clear cluster_master on election so that slaves will accept a new master

[l2tpns.git] / cluster.c

diff --git a/cluster.c b/cluster.c
index a8c9310..8f9c01e 100644 (file)
--- a/cluster.c
+++ b/cluster.c
@@ -1,8 +1,12 @@
 // L2TPNS Clustering Stuff
 
 // L2TPNS Clustering Stuff
 

-char const *cvs_id_cluster = "$Id: cluster.c,v 1.7 2004/07/07 09:09:53 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.12 2005/07/03 02:58:06 bodea Exp $";

 
 #include <stdio.h>
 
 #include <stdio.h>

+#include <stdlib.h>
+#include <stdarg.h>
+#include <unistd.h>
+#include <inttypes.h>

 #include <sys/file.h>
 #include <sys/stat.h>
 #include <sys/socket.h>
 #include <sys/file.h>
 #include <sys/stat.h>
 #include <sys/socket.h>


@@ -13,10 +17,6 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.7 2004/07/07 09:09:53 bodea Exp
 #include <string.h>
 #include <malloc.h>
 #include <errno.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 <libcli.h>
 
 #include "l2tpns.h"


@@ -40,7 +40,7 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.7 2004/07/07 09:09:53 bodea Exp
 // Module variables.
 int cluster_sockfd = 0;                // The filedescriptor for the cluster communications port.
 
 // Module variables.
 int cluster_sockfd = 0;                // The filedescriptor for the cluster communications port.
 

-ipt my_address = 0;            // The network address of my ethernet port.
+in_addr_t my_address = 0;              // The network address of my ethernet port.

 static int walk_session_number = 0;    // The next session to send when doing the slow table walk.
 static int walk_tunnel_number = 0;     // The next tunnel to send when doing the slow table walk.
 
 static int 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.
 


@@ -60,16 +60,15 @@ static struct {
                                // we can re-transmit if needed.
 
 static struct {
                                // we can re-transmit if needed.
 
 static struct {

-       u32 peer;
-       time_t  basetime;
-       clockt  timestamp;
-       int     uptodate;
+       in_addr_t peer;
+       uint32_t basetime;
+       clockt timestamp;
+       int uptodate;

 } peers[CLUSTER_MAX_SIZE];     // List of all the peers we've heard from.
 static int num_peers;          // Number of peers in list.
 } peers[CLUSTER_MAX_SIZE];     // List of all the peers we've heard from.
 static int num_peers;          // Number of peers in list.

-static int have_peers;         // At least one up to date peer

 
 

-int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
-int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
+static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);

 
 //
 // Create a listening socket
 
 //
 // Create a listening socket


@@ -78,65 +77,71 @@ int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
 //
 int cluster_init()
 {
 //
 int cluster_init()
 {

-    struct sockaddr_in addr;
-    struct sockaddr_in interface_addr;
-    struct ip_mreq mreq;
-    struct ifreq   ifr;
-    int opt = 0;
+       struct sockaddr_in addr;
+       struct sockaddr_in interface_addr;
+       struct ip_mreq mreq;
+       struct ifreq   ifr;
+       int opt;

 
 

-    config->cluster_undefined_sessions = MAXSESSION-1;
-    config->cluster_undefined_tunnels = MAXTUNNEL-1;
+       config->cluster_undefined_sessions = MAXSESSION-1;
+       config->cluster_undefined_tunnels = MAXTUNNEL-1;

 
 

-    if (!config->cluster_address)
-       return 0;
-    if (!*config->cluster_interface)
-       return 0;
+       if (!config->cluster_address)
+               return 0;
+       if (!*config->cluster_interface)
+               return 0;

 
 

-    cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
+       cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);

 
 

-    memset(&addr, 0, sizeof(addr));
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_addr.s_addr = INADDR_ANY;
-    setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
+       memset(&addr, 0, sizeof(addr));
+       addr.sin_family = AF_INET;
+       addr.sin_port = htons(CLUSTERPORT);
+       addr.sin_addr.s_addr = INADDR_ANY;
+       setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));

 
 

-    if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
-    {
-       log(0, 0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
-       return -1;
-    }
+       opt = fcntl(cluster_sockfd, F_GETFL, 0);
+       fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK);

 
 

-    strcpy(ifr.ifr_name, config->cluster_interface);
-    if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0) {
-       log(0, 0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
-       return -1;
-    }
+       if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
+       {
+               LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
+               return -1;
+       }

 
 

-    memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr) );
-    my_address = interface_addr.sin_addr.s_addr;
+       strcpy(ifr.ifr_name, config->cluster_interface);
+       if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0)
+       {
+               LOG(0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
+               return -1;
+       }

 
 

-               // Join multicast group.
-    mreq.imr_multiaddr.s_addr = config->cluster_address;
-    mreq.imr_interface = interface_addr.sin_addr;
+       memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr));
+       my_address = interface_addr.sin_addr.s_addr;

 
 

+       // Join multicast group.
+       mreq.imr_multiaddr.s_addr = config->cluster_address;
+       mreq.imr_interface = interface_addr.sin_addr;

 
 

-    opt = 0;   // Turn off multicast loopback.
-    setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));

 
 

-    if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0) {
-       log(0, 0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
-       return -1;
-    }
+       opt = 0;        // Turn off multicast loopback.
+       setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));

 
 

-    if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0) {
-       log(0, 0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
-       return -1;
-    }
+       if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0)
+       {
+               LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
+               return -1;
+       }

 
 

-    config->cluster_last_hb = TIME;
-    config->cluster_seq_number = -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;
+       return cluster_sockfd;

 }
 
 
 }
 
 


@@ -145,26 +150,26 @@ int cluster_init()
 // address ).
 //
 
 // address ).
 //
 

-int cluster_send_data(void *data, int datalen)
+static int cluster_send_data(void *data, int datalen)

 {
 {

-    struct sockaddr_in addr = {0};
+       struct sockaddr_in addr = {0};

 
 

-    if (!cluster_sockfd) return -1;
-    if (!config->cluster_address) return 0;
+       if (!cluster_sockfd) return -1;
+       if (!config->cluster_address) return 0;

 
 

-    addr.sin_addr.s_addr = config->cluster_address;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_family = AF_INET;
+       addr.sin_addr.s_addr = config->cluster_address;
+       addr.sin_port = htons(CLUSTERPORT);
+       addr.sin_family = AF_INET;

 
 

-    log(5,0,0,0, "Cluster send data: %d bytes\n", datalen);
+       LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen);

 
 

-    if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
-    {
-       log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
-       return -1;
-    }
+       if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
+       {
+               LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
+               return -1;
+       }

 
 

-    return 0;
+       return 0;

 }
 
 //
 }
 
 //


@@ -172,21 +177,42 @@ int cluster_send_data(void *data, int datalen)
 // Maintains the format. Assumes that the caller
 // has passed in a big enough buffer!
 //
 // Maintains the format. Assumes that the caller
 // has passed in a big enough buffer!
 //

-static void add_type(char ** p, int type, int more, char * data, int size)
+static void add_type(char **p, int type, int more, char *data, int size)

 {
 {

-       * ( (u32*)(*p) ) = type;
-       *p += sizeof(u32);
+       *((uint32_t *) (*p)) = type;
+       *p += sizeof(uint32_t);

 
 

-       * ( (u32*)(*p) ) = more;
-       *p += sizeof(u32);
+       *((uint32_t *)(*p)) = more;
+       *p += sizeof(uint32_t);

 
        if (data && size > 0) {
                memcpy(*p, data, size);
 
        if (data && size > 0) {
                memcpy(*p, data, size);

-               (*p) += size;
+               *p += size;

        }
 }
 
        }
 }
 

-void cluster_uptodate(void)
+// advertise our presence via BGP or gratuitous ARP
+static void advertise_routes(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.
+}
+
+// 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_iam_uptodate)
                return;


@@ -196,55 +222,48 @@ void cluster_uptodate(void)
 
        config->cluster_iam_uptodate = 1;
 
 
        config->cluster_iam_uptodate = 1;
 

-       log(0,0,0,0, "Now uptodate with master.\n");
-
-#ifdef BGP
-       if (bgp_configured)
-               bgp_enable_routing(1);
-       else
-#endif /* BGP */
-               if (config->send_garp)
-                       send_garp(config->bind_address);        // Start taking traffic.
+       LOG(0, 0, 0, "Now uptodate with master.\n");
+       advertise_routes();

 }
 
 //
 // Send a unicast UDP packet to a peer with 'data' as the
 // contents.
 //
 }
 
 //
 // Send a unicast UDP packet to a peer with 'data' as the
 // contents.
 //

-int peer_send_data(u32 peer, char * data, int size)
+static int peer_send_data(in_addr_t peer, char *data, int size)

 {
 {

-    struct sockaddr_in addr = {0};
+       struct sockaddr_in addr = {0};

 
 

-    if (!cluster_sockfd) return -1;
-    if (!config->cluster_address) return 0;
+       if (!cluster_sockfd) return -1;
+       if (!config->cluster_address) return 0;

 
 

-    if (!peer) // Odd??
-       return -1;
+       if (!peer)      // Odd??
+               return -1;

 
 

-    addr.sin_addr.s_addr = peer;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_family = AF_INET;
+       addr.sin_addr.s_addr = peer;
+       addr.sin_port = htons(CLUSTERPORT);
+       addr.sin_family = AF_INET;

 
 

-    log_hex(5, "Peer send", data, size);
+       LOG_HEX(5, "Peer send", data, size);

 
 

-    if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
-    {
-       log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
-       return -1;
-    }
+       if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
+       {
+               LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
+               return -1;
+       }

 
 

-    return 0;
+       return 0;

 }
 
 //
 // Send a structured message to a peer with a single element of type 'type'.
 //
 }
 
 //
 // Send a structured message to a peer with a single element of type 'type'.
 //

-int peer_send_message(u32 peer, int type, int more, char * data, int size)
+static int peer_send_message(in_addr_t peer, int type, int more, char *data, int size)

 {
        char buf[65536];        // Vast overkill.
 {
        char buf[65536];        // Vast overkill.

-       char * p = buf;
+       char *p = buf;

 
 

-       log(4,0,0,0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
+       LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);

        add_type(&p, type, more, data, size);
 
        return peer_send_data(peer, buf, (p-buf) );
        add_type(&p, type, more, data, size);
 
        return peer_send_data(peer, buf, (p-buf) );


@@ -254,25 +273,24 @@ int peer_send_message(u32 peer, int type, int more, char * data, int size)
 // Forward a state changing packet to the master.
 //
 // The master just processes the payload as if it had
 // Forward a state changing packet to the master.
 //
 // The master just processes the payload as if it had

-// received it off the tap device.
+// received it off the tun device.

 //
 //

-int master_forward_packet(char * data, int size, u32 addr, int port)
+int master_forward_packet(char *data, int size, in_addr_t addr, int port)

 {
        char buf[65536];        // Vast overkill.
 {
        char buf[65536];        // Vast overkill.

-       char * p = buf;
+       char *p = buf;

 
        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;
 
 
        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;
 

-       log(4,0,0,0,    "Forwarding packet from %s to master (size %d)\n", inet_toa(addr), size);
+       LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);

 
        STAT(c_forwarded);
 
        STAT(c_forwarded);

-       add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) );
+       add_type(&p, C_FORWARD, addr, (char *) &port, sizeof(port));

        memcpy(p, data, size);
        p += size;
 
        memcpy(p, data, size);
        p += size;
 

-       return peer_send_data(config->cluster_master_address, buf, (p-buf) );
-
+       return peer_send_data(config->cluster_master_address, buf, (p - buf));

 }
 
 //
 }
 
 //


@@ -282,15 +300,15 @@ int master_forward_packet(char * data, int size, u32 addr, int port)
 // token bucket queue, and lets normal processing take care
 // of it.
 //
 // token bucket queue, and lets normal processing take care
 // of it.
 //

-int master_throttle_packet(int tbfid, char * data, int size)
+int master_throttle_packet(int tbfid, char *data, int size)

 {
        char buf[65536];        // Vast overkill.
 {
        char buf[65536];        // Vast overkill.

-       char * p = buf;
+       char *p = buf;

 
        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;
 
 
        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;
 

-       log(4,0,0,0,    "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
+       LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);

 
        add_type(&p, C_THROTTLE, tbfid, data, size);
 
 
        add_type(&p, C_THROTTLE, tbfid, data, size);
 


@@ -315,7 +333,7 @@ int master_garden_packet(sessionidt s, char *data, int size)
        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;
 
        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;
 

-       log(4,0,0,0,    "Walled garden packet to master (size %d)\n", size);
+       LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);

 
        add_type(&p, C_GARDEN, s, data, size);
 
 
        add_type(&p, C_GARDEN, s, data, size);
 


@@ -327,27 +345,21 @@ int master_garden_packet(sessionidt s, char *data, int size)
 // Send a chunk of data as a heartbeat..
 // We save it in the history buffer as we do so.
 //
 // Send a chunk of data as a heartbeat..
 // We save it in the history buffer as we do so.
 //

-static void send_heartbeat(int seq, char * data, int size)
+static void send_heartbeat(int seq, char *data, int size)

 {
 {

-    int i;
-    static int last_seq = -1;
-
-    if (last_seq != -1 && (seq != (last_seq+1)%HB_MAX_SEQ) ) {
-       log(0,0,0,0, "FATAL: Sequence number skipped! (%d != %d)\n",
-               seq, last_seq);
-    } 
-    last_seq = seq;
-
-    if (size > sizeof(past_hearts[0].data)) {
-       log(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
-       kill(0, SIGTERM);
-       exit(1);
-    }
-    i = seq % HB_HISTORY_SIZE;
-    past_hearts[i].seq = seq;
-    past_hearts[i].size = size;
-    memcpy(&past_hearts[i].data, data, size);  // Save it.
-    cluster_send_data(data, size);
+       int i;
+
+       if (size > sizeof(past_hearts[0].data))
+       {
+               LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
+               kill(0, SIGTERM);
+               exit(1);
+       }
+       i = seq % HB_HISTORY_SIZE;
+       past_hearts[i].seq = seq;
+       past_hearts[i].size = size;
+       memcpy(&past_hearts[i].data, data, size);       // Save it.
+       cluster_send_data(data, size);

 }
 
 //
 }
 
 //


@@ -362,7 +374,7 @@ void cluster_send_ping(time_t basetime)
        if (config->cluster_iam_master && basetime)             // We're heartbeating so no need to ping.
                return;
 
        if (config->cluster_iam_master && basetime)             // We're heartbeating so no need to ping.
                return;
 

-       log(5,0,0,0, "Sending cluster ping...\n");
+       LOG(5, 0, 0, "Sending cluster ping...\n");

 
        x.ver = 1;
        x.addr = config->bind_address;
 
        x.ver = 1;
        x.addr = config->bind_address;


@@ -375,12 +387,12 @@ void cluster_send_ping(time_t basetime)
 
 //
 // Walk the session counters looking for non-zero ones to send
 
 //
 // Walk the session counters looking for non-zero ones to send

-// to the master. We send up to 100 of them at one time.
-// We examine a maximum of 2000 sessions.
+// to the master. We send up to 600 of them at one time.
+// We examine a maximum of 3000 sessions.

 // (50k max session should mean that we normally
 // examine the entire session table every 25 seconds).
 
 // (50k max session should mean that we normally
 // examine the entire session table every 25 seconds).
 

-#define MAX_B_RECS (400)
+#define MAX_B_RECS (600)

 void master_update_counts(void)
 {
        int i, c;
 void master_update_counts(void)
 {
        int i, c;


@@ -402,18 +414,18 @@ void master_update_counts(void)
                if ( walk_session_number > config->cluster_highest_sessionid)
                        walk_session_number = 1;
 
                if ( walk_session_number > config->cluster_highest_sessionid)
                        walk_session_number = 1;
 

-               if (!sess_count[walk_session_number].cin && !sess_count[walk_session_number].cout)
+               if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)

                        continue; // Unused. Skip it.
 
                b[c].sid = walk_session_number;
                        continue; // Unused. Skip it.
 
                b[c].sid = walk_session_number;

-               b[c].in = sess_count[walk_session_number].cin;
-               b[c].out = sess_count[walk_session_number].cout;
+               b[c].in = sess_local[walk_session_number].cin;
+               b[c].out = sess_local[walk_session_number].cout;

 
                if (++c > MAX_B_RECS)   // Send a max of 400 elements in a packet.
                        break;
 
                        // Reset counters.
 
                if (++c > MAX_B_RECS)   // Send a max of 400 elements in a packet.
                        break;
 
                        // Reset counters.

-               sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
+               sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;

        }
 
        if (!c)         // Didn't find any that changes. Get out of here!
        }
 
        if (!c)         // Didn't find any that changes. Get out of here!


@@ -421,59 +433,103 @@ void master_update_counts(void)
 
 
                        // Forward the data to the master.
 
 
                        // Forward the data to the master.

-       log(4,0,0,0, "Sending byte counters to master (%d elements)\n", c);
-       peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
+       LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
+       peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);

        return;
 }
 
        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++;
+               else
+                       config->cluster_iam_uptodate = 0; // Start fast heartbeats
+       }
+
+       // 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();
+       }
+}
+

 //
 // 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)
 {
 //
 // Check that we have a master. If it's been too
 // long since we heard from a master then hold an election.
 //
 void cluster_check_master(void)
 {

-       int i, count, tcount, high_sid = 0;
+       int i, count, tcount, high_unique_id = 0;

        int last_free = 0;
        int last_free = 0;

-       int had_peers = have_peers;

        clockt t = TIME;
        static int probed = 0;
        clockt t = TIME;
        static int probed = 0;

+       int have_peers;

 
 

-       if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
+       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 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 (config->cluster_master_address
+           && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))

        {
        {

-               // 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 (config->cluster_master_address
-                   && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
+               if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval)))

                {
                {

-                       if (!probed)
-                       {
-                               probed = 1;
-                               log(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
-                                   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 (!config->cluster_iam_master)
-                       return;         // Everything's ok. return.
+                       probed = TIME;
+                       LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
+                               0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));

 
 

-               // Master needs to check peer state
+                       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;

        }
 
        }
 

-       config->cluster_last_hb = TIME + 1;
+       if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
+               return; // Everything's ok!

 
 

-       if (config->cluster_iam_master)
-               config->cluster_iam_uptodate = 1;       // cleared in loop below
-       else
-               log(0,0,0,0, "Master timed out! Holding election...\n");
+       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");

 
 

+       // In the process of shutting down, can't be master
+       if (main_quit)
+               return;

 
        for (i = have_peers = 0; i < num_peers; i++)
        {
 
        for (i = have_peers = 0; i < num_peers; i++)
        {


@@ -483,37 +539,19 @@ void cluster_check_master(void)
                if (!peers[i].basetime)
                        continue;       // Shutdown peer! Skip them.
 
                if (!peers[i].basetime)
                        continue;       // Shutdown peer! Skip them.
 

-               if (peers[i].uptodate)
-                       have_peers = 1;
-
-               if (config->cluster_iam_master)
-               {
-                       if (!peers[i].uptodate)
-                               config->cluster_iam_uptodate = 0; // Start fast heartbeats
-
-                       continue;
-               }
-

                if (peers[i].basetime < basetime) {
                if (peers[i].basetime < basetime) {

-                       log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
+                       LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));

                        return;         // They'll win the election. Get out of here.
                }
 
                if (peers[i].basetime == basetime &&
                        peers[i].peer > my_address) {
                        return;         // They'll win the election. Get out of here.
                }
 
                if (peers[i].basetime == basetime &&
                        peers[i].peer > my_address) {

-                       log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
+                       LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));

                        return;         // They'll win the election. Wait for them to come up.
                }
                        return;         // They'll win the election. Wait for them to come up.
                }

-       }

 
 

-       if (config->cluster_iam_master)         // If we're the master, we've already won
-       {
-#ifdef BGP
-               // master lost all slaves, need to handle traffic ourself
-               if (bgp_configured && had_peers && !have_peers)
-                       bgp_enable_routing(1);
-#endif /* BGP */
-               return;
+               if (peers[i].uptodate)
+                       have_peers++;

        }
 
                // Wow. it's been ages since I last heard a heartbeat
        }
 
                // Wow. it's been ages since I last heard a heartbeat


@@ -521,14 +559,13 @@ void cluster_check_master(void)
                // to become a master!!!
 
        config->cluster_iam_master = 1;
                // to become a master!!!
 
        config->cluster_iam_master = 1;

-       config->cluster_master_address = 0;

 
 

-       log(0,0,0,0, "I am declaring myself the master!\n");
+       LOG(0, 0, 0, "I am declaring myself the master!\n");

 
 

-#ifdef BGP
-       if (bgp_configured && have_peers)
-               bgp_enable_routing(0); /* stop handling traffic */
-#endif /* BGP */
+       if (have_peers < config->cluster_master_min_adv)
+               advertise_routes();
+       else
+               withdraw_routes();

 
        if (config->cluster_seq_number == -1)
                config->cluster_seq_number = 0;
 
        if (config->cluster_seq_number == -1)
                config->cluster_seq_number = 0;


@@ -553,7 +590,7 @@ void cluster_check_master(void)
                // Re-string the free list.
                // Find the ID of the highest session.
        last_free = 0;
                // Re-string the free list.
                // Find the ID of the highest session.
        last_free = 0;

-       high_sid = 0;
+       high_unique_id = 0;

        config->cluster_highest_sessionid = 0;
        for (i = 0, count = 0; i < MAXSESSION; ++i) {
                if (session[i].tunnel == T_UNDEF) {
        config->cluster_highest_sessionid = 0;
        for (i = 0, count = 0; i < MAXSESSION; ++i) {
                if (session[i].tunnel == T_UNDEF) {


@@ -561,45 +598,42 @@ void cluster_check_master(void)
                        ++count;
                }
 
                        ++count;
                }
 

-               if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
+               if (!session[i].opened) { // Unused session. Add to free list.
+                       memset(&session[i], 0, sizeof(session[i]));
+                       session[i].tunnel = T_FREE;

                        session[last_free].next = i;
                        session[i].next = 0;
                        last_free = i;
                        session[last_free].next = i;
                        session[i].next = 0;
                        last_free = i;

+                       continue;

                }
 
                }
 

-                       // Reset all the idle timeouts..
+                       // Reset idle timeouts..

                session[i].last_packet = time_now;
 
                session[i].last_packet = time_now;
 

+                       // Reset die relative to our uptime rather than the old master's
+               if (session[i].die) session[i].die = TIME;
+

                        // Accumulate un-sent byte counters.
                        // Accumulate un-sent byte counters.

-               session[i].cin += sess_count[i].cin;
-               session[i].cout += sess_count[i].cout;
-               session[i].total_cin += sess_count[i].cin;
-               session[i].total_cout += sess_count[i].cout;
+               session[i].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;

 
 

-               sess_count[i].cin = sess_count[i].cout = 0;
+               sess_local[i].cin = sess_local[i].cout = 0;

 
                session[i].radius = 0;  // Reset authentication as the radius blocks aren't up to date.
 
 
                session[i].radius = 0;  // Reset authentication as the radius blocks aren't up to date.
 

-               if (session[i].sid >= high_sid) // This is different to the index into the session table!!!
-                       high_sid = session[i].sid+1;
-
+               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.
 
                session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.

-               throttle_session(i, session[i].throttle);
-
-// I'm unsure about this. --mo
-// It's potentially a good thing, but it could send a
-// LOT of packets.
-//             if (session[i].throttle)
-//                     cluster_send_session(s);        // Tell the slaves about the new tbf indexes.
-
-               if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
-                       config->cluster_highest_sessionid = i;
+               throttle_session(i, session[i].throttle_in, session[i].throttle_out);

 
 

+               config->cluster_highest_sessionid = i;

        }
 
        session[last_free].next = 0;    // End of chain.
        }
 
        session[last_free].next = 0;    // End of chain.

-       last_sid = high_sid;            // Keep track of the highest used session ID.
+       last_id = high_unique_id;       // Keep track of the highest used session ID.

 
        become_master();
 
 
        become_master();
 


@@ -607,11 +641,11 @@ void cluster_check_master(void)
 
                // If we're not the very first master, this is a big issue!
        if(count>0)
 
                // If we're not the very first master, this is a big issue!
        if(count>0)

-               log(0,0,0,0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
+               LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);

 
        config->cluster_undefined_sessions = 0;
        config->cluster_undefined_tunnels = 0;
 
        config->cluster_undefined_sessions = 0;
        config->cluster_undefined_tunnels = 0;

-       config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
+       config->cluster_iam_uptodate = 1; // assume all peers are up-to-date

 
        // FIXME. We need to fix up the tunnel control message
        // queue here! There's a number of other variables we
 
        // FIXME. We need to fix up the tunnel control message
        // queue here! There's a number of other variables we


@@ -644,12 +678,12 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
        config->cluster_undefined_sessions = 0;
        for (i = 1 ; i < MAXSESSION; ++i) {
                if (i > highsession) {
        config->cluster_undefined_sessions = 0;
        for (i = 1 ; i < MAXSESSION; ++i) {
                if (i > highsession) {

-                       session[i].tunnel = 0; // Defined.
+                       if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.

                        continue;
                }
                        continue;
                }

-               if (session[i].tunnel != T_UNDEF)
-                       continue;
-               ++config->cluster_undefined_sessions;
+
+               if (session[i].tunnel == T_UNDEF)
+                       ++config->cluster_undefined_sessions;

        }
 
                // Clear out defined tunnels, counting the number of
        }
 
                // Clear out defined tunnels, counting the number of


@@ -657,17 +691,17 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
        config->cluster_undefined_tunnels = 0;
        for (i = 1 ; i < MAXTUNNEL; ++i) {
                if (i > hightunnel) {
        config->cluster_undefined_tunnels = 0;
        for (i = 1 ; i < MAXTUNNEL; ++i) {
                if (i > hightunnel) {

-                       tunnel[i].state = TUNNELFREE; // Defined.
+                       if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.

                        continue;
                }
                        continue;
                }

-               if (tunnel[i].state != TUNNELUNDEF)
-                       continue;
-               ++config->cluster_undefined_tunnels;
+
+               if (tunnel[i].state == TUNNELUNDEF)
+                       ++config->cluster_undefined_tunnels;

        }
 
 
        if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
        }
 
 
        if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {

-               log(2,0,0,0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
+               LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",

                        config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
                return;
        }
                        config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
                return;
        }


@@ -678,13 +712,13 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig
                cluster_uptodate();
 }
 
                cluster_uptodate();
 }
 

-int hb_add_type(char **p, int type, int id)
+static int hb_add_type(char **p, int type, int id)

 {
        switch (type) {
                case C_CSESSION: { // Compressed C_SESSION.
 {
        switch (type) {
                case C_CSESSION: { // Compressed C_SESSION.

-                       u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
-                       u8 *d = (u8 *) &session[id];
-                       u8 *orig = d;
+                       uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
+                       uint8_t *d = (uint8_t *) &session[id];
+                       uint8_t *orig = d;

                        int size;
 
                        size = rle_compress( &d,  sizeof(sessiont), c, sizeof(c) );
                        int size;
 
                        size = rle_compress( &d,  sizeof(sessiont), c, sizeof(c) );


@@ -692,19 +726,19 @@ int hb_add_type(char **p, int type, int id)
                                // Did we compress the full structure, and is the size actually
                                // reduced??
                        if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
                                // Did we compress the full structure, and is the size actually
                                // reduced??
                        if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {

-                               add_type(p, C_CSESSION, id, (char*) c, size);
+                               add_type(p, C_CSESSION, id, (char *) c, size);

                                break;
                        }
                        // Failed to compress : Fall through.
                }
                case C_SESSION: add_type(p, C_SESSION, id,
                                break;
                        }
                        // Failed to compress : Fall through.
                }
                case C_SESSION: add_type(p, C_SESSION, id,

-                       (char*) &session[id], sizeof(sessiont));
+                       (char *) &session[id], sizeof(sessiont));

                        break;
 
                case C_CTUNNEL: { // Compressed C_TUNNEL
                        break;
 
                case C_CTUNNEL: { // Compressed C_TUNNEL

-                       u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
-                       u8 *d = (u8 *) &tunnel[id];
-                       u8 *orig = d;
+                       uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
+                       uint8_t *d = (uint8_t *) &tunnel[id];
+                       uint8_t *orig = d;

                        int size;
 
                        size = rle_compress( &d,  sizeof(tunnelt), c, sizeof(c) );
                        int size;
 
                        size = rle_compress( &d,  sizeof(tunnelt), c, sizeof(c) );


@@ -718,10 +752,10 @@ int hb_add_type(char **p, int type, int id)
                        // Failed to compress : Fall through.
                }
                case C_TUNNEL: add_type(p, C_TUNNEL, id,
                        // Failed to compress : Fall through.
                }
                case C_TUNNEL: add_type(p, C_TUNNEL, id,

-                       (char*) &tunnel[id], sizeof(tunnelt));
+                       (char *) &tunnel[id], sizeof(tunnelt));

                        break;
                default:
                        break;
                default:

-                       log(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
+                       LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);

                        kill(0, SIGTERM);
                        exit(1);
        }
                        kill(0, SIGTERM);
                        exit(1);
        }


@@ -741,6 +775,8 @@ void cluster_heartbeat()
        if (!config->cluster_iam_master)        // Only the master does this.
                return;
 
        if (!config->cluster_iam_master)        // Only the master does this.
                return;
 

+       config->cluster_table_version += config->cluster_num_changes;
+

        // Fill out the heartbeat header.
        memset(&h, 0, sizeof(h));
 
        // Fill out the heartbeat header.
        memset(&h, 0, sizeof(h));
 


@@ -756,15 +792,16 @@ void cluster_heartbeat()
        h.size_tunn = sizeof(tunnelt);
        h.interval = config->cluster_hb_interval;
        h.timeout  = config->cluster_hb_timeout;
        h.size_tunn = sizeof(tunnelt);
        h.interval = config->cluster_hb_interval;
        h.timeout  = config->cluster_hb_timeout;

+       h.table_version = config->cluster_table_version;

 
 

-       add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h));
+       add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));

 
        for (i = 0; i < config->cluster_num_changes; ++i) {
                hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
        }
 
        if (p > (buff + sizeof(buff))) {        // Did we somehow manage to overun the buffer?
 
        for (i = 0; i < config->cluster_num_changes; ++i) {
                hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
        }
 
        if (p > (buff + sizeof(buff))) {        // Did we somehow manage to overun the buffer?

-               log(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
+               LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));

                kill(0, SIGTERM);
                exit(1);
        }
                kill(0, SIGTERM);
                exit(1);
        }


@@ -772,7 +809,7 @@ void cluster_heartbeat()
                //
                // Fill out the packet with sessions from the session table...
                // (not forgetting to leave space so we can get some tunnels in too )
                //
                // Fill out the packet with sessions from the session table...
                // (not forgetting to leave space so we can get some tunnels in too )

-       while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
+       while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {

 
                if (!walk_session_number)       // session #0 isn't valid.
                        ++walk_session_number;
 
                if (!walk_session_number)       // session #0 isn't valid.
                        ++walk_session_number;


@@ -792,7 +829,7 @@ void cluster_heartbeat()
                // than the session table. This is good because stuffing up a 
                // tunnel is a much bigger deal than stuffing up a session.
                //
                // than the session table. This is good because stuffing up a 
                // tunnel is a much bigger deal than stuffing up a session.
                //

-       while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
+       while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {

 
                if (!walk_tunnel_number)        // tunnel #0 isn't valid.
                        ++walk_tunnel_number;
 
                if (!walk_tunnel_number)        // tunnel #0 isn't valid.
                        ++walk_tunnel_number;


@@ -809,14 +846,16 @@ void cluster_heartbeat()
                //
                // Did we do something wrong?
        if (p > (buff + sizeof(buff))) {        // Did we somehow manage to overun the buffer?
                //
                // Did we do something wrong?
        if (p > (buff + sizeof(buff))) {        // Did we somehow manage to overun the buffer?

-               log(0,0,0,0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
+               LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));

                kill(0, SIGTERM);
                exit(1);
        }
 
                kill(0, SIGTERM);
                exit(1);
        }
 

-       log(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
-                       h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
-                       config->cluster_highest_tunnelid, (p-buff));
+       LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
+                    "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
+           HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
+           count, tcount, config->cluster_highest_sessionid,
+           config->cluster_highest_tunnelid, (int) (p - buff));

 
        config->cluster_num_changes = 0;
 
 
        config->cluster_num_changes = 0;
 


@@ -828,7 +867,7 @@ void cluster_heartbeat()
 //
 // A structure of type 'type' has changed; Add it to the queue to send.
 //
 //
 // A structure of type 'type' has changed; Add it to the queue to send.
 //

-int type_changed(int type, int id)
+static int type_changed(int type, int id)

 {
        int i;
 
 {
        int i;
 


@@ -852,7 +891,7 @@ int type_changed(int type, int id)
 int cluster_send_session(int sid)
 {
        if (!config->cluster_iam_master) {
 int cluster_send_session(int sid)
 {
        if (!config->cluster_iam_master) {

-               log(0,0,sid,0, "I'm not a master, but I just tried to change a session!\n");
+               LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");

                return -1;
        }
 
                return -1;
        }
 


@@ -863,7 +902,7 @@ int cluster_send_session(int sid)
 int cluster_send_tunnel(int tid)
 {
        if (!config->cluster_iam_master) {
 int cluster_send_tunnel(int tid)
 {
        if (!config->cluster_iam_master) {

-               log(0,0,0,tid, "I'm not a master, but I just tried to change a tunnel!\n");
+               LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");

                return -1;
        }
 
                return -1;
        }
 


@@ -876,34 +915,38 @@ int cluster_send_tunnel(int tid)
 // missed a packet. We'll resend it every packet since
 // the last one it's seen.
 //
 // missed a packet. We'll resend it every packet since
 // the last one it's seen.
 //

-int cluster_catchup_slave(int seq, u32 slave)
+static int cluster_catchup_slave(int seq, in_addr_t slave)

 {
        int s;
        int diff;
 
 {
        int s;
        int diff;
 

-       log(1,0,0,0, "Slave %s sent LASTSEEN with seq %d\n", inet_toa(slave), seq);
+       LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
+       if (!config->cluster_iam_master) {
+               LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
+                       fmtaddr(config->cluster_master_address, 0));
+
+               peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0);
+               return 0;
+       }

 
        diff = config->cluster_seq_number - seq;        // How many packet do we need to send?
        if (diff < 0)
                diff += HB_MAX_SEQ;
 
        if (diff >= HB_HISTORY_SIZE) {  // Ouch. We don't have the packet to send it!
 
        diff = config->cluster_seq_number - seq;        // How many packet do we need to send?
        if (diff < 0)
                diff += HB_MAX_SEQ;
 
        if (diff >= HB_HISTORY_SIZE) {  // Ouch. We don't have the packet to send it!

-               log(0,0,0,0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
+               LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",

                        seq, config->cluster_seq_number);
                return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
        }
 
                        seq, config->cluster_seq_number);
                return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
        }
 

+       LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) );
+

                // Now resend every packet that it missed, in order.
        while (seq != config->cluster_seq_number) {
                // 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) {
                if (seq != past_hearts[s].seq) {

-                       int i;
-                       log(0,0,0,0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
-                               inet_toa(slave), seq, past_hearts[s].seq, s, config->cluster_seq_number);
-
-                       for (i = 0; i < HB_HISTORY_SIZE; ++i) {
-                               log(0,0,0,0, "\tentry %3d: seq %d (size %d)\n", i, past_hearts[s].seq, past_hearts[s].size);
-                       }
+                       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);
                        return -1;      // What to do here!?
                }
                peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);


@@ -916,16 +959,27 @@ int cluster_catchup_slave(int seq, u32 slave)
 // We've heard from another peer! Add it to the list
 // that we select from at election time.
 //
 // We've heard from another peer! Add it to the list
 // that we select from at election time.
 //

-int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
+static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)

 {
        int i;
 {
        int i;

-       u32 clusterid;
+       in_addr_t clusterid;
+       pingt p;
+
+       // Allow for backward compatability.
+       // Just the ping packet into a new structure to allow
+       // for the possibility that we might have received
+       // more or fewer elements than we were expecting.
+       if (size > sizeof(p))
+               size = sizeof(p);

 
 

-       clusterid = p->addr;
+       memset( (void *) &p, 0, sizeof(p) );
+       memcpy( (void *) &p, (void *) pp, size);
+
+       clusterid = p.addr;

        if (clusterid != config->bind_address)
        {
                // Is this for us?
        if (clusterid != config->bind_address)
        {
                // Is this for us?

-               log(4,0,0,0, "Skipping ping from %s (different cluster)\n", inet_toa(peer));
+               LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));

                return 0;
        }
 
                return 0;
        }
 


@@ -937,12 +991,15 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
                // This peer already exists. Just update the timestamp.
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;
                // This peer already exists. Just update the timestamp.
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;

-               peers[i].uptodate = !p->undef;
+               peers[i].uptodate = !p.undef;

                break;
        }
 
                break;
        }
 

-               // Is this the master shutting down??
-       if (peer == config->cluster_master_address && !basetime) {
+       // Is this the master shutting down??
+       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();
                config->cluster_master_address = 0;
                config->cluster_last_hb = 0; // Force an election.
                cluster_check_master();


@@ -951,7 +1008,7 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
 
        if (i >= num_peers)
        {
 
        if (i >= num_peers)
        {

-               log(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));
+               LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));

 
                // Not found. Is there a stale slot to re-use?
                for (i = 0; i < num_peers ; ++i)
 
                // Not found. Is there a stale slot to re-use?
                for (i = 0; i < num_peers ; ++i)


@@ -966,36 +1023,35 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
                if (i >= CLUSTER_MAX_SIZE)
                {
                        // Too many peers!!
                if (i >= CLUSTER_MAX_SIZE)
                {
                        // Too many peers!!

-                       log(0,0,0,0, "Tried to add %s as a peer, but I already have %d of them!\n", inet_toa(peer), i);
+                       LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);

                        return -1;
                }
 
                peers[i].peer = peer;
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;
                        return -1;
                }
 
                peers[i].peer = peer;
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;

-               peers[i].uptodate = !p->undef;
+               peers[i].uptodate = !p.undef;

                if (i == num_peers)
                        ++num_peers;
 
                if (i == num_peers)
                        ++num_peers;
 

-               log(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer), num_peers);
+               LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);

        }
 
        }
 

-       if (peers[i].uptodate)
-       {
-#ifdef BGP
-               /* drop routes if we've now got a peer */
-               if (config->cluster_iam_master && bgp_configured && !have_peers)
-                       bgp_enable_routing(0);
-#endif /* BGP */
-               have_peers = 1;
-       }
-       else if (config->cluster_iam_master)
-       {
-               config->cluster_iam_uptodate = 0; // increase heart-rate...
-       }
+       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;

 
 

-       return 1;
+       LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0),
+               fmtaddr(master, 1));
+
+       config->cluster_master_address = master;
+       cluster_check_master();
+       return 0;

 }
 
 /* Handle the slave updating the byte counters for the master. */
 }
 
 /* Handle the slave updating the byte counters for the master. */


@@ -1003,20 +1059,20 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
 // Note that we don't mark the session as dirty; We rely on
 // the slow table walk to propogate this back out to the slaves.
 //
 // Note that we don't mark the session as dirty; We rely on
 // the slow table walk to propogate this back out to the slaves.
 //

-int cluster_handle_bytes(char * data, int size)
+static int cluster_handle_bytes(char *data, int size)

 {
 {

-       bytest * b;
+       bytest *b;

 
 

-       b = (bytest*) data;
+       b = (bytest *) data;

 
 

-       log(3,0,0,0, "Got byte counter update (size %d)\n", size);
+       LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);

 
                                /* Loop around, adding the byte
                                counts to each of the sessions. */
 
        while (size >= sizeof(*b) ) {
                if (b->sid > MAXSESSION) {
 
                                /* Loop around, adding the byte
                                counts to each of the sessions. */
 
        while (size >= sizeof(*b) ) {
                if (b->sid > MAXSESSION) {

-                       log(0,0,0,0, "Got C_BYTES with session #%d!\n", b->sid);
+                       LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);

                        return -1; /* Abort processing */
                }
 
                        return -1; /* Abort processing */
                }
 


@@ -1025,14 +1081,16 @@ int cluster_handle_bytes(char * data, int size)
 
                session[b->sid].cin += b->in;
                session[b->sid].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!
+
+               if (b->in)
+                       session[b->sid].last_packet = time_now; // Reset idle timer!

 
                size -= sizeof(*b);
                ++b;
        }
 
        if (size != 0)
 
                size -= sizeof(*b);
                ++b;
        }
 
        if (size != 0)

-               log(0,0,0,0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
+               LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);

 
        return size;
 }
 
        return size;
 }


@@ -1040,24 +1098,24 @@ int cluster_handle_bytes(char * data, int size)
 //
 // Handle receiving a session structure in a heartbeat packet.
 //
 //
 // Handle receiving a session structure in a heartbeat packet.
 //

-static int cluster_recv_session(int more , u8 * p)
+static int cluster_recv_session(int more, uint8_t *p)

 {
        if (more >= MAXSESSION) {
 {
        if (more >= MAXSESSION) {

-               log(0,0,0,0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
+               LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");

                return -1;
        }
 
        if (session[more].tunnel == T_UNDEF) {
                if (config->cluster_iam_uptodate) { // Sanity.
                return -1;
        }
 
        if (session[more].tunnel == T_UNDEF) {
                if (config->cluster_iam_uptodate) { // Sanity.

-                       log(0,0,0,0, "I thought I was uptodate but I just found an undefined session!\n");
+                       LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");

                } else {
                        --config->cluster_undefined_sessions;
                }
        }
 
                } else {
                        --config->cluster_undefined_sessions;
                }
        }
 

-       load_session(more, (sessiont*) p);      // Copy session into session table..
+       load_session(more, (sessiont *) p);     // Copy session into session table..

 
 

-       log(5,0,more,0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
+       LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);

 
        if (!config->cluster_iam_uptodate)
                cluster_uptodate();     // Check to see if we're up to date.
 
        if (!config->cluster_iam_uptodate)
                cluster_uptodate();     // Check to see if we're up to date.


@@ -1065,16 +1123,16 @@ static int cluster_recv_session(int more , u8 * p)
        return 0;
 }
 
        return 0;
 }
 

-static int cluster_recv_tunnel(int more, u8 *p)
+static int cluster_recv_tunnel(int more, uint8_t *p)

 {
        if (more >= MAXTUNNEL) {
 {
        if (more >= MAXTUNNEL) {

-               log(0,0,0,0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
+               LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");

                return -1;
        }
 
        if (tunnel[more].state == TUNNELUNDEF) {
                if (config->cluster_iam_uptodate) { // Sanity.
                return -1;
        }
 
        if (tunnel[more].state == TUNNELUNDEF) {
                if (config->cluster_iam_uptodate) { // Sanity.

-                       log(0,0,0,0, "I thought I was uptodate but I just found an undefined tunnel!\n");
+                       LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");

                } else {
                        --config->cluster_undefined_tunnels;
                }
                } else {
                        --config->cluster_undefined_tunnels;
                }


@@ -1089,7 +1147,7 @@ static int cluster_recv_tunnel(int more, u8 *p)
        tunnel[more].controls = tunnel[more].controle = NULL;
        tunnel[more].controlc = 0;
 
        tunnel[more].controls = tunnel[more].controle = NULL;
        tunnel[more].controlc = 0;
 

-       log(5,0,0,more, "Received tunnel update\n");
+       LOG(5, 0, more, "Received tunnel update\n");

 
        if (!config->cluster_iam_uptodate)
                cluster_uptodate();     // Check to see if we're up to date.
 
        if (!config->cluster_iam_uptodate)
                cluster_uptodate();     // Check to see if we're up to date.


@@ -1101,19 +1159,21 @@ static int cluster_recv_tunnel(int more, u8 *p)
 //
 // Process a heartbeat..
 //
 //
 // Process a heartbeat..
 //

-static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
+// v3: added interval, timeout
+// v4: added table_version
+static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)

 {
 {

-       heartt * h;
+       heartt *h;

        int s = size - (p-data);
        int i, type;
 
        int s = size - (p-data);
        int i, type;
 

-#if HB_VERSION != 3
+#if HB_VERSION != 4

 # error "need to update cluster_process_heartbeat()"
 #endif
 
 # error "need to update cluster_process_heartbeat()"
 #endif
 

-       // we handle version 2+
-       if (more < 2 || more > HB_VERSION) {
-               log(0,0,0,0, "Received a heartbeat version that I don't support (%d)!\n", more);
+       // 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??
        }
 
                return -1; // Ignore it??
        }
 


@@ -1121,7 +1181,7 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
        if (s < sizeof(*h))
                goto shortpacket;
 
        if (s < sizeof(*h))
                goto shortpacket;
 

-       h = (heartt*) p;
+       h = (heartt *) p;

        p += sizeof(*h);
        s -= sizeof(*h);
 
        p += sizeof(*h);
        s -= sizeof(*h);
 


@@ -1131,33 +1191,87 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
        if (config->cluster_iam_master) {       // Sanity...
                                // Note that this MUST match the election process above!
 
        if (config->cluster_iam_master) {       // Sanity...
                                // Note that this MUST match the election process above!
 

-               log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
+               LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));

                if (!h->basetime) {
                if (!h->basetime) {

-                       log(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr) );
+                       LOG(0, 0, 0, "Heartbeat with zero basetime!  Ignoring\n");

                        return -1; // Skip it.
                }
                        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) {
                if (basetime > h->basetime) {

-                       log(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr));
+                       LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");

                        kill(0, SIGTERM);
                        exit(1);
                }
                        kill(0, SIGTERM);
                        exit(1);
                }

-               if (basetime == h->basetime && my_address < addr) { // Tie breaker.
-                       log(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
+
+               if (basetime < h->basetime)
+                       return -1;
+
+               if (my_address < addr) { // Tie breaker.
+                       LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");

                        kill(0, SIGTERM);
                        exit(1);
                }
                        kill(0, SIGTERM);
                        exit(1);
                }

+
+                       //
+                       // Send it a unicast heartbeat to see give it a chance to die.
+                       // NOTE: It's actually safe to do seq-number - 1 without checking
+                       // for wrap around.
+                       //
+               cluster_catchup_slave(config->cluster_seq_number - 1, addr);
+

                return -1; // Skip it.
        }
 
                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!!
 
        if (config->cluster_seq_number != h->seq) {     // Out of sequence heartbeat!
        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!

-               log(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);
+               static int lastseen_seq = 0;
+               static time_t lastseen_time = 0;

 
 

-               peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 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!!
 
 
                config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
 


@@ -1178,47 +1292,44 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
                        // that the free session pointer is correct.
        cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
 
                        // that the free session pointer is correct.
        cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
 

-       if (more > 2) // reserved section of heartt was not initialized prior to v3
+       if (h->interval != config->cluster_hb_interval)

        {
        {

-               if (h->interval != config->cluster_hb_interval)
-               {
-                       log(2, 0, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
-                           h->interval, config->cluster_hb_interval);
+               LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
+                       h->interval, config->cluster_hb_interval);

 
 

-                       config->cluster_hb_interval = h->interval;
-               }
+               config->cluster_hb_interval = h->interval;
+       }

 
 

-               if (h->timeout != config->cluster_hb_timeout)
-               {
-                       log(2, 0, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
-                           h->timeout, config->cluster_hb_timeout);
+       if (h->timeout != config->cluster_hb_timeout)
+       {
+               LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
+                       h->timeout, config->cluster_hb_timeout);

 
 

-                       config->cluster_hb_timeout = h->timeout;
-               }
+               config->cluster_hb_timeout = h->timeout;

        }
 
                // Ok. process the packet...
        while ( s > 0) {
 
        }
 
                // Ok. process the packet...
        while ( s > 0) {
 

-               type = * ((u32*) p);
-               p += sizeof(u32);
-               s -= sizeof(u32);
+               type = *((uint32_t *) p);
+               p += sizeof(uint32_t);
+               s -= sizeof(uint32_t);

 
 

-               more = * ((u32*) p);
-               p += sizeof(u32);
-               s -= sizeof(u32);
+               more = *((uint32_t *) p);
+               p += sizeof(uint32_t);
+               s -= sizeof(uint32_t);

 
                switch (type) {
                        case C_CSESSION: { // Compressed session structure.
 
                switch (type) {
                        case C_CSESSION: { // Compressed session structure.

-                               u8 c [ sizeof(sessiont) + 2];
+                               uint8_t c[ sizeof(sessiont) + 2];

                                int size;
                                int size;

-                               u8 * orig_p = p;
+                               uint8_t *orig_p = p;

 
 

-                               size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
+                               size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );

                                s -= (p - orig_p);
 
                                if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
                                s -= (p - orig_p);
 
                                if (size != sizeof(sessiont) ) { // Ouch! Very very bad!

-                                       log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
+                                       LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");

                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }
                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }


@@ -1237,15 +1348,15 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
                                break;
 
                        case C_CTUNNEL: { // Compressed tunnel structure.
                                break;
 
                        case C_CTUNNEL: { // Compressed tunnel structure.

-                               u8 c [ sizeof(tunnelt) + 2];
+                               uint8_t c[ sizeof(tunnelt) + 2];

                                int size;
                                int size;

-                               u8 * orig_p = p;
+                               uint8_t *orig_p = p;

 
 

-                               size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
+                               size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));

                                s -= (p - orig_p);
 
                                if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
                                s -= (p - orig_p);
 
                                if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!

-                                       log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
+                                       LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");

                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }
                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }


@@ -1264,24 +1375,25 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32
                                s -= sizeof(tunnel[more]);
                                break;
                        default:
                                s -= sizeof(tunnel[more]);
                                break;
                        default:

-                               log(0,0,0,0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
+                               LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);

                                return -1; // can't process any more of the packet!!
                }
        }
                                return -1; // can't process any more of the packet!!
                }
        }

+

        if (config->cluster_master_address != addr)
        {
        if (config->cluster_master_address != addr)
        {

-               char *str;
-               str = strdup(inet_toa(config->cluster_master_address));
-               log(0,0,0,0, "My master just changed from %s to %s!\n", str, inet_toa(addr));
-               if (str) free(str);
+               LOG(0, 0, 0, "My master just changed from %s to %s!\n",
+                       fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
+
+               config->cluster_master_address = addr;

        }
 
        }
 

-       config->cluster_master_address = addr;

        config->cluster_last_hb = TIME; // Successfully received a heartbeat!
        config->cluster_last_hb = TIME; // Successfully received a heartbeat!

+       config->cluster_table_version = h->table_version;

        return 0;
 
 shortpacket:
        return 0;
 
 shortpacket:

-       log(0,0,0,0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
+       LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");

        return -1;
 }
 
        return -1;
 }
 


@@ -1289,16 +1401,16 @@ shortpacket:
 // We got a packet on the cluster port!
 // Handle pings, lastseens, and heartbeats!
 //
 // We got a packet on the cluster port!
 // Handle pings, lastseens, and heartbeats!
 //

-int processcluster(char * data, int size, u32 addr)
+int processcluster(char *data, int size, in_addr_t addr)

 {
        int type, more;
 {
        int type, more;

-       char * p = data;
+       char *p = data;

        int s = size;
 
        if (addr == my_address)
                return -1;      // Ignore it. Something looped back the multicast!
 
        int s = size;
 
        if (addr == my_address)
                return -1;      // Ignore it. Something looped back the multicast!
 

-       log(5,0,0,0, "Process cluster: %d bytes from %s\n", size, inet_toa(addr));
+       LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));

 
        if (s <= 0)     // Any data there??
                return -1;
 
        if (s <= 0)     // Any data there??
                return -1;


@@ -1306,42 +1418,45 @@ int processcluster(char * data, int size, u32 addr)
        if (s < 8)
                goto shortpacket;
 
        if (s < 8)
                goto shortpacket;
 

-       type = * ((u32*) p);
-       p += sizeof(u32);
-       s -= sizeof(u32);
+       type = *((uint32_t *) p);
+       p += sizeof(uint32_t);
+       s -= sizeof(uint32_t);

 
 

-       more = * ((u32*) p);
-       p += sizeof(u32);
-       s -= sizeof(u32);
+       more = *((uint32_t *) p);
+       p += sizeof(uint32_t);
+       s -= sizeof(uint32_t);

 
        switch (type) {
 
        switch (type) {

-       case C_PING:    // Update the peers table.
-               return cluster_add_peer(addr, more, (pingt*)p);
+       case C_PING: // Update the peers table.
+               return cluster_add_peer(addr, more, (pingt *) p, s);

 
 

-       case C_LASTSEEN:        // Catch up a slave (slave missed a packet).
+       case C_MASTER: // Our master is wrong
+               return cluster_set_master(addr, more);
+
+       case C_LASTSEEN: // Catch up a slave (slave missed a packet).

                return cluster_catchup_slave(more, addr);
 
        case C_FORWARD: { // Forwarded control packet. pass off to processudp.
                struct sockaddr_in a;
                a.sin_addr.s_addr = more;
 
                return cluster_catchup_slave(more, addr);
 
        case C_FORWARD: { // Forwarded control packet. pass off to processudp.
                struct sockaddr_in a;
                a.sin_addr.s_addr = more;
 

-               a.sin_port = * (int*) p;
+               a.sin_port = *(int *) p;

                s -= sizeof(int);
                p += sizeof(int);
 
                if (!config->cluster_iam_master) { // huh?
                s -= sizeof(int);
                p += sizeof(int);
 
                if (!config->cluster_iam_master) { // huh?

-                       log(0,0,0,0, "I'm not the master, but I got a C_FORWARD from %s?\n", inet_toa(addr));
+                       LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));

                        return -1;
                }
 
                        return -1;
                }
 

-               log(4,0,0,0, "Got a forwarded packet... (%s:%d)\n", inet_toa(more), a.sin_port);
+               LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);

                STAT(recv_forward);
                processudp(p, s, &a);
                return 0;
        }
        case C_THROTTLE: {      // Receive a forwarded packet from a slave.
                if (!config->cluster_iam_master) {
                STAT(recv_forward);
                processudp(p, s, &a);
                return 0;
        }
        case C_THROTTLE: {      // Receive a forwarded packet from a slave.
                if (!config->cluster_iam_master) {

-                       log(0,0,0,0, "I'm not the master, but I got a C_THROTTLE from %s?\n", inet_toa(addr));
+                       LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));

                        return -1;
                }
 
                        return -1;
                }
 


@@ -1351,7 +1466,7 @@ int processcluster(char * data, int size, u32 addr)
        case C_GARDEN:
                // Receive a walled garden packet from a slave.
                if (!config->cluster_iam_master) {
        case C_GARDEN:
                // Receive a walled garden packet from a slave.
                if (!config->cluster_iam_master) {

-                       log(0,0,0,0, "I'm not the master, but I got a C_GARDEN from %s?\n", inet_toa(addr));
+                       LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));

                        return -1;
                }
 
                        return -1;
                }
 


@@ -1359,41 +1474,46 @@ int processcluster(char * data, int size, u32 addr)
                return 0;
 
        case C_BYTES:
                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).
                if (config->cluster_iam_master) {
                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(0,0,0,0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", inet_toa(addr), more);
+                       LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);

                        return -1;
                }
                if (more != config->cluster_seq_number) {
                        return -1;
                }
                if (more != config->cluster_seq_number) {

-                       log(0,0,0,0, "The master asked us to die but the seq number didn't match!?\n");
+                       LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");

                        return -1;
                }
 
                if (addr != config->cluster_master_address) {
                        return -1;
                }
 
                if (addr != config->cluster_master_address) {

-                       log(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
-                               inet_toa(addr), config->cluster_master_address);
+                       LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
+                               fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));

                        // We can only warn about it. The master might really have switched!
                }
 
                        // We can only warn about it. The master might really have switched!
                }
 

-               log(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
+               LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");

                kill(0, SIGTERM);
                exit(0);        // Lets be paranoid;
                return -1;              // Just signalling the compiler.
 
        case C_HEARTBEAT:
                kill(0, SIGTERM);
                exit(0);        // Lets be paranoid;
                return -1;              // Just signalling the compiler.
 
        case C_HEARTBEAT:

-               log(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));
+               LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));

                return cluster_process_heartbeat(data, size, more, p, addr);
 
        default:
                return cluster_process_heartbeat(data, size, more, p, addr);
 
        default:

-               log(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type);
+               LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);

                return -1;
        }
        return 0;
 
 shortpacket:
                return -1;
        }
        return 0;
 
 shortpacket:

-       log(0,0,0,0, "I got an cluster heartbeat packet! This means I'm probably out of sync!!\n");
+       LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");

        return -1;
 }
 
        return -1;
 }
 


@@ -1406,21 +1526,25 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
        if (CLI_HELP_REQUESTED)
                return CLI_HELP_NO_ARGS;
 
        if (CLI_HELP_REQUESTED)
                return CLI_HELP_NO_ARGS;
 

-        cli_print(cli, "Cluster status   : %s", config->cluster_iam_master ? "Master" : "Slave" );
-       cli_print(cli, "My address       : %s", inet_toa(my_address));
-       cli_print(cli, "VIP address      : %s", inet_toa(config->bind_address));
-       cli_print(cli, "Multicast address: %s", inet_toa(config->cluster_address));
+       cli_print(cli, "Cluster status   : %s", config->cluster_iam_master ? "Master" : "Slave" );
+       cli_print(cli, "My address       : %s", fmtaddr(my_address, 0));
+       cli_print(cli, "VIP address      : %s", fmtaddr(config->bind_address, 0));
+       cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));

        cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
 
        cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
 

-        if (!config->cluster_iam_master) {
+       if (!config->cluster_iam_master) {

                cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",
                cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",

-                       config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
+                       config->cluster_master_address
+                               ? fmtaddr(config->cluster_master_address, 0)
+                               : "Not defined",

                        0.1 * (TIME - config->cluster_last_hb));
                        0.1 * (TIME - config->cluster_last_hb));

-                cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");
+               cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");
+               cli_print(cli, "Table version #  : %" PRIu64, config->cluster_table_version);

                cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
                cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
                cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
        } else {
                cli_print(cli, "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, "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);


@@ -1431,7 +1555,7 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
        if (num_peers)
                cli_print(cli, "%20s  %10s %8s", "Address", "Basetime", "Age");
        for (i = 0; i < num_peers; ++i) {
        if (num_peers)
                cli_print(cli, "%20s  %10s %8s", "Address", "Basetime", "Age");
        for (i = 0; i < num_peers; ++i) {

-               cli_print(cli, "%20s  %10d %8d", inet_toa(peers[i].peer),
+               cli_print(cli, "%20s  %10u %8d", fmtaddr(peers[i].peer, 0),

                        peers[i].basetime, TIME - peers[i].timestamp);
        }
        return CLI_OK;
                        peers[i].basetime, TIME - peers[i].timestamp);
        }
        return CLI_OK;


@@ -1456,11 +1580,11 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
 //
 // Worst case is a 50% expansion in space required (trying to
 // compress { 0x00, 0x01 } * N )
 //
 // Worst case is a 50% expansion in space required (trying to
 // compress { 0x00, 0x01 } * N )

-int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)

 {
        int count;
        int orig_dsize = dsize;
 {
        int count;
        int orig_dsize = dsize;

-       u8 * x,*src;
+       uint8_t *x, *src;

        src = *src_p;
 
        while (ssize > 0 && dsize > 2) {
        src = *src_p;
 
        while (ssize > 0 && dsize > 2) {


@@ -1502,11 +1626,11 @@ int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
 // Return the number of dst bytes used.
 // Updates the 'src_p' pointer to point to the
 // first un-used byte.
 // Return the number of dst bytes used.
 // Updates the 'src_p' pointer to point to the
 // first un-used byte.

-int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
+static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)

 {
        int count;
        int orig_dsize = dsize;
 {
        int count;
        int orig_dsize = dsize;

-       char * src = *src_p;
+       char *src = *src_p;

 
        while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
                count = *src++; --ssize;  // get the count byte from the source.
 
        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.