[BEHAVE] General PMTUD black hole detection in translators

[Iljitsch van Beijnum <[e-mail]>]

If we want to avoid cases where there is a link with a reduced MTU between the IPv6 hosts and a translator that triggers path MTU discovery black holes towards certain IPv4 servers, we could add a form of PMTUD black hole detection to translators.

It would work like this: the translator of course translates too big messages from the IPv6 side to IPv4. If it remembers that it did this and then sees that the IPv4 host doesn't lower its packet sizes within a reasonable timeframe, it can then when new sessions are set up from the same IPv6 host towards the same IPv4 host, adjust the MSS to the value that was present in the earlier too big message.

So then the first TCP session would time out but subsequent ones would have their MSS clamped and those would work. This would only kick in when there is an actual black hole so there is no impact for well behaved systems. The initial session would still fail so there would still be good reason for PMTUD breakers to fix their systems, but after the initial failure, the user would have something that works.