Christian Vogt wrote: >>Anyway, checksum of ICMP error generated against ICMP echo must >>still be changed, and the error packet may (though does not usually >>have to) be fragmented. > Sure, applications that use address referrals don't work well through > NAT; we mentioned this earlier. ICMP is one of these "applications". You miss the point. People expect NAT support ICMP. But, NAT for ICMPv6 does not need state, unless it is ICMP error against ICMP echo, contains large routing header, is fragmented and the first fragment does not contain checksum. It is an example of how IPv6 was bloated to have many useless features ignoring corner cases of interactions of the features. Masataka Ohta _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf