Hi Rich, Peter, all,

Thanks for the comments. My suggestions were more related to X.509 CRLs, which 
are the ones we standardized in PKIX. 

Personally, I do not think that CRLSet is something that we should really be 
concerned with at IETF. In particular, I think that CRLSet is a bad and fast 
fix to a real problem that (a) is available from one single vendor, (b) 
completely ignores the golden rule about PKIs by substituting the real source 
of trust (the CA) with an application-level (chrome) decision (not even 
user-driven!), and (c) is an ad-hoc not-standardized solution.

For client side certificates, I think that checking the revocation status in 
that environment carries even more privacy concerns. This might be true 
especially when the certificate to be checked is issued by a CA that is not 
connected to the services offered by the server.

For example, if you are checking the revocation from a 3rd party CA (e.g. for 
granting access to a building), checking the revocation status of that 
certificate might reveal the user's position or even its activity (e.g. logging 
into her e-mail or accessing a building) to the CA's revocation servers - they 
have no business knowing that information and that info might even be disclosed 
to passive attackers that monitor the traffic. Another example, can be when 
using client-side certificates for WiFi roaming.

Therefore I think that also for client-side certificates the privacy issue is 
quite relevant.

Just my two cents...

Cheers,
Max


On Dec 1, 2014, at 10:24 PM, Salz, Rich <[email protected]> wrote:

>>>    Certificate Revocation Lists (CRLs) provide the most supported
>>>    revocationinformation distribution mechanism. Although updates to
>>>    the CRLformat (e.g., partitioned CRLs, delta CRLs) have been defined
>>>    toaddress scalability issues, they are rarely used in favor of more
>>>    compact formats. However, it is important to notice that the use of
>>>    CRLs might provide better privacy-preserving properties than other
>>>    protocols (e.g., OCSP), especially when considering the validation of
>>>    client-side certificate.
> 
> Are you using CRL as a generic term -- to include tihngs like CRLset, for 
> example?  If so, then this makes sense. But if so, then I think the wording 
> needs to be tweaked, since portioning and delta CRL's are only for the 
> classic X.509 CRL structures.But then if it does include generic, then Chrome 
> support for CRLSet is a pretty widely deployed use on the Web.  The private 
> point raised is important. Not sure if it's important for client certificates 
> (since you've already decided to give the server your cert); I do think it is 
> VERY important for server-side certificates, where not letting national-scale 
> adversaries know where you might be connecting is (well, er) key.
> 
> --  
> Principal Security Engineer, Akamai Technologies
> IM: [email protected] Twitter: RichSalz
> 

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

Reply via email to