add throttle_buckets

[l2tpns.git] / cluster.c

diff --git a/cluster.c b/cluster.c
index 7c14160..2c8da60 100644 (file)
--- a/cluster.c
+++ b/cluster.c
@@ -1,6 +1,6 @@
 // L2TPNS Clustering Stuff
 
 // L2TPNS Clustering Stuff
 

-char const *cvs_id_cluster = "$Id: cluster.c,v 1.4 2004/06/28 02:43:13 fred_nerk Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.13 2004/10/29 04:25:40 bodea Exp $";

 
 #include <stdio.h>
 #include <sys/file.h>
 
 #include <stdio.h>
 #include <sys/file.h>


@@ -44,8 +44,6 @@ ipt 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.
 

-static int hsess, fsess;       // Saved copies of the highest used session id, and the first free one.
-

 #define MAX_HEART_SIZE (8192)  // Maximum size of heartbeat packet. Must be less than max IP packet size :)
 #define MAX_CHANGES  (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2)    // Assumes a session is the biggest type!
 
 #define MAX_HEART_SIZE (8192)  // Maximum size of heartbeat packet. Must be less than max IP packet size :)
 #define MAX_CHANGES  (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2)    // Assumes a session is the biggest type!
 


@@ -68,7 +66,6 @@ static struct {
        int     uptodate;
 } peers[CLUSTER_MAX_SIZE];     // List of all the peers we've heard from.
 static int num_peers;          // Number of peers in list.
        int     uptodate;
 } peers[CLUSTER_MAX_SIZE];     // List of all the peers we've heard from.
 static int num_peers;          // Number of peers in list.

-static int have_peers;         // At least one peer

 
 int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
 int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
 
 int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
 int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);


@@ -80,65 +77,68 @@ 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 = 0;

 
 

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

 
 

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

 
 

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

 
 

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

 
 

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

 
 

-    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;
-    }
+       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;
+       }

 
 

-    memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr) );
-    my_address = interface_addr.sin_addr.s_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;
+       // 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));
+       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;
-    }
+       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;
+       }

 
 

-    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_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;
+       }

 
 

-    config->cluster_last_hb = TIME;
-    config->cluster_seq_number = -1;
+       config->cluster_last_hb = TIME;
+       config->cluster_seq_number = -1;

 
 

-    return cluster_sockfd;
+       return cluster_sockfd;

 }
 
 
 }
 
 


@@ -149,26 +149,24 @@ int cluster_init()
 
 int cluster_send_data(void *data, int datalen)
 {
 
 int cluster_send_data(void *data, int datalen)
 {

-    struct sockaddr_in addr = {0};
-
-    if (!cluster_sockfd) return -1;
-    if (!config->cluster_address) return 0;
+       struct sockaddr_in addr = {0};

 
 

-    addr.sin_addr.s_addr = config->cluster_address;
-    addr.sin_port = htons(CLUSTERPORT);
-    addr.sin_family = AF_INET;
+       if (!cluster_sockfd) return -1;
+       if (!config->cluster_address) return 0;

 
 

-//    log_hex(4, "Cluster send", data, datalen); // VERY big data packets. How about we don't..
+       addr.sin_addr.s_addr = config->cluster_address;
+       addr.sin_port = htons(CLUSTERPORT);
+       addr.sin_family = AF_INET;

 
 

-    log(5,0,0,0, "Cluster send data: %d bytes\n", datalen);
+       log(5,0,0,0, "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, 0, "sendto: %s\n", strerror(errno));
+               return -1;
+       }

 
 

-    return 0;
+       return 0;

 }
 
 //
 }
 
 //


@@ -202,18 +200,13 @@ void cluster_uptodate(void)
 
        log(0,0,0,0, "Now uptodate with master.\n");
 
 
        log(0,0,0,0, "Now uptodate with master.\n");
 

-               // If we're not a master, or if we have no slaves
-               // then start taking traffic..
-       if (!config->cluster_iam_master || !have_peers)
-       {

 #ifdef BGP
 #ifdef BGP

-               if (bgp_configured)
-                       bgp_enable_routing(1);
-               else
+       if (bgp_configured)
+               bgp_enable_routing(1);
+       else

 #endif /* BGP */
 #endif /* BGP */

-                       if (config->send_garp)
-                               send_garp(config->bind_address);        // Start taking traffic.
-       }
+               if (config->send_garp)
+                       send_garp(config->bind_address);        // Start taking traffic.

 }
 
 //
 }
 
 //


@@ -222,27 +215,27 @@ void cluster_uptodate(void)
 //
 int peer_send_data(u32 peer, char * data, int size)
 {
 //
 int peer_send_data(u32 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, 0, "sendto: %s\n", strerror(errno));
+               return -1;
+       }

 
 

-    return 0;
+       return 0;

 }
 
 //
 }
 
 //


@@ -263,12 +256,12 @@ 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, u32 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;


@@ -291,10 +284,10 @@ 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;


@@ -338,17 +331,19 @@ int master_garden_packet(sessionidt s, char *data, int size)
 //
 static void send_heartbeat(int seq, char * data, int size)
 {
 //
 static void send_heartbeat(int seq, char * data, int size)
 {

-    int i;
-
-    if (size > sizeof(past_hearts[0].data)) {
-       log(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
-       kill(0, SIGTERM);
-    }
-    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(0,0,0,0, "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);

 }
 
 //
 }
 
 //


@@ -376,12 +371,12 @@ 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;


@@ -427,33 +422,97 @@ void master_update_counts(void)
        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 = 1;
+
+               if (!peers[i].uptodate)
+                       config->cluster_iam_uptodate = 0; // Start fast heartbeats
+       }
+
+#ifdef BGP
+       // master lost all slaves, need to handle traffic ourself
+       if (bgp_configured && had_peers && !have_peers)
+               bgp_enable_routing(1);
+       else if (bgp_configured && !had_peers && have_peers)
+               bgp_enable_routing(0);
+#endif /* BGP */
+}
+

 //
 // Check that we have a master. If it's been too
 // long since we heard from a master then hold an election.
 //
 void cluster_check_master(void)
 {
 //
 // 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;
        clockt t = TIME;

+       static int probed = 0;

 
 

-       if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout) )
-               return;         // Everything's ok. return.
+       if (config->cluster_iam_master)
+               return;         // Only runs on the slaves...

 
 

-       if (!config->cluster_iam_master)
-               log(0,0,0,0, "Master timed out! Holding election...\n");
+       // If the master is late (missed 2 hearbeats by a second and a
+       // hair) it may be that the switch has dropped us from the
+       // multicast group, try unicasting one probe to the master
+       // which will hopefully respond with a unicast heartbeat that
+       // will allow us to limp along until the querier next runs.
+       if (TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
+       {
+               if (!probed && config->cluster_master_address)
+               {
+                       probed = 1;
+                       log(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
+                               0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));

 
 

-       config->cluster_last_hb = TIME + 1;
+                       peer_send_message(config->cluster_master_address,
+                               C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+               }
+       } else {        // We got a recent heartbeat; reset the probe flag.
+               probed = 0;
+       }
+
+       if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
+               return; // Everything's ok!
+
+       config->cluster_last_hb = TIME + 1;     // Just the one election thanks.
+
+       log(0,0,0,0, "Master timed out! Holding election...\n");

 
 

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

                if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
                        continue;       // Stale peer! Skip them.
 
                if (!peers[i].basetime)
                        continue;       // Shutdown peer! Skip them.
 
                if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
                        continue;       // Stale peer! Skip them.
 
                if (!peers[i].basetime)
                        continue;       // Shutdown peer! Skip them.
 

-               have_peers = 1;

                if (peers[i].basetime < basetime) {
                        log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
                        return;         // They'll win the election. Get out of here.
                if (peers[i].basetime < basetime) {
                        log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
                        return;         // They'll win the election. Get out of here.


@@ -466,16 +525,6 @@ void cluster_check_master(void)
                }
        }
 
                }
        }
 

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

                // Wow. it's been ages since I last heard a heartbeat
                // and I'm better than an of my peers so it's time
                // to become a master!!!
                // Wow. it's been ages since I last heard a heartbeat
                // and I'm better than an of my peers so it's time
                // to become a master!!!


@@ -485,11 +534,6 @@ void cluster_check_master(void)
 
        log(0,0,0,0, "I am declaring myself the master!\n");
 
 
        log(0,0,0,0, "I am declaring myself the master!\n");
 

-#ifdef BGP
-       if (bgp_configured && have_peers)
-               bgp_enable_routing(0); /* stop handling traffic */
-#endif /* BGP */
-

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


@@ -513,7 +557,7 @@ 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) {


@@ -540,8 +584,8 @@ void cluster_check_master(void)
 
                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.


@@ -559,7 +603,7 @@ void cluster_check_master(void)
        }
 
        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();
 


@@ -571,12 +615,11 @@ void cluster_check_master(void)
 
        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

 
 

-               //
-               // FIXME. We need to fix up the tunnel control message
-               // queue here! There's a number of other variables we
-               // should also update.
-       cluster_uptodate();
+       // FIXME. We need to fix up the tunnel control message
+       // queue here! There's a number of other variables we
+       // should also update.

 }
 
 
 }
 
 


@@ -684,6 +727,7 @@ int hb_add_type(char **p, int type, int id)
                default:
                        log(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
                        kill(0, SIGTERM);
                default:
                        log(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
                        kill(0, SIGTERM);

+                       exit(1);

        }
        return 0;
 }
        }
        return 0;
 }


@@ -691,31 +735,33 @@ int hb_add_type(char **p, int type, int id)
 //
 // Send a heartbeat, incidently sending out any queued changes..
 //
 //
 // Send a heartbeat, incidently sending out any queued changes..
 //

-void cluster_heartbeat(int highsession, int freesession, int hightunnel)
+void cluster_heartbeat()

 {
        int i, count = 0, tcount = 0;
        char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
        heartt h;
 {
        int i, count = 0, tcount = 0;
        char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
        heartt h;

-       char * p = buff;
+       char *p = buff;

 
        if (!config->cluster_iam_master)        // Only the master does this.
                return;
 
 
        if (!config->cluster_iam_master)        // Only the master does this.
                return;
 

-       hsess = highsession;
-       fsess = freesession;
-               // Fill out the heartbeat header.
+       // Fill out the heartbeat header.
+       memset(&h, 0, sizeof(h));
+

        h.version = HB_VERSION;
        h.seq = config->cluster_seq_number;
        h.basetime = basetime;
        h.clusterid = config->bind_address;     // Will this do??
        h.basetime = basetime;
        h.version = HB_VERSION;
        h.seq = config->cluster_seq_number;
        h.basetime = basetime;
        h.clusterid = config->bind_address;     // Will this do??
        h.basetime = basetime;

-       h.highsession = highsession;
-       h.freesession = freesession;
-       h.hightunnel = hightunnel;
+       h.highsession = config->cluster_highest_sessionid;
+       h.freesession = sessionfree;
+       h.hightunnel = config->cluster_highest_tunnelid;

        h.size_sess = sizeof(sessiont);         // Just in case.
        h.size_tunn = sizeof(tunnelt);
        h.size_sess = sizeof(sessiont);         // Just in case.
        h.size_tunn = sizeof(tunnelt);

+       h.interval = config->cluster_hb_interval;
+       h.timeout  = config->cluster_hb_timeout;

 
 

-       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);
 
        for (i = 0; i < config->cluster_num_changes; ++i) {
                hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);


@@ -724,6 +770,7 @@ void cluster_heartbeat(int highsession, int freesession, int hightunnel)
        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);
                kill(0, SIGTERM);
        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);
                kill(0, SIGTERM);

+               exit(1);

        }
 
                //
        }
 
                //


@@ -734,11 +781,11 @@ void cluster_heartbeat(int highsession, int freesession, int hightunnel)
                if (!walk_session_number)       // session #0 isn't valid.
                        ++walk_session_number;
 
                if (!walk_session_number)       // session #0 isn't valid.
                        ++walk_session_number;
 

-               if (count >= highsession)       // If we're a small cluster, don't go wild.
+               if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.

                        break;
 
                hb_add_type(&p, C_CSESSION, walk_session_number);
                        break;
 
                hb_add_type(&p, C_CSESSION, walk_session_number);

-               walk_session_number = (1+walk_session_number)%(highsession+1);  // +1 avoids divide by zero.
+               walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1);    // +1 avoids divide by zero.

 
                ++count;                        // Count the number of extra sessions we're sending.
        }
 
                ++count;                        // Count the number of extra sessions we're sending.
        }


@@ -768,10 +815,11 @@ void cluster_heartbeat(int highsession, int freesession, int hightunnel)
        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);
                kill(0, SIGTERM);
        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);
                kill(0, SIGTERM);

+               exit(1);

        }
 
        }
 

-       log(3,0,0,0, "Sending heartbeat with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
-                       config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
+       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));
 
        config->cluster_num_changes = 0;
                        config->cluster_highest_tunnelid, (p-buff));
 
        config->cluster_num_changes = 0;


@@ -798,7 +846,7 @@ int type_changed(int type, int id)
        ++config->cluster_num_changes;
 
        if (config->cluster_num_changes > MAX_CHANGES)
        ++config->cluster_num_changes;
 
        if (config->cluster_num_changes > MAX_CHANGES)

-               cluster_heartbeat(config->cluster_highest_sessionid, fsess, config->cluster_highest_tunnelid);
+               cluster_heartbeat(); // flush now

 
        return 1;
 }
 
        return 1;
 }


@@ -867,12 +915,23 @@ 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)
+int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)

 {
        int i;
        u32 clusterid;
 {
        int i;
        u32 clusterid;

+       pingt p;

 
 

-       clusterid = p->addr;
+       // Allow for backward compatability.
+       // Just the ping packet into a new structure to allow
+       // for the possibility that we might have received
+       // more or fewer elements than we were expecting.
+       if (size > sizeof(p))
+               size = sizeof(p);
+
+       memset( (void*) &p, 0, sizeof(p) );
+       memcpy( (void*) &p, (void*) pp, size);
+
+       clusterid = p.addr;

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


@@ -880,14 +939,6 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
                return 0;
        }
 
                return 0;
        }
 

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

        for (i = 0; i < num_peers ; ++i)
        {
                if (peers[i].peer != peer)
        for (i = 0; i < num_peers ; ++i)
        {
                if (peers[i].peer != peer)


@@ -896,9 +947,19 @@ 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;

                break;
        }
 
                break;
        }
 

+       // Is this the master shutting down??
+       if (peer == config->cluster_master_address && !basetime) {
+               log(3,0,0,0, "Master %s shutting down...\n", inet_toa(config->cluster_master_address));
+               config->cluster_master_address = 0;
+               config->cluster_last_hb = 0; // Force an election.
+               cluster_check_master();
+               return 0;
+       }
+

        if (i >= num_peers)
        {
                log(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));
        if (i >= num_peers)
        {
                log(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));


@@ -906,8 +967,9 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
                // 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)
                {

-                       if (peers[i].peer != peer)
-                               continue;
+                       if (!peers[i].basetime) // Shutdown
+                               break;
+

                        if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
                                break;
                }
                        if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
                                break;
                }


@@ -922,20 +984,13 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
                peers[i].peer = peer;
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;
                peers[i].peer = peer;
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;

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

                if (i == num_peers)
                        ++num_peers;
 
                log(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer), num_peers);
        }
 
                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);
        }
 

-#ifdef BGP
-       /* drop routes if we've now got a peer */
-       if (bgp_configured && config->cluster_iam_master && !have_peers)
-               bgp_enable_routing(0);
-#endif /* BGP */
-
-       have_peers = 1;
-

        return 1;
 }
 
        return 1;
 }
 


@@ -1002,6 +1057,7 @@ static int cluster_recv_session(int more , u8 * p)
 
        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.

+

        return 0;
 }
 
        return 0;
 }
 


@@ -1039,23 +1095,28 @@ static int cluster_recv_tunnel(int more, u8 *p)
 
 
 //
 
 
 //

-// Process a version one heartbeat..
+// Process a heartbeat..

 //
 //

-static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u32 addr)
+static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)

 {
        heartt * h;
        int s = size - (p-data);
        int i, type;
 
 {
        heartt * h;
        int s = size - (p-data);
        int i, type;
 

-       if (more != HB_VERSION) {
-               log(0,0,0,0, "Received a heartbeat version that I don't understand!\n");
+#if HB_VERSION != 3
+# 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);

                return -1; // Ignore it??
        }
                return -1; // Ignore it??
        }

+

                // Ok. It's a heartbeat packet from a cluster master!
        if (s < sizeof(*h))
                goto shortpacket;
 
                // Ok. It's a heartbeat packet from a cluster master!
        if (s < sizeof(*h))
                goto shortpacket;
 

-

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


@@ -1068,16 +1129,18 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
 
                log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
                if (!h->basetime) {
 
                log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
                if (!h->basetime) {

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

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

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

                        kill(0, SIGTERM);
                        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");
                        kill(0, SIGTERM);
                }
                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");
                        kill(0, SIGTERM);

+                       exit(1);

                }
                return -1; // Skip it.
        }
                }
                return -1; // Skip it.
        }


@@ -1111,6 +1174,25 @@ static int cluster_process_heartbeat_v2(u8 * data, int size, int more, u8 * p, u
                        // that the free session pointer is correct.
        cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
 
                        // 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)
+               {
+                       log(2, 0, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
+                               h->interval, config->cluster_hb_interval);
+
+                       config->cluster_hb_interval = h->interval;
+               }
+
+               if (h->timeout != config->cluster_hb_timeout)
+               {
+                       log(2, 0, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
+                               h->timeout, config->cluster_hb_timeout);
+
+                       config->cluster_hb_timeout = h->timeout;
+               }
+       }
+

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


@@ -1230,7 +1312,7 @@ int processcluster(char * data, int size, u32 addr)
 
        switch (type) {
        case C_PING:    // Update the peers table.
 
        switch (type) {
        case C_PING:    // Update the peers table.

-               return cluster_add_peer(addr, more, (pingt*)p);
+               return cluster_add_peer(addr, more, (pingt*)p, s);

 
        case C_LASTSEEN:        // Catch up a slave (slave missed a packet).
                return cluster_catchup_slave(more, addr);
 
        case C_LASTSEEN:        // Catch up a slave (slave missed a packet).
                return cluster_catchup_slave(more, addr);


@@ -1298,16 +1380,16 @@ int processcluster(char * data, int size, u32 addr)
 
        case C_HEARTBEAT:
                log(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));
 
        case C_HEARTBEAT:
                log(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));

-
-               return cluster_process_heartbeat_v2(data, size, more, p, addr);
+               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);
                return -1;
        }
        return 0;
 
        default:
                log(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type);
                return -1;
        }
        return 0;

+

 shortpacket:
 shortpacket:

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

        return -1;
 }
 
        return -1;
 }
 


@@ -1320,17 +1402,17 @@ 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, "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, "Multicast i'face : %s", config->cluster_interface);
 
        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, "Multicast i'face : %s", config->cluster_interface);
 

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

                cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",
                        config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
                        0.1 * (TIME - config->cluster_last_hb));
                cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",
                        config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
                        0.1 * (TIME - config->cluster_last_hb));

-                cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");
+               cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");

                cli_print(cli, "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);
                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);