Fix agent crash in FlowHandler::Run

Issue:
------
for a flow on re-evaluation if it moves from non-NAT to
NAT, instead of marking it as short flow and getting
removed it was getting re-enqueued to partition 0 with
IPC pointer being reset, causing NULL ptr access while
processing the re-enqueued message

Fix:
----
do not allow re-enqueuing of a flow update message

Closes-Bug: 1576818
Change-Id: I96a42eda7ae941da022915fd3bfe8c114a1fe437

src/vnsw/agent/pkt/flow_handler.cc

@@ -61,11 +61,16 @@ bool FlowHandler::Run() {
     PktFlowInfo info(agent_, pkt_info_,
                      flow_proto_->GetTable(flow_table_index_));
     std::auto_ptr<FlowTaskMsg> ipc;
+    bool allow_reentrant = true;
 
     if (pkt_info_->type == PktType::INVALID) {
         info.SetPktInfo(pkt_info_);
         info.l3_flow = pkt_info_->l3_forwarding = IsL3ModeFlow();
     } else if (pkt_info_->type == PktType::MESSAGE) {
+        // we don't allow reentrancy to different partition if it is
+        // a reevaluation for an existing flow which will only exist
+        // in this partition
+        allow_reentrant = false;
         ipc = std::auto_ptr<FlowTaskMsg>(static_cast<FlowTaskMsg *>(pkt_info_->ipc));
         pkt_info_->ipc = NULL;
         FlowEntry *fe = ipc->fe_ptr.get();
@@ -115,8 +120,8 @@ bool FlowHandler::Run() {
 
     // Flows that change port-numbers are always processed in thread-0.
     // Identify flows that change port and enqueue to thread-0
-    if (((pkt_info_->sport != info.nat_sport) ||
-         (pkt_info_->dport != info.nat_dport))) {
+    if (allow_reentrant && ((pkt_info_->sport != info.nat_sport) ||
+                            (pkt_info_->dport != info.nat_dport))) {
         if ((info.nat_sport != 0 || info.nat_dport != 0)) {
             if (flow_table_index_ != FlowTable::kPortNatFlowTableInstance) {
                 // Enqueue flow evaluation to