convert manual.html to docbook
[l2tpns.git] / cluster.c
index 8175119..abe2a63 100644 (file)
--- a/cluster.c
+++ b/cluster.c
@@ -1,6 +1,6 @@
 // L2TPNS Clustering Stuff
 
-char const *cvs_id_cluster = "$Id: cluster.c,v 1.33 2005/04/01 08:55:29 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.51 2006/04/27 09:53:49 bodea Exp $";
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -38,10 +38,11 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.33 2005/04/01 08:55:29 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.
 
 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
 
@@ -56,7 +57,7 @@ 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.
 
@@ -85,6 +86,7 @@ int cluster_init()
        int opt;
 
        config->cluster_undefined_sessions = MAXSESSION-1;
+       config->cluster_undefined_bundles = MAXBUNDLE-1;
        config->cluster_undefined_tunnels = MAXTUNNEL-1;
 
        if (!config->cluster_address)
@@ -127,6 +129,15 @@ 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));
@@ -178,7 +189,7 @@ 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)
 {
        *((uint32_t *) (*p)) = type;
        *p += sizeof(uint32_t);
@@ -193,7 +204,7 @@ static void add_type(char **p, int type, int more, char *data, int size)
 }
 
 // advertise our presence via BGP or gratuitous ARP
-static void advertise(void)
+static void advertise_routes(void)
 {
 #ifdef BGP
        if (bgp_configured)
@@ -204,25 +215,34 @@ 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(in_addr_t peer, char *data, int size)
+static int peer_send_data(in_addr_t peer, uint8_t *data, int size)
 {
        struct sockaddr_in addr = {0};
 
@@ -250,10 +270,10 @@ static int peer_send_data(in_addr_t peer, char *data, int size)
 //
 // 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)
+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);
@@ -261,16 +281,11 @@ static int peer_send_message(in_addr_t peer, int type, int more, char *data, int
        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)
+// send a packet to the master
+static int _forward_packet(uint8_t *data, int size, in_addr_t addr, int port, int type)
 {
-       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;
@@ -278,13 +293,30 @@ int master_forward_packet(char *data, int size, in_addr_t 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));
+       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(uint8_t *data, int size, in_addr_t addr, int port)
+{
+       return _forward_packet(data, size, addr, port, C_FORWARD);
+}
+
+// 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);
+}
+
 //
 // Forward a throttled packet to the master for handling.
 //
@@ -292,10 +324,10 @@ int master_forward_packet(char *data, int size, in_addr_t 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;
@@ -317,10 +349,10 @@ 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;
@@ -337,7 +369,7 @@ int master_garden_packet(sessionidt s, char *data, int size)
 // Send a chunk of data as a heartbeat..
 // We save it in the history buffer as we do so.
 //
-static void send_heartbeat(int seq, char *data, int size)
+static void send_heartbeat(int seq, uint8_t *data, int size)
 {
        int i;
 
@@ -359,8 +391,8 @@ 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.
@@ -373,7 +405,7 @@ void cluster_send_ping(time_t basetime)
        x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
        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) );
 }
 
@@ -393,10 +425,16 @@ 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;
+       // C_BYTES format changed in 2.1.0 (cluster version 5)
+       // during upgrade from previous versions, hang onto our counters
+       // for a bit until the new master comes up
+       if (config->cluster_last_hb_ver < 5)
+               return;
+
+       i = MAX_B_RECS * 5; // Examine max 3000 sessions;
        if (config->cluster_highest_sessionid > i)
                i = config->cluster_highest_sessionid;
 
@@ -407,17 +445,20 @@ void master_update_counts(void)
                        walk_session_number = 1;
 
                if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
-                       continue; // Unused. Skip it.
+                       continue; // Unchanged. Skip it.
 
                b[c].sid = walk_session_number;
-               b[c].in = sess_local[walk_session_number].cin;
-               b[c].out = sess_local[walk_session_number].cout;
-
-               if (++c > MAX_B_RECS)   // Send a max of 400 elements in a packet.
-                       break;
+               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_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!
@@ -426,7 +467,7 @@ 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;
 }
 
@@ -457,17 +498,22 @@ 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();
+       }
 }
 
 //
@@ -476,24 +522,26 @@ void cluster_check_slaves(void)
 //
 void cluster_check_master(void)
 {
-       int i, count, tcount, high_unique_id = 0;
+       int i, count, tcount, bcount, 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)));
 
@@ -508,10 +556,15 @@ 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.
@@ -529,6 +582,9 @@ 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
@@ -536,10 +592,14 @@ void cluster_check_master(void)
                // 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 (have_peers < config->cluster_master_min_adv)
+               advertise_routes();
+       else
+               withdraw_routes();
+
        if (config->cluster_seq_number == -1)
                config->cluster_seq_number = 0;
 
@@ -556,6 +616,19 @@ 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, bcount = 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.
@@ -586,15 +659,19 @@ void cluster_check_master(void)
                        // Reset die relative to our uptime rather than the old master's
                if (session[i].die) session[i].die = TIME;
 
-                       // Accumulate un-sent byte counters.
-               session[i].cin += sess_local[i].cin;
-               session[i].cout += sess_local[i].cout;
-               session[i].total_cin += sess_local[i].cin;
-               session[i].total_cout += sess_local[i].cout;
+                       // 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;
+
+               session[i].pin += sess_local[i].pin;
+               session[i].pout += sess_local[i].pout;
 
                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;
@@ -613,16 +690,14 @@ void cluster_check_master(void)
        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.
@@ -637,7 +712,7 @@ 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;
 
@@ -646,7 +721,7 @@ 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
@@ -662,6 +737,19 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
                        ++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;
+               }
+
+               if (bundle[i].state == BUNDLEUNDEF)
+                       ++config->cluster_undefined_bundles;
+       }
+
                // Clear out defined tunnels, counting the number of
                // undefs remaining.
        config->cluster_undefined_tunnels = 0;
@@ -676,9 +764,9 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
        }
 
 
-       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;
        }
 
@@ -688,7 +776,7 @@ 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.
@@ -702,13 +790,34 @@ 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
@@ -727,8 +836,8 @@ 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);
@@ -743,10 +852,10 @@ 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;
@@ -764,13 +873,15 @@ 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);
@@ -819,6 +930,22 @@ 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_CTUNNEL, walk_bundle_number);
+               walk_tunnel_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?
@@ -828,9 +955,9 @@ void cluster_heartbeat()
        }
 
        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",
+                    "(%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, tcount, config->cluster_highest_sessionid,
+           count, bcount, tcount, config->cluster_highest_sessionid, config->cluster_highest_bundleid,
            config->cluster_highest_tunnelid, (int) (p - buff));
 
        config->cluster_num_changes = 0;
@@ -879,6 +1006,17 @@ int cluster_send_session(int sid)
        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)
 {
@@ -902,6 +1040,13 @@ static int cluster_catchup_slave(int seq, in_addr_t slave)
        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)
@@ -913,9 +1058,11 @@ static int cluster_catchup_slave(int seq, in_addr_t slave)
                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);
@@ -968,12 +1115,13 @@ static int cluster_add_peer(in_addr_t 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)
@@ -1010,12 +1158,36 @@ static int cluster_add_peer(in_addr_t 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;
 
@@ -1032,12 +1204,17 @@ 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 += b->in;
-               session[b->sid].cout += b->out;
-               session[b->sid].last_packet = time_now; // Reset idle timer!
+               session[b->sid].cin_delta += b->cin;
+               session[b->sid].cout_delta += b->cout;
+
+               if (b->cin)
+                       session[b->sid].last_packet = time_now; // Reset idle timer!
 
                size -= sizeof(*b);
                ++b;
@@ -1077,6 +1254,31 @@ static int cluster_recv_session(int more, uint8_t *p)
        return 0;
 }
 
+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) {
@@ -1110,24 +1312,142 @@ static int cluster_recv_tunnel(int more, uint8_t *p)
 }
 
 
+// pre v5 heartbeat session structure
+struct oldsession {
+       sessionidt next;
+       sessionidt far;
+       tunnelidt tunnel;
+       in_addr_t ip;
+       int ip_pool_index;
+       unsigned long unique_id;
+       uint16_t nr;
+       uint16_t ns;
+       uint32_t magic;
+       uint32_t cin, cout;
+       uint32_t pin, pout;
+       uint32_t total_cin;
+       uint32_t total_cout;
+       uint32_t id;
+       uint16_t throttle_in;
+       uint16_t throttle_out;
+       clockt opened;
+       clockt die;
+       time_t last_packet;
+       in_addr_t dns1, dns2;
+       routet route[MAXROUTE];
+       uint16_t radius;
+       uint16_t mru;
+       uint16_t tbf_in;
+       uint16_t tbf_out;
+       uint8_t l2tp_flags;
+       uint8_t reserved_old_snoop;
+       uint8_t walled_garden;
+       uint8_t flags1;
+       char random_vector[MAXTEL];
+       int random_vector_length;
+       char user[129];
+       char called[MAXTEL];
+       char calling[MAXTEL];
+       uint32_t tx_connect_speed;
+       uint32_t rx_connect_speed;
+       uint32_t flags;
+#define SF_IPCP_ACKED  1       // Has this session seen an IPCP Ack?
+#define SF_LCP_ACKED   2       // LCP negotiated
+#define SF_CCP_ACKED   4       // CCP negotiated
+       in_addr_t snoop_ip;
+       uint16_t snoop_port;
+       uint16_t sid;
+       uint8_t filter_in;
+       uint8_t filter_out;
+       char reserved[18];
+};
+
+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->l2tp_flags;
+       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->total_cin;
+       new.cout = old->total_cout;
+       new.cin_delta = old->cin;
+       new.cout_delta = old->cout;
+       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.last_packet = old->last_packet;
+       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.snoop_ip = old->snoop_ip;
+       new.snoop_port = old->snoop_port;
+       new.walled_garden = old->walled_garden;
+
+       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]));
+
+       if (new.opened)
+       {
+               new.ppp.phase = Establish;
+               if (old->flags & (SF_IPCP_ACKED|SF_LCP_ACKED))
+               {
+                       new.ppp.phase = Network;
+                       new.ppp.lcp   = Opened;
+                       new.ppp.ipcp  = (old->flags & SF_IPCP_ACKED) ? Opened : Starting;
+                       new.ppp.ccp   = (old->flags & SF_CCP_ACKED)  ? Opened : Stopped;
+               }
+
+               // no PPPv6 in old session
+               new.ppp.ipv6cp = Stopped;
+       }
+
+       return (uint8_t *) &new;
+}
+
 //
 // Process a heartbeat..
 //
 // v3: added interval, timeout
 // v4: added table_version
+// v5: added ipv6, re-ordered session structure
 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;
+       int hb_ver = more;
 
-#if HB_VERSION != 4
+#if HB_VERSION != 5
 # 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);
+       // we handle versions 3 through 5
+       if (hb_ver < 3 || 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??
        }
 
@@ -1151,7 +1471,7 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
                        return -1; // Skip it.
                }
 
-               if (more >= 4) {
+               if (hb_ver >= 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);
@@ -1178,13 +1498,37 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
                        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!
                static int lastseen_seq = 0;
@@ -1221,7 +1565,7 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
 
                        // 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 (h->interval != config->cluster_hb_interval)
        {
@@ -1259,6 +1603,18 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
                                size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
                                s -= (p - orig_p);
 
+                               // session struct changed with v5
+                               if (hb_ver < 5)
+                               {
+                                       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!
@@ -1269,6 +1625,18 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
                                break;
                        }
                        case C_SESSION:
+                               if (hb_ver < 5)
+                               {
+                                       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;
 
@@ -1287,7 +1655,7 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
                                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");
+                                       LOG(0, 0, 0, "DANGER: Received a CTUNNEL that didn't decompress correctly!\n");
                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }
@@ -1305,6 +1673,34 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
                                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!!
@@ -1332,10 +1728,10 @@ shortpacket:
 // We got a packet on the cluster port!
 // Handle pings, lastseens, and heartbeats!
 //
-int processcluster(char *data, int size, in_addr_t 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)
@@ -1357,31 +1753,51 @@ int processcluster(char *data, int size, in_addr_t addr)
        p += sizeof(uint32_t);
        s -= sizeof(uint32_t);
 
-       switch (type) {
-       case C_PING:    // Update the peers table.
+       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_MASTER: // Our master is wrong
+               return cluster_set_master(addr, more);
 
-       case C_FORWARD: { // Forwarded control packet. pass off to processudp.
-               struct sockaddr_in a;
-               a.sin_addr.s_addr = more;
+       case C_LASTSEEN: // Catch up a slave (slave missed a packet).
+               return cluster_catchup_slave(more, addr);
 
-               a.sin_port = *(int *) p;
-               s -= sizeof(int);
-               p += sizeof(int);
+       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));
 
-               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;
                }
+               else
+               {
+                       struct sockaddr_in a;
+                       a.sin_addr.s_addr = more;
+
+                       a.sin_port = *(int *) p;
+                       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);
+
+                       return 0;
+               }
 
-               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));
@@ -1402,6 +1818,11 @@ int processcluster(char *data, int size, in_addr_t 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).
@@ -1465,11 +1886,13 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
                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);
        }
@@ -1553,7 +1976,7 @@ 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.