X-Git-Url: http://git.sameswireless.fr/l2tpns.git/blobdiff_plain/a8434db159038c2fb6b0fb12a28105f3ccba64c9..071f1d77ebc8cd92cc4896ad10959234910c2b7e:/cluster.c?ds=inline

diff --git a/cluster.c b/cluster.c
index 19666cd..9cead15 100644
--- a/cluster.c
+++ b/cluster.c
@@ -1,8 +1,12 @@
 // L2TPNS Clustering Stuff
 
-char const *cvs_id_cluster = "$Id: cluster.c,v 1.5 2004/07/02 07:30:43 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.36 2005/05/07 10:14:33 bodea Exp $";
 
 #include <stdio.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <unistd.h>
+#include <inttypes.h>
 #include <sys/file.h>
 #include <sys/stat.h>
 #include <sys/socket.h>
@@ -13,10 +17,6 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.5 2004/07/02 07:30:43 bodea Exp
 #include <string.h>
 #include <malloc.h>
 #include <errno.h>
-#include <stdlib.h>
-#include <stdarg.h>
-#include <unistd.h>
-#include <stdio.h>
 #include <libcli.h>
 
 #include "l2tpns.h"
@@ -38,13 +38,12 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.5 2004/07/02 07:30:43 bodea Exp
  */
 
 // Module variables.
-int cluster_sockfd = 0;		// The filedescriptor for the cluster communications port.
+int cluster_sockfd = 0;			// The filedescriptor for the cluster communications port.
 
-ipt my_address = 0;		// The network address of my ethernet port.
+in_addr_t my_address = 0;		// The network address of my ethernet port.
 static int walk_session_number = 0;	// The next session to send when doing the slow table walk.
 static int walk_tunnel_number = 0;	// The next tunnel to send when doing the slow table walk.
-
-static int hsess, fsess;	// Saved copies of the highest used session id, and the first free one.
+int forked = 0;				// Sanity check: CLI must not diddle with heartbeat table
 
 #define MAX_HEART_SIZE (8192)	// Maximum size of heartbeat packet. Must be less than max IP packet size :)
 #define MAX_CHANGES  (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2)	// Assumes a session is the biggest type!
@@ -62,16 +61,15 @@ static struct {
 				// we can re-transmit if needed.
 
 static struct {
-	u32 peer;
-	time_t	basetime;
-	clockt	timestamp;
-	int	uptodate;
+	in_addr_t peer;
+	uint32_t basetime;
+	clockt timestamp;
+	int uptodate;
 } peers[CLUSTER_MAX_SIZE];	// List of all the peers we've heard from.
 static int num_peers;		// Number of peers in list.
-static int have_peers;		// At least one peer
 
-int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
-int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
+static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
 
 //
 // Create a listening socket
@@ -80,65 +78,71 @@ int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
 //
 int cluster_init()
 {
-    struct sockaddr_in addr;
-    struct sockaddr_in interface_addr;
-    struct ip_mreq mreq;
-    struct ifreq   ifr;
-    int opt = 0;
+	struct sockaddr_in addr;
+	struct sockaddr_in interface_addr;
+	struct ip_mreq mreq;
+	struct ifreq   ifr;
+	int opt;
 
-    config->cluster_undefined_sessions = MAXSESSION-1;
-    config->cluster_undefined_tunnels = MAXTUNNEL-1;
+	config->cluster_undefined_sessions = MAXSESSION-1;
+	config->cluster_undefined_tunnels = MAXTUNNEL-1;
 
-    if (!config->cluster_address)
-	return 0;
-    if (!*config->cluster_interface)
-	return 0;
+	if (!config->cluster_address)
+		return 0;
+	if (!*config->cluster_interface)
+		return 0;
 
-    cluster_sockfd = socket(AF_INET, SOCK_DGRAM, UDP);
+	cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
 
-    memset(&addr, 0, sizeof(addr));
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_addr.s_addr = INADDR_ANY;
-    setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
+	memset(&addr, 0, sizeof(addr));
+	addr.sin_family = AF_INET;
+	addr.sin_port = htons(CLUSTERPORT);
+	addr.sin_addr.s_addr = INADDR_ANY;
+	setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
 
-    if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
-    {
-	log(0, 0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
-	return -1;
-    }
+	opt = fcntl(cluster_sockfd, F_GETFL, 0);
+	fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK);
 
-    strcpy(ifr.ifr_name, config->cluster_interface);
-    if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0) {
-	log(0, 0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
-	return -1;
-    }
+	if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
+	{
+		LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
+		return -1;
+	}
 
-    memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr) );
-    my_address = interface_addr.sin_addr.s_addr;
+	strcpy(ifr.ifr_name, config->cluster_interface);
+	if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0)
+	{
+		LOG(0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
+		return -1;
+	}
 
-		// Join multicast group.
-    mreq.imr_multiaddr.s_addr = config->cluster_address;
-    mreq.imr_interface = interface_addr.sin_addr;
+	memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr));
+	my_address = interface_addr.sin_addr.s_addr;
 
+	// Join multicast group.
+	mreq.imr_multiaddr.s_addr = config->cluster_address;
+	mreq.imr_interface = interface_addr.sin_addr;
 
-    opt = 0;	// Turn off multicast loopback.
-    setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));
 
-    if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0) {
-	log(0, 0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
-	return -1;
-    }
+	opt = 0;	// Turn off multicast loopback.
+	setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));
 
-    if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0) {
-	log(0, 0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
-	return -1;
-    }
+	if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0)
+	{
+		LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
+		return -1;
+	}
+
+	if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
+	{
+		LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
+		return -1;
+	}
 
-    config->cluster_last_hb = TIME;
-    config->cluster_seq_number = -1;
+	config->cluster_last_hb = TIME;
+	config->cluster_seq_number = -1;
 
-    return cluster_sockfd;
+	return cluster_sockfd;
 }
 
 
@@ -147,28 +151,26 @@ int cluster_init()
 // address ).
 //
 
-int cluster_send_data(void *data, int datalen)
+static int cluster_send_data(void *data, int datalen)
 {
-    struct sockaddr_in addr = {0};
-
-    if (!cluster_sockfd) return -1;
-    if (!config->cluster_address) return 0;
+	struct sockaddr_in addr = {0};
 
-    addr.sin_addr.s_addr = config->cluster_address;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_family = AF_INET;
+	if (!cluster_sockfd) return -1;
+	if (!config->cluster_address) return 0;
 
-//    log_hex(4, "Cluster send", data, datalen); // VERY big data packets. How about we don't..
+	addr.sin_addr.s_addr = config->cluster_address;
+	addr.sin_port = htons(CLUSTERPORT);
+	addr.sin_family = AF_INET;
 
-    log(5,0,0,0, "Cluster send data: %d bytes\n", datalen);
+	LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen);
 
-    if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
-    {
-	log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
-	return -1;
-    }
+	if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
+	{
+		LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
+		return -1;
+	}
 
-    return 0;
+	return 0;
 }
 
 //
@@ -176,21 +178,33 @@ int cluster_send_data(void *data, int datalen)
 // Maintains the format. Assumes that the caller
 // has passed in a big enough buffer!
 //
-static void add_type(char ** p, int type, int more, char * data, int size)
+static void add_type(char **p, int type, int more, char *data, int size)
 {
-	* ( (u32*)(*p) ) = type;
-	*p += sizeof(u32);
+	*((uint32_t *) (*p)) = type;
+	*p += sizeof(uint32_t);
 
-	* ( (u32*)(*p) ) = more;
-	*p += sizeof(u32);
+	*((uint32_t *)(*p)) = more;
+	*p += sizeof(uint32_t);
 
 	if (data && size > 0) {
 		memcpy(*p, data, size);
-		(*p) += size;
+		*p += size;
 	}
 }
 
-void cluster_uptodate(void)
+// advertise our presence via BGP or gratuitous ARP
+static void advertise(void)
+{
+#ifdef BGP
+	if (bgp_configured)
+		bgp_enable_routing(1);
+	else
+#endif /* BGP */
+		if (config->send_garp)
+			send_garp(config->bind_address);	// Start taking traffic.
+}
+
+static void cluster_uptodate(void)
 {
 	if (config->cluster_iam_uptodate)
 		return;
@@ -200,60 +214,48 @@ void cluster_uptodate(void)
 
 	config->cluster_iam_uptodate = 1;
 
-	log(0,0,0,0, "Now uptodate with master.\n");
-
-		// If we're not a master, or if we have no slaves
-		// then start taking traffic..
-	if (!config->cluster_iam_master || !have_peers)
-	{
-#ifdef BGP
-		if (bgp_configured)
-			bgp_enable_routing(1);
-		else
-#endif /* BGP */
-			if (config->send_garp)
-				send_garp(config->bind_address);	// Start taking traffic.
-	}
+	LOG(0, 0, 0, "Now uptodate with master.\n");
+	advertise();
 }
 
 //
 // Send a unicast UDP packet to a peer with 'data' as the
 // contents.
 //
-int peer_send_data(u32 peer, char * data, int size)
+static int peer_send_data(in_addr_t peer, char *data, int size)
 {
-    struct sockaddr_in addr = {0};
+	struct sockaddr_in addr = {0};
 
-    if (!cluster_sockfd) return -1;
-    if (!config->cluster_address) return 0;
+	if (!cluster_sockfd) return -1;
+	if (!config->cluster_address) return 0;
 
-    if (!peer)	// Odd??
-	return -1;
+	if (!peer)	// Odd??
+		return -1;
 
-    addr.sin_addr.s_addr = peer;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_family = AF_INET;
+	addr.sin_addr.s_addr = peer;
+	addr.sin_port = htons(CLUSTERPORT);
+	addr.sin_family = AF_INET;
 
-    log_hex(5, "Peer send", data, size);
+	LOG_HEX(5, "Peer send", data, size);
 
-    if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
-    {
-	log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
-	return -1;
-    }
+	if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
+	{
+		LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
+		return -1;
+	}
 
-    return 0;
+	return 0;
 }
 
 //
 // Send a structured message to a peer with a single element of type 'type'.
 //
-int peer_send_message(u32 peer, int type, int more, char * data, int size)
+static int peer_send_message(in_addr_t peer, int type, int more, char *data, int size)
 {
 	char buf[65536];	// Vast overkill.
-	char * p = buf;
+	char *p = buf;
 
-	log(4,0,0,0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
+	LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
 	add_type(&p, type, more, data, size);
 
 	return peer_send_data(peer, buf, (p-buf) );
@@ -263,25 +265,24 @@ int peer_send_message(u32 peer, int type, int more, char * data, int size)
 // Forward a state changing packet to the master.
 //
 // The master just processes the payload as if it had
-// received it off the tap device.
+// received it off the tun device.
 //
-int master_forward_packet(char * data, int size, u32 addr, int port)
+int master_forward_packet(char *data, int size, in_addr_t addr, int port)
 {
 	char buf[65536];	// Vast overkill.
-	char * p = buf;
+	char *p = buf;
 
 	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
 		return -1;
 
-	log(4,0,0,0,	"Forwarding packet from %s to master (size %d)\n", inet_toa(addr), size);
+	LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
 
 	STAT(c_forwarded);
-	add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) );
+	add_type(&p, C_FORWARD, addr, (char *) &port, sizeof(port));
 	memcpy(p, data, size);
 	p += size;
 
-	return peer_send_data(config->cluster_master_address, buf, (p-buf) );
-
+	return peer_send_data(config->cluster_master_address, buf, (p - buf));
 }
 
 //
@@ -291,15 +292,15 @@ int master_forward_packet(char * data, int size, u32 addr, int port)
 // token bucket queue, and lets normal processing take care
 // of it.
 //
-int master_throttle_packet(int tbfid, char * data, int size)
+int master_throttle_packet(int tbfid, char *data, int size)
 {
 	char buf[65536];	// Vast overkill.
-	char * p = buf;
+	char *p = buf;
 
 	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
 		return -1;
 
-	log(4,0,0,0,	"Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
+	LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
 
 	add_type(&p, C_THROTTLE, tbfid, data, size);
 
@@ -324,7 +325,7 @@ int master_garden_packet(sessionidt s, char *data, int size)
 	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
 		return -1;
 
-	log(4,0,0,0,	"Walled garden packet to master (size %d)\n", size);
+	LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
 
 	add_type(&p, C_GARDEN, s, data, size);
 
@@ -336,20 +337,21 @@ int master_garden_packet(sessionidt s, char *data, int size)
 // Send a chunk of data as a heartbeat..
 // We save it in the history buffer as we do so.
 //
-static void send_heartbeat(int seq, char * data, int size)
+static void send_heartbeat(int seq, char *data, int size)
 {
-    int i;
-
-    if (size > sizeof(past_hearts[0].data)) {
-	log(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
-	kill(0, SIGTERM);
-	exit(1);
-    }
-    i = seq % HB_HISTORY_SIZE;
-    past_hearts[i].seq = seq;
-    past_hearts[i].size = size;
-    memcpy(&past_hearts[i].data, data, size);	// Save it.
-    cluster_send_data(data, size);
+	int i;
+
+	if (size > sizeof(past_hearts[0].data))
+	{
+		LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
+		kill(0, SIGTERM);
+		exit(1);
+	}
+	i = seq % HB_HISTORY_SIZE;
+	past_hearts[i].seq = seq;
+	past_hearts[i].size = size;
+	memcpy(&past_hearts[i].data, data, size);	// Save it.
+	cluster_send_data(data, size);
 }
 
 //
@@ -364,7 +366,7 @@ void cluster_send_ping(time_t basetime)
 	if (config->cluster_iam_master && basetime)		// We're heartbeating so no need to ping.
 		return;
 
-	log(5,0,0,0, "Sending cluster ping...\n");
+	LOG(5, 0, 0, "Sending cluster ping...\n");
 
 	x.ver = 1;
 	x.addr = config->bind_address;
@@ -377,12 +379,12 @@ void cluster_send_ping(time_t basetime)
 
 //
 // Walk the session counters looking for non-zero ones to send
-// to the master. We send up to 100 of them at one time.
-// We examine a maximum of 2000 sessions.
+// to the master. We send up to 600 of them at one time.
+// We examine a maximum of 3000 sessions.
 // (50k max session should mean that we normally
 // examine the entire session table every 25 seconds).
 
-#define MAX_B_RECS (400)
+#define MAX_B_RECS (600)
 void master_update_counts(void)
 {
 	int i, c;
@@ -404,18 +406,18 @@ void master_update_counts(void)
 		if ( walk_session_number > config->cluster_highest_sessionid)
 			walk_session_number = 1;
 
-		if (!sess_count[walk_session_number].cin && !sess_count[walk_session_number].cout)
+		if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
 			continue; // Unused. Skip it.
 
 		b[c].sid = walk_session_number;
-		b[c].in = sess_count[walk_session_number].cin;
-		b[c].out = sess_count[walk_session_number].cout;
+		b[c].in = sess_local[walk_session_number].cin;
+		b[c].out = sess_local[walk_session_number].cout;
 
 		if (++c > MAX_B_RECS)	// Send a max of 400 elements in a packet.
 			break;
 
 			// Reset counters.
-		sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
+		sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
 	}
 
 	if (!c)		// Didn't find any that changes. Get out of here!
@@ -423,60 +425,112 @@ void master_update_counts(void)
 
 
 			// Forward the data to the master.
-	log(4,0,0,0, "Sending byte counters to master (%d elements)\n", c);
-	peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
+	LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
+	peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);
 	return;
 }
 
+//
+// On the master, check how our slaves are going. If
+// one of them's not up-to-date we'll heartbeat faster.
+// If we don't have any of them, then we need to turn
+// on our own packet handling!
+//
+void cluster_check_slaves(void)
+{
+	int i;
+	static int have_peers = 0;
+	int had_peers = have_peers;
+	clockt t = TIME;
+
+	if (!config->cluster_iam_master)
+		return;		// Only runs on the master...
+
+	config->cluster_iam_uptodate = 1;	// cleared in loop below
+
+	for (i = have_peers = 0; i < num_peers; i++)
+	{
+		if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
+			continue;	// Stale peer! Skip them.
+
+		if (!peers[i].basetime)
+			continue;	// Shutdown peer! Skip them.
+
+		if (peers[i].uptodate)
+			have_peers = 1;
+
+		if (!peers[i].uptodate)
+			config->cluster_iam_uptodate = 0; // Start fast heartbeats
+	}
+
+#ifdef BGP
+	// in a cluster, withdraw/add routes when we get a peer/lose all peers
+	if (bgp_configured && have_peers != had_peers)
+		bgp_enable_routing(!have_peers);
+#endif /* BGP */
+}
+
 //
 // Check that we have a master. If it's been too
 // long since we heard from a master then hold an election.
 //
 void cluster_check_master(void)
 {
-	int i, count, tcount, high_sid = 0;
+	int i, count, tcount, high_unique_id = 0;
 	int last_free = 0;
-	int had_peers = have_peers;
 	clockt t = TIME;
+	static int probed = 0;
 
-	if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout) )
-		return;		// Everything's ok. return.
+	if (config->cluster_iam_master)
+		return;		// Only runs on the slaves...
 
-	if (!config->cluster_iam_master)
-		log(0,0,0,0, "Master timed out! Holding election...\n");
+	// If the master is late (missed 2 hearbeats by a second and a
+	// hair) it may be that the switch has dropped us from the
+	// multicast group, try unicasting one probe to the master
+	// which will hopefully respond with a unicast heartbeat that
+	// will allow us to limp along until the querier next runs.
+	if (TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
+	{
+		if (!probed && config->cluster_master_address)
+		{
+			probed = 1;
+			LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
+				0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
 
-	config->cluster_last_hb = TIME + 1;
+			peer_send_message(config->cluster_master_address,
+				C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+		}
+	} else {	// We got a recent heartbeat; reset the probe flag.
+		probed = 0;
+	}
+
+	if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
+		return;	// Everything's ok!
+
+	config->cluster_last_hb = TIME + 1;	// Just the one election thanks.
 
-	for (i = have_peers = 0; i < num_peers ; ++i) {
+	LOG(0, 0, 0, "Master timed out! Holding election...\n");
+
+	for (i = 0; i < num_peers; i++)
+	{
 		if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
 			continue;	// Stale peer! Skip them.
 
 		if (!peers[i].basetime)
 			continue;	// Shutdown peer! Skip them.
 
-		have_peers = 1;
 		if (peers[i].basetime < basetime) {
-			log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
+			LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
 			return;		// They'll win the election. Get out of here.
 		}
 
 		if (peers[i].basetime == basetime &&
 			peers[i].peer > my_address) {
-			log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
+			LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
 			return;		// They'll win the election. Wait for them to come up.
 		}
 	}
 
-	if (config->cluster_iam_master)		// If we're the master, we've already won
-	{
-#ifdef BGP
-		// master lost all slaves, need to handle traffic ourself
-		if (bgp_configured && had_peers && !have_peers)
-			bgp_enable_routing(1);
-#endif /* BGP */
-		return;
-	}
-
 		// Wow. it's been ages since I last heard a heartbeat
 		// and I'm better than an of my peers so it's time
 		// to become a master!!!
@@ -484,12 +538,7 @@ void cluster_check_master(void)
 	config->cluster_iam_master = 1;
 	config->cluster_master_address = 0;
 
-	log(0,0,0,0, "I am declaring myself the master!\n");
-
-#ifdef BGP
-	if (bgp_configured && have_peers)
-		bgp_enable_routing(0); /* stop handling traffic */
-#endif /* BGP */
+	LOG(0, 0, 0, "I am declaring myself the master!\n");
 
 	if (config->cluster_seq_number == -1)
 		config->cluster_seq_number = 0;
@@ -514,7 +563,7 @@ void cluster_check_master(void)
 		// Re-string the free list.
 		// Find the ID of the highest session.
 	last_free = 0;
-	high_sid = 0;
+	high_unique_id = 0;
 	config->cluster_highest_sessionid = 0;
 	for (i = 0, count = 0; i < MAXSESSION; ++i) {
 		if (session[i].tunnel == T_UNDEF) {
@@ -522,45 +571,42 @@ void cluster_check_master(void)
 			++count;
 		}
 
-		if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
+		if (!session[i].opened) { // Unused session. Add to free list.
+			memset(&session[i], 0, sizeof(session[i]));
+			session[i].tunnel = T_FREE;
 			session[last_free].next = i;
 			session[i].next = 0;
 			last_free = i;
+			continue;
 		}
 
-			// Reset all the idle timeouts..
+			// Reset idle timeouts..
 		session[i].last_packet = time_now;
 
-			// Accumulate un-sent byte counters.
-		session[i].cin += sess_count[i].cin;
-		session[i].cout += sess_count[i].cout;
-		session[i].total_cin += sess_count[i].cin;
-		session[i].total_cout += sess_count[i].cout;
+			// Reset die relative to our uptime rather than the old master's
+		if (session[i].die) session[i].die = TIME;
 
-		sess_count[i].cin = sess_count[i].cout = 0;
+			// Accumulate un-sent byte counters.
+		session[i].cin += sess_local[i].cin;
+		session[i].cout += sess_local[i].cout;
+		session[i].total_cin += sess_local[i].cin;
+		session[i].total_cout += sess_local[i].cout;
 
-		session[i].radius = 0;	// Reset authentication as the radius blocks aren't up to date.
+		sess_local[i].cin = sess_local[i].cout = 0;
 
-		if (session[i].sid >= high_sid)	// This is different to the index into the session table!!!
-			high_sid = session[i].sid+1;
+		sess_local[i].radius = 0;	// Reset authentication as the radius blocks aren't up to date.
 
+		if (session[i].unique_id >= high_unique_id)	// This is different to the index into the session table!!!
+			high_unique_id = session[i].unique_id+1;
 
 		session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
-		throttle_session(i, session[i].throttle);
-
-// I'm unsure about this. --mo
-// It's potentially a good thing, but it could send a
-// LOT of packets.
-//		if (session[i].throttle)
-//			cluster_send_session(s);	// Tell the slaves about the new tbf indexes.
-
-		if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
-			config->cluster_highest_sessionid = i;
+		throttle_session(i, session[i].throttle_in, session[i].throttle_out);
 
+		config->cluster_highest_sessionid = i;
 	}
 
 	session[last_free].next = 0;	// End of chain.
-	last_sid = high_sid;		// Keep track of the highest used session ID.
+	last_id = high_unique_id;	// Keep track of the highest used session ID.
 
 	become_master();
 
@@ -568,16 +614,18 @@ void cluster_check_master(void)
 
 		// If we're not the very first master, this is a big issue!
 	if(count>0)
-		log(0,0,0,0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
+		LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
 
 	config->cluster_undefined_sessions = 0;
 	config->cluster_undefined_tunnels = 0;
+	config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
 
-		//
-		// FIXME. We need to fix up the tunnel control message
-		// queue here! There's a number of other variables we
-		// should also update.
-	cluster_uptodate();
+	if (!num_peers) // lone master
+		advertise();
+
+	// FIXME. We need to fix up the tunnel control message
+	// queue here! There's a number of other variables we
+	// should also update.
 }
 
 
@@ -606,12 +654,12 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
 	config->cluster_undefined_sessions = 0;
 	for (i = 1 ; i < MAXSESSION; ++i) {
 		if (i > highsession) {
-			session[i].tunnel = 0; // Defined.
+			if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
 			continue;
 		}
-		if (session[i].tunnel != T_UNDEF)
-			continue;
-		++config->cluster_undefined_sessions;
+
+		if (session[i].tunnel == T_UNDEF)
+			++config->cluster_undefined_sessions;
 	}
 
 		// Clear out defined tunnels, counting the number of
@@ -619,17 +667,17 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
 	config->cluster_undefined_tunnels = 0;
 	for (i = 1 ; i < MAXTUNNEL; ++i) {
 		if (i > hightunnel) {
-			tunnel[i].state = TUNNELFREE; // Defined.
+			if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
 			continue;
 		}
-		if (tunnel[i].state != TUNNELUNDEF)
-			continue;
-		++config->cluster_undefined_tunnels;
+
+		if (tunnel[i].state == TUNNELUNDEF)
+			++config->cluster_undefined_tunnels;
 	}
 
 
 	if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
-		log(2,0,0,0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
+		LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
 			config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
 		return;
 	}
@@ -640,13 +688,13 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
 		cluster_uptodate();
 }
 
-int hb_add_type(char **p, int type, int id)
+static int hb_add_type(char **p, int type, int id)
 {
 	switch (type) {
 		case C_CSESSION: { // Compressed C_SESSION.
-			u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
-			u8 *d = (u8 *) &session[id];
-			u8 *orig = d;
+			uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
+			uint8_t *d = (uint8_t *) &session[id];
+			uint8_t *orig = d;
 			int size;
 
 			size = rle_compress( &d,  sizeof(sessiont), c, sizeof(c) );
@@ -654,19 +702,19 @@ int hb_add_type(char **p, int type, int id)
 				// Did we compress the full structure, and is the size actually
 				// reduced??
 			if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
-				add_type(p, C_CSESSION, id, (char*) c, size);
+				add_type(p, C_CSESSION, id, (char *) c, size);
 				break;
 			}
 			// Failed to compress : Fall through.
 		}
 		case C_SESSION: add_type(p, C_SESSION, id,
-			(char*) &session[id], sizeof(sessiont));
+			(char *) &session[id], sizeof(sessiont));
 			break;
 
 		case C_CTUNNEL: { // Compressed C_TUNNEL
-			u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
-			u8 *d = (u8 *) &tunnel[id];
-			u8 *orig = d;
+			uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
+			uint8_t *d = (uint8_t *) &tunnel[id];
+			uint8_t *orig = d;
 			int size;
 
 			size = rle_compress( &d,  sizeof(tunnelt), c, sizeof(c) );
@@ -680,11 +728,12 @@ int hb_add_type(char **p, int type, int id)
 			// Failed to compress : Fall through.
 		}
 		case C_TUNNEL: add_type(p, C_TUNNEL, id,
-			(char*) &tunnel[id], sizeof(tunnelt));
+			(char *) &tunnel[id], sizeof(tunnelt));
 			break;
 		default:
-			log(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
+			LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
 			kill(0, SIGTERM);
+			exit(1);
 	}
 	return 0;
 }
@@ -692,54 +741,60 @@ int hb_add_type(char **p, int type, int id)
 //
 // Send a heartbeat, incidently sending out any queued changes..
 //
-void cluster_heartbeat(int highsession, int freesession, int hightunnel)
+void cluster_heartbeat()
 {
 	int i, count = 0, tcount = 0;
 	char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
 	heartt h;
-	char * p = buff;
+	char *p = buff;
 
 	if (!config->cluster_iam_master)	// Only the master does this.
 		return;
 
-	hsess = highsession;
-	fsess = freesession;
-		// Fill out the heartbeat header.
+	config->cluster_table_version += config->cluster_num_changes;
+
+	// Fill out the heartbeat header.
+	memset(&h, 0, sizeof(h));
+
 	h.version = HB_VERSION;
 	h.seq = config->cluster_seq_number;
 	h.basetime = basetime;
 	h.clusterid = config->bind_address;	// Will this do??
 	h.basetime = basetime;
-	h.highsession = highsession;
-	h.freesession = freesession;
-	h.hightunnel = hightunnel;
+	h.highsession = config->cluster_highest_sessionid;
+	h.freesession = sessionfree;
+	h.hightunnel = config->cluster_highest_tunnelid;
 	h.size_sess = sizeof(sessiont);		// Just in case.
 	h.size_tunn = sizeof(tunnelt);
+	h.interval = config->cluster_hb_interval;
+	h.timeout  = config->cluster_hb_timeout;
+	h.table_version = config->cluster_table_version;
 
-	add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h) );
+	add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
 
 	for (i = 0; i < config->cluster_num_changes; ++i) {
 		hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
 	}
 
 	if (p > (buff + sizeof(buff))) {	// Did we somehow manage to overun the buffer?
-		log(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
+		LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
 		kill(0, SIGTERM);
+		exit(1);
 	}
 
 		//
 		// Fill out the packet with sessions from the session table...
 		// (not forgetting to leave space so we can get some tunnels in too )
-	while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
+	while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
 
 		if (!walk_session_number)	// session #0 isn't valid.
 			++walk_session_number;
 
-		if (count >= highsession)	// If we're a small cluster, don't go wild.
+		if (count >= config->cluster_highest_sessionid)	// If we're a small cluster, don't go wild.
 			break;
 
 		hb_add_type(&p, C_CSESSION, walk_session_number);
-		walk_session_number = (1+walk_session_number)%(highsession+1);	// +1 avoids divide by zero.
+		walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1);	// +1 avoids divide by zero.
 
 		++count;			// Count the number of extra sessions we're sending.
 	}
@@ -750,7 +805,7 @@ void cluster_heartbeat(int highsession, int freesession, int hightunnel)
 		// than the session table. This is good because stuffing up a 
 		// tunnel is a much bigger deal than stuffing up a session.
 		//
-	while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
+	while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
 
 		if (!walk_tunnel_number)	// tunnel #0 isn't valid.
 			++walk_tunnel_number;
@@ -767,13 +822,16 @@ void cluster_heartbeat(int highsession, int freesession, int hightunnel)
 		//
 		// Did we do something wrong?
 	if (p > (buff + sizeof(buff))) {	// Did we somehow manage to overun the buffer?
-		log(0,0,0,0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
+		LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
 		kill(0, SIGTERM);
+		exit(1);
 	}
 
-	log(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
-			h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
-			config->cluster_highest_tunnelid, (p-buff));
+	LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
+		     "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
+	    HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
+	    count, tcount, config->cluster_highest_sessionid,
+	    config->cluster_highest_tunnelid, (int) (p - buff));
 
 	config->cluster_num_changes = 0;
 
@@ -785,7 +843,7 @@ void cluster_heartbeat(int highsession, int freesession, int hightunnel)
 //
 // A structure of type 'type' has changed; Add it to the queue to send.
 //
-int type_changed(int type, int id)
+static int type_changed(int type, int id)
 {
 	int i;
 
@@ -799,7 +857,7 @@ int type_changed(int type, int id)
 	++config->cluster_num_changes;
 
 	if (config->cluster_num_changes > MAX_CHANGES)
-		cluster_heartbeat(config->cluster_highest_sessionid, fsess, config->cluster_highest_tunnelid);
+		cluster_heartbeat(); // flush now
 
 	return 1;
 }
@@ -809,7 +867,12 @@ int type_changed(int type, int id)
 int cluster_send_session(int sid)
 {
 	if (!config->cluster_iam_master) {
-		log(0,0,sid,0, "I'm not a master, but I just tried to change a session!\n");
+		LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
+		return -1;
+	}
+
+	if (forked) {
+		LOG(0, sid, 0, "cluster_send_session called from child process!\n");
 		return -1;
 	}
 
@@ -820,7 +883,7 @@ int cluster_send_session(int sid)
 int cluster_send_tunnel(int tid)
 {
 	if (!config->cluster_iam_master) {
-		log(0,0,0,tid, "I'm not a master, but I just tried to change a tunnel!\n");
+		LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
 		return -1;
 	}
 
@@ -833,19 +896,19 @@ int cluster_send_tunnel(int tid)
 // missed a packet. We'll resend it every packet since
 // the last one it's seen.
 //
-int cluster_catchup_slave(int seq, u32 slave)
+static int cluster_catchup_slave(int seq, in_addr_t slave)
 {
 	int s;
 	int diff;
 
-	log(1,0,0,0, "Slave %s sent LASTSEEN with seq %d\n", inet_toa(slave), seq);
+	LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
 
 	diff = config->cluster_seq_number - seq;	// How many packet do we need to send?
 	if (diff < 0)
 		diff += HB_MAX_SEQ;
 
 	if (diff >= HB_HISTORY_SIZE) {	// Ouch. We don't have the packet to send it!
-		log(0,0,0,0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
+		LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
 			seq, config->cluster_seq_number);
 		return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
 	}
@@ -854,8 +917,8 @@ int cluster_catchup_slave(int seq, u32 slave)
 	while (seq != config->cluster_seq_number) {
 		s = seq%HB_HISTORY_SIZE;
 		if (seq != past_hearts[s].seq) {
-			log(0,0,0,0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
-				inet_toa(slave), seq, past_hearts[s].seq, s, config->cluster_seq_number);
+			LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
+				fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
 			return -1;	// What to do here!?
 		}
 		peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
@@ -868,24 +931,27 @@ int cluster_catchup_slave(int seq, u32 slave)
 // We've heard from another peer! Add it to the list
 // that we select from at election time.
 //
-int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
+static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
 {
 	int i;
-	u32 clusterid;
+	in_addr_t clusterid;
+	pingt p;
+
+	// Allow for backward compatability.
+	// Just the ping packet into a new structure to allow
+	// for the possibility that we might have received
+	// more or fewer elements than we were expecting.
+	if (size > sizeof(p))
+		size = sizeof(p);
 
-	clusterid = p->addr;
+	memset( (void *) &p, 0, sizeof(p) );
+	memcpy( (void *) &p, (void *) pp, size);
+
+	clusterid = p.addr;
 	if (clusterid != config->bind_address)
 	{
 		// Is this for us?
-		log(4,0,0,0, "Skipping ping from %s (different cluster)\n", inet_toa(peer));
-		return 0;
-	}
-
-		// Is this the master shutting down??
-	if (peer == config->cluster_master_address && !basetime) {
-		config->cluster_master_address = 0;
-		config->cluster_last_hb = 0; // Force an election.
-		cluster_check_master();
+		LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
 		return 0;
 	}
 
@@ -897,18 +963,29 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
 		// This peer already exists. Just update the timestamp.
 		peers[i].basetime = basetime;
 		peers[i].timestamp = TIME;
+		peers[i].uptodate = !p.undef;
 		break;
 	}
 
+	// Is this the master shutting down??
+	if (peer == config->cluster_master_address && !basetime) {
+		LOG(3, 0, 0, "Master %s shutting down...\n", fmtaddr(config->cluster_master_address, 0));
+		config->cluster_master_address = 0;
+		config->cluster_last_hb = 0; // Force an election.
+		cluster_check_master();
+		return 0;
+	}
+
 	if (i >= num_peers)
 	{
-		log(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));
+		LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
 
 		// Not found. Is there a stale slot to re-use?
 		for (i = 0; i < num_peers ; ++i)
 		{
-			if (peers[i].peer != peer)
-				continue;
+			if (!peers[i].basetime) // Shutdown
+				break;
+
 			if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
 				break;
 		}
@@ -916,27 +993,20 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
 		if (i >= CLUSTER_MAX_SIZE)
 		{
 			// Too many peers!!
-			log(0,0,0,0, "Tried to add %s as a peer, but I already have %d of them!\n", inet_toa(peer), i);
+			LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
 			return -1;
 		}
 
 		peers[i].peer = peer;
 		peers[i].basetime = basetime;
 		peers[i].timestamp = TIME;
+		peers[i].uptodate = !p.undef;
 		if (i == num_peers)
 			++num_peers;
 
-		log(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer), num_peers);
+		LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
 	}
 
-#ifdef BGP
-	/* drop routes if we've now got a peer */
-	if (bgp_configured && config->cluster_iam_master && !have_peers)
-		bgp_enable_routing(0);
-#endif /* BGP */
-
-	have_peers = 1;
-
 	return 1;
 }
 
@@ -945,20 +1015,20 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
 // Note that we don't mark the session as dirty; We rely on
 // the slow table walk to propogate this back out to the slaves.
 //
-int cluster_handle_bytes(char * data, int size)
+static int cluster_handle_bytes(char *data, int size)
 {
-	bytest * b;
+	bytest *b;
 
-	b = (bytest*) data;
+	b = (bytest *) data;
 
-	log(3,0,0,0, "Got byte counter update (size %d)\n", size);
+	LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
 
 				/* Loop around, adding the byte
 				counts to each of the sessions. */
 
 	while (size >= sizeof(*b) ) {
 		if (b->sid > MAXSESSION) {
-			log(0,0,0,0, "Got C_BYTES with session #%d!\n", b->sid);
+			LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
 			return -1; /* Abort processing */
 		}
 
@@ -967,14 +1037,16 @@ int cluster_handle_bytes(char * data, int size)
 
 		session[b->sid].cin += b->in;
 		session[b->sid].cout += b->out;
-		session[b->sid].last_packet = time_now; // Reset idle timer!
+
+		if (b->in)
+			session[b->sid].last_packet = time_now; // Reset idle timer!
 
 		size -= sizeof(*b);
 		++b;
 	}
 
 	if (size != 0)
-		log(0,0,0,0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
+		LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
 
 	return size;
 }
@@ -982,40 +1054,41 @@ int cluster_handle_bytes(char * data, int size)
 //
 // Handle receiving a session structure in a heartbeat packet.
 //
-static int cluster_recv_session(int more , u8 * p)
+static int cluster_recv_session(int more, uint8_t *p)
 {
 	if (more >= MAXSESSION) {
-		log(0,0,0,0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
+		LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
 		return -1;
 	}
 
 	if (session[more].tunnel == T_UNDEF) {
 		if (config->cluster_iam_uptodate) { // Sanity.
-			log(0,0,0,0, "I thought I was uptodate but I just found an undefined session!\n");
+			LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
 		} else {
 			--config->cluster_undefined_sessions;
 		}
 	}
 
-	load_session(more, (sessiont*) p);	// Copy session into session table..
+	load_session(more, (sessiont *) p);	// Copy session into session table..
 
-	log(5,0,more,0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
+	LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
 
 	if (!config->cluster_iam_uptodate)
 		cluster_uptodate();	// Check to see if we're up to date.
+
 	return 0;
 }
 
-static int cluster_recv_tunnel(int more, u8 *p)
+static int cluster_recv_tunnel(int more, uint8_t *p)
 {
 	if (more >= MAXTUNNEL) {
-		log(0,0,0,0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
+		LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
 		return -1;
 	}
 
 	if (tunnel[more].state == TUNNELUNDEF) {
 		if (config->cluster_iam_uptodate) { // Sanity.
-			log(0,0,0,0, "I thought I was uptodate but I just found an undefined tunnel!\n");
+			LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
 		} else {
 			--config->cluster_undefined_tunnels;
 		}
@@ -1030,7 +1103,7 @@ static int cluster_recv_tunnel(int more, u8 *p)
 	tunnel[more].controls = tunnel[more].controle = NULL;
 	tunnel[more].controlc = 0;
 
-	log(5,0,0,more, "Received tunnel update\n");
+	LOG(5, 0, more, "Received tunnel update\n");
 
 	if (!config->cluster_iam_uptodate)
 		cluster_uptodate();	// Check to see if we're up to date.
@@ -1040,24 +1113,31 @@ static int cluster_recv_tunnel(int more, u8 *p)
 
 
 //
-// Process a version one heartbeat..
+// Process a heartbeat..
 //
-static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u32 addr)
+// v3: added interval, timeout
+// v4: added table_version
+static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
 {
-	heartt * h;
+	heartt *h;
 	int s = size - (p-data);
 	int i, type;
 
-	if (more != HB_VERSION) {
-		log(0,0,0,0, "Received a heartbeat version that I don't understand!\n");
+#if HB_VERSION != 4
+# error "need to update cluster_process_heartbeat()"
+#endif
+
+	// we handle versions 3 through 4
+	if (more < 3 || more > HB_VERSION) {
+		LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
 		return -1; // Ignore it??
 	}
+
 		// Ok. It's a heartbeat packet from a cluster master!
 	if (s < sizeof(*h))
 		goto shortpacket;
 
-
-	h = (heartt*) p;
+	h = (heartt *) p;
 	p += sizeof(*h);
 	s -= sizeof(*h);
 
@@ -1067,21 +1147,39 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
 	if (config->cluster_iam_master) {	// Sanity...
 				// Note that this MUST match the election process above!
 
-		log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
+		LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
 		if (!h->basetime) {
-			log(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr) );
+			LOG(0, 0, 0, "Heartbeat with zero basetime!  Ignoring\n");
 			return -1; // Skip it.
 		}
+
+		if (more >= 4) {
+			if (h->table_version > config->cluster_table_version) {
+				LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
+					h->table_version, config->cluster_table_version);
+
+				kill(0, SIGTERM);
+				exit(1);
+			}
+			if (h->table_version < config->cluster_table_version)
+			    	return -1;
+		}
+
 		if (basetime > h->basetime) {
-			log(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr));
+			LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
 			kill(0, SIGTERM);
 			exit(1);
 		}
-		if (basetime == h->basetime && my_address < addr) { // Tie breaker.
-			log(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
+
+		if (basetime < h->basetime)
+			return -1;
+
+		if (my_address < addr) { // Tie breaker.
+			LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
 			kill(0, SIGTERM);
 			exit(1);
 		}
+
 		return -1; // Skip it.
 	}
 
@@ -1091,9 +1189,22 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
 	config->cluster_last_hb = TIME;	// Reset to ensure that we don't become master!!
 
 	if (config->cluster_seq_number != h->seq) {	// Out of sequence heartbeat!
-		log(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);
+		static int lastseen_seq = 0;
+		static time_t lastseen_time = 0;
+
+		// limit to once per second for a particular seq#
+		int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
 
-		peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+		LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d.  %s.\n",
+			h->seq, config->cluster_seq_number,
+			ask ? "Asking for resend" : "Ignoring");
+
+		if (ask)
+		{
+			lastseen_seq = config->cluster_seq_number;
+			lastseen_time = time_now;
+			peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+		}
 
 		config->cluster_last_hb = TIME;	// Reset to ensure that we don't become master!!
 
@@ -1114,28 +1225,44 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
 			// that the free session pointer is correct.
 	cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
 
+	if (h->interval != config->cluster_hb_interval)
+	{
+		LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
+			h->interval, config->cluster_hb_interval);
+
+		config->cluster_hb_interval = h->interval;
+	}
+
+	if (h->timeout != config->cluster_hb_timeout)
+	{
+		LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
+			h->timeout, config->cluster_hb_timeout);
+
+		config->cluster_hb_timeout = h->timeout;
+	}
+
 		// Ok. process the packet...
 	while ( s > 0) {
 
-		type = * ((u32*) p);
-		p += sizeof(u32);
-		s -= sizeof(u32);
+		type = *((uint32_t *) p);
+		p += sizeof(uint32_t);
+		s -= sizeof(uint32_t);
 
-		more = * ((u32*) p);
-		p += sizeof(u32);
-		s -= sizeof(u32);
+		more = *((uint32_t *) p);
+		p += sizeof(uint32_t);
+		s -= sizeof(uint32_t);
 
 		switch (type) {
 			case C_CSESSION: { // Compressed session structure.
-				u8 c [ sizeof(sessiont) + 2];
+				uint8_t c[ sizeof(sessiont) + 2];
 				int size;
-				u8 * orig_p = p;
+				uint8_t *orig_p = p;
 
-				size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
+				size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
 				s -= (p - orig_p);
 
 				if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
-					log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
+					LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
 						// Now what? Should exit! No-longer up to date!
 					break;
 				}
@@ -1154,15 +1281,15 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
 				break;
 
 			case C_CTUNNEL: { // Compressed tunnel structure.
-				u8 c [ sizeof(tunnelt) + 2];
+				uint8_t c[ sizeof(tunnelt) + 2];
 				int size;
-				u8 * orig_p = p;
+				uint8_t *orig_p = p;
 
-				size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
+				size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
 				s -= (p - orig_p);
 
 				if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
-					log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
+					LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
 						// Now what? Should exit! No-longer up to date!
 					break;
 				}
@@ -1181,24 +1308,25 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
 				s -= sizeof(tunnel[more]);
 				break;
 			default:
-				log(0,0,0,0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
+				LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
 				return -1; // can't process any more of the packet!!
 		}
 	}
+
 	if (config->cluster_master_address != addr)
 	{
-		char *str;
-		str = strdup(inet_toa(config->cluster_master_address));
-		log(0,0,0,0, "My master just changed from %s to %s!\n", str, inet_toa(addr));
-		if (str) free(str);
+		LOG(0, 0, 0, "My master just changed from %s to %s!\n",
+			fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
+
+		config->cluster_master_address = addr;
 	}
 
-	config->cluster_master_address = addr;
 	config->cluster_last_hb = TIME;	// Successfully received a heartbeat!
+	config->cluster_table_version = h->table_version;
 	return 0;
 
 shortpacket:
-	log(0,0,0,0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
+	LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
 	return -1;
 }
 
@@ -1206,16 +1334,16 @@ shortpacket:
 // We got a packet on the cluster port!
 // Handle pings, lastseens, and heartbeats!
 //
-int processcluster(char * data, int size, u32 addr)
+int processcluster(char *data, int size, in_addr_t addr)
 {
 	int type, more;
-	char * p = data;
+	char *p = data;
 	int s = size;
 
 	if (addr == my_address)
 		return -1;	// Ignore it. Something looped back the multicast!
 
-	log(5,0,0,0, "Process cluster: %d bytes from %s\n", size, inet_toa(addr));
+	LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
 
 	if (s <= 0)	// Any data there??
 		return -1;
@@ -1223,42 +1351,47 @@ int processcluster(char * data, int size, u32 addr)
 	if (s < 8)
 		goto shortpacket;
 
-	type = * ((u32*) p);
-	p += sizeof(u32);
-	s -= sizeof(u32);
+	type = *((uint32_t *) p);
+	p += sizeof(uint32_t);
+	s -= sizeof(uint32_t);
 
-	more = * ((u32*) p);
-	p += sizeof(u32);
-	s -= sizeof(u32);
+	more = *((uint32_t *) p);
+	p += sizeof(uint32_t);
+	s -= sizeof(uint32_t);
 
 	switch (type) {
 	case C_PING:	// Update the peers table.
-		return cluster_add_peer(addr, more, (pingt*)p);
+		return cluster_add_peer(addr, more, (pingt *) p, s);
 
 	case C_LASTSEEN:	// Catch up a slave (slave missed a packet).
+		if (!config->cluster_iam_master) { // huh?
+			LOG(0, 0, 0, "I'm not the master, but I got a C_LASTSEEN from %s?\n", fmtaddr(addr, 0));
+			return -1;
+		}
+
 		return cluster_catchup_slave(more, addr);
 
 	case C_FORWARD: { // Forwarded control packet. pass off to processudp.
 		struct sockaddr_in a;
 		a.sin_addr.s_addr = more;
 
-		a.sin_port = * (int*) p;
+		a.sin_port = *(int *) p;
 		s -= sizeof(int);
 		p += sizeof(int);
 
 		if (!config->cluster_iam_master) { // huh?
-			log(0,0,0,0, "I'm not the master, but I got a C_FORWARD from %s?\n", inet_toa(addr));
+			LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
 			return -1;
 		}
 
-		log(4,0,0,0, "Got a forwarded packet... (%s:%d)\n", inet_toa(more), a.sin_port);
+		LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
 		STAT(recv_forward);
 		processudp(p, s, &a);
 		return 0;
 	}
 	case C_THROTTLE: {	// Receive a forwarded packet from a slave.
 		if (!config->cluster_iam_master) {
-			log(0,0,0,0, "I'm not the master, but I got a C_THROTTLE from %s?\n", inet_toa(addr));
+			LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
 			return -1;
 		}
 
@@ -1268,7 +1401,7 @@ int processcluster(char * data, int size, u32 addr)
 	case C_GARDEN:
 		// Receive a walled garden packet from a slave.
 		if (!config->cluster_iam_master) {
-			log(0,0,0,0, "I'm not the master, but I got a C_GARDEN from %s?\n", inet_toa(addr));
+			LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
 			return -1;
 		}
 
@@ -1280,37 +1413,37 @@ int processcluster(char * data, int size, u32 addr)
 
 	case C_KILL:	// The master asked us to die!? (usually because we're too out of date).
 		if (config->cluster_iam_master) {
-			log(0,0,0,0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", inet_toa(addr), more);
+			LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
 			return -1;
 		}
 		if (more != config->cluster_seq_number) {
-			log(0,0,0,0, "The master asked us to die but the seq number didn't match!?\n");
+			LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
 			return -1;
 		}
 
 		if (addr != config->cluster_master_address) {
-			log(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
-				inet_toa(addr), config->cluster_master_address);
+			LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
+				fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
 			// We can only warn about it. The master might really have switched!
 		}
 
-		log(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
+		LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
 		kill(0, SIGTERM);
 		exit(0);	// Lets be paranoid;
 		return -1;		// Just signalling the compiler.
 
 	case C_HEARTBEAT:
-		log(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));
-
-		return cluster_process_heartbeat_v2(data, size, more, p, addr);
+		LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
+		return cluster_process_heartbeat(data, size, more, p, addr);
 
 	default:
-		log(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type);
+		LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
 		return -1;
 	}
 	return 0;
+
 shortpacket:
-	log(0,0,0,0, "I got an cluster heartbeat packet! This means I'm probably out of sync!!\n");
+	LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
 	return -1;
 }
 
@@ -1323,21 +1456,25 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
 	if (CLI_HELP_REQUESTED)
 		return CLI_HELP_NO_ARGS;
 
-        cli_print(cli, "Cluster status   : %s", config->cluster_iam_master ? "Master" : "Slave" );
-	cli_print(cli, "My address       : %s", inet_toa(my_address));
-	cli_print(cli, "VIP address      : %s", inet_toa(config->bind_address));
-	cli_print(cli, "Multicast address: %s", inet_toa(config->cluster_address));
+	cli_print(cli, "Cluster status   : %s", config->cluster_iam_master ? "Master" : "Slave" );
+	cli_print(cli, "My address       : %s", fmtaddr(my_address, 0));
+	cli_print(cli, "VIP address      : %s", fmtaddr(config->bind_address, 0));
+	cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
 	cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
 
-        if (!config->cluster_iam_master) {
+	if (!config->cluster_iam_master) {
 		cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",
-			config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
+			config->cluster_master_address
+				? fmtaddr(config->cluster_master_address, 0)
+				: "Not defined",
 			0.1 * (TIME - config->cluster_last_hb));
-                cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");
+		cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");
+		cli_print(cli, "Table version #  : %" PRIu64, config->cluster_table_version);
 		cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
 		cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
 		cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
 	} else {
+		cli_print(cli, "Table version #  : %" PRIu64, config->cluster_table_version);
 		cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
 		cli_print(cli, "Highest session  : %d", config->cluster_highest_sessionid);
 		cli_print(cli, "Highest tunnel   : %d", config->cluster_highest_tunnelid);
@@ -1348,7 +1485,7 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
 	if (num_peers)
 		cli_print(cli, "%20s  %10s %8s", "Address", "Basetime", "Age");
 	for (i = 0; i < num_peers; ++i) {
-		cli_print(cli, "%20s  %10d %8d", inet_toa(peers[i].peer),
+		cli_print(cli, "%20s  %10u %8d", fmtaddr(peers[i].peer, 0),
 			peers[i].basetime, TIME - peers[i].timestamp);
 	}
 	return CLI_OK;
@@ -1373,11 +1510,11 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
 //
 // Worst case is a 50% expansion in space required (trying to
 // compress { 0x00, 0x01 } * N )
-int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
 {
 	int count;
 	int orig_dsize = dsize;
-	u8 * x,*src;
+	uint8_t *x, *src;
 	src = *src_p;
 
 	while (ssize > 0 && dsize > 2) {
@@ -1419,11 +1556,11 @@ int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
 // Return the number of dst bytes used.
 // Updates the 'src_p' pointer to point to the
 // first un-used byte.
-int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
+static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
 {
 	int count;
 	int orig_dsize = dsize;
-	char * src = *src_p;
+	char *src = *src_p;
 
 	while (ssize >0 && dsize > 0) {	// While there's more to decompress, and there's room in the decompress buffer...
 		count = *src++; --ssize;  // get the count byte from the source.