* Wed Jun 23 2004 David Parrish <david@dparrish.com> 2.0.0
[l2tpns.git] / l2tpns.c
index 24f62d5..ebf71ad 100644 (file)
--- a/l2tpns.c
+++ b/l2tpns.c
@@ -16,7 +16,6 @@
 #include <sys/mman.h>
 #include <netdb.h>
 #include <netinet/in.h>
-#include <signal.h>
 #include <stdarg.h>
 #include <stdlib.h>
 #include <stdio.h>
 #include <sys/stat.h>
 #include <sys/time.h>
 #include <sys/resource.h>
-#include <sys/types.h>
 #include <sys/wait.h>
 #include <linux/if.h>
 #include <stddef.h>
 #include <time.h>
 #include <dlfcn.h>
 #include <unistd.h>
+#include <sched.h>
+#include <sys/sysinfo.h>
+#include <libcli.h>
+
 #include "md5.h"
 #include "l2tpns.h"
 #include "cluster.h"
@@ -42,6 +44,7 @@
 #include "constants.h"
 #include "control.h"
 #include "util.h"
+#include "tbf.h"
 
 // Globals
 struct configt *config = NULL; // all configuration
@@ -49,11 +52,10 @@ int tapfd = -1;                     // tap interface file handle
 int udpfd = -1;                        // UDP file handle
 int controlfd = -1;            // Control signal handle
 int snoopfd = -1;              // UDP file handle for sending out intercept data
-int radfd = -1;                        // RADIUS requests file handle
+int *radfds = NULL;            // RADIUS requests file handles
 int ifrfd = -1;                        // File descriptor for routing, etc
 time_t basetime = 0;           // base clock
 char hostname[1000] = "";      // us.
-ipt myip = 0;                  // MY IP
 u16 tapmac[3];                 // MAC of tap interface
 int tapidx;                    // ifr_ifindex of tap device
 u32 sessionid = 0;             // session id for radius accounting
@@ -61,22 +63,28 @@ int syslog_log = 0;         // are we logging to syslog
 FILE *log_stream = NULL;
 struct sockaddr_in snoop_addr = {0};
 extern int cluster_sockfd;
-unsigned long last_sid = 0;
-int handle_interface = 0;
+u32 last_sid = 0;
 int clifd = 0;
 sessionidt *cli_session_kill = NULL;
 tunnelidt *cli_tunnel_kill = NULL;
 static void *ip_hash[256];
-unsigned long udp_tx = 0, udp_rx = 0, udp_rx_pkt = 0;
-unsigned long eth_tx = 0, eth_rx = 0, eth_rx_pkt = 0;
-unsigned int ip_pool_size = 0;
-time_t time_now;
+u32 udp_tx = 0, udp_rx = 0, udp_rx_pkt = 0;
+u32 eth_tx = 0, eth_rx = 0, eth_rx_pkt = 0;
+u32 ip_pool_size = 1;
+time_t time_now = 0;
 char time_now_string[64] = {0};
 char main_quit = 0;
 char *_program_name = NULL;
 linked_list *loaded_plugins;
 linked_list *plugins[MAX_PLUGIN_TYPES];
 
+#ifdef BGP
+#include "bgp.h"
+struct bgp_peer *bgp_peers = 0;
+struct bgp_route_list *bgp_routes = 0;
+int bgp_configured = 0;
+#endif /* BGP */
+
 #define membersize(STRUCT, MEMBER) sizeof(((STRUCT *)0)->MEMBER)
 #define CONFIG(NAME, MEMBER, TYPE) { NAME, offsetof(struct configt, MEMBER), membersize(struct configt, MEMBER), TYPE }
 
@@ -87,18 +95,29 @@ struct config_descriptt config_values[] = {
        CONFIG("primary_dns", default_dns1, IP),
        CONFIG("secondary_dns", default_dns2, IP),
        CONFIG("save_state", save_state, BOOL),
-       CONFIG("snoop_host", snoop_destination_host, IP),
-       CONFIG("snoop_port", snoop_destination_port, SHORT),
        CONFIG("primary_radius", radiusserver[0], IP),
        CONFIG("secondary_radius", radiusserver[1], IP),
        CONFIG("radius_accounting", radius_accounting, BOOL),
        CONFIG("radius_secret", radiussecret, STRING),
        CONFIG("bind_address", bind_address, IP),
-       CONFIG("cluster_master", cluster_address, IP),
+       CONFIG("send_garp", send_garp, BOOL),
        CONFIG("throttle_speed", rl_rate, UNSIGNED_LONG),
        CONFIG("accounting_dir", accounting_dir, STRING),
        CONFIG("setuid", target_uid, INT),
        CONFIG("dump_speed", dump_speed, BOOL),
+       CONFIG("cleanup_interval", cleanup_interval, INT),
+       CONFIG("multi_read_count", multi_read_count, INT),
+       CONFIG("scheduler_fifo", scheduler_fifo, BOOL),
+       CONFIG("icmp_rate", icmp_rate, INT),
+       CONFIG("cluster_address", cluster_address, IP),
+       CONFIG("cluster_interface", cluster_interface, STRING),
+#ifdef BGP
+       CONFIG("as_number", as_number, SHORT),
+       CONFIG("bgp_peer1", bgp_peer[0], STRING),
+       CONFIG("bgp_peer1_as", bgp_peer_as[0], SHORT),
+       CONFIG("bgp_peer2", bgp_peer[1], STRING),
+       CONFIG("bgp_peer2_as", bgp_peer_as[1], SHORT),
+#endif /* BGP */
        { NULL, 0, 0, 0 },
 };
 
@@ -113,11 +132,15 @@ char *plugin_functions[] = {
        "plugin_kill_session",
        "plugin_control",
        "plugin_radius_response",
+       "plugin_become_master",
+       "plugin_new_session_master",
 };
+
 #define max_plugin_functions (sizeof(plugin_functions) / sizeof(char *))
 
 tunnelt *tunnel = NULL;                // 1000 * 45 = 45000 = 45k
 sessiont *session = NULL;      // 5000 * 213 = 1065000 = 1 Mb
+sessioncountt *sess_count = NULL;
 radiust *radius = NULL;
 ippoolt *ip_address_pool = NULL;
 controlt *controlfree = 0;
@@ -125,14 +148,12 @@ struct Tstats *_statistics = NULL;
 #ifdef RINGBUFFER
 struct Tringbuffer *ringbuffer = NULL;
 #endif
-tbft *filter_buckets = NULL;
 
 void sigalrm_handler(int);
 void sighup_handler(int);
 void sigterm_handler(int);
 void sigquit_handler(int);
 void sigchild_handler(int);
-void sigsegv_handler(int);
 void read_config_file();
 void read_state();
 void dump_state();
@@ -140,6 +161,9 @@ void tunnel_clean();
 tunnelidt new_tunnel();
 void update_config();
 
+static void cache_ipmap(ipt ip, int s);
+static void uncache_ipmap(ipt ip);
+
 // return internal time (10ths since run)
 clockt now(void)
 {
@@ -157,8 +181,8 @@ clockt backoff(u8 try)
 
 void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format, ...)
 {
-       static char message[65535] = {0};
-       static char message2[65535] = {0};
+       static char message[65536] = {0};
+       static char message2[65536] = {0};
        va_list ap;
 
 #ifdef RINGBUFFER
@@ -187,7 +211,7 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
        {
                vsnprintf(message2, 65535, format, ap);
                snprintf(message, 65535, "%s %02d/%02d %s", time_now_string, t, s, message2);
-               fprintf(log_stream, message);
+               fprintf(log_stream, "%s", message);
        }
        else if (syslog_log)
        {
@@ -201,7 +225,7 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
 void _log_hex(int level, ipt address, sessionidt s, tunnelidt t, const char *title, const char *data, int maxsize)
 {
        int i, j;
-       unsigned const char *d = (unsigned const char *)data;
+       const u8 *d = (const u8 *)data;
 
        if (config->debug < level) return;
 
@@ -251,9 +275,22 @@ void _log_hex(int level, ipt address, sessionidt s, tunnelidt t, const char *tit
 
 
 // Add a route
-void routeset(ipt ip, ipt mask, ipt gw, u8 add)
+//
+// This adds it to the routing table, advertises it
+// via iBGP if enabled, and stuffs it into the
+// 'sessionbyip' cache.
+//
+// 'ip' and 'mask' must be in _host_ order.
+//
+void routeset(sessionidt s, ipt ip, ipt mask, ipt gw, u8 add)
 {
        struct rtentry r;
+       int i;
+
+       if (!mask) mask = 0xffffffff;
+
+       ip = ip & mask;         // Force the ip to be the first one in the route.
+
        memset(&r, 0, sizeof(r));
        r.rt_dev = config->tapdevice;
        r.rt_dst.sa_family = AF_INET;
@@ -261,14 +298,44 @@ void routeset(ipt ip, ipt mask, ipt gw, u8 add)
        r.rt_gateway.sa_family = AF_INET;
        *(u32 *) & (((struct sockaddr_in *) &r.rt_gateway)->sin_addr.s_addr) = htonl(gw);
        r.rt_genmask.sa_family = AF_INET;
-       *(u32 *) & (((struct sockaddr_in *) &r.rt_genmask)->sin_addr.s_addr) = htonl(mask ? mask : 0xFFFFFFF);
+       *(u32 *) & (((struct sockaddr_in *) &r.rt_genmask)->sin_addr.s_addr) = htonl(mask);
        r.rt_flags = (RTF_UP | RTF_STATIC);
        if (gw)
                r.rt_flags |= RTF_GATEWAY;
-       else
+       else if (mask == 0xffffffff)
                r.rt_flags |= RTF_HOST;
-       if (ioctl(ifrfd, add ? SIOCADDRT : SIOCDELRT, (void *) &r) < 0) perror("routeset");
-       log(1, ip, 0, 0, "Route %s %u.%u.%u.%u/%u.%u.%u.%u %u.%u.%u.%u\n", add ? "Add" : "Del", ip >> 24, ip >> 16 & 255, ip >> 8 & 255, ip & 255, mask >> 24, mask >> 16 & 255, mask >> 8 & 255, mask & 255, gw >> 24, gw >> 16 & 255, gw >> 8 & 255, gw & 255);
+       if (ioctl(ifrfd, add ? SIOCADDRT : SIOCDELRT, (void *) &r) < 0)
+               log(0, 0, 0, 0, "routeset() error in ioctl: %s\n", strerror(errno));
+
+       log(1, ip, 0, 0, "Route %s %u.%u.%u.%u/%u.%u.%u.%u %u.%u.%u.%u\n",
+           add ? "add" : "del",
+           ip   >> 24, ip   >> 16 & 0xff, ip   >> 8 & 0xff, ip   & 0xff,
+           mask >> 24, mask >> 16 & 0xff, mask >> 8 & 0xff, mask & 0xff,
+           gw   >> 24, gw   >> 16 & 0xff, gw   >> 8 & 0xff, gw   & 0xff);
+
+#ifdef BGP
+       if (add)
+               bgp_add_route(htonl(ip), htonl(mask));
+       else
+               bgp_del_route(htonl(ip), htonl(mask));
+#endif /* BGP */
+
+               // Add/Remove the IPs to the 'sessionbyip' cache.
+               // Note that we add the zero address in the case of 
+               // a network route. Roll on CIDR.
+
+               // Note that 's == 0' implies this is the address pool.
+               // We still cache it here, because it will pre-fill
+               // the malloc'ed tree.
+
+       if (s)
+       {
+               if (!add)       // Are we deleting a route?
+                       s = 0;  // Caching the session as '0' is the same as uncaching.
+
+               for (i = ip; (i&mask) == (ip&mask) ; ++i)
+                       cache_ipmap(i, s);
+       }
 }
 
 // Set up TAP interface
@@ -283,42 +350,53 @@ void inittap(void)
        if (tapfd < 0)
        {                          // fatal
                log(0, 0, 0, 0, "Can't open %s: %s\n", TAPDEVICE, strerror(errno));
-               exit(-1);
+               exit(1);
+       }
+       {
+               int flags = fcntl(tapfd, F_GETFL, 0);
+               fcntl(tapfd, F_SETFL, flags | O_NONBLOCK);
        }
        if (ioctl(tapfd, TUNSETIFF, (void *) &ifr) < 0)
        {
                log(0, 0, 0, 0, "Can't set tap interface: %s\n", strerror(errno));
-               exit(-1);
+               exit(1);
        }
        assert(strlen(ifr.ifr_name) < sizeof(config->tapdevice));
        strncpy(config->tapdevice, ifr.ifr_name, sizeof(config->tapdevice) - 1);
        ifrfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 
        sin.sin_family = AF_INET;
-       sin.sin_addr.s_addr = handle_interface ? config->bind_address : 0x01010101; // 1.1.1.1
+       sin.sin_addr.s_addr = config->bind_address ? config->bind_address : 0x01010101; // 1.1.1.1
        memcpy(&ifr.ifr_addr, &sin, sizeof(struct sockaddr));
 
        if (ioctl(ifrfd, SIOCSIFADDR, (void *) &ifr) < 0)
        {
-               perror("set tap addr");
-               exit( -1);
+               log(0, 0, 0, 0, "Error setting tun address: %s\n", strerror(errno));
+               exit(1);
+       }
+       /* Bump up the qlen to deal with bursts from the network */
+       ifr.ifr_qlen = 1000;
+       if (ioctl(ifrfd, SIOCSIFTXQLEN, (void *) &ifr) < 0)
+       {
+               log(0, 0, 0, 0, "Error setting tun queue length: %s\n", strerror(errno));
+               exit(1);
        }
        ifr.ifr_flags = IFF_UP;
        if (ioctl(ifrfd, SIOCSIFFLAGS, (void *) &ifr) < 0)
        {
-               perror("set tap flags");
-               exit( -1);
+               log(0, 0, 0, 0, "Error setting tun flags: %s\n", strerror(errno));
+               exit(1);
        }
        if (ioctl(ifrfd, SIOCGIFHWADDR, (void *) &ifr) < 0)
        {
-               perror("get tap hwaddr");
-               exit( -1);
+               log(0, 0, 0, 0, "Error setting tun hardware address: %s\n", strerror(errno));
+               exit(1);
        }
        memcpy(&tapmac, 2 + (u8 *) & ifr.ifr_hwaddr, 6);
        if (ioctl(ifrfd, SIOCGIFINDEX, (void *) &ifr) < 0)
        {
-               perror("get tap ifindex");
-               exit( -1);
+               log(0, 0, 0, 0, "Error setting tun ifindex: %s\n", strerror(errno));
+               exit(1);
        }
        tapidx = ifr.ifr_ifindex;
 }
@@ -336,12 +414,17 @@ void initudp(void)
        addr.sin_addr.s_addr = config->bind_address;
        udpfd = socket(AF_INET, SOCK_DGRAM, UDP);
        setsockopt(udpfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
+       {
+               int flags = fcntl(udpfd, F_GETFL, 0);
+               fcntl(udpfd, F_SETFL, flags | O_NONBLOCK);
+       }
        if (bind(udpfd, (void *) &addr, sizeof(addr)) < 0)
        {
-               perror("udp bind");
-               exit( -1);
+               log(0, 0, 0, 0, "Error in UDP bind: %s\n", strerror(errno));
+               exit(1);
        }
        snoopfd = socket(AF_INET, SOCK_DGRAM, UDP);
+       snoop_addr.sin_family = AF_INET;
 
        // Control
        memset(&addr, 0, sizeof(addr));
@@ -351,31 +434,62 @@ void initudp(void)
        setsockopt(controlfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
        if (bind(controlfd, (void *) &addr, sizeof(addr)) < 0)
        {
-               perror("bind");
-               exit(-1);
+               log(0, 0, 0, 0, "Error in control bind: %s\n", strerror(errno));
+               exit(1);
        }
 }
 
-// Find session by IP, 0 for not found
-sessionidt sessionbyip(ipt ip)
+//
+// Find session by IP, < 1 for not found
+//
+// Confusingly enough, this 'ip' must be
+// in _network_ order. This being the common
+// case when looking it up from IP packet headers.
+//
+// We actually use this cache for two things.
+// #1. For used IP addresses, this maps to the
+// session ID that it's used by.
+// #2. For un-used IP addresses, this maps to the
+// index into the pool table that contains that
+// IP address.
+//
+
+int lookup_ipmap(ipt ip)
 {
-       unsigned char *a = (unsigned char *)&ip;
+       u8 *a = (u8 *)&ip;
        char **d = (char **) ip_hash;
-
-#ifdef STAT_CALLS
-       STAT(call_sessionbyip);
-#endif
+       int s;
 
        if (!(d = (char **) d[(size_t) *a++])) return 0;
        if (!(d = (char **) d[(size_t) *a++])) return 0;
        if (!(d = (char **) d[(size_t) *a++])) return 0;
 
-       return (ipt) d[(size_t) *a];
+       s = (ipt) d[(size_t) *a];
+       return s;
+}
+
+sessionidt sessionbyip(ipt ip)
+{
+       int s = lookup_ipmap(ip);
+
+#ifdef STAT_CALLS
+       STAT(call_sessionbyip);
+#endif
+       if (s > 0 && s < MAXSESSION && session[s].tunnel)
+               return s;
+       return 0;
 }
 
-void cache_sessionid(ipt ip, sessionidt s)
+//
+// Take an IP address in HOST byte order and
+// add it to the sessionid by IP cache.
+//
+// (It's actually cached in network order)
+//
+static void cache_ipmap(ipt ip, int s)
 {
-       unsigned char *a = (unsigned char *) &ip;
+       ipt nip = htonl(ip);            // MUST be in network order. I.e. MSB must in be ((char*)(&ip))[0]
+       u8 *a = (u8 *) &nip;
        char **d = (char **) ip_hash;
        int i;
 
@@ -390,66 +504,105 @@ void cache_sessionid(ipt ip, sessionidt s)
                d = (char **) d[(size_t) a[i]];
        }
 
-       log(4, ip, s, session[s].tunnel, "Caching session ID %d for ip address\n", s);
        d[(size_t) a[3]] = (char *)((int)s);
+
+       if (s > 0)
+               log(4, ip, s, session[s].tunnel, "Caching ip address %s\n", inet_toa(nip));
+       else if (s == 0)
+               log(4, ip, 0, 0, "Un-caching ip address %s\n", inet_toa(nip));
+       // else a map to an ip pool index.
 }
 
-void uncache_sessionid(ipt ip)
+static void uncache_ipmap(ipt ip)
 {
-       unsigned char *a = (unsigned char *) &ip;
-       char **d = (char **) ip_hash;
-       int i;
+       cache_ipmap(ip, 0);     // Assign it to the NULL session.
+}
 
-       for (i = 0; i < 3; i++)
-       {
-               if (!d[(size_t) a[i]]) return;
-               d = (char **) d[(size_t) a[i]];
+//
+// CLI list to dump current ipcache.
+//
+int cmd_show_ipcache(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       char **d = (char **) ip_hash, **e, **f, **g;
+       int i, j, k, l;
+       int count = 0;
+
+        cli_print(cli, "%7s %s", "Sess#", "IP Address");
+
+       for (i = 0; i < 256; ++i) {
+               if (!d[i]) continue;
+               e = (char**) d[i];
+               for (j = 0; j < 256; ++j) {
+                       if (!e[j]) continue;
+                       f = (char**) e[j];
+                       for (k = 0; k < 256; ++k) {
+                               if (!f[k]) continue;
+                               g = (char**)f[k];
+                               for (l = 0; l < 256; ++l) {
+                                       if (!g[l]) continue;
+                                       cli_print(cli, "%7d %d.%d.%d.%d",  (int) g[l], i, j, k, l);
+                                       ++count;
+                               }
+                       }
+               }
        }
-       d[(size_t) a[3]] = NULL;
+       cli_print(cli, "%d entries in cache", count);
+       return CLI_OK;
 }
 
+
 // Find session by username, 0 for not found
 // walled garden users aren't authenticated, so the username is
 // reasonably useless. Ignore them to avoid incorrect actions
+//
+// This is VERY inefficent. Don't call it often. :)
+//
 sessionidt sessionbyuser(char *username)
 {
        int s;
 #ifdef STAT_CALLS
        STAT(call_sessionbyuser);
 #endif
-       for (s = 1; s < MAXSESSION && (session[s].walled_garden || strncmp(session[s].user, username, 128)); s++);
-       if (s < MAXSESSION)
-               return s;
-       return 0;
+       for (s = 1; s < MAXSESSION ; ++s) {
+               if (session[s].walled_garden)
+                       continue;               // Skip walled garden users.
+
+               if (!strncmp(session[s].user, username, 128))
+                       return s;
+
+       }
+       return 0;       // Not found.
 }
 
 void send_garp(ipt ip)
 {
-    int s;
-    struct ifreq ifr;
-    unsigned char mac[6];
+       int s;
+       struct ifreq ifr;
+       u8 mac[6];
 
-    s = socket(PF_INET, SOCK_DGRAM, 0);
-    if (s < 0)
-    {
-      perror("socket");
-      exit(-1);
-    }
-    memset(&ifr, 0, sizeof(ifr));
-    strncpy(ifr.ifr_name, "eth0", sizeof(ifr.ifr_name) - 1);
-    if (ioctl(s, SIOCGIFHWADDR, &ifr) < 0)
-    {
-      perror("get eth0 hwaddr");
-      exit(-1);
-    }
-    memcpy(mac, &ifr.ifr_hwaddr.sa_data, 6*sizeof(char));
-    if (ioctl(s, SIOCGIFINDEX, &ifr) < 0)
-    {
-      perror("get eth0 ifindex");
-      exit(-1);
-    }
-    close(s);
-    sendarp(ifr.ifr_ifindex, mac, ip);
+       s = socket(PF_INET, SOCK_DGRAM, 0);
+       if (s < 0)
+       {
+               log(0, 0, 0, 0, "Error creating socket for GARP: %s\n", strerror(errno));
+               return;
+       }
+       memset(&ifr, 0, sizeof(ifr));
+       strncpy(ifr.ifr_name, "eth0", sizeof(ifr.ifr_name) - 1);
+       if (ioctl(s, SIOCGIFHWADDR, &ifr) < 0)
+       {
+               log(0, 0, 0, 0, "Error getting eth0 hardware address for GARP: %s\n", strerror(errno));
+               close(s);
+               return;
+       }
+       memcpy(mac, &ifr.ifr_hwaddr.sa_data, 6*sizeof(char));
+       if (ioctl(s, SIOCGIFINDEX, &ifr) < 0)
+       {
+               log(0, 0, 0, 0, "Error getting eth0 interface index for GARP: %s\n", strerror(errno));
+               close(s);
+               return;
+       }
+       close(s);
+       sendarp(ifr.ifr_ifindex, mac, ip);
 }
 
 // Find session by username, 0 for not found
@@ -466,20 +619,6 @@ sessionidt sessionidtbysessiont(sessiont *s)
        return val;
 }
 
-// send gratuitous ARP to set ARP table for newly allocated IP
-void sessionsendarp(sessionidt s)
-{
-        unsigned char mac[6];
-#ifdef STAT_CALLS
-       STAT(call_sendarp);
-#endif
-       *(u16 *) (mac + 0) = htons(tapmac[0]); // set source address
-       *(u16 *) (mac + 2) = htons(tapmac[1]);
-       *(u16 *) (mac + 4) = htons(tapmac[2]);
-       sendarp(tapidx, mac, session[s].ip);
-       STAT(arp_sent);
-}
-
 // Handle ARP requests
 void processarp(u8 * buf, int len)
 {
@@ -532,27 +671,27 @@ void processarp(u8 * buf, int len)
                STAT(arp_errors);
                return;
        }
-       ip = ntohl(*(u32 *) (buf + 42));
+       ip = *(u32 *) (buf + 42);
        // look up session
-       s = sessionbyip(htonl(ip));
+       s = sessionbyip(ip);
        if (s)
        {
-               log(3, ip, s, session[s].tunnel, "ARP reply for %u.%u.%u.%u\n", ip >> 24, ip >> 16 & 255, ip >> 8 & 255, ip & 255, session[s].tunnel, s);
+               log(3, ntohl(ip), s, session[s].tunnel, "ARP reply for %s\n", inet_toa(ip));
                memcpy(buf + 4, buf + 10, 6); // set destination as source
-               *(u16 *) (buf + 10) = htons(tapmac[0]); // set soucre address
+               *(u16 *) (buf + 10) = htons(tapmac[0]); // set source address
                *(u16 *) (buf + 12) = htons(tapmac[1]);
                *(u16 *) (buf + 14) = htons(tapmac[2]);
                *(u16 *) (buf + 24) = htons(0x0002); // ARP reply
                memcpy(buf + 26, buf + 10, 6); // sender ethernet
                memcpy(buf + 36, buf + 4, 6); // target ethernet
                *(u32 *) (buf + 42) = *(u32 *) (buf + 32); // target IP
-               *(u32 *) (buf + 32) = htonl(ip); // sender IP
+               *(u32 *) (buf + 32) = ip; // sender IP
                write(tapfd, buf, len);
                STAT(arp_replies);
        }
        else
        {
-               log(3, ip, 0, 0, "ARP request for unknown IP %u.%u.%u.%u\n", ip >> 24, ip >> 16 & 255, ip >> 8 & 255, ip & 255);
+               log(3, ntohl(ip), 0, 0, "ARP request for unknown IP %s\n", inet_toa(ip));
                STAT(arp_discarded);
        }
 }
@@ -565,9 +704,20 @@ void tunnelsend(u8 * buf, u16 l, tunnelidt t)
 #ifdef STAT_CALLS
        STAT(call_tunnelsend);
 #endif
+       if (!t)
+       {
+               static int backtrace_count = 0;
+               log(0, 0, 0, t, "tunnelsend called with 0 as tunnel id\n");
+               STAT(tunnel_tx_errors);
+               log_backtrace(backtrace_count, 5)
+               return;
+       }
+
        if (!tunnel[t].ip)
        {
+               static int backtrace_count = 0;
                log(1, 0, 0, t, "Error sending data out tunnel: no remote endpoint (tunnel not set up)\n");
+               log_backtrace(backtrace_count, 5)
                STAT(tunnel_tx_errors);
                return;
        }
@@ -604,23 +754,43 @@ void tunnelsend(u8 * buf, u16 l, tunnelidt t)
        INC_STAT(tunnel_tx_bytes, l);
 }
 
+//
+// Tiny helper function to write data to
+// the 'tun' device.
+//
+int tun_write(u8 * data, int size)
+{
+       return write(tapfd, data, size);
+}
+
 // process outgoing (to tunnel) IP
+//
 void processipout(u8 * buf, int len)
 {
        sessionidt s;
        sessiont *sp;
        tunnelidt t;
        ipt ip;
-       u8 b[MAXETHER];
+
+       char * data = buf;      // Keep a copy of the originals.
+       int size = len;
+
+       u8 b[MAXETHER + 20];
 #ifdef STAT_CALLS
        STAT(call_processipout);
 #endif
-       if (len < 38)
+       if (len < MIN_IP_SIZE)
        {
                log(1, 0, 0, 0, "Short IP, %d bytes\n", len);
                STAT(tunnel_tx_errors);
                return;
        }
+       if (len >= MAXETHER)
+       {
+               log(1, 0, 0, 0, "Oversize IP packet %d bytes\n", len);
+               STAT(tunnel_tx_errors);
+               return;
+       }
 
        // Skip the tun header
        buf += 4;
@@ -636,33 +806,112 @@ void processipout(u8 * buf, int len)
        ip = *(u32 *)(buf + 16);
        if (!(s = sessionbyip(ip)))
        {
-               log(4, 0, 0, 0, "IP: Sending ICMP host unreachable to %s\n", inet_toa(*(u32 *)(buf + 12)));
-               host_unreachable(*(u32 *)(buf + 12), *(u16 *)(buf + 4), ip, buf, (len < 64) ? 64 : len);
+               // Is this a packet for a session that doesn't exist?
+               static int rate = 0;    // Number of ICMP packets we've sent this second.
+               static int last = 0;    // Last time we reset the ICMP packet counter 'rate'.
+
+               if (last != time_now) {
+                       last = time_now;
+                       rate = 0;
+               }
+
+               if (rate++ < config->icmp_rate) // Only send a max of icmp_rate per second.
+               {
+                       log(4, 0, 0, 0, "IP: Sending ICMP host unreachable to %s\n", inet_toa(*(u32 *)(buf + 12)));
+                       host_unreachable(*(u32 *)(buf + 12), *(u16 *)(buf + 4), ip, buf, (len < 64) ? 64 : len);
+               }
                return;
        }
        t = session[s].tunnel;
        sp = &session[s];
 
-       // Snooping this session, send it to ASIO
-       if (sp->snoop) snoop_send_packet(buf, len);
+       if (sp->tbf_out)
+       {
+               // Are we throttling this session?
+               if (config->cluster_iam_master)
+                       tbf_queue_packet(sp->tbf_out, data, size);
+               else
+                       master_throttle_packet(sp->tbf_out, data, size);
+               return;
+       }
+       else if (sp->walled_garden && !config->cluster_iam_master)
+       {
+               // We are walled-gardening this
+               master_garden_packet(s, data, size);
+               return;
+       }
+
+       // Snooping this session, send it to intercept box
+       if (sp->snoop_ip && sp->snoop_port)
+               snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
 
        log(5, session[s].ip, s, t, "Ethernet -> Tunnel (%d bytes)\n", len);
 
-       // Plugin hook
+       // Add on L2TP header
        {
-               struct param_packet_rx packet = { &tunnel[t], &session[s], buf, len };
-               run_plugins(PLUGIN_PACKET_RX, &packet);
+               u8 *p = makeppp(b, sizeof(b), buf, len, t, s, PPPIP);
+               if (!p) {
+                       log(3, session[s].ip, s, t, "failed to send packet in processipout.\n");
+                       return;
+               }
+               tunnelsend(b, len + (p-b), t); // send it...
+       }
+
+       sp->cout += len; // byte count
+       sp->total_cout += len; // byte count
+       sp->pout++;
+       udp_tx += len;
+       sess_count[s].cout += len;      // To send to master..
+}
+
+//
+// Helper routine for the TBF filters.
+// Used to send queued data in to the user!
+//
+void send_ipout(sessionidt s, u8 *buf, int len)
+{
+       sessiont *sp;
+       tunnelidt t;
+       ipt ip;
+
+       u8 b[MAXETHER + 20];
+
+       if (len < 0 || len > MAXETHER) {
+               log(1,0,0,0, "Odd size IP packet: %d bytes\n", len);
+               return;
        }
 
+       // Skip the tun header
+       buf += 4;
+       len -= 4;
+
+       ip = *(u32 *)(buf + 16);
+
+       if (!session[s].ip)
+               return;
+       t = session[s].tunnel;
+       sp = &session[s];
+
+       log(5, session[s].ip, s, t, "Ethernet -> Tunnel (%d bytes)\n", len);
+
+       // Snooping this session, send it to ASIO
+       if (sp->snoop_ip && sp->snoop_port)
+               snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
+
        // Add on L2TP header
        {
-               u8 *p = makeppp(b, buf, len, t, s, PPPIP);
+               u8 *p = makeppp(b, sizeof(b),  buf, len, t, s, PPPIP);
+               if (!p) {
+                       log(3, session[s].ip, s, t, "failed to send packet in send_ipout.\n");
+                       return;
+               }
                tunnelsend(b, len + (p-b), t); // send it...
-               sp->cout += len; // byte count
-               sp->total_cout += len; // byte count
-               sp->pout++;
-               udp_tx += len;
        }
+       sp->cout += len; // byte count
+       sp->total_cout += len; // byte count
+       sp->pout++;
+       udp_tx += len;
+       sess_count[s].cout += len;      // To send to master..
 }
 
 // add an AVP (16 bit)
@@ -729,11 +978,13 @@ controlt *controlnew(u16 mtype)
 }
 
 // send zero block if nothing is waiting
+// (ZLB send).
 void controlnull(tunnelidt t)
 {
        u8 buf[12];
-       if (tunnel[t].controlc)
+       if (tunnel[t].controlc) // Messages queued; They will carry the ack.
                return;
+
        *(u16 *) (buf + 0) = htons(0xC802); // flags/ver
        *(u16 *) (buf + 2) = htons(12); // length
        *(u16 *) (buf + 4) = htons(tunnel[t].far); // tunnel
@@ -766,6 +1017,61 @@ void controladd(controlt * c, tunnelidt t, sessionidt s)
        }
 }
 
+//
+// Throttle or Unthrottle a session
+//
+// Throttle the data folling through a session
+// to be no more than 'throttle' kbit/sec each way.
+//
+int throttle_session(sessionidt s, int throttle)
+{
+       if (!session[s].tunnel)
+               return 0;       // No-one home.
+
+        if (!*session[s].user)
+                return 0; // User not logged in
+
+       if (throttle) {
+               if (session[s].tbf_in || session[s].tbf_out) {
+                       if (throttle == session[s].throttle)
+                               return 1;
+
+                       // Currently throttled but the rate is changing.
+
+                       free_tbf(session[s].tbf_in);
+                       free_tbf(session[s].tbf_out);
+               }
+
+               session[s].tbf_in = new_tbf(s, throttle*1024/4, throttle*1024/8, send_ipin);
+               session[s].tbf_out = new_tbf(s, throttle*1024/4, throttle*1024/8, send_ipout);
+
+               if (throttle != session[s].throttle) {  // Changed. Flood to slaves.
+                       session[s].throttle = throttle;
+                       cluster_send_session(s);
+               }
+
+               return 1;
+       }
+
+       // else Unthrottling.
+
+       if (!session[s].tbf_in && !session[s].tbf_out && !session[s].throttle)
+               return 0;
+
+       free_tbf(session[s].tbf_in);
+       session[s].tbf_in = 0;
+
+       free_tbf(session[s].tbf_out);
+       session[s].tbf_out = 0;
+
+       if (throttle != session[s].throttle) {  // Changed. Flood to slaves.
+               session[s].throttle = throttle;
+               cluster_send_session(s);
+       }
+
+       return 0;
+}
+
 // start tidy shutdown of session
 void sessionshutdown(sessionidt s, char *reason)
 {
@@ -776,7 +1082,10 @@ void sessionshutdown(sessionidt s, char *reason)
        STAT(call_sessionshutdown);
 #endif
        if (!session[s].tunnel)
+       {
+               log(3, session[s].ip, s, session[s].tunnel, "Called sessionshutdown on a session with no tunnel.\n");
                return;                   // not a live session
+       }
 
        if (!session[s].die)
                log(2, 0, s, session[s].tunnel, "Shutting down session %d: %s\n", s, reason);
@@ -790,7 +1099,7 @@ void sessionshutdown(sessionidt s, char *reason)
 
        // RADIUS Stop message
        if (session[s].opened && !walled_garden && !dead) {
-               u8 r = session[s].radius;
+               u16 r = session[s].radius;
                if (!r)
                {
                        if (!(r = radiusnew(s)))
@@ -811,21 +1120,23 @@ void sessionshutdown(sessionidt s, char *reason)
 
        if (session[s].ip)
        {                          // IP allocated, clear and unroute
-               u8 r;
-               if (session[s].route[0].ip)
+               int r;
+               for (r = 0; r < MAXROUTE && session[s].route[r].ip; r++)
                {
-                       routeset(session[s].ip, 0, 0, 0);
-                       for (r = 0; r < MAXROUTE; r++)
-                       {
-                               if (session[s].route[r].ip)
-                               {
-                                       routeset(session[s].route[r].ip, session[s].route[r].mask, session[s].ip, 0);
-                                       session[s].route[r].ip = 0;
-                               }
-                       }
+                       routeset(s, session[s].route[r].ip, session[s].route[r].mask, session[s].ip, 0);
+                       session[s].route[r].ip = 0;
+               }
+
+               if (session[s].ip_pool_index == -1) // static ip
+               {
+                       routeset(s, session[s].ip, 0, 0, 0);    // Delete route.
+                       session[s].ip = 0;
                }
-               if (session[s].throttle) throttle_session(s, 0); session[s].throttle = 0;
-               free_ip_address(s);
+               else
+                       free_ip_address(s);
+
+               if (session[s].throttle)        // Unthrottle if throttled.
+                       throttle_session(s, 0);
        }
        {                            // Send CDN
                controlt *c = controlnew(14); // sending CDN
@@ -839,7 +1150,7 @@ void sessionshutdown(sessionidt s, char *reason)
 void sendipcp(tunnelidt t, sessionidt s)
 {
        u8 buf[MAXCONTROL];
-       u8 r = session[s].radius;
+       u16 r = session[s].radius;
        u8 *q;
 #ifdef STAT_CALLS
        STAT(call_sendipcp);
@@ -857,14 +1168,21 @@ void sendipcp(tunnelidt t, sessionidt s)
                sessionshutdown(s, "No reply on IPCP");
                return;
        }
-       q = makeppp(buf, 0, 0, t, s, PPPIPCP);
+
+       q = makeppp(buf,sizeof(buf), 0, 0, t, s, PPPIPCP);
+       if (!q) {
+               log(3, session[s].ip, s, t, "failed to send packet in sendipcp.\n");
+               return;
+       }
+
        *q = ConfigReq;
-       q[1] = r;                    // ID, dont care, we only send one type of request
+       q[1] = r << RADIUS_SHIFT;                    // ID, dont care, we only send one type of request
        *(u16 *) (q + 2) = htons(10);
        q[4] = 3;
        q[5] = 6;
-       *(u32 *) (q + 6) = htonl(myip ? myip : session[s].ip); // send my IP (use theirs if I dont have one)
+       *(u32 *) (q + 6) = config->bind_address; // send my IP
        tunnelsend(buf, 10 + (q - buf), t); // send it
+       session[s].flags &= ~SF_IPCP_ACKED;     // Clear flag.
 }
 
 // kill a session now
@@ -876,10 +1194,14 @@ void sessionkill(sessionidt s, char *reason)
        sessionshutdown(s, reason);  // close radius/routes, etc.
        if (session[s].radius)
                radiusclear(session[s].radius, 0); // cant send clean accounting data, session is killed
+       log(2, 0, s, session[s].tunnel, "Kill session %d (%s): %s\n", s, session[s].user, reason);
+
+       throttle_session(s, 0);         // Force session to be un-throttle. Free'ing TBF structures.
+
        memset(&session[s], 0, sizeof(session[s]));
+       session[s].tunnel = T_FREE;     // Mark it as free.
        session[s].next = sessionfree;
        sessionfree = s;
-       log(2, 0, s, session[s].tunnel, "Kill session %d: %s\n", s, reason);
        cluster_send_session(s);
 }
 
@@ -996,12 +1318,6 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                STAT(tunnel_rx_errors);
                return;
        }
-       if (s && !session[s].tunnel)
-       {
-               log(1, ntohl(addr->sin_addr.s_addr), s, t, "UDP packet contains session %d but no session[%d].tunnel exists (LAC said tunnel = %d). Dropping packet.\n", s, s, t);
-               STAT(tunnel_rx_errors);
-               return;
-       }
        if (*buf & 0x08)
        {                          // ns/nr
                ns = ntohs(*(u16 *) p);
@@ -1034,17 +1350,39 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                int requestchap = 0;    // do we request PAP instead of original CHAP request?
                char called[MAXTEL] = ""; // called number
                char calling[MAXTEL] = ""; // calling number
+
+               if (!config->cluster_iam_master) {
+                       master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                       return;
+               }
+
                if ((*buf & 0xCA) != 0xC8)
                {
                        log(1, ntohl(addr->sin_addr.s_addr), s, t, "Bad control header %02X\n", *buf);
                        STAT(tunnel_rx_errors);
                        return;
                }
-               log(3, ntohl(addr->sin_addr.s_addr), s, t, "Control message (%d bytes): %d ns %d nr %d ns %d nr %d\n",
+               log(3, ntohl(addr->sin_addr.s_addr), s, t, "Control message (%d bytes): (unacked %d) l-ns %d l-nr %d r-ns %d r-nr %d\n",
                                l, tunnel[t].controlc, tunnel[t].ns, tunnel[t].nr, ns, nr);
                // if no tunnel specified, assign one
                if (!t)
                {
+                       int i;
+
+                               //
+                               // Is this a duplicate of the first packet? (SCCRQ)
+                               //
+                       for ( i = 1; i <= config->cluster_highest_tunnelid ; ++i) {
+                               if (tunnel[t].state != TUNNELOPENING ||
+                                       tunnel[t].ip != ntohl(*(ipt *) & addr->sin_addr) ||
+                                       tunnel[t].port != ntohs(addr->sin_port) )
+                                       continue;
+                               t = i;
+                               break;
+                       }
+               }
+
+               if (!t) {
                        if (!(t = new_tunnel()))
                        {
                                log(1, ntohl(addr->sin_addr.s_addr), 0, 0, "No more tunnels\n");
@@ -1065,29 +1403,36 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                // check sequence of this message
                {
                        int skip = tunnel[t].window; // track how many in-window packets are still in queue
-                       if (tunnel[t].controlc)
-                       {                  // some to clear maybe
-                               while (tunnel[t].controlc && (((tunnel[t].ns - tunnel[t].controlc) - nr) & 0x8000))
-                               {
-                                       controlt *c = tunnel[t].controls;
-                                       tunnel[t].controls = c->next;
-                                       tunnel[t].controlc--;
-                                       c->next = controlfree;
-                                       controlfree = c;
-                                       skip--;
-                                       tunnel[t].try = 0; // we have progress
-                               }
+                               // some to clear maybe?
+                       while (tunnel[t].controlc && (((tunnel[t].ns - tunnel[t].controlc) - nr) & 0x8000))
+                       {
+                               controlt *c = tunnel[t].controls;
+                               tunnel[t].controls = c->next;
+                               tunnel[t].controlc--;
+                               c->next = controlfree;
+                               controlfree = c;
+                               skip--;
+                               tunnel[t].try = 0; // we have progress
                        }
-                       if (tunnel[t].nr < ns && tunnel[t].nr != 0)
+
+                               // If the 'ns' just received is not the 'nr' we're
+                               // expecting, just send an ack and drop it.
+                               //
+                               // if 'ns' is less, then we got a retransmitted packet.
+                               // if 'ns' is greater than missed a packet. Either way
+                               // we should ignore it.
+                       if (ns != tunnel[t].nr)
                        {
                                // is this the sequence we were expecting?
-                               log(1, ntohl(addr->sin_addr.s_addr), 0, t, "   Out of sequence tunnel %d, (%d not %d)\n", t, ns, tunnel[t].nr);
+                               log(1, ntohl(addr->sin_addr.s_addr), 0, t, "   Out of sequence tunnel %d, (%d is not the expected %d)\n", t, ns, tunnel[t].nr);
                                STAT(tunnel_rx_errors);
-//                             controlnull(t);
+
+                               if (l)  // Is this not a ZLB?
+                                       controlnull(t);
                                return;
                        }
                        // receiver advance (do here so quoted correctly in any sends below)
-                       if (l) tunnel[t].nr++;
+                       if (l) tunnel[t].nr = (ns + 1);
                        if (skip < 0) skip = 0;
                        if (skip < tunnel[t].controlc)
                        {
@@ -1143,7 +1488,7 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                fatal = flags;
                                                continue;
                                        }
-                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Hidden AVP\n");
+                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "Hidden AVP\n");
                                }
                                if (*b & 0x3C)
                                {
@@ -1294,7 +1639,7 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                memcpy(tmp, b, (n >= 30) ? 30 : n);
                                                session[s].tx_connect_speed = atol(tmp);
                                        }
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   TX connect speed <%d>\n",
+                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   TX connect speed <%u>\n",
                                                        session[s].tx_connect_speed);
                                        break;
                                case 38:    // rx connect speed
@@ -1309,7 +1654,7 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                memcpy(tmp, b, (n >= 30) ? 30 : n);
                                                session[s].rx_connect_speed = atol(tmp);
                                        }
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   RX connect speed <%d>\n",
+                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   RX connect speed <%u>\n",
                                                        session[s].rx_connect_speed);
                                        break;
                                case 25:    // Physical Channel ID
@@ -1337,7 +1682,7 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                case 31:    // Proxy Authentication Challenge
                                        {
                                                memcpy(radius[session[s].radius].auth, b, 16);
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Proxy Auth Challenge (%X)\n", radius[session[s].radius].auth);
+                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Proxy Auth Challenge\n");
                                                break;
                                        }
                                case 32:    // Proxy Authentication ID
@@ -1457,18 +1802,21 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        }
                                        else
                                        {
-                                               u8 r;
+                                               u16 r;
                                                controlt *c;
 
                                                s = sessionfree;
                                                sessionfree = session[s].next;
                                                memset(&session[s], 0, sizeof(session[s]));
 
+                                               if (s > config->cluster_highest_sessionid)
+                                                       config->cluster_highest_sessionid = s;
+
                                                // make a RADIUS session
                                                if (!(r = radiusnew(s)))
                                                {
                                                        log(1, ntohl(addr->sin_addr.s_addr), s, t, "No free RADIUS sessions for ICRQ\n");
-                                                       sessionkill(s, "no free RADIUS sesions");
+//                                                     sessionkill(s, "no free RADIUS sesions");
                                                        return;
                                                }
 
@@ -1498,7 +1846,7 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        break;
                                case 12:      // ICCN
                                        session[s].magic = amagic; // set magic number
-                                       session[s].flags = aflags; // set flags received
+                                       session[s].l2tp_flags = aflags; // set flags received
                                        log(3, ntohl(addr->sin_addr.s_addr), s, t, "Magic %X Flags %X\n", amagic, aflags);
                                        controlnull(t); // ack
                                        // In CHAP state, request PAP instead
@@ -1533,11 +1881,6 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                u16 prot;
 
                log_hex(5, "Receive Tunnel Data", p, l);
-               if (session[s].die)
-               {
-                       log(3, ntohl(addr->sin_addr.s_addr), s, t, "Session %d is closing. Don't process PPP packets\n", s);
-                       return;              // closing session, PPP not processed
-               }
                if (l > 2 && p[0] == 0xFF && p[1] == 0x03)
                {                     // HDLC address header, discard
                        p += 2;
@@ -1560,34 +1903,83 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                        p += 2;
                        l -= 2;
                }
+
+               if (s && !session[s].tunnel)    // Is something wrong??
+               {
+                       if (!config->cluster_iam_master)
+                       {
+                               // Pass it off to the master to deal with..
+                               master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                               return;
+                       }
+
+
+                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "UDP packet contains session %d "
+                                       "but no session[%d].tunnel exists (LAC said"
+                                       " tunnel = %d). Dropping packet.\n", s, s, t);
+                       STAT(tunnel_rx_errors);
+                       return;
+               }
+
+               if (session[s].die)
+               {
+                       log(3, ntohl(addr->sin_addr.s_addr), s, t, "Session %d is closing. Don't process PPP packets\n", s);
+// I'm pretty sure this isn't right -- mo.
+//                     return;              // closing session, PPP not processed
+               }
                if (prot == PPPPAP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processpap(t, s, p, l);
                }
                else if (prot == PPPCHAP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processchap(t, s, p, l);
                }
                else if (prot == PPPLCP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processlcp(t, s, p, l);
                }
                else if (prot == PPPIPCP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processipcp(t, s, p, l);
                }
                else if (prot == PPPCCP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processccp(t, s, p, l);
                }
                else if (prot == PPPIP)
                {
-                       session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master)
+                       {
+                               // We're a slave. Should we forward this packet to the master?
+
+                               // Is this a walled garden session, or something that needs it's
+                               // idle time updated??
+
+                               // Maintain the idle timeouts on the master. If this would
+                               // significantly reset the idletimeout, run it via the master
+                               // to refresh the master's idle timer.
+                               // Not sure this is ideal: It may re-order packets.
+
+                               if (session[s].walled_garden || (session[s].last_packet + (ECHO_TIMEOUT/2)) < time_now)
+                               {
+                                       master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                                       session[s].last_packet = time_now;
+                                       return;
+                               }
+                               // fall through to processipin.
+                       } else
+                               session[s].last_packet = time_now;
                        processipin(t, s, p, l);
                }
                else
@@ -1617,47 +2009,247 @@ void processtap(u8 * buf, int len)
        }
        if (*(u16 *) (buf + 2) == htons(PKTARP)) // ARP
                processarp(buf, len);
-       else if (*(u16 *) (buf + 2) == htons(PKTIP)) // ARP
+       else if (*(u16 *) (buf + 2) == htons(PKTIP)) // IP
                processipout(buf, len);
-       else
-       {
-               log(1, 0, 0, 0, "Unexpected tap packet %04X, %d bytes\n", ntohs(*(u16 *) (buf + 2)), len);
-       }
+       // Else discard.
 }
 
-// main loop - gets packets on tap or udp and processes them
-void mainloop(void)
+//
+// Maximum number of actions to complete.
+// This is to avoid sending out too many packets
+// at once.
+#define MAX_ACTIONS 500
+
+int regular_cleanups(void)
 {
-       fd_set cr;
-       int cn;
-       u8 buf[65536];
-       struct timeval to;
-       clockt slow = now();       // occasional functions like session/tunnel expiry, tunnel hello, etc
-       clockt next_acct = slow + ACCT_TIME;
-       clockt next_cluster_ping = slow + 50;
-       to.tv_sec = 1;
-       to.tv_usec = 0;
-       log(4, 0, 0, 0, "Beginning of main loop. udpfd=%d, tapfd=%d, radfd=%d, cluster_sockfd=%d, controlfd=%d\n",
-                       udpfd, tapfd, radfd, cluster_sockfd, controlfd);
+       static sessionidt s = 0;        // Next session to check for actions on.
+       tunnelidt t;
+       int count=0,i;
+       u16 r;
+       static clockt next_acct = 0;
 
-       FD_ZERO(&cr);
-       FD_SET(udpfd, &cr);
-       FD_SET(tapfd, &cr);
-       FD_SET(radfd, &cr);
-       FD_SET(controlfd, &cr);
-       FD_SET(clifd, &cr);
-       if (cluster_sockfd) FD_SET(cluster_sockfd, &cr);
+       log(3, 0, 0, 0, "Begin regular cleanup\n");
+
+       for (r = 1; r < MAXRADIUS; r++)
+       {
+               if (!radius[r].state)
+                       continue;
+               if (radius[r].retry)
+               {
+                       if (radius[r].retry <= config->current_time)
+                               radiusretry(r);
+               } else
+                       radius[r].retry = backoff(radius[r].try+1);     // Is this really needed? --mo
+       }
+       for (t = 1; t < config->cluster_highest_tunnelid; t++)
+       {
+               // check for expired tunnels
+               if (tunnel[t].die && tunnel[t].die <= config->current_time)
+               {
+                       STAT(tunnel_timeout);
+                       tunnelkill(t, "Expired");
+                       continue;
+               }
+               // check for message resend
+               if (tunnel[t].retry && tunnel[t].controlc)
+               {
+                       // resend pending messages as timeout on reply
+                       if (tunnel[t].retry <= config->current_time)
+                       {
+                               controlt *c = tunnel[t].controls;
+                               u8 w = tunnel[t].window;
+                               tunnel[t].try++; // another try
+                               if (tunnel[t].try > 5)
+                                       tunnelkill(t, "Timeout on control message"); // game over
+                               else
+                                       while (c && w--)
+                                       {
+                                               tunnelsend(c->buf, c->length, t);
+                                               c = c->next;
+                                       }
+                       }
+               }
+               // Send hello
+               if (tunnel[t].state == TUNNELOPEN && tunnel[t].lastrec < config->current_time + 600)
+               {
+                       controlt *c = controlnew(6); // sending HELLO
+                       controladd(c, t, 0); // send the message
+                       log(3, tunnel[t].ip, 0, t, "Sending HELLO message\n");
+               }
+       }
+
+       // Check for sessions that have been killed from the CLI
+       if (cli_session_kill[0])
+       {
+               int i;
+               for (i = 0; i < MAXSESSION && cli_session_kill[i]; i++)
+               {
+                       log(2, 0, cli_session_kill[i], 0, "Dropping session by CLI\n");
+                       sessionshutdown(cli_session_kill[i], "Requested by administrator");
+                       cli_session_kill[i] = 0;
+               }
+       }
+       // Check for tunnels that have been killed from the CLI
+       if (cli_tunnel_kill[0])
+       {
+               int i;
+               for (i = 1; i < MAXTUNNEL && cli_tunnel_kill[i]; i++)
+               {
+                       log(2, 0, cli_tunnel_kill[i], 0, "Dropping tunnel by CLI\n");
+                       tunnelshutdown(cli_tunnel_kill[i], "Requested by administrator");
+                       cli_tunnel_kill[i] = 0;
+               }
+       }
+
+       count = 0;
+       for (i = 1; i < config->cluster_highest_sessionid; i++)
+       {
+
+               s++;
+               if (s >= config->cluster_highest_sessionid)
+                       s = 1;
+
+               if (!session[s].tunnel) // Session isn't in use
+                       continue;
+
+               if (!session[s].die && session[s].ip && !(session[s].flags & SF_IPCP_ACKED) )
+               {
+                       // IPCP has not completed yet. Resend
+                       log(3, session[s].ip, s, session[s].tunnel, "No ACK for initial IPCP ConfigReq... resending\n");
+                       sendipcp(session[s].tunnel, s);
+               }
+
+               // check for expired sessions
+               if (session[s].die && session[s].die <= config->current_time)
+               {
+                       sessionkill(s, "Expired");
+                       if (++count >= MAX_ACTIONS) break;
+                       continue;
+               }
+
+               // Drop sessions who have not responded within IDLE_TIMEOUT seconds
+               if (session[s].last_packet && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
+               {
+                       sessionkill(s, "No response to LCP ECHO requests");
+                       STAT(session_timeout);
+                       if (++count >= MAX_ACTIONS) break;
+                       continue;
+               }
+
+               // No data in IDLE_TIMEOUT seconds, send LCP ECHO
+               if (session[s].user[0] && (time_now - session[s].last_packet >= ECHO_TIMEOUT))
+               {
+                       u8 b[MAXCONTROL] = {0};
+
+                       u8 *q = makeppp(b, sizeof(b), 0, 0, session[s].tunnel, s, PPPLCP);
+                       if (!q) {
+                               log(3, session[s].ip, s, t, "failed to send ECHO packet.\n");
+                               continue;
+                       }
+
+                       *q = EchoReq;
+                       *(u8 *)(q + 1) = (time_now % 255); // ID
+                       *(u16 *)(q + 2) = htons(8); // Length
+                       *(u32 *)(q + 4) = 0; // Magic Number (not supported)
+
+                       log(4, session[s].ip, s, session[s].tunnel, "No data in %d seconds, sending LCP ECHO\n",
+                                       (int)(time_now - session[s].last_packet));
+                       tunnelsend(b, 24, session[s].tunnel); // send it
+                       if (++count >= MAX_ACTIONS) break;
+                       continue;
+               }
+       }
+       if (config->accounting_dir && next_acct <= config->current_time)
+       {
+               // Dump accounting data
+               next_acct = config->current_time + ACCT_TIME;
+               dump_acct_info();
+       }
+
+       if (count >= MAX_ACTIONS)
+               return 1;       // Didn't finish!
+
+       log(3, 0, 0, 0, "End regular cleanup (%d actions), next in %d seconds\n", count, config->cleanup_interval);
+       return 0;
+}
+
+
+//
+// Are we in the middle of a tunnel update, or radius
+// requests??
+//
+int still_busy(void)
+{
+       int i;
+       static int last_talked = 0;
+       for (i = config->cluster_highest_tunnelid ; i > 0 ; --i) {
+               if (!tunnel[i].controlc)
+                       continue;
+
+               if (last_talked != config->current_time) {
+                       log(2,0,0,0, "Tunnel %d still has an-acked control messages.\n", i);
+                       last_talked = config->current_time;
+               }
+               return 1;
+       }
+
+       for (i = 1; i < MAXRADIUS; i++)
+       {
+               if (radius[i].state == RADIUSNULL)
+                       continue;
+               if (radius[i].state == RADIUSWAIT)
+                       continue;
+
+               if (last_talked != config->current_time) {
+                       log(2,0,0,0, "Radius session %d is still busy (sid %d)\n", i, radius[i].session);
+                       last_talked = config->current_time;
+               }
+               return 1;
+       }
+
+       return 0;
+}
+
+// main loop - gets packets on tap or udp and processes them
+void mainloop(void)
+{
+       fd_set cr;
+       int cn, i;
+       u8 buf[65536];
+       struct timeval to;
+       time_t next_cluster_ping = 0;   // default 1 second pings.
+       clockt next_clean = time_now + config->cleanup_interval;
+
+       log(4, 0, 0, 0, "Beginning of main loop. udpfd=%d, tapfd=%d, cluster_sockfd=%d, controlfd=%d\n",
+                       udpfd, tapfd, cluster_sockfd, controlfd);
+
+       FD_ZERO(&cr);
+       FD_SET(udpfd, &cr);
+       FD_SET(tapfd, &cr);
+       FD_SET(controlfd, &cr);
+       FD_SET(clifd, &cr);
+       if (cluster_sockfd) FD_SET(cluster_sockfd, &cr);
        cn = udpfd;
-       if (cn < radfd) cn = radfd;
        if (cn < tapfd) cn = tapfd;
        if (cn < controlfd) cn = controlfd;
        if (cn < clifd) cn = clifd;
        if (cn < cluster_sockfd) cn = cluster_sockfd;
+       for (i = 0; i < config->num_radfds; i++)
+       {
+               if (!radfds[i]) continue;
+               FD_SET(radfds[i], &cr);
+               if (radfds[i] > cn)
+                       cn = radfds[i];
+       }
 
-       while (!main_quit)
+       while (!main_quit || still_busy())
        {
                fd_set r;
                int n = cn;
+#ifdef BGP
+               fd_set w;
+               int bgp_set[BGP_NUM_PEERS];
+#endif /* BGP */
 
                if (config->reload_config)
                {
@@ -1666,30 +2258,81 @@ void mainloop(void)
                }
 
                memcpy(&r, &cr, sizeof(fd_set));
-               n = select(n + 1, &r, 0, 0, &to);
-               if (n < 0)
+               to.tv_sec = 0;
+               to.tv_usec = 100000; // 1/10th of a second.
+
+#ifdef BGP
+               FD_ZERO(&w);
+               for (i = 0; i < BGP_NUM_PEERS; i++)
                {
-                       if (errno != EINTR)
+                       bgp_set[i] = bgp_select_state(&bgp_peers[i]);
+                       if (bgp_set[i] & 1)
                        {
-                               perror("select");
-                               exit( -1);
+                               FD_SET(bgp_peers[i].sock, &r);
+                               if (bgp_peers[i].sock > n)
+                                       n = bgp_peers[i].sock;
+                       }
+
+                       if (bgp_set[i] & 2)
+                       {
+                               FD_SET(bgp_peers[i].sock, &w);
+                               if (bgp_peers[i].sock > n)
+                                       n = bgp_peers[i].sock;
                        }
                }
+
+               n = select(n + 1, &r, &w, 0, &to);
+#else /* BGP */
+               n = select(n + 1, &r, 0, 0, &to);
+#endif /* BGP */
+
+               config->current_time = now();
+               if (n < 0)
+               {
+                       if (errno == EINTR)
+                               continue;
+
+                       log(0, 0, 0, 0, "Error returned from select(): %s\n", strerror(errno));
+                       main_quit++;
+                       break;
+               }
                else if (n)
                {
                        struct sockaddr_in addr;
                        int alen = sizeof(addr);
                        if (FD_ISSET(udpfd, &r))
-                               processudp(buf, recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen), &addr);
-                       else if (FD_ISSET(tapfd, &r))
-                               processtap(buf, read(tapfd, buf, sizeof(buf)));
-                       else if (FD_ISSET(radfd, &r))
-                               processrad(buf, recv(radfd, buf, sizeof(buf), 0));
-                       else if (FD_ISSET(cluster_sockfd, &r))
-                               processcluster(buf, recvfrom(cluster_sockfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen));
-                       else if (FD_ISSET(controlfd, &r))
+                       {
+                               int c, n;
+                               for (c = 0; c < config->multi_read_count; c++)
+                               {
+                                       if ((n = recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen)) > 0)
+                                               processudp(buf, n, &addr);
+                                       else
+                                               break;
+                               }
+                       }
+                       if (FD_ISSET(tapfd, &r))
+                       {
+                               int c, n;
+                               for (c = 0; c < config->multi_read_count; c++)
+                               {
+                                       if ((n = read(tapfd, buf, sizeof(buf))) > 0)
+                                               processtap(buf, n);
+                                       else
+                                               break;
+                               }
+                       }
+                       for (i = 0; i < config->num_radfds; i++)
+                               if (FD_ISSET(radfds[i], &r))
+                                       processrad(buf, recv(radfds[i], buf, sizeof(buf), 0), i);
+                       if (FD_ISSET(cluster_sockfd, &r)) {
+                               int size;
+                               size = recvfrom(cluster_sockfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen);
+                               processcluster(buf, size, addr.sin_addr.s_addr);
+                       }
+                       if (FD_ISSET(controlfd, &r))
                                processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr);
-                       else if (FD_ISSET(clifd, &r))
+                       if (FD_ISSET(clifd, &r))
                        {
                                struct sockaddr_in addr;
                                int sockfd;
@@ -1706,144 +2349,67 @@ void mainloop(void)
                                        close(sockfd);
                                }
                        }
-                       else
-                       {
-                               log(1, 0, 0, 0, "Main select() loop returned %d, but no fds have data waiting\n", n);
-                               continue;
-                       }
                }
-               else if (n == 0) {                       // handle timeouts
-                       clockt when = now();
-                       clockt best = when + 100; // default timeout
-                       sessionidt s;
-                       tunnelidt t;
-                       u8 r;
-
-                       for (r = 1; r < MAXRADIUS; r++)
-                               if (radius[r].state && radius[r].retry)
-                               {
-                                       if (radius[r].retry <= when)
-                                               radiusretry(r);
-                                       if (radius[r].retry && radius[r].retry < best)
-                                               best = radius[r].retry;
-                               }
-                       for (t = 1; t < MAXTUNNEL; t++)
-                       {
-                               // check for expired tunnels
-                               if (tunnel[t].die && tunnel[t].die <= when)
-                               {
-                                       STAT(tunnel_timeout);
-                                       tunnelkill(t, "Expired");
-                                       continue;
-                               }
-                               // check for message resend
-                               if (tunnel[t].retry && tunnel[t].controlc)
-                               {
-                                       // resend pending messages as timeout on reply
-                                       if (tunnel[t].retry <= when)
-                                       {
-                                               controlt *c = tunnel[t].controls;
-                                               u8 w = tunnel[t].window;
-                                               tunnel[t].try++; // another try
-                                               if (tunnel[t].try > 5)
-                                                       tunnelkill(t, "Timeout on control message"); // game over
-                                               else
-                                                       while (c && w--)
-                                                       {
-                                                               tunnelsend(c->buf, c->length, t);
-                                                               c = c->next;
-                                                       }
-                                       }
-                                       if (tunnel[t].retry && tunnel[t].retry < best)
-                                               best = tunnel[t].retry;
-                               }
-                               // Send hello
-                               if (tunnel[t].state == TUNNELOPEN && tunnel[t].lastrec < when + 600)
-                               {
-                                       controlt *c = controlnew(6); // sending HELLO
-                                       controladd(c, t, 0); // send the message
-                                       log(3, tunnel[t].ip, 0, t, "Sending HELLO message\n");
-                               }
-                       }
-
-                       // Check for sessions that have been killed from the CLI
-                       if (cli_session_kill[0])
-                       {
-                               int i;
-                               for (i = 1; i < MAXSESSION && cli_session_kill[i]; i++)
-                               {
-                                       log(2, 0, cli_session_kill[i], 0, "Dropping session by CLI\n");
-                                       sessionshutdown(cli_session_kill[i], "Requested by administrator");
-                                       cli_session_kill[i] = 0;
-                               }
-                       }
-                       // Check for tunnels that have been killed from the CLI
-                       if (cli_tunnel_kill[0])
-                       {
-                               int i;
-                               for (i = 1; i < MAXTUNNEL && cli_tunnel_kill[i]; i++)
-                               {
-                                       log(2, 0, cli_tunnel_kill[i], 0, "Dropping tunnel by CLI\n");
-                                       tunnelshutdown(cli_tunnel_kill[i], "Requested by administrator");
-                                       cli_tunnel_kill[i] = 0;
-                               }
-                       }
 
-                       for (s = 1; s < MAXSESSION; s++)
-                       {
-                               // check for expired sessions
-                               if (session[s].die && session[s].die <= when)
-                               {
-                                       sessionkill(s, "Expired");
-                                       continue;
-                               }
-
-                               // Drop sessions who have not responded within IDLE_TIMEOUT seconds
-                               if (session[s].last_packet && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
-                               {
-                                       sessionkill(s, "No response to LCP ECHO requests");
-                                       STAT(session_timeout);
-                                       continue;
-                               }
+                       // Runs on every machine (master and slaves).
+               if (cluster_sockfd && next_cluster_ping <= time_now)
+               {
+                       // Check to see which of the cluster is still alive..
+                       next_cluster_ping = time_now + 1;
+                       cluster_send_ping(basetime);
 
-                               // No data in IDLE_TIMEOUT seconds, send LCP ECHO
-                               if (session[s].user[0] && (time_now - session[s].last_packet >= ECHO_TIMEOUT))
-                               {
-                                       u8 b[MAXCONTROL] = {0};
-                                       u8 *q = makeppp(b, 0, 0, session[s].tunnel, s, PPPLCP);
+                       cluster_check_master();
 
-                                       *q = EchoReq;
-                                       *(u8 *)(q + 1) = (time_now % 255); // ID
-                                       *(u16 *)(q + 2) = htons(8); // Length
-                                       *(u32 *)(q + 4) = 0; // Magic Number (not supported)
+                       cluster_heartbeat(config->cluster_highest_sessionid, sessionfree, config->cluster_highest_tunnelid);    // Only does anything if we're a master.
+                       master_update_counts();         // If we're a slave, send our byte counters to our master.
+               }
 
-                                       log(4, session[s].ip, s, session[s].tunnel, "No data in %d seconds, sending LCP ECHO\n",
-                                                       time_now - session[s].last_packet);
-                                       tunnelsend(b, 24, session[s].tunnel); // send it
-                                       continue;
-                               }
-                       }
-                       if (config->accounting_dir && next_acct <= when)
-                       {
-                               // Dump accounting data
-                               next_acct = when + ACCT_TIME;
-                               dump_acct_info();
+                       // Run token bucket filtering queue..
+                       // Only run it every 1/10th of a second.
+                       // Runs on all machines both master and slave.
+               {
+                       static clockt last_run = 0;
+                       if (last_run != config->current_time) {
+                               last_run = config->current_time;
+                               tbf_run_timer();
                        }
+               }
 
-                       if (cluster_sockfd && next_cluster_ping <= when)
-                       {
-                               // Dump accounting data
-                               next_cluster_ping = when + 50;
-                               cluster_send_message(config->cluster_address, config->bind_address, C_PING, hostname, strlen(hostname));
+               /* Handle timeouts. Make sure that this gets run anyway, even if there was
+                * something to read, else under load this will never actually run....
+                *
+                */
+               if (config->cluster_iam_master && next_clean <= time_now) {
+                       if (regular_cleanups()) { // Did it finish?
+                               next_clean = time_now + 1 ;     // Didn't finish. Check quickly.
+                       } else {
+                               next_clean = time_now + config->cleanup_interval; // Did. Move to next interval.
                        }
+               }
 
-                       if (best <= when)
-                               best = when + 1;  // should not really happen
-                       to.tv_sec = (best - when) / 10;
-                       to.tv_usec = 100000 * ((best - when) % 10);
-                       log(5, 0, 0, 0, "Next time check in %d.%d seconds\n", (best - when) / 10, ((best - when) % 10));
+#ifdef BGP
+               for (i = 0; i < BGP_NUM_PEERS; i++)
+               {
+                       bgp_process(&bgp_peers[i],
+                               bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &r) : 0,
+                               bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &w) : 0);
                }
+#endif /* BGP */
+       }
+
+               // Are we the master and shutting down??
+       if (config->cluster_iam_master) {
+
+               cluster_heartbeat(config->cluster_highest_sessionid, sessionfree,
+                       config->cluster_highest_tunnelid);      // Flush any queued changes..
        }
+
+               // Ok. Notify everyone we're shutting down. If we're
+               // the master, this will force an election.
+       cluster_send_ping(0);
+
+       //
+       // Important!!! We MUST not process any packets past this point!
 }
 
 // Init data structures
@@ -1878,6 +2444,14 @@ void initdata(void)
                log(0, 0, 0, 0, "Error doing mmap for sessions: %s\n", strerror(errno));
                exit(1);
        }
+
+       sess_count = mmap(NULL, sizeof(sessioncountt) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
+       if (sess_count == MAP_FAILED)
+       {
+               log(0, 0, 0, 0, "Error doing mmap for sessions_count: %s\n", strerror(errno));
+               exit(1);
+       }
+
        radius = mmap(NULL, sizeof(radiust) * MAXRADIUS, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
        if (radius == MAP_FAILED)
        {
@@ -1915,48 +2489,52 @@ void initdata(void)
        }
        memset(cli_tunnel_kill, 0, sizeof(tunnelidt) * MAXSESSION);
 
-       filter_buckets = mmap(NULL, sizeof(tbft) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (filter_buckets == MAP_FAILED)
-       {
-               log(0, 0, 0, 0, "Error doing mmap for filter buckets: %s\n", strerror(errno));
-               exit(1);
-       }
-       memset(filter_buckets, 0, sizeof(tbft) * MAXSESSION);
-
        memset(tunnel, 0, sizeof(tunnelt) * MAXTUNNEL);
        memset(session, 0, sizeof(sessiont) * MAXSESSION);
        memset(radius, 0, sizeof(radiust) * MAXRADIUS);
        memset(ip_address_pool, 0, sizeof(ippoolt) * MAXIPPOOL);
-       for (i = 1; i < MAXSESSION - 1; i++)
+
+               // Put all the sessions on the free list marked as undefined.
+       for (i = 1; i < MAXSESSION - 1; i++) {
                session[i].next = i + 1;
+               session[i].tunnel = T_UNDEF;    // mark it as not filled in.
+       }
        session[MAXSESSION - 1].next = 0;
        sessionfree = 1;
+
+               // Mark all the tunnels as undefined (waiting to be filled in by a download).
+       for (i = 1; i < MAXTUNNEL- 1; i++) {
+               tunnel[i].state = TUNNELUNDEF;  // mark it as not filled in.
+       }
+
        if (!*hostname)
        {
-               char *p;
                // Grab my hostname unless it's been specified
                gethostname(hostname, sizeof(hostname));
-               {
-                       struct hostent *h = gethostbyname(hostname);
-                       if (h)
-                               myip = ntohl(*(u32 *) h->h_addr);
-               }
-
-               if ((p = strstr(hostname, ".optusnet.com.au"))) *p = 0;
        }
        _statistics->start_time = _statistics->last_reset = time(NULL);
+
+#ifdef BGP
+       bgp_peers = mmap(NULL, sizeof(struct bgp_peer) * BGP_NUM_PEERS, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
+       if (bgp_peers == MAP_FAILED)
+       {
+               log(0, 0, 0, 0, "Error doing mmap for bgp: %s\n", strerror(errno));
+               exit(1);
+       }
+#endif /* BGP */
 }
 
 void initiptables(void)
 {
        /* Flush the tables here so that we have a clean slate */
-       system("iptables -t nat -F l2tpns");
-       system("iptables -t mangle -F l2tpns");
+
+// Not needed. 'nat' is setup by garden.c
+// mangle isn't used (as throttling is done by tbf inhouse).
 }
 
 int assign_ip_address(sessionidt s)
 {
-       unsigned i;
+       u32 i;
        int best = -1;
        clockt best_time = time_now;
        char *u = session[s].user;
@@ -1965,7 +2543,7 @@ int assign_ip_address(sessionidt s)
 #ifdef STAT_CALLS
        STAT(call_assign_ip_address);
 #endif
-       for (i = 0; i < ip_pool_size; i++)
+       for (i = 1; i < ip_pool_size; i++)
        {
                if (!ip_address_pool[i].address || ip_address_pool[i].assigned)
                        continue;
@@ -1991,10 +2569,11 @@ int assign_ip_address(sessionidt s)
                return 0;
        }
 
-       session[s].ip = ntohl(ip_address_pool[best].address);
+       session[s].ip = ip_address_pool[best].address;
        session[s].ip_pool_index = best;
        ip_address_pool[best].assigned = 1;
        ip_address_pool[best].last = time_now;
+       ip_address_pool[best].session = s;
        if (session[s].walled_garden)
                /* Don't track addresses of users in walled garden (note: this
                   means that their address isn't "sticky" even if they get
@@ -2005,7 +2584,7 @@ int assign_ip_address(sessionidt s)
 
        STAT(ip_allocated);
        log(4, ip_address_pool[best].address, s, session[s].tunnel,
-               "assign_ip_address(): %s ip address %lu from pool\n", reuse ? "Reusing" : "Allocating", best);
+               "assign_ip_address(): %s ip address %d from pool\n", reuse ? "Reusing" : "Allocating", best);
 
        return 1;
 }
@@ -2014,39 +2593,150 @@ void free_ip_address(sessionidt s)
 {
        int i = session[s].ip_pool_index;
 
-#ifdef STAT_CALLS
-       STAT(call_free_ip_address);
-#endif
-
        if (!session[s].ip)
                return; // what the?
 
+       if (i < 0)      // Is this actually part of the ip pool?
+               i = 0;
+
        STAT(ip_freed);
-       uncache_sessionid(session[s].ip);
+       cache_ipmap(session[s].ip, -i); // Change the mapping to point back to the ip pool index.
        session[s].ip = 0;
        ip_address_pool[i].assigned = 0;
+       ip_address_pool[i].session = 0;
        ip_address_pool[i].last = time_now;
+
+#ifdef STAT_CALLS
+       STAT(call_free_ip_address);
+#endif
+}
+
+//
+// Fsck the address pool against the session table.
+// Normally only called when we become a master.
+//
+// This isn't perfect: We aren't keep tracking of which
+// users used to have an IP address.
+//
+void rebuild_address_pool(void)
+{
+       int i;
+
+               //
+               // Zero the IP pool allocation, and build
+               // a map from IP address to pool index.
+       for (i = 1; i < MAXIPPOOL; ++i) {
+               ip_address_pool[i].assigned = 0;
+               ip_address_pool[i].session = 0;
+               if (!ip_address_pool[i].address)
+                       continue;
+
+               cache_ipmap(ip_address_pool[i].address, -i);    // Map pool IP to pool index.
+       }
+
+       for (i = 0; i < MAXSESSION; ++i) {
+               int ipid;
+               if (!session[i].ip || !session[i].tunnel)
+                       continue;
+               ipid = - lookup_ipmap(htonl(session[i].ip));
+
+               if (session[i].ip_pool_index < 0) {     // Not allocated out of the pool.
+                       if (ipid < 1)                   // Not found in the pool either? good.
+                               continue;
+
+                       log(0, 0, i, 0, "Session %d has an IP address (%s) that was marked static, but is in the pool (%d)!",
+                               i, inet_toa(session[i].ip), ipid);
+
+                       // Fall through and process it as part of the pool.
+               }
+
+
+               if (ipid > MAXIPPOOL || ipid < 0) {
+                       log(0, 0, i, 0, "Session %d has a pool IP that's not found in the pool! (%d)\n", i, ipid);
+                       ipid = -1;
+                       session[i].ip_pool_index = ipid;
+                       continue;
+               }
+
+               ip_address_pool[ipid].assigned = 1;
+               ip_address_pool[ipid].session = i;
+               ip_address_pool[ipid].last = time_now;
+               strncpy(ip_address_pool[ipid].user, session[i].user, sizeof(ip_address_pool[ipid].user) - 1);
+               session[i].ip_pool_index = ipid;
+               cache_ipmap(session[i].ip, i);  // Fix the ip map.
+       }
+}
+
+//
+// Fix the address pool to match a changed session.
+// (usually when the master sends us an update).
+void fix_address_pool(int sid)
+{
+       int ipid;
+
+       ipid = session[sid].ip_pool_index;
+
+       if (ipid > ip_pool_size)
+               return;         // Ignore it. rebuild_address_pool will fix it up.
+
+       if (ip_address_pool[ipid].address != session[sid].ip)
+               return;         // Just ignore it. rebuild_address_pool will take care of it.
+
+       ip_address_pool[ipid].assigned = 1;
+       ip_address_pool[ipid].session = sid;
+       ip_address_pool[ipid].last = time_now;
+       strncpy(ip_address_pool[ipid].user, session[sid].user, sizeof(ip_address_pool[ipid].user) - 1);
+}
+
+//
+// Add a block of addresses to the IP pool to hand out.
+//
+void add_to_ip_pool(u32 addr, u32 mask)
+{
+       int i;
+       if (mask == 0)
+               mask = 0xffffffff;      // Host route only.
+
+       addr &= mask;
+
+       if (ip_pool_size >= MAXIPPOOL)  // Pool is full!
+               return ;
+
+       for (i = addr ;(i & mask) == addr; ++i)
+       {
+               if ((i & 0xff) == 0 || (i&0xff) == 255)
+                       continue;       // Skip 0 and broadcast addresses.
+
+               ip_address_pool[ip_pool_size].address = i;
+               ip_address_pool[ip_pool_size].assigned = 0;
+               ++ip_pool_size;
+               if (ip_pool_size >= MAXIPPOOL)
+               {
+                       log(0,0,0,0, "Overflowed IP pool adding %s\n", inet_toa(htonl(addr)) );
+                       return;
+               }
+       }
 }
 
 // Initialize the IP address pool
 void initippool()
 {
        FILE *f;
-       char *buf, *p;
-       int pi = 0;
+       char *p;
+       char buf[4096];
        memset(ip_address_pool, 0, sizeof(ip_address_pool));
 
        if (!(f = fopen(IPPOOLFILE, "r")))
        {
                log(0, 0, 0, 0, "Can't load pool file " IPPOOLFILE ": %s\n", strerror(errno));
-               exit(-1);
+               exit(1);
        }
 
-       buf = (char *)malloc(4096);
-
-       while (pi < MAXIPPOOL && fgets(buf, 4096, f))
+       while (ip_pool_size < MAXIPPOOL && fgets(buf, 4096, f))
        {
-               char* pool = buf;
+               char *pool = buf;
+               buf[4095] = 0;  // Force it to be zero terminated/
+
                if (*buf == '#' || *buf == '\n')
                        continue; // Skip comments / blank lines
                if ((p = (char *)strrchr(buf, '\n'))) *p = 0;
@@ -2058,7 +2748,7 @@ void initippool()
                        if (src == INADDR_NONE)
                        {
                                log(0, 0, 0, 0, "Invalid address pool IP %s", buf);
-                               exit(-1);
+                               exit(1);
                        }
                        // This entry is for a specific IP only
                        if (src != config->bind_address)
@@ -2070,60 +2760,44 @@ void initippool()
                {
                        // It's a range
                        int numbits = 0;
-                       unsigned long start = 0, end = 0, mask = 0, ip;
-                       struct rtentry r;
+                       u32 start = 0, mask = 0;
 
                        log(2, 0, 0, 0, "Adding IP address range %s\n", buf);
                        *p++ = 0;
                        if (!*p || !(numbits = atoi(p)))
                        {
-                               log(0, 0, 0, 0, "Invalid pool range %s/\n", buf, p);
+                               log(0, 0, 0, 0, "Invalid pool range %s\n", buf);
                                continue;
                        }
-                       start = end = ntohl(inet_addr(pool));
-                       mask = (unsigned long)(pow(2, numbits) - 1) << (32 - numbits);
-                       start &= mask;
-                       end = start + (int)(pow(2, (32 - numbits))) - 1;
-                       for (ip = (start + 1); ip < end && pi < MAXIPPOOL; ip++)
-                       {
-                               if ((ip & 0xFF) == 0 || (ip & 0xFF) == 255)
-                                       continue;
-                               ip_address_pool[pi++].address = htonl(ip);
-                       }
+                       start = ntohl(inet_addr(pool));
+                       mask = (u32)(pow(2, numbits) - 1) << (32 - numbits);
 
                        // Add a static route for this pool
-                       log(5, 0, 0, 0, "Adding route for address pool %s/%d\n", inet_toa(htonl(start)), 32+mask);
-                       memset(&r, 0, sizeof(r));
-                       r.rt_dev = config->tapdevice;
-                       r.rt_dst.sa_family = AF_INET;
-                       *(u32 *) & (((struct sockaddr_in *) &r.rt_dst)->sin_addr.s_addr) = htonl(start);
-                       r.rt_genmask.sa_family = AF_INET;
-                       *(u32 *) & (((struct sockaddr_in *) &r.rt_genmask)->sin_addr.s_addr) = htonl(mask);
-                       r.rt_flags = (RTF_UP | RTF_STATIC);
-                       if (ioctl(ifrfd, SIOCADDRT, (void *) &r) < 0)
-                       {
-                               log(0, 0, 0, 0, "Error adding ip address pool route %s/%d: %s\n", inet_toa(start), mask, strerror(errno));
-                       }
+                       log(5, 0, 0, 0, "Adding route for address pool %s/%u\n", inet_toa(htonl(start)), 32 + mask);
+                       routeset(0, start, mask, 0, 1);
+
+                       add_to_ip_pool(start, mask);
                }
                else
                {
                        // It's a single ip address
-                       ip_address_pool[pi++].address = inet_addr(pool);
+                       add_to_ip_pool(inet_addr(pool), 0);
                }
        }
-
-       free(buf);
        fclose(f);
-       log(1, 0, 0, 0, "IP address pool is %d addresses\n", pi);
-       ip_pool_size = pi;
+       log(1, 0, 0, 0, "IP address pool is %d addresses\n", ip_pool_size - 1);
 }
 
-void snoop_send_packet(char *packet, u16 size)
+void snoop_send_packet(char *packet, u16 size, ipt destination, u16 port)
 {
-       if (!snoop_addr.sin_port || snoopfd <= 0 || size <= 0 || !packet)
+       if (!destination || !port || snoopfd <= 0 || size <= 0 || !packet)
                return;
 
-       log(5, 0, 0, 0, "Snooping packet at %p (%d bytes) to %s:%d\n", packet, size, inet_toa(snoop_addr.sin_addr.s_addr), htons(snoop_addr.sin_port));
+       snoop_addr.sin_addr.s_addr = destination;
+       snoop_addr.sin_port = ntohs(port);
+
+       log(5, 0, 0, 0, "Snooping packet at %p (%d bytes) to %s:%d\n",
+                       packet, size, inet_toa(snoop_addr.sin_addr.s_addr), htons(snoop_addr.sin_port));
        if (sendto(snoopfd, packet, size, MSG_DONTWAIT | MSG_NOSIGNAL, (void *) &snoop_addr, sizeof(snoop_addr)) < 0)
                log(0, 0, 0, 0, "Error sending intercept packet: %s\n", strerror(errno));
        STAT(packets_snooped);
@@ -2145,7 +2819,7 @@ void dump_acct_info()
 
     for (i = 0; i < MAXSESSION; i++)
     {
-       if (!session[i].opened || !session[i].cin || !session[i].cout || !*session[i].user || session[i].walled_garden)
+       if (!session[i].opened || !session[i].ip || !session[i].cin || !session[i].cout || !*session[i].user || session[i].walled_garden)
                continue;
        if (!f)
        {
@@ -2167,12 +2841,12 @@ void dump_acct_info()
        }
 
        log(4, 0, 0, 0, "Dumping accounting information for %s\n", session[i].user);
-       fprintf(f, "%s %s %d %lu %lu\n",
+       fprintf(f, "%s %s %d %u %u\n",
            session[i].user,                            // username
            inet_toa(htonl(session[i].ip)),             // ip
            (session[i].throttle) ? 2 : 1,              // qos
-           (unsigned long)session[i].cin,              // uptxoctets
-           (unsigned long)session[i].cout);            // downrxoctets
+           (u32)session[i].cin,                // uptxoctets
+           (u32)session[i].cout);              // downrxoctets
 
        session[i].pin = session[i].cin = 0;
        session[i].pout = session[i].cout = 0;
@@ -2188,44 +2862,33 @@ int main(int argc, char *argv[])
 
        _program_name = strdup(argv[0]);
 
-       {
-               struct rlimit rlim;
-               rlim.rlim_cur = RLIM_INFINITY;
-               rlim.rlim_max = RLIM_INFINITY;
-               // Remove the maximum core size
-               setrlimit(RLIMIT_CORE, &rlim);
-               // Make core dumps go to /tmp
-               chdir("/tmp");
-       }
-
        time(&basetime);             // start clock
-       // scan args
 
+       // scan args
        while ((o = getopt(argc, argv, "vc:h:a:")) >= 0)
        {
                switch (o)
                {
+                       case 'd':
+                               // Double fork to detach from terminal
+                               if (fork()) exit(0);
+                               if (fork()) exit(0);
+                               break;
                        case 'v':
                                config->debug++;
                                break;
-                       case 'c':
-                               strncpy(config->config_file, optarg, sizeof(config->config_file) - 1);
-                               break;
                        case 'h':
                                strncpy(hostname, optarg, 999);
                                break;
-                       case 'a':
-                               myip = inet_addr(optarg);
-                               if (myip == INADDR_NONE) {
-                                       log(0, 0, 0, 0, "Invalid ip %s\n", optarg);
-                                       exit(-1);
-                               }
-                               config->bind_address = myip;
-                               handle_interface = 1;
-                               break;
                        case '?':
                        default:
-                               printf("Args are:\n\t-c <file>\tConfig file\n\t-h <hostname>\tForce hostname\n\t-a <address>\tUse specific address\n\t-v\t\tDebug\n");
+                               printf("Args are:\n"
+                                      "\t-d\tDetach from terminal\n"
+                                      "\t-c <file>\tConfig file\n"
+                                      "\t-h <hostname>\tForce hostname\n"
+                                      "\t-a <address>\tUse specific address\n"
+                                      "\t-v\t\tDebug\n");
+
                                return (0);
                                break;
                }
@@ -2234,20 +2897,71 @@ int main(int argc, char *argv[])
        // Start the timer routine off
        time(&time_now);
        strftime(time_now_string, 64, "%Y-%m-%d %H:%M:%S", localtime(&time_now));
+       signal(SIGALRM, sigalrm_handler);
+       siginterrupt(SIGALRM, 0);
 
        initiptables();
        initplugins();
        initdata();
+       init_tbf();
        init_cli();
        read_config_file();
-       log(0, 0, 0, 0, "$Id: l2tpns.c,v 1.3 2004-03-05 00:22:06 fred_nerk Exp $\n(c) Copyright 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced\n");
 
-       /* Start up the cluster first, so that we don't have two machines with
-        * the same IP at once.
-        * This is still racy, but the second GARP should fix that
-        */
-       cluster_init(config->bind_address, 0);
-       cluster_send_message(config->cluster_address, config->bind_address, C_HELLO, hostname, strlen(hostname));
+       log(0, 0, 0, 0, "$Id: l2tpns.c,v 1.8 2004-06-23 03:52:24 fred_nerk Exp $\n"
+                       "(c) Copyright 2003, 2004 Optus Internet Engineering\n"
+                       "(c) Copyright 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced\n");
+       {
+               struct rlimit rlim;
+               rlim.rlim_cur = RLIM_INFINITY;
+               rlim.rlim_max = RLIM_INFINITY;
+               // Remove the maximum core size
+               if (setrlimit(RLIMIT_CORE, &rlim) < 0)
+                       log(0, 0, 0, 0, "Can't set ulimit: %s\n", strerror(errno));
+               // Make core dumps go to /tmp
+               chdir("/tmp");
+       }
+
+       if (config->scheduler_fifo)
+       {
+               int ret;
+               struct sched_param params = {0};
+               params.sched_priority = 1;
+
+               if (get_nprocs() < 2)
+               {
+                       log(0, 0, 0, 0, "Not using FIFO scheduler, there is only 1 processor in the system.\n");
+                       config->scheduler_fifo = 0;
+               }
+               else
+               {
+                       if ((ret = sched_setscheduler(0, SCHED_FIFO, &params)) == 0)
+                       {
+                               log(1, 0, 0, 0, "Using FIFO scheduler. Say goodbye to any other processes running\n");
+                       }
+                       else
+                       {
+                               log(0, 0, 0, 0, "Error setting scheduler to FIFO: %s\n", strerror(errno));
+                               config->scheduler_fifo = 0;
+                       }
+               }
+       }
+
+       /* Set up the cluster communications port. */
+       if (cluster_init(config->bind_address) < 0)
+               exit(1);
+
+#ifdef BGP
+       signal(SIGPIPE, SIG_IGN);
+       bgp_setup(config->as_number);
+       bgp_add_route(config->bind_address, 0xffffffff);
+        if (*config->bgp_peer[0])
+               bgp_start(&bgp_peers[0], config->bgp_peer[0],
+                   config->bgp_peer_as[0], 0); /* 0 = routing disabled */
+
+        if (*config->bgp_peer[1])
+               bgp_start(&bgp_peers[1], config->bgp_peer[1],
+                   config->bgp_peer_as[1], 0);
+#endif /* BGP */
 
        inittap();
        log(1, 0, 0, 0, "Set up on interface %s\n", config->tapdevice);
@@ -2255,24 +2969,14 @@ int main(int argc, char *argv[])
        initudp();
        initrad();
        initippool();
-       init_rl();
-       if (handle_interface) {
-               send_garp(config->bind_address);
-       }
-
-       // If NOSTATEFILE exists, we will ignore any updates from the cluster master for this execution
-       if (!unlink(NOSTATEFILE))
-           config->ignore_cluster_updates = 1;
 
        read_state();
 
-       signal(SIGALRM, sigalrm_handler);
        signal(SIGHUP, sighup_handler);
        signal(SIGTERM, sigterm_handler);
        signal(SIGINT, sigterm_handler);
        signal(SIGQUIT, sigquit_handler);
        signal(SIGCHLD, sigchild_handler);
-       signal(SIGSEGV, sigsegv_handler);
 
        alarm(1);
 
@@ -2281,6 +2985,24 @@ int main(int argc, char *argv[])
                setuid(config->target_uid);
 
        mainloop();
+
+#ifdef BGP
+       /* try to shut BGP down cleanly; with luck the sockets will be
+          writable since we're out of the select */
+       {
+               int i;
+               for (i = 0; i < BGP_NUM_PEERS; i++)
+                       if (bgp_peers[i].state == Established)
+                               bgp_stop(&bgp_peers[i]);
+       }
+#endif /* BGP */
+
+       /* remove plugins (so cleanup code gets run) */
+       plugins_done();
+
+       /* kill CLI children */
+       signal(SIGTERM, SIG_IGN);
+       kill(0, SIGTERM);
        return 0;
 }
 
@@ -2300,7 +3022,7 @@ void sigalrm_handler(int junk)
        // Log current traffic stats
 
        snprintf(config->bandwidth, sizeof(config->bandwidth),
-               "UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%lu OUT:%lu",
+               "UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%u OUT:%u",
                (udp_rx / 1024.0 / 1024.0 * 8),
                (eth_tx / 1024.0 / 1024.0 * 8),
                (eth_rx / 1024.0 / 1024.0 * 8),
@@ -2333,12 +3055,12 @@ void sigterm_handler(int junk)
        log(1, 0, 0, 0, "Shutting down cleanly\n");
        if (config->save_state)
                dump_state();
+
        main_quit++;
 }
 
 void sigquit_handler(int junk)
 {
-       FILE *f;
        int i;
 
        log(1, 0, 0, 0, "Shutting down without saving sessions\n");
@@ -2353,12 +3075,6 @@ void sigquit_handler(int junk)
                        tunnelshutdown(i, "L2TPNS Closing");
        }
 
-       cluster_send_goodbye();
-
-       // Touch a file which says not to reload the state
-       f = fopen(NOSTATEFILE, "w");
-       if (f) fclose(f);
-
        main_quit++;
 }
 
@@ -2368,14 +3084,6 @@ void sigchild_handler(int signal)
            ;
 }
 
-void sigsegv_handler(int signal)
-{
-       log(0, 0, 0, 0, "----------------------------------------------\n");
-       log(0, 0, 0, 0, "- SEGFAULT!                                  -\n");
-       log(0, 0, 0, 0, "----------------------------------------------\n");
-       _exit(0);
-}
-
 void read_state()
 {
     struct stat sb;
@@ -2386,17 +3094,16 @@ void read_state()
     u32 buf[2];
 
     if (!config->save_state)
-       return;
-
-    // Ignore saved state if NOSTATEFILE exists
-    if (config->ignore_cluster_updates)
     {
        unlink(STATEFILE);
        return;
     }
 
     if (stat(STATEFILE, &sb) < 0)
+    {
+       unlink(STATEFILE);
        return;
+    }
 
     if (sb.st_mtime < (time(NULL) - 60))
     {
@@ -2488,13 +3195,13 @@ void read_state()
 
     for (i = 0; i < MAXSESSION; i++)
     {
-       session[i].tbf = 0;
-       session[i].throttle = 0;
+       session[i].tbf_in = 0;
+       session[i].tbf_out = 0;
        if (session[i].opened)
        {
            log(2, 0, i, 0, "Loaded active session for user %s\n", session[i].user);
-           if (session[i].ip && session[i].ip != 0xFFFFFFFE)
-               sessionsetup(session[i].tunnel, i, 0);
+           if (session[i].ip)
+               sessionsetup(session[i].tunnel, i);
        }
     }
 
@@ -2582,10 +3289,6 @@ void update_config()
 {
        int i;
 
-       snoop_addr.sin_family = AF_INET;
-       snoop_addr.sin_addr.s_addr = config->snoop_destination_host;
-       snoop_addr.sin_port = htons(config->snoop_destination_port);
-
        // Update logging
        closelog();
        syslog_log = 0;
@@ -2596,9 +3299,18 @@ void update_config()
        }
        if (*config->log_filename)
        {
-               if (strstr(config->log_filename, "file:") == config->log_filename)
+               if (strstr(config->log_filename, "syslog:") == config->log_filename)
                {
-                       if ((log_stream = fopen((char *)(config->log_filename + 5), "a")))
+                       char *p = config->log_filename + 7;
+                       if (*p)
+                       {
+                               openlog("l2tpns", LOG_PID, facility_value(p));
+                               syslog_log = 1;
+                       }
+               }
+               else if (strchr(config->log_filename, '/') == config->log_filename)
+               {
+                       if ((log_stream = fopen((char *)(config->log_filename), "a")))
                        {
                                fseek(log_stream, 0, SEEK_END);
                                setbuf(log_stream, NULL);
@@ -2609,15 +3321,6 @@ void update_config()
                                setbuf(log_stream, NULL);
                        }
                }
-               else if (strstr(config->log_filename, "syslog:") == config->log_filename)
-               {
-                       char *p = config->log_filename + 7;
-                       if (*p)
-                       {
-                               openlog("l2tpns", LOG_PID, facility_value(p));
-                               syslog_log = 1;
-                       }
-               }
        }
        else
        {
@@ -2636,6 +3339,8 @@ void update_config()
                log(0, 0, 0, 0, "No RADIUS servers defined!\n");
        }
 
+       config->num_radfds = 2 << RADIUS_SHIFT;
+
        // Update plugins
        for (i = 0; i < MAXPLUGINS; i++)
        {
@@ -2653,6 +3358,12 @@ void update_config()
                }
        }
        memcpy(config->old_plugins, config->plugins, sizeof(config->plugins));
+       if (!config->cleanup_interval) config->cleanup_interval = 10;
+       if (!config->multi_read_count) config->multi_read_count = 1;
+       if (!config->cluster_address) config->cluster_address = inet_addr(DEFAULT_MCAST_ADDR);
+       if (!*config->cluster_interface)
+               strncpy(config->cluster_interface, DEFAULT_MCAST_INTERFACE, sizeof(config->cluster_interface) - 1);
+
        config->reload_config = 0;
 }
 
@@ -2671,84 +3382,167 @@ void read_config_file()
        log(3, 0, 0, 0, "Done reading config file\n");
        fclose(f);
        update_config();
-       log_stream = NULL;
 }
 
-int sessionsetup(tunnelidt t, sessionidt s, u8 routes)
+int sessionsetup(tunnelidt t, sessionidt s)
 {
        // A session now exists, set it up
        ipt ip;
        char *user;
        sessionidt i;
+       int r;
 #ifdef STAT_CALLS
        STAT(call_sessionsetup);
 #endif
+
        log(3, session[s].ip, s, t, "Doing session setup for session\n");
-       if (!session[s].ip) {
-               log(0, session[s].ip, s, t, "VERY VERY BAD! sessionsetup() called with no session[s].ip\n");
-               return 1;
+
+       if (!session[s].ip || session[s].ip == 0xFFFFFFFE)
+       {
+               assign_ip_address(s);
+               if (session[s].ip)
+                       log(3, 0, s, t, "   No IP allocated. Assigned %s from pool\n",
+                                       inet_toa(htonl(session[s].ip)));
+               else
+                       log(0, 0, s, t, "   No IP allocated. The IP address pool is FULL!\n");
        }
 
+
        // Make sure this is right
        session[s].tunnel = t;
+
        // zap old sessions with same IP and/or username
-       // Don't kill walled garden sessions - doing so leads to a DoS
+       // Don't kill gardened sessions - doing so leads to a DoS
        // from someone who doesn't need to know the password
-       ip = session[s].ip;
-       user = session[s].user;
-       for (i = 1; i < MAXSESSION; i++)
        {
-               if (i == s) continue;
-               if (ip == session[i].ip) sessionkill(i, "Duplicate IP address");
-               if (!session[s].walled_garden && !session[i].walled_garden && strcasecmp(user, session[i].user) == 0)
-                       sessionkill(i, "Duplicate session for user");
-       }
-
-       if (routes)
-       {
-               if (session[s].route[routes].ip && session[s].route[routes].mask)
+               ip = session[s].ip;
+               user = session[s].user;
+               for (i = 1; i <= config->cluster_highest_sessionid; i++)
                {
-                       log(2, session[s].ip, s, t, "Routing session\n");
-                       routeset(session[s].ip, 0, 0, 1);
-                       while (routes--)
-                               routeset(session[s].route[routes].ip, session[s].route[routes].mask,
-                                               session[s].ip, 1);
+                       if (i == s) continue;
+                       if (ip == session[i].ip) sessionkill(i, "Duplicate IP address");
+                       if (!session[s].walled_garden && !session[i].walled_garden && strcasecmp(user, session[i].user) == 0)
+                               sessionkill(i, "Duplicate session for users");
                }
        }
-       sessionsendarp(s);
-       if (!session[s].sid)
-               sendipcp(t, s);
 
-       // Force throttling on or off
-       // This has the advantage of cleaning up after another throttled user who may have left
-       // firewall rules lying around
-       throttle_session(s, session[s].throttle);
+               // Add the route for this session.
+               //
+               // Static IPs need to be routed. Anything else
+               // is part of the IP address pool and is already routed,
+               // it just needs to be added to the IP cache.
+       if (session[s].ip_pool_index == -1) // static ip
+               routeset(s, session[s].ip, 0, 0, 1);
+       else
+               cache_ipmap(session[s].ip, s);
+
+       for (r = 0; r < MAXROUTE && session[s].route[r].ip; r++)
+               routeset(s, session[s].route[r].ip, session[s].route[r].mask, session[s].ip, 1);
+
+       if (!session[s].sid) {  // did this session just finish radius?
+               log(3, session[s].ip, s, t, "Sending initial IPCP to client\n");
+               sendipcp(t, s);
+               session[s].sid = ++last_sid;
+       }
 
+       // Run the plugin's against this new session.
        {
                struct param_new_session data = { &tunnel[t], &session[s] };
                run_plugins(PLUGIN_NEW_SESSION, &data);
        }
 
-       if (!session[s].sid)
-               session[s].sid = ++last_sid;
-
-       cache_sessionid(htonl(session[s].ip), s);
+       // Force throttling on or off (Actually : refresh the current throttling status)
+       // This has the advantage of cleaning up after another throttled user who may have left
+       // firewall rules lying around
+       throttle_session(s, session[s].throttle);
 
-       cluster_send_session(s);
        session[s].last_packet = time_now;
+
        {
                char *sessionip, *tunnelip;
-               sessionip = strdup(inet_toa(ntohl(session[s].ip)));
-               tunnelip = strdup(inet_toa(ntohl(tunnel[t].ip)));
+               sessionip = strdup(inet_toa(htonl(session[s].ip)));
+               tunnelip = strdup(inet_toa(htonl(tunnel[t].ip)));
                log(2, session[s].ip, s, t, "Login by %s at %s from %s (%s)\n",
                                session[s].user, sessionip, tunnelip, tunnel[t].hostname);
                if (sessionip) free(sessionip);
                if (tunnelip) free(tunnelip);
        }
 
+       cluster_send_session(s);        // Mark it as dirty, and needing to the flooded to the cluster.
+
        return 1;       // RADIUS OK and IP allocated, done...
 }
 
+//
+// This session just got dropped on us by the master or something.
+// Make sure our tables up up to date...
+//
+int load_session(sessionidt s, sessiont *new)
+{
+       int i;
+
+               // Sanity checks.
+       if (new->ip_pool_index >= MAXIPPOOL ||
+               new->tunnel >= MAXTUNNEL) {
+               log(0,0,s,0, "Strange session update received!\n");
+                       // FIXME! What to do here?
+               return 0;
+       }
+
+               //
+               // Ok. All sanity checks passed. Now we're committed to
+               // loading the new session.
+               //
+
+       session[s].tunnel = new->tunnel; // For logging in cache_ipmap
+
+
+       if (new->ip != session[s].ip)   // Changed ip. fix up hash tables.
+       {
+               if (session[s].ip)      // If there's an old one, remove it.
+               {
+                       if (session[s].ip_pool_index == -1) // static IP
+                               routeset(s, session[s].ip, 0, 0, 0);
+                       else            // It's part of the IP pool, add it manually.
+                               uncache_ipmap(session[s].ip);
+               }
+
+               if (new->ip) {          // If there's a new one, add it.
+                       if (new->ip_pool_index == -1)
+                               routeset(s, new->ip, 0, 0, 1);
+                       else
+                               cache_ipmap(new->ip, s);
+               }
+       }
+
+               // Add routes for the session if they're new.
+       for (i = 0; i < MAXROUTE && (session[s].route[i].ip || new->route[i].ip); i++)
+       {
+               if (new->route[i].ip == session[s].route[i].ip &&
+                   new->route[i].mask == session[s].route[i].mask)
+                       continue;
+
+               if (session[s].route[i].ip) // Remove the old one if it exists.
+                       routeset(s, session[s].route[i].ip, session[s].route[i].mask, session[s].route[i].ip, 0);
+
+               if (new->route[i].ip)   // Add the new one if it exists.
+                       routeset(s, new->route[i].ip, new->route[i].mask, new->ip, 1);
+       }
+
+
+
+       if (new->tunnel && s > config->cluster_highest_sessionid)       // Maintain this in the slave. It's used
+                                       // for walking the sessions to forward byte counts to the master.
+               config->cluster_highest_sessionid = s;
+
+       memcpy(&session[s], new, sizeof(session[s]));   // Copy over..
+
+               // Do fixups into address pool.
+       if (new->ip_pool_index != -1)
+               fix_address_pool(s);
+       return 1;
+}
+
 #ifdef RINGBUFFER
 void ringbuffer_dump(FILE *stream)
 {
@@ -2774,28 +3568,33 @@ void initplugins()
                plugins[i] = ll_init();
 }
 
+static void *open_plugin(char *plugin_name, int load)
+{
+       char path[256] = "";
+
+       snprintf(path, 256, PLUGINDIR "/%s.so", plugin_name);
+       log(2, 0, 0, 0, "%soading plugin from %s\n", load ? "L" : "Un-l", path);
+       return dlopen(path, RTLD_NOW);
+}
+
 void add_plugin(char *plugin_name)
 {
-       void *p;
+       static struct pluginfuncs funcs = {
+               _log,
+               _log_hex,
+               inet_toa,
+               sessionbyuser,
+               sessiontbysessionidt,
+               sessionidtbysessiont,
+               sessionkill,
+               radiusnew,
+               radiussend,
+       };
+
+       void *p = open_plugin(plugin_name, 1);
        int (*initfunc)(struct pluginfuncs *);
-       char path[256] = {0};
        int i;
-       struct pluginfuncs funcs;
-
-       funcs._log = _log;
-       funcs._log_hex = _log_hex;
-       funcs.inet_toa = inet_toa;
-       funcs.get_session_by_username = sessionbyuser;
-       funcs.get_session_by_id = sessiontbysessionidt;
-       funcs.get_id_by_session = sessionidtbysessiont;
-       funcs.sessionkill = sessionkill;
-       funcs.radiusnew = radiusnew;
-       funcs.radiussend = radiussend;
-
-       snprintf(path, 256, "%s/%s.so", LIBDIR, plugin_name);
-
-       log(2, 0, 0, 0, "Loading plugin from %s\n", path);
-       p = dlopen(path, RTLD_NOW);
+
        if (!p)
        {
                log(1, 0, 0, 0, "   Plugin load failed: %s\n", dlerror());
@@ -2812,70 +3611,66 @@ void add_plugin(char *plugin_name)
                int *v = dlsym(p, "__plugin_api_version");
                if (!v || *v != PLUGIN_API_VERSION)
                {
-                       log(1, 0, 0, 0, "   Plugin load failed: API version mismatch\n", dlerror());
+                       log(1, 0, 0, 0, "   Plugin load failed: API version mismatch: %s\n", dlerror());
                        dlclose(p);
                        return;
                }
        }
 
-       initfunc = dlsym(p, "plugin_init");
-       if (!initfunc)
+       if ((initfunc = dlsym(p, "plugin_init")))
        {
-               log(1, 0, 0, 0, "   Plugin load failed: function plugin_init() does not exist.\n", dlerror());
-               dlclose(p);
-               return;
+               if (!initfunc(&funcs))
+               {
+                       log(1, 0, 0, 0, "   Plugin load failed: plugin_init() returned FALSE: %s\n", dlerror());
+                       dlclose(p);
+                       return;
+               }
        }
 
-       if (!initfunc(&funcs))
-       {
-               log(1, 0, 0, 0, "   Plugin load failed: plugin_init() returned FALSE.\n", dlerror());
-               dlclose(p);
-               return;
-       }
+       ll_push(loaded_plugins, p);
 
        for (i = 0; i < max_plugin_functions; i++)
        {
                void *x;
-               if (!plugin_functions[i]) continue;
-               if ((x = dlsym(p, plugin_functions[i])))
+               if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
                {
                        log(3, 0, 0, 0, "   Supports function \"%s\"\n", plugin_functions[i]);
                        ll_push(plugins[i], x);
                }
        }
+
        log(2, 0, 0, 0, "   Loaded plugin %s\n", plugin_name);
 }
 
+static void run_plugin_done(void *plugin)
+{
+       int (*donefunc)(void) = dlsym(plugin, "plugin_done");
+
+       if (donefunc)
+               donefunc();
+}
+
 void remove_plugin(char *plugin_name)
 {
-       void *p;
-       int (*donefunc)();
-       char path[256] = {0};
+       void *p = open_plugin(plugin_name, 0);
        int i;
 
-       snprintf(path, 256, "%s/%s.so", LIBDIR, plugin_name);
-
-       log(2, 0, 0, 0, "Removing plugin %s\n", plugin_name);
-       // Get the existing pointer
-       p = dlopen(path, RTLD_LAZY);
-       if (!p) return;
+       if (!p)
+               return;
 
        for (i = 0; i < max_plugin_functions; i++)
        {
                void *x;
-               if (!plugin_functions[i]) continue;
-               if ((x = dlsym(p, plugin_functions[i]))) ll_delete(plugins[i], x);
+               if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
+                       ll_delete(plugins[i], x);
        }
 
        if (ll_contains(loaded_plugins, p))
        {
                ll_delete(loaded_plugins, p);
-
-               donefunc = dlsym(p, "plugin_done");
-               if (donefunc) donefunc();
+               run_plugin_done(p);
        }
 
-       dlclose(p);
        dlclose(p);
        log(2, 0, 0, 0, "Removed plugin %s\n", plugin_name);
 }
@@ -2896,6 +3691,15 @@ int run_plugins(int plugin_type, void *data)
        return 1;
 }
 
+void plugins_done()
+{
+       void *p;
+
+       ll_reset(loaded_plugins);
+       while ((p = ll_next(loaded_plugins)))
+               run_plugin_done(p);
+}
+
 void processcontrol(u8 * buf, int len, struct sockaddr_in *addr)
 {
        char *resp;
@@ -2917,18 +3721,7 @@ void processcontrol(u8 * buf, int len, struct sockaddr_in *addr)
        param.response = resp;
        param.response_length = l;
 
-       if (param.type == PKT_LOAD_PLUGIN && param.data_length)
-       {
-               add_plugin(param.data);
-       }
-       else if (param.type == PKT_UNLOAD_PLUGIN && param.data_length)
-       {
-               remove_plugin(param.data);
-       }
-       else
-       {
-               run_plugins(PLUGIN_CONTROL, &param);
-       }
+       run_plugins(PLUGIN_CONTROL, &param);
 
        if (param.send_response)
        {
@@ -2975,6 +3768,8 @@ tunnelidt new_tunnel()
                if (tunnel[i].state == TUNNELFREE)
                {
                        log(4, 0, 0, i, "Assigning tunnel ID %d\n", i);
+                       if (i > config->cluster_highest_tunnelid)
+                               config->cluster_highest_tunnelid = i;
                        return i;
                }
        }
@@ -2982,3 +3777,91 @@ tunnelidt new_tunnel()
        return 0;
 }
 
+//
+// We're becoming the master. Do any required setup..
+//
+// This is principally telling all the plugins that we're
+// now a master, and telling them about all the sessions
+// that are active too..
+//
+void become_master(void)
+{
+       int s;
+       run_plugins(PLUGIN_BECOME_MASTER, NULL);
+
+       for (s = 0; s < config->cluster_highest_sessionid ; ++s) {
+               if (!session[s].tunnel) // Not an in-use session.
+                       continue;
+
+               run_plugins(PLUGIN_NEW_SESSION_MASTER, &session[s]);
+       }
+}
+
+
+
+int cmd_show_hist_idle(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       int s, i;
+       int count = 0;
+       int buckets[64];
+
+       time(&time_now);
+       for (i = 0; i < 64;++i) buckets[i] = 0;
+
+       for (s = 0; s < config->cluster_highest_sessionid ; ++s) {
+               int idle;
+               if (!session[s].tunnel)
+                       continue;
+
+               idle = time_now - session[s].last_packet;
+               idle /= 5 ; // In multiples of 5 seconds.
+               if (idle < 0)
+                       idle = 0;
+               if (idle > 63)
+                       idle = 63;
+
+               ++count;
+               ++buckets[idle];
+       }
+
+       for (i = 0; i < 63; ++i) {
+               cli_print(cli, "%3d seconds  : %7.2f%% (%6d)", i * 5, (double) buckets[i] * 100.0 / count , buckets[i]);
+       }
+       cli_print(cli, "lots of secs : %7.2f%% (%6d)", (double) buckets[63] * 100.0 / count , buckets[i]);
+       cli_print(cli, "%d total sessions open.", count);
+       return CLI_OK;
+}
+
+int cmd_show_hist_open(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       int s, i;
+       int count = 0;
+       int buckets[64];
+
+       time(&time_now);
+       for (i = 0; i < 64;++i) buckets[i] = 0;
+
+       for (s = 0; s < config->cluster_highest_sessionid ; ++s) {
+               int open = 0, d;
+               if (!session[s].tunnel)
+                       continue;
+
+               d = time_now - session[s].opened;
+               if (d < 0)
+                       d = 0;
+               while (d > 1 && open < 32) {
+                       ++open;
+                       d >>= 1; // half.
+               }
+               ++count;
+               ++buckets[open];
+       }
+
+       s = 1;
+       for (i = 0; i  < 30; ++i) {
+               cli_print(cli, " < %8d seconds : %7.2f%% (%6d)", s, (double) buckets[i] * 100.0 / count , buckets[i]);
+               s <<= 1;
+       }
+       cli_print(cli, "%d total sessions open.", count);
+       return CLI_OK;
+}