On Fri, October 12, 2012 1:57 am, Nicholas Weaver wrote:
>
>  On Oct 12, 2012, at 9:45 AM, Tony Finch wrote:
>
> > Marc Lampo <[email protected]> wrote:
> >
> >> Before the draft was adopted as RFC I asked how to cope with proxies.
> >
> > Why are proxies a problem for DANE in particular, rather than TLS in
> > general?
>
>  Because how the proxies work is they add an additional root key into the
>  (victim) browser during some form of configuration.   DANE makes the
>  injected root key no longer valid.
>
>  Of course, certificate pinning and other techniques has the same problem,
>  and IMO, killing the idea of TLS proxies is a feature, so DANE's general
>  incompatibility with TLS proxies is a feature, not a bug.

Public Key Pinning, as implemented in Chromium, only evaluates pins
against publicly trusted root certificates (that is, roots which
participate in the various platforms' root certificate programs).

When the certificate chain that was validated in the client does not
terminate in a "publicy trusted" anchor, pinning is not enforced. If you
have sufficient privileges to install a local trust anchor on a users
machine - whether by means technical or policy - then such local settings
are respected by Chromium.

For users who are not operating in such scenarios, public key pinning
offers a robust defense against future Diginotar-like misissuance. For
users who do operate in such environments, it is presumed (albeit
unlikely) that the (enterprise) endpoint terminating the TLS connection is
performing these checks. In the reality that many such solutions are not,
proposals such as
http://tools.ietf.org/html/draft-mcgrew-tls-proxy-server-01 or
http://tools.ietf.org/html/draft-rpeon-httpbis-exproxy-00 would allow
clients to implement such security policies.

Such scenarios are not strictly relegated to "enterprise" scenarios, where
there are effectively three parties participating in the end-to-end
session (user, enterprise, site). It's unfortunately common that a number
of antivirus vendors have begun shipping local "TLS proxies" as part of
their traffic inspection anti-virus/user-protection platforms. These are
solutions that the user directly opts in to (by choice of installing the
antivirus and configuring it to do so). Thus, I'm not sure that RFC 1984
really comes into play here - this is the user wanting to configure a
local policy that can be respected regardless of the applications running
on their machine.

For DANE, presumably solutions would use some form of DNSSEC rewriting,
with DLV, to accomplish the equivalent. That, or individual applications
would have explicit options to configure such scenarios. Of course,
without some form of central trust mechanism - as exists on Windows and OS
X, and as does not exist on most Linux-based distros - it's likely that
many applications will end up trying to roll their own settings, and
getting them wrong or being otherwise unmaintainable at scale. The same as
it is today with application support for explicit (not TLS) proxies such
as SOCKS or HTTP.

_______________________________________________
dane mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dane

Reply via email to