make log a conditional macro for _log
[l2tpns.git] / l2tpns.c
index b988c7f..c0e7b7a 100644 (file)
--- a/l2tpns.c
+++ b/l2tpns.c
@@ -1,8 +1,11 @@
 // L2TP Network Server
 // Adrian Kennard 2002
-// (c) Copyrigth 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd)
+// Copyright (c) 2003, 2004 Optus Internet Engineering
+// Copyright (c) 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced
 // vim: sw=8 ts=8
 
+char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.39 2004/11/03 13:23:58 bodea Exp $";
+
 #include <arpa/inet.h>
 #include <assert.h>
 #include <errno.h>
 #include <sys/mman.h>
 #include <netdb.h>
 #include <netinet/in.h>
-#include <signal.h>
 #include <stdarg.h>
 #include <stdlib.h>
 #include <stdio.h>
-#define __USE_GNU
 #include <string.h>
 #include <sys/ioctl.h>
 #include <sys/socket.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"
 #include "constants.h"
 #include "control.h"
 #include "util.h"
+#include "tbf.h"
 
 // Globals
 struct configt *config = NULL; // all configuration
-int tapfd = -1;                        // tap interface file handle
+int tunfd = -1;                        // tun interface file handle. (network device)
 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
+int tunidx;                    // ifr_ifindex of tun device
 u32 sessionid = 0;             // session id for radius accounting
 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;
-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;
-char time_now_string[64] = {0};
-char main_quit = 0;
+FILE *log_stream = NULL;       // file handle for direct logging (i.e. direct into file, not via syslog).
+extern int cluster_sockfd;     // Intra-cluster communications socket.
+u32 last_id = 0;               // Last used PPP SID. Can I kill this?? -- mo
+int clifd = 0;                 // Socket listening for CLI connections.
+
+struct cli_session_actions *cli_session_actions = NULL;        // Pending session changes requested by CLI
+struct cli_tunnel_actions *cli_tunnel_actions = NULL;  // Pending tunnel changes required by CLI
+
+static void *ip_hash[256];     // Mapping from IP address to session structures.
+
+u32 udp_tx = 0, udp_rx = 0, udp_rx_pkt = 0;    // Global traffic counters.
+u32 eth_tx = 0, eth_rx = 0, eth_rx_pkt = 0;
+u32 ip_pool_size = 1;          // Size of the pool of addresses used for dynamic address allocation.
+time_t time_now = 0;           // Current time in seconds since epoch.
+char time_now_string[64] = {0};        // Current time as a string.
+char main_quit = 0;            // True if we're in the process of exiting.
 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 }
 
 struct config_descriptt config_values[] = {
        CONFIG("debug", debug, INT),
        CONFIG("log_file", log_filename, STRING),
+       CONFIG("pid_file", pid_file, STRING),
        CONFIG("l2tp_secret", l2tpsecret, STRING),
        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("primary_radius_port", radiusport[0], SHORT),
+       CONFIG("secondary_radius_port", radiusport[1], SHORT),
        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("throttle_buckets", num_tbfs, INT),
        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("lock_pages", lock_pages, BOOL),
+       CONFIG("icmp_rate", icmp_rate, INT),
+       CONFIG("cluster_address", cluster_address, IP),
+       CONFIG("cluster_interface", cluster_interface, STRING),
+       CONFIG("cluster_hb_interval", cluster_hb_interval, INT),
+       CONFIG("cluster_hb_timeout", cluster_hb_timeout, INT),
+#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 },
 };
 
@@ -113,34 +142,40 @@ 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
-radiust *radius = NULL;
-ippoolt *ip_address_pool = NULL;
+tunnelt *tunnel = NULL;                        // Array of tunnel structures.
+sessiont *session = NULL;              // Array of session structures.
+sessioncountt *sess_count = NULL;      // Array of partial per-session traffic counters.
+radiust *radius = NULL;                        // Array of radius structures.
+ippoolt *ip_address_pool = NULL;       // Array of dynamic IP addresses.
 controlt *controlfree = 0;
 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();
 void tunnel_clean();
 tunnelidt new_tunnel();
 void update_config();
+int unhide_avp(u8 *avp, tunnelidt t, sessionidt s, u16 length);
 
-// return internal time (10ths since run)
+static void cache_ipmap(ipt ip, int s);
+static void uncache_ipmap(ipt ip);
+
+// return internal time (10ths since process startup)
 clockt now(void)
 {
        struct timeval t;
@@ -149,16 +184,22 @@ clockt now(void)
 }
 
 // work out a retry time based on try number
+// This is a straight bounded exponential backoff.
+// Maximum re-try time is 32 seconds. (2^5).
 clockt backoff(u8 try)
 {
        if (try > 5) try = 5;                  // max backoff
        return now() + 10 * (1 << try);
 }
 
+
+//
+// Log a debug message.
+//
 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 +228,7 @@ 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)
        {
@@ -198,17 +239,17 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
        va_end(ap);
 }
 
-void _log_hex(int level, ipt address, sessionidt s, tunnelidt t, const char *title, const char *data, int maxsize)
+void _log_hex(int level, 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;
 
        // No support for log_hex to syslog
        if (log_stream)
        {
-               log(level, address, s, t, "%s (%d bytes):\n", title, maxsize);
+               _log(level, 0, 0, 0, "%s (%d bytes):\n", title, maxsize);
                setvbuf(log_stream, NULL, _IOFBF, 16384);
 
                for (i = 0; i < maxsize; )
@@ -251,76 +292,126 @@ 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 &= mask;             // Force the ip to be the first one in the route.
+
        memset(&r, 0, sizeof(r));
-       r.rt_dev = config->tapdevice;
+       r.rt_dev = config->tundevice;
        r.rt_dst.sa_family = AF_INET;
        *(u32 *) & (((struct sockaddr_in *) &r.rt_dst)->sin_addr.s_addr) = htonl(ip);
        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);
+
+       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);
+
+       if (ioctl(ifrfd, add ? SIOCADDRT : SIOCDELRT, (void *) &r) < 0)
+               log(0, 0, 0, 0, "routeset() error in ioctl: %s\n", strerror(errno));
+
+#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
-void inittap(void)
+//
+// Set up TUN interface
+void inittun(void)
 {
        struct ifreq ifr;
        struct sockaddr_in sin = {0};
        memset(&ifr, 0, sizeof(ifr));
        ifr.ifr_flags = IFF_TUN;
 
-       tapfd = open(TAPDEVICE, O_RDWR);
-       if (tapfd < 0)
+       tunfd = open(TUNDEVICE, O_RDWR);
+       if (tunfd < 0)
        {                          // fatal
-               log(0, 0, 0, 0, "Can't open %s: %s\n", TAPDEVICE, strerror(errno));
-               exit(-1);
+               log(0, 0, 0, 0, "Can't open %s: %s\n", TUNDEVICE, strerror(errno));
+               exit(1);
+       }
+       {
+               int flags = fcntl(tunfd, F_GETFL, 0);
+               fcntl(tunfd, F_SETFL, flags | O_NONBLOCK);
        }
-       if (ioctl(tapfd, TUNSETIFF, (void *) &ifr) < 0)
+       if (ioctl(tunfd, TUNSETIFF, (void *) &ifr) < 0)
        {
-               log(0, 0, 0, 0, "Can't set tap interface: %s\n", strerror(errno));
-               exit(-1);
+               log(0, 0, 0, 0, "Can't set tun interface: %s\n", strerror(errno));
+               exit(1);
        }
-       assert(strlen(ifr.ifr_name) < sizeof(config->tapdevice));
-       strncpy(config->tapdevice, ifr.ifr_name, sizeof(config->tapdevice) - 1);
+       assert(strlen(ifr.ifr_name) < sizeof(config->tundevice));
+       strncpy(config->tundevice, ifr.ifr_name, sizeof(config->tundevice) - 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);
        }
-       ifr.ifr_flags = IFF_UP;
-       if (ioctl(ifrfd, SIOCSIFFLAGS, (void *) &ifr) < 0)
+       /* Bump up the qlen to deal with bursts from the network */
+       ifr.ifr_qlen = 1000;
+       if (ioctl(ifrfd, SIOCSIFTXQLEN, (void *) &ifr) < 0)
        {
-               perror("set tap flags");
-               exit( -1);
+               log(0, 0, 0, 0, "Error setting tun queue length: %s\n", strerror(errno));
+               exit(1);
        }
-       if (ioctl(ifrfd, SIOCGIFHWADDR, (void *) &ifr) < 0)
+       ifr.ifr_flags = IFF_UP;
+       if (ioctl(ifrfd, SIOCSIFFLAGS, (void *) &ifr) < 0)
        {
-               perror("get tap hwaddr");
-               exit( -1);
+               log(0, 0, 0, 0, "Error setting tun flags: %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;
+       tunidx = ifr.ifr_ifindex;
 }
 
 // set up UDP port
@@ -334,14 +425,18 @@ void initudp(void)
        addr.sin_family = AF_INET;
        addr.sin_port = htons(L2TPPORT);
        addr.sin_addr.s_addr = config->bind_address;
-       udpfd = socket(AF_INET, SOCK_DGRAM, UDP);
+       udpfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_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);
+       snoopfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
 
        // Control
        memset(&addr, 0, sizeof(addr));
@@ -351,31 +446,60 @@ 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);
+       CSTAT(call_sessionbyip);
+
+       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 +514,116 @@ 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++)
+//
+// 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;
+
+       if (CLI_HELP_REQUESTED)
+               return CLI_HELP_NO_ARGS;
+
+       cli_print(cli, "%7s %s", "Sess#", "IP Address");
+
+       for (i = 0; i < 256; ++i)
        {
-               if (!d[(size_t) a[i]]) return;
-               d = (char **) d[(size_t) a[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;
+       CSTAT(call_sessionbyuser);
+
+       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];
-
-    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);
+       int s;
+       struct ifreq ifr;
+       u8 mac[6];
+
+       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
@@ -466,108 +640,27 @@ sessionidt sessionidtbysessiont(sessiont *s)
        return val;
 }
 
-// send gratuitous ARP to set ARP table for newly allocated IP
-void sessionsendarp(sessionidt s)
+// actually send a control message for a specific tunnel
+void tunnelsend(u8 * buf, u16 l, tunnelidt t)
 {
-        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);
-}
+       struct sockaddr_in addr;
 
-// Handle ARP requests
-void processarp(u8 * buf, int len)
-{
-       ipt ip;
-       sessionidt s;
+       CSTAT(call_tunnelsend);
 
-#ifdef STAT_CALLS
-       STAT(call_processarp);
-#endif
-       STAT(arp_recv);
-       if (len != 46)
-       {
-               log(0, 0, 0, 0, "Unexpected length ARP %d bytes\n", len);
-               STAT(arp_errors);
-               return;
-       }
-       if (*(u16 *) (buf + 16) != htons(PKTARP))
-       {
-               log(0, 0, 0, 0, "Unexpected ARP type %04X\n", ntohs(*(u16 *) (buf + 16)));
-               STAT(arp_errors);
-               return;
-       }
-       if (*(u16 *) (buf + 18) != htons(0x0001))
-       {
-               log(0, 0, 0, 0, "Unexpected ARP hard type %04X\n", ntohs(*(u16 *) (buf + 18)));
-               STAT(arp_errors);
-               return;
-       }
-       if (*(u16 *) (buf + 20) != htons(PKTIP))
-       {
-               log(0, 0, 0, 0, "Unexpected ARP prot type %04X\n", ntohs(*(u16 *) (buf + 20)));
-               STAT(arp_errors);
-               return;
-       }
-       if (buf[22] != 6)
+       if (!t)
        {
-               log(0, 0, 0, 0, "Unexpected ARP hard len %d\n", buf[22]);
-               STAT(arp_errors);
-               return;
-       }
-       if (buf[23] != 4)
-       {
-               log(0, 0, 0, 0, "Unexpected ARP prot len %d\n", buf[23]);
-               STAT(arp_errors);
-               return;
-       }
-       if (*(u16 *) (buf + 24) != htons(0x0001))
-       {
-               log(0, 0, 0, 0, "Unexpected ARP op %04X\n", ntohs(*(u16 *) (buf + 24)));
-               STAT(arp_errors);
+               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;
        }
-       ip = ntohl(*(u32 *) (buf + 42));
-       // look up session
-       s = sessionbyip(htonl(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);
-               memcpy(buf + 4, buf + 10, 6); // set destination as source
-               *(u16 *) (buf + 10) = htons(tapmac[0]); // set soucre 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
-               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);
-               STAT(arp_discarded);
-       }
-}
-
-// actually send a control message for a specific tunnel
-void tunnelsend(u8 * buf, u16 l, tunnelidt t)
-{
-       struct sockaddr_in addr;
 
-#ifdef STAT_CALLS
-       STAT(call_tunnelsend);
-#endif
        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 +697,43 @@ 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(tunfd, data, size);
+}
+
 // process outgoing (to tunnel) IP
+//
 void processipout(u8 * buf, int len)
 {
        sessionidt s;
        sessiont *sp;
        tunnelidt t;
        ipt ip;
-       u8 b[MAXETHER];
-#ifdef STAT_CALLS
-       STAT(call_processipout);
-#endif
-       if (len < 38)
+
+       char * data = buf;      // Keep a copy of the originals.
+       int size = len;
+
+       u8 b[MAXETHER + 20];
+
+       CSTAT(call_processipout);
+
+       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 +749,116 @@ 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
+       {
+               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)
        {
-               struct param_packet_rx packet = { &tunnel[t], &session[s], buf, len };
-               run_plugins(PLUGIN_PACKET_RX, &packet);
+               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.
+       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 +925,13 @@ 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
@@ -766,19 +964,69 @@ void controladd(controlt * c, tunnelidt t, sessionidt s)
        }
 }
 
+//
+// Throttle or Unthrottle a session
+//
+// Throttle the data from/to through a session to no more than
+// 'rate_in' kbit/sec in (from user) or 'rate_out' kbit/sec out (to
+// user).
+//
+// If either value is -1, the current value is retained for that
+// direction.
+//
+void throttle_session(sessionidt s, int rate_in, int rate_out)
+{
+       if (!session[s].tunnel)
+               return; // No-one home.
+
+       if (!*session[s].user)
+               return; // User not logged in
+
+       if (rate_in >= 0)
+       {
+               int bytes = rate_in * 1024 / 8; // kbits to bytes
+               if (session[s].tbf_in)
+                       free_tbf(session[s].tbf_in);
+
+               if (rate_in > 0)
+                       session[s].tbf_in = new_tbf(s, bytes * 2, bytes, send_ipin);
+               else
+                       session[s].tbf_in = 0;
+
+               session[s].throttle_in = rate_in;
+       }
+
+       if (rate_out >= 0)
+       {
+               int bytes = rate_out * 1024 / 8;
+               if (session[s].tbf_out)
+                       free_tbf(session[s].tbf_out);
+
+               if (rate_out > 0)
+                       session[s].tbf_out = new_tbf(s, bytes * 2, bytes, send_ipout);
+               else
+                       session[s].tbf_out = 0;
+
+               session[s].throttle_out = rate_out;
+       }
+}
+
 // start tidy shutdown of session
 void sessionshutdown(sessionidt s, char *reason)
 {
        int dead = session[s].die;
        int walled_garden = session[s].walled_garden;
 
-#ifdef STAT_CALLS
-       STAT(call_sessionshutdown);
-#endif
+
+       CSTAT(call_sessionshutdown);
+
        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)
+       if (!dead)
                log(2, 0, s, session[s].tunnel, "Shutting down session %d: %s\n", s, reason);
 
        session[s].die = now() + 150; // Clean up in 15 seconds
@@ -789,8 +1037,9 @@ void sessionshutdown(sessionidt s, char *reason)
        }
 
        // RADIUS Stop message
-       if (session[s].opened && !walled_garden && !dead) {
-               u8 r = session[s].radius;
+       if (session[s].opened && !walled_garden && !dead)
+       {
+               u16 r = session[s].radius;
                if (!r)
                {
                        if (!(r = radiusnew(s)))
@@ -811,75 +1060,95 @@ 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_in || session[s].throttle_out) // Unthrottle if throttled.
+               throttle_session(s, 0, 0);
+
        {                            // Send CDN
                controlt *c = controlnew(14); // sending CDN
                control16(c, 1, 3, 1);    // result code (admin reasons - TBA make error, general error, add message
                control16(c, 14, s, 1);   // assigned session (our end)
                controladd(c, session[s].tunnel, s); // send the message
        }
+
        cluster_send_session(s);
 }
 
 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);
-#endif
+
+       CSTAT(call_sendipcp);
+
        if (!r)
                r = radiusnew(s);
+
        if (radius[r].state != RADIUSIPCP)
        {
                radius[r].state = RADIUSIPCP;
                radius[r].try = 0;
        }
+
        radius[r].retry = backoff(radius[r].try++);
        if (radius[r].try > 10)
        {
+               radiusclear(r, s);      // Clear radius session.
                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 ? config->bind_address : my_address; // send my IP
        tunnelsend(buf, 10 + (q - buf), t); // send it
+       session[s].flags &= ~SF_IPCP_ACKED;     // Clear flag.
 }
 
 // kill a session now
 void sessionkill(sessionidt s, char *reason)
 {
-#ifdef STAT_CALLS
-       STAT(call_sessionkill);
-#endif
+
+       CSTAT(call_sessionkill);
+
+       session[s].die = now();
        sessionshutdown(s, reason);  // close radius/routes, etc.
        if (session[s].radius)
-               radiusclear(session[s].radius, 0); // cant send clean accounting data, session is killed
+               radiusclear(session[s].radius, s); // 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);
+
        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);
+       cli_session_actions[s].action = 0;
        cluster_send_session(s);
 }
 
@@ -888,9 +1157,8 @@ void tunnelkill(tunnelidt t, char *reason)
 {
        sessionidt s;
        controlt *c;
-#ifdef STAT_CALLS
-       STAT(call_tunnelkill);
-#endif
+
+       CSTAT(call_tunnelkill);
 
        tunnel[t].state = TUNNELDIE;
 
@@ -910,19 +1178,18 @@ void tunnelkill(tunnelidt t, char *reason)
 
        // free tunnel
        tunnelclear(t);
-       cluster_send_tunnel(t);
        log(1, 0, 0, t, "Kill tunnel %d: %s\n", t, reason);
-       tunnel[t].die = 0;
-       tunnel[t].state = TUNNELFREE;
+       cli_tunnel_actions[s].action = 0;
+       cluster_send_tunnel(t);
 }
 
 // shut down a tunnel cleanly
 void tunnelshutdown(tunnelidt t, char *reason)
 {
        sessionidt s;
-#ifdef STAT_CALLS
-       STAT(call_tunnelshutdown);
-#endif
+
+       CSTAT(call_tunnelshutdown);
+
        if (!tunnel[t].last || !tunnel[t].far || tunnel[t].state == TUNNELFREE)
        {
                // never set up, can immediately kill
@@ -955,9 +1222,9 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
        u16 l = len, t = 0, s = 0, ns = 0, nr = 0;
        u8 *p = buf + 2;
 
-#ifdef STAT_CALLS
-       STAT(call_processudp);
-#endif
+
+       CSTAT(call_processudp);
+
        udp_rx += len;
        udp_rx_pkt++;
        log_hex(5, "UDP Data", buf, len);
@@ -996,12 +1263,6 @@ 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);
@@ -1034,15 +1295,40 @@ 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[i].state != TUNNELOPENING ||
+                                       tunnel[i].ip != ntohl(*(ipt *) & addr->sin_addr) ||
+                                       tunnel[i].port != ntohs(addr->sin_port) )
+                                       continue;
+                               t = i;
+                               break;
+                       }
+               }
+
                if (!t)
                {
                        if (!(t = new_tunnel()))
@@ -1065,29 +1351,36 @@ 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 +1436,14 @@ 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");
+                                       // Unhide the AVP
+                                       n = unhide_avp(b, t, s, n);
+                                       if (n == 0)
+                                       {
+                                               fatal = flags;
+                                               continue;
+                                       }
                                }
                                if (*b & 0x3C)
                                {
@@ -1176,17 +1476,21 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        {
                                                u16 rescode = ntohs(*(u16 *)(b));
                                                const char* resdesc = "(unknown)";
-                                               if (message == 4) { /* StopCCN */
+                                               if (message == 4)
+                                               { /* StopCCN */
                                                        if (rescode <= MAX_STOPCCN_RESULT_CODE)
                                                                resdesc = stopccn_result_codes[rescode];
-                                               } else if (message == 14) { /* CDN */
+                                               }
+                                               else if (message == 14)
+                                               { /* CDN */
                                                        if (rescode <= MAX_CDN_RESULT_CODE)
                                                                resdesc = cdn_result_codes[rescode];
                                                }
 
                                                log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Result Code %d: %s\n",
                                                        rescode, resdesc);
-                                               if (n >= 4) {
+                                               if (n >= 4)
+                                               {
                                                        u16 errcode = ntohs(*(u16 *)(b + 2));
                                                        const char* errdesc = "(unknown)";
                                                        if (errcode <= MAX_ERROR_CODE)
@@ -1194,13 +1498,10 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                        log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Error Code %d: %s\n",
                                                                errcode, errdesc);
                                                }
-                                               if (n > 4) {
-                                                       /* %*s doesn't work?? */
-                                                       char *buf = (char *)strndup(b+4, n-4);
-                                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Error String: %s\n",
-                                                               buf);
-                                                       free(buf);
-                                               }
+                                               if (n > 4)
+                                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Error String: %.*s\n",
+                                                               n-4, b+4);
+
                                                break;
                                        }
                                        break;
@@ -1237,8 +1538,8 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        // TBA - to send to RADIUS
                                        break;
                                case 8:     // vendor name
-                                       memset(tunnel[t].vendor, 0, 128);
-                                       memcpy(tunnel[t].vendor, b, (n >= 127) ? 127 : n);
+                                       memset(tunnel[t].vendor, 0, sizeof(tunnel[t].vendor));
+                                       memcpy(tunnel[t].vendor, b, (n >= sizeof(tunnel[t].vendor) - 1) ? sizeof(tunnel[t].vendor) - 1 : n);
                                        log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Vendor name = \"%s\"\n", tunnel[t].vendor);
                                        break;
                                case 9:     // assigned tunnel
@@ -1257,6 +1558,11 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                build_chap_response(b, 2, n, &chapresponse);
                                        }
                                        break;
+                               case 13:    // Response
+                                       // Why did they send a response? We never challenge.
+                                       log(2, ntohl(addr->sin_addr.s_addr), s, t, "   received unexpected challenge response\n");
+                                       break;
+
                                case 14:    // assigned session
                                        asession = session[s].far = ntohs(*(u16 *) (b));
                                        log(4, ntohl(addr->sin_addr.s_addr), s, t, "   assigned session = %d\n", asession);
@@ -1282,6 +1588,8 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        memcpy(calling, b, (n >= MAXTEL) ? (MAXTEL-1) : n);
                                        log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Calling <%s>\n", calling);
                                        break;
+                               case 23:    // subtype
+                                       break;
                                case 24:    // tx connect speed
                                        if (n == 4)
                                        {
@@ -1294,7 +1602,7 @@ 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 +1617,7 @@ 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 +1645,7 @@ 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
@@ -1457,13 +1765,16 @@ 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)))
                                                {
@@ -1497,8 +1808,9 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        // TBA
                                        break;
                                case 12:      // ICCN
+                                       if (amagic == 0) amagic = time_now;
                                        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 +1845,6 @@ 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 +1867,83 @@ 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
@@ -1598,66 +1954,316 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
        }
 }
 
-// read and process packet on tap
-void processtap(u8 * buf, int len)
+// read and process packet on tun
+void processtun(u8 * buf, int len)
 {
-       log_hex(5, "Receive TAP Data", buf, len);
-       STAT(tap_rx_packets);
-       INC_STAT(tap_rx_bytes, len);
-#ifdef STAT_CALLS
-       STAT(call_processtap);
-#endif
+       log_hex(5, "Receive TUN Data", buf, len);
+       STAT(tun_rx_packets);
+       INC_STAT(tun_rx_bytes, len);
+
+       CSTAT(call_processtun);
+
        eth_rx_pkt++;
        eth_rx += len;
        if (len < 22)
        {
-               log(1, 0, 0, 0, "Short tap packet %d bytes\n", len);
-               STAT(tap_rx_errors);
+               log(1, 0, 0, 0, "Short tun packet %d bytes\n", len);
+               STAT(tun_rx_errors);
                return;
        }
-       if (*(u16 *) (buf + 2) == htons(PKTARP)) // ARP
-               processarp(buf, len);
-       else if (*(u16 *) (buf + 2) == htons(PKTIP)) // ARP
+
+       if (*(u16 *) (buf + 2) == htons(PKTIP)) // IP
                processipout(buf, len);
-       else
+       // Else discard.
+}
+
+//
+// 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)
+{
+       static sessionidt s = 0;        // Next session to check for actions on.
+       tunnelidt t;
+       int count=0,i;
+       u16 r;
+       static clockt next_acct = 0;
+       int a;
+
+       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 <= 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 <= 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 <= 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 < 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 tunnel changes requested from the CLI
+               if ((a = cli_tunnel_actions[t].action))
+               {
+                       cli_tunnel_actions[t].action = 0;
+                       if (a & CLI_TUN_KILL)
+                       {
+                               log(2, tunnel[t].ip, 0, t, "Dropping tunnel by CLI\n");
+                               tunnelshutdown(t, "Requested by administrator");
+                       }
+               }
+
+       }
+
+       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 <= 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;
+               }
+
+               // Check for actions requested from the CLI
+               if ((a = cli_session_actions[s].action))
+               {
+                       int send = 0;
+
+                       cli_session_actions[s].action = 0;
+                       if (a & CLI_SESS_KILL)
+                       {
+                               log(2, 0, s, session[s].tunnel, "Dropping session by CLI\n");
+                               sessionshutdown(s, "Requested by administrator");
+                               a = 0; // dead, no need to check for other actions
+                       }
+
+                       if (a & CLI_SESS_NOSNOOP)
+                       {
+                               log(2, 0, s, session[s].tunnel, "Unsnooping session by CLI\n");
+                               session[s].snoop_ip = 0;
+                               session[s].snoop_port = 0;
+                               send++;
+                       }
+                       else if (a & CLI_SESS_SNOOP)
+                       {
+                               log(2, 0, s, session[s].tunnel, "Snooping session by CLI (to %s:%d)\n",
+                                   inet_toa(cli_session_actions[s].snoop_ip), cli_session_actions[s].snoop_port);
+
+                               session[s].snoop_ip = cli_session_actions[s].snoop_ip;
+                               session[s].snoop_port = cli_session_actions[s].snoop_port;
+                               send++;
+                       }
+
+                       if (a & CLI_SESS_NOTHROTTLE)
+                       {
+                               log(2, 0, s, session[s].tunnel, "Un-throttling session by CLI\n");
+                               throttle_session(s, 0, 0);
+                               send++;
+                       }
+                       else if (a & CLI_SESS_THROTTLE)
+                       {
+                               log(2, 0, s, session[s].tunnel, "Throttling session by CLI (to %dkb/s up and %dkb/s down)\n",
+                                   cli_session_actions[s].throttle_in,
+                                   cli_session_actions[s].throttle_out);
+
+                               throttle_session(s, cli_session_actions[s].throttle_in, cli_session_actions[s].throttle_out);
+                               send++;
+                       }
+
+                       if (send)
+                               cluster_send_session(s);
+
+                       if (++count >= MAX_ACTIONS) break;
+               }
+       }
+
+       if (*config->accounting_dir && next_acct <= TIME)
+       {
+               // Dump accounting data
+               next_acct = 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 clockt last_talked = 0;
+       static clockt start_busy_wait = 0;
+       if (start_busy_wait == 0)
+               start_busy_wait = TIME;
+
+       for (i = config->cluster_highest_tunnelid ; i > 0 ; --i)
+       {
+               if (!tunnel[i].controlc)
+                       continue;
+
+               if (last_talked != TIME)
+               {
+                       log(2,0,0,0, "Tunnel %d still has un-acked control messages.\n", i);
+                       last_talked = TIME;
+               }
+               return 1;
+       }
+
+       // We stop waiting for radius after BUSY_WAIT_TIME 1/10th seconds
+       if (abs(TIME - start_busy_wait) > BUSY_WAIT_TIME)
+       {
+               log(1, 0, 0, 0, "Giving up waiting for RADIUS to be empty. Shutting down anyway.\n");
+               return 0;
+       }
+
+       for (i = 1; i < MAXRADIUS; i++)
        {
-               log(1, 0, 0, 0, "Unexpected tap packet %04X, %d bytes\n", ntohs(*(u16 *) (buf + 2)), len);
+               if (radius[i].state == RADIUSNULL)
+                       continue;
+               if (radius[i].state == RADIUSWAIT)
+                       continue;
+
+               if (last_talked != TIME)
+               {
+                       log(2,0,0,0, "Radius session %d is still busy (sid %d)\n", i, radius[i].session);
+                       last_talked = TIME;
+               }
+               return 1;
        }
+
+       return 0;
 }
 
-// main loop - gets packets on tap or udp and processes them
+static fd_set readset;
+static int readset_n = 0;
+
+// main loop - gets packets on tun or udp and processes them
 void mainloop(void)
 {
-       fd_set cr;
-       int cn;
+       int i;
        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);
-
-       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);
-       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;
-
-       while (!main_quit)
+       clockt next_cluster_ping = 0;   // send initial ping immediately
+       time_t next_clean = time_now + config->cleanup_interval;
+
+       log(4, 0, 0, 0, "Beginning of main loop. udpfd=%d, tunfd=%d, cluster_sockfd=%d, controlfd=%d\n",
+                       udpfd, tunfd, cluster_sockfd, controlfd);
+
+       FD_ZERO(&readset);
+       FD_SET(udpfd, &readset);
+       FD_SET(tunfd, &readset);
+       FD_SET(controlfd, &readset);
+       FD_SET(clifd, &readset);
+       if (cluster_sockfd) FD_SET(cluster_sockfd, &readset);
+       readset_n = udpfd;
+       if (tunfd > readset_n)          readset_n = tunfd;
+       if (controlfd > readset_n)      readset_n = controlfd;
+       if (clifd > readset_n)          readset_n = clifd;
+       if (cluster_sockfd > readset_n) readset_n = cluster_sockfd;
+
+       while (!main_quit || still_busy())
        {
                fd_set r;
-               int n = cn;
+               int n = readset_n;
+#ifdef BGP
+               fd_set w;
+               int bgp_set[BGP_NUM_PEERS];
+#endif /* BGP */
 
                if (config->reload_config)
                {
@@ -1665,31 +2271,89 @@ void mainloop(void)
                        update_config();
                }
 
-               memcpy(&r, &cr, sizeof(fd_set));
-               n = select(n + 1, &r, 0, 0, &to);
-               if (n < 0)
+               memcpy(&r, &readset, sizeof(fd_set));
+               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)
+                       {
+                               FD_SET(bgp_peers[i].sock, &r);
+                               if (bgp_peers[i].sock > n)
+                                       n = bgp_peers[i].sock;
+                       }
+
+                       if (bgp_set[i] & 2)
                        {
-                               perror("select");
-                               exit( -1);
+                               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 */
+
+               TIME = now();
+               if (n < 0)
+               {
+                       if (errno == EINTR ||
+                           errno == ECHILD) // EINTR was clobbered by sigchild_handler()
+                               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(tunfd, &r))
+                       {
+                               int c, n;
+                               for (c = 0; c < config->multi_read_count; c++)
+                               {
+                                       if ((n = read(tunfd, buf, sizeof(buf))) > 0)
+                                               processtun(buf, n);
+                                       else
+                                               break;
+                               }
+                       }
+
+                       if (config->cluster_iam_master)
+                               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;
@@ -1706,266 +2370,209 @@ void mainloop(void)
                                        close(sockfd);
                                }
                        }
+               }
+
+                       // Runs on every machine (master and slaves).
+               if (cluster_sockfd && next_cluster_ping <= TIME)
+               {
+                       // Check to see which of the cluster is still alive..
+
+                       cluster_send_ping(basetime);    // Only does anything if we're a slave
+                       cluster_check_master();         // ditto.
+
+                       cluster_heartbeat();            // Only does anything if we're a master.
+                       cluster_check_slaves();         // ditto.
+
+                       master_update_counts();         // If we're a slave, send our byte counters to our master.
+
+                       if (config->cluster_iam_master && !config->cluster_iam_uptodate)
+                               next_cluster_ping = TIME + 1; // out-of-date slaves, do fast updates
                        else
-                       {
-                               log(1, 0, 0, 0, "Main select() loop returned %d, but no fds have data waiting\n", n);
-                               continue;
-                       }
+                               next_cluster_ping = TIME + config->cluster_hb_interval;
                }
-               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++)
+                       // 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 != TIME)
                        {
-                               // 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");
-                               }
+                               last_run = TIME;
+                               tbf_run_timer();
                        }
+               }
 
-                       // Check for sessions that have been killed from the CLI
-                       if (cli_session_kill[0])
+               /* 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())
                        {
-                               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;
-                               }
+                               // Did it finish?
+                               next_clean = time_now + 1 ;     // Didn't finish. Check quickly.
                        }
-                       // Check for tunnels that have been killed from the CLI
-                       if (cli_tunnel_kill[0])
+                       else
                        {
-                               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;
-                               }
+                               next_clean = time_now + config->cleanup_interval; // Did. Move to next interval.
                        }
+               }
 
-                       for (s = 1; s < MAXSESSION; s++)
-                       {
-                               // check for expired sessions
-                               if (session[s].die && session[s].die <= when)
-                               {
-                                       sessionkill(s, "Expired");
-                                       continue;
-                               }
+#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 */
+       }
 
-                               // 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;
-                               }
+               // Are we the master and shutting down??
+       if (config->cluster_iam_master)
+               cluster_heartbeat(); // Flush any queued changes..
 
-                               // 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);
+               // Ok. Notify everyone we're shutting down. If we're
+               // the master, this will force an election.
+       cluster_send_ping(0);
 
-                                       *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",
-                                                       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();
-                       }
-
-                       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));
-                       }
-
-                       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));
-               }
-       }
-}
+       //
+       // Important!!! We MUST not process any packets past this point!
+}
 
 // Init data structures
-void initdata(void)
+void initdata(int optdebug, char *optconfig)
 {
        int i;
 
-       _statistics = mmap(NULL, sizeof(struct Tstats), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (_statistics == MAP_FAILED)
+       if (!(_statistics = shared_malloc(sizeof(struct Tstats))))
        {
-               log(0, 0, 0, 0, "Error doing mmap for _statistics: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for _statistics: %s\n", strerror(errno));
                exit(1);
        }
-       config = mmap(NULL, sizeof(struct configt), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (config == MAP_FAILED)
+       if (!(config = shared_malloc(sizeof(struct configt))))
        {
-               log(0, 0, 0, 0, "Error doing mmap for configuration: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for configuration: %s\n", strerror(errno));
                exit(1);
        }
        memset(config, 0, sizeof(struct configt));
        time(&config->start_time);
-       strncpy(config->config_file, CONFIGFILE, sizeof(config->config_file) - 1);
-       tunnel = mmap(NULL, sizeof(tunnelt) * MAXTUNNEL, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (tunnel == MAP_FAILED)
+       strncpy(config->config_file, optconfig, strlen(optconfig));
+       config->debug = optdebug;
+       config->num_tbfs = MAXTBFS;
+       config->rl_rate = 28; // 28kbps
+
+       if (!(tunnel = shared_malloc(sizeof(tunnelt) * MAXTUNNEL)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for tunnels: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for tunnels: %s\n", strerror(errno));
                exit(1);
        }
-       session = mmap(NULL, sizeof(sessiont) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (session == MAP_FAILED)
+       if (!(session = shared_malloc(sizeof(sessiont) * MAXSESSION)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for sessions: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for sessions: %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)
+
+       if (!(sess_count = shared_malloc(sizeof(sessioncountt) * MAXSESSION)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for sessions_count: %s\n", strerror(errno));
                exit(1);
        }
-       ip_address_pool = mmap(NULL, sizeof(ippoolt) * MAXIPPOOL, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (ip_address_pool == MAP_FAILED)
+
+       if (!(radius = shared_malloc(sizeof(radiust) * MAXRADIUS)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for radius: %s\n", strerror(errno));
                exit(1);
        }
-#ifdef RINGBUFFER
-       ringbuffer = mmap(NULL, sizeof(struct Tringbuffer), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (ringbuffer == MAP_FAILED)
+
+       if (!(ip_address_pool = shared_malloc(sizeof(ippoolt) * MAXIPPOOL)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for ip_address_pool: %s\n", strerror(errno));
                exit(1);
        }
-       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
-#endif
 
-       cli_session_kill = mmap(NULL, sizeof(sessionidt) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (cli_session_kill == MAP_FAILED)
+#ifdef RINGBUFFER
+       if (!(ringbuffer = shared_malloc(sizeof(struct Tringbuffer))))
        {
-               log(0, 0, 0, 0, "Error doing mmap for cli session kill: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for ringbuffer: %s\n", strerror(errno));
                exit(1);
        }
-       memset(cli_session_kill, 0, sizeof(sessionidt) * MAXSESSION);
-       cli_tunnel_kill = mmap(NULL, sizeof(tunnelidt) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (cli_tunnel_kill == MAP_FAILED)
+       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
+#endif
+
+       if (!(cli_session_actions = shared_malloc(sizeof(struct cli_session_actions) * MAXSESSION)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for cli tunnel kill: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for cli session actions: %s\n", strerror(errno));
                exit(1);
        }
-       memset(cli_tunnel_kill, 0, sizeof(tunnelidt) * MAXSESSION);
+       memset(cli_session_actions, 0, sizeof(struct cli_session_actions) * MAXSESSION);
 
-       filter_buckets = mmap(NULL, sizeof(tbft) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (filter_buckets == MAP_FAILED)
+       if (!(cli_tunnel_actions = shared_malloc(sizeof(struct cli_tunnel_actions) * MAXSESSION)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for filter buckets: %s\n", strerror(errno));
+               log(0, 0, 0, 0, "Error doing malloc for cli tunnel actions: %s\n", strerror(errno));
                exit(1);
        }
-       memset(filter_buckets, 0, sizeof(tbft) * MAXSESSION);
+       memset(cli_tunnel_actions, 0, sizeof(struct cli_tunnel_actions) * 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);
+
+               // 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;
+               if ((p = strchr(hostname, '.'))) *p = 0;
        }
+
        _statistics->start_time = _statistics->last_reset = time(NULL);
+
+#ifdef BGP
+       if (!(bgp_peers = shared_malloc(sizeof(struct bgp_peer) * BGP_NUM_PEERS)))
+       {
+               log(0, 0, 0, 0, "Error doing malloc 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;
+       time_t best_time = time_now;
        char *u = session[s].user;
        char reuse = 0;
 
-#ifdef STAT_CALLS
-       STAT(call_assign_ip_address);
-#endif
-       for (i = 0; i < ip_pool_size; i++)
+
+       CSTAT(call_assign_ip_address);
+
+       for (i = 1; i < ip_pool_size; i++)
        {
                if (!ip_address_pool[i].address || ip_address_pool[i].assigned)
                        continue;
@@ -1991,10 +2598,11 @@ 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 +2613,7 @@ 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 +2622,155 @@ 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;
+
+
+       CSTAT(call_free_ip_address);
+
+}
+
+//
+// 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)!\n",
+                               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;
@@ -2057,8 +2781,8 @@ void initippool()
                        src = inet_addr(buf);
                        if (src == INADDR_NONE)
                        {
-                               log(0, 0, 0, 0, "Invalid address pool IP %s", buf);
-                               exit(-1);
+                               log(0, 0, 0, 0, "Invalid address pool IP %s\n", buf);
+                               exit(1);
                        }
                        // This entry is for a specific IP only
                        if (src != config->bind_address)
@@ -2070,60 +2794,46 @@ 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)
+       struct sockaddr_in snoop_addr = {0};
+       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_family = AF_INET;
+       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);
@@ -2131,102 +2841,95 @@ void snoop_send_packet(char *packet, u16 size)
 
 void dump_acct_info()
 {
-    char filename[1024];
-    char timestr[64];
-    time_t t = time(NULL);
-    int i;
-    FILE *f = NULL;
+       char filename[1024];
+       char timestr[64];
+       time_t t = time(NULL);
+       int i;
+       FILE *f = NULL;
 
-#ifdef STAT_CALLS
-       STAT(call_dump_acct_info);
-#endif
-    strftime(timestr, 64, "%Y%m%d%H%M%S", localtime(&t));
-    snprintf(filename, 1024, "%s/%s", config->accounting_dir, timestr);
 
-    for (i = 0; i < MAXSESSION; i++)
-    {
-       if (!session[i].opened || !session[i].cin || !session[i].cout || !*session[i].user || session[i].walled_garden)
-               continue;
-       if (!f)
+       CSTAT(call_dump_acct_info);
+
+       strftime(timestr, 64, "%Y%m%d%H%M%S", localtime(&t));
+       snprintf(filename, 1024, "%s/%s", config->accounting_dir, timestr);
+
+       for (i = 0; i < MAXSESSION; i++)
        {
-           time_t now = time(NULL);
-           if (!(f = fopen(filename, "w")))
-           {
-                   log(0, 0, 0, 0, "Can't write accounting info to %s: %s\n", filename, strerror(errno));
-                   return;
-           }
-           log(3, 0, 0, 0, "Dumping accounting information to %s\n", filename);
-           fprintf(f, "# dslwatch.pl dump file V1.01\n"
-               "# host: %s\n"
-               "# time: %ld\n"
-               "# uptime: %ld\n"
-               "# format: username ip qos uptxoctets downrxoctets\n",
-                   hostname,
-                   now,
-                   now - basetime);
-       }
-
-       log(4, 0, 0, 0, "Dumping accounting information for %s\n", session[i].user);
-       fprintf(f, "%s %s %d %lu %lu\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
-
-       session[i].pin = session[i].cin = 0;
-       session[i].pout = session[i].cout = 0;
-    }
-
-    if (f) fclose(f);
+               if (!session[i].opened || !session[i].ip || !(session[i].cin || session[i].cout) || !*session[i].user || session[i].walled_garden)
+                       continue;
+               if (!f)
+               {
+                       time_t now = time(NULL);
+                       if (!(f = fopen(filename, "w")))
+                       {
+                               log(0, 0, 0, 0, "Can't write accounting info to %s: %s\n", filename, strerror(errno));
+                               return ;
+                       }
+                       log(3, 0, 0, 0, "Dumping accounting information to %s\n", filename);
+                       fprintf(f, "# dslwatch.pl dump file V1.01\n"
+                               "# host: %s\n"
+                               "# time: %ld\n"
+                               "# uptime: %ld\n"
+                               "# format: username ip qos uptxoctets downrxoctets\n",
+                               hostname,
+                               now,
+                               now - basetime);
+               }
+
+               log(4, 0, 0, 0, "Dumping accounting information for %s\n", session[i].user);
+               fprintf(f, "%s %s %d %u %u\n",
+                       session[i].user,                                                // username
+                       inet_toa(htonl(session[i].ip)),                                 // ip
+                       (session[i].throttle_in || session[i].throttle_out) ? 2 : 1,    // qos
+                       (u32)session[i].cin,                                            // uptxoctets
+                       (u32)session[i].cout);                                          // downrxoctets
+
+               session[i].pin = session[i].cin = 0;
+               session[i].pout = session[i].cout = 0;
+       }
+
+       if (f)
+               fclose(f);
 }
 
 // Main program
 int main(int argc, char *argv[])
 {
        int o;
+       int optdebug = 0;
+       char *optconfig = CONFIGFILE;
 
        _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
-       initdata();
 
        // scan args
-       while ((o = getopt(argc, argv, "vc:h:a:")) >= 0)
+       while ((o = getopt(argc, argv, "dvc:h:")) >= 0)
        {
                switch (o)
                {
+                       case 'd':
+                               // Double fork to detach from terminal
+                               if (fork()) exit(0);
+                               if (fork()) exit(0);
+                               break;
                        case 'v':
-                               config->debug++;
+                               optdebug++;
                                break;
                        case 'c':
-                               strncpy(config->config_file, optarg, sizeof(config->config_file) - 1);
+                               optconfig = optarg;
                                break;
                        case 'h':
-                               strncpy(hostname, optarg, 999);
+                               snprintf(hostname, sizeof(hostname), "%s", optarg);
                                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;
                }
@@ -2234,46 +2937,97 @@ 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));
+       strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
+       signal(SIGALRM, sigalrm_handler);
+       siginterrupt(SIGALRM, 0);
 
        initiptables();
        initplugins();
-       init_cli();
+       initdata(optdebug, optconfig);
+
+       init_cli(hostname);
        read_config_file();
-       log(1, 0, 0, 0, "L2TPNS Version 1.1.0 - http://l2tpns.sourceforge.net/\n");
-       log(1, 0, 0, 0, "Licensed under the GPL\n");
+       init_tbf(config->num_tbfs);
 
-       /* 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, "L2TPNS version " VERSION "\n");
+       log(0, 0, 0, 0, "Copyright (c) 2003, 2004 Optus Internet Engineering\n");
+       log(0, 0, 0, 0, "Copyright (c) 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");
+       }
 
-       inittap();
-       log(1, 0, 0, 0, "Set up on interface %s\n", config->tapdevice);
+       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 */
+
+       inittun();
+       log(1, 0, 0, 0, "Set up on interface %s\n", config->tundevice);
 
        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);
+
+       // Prevent us from getting paged out
+       if (config->lock_pages)
+       {
+               if (!mlockall(MCL_CURRENT))
+                       log(1, 0, 0, 0, "Locking pages into memory\n");
+               else
+                       log(0, 0, 0, 0, "Can't lock pages: %s\n", strerror(errno));
+       }
 
        alarm(1);
 
@@ -2282,6 +3036,28 @@ 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();
+
+       // Remove the PID file if we wrote it
+       if (config->wrote_pid && *config->pid_file == '/')
+               unlink(config->pid_file);
+
+       /* kill CLI children */
+       signal(SIGTERM, SIG_IGN);
+       kill(0, SIGTERM);
        return 0;
 }
 
@@ -2301,7 +3077,7 @@ 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),
@@ -2318,7 +3094,7 @@ void sigalrm_handler(int junk)
 
        // Update the internal time counter
        time(&time_now);
-       strftime(time_now_string, 64, "%Y-%m-%d %H:%M:%S", localtime(&time_now));
+       strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
        alarm(1);
 
        {
@@ -2334,12 +3110,12 @@ 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");
@@ -2354,12 +3130,6 @@ 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++;
 }
 
@@ -2369,179 +3139,180 @@ 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;
-    int i;
-    ippoolt itmp;
-    FILE *f;
-    char magic[sizeof(DUMP_MAGIC)-1];
-    u32 buf[2];
+       struct stat sb;
+       int i;
+       ippoolt itmp;
+       FILE *f;
+       char magic[sizeof(DUMP_MAGIC) - 1];
+       u32 buf[2];
 
-    if (!config->save_state)
-       return;
+       if (!config->save_state)
+       {
+               unlink(STATEFILE);
+               return ;
+       }
 
-    // Ignore saved state if NOSTATEFILE exists
-    if (config->ignore_cluster_updates)
-    {
-       unlink(STATEFILE);
-       return;
-    }
+       if (stat(STATEFILE, &sb) < 0)
+       {
+               unlink(STATEFILE);
+               return ;
+       }
 
-    if (stat(STATEFILE, &sb) < 0)
-       return;
+       if (sb.st_mtime < (time(NULL) - 60))
+       {
+               log(0, 0, 0, 0, "State file is too old to read, ignoring\n");
+               unlink(STATEFILE);
+               return ;
+       }
 
-    if (sb.st_mtime < (time(NULL) - 60))
-    {
-       log(0, 0, 0, 0, "State file is too old to read, ignoring\n");
+       f = fopen(STATEFILE, "r");
        unlink(STATEFILE);
-       return;
-    }
-
-    f = fopen(STATEFILE, "r");
-    unlink(STATEFILE);
-
-    if (!f)
-    {
-       log(0, 0, 0, 0, "Can't read state file: %s\n", strerror(errno));
-       exit(1);
-    }
-
-    if (fread(magic, sizeof(magic), 1, f) != 1 || strncmp(magic, DUMP_MAGIC, sizeof(magic)))
-    {
-       log(0, 0, 0, 0, "Bad state file magic\n");
-       exit(1);
-    }
-
-    log(1, 0, 0, 0, "Reading state information\n");
-    if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] > MAXIPPOOL || buf[1] != sizeof(ippoolt))
-    {
-       log(0, 0, 0, 0, "Error/mismatch reading ip pool header from state file\n");
-       exit(1);
-    }
-
-    if (buf[0] > ip_pool_size)
-    {
-       log(0, 0, 0, 0, "ip pool has shrunk!  state = %d, current = %d\n", buf[0], ip_pool_size);
-       exit(1);
-    }
-
-    log(2, 0, 0, 0, "Loading %u ip addresses\n", buf[0]);
-    for (i = 0; i < buf[0]; i++)
-    {
-       if (fread(&itmp, sizeof(itmp), 1, f) != 1)
-       {
-           log(0, 0, 0, 0, "Error reading ip %d from state file: %s\n", i, strerror(errno));
-           exit(1);
-       }
-
-       if (itmp.address != ip_address_pool[i].address)
-       {
-           log(0, 0, 0, 0, "Mismatched ip %d from state file: pool may only be extended\n", i);
-           exit(1);
-       }
-
-       memcpy(&ip_address_pool[i], &itmp, sizeof(itmp));
-    }
-
-    if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXTUNNEL || buf[1] != sizeof(tunnelt))
-    {
-       log(0, 0, 0, 0, "Error/mismatch reading tunnel header from state file\n");
-       exit(1);
-    }
-
-    log(2, 0, 0, 0, "Loading %u tunnels\n", MAXTUNNEL);
-    if (fread(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL)
-    {
-       log(0, 0, 0, 0, "Error reading tunnel data from state file\n");
-       exit(1);
-    }
-
-    for (i = 0; i < MAXTUNNEL; i++)
-    {
-       tunnel[i].controlc = 0;
-       tunnel[i].controls = NULL;
-       tunnel[i].controle = NULL;
-       if (*tunnel[i].hostname)
-           log(3, 0, 0, 0, "Created tunnel for %s\n", tunnel[i].hostname);
-    }
-
-    if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXSESSION || buf[1] != sizeof(sessiont))
-    {
-       log(0, 0, 0, 0, "Error/mismatch reading session header from state file\n");
-       exit(1);
-    }
-
-    log(2, 0, 0, 0, "Loading %u sessions\n", MAXSESSION);
-    if (fread(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION)
-    {
-       log(0, 0, 0, 0, "Error reading session data from state file\n");
-       exit(1);
-    }
-
-    for (i = 0; i < MAXSESSION; i++)
-    {
-       session[i].tbf = 0;
-       session[i].throttle = 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);
-       }
-    }
-
-    fclose(f);
-    log(0, 0, 0, 0, "Loaded saved state information\n");
+
+       if (!f)
+       {
+               log(0, 0, 0, 0, "Can't read state file: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       if (fread(magic, sizeof(magic), 1, f) != 1 || strncmp(magic, DUMP_MAGIC, sizeof(magic)))
+       {
+               log(0, 0, 0, 0, "Bad state file magic\n");
+               exit(1);
+       }
+
+       log(1, 0, 0, 0, "Reading state information\n");
+       if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] > MAXIPPOOL || buf[1] != sizeof(ippoolt))
+       {
+               log(0, 0, 0, 0, "Error/mismatch reading ip pool header from state file\n");
+               exit(1);
+       }
+
+       if (buf[0] > ip_pool_size)
+       {
+               log(0, 0, 0, 0, "ip pool has shrunk!  state = %d, current = %d\n", buf[0], ip_pool_size);
+               exit(1);
+       }
+
+       log(2, 0, 0, 0, "Loading %u ip addresses\n", buf[0]);
+       for (i = 0; i < buf[0]; i++)
+       {
+               if (fread(&itmp, sizeof(itmp), 1, f) != 1)
+               {
+                       log(0, 0, 0, 0, "Error reading ip %d from state file: %s\n", i, strerror(errno));
+                       exit(1);
+               }
+
+               if (itmp.address != ip_address_pool[i].address)
+               {
+                       log(0, 0, 0, 0, "Mismatched ip %d from state file: pool may only be extended\n", i);
+                       exit(1);
+               }
+
+               memcpy(&ip_address_pool[i], &itmp, sizeof(itmp));
+       }
+
+       if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXTUNNEL || buf[1] != sizeof(tunnelt))
+       {
+               log(0, 0, 0, 0, "Error/mismatch reading tunnel header from state file\n");
+               exit(1);
+       }
+
+       log(2, 0, 0, 0, "Loading %u tunnels\n", MAXTUNNEL);
+       if (fread(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL)
+       {
+               log(0, 0, 0, 0, "Error reading tunnel data from state file\n");
+               exit(1);
+       }
+
+       for (i = 0; i < MAXTUNNEL; i++)
+       {
+               tunnel[i].controlc = 0;
+               tunnel[i].controls = NULL;
+               tunnel[i].controle = NULL;
+               if (*tunnel[i].hostname)
+                       log(3, 0, 0, 0, "Created tunnel for %s\n", tunnel[i].hostname);
+       }
+
+       if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXSESSION || buf[1] != sizeof(sessiont))
+       {
+               log(0, 0, 0, 0, "Error/mismatch reading session header from state file\n");
+               exit(1);
+       }
+
+       log(2, 0, 0, 0, "Loading %u sessions\n", MAXSESSION);
+       if (fread(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION)
+       {
+               log(0, 0, 0, 0, "Error reading session data from state file\n");
+               exit(1);
+       }
+
+       for (i = 0; i < MAXSESSION; i++)
+       {
+               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)
+                               sessionsetup(session[i].tunnel, i);
+               }
+       }
+
+       fclose(f);
+       log(0, 0, 0, 0, "Loaded saved state information\n");
 }
 
 void dump_state()
 {
-    FILE *f;
-    u32 buf[2];
+       FILE *f;
+       u32 buf[2];
 
-    if (!config->save_state)
-       return;
+       if (!config->save_state)
+               return;
 
-    do {
-       if (!(f = fopen(STATEFILE, "w")))
-           break;
+       do
+       {
+               if (!(f = fopen(STATEFILE, "w")))
+                       break;
 
-       log(1, 0, 0, 0, "Dumping state information\n");
+               log(1, 0, 0, 0, "Dumping state information\n");
 
-       if (fwrite(DUMP_MAGIC, sizeof(DUMP_MAGIC)-1, 1, f) != 1) break;
+               if (fwrite(DUMP_MAGIC, sizeof(DUMP_MAGIC) - 1, 1, f) != 1)
+                       break;
 
-       log(2, 0, 0, 0, "Dumping %u ip addresses\n", ip_pool_size);
-       buf[0] = ip_pool_size;
-       buf[1] = sizeof(ippoolt);
-       if (fwrite(buf, sizeof(buf), 1, f) != 1) break;
-       if (fwrite(ip_address_pool, sizeof(ippoolt), ip_pool_size, f) != ip_pool_size) break;
+               log(2, 0, 0, 0, "Dumping %u ip addresses\n", ip_pool_size);
+               buf[0] = ip_pool_size;
+               buf[1] = sizeof(ippoolt);
+               if (fwrite(buf, sizeof(buf), 1, f) != 1)
+                       break;
+               if (fwrite(ip_address_pool, sizeof(ippoolt), ip_pool_size, f) != ip_pool_size)
+                       break;
 
-       log(2, 0, 0, 0, "Dumping %u tunnels\n", MAXTUNNEL);
-       buf[0] = MAXTUNNEL;
-       buf[1] = sizeof(tunnelt);
-       if (fwrite(buf, sizeof(buf), 1, f) != 1) break;
-       if (fwrite(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL) break;
+               log(2, 0, 0, 0, "Dumping %u tunnels\n", MAXTUNNEL);
+               buf[0] = MAXTUNNEL;
+               buf[1] = sizeof(tunnelt);
+               if (fwrite(buf, sizeof(buf), 1, f) != 1)
+                       break;
+               if (fwrite(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL)
+                       break;
 
-       log(2, 0, 0, 0, "Dumping %u sessions\n", MAXSESSION);
-       buf[0] = MAXSESSION;
-       buf[1] = sizeof(sessiont);
-       if (fwrite(buf, sizeof(buf), 1, f) != 1) break;
-       if (fwrite(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION) break;
+               log(2, 0, 0, 0, "Dumping %u sessions\n", MAXSESSION);
+               buf[0] = MAXSESSION;
+               buf[1] = sizeof(sessiont);
+               if (fwrite(buf, sizeof(buf), 1, f) != 1)
+                       break;
+               if (fwrite(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION)
+                       break;
 
-       if (fclose(f) == 0) return; // OK
-    } while (0);
+               if (fclose(f) == 0)
+                       return ; // OK
+       }
+       while (0);
 
-    log(0, 0, 0, 0, "Can't write state information: %s\n", strerror(errno));
-    unlink(STATEFILE);
+       log(0, 0, 0, 0, "Can't write state information: %s\n", strerror(errno));
+       unlink(STATEFILE);
 }
 
 void build_chap_response(char *challenge, u8 id, u16 challenge_length, char **challenge_response)
@@ -2570,22 +3341,20 @@ void build_chap_response(char *challenge, u8 id, u16 challenge_length, char **ch
 
 static int facility_value(char *name)
 {
-    int i;
-    for (i = 0; facilitynames[i].c_name; i++)
-    {
-       if (strcmp(facilitynames[i].c_name, name) == 0)
-           return facilitynames[i].c_val;
-    }
-    return 0;
+       int i;
+       for (i = 0; facilitynames[i].c_name; i++)
+       {
+               if (strcmp(facilitynames[i].c_name, name) == 0)
+                       return facilitynames[i].c_val;
+       }
+       return 0;
 }
 
 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);
+       static int timeout = 0;
+       static int interval = 0;
 
        // Update logging
        closelog();
@@ -2597,9 +3366,18 @@ 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);
@@ -2610,15 +3388,6 @@ 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
        {
@@ -2630,13 +3399,29 @@ void update_config()
        // Update radius
        config->numradiusservers = 0;
        for (i = 0; i < MAXRADSERVER; i++)
-               if (config->radiusserver[i]) config->numradiusservers++;
+               if (config->radiusserver[i])
+               {
+                       config->numradiusservers++;
+                       // Set radius port: if not set, take the port from the
+                       // first radius server.  For the first radius server,
+                       // take the #defined default value from l2tpns.h
+
+                       // test twice, In case someone works with
+                       // a secondary radius server without defining
+                       // a primary one, this will work even then.
+                       if (i>0 && !config->radiusport[i])
+                               config->radiusport[i] = config->radiusport[i-1];
+                       if (!config->radiusport[i])
+                               config->radiusport[i] = RADPORT;
+               }
 
        if (!config->numradiusservers)
        {
                log(0, 0, 0, 0, "No RADIUS servers defined!\n");
        }
 
+       config->num_radfds = 2 << RADIUS_SHIFT;
+
        // Update plugins
        for (i = 0; i < MAXPLUGINS; i++)
        {
@@ -2654,6 +3439,54 @@ 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 = 10;
+       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);
+
+       if (!config->cluster_hb_interval)
+               config->cluster_hb_interval = PING_INTERVAL;    // Heartbeat every 0.5 seconds.
+
+       if (!config->cluster_hb_timeout)
+               config->cluster_hb_timeout = HB_TIMEOUT;        // 10 missed heartbeat triggers an election.
+
+       if (interval != config->cluster_hb_interval || timeout != config->cluster_hb_timeout)
+       {
+               // Paranoia:  cluster_check_master() treats 2 x interval + 1 sec as
+               // late, ensure we're sufficiently larger than that
+               int t = 4 * config->cluster_hb_interval + 11;
+
+               if (config->cluster_hb_timeout < t)
+               {
+                       log(0,0,0,0, "Heartbeat timeout %d too low, adjusting to %d\n", config->cluster_hb_timeout, t);
+                       config->cluster_hb_timeout = t;
+               }
+
+               // Push timing changes to the slaves immediately if we're the master
+               if (config->cluster_iam_master)
+                       cluster_heartbeat();
+
+               interval = config->cluster_hb_interval;
+               timeout = config->cluster_hb_timeout;
+       }
+
+       // Write PID file
+       if (*config->pid_file == '/' && !config->wrote_pid)
+       {
+               FILE *f;
+               if ((f = fopen(config->pid_file, "w")))
+               {
+                       fprintf(f, "%d\n", getpid());
+                       fclose(f);
+                       config->wrote_pid = 1;
+               }
+               else
+               {
+                       log(0, 0, 0, 0, "Can't write to PID file %s: %s\n", config->pid_file, strerror(errno));
+               }
+       }
+
        config->reload_config = 0;
 }
 
@@ -2662,7 +3495,8 @@ void read_config_file()
        FILE *f;
 
        if (!config->config_file) return;
-       if (!(f = fopen(config->config_file, "r"))) {
+       if (!(f = fopen(config->config_file, "r")))
+       {
                fprintf(stderr, "Can't open config file %s: %s\n", config->config_file, strerror(errno));
                return;
        }
@@ -2672,84 +3506,186 @@ 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;
-#ifdef STAT_CALLS
-       STAT(call_sessionsetup);
-#endif
+       int r;
+
+       CSTAT(call_sessionsetup);
+
        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");
+                       sessionshutdown(s, "No IP addresses available");
+                       return 0;
+               }
        }
 
+
        // 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].unique_id)
+       {
+               // did this session just finish radius?
+               log(3, session[s].ip, s, t, "Sending initial IPCP to client\n");
+               sendipcp(t, s);
+               session[s].unique_id = ++last_id;
+       }
 
+       // 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);
+       // Allocate TBFs if throttled
+       if (session[s].throttle_in || session[s].throttle_out)
+               throttle_session(s, session[s].throttle_in, session[s].throttle_out);
 
-       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.
+               {
+                       // Remove any routes if the IP has changed
+                       for (i = 0; i < MAXROUTE && session[s].route[i].ip; i++)
+                       {
+                               routeset(s, session[s].route[i].ip, session[s].route[i].mask, session[s].ip, 0);
+                               session[s].route[i].ip = 0;
+                       }
+
+                       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, remove 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);
+               }
+       }
+
+       // Update routed networks
+       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].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;
+
+       // TEMP: old session struct used a u32 to define the throttle
+       // speed for both up/down, new uses a u16 for each.  Deal with
+       // sessions from an old master for migration.
+       if (new->throttle_out == 0 && new->tbf_out)
+               new->throttle_out = new->throttle_in;
+
+       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)
 {
@@ -2775,28 +3711,33 @@ 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());
@@ -2813,70 +3754,66 @@ 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);
 }
@@ -2897,15 +3834,27 @@ 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;
        int l;
        struct param_control param = { buf, len, ntohl(addr->sin_addr.s_addr), ntohs(addr->sin_port), NULL, 0, 0 };
 
-       log(4, ntohl(addr->sin_addr.s_addr), 0, 0, "Received ");
-       if (log_stream)
+
+       if (log_stream && config->debug >= 4)
+       {
+               log(4, ntohl(addr->sin_addr.s_addr), 0, 0, "Received ");
                dump_packet(buf, log_stream);
+       }
 
        resp = calloc(1400, 1);
        l = new_packet(PKT_RESP_ERROR, resp);
@@ -2918,18 +3867,7 @@ 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)
        {
@@ -2976,6 +3914,8 @@ 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;
                }
        }
@@ -2983,3 +3923,191 @@ 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, i;
+       run_plugins(PLUGIN_BECOME_MASTER, NULL);
+
+       // running a bunch of iptables commands is slow and can cause
+       // the master to drop tunnels on takeover--kludge around the
+       // problem by forking for the moment (note: race)
+       if (!fork_and_close())
+       {
+               for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
+               {
+                       if (!session[s].tunnel) // Not an in-use session.
+                               continue;
+
+                       run_plugins(PLUGIN_NEW_SESSION_MASTER, &session[s]);
+               }
+               exit(0);
+       }
+
+       // add radius fds
+       for (i = 0; i < config->num_radfds; i++)
+       {
+               if (!radfds[i]) continue;
+               FD_SET(radfds[i], &readset);
+               if (radfds[i] > readset_n)
+                       readset_n = radfds[i];
+       }
+}
+
+int cmd_show_hist_idle(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       int s, i;
+       int count = 0;
+       int buckets[64];
+
+       if (CLI_HELP_REQUESTED)
+               return CLI_HELP_NO_ARGS;
+
+       time(&time_now);
+       for (i = 0; i < 64;++i) buckets[i] = 0;
+
+       for (s = 1; 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];
+
+       if (CLI_HELP_REQUESTED)
+               return CLI_HELP_NO_ARGS;
+
+       time(&time_now);
+       for (i = 0; i < 64;++i) buckets[i] = 0;
+
+       for (s = 1; 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;
+}
+
+/* Unhide an avp.
+ *
+ * This unencodes the AVP using the L2TP CHAP secret and the
+ * previously stored random vector. It replaces the hidden data with
+ * the cleartext data and returns the length of the cleartext data
+ * (including the AVP "header" of 6 bytes).
+ *
+ * Based on code from rp-l2tpd by Roaring Penguin Software Inc.
+ */
+int unhide_avp(u8 *avp, tunnelidt t, sessionidt s, u16 length)
+{
+       MD5_CTX ctx;
+       u8 *cursor;
+       u8 digest[16];
+       u8 working_vector[16];
+       uint16_t hidden_length;
+       u8 type[2];
+       size_t done, todo;
+       u8 *output;
+
+       // Find the AVP type.
+       type[0] = *(avp + 4);
+       type[1] = *(avp + 5);
+
+       // Line up with the hidden data
+       cursor = output = avp + 6;
+
+       // Compute initial pad
+       MD5Init(&ctx);
+       MD5Update(&ctx, type, 2);
+       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
+       MD5Update(&ctx, session[s].random_vector, session[s].random_vector_length);
+       MD5Final(digest, &ctx);
+
+       // Get hidden length
+       hidden_length = ((uint16_t) (digest[0] ^ cursor[0])) * 256 + (uint16_t) (digest[1] ^ cursor[1]);
+
+       // Keep these for later use
+       working_vector[0] = *cursor;
+       working_vector[1] = *(cursor + 1);
+       cursor += 2;
+
+       if (hidden_length > length - 8)
+       {
+               log(1, 0, s, t, "Hidden length %d too long in AVP of length %d\n", (int) hidden_length, (int) length);
+               return 0;
+       }
+
+       /* Decrypt remainder */
+       done = 2;
+       todo = hidden_length;
+       while (todo)
+       {
+               working_vector[done] = *cursor;
+               *output = digest[done] ^ *cursor;
+               ++output;
+               ++cursor;
+               --todo;
+               ++done;
+               if (done == 16 && todo)
+               {
+                       // Compute new digest
+                       done = 0;
+                       MD5Init(&ctx);
+                       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
+                       MD5Update(&ctx, &working_vector, 16);
+                       MD5Final(digest, &ctx);
+               }
+       }
+
+       return hidden_length + 6;
+}
+