Telechat Review of draft-ietf-pim-ipv6-zeroconf-assignment-13
review-ietf-pim-ipv6-zeroconf-assignment-13-intdir-telechat-box-2026-10-20-00
review-ietf-pim-ipv6-zeroconf-assignment-13-intdir-telechat-box-2026-10-20-00
I am the assigned int-dir reviewer for this draft. These comments were written with the intent of improving the Internet area aspects of the IETF drafts. Please wait for direction from your document shepherd or AD before posting a new version of the draft. For background on int-dir, please see the [FAQ](https://wiki.ietf.org/en/group/intdir). I previously reviewed revision 07 of this document. Revision 13 is substantially improved and addresses most of the points raised in that review. In particular, the document now: * specifies the construction of the eth-addr.arpa name; * identifies the use of RFC 4489 for constructing the IPv6 multicast address; * requires the selected group ID to be retained in persistent storage; * describes how to handle a network partition repair; * describes the denial-of-service attacks possible through conflict responses and veto records; and * explains the limited trust model inherited from mDNS. I do have some substantive concerns though, which I will now describe. It's possible they are due to my own misunderstanding, in which case I suggest adding clarifying text. But it also might be the case that these are genuine issues and so more work is needed. Use with multiple subnets ------------------------- In my first review I said that blindly forwarding mDNS between subnets is problematic for a number of reasons, which is why DNSSD is defining how mDNS and unicast DNS can work together across multiple links. I suggested reframing the document to explicitly only define link-local behaviour, and leave the question of how to handle multiple links to a future document. Revision -13 now says "The protocol can be extended across multiple subnets if PTR records are distributed between subnets (for example, by using an mDNS reflector). Distributing mDNS records between subnets is an ongoing area of research and improvement ([RFC7558])." This term mDNS reflector is not defined, but sounds like proxying or bridging of mDNS. RFC7558 sets out the requirements for scalable DNS-SD and considers that sending multicast packets between links is not scalable. It calls for a unicast protocol suite for multi-link discovery. If section 4 is to remain in the document, it should be changed to describe how this should be done, e.g. by referring to particular outputs of the DNSSD working group. However this may conflict with section 6.1 part 3 which prohibits unicast. My recommendation is that the section 4 topic is presented to the DNSSD working group where the relevant experts are present and perhaps the outlines of a multi-subnet and multi-link solution may become clear. Alternatively this document could be reframed to define only link-local behaviour. Network infrastructure ---------------------- In my first review I said "What is this infrastructure component? It hasn't been defined." It's still not defined, but at least the first paragraph of section 3 says that switches are an example of network infrastructure. So I will assume that the term means switches. Section 3 then gives requirements to network infrastructure: "When a network infrastructure component detects a collision it cannot resolve, it triggers a conflict with the application by publishing a veto record." Are special switches now required to do this? This seems in conflict with https://www.rfc-editor.org/info/rfc10019/#name-solution-requirements which says the solution needs to operate in "dynamic, infrastructure-free environments". Ethernet address uniqueness --------------------------- Section 5's evaluations says: Unique Address Assignment: Use of the protocol results in a unique address being assigned to the multicast group at both the network and link layers. It's true that within the allocated group-ID range, different group IDs produce different ethernet multicast addresses. However, I think an ethernet address can still collide with the ethernet mapping of an IPv6 multicast address allocated outside this protocol. Section 2.7 makes host monitoring for this condition optional. Section 3 provides veto records as a detection mechanism, but this depends on the aforementioned network infrastructure components detecting the collision and implementing the vetos. The document does not seem to require such components to be present. Consequently, in a deployment without optional host monitoring and without an infrastructure component declaring veto records, the protocol does not appear to guarantee detection or resolution of such an ethernet collision. Unclear text ------------ Section 2.9 says "Note that retaining the value does not permit the application to skip any of the preceding steps." If I understand correctly, it enables it to skip generating a random group ID in 2.1. Perhaps this sentence should say does not permit the application to skip the steps in sections 2.2 to 2.8. Optional editorial ------------------ Personally I suggest it would be better if the introduction was a little shorter: specifically the final three paragraphs of section 1 could be placed elsewhere in the document. Some of it is already covered in section 8. This is completely optional though, with the aim of making it slightly easier for readers.