Sam,
Thanks for picking this up. Unlike my other two concerns with this draft,
I think we have a longer discussion ahead of us on this one.
Your summary of my concern is on the mark:
David's main concern is that bad security will get registered.
I understand the response to be two-fold:
1) It doesn't matter; the controlling registry for crypto algorithm usage
is the protocol-specific registry, not the key table database registry.
2) The guidance for the Expert Reviewer will be very difficult to write.
I'm not convinced by either of these, sorry.
I'm concerned that the two-registry subtlety in 1) will be lost on
implementers, especially because (as mentioned in the IESG thread), this
key table database is likely to see use beyond routing protocols. Among
other things, it's being proposed as a general mechanism for keying RSVP,
not just RSVP-TE (I'm one of the co-chairs of the tsvwg WG that's
responsible for RSVP). That key table databases registry is also likely
to be a place that designers of new protocols look to figure out what to
use for security.
As for cleartext passwords:
Also, some routing protocols are protected by cleartext passwords sent
over the network. We want to be able to manage that, so we will be
registering plaintext password in these registries.
I don't think anyone will come up with anything worse than that.
I read the first sentence in Section 2 of this draft as excluding
cleartext passwords:
The database is characterized as a table, where each row represents
a single long-lived symmetric cryptographic key.
If someone wants to argue that a cleartext password is a "long-lived
symmetric cryptographic key", I'll go break out the popcorn and watch
w/amusement :-).
Seriously, if the intention is to include cleartext passwords, then I
think some more rewriting is in order, and I would suggest checking
directly with the Security ADs before going there.
As for 2), the fact that it will be difficult (with which I agree)
doesn't imply that it isn't necessary or shouldn't be done. IMHO, we
really should be setting a bar that says that this sort of IETF
imprimatur of approval of a crypto algorithm actually means something.
I appreciate that FCFS provides an easier path forward, however I'm
reminded by analogy of something I learned from my grad school
software engineering professor:
I can make the code run arbitrarily fast ...
... if it doesn't have to be correct.
I'm rather uncomfortable with this use of process expediency as a
rationale for avoiding a technical concern.
This may ultimately be an issue that the IESG needs to sort out, as the
level of security for IETF protocols and concerns about "vanity crypto"
extend well beyond the karp WG, but discussion ought to start here.
Thanks,
--David
-----Original Message-----
From: Sam Hartman [mailto:hartmans-ietfc(_at_)mit(_dot_)edu]
Sent: Wednesday, August 14, 2013 3:19 PM
To: Black, David
Cc: housley(_at_)vigilsec(_dot_)com; tim(_dot_)polk(_at_)nist(_dot_)gov;
Dacheng Zhang
(zhangdacheng(_at_)huawei(_dot_)com); General Area Review Team
(gen-art(_at_)ietf(_dot_)org);
karp(_at_)ietf(_dot_)org; ietf(_at_)ietf(_dot_)org
Subject: Re: [karp] IANA policy for draft-ietf-karp-crypto-key-table-08
David, as we mentioned in the IESG thread, we seem to have dropped the
response to your comments about IANA actions.
WG:
From the genart review:
[9] I suggest Expert Review for the new IANA registries, not just
First Come First Served, so that someone with a security "clue" can
check that the proposed registrations are reasonable.
Stephen has filed a related DISCUSS position. He's confused why we need
a registry for KDFs or algorithms.
He argues that the protocols should already have such a registry. He
argues that it would be non-sensical to register a value in this
registry but not the protocol registry.
In a somewhat related discussion, multiple people have asked what the
scope of this document is. Are we defining something for routing
protocols? Any security protocol in the world? Something in-between?
IU'm going to make two responses:
1)
I think FCFS is not harmful for these registries.
David's main concern is that bad security will get registered.
I'll point out that these registries are not about what security you can
use with a routing protocol, but about what security you can configure
from a management standpoint.
Registering rot13 or similarly questionable security here wouldn't mean
I could use it with a routing protocol, only that I could ask a system
to do so. If ROT13 was not actually in the security-specific registries
for the protocols in question there'd be no way to send a
rot13-transformed message.
I think people wanting to use bad security in routing protocols will
focus on specifying how to use the security for the protocols, and
that's the appropriate place to do any gateway review.
Yeah, I guess with FCFS it's possible someone could register here and
then later realize they cannot get their md4 security approved in the
actual protocol registry document.
That might be confusing but doesn't seem very harmful.
Also, some routing protocols are protected by cleartext passwords sent
over the network. We want to be able to manage that, so we will be
registering plaintext password in these registries.
I don't think anyone will come up with anything worse than that.
Finally, I think a lot of us have begun to question the value in
security review for codepoint assignment. Some security WGs care a lot;
some don't seem to care much at all.
2) Why I prefer FCFS or at least would object strongly to expert
review.
If we're going to say we want expert review I want us to give expert
instructions at least good enough that I believe I could answer the
question of what registrations to approve if I were appointed the
expert.
I don't think we could come to consensus on those instructions very
easily.
In particular, I think it would be challenging for us to describe what
security protocols the protocol registry applies to and which ones it
does not.
I'm totally fine publishing this document knowing it will be used for
routing protocols but not knowing what beyond routing protocols it will
be used for.
If we do that FCFS makes a lot of sense.
So, my recommendation is that we keep our current registration policy.