state loss detection

[Iljitsch van Beijnum <iljitsch@muada.com>]

Hi,

It just occurred to me that the state loss problem may not be as  
intractable as we thought.

The problem is that when (for instance) a host taking part in the  
shim6 protocol reboots, and it sees shimmed packets when it gets back  
on its feet again, it gets confused. If the packets that are  
rewritten by the shim only have their addresses rewritten, the  
checksum on those packets will be invalid, the packet will be dropped  
and the source host doesn't get to hear about it so the failure won't  
be detected until there is a timeout.

However, as I wrote in a draft a while ago, it may be a good idea to  
rewrite the checksum too, in order to avoid confusing firewalls and  
the like unnecessarily, and to maintain compatibility with NICs that  
do checksum offloading. (Although I'm not familiar with ones that do  
this for IPv6.)

With the checksum rewritten as well, the packet will flow through the  
shim layer without any action as there is no state, and the upper  
layer will generate an error, such as a TCP RST.

In and of itself this doesn't provide the host that still has state  
with the information it needs to recover from the remote state loss  
(it could have been a legitimate RST that was shimmed), but it  
wouldn't be huge imposition to send a shim state refresh when this  
happens.

For TCP and UDP where the host with the state loss used an ephemeral  
port this should work well. I'm not sure what happens when an UDP  
server does when it sees packets for an unknown "session". For non- 
UDP/TCP protocols the transport protocol would probably have to  
trigger a shim state refresh.