Re: [dns-privacy] Working Group Last Call for draft-ietf-dprive-bcp-op

[Sara Dickinson <[e-mail]>]

> On 16 Aug 2019, at 15:42, Stephen Farrell <[e-mail]> wrote:
> 
> Hiya,
> 
> I had an hour so gave this a read. I like the draft
> but think it's not ready quite yet. Depending on the
> answer to my question at #1 below, that could all be
> fixed before or just after the end of the WGLC though
> but it might take a bit longer as there's more than a
> teeny bit of work needed I'd say.
> 
> Substantive comments and nits below.

Hi Stephen,

Thanks so much for the detailed review!

> 
> Cheers,
> S.
> 
> 
> #1: I wonder would we be better to wait until some real
> operators have published DPPPS documents before making
> this an RFC? I'd expect that there'll be some changes
> when/if that happens (I'm assuming it's not yet happened
> but didn't check.) If some such operator has said they'd
> do that, but want to wait 'till there's an RFC, then
> that's a good enough reason to go ahead now. If we just
> don't know, then going ahead and recording this in an
> RFC is probably ok, but risks this not being followed
> much or at all.

I admit we have a chicken and egg problem. Cloudflare, Google, Quad9 and several other operators have all published ‘privacy statements’ but in very differing formats (see
https://dnsprivacy.org/wiki/display/DP/Comparison+of+policy+and+privacy+statements,
https://dnsprivacy.org/wiki/display/DP/DNS+Privacy+Public+Resolvers and
https://dnsprivacy.org/wiki/display/DP/DNS+Privacy+Test+Servers).

I don’t really see any motivation for them to update these until there is an RFC to follow and also I fully expect to bis this document in the foreseeable future as the landscape evolves. The consensus at IETF105 seemed to be to get on and publish this in preference to waiting some unknown time for operator movement.

> 
> #2: I think a sample DPPPS document should at least
> exist and be referenced from this or included as an
> appendix. Is there one somewhere? If not, then I think
> we really ought fix that before publication.

I like that idea. I think an example in the Appendix could work well and could even be based on one of the existing policies :-)

> 
> #3: Perhaps we should ask the pearg and httpbis folks if
> there's any more/better guidance to add to 5.1.3.2?
> IIRC a pearg presentation at IETF105 seems to have
> indicated that some h2 session management stuff might
> also assist with traffic analysis. Checking that out
> before publication would seem worthwhile in case there's
> more to usefully say here.

I’ve mentioned this off-list to a couple fo PEARG folks, but I think sending mail to those groups makes sense too.

> 
> #4: 5.1.4: the lowercase "must" for DNSSEC validation is
> odd. I'd like a "MUST" but perhaps that's not feasible
> now?

There was a discussion early in the life of the document about how prescriptive it could/should be because it was a BCP and the approach taken after that was not to use normative language throughout the document and instead just to use the 3 categories defined at the end of section 5 with a SHOULD there:

“This document does not specify policy only best practice, however for
   DNS Privacy services to be considered compliant with these best
   practice guidelines they SHOULD implement (where appropriate) all:

   o  Threat mitigations to be minimally compliant

   o  Optimizations to be moderately compliant

   o  Additional options to be maximally compliant"

DNSSEC is classed as a Threat Mitigation so it falls into the top category. If folks want to revisit this general approach then now is the time :-)

> 
> #5: given that 1% of zones are signed, and that DNSSEC
> answers are big, do we think padding is sufficient to
> prevent an observer from knowing when a qname is in the
> 1%? If so, great! (But really?;-) If not, is that a
> problem worth mentioning? And is there anything one could
> do about it?

It is a good question, but I’m not sure this document should do anything other than recommend following the guidelines in RFC8467? Also, I think that only applies if the client is directly requesting the DNSSEC records from the resolver so it can do validation itself (which virtually none do today…..), otherwise the resolver just sends the same answer with the AD set appropriately for both signed and unsigned responses.

FYI: My understanding is that an update to RFC8467 is highly likely given the recent research on website fingerprinting presented in PEARG at IETF 105:
https://datatracker.ietf.org/meeting/105/materials/slides-105-pearg-encrypted-dns-privacy-a-traffic-analysis-perspective

> 
> #6: I don't think mention of "consent" is a good thing
> to positively recommend operators do as there is (afaik)
> no good way to really do that. The idea that whatever is
> behind a DNS stub client can meaningfully "consent" to
> something is really just broken IMO.  (Mentioning that
> one cannot assume consent is fine, but saying to go get
> it, when that makes no sense, isn't fine IMO.) Same
> applies to 6.1.1 item 7.

For the subset of operators with ‘contracts' with users e.g. ISPs it seems entirely possible to obtain consent? For others, I agree - it is untested.

> 
> #7: 6.1.1, item 3: "sold or rented" - heh - good luck
> with that;-) Why not just say doing that's a "MUST NOT"?
> What operator adhering to this BCP would really be
> selling or renting the data? I would hope none, and that
> the BCP calls that out as non-conforming.

So Section 5.3.3 is actually the part that describes what operators need to do for compliance and that does say “should not provide identifiable data to third-parties  without explicit consent from clients"

Section 6.1.1 is just stating what a DPPPS should contain, regardless of the level of compliance of the operator. I see a DPPPS as serving to let client know how an operator might deviate from complying from the BCP so shouldn’t necessarily be limited to only those that do comply?

> 
> #8: 6.1.1, item 6: That all looks too heavy to actually
> be done by any of the bigger folks who use anycast - are
> we really expecting them to list all the local laws that
> might apply?  What use would that be to someone
> selecting an operator really? While some information
> about jurisdictions ought be part of this, I think that
> needs more thought.

Item 6.1.2?

I didn’t read the section that way so it probably needs some clarification. I understood points 1 and 2 to apply to just the legal entity of the operator (e.g. Google would specify US) and point 3 is just a list of the counties the anycast servers are located in. Would be interested to hear other views on this.

> 
> nits:
> 
> - abstract: Is "DNS operators" right? Maybe this is
>  really just "DNS recursive resolver operators" and
> doesn't include advice for authoritative server
> operators? Probably better to have the abstract
> correct.

Section 2 makes the scope clear ‘'However this
   document is limited in scope to best practice considerations for the
   provision of DNS privacy services by servers (recursive resolvers) to
   clients (stub resolvers or forwarders).”

but it wouldn’t hurt to tidy up the abstract :-)

> 
> - DPPPS: that's not a great term, hard to say and hard
> to remember how many P's - I'd suggest a bit of creative
> bacronyming might be worthwhile - I'll start the bikeshed
> painting with "DNS Recursive Operator Privacy statement"
> or "DROP statement":-)

It is a total mouthful but the idea was to roughly follow but not collide with RFC6841 which describes a DP (DNSSEC Policies) and a DPS (DNSSEC Practice Statement). I’m liking DROP, but let the bikeshedding begin :-)

> 
> - I don't understand the bullet list just before section
>  5.1, nor the preceding paragraph - what's it trying
> to say?

It goes back again to the discussion about how prescriptive the document should be. The very first drafts had those 3 categories of actions as MUST/SHOULD/MAY options for operators in each section but there was push back against it so the current text is an attempt to convey more of a ’traffic light’ system of compliance.

> 
> - 5.1.2: X.509 probably needs a reference to rfc5280,
>  but where do I go to find a definition of an "SPKI
> pinset"? Later you point to rfc7858 for that but I don't
> find the term "pinset" used in there - "pin set" (with a
> space) is used in appendix A there, so maybe that's the
> reference you want.

Yup - it should be ‘pin set’ and this should also reference RFC8310 for authentication methods based on pin sets.

> 
> - 5.1.2: (sadly) maybe it'd be better to remove the
>  mention of the DNSSEC chain extension draft. I suspect
> it'd just confuse readers of an RFC later.

Ack. I lived in hope of its resurrection but since it hasn’t happened I’ll yank that text.

> 
> - 5.1.2: Is anyone publishing TLSA for DoT/DoH servers?
>  If so, great! If not, "can also consider" mightn't be
> worth saying.

Only a few of the individual test servers at the moment, none of the big players to my knowledge. That is why is is only an optimisation...

> 
> - 5.1.3.1: the link at [1] seems broken? I didn't check
>  all others, but someone should:-)

Will fix.

> 
> - 5.1.3.1: Maybe say using RFC8467 or a successor
>  specification? It's pretty early days to be very
> confident in that RFC. (Much though I like it.)

Will do.

> 
> - 5.1.7: Is "Operator" in the section heading here
>  really the operator of the DNS privacy service?
> Better to be clear about that. People might think you
> mean n/w operators between the stub and recursive.

Right, this section is designed to inform the DNS privacy service operator (I’ll update the heading)

> 
> - 5.2.3: The content of [10] really needs to be in the
>  document body, or else the text above it needs to
> change.

Ack - will see what SVG magic can be done or will convert to ASCII art...

> 
> - 5.3.1: The "NOTE" that the authors may add something
>  later needs fixing.

Indeed - I’ll follow up with Roland as I believe he was aware of work on this.

> 
> - 5.3.2: The "pitfalls" link reference is broken I
>  think.

Yup - will fix…..

> 
> - 6.1.1, item 2: add "and for how long and how those
>  logs are handled" (and maybe more) - this bullet also
> overlaps with item 3, so maybe it'd be better to
> re-organise this list a bit?

Fair point. I think merging 2 into 3 would be best?

> 
> - section 8: A "TODO" there isn't gonna work really:-)

I thought I had zapped them all but missed that one…. thanks. I’m not sure we have much more to say beyond what is in RFC7766 so I think the intention had been to just remove that unless anyone wants to suggest text?

FYI - I am out of the office next week so will follow up on any further discussion as soon as I am back!

Best regards

Sara.