Hi, This is a followup on my Gen-ART review of draft-ietf-dhc-forcerenew-nonce-04, based on my previous review of version 03. In summary, this version is improved, but I still don't think it's ready for publication. On Feb 6, 2012, at 5:17 PM, Ben Campbell wrote: > I am the assigned Gen-ART reviewer for this draft. For background on Gen-ART, please see the FAQ at . > > Please resolve these comments along with any other Last Call comments you may receive. > > Document: draft-ietf-dhc-forcerenew-nonce-03 > Reviewer: Ben Campbell > Review Date: 2012-02-06 > IETF LC End Date: 2012-02-06 > > Summary:This draft is not quite ready for publication as a proposed standard. There are some potentially significant issues that should be addressed first. > > [Note: Hopefully this draft has had or will have a SecDir review, since it seems ripe for significant security implications.] > I didn't see a comment about whether this happened. I still recommend a SecDir review if one has not yet occurred. > > *** Major issues: > > -- I admit to not being a DHCP expert, but If I understand this draft correctly, it proposes to send what is effectively a secret-key in a DHCPACK request, then use that key to authenticate a force renew message. It seems like any eavesdropper could sniff that key, and use it to spoof force renew requests. The introduction mentions that there may be some environments where the use of RFC3118 authentication could be relaxed, and offers an example of such an environment. But nowhere does this draft appear to be limited in scope to such environments. > > I think some additional text in (perhaps in the security considerations) is needed to explain either why the vulnerability to eavesdroppers is either okay in general, or limits the mechanism's use to environment where it is okay. It also seems like that, in the best case, this mechanism proves only that a Forcerenew request comes from the same DHCP server as in the original transaction, but otherwise does not prove anything about the identity of that server. If so, it would be worth mentioning it. > This is improved by some changes to the introduction and the security considerations. But I still don't see a clear, normative statement about what environments this can be safely used in. I know Ted objected to Roberta's specific proposal, but I think to move forward this needs _some_ clear, concrete guidance about when to use or not to use it, or when 3118 authentication is more appropriate. Don't leave it up to the reader to draw the right conclusion. > -- The mechanism appears to be limited to HMAC-MD5, and there does not appear to be any way of selecting other algorithms. Is HMAC-MD5 really sufficient as the only choice? Is there some expectation that stronger mechanisms or algorithm extensibility would be too expensive? (Perhaps the extensibility method would be to specify another mechanism that's identical except for a different HMAC algorithm. But if that's the intent it should be mentioned.) There's a statement in the security considerations that "... security of the nonce in the case of on-link attacks isn't relevant." It would help drive the point home to add something to the effect that "Therefore HMAC-MD5 is by definition adequate for the purpose, and there is no need for an extensible HMAC mechanism." > > *** Minor issues: > > -- Section 1 " > In such environments the mandatory coupling between FORCERENEW and DHCP Authentication [RFC3118] can be relaxed." > > It's not clear to me what connection that assertion has with this draft. Is there an intent that the proposed mechanisms be used only in such environments? I don't find any language scoping this proposal to any particular environment. See comment under the first "major issue". I still think we need normative, concrete guidance about when to use or not to use this mechanism. [...] > > -- section 3.1.3, 2nd paragraph: "The server SHOULD NOT include this option unless the client has indicated its capability in a DHCP Discovery message." > > Why not; what harm would it do? And on the other hand, if you want to discourage it, why not go all the way to "MUST NOT"? > I don't think my comment has been addressed by the discussion so far. I understand that the server should not insert a nonce unless the client has advertised support--but is it the author's intent that the server should not _advertise_ support unless the client has first advertised support? Additionally, this update added the following text to the section: > The > server SHOULD NOT include the nonce in an ACK when responding to > a renew unless a nonce was generated. This minimizes the number of > times the nonce is sent over the wire. I don't understand the first sentence. Don't include a nonce unless you generated it first? That seems like a tautology. Do you mean to say unless the nonce was sent in a previous message? Or unless a _new_ nonce was generated? [...] > > -- section 3.1.4, 1st paragraph: "DHCP servers that support Forcerenew nonce Protocol authentication MUST include the DHCP Forcerenew Nonce protocol authentication option…" > > Only if the client advertised it, right? Otherwise, this seems to conflict with the previous SHOULD NOT. The text is updated, but it still conflicts with the previous statement in 3.1.3, paragraph 2 that says the server SHOULD NOT insert the option unless the client first inserted it. (See comment above). [...] > > -- section 6: > > You mention this mechanism is vulnerable to MiTM attacks. Why is this okay? Are there some environments where it is good enough and others where it is not? (Also, do they really need to be MitM attackers? Seems like any eavesdropper could learn the nonce.) The text still refers to a MiTM attacker. Do you mean "on-link" attacker? Otherwise, you've added text to motivate the mechanism for a specific use case. But I'd still like to see concrete guidance about when this draft is appropriate to use. > > *** Nits/editorial comments: > > -- Abstract, second sentence: > > I have trouble parsing this sentence. Are the propositions correct? Partially fixed. Should "on the initial ACK" be "in the initial ACK"? > > -- Section 3.1.1,4th paragraph: "… configured to require …" > > Do you mean configured to "require", or configured to "use"? I would normally take "requires" to mean that the server would not work with clients that don't advertise support for the mechanism. You changed the text from "configured to require" to "requires". I'm still confused on whether "requires" means that the server will refuse to service clients that don't advertise support. This is in contrast to "use", which would mean to me the server would invoke the option if the client supports it, but still work with clients that don't support it. [...]