Fully launch inner experiment (was just preloading cause of frisbee problems).

Do not die when turning firewall rules back on fails. This is a transient
error I do not understand yet.
When firewalled and paniced, skip clean shutdown of inner nodes since
they are going to be powered off anyway later, and besides, the control
network is shut off, so no way to talk to inner boss anyway.

tbsetup/elabinelab.in

@@ -397,7 +397,7 @@ if (defined($elabinelab_eid)) {
     print "Starting experiment $pid/$elabinelab_eid on inner emulab.\n";
     TBDebugTimeStamp("Starting inner experiment");
     system("$SSH -host $bossnode 'sudo -u $dbuid /usr/testbed/bin/batchexp ".
-	   "  -q -i $optarg -f -S \"ElabInElab Experiment\" ".
+	   "  -q -i $optarg -S \"ElabInElab Experiment\" ".
 	   "  -L \"ElabInElab ElabInElab\" -E \"ElabInElab Experiment\" ".
 	   "  -p $pid -e $elabinelab_eid $nsfilename'");
     
@@ -413,8 +413,8 @@ if ($firewalled) {
     $UID = 0;
     system("$SSH -host $firewall ipfw delete 1");
     if ($?) {
-	die("*** $0:\n".
-	    "    Error turning on firewall rules ($firewall)!\n");
+	print STDERR "*** $0:\n".
+	             "    Error turning back on firewall rules ($firewall)!\n";
     }
     $UID = $SAVEUID;
 }
@@ -739,6 +739,15 @@ sub TearDownEmulab()
     my $tbdir      = "/usr/testbed";
     my $wap        = "$tbdir/sbin/withadminprivs";
     my $nodereboot = "$tbdir/bin/node_reboot";
+    my $paniced;
+
+    #
+    # If firewalled, check to see if paniced. Right now that means the nodes
+    # are going to be powered off, so need to do the clean shutdown dance.
+    # 
+    if ($firewalled) {
+	TBExptGetPanicBit($pid, $eid, \$paniced);
+    }
         
     #
     # We want to rebuild the DHCPD file so that when we reboot the inner nodes
@@ -771,6 +780,14 @@ sub TearDownEmulab()
     }
     $EUID = 0;
 
+    #
+    # If panic set, just return; nodes are going to be powered down.
+    #
+    if ($firewalled and $paniced) {
+	print "Skipping clean shutdown cause panic button pressed.\n";
+	return 0;
+    }
+
     #
     # When the nodes reboot, we want them to do something reasonable. We
     # have no idea what is loaded on the disk, so they should go into an