Huge thanks to our Platinum Members Endace and LiveAction,
and our Silver Member Veeam, for supporting the Wireshark Foundation and project.

Wireshark-dev: Re: [Wireshark-dev] Using find_conversation with multiple conversations conducte

From: Graham Bloice <graham.bloice@xxxxxxxxxxxxx>
Date: Fri, 26 Jun 2009 10:40:19 +0100
kelvin.proctor@xxxxxxxxxxxxxxxxxxxxxxxxx wrote:

Hi Wireshark Dev List,

We have a network with DNP3.0 that is using UDP.  We have a gateway device that communicates to a collections of RTUs (RTU = Remote Terminal Unit, a device on the DNP3.0 network) over a serial radio-modem network and we talk to the gateway using UDP / IP.

The DNP3.0 dissector uses the standard calls to find_conversation etc... and fragment_add_seq_next etc... to do application layer frame re-assembly as follows:

      if (! (tr_fir && tr_fin))
      {
        /* A fragmented packet */
        pinfo->fragmented = TRUE;

        /* Look up the conversation to get the fragment reassembly id */
        conversation = find_conversation(pinfo->fd->num, &pinfo->src, &pinfo->dst,
          pinfo->ptype, pinfo->srcport, pinfo->destport, 0);

        if (conversation == NULL) {
          /* No conversation yet, so make one */
          conversation = conversation_new(pinfo->fd->num,  &pinfo->src, &pinfo->dst, pinfo->ptype,
            pinfo->srcport, pinfo->destport, 0);
        }

        conv_data_ptr = (dnp3_conv_t*)conversation_get_proto_data(conversation, proto_dnp3);

        if (conv_data_ptr == NULL) {
          /* New data structure required */
          conv_data_ptr = se_alloc(sizeof(dnp3_conv_t));

          /*** Increment static global fragment reassembly id ***/
          conv_data_ptr->conv_seq_number = seq_number++;

          conversation_add_proto_data(conversation, proto_dnp3, (void *)conv_data_ptr);
        }
        conv_seq_number = conv_data_ptr->conv_seq_number;

        /*
        * Add the frame to
        * whatever reassembly is in progress, if any, and see
        * if it's done.
        */

        frag_msg = fragment_add_seq_next(al_tvb, 0, pinfo, conv_seq_number,
            al_fragment_table,
            al_reassembled_table,
            tvb_reported_length(al_tvb), /* As this is a constructed tvb, all of it is ok */
            !tr_fin);

        next_tvb = process_reassembled_data(al_tvb, 0, pinfo,
            "Reassembled DNP 3.0 Application Layer message", frag_msg, &dnp3_frag_items,
            &update_col_info, tr_tree);

Unfortunately this does not take into account the fact that over the same IP address / port number pairs we may receive messages from many different source RTUs that should be treated as separate conversations.  The problem we have at the moment is that we are getting fragment re-assembly occurring for fragments from difference RTUs.

Is there a standard way to handle situations like this?  If so can someone point me to the correct dissector to look at?

We can provide packet captures that demonstrate the issue if desired.   It is considered acceptable to attach a .pcap file to a posting to this mailing list?

Thanks in advance,

Kelvin Proctor

Kelvin,

>From your description I think you have multiple concurrent "conversations" from the RTU's via the gateway.  Unfortunately as I understand it the conversation support in wireshark expects the conversations to be sequential.  I struggled to get this part of the dissector working so will need a greater mind than mine to suggest a solution.  I presume something could be done with the RTU address as an additional identifying parameter for the conversation, so maybe conversation support could be extended to allow searching existing conversations with additional info other than the IP\port combo.

BTW, how does the dissector cope with fragmented messages over UDP or does your gateway always send complete data link packets?

I think you should raise a bug on bugzilla (https://bugs.wireshark.org/bugzilla/) noting that it's the conversation support that's lacking and attach a pcap file illustrating the problem.

-- 
Regards,

Graham Bloice