It feels like allowing wildcards will help with adoption, because web
domains [that use them] can then reuse their existing certificates for
STARTTLS. Deployment and cert reuse are essential to making progress;
creating key-management challenges for small companies will mean that
traffic continues in the clear.

I agree that 404 should be a soft fail for the reasons you listed, Dan.

/m



--
Mark E. Risher |   [email protected] |  650-253-3123

On Wed, Sep 7, 2016 at 2:46 AM, Daniel Margolis <[email protected]>
wrote:

> If we forbid wildcards, the argument against the two level hostname goes
> away. That is, the only argument against "policy.mta-sts" is that it breaks
> "example.com" wildcards! I think there's a reasonable argument for going
> that direction, as you say, though, hence my question for input here. Do we
> prefer to make deployment and cert reuse easy at slight risk of attacks as
> described?
>
> Regarding 404s (a slight tangent now), I am not sure we do want to treat
> that as revocation. To me, a 404 runs a somewhat higher risk of operational
> mix-ups, and revocation can be had more explicitly but serving a policy
> with a very short max-age (which we could make explicit, i.e. "max-age of
> zero equals revocation").
>
> On Sep 6, 2016 22:45, "Viktor Dukhovni" <[email protected]> wrote:
>
>> On Tue, Sep 06, 2016 at 10:25:55PM +0200, Daniel Margolis wrote:
>>
>> > I may be missing something, but surely anything an attacker can do to a
>> SRV
>> > record they can do to a CNAME or A record for a host itself.
>>
>> Generally, with SRV records, one would expect that the reference
>> identifier for certificate name checks is the target hostname, not
>> the service domain.
>>
>> Note also that the nexthop domain itself (example.com vs.
>> mta-sts.example.com or similar) typically operates multiple https
>> services, many of which would deny (404) the existence of the STS
>> policy resource, and thus clear the policy cache.
>>
>>
>> > So hardcoding the hostname itself doesn't do much against attackers who
>> > can arbitrarily inject DNS records. (The defense here is really in the
>> > SSL cert validation and the assumption that nobody is serving untrusted
>> > content on .well-known/...)
>>
>> It is not that simple.  The target URI has be verifiably the right
>> URI for STS policy lookup, not just some random web service associated
>> with the domain.
>>
>> Furthermore, STS is subject downgrade attacks via MITM redirection
>> to unintended HTTPS servers that have a wildcard cert for the
>> domain, and so the STS spec may need to specifically forbid wildcard
>> certificate matching.
>>
>> A bad alternative is to not support authenticated policy revocation
>> via a 404 response, treating that as a transient lookup error, but
>> this is not a good outcome.  You might need to extend the spec to
>> support serving an explicit policy revocation response.
>>
>> --
>>         Viktor.
>>
>> _______________________________________________
>> Uta mailing list
>> [email protected]
>> https://www.ietf.org/mailman/listinfo/uta
>>
>
> _______________________________________________
> Uta mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/uta
>
>
_______________________________________________
Uta mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/uta

Reply via email to