X-Git-Url: http://git.sameswireless.fr/l2tpns.git/blobdiff_plain/2f5b811429a2690c2ce44d09cc571c3230420c1b..c1d1c2a8c4f22c3911d392393072327b0e319c5e:/cluster.c

diff --git a/cluster.c b/cluster.c
index fdfe2dd..651307a 100644
--- a/cluster.c
+++ b/cluster.c
@@ -1,8 +1,10 @@
 // L2TPNS Clustering Stuff
 
-char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004-11-29 02:17:17 bodea Exp $";
-
 #include <stdio.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <unistd.h>
+#include <inttypes.h>
 #include <sys/file.h>
 #include <sys/stat.h>
 #include <sys/socket.h>
@@ -13,16 +15,13 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004-11-29 02:17:17 bodea Ex
 #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"
 #include "cluster.h"
 #include "util.h"
 #include "tbf.h"
+#include "pppoe.h"
 
 #ifdef BGP
 #include "bgp.h"
@@ -38,11 +37,13 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004-11-29 02:17:17 bodea Ex
  */
 
 // Module variables.
-int cluster_sockfd = 0;		// The filedescriptor for the cluster communications port.
+extern int cluster_sockfd;		// 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_bundle_number = 0;	// The next bundle 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.
+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!
@@ -55,20 +56,20 @@ static struct {
 static struct {
 	int seq;
 	int size;
-	char data[MAX_HEART_SIZE];
+	uint8_t data[MAX_HEART_SIZE];
 } past_hearts[HB_HISTORY_SIZE];	// Ring buffer of heartbeats that we've recently sent out. Needed so
 				// 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 rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
-static 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
@@ -81,9 +82,10 @@ int cluster_init()
 	struct sockaddr_in interface_addr;
 	struct ip_mreq mreq;
 	struct ifreq   ifr;
-	int opt = 0;
+	int opt;
 
 	config->cluster_undefined_sessions = MAXSESSION-1;
+	config->cluster_undefined_bundles = MAXBUNDLE-1;
 	config->cluster_undefined_tunnels = MAXTUNNEL-1;
 
 	if (!config->cluster_address)
@@ -99,6 +101,9 @@ int cluster_init()
 	addr.sin_addr.s_addr = INADDR_ANY;
 	setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
 
+	opt = fcntl(cluster_sockfd, F_GETFL, 0);
+	fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK);
+
 	if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
 	{
 		LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
@@ -123,13 +128,22 @@ int cluster_init()
 	opt = 0;	// Turn off multicast loopback.
 	setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));
 
+	if (config->cluster_mcast_ttl != 1)
+	{
+		uint8_t ttl = 0;
+		if (config->cluster_mcast_ttl > 0)
+			ttl = config->cluster_mcast_ttl < 256 ? config->cluster_mcast_ttl : 255;
+
+		setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl));
+	}
+
 	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)
+	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;
@@ -174,22 +188,22 @@ static 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(uint8_t **p, int type, int more, uint8_t *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;
 	}
 }
 
 // advertise our presence via BGP or gratuitous ARP
-static void advertise(void)
+static void advertise_routes(void)
 {
 #ifdef BGP
 	if (bgp_configured)
@@ -200,25 +214,34 @@ static void advertise(void)
 			send_garp(config->bind_address);	// Start taking traffic.
 }
 
+// withdraw our routes (BGP only)
+static void withdraw_routes(void)
+{
+#ifdef BGP
+	if (bgp_configured)
+		bgp_enable_routing(0);
+#endif /* BGP */
+}
+
 static void cluster_uptodate(void)
 {
 	if (config->cluster_iam_uptodate)
 		return;
 
-	if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
+	if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels || config->cluster_undefined_bundles)
 		return;
 
 	config->cluster_iam_uptodate = 1;
 
 	LOG(0, 0, 0, "Now uptodate with master.\n");
-	advertise();
+	advertise_routes();
 }
 
 //
 // Send a unicast UDP packet to a peer with 'data' as the
 // contents.
 //
-static int peer_send_data(u32 peer, char * data, int size)
+static int peer_send_data(in_addr_t peer, uint8_t *data, int size)
 {
 	struct sockaddr_in addr = {0};
 
@@ -246,10 +269,10 @@ static int peer_send_data(u32 peer, char * data, int size)
 //
 // Send a structured message to a peer with a single element of type 'type'.
 //
-static 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, uint8_t *data, int size)
 {
-	char buf[65536];	// Vast overkill.
-	char * p = buf;
+	uint8_t buf[65536];	// Vast overkill.
+	uint8_t *p = buf;
 
 	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);
@@ -257,16 +280,36 @@ static int peer_send_message(u32 peer, int type, int more, char * data, int size
 	return peer_send_data(peer, buf, (p-buf) );
 }
 
-//
+// send a packet to the master
+static int _forward_packet(uint8_t *data, int size, in_addr_t addr, int port, int type)
+{
+	uint8_t buf[65536];	// Vast overkill.
+	uint8_t *p = buf;
+
+	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
+		return -1;
+
+	LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
+
+	STAT(c_forwarded);
+	add_type(&p, type, addr, (uint8_t *) &port, sizeof(port)); // ick. should be uint16_t
+	memcpy(p, data, size);
+	p += size;
+
+	return peer_send_data(config->cluster_master_address, buf, (p - buf));
+}
+
+// 
 // Forward a state changing packet to the master.
 //
 // The master just processes the payload as if it had
 // received it off the tun device.
-//
-int master_forward_packet(char *data, int size, u32 addr, int port)
+//(note: THIS ROUTINE WRITES TO pack[-6]).
+int master_forward_packet(uint8_t *data, int size, in_addr_t addr, uint16_t port, uint16_t indexudp)
 {
-	char buf[65536];	// Vast overkill.
-	char *p = buf;
+	uint8_t *p = data - (3 * sizeof(uint32_t));
+	uint8_t *psave = p;
+	uint32_t indexandport = port | ((indexudp << 16) & 0xFFFF0000);
 
 	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
 		return -1;
@@ -274,12 +317,33 @@ int master_forward_packet(char *data, int size, u32 addr, int port)
 	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) );
-	memcpy(p, data, size);
-	p += size;
+	add_type(&p, C_FORWARD, addr, (uint8_t *) &indexandport, sizeof(indexandport));
 
-	return peer_send_data(config->cluster_master_address, buf, (p-buf) );
+	return peer_send_data(config->cluster_master_address, psave, size + (3 * sizeof(uint32_t)));
+}
+
+// Forward PPPOE packet to the master.
+//(note: THIS ROUTINE WRITES TO pack[-4]).
+int master_forward_pppoe_packet(uint8_t *data, int size, uint8_t codepad)
+{
+	uint8_t *p = data - (2 * sizeof(uint32_t));
+	uint8_t *psave = p;
+
+	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
+		return -1;
+
+	LOG(4, 0, 0, "Forward PPPOE packet to master, code %s (size %d)\n", get_string_codepad(codepad), size);
+
+	STAT(c_forwarded);
+	add_type(&p, C_PPPOE_FORWARD, codepad, NULL, 0);
 
+	return peer_send_data(config->cluster_master_address, psave, size + (2 * sizeof(uint32_t)));
+}
+
+// Forward a DAE RADIUS packet to the master.
+int master_forward_dae_packet(uint8_t *data, int size, in_addr_t addr, int port)
+{
+	return _forward_packet(data, size, addr, port, C_FORWARD_DAE);
 }
 
 //
@@ -289,10 +353,10 @@ 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, uint8_t *data, int size)
 {
-	char buf[65536];	// Vast overkill.
-	char *p = buf;
+	uint8_t buf[65536];	// Vast overkill.
+	uint8_t *p = buf;
 
 	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
 		return -1;
@@ -314,10 +378,10 @@ int master_throttle_packet(int tbfid, char *data, int size)
 //
 // (Note that this must be called with the tun header
 // as the start of the data).
-int master_garden_packet(sessionidt s, char *data, int size)
+int master_garden_packet(sessionidt s, uint8_t *data, int size)
 {
-	char buf[65536];	// Vast overkill.
-	char *p = buf;
+	uint8_t buf[65536];	// Vast overkill.
+	uint8_t *p = buf;
 
 	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
 		return -1;
@@ -330,11 +394,33 @@ int master_garden_packet(sessionidt s, char *data, int size)
 
 }
 
+//
+// Forward a MPPP packet to the master for handling.
+//
+// (Note that this must be called with the tun header
+// as the start of the data).
+// (i.e. this routine writes to data[-8]).
+int master_forward_mppp_packet(sessionidt s, uint8_t *data, int size)
+{
+	uint8_t *p = data - (2 * sizeof(uint32_t));
+	uint8_t *psave = p;
+
+	if (!config->cluster_master_address) // No election has been held yet. Just skip it.
+		return -1;
+
+	LOG(4, 0, 0, "Forward MPPP packet to master (size %d)\n", size);
+
+	add_type(&p, C_MPPP_FORWARD, s, NULL, 0);
+
+	return peer_send_data(config->cluster_master_address, psave, size + (2 * sizeof(uint32_t)));
+
+}
+
 //
 // 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, uint8_t *data, int size)
 {
 	int i;
 
@@ -356,8 +442,8 @@ static void send_heartbeat(int seq, char * data, int size)
 //
 void cluster_send_ping(time_t basetime)
 {
-	char buff[100 + sizeof(pingt)];
-	char *p = buff;
+	uint8_t buff[100 + sizeof(pingt)];
+	uint8_t *p = buff;
 	pingt x;
 
 	if (config->cluster_iam_master && basetime)		// We're heartbeating so no need to ping.
@@ -367,10 +453,10 @@ void cluster_send_ping(time_t basetime)
 
 	x.ver = 1;
 	x.addr = config->bind_address;
-	x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
+	x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels + config->cluster_undefined_bundles;
 	x.basetime = basetime;
 
-	add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
+	add_type(&p, C_PING, basetime, (uint8_t *) &x, sizeof(x));
 	cluster_send_data(buff, (p-buff) );
 }
 
@@ -390,10 +476,10 @@ void master_update_counts(void)
 	if (config->cluster_iam_master)		// Only happens on the slaves.
 		return;
 
-	if (!config->cluster_master_address)		// If we don't have a master, skip it for a while.
+	if (!config->cluster_master_address)	// If we don't have a master, skip it for a while.
 		return;
 
-	i = MAX_B_RECS * 5; // Examine max 2000 sessions;
+	i = MAX_B_RECS * 5; // Examine max 3000 sessions;
 	if (config->cluster_highest_sessionid > i)
 		i = config->cluster_highest_sessionid;
 
@@ -403,18 +489,21 @@ 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)
-			continue; // Unused. Skip it.
+		if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
+			continue; // Unchanged. 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;
-
-		if (++c > MAX_B_RECS)	// Send a max of 400 elements in a packet.
-			break;
+		b[c].pin = sess_local[walk_session_number].pin;
+		b[c].pout = sess_local[walk_session_number].pout;
+		b[c].cin = sess_local[walk_session_number].cin;
+		b[c].cout = sess_local[walk_session_number].cout;
 
 			// Reset counters.
-		sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
+		sess_local[walk_session_number].pin = sess_local[walk_session_number].pout = 0;
+		sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
+
+		if (++c > MAX_B_RECS)	// Send a max of 600 elements in a packet.
+			break;
 	}
 
 	if (!c)		// Didn't find any that changes. Get out of here!
@@ -423,7 +512,7 @@ void master_update_counts(void)
 
 			// Forward the data to the master.
 	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);
+	peer_send_message(config->cluster_master_address, C_BYTES, c, (uint8_t *) &b, sizeof(b[0]) * c);
 	return;
 }
 
@@ -454,17 +543,22 @@ void cluster_check_slaves(void)
 			continue;	// Shutdown peer! Skip them.
 
 		if (peers[i].uptodate)
-			have_peers = 1;
-
-		if (!peers[i].uptodate)
+			have_peers++;
+		else
 			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 */
+	// in a cluster, withdraw/add routes when we get a peer/lose peers
+	if (have_peers != had_peers)
+	{
+		if (had_peers < config->cluster_master_min_adv &&
+		    have_peers >= config->cluster_master_min_adv)
+			withdraw_routes();
+
+		else if (had_peers >= config->cluster_master_min_adv &&
+		    have_peers < config->cluster_master_min_adv)
+			advertise_routes();
+	}
 }
 
 //
@@ -473,24 +567,26 @@ void cluster_check_slaves(void)
 //
 void cluster_check_master(void)
 {
-	int i, count, tcount, high_unique_id = 0;
+	int i, count, high_unique_id = 0;
 	int last_free = 0;
 	clockt t = TIME;
 	static int probed = 0;
+	int have_peers;
 
 	if (config->cluster_iam_master)
 		return;		// Only runs on the slaves...
 
 	// 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
+	// multicast group, try unicasting probes 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 (config->cluster_master_address
+	    && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
 	{
-		if (!probed && config->cluster_master_address)
+		if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval)))
 		{
-			probed = 1;
+			probed = TIME;
 			LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
 				0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
 
@@ -505,10 +601,15 @@ void cluster_check_master(void)
 		return;	// Everything's ok!
 
 	config->cluster_last_hb = TIME + 1;	// Just the one election thanks.
+	config->cluster_master_address = 0;
 
 	LOG(0, 0, 0, "Master timed out! Holding election...\n");
 
-	for (i = 0; i < num_peers; i++)
+	// In the process of shutting down, can't be master
+	if (main_quit)
+		return;
+
+	for (i = have_peers = 0; i < num_peers; i++)
 	{
 		if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
 			continue;	// Stale peer! Skip them.
@@ -526,6 +627,9 @@ void cluster_check_master(void)
 			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 (peers[i].uptodate)
+			have_peers++;
 	}
 
 		// Wow. it's been ages since I last heard a heartbeat
@@ -533,10 +637,15 @@ void cluster_check_master(void)
 		// to become a master!!!
 
 	config->cluster_iam_master = 1;
-	config->cluster_master_address = 0;
+	pppoe_send_garp(); // gratuitous arp of the pppoe interface
 
 	LOG(0, 0, 0, "I am declaring myself the master!\n");
 
+	if (have_peers < config->cluster_master_min_adv)
+		advertise_routes();
+	else
+		withdraw_routes();
+
 	if (config->cluster_seq_number == -1)
 		config->cluster_seq_number = 0;
 
@@ -545,7 +654,7 @@ void cluster_check_master(void)
 		// Count the highest used tunnel number as well.
 		//
 	config->cluster_highest_tunnelid = 0;
-	for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
+	for (i = 0; i < MAXTUNNEL; ++i) {
 		if (tunnel[i].state == TUNNELUNDEF)
 			tunnel[i].state = TUNNELFREE;
 
@@ -553,6 +662,19 @@ void cluster_check_master(void)
 			config->cluster_highest_tunnelid = i;
 	}
 
+		//
+                // Go through and mark all the bundles as defined.
+                // Count the highest used bundle number as well.
+                //
+        config->cluster_highest_bundleid = 0;
+        for (i = 0; i < MAXBUNDLE; ++i) {
+                if (bundle[i].state == BUNDLEUNDEF)
+                        bundle[i].state = BUNDLEFREE;
+
+                if (bundle[i].state != BUNDLEFREE && i > config->cluster_highest_bundleid)
+                        config->cluster_highest_bundleid = i;
+        }
+
 		//
 		// Go through and mark all the sessions as being defined.
 		// reset the idle timeouts.
@@ -568,54 +690,60 @@ 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..
-		session[i].last_packet = time_now;
+			// Reset idle timeouts..
+		session[i].last_packet = session[i].last_data = time_now;
+
+			// Reset die relative to our uptime rather than the old master's
+		if (session[i].die) session[i].die = TIME;
+
+			// Accumulate un-sent byte/packet counters.
+		increment_counter(&session[i].cin, &session[i].cin_wrap, sess_local[i].cin);
+		increment_counter(&session[i].cout, &session[i].cout_wrap, sess_local[i].cout);
+		session[i].cin_delta += sess_local[i].cin;
+		session[i].cout_delta += sess_local[i].cout;
 
-			// 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;
+		session[i].pin += sess_local[i].pin;
+		session[i].pout += sess_local[i].pout;
 
-		sess_count[i].cin = sess_count[i].cout = 0;
+		sess_local[i].cin = sess_local[i].cout = 0;
+		sess_local[i].pin = sess_local[i].pout = 0;
 
-		session[i].radius = 0;	// Reset authentication as the radius blocks aren't up to date.
+		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_in, session[i].throttle_out);
 
-		if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
-			config->cluster_highest_sessionid = i;
+		config->cluster_highest_sessionid = i;
 	}
 
 	session[last_free].next = 0;	// End of chain.
-	last_id = high_unique_id;		// Keep track of the highest used session ID.
+	last_id = high_unique_id;	// Keep track of the highest used session ID.
 
 	become_master();
 
 	rebuild_address_pool();
 
 		// If we're not the very first master, this is a big issue!
-	if(count>0)
+	if (count > 0)
 		LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
 
 	config->cluster_undefined_sessions = 0;
+	config->cluster_undefined_bundles = 0;
 	config->cluster_undefined_tunnels = 0;
 	config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
 
-	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.
@@ -630,7 +758,7 @@ void cluster_check_master(void)
 // we fix it up here, and we ensure that the 'first free session'
 // pointer is valid.
 //
-static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
+static void cluster_check_sessions(int highsession, int freesession_ptr, int highbundle, int hightunnel)
 {
 	int i;
 
@@ -639,7 +767,7 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
 	if (config->cluster_iam_uptodate)
 		return;
 
-	if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
+	if (highsession > config->cluster_undefined_sessions && highbundle > config->cluster_undefined_bundles && hightunnel > config->cluster_undefined_tunnels)
 		return;
 
 		// Clear out defined sessions, counting the number of
@@ -647,12 +775,25 @@ 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)
+
+		if (session[i].tunnel == T_UNDEF)
+			++config->cluster_undefined_sessions;
+	}
+
+		// Clear out defined bundles, counting the number of
+		// undefs remaining.
+	config->cluster_undefined_bundles = 0;
+	for (i = 1 ; i < MAXBUNDLE; ++i) {
+		if (i > highbundle) {
+			if (bundle[i].state == BUNDLEUNDEF) bundle[i].state = BUNDLEFREE; // Defined.
 			continue;
-		++config->cluster_undefined_sessions;
+		}
+
+		if (bundle[i].state == BUNDLEUNDEF)
+			++config->cluster_undefined_bundles;
 	}
 
 		// Clear out defined tunnels, counting the number of
@@ -660,18 +801,18 @@ 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, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
-			config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
+	if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels || config->cluster_undefined_bundles) {
+		LOG(2, 0, 0, "Cleared undefined sessions/bundles/tunnels. %d sess (high %d), %d bund (high %d), %d tunn (high %d)\n",
+			config->cluster_undefined_sessions, highsession, config->cluster_undefined_bundles, highbundle, config->cluster_undefined_tunnels, hightunnel);
 		return;
 	}
 
@@ -681,13 +822,13 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
 		cluster_uptodate();
 }
 
-static int hb_add_type(char **p, int type, int id)
+static int hb_add_type(uint8_t **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) );
@@ -695,19 +836,40 @@ static 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, c, size);
 				break;
 			}
 			// Failed to compress : Fall through.
 		}
-		case C_SESSION: add_type(p, C_SESSION, id,
-			(char*) &session[id], sizeof(sessiont));
+		case C_SESSION:
+		    	add_type(p, C_SESSION, id, (uint8_t *) &session[id], sizeof(sessiont));
+			break;
+
+		case C_CBUNDLE: { // Compressed C_BUNDLE
+                        uint8_t c[sizeof(bundlet) * 2]; // Bigger than worst case.
+                        uint8_t *d = (uint8_t *) &bundle[id];
+                        uint8_t *orig = d;
+                        int size;
+
+                        size = rle_compress( &d,  sizeof(bundlet), c, sizeof(c) );
+
+                                // Did we compress the full structure, and is the size actually
+                                // reduced??
+                        if ( (d - orig) == sizeof(bundlet) && size < sizeof(bundlet) ) {
+                                add_type(p, C_CBUNDLE, id, c, size);
+                                break;
+                        }
+                        // Failed to compress : Fall through.
+                }
+
+		case C_BUNDLE:
+		    	add_type(p, C_BUNDLE, id, (uint8_t *) &bundle[id], sizeof(bundlet));
 			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) );
@@ -720,8 +882,8 @@ static 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));
+		case C_TUNNEL:
+		    	add_type(p, C_TUNNEL, id, (uint8_t *) &tunnel[id], sizeof(tunnelt));
 			break;
 		default:
 			LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
@@ -736,14 +898,16 @@ static int hb_add_type(char **p, int type, int id)
 //
 void cluster_heartbeat()
 {
-	int i, count = 0, tcount = 0;
-	char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
+	int i, count = 0, tcount = 0, bcount = 0;
+	uint8_t buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
 	heartt h;
-	char *p = buff;
+	uint8_t *p = buff;
 
 	if (!config->cluster_iam_master)	// Only the master does this.
 		return;
 
+	config->cluster_table_version += config->cluster_num_changes;
+
 	// Fill out the heartbeat header.
 	memset(&h, 0, sizeof(h));
 
@@ -755,19 +919,22 @@ void cluster_heartbeat()
 	h.highsession = config->cluster_highest_sessionid;
 	h.freesession = sessionfree;
 	h.hightunnel = config->cluster_highest_tunnelid;
+	h.highbundle = config->cluster_highest_bundleid;
 	h.size_sess = sizeof(sessiont);		// Just in case.
+	h.size_bund = sizeof(bundlet);
 	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, (uint8_t *) &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, "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);
 	}
@@ -775,7 +942,7 @@ void cluster_heartbeat()
 		//
 		// 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;
@@ -795,7 +962,7 @@ void cluster_heartbeat()
 		// 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;
@@ -809,17 +976,34 @@ void cluster_heartbeat()
 		++tcount;
 	}
 
+		//
+		// Fill out the packet with bundles from the bundle table...
+	while ( (p + sizeof(uint32_t) * 2 + sizeof(bundlet) ) < (buff + MAX_HEART_SIZE) ) {
+
+		if (!walk_bundle_number)        // bundle #0 isn't valid.
+			++walk_bundle_number;
+
+		if (bcount >= config->cluster_highest_bundleid)
+			break;
+
+		hb_add_type(&p, C_CBUNDLE, walk_bundle_number);
+		walk_bundle_number = (1+walk_bundle_number)%(config->cluster_highest_bundleid+1);	// +1 avoids divide by zero.
+		++bcount;
+        }
+
 		//
 		// Did we do something wrong?
 	if (p > (buff + sizeof(buff))) {	// Did we somehow manage to overun the buffer?
-		LOG(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, "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(4, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
+		     "(%d x-sess, %d x-bundles, %d x-tunnels, %d highsess, %d highbund, %d hightun, size %d)\n",
+	    HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
+	    count, bcount, tcount, config->cluster_highest_sessionid, config->cluster_highest_bundleid,
+	    config->cluster_highest_tunnelid, (int) (p - buff));
 
 	config->cluster_num_changes = 0;
 
@@ -836,12 +1020,20 @@ static int type_changed(int type, int id)
 	int i;
 
 	for (i = 0 ; i < config->cluster_num_changes ; ++i)
-		if ( cluster_changes[i].id == id &&
-			cluster_changes[i].type == type)
-			return 0;	// Already marked for change.
+	{
+		if ( cluster_changes[i].id == id && cluster_changes[i].type == type)
+		{
+			// Already marked for change, remove it
+			--config->cluster_num_changes;
+			memmove(&cluster_changes[i],
+					&cluster_changes[i+1],
+					(config->cluster_num_changes - i) * sizeof(cluster_changes[i]));
+			break;
+		}
+	}
 
-	cluster_changes[i].type = type;
-	cluster_changes[i].id = id;
+	cluster_changes[config->cluster_num_changes].type = type;
+	cluster_changes[config->cluster_num_changes].id = id;
 	++config->cluster_num_changes;
 
 	if (config->cluster_num_changes > MAX_CHANGES)
@@ -850,7 +1042,6 @@ static int type_changed(int type, int id)
 	return 1;
 }
 
-
 // A particular session has been changed!
 int cluster_send_session(int sid)
 {
@@ -859,9 +1050,25 @@ int cluster_send_session(int sid)
 		return -1;
 	}
 
+	if (forked) {
+		LOG(0, sid, 0, "cluster_send_session called from child process!\n");
+		return -1;
+	}
+
 	return type_changed(C_CSESSION, sid);
 }
 
+// A particular bundle has been changed!
+int cluster_send_bundle(int bid)
+{
+	if (!config->cluster_iam_master) {
+		LOG(0, 0, bid, "I'm not a master, but I just tried to change a bundle!\n");
+		return -1;
+	}
+
+	return type_changed(C_CBUNDLE, bid);
+}
+
 // A particular tunnel has been changed!
 int cluster_send_tunnel(int tid)
 {
@@ -879,26 +1086,35 @@ int cluster_send_tunnel(int tid)
 // missed a packet. We'll resend it every packet since
 // the last one it's seen.
 //
-static 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, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
+	if (!config->cluster_iam_master) {
+		LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
+			fmtaddr(config->cluster_master_address, 0));
+
+		peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0);
+		return 0;
+	}
 
 	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, "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.
 	}
 
+	LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) );
+
 		// Now resend every packet that it missed, in order.
 	while (seq != config->cluster_seq_number) {
-		s = seq%HB_HISTORY_SIZE;
+		s = seq % HB_HISTORY_SIZE;
 		if (seq != past_hearts[s].seq) {
 			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);
@@ -914,10 +1130,10 @@ static 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.
 //
-static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
+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.
@@ -927,8 +1143,8 @@ static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
 	if (size > sizeof(p))
 		size = sizeof(p);
 
-	memset( (void*) &p, 0, sizeof(p) );
-	memcpy( (void*) &p, (void*) pp, size);
+	memset( (void *) &p, 0, sizeof(p) );
+	memcpy( (void *) &p, (void *) pp, size);
 
 	clusterid = p.addr;
 	if (clusterid != config->bind_address)
@@ -951,12 +1167,13 @@ static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
 	}
 
 	// 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));
+	if (peer == config->cluster_master_address) {
+		LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0),
+			basetime ? "has restarted!" : "shutting down...");
+
 		config->cluster_master_address = 0;
 		config->cluster_last_hb = 0; // Force an election.
 		cluster_check_master();
-		return 0;
 	}
 
 	if (i >= num_peers)
@@ -993,16 +1210,40 @@ static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
 	return 1;
 }
 
+// A slave responds with C_MASTER when it gets a message which should have gone to a master.
+static int cluster_set_master(in_addr_t peer, in_addr_t master)
+{
+	if (config->cluster_iam_master)	// Sanity...
+		return 0;
+
+	LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0),
+		fmtaddr(master, 1));
+
+	config->cluster_master_address = master;
+	if (master)
+	{
+		// catchup with new master
+		peer_send_message(master, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+
+		// delay next election
+		config->cluster_last_hb = TIME;
+	}
+
+	// run election (or reset "probed" if master was set)
+	cluster_check_master();
+	return 0;
+}
+
 /* Handle the slave updating the byte counters for the master. */
 //
 // 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.
 //
-static int cluster_handle_bytes(char * data, int size)
+static int cluster_handle_bytes(uint8_t *data, int size)
 {
-	bytest * b;
+	bytest *b;
 
-	b = (bytest*) data;
+	b = (bytest *) data;
 
 	LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
 
@@ -1015,12 +1256,19 @@ static int cluster_handle_bytes(char * data, int size)
 			return -1; /* Abort processing */
 		}
 
-		session[b->sid].total_cin += b->in;
-		session[b->sid].total_cout += b->out;
+		session[b->sid].pin += b->pin;
+		session[b->sid].pout += b->pout;
+
+		increment_counter(&session[b->sid].cin, &session[b->sid].cin_wrap, b->cin);
+		increment_counter(&session[b->sid].cout, &session[b->sid].cout_wrap, b->cout);
+
+		session[b->sid].cin_delta += b->cin;
+		session[b->sid].cout_delta += b->cout;
 
-		session[b->sid].cin += b->in;
-		session[b->sid].cout += b->out;
-		session[b->sid].last_packet = time_now; // Reset idle timer!
+		if (b->cin)
+			session[b->sid].last_packet = session[b->sid].last_data = time_now;
+		else if (b->cout)
+			session[b->sid].last_data = time_now;
 
 		size -= sizeof(*b);
 		++b;
@@ -1035,7 +1283,7 @@ static 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, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
@@ -1050,7 +1298,7 @@ static int cluster_recv_session(int more , u8 * p)
 		}
 	}
 
-	load_session(more, (sessiont*) p);	// Copy session into session table..
+	load_session(more, (sessiont *) p);	// Copy session into session table..
 
 	LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
 
@@ -1060,7 +1308,32 @@ static int cluster_recv_session(int more , u8 * p)
 	return 0;
 }
 
-static int cluster_recv_tunnel(int more, u8 *p)
+static int cluster_recv_bundle(int more, uint8_t *p)
+{
+	if (more >= MAXBUNDLE) {
+		LOG(0, 0, 0, "DANGER: Received a bundle id > MAXBUNDLE!\n");
+		return -1;
+	}
+
+	if (bundle[more].state == BUNDLEUNDEF) {
+		if (config->cluster_iam_uptodate) { // Sanity.
+			LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined bundle!\n");
+		} else {
+			--config->cluster_undefined_bundles;
+		}
+	}
+
+	memcpy(&bundle[more], p, sizeof(bundle[more]) );
+
+	LOG(5, 0, more, "Received bundle update\n");
+
+	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, uint8_t *p)
 {
 	if (more >= MAXTUNNEL) {
 		LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
@@ -1093,30 +1366,163 @@ static int cluster_recv_tunnel(int more, u8 *p)
 }
 
 
+// pre v6 heartbeat session structure
+struct oldsession {
+	sessionidt next;
+	sessionidt far;
+	tunnelidt tunnel;
+	uint8_t flags;
+	struct {
+		uint8_t phase;
+		uint8_t lcp:4;
+		uint8_t ipcp:4;
+		uint8_t ipv6cp:4;
+		uint8_t ccp:4;
+	} ppp;
+	char reserved_1[2];
+	in_addr_t ip;
+	int ip_pool_index;
+	uint32_t unique_id;
+	char reserved_2[4];
+	uint32_t magic;
+	uint32_t pin, pout;
+	uint32_t cin, cout;
+	uint32_t cin_wrap, cout_wrap;
+	uint32_t cin_delta, cout_delta;
+	uint16_t throttle_in;
+	uint16_t throttle_out;
+	uint8_t filter_in;
+	uint8_t filter_out;
+	uint16_t mru;
+	clockt opened;
+	clockt die;
+	uint32_t session_timeout;
+	uint32_t idle_timeout;
+	time_t last_packet;
+	time_t last_data;
+	in_addr_t dns1, dns2;
+	routet route[MAXROUTE];
+	uint16_t tbf_in;
+	uint16_t tbf_out;
+	int random_vector_length;
+	uint8_t random_vector[MAXTEL];
+	char user[MAXUSER];
+	char called[MAXTEL];
+	char calling[MAXTEL];
+	uint32_t tx_connect_speed;
+	uint32_t rx_connect_speed;
+	clockt timeout;
+	uint32_t mrru;
+	uint8_t mssf;
+	epdist epdis;
+	bundleidt bundle;
+	in_addr_t snoop_ip;
+	uint16_t snoop_port;
+	uint8_t walled_garden;
+	uint8_t ipv6prefixlen;
+	struct in6_addr ipv6route;
+	char reserved_3[11];
+};
+
+static uint8_t *convert_session(struct oldsession *old)
+{
+	static sessiont new;
+	int i;
+
+	memset(&new, 0, sizeof(new));
+
+	new.next = old->next;
+	new.far = old->far;
+	new.tunnel = old->tunnel;
+	new.flags = old->flags;
+	new.ppp.phase = old->ppp.phase;
+	new.ppp.lcp = old->ppp.lcp;
+	new.ppp.ipcp = old->ppp.ipcp;
+	new.ppp.ipv6cp = old->ppp.ipv6cp;
+	new.ppp.ccp = old->ppp.ccp;
+	new.ip = old->ip;
+	new.ip_pool_index = old->ip_pool_index;
+	new.unique_id = old->unique_id;
+	new.magic = old->magic;
+	new.pin = old->pin;
+	new.pout = old->pout;
+	new.cin = old->cin;
+	new.cout = old->cout;
+	new.cin_wrap = old->cin_wrap;
+	new.cout_wrap = old->cout_wrap;
+	new.cin_delta = old->cin_delta;
+	new.cout_delta = old->cout_delta;
+	new.throttle_in = old->throttle_in;
+	new.throttle_out = old->throttle_out;
+	new.filter_in = old->filter_in;
+	new.filter_out = old->filter_out;
+	new.mru = old->mru;
+	new.opened = old->opened;
+	new.die = old->die;
+	new.session_timeout = old->session_timeout;
+	new.idle_timeout = old->idle_timeout;
+	new.last_packet = old->last_packet;
+	new.last_data = old->last_data;
+	new.dns1 = old->dns1;
+	new.dns2 = old->dns2;
+	new.tbf_in = old->tbf_in;
+	new.tbf_out = old->tbf_out;
+	new.random_vector_length = old->random_vector_length;
+	new.tx_connect_speed = old->tx_connect_speed;
+	new.rx_connect_speed = old->rx_connect_speed;
+	new.timeout = old->timeout;
+	new.mrru = old->mrru;
+	new.mssf = old->mssf;
+	new.epdis = old->epdis;
+	new.bundle = old->bundle;
+	new.snoop_ip = old->snoop_ip;
+	new.snoop_port = old->snoop_port;
+	new.walled_garden = old->walled_garden;
+	new.ipv6prefixlen = old->ipv6prefixlen;
+	new.ipv6route = old->ipv6route;
+
+	memcpy(new.random_vector, old->random_vector, sizeof(new.random_vector));
+	memcpy(new.user, old->user, sizeof(new.user));
+	memcpy(new.called, old->called, sizeof(new.called));
+	memcpy(new.calling, old->calling, sizeof(new.calling));
+
+	for (i = 0; i < MAXROUTE; i++)
+		memcpy(&new.route[i], &old->route[i], sizeof(new.route[i]));
+
+	return (uint8_t *) &new;
+}
+
 //
 // Process a heartbeat..
 //
-static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
+// v6: added RADIUS class attribute, re-ordered session structure
+// v7: added tunnelt attribute at the end of struct (tunnelt size change)
+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;
+	int hb_ver = more;
 
-#if HB_VERSION != 3
+#if HB_VERSION != 7
 # error "need to update cluster_process_heartbeat()"
 #endif
 
-	// we handle version 2+
-	if (more < 2 || more > HB_VERSION) {
-		LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
+	// we handle versions 5 through 7
+	if (hb_ver < 5 || hb_ver > HB_VERSION) {
+		LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", hb_ver);
 		return -1; // Ignore it??
 	}
 
-		// Ok. It's a heartbeat packet from a cluster master!
+	if (size > sizeof(past_hearts[0].data)) {
+		LOG(0, 0, 0, "Received an oversize heartbeat from %s (%d)!\n", fmtaddr(addr, 0), size);
+		return -1;
+	}
+
 	if (s < sizeof(*h))
 		goto shortpacket;
 
-	h = (heartt*) p;
+	h = (heartt *) p;
 	p += sizeof(*h);
 	s -= sizeof(*h);
 
@@ -1126,33 +1532,87 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
 	if (config->cluster_iam_master) {	// Sanity...
 				// Note that this MUST match the election process above!
 
-		LOG(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, "Heartbeat from addr %s with zero basetime!\n", fmtaddr(addr, 0));
+			LOG(0, 0, 0, "Heartbeat with zero basetime!  Ignoring\n");
 			return -1; // Skip it.
 		}
+
+		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, "They're (%s) an older master than me so I'm gone!\n", fmtaddr(addr, 0));
+			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.
+
+		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);
 		}
+
+			//
+			// Send it a unicast heartbeat to see give it a chance to die.
+			// NOTE: It's actually safe to do seq-number - 1 without checking
+			// for wrap around.
+			//
+		cluster_catchup_slave(config->cluster_seq_number - 1, addr);
+
 		return -1; // Skip it.
 	}
 
+		//
+		// Try and guard against a stray master appearing.
+		//
+		// Ignore heartbeats received from another master before the
+		// timeout (less a smidgen) for the old master has elapsed.
+		//
+		// Note that after a clean failover, the cluster_master_address
+		// is cleared, so this doesn't run. 
+		//
+	if (config->cluster_master_address && addr != config->cluster_master_address) {
+		    LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
+			    fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1),
+			    0.1 * (TIME - config->cluster_last_hb));
+		    return -1; // ignore
+	}
+
 	if (config->cluster_seq_number == -1)	// Don't have one. Just align to the master...
 		config->cluster_seq_number = h->seq;
 
 	config->cluster_last_hb = TIME;	// Reset to ensure that we don't become master!!
+	config->cluster_last_hb_ver = hb_ver; // remember what cluster version the master is using
 
 	if (config->cluster_seq_number != h->seq) {	// Out of sequence heartbeat!
-		LOG(1, 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!!
 
@@ -1171,47 +1631,56 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
 
 			// Check that we don't have too many undefined sessions, and
 			// that the free session pointer is correct.
-	cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
+	cluster_check_sessions(h->highsession, h->freesession, h->highbundle, h->hightunnel);
 
-	if (more > 2) // reserved section of heartt was not initialized prior to v3
+	if (h->interval != config->cluster_hb_interval)
 	{
-		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);
+		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;
-		}
+		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);
+	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;
-		}
+		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);
 
+				// session struct changed with v5
+				if (hb_ver < 6)
+				{
+					if (size != sizeof(struct oldsession)) {
+						LOG(0, 0, 0, "DANGER: Received a v%d CSESSION that didn't decompress correctly!\n", hb_ver);
+							// Now what? Should exit! No-longer up to date!
+						break;
+					}
+					cluster_recv_session(more, convert_session((struct oldsession *) c));
+					break;
+				}
+
 				if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
 					LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
 						// Now what? Should exit! No-longer up to date!
@@ -1222,6 +1691,18 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
 				break;
 			}
 			case C_SESSION:
+				if (hb_ver < 6)
+				{
+					if (s < sizeof(struct oldsession))
+						goto shortpacket;
+
+					cluster_recv_session(more, convert_session((struct oldsession *) p));
+
+					p += sizeof(struct oldsession);
+					s -= sizeof(struct oldsession);
+					break;
+				}
+
 				if ( s < sizeof(session[more]))
 					goto shortpacket;
 
@@ -1232,15 +1713,17 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
 				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, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
+				if ( ((hb_ver >= HB_VERSION) && (size != sizeof(tunnelt))) ||
+					 ((hb_ver < HB_VERSION) && (size > sizeof(tunnelt))) )
+				{ // Ouch! Very very bad!
+					LOG(0, 0, 0, "DANGER: Received a CTUNNEL that didn't decompress correctly!\n");
 						// Now what? Should exit! No-longer up to date!
 					break;
 				}
@@ -1258,6 +1741,34 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
 				p += sizeof(tunnel[more]);
 				s -= sizeof(tunnel[more]);
 				break;
+
+			case C_CBUNDLE: { // Compressed bundle structure.
+				uint8_t c[ sizeof(bundlet) + 2];
+				int size;
+				uint8_t *orig_p = p;
+
+				size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
+				s -= (p - orig_p);
+
+				if (size != sizeof(bundlet) ) { // Ouch! Very very bad!
+					LOG(0, 0, 0, "DANGER: Received a CBUNDLE that didn't decompress correctly!\n");
+						// Now what? Should exit! No-longer up to date!
+					break;
+				}
+
+				cluster_recv_bundle(more, c);
+				break;
+
+			}
+			case C_BUNDLE:
+				if ( s < sizeof(bundle[more]))
+                                        goto shortpacket;
+
+                                cluster_recv_bundle(more, p);
+
+                                p += sizeof(bundle[more]);
+                                s -= sizeof(bundle[more]);
+                                break;
 			default:
 				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!!
@@ -1273,6 +1784,7 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
 	}
 
 	config->cluster_last_hb = TIME;	// Successfully received a heartbeat!
+	config->cluster_table_version = h->table_version;
 	return 0;
 
 shortpacket:
@@ -1284,10 +1796,10 @@ shortpacket:
 // We got a packet on the cluster port!
 // Handle pings, lastseens, and heartbeats!
 //
-int processcluster(char * data, int size, u32 addr)
+int processcluster(uint8_t *data, int size, in_addr_t addr)
 {
 	int type, more;
-	char * p = data;
+	uint8_t *p = data;
 	int s = size;
 
 	if (addr == my_address)
@@ -1301,39 +1813,82 @@ 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, s);
+	switch (type)
+	{
+	case C_PING: // Update the peers table.
+		return cluster_add_peer(addr, more, (pingt *) p, s);
 
-	case C_LASTSEEN:	// Catch up a slave (slave missed a packet).
+	case C_MASTER: // Our master is wrong
+	    	return cluster_set_master(addr, more);
+
+	case C_LASTSEEN: // Catch up a slave (slave missed a packet).
 		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;
+	case C_FORWARD: // Forwarded control packet. pass off to processudp.
+	case C_FORWARD_DAE: // Forwarded DAE packet. pass off to processdae.
+		if (!config->cluster_iam_master)
+		{
+			LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD%s from %s?\n",
+				type == C_FORWARD_DAE ? "_DAE" : "", fmtaddr(addr, 0));
+
+			return -1;
+		}
+		else
+		{
+			struct sockaddr_in a;
+			uint16_t indexudp;
+			a.sin_addr.s_addr = more;
+
+			a.sin_port = (*(int *) p) & 0xFFFF;
+			indexudp = ((*(int *) p) >> 16) & 0xFFFF;
+			s -= sizeof(int);
+			p += sizeof(int);
+
+			LOG(4, 0, 0, "Got a forwarded %spacket... (%s:%d)\n",
+				type == C_FORWARD_DAE ? "DAE " : "", fmtaddr(more, 0), a.sin_port);
+
+			STAT(recv_forward);
+			if (type == C_FORWARD_DAE)
+			{
+				struct in_addr local;
+				local.s_addr = config->bind_address ? config->bind_address : my_address;
+				processdae(p, s, &a, sizeof(a), &local);
+			}
+			else
+				processudp(p, s, &a, indexudp);
 
-		a.sin_port = * (int*) p;
-		s -= sizeof(int);
-		p += sizeof(int);
+			return 0;
+		}
+	case C_PPPOE_FORWARD:
+		if (!config->cluster_iam_master)
+		{
+			LOG(0, 0, 0, "I'm not the master, but I got a C_PPPOE_FORWARD from %s?\n", fmtaddr(addr, 0));
+			return -1;
+		}
+		else
+		{
+			pppoe_process_forward(p, s, addr);
+			return 0;
+		}
 
-		if (!config->cluster_iam_master) { // huh?
-			LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
+	case C_MPPP_FORWARD:
+		// Receive a MPPP packet from a slave.
+		if (!config->cluster_iam_master) {
+			LOG(0, 0, 0, "I'm not the master, but I got a C_MPPP_FORWARD from %s?\n", fmtaddr(addr, 0));
 			return -1;
 		}
 
-		LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
-		STAT(recv_forward);
-		processudp(p, s, &a);
+		processipout(p, s);
 		return 0;
-	}
+
 	case C_THROTTLE: {	// Receive a forwarded packet from a slave.
 		if (!config->cluster_iam_master) {
 			LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
@@ -1354,6 +1909,11 @@ int processcluster(char * data, int size, u32 addr)
 		return 0;
 
 	case C_BYTES:
+		if (!config->cluster_iam_master) {
+			LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr, 0));
+			return -1;
+		}
+
 		return cluster_handle_bytes(p, s);
 
 	case C_KILL:	// The master asked us to die!? (usually because we're too out of date).
@@ -1414,12 +1974,16 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
 				: "Not defined",
 			0.1 * (TIME - config->cluster_last_hb));
 		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 bundles undefined of %d", config->cluster_undefined_bundles, config->cluster_highest_bundleid);
 		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 bundle   : %d", config->cluster_highest_bundleid);
 		cli_print(cli, "Highest tunnel   : %d", config->cluster_highest_tunnelid);
 		cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
 	}
@@ -1428,7 +1992,7 @@ 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", fmtaddr(peers[i].peer, 0),
+		cli_print(cli, "%20s  %10u %8d", fmtaddr(peers[i].peer, 0),
 			peers[i].basetime, TIME - peers[i].timestamp);
 	}
 	return CLI_OK;
@@ -1453,11 +2017,11 @@ 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 )
-static 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) {
@@ -1499,11 +2063,11 @@ static 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.
-static 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;
+	uint8_t *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.