Hi Yaron, I see your point. IKEv2-bis completely punts the issue and so you want to too. If that's your tact you might want to mention _how_ an implementation is supposed to interoperate with a AAA server, like an RFC that defines this behavior. If there is no RFC to define the behavior you want to refer to then I think you (and IKEv2-bis!!!) still have a problem. At the absolute minimum I'd at least like to see some text that describes how to deal with this problem in the draft, something along the lines of, "if you implement this scheme then, in lieu of some external indication of the authenticated identity of the client, the security policy and access control provided to the client MUST be based that afforded by the lowest ordered entry in the SPD (see RFC 4301)." That way the lowest common denominator rule will apply to what is, effectively, an unknown entity. You also seem to be treating the authentication issues with gateway identity identically. You treat the IKEv2-with-EAP case and the EAP-only case as the "lying NAS" problem. But that isn't true. In the former the client gets an authenticated identity. It _might_ be different than the identity the gateway presents to the AAA server, and hence the "lying NAS" issue, but it's an authenticated identity. In the EAP-only scheme the identity is completely unauthenticated and can be anything the gateway decides to make it. This becomes the "lying NAS" problem on stilts. I'd be happy with text along the lines as that I supplied above that talks about this problem but from the client point-of-view. Regarding "channel bindings". It's not just that this capability hasn't been defined in an interoperable manner across the many EAP methods with a "y" in that table. It hasn't been defined at all anywhere! What your draft says is: "EAP methods that provide what is called 'connection binding' or 'channel binding' transport some identity or identities of the gateway (or WLAN access point/NAS) inside the EAP method. Then the AAA server can check that it is indeed sending the key to the gateway expected by the client." And that just is not true. EAP methods that support "channel binding" as indicated by the table do not transport anything anywhere. And EAP-SAKE says it doesn't support channel bindings so you might want to fix that table. The paragraph you quote says, "the method provides protection against certain types of attacks" but it does no such thing. All the method might provide is a way to negotiate an encryption and integrity algorithm and that's a far cry from providing protection against this specific attack. This draft has security issues. Some you don't mention because another RFC that has the same issues doesn't mention anything, some you conflate, and some you infer that the problem has been solved by some other protocol capability when it hasn't. If the problems can't be solved in this draft, or you think they shouldn't be solved here, then at least say what those problems are so a casual reader or implementer knows what he's getting into if he decides to implement this. regards, Dan. On Thu, May 27, 2010 1:43 pm, Yaron Sheffer wrote: > Hi Dan, > > Thanks for reviewing this document. Responding to your main point: > > The need to correlate the client's EAP and IKE identity is not new to > this protocol. It has to be done in exactly the same manner, for exactly > the same reasons, in today's IKEv2. This is what IKEv2-bis says about it: > > When the initiator authentication uses EAP, it is possible that the > contents of the IDi payload is used only for AAA routing purposes and > selecting which EAP method to use. This value may be different from > the identity authenticated by the EAP method. It is important that > policy lookups and access control decisions use the actual > authenticated identity. Often the EAP server is implemented in a > separate AAA server that communicates with the IKEv2 responder. In > this case, the authenticated identity, if different from that in the > IDi payload, has to be sent from the AAA server to the IKEv2 > responder. > > I don't see a reason to add a lengthy RFC 4301-analysis for a problem > that already exists in the base protocol. > > As to the equivalent problem for the server's identity, this is the > "lying NAS" issue. I agree that it is not solved well today, and the > draft says so quite clearly: > > This third, optional property of the method provides protection > against certain types of attacks (see Section 6.2), and therefore in > some scenarios, methods that allow for channel binding are to be > preferred. It is noted that at the time of writing, even when such > capabilities are provided, they are not fully specified in an > interoperable manner. > > The last sentence of this paragraph says that we *do not* have a > workable solution yet. Perhaps we should have worded it more strongly. > > Please let me know if you still request additional clarifications in the > draft. > > Thanks, > Yaron > > On 05/27/2010 12:22 AM, Dan Harkins wrote: >> >> Hello, >> >> Here are some comments on this draft for IETF LC. There are some >> editorial nits, some errors that might be considered editorial but >> there is a serious issue with the Security Considerations that I think >> needs addressing. >> >> I'll start with the security issue since it's the most serious (and >> some may stop reading after getting to the editorial complaints). >> >> Issue in Security Considerations: >> >> In IKEv2 both sides present identities in the form of an ID >> payload. >> EAP also has an identity exchange that is not useful for >> authentication >> so EAP methods typically include their own, separate, identity >> exchange. In many cases that identity is in a protected channel >> from >> the EAP peer to the EAP server. When the EAP server is not >> co-located >> with the IKEv2 implementation (i.e. EAP is offloaded to a separate >> server) the identities that are actually authenticated are unknown >> and/or unverified. >> >> Section 6.2 mentions this from the client's point-of-view-- "after >> the client has verified the AUTH payload sent by the IKEv2 gateway, >> it knows that it is talking to SOME gateway trusted by the home AAA >> server, but not which one." This is used as a segue to the "lying >> NAS problem", which I have note below is not really solved so this >> problem isn't really addressed properly. >> >> But the problem is worse. The point-of-view of the gateway is never >> mentioned. The gateway may know some anonymous identity, or an >> identity that is colored or obfuscated that is useful only by the >> EAP server to determine which EAP method to offer. It does not know >> what the real client identity is. The identity that gets >> authenticated >> is unknown to the gateway! >> >> Unfortunately, it is the authenticated identity that the gateway >> must >> use to look up an entry in the IPsec Security Policy Datbase (see >> RFC 4301) that says what kind of security (bypass, drop, protect >> rules, etc) to apply to the client's packets. It is therefore not >> possible to properly support RFC 4301 when using this EAP-only >> method of authentication. >> >> Furthermore, the only identity available to the gateway can be >> forged so the client can achieve more favorable access to the >> protected network (a more "generous" security policy) than the >> gateway would have given it had it known its _real_ identity. >> >> I think there has to be a top-down analysis of RFC 4301 and how >> this >> scheme impacts it. Each part of RFC 4301 that cannot be supported >> or can only be supported in a limited manner must be spelled out >> and >> the impact of the lack, or limit, of support has to be presented in >> this draft's Security Considerations so a reader/implementer knows >> what he's getting into if he decides to implement this scheme. >> >> I would advise a DISCUSS on this draft until this has been worked >> out. >> >> Errors: >> >> - the draft states "IEEE 802.11i uses EAP without any PKI for >> authenticating the WLAN access points." That is just plain wrong. >> >> IEEE 802.11 is agnostic about particular EAP methods but requires >> they provide mutual authentication and key generation. The Wi-Fi >> Alliance certifies IEEE 802.11 implementations (and the EAP methods >> they use). Of the EAP methods certified for use in IEEE 802.11-- >> EAP-TLS, EAP-TTLS w/MSCHAPv2, PEAPv0 w/MSCHAPv2, PEAPv1 w/GTC, >> and EAP-SIM-- only EAP-SIM does not require a PKI. Having been >> involved in that particular industry for most of the past decade >> I can assure the authors that PKIs of some sort are used in most >> every deployment of IEEE 802.11i. I have yet to see a pure EAP-SIM >> deployment anywhere. >> >> - in section 6.2 the draft mentions the "lying NAS problem" and says >> that "EAP methods that provide what is called 'connection binding' >> or 'channel binding' transport some identity or identities of the >> gateway (or WLAN access point/NAS)." The table in section 4 lists >> EAP methods that supposedly support "channel binding" but none of >> methods with a "yes" in that column do what section 6.2 says. >> >> The closest is EAP-SAKE (RFC 4763) which includes identities in a >> MIC but that won't solve the "lying NAS problem" and RFC 4763 even >> says, "EAP-SAKE does not claim channel binding" (so the table in >> section 4 is wrong). The others provide a protected channel but >> don't >> say how this can be used to actually solve the "lying NAS problem". >> The draft really needs to say that this problem has not been solved >> yet and remains an issue that must be taken into account if someone >> decides to implement this scheme. >> >> Editorial nits: >> >> - this draft really does not justify its existence well at all. I >> know the WG decided to pursue this work item but the reasoning >> around that decision does not come through in this draft. >> >> The introduction says that "[P]ublic key signatures and shared >> secrets are not flexible enough to meet the requirements of >> many deployment scenarios." OK, so there's some unnamed credential >> needed (probably a token card or some such). This is handled using >> the existing EAP support in IKEv2 so why is this scheme being >> defined? >> >> Then the draft mentions as justification that "deployment of a PKI >> is required" and "in many cases this is not realistic." If that's >> the case then why are EAP methods that require a PKI listed in >> the table of acceptable EAP methods for this technique? >> >> In section 2 the draft says "Offering EAP based authentication has >> the advantage that multiple different authentication and key >> exchange protocols are available with EAP with different security >> properties (such as strong password based protocols, protocols >> offering user identity confidentiality and many more)." OK, fine >> but all that is easily satisfied with the existing EAP support in >> IKEv2 so why is this scheme being defined? >> >> > _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf