[HECnet] Multinet over TCP -- protocol and observations

[Paul_Koning at Dell.com]

With help from Bob Armstrong and Peter Lothberg, I have a reverse engineered Python implementation of the Multinet tunnel over TCP working now.   

Peter forwarded an old HECnet message that roughly describes the protocol, but some details were missing and some were not quite accurate.   For the benefit of anyone else who wants to implement this, here is the protocol description plus some implementation notes.

	paul

                                       

Multinet protocol for DECnet

A Multinet tunnel runs the Point to Point mode of the DECnet protocol. Traffic may be carried over a TCP connection, or over UDP packets.   The operation is nearly the same for both.

In TCP mode, one end is designated the connecting end and the other the listening end.   In UDP mode, operation is symmetric since there are no connections.   In either case, the port number may be configured; the default is 700. 

Multinet puts a 4 byte header on each DECnet packet.   For TCP, the first two bytes are the DECnet packet length, little endian.   The TCP data stream consists of these packets with headers.   

For UDP, the first two byte are a sequence number, little endian.   The next two bytes appear to be unused and were observed to contain zero.   Each DECnet packet, with its header, is sent as a separate UDP packet.   Some Multinet implementations can be configured to check the sequence number in UDP mode; it is not clear how this works (in particular, how it is initialized).   Linux does not, and things seem to work fine without this extra check.

Multinet tunneling does not obey many of the requirements that the DECnet Routing architecture imposes on point to point datalinks.   The most obvious issue is that routing layer point to point init messages (control packet, type code 0) may appear unexpectedly.   This happens even in TCP mode where it might be expected that the TCP connection would be closed and re-established before the init message.

A conforming DECnet implementation reacts to such an unexpected init message by restarting the circuit, sending an init message, and expecting one in return.   Multinet does not expect this, and if the peer operates according to the standard, initialization may require many cycles before it finally succeeds.   As a workaround, the response to an unexpected init should be to set the routing initialization sublayer state machine to the DS state (without restarting the TCP connection, if any), and processing the unexpected init message as if it had been received in that state.

In testing with a DECnet/VMS system, it was observed to send an Init message requesting verification even though verification was not called for in the circuit parameters.   It may be that this was due to the routing architecture requirement to use verification in areas other than area 1.   A null verification message (empty FCNVAL field) was accepted by the VMS system in that test.