Merge pull request #334 from Arceliar/bugfix

Bugfix

misc/run-twolink-test

@@ -0,0 +1,33 @@
+#!/bin/bash
+
+# Connects nodes in two namespaces by two links with different bandwidth (10mbit and 100mbit)
+
+ip netns add node1
+ip netns add node2
+
+ip link add veth11 type veth peer name veth21
+ip link set veth11 netns node1 up
+ip link set veth21 netns node2 up
+
+ip link add veth12 type veth peer name veth22
+ip link set veth12 netns node1 up
+ip link set veth22 netns node2 up
+
+ip netns exec node1 tc qdisc add dev veth11 root tbf rate 10mbit burst 8192 latency 1ms
+ip netns exec node2 tc qdisc add dev veth21 root tbf rate 10mbit burst 8192 latency 1ms
+
+ip netns exec node1 tc qdisc add dev veth12 root tbf rate 100mbit burst 8192 latency 1ms
+ip netns exec node2 tc qdisc add dev veth22 root tbf rate 100mbit burst 8192 latency 1ms
+
+echo '{AdminListen: "unix://node1.sock"}' | ip netns exec node1 env PPROFLISTEN=localhost:6060 ./yggdrasil -logging "info,warn,error,debug" -useconf &> node1.log &
+echo '{AdminListen: "unix://node2.sock"}' | ip netns exec node2 env PPROFLISTEN=localhost:6060 ./yggdrasil -logging "info,warn,error,debug" -useconf &> node2.log &
+
+echo "Started, to continue you should (possibly w/ sudo):"
+echo "kill" $(jobs -p)
+wait
+
+ip netns delete node1
+ip netns delete node2
+
+ip link delete veth11
+ip link delete veth12

src/yggdrasil/link.go

@@ -216,6 +216,8 @@ func (intf *linkInterface) handler() error {
 				case signalReady <- struct{}{}:
 				default:
 				}
+				intf.link.core.log.Debugf("Sending packet to %s: %s, source %s",
+					strings.ToUpper(intf.info.linkType), themString, intf.info.local)
 			}
 		}
 	}()
@@ -235,18 +237,21 @@ func (intf *linkInterface) handler() error {
 		recvTimer := time.NewTimer(recvTime)
 		defer util.TimerStop(recvTimer)
 		for {
+			intf.link.core.log.Debugf("State of %s: %s, source %s :: isAlive %t isReady %t sendTimerRunning %t recvTimerRunning %t",
+				strings.ToUpper(intf.info.linkType), themString, intf.info.local,
+				isAlive, isReady, sendTimerRunning, recvTimerRunning)
 			select {
 			case gotMsg, ok := <-signalAlive:
 				if !ok {
 					return
 				}
-				if !isAlive {
+				util.TimerStop(recvTimer)
+				recvTimerRunning = false
 				isAlive = true
 				if !isReady {
 					// (Re-)enable in the switch
-						isReady = true
 					intf.link.core.switchTable.idleIn <- intf.peer.port
-					}
+					isReady = true
 				}
 				if gotMsg && !sendTimerRunning {
 					// We got a message
@@ -255,6 +260,10 @@ func (intf *linkInterface) handler() error {
 					sendTimer.Reset(sendTime)
 					sendTimerRunning = true
 				}
+				if !gotMsg {
+					intf.link.core.log.Debugf("Received ack from %s: %s, source %s",
+						strings.ToUpper(intf.info.linkType), themString, intf.info.local)
+				}
 			case sentMsg, ok := <-signalSent:
 				// Stop any running ack timer
 				if !ok {
@@ -273,12 +282,13 @@ func (intf *linkInterface) handler() error {
 				if !ok {
 					return
 				}
-				if !isAlive || !isReady {
+				if !isAlive {
 					// Disable in the switch
 					isReady = false
 				} else {
 					// Keep enabled in the switch
 					intf.link.core.switchTable.idleIn <- intf.peer.port
+					isReady = true
 				}
 			case <-sendTimer.C:
 				// We haven't sent anything, so signal a send of a 0 packet to let them know we're alive

src/yggdrasil/switch.go

@@ -780,6 +780,7 @@ func (t *switchTable) doWorker() {
 	t.queues.bufs = make(map[string]switch_buffer) // Packets per PacketStreamID (string)
 	idle := make(map[switchPort]struct{})          // this is to deduplicate things
 	for {
+		t.core.log.Debugf("Switch state: idle = %d, buffers = %d", len(idle), len(t.queues.bufs))
 		select {
 		case bytes := <-t.packetIn:
 			// Try to send it somewhere (or drop it if it's corrupt or at a dead end)