Re: [websec] #42: STS exception for CRL fetching

[Alexey Melnikov <[e-mail]>]

On 16 Jun 2012, at 20:19, Tobias Gondrom <[e-mail]> wrote:

> <hat="individual">

Same here in this message.

> +1. I would agree with Jeff on that.
> (Mostly because I see no other good solution.)
> Any other comments/ideas on this one?

I tend to think that "don't use this hammer if it doesn't suit you" is the best answer here. Exceptions are bad (hard to handle properly).

> 
> Best regards, Tobias
> 
> On 15/06/12 02:06, =JeffH wrote:
>> > This is about fetching CRLs from a domain that happens to be the same as
>> > that of a website.
>> >
>> > Obviously you can't get a CRL or an OCSP response over HTTPS. Jeff's
>> > response was that they should use a different domain name for the CRLs (if
>> > they want to deploy HSTS)
>> >
>> > Obviously, it's too late to change AIA or CDP in existing certificates. But
>> > I think it goes deeper. HSTS affects what the browser is doing. Different
>> > resources from the same domain should all be protected by TLS. But we don't
>> > expect this to affect things that are outside the browser, like email or
>> > system updates. IMO the fetching of CRLs or OCSP responses is not part of
>> > the browsing, but part of the HTTPS handshake. The fact that some browsers
>> > implement both is besides the point. Internet Explorer uses an OS library to
>> > do the TLS handshake, including any checking of revocation. In fact getting
>> > the CRL fetch function to apply the HSTS policy would require extra effort
>> > from the browser implementer.
>> >
>> > I think we should simply say that HSTS does not apply to non-content.
>> > Fetching CRLs or browser software updates is not content, and HSTS should
>> > not apply to it.
>> 
>> Unfortunately, we would then have to define what comprises "content", and that's a huge rathole I think we should unequivocally avoid.
>> 
>> I've taken a (ad-hoc, limited, unscientific) look at the AIA and CDP in some existing certificates (from major CAs), and all of the ones I checked used a dedicated subdomain for their OCSP responder, e.g...
>> 
>>  evsecure-crl.verisign.com
>>  evsecure-ocsp.verisign.com
>>  SVRSecure-G3-crl.verisign.com
>>  ocsp.verisign.com
>>  ocsp.thawte.com
>>  crl.thawte.com
>> 
>> 
>> So in order not to cause themselves and their customers problems, CAs simply shouldn't issue HSTS policy with "includsubdomains" from their top-level domain name, and they will be fine. This is already discussed in S 10.3 of the HSTS draft.
>> 
>> HTH,
>> 
>> =JeffH