possible!

  Thanks.  Cut & paste is often easier than thinking :)

> I think it's good to go, except for some minor comments. One question I had is whether TLS-PWD was considered,
> in light of knowing administrators will use weak PSKs. I guess it would complicate things a bit, as TLS-PWD is only
> available for TLS 1.3.

  I'm happy to allow for TLS-PWD?  But I don't know much about it.  And on quick inspection, I can't find references to it for OpenSSL.

> Comments:
> 
>         PSKs MUST be uniformly random.
> 
> Of course, if they are "uniformly" random, they are not truly random :) Maybe remove "uniformly" ?

  Sure.

>         In order to guide Implementers [...] Perl script
> 
> It made me smile to believe people think perl is still used and generally available :P It also uses a
> Crypt:URandom() module which I would not be sure if it was cryptographically secure, or just a thin
> wrapper over /dev/urandom, which on linux is less secure in certain circumstances.
> 
> How about python? :)

  I'll use that.

>         Implementations SHOULD NOT provide user interfaces which
>         allow both PSKs and shared secrets to be entered at the same
>         time. There is too much of a temptation for administrators to
>         enter the same value in both fields, which would violate the
>         limitations given above.
> 
> I think you should ignore the UI part and enforce things harder:
> 
>         Implementations MUST reject configuration attempts that try to
>          use the same value for PSK and shared secret.

  I'll add that text.  I think it's still useful to comment on the UI part, tho.  I'm thinking of an administration interface which has a "shared secret' entry box for RADIUS/UDP, and a checkbox for TLS.  If the TLS checkbox is clicked, then the "shared secret" field should _go away_ and be replaced by a PSK entry box.

  i.e. where possible, make it difficult (or impossible) to enter both values at the same time.

> Nits:
> 
>         This output may be 256, 384, or 512 bits in length. Never the less,
> 
> Reading a bunch of numbers and then seeing "never the less" made me think of
> a less than XXX amount syntax. Would using "nevertheless" be better?

  Yes.

>         There are few, if any, use-cases for
> 
> How about: There are no known use-cases for

  Yes.

> implements should allow -> implementers should allow
> 
> Remove "Returning to the subject of IP address lookups,"

  Done, and new draft submitted shortly.

  Alan DeKok.