Re: [Trans] Threat model outline, attack model

Stephen Kent <kent@bbn.com> Mon, 15 September 2014 19:03 UTC

Return-Path: <kent@bbn.com>
X-Original-To: trans@ietfa.amsl.com
Delivered-To: trans@ietfa.amsl.com
Received: from localhost (ietfa.amsl.com [127.0.0.1]) by ietfa.amsl.com (Postfix) with ESMTP id 357EA1A877A for <trans@ietfa.amsl.com>; Mon, 15 Sep 2014 12:03:34 -0700 (PDT)
X-Virus-Scanned: amavisd-new at amsl.com
X-Spam-Flag: NO
X-Spam-Score: -5.853
X-Spam-Level:
X-Spam-Status: No, score=-5.853 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, RCVD_IN_DNSWL_MED=-2.3, RP_MATCHES_RCVD=-1.652, SPF_PASS=-0.001] autolearn=ham
Received: from mail.ietf.org ([4.31.198.44]) by localhost (ietfa.amsl.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id buf59W4Zo-Hr for <trans@ietfa.amsl.com>; Mon, 15 Sep 2014 12:03:21 -0700 (PDT)
Received: from smtp.bbn.com (smtp.bbn.com [128.33.0.80]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (No client certificate requested) by ietfa.amsl.com (Postfix) with ESMTPS id 4BC101A700D for <trans@ietf.org>; Mon, 15 Sep 2014 11:53:05 -0700 (PDT)
Received: from dommiel.bbn.com ([192.1.122.15]:60313 helo=comsec.home) by smtp.bbn.com with esmtp (Exim 4.77 (FreeBSD)) (envelope-from <kent@bbn.com>) id 1XTbOa-000Lcf-7o for trans@ietf.org; Mon, 15 Sep 2014 14:53:04 -0400
Message-ID: <5417358E.7060907@bbn.com>
Date: Mon, 15 Sep 2014 14:53:02 -0400
From: Stephen Kent <kent@bbn.com>
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:24.0) Gecko/20100101 Thunderbird/24.6.0
MIME-Version: 1.0
To: trans@ietf.org
References: <5411E511.1040605@bbn.com> <BEAB70F21EC44221A49E95FC448603FC@gmail.com>
In-Reply-To: <BEAB70F21EC44221A49E95FC448603FC@gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: http://mailarchive.ietf.org/arch/msg/trans/6yuutVDdKc-kemr2ZmVsWfxBx6A
Subject: Re: [Trans] Threat model outline, attack model
X-BeenThere: trans@ietf.org
X-Mailman-Version: 2.1.15
Precedence: list
List-Id: Public Notary Transparency working group discussion list <trans.ietf.org>
List-Unsubscribe: <https://www.ietf.org/mailman/options/trans>, <mailto:trans-request@ietf.org?subject=unsubscribe>
List-Archive: <http://www.ietf.org/mail-archive/web/trans/>
List-Post: <mailto:trans@ietf.org>
List-Help: <mailto:trans-request@ietf.org?subject=help>
List-Subscribe: <https://www.ietf.org/mailman/listinfo/trans>, <mailto:trans-request@ietf.org?subject=subscribe>
X-List-Received-Date: Mon, 15 Sep 2014 19:03:34 -0000

Katriel,
> I’ve also been thinking about a threat model, though your writing is much clearer than mine!
thanks.
> tl;dr: (a) There’s a lot of text about the details of revocation that I thought was out of scope for CT. (b) A client specification may answer many of the questions you raise about Monitors.
well, revocation, per se, may not be in scope for CT, but unless we can 
specify what actions
can/will be taken in response to the detection afforded by CT, this 
seems like a hard sell.
> On Thursday, 11 September 2014 at 19:08, Stephen Kent wrote:
>> Certificate Transparency (CT) is intended to detect and mitigate...
> Is it intended to mitigate problems? (-04) says so but immediately follows it with “the logs do not themselves prevent misissue, but they ensure that interested parties (particularly those named in certificates) can detect such misissuance”
fair point. I'll revise that text if there is general agreement to do so.
> A fairer introduction might be “CT is intended to allow subjects to detect…"
Well, it's not necessarily subjects, in the PKI sense. 6279-bis suggest 
that a Subject
can ask a Monitor to look for certs that have been issued without the 
authorization
of the Subject. Only if every Subject were to perform Monitoring for 
itself would your
text be accurate. Still, this will be easy to fix.
>> A certificate is characterized as mis-issued if the certificate is issued to an entity that is not authorized to represent the host (web server) named in the certificate Subject field or Subject Alternative Name extension.
> Or doesn’t follow certain regulations, or some other definition (e.g. comes from a root that the CA can but does not use to issue certificates). I think it should be up to individual Monitors to define what they consider to be mis-issued, though it makes sense to give a minimal criterion, and hence that the text should say something like “In the following, a certificate is characterised as mis-issued if… but any Monitor may choose to use an alternative definition, since CT is definition-agnostic."
"regulations?" We need to be precise. Ben replied with a much more 
general characterization
of the term, which I believe is derived from CABF docs. I agree that we 
don't want each
Monitor to make up its own criteria; I don't agree that a "minimal" set 
of criteria is a
good idea. I would like to see uniformity in the criteria, based on 
published standards,
so that everyone knows what will be OK, and what is not OK.
>>   
>>   
>> Certificate mis-issuance may arise in one of several ways. The ways that CT helps detect and remedy mis-issuance depends on the context of the mis-issuance.
> I think the key distinction to be made is between contexts that cause the certificate to be submitted to a non-malicious Log Server, and those that do not. As you say, in the former case a Monitor tracking the targeted Subject will detect the mis-issued certificate and set off the out-of-scope revocation process. In the latter, TLS clients which are not backwards compatible will refuse to accept the certificate, while TLS clients which do not yet enforce CT will act as they do today.
Submission to a log server, or lack thereof, is part of the analysis, as 
noted later. We may disagree
on what is the best way to structure the taxonomy.
> Once the certificate is detected, is the revocation process (e.g. what data the CA requires) in scope for this document?
I included the topic of what info is needed to help justify what is 
covered by an SCT. The
(initial) analysis I performed noted that the serial number is not 
required in all cases,
and it may not suffice in others (a malicious CA). So, to the extent 
that we are discussing
what data needs to be covered by an SCT, this part of the analysis is 
needed, and thus in scope.
>>   
>> 1. If a CA submits the bogus certificate to logs, but these logs are not watched by a Monitor that is tracking the targeted Subject, CT will not mitigate a mis-issuance attack. It is not clear whether every Monitor MUST offer to track every Subject that requests its certificates be monitored. Absent such a guarantee, how do TLS clients and CAs know which set of Monitors will provide “sufficient” coverage. Unless these details are addressed, use of CT does not mitigation mis-issuance even when certificates are logged.
> I understood Monitors to be run by Subjects (or paid by Subjects to do their tracking for them), so I see no reason why they MUST offer to track anybody. It’s log servers for which you need sufficient coverage, though, not Monitors: one of the latter will suffice as long as it watches all the former. I don’t know what the current definition of “the set of all log servers” is (*); for the moment I assume it is defined by Google’s list for Chrome.
I don't recall that the text made it clear who was running Monitors. if 
every Subject had to
run its own Monitor, it might be hard to argue that CT will provide very 
wide coverage in the
near term. Yet, near term, broad coverage is precisely the argument made 
to justify the need for
pre-certs. So, there is an important question here. I could easily 
envision Monitors that are
paid by Subjects to watch over certs. But, if that is part of the model, 
the doc needs to say
so. One Monitor, with a "pay for protection" business model probably 
would NOT suffice. Also,
relying on one company's list to define the set of log servers is not 
the sort approach we follow
in IETF standards, so ...

> I agree it’s important to specify how Monitors can find out the set of log servers. (e.g. it’s Very Bad if I run my own monitor but don’t update its list, since then a mis-issued certificate submitted to a new log server will pass me by.)
>>   
>> 3. If a TLS client is to reject a certificate that lacks an embedded SCT, or is not accompanied by a post-issuance SCT, this behavior needs to be defined in a way that is compatible with incremental deployment. Issuing a warning to a (human) user is probably insufficient, based on experience with warnings displayed for expired certificates, lack of certificate revocation status information, and similar errors that violate RFC 5280 path validation rules.
> What’s the precedent for defining incremental deployment in this type of document? I suppose the wording would be something like “TLS clients wishing to respect incremental deployment MAY choose to accept certificates without embedded SCTs, but…"
The IETF hates flag days; we almost never approve a protocol design that 
relies on all instances of
something to switch to a new something in unison. We're very big on 
backwards compatibility :-).
>>   
>> 4. The targeted Subject might request the parent or the offending CA to revoke the certificate of the non-cooperative CA. However, a request of this sort may be rejected, e.g., because of the potential for significant collateral damage.
> Again, is this in scope for CT? This issue exists in today’s infrastructure just as much (c.f. when people removed DigiNotar from trust lists).
The analysis needs to address what happens in response to detection of 
mis-issuance, so that the
community can decide if the burden of deploying a big system of this 
sort is justified by the
benefits.
>>   
>>   
>>   
>> Absent a protocol (a so-called “gossip” protocol) that enables Monitors to verify that data from logs are consistent, CT does not provide protection against logs that may conspire with, or be victims of, attackers effecting certificate mis-issuance. Even when a gossip protocol is deployed, it is necessary to describe how the CT system will deal with a mis-behaving or compromised log. For example, will there be a mechanism to alert all TLS clients to reject SCTs issued by such a log.
> I believe this is the same question as at (*): who controls the list of trusted logs? One assumes that anybody doing so (e.g. Chrome) would also run a Monitor, and thus detect misissuances and remove said logs from their list the same way they would have removed DigiNotar.
Chrome is a browser; Google is the vendor. If a vendor operates a 
Monitor, it can detect some
types of mis-issuance where knowledge of Subject cert info is not 
needed. For example, if
the cert issuance criteria require a minimum of 1024-bit RSA keys, any 
Monitor can detect a
cert with a 512-bit key, and flag it as "bad." But, if one is trying to 
detect that a cert
for foo.com was issued to the wrong entity, one needs to have 
authoritative info about the
real foo.com cert, or knowledge that foo.com has never requested a cert. 
A browser vendor
would have that info for itself, but not for web sites in general. So, I 
disagree with your
analysis that anyone who runs a log, such as a browser vendor, is in a 
position to detect
the form of mis-issuance that I had assumed was the primary motivation 
for CT. I agree that
any entity running a log could choose to run a Monitor, but by itself, 
one Monitor cannot
necessarily discover a misbehaving log.
> I think a section specifying the actions of TLS clients should answer a lot of these questions. Such a section would say in part that clients must keep lists of CAs and log servers, maintained either by themselves or by a trusted third party, and that the signatures on the cert and the SCT should be from entities in the respective lists. Whoever maintains the list undertakes to remove both bad CAs and bad log servers if they are detected.
Yes, these are topics that should be addressed in the description of TLS 
client behavior.
>>   
>> Monitors also play a critical role in detecting certificate mis-issuance, for Subjects that have requested monitoring of their certificates. Thus Monitors represent another target for adversaries who wish to effect certificate mis-issuance. If a Monitor is compromised by, or is complicit with, an attacker, it will fail to alert a Subject to a mis-issued certificate targeting the Subject. This raises the question of whether a Subject needs to request certificate monitoring from multiple sources to guard against such failures.
> Subjects must choose a trusted Monitor because the Monitor is by definition the thing that watches the logs; they can do this themselves or they can delegate to a third party, in which case they rely on that party not to be compromised. This doesn’t seem particular to CT: if I install a burglar alarm either I have to monitor it, or I have to pay someone else to monitor it and trust them to respond if it goes off, or I have to accept that the alarm might be ignored.
Your analogy is reasonable. Nonetheless, the topic of trust in Monitors 
has a place in the
attack analysis.

Steve