SNMP locking up problem

[Won-Sik Kim]

 

Hi all,

 

I develop snmp agent which use agentX in Linux.

 

 

I found a SNMP locking problem.

 

1. which kind of locking ?

    a. Every snmp get/set/getnext pdu to snmpd are discarded.

    b. But, snmpd shows no problem with processing any signal.

    c. It looks like main loop is also OK.

    d. I am sure that snmpd is not in infinite loop status when this problem occur.

 

2. when this problem occur ?

    a. send a lot of snmpset pdu periodically to sub-agent 'A'. (I think this item is necessary

        to make this problem)

    b. Also, send a lot of snmp getnext/getbulk pdu to sub-agent 'A'

    c. do 'link down' and 'link up' between master agent and sub-agent 'A' continuously.

        This causes master agent calls close_agentx_session() and open_agentx_session()

        periodically. (in 'link down' status, no ping connection between master and sub-agent) 

    d. Then, some time, SNMP locking problem occur. (randomly) Some times, I have to

        do 'link down' and 'link up' more than 30 times to see this problem. But, some times,

        I see this problem with only 1 or 2 'link down' and 'link up' .

    e. mainly, this locking problem occur when do 'link down' command. (this means

        locking problem occur when snmpd calls close_agentx_session())

    f.  only one remedy to this locking problem is 'kill master agent, and restart it'

 

3. what did I do to solve this problem ?

    a. I removed following two lines from file agent/snmp_agent.c

        *** agent/snmp_agent.c

        *** line 3326

        int
        netsnmp_remove_delegated_requests_for_session(netsnmp_session *sess)
       {
           netsnmp_agent_session *asp;
           int count = 0;

 

           for (asp = agent_delegated_list; asp; asp = asp->next) {
           /*
            * check each request
            */
               netsnmp_request_info *request;
               for(request = asp->requests; request; request = request->next) {
                /*
                 * check session
                 */
                    netsnmp_assert(NULL!=request->subtree);
-                    if(request->subtree->session != sess)
-                        continue;

 

                     /*
                      * matched! mark request as done
                      */
                     netsnmp_request_set_error(request, SNMP_ERR_GENERR);
                     ++count;
                }
            }

      b. because of removing two lines above, it seems, locking problem is solved.

          but, this solution causes all delegated snmp set requests to any sub-agent

          return error when one of sub-agent disconnected. but, this new small bug

          is much better than SNMP locking problem.

 

 

So, is there somebody who has any idea about my SNMP locking problem ?

 

I am not sure my remedy is fine.

 

Thanks in advance.

 

Regards,

 

Won-Sik