Re: [BEHAVE] [NAT64] Hairpinning loop

[Brian E Carpenter <brian.e.carpenter at gmail.com>]

Commenting on my own comment...

On 2009-11-08 12:09, Brian E Carpenter wrote:
> On 2009-11-08 01:56, Simon Perreault wrote:
>> Brian E Carpenter wrote:
>>> On reading section 6 a few times, I couldn't figure out a case where
>>> sending a hairpin packet would actually have any value to a user.
>>> There will always be another way of reaching the intended destination
>>> with a native IPv6 address.
>> RFC 4787 section 6, RFC 5382 section 7.2, and RFC 5508 section 5 address
>> this for the IPv4-IPv4 case.
>>
>> Does this still apply for IPv6-IPv4?
> 
> I abstained from IPv4-IPv4 NAT discussions years ago, so I'm not
> familiar with those scenarios. However, since the normal mode for
> IPv6 is multiple addresses per host, I would suggest that any
> communication that occurs between IPv6 hosts behind the same NAT64
> SHOULD occur using some other form of address, so it is perfectly
> OK to specify that hairpin packets SHOULD be dropped. It is not
> the same as IPv4.
> 
>     Brian

I should have read Dan's message first. OK, if a via-IPv4 referral
has taken place, the host generating the hairpin packet might
have no choice (but see the GROBJ BOF). But since we have a looping
issue, I would modify what I just wrote to "hairpin packets MAY
be dropped" so that an implementor has the freedom to choose.

   Brian