>>>>> On Wed, 4 Aug 2004 11:15:36 -0700 (PDT), >>>>> Erik Nordmark <[EMAIL PROTECTED]> said:
>> To make it sure, I'm going to talk about the following bullet of >> Section 3.1: >> >> * (adds to 7.2.6) The Optimistic node MAY send an unsolicited >> Neighbour Advertisement to All Nodes when it first configures an >> address. The Override flag on this advertisement MUST be cleared >> (O=0). >> >> Hmm, I'd still like to remove this bullet, since I don't want to see >> new implementations introduce the optimization just because it's a >> "MAY", without considering the rationale and possible bad effects. > Sorry, I missed that one. > I think the bullet should be reworded to not in any sense recommend sending > such things. > Saying that it isn't illegal for optimistic nodes to send unsolicited NAs > as long as the Override flag is clear should be ok, but the text above talks > about doing something when an address is configured which we should avoid. My meta-level concern is that if we explicitly mentioned the unsolicited NA the intent of the NA would be rather unclear. In fact, even rev01 of the optimistic-dad draft doesn't talk about the real intent (i.e., a signal for mobility-supporting routers). We only understood the author's real intent through the discussion in the wg last call. After understanding the real intent, we seem to have agreed that we should not override the semantics of unsolicited NA for such a specialized purpose. So, we now don't have a reason to send such an NA for the original purpose. Meanwhile, RFC2461 (and 2461bis) talks about a case where a node may want to send unsolicited NAs (e.g., in Section 7.2.6): the case of link-layer address change. Of course, this does not match the optimistic-DAD case either. Now, why would we bother to mention such an NA at all, while we do not have a reason to send it? If the optimistic-DAD document does not mention it at all, then new implementors would even never imagine to send it, since there is no documented reason to do so. Considering the fact there is (reportedly) an implementation that sends such unsolicited NAs for the reason which is now almost deprecated, I'd rather add an explicit note that discourages the behavior. Otherwise, new implementors may see the source code or the behavior and try to do the same thing without considering the background motivation, which is now actually obsolete. One may perhaps worry about the case where a new implementation happens to send an NA (for an optimistic address), either solicited or unsolicited, setting the override flag ON. And I guess your suggested new text tries to avoid the behavior explicitly. I personally think we don't have to worry about it, since Section 2.3 of draft-ietf-ipv6-optimistic-dad-01.txt prohibits the behavior: * clearing the 'Override' flag in Neighbour Advertisements for Optimistic addresses, which prevents neighbours from overriding their existing NC entries. The 'Override' flag is already defined [RFC2461] and used for Proxy Neighbour Advertisement. We could still note the same thing if we really want to care about it, but then we should simply concentrate on the override flag without mentioning solicited/unsolicited, e.g. An Optimistic node MUST NOT send any Neighbor Advertisements when the address is in Optimistic state, unless the Override flag on the advertisement is zero. *** conclusion *** So my suggestion is: 1. remove the following bullet of Section 3.1: * (adds to 7.2.6) The Optimistic node MAY send an unsolicited Neighbour Advertisement to All Nodes when it first configures an address. The Override flag on this advertisement MUST be cleared (O=0). And then (optionally) 2. add the following somewhere (perhaps not in Section 3, since it's not a modification to the existing specifications): * The Optimistic node MUST NOT send an unsolicited Neighbor Advertisement containing an Optimistic address in the target address field. This is to prohibit new implementations that have seen existing one using the obsolete behavior from doing the same thing without considering the real reason for it. I prefer 'MUST NOT', but this might be too strong and I can live with 'SHOULD NOT'. And then (optionally) 3. add the following somewhere (perhaps not in Section 3, since it's not a modification to the existing specifications): * The Optimistic node MUST NOT send any Neighbor Advertisement containing an Optimistic address in the target address field, unless the Override flag on the advertisement is zero. if we really want to avoid an NA that happens to set the override flag ON. JINMEI, Tatuya Communication Platform Lab. Corporate R&D Center, Toshiba Corp. [EMAIL PROTECTED] -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------