W dniu 08.07.2019 o 19:20, Wessels, Duane pisze:>> On Jul 8, 2019, at
9:20 AM, Joe Abley <[email protected]> wrote:
>>
>>
>> As far as this particular idea goes, I mentioned before what had given me 
>> pause: we're talking about taking a protocol where every RRSet in a zone to 
>> date has been public and is made available in DNS responses. Any server that 
>> doesn't implement this new mechanism would presumably treat the new covert 
>> RRTypes as they would any other unknown/opaque type and make the data public.
>>
>> There is hence an operational risk that data will leak (e.g. by 
>> configuration changes, software downgrades that are pragmatic necessities, 
>> side systems that publish zone data in ways other than the DNS).
>>
>> By keeping data that is already exchanged over a (manual) out-of-band 
>> channel separate, and not packaging them up with zone data, the existing 
>> segregation of private vs. public is preserved and the task is simply to 
>> automate a process that is currently manual.
> 
> I share this concern raised by Joe.  I agree that it can be useful to 
> exchange configuration/provisioning data this way, but leaks seem almost 
> inevitable as proposed.
We tried to include as many safeguards as possible, but I don't think
we'd ever be able to avoid footie-shootie. I'd just add a section in
"Security consideration" with summary of the issues stated above.

> I'll probably regret this, but what about a COVERT class, instead type RR 
> type?But then a) we'll end up with a dual-class zones b) nobody would
implement it, as I don't think anything but BIND supports classes other
than IN (correct me if I'm wrong?)

Witold

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

Reply via email to