This is a re-review. My answer below, sent on Jan. 16th,  also applies to version -07.

Summary: Almost ready

Most of my initial comments have been addressed by the authors.

I made a few suggestions/clarifications (below), in addition to my initial review.

I would insist on implementing the first one, 2), beginning with:
	"I’d like you extend a little the the bullet (b) of introduction: … »

but let the authors judge what they prefer to do for the next ones.

Regards,

  Vincent



> Le 16 janv. 2018 à 09:50, Vincent Roca <vincent.roca@inria.fr> a écrit :
> 
> Hello Rolf,
> 
>> thanks for the detailed review and sorry for the belated reply. We have submitted a new version based on these comments and recommendations. Please see inline:
> 
> You’re welcome.
> 
>>> 2) Another point: I'd like to see the item (b) of introduction
>>> developed as this is the key issue with multicast/broadcast traffic.
>>> Without encryption, application messages are vulnerable to
>>> eavesdropping and all kinds of analyses. A group key could be
>>> negociated and used for encrypting this traffic, but this requires a
>>> more complex setup mechanism (typically a key management system,
>>> probably with re-keying and all the stuff) and requires a preliminary
>>> association of a terminal that is not necessarily compatible with the
>>> application-layer protocol features.
>>> The text could say that if possible, multicast/broadcast traffic
>>> encryption should be used, otherwise risks should be minimised as
>>> discussed in this document. And of course, it's always good to remind
>>> the designers that unicast traffic should be always encrypted since
>>> there's no good reason not to do that.
>> 
>> 
>> We talk about this in the security considerations section and come to the same conclusion as you do:
>> 
>> "It should be noted that certain applications could make use of
>>   existing mechanisms to protect multicast traffic such as the ones
>>   defined in [RFC5374].  Examples of such applications can be found in
>>   Appendix A. of [RFC5374].  Given the required infrastructure and
>>   assumptions about these applications and the security infrastructure,
>>   many applications will not be able to make use of such mechanisms."
>> 
>> In which way would you expand on this?
> 
> I’d like you extend a little the the bullet (b) of introduction:
> (b) Encryption is more difficult when broadcast/multicast messages,
>         leaving content of these messages in the clear and making it
>         easier to spoof and replay them.
> It suggests that because it’s difficult, it won’t be used. This is not the
> message the I-D should carry. It’s difficult (add a reference here), sure,
> but not impossible. You can also briefly explain why it’s more difficult with
> multicast/broadcast.
> 
> 
>>> 3) I'd like to suggest another recommendation: when applicable, limit
>>> or avoid user-originated broadcast/multicast messages, preferring
>>> infrastructure originated messages.
>>> An example: in a discovery mechanism, avoid terminal sollicitation as
>>> much as possible and limit oneself to infrastructure announcement
>>> broadcast messages, even if it can include some extra delay (controled
>>> with the announcement frequency). This is exactly what happened in
>>> Wifi AP discovery, where the passive discovery scheme, AP based, is
>>> now prefered to the active scheme.
>>> Said differently, multicast/broadcast is not per se the issue, when
>>> originating from the infrastructure it is generaly safe.
>> 
>> I think I disagree somewhat (and it is not a generally applicable use case, e.g. if resources on a host are offered there is no general way to delegate announcements of these to the infrastructure). Saying that infrastructure announcements are generally safe requires that a receiver can validate that a message came from the infrastructure which is a problem (e.g. rogue DHCP servers). A lot of deployments solve some of this through infrastructure means (e.g. DHCP snooping), but we have documented that. Also, whatever comes from the infrastructure requires a standard so that devices/apps understand it. We have text around that, too.
> 
> Let me clarify what I had in mind. A few years ago, discovery of a known WiFi AP
> from a device was accelerated through the use of an active mechanism, with the
> device sending probes to potentially detects those know WiFi APs. That was
> awful from a privacy point of view as highlighted by researchers. The trend was to
> move to a passive approach, relying mostly on AP announcements (in addition to
> MAC address randomisation). The privacy benefits while doing so are terrible.
> See e.g., my colleague Mathieu publication on the topic: https://hal.inria.fr/hal-00747825 <https://hal.inria.fr/hal-00747825>
> This is perfectly in line with your  I-D. How privacy can be compromised by using
> a broadcast based protocol. 
> 
> Having said that, I agree with you that rogue announcements raise security concerns
> (but that’s true for any message sent), and it does not work for resources offered by
> a host. I agree, this is not a universal approach. I let you judge whether it makes sense
> to talk about it or not.
> 
> 
>>> 4) A more general comment, related to item (a) of intro:
>>> Broadcast/multicast can simplify user surveillance, I agree. However,
>>> when I'm looking at all the practicies of companies working in the
>>> advertising area (in particular A&A companies, sitting between
>>> smartphone users and advertisers, domain that I know a little bit),
>>> the techniques deployed are incredibly smart, some of them bypassing
>>> technical protections set up by the OS manufacturers, others relying
>>> on hidden technical features. It's impressive how inventive those
>>> companies can be.
>>> My point is that even if traffic is 100% unicast, as long as a
>>> wireless medium is used, anyone wanting to monitor users in order to
>>> profile them will set up the required hardware/software configuration.
>>> This is trivial and extremely cheap nowadays (except perhaps for 4G
>>> connections), and most of the time this is a purely passive
>>> monitoring. The unicast/multicast/broadcast nature of traffic has no
>>> importance as long as you are within wireless network range.
>> 
>> 
>> I think there is little that the IETF can do about that. Also "bypassing technical protection" and using "hidden technical features" is nothing you can protect against in general and I assume is often not by design.
> 
> I agree and I don’t want to question the importance of this I-D here.
> But it remains that wireless communications are so easy to monitor.
> 
> 
>>> *Various:*
>>> ** the authors use the term "host name" to refer to a higher layer
>>> name of a device. To me this term refers to the DNS notion of
>>> "hostname", i.e. something different, hence confusion. I'd recommend
>>> to use "device name" in order to avoid it.
>> 
>> Well, RFC 7719 even admits the term is ill defined and does refer to various things. Also, for the protocols we looked at and OS configurations, sometimes there are different host namens being used for different purposes. We would like to stick with host names (also since RFC8117 useses the term which is highly relevant in this context).
> 
> I was not aware of those RFCs. Do what you believe is the most appropriate.
> 
> Cheers,
> 
>    Vincent