Re: Control of domain certs: an example

[Ian G]

Hendrik,
in a lot of the below, you are disagreeing in
the detail of what I wrote, but agreeing in
the big picture.  What I was trying to say
was that looking at whois is a roughly bad
test, like all the others, and ruling it out
isn't going to be a useful addition to the
policy.
On the specific issue of RFC 2821, my point
was to show that RFC 2821 may be a written
down document that you all abide by, but in
the real world it means nothing much more
or better than anything else.  Looking at
the postmaster address is just another 'bad
test' among dozens of others.
Literally, all these tests are bad.  They
are all fakeable in some sense or other.  So
the only thing that a CA can do is do more
and more of the tests, and hope that this
raises the barrier - the cost - enough such
that it isn't worth it to an attacker.
In such a world, Frank's policy really has a
lot of trouble deciding what to rule out and
what to rule in.
iang


Hendrik Weimer wrote:
Ian G <[EMAIL PROTECTED]> writes:

The way identity and control tests fundamentally work
is that they take as many pieces of information as
they can get economically and confirm each one as being
consistent.

This only works if an attacker is unable to fake a consistent picture.

So for identity and control purposes, looking at the whois registry
quite a reasonable idea.

It may be a good idea in some cases, and in other cases it is
not. Therefore I think that looking at whois information does not
provide a generally reliable way to increase the security of the
identification process.

It's when the number of tests sink down to 1 that it
gets a bit silly, but unfortunately, it's very hard to
say where the line is to be drawn.  In some countries,
looking at the whois will be quite authoritive.

Maybe. But it is definitely out of the scope of a browser's CA policy
to address such issues.

Probably the only way to make sure that only the domain owner
receives the mail is to send it to [EMAIL PROTECTED]
How does that make you sure?

RFC 2821.

Speaking as an ordinary domain owner, I don't receive email at
[EMAIL PROTECTED]  The only reason I know this is that there is about
one and only one place on the Internet that tells me my mail is not
being accepted because I don't receive email to postmaster!  I've
never bothered to change this because it isn't obvious - and all the
other domains I know of (across about 6-10 different
administrations) probably don't have it set either.

Well, if you have misconfigured your system, you cannot participate in
such an authentication process. Doesn't seem to be too harsh to me.

However, the other problems (MITM, password sniffing) mentioned in
Juraj Bednar's article still persist. It seems to me that it is a
really bad idea if domain control certs are included in browsers
the same way as certs with proper verification.
They are already there, aren't they?  One would need to
propose that they be taken out.

I have already thrown out some, but I don't have a complete list.

Another problem that came to my mind is how to deal with malicious
CAs. One way would be to make the costs of getting included in the
browsers higher than the possible profits out of malicious use of the
CA. But this is not really desirable and probably only a short-term
solution since the possible profits are increasing over time.
Yes, that whole area is a bit of a deal killer.  The
cost of a CA is in the order of 100k, more and less.
(Anybody got any figures on that?)

The cost to implement and run a CA that meets the requirements set in
the policy draft is probably lower by one order of magnitude.
Hendrik

--
News and views on what matters in finance+crypto:
        http://financialcryptography.com/
_______________________________________________
mozilla-crypto mailing list
mozilla-crypto@mozilla.org
http://mail.mozilla.org/listinfo/mozilla-crypto