On 09 Aug 2014, at 04:42, Randy Bush <[email protected]> wrote:

>>>> The question was about why, in this effort, we are using 3779
>>>> validation rules
>>> because we understand how they work formally from considerable
>>> experience with PKIs.  they are deployed and working today.
>> Well ok. Where else are the 3779 rules used?
> 
> the validation for 3779 is essentially the same as for X.509, a rigid
> hierarchy.  

Personally I am not challenging this general approach.

The *one* thing I (and I believe we..) challenge is whether the an overclaimed 
resource should invalidate a complete certificate, instead of invalidating just 
the resources at hand, but allowing the remainder.


>>> the wg is considering an other validation process.  it is still a bit
>>> wobbly, and the need for it is poorly motivated.  if you remember, i
>>> advocated looking at this as a work item, and came up with the one
>>> rather subtle motivation for it.
>> I’m not seeing the relevance of this history lesson to the question
>> that was asked. How does this relate to the technical reasons of why
>> we are using 3779 rules?
> 
> because they are working.  you are proposing a change to a deployed
> rfc.  what is the motivation for the change?  be technically explicit.

Resource certificates are always evaluated in context. Under RFC3779 we accept 
the resources listed for the trust anchor as-is, but for everything down the 
chain we insist that *everything* is also held by the parent. I.e. we know 
exactly which resources we are willing to accept. So we do not believe the 
listed resources at face value.

I understand that conceptually this is an attractive simple model. However I 
believe that are reasons why certificates can end up overclaiming, and 
*importantly* this is not always something that a CA has full control over.

Essentially I see two causes, that can be caused by various reasons:
1 = the parent certificate is shrunk and published before the, now 
overclaiming, certificate is shrunk and re-published
2 = the certificate is grown and republished, before the grown parent 
certificate is republished, so it looks overclaiming

Some possible causes, there may be more:

The following two we may be able to mitigate technically, because we are 
dealing with co-operating parties:
  @1 There is a mis-timing in certificate shrinking in a transfer between 
co-operating parties.
  @2 There is a mis-timing in the parent publishing a cert with extended 
resource, and issuing it to the child, and the child using those resources in 
turn.

But this is the one that has me scared. It is not the issuing CA being sloppy, 
it’s a problem between them and *their* parent:
@1 The *grand*-parent shrinks the *parent* certificate without the parent 
knowing about this, and some of these resources appear on the, now 
overclaiming, child certificate. The grand-parent and parent are not involved 
in an active transfer process. They may not agree that the resource in question 
should be removed, or the grand-parent may have shrunk these resources in error.

By rejecting the overclaiming child certificate completely I think we are being 
too harsh, or pedantic even. We know better, so we are just not going to trust 
this one. But.. there is a very real possibility that we actually *do* know 
better than the issuing CA at this point. So, what exactly is the problem with 
accepting the remaining resources? i.e. the intersection between the parent 
certificate’s resources and this certificate. We know the context, this 
evaluation is very easy and well-defined. If the CA really intended that the 
remaining resources should not be tied to the certificate, they would have 
revoked. If the grand-parent really intended that those resources should not be 
certified anymore, they would have removed them as well.


> the point of the history lesson of the rirs blocking the definition and
> documentation of transfer is that it seems to be the only serious reason
> for the change those very same rirs are proposing.  so, if you want your
> proposal to change rfced running code to be taken seriously, you had
> best document it technically, not just throw perjoratives and hyperbole.

I think a formal specification for transfers can mitigate the problem of 
transfers between co-operative CAs, but this can not mitigate errors.

But, more importantly our transfer policies are set in the address policy 
working group, and our implementation is based on the directives that we get 
from our community there. Other RIRs have similar, but slightly different 
policies. The ‘rules’ regarding these transfers can be changed at any time. As 
an RIR employee I feel a lot of reluctance against engaging in a technical 
definition of a transfer, as it can be perceived to compete with our community 
process. 


> i just want clear and well-understood technical reasons for a
> change to deployed running documented code that seems to be working.

I hope that the third case above (CA unaware that its own certificate has 
shrunk) made my motivations more clear.

I understand that changing running code is a pain, but I don’t think it’s 
impossible. We have running code that warns about overclaiming, but accepts 
remaining.

Tim



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

Reply via email to