Replicate peer endpoints and forget paths if we have them -- this allows two clusters to talk to each other, whereas forgetting all paths does not.

5 files changed, 58 insertions, 28 deletions

diff --git a/node/Cluster.cpp b/node/Cluster.cpp
index b2f3d585..0797d83d 100644
--- a/node/Cluster.cpp
+++ b/node/Cluster.cpp
@@ -210,22 +210,30 @@ void Cluster::handleIncomingStateMessage(const void *msg,unsigned int len)
 						}	break;
 
 						case STATE_MESSAGE_HAVE_PEER: {
-							try {
-								Identity id;
-								ptr += id.deserialize(dmsg,ptr);
-								if (id) {
-									RR->topology->saveIdentity(id);
-									{
-										Mutex::Lock _l2(_peerAffinities_m);
-										_PA &pa = _peerAffinities[id.address()];
-										pa.ts = RR->node->now();
-										pa.mid = fromMemberId;
-			 						}
-			 						TRACE("[%u] has %s",(unsigned int)fromMemberId,id.address().toString().c_str());
-			 					}
-							} catch ( ... ) {
-								// ignore invalid identities
-							}
+							Identity id;
+							InetAddress physicalAddress;
+							ptr += id.deserialize(dmsg,ptr);
+							ptr += physicalAddress.deserialize(dmsg,ptr);
+							if (id) {
+								// Forget any paths that we have to this peer at its address
+								if (physicalAddress) {
+									SharedPtr<Peer> myPeerRecord(RR->topology->getPeer(id.address()));
+									if (myPeerRecord)
+										myPeerRecord->removePathByAddress(physicalAddress);
+								}
+
+								// Always save identity to update file time
+								RR->topology->saveIdentity(id);
+
+								// Set peer affinity to its new home
+								{
+									Mutex::Lock _l2(_peerAffinities_m);
+									_PA &pa = _peerAffinities[id.address()];
+									pa.ts = RR->node->now();
+									pa.mid = fromMemberId;
+		 						}
+		 						TRACE("[%u] has %s @ %s",(unsigned int)fromMemberId,id.address().toString().c_str(),physicalAddress.toString().c_str());
+		 					}
 						}	break;
 
 						case STATE_MESSAGE_MULTICAST_LIKE: {
@@ -396,7 +404,7 @@ bool Cluster::sendViaCluster(const Address &fromPeerAddress,const Address &toPee
 	return true;
 }
 
-void Cluster::replicateHavePeer(const Identity &peerId)
+void Cluster::replicateHavePeer(const Identity &peerId,const InetAddress &physicalAddress)
 {
 	const uint64_t now = RR->node->now();
 	{	// Use peer affinity table to track our own last announce time for peers
@@ -405,7 +413,7 @@ void Cluster::replicateHavePeer(const Identity &peerId)
 		if (pa.mid != _id) {
 			pa.ts = now;
 			pa.mid = _id;
-		} else if ((now - pa.ts) >= ZT_CLUSTER_HAVE_PEER_ANNOUNCE_PERIOD) {
+		} else if ((now - pa.ts) < ZT_CLUSTER_HAVE_PEER_ANNOUNCE_PERIOD) {
 			return;
 		} else {
 			pa.ts = now;
@@ -415,6 +423,7 @@ void Cluster::replicateHavePeer(const Identity &peerId)
 	// announcement
 	Buffer<4096> buf;
 	peerId.serialize(buf,false);
+	physicalAddress.serialize(buf);
 	{
 		Mutex::Lock _l(_memberIds_m);
 		for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
diff --git a/node/Cluster.hpp b/node/Cluster.hpp
index 42a26c7f..c3367d57 100644
--- a/node/Cluster.hpp
+++ b/node/Cluster.hpp
@@ -117,6 +117,10 @@ public:
 		/**
 		 * Cluster member has this peer:
 		 *   <[...] binary serialized peer identity>
+		 *   <[...] binary serialized peer remote physical address>
+		 *
+		 * Clusters send this message when they learn a path to a peer. The
+		 * replicated physical address is the one learned.
 		 */
 		STATE_MESSAGE_HAVE_PEER = 2,
 
@@ -225,8 +229,9 @@ public:
 	 * Advertise to the cluster that we have this peer
 	 *
 	 * @param peerId Identity of peer that we have
+	 * @param physicalAddress Physical address of peer (from our POV)
 	 */
-	void replicateHavePeer(const Identity &peerId);
+	void replicateHavePeer(const Identity &peerId,const InetAddress &physicalAddress);
 
 	/**
 	 * Advertise a multicast LIKE to the cluster
diff --git a/node/Constants.hpp b/node/Constants.hpp
index bef1183a..4b06db44 100644
--- a/node/Constants.hpp
+++ b/node/Constants.hpp
@@ -317,7 +317,7 @@
 /**
  * Minimum delay between attempts to confirm new paths to peers (to avoid HELLO flooding)
  */
-#define ZT_MIN_PATH_CONFIRMATION_INTERVAL 5000
+#define ZT_MIN_PATH_CONFIRMATION_INTERVAL 1000
 
 /**
  * Interval between direct path pushes in milliseconds
@@ -350,7 +350,7 @@
 /**
  * Maximum number of endpoints to contact per address type (to limit pushes like GitHub issue #235)
  */
-#define ZT_PUSH_DIRECT_PATHS_MAX_ENDPOINTS_PER_TYPE 2
+#define ZT_PUSH_DIRECT_PATHS_MAX_ENDPOINTS_PER_TYPE 4
 
 /**
  * A test pseudo-network-ID that can be joined
diff --git a/node/Peer.cpp b/node/Peer.cpp
index 99e2156e..99eb32c7 100644
--- a/node/Peer.cpp
+++ b/node/Peer.cpp
@@ -140,13 +140,10 @@ void Peer::received(
 			_lastMulticastFrame = now;
 
 #ifdef ZT_ENABLE_CLUSTER
-		// If we're in cluster mode and there's a better endpoint, stop here and don't
-		// learn or confirm paths. Also reset any existing paths, since they should
-		// go there and no longer talk to us here.
-		if (redirectTo) {
-			_numPaths = 0;
+		// If we think this peer belongs elsewhere, don't learn this path or
+		// do other connection init stuff.
+		if (redirectTo)
 			return;
-		}
 #endif
 
 		if ((now - _lastAnnouncedTo) >= ((ZT_MULTICAST_LIKE_EXPIRE / 2) - 1000)) {
@@ -206,7 +203,7 @@ void Peer::received(
 
 #ifdef ZT_ENABLE_CLUSTER
 	if ((RR->cluster)&&(pathIsConfirmed))
-		RR->cluster->replicateHavePeer(_id);
+		RR->cluster->replicateHavePeer(_id,remoteAddr);
 #endif
 
 	if (needMulticastGroupAnnounce) {
diff --git a/node/Peer.hpp b/node/Peer.hpp
index aa75b3f4..69343f20 100644
--- a/node/Peer.hpp
+++ b/node/Peer.hpp
@@ -413,6 +413,25 @@ public:
 	void clean(const RuntimeEnvironment *RR,uint64_t now);
 
 	/**
+	 * Remove all paths with this remote address
+	 *
+	 * @param addr Remote address to remove
+	 */
+	inline void removePathByAddress(const InetAddress &addr)
+	{
+		Mutex::Lock _l(_lock);
+		unsigned int np = _numPaths;
+		unsigned int x = 0;
+		unsigned int y = 0;
+		while (x < np) {
+			if (_paths[x].address() != addr)
+				_paths[y++] = _paths[x];
+			++x;
+		}
+		_numPaths = y;
+	}
+
+	/**
 	 * Find a common set of addresses by which two peers can link, if any
 	 *
 	 * @param a Peer A