fix clear counters; show counters alignment
[l2tpns.git] / cluster.c
index 9a988e2..093d944 100644 (file)
--- a/cluster.c
+++ b/cluster.c
@@ -1,8 +1,14 @@
 // L2TPNS Clustering Stuff
-// $Id: cluster.c,v 1.2 2004/03/05 00:09:03 fred_nerk Exp $ #include <stdio.h> #include <sys/file.h>
 
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.1 2005/01/06 01:39:23 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/types.h>
 #include <sys/socket.h>
 #include <netinet/in.h>
 #include <arpa/inet.h>
 #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"
 
-int cluster_sockfd = 0;
-int cluster_server = 0;
-uint32_t vip_address;
-extern int debug;
-void _log_hex(int level, const char *title, const char *data, int maxsize);
-#define log_hex(a,b,c,d)
-#ifndef log_hex
-#define log_hex(a,b,c,d) do{if (a > debug) _log_hex(a,b,c,d);}while (0)
+#ifdef BGP
+#include "bgp.h"
 #endif
+/*
+ * All cluster packets have the same format.
+ *
+ * One or more instances of
+ *     a 32 bit 'type' id.
+ *     a 32 bit 'extra' data dependant on the 'type'.
+ *     zero or more bytes of structure data, dependant on the type.
+ *
+ */
+
+// Module variables.
+int cluster_sockfd = 0;                // The filedescriptor for the cluster communications 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.
 
+#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!
+
+static struct {
+       int type;
+       int id;
+} cluster_changes[MAX_CHANGES];        // Queue of changed structures that need to go out when next heartbeat.
+
+static struct {
+       int seq;
+       int size;
+       char 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 {
+       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(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
-int cluster_init(uint32_t bind_address, int server)
+//
+// This joins the cluster multi-cast group.
+//
+int cluster_init()
+{
+       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;
+
+       if (!config->cluster_address)
+               return 0;
+       if (!*config->cluster_interface)
+               return 0;
+
+       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));
+
+       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));
+               return -1;
+       }
+
+       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;
+       }
+
+       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, "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;
+
+       return cluster_sockfd;
+}
+
+
+//
+// Send a chunk of data to the entire cluster (usually via the multicast
+// address ).
+//
+
+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;
+
+       addr.sin_addr.s_addr = config->cluster_address;
+       addr.sin_port = htons(CLUSTERPORT);
+       addr.sin_family = AF_INET;
+
+       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, "sendto: %s\n", strerror(errno));
+               return -1;
+       }
+
+       return 0;
+}
+
+//
+// Add a chunk of data to a heartbeat packet.
+// 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)
+{
+       *((uint32_t *) (*p)) = type;
+       *p += sizeof(uint32_t);
+
+       *((uint32_t *)(*p)) = more;
+       *p += sizeof(uint32_t);
+
+       if (data && size > 0) {
+               memcpy(*p, data, size);
+               *p += size;
+       }
+}
+
+// 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;
+
+       if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
+               return;
+
+       config->cluster_iam_uptodate = 1;
+
+       LOG(0, 0, 0, "Now uptodate with master.\n");
+       advertise();
+}
+
+//
+// Send a unicast UDP packet to a peer with 'data' as the
+// contents.
+//
+static int peer_send_data(in_addr_t peer, char *data, int size)
+{
+       struct sockaddr_in addr = {0};
+
+       if (!cluster_sockfd) return -1;
+       if (!config->cluster_address) return 0;
+
+       if (!peer)      // Odd??
+               return -1;
+
+       addr.sin_addr.s_addr = peer;
+       addr.sin_port = htons(CLUSTERPORT);
+       addr.sin_family = AF_INET;
+
+       LOG_HEX(5, "Peer send", data, size);
+
+       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;
+}
+
+//
+// Send a structured message to a peer with a single element of type 'type'.
+//
+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;
+
+       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) );
+}
+
+//
+// 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, in_addr_t addr, int port)
+{
+       char buf[65536];        // Vast overkill.
+       char *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, C_FORWARD, addr, (char *) &port, sizeof(port));
+       memcpy(p, data, size);
+       p += size;
+
+       return peer_send_data(config->cluster_master_address, buf, (p - buf));
+}
+
+//
+// Forward a throttled packet to the master for handling.
+//
+// The master just drops the packet into the appropriate
+// token bucket queue, and lets normal processing take care
+// of it.
+//
+int master_throttle_packet(int tbfid, char *data, int size)
 {
-    struct sockaddr_in addr;
+       char buf[65536];        // Vast overkill.
+       char *p = buf;
+
+       if (!config->cluster_master_address) // No election has been held yet. Just skip it.
+               return -1;
 
-    vip_address = bind_address;
-    cluster_server = !!server;
+       LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
 
-    cluster_sockfd = socket(AF_INET, SOCK_DGRAM, UDP);
+       add_type(&p, C_THROTTLE, tbfid, data, size);
 
-    memset(&addr, 0, sizeof(addr));
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(cluster_server ? CLUSTERPORT : CLUSTERCLIENTPORT);
-    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)
-    {
-       perror("bind");
-       exit(-1);
-    }
+       return peer_send_data(config->cluster_master_address, buf, (p-buf) );
 
-    return cluster_sockfd;
 }
 
-int cluster_send_message(unsigned long ip_address, uint32_t vip, char type, void *data, int datalen)
+//
+// Forward a walled garden packet to the master for handling.
+//
+// The master just writes the packet straight to the tun
+// device (where is will normally loop through the
+// firewall rules, and come back in on the tun device)
+//
+// (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)
 {
-    size_t l = 1 + sizeof(uint32_t) + datalen;
-    char *buf = NULL;
-    struct sockaddr_in addr = {0};
+       char buf[65536];        // Vast overkill.
+       char *p = buf;
+
+       if (!config->cluster_master_address) // No election has been held yet. Just skip it.
+               return -1;
+
+       LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
+
+       add_type(&p, C_GARDEN, s, data, size);
+
+       return peer_send_data(config->cluster_master_address, buf, (p-buf));
+
+}
+
+//
+// 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)
+{
+       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);
+}
+
+//
+// Send an 'i am alive' message to every machine in the cluster.
+//
+void cluster_send_ping(time_t basetime)
+{
+       char buff[100 + sizeof(pingt)];
+       char *p = buff;
+       pingt x;
+
+       if (config->cluster_iam_master && basetime)             // We're heartbeating so no need to ping.
+               return;
+
+       LOG(5, 0, 0, "Sending cluster ping...\n");
+
+       x.ver = 1;
+       x.addr = config->bind_address;
+       x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
+       x.basetime = basetime;
+
+       add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
+       cluster_send_data(buff, (p-buff) );
+}
+
+//
+// Walk the session counters looking for non-zero ones to send
+// 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 (600)
+void master_update_counts(void)
+{
+       int i, c;
+       bytest b[MAX_B_RECS+1];
+
+       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.
+               return;
+
+       i = MAX_B_RECS * 5; // Examine max 2000 sessions;
+       if (config->cluster_highest_sessionid > i)
+               i = config->cluster_highest_sessionid;
+
+       for ( c = 0; i > 0 ; --i) {
+                       // Next session to look at.
+               walk_session_number++;
+               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.
+
+               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;
+
+                       // Reset counters.
+               sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
+       }
+
+       if (!c)         // Didn't find any that changes. Get out of here!
+               return;
+
+
+                       // 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);
+       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_unique_id = 0;
+       int last_free = 0;
+       clockt t = TIME;
+       static int probed = 0;
+
+       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
+       // 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)));
+
+                       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.
+
+       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.
+
+               if (peers[i].basetime < basetime) {
+                       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, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
+                       return;         // They'll win the election. Wait for them to come up.
+               }
+       }
+
+               // 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!!!
+
+       config->cluster_iam_master = 1;
+       config->cluster_master_address = 0;
+
+       LOG(0, 0, 0, "I am declaring myself the master!\n");
+
+       if (config->cluster_seq_number == -1)
+               config->cluster_seq_number = 0;
+
+               //
+               // Go through and mark all the tunnels as defined.
+               // Count the highest used tunnel number as well.
+               //
+       config->cluster_highest_tunnelid = 0;
+       for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
+               if (tunnel[i].state == TUNNELUNDEF)
+                       tunnel[i].state = TUNNELFREE;
+
+               if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
+                       config->cluster_highest_tunnelid = i;
+       }
+
+               //
+               // Go through and mark all the sessions as being defined.
+               // reset the idle timeouts.
+               // add temporary byte counters to permanent ones.
+               // Re-string the free list.
+               // Find the ID of the highest session.
+       last_free = 0;
+       high_unique_id = 0;
+       config->cluster_highest_sessionid = 0;
+       for (i = 0, count = 0; i < MAXSESSION; ++i) {
+               if (session[i].tunnel == T_UNDEF) {
+                       session[i].tunnel = T_FREE;
+                       ++count;
+               }
+
+               if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
+                       session[last_free].next = i;
+                       session[i].next = 0;
+                       last_free = i;
+               }
+
+                       // Reset all the 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;
+
+               sess_count[i].cin = sess_count[i].cout = 0;
+
+               session[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;
+       }
+
+       session[last_free].next = 0;    // End of chain.
+       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)
+               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
+
+       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.
+}
+
+
+//
+// Check that our session table is validly matching what the
+// master has in mind.
+//
+// In particular, if we have too many sessions marked 'undefined'
+// 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)
+{
+       int i;
+
+       sessionfree = freesession_ptr;  // Keep the freesession ptr valid.
+
+       if (config->cluster_iam_uptodate)
+               return;
+
+       if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
+               return;
+
+               // Clear out defined sessions, counting the number of
+               // undefs remaining.
+       config->cluster_undefined_sessions = 0;
+       for (i = 1 ; i < MAXSESSION; ++i) {
+               if (i > highsession) {
+                       session[i].tunnel = 0; // Defined.
+                       continue;
+               }
+               if (session[i].tunnel != T_UNDEF)
+                       continue;
+               ++config->cluster_undefined_sessions;
+       }
+
+               // Clear out defined tunnels, counting the number of
+               // undefs remaining.
+       config->cluster_undefined_tunnels = 0;
+       for (i = 1 ; i < MAXTUNNEL; ++i) {
+               if (i > hightunnel) {
+                       tunnel[i].state = TUNNELFREE; // Defined.
+                       continue;
+               }
+               if (tunnel[i].state != TUNNELUNDEF)
+                       continue;
+               ++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);
+               return;
+       }
+
+               // Are we up to date?
+
+       if (!config->cluster_iam_uptodate)
+               cluster_uptodate();
+}
+
+static int hb_add_type(char **p, int type, int id)
+{
+       switch (type) {
+               case C_CSESSION: { // Compressed C_SESSION.
+                       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) );
+
+                               // 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);
+                               break;
+                       }
+                       // Failed to compress : Fall through.
+               }
+               case C_SESSION: add_type(p, C_SESSION, id,
+                       (char *) &session[id], sizeof(sessiont));
+                       break;
+
+               case C_CTUNNEL: { // Compressed C_TUNNEL
+                       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) );
+
+                               // Did we compress the full structure, and is the size actually
+                               // reduced??
+                       if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
+                               add_type(p, C_CTUNNEL, id, c, size);
+                               break;
+                       }
+                       // Failed to compress : Fall through.
+               }
+               case C_TUNNEL: add_type(p, C_TUNNEL, id,
+                       (char *) &tunnel[id], sizeof(tunnelt));
+                       break;
+               default:
+                       LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
+                       kill(0, SIGTERM);
+                       exit(1);
+       }
+       return 0;
+}
+
+//
+// Send a heartbeat, incidently sending out any queued changes..
+//
+void cluster_heartbeat()
+{
+       int i, count = 0, tcount = 0;
+       char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
+       heartt h;
+       char *p = buff;
+
+       if (!config->cluster_iam_master)        // Only the master does this.
+               return;
 
-    if (!cluster_sockfd) return -1;
-    if (!ip_address) return 0;
+       config->cluster_table_version += config->cluster_num_changes;
 
-    buf = calloc(l, 1);
-    *(uint32_t *)(buf) = htonl(vip);
-    *(char *)(buf+sizeof(uint32_t)) = type;
+       // Fill out the heartbeat header.
+       memset(&h, 0, sizeof(h));
 
-    if (data && datalen > 0)
-       memcpy((char *)(buf + sizeof(uint32_t) + 1), data, datalen);
+       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 = 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;
 
-    addr.sin_addr.s_addr = ip_address;
-    addr.sin_port = htons(cluster_server ? CLUSTERCLIENTPORT : CLUSTERPORT);
-    addr.sin_family = AF_INET;
+       add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
 
-    log_hex(4, "Cluster send", buf, l);
+       for (i = 0; i < config->cluster_num_changes; ++i) {
+               hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
+       }
 
-    if (sendto(cluster_sockfd, buf, l, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
-    {
-       perror("sendto");
-       free(buf);
+       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", (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(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
+
+               if (!walk_session_number)       // session #0 isn't valid.
+                       ++walk_session_number;
+
+               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)%(config->cluster_highest_sessionid+1);    // +1 avoids divide by zero.
+
+               ++count;                        // Count the number of extra sessions we're sending.
+       }
+
+               //
+               // Fill out the packet with tunnels from the tunnel table...
+               // This effectively means we walk the tunnel table more quickly
+               // 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(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
+
+               if (!walk_tunnel_number)        // tunnel #0 isn't valid.
+                       ++walk_tunnel_number;
+
+               if (tcount >= config->cluster_highest_tunnelid)
+                       break;
+
+               hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
+               walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1);       // +1 avoids divide by zero.
+
+               ++tcount;
+       }
+
+               //
+               // 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", (int) (p - buff));
+               kill(0, SIGTERM);
+               exit(1);
+       }
+
+       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;
+
+       send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
+
+       config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
+}
+
+//
+// A structure of type 'type' has changed; Add it to the queue to send.
+//
+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.
+
+       cluster_changes[i].type = type;
+       cluster_changes[i].id = id;
+       ++config->cluster_num_changes;
+
+       if (config->cluster_num_changes > MAX_CHANGES)
+               cluster_heartbeat(); // flush now
+
+       return 1;
+}
+
+
+// A particular session has been changed!
+int cluster_send_session(int sid)
+{
+       if (!config->cluster_iam_master) {
+               LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
+               return -1;
+       }
+
+       return type_changed(C_CSESSION, sid);
+}
+
+// A particular tunnel has been changed!
+int cluster_send_tunnel(int tid)
+{
+       if (!config->cluster_iam_master) {
+               LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
+               return -1;
+       }
+
+       return type_changed(C_CTUNNEL, tid);
+}
+
+
+//
+// We're a master, and a slave has just told us that it's
+// missed a packet. We'll resend it every packet since
+// the last one it's seen.
+//
+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);
+
+       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 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.
+       }
+
+               // Now resend every packet that it missed, in order.
+       while (seq != config->cluster_seq_number) {
+               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);
+                       return -1;      // What to do here!?
+               }
+               peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
+               seq = (seq+1)%HB_MAX_SEQ;       // Increment to next seq number.
+       }
+       return 0; // All good!
+}
+
+//
+// We've heard from another peer! Add it to the list
+// that we select from at election time.
+//
+static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
+{
+       int i;
+       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);
+
+       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, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
+               return 0;
+       }
+
+       for (i = 0; i < num_peers ; ++i)
+       {
+               if (peers[i].peer != peer)
+                       continue;
+
+               // 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, "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].basetime) // Shutdown
+                               break;
+
+                       if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
+                               break;
+               }
+
+               if (i >= CLUSTER_MAX_SIZE)
+               {
+                       // Too many peers!!
+                       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, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
+       }
+
+       return 1;
+}
+
+/* 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)
+{
+       bytest *b;
+
+       b = (bytest *) data;
+
+       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, "Got C_BYTES with session #%d!\n", b->sid);
+                       return -1; /* Abort processing */
+               }
+
+               session[b->sid].total_cin += b->in;
+               session[b->sid].total_cout += b->out;
+
+               session[b->sid].cin += b->in;
+               session[b->sid].cout += b->out;
+               session[b->sid].last_packet = time_now; // Reset idle timer!
+
+               size -= sizeof(*b);
+               ++b;
+       }
+
+       if (size != 0)
+               LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
+
+       return size;
+}
+
+//
+// Handle receiving a session structure in a heartbeat packet.
+//
+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");
+               return -1;
+       }
+
+       if (session[more].tunnel == T_UNDEF) {
+               if (config->cluster_iam_uptodate) { // Sanity.
+                       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..
+
+       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, uint8_t *p)
+{
+       if (more >= MAXTUNNEL) {
+               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, "I thought I was uptodate but I just found an undefined tunnel!\n");
+               } else {
+                       --config->cluster_undefined_tunnels;
+               }
+       }
+
+       memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
+
+               //
+               // Clear tunnel control messages. These are dynamically allocated.
+               // If we get unlucky, this may cause the tunnel to drop!
+               //
+       tunnel[more].controls = tunnel[more].controle = NULL;
+       tunnel[more].controlc = 0;
+
+       LOG(5, 0, more, "Received tunnel update\n");
+
+       if (!config->cluster_iam_uptodate)
+               cluster_uptodate();     // Check to see if we're up to date.
+
+       return 0;
+}
+
+
+//
+// Process a heartbeat..
+//
+// 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;
+       int s = size - (p-data);
+       int i, type;
+
+#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;
+       p += sizeof(*h);
+       s -= sizeof(*h);
+
+       if (h->clusterid != config->bind_address)
+               return -1;      // It's not part of our cluster.
+
+       if (config->cluster_iam_master) {       // Sanity...
+                               // Note that this MUST match the election process above!
+
+               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 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, "They're an older master than me so I'm gone!\n");
+                       kill(0, SIGTERM);
+                       exit(1);
+               }
+
+               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.
+       }
+
+       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!!
+
+       if (config->cluster_seq_number != h->seq) {     // Out of sequence heartbeat!
+               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);
+
+               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!!
+
+                       // Just drop the packet. The master will resend it as part of the catchup.
+
+               return 0;
+       }
+               // Save the packet in our buffer.
+               // This is needed in case we become the master.
+       config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
+       i = h->seq % HB_HISTORY_SIZE;
+       past_hearts[i].seq = h->seq;
+       past_hearts[i].size = size;
+       memcpy(&past_hearts[i].data, data, size);       // Save it.
+
+
+                       // 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);
+
+       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 = *((uint32_t *) p);
+               p += sizeof(uint32_t);
+               s -= sizeof(uint32_t);
+
+               more = *((uint32_t *) p);
+               p += sizeof(uint32_t);
+               s -= sizeof(uint32_t);
+
+               switch (type) {
+                       case C_CSESSION: { // Compressed session structure.
+                               uint8_t c[ sizeof(sessiont) + 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(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!
+                                       break;
+                               }
+
+                               cluster_recv_session(more, c);
+                               break;
+                       }
+                       case C_SESSION:
+                               if ( s < sizeof(session[more]))
+                                       goto shortpacket;
+
+                               cluster_recv_session(more, p);
+
+                               p += sizeof(session[more]);
+                               s -= sizeof(session[more]);
+                               break;
+
+                       case C_CTUNNEL: { // Compressed tunnel structure.
+                               uint8_t c[ sizeof(tunnelt) + 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(tunnelt) ) { // 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!
+                                       break;
+                               }
+
+                               cluster_recv_tunnel(more, c);
+                               break;
+
+                       }
+                       case C_TUNNEL:
+                               if ( s < sizeof(tunnel[more]))
+                                       goto shortpacket;
+
+                               cluster_recv_tunnel(more, p);
+
+                               p += sizeof(tunnel[more]);
+                               s -= sizeof(tunnel[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!!
+               }
+       }
+
+       if (config->cluster_master_address != addr)
+       {
+               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_last_hb = TIME; // Successfully received a heartbeat!
+       config->cluster_table_version = h->table_version;
+       return 0;
+
+shortpacket:
+       LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
        return -1;
-    }
-    free(buf);
+}
+
+//
+// We got a packet on the cluster port!
+// Handle pings, lastseens, and heartbeats!
+//
+int processcluster(char *data, int size, in_addr_t addr)
+{
+       int type, more;
+       char *p = data;
+       int s = size;
+
+       if (addr == my_address)
+               return -1;      // Ignore it. Something looped back the multicast!
+
+       LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
+
+       if (s <= 0)     // Any data there??
+               return -1;
+
+       if (s < 8)
+               goto shortpacket;
+
+       type = *((uint32_t *) p);
+       p += sizeof(uint32_t);
+       s -= sizeof(uint32_t);
+
+       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);
+
+       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;
+
+               a.sin_port = *(int *) p;
+               s -= sizeof(int);
+               p += sizeof(int);
+
+               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));
+                       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);
+               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));
+                       return -1;
+               }
+
+               tbf_queue_packet(more, p, s);   // The TBF id tells wether it goes in or out.
+               return 0;
+       }
+       case C_GARDEN:
+               // Receive a walled garden packet from a slave.
+               if (!config->cluster_iam_master) {
+                       LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
+                       return -1;
+               }
+
+               tun_write(p, s);
+               return 0;
+
+       case C_BYTES:
+               return cluster_handle_bytes(p, s);
+
+       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, "_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, "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, "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, "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, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
+               return cluster_process_heartbeat(data, size, more, p, addr);
+
+       default:
+               LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
+               return -1;
+       }
+       return 0;
+
+shortpacket:
+       LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
+       return -1;
+}
+
+//====================================================================================================
+
+int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       int i;
+
+       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", 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) {
+               cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",
+                       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, "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);
+               cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
+       }
+       cli_print(cli, "%d peers.", num_peers);
+
+       if (num_peers)
+               cli_print(cli, "%20s  %10s %8s", "Address", "Basetime", "Age");
+       for (i = 0; i < num_peers; ++i) {
+               cli_print(cli, "%20s  %10u %8d", fmtaddr(peers[i].peer, 0),
+                       peers[i].basetime, TIME - peers[i].timestamp);
+       }
+       return CLI_OK;
+}
+
+//
+// Simple run-length-encoding compression.
+// Format is
+//     1 byte < 128 = count of non-zero bytes following.       // Not legal to be zero.
+//     n non-zero bytes;
+// or
+//     1 byte > 128 = (count - 128) run of zero bytes.         //
+//   repeat.
+//   count == 0 indicates end of compressed stream.
+//
+// Compress from 'src' into 'dst'. return number of bytes
+// used from 'dst'.
+// Updates *src_p to indicate 1 past last bytes used.
+//
+// We could get an extra byte in the zero runs by storing (count-1)
+// but I'm playing it safe.
+//
+// Worst case is a 50% expansion in space required (trying to
+// compress { 0x00, 0x01 } * N )
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
+{
+       int count;
+       int orig_dsize = dsize;
+       uint8_t *x, *src;
+       src = *src_p;
+
+       while (ssize > 0 && dsize > 2) {
+               count = 0;
+               x = dst++; --dsize;     // Reserve space for count byte..
+
+               if (*src) {             // Copy a run of non-zero bytes.
+                       while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
+                               *dst++ = *src++;
+                               --dsize; --ssize;
+                               ++count;
+                       }
+                       *x = count;     // Store number of non-zero bytes. Guarenteed to be non-zero!
+
+               } else {                // Compress a run of zero bytes.
+                       while (*src == 0 && count < 127 && ssize > 0) {
+                               ++src;
+                               --ssize;
+                               ++count;
+                       }
+                       *x = count | 0x80 ;
+               }
+       }
+
+       *dst++ = 0x0; // Add Stop byte.
+       --dsize;
+
+       *src_p = src;
+       return (orig_dsize - dsize);
+}
+
+//
+// Decompress the buffer into **p.
+// 'psize' is the size of the decompression buffer available.
+//
+// Returns the number of bytes decompressed.
+//
+// Decompresses from '*src_p' into 'dst'.
+// Return the number of dst bytes used.
+// Updates the 'src_p' pointer to point to the
+// first un-used byte.
+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;
+
+       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.
+               if (count == 0x0)       // End marker reached? If so, finish.
+                       break;
 
-    return 0;
+               if (count & 0x80) {     // Decompress a run of zeros
+                       for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
+                               *dst++ = 0x0;
+                               --dsize;
+                       }
+               } else {                // Copy run of non-zero bytes.
+                       for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
+                               *dst++ = *src++;
+                               --ssize; --dsize;
+                       }
+               }
+       }
+       *src_p = src;
+       return (orig_dsize - dsize);
 }