Re: [websec] Decouple HSTS's two orthogonal effects?

Hello Matt and Tom,

I also do not share your view. HSTS has one goal and that is to ensure a
secure and trusted connection to prevent MiM and spoofing attacks. For
this a strict behavior must be enforced. That means only https
connections are allowed and self-signed certificates must be rejected.
Otherwise an attacker uses exactly this to bypass the protection.

Considering Tom's scenarios of self-signed certificates for intranet-site:
- As devdatta writes you should in that case deploy the certs  to your
browser.
- And second, in the long-run there might also be the additional use of
stored certs in DNSSEC (see dane WG, "DNS-based Authentication of Named
Entities"), though I admit that this is still a longer way to go.

BR, Tobias

On 03/24/2011 05:18 AM, Devdatta Akhawe wrote:
> What design principle is this ?
>
> Seems to me that a design principle should be `don't mix orthogonal
> design goals'. If a single goal has multiple effects, so be it. The
> design goal is ensuring that interaction with website is not
> vulnerable to MITM attacks. Since users click through cert warnings,
> showing warnings is not enough (hence the effect 2). And because of
> the way the web platform is designed, just switching half-way to SSL
> doesn't really buy much security, hence effect 1. But the design goal
> is just 1.
>
> If the use cases you mentioned are compelling, then they should be
> handled via a separate mechanism. I personally don't find them too
> compelling. For Tom's intranet example, (or imagine a large
> enterprise), the right thing might be to just add the cert issuer to
> the browser CA store.
>
>
> =devdatta
>
>
> On 23 March 2011 21:22, Tom Ritter <tom@ritter.vg> wrote:
>> I'm new to this list, so apologies if this has been discussed - but I
>> was surprised and dismayed when I learned of the coupling of these two
>> effects.  A very common situation I find myself in is adding an
>> exception for a self-signed certificate or otherwise untrusted
>> certificate, either for a personal machine, development machines, or
>> intranet machines.  The individual cert is verified by fingerprint
>> using out-of-band channels, added, and doesn't change for a while -
>> and we'd like to use HSTS on these boxes.  But based off my simple
>> testing in Chrome - coupling these effects prevents this workflow.
>>
>> -tom
>> _______________________________________________
>> websec mailing list
>> websec@ietf.org
>> https://www.ietf.org/mailman/listinfo/websec
>>
> _______________________________________________
> websec mailing list
> websec@ietf.org
> https://www.ietf.org/mailman/listinfo/websec