[guardian-dev] WOT and Authentication Research

Patrick Baxter patch at cs.ucsb.edu
Tue Dec 4 00:55:40 CET 2012


Yup, Sovereign Keys is awesome. I hadn't looked it up since thinking
more about the importance of having a single mapping but on a quick
re-read I understand it as follows:

Sovereign keys has a very strict requirement for changing this mapping
as domain names should. ie. Only a key revocation can change the
mapping, and thus, a simple append only structure works great. If
domain operator loses control for their domain name, they can revoke
their key. This works as long as each party is trusted to never lose
access to both their key AND revoke cert.

You are also relying on these keyservers to establish validity of
domain names through possibly DNSSEC and existing CAs. I'm OK with
this and would gladly use it over the CA system and this would be an
excellent start. I think you might be able to do more, but its up for
debate.

DNS is unique compared to common user-user authentications in that it
is one-way authentication. A user validates a key for a TLS
connection, but the TLS connection doesn't care what identity connects
to it.

Using a WOT between DNS owners and user keys could be used to find
valid paths, but since the keyservers in this case already plan to
maintain a fairly exhaustive list of domain names, it might not be
that valuable. Particularly, because users don't have any meaningful
way to capture strong validity of a Domain owners key ( maybe if they
know the admin who captures verification with people he knows through
ZTRP...)

Although you could allow users to publish a special type of signature
for every site they visit. It wouldn't establish validity, but you
could see if their is a consenus among what users are seeing when they
visit a website. Now you have a consensus on network perspectives.
This could be used to prune any initial bad seeds in the mapping. If
Google is seen all over the world as having cert X but the newly
installed keyservers say its Y, I'm likely to believe the massive
network perspective.

One question I have about Sovereign Keys:
1) It might be an easy assumption to make that a small subset of
domain admins might lose access to their revoke certificate and
private key. How can they appeal this? If this append only structure
is *absolutely* immutable in its mappings without a revocation cert,
then they shouldn't be able to use DNSSEC or a CA to re-establish a
mapping. Some resolution protocol is needed to not lock a domain out
of all future TLS communication. The doc suggests just using extreme
caution in preventing this. Storing revoke certs in key-escrow and
things like this. Maybe this is a good/necessary security trade off?

I'd like to see this enforcement of a single mapping applicable to
each domain in a way that makes sense, I tend to agree that a WOT
wouldn't make much sense for DNS, but I think it adds a lot of value
for mediums with user-to-user interaction.

So, ideally, you have a trusted set of keyservers to which you have
somewhat sercurly bootstrapped (and pinned) keys to talk with. You
trust them to be responsible in establishing single mappings in a way
that balances the ability to change mappings with the ability to
prevent malicious or bad mappings from being populated. You rely on a
WOT built through user interaction to provide independent validity
beyond "mapping exists on keyserver".

A question I'm not sure about here though is: For a keyserver that is
establishing mappings, what happens when someone in the mapping
publishes a bad signature (ie a signature to a mapping that is not
considered the standard)? Maybe keep track of these signatures, but
don't present them for validation. This way you can have a reference
count of what the communitiy is seeing as the mapping in real-ilfe.
Again, this is the idea of consensus around a mapping.

So I guess the point is that any project like convergence.io/tack.io
and Soverign Keys that forces domain name owners to establish their
own keys is a huge step in the right direction. I like the Soveirgn
Keys approach of establishing a set of keyservers that check each
other. Convergence captures network perspectives. Combine them? Also,
I really like the idea of failing back to an .onion address of the
public key in the presence of a MITM.

Also, I just realized that when I've been trying to separate the
different characteristics of email authentication vs domain
authentication and referring to them as 'communication domains' for
lack of a better reason. This is confusing. All the accounts are
addressable by domain names. The difference is that an email account
is a personal account separately authenticated that is just addressed
with and @domain.tld. Its uniqueness is guaranteed by the provider and
not the domain registration system. Overall, the separation of
mappings is based on the communication medium that needs
authentication.

On Sun, Dec 2, 2012 at 10:25 PM, Miron (devrandom)
<c1.android at niftybox.net> wrote:
> Hi Patrick,
>
> Have you seen EFF's Sovereign Keys project?  It attempts to establish a distributed single-mapping database of cert <-> domain.
>
> Also see the schemes in https://en.bitcoin.it/wiki/BIP_0015, altough they create new handles rather than try to capture existing ones.



More information about the Gnupg-users mailing list