Hi, Trying to nudge the discussion on ICMP errors, wanting to nail this down soon...
On Sep 3, 2010, at 1:51 PM, Gerrit Renker wrote: > a. It would be good to specify how UDP ICMP errors are translated into DCCP > semantics, in particular the "administratively prohibited" variants of > Destination Unreachable (RFC 1122, 3.2.2.1), and how these options are > to be passed on to UDP. Do we assume that UDP implementations are modified to be able to route the full ICMP messages intended for DCCP-UDP? Or that there is some other trick to capture the full ICMP messages? If yes, it might be good to be clear on that. Do we want to discuss the case where DCCP-UDP sits on top of an unmodified UDP implementation (e.g., as a user-space implementation using a standard UDP socket)? Then the ICMP handling becomes more limited. A connected UDP socket might be able to pass some ICMP errors, assuming that there is a way to route the error to the right DCCP connection (each DCCP connection would perhaps need to map to a different UDP "connection"). Some information might be lost, such as the reason for host unreachable, or the next-hop MTU in packet too big message. So implementing path MTU discovery might be difficult (there would also need to be a way to enforce the DF bit). Then there is the question of how different ICMP errors are translated to DCCP. "port unreachable" could mean that the other end does not support DCCP-UDP encapsulation, but can other error types be passed up as such? - Pasi
