remove defunct PLUGIN_PACKET_{RX,TX} constants
[l2tpns.git] / l2tpns.c
index ea21dda..fb83200 100644 (file)
--- a/l2tpns.c
+++ b/l2tpns.c
@@ -1,38 +1,45 @@
 // L2TP Network Server
 // Adrian Kennard 2002
-// (c) Copyrigth 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd)
+// Copyright (c) 2003, 2004, 2005, 2006 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 1.166 2006/05/16 06:46:37 bodea Exp $";
+
 #include <arpa/inet.h>
 #include <assert.h>
 #include <errno.h>
 #include <fcntl.h>
 #include <linux/if_tun.h>
+#define SYSLOG_NAMES
+#include <syslog.h>
 #include <malloc.h>
 #include <math.h>
 #include <net/route.h>
 #include <sys/mman.h>
 #include <netdb.h>
 #include <netinet/in.h>
-#include <signal.h>
+#include <netinet/ip6.h>
 #include <stdarg.h>
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
+#include <ctype.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>
-#ifdef HAVE_LIBCLI
+#include <sched.h>
+#include <sys/sysinfo.h>
 #include <libcli.h>
-#endif
+
 #include "md5.h"
 #include "l2tpns.h"
 #include "cluster.h"
 #include "constants.h"
 #include "control.h"
 #include "util.h"
+#include "tbf.h"
 
-ipt radiusserver[MAXRADSERVER]; // radius servers
-u8 numradiusservers = 0;  // how many radius servers
+#ifdef BGP
+#include "bgp.h"
+#endif
 
 // Globals
-char tapdevice[10] = "";    // tap device name
-int tapfd = -1;            // tap interface file handle
-int udpfd = -1;            // UDP file handle
-int controlfd = -1;        // Control signal handle
-int snoopfd = -1;          // UDP file handle for sending out intercept data
-int radfd = -1;            // RADIUS requests file handle
-int ifrfd = -1;            // File descriptor for routing, etc
-char debug = 0;                // debug leveL
-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
-char *radiussecret = 0;      // RADIUS secret
-char *l2tpsecret = 0;        // L2TP secret
-u32 sessionid = 0;         // session id for radius accounting
-char   *snoop_destination_host = NULL;
-u16    snoop_destination_port = 0;
-char *log_filename = NULL;
-char *config_file = CONFIGFILE;
-FILE *log_stream = NULL;
-unsigned long default_dns1 = 0, default_dns2 = 0;
-struct sockaddr_in snoop_addr = {0};
-extern unsigned long rl_rate;
-extern int cluster_sockfd;
-unsigned long last_sid = 0;
-int config_save_state = 0;
-int radius_accounting = 0;
-char *accounting_dir = NULL;
-uint32_t cluster_address = 0;
-uint32_t bind_address = INADDR_ANY;
-int handle_interface = 0;
-#ifdef HAVE_LIBCLI
-pid_t cli_pid = 0;
-int clifd = 0;
-sessionidt *cli_session_kill = NULL;
-tunnelidt *cli_tunnel_kill = NULL;
-#endif
-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_index = 0;
-unsigned int ip_pool_size = 0;
-time_t time_now;
-char time_now_string[64] = {0};
-char main_quit = 0;
-int dump_speed = 0;
-int target_uid = 500;
-char *_program_name = NULL;
+configt *config = NULL;                // all configuration
+int tunfd = -1;                        // tun interface file handle. (network device)
+int udpfd = -1;                        // UDP file handle
+int controlfd = -1;            // Control signal handle
+int clifd = -1;                        // Socket listening for CLI connections.
+int daefd = -1;                        // Socket listening for DAE connections.
+int snoopfd = -1;              // UDP file handle for sending out intercept data
+int *radfds = NULL;            // RADIUS requests file handles
+int ifrfd = -1;                        // File descriptor for routing, etc
+int ifr6fd = -1;               // File descriptor for IPv6 routing, etc
+int rand_fd = -1;              // Random data source
+int cluster_sockfd = -1;       // Intra-cluster communications socket.
+int epollfd = -1;              // event polling
+time_t basetime = 0;           // base clock
+char hostname[1000] = "";      // us.
+static int tunidx;             // ifr_ifindex of tun device
+static int syslog_log = 0;     // are we logging to syslog
+static FILE *log_stream = 0;   // file handle for direct logging (i.e. direct into file, not via syslog).
+uint32_t last_id = 0;          // Unique ID for radius accounting
+
+// calculated from config->l2tp_mtu
+uint16_t MRU = 0;              // PPP MRU
+uint16_t MSS = 0;              // TCP MSS
+
+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.
+struct ipv6radix {
+       int sess;
+       struct ipv6radix *branch;
+} ipv6_hash[256];              // Mapping from IPv6 address to session structures.
+
+// Traffic counters.
+static uint32_t udp_rx = 0, udp_rx_pkt = 0, udp_tx = 0;
+static uint32_t eth_rx = 0, eth_rx_pkt = 0;
+uint32_t eth_tx = 0;
+
+static uint32_t 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.
+static char time_now_string[64] = {0}; // Current time as a string.
+static int time_changed = 0;           // time_now changed
+char main_quit = 0;                    // True if we're in the process of exiting.
+static char main_reload = 0;           // Re-load pending
 linked_list *loaded_plugins;
 linked_list *plugins[MAX_PLUGIN_TYPES];
 
-char *plugin_functions[] = {
+#define membersize(STRUCT, MEMBER) sizeof(((STRUCT *)0)->MEMBER)
+#define CONFIG(NAME, MEMBER, TYPE) { NAME, offsetof(configt, MEMBER), membersize(configt, MEMBER), TYPE }
+
+config_descriptt config_values[] = {
+       CONFIG("debug", debug, INT),
+       CONFIG("log_file", log_filename, STRING),
+       CONFIG("pid_file", pid_file, STRING),
+       CONFIG("random_device", random_device, STRING),
+       CONFIG("l2tp_secret", l2tp_secret, STRING),
+       CONFIG("l2tp_mtu", l2tp_mtu, INT),
+       CONFIG("ppp_restart_time", ppp_restart_time, INT),
+       CONFIG("ppp_max_configure", ppp_max_configure, INT),
+       CONFIG("ppp_max_failure", ppp_max_failure, INT),
+       CONFIG("primary_dns", default_dns1, IPv4),
+       CONFIG("secondary_dns", default_dns2, IPv4),
+       CONFIG("primary_radius", radiusserver[0], IPv4),
+       CONFIG("secondary_radius", radiusserver[1], IPv4),
+       CONFIG("primary_radius_port", radiusport[0], SHORT),
+       CONFIG("secondary_radius_port", radiusport[1], SHORT),
+       CONFIG("radius_accounting", radius_accounting, BOOL),
+       CONFIG("radius_interim", radius_interim, INT),
+       CONFIG("radius_secret", radiussecret, STRING),
+       CONFIG("radius_authtypes", radius_authtypes_s, STRING),
+       CONFIG("radius_dae_port", radius_dae_port, SHORT),
+       CONFIG("allow_duplicate_users", allow_duplicate_users, BOOL),
+       CONFIG("guest_account", guest_user, STRING),
+       CONFIG("bind_address", bind_address, IPv4),
+       CONFIG("peer_address", peer_address, IPv4),
+       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("dump_speed", dump_speed, BOOL),
+       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("packet_limit", max_packets, INT),
+       CONFIG("cluster_address", cluster_address, IPv4),
+       CONFIG("cluster_interface", cluster_interface, STRING),
+       CONFIG("cluster_mcast_ttl", cluster_mcast_ttl, INT),
+       CONFIG("cluster_hb_interval", cluster_hb_interval, INT),
+       CONFIG("cluster_hb_timeout", cluster_hb_timeout, INT),
+       CONFIG("cluster_master_min_adv", cluster_master_min_adv, INT),
+       CONFIG("ipv6_prefix", ipv6_prefix, IPv6),
+       { NULL, 0, 0, 0 },
+};
+
+static char *plugin_functions[] = {
        NULL,
        "plugin_pre_auth",
        "plugin_post_auth",
        "plugin_packet_rx",
        "plugin_packet_tx",
        "plugin_timer",
-       "plugin_config",
        "plugin_new_session",
        "plugin_kill_session",
        "plugin_control",
        "plugin_radius_response",
+       "plugin_radius_reset",
+       "plugin_radius_account",
+       "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;
-tunnelidt tunnelfree;           // free list link heads
-sessionidt sessionfree = 0;
-u8 radiusfree;
-controlt *controlfree = 0;
+// Counters for shutdown sessions
+static sessiont shut_acct[8192];
+static sessionidt shut_acct_n = 0;
+
+tunnelt *tunnel = NULL;                        // Array of tunnel structures.
+bundlet *bundle = NULL;                        // Array of bundle structures.
+fragmentationt *frag = NULL;           // Array of fragmentation structures.
+sessiont *session = NULL;              // Array of session structures.
+sessionlocalt *sess_local = NULL;      // Array of local per-session counters.
+radiust *radius = NULL;                        // Array of radius structures.
+ippoolt *ip_address_pool = NULL;       // Array of dynamic IP addresses.
+ip_filtert *ip_filters = NULL;         // Array of named filters.
+static 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();
-
-// return internal time (10ths since run)
-clockt now(void)
+static void cache_ipmap(in_addr_t ip, int s);
+static void uncache_ipmap(in_addr_t ip);
+static void cache_ipv6map(struct in6_addr ip, int prefixlen, int s);
+static void free_ip_address(sessionidt s);
+static void dump_acct_info(int all);
+static void sighup_handler(int sig);
+static void shutdown_handler(int sig);
+static void sigchild_handler(int sig);
+static void build_chap_response(uint8_t *challenge, uint8_t id, uint16_t challenge_length, uint8_t **challenge_response);
+static void update_config(void);
+static void read_config_file(void);
+static void initplugins(void);
+static int add_plugin(char *plugin_name);
+static int remove_plugin(char *plugin_name);
+static void plugins_done(void);
+static void processcontrol(uint8_t *buf, int len, struct sockaddr_in *addr, int alen, struct in_addr *local);
+static tunnelidt new_tunnel(void);
+static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vector, size_t vec_len);
+static void bundleclear(bundleidt b);
+
+// on slaves, alow BGP to withdraw cleanly before exiting
+#define QUIT_DELAY     5
+
+// quit actions (master)
+#define QUIT_FAILOVER  1 // SIGTERM: exit when all control messages have been acked (for cluster failover)
+#define QUIT_SHUTDOWN  2 // SIGQUIT: shutdown sessions/tunnels, reject new connections
+
+// return internal time (10ths since process startup), set f if given
+// as a side-effect sets time_now, and time_changed
+static clockt now(double *f)
 {
        struct timeval t;
        gettimeofday(&t, 0);
+       if (f) *f = t.tv_sec + t.tv_usec / 1000000.0;
+       if (t.tv_sec != time_now)
+       {
+           time_now = t.tv_sec;
+           time_changed++;
+       }
        return (t.tv_sec - basetime) * 10 + t.tv_usec / 100000 + 1;
 }
 
 // work out a retry time based on try number
-clockt backoff(u8 try)
+// This is a straight bounded exponential backoff.
+// Maximum re-try time is 32 seconds. (2^5).
+clockt backoff(uint8_t try)
 {
        if (try > 5) try = 5;                  // max backoff
-       return now() + 10 * (1 << try);
+       return now(NULL) + 10 * (1 << try);
 }
 
-void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format, ...)
+
+//
+// Log a debug message.  Typically called via the LOG macro
+//
+void _log(int level, sessionidt s, tunnelidt t, const char *format, ...)
 {
+       static char message[65536] = {0};
        va_list ap;
 
 #ifdef RINGBUFFER
@@ -166,7 +257,6 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
                                ringbuffer->head = 0;
 
                ringbuffer->buffer[ringbuffer->tail].level = level;
-               ringbuffer->buffer[ringbuffer->tail].address = address;
                ringbuffer->buffer[ringbuffer->tail].session = s;
                ringbuffer->buffer[ringbuffer->tail].tunnel = t;
                va_start(ap, format);
@@ -175,146 +265,351 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
        }
 #endif
 
-       if (debug < level) return;
-
-       if (!log_stream && log_filename)
-       {
-               if ((log_stream = fopen(log_filename, "a")))
-               fseek(log_stream, 0, SEEK_END);
-               setbuf(log_stream, NULL);
-       }
-       if (!log_stream)
-       {
-               log_stream = stderr;
-               setbuf(log_stream, NULL);
-       }
+       if (config->debug < level) return;
 
        va_start(ap, format);
-       fprintf(log_stream, "%s %02d/%02d ", time_now_string, t, s);
-       vfprintf(log_stream, format, ap);
+       vsnprintf(message, sizeof(message), format, ap);
+
+       if (log_stream)
+               fprintf(log_stream, "%s %02d/%02d %s", time_now_string, t, s, message);
+       else if (syslog_log)
+               syslog(level + 2, "%02d/%02d %s", t, s, message); // We don't need LOG_EMERG or LOG_ALERT
+
        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 uint8_t *data, int maxsize)
 {
-       unsigned int i, j;
-       unsigned const char *d = (unsigned const char *)data;
+       int i, j;
+       const uint8_t *d = data;
 
-       if (debug < level) return;
-       log(level, address, s, t, "%s (%d bytes):\n", title, maxsize);
-       setvbuf(log_stream, NULL, _IOFBF, 16384);
-       for (i = 0; i < maxsize; )
+       if (config->debug < level) return;
+
+       // No support for _log_hex to syslog
+       if (log_stream)
        {
-               fprintf(log_stream, "%4X: ", i);
-               for (j = i; j < maxsize && j < (i + 16); j++)
+               _log(level, 0, 0, "%s (%d bytes):\n", title, maxsize);
+               setvbuf(log_stream, NULL, _IOFBF, 16384);
+
+               for (i = 0; i < maxsize; )
                {
-                       fprintf(log_stream, "%02X ", d[j]);
-                       if (j == i + 7)
-                               fputs(": ", log_stream);
+                       fprintf(log_stream, "%4X: ", i);
+                       for (j = i; j < maxsize && j < (i + 16); j++)
+                       {
+                               fprintf(log_stream, "%02X ", d[j]);
+                               if (j == i + 7)
+                                       fputs(": ", log_stream);
+                       }
+
+                       for (; j < i + 16; j++)
+                       {
+                               fputs("   ", log_stream);
+                               if (j == i + 7)
+                                       fputs(": ", log_stream);
+                       }
+
+                       fputs("  ", log_stream);
+                       for (j = i; j < maxsize && j < (i + 16); j++)
+                       {
+                               if (d[j] >= 0x20 && d[j] < 0x7f && d[j] != 0x20)
+                                       fputc(d[j], log_stream);
+                               else
+                                       fputc('.', log_stream);
+
+                               if (j == i + 7)
+                                       fputs("  ", log_stream);
+                       }
+
+                       i = j;
+                       fputs("\n", log_stream);
                }
 
-               for (; j < i + 16; j++)
+               fflush(log_stream);
+               setbuf(log_stream, NULL);
+       }
+}
+
+// update a counter, accumulating 2^32 wraps
+void increment_counter(uint32_t *counter, uint32_t *wrap, uint32_t delta)
+{
+       uint32_t new = *counter + delta;
+       if (new < *counter)
+               (*wrap)++;
+
+       *counter = new;
+}
+
+// initialise the random generator
+static void initrandom(char *source)
+{
+       static char path[sizeof(config->random_device)] = "*undefined*";
+
+       // reinitialise only if we are forced to do so or if the config has changed
+       if (source && !strncmp(path, source, sizeof(path)))
+               return;
+
+       // close previous source, if any
+       if (rand_fd >= 0)
+               close(rand_fd);
+
+       rand_fd = -1;
+
+       if (source)
+       {
+               // register changes
+               snprintf(path, sizeof(path), "%s", source);
+
+               if (*path == '/')
                {
-                       fputs("   ", log_stream);
-                       if (j == i + 7)
-                               fputs(": ", log_stream);
+                       rand_fd = open(path, O_RDONLY|O_NONBLOCK);
+                       if (rand_fd < 0)
+                               LOG(0, 0, 0, "Error opening the random device %s: %s\n",
+                                       path, strerror(errno));
                }
+       }
+}
+
+// fill buffer with random data
+void random_data(uint8_t *buf, int len)
+{
+       int n = 0;
 
-               fputs("  ", log_stream);
-               for (j = i; j < maxsize && j < (i + 16); j++)
+       CSTAT(random_data);
+       if (rand_fd >= 0)
+       {
+               n = read(rand_fd, buf, len);
+               if (n >= len) return;
+               if (n < 0)
                {
-                       if (d[j] >= 0x20 && d[j] < 0x7f && d[j] != 0x20)
-                               fputc(d[j], log_stream);
-                       else
-                               fputc('.', log_stream);
+                       if (errno != EAGAIN)
+                       {
+                               LOG(0, 0, 0, "Error reading from random source: %s\n",
+                                       strerror(errno));
 
-                       if (j == i + 7)
-                               fputs("  ", log_stream);
-               }
+                               // fall back to rand()
+                               initrandom(NULL);
+                       }
 
-               i = j;
-               fputs("\n", log_stream);
+                       n = 0;
+               }
        }
-       fflush(log_stream);
-       setbuf(log_stream, NULL);
-}
 
+       // append missing data
+       while (n < len)
+               // not using the low order bits from the prng stream
+               buf[n++] = (rand() >> 4) & 0xff;
+}
 
 // Add a route
-void routeset(ipt ip, ipt mask, ipt gw, u8 add)
+//
+// This adds it to the routing table, advertises it
+// via BGP if enabled, and stuffs it into the
+// 'sessionbyip' cache.
+//
+// 'ip' and 'mask' must be in _host_ order.
+//
+static void routeset(sessionidt s, in_addr_t ip, in_addr_t mask, in_addr_t gw, int 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 = 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);
+       *(uint32_t *) & (((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);
+       *(uint32_t *) & (((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 ? : 0xFFFFFFF);
+       *(uint32_t *) & (((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, s, 0, "Route %s %s/%s%s%s\n", add ? "add" : "del",
+           fmtaddr(htonl(ip), 0), fmtaddr(htonl(mask), 1),
+           gw ? " via" : "", gw ? fmtaddr(htonl(gw), 2) : "");
+
+       if (ioctl(ifrfd, add ? SIOCADDRT : SIOCDELRT, (void *) &r) < 0)
+               LOG(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);
+       }
+}
+
+void route6set(sessionidt s, struct in6_addr ip, int prefixlen, int add)
+{
+       struct in6_rtmsg rt;
+       char ipv6addr[INET6_ADDRSTRLEN];
+
+       if (ifr6fd < 0)
+       {
+               LOG(0, 0, 0, "Asked to set IPv6 route, but IPv6 not setup.\n");
+               return;
+       }
+
+       memset(&rt, 0, sizeof(rt));
+
+       memcpy(&rt.rtmsg_dst, &ip, sizeof(struct in6_addr));
+       rt.rtmsg_dst_len = prefixlen;
+       rt.rtmsg_metric = 1;
+       rt.rtmsg_flags = RTF_UP;
+       rt.rtmsg_ifindex = tunidx;
+
+       LOG(1, 0, 0, "Route %s %s/%d\n",
+           add ? "add" : "del",
+           inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN),
+           prefixlen);
+
+       if (ioctl(ifr6fd, add ? SIOCADDRT : SIOCDELRT, (void *) &rt) < 0)
+               LOG(0, 0, 0, "route6set() error in ioctl: %s\n",
+                               strerror(errno));
+
+       // FIXME: need to add BGP routing (RFC2858)
+
+       if (s)
+       {
+               if (!add)       // Are we deleting a route?
+                       s = 0;  // Caching the session as '0' is the same as uncaching.
+
+               cache_ipv6map(ip, prefixlen, s);
+       }
+       
+       return;
 }
 
-// Set up TAP interface
-void inittap(void)
+// defined in linux/ipv6.h, but tricky to include from user-space
+// TODO: move routing to use netlink rather than ioctl
+struct in6_ifreq {
+       struct in6_addr ifr6_addr;
+       __u32 ifr6_prefixlen;
+       unsigned int ifr6_ifindex;
+};
+
+//
+// Set up TUN interface
+static void inittun(void)
 {
        struct ifreq ifr;
+       struct in6_ifreq ifr6;
        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, "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, "Can't set tun interface: %s\n", strerror(errno));
+               exit(1);
        }
-       assert(strlen(ifr.ifr_name) < sizeof(tapdevice));
-       strcpy(tapdevice, ifr.ifr_name);
+       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 ? 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, "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)
+       {
+               LOG(0, 0, 0, "Error setting tun queue length: %s\n", strerror(errno));
+               exit(1);
+       }
+       /* set MTU to modem MRU */
+       ifr.ifr_mtu = MRU;
+       if (ioctl(ifrfd, SIOCSIFMTU, (void *) &ifr) < 0)
        {
-               perror("set tap flags");
-               exit( -1);
+               LOG(0, 0, 0, "Error setting tun MTU: %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, "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, "Error getting tun ifindex: %s\n", strerror(errno));
+               exit(1);
+       }
+       tunidx = ifr.ifr_ifindex;
+
+       // Only setup IPv6 on the tun device if we have a configured prefix
+       if (config->ipv6_prefix.s6_addr[0]) {
+               ifr6fd = socket(PF_INET6, SOCK_DGRAM, 0);
+
+               // Link local address is FE80::1
+               memset(&ifr6.ifr6_addr, 0, sizeof(ifr6.ifr6_addr));
+               ifr6.ifr6_addr.s6_addr[0] = 0xFE;
+               ifr6.ifr6_addr.s6_addr[1] = 0x80;
+               ifr6.ifr6_addr.s6_addr[15] = 1;
+               ifr6.ifr6_prefixlen = 64;
+               ifr6.ifr6_ifindex = ifr.ifr_ifindex;
+               if (ioctl(ifr6fd, SIOCSIFADDR, (void *) &ifr6) < 0)
+               {
+                       LOG(0, 0, 0, "Error setting tun IPv6 link local address:"
+                               " %s\n", strerror(errno));
+               }
+
+               // Global address is prefix::1
+               memset(&ifr6.ifr6_addr, 0, sizeof(ifr6.ifr6_addr));
+               ifr6.ifr6_addr = config->ipv6_prefix;
+               ifr6.ifr6_addr.s6_addr[15] = 1;
+               ifr6.ifr6_prefixlen = 64;
+               ifr6.ifr6_ifindex = ifr.ifr_ifindex;
+               if (ioctl(ifr6fd, SIOCSIFADDR, (void *) &ifr6) < 0)
+               {
+                       LOG(0, 0, 0, "Error setting tun IPv6 global address: %s\n",
+                               strerror(errno));
+               }
        }
-       tapidx = ifr.ifr_ifindex;
 }
 
-// set up UDP port
-void initudp(void)
+// set up UDP ports
+static void initudp(void)
 {
        int on = 1;
        struct sockaddr_in addr;
@@ -323,382 +618,890 @@ void initudp(void)
        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
        addr.sin_port = htons(L2TPPORT);
-       addr.sin_addr.s_addr = bind_address;
-       udpfd = socket(AF_INET, SOCK_DGRAM, UDP);
+       addr.sin_addr.s_addr = config->bind_address;
+       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("bind");
-               exit( -1);
+               LOG(0, 0, 0, "Error in UDP bind: %s\n", strerror(errno));
+               exit(1);
        }
-       snoopfd = socket(AF_INET, SOCK_DGRAM, UDP);
 
        // Control
        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
-       addr.sin_port = htons(1702);
-       controlfd = socket(AF_INET, SOCK_DGRAM, 17);
+       addr.sin_port = htons(NSCTL_PORT);
+       controlfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
        setsockopt(controlfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
+       setsockopt(controlfd, SOL_IP, IP_PKTINFO, &on, sizeof(on)); // recvfromto
        if (bind(controlfd, (void *) &addr, sizeof(addr)) < 0)
        {
-               perror("bind");
-               exit(-1);
+               LOG(0, 0, 0, "Error in control bind: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       // Dynamic Authorization Extensions to RADIUS
+       memset(&addr, 0, sizeof(addr));
+       addr.sin_family = AF_INET;
+       addr.sin_port = htons(config->radius_dae_port);
+       daefd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
+       setsockopt(daefd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
+       setsockopt(daefd, SOL_IP, IP_PKTINFO, &on, sizeof(on)); // recvfromto
+       if (bind(daefd, (void *) &addr, sizeof(addr)) < 0)
+       {
+               LOG(0, 0, 0, "Error in DAE bind: %s\n", strerror(errno));
+               exit(1);
        }
+
+       // Intercept
+       snoopfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
 }
 
-// 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.
+//
+
+static int lookup_ipmap(in_addr_t ip)
 {
-       unsigned char *a = (char *)&ip;
-       char **d = (char **) ip_hash;
+       uint8_t *a = (uint8_t *) &ip;
+       uint8_t **d = (uint8_t **) ip_hash;
 
-#ifdef STAT_CALLS
-       STAT(call_sessionbyip);
-#endif
+       if (!(d = (uint8_t **) d[(size_t) *a++])) return 0;
+       if (!(d = (uint8_t **) d[(size_t) *a++])) return 0;
+       if (!(d = (uint8_t **) d[(size_t) *a++])) return 0;
+
+       return (int) (intptr_t) d[(size_t) *a];
+}
+
+static int lookup_ipv6map(struct in6_addr ip)
+{
+       struct ipv6radix *curnode;
+       int i;
+       int s;
+       char ipv6addr[INET6_ADDRSTRLEN];
+
+       curnode = &ipv6_hash[ip.s6_addr[0]];
+       i = 1;
+       s = curnode->sess;
+
+       while (s == 0 && i < 15 && curnode->branch != NULL)
+       {
+               curnode = &curnode->branch[ip.s6_addr[i]];
+               s = curnode->sess;
+               i++;
+       }
+
+       LOG(4, s, session[s].tunnel, "Looking up address %s and got %d\n",
+                       inet_ntop(AF_INET6, &ip, ipv6addr,
+                               INET6_ADDRSTRLEN),
+                       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 s;
+}
+
+sessionidt sessionbyip(in_addr_t ip)
+{
+       int s = lookup_ipmap(ip);
+       CSTAT(sessionbyip);
+
+       if (s > 0 && s < MAXSESSION && session[s].opened)
+               return (sessionidt) s;
+
+       return 0;
+}
+
+sessionidt sessionbyipv6(struct in6_addr ip)
+{
+       int s;
+       CSTAT(sessionbyipv6);
+
+       if (!memcmp(&config->ipv6_prefix, &ip, 8) ||
+               (ip.s6_addr[0] == 0xFE &&
+                ip.s6_addr[1] == 0x80 &&
+                ip.s6_addr16[1] == 0 &&
+                ip.s6_addr16[2] == 0 &&
+                ip.s6_addr16[3] == 0)) {
+               s = lookup_ipmap(*(in_addr_t *) &ip.s6_addr[8]);
+       } else {
+               s = lookup_ipv6map(ip);
+       }
+
+       if (s > 0 && s < MAXSESSION && session[s].opened)
+               return s;
 
-       return (ipt) d[(size_t) *a];
+       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(in_addr_t ip, int s)
 {
-       unsigned char *a = (char *) &ip;
-       char **d = (char **) ip_hash;
+       in_addr_t nip = htonl(ip);      // MUST be in network order. I.e. MSB must in be ((char *) (&ip))[0]
+       uint8_t *a = (uint8_t *) &nip;
+       uint8_t **d = (uint8_t **) ip_hash;
        int i;
 
        for (i = 0; i < 3; i++)
        {
                if (!d[(size_t) a[i]])
                {
-                       if (!(d[(size_t) a[i]] = calloc(256, sizeof (void *))))
+                       if (!(d[(size_t) a[i]] = calloc(256, sizeof(void *))))
                                return;
                }
 
-               d = (char **) d[(size_t) a[i]];
+               d = (uint8_t **) 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);
+       d[(size_t) a[3]] = (uint8_t *) (intptr_t) s;
+
+       if (s > 0)
+               LOG(4, s, session[s].tunnel, "Caching ip address %s\n", fmtaddr(nip, 0));
+
+       else if (s == 0)
+               LOG(4, 0, 0, "Un-caching ip address %s\n", fmtaddr(nip, 0));
+       // else a map to an ip pool index.
 }
 
-void uncache_sessionid(ipt ip)
+static void uncache_ipmap(in_addr_t ip)
+{
+       cache_ipmap(ip, 0);     // Assign it to the NULL session.
+}
+
+static void cache_ipv6map(struct in6_addr ip, int prefixlen, int s)
 {
-       unsigned char *a = (char *) &ip;
-       char **d = (char **) ip_hash;
        int i;
+       int bytes;
+       struct ipv6radix *curnode;
+       char ipv6addr[INET6_ADDRSTRLEN];
 
-       for (i = 0; i < 3; i++)
+       curnode = &ipv6_hash[ip.s6_addr[0]];
+
+       bytes = prefixlen >> 3;
+       i = 1;
+       while (i < bytes) {
+               if (curnode->branch == NULL)
+               {
+                       if (!(curnode->branch = calloc(256,
+                                       sizeof (struct ipv6radix))))
+                               return;
+               }
+               curnode = &curnode->branch[ip.s6_addr[i]];
+               i++;
+       }
+
+       curnode->sess = s;
+
+       if (s > 0)
+               LOG(4, s, session[s].tunnel, "Caching ip address %s/%d\n",
+                               inet_ntop(AF_INET6, &ip, ipv6addr, 
+                                       INET6_ADDRSTRLEN),
+                               prefixlen);
+       else if (s == 0)
+               LOG(4, 0, 0, "Un-caching ip address %s/%d\n",
+                               inet_ntop(AF_INET6, &ip, ipv6addr, 
+                                       INET6_ADDRSTRLEN),
+                               prefixlen);
+}
+
+//
+// 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) (intptr_t) 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'd users aren't authenticated, so the username is
+// 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(sessionbyuser);
 
-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));
-    strcpy(ifr.ifr_name, "eth0");
-    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);
-}
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
+       {
+               if (!session[s].opened)
+                       continue;
 
-// Find session by username, 0 for not found
-sessiont *sessiontbysessionidt(sessionidt s)
-{
-       if (!s || s > MAXSESSION) return NULL;
-       return &session[s];
-}
+               if (session[s].walled_garden)
+                       continue;               // Skip walled garden users.
 
-sessionidt sessionidtbysessiont(sessiont *s)
-{
-       sessionidt val = s-session;
-       if (s < session || val > MAXSESSION) return 0;
-       return val;
-}
+               if (!strncmp(session[s].user, username, 128))
+                       return s;
 
-// send gratuitous ARP to set ARP table for newly allocated IP
-void sessionsendarp(sessionidt s)
-{
-        unsigned char mac[6];
-#ifdef STAT_CALLS
-       STAT(call_sendarp);
-#endif
-       *(u16 *) (mac + 0) = htons(tapmac[0]); // set source address
-       *(u16 *) (mac + 2) = htons(tapmac[1]);
-       *(u16 *) (mac + 4) = htons(tapmac[2]);
-       sendarp(tapidx, mac, session[s].ip);
-       STAT(arp_sent);
+       }
+       return 0;       // Not found.
 }
 
-// Handle ARP requests
-void processarp(u8 * buf, int len)
+void send_garp(in_addr_t ip)
 {
-       ipt ip;
-       sessionidt s;
+       int s;
+       struct ifreq ifr;
+       uint8_t mac[6];
 
-#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))
+       s = socket(PF_INET, SOCK_DGRAM, 0);
+       if (s < 0)
        {
-               log(0, 0, 0, 0, "Unexpected ARP prot type %04X\n", ntohs(*(u16 *) (buf + 20)));
-               STAT(arp_errors);
+               LOG(0, 0, 0, "Error creating socket for GARP: %s\n", strerror(errno));
                return;
        }
-       if (buf[22] != 6)
-       {
-               log(0, 0, 0, 0, "Unexpected ARP hard len %d\n", buf[22]);
-               STAT(arp_errors);
-               return;
-       }
-       if (buf[23] != 4)
+       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, "Unexpected ARP prot len %d\n", buf[23]);
-               STAT(arp_errors);
+               LOG(0, 0, 0, "Error getting eth0 hardware address for GARP: %s\n", strerror(errno));
+               close(s);
                return;
        }
-       if (*(u16 *) (buf + 24) != htons(0x0001))
+       memcpy(mac, &ifr.ifr_hwaddr.sa_data, 6*sizeof(char));
+       if (ioctl(s, SIOCGIFINDEX, &ifr) < 0)
        {
-               log(0, 0, 0, 0, "Unexpected ARP op %04X\n", ntohs(*(u16 *) (buf + 24)));
-               STAT(arp_errors);
+               LOG(0, 0, 0, "Error getting eth0 interface index for GARP: %s\n", strerror(errno));
+               close(s);
                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);
-       }
+       close(s);
+       sendarp(ifr.ifr_ifindex, mac, ip);
+}
+
+static sessiont *sessiontbysessionidt(sessionidt s)
+{
+       if (!s || s >= MAXSESSION) return NULL;
+       return &session[s];
+}
+
+static sessionidt sessionidtbysessiont(sessiont *s)
+{
+       sessionidt val = s-session;
+       if (s < session || val >= MAXSESSION) return 0;
+       return val;
 }
 
 // actually send a control message for a specific tunnel
-void tunnelsend(u8 * buf, u16 l, tunnelidt t)
+void tunnelsend(uint8_t * buf, uint16_t l, tunnelidt t)
 {
        struct sockaddr_in addr;
 
-#ifdef STAT_CALLS
-       STAT(call_tunnelsend);
-#endif
+       CSTAT(tunnelsend);
+
+       if (!t)
+       {
+               LOG(0, 0, t, "tunnelsend called with 0 as tunnel id\n");
+               STAT(tunnel_tx_errors);
+               return;
+       }
+
        if (!tunnel[t].ip)
        {
-               log(1, 0, 0, t, "Error sending data out tunnel: no remote endpoint (tunnel not set up)\n");
+               LOG(1, 0, t, "Error sending data out tunnel: no remote endpoint (tunnel not set up)\n");
                STAT(tunnel_tx_errors);
                return;
        }
+
        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
-       *(u32 *) & addr.sin_addr = htonl(tunnel[t].ip);
+       *(uint32_t *) & addr.sin_addr = htonl(tunnel[t].ip);
        addr.sin_port = htons(tunnel[t].port);
 
        // sequence expected, if sequence in message
-       if (*buf & 0x08) *(u16 *) (buf + ((*buf & 0x40) ? 10 : 8)) = htons(tunnel[t].nr);
+       if (*buf & 0x08) *(uint16_t *) (buf + ((*buf & 0x40) ? 10 : 8)) = htons(tunnel[t].nr);
 
        // If this is a control message, deal with retries
        if (*buf & 0x80)
        {
                tunnel[t].last = time_now; // control message sent
                tunnel[t].retry = backoff(tunnel[t].try); // when to resend
-               if (tunnel[t].try > 1)
+               if (tunnel[t].try)
                {
                        STAT(tunnel_retries);
-                       log(3, tunnel[t].ip, 0, t, "Control message resend try %d\n", tunnel[t].try);
+                       LOG(3, 0, t, "Control message resend try %d\n", tunnel[t].try);
                }
        }
 
        if (sendto(udpfd, buf, l, 0, (void *) &addr, sizeof(addr)) < 0)
        {
-               log(0, tunnel[t].ip, ntohs((*(u16 *) (buf + 6))), t, "Error sending data out tunnel: %s (udpfd=%d, buf=%p, len=%d, dest=%s)\n",
+               LOG(0, ntohs((*(uint16_t *) (buf + 6))), t, "Error sending data out tunnel: %s (udpfd=%d, buf=%p, len=%d, dest=%s)\n",
                                strerror(errno), udpfd, buf, l, inet_ntoa(addr.sin_addr));
                STAT(tunnel_tx_errors);
                return;
        }
 
-       log_hex(5, "Send Tunnel Data", buf, l);
+       LOG_HEX(5, "Send Tunnel Data", buf, l);
        STAT(tunnel_tx_packets);
        INC_STAT(tunnel_tx_bytes, l);
 }
 
-// process outgoing (to tunnel) IP
-void processipout(u8 * buf, int len)
+//
+// Tiny helper function to write data to
+// the 'tun' device.
+//
+int tun_write(uint8_t * data, int size)
 {
-       sessionidt s;
-       sessiont *sp;
-       tunnelidt t;
-       ipt ip;
-       u8 b[MAXETHER];
-#ifdef STAT_CALLS
-       STAT(call_processipout);
-#endif
-       if (len < 38)
-       {
-               log(1, 0, 0, 0, "Short IP, %d bytes\n", len);
-               STAT(tunnel_tx_errors);
+       return write(tunfd, data, size);
+}
+
+// adjust tcp mss to avoid fragmentation (called only for tcp packets with syn set)
+void adjust_tcp_mss(sessionidt s, tunnelidt t, uint8_t *buf, int len, uint8_t *tcp)
+{
+       int d = (tcp[12] >> 4) * 4;
+       uint8_t *mss = 0;
+       uint8_t *opts;
+       uint8_t *data;
+       uint16_t orig;
+       uint32_t sum;
+
+       if ((tcp[13] & 0x3f) & ~(TCP_FLAG_SYN|TCP_FLAG_ACK)) // only want SYN and SYN,ACK
                return;
-       }
 
-       // Skip the tun header
-       buf += 4;
+       if (tcp + d > buf + len) // short?
+               return;
+
+       opts = tcp + 20;
+       data = tcp + d;
+
+       while (opts < data)
+       {
+               if (*opts == 2 && opts[1] == 4) // mss option (2), length 4
+               {
+                       mss = opts + 2;
+                       if (mss + 2 > data) return; // short?
+                       break;
+               }
+
+               if (*opts == 0) return; // end of options
+               if (*opts == 1 || !opts[1]) // no op (one byte), or no length (prevent loop)
+                       opts++;
+               else
+                       opts += opts[1]; // skip over option
+       }
+
+       if (!mss) return; // not found
+       orig = ntohs(*(uint16_t *) mss);
+
+       if (orig <= MSS) return; // mss OK
+
+       LOG(5, s, t, "TCP: %s:%u -> %s:%u SYN%s: adjusted mss from %u to %u\n",
+               fmtaddr(*(in_addr_t *) (buf + 12), 0), ntohs(*(uint16_t *) tcp),
+               fmtaddr(*(in_addr_t *) (buf + 16), 1), ntohs(*(uint16_t *) (tcp + 2)),
+               (tcp[13] & TCP_FLAG_ACK) ? ",ACK" : "", orig, MSS);
+
+       // set mss
+       *(int16_t *) mss = htons(MSS);
+
+       // adjust checksum (see rfc1141)
+       sum = orig + (~MSS & 0xffff);
+       sum += ntohs(*(uint16_t *) (tcp + 16));
+       sum = (sum & 0xffff) + (sum >> 16);
+       *(uint16_t *) (tcp + 16) = htons(sum + (sum >> 16));
+}
+
+void processmpframe(sessionidt s, tunnelidt t, uint8_t *p, uint16_t l, uint8_t extra)
+{
+       uint16_t proto;
+       if (extra) {
+               // Skip the four extra bytes
+               p += 4;
+               l -= 4;
+       }
+
+       // Process this frame
+       if (*p & 1)
+       {
+               proto = *p++;
+               l--;
+       }
+       else
+       {
+               proto = ntohs(*(uint16_t *) p);
+               p += 2;
+               l -= 2;
+       }
+       if (proto == PPPIP)
+       {
+               if (session[s].die)
+               {
+                       LOG(4, s, t, "MPPP: Session %u is closing.  Don't process PPP packets\n", s);
+                       return;              // closing session, PPP not processed
+               }
+               session[s].last_packet = time_now;
+               processipin(s, t, p, l);
+       }
+       else if (proto == PPPIPV6 && config->ipv6_prefix.s6_addr[0])
+       {
+               if (session[s].die)
+               {
+                       LOG(4, s, t, "MPPP: Session %u is closing.  Don't process PPP packets\n", s);
+                       return;              // closing session, PPP not processed
+               }
+
+               session[s].last_packet = time_now;
+               processipv6in(s, t, p, l);
+       }
+       else
+       {
+               LOG(2, s, t, "MPPP: Unsupported MP protocol 0x%04X received\n",proto);
+       }
+}
+
+// process outgoing (to tunnel) IP
+//
+static void processipout(uint8_t *buf, int len)
+{
+       sessionidt s;
+       sessiont *sp;
+       tunnelidt t;
+       in_addr_t ip;
+
+       uint8_t *data = buf;    // Keep a copy of the originals.
+       int size = len;
+
+       uint8_t b1[MAXETHER + 20];
+       uint8_t b2[MAXETHER + 20];
+
+       CSTAT(processipout);
+
+       if (len < MIN_IP_SIZE)
+       {
+               LOG(1, 0, 0, "Short IP, %d bytes\n", len);
+               STAT(tun_rx_errors);
+               return;
+       }
+       if (len >= MAXETHER)
+       {
+               LOG(1, 0, 0, "Oversize IP packet %d bytes\n", len);
+               STAT(tun_rx_errors);
+               return;
+       }
+
+       // Skip the tun header
+       buf += 4;
        len -= 4;
 
        // Got an IP header now
-       if (*(u8 *)(buf) >> 4 != 4)
+       if (*(uint8_t *)(buf) >> 4 != 4)
        {
-               log(1, 0, 0, 0, "IP: Don't understand anything except IPv4\n");
+               LOG(1, 0, 0, "IP: Don't understand anything except IPv4\n");
                return;
        }
 
-       ip = *(u32 *)(buf + 16);
+       ip = *(uint32_t *)(buf + 16);
        if (!(s = sessionbyip(ip)))
        {
-//             log(4, 0, 0, 0, "IP: Can't find session for IP %s\n", inet_toa(ip));
+               // 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, "IP: Sending ICMP host unreachable to %s\n", fmtaddr(*(in_addr_t *)(buf + 12), 0));
+                       host_unreachable(*(in_addr_t *)(buf + 12), *(uint16_t *)(buf + 4),
+                               config->bind_address ? config->bind_address : my_address, buf, len);
+               }
+               return;
+       }
+       t = session[s].tunnel;
+       sp = &session[s];
+
+       // DoS prevention: enforce a maximum number of packets per 0.1s for a session
+       if (config->max_packets > 0)
+       {
+               if (sess_local[s].last_packet_out == TIME)
+               {
+                       int max = config->max_packets;
+
+                       // All packets for throttled sessions are handled by the
+                       // master, so further limit by using the throttle rate.
+                       // A bit of a kludge, since throttle rate is in kbps,
+                       // but should still be generous given our average DSL
+                       // packet size is 200 bytes: a limit of 28kbps equates
+                       // to around 180 packets per second.
+                       if (!config->cluster_iam_master && sp->throttle_out && sp->throttle_out < max)
+                               max = sp->throttle_out;
+
+                       if (++sess_local[s].packets_out > max)
+                       {
+                               sess_local[s].packets_dropped++;
+                               return;
+                       }
+               }
+               else
+               {
+                       if (sess_local[s].packets_dropped)
+                       {
+                               INC_STAT(tun_rx_dropped, sess_local[s].packets_dropped);
+                               LOG(3, s, t, "Dropped %u/%u packets to %s for %suser %s\n",
+                                       sess_local[s].packets_dropped, sess_local[s].packets_out,
+                                       fmtaddr(ip, 0), sp->throttle_out ? "throttled " : "",
+                                       sp->user);
+                       }
+
+                       sess_local[s].last_packet_out = TIME;
+                       sess_local[s].packets_out = 1;
+                       sess_local[s].packets_dropped = 0;
+               }
+       }
+
+       // run access-list if any
+       if (session[s].filter_out && !ip_filter(buf, len, session[s].filter_out - 1))
+               return;
+
+       // adjust MSS on SYN and SYN,ACK packets with options
+       if ((ntohs(*(uint16_t *) (buf + 6)) & 0x1fff) == 0 && buf[9] == IPPROTO_TCP) // first tcp fragment
+       {
+               int ihl = (buf[0] & 0xf) * 4; // length of IP header
+               if (len >= ihl + 20 && (buf[ihl + 13] & TCP_FLAG_SYN) && ((buf[ihl + 12] >> 4) > 5))
+                       adjust_tcp_mss(s, t, buf, len, buf + ihl);
+       }
+
+       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;
+       }
+
+       if (sp->walled_garden && !config->cluster_iam_master)
+       {
+               // We are walled-gardening this
+               master_garden_packet(s, data, size);
+               return;
+       }
+
+       // Add on L2TP header
+       {
+               bundleidt bid = 0;
+               if (session[s].bundle && bundle[session[s].bundle].num_of_links > 1)
+               {
+                       bid = session[s].bundle;
+                       s = bundle[bid].members[bundle[bid].current_ses = ++bundle[bid].current_ses % bundle[bid].num_of_links];
+                       LOG(4, s, t, "MPPP: (1)Session number becomes: %u\n", s);
+                       if (len > 256)
+                       {
+                               // Partition the packet to 2 fragments
+                               uint32_t frag1len = len / 2;
+                               uint32_t frag2len = len - frag1len;
+                               uint8_t *p = makeppp(b1, sizeof(b1), buf, frag1len, s, t, PPPIP, 0, bid, MP_BEGIN);
+                               uint8_t *q;
+
+                               if (!p) return;
+                               tunnelsend(b1, frag1len + (p-b1), t); // send it...
+                               s = bundle[bid].members[bundle[bid].current_ses = ++bundle[bid].current_ses % bundle[bid].num_of_links];
+                               LOG(4, s, t, "MPPP: (2)Session number becomes: %u\n", s);
+                               q = makeppp(b2, sizeof(b2), buf+frag1len, frag2len, s, t, PPPIP, 0, bid, MP_END);
+                               if (!q) return;
+                               tunnelsend(b2, frag2len + (q-b2), t); // send it...
+                       }
+                       else {
+                               // Send it as one frame
+                               uint8_t *p = makeppp(b1, sizeof(b1), buf, len, s, t, PPPIP, 0, bid, MP_BOTH_BITS);
+                               if (!p) return;
+                               tunnelsend(b1, len + (p-b1), t); // send it...
+                       }
+               }
+               else
+               {
+                       uint8_t *p = makeppp(b1, sizeof(b1), buf, len, s, t, PPPIP, 0, 0, 0);
+                       if (!p) return;
+                       tunnelsend(b1, len + (p-b1), t); // send it...
+               }
+       }
+
+       // 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);
+
+       increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
+       sp->cout_delta += len;
+       sp->pout++;
+       udp_tx += len;
+
+       sess_local[s].cout += len;      // To send to master..
+       sess_local[s].pout++;
+}
+
+// process outgoing (to tunnel) IPv6
+//
+static void processipv6out(uint8_t * buf, int len)
+{
+       sessionidt s;
+       sessiont *sp;
+       tunnelidt t;
+       in_addr_t ip;
+       struct in6_addr ip6;
+
+       uint8_t *data = buf;    // Keep a copy of the originals.
+       int size = len;
+
+       uint8_t b[MAXETHER + 20];
+
+       CSTAT(processipv6out);
+
+       if (len < MIN_IP_SIZE)
+       {
+               LOG(1, 0, 0, "Short IPv6, %d bytes\n", len);
+               STAT(tunnel_tx_errors);
+               return;
+       }
+       if (len >= MAXETHER)
+       {
+               LOG(1, 0, 0, "Oversize IPv6 packet %d bytes\n", len);
+               STAT(tunnel_tx_errors);
+               return;
+       }
+
+       // Skip the tun header
+       buf += 4;
+       len -= 4;
+
+       // Got an IP header now
+       if (*(uint8_t *)(buf) >> 4 != 6)
+       {
+               LOG(1, 0, 0, "IP: Don't understand anything except IPv6\n");
+               return;
+       }
+
+       ip6 = *(struct in6_addr *)(buf+24);
+       s = sessionbyipv6(ip6);
+
+       if (s == 0)
+       {
+               ip = *(uint32_t *)(buf + 32);
+               s = sessionbyip(ip);
+       }
+       
+       if (s == 0)
+       {
+               // 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.
+               {
+                       // FIXME: Should send icmp6 host unreachable
+               }
                return;
        }
+       if (session[s].bundle && bundle[session[s].bundle].num_of_links > 1)
+       {
+               bundleidt bid = session[s].bundle;
+               s = bundle[bid].members[bundle[bid].current_ses = ++bundle[bid].current_ses % bundle[bid].num_of_links];
+               LOG(3, s, session[s].tunnel, "MPPP: Session number becomes: %u\n", s);
+       }
        t = session[s].tunnel;
        sp = &session[s];
 
-       // Snooping this session, send it to ASIO
-       if (sp->snoop) snoop_send_packet(buf, len);
+       // FIXME: add DoS prevention/filters?
+
+       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;
+       }
+
+       LOG(5, s, t, "Ethernet -> Tunnel (%d bytes)\n", len);
+
+       // Add on L2TP header
+       {
+               uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIPV6, 0, 0, 0);
+               if (!p) return;
+               tunnelsend(b, len + (p-b), t); // send it...
+       }
+
+       // 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);
+
+       increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
+       sp->cout_delta += len;
+       sp->pout++;
+       udp_tx += len;
+
+       sess_local[s].cout += len;      // To send to master..
+       sess_local[s].pout++;
+}
+
+//
+// Helper routine for the TBF filters.
+// Used to send queued data in to the user!
+//
+static void send_ipout(sessionidt s, uint8_t *buf, int len)
+{
+       sessiont *sp;
+       tunnelidt t;
+       in_addr_t ip;
 
-       log(5, session[s].ip, s, t, "Ethernet -> Tunnel (%d bytes)\n", len);
+       uint8_t b[MAXETHER + 20];
 
-       // Plugin hook
+       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, "Odd size IP packet: %d bytes\n", len);
+               return;
        }
 
+       // Skip the tun header
+       buf += 4;
+       len -= 4;
+
+       ip = *(in_addr_t *)(buf + 16);
+
+       if (!session[s].ip)
+               return;
+
+       t = session[s].tunnel;
+       sp = &session[s];
+
+       LOG(5, s, t, "Ethernet -> Tunnel (%d bytes)\n", len);
+
        // Add on L2TP header
        {
-               u8 *p = makeppp(b, buf, len, t, s, PPPIP);
+               uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIP, 0, 0, 0);
+               if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
-               sp->cout += len; // byte count
-               sp->pout++;
-               udp_tx += len;
        }
+
+       // Snooping this session.
+       if (sp->snoop_ip && sp->snoop_port)
+               snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
+
+       increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
+       sp->cout_delta += len;
+       sp->pout++;
+       udp_tx += len;
+
+       sess_local[s].cout += len;      // To send to master..
+       sess_local[s].pout++;
 }
 
 // add an AVP (16 bit)
-void control16(controlt * c, u16 avp, u16 val, u8 m)
+static void control16(controlt * c, uint16_t avp, uint16_t val, uint8_t m)
 {
-       u16 l = (m ? 0x8008 : 0x0008);
-       *(u16 *) (c->buf + c->length + 0) = htons(l);
-       *(u16 *) (c->buf + c->length + 2) = htons(0);
-       *(u16 *) (c->buf + c->length + 4) = htons(avp);
-       *(u16 *) (c->buf + c->length + 6) = htons(val);
+       uint16_t l = (m ? 0x8008 : 0x0008);
+       *(uint16_t *) (c->buf + c->length + 0) = htons(l);
+       *(uint16_t *) (c->buf + c->length + 2) = htons(0);
+       *(uint16_t *) (c->buf + c->length + 4) = htons(avp);
+       *(uint16_t *) (c->buf + c->length + 6) = htons(val);
        c->length += 8;
 }
 
 // add an AVP (32 bit)
-void control32(controlt * c, u16 avp, u32 val, u8 m)
+static void control32(controlt * c, uint16_t avp, uint32_t val, uint8_t m)
 {
-       u16 l = (m ? 0x800A : 0x000A);
-       *(u16 *) (c->buf + c->length + 0) = htons(l);
-       *(u16 *) (c->buf + c->length + 2) = htons(0);
-       *(u16 *) (c->buf + c->length + 4) = htons(avp);
-       *(u32 *) (c->buf + c->length + 6) = htonl(val);
+       uint16_t l = (m ? 0x800A : 0x000A);
+       *(uint16_t *) (c->buf + c->length + 0) = htons(l);
+       *(uint16_t *) (c->buf + c->length + 2) = htons(0);
+       *(uint16_t *) (c->buf + c->length + 4) = htons(avp);
+       *(uint32_t *) (c->buf + c->length + 6) = htonl(val);
        c->length += 10;
 }
 
-// add an AVP (32 bit)
-void controls(controlt * c, u16 avp, char *val, u8 m)
+// add an AVP (string)
+static void controls(controlt * c, uint16_t avp, char *val, uint8_t m)
 {
-       u16 l = ((m ? 0x8000 : 0) + strlen(val) + 6);
-       *(u16 *) (c->buf + c->length + 0) = htons(l);
-       *(u16 *) (c->buf + c->length + 2) = htons(0);
-       *(u16 *) (c->buf + c->length + 4) = htons(avp);
+       uint16_t l = ((m ? 0x8000 : 0) + strlen(val) + 6);
+       *(uint16_t *) (c->buf + c->length + 0) = htons(l);
+       *(uint16_t *) (c->buf + c->length + 2) = htons(0);
+       *(uint16_t *) (c->buf + c->length + 4) = htons(avp);
        memcpy(c->buf + c->length + 6, val, strlen(val));
        c->length += 6 + strlen(val);
 }
 
 // add a binary AVP
-void controlb(controlt * c, u16 avp, char *val, unsigned int len, u8 m)
+static void controlb(controlt * c, uint16_t avp, uint8_t *val, unsigned int len, uint8_t m)
 {
-       u16 l = ((m ? 0x8000 : 0) + len + 6);
-       *(u16 *) (c->buf + c->length + 0) = htons(l);
-       *(u16 *) (c->buf + c->length + 2) = htons(0);
-       *(u16 *) (c->buf + c->length + 4) = htons(avp);
+       uint16_t l = ((m ? 0x8000 : 0) + len + 6);
+       *(uint16_t *) (c->buf + c->length + 0) = htons(l);
+       *(uint16_t *) (c->buf + c->length + 2) = htons(0);
+       *(uint16_t *) (c->buf + c->length + 4) = htons(avp);
        memcpy(c->buf + c->length + 6, val, len);
        c->length += 6 + len;
 }
 
 // new control connection
-controlt *controlnew(u16 mtype)
+static controlt *controlnew(uint16_t mtype)
 {
        controlt *c;
        if (!controlfree)
@@ -710,42 +1513,46 @@ controlt *controlnew(u16 mtype)
        }
        assert(c);
        c->next = 0;
-       *(u16 *) (c->buf + 0) = htons(0xC802); // flags/ver
+       *(uint16_t *) (c->buf + 0) = htons(0xC802); // flags/ver
        c->length = 12;
        control16(c, 0, mtype, 1);
        return c;
 }
 
 // send zero block if nothing is waiting
-void controlnull(tunnelidt t)
+// (ZLB send).
+static void controlnull(tunnelidt t)
 {
-       u8 buf[12];
-       if (tunnel[t].controlc)
+       uint8_t buf[12];
+       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
-       *(u16 *) (buf + 6) = htons(0); // session
-       *(u16 *) (buf + 8) = htons(tunnel[t].ns); // sequence
-       *(u16 *) (buf + 10) = htons(tunnel[t].nr); // sequence
+
+       *(uint16_t *) (buf + 0) = htons(0xC802); // flags/ver
+       *(uint16_t *) (buf + 2) = htons(12); // length
+       *(uint16_t *) (buf + 4) = htons(tunnel[t].far); // tunnel
+       *(uint16_t *) (buf + 6) = htons(0); // session
+       *(uint16_t *) (buf + 8) = htons(tunnel[t].ns); // sequence
+       *(uint16_t *) (buf + 10) = htons(tunnel[t].nr); // sequence
        tunnelsend(buf, 12, t);
 }
 
 // add a control message to a tunnel, and send if within window
-void controladd(controlt * c, tunnelidt t, sessionidt s)
+static void controladd(controlt *c, sessionidt far, tunnelidt t)
 {
-       *(u16 *) (c->buf + 2) = htons(c->length); // length
-       *(u16 *) (c->buf + 4) = htons(tunnel[t].far); // tunnel
-       *(u16 *) (c->buf + 6) = htons(s ? session[s].far : 0); // session
-       *(u16 *) (c->buf + 8) = htons(tunnel[t].ns); // sequence
+       *(uint16_t *) (c->buf + 2) = htons(c->length); // length
+       *(uint16_t *) (c->buf + 4) = htons(tunnel[t].far); // tunnel
+       *(uint16_t *) (c->buf + 6) = htons(far); // session
+       *(uint16_t *) (c->buf + 8) = htons(tunnel[t].ns); // sequence
        tunnel[t].ns++;              // advance sequence
        // link in message in to queue
        if (tunnel[t].controlc)
                tunnel[t].controle->next = c;
        else
                tunnel[t].controls = c;
+
        tunnel[t].controle = c;
        tunnel[t].controlc++;
+
        // send now if space in window
        if (tunnel[t].controlc <= tunnel[t].window)
        {
@@ -754,131 +1561,347 @@ 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].opened)
+               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;
+       }
+}
+
+// add/remove filters from session (-1 = no change)
+void filter_session(sessionidt s, int filter_in, int filter_out)
+{
+       if (!session[s].opened)
+               return; // No-one home.
+
+       if (!*session[s].user)
+               return; // User not logged in
+
+       // paranoia
+       if (filter_in > MAXFILTER) filter_in = -1;
+       if (filter_out > MAXFILTER) filter_out = -1;
+       if (session[s].filter_in > MAXFILTER) session[s].filter_in = 0;
+       if (session[s].filter_out > MAXFILTER) session[s].filter_out = 0;
+
+       if (filter_in >= 0)
+       {
+               if (session[s].filter_in)
+                       ip_filters[session[s].filter_in - 1].used--;
+
+               if (filter_in > 0)
+                       ip_filters[filter_in - 1].used++;
+
+               session[s].filter_in = filter_in;
+       }
+
+       if (filter_out >= 0)
+       {
+               if (session[s].filter_out)
+                       ip_filters[session[s].filter_out - 1].used--;
+
+               if (filter_out > 0)
+                       ip_filters[filter_out - 1].used++;
+
+               session[s].filter_out = filter_out;
+       }
+}
+
 // start tidy shutdown of session
-void sessionshutdown(sessionidt s, char *reason)
+void sessionshutdown(sessionidt s, char const *reason, int cdn_result, int cdn_error, int term_cause)
 {
-       int dead = session[s].die;
        int walled_garden = session[s].walled_garden;
 
-#ifdef STAT_CALLS
-       STAT(call_sessionshutdown);
-#endif
-       if (!session[s].tunnel)
-               return;                   // not a live session
 
-       if (!session[s].die)
-               log(2, 0, s, session[s].tunnel, "Shutting down session %d: %s\n", s, reason);
+       CSTAT(sessionshutdown);
 
-       session[s].die = now() + 150; // Clean up in 15 seconds
+       if (!session[s].opened)
+       {
+               LOG(3, s, session[s].tunnel, "Called sessionshutdown on an unopened session.\n");
+               return;                   // not a live session
+       }
 
+       if (!session[s].die)
        {
                struct param_kill_session data = { &tunnel[session[s].tunnel], &session[s] };
+               LOG(2, s, session[s].tunnel, "Shutting down session %u: %s\n", s, reason);
                run_plugins(PLUGIN_KILL_SESSION, &data);
        }
 
-       // RADIUS Stop message
-       if (session[s].opened && !walled_garden && !dead) {
-               u8 r = session[s].radius;
-               if (!r)
+       if (session[s].ip && !walled_garden && !session[s].die)
+       {
+               // RADIUS Stop message
+               uint16_t r = radiusnew(s);
+               if (r)
                {
-                       if (!radiusfree)
+                       // stop, if not already trying
+                       if (radius[r].state != RADIUSSTOP)
                        {
-                               log(1, 0, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");
-                               STAT(radius_overflow);
-                       } else {
-                               int n;
-                               r = radiusnew(s);
-                               for (n = 0; n < 15; n++)
-                                       radius[r].auth[n] = rand();
+                               radius[r].term_cause = term_cause;
+                               radius[r].term_msg = reason;
+                               radiussend(r, RADIUSSTOP);
                        }
                }
-               if (r && radius[r].state != RADIUSSTOP)
-                       radiussend(r, RADIUSSTOP); // stop, if not already trying
+               else
+                       LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");
+
+               // Save counters to dump to accounting file
+               if (*config->accounting_dir && shut_acct_n < sizeof(shut_acct) / sizeof(*shut_acct))
+                       memcpy(&shut_acct[shut_acct_n++], &session[s], sizeof(session[s]));
        }
 
        if (session[s].ip)
        {                          // IP allocated, clear and unroute
-               u8 r;
-               if (session[s].route[0].ip)
+               int r;
+               int routed = 0;
+               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;
-                               }
-                       }
+                       if ((session[s].ip & session[s].route[r].mask) ==
+                           (session[s].route[r].ip & session[s].route[r].mask))
+                               routed++;
+
+                       routeset(s, session[s].route[r].ip, session[s].route[r].mask, 0, 0);
+                       session[s].route[r].ip = 0;
+               }
+
+               if (session[s].ip_pool_index == -1) // static ip
+               {
+                       if (!routed) routeset(s, session[s].ip, 0, 0, 0);
+                       session[s].ip = 0;
                }
-               if (session[s].throttle) throttle_session(s, 0); session[s].throttle = 0;
-               free_ip_address(session[s].ip);
-               session[s].ip = 0;
+               else
+                       free_ip_address(s);
+
+               // unroute IPv6, if setup
+               if (session[s].ppp.ipv6cp == Opened && session[s].ipv6prefixlen)
+                       route6set(s, session[s].ipv6route, session[s].ipv6prefixlen, 0);
        }
+
+       if (session[s].throttle_in || session[s].throttle_out) // Unthrottle if throttled.
+               throttle_session(s, 0, 0);
+
+       if (cdn_result)
        {                            // Send CDN
                controlt *c = controlnew(14); // sending CDN
-               control16(c, 1, 3, 1);    // result code (admin reasons - TBA make error, general error, add message
+               if (cdn_error)
+               {
+                       uint8_t buf[4];
+                       *(uint16_t *) buf     = htons(cdn_result);
+                       *(uint16_t *) (buf+2) = htons(cdn_error);
+                       controlb(c, 1, buf, 4, 1);
+               }
+               else
+                       control16(c, 1, cdn_result, 1);
+
                control16(c, 14, s, 1);   // assigned session (our end)
-               controladd(c, session[s].tunnel, s); // send the message
+               controladd(c, session[s].far, session[s].tunnel); // send the message
        }
-       cluster_send_session(s);
-}
 
-void sendipcp(tunnelidt t, sessionidt s)
+       if (!session[s].die)
+               session[s].die = TIME + 150; // Clean up in 15 seconds
+
+       // update filter refcounts
+       if (session[s].filter_in) ip_filters[session[s].filter_in - 1].used--;
+       if (session[s].filter_out) ip_filters[session[s].filter_out - 1].used--;
+
+       // clear PPP state
+       memset(&session[s].ppp, 0, sizeof(session[s].ppp));
+       sess_local[s].lcp.restart = 0;
+       sess_local[s].ipcp.restart = 0;
+       sess_local[s].ipv6cp.restart = 0;
+       sess_local[s].ccp.restart = 0;
+
+       cluster_send_session(s);
+}
+
+void sendipcp(sessionidt s, tunnelidt t)
 {
-       u8 buf[MAXCONTROL];
-       u8 r = session[s].radius;
-       u8 *q;
-#ifdef STAT_CALLS
-       STAT(call_sendipcp);
-#endif
-       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)
+       uint8_t buf[MAXETHER];
+       uint8_t *q;
+
+       CSTAT(sendipcp);
+       LOG(3, s, t, "IPCP: send ConfigReq\n");
+
+       if (!session[s].unique_id)
        {
-               sessionshutdown(s, "No reply on IPCP");
-               return;
+               if (!++last_id) ++last_id; // skip zero
+               session[s].unique_id = last_id;
        }
-       q = makeppp(buf, 0, 0, t, s, PPPIPCP);
+
+       q = makeppp(buf, sizeof(buf), 0, 0, s, t, PPPIPCP, 0, 0, 0);
+       if (!q) return;
+
        *q = ConfigReq;
-       q[1] = r;                    // 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 ? : session[s].ip); // send my IP (use theirs if I dont have one)
+       q[1] = session[s].unique_id & 0xf;      // ID, dont care, we only send one type of request
+       *(uint16_t *) (q + 2) = htons(10);      // packet length
+       q[4] = 3;                               // ip address option
+       q[5] = 6;                               // option length
+       *(in_addr_t *) (q + 6) = config->peer_address ? config->peer_address :
+                                config->bind_address ? config->bind_address :
+                                my_address; // send my IP
+
        tunnelsend(buf, 10 + (q - buf), t); // send it
+       restart_timer(s, ipcp);
 }
 
-// kill a session now
-void sessionkill(sessionidt s, char *reason)
+void sendipv6cp(sessionidt s, tunnelidt t)
+{
+       uint8_t buf[MAXETHER];
+       uint8_t *q;
+
+       CSTAT(sendipv6cp);
+       LOG(3, s, t, "IPV6CP: send ConfigReq\n");
+
+       q = makeppp(buf, sizeof(buf), 0, 0, s, t, PPPIPV6CP, 0, 0, 0);
+       if (!q) return;
+
+       *q = ConfigReq;
+       q[1] = session[s].unique_id & 0xf;      // ID, don't care, we
+                                               // only send one type
+                                               // of request
+       *(uint16_t *) (q + 2) = htons(14);
+       q[4] = 1;                               // interface identifier option
+       q[5] = 10;                              // option length
+       *(uint32_t *) (q + 6) = 0;              // We'll be prefix::1
+       *(uint32_t *) (q + 10) = 0;
+       q[13] = 1;
+
+       tunnelsend(buf, 14 + (q - buf), t);     // send it
+       restart_timer(s, ipv6cp);
+}
+
+static void sessionclear(sessionidt s)
 {
-#ifdef STAT_CALLS
-       STAT(call_sessionkill);
-#endif
-       sessionshutdown(s, reason);  // close radius/routes, etc.
-       if (session[s].radius)
-               radius[session[s].radius].session = 0; // cant send clean accounting data, session is killed
        memset(&session[s], 0, sizeof(session[s]));
+       memset(&sess_local[s], 0, sizeof(sess_local[s]));
+       memset(&cli_session_actions[s], 0, sizeof(cli_session_actions[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);
+}
+
+// kill a session now
+void sessionkill(sessionidt s, char *reason)
+{
+       bundleidt b;
+
+       CSTAT(sessionkill);
+
+       if (!session[s].opened) // not alive
+               return;
+
+       if (session[s].next)
+       {
+               LOG(0, s, session[s].tunnel, "Tried to kill a session with next pointer set (%u)\n", session[s].next);
+               return;
+       }
+
+       session[s].die = TIME;
+       sessionshutdown(s, reason, CDN_ADMIN_DISC, TERM_ADMIN_RESET);  // close radius/routes, etc.
+       if (sess_local[s].radius)
+               radiusclear(sess_local[s].radius, s); // cant send clean accounting data, session is killed
+
+       LOG(2, s, session[s].tunnel, "Kill session %u (%s): %s\n", s, session[s].user, reason);
+       if ((b = session[s].bundle))
+       {
+               // This session was part of a bundle
+               bundle[b].num_of_links--;
+               LOG(3, s, 0, "MPPP: Dropping member link: %u from bundle %u\n", s, b);
+               if (bundle[b].num_of_links == 0)
+               {
+                       bundleclear(b);
+                       LOG(3, s, 0, "MPPP: Kill bundle: %u (No remaing member links)\n", b);
+               }
+               else
+               {
+                       // Adjust the members array to accomodate the new change
+                       uint8_t mem_num = 0;
+                       // It should be here num_of_links instead of num_of_links-1 (previous instruction "num_of_links--")
+                       if (bundle[b].members[bundle[b].num_of_links] != s)
+                       {
+                               uint8_t ml;
+                               for (ml = 0; ml<bundle[b].num_of_links; ml++)
+                               {
+                                       if (bundle[b].members[ml] == s)
+                                       {
+                                               mem_num = ml;
+                                               break;
+                                       }
+                               }
+                               bundle[b].members[mem_num] = bundle[b].members[bundle[b].num_of_links];
+                               LOG(3, s, 0, "MPPP: Adjusted member links array\n");
+                       }
+               }
+               cluster_send_bundle(b);
+       }
+       sessionclear(s);
        cluster_send_session(s);
 }
 
+static void tunnelclear(tunnelidt t)
+{
+       if (!t) return;
+       memset(&tunnel[t], 0, sizeof(tunnel[t]));
+       tunnel[t].state = TUNNELFREE;
+}
+
+static void bundleclear(bundleidt b)
+{
+       if (!b) return;
+       memset(&bundle[b], 0, sizeof(bundle[b]));
+       bundle[b].state = BUNDLEFREE;
+}
+
 // kill a tunnel now
-void tunnelkill(tunnelidt t, char *reason)
+static void tunnelkill(tunnelidt t, char *reason)
 {
        sessionidt s;
        controlt *c;
-#ifdef STAT_CALLS
-       STAT(call_tunnelkill);
-#endif
+
+       CSTAT(tunnelkill);
+
+       tunnel[t].state = TUNNELDIE;
+
        // free control messages
        while ((c = tunnel[t].controls))
        {
@@ -889,206 +1912,250 @@ void tunnelkill(tunnelidt t, char *reason)
                controlfree = c;
        }
        // kill sessions
-       for (s = 0; s < MAXSESSION; s++)
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                if (session[s].tunnel == t)
                        sessionkill(s, reason);
+
        // free tunnel
-       memset(&tunnel[t], 0, sizeof(tunnel[t]));
-       tunnel[t].next = tunnelfree;
+       tunnelclear(t);
+       LOG(1, 0, t, "Kill tunnel %u: %s\n", t, reason);
+       cli_tunnel_actions[t].action = 0;
        cluster_send_tunnel(t);
-       log(1, 0, 0, t, "Kill tunnel %d: %s\n", t, reason);
-       tunnelfree = t;
 }
 
-// shut down a tunnel
-void tunnelshutdown(tunnelidt t, char *reason)
+// shut down a tunnel cleanly
+static void tunnelshutdown(tunnelidt t, char *reason, int result, int error, char *msg)
 {
        sessionidt s;
-#ifdef STAT_CALLS
-       STAT(call_tunnelshutdown);
-#endif
-       if (!tunnel[t].last || !tunnel[t].far)
-       {                          // never set up, can immediately kill
+
+       CSTAT(tunnelshutdown);
+
+       if (!tunnel[t].last || !tunnel[t].far || tunnel[t].state == TUNNELFREE)
+       {
+               // never set up, can immediately kill
                tunnelkill(t, reason);
                return;
        }
-       log(1, 0, 0, t, "Shutting down tunnel %d (%s)\n", t, reason);
+       LOG(1, 0, t, "Shutting down tunnel %u (%s)\n", t, reason);
+
        // close session
-       for (s = 0; s < MAXSESSION; s++)
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                if (session[s].tunnel == t)
-                       sessionkill(s, reason);
-       tunnel[t].die = now() + 700; // Clean up in 70 seconds
+                       sessionshutdown(s, reason, CDN_NONE, TERM_ADMIN_RESET);
+
+       tunnel[t].state = TUNNELDIE;
+       tunnel[t].die = TIME + 700; // Clean up in 70 seconds
        cluster_send_tunnel(t);
        // TBA - should we wait for sessions to stop?
-       {                            // Send StopCCN
-               controlt *c = controlnew(4); // sending StopCCN
-               control16(c, 1, 1, 1);    // result code (admin reasons - TBA make error, general error, add message
-               control16(c, 9, t, 1);    // assigned tunnel (our end)
-               controladd(c, t, 0);      // send the message
+       if (result) 
+       {
+               controlt *c = controlnew(4);    // sending StopCCN
+               if (error)
+               {
+                       uint8_t buf[64];
+                       int l = 4;
+                       *(uint16_t *) buf     = htons(result);
+                       *(uint16_t *) (buf+2) = htons(error);
+                       if (msg)
+                       {
+                               int m = strlen(msg);
+                               if (m + 4 > sizeof(buf))
+                                   m = sizeof(buf) - 4;
+
+                               memcpy(buf+4, msg, m);
+                               l += m;
+                       }
+
+                       controlb(c, 1, buf, l, 1);
+               }
+               else
+                       control16(c, 1, result, 1);
+
+               control16(c, 9, t, 1);          // assigned tunnel (our end)
+               controladd(c, 0, t);            // send the message
        }
 }
 
 // read and process packet on tunnel (UDP)
-void processudp(u8 * buf, int len, struct sockaddr_in *addr)
+void processudp(uint8_t *buf, int len, struct sockaddr_in *addr)
 {
-       char *chapresponse = NULL;
-       u16 l = len, t = 0, s = 0, ns = 0, nr = 0;
-       u8 *p = buf + 2;
+       uint8_t *chapresponse = NULL;
+       uint16_t l = len, t = 0, s = 0, ns = 0, nr = 0;
+       uint8_t *p = buf + 2;
+
+
+       CSTAT(processudp);
 
-#ifdef STAT_CALLS
-       STAT(call_processudp);
-#endif
        udp_rx += len;
        udp_rx_pkt++;
-       log_hex(5, "UDP Data", buf, len);
+       LOG_HEX(5, "UDP Data", buf, len);
        STAT(tunnel_rx_packets);
        INC_STAT(tunnel_rx_bytes, len);
        if (len < 6)
        {
-               log(1, ntohl(addr->sin_addr.s_addr), 0, 0, "Short UDP, %d bytes\n", len);
+               LOG(1, 0, 0, "Short UDP, %d bytes\n", len);
                STAT(tunnel_rx_errors);
                return;
        }
        if ((buf[1] & 0x0F) != 2)
        {
-               log(1, ntohl(addr->sin_addr.s_addr), 0, 0, "Bad L2TP ver %d\n", (buf[1] & 0x0F) != 2);
+               LOG(1, 0, 0, "Bad L2TP ver %d\n", buf[1] & 0x0F);
                STAT(tunnel_rx_errors);
                return;
        }
        if (*buf & 0x40)
        {                          // length
-               l = ntohs(*(u16 *) p);
+               l = ntohs(*(uint16_t *) p);
                p += 2;
        }
-       t = ntohs(*(u16 *) p);
+       t = ntohs(*(uint16_t *) p);
        p += 2;
-       s = ntohs(*(u16 *) p);
+       s = ntohs(*(uint16_t *) p);
        p += 2;
        if (s >= MAXSESSION)
        {
-               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Received UDP packet with invalid session ID\n");
+               LOG(1, s, t, "Received UDP packet with invalid session ID\n");
                STAT(tunnel_rx_errors);
                return;
        }
        if (t >= MAXTUNNEL)
        {
-               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Received UDP packet with invalid tunnel ID\n");
-               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);
+               LOG(1, s, t, "Received UDP packet with invalid tunnel ID\n");
                STAT(tunnel_rx_errors);
                return;
        }
        if (*buf & 0x08)
        {                          // ns/nr
-               ns = ntohs(*(u16 *) p);
+               ns = ntohs(*(uint16_t *) p);
                p += 2;
-               nr = ntohs(*(u16 *) p);
+               nr = ntohs(*(uint16_t *) p);
                p += 2;
        }
        if (*buf & 0x02)
        {                          // offset
-               u16 o = ntohs(*(u16 *) p);
+               uint16_t o = ntohs(*(uint16_t *) p);
                p += o + 2;
        }
        if ((p - buf) > l)
        {
-               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Bad length %d>%d\n", (p - buf), l);
+               LOG(1, s, t, "Bad length %d>%d\n", (int) (p - buf), l);
                STAT(tunnel_rx_errors);
                return;
        }
        l -= (p - buf);
-       if (t) tunnel[t].last = time_now;
+
+       // used to time out old tunnels
+       if (t && tunnel[t].state == TUNNELOPEN)
+               tunnel[t].lastrec = time_now;
+
        if (*buf & 0x80)
        {                          // control
-               u16 message = 0xFFFF; // message type
-               u8 fatal = 0;
-               u8 mandatorymessage = 0;
-               u8 chap = 0;      // if CHAP being used
-               u16 asession = 0;  // assigned session
-               u32 amagic = 0;    // magic number
-               u8 aflags = 0;    // flags from last LCF
-               u16 version = 0x0100; // protocol version (we handle 0.0 as well and send that back just in case)
-               int requestchap = 0;    // do we request PAP instead of original CHAP request?
-               char called[MAXTEL] = ""; // called number
-               char calling[MAXTEL] = ""; // calling number
-               if ((*buf & 0xCA) != 0xC8)
-               {
-                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Bad control header %02X\n", *buf);
+               uint16_t message = 0xFFFF;      // message type
+               uint8_t fatal = 0;
+               uint8_t mandatory = 0;
+               uint16_t asession = 0;          // assigned session
+               uint32_t amagic = 0;            // magic number
+               uint8_t aflags = 0;             // flags from last LCF
+               uint16_t version = 0x0100;      // protocol version (we handle 0.0 as well and send that back just in case)
+               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;
+               }
+
+               // control messages must have bits 0x80|0x40|0x08
+               // (type, length and sequence) set, and bits 0x02|0x01
+               // (offset and priority) clear
+               if ((*buf & 0xCB) != 0xC8)
+               {
+                       LOG(1, 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",
-                               l, tunnel[t].controlc, tunnel[t].ns, tunnel[t].nr, ns, nr);
-               // if no tunnel specified, assign one
-               if (!t)
+
+               // check for duplicate tunnel open message
+               if (!t && ns == 0)
                {
-                       /*
-                       ipt ip = ntohl(*(ipt *) & addr->sin_addr);
-                       portt port = ntohs(addr->sin_port);
+                       int i;
 
-                       // find existing tunnel that was not fully set up
-                       for (t = 0; t < MAXTUNNEL; t++)
+                               //
+                               // Is this a duplicate of the first packet? (SCCRQ)
+                               //
+                       for (i = 1; i <= config->cluster_highest_tunnelid ; ++i)
                        {
-                               if ((tunnel[t].ip == ip && tunnel[t].port == port) &&
-                                       (!tunnel[t].die || !tunnel[t].hostname[0]))
-                               {
-                                       char buf[600] = {0};
-                                       snprintf(buf, 600, "Duplicate tunnel with %d. ip=%u port=%d die=%d hostname=%s",
-                                                       t, tunnel[t].ip, tunnel[t].port, tunnel[t].die, tunnel[t].hostname);
-                                       tunnelshutdown(t, buf);
-                                       break;
-                               }
+                               if (tunnel[i].state != TUNNELOPENING ||
+                                       tunnel[i].ip != ntohl(*(in_addr_t *) & addr->sin_addr) ||
+                                       tunnel[i].port != ntohs(addr->sin_port) )
+                                       continue;
+                               t = i;
+                               LOG(3, s, t, "Duplicate SCCRQ?\n");
+                               break;
                        }
-                       */
+               }
+
+               LOG(3, s, t, "Control message (%d bytes): (unacked %d) l-ns %u l-nr %u r-ns %u r-nr %u\n",
+                       l, tunnel[t].controlc, tunnel[t].ns, tunnel[t].nr, ns, nr);
 
-                       t = tunnelfree;
-                       if (!t)
+               // if no tunnel specified, assign one
+               if (!t)
+               {
+                       if (!(t = new_tunnel()))
                        {
-                               log(1, ntohl(addr->sin_addr.s_addr), 0, 0, "No more tunnels\n");
+                               LOG(1, 0, 0, "No more tunnels\n");
                                STAT(tunnel_overflow);
                                return;
                        }
-                       tunnelfree = tunnel[t].next;
-                       memset(&tunnel[t], 0, sizeof(tunnelt));
-                       tunnel[t].ip = ntohl(*(ipt *) & addr->sin_addr);
+                       tunnelclear(t);
+                       tunnel[t].ip = ntohl(*(in_addr_t *) & addr->sin_addr);
                        tunnel[t].port = ntohs(addr->sin_port);
                        tunnel[t].window = 4; // default window
-                       log(1, ntohl(addr->sin_addr.s_addr), 0, t, "   New tunnel from %u.%u.%u.%u/%u ID %d\n", tunnel[t].ip >> 24, tunnel[t].ip >> 16 & 255, tunnel[t].ip >> 8 & 255, tunnel[t].ip & 255, tunnel[t].port, t);
                        STAT(tunnel_created);
+                       LOG(1, 0, t, "   New tunnel from %s:%u ID %u\n",
+                               fmtaddr(htonl(tunnel[t].ip), 0), tunnel[t].port, t);
+               }
+
+                       // 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?
+                       STAT(tunnel_rx_errors);
+                       LOG(1, 0, t, "   Out of sequence tunnel %u, (%u is not the expected %u)\n",
+                               t, ns, tunnel[t].nr);
+
+                       if (l)  // Is this not a ZLB?
+                               controlnull(t);
+                       return;
                }
-               {                       // check sequence of this message
+
+               // 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
-                               }
-                       }
-                       if (tunnel[t].nr < ns && tunnel[t].nr != 0)
+                               // some to clear maybe?
+                       while (tunnel[t].controlc > 0 && (((tunnel[t].ns - tunnel[t].controlc) - nr) & 0x8000))
                        {
-                               // 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);
-                               STAT(tunnel_rx_errors);
-//                             controlnull(t);
-                               return;
+                               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 (l)
-                               tunnel[t].nr++;   // receiver advance (do here so quoted correctly in any sends below)
-                       if (skip < 0)
-                               skip = 0;
+
+                       // receiver advance (do here so quoted correctly in any sends below)
+                       if (l) tunnel[t].nr = (ns + 1);
+                       if (skip < 0) skip = 0;
                        if (skip < tunnel[t].controlc)
-                       {                  // some control packets can now be sent that were previous stuck out of window
+                       {
+                               // some control packets can now be sent that were previous stuck out of window
                                int tosend = tunnel[t].window - skip;
                                controlt *c = tunnel[t].controls;
                                while (c && skip)
@@ -1109,330 +2176,444 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                }
                if (l)
                {                     // if not a null message
+                       int result = 0;
+                       int error = 0;
+                       char *msg = 0;
+
+                       // default disconnect cause/message on receipt
+                       // of CDN (set to more specific value from
+                       // attribute 46 if present below).
+                       int disc_cause = TERM_NAS_REQUEST;
+                       char const *disc_reason = "Closed (Received CDN).";
+
                        // process AVPs
-                       while (l && !(fatal & 0x80))
+                       while (l && !(fatal & 0x80)) // 0x80 = mandatory AVP
                        {
-                               u16 n = (ntohs(*(u16 *) p) & 0x3FF);
-                               u8 *b = p;
-                               u8 flags = *p;
-                               u16 mtype;
-                               p += n;       // next
-                               if (l < n)
+                               uint16_t n = (ntohs(*(uint16_t *) p) & 0x3FF);
+                               uint8_t *b = p;
+                               uint8_t flags = *p;
+                               uint16_t mtype;
+
+                               if (n > l)
                                {
-                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Invalid length in AVP\n");
+                                       LOG(1, s, t, "Invalid length in AVP\n");
                                        STAT(tunnel_rx_errors);
-                                       fatal = flags;
                                        return;
                                }
+                               p += n;       // next
                                l -= n;
+                               if (flags & 0x3C) // reserved bits, should be clear
+                               {
+                                       LOG(1, s, t, "Unrecognised AVP flags %02X\n", *b);
+                                       fatal = flags;
+                                       result = 2; // general error
+                                       error = 3; // reserved field non-zero
+                                       msg = 0;
+                                       continue; // next
+                               }
+                               b += 2;
+                               if (*(uint16_t *) (b))
+                               {
+                                       LOG(2, s, t, "Unknown AVP vendor %u\n", ntohs(*(uint16_t *) (b)));
+                                       fatal = flags;
+                                       result = 2; // general error
+                                       error = 6; // generic vendor-specific error
+                                       msg = "unsupported vendor-specific";
+                                       continue; // next
+                               }
+                               b += 2;
+                               mtype = ntohs(*(uint16_t *) (b));
+                               b += 2;
+                               n -= 6;
+
                                if (flags & 0x40)
                                {
+                                       uint16_t orig_len;
+
                                        // handle hidden AVPs
-                                       if (!l2tpsecret)
+                                       if (!*config->l2tp_secret)
                                        {
-                                               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Hidden AVP requested, but no L2TP secret.\n");
+                                               LOG(1, s, t, "Hidden AVP requested, but no L2TP secret.\n");
                                                fatal = flags;
+                                               result = 2; // general error
+                                               error = 6; // generic vendor-specific error
+                                               msg = "secret not specified";
                                                continue;
                                        }
                                        if (!session[s].random_vector_length)
                                        {
-                                               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Hidden AVP requested, but no random vector.\n");
+                                               LOG(1, s, t, "Hidden AVP requested, but no random vector.\n");
                                                fatal = flags;
+                                               result = 2; // general error
+                                               error = 6; // generic
+                                               msg = "no random vector";
                                                continue;
                                        }
-                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Hidden AVP\n");
-                               }
-                               if (*b & 0x3C)
-                               {
-                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Unrecognised AVP flags %02X\n", *b);
-                                       fatal = flags;
-                                       continue; // next
-                               }
-                               b += 2;
-                               if (*(u16 *) (b))
-                               {
-                                       log(2, ntohl(addr->sin_addr.s_addr), s, t, "Unknown AVP vendor %d\n", ntohs(*(u16 *) (b)));
-                                       fatal = flags;
-                                       continue; // next
+                                       if (n < 8)
+                                       {
+                                               LOG(2, s, t, "Short hidden AVP.\n");
+                                               fatal = flags;
+                                               result = 2; // general error
+                                               error = 2; // length is wrong
+                                               msg = 0;
+                                               continue;
+                                       }
+
+                                       // Unhide the AVP
+                                       unhide_value(b, n, mtype, session[s].random_vector, session[s].random_vector_length);
+
+                                       orig_len = ntohs(*(uint16_t *) b);
+                                       if (orig_len > n + 2)
+                                       {
+                                               LOG(1, s, t, "Original length %d too long in hidden AVP of length %d; wrong secret?\n",
+                                                   orig_len, n);
+
+                                               fatal = flags;
+                                               result = 2; // general error
+                                               error = 2; // length is wrong
+                                               msg = 0;
+                                               continue;
+                                       }
+
+                                       b += 2;
+                                       n = orig_len;
                                }
-                               b += 2;
-                               mtype = ntohs(*(u16 *) (b));
-                               b += 2;
-                               n -= 6;
 
-                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   AVP %d (%s) len %d\n", mtype, avpnames[mtype], n);
+                               LOG(4, s, t, "   AVP %u (%s) len %d%s%s\n", mtype, l2tp_avp_name(mtype), n,
+                                       flags & 0x40 ? ", hidden" : "", flags & 0x80 ? ", mandatory" : "");
+
                                switch (mtype)
                                {
                                case 0:     // message type
-                                       message = ntohs(*(u16 *) b);
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Message type = %d (%s)\n", *b,
-                                                       l2tp_message_types[message]);
-                                       mandatorymessage = flags;
+                                       message = ntohs(*(uint16_t *) b);
+                                       mandatory = flags & 0x80;
+                                       LOG(4, s, t, "   Message type = %u (%s)\n", *b, l2tp_code(message));
                                        break;
                                case 1:     // result code
                                        {
-                                               u16 rescode = ntohs(*(u16 *)(b));
+                                               uint16_t rescode = ntohs(*(uint16_t *) b);
                                                const char* resdesc = "(unknown)";
-                                               if (message == 4) { /* StopCCN */
-                                                       if (rescode <= MAX_STOPCCN_RESULT_CODE)
-                                                               resdesc = stopccn_result_codes[rescode];
-                                               } else if (message == 14) { /* CDN */
-                                                       if (rescode <= MAX_CDN_RESULT_CODE)
-                                                               resdesc = cdn_result_codes[rescode];
+                                               if (message == 4)
+                                               { /* StopCCN */
+                                                       resdesc = l2tp_stopccn_result_code(rescode);
                                                }
-
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Result Code %d: %s\n",
-                                                       rescode, resdesc);
-                                               if (n >= 4) {
-                                                       u16 errcode = ntohs(*(u16 *)(b + 2));
-                                                       const char* errdesc = "(unknown)";
-                                                       if (errcode <= MAX_ERROR_CODE)
-                                                               errdesc = error_codes[errcode];
-                                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Error Code %d: %s\n",
-                                                               errcode, errdesc);
+                                               else if (message == 14)
+                                               { /* CDN */
+                                                       resdesc = l2tp_cdn_result_code(rescode);
                                                }
-                                               if (n > 4) {
-                                                       /* %*s doesn't work?? */
-                                                       char buf[n-4+2];
-                                                       memcpy(buf, b+4, n-4);
-                                                       buf[n-4+1] = '\0';
-                                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Error String: %s\n",
-                                                               buf);
+
+                                               LOG(4, s, t, "   Result Code %u: %s\n", rescode, resdesc);
+                                               if (n >= 4)
+                                               {
+                                                       uint16_t errcode = ntohs(*(uint16_t *)(b + 2));
+                                                       LOG(4, s, t, "   Error Code %u: %s\n", errcode, l2tp_error_code(errcode));
                                                }
+                                               if (n > 4)
+                                                       LOG(4, s, t, "   Error String: %.*s\n", n-4, b+4);
+
                                                break;
                                        }
                                        break;
                                case 2:     // protocol version
                                        {
-                                               version = ntohs(*(u16 *) (b));
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Protocol version = %d\n", version);
+                                               version = ntohs(*(uint16_t *) (b));
+                                               LOG(4, s, t, "   Protocol version = %u\n", version);
                                                if (version && version != 0x0100)
                                                {   // allow 0.0 and 1.0
-                                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "   Bad protocol version %04X\n",
-                                                                       version);
+                                                       LOG(1, s, t, "   Bad protocol version %04X\n", version);
                                                        fatal = flags;
+                                                       result = 5; // unspported protocol version
+                                                       error = 0x0100; // supported version
+                                                       msg = 0;
                                                        continue; // next
                                                }
                                        }
                                        break;
                                case 3:     // framing capabilities
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Framing capabilities\n");
                                        break;
                                case 4:     // bearer capabilities
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Bearer capabilities\n");
                                        break;
                                case 5:         // tie breaker
-                                       // We never open tunnels, so we don't
-                                       // care about tie breakers
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Tie breaker\n");
+                                       // We never open tunnels, so we don't care about tie breakers
                                        continue;
                                case 6:     // firmware revision
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Firmware revision\n");
                                        break;
                                case 7:     // host name
-                                       memset(tunnel[t].hostname, 0, 128);
-                                       memcpy(tunnel[t].hostname, b, (n >= 127) ? 127 : n);
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Tunnel hostname = \"%s\"\n", tunnel[t].hostname);
+                                       memset(tunnel[t].hostname, 0, sizeof(tunnel[t].hostname));
+                                       memcpy(tunnel[t].hostname, b, (n < sizeof(tunnel[t].hostname)) ? n : sizeof(tunnel[t].hostname) - 1);
+                                       LOG(4, s, t, "   Tunnel hostname = \"%s\"\n", tunnel[t].hostname);
                                        // 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);
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Vendor name = \"%s\"\n", tunnel[t].vendor);
+                                       memset(tunnel[t].vendor, 0, sizeof(tunnel[t].vendor));
+                                       memcpy(tunnel[t].vendor, b, (n < sizeof(tunnel[t].vendor)) ? n : sizeof(tunnel[t].vendor) - 1);
+                                       LOG(4, s, t, "   Vendor name = \"%s\"\n", tunnel[t].vendor);
                                        break;
                                case 9:     // assigned tunnel
-                                       tunnel[t].far = ntohs(*(u16 *) (b));
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Remote tunnel id = %d\n", tunnel[t].far);
+                                       tunnel[t].far = ntohs(*(uint16_t *) (b));
+                                       LOG(4, s, t, "   Remote tunnel id = %u\n", tunnel[t].far);
                                        break;
                                case 10:    // rx window
-                                       tunnel[t].window = ntohs(*(u16 *) (b));
+                                       tunnel[t].window = ntohs(*(uint16_t *) (b));
                                        if (!tunnel[t].window)
                                                tunnel[t].window = 1; // window of 0 is silly
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   rx window = %d\n", tunnel[t].window);
+                                       LOG(4, s, t, "   rx window = %u\n", tunnel[t].window);
                                        break;
                                case 11:        // Challenge
                                        {
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   LAC requested CHAP authentication for tunnel\n");
+                                               LOG(4, s, t, "   LAC requested CHAP authentication for tunnel\n");
                                                build_chap_response(b, 2, n, &chapresponse);
                                        }
                                        break;
+                               case 13:    // Response
+                                       // Why did they send a response? We never challenge.
+                                       LOG(2, 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);
+                                       asession = session[s].far = ntohs(*(uint16_t *) (b));
+                                       LOG(4, s, t, "   assigned session = %u\n", asession);
                                        break;
                                case 15:    // call serial number
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   call serial number = %d\n", ntohl(*(u32 *)b));
+                                       LOG(4, s, t, "   call serial number = %u\n", ntohl(*(uint32_t *)b));
                                        break;
                                case 18:    // bearer type
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   bearer type = %d\n", ntohl(*(u32 *)b));
+                                       LOG(4, s, t, "   bearer type = %u\n", ntohl(*(uint32_t *)b));
                                        // TBA - for RADIUS
                                        break;
                                case 19:    // framing type
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   framing type = %d\n", ntohl(*(u32 *)b));
+                                       LOG(4, s, t, "   framing type = %u\n", ntohl(*(uint32_t *)b));
                                        // TBA
                                        break;
                                case 21:    // called number
-                                       memset(called, 0, MAXTEL);
-                                       memcpy(called, b, (n >= MAXTEL) ? (MAXTEL-1) : n);
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Called <%s>\n", called);
+                                       memset(called, 0, sizeof(called));
+                                       memcpy(called, b, (n < sizeof(called)) ? n : sizeof(called) - 1);
+                                       LOG(4, s, t, "   Called <%s>\n", called);
                                        break;
                                case 22:    // calling number
-                                       memset(calling, 0, MAXTEL);
-                                       memcpy(calling, b, (n >= MAXTEL) ? (MAXTEL-1) : n);
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Calling <%s>\n", calling);
+                                       memset(calling, 0, sizeof(calling));
+                                       memcpy(calling, b, (n < sizeof(calling)) ? n : sizeof(calling) - 1);
+                                       LOG(4, s, t, "   Calling <%s>\n", calling);
+                                       break;
+                               case 23:    // subtype
                                        break;
                                case 24:    // tx connect speed
                                        if (n == 4)
                                        {
-                                               session[s].tx_connect_speed = ntohl(*(u32 *)b);
+                                               session[s].tx_connect_speed = ntohl(*(uint32_t *)b);
                                        }
                                        else
                                        {
                                                // AS5300s send connect speed as a string
-                                               char tmp[30] = {0};
-                                               memcpy(tmp, b, (n >= 30) ? 30 : n);
+                                               char tmp[30];
+                                               memset(tmp, 0, sizeof(tmp));
+                                               memcpy(tmp, b, (n < sizeof(tmp)) ? n : sizeof(tmp) - 1);
                                                session[s].tx_connect_speed = atol(tmp);
                                        }
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   TX connect speed <%d>\n",
-                                                       session[s].tx_connect_speed);
+                                       LOG(4, s, t, "   TX connect speed <%u>\n", session[s].tx_connect_speed);
                                        break;
                                case 38:    // rx connect speed
                                        if (n == 4)
                                        {
-                                               session[s].rx_connect_speed = ntohl(*(u32 *)b);
+                                               session[s].rx_connect_speed = ntohl(*(uint32_t *)b);
                                        }
                                        else
                                        {
                                                // AS5300s send connect speed as a string
-                                               char tmp[30] = {0};
-                                               memcpy(tmp, b, (n >= 30) ? 30 : n);
+                                               char tmp[30];
+                                               memset(tmp, 0, sizeof(tmp));
+                                               memcpy(tmp, b, (n < sizeof(tmp)) ? n : sizeof(tmp) - 1);
                                                session[s].rx_connect_speed = atol(tmp);
                                        }
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   RX connect speed <%d>\n",
-                                                       session[s].rx_connect_speed);
+                                       LOG(4, s, t, "   RX connect speed <%u>\n", session[s].rx_connect_speed);
                                        break;
                                case 25:    // Physical Channel ID
                                        {
-                                               u32 tmp = ntohl(*(u32 *)b);
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Physical Channel ID <%X>\n", tmp);
+                                               uint32_t tmp = ntohl(*(uint32_t *) b);
+                                               LOG(4, s, t, "   Physical Channel ID <%X>\n", tmp);
                                                break;
                                        }
                                case 29:    // Proxy Authentication Type
                                        {
-                                               u16 authtype = ntohs(*(u16 *)b);
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Proxy Auth Type %d (%s)\n",
-                                                       authtype, authtypes[authtype]);
-                                               requestchap = (authtype == 2);
+                                               uint16_t atype = ntohs(*(uint16_t *)b);
+                                               LOG(4, s, t, "   Proxy Auth Type %u (%s)\n", atype, ppp_auth_type(atype));
                                                break;
                                        }
                                case 30:    // Proxy Authentication Name
                                        {
-                                               char authname[64] = {0};
-                                               memcpy(authname, b, (n > 63) ? 63 : n);
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Proxy Auth Name (%s)\n",
+                                               char authname[64];
+                                               memset(authname, 0, sizeof(authname));
+                                               memcpy(authname, b, (n < sizeof(authname)) ? n : sizeof(authname) - 1);
+                                               LOG(4, s, t, "   Proxy Auth Name (%s)\n",
                                                        authname);
                                                break;
                                        }
                                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, s, t, "   Proxy Auth Challenge\n");
                                                break;
                                        }
                                case 32:    // Proxy Authentication ID
                                        {
-                                               u16 authid = ntohs(*(u16 *)(b));
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Proxy Auth ID (%d)\n",
-                                                       authid);
-                                               if (session[s].radius)
-                                                       radius[session[s].radius].id = authid;
+                                               uint16_t authid = ntohs(*(uint16_t *)(b));
+                                               LOG(4, s, t, "   Proxy Auth ID (%u)\n", authid);
                                                break;
                                        }
                                case 33:    // Proxy Authentication Response
-                                       {
-                                               char authresp[64] = {0};
-                                               memcpy(authresp, b, (n > 63) ? 63 : n);
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Proxy Auth Response\n");
-                                               break;
-                                       }
-                               case 27:    // last send lcp
+                                       LOG(4, s, t, "   Proxy Auth Response\n");
+                                       break;
+                               case 27:    // last sent lcp
                                        {        // find magic number
-                                               u8 *p = b, *e = p + n;
-                                               while (p < e && p[1])
+                                               uint8_t *p = b, *e = p + n;
+                                               while (p + 1 < e && p[1] && p + p[1] <= e)
                                                {
-                                                       if (*p == 5 && p[1] == 6)
-                                                               amagic = ntohl(*(u32 *) (p + 2));
-                                                       else if (*p == 3 && p[1] == 5 && *(u16 *) (p + 2) == htons(PPPCHAP) && p[4] == 5)
-                                                               chap = 1;
-                                                       else if (*p == 7)
-                                                               aflags |= SESSIONPFC;
-                                                       else if (*p == 8)
-                                                               aflags |= SESSIONACFC;
+                                                       if (*p == 5 && p[1] == 6) // Magic-Number
+                                                               amagic = ntohl(*(uint32_t *) (p + 2));
+                                                       else if (*p == 7) // Protocol-Field-Compression
+                                                               aflags |= SESSION_PFC;
+                                                       else if (*p == 8) // Address-and-Control-Field-Compression
+                                                               aflags |= SESSION_ACFC;
                                                        p += p[1];
                                                }
-
-                                               {
-                                                       char tmp[500] = {0};
-                                                       tmp[0] = ConfigReq;
-                                                       memcpy((tmp + 1), b, n);
-                                               }
                                        }
                                        break;
                                case 28:    // last recv lcp confreq
-                                       {
-                                               char tmp[500] = {0};
-                                               tmp[0] = ConfigReq;
-                                               memcpy((tmp + 1), b, n);
-                                               break;
-                                       }
+                                       break;
                                case 26:    // Initial Received LCP CONFREQ
-                                       {
-                                               char tmp[500] = {0};
-                                               tmp[0] = ConfigReq;
-                                               memcpy((tmp + 1), b, n);
-                                       }
                                        break;
                                case 39:    // seq required - we control it as an LNS anyway...
                                        break;
                                case 36:    // Random Vector
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Random Vector received. Enabled AVP Hiding.\n");
+                                       LOG(4, s, t, "   Random Vector received.  Enabled AVP Hiding.\n");
                                        memset(session[s].random_vector, 0, sizeof(session[s].random_vector));
+                                       if (n > sizeof(session[s].random_vector))
+                                               n = sizeof(session[s].random_vector);
                                        memcpy(session[s].random_vector, b, n);
                                        session[s].random_vector_length = n;
                                        break;
+                               case 46:    // ppp disconnect cause
+                                       if (n >= 5)
+                                       {
+                                               uint16_t code = ntohs(*(uint16_t *) b);
+                                               uint16_t proto = ntohs(*(uint16_t *) (b + 2));
+                                               uint8_t dir = *(b + 4);
+
+                                               LOG(4, s, t, "   PPP disconnect cause "
+                                                       "(code=%u, proto=%04X, dir=%u, msg=\"%.*s\")\n",
+                                                       code, proto, dir, n - 5, b + 5);
+
+                                               switch (code)
+                                               {
+                                               case 1: // admin disconnect
+                                                       disc_cause = TERM_ADMIN_RESET;
+                                                       disc_reason = "Administrative disconnect";
+                                                       break;
+                                               case 3: // lcp terminate
+                                                       if (dir != 2) break; // 1=peer (LNS), 2=local (LAC)
+                                                       disc_cause = TERM_USER_REQUEST;
+                                                       disc_reason = "Normal disconnection";
+                                                       break;
+                                               case 4: // compulsory encryption unavailable
+                                                       if (dir != 1) break; // 1=refused by peer, 2=local
+                                                       disc_cause = TERM_USER_ERROR;
+                                                       disc_reason = "Compulsory encryption refused";
+                                                       break;
+                                               case 5: // lcp: fsm timeout
+                                                       disc_cause = TERM_PORT_ERROR;
+                                                       disc_reason = "LCP: FSM timeout";
+                                                       break;
+                                               case 6: // lcp: no recognisable lcp packets received
+                                                       disc_cause = TERM_PORT_ERROR;
+                                                       disc_reason = "LCP: no recognisable LCP packets";
+                                                       break;
+                                               case 7: // lcp: magic-no error (possibly looped back)
+                                                       disc_cause = TERM_PORT_ERROR;
+                                                       disc_reason = "LCP: magic-no error (possible loop)";
+                                                       break;
+                                               case 8: // lcp: echo request timeout
+                                                       disc_cause = TERM_PORT_ERROR;
+                                                       disc_reason = "LCP: echo request timeout";
+                                                       break;
+                                               case 13: // auth: fsm timeout
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = "Authentication: FSM timeout";
+                                                       break;
+                                               case 15: // auth: unacceptable auth protocol
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = "Unacceptable authentication protocol";
+                                                       break;
+                                               case 16: // auth: authentication failed
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = "Authentication failed";
+                                                       break;
+                                               case 17: // ncp: fsm timeout
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = "NCP: FSM timeout";
+                                                       break;
+                                               case 18: // ncp: no ncps available
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = "NCP: no NCPs available";
+                                                       break;
+                                               case 19: // ncp: failure to converge on acceptable address
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = (dir == 1)
+                                                               ? "NCP: too many Configure-Naks received from peer"
+                                                               : "NCP: too many Configure-Naks sent to peer";
+                                                       break;
+                                               case 20: // ncp: user not permitted to use any address
+                                                       disc_cause = TERM_SERVICE_UNAVAILABLE;
+                                                       disc_reason = (dir == 1)
+                                                               ? "NCP: local link address not acceptable to peer"
+                                                               : "NCP: remote link address not acceptable";
+                                                       break;
+                                               }
+                                       }
+                                       break;
                                default:
-                                       log(2, ntohl(addr->sin_addr.s_addr), s, t, "   Unknown AVP type %d\n", mtype);
-                                       fatal = flags;
-                                       continue; // next
+                                       {
+                                               static char e[] = "unknown AVP 0xXXXX";
+                                               LOG(2, s, t, "   Unknown AVP type %u\n", mtype);
+                                               fatal = flags;
+                                               result = 2; // general error
+                                               error = 8; // unknown mandatory AVP
+                                               sprintf((msg = e) + 14, "%04x", mtype);
+                                               continue; // next
+                                       }
                                }
                        }
                        // process message
                        if (fatal & 0x80)
-                               tunnelshutdown(t, "Unknown Mandatory AVP");
+                               tunnelshutdown(t, "Invalid mandatory AVP", result, error, msg);
                        else
                                switch (message)
                                {
                                case 1:       // SCCRQ - Start Control Connection Request
+                                       tunnel[t].state = TUNNELOPENING;
+                                       if (main_quit != QUIT_SHUTDOWN)
                                        {
                                                controlt *c = controlnew(2); // sending SCCRP
                                                control16(c, 2, version, 1); // protocol version
                                                control32(c, 3, 3, 1); // framing
-                                               controls(c, 7, tunnel[t].hostname, 1); // host name (TBA)
+                                               controls(c, 7, hostname, 1); // host name
                                                if (chapresponse) controlb(c, 13, chapresponse, 16, 1); // Challenge response
                                                control16(c, 9, t, 1); // assigned tunnel
-                                               controladd(c, t, s); // send the resply
+                                               controladd(c, 0, t); // send the resply
+                                       }
+                                       else
+                                       {
+                                               tunnelshutdown(t, "Shutting down", 6, 0, 0);
                                        }
                                        break;
                                case 2:       // SCCRP
-                                       // TBA
+                                       tunnel[t].state = TUNNELOPEN;
                                        break;
                                case 3:       // SCCN
+                                       tunnel[t].state = TUNNELOPEN;
                                        controlnull(t); // ack
                                        break;
                                case 4:       // StopCCN
                                        controlnull(t); // ack
-                                       tunnelshutdown(t, "Stopped"); // Shut down cleanly
-                                       tunnelkill(t, "Stopped"); // Immediately force everything dead
+                                       tunnelshutdown(t, "Stopped", 0, 0, 0); // Shut down cleanly
                                        break;
                                case 6:       // HELLO
                                        controlnull(t); // simply ACK
@@ -1447,77 +2628,84 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        // TBA
                                        break;
                                case 10:      // ICRQ
-                                       if (!sessionfree)
-                                       {
-                                               STAT(session_overflow);
-                                               tunnelshutdown(t, "No free sessions");
-                                       }
-                                       else
+                                       if (sessionfree && main_quit != QUIT_SHUTDOWN)
                                        {
-                                               u8 r;
-                                               controlt *c;
-
-                                               // make a RADIUS session
-                                               if (!radiusfree)
-                                               {
-                                                       STAT(radius_overflow);
-                                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "No free RADIUS sessions for ICRQ\n");
-                                                       return;
-                                               }
+                                               controlt *c = controlnew(11); // ICRP
 
-                                               c = controlnew(11); // sending ICRP
                                                s = sessionfree;
                                                sessionfree = session[s].next;
                                                memset(&session[s], 0, sizeof(session[s]));
-                                               session[s].id = sessionid++;
-                                               session[s].opened = time(NULL);
+
+                                               if (s > config->cluster_highest_sessionid)
+                                                       config->cluster_highest_sessionid = s;
+
+                                               session[s].opened = time_now;
                                                session[s].tunnel = t;
                                                session[s].far = asession;
-                                               log(3, ntohl(addr->sin_addr.s_addr), s, t, "New session (%d/%d)\n", tunnel[t].far, session[s].far);
+                                               session[s].last_packet = time_now;
+                                               LOG(3, s, t, "New session (%u/%u)\n", tunnel[t].far, session[s].far);
                                                control16(c, 14, s, 1); // assigned session
-                                               controladd(c, t, s); // send the reply
-                                               r = radiusfree;
-                                               radiusfree = radius[r].next;
-                                               memset(&radius[r], 0, sizeof(radius[r]));
-                                               session[s].radius = r;
-                                               radius[r].session = s;
+                                               controladd(c, asession, t); // send the reply
+
+                                               strncpy(session[s].called, called, sizeof(session[s].called) - 1);
+                                               strncpy(session[s].calling, calling, sizeof(session[s].calling) - 1);
+
+                                               session[s].ppp.phase = Establish;
+                                               session[s].ppp.lcp = Starting;
+
+                                               STAT(session_created);
+                                               break;
+                                       }
+
+                                       {
+                                               controlt *c = controlnew(14); // CDN
+                                               if (!sessionfree)
                                                {
-                                                       // Generate a random challenge
-                                                       int n;
-                                                       for (n = 0; n < 15; n++)
-                                                               radius[r].auth[n] = rand();
+                                                       STAT(session_overflow);
+                                                       LOG(1, 0, t, "No free sessions\n");
+                                                       control16(c, 1, 4, 0); // temporary lack of resources
                                                }
-                                               strcpy(radius[r].calling, calling);
-                                               strcpy(session[s].called, called);
-                                               strcpy(session[s].calling, calling);
-                                               STAT(session_created);
+                                               else
+                                                       control16(c, 1, 2, 7); // shutting down, try another
+
+                                               controladd(c, asession, t); // send the message
                                        }
-                                       break;
+                                       return;
                                case 11:      // ICRP
                                        // 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
-                                       log(3, ntohl(addr->sin_addr.s_addr), s, t, "Magic %X Flags %X\n", amagic, aflags);
+                                       session[s].mru = PPPoE_MRU; // default
                                        controlnull(t); // ack
-                                       // In CHAP state, request PAP instead
-                                       if (requestchap)
-                                               initlcp(t, s);
+
+                                       // start LCP
+                                       sess_local[s].lcp_authtype = config->radius_authprefer;
+                                       sess_local[s].ppp_mru = MRU;
+
+                                       // Set multilink options before sending initial LCP packet
+                                       sess_local[s].mp_mrru = 1614;
+                                       sess_local[s].mp_epdis = config->bind_address ? config->bind_address : my_address;
+
+                                       sendlcp(s, t);
+                                       change_state(s, lcp, RequestSent);
                                        break;
+
                                case 14:      // CDN
                                        controlnull(t); // ack
-                                       sessionshutdown(s, "Closed (Received CDN)");
+                                       sessionshutdown(s, disc_reason, CDN_NONE, disc_cause);
                                        break;
                                case 0xFFFF:
-                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Missing message type\n");
+                                       LOG(1, s, t, "Missing message type\n");
                                        break;
                                default:
                                        STAT(tunnel_rx_errors);
-                                       if (mandatorymessage & 0x80)
-                                               tunnelshutdown(t, "Unknown message");
+                                       if (mandatory)
+                                               tunnelshutdown(t, "Unknown message type", 2, 6, "unknown message type");
                                        else
-                                               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Unknown message type %d\n", message);
+                                               LOG(1, s, t, "Unknown message type %u\n", message);
                                        break;
                                }
                        if (chapresponse) free(chapresponse);
@@ -1525,19 +2713,14 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                }
                else
                {
-                       log(4, 0, s, t, "   Got a ZLB ack\n");
+                       LOG(4, s, t, "   Got a ZLB ack\n");
                }
        }
        else
        {                          // data
-               u16 prot;
+               uint16_t proto;
 
-               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
-               }
+               LOG_HEX(5, "Receive Tunnel Data", p, l);
                if (l > 2 && p[0] == 0xFF && p[1] == 0x03)
                {                     // HDLC address header, discard
                        p += 2;
@@ -1545,1444 +2728,2959 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                }
                if (l < 2)
                {
-                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Short ppp length %d\n", l);
+                       LOG(1, s, t, "Short ppp length %d\n", l);
                        STAT(tunnel_rx_errors);
                        return;
                }
                if (*p & 1)
                {
-                       prot = *p++;
+                       proto = *p++;
                        l--;
                }
                else
                {
-                       prot = ntohs(*(u16 *) p);
+                       proto = ntohs(*(uint16_t *) p);
                        p += 2;
                        l -= 2;
                }
-               if (prot == PPPPAP)
-               {
-                       session[s].last_packet = time_now;
-                       processpap(t, s, p, l);
-               }
-               else if (prot == PPPCHAP)
+
+               if (s && !session[s].opened)    // Is something wrong??
                {
-                       session[s].last_packet = time_now;
-                       processchap(t, s, p, l);
+                       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, s, t, "UDP packet contains session which is not opened.  Dropping packet.\n");
+                       STAT(tunnel_rx_errors);
+                       return;
+               }
+
+               if (proto == 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(s, t, p, l);
+               }
+               else if (proto == 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(s, t, p, l);
+               }
+               else if (proto == 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(s, t, p, l);
+               }
+               else if (proto == 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(s, t, p, l);
+               }
+               else if (proto == PPPIPV6CP && config->ipv6_prefix.s6_addr[0])
+               {
+                       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; }
+                       processipv6cp(s, t, p, l);
+               }
+               else if (proto == 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(s, t, p, l);
                }
-               else if (prot == PPPLCP)
+               else if (proto == PPPIP)
                {
+                       if (session[s].die)
+                       {
+                               LOG(4, s, t, "Session %u is closing.  Don't process PPP packets\n", s);
+                               return;              // closing session, PPP not processed
+                       }
+
                        session[s].last_packet = time_now;
-                       processlcp(t, s, p, l);
+                       if (session[s].walled_garden && !config->cluster_iam_master)
+                       {
+                               master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                               return;
+                       }
+
+                       processipin(s, t, p, l);
                }
-               else if (prot == PPPIPCP)
+               else if (proto == PPPMP)
                {
+                       if (session[s].die)
+                       {
+                               LOG(4, s, t, "Session %u is closing.  Don't process PPP packets\n", s);
+                               return;              // closing session, PPP not processed
+                       }
+
                        session[s].last_packet = time_now;
-                       processipcp(t, s, p, l);
+                       if (session[s].walled_garden && !config->cluster_iam_master)
+                       {
+                               master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                               return;
+                       }
+
+                       processmpin(s, t, p, l);
                }
-               else if (prot == PPPCCP)
+               else if (proto == PPPIPV6 && config->ipv6_prefix.s6_addr[0])
                {
+                       if (session[s].die)
+                       {
+                               LOG(4, s, t, "Session %u is closing.  Don't process PPP packets\n", s);
+                               return;              // closing session, PPP not processed
+                       }
+
                        session[s].last_packet = time_now;
-                       processccp(t, s, p, l);
+                       if (session[s].walled_garden && !config->cluster_iam_master)
+                       {
+                               master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                               return;
+                       }
+
+                       processipv6in(s, t, p, l);
                }
-               else if (prot == PPPIP)
+               else if (session[s].ppp.lcp == Opened)
                {
                        session[s].last_packet = time_now;
-                       processipin(t, s, p, l);
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
+                       protoreject(s, t, p, l, proto);
                }
                else
                {
-                       STAT(tunnel_rx_errors);
-                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Unknown PPP protocol %04X\n", prot);
+                       LOG(2, s, t, "Unknown PPP protocol 0x%04X received in LCP %s state\n",
+                               proto, ppp_state(session[s].ppp.lcp));
                }
        }
 }
 
-// read and process packet on tap
-void processtap(u8 * buf, int len)
+// read and process packet on tun
+static void processtun(uint8_t * 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(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, "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 (*(uint16_t *) (buf + 2) == htons(PKTIP)) // IPv4
                processipout(buf, len);
-       else
-       {
-               log(1, 0, 0, 0, "Unexpected tap packet %04X, %d bytes\n", ntohs(*(u16 *) (buf + 2)), len);
-       }
+       else if (*(uint16_t *) (buf + 2) == htons(PKTIPV6) // IPV6
+           && config->ipv6_prefix.s6_addr[0])
+               processipv6out(buf, len);
+
+       // Else discard.
 }
 
-// main loop - gets packets on tap or udp and processes them
-void mainloop(void)
+// Handle retries, timeouts.  Runs every 1/10th sec, want to ensure
+// that we look at the whole of the tunnel, radius and session tables
+// every second
+static void regular_cleanups(double period)
 {
-       fd_set cr;
-       int cn;
-       u8 buf[65536];
-       struct timeval to;
-       clockt slow = now();       // occasional functions like session/tunnel expiry, tunnel hello, etc
-       clockt next_acct = slow + ACCT_TIME;
-       clockt next_cluster_ping = slow + 50;
-       to.tv_sec = 1;
-       to.tv_usec = 0;
-       log(4, 0, 0, 0, "Beginning of main loop. udpfd=%d, tapfd=%d, radfd=%d, cluster_sockfd=%d, controlfd=%d\n",
-                       udpfd, tapfd, radfd, cluster_sockfd, controlfd);
+       // Next tunnel, radius and session to check for actions on.
+       static tunnelidt t = 0;
+       static int r = 0;
+       static sessionidt s = 0;
 
-       FD_ZERO(&cr);
-       FD_SET(udpfd, &cr);
-       FD_SET(tapfd, &cr);
-       FD_SET(radfd, &cr);
-       FD_SET(controlfd, &cr);
-#ifdef HAVE_LIBCLI
-       FD_SET(clifd, &cr);
-#endif
-       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;
-#ifdef HAVE_LIBCLI
-       if (cn < clifd) cn = clifd;
-#endif
-       if (cn < cluster_sockfd) cn = cluster_sockfd;
+       int t_actions = 0;
+       int r_actions = 0;
+       int s_actions = 0;
+
+       int t_slice;
+       int r_slice;
+       int s_slice;
+
+       int i;
+       int a;
+
+       // divide up tables into slices based on the last run
+       t_slice = config->cluster_highest_tunnelid  * period;
+       r_slice = (MAXRADIUS - 1)                   * period;
+       s_slice = config->cluster_highest_sessionid * period;
+
+       if (t_slice < 1)
+           t_slice = 1;
+       else if (t_slice > config->cluster_highest_tunnelid)
+           t_slice = config->cluster_highest_tunnelid;
+
+       if (r_slice < 1)
+           r_slice = 1;
+       else if (r_slice > (MAXRADIUS - 1))
+           r_slice = MAXRADIUS - 1;
 
-       while (!main_quit)
+       if (s_slice < 1)
+           s_slice = 1;
+       else if (s_slice > config->cluster_highest_sessionid)
+           s_slice = config->cluster_highest_sessionid;
+
+       LOG(4, 0, 0, "Begin regular cleanup (last %f seconds ago)\n", period);
+
+       for (i = 0; i < t_slice; i++)
        {
-               fd_set r;
-               int n = cn;
-               memcpy(&r, &cr, sizeof(fd_set));
-               n = select(n + 1, &r, 0, 0, &to);
-               if (n < 0)
+               t++;
+               if (t > config->cluster_highest_tunnelid)
+                       t = 1;
+
+               // check for expired tunnels
+               if (tunnel[t].die && tunnel[t].die <= TIME)
                {
-                       if (errno != EINTR)
+                       STAT(tunnel_timeout);
+                       tunnelkill(t, "Expired");
+                       t_actions++;
+                       continue;
+               }
+               // check for message resend
+               if (tunnel[t].retry && tunnel[t].controlc)
+               {
+                       // resend pending messages as timeout on reply
+                       if (tunnel[t].retry <= TIME)
                        {
-                               perror("select");
-                               exit( -1);
+                               controlt *c = tunnel[t].controls;
+                               uint8_t 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;
+                                       }
+
+                               t_actions++;
                        }
                }
-               else if (n)
+               // Send hello
+               if (tunnel[t].state == TUNNELOPEN && !tunnel[t].controlc && (time_now - tunnel[t].lastrec) > 60)
                {
-                       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))
-                               processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr);
-#ifdef HAVE_LIBCLI
-                       else if (FD_ISSET(clifd, &r))
-                       {
-                               struct sockaddr_in addr;
-                               int sockfd;
-                               int len = sizeof(addr);
-
-                               if ((sockfd = accept(clifd, (struct sockaddr *)&addr, &len)) <= 0)
-                               {
-                                       log(0, 0, 0, 0, "accept error: %s\n", strerror(errno));
-                                       continue;
-                               }
-                               else
+                       controlt *c = controlnew(6); // sending HELLO
+                       controladd(c, 0, t); // send the message
+                       LOG(3, 0, t, "Sending HELLO message\n");
+                       t_actions++;
+               }
+
+               // 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, 0, t, "Dropping tunnel by CLI\n");
+                               tunnelshutdown(t, "Requested by administrator", 1, 0, 0);
+                               t_actions++;
+                       }
+               }
+       }
+
+       for (i = 0; i < r_slice; i++)
+       {
+               r++;
+               if (r >= MAXRADIUS)
+                       r = 1;
+
+               if (!radius[r].state)
+                       continue;
+
+               if (radius[r].retry <= TIME)
+               {
+                       radiusretry(r);
+                       r_actions++;
+               }
+       }
+
+       for (i = 0; i < s_slice; i++)
+       {
+               s++;
+               if (s > config->cluster_highest_sessionid)
+                       s = 1;
+
+               if (!session[s].opened) // Session isn't in use
+                       continue;
+
+               // check for expired sessions
+               if (session[s].die)
+               {
+                       if (session[s].die <= TIME)
+                       {
+                               sessionkill(s, "Expired");
+                               s_actions++;
+                       }
+                       continue;
+               }
+
+               // check for timed out sessions
+               if (session[s].timeout)
+               {
+                       bundleidt bid = session[s].bundle;
+                       if (bid)
+                       {
+                               clockt curr_time = time_now;
+                               if (curr_time - bundle[bid].last_check >= 1)
                                {
-                                       cli_do(sockfd);
-                                       close(sockfd);
+                                       bundle[bid].online_time += (curr_time-bundle[bid].last_check)*bundle[bid].num_of_links;
+                                       bundle[bid].last_check = curr_time;
+                                       if (bundle[bid].online_time >= session[s].timeout)
+                                       {
+                                               int ses;
+                                               for (ses = bundle[bid].num_of_links - 1; ses >= 0; ses--)
+                                               {
+                                                       sessionshutdown(bundle[bid].members[ses], "Session timeout", CDN_ADMIN_DISC, TERM_SESSION_TIMEOUT);
+                                                       s_actions++;
+                                                       continue;
+                                               }
+                                       }
                                }
                        }
-#endif
-                       else
+                       else if (session[s].timeout <= time_now - session[s].opened)
                        {
-                               log(1, 0, 0, 0, "Main select() loop returned %d, but no fds have data waiting\n", n);
+                               sessionshutdown(s, "Session timeout", CDN_ADMIN_DISC, TERM_SESSION_TIMEOUT);
+                               s_actions++;
                                continue;
                        }
                }
-               else if (n == 0) {                       // handle timeouts
-                       clockt when = now();
-                       clockt best = when + 100; // default timeout
-                       sessionidt s;
-                       tunnelidt t;
-                       u8 r;
-                       for (r = 0; r < MAXRADIUS; r++)
-                               if (radius[r].state && radius[r].retry)
+
+               // PPP timeouts
+               if (sess_local[s].lcp.restart <= time_now)
+               {
+                       int next_state = session[s].ppp.lcp;
+                       switch (session[s].ppp.lcp)
+                       {
+                       case RequestSent:
+                       case AckReceived:
+                               next_state = RequestSent;
+
+                       case AckSent:
+                               if (sess_local[s].lcp.conf_sent < config->ppp_max_configure)
                                {
-                                       if (radius[r].retry <= when)
-                                               radiusretry(r);
-                                       if (radius[r].retry && radius[r].retry < best)
-                                               best = radius[r].retry;
+                                       LOG(3, s, session[s].tunnel, "No ACK for LCP ConfigReq... resending\n");
+                                       sendlcp(s, session[s].tunnel);
+                                       change_state(s, lcp, next_state);
                                }
-                       for (t = 0; t < MAXTUNNEL; t++)
-                       {
-                               // check for expired tunnels
-                               if (tunnel[t].die && tunnel[t].die <= when)
+                               else
                                {
-                                       STAT(tunnel_timeout);
-                                       tunnelkill(t, "Expired");
-                                       continue;
+                                       sessionshutdown(s, "No response to LCP ConfigReq.", CDN_ADMIN_DISC, TERM_LOST_SERVICE);
+                                       STAT(session_timeout);
                                }
-                               // check for message resend
-                               if (tunnel[t].retry && tunnel[t].controlc)
+
+                               s_actions++;
+                       }
+
+                       if (session[s].die)
+                               continue;
+               }
+
+               if (sess_local[s].ipcp.restart <= time_now)
+               {
+                       int next_state = session[s].ppp.ipcp;
+                       switch (session[s].ppp.ipcp)
+                       {
+                       case RequestSent:
+                       case AckReceived:
+                               next_state = RequestSent;
+
+                       case AckSent:
+                               if (sess_local[s].ipcp.conf_sent < config->ppp_max_configure)
                                {
-                                       // 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;
+                                       LOG(3, s, session[s].tunnel, "No ACK for IPCP ConfigReq... resending\n");
+                                       sendipcp(s, session[s].tunnel);
+                                       change_state(s, ipcp, next_state);
                                }
-                               // Send hello
-                               if (tunnel[t].ip && !tunnel[t].die && tunnel[t].last < when + 600 && !tunnel[t].controlc)
+                               else
                                {
-                                       controlt *c = controlnew(6); // sending HELLO
-                                       controladd(c, t, 0); // send the message
-                                       log(3, tunnel[t].ip, 0, t, "Sending HELLO message\n");
+                                       sessionshutdown(s, "No response to IPCP ConfigReq.", CDN_ADMIN_DISC, TERM_LOST_SERVICE);
+                                       STAT(session_timeout);
                                }
+
+                               s_actions++;
                        }
 
-#ifdef HAVE_LIBCLI
-                       // Check for sessions that have been killed from the CLI
-                       if (cli_session_kill[0])
+                       if (session[s].die)
+                               continue;
+               }
+
+               if (sess_local[s].ipv6cp.restart <= time_now)
+               {
+                       int next_state = session[s].ppp.ipv6cp;
+                       switch (session[s].ppp.ipv6cp)
                        {
-                               int i;
-                               for (i = 0; i < MAXSESSION && cli_session_kill[i]; i++)
+                       case RequestSent:
+                       case AckReceived:
+                               next_state = RequestSent;
+
+                       case AckSent:
+                               if (sess_local[s].ipv6cp.conf_sent < config->ppp_max_configure)
                                {
-                                       log(2, 0, cli_session_kill[i], 0, "Dropping session by CLI\n");
-                                       sessionshutdown(cli_session_kill[i], "Requested by CLI");
-                                       cli_session_kill[i] = 0;
+                                       LOG(3, s, session[s].tunnel, "No ACK for IPV6CP ConfigReq... resending\n");
+                                       sendipv6cp(s, session[s].tunnel);
+                                       change_state(s, ipv6cp, next_state);
                                }
-                       }
-                       // Check for tunnels that have been killed from the CLI
-                       if (cli_tunnel_kill[0])
-                       {
-                               int i;
-                               for (i = 0; i < MAXTUNNEL && cli_tunnel_kill[i]; i++)
+                               else
                                {
-                                       log(2, 0, cli_tunnel_kill[i], 0, "Dropping tunnel by CLI\n");
-                                       tunnelshutdown(cli_tunnel_kill[i], "Requested by CLI");
-                                       cli_tunnel_kill[i] = 0;
+                                       LOG(3, s, session[s].tunnel, "No ACK for IPV6CP ConfigReq\n");
+                                       change_state(s, ipv6cp, Stopped);
                                }
+
+                               s_actions++;
                        }
-#endif
+               }
 
-                       for (s = 0; s < MAXSESSION; s++)
+               if (sess_local[s].ccp.restart <= time_now)
+               {
+                       int next_state = session[s].ppp.ccp;
+                       switch (session[s].ppp.ccp)
                        {
-                               // check for expired sessions
-                               if (session[s].die && session[s].die <= when)
+                       case RequestSent:
+                       case AckReceived:
+                               next_state = RequestSent;
+
+                       case AckSent:
+                               if (sess_local[s].ccp.conf_sent < config->ppp_max_configure)
                                {
-                                       STAT(session_timeout);
-                                       sessionkill(s, "Expired");
-                                       continue;
+                                       LOG(3, s, session[s].tunnel, "No ACK for CCP ConfigReq... resending\n");
+                                       sendccp(s, session[s].tunnel);
+                                       change_state(s, ccp, next_state);
                                }
-
-                               // Drop sessions who have not responded within IDLE_TIMEOUT seconds
-                               if (session[s].user[0] && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
+                               else
                                {
-                                       sessionkill(s, "No response to LCP ECHO requests");
-                                       continue;
+                                       LOG(3, s, session[s].tunnel, "No ACK for CCP ConfigReq\n");
+                                       change_state(s, ccp, Stopped);
                                }
 
-                               // 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);
+                               s_actions++;
+                       }
+               }
 
-                                       *q = EchoReq;
-                                       *(u8 *)(q + 1) = (time_now % 255); // ID
-                                       *(u16 *)(q + 2) = htons(8); // Length
-                                       *(u32 *)(q + 4) = 0; // Magic Number (not supported)
+               // Drop sessions who have not responded within IDLE_TIMEOUT seconds
+               if (session[s].last_packet && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
+               {
+                       sessionshutdown(s, "No response to LCP ECHO requests.", CDN_ADMIN_DISC, TERM_LOST_SERVICE);
+                       STAT(session_timeout);
+                       s_actions++;
+                       continue;
+               }
 
-                                       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;
-                               }
+               // No data in ECHO_TIMEOUT seconds, send LCP ECHO
+               if (session[s].ppp.phase >= Establish && (time_now - session[s].last_packet >= ECHO_TIMEOUT) &&
+                       (time_now - sess_local[s].last_echo >= ECHO_TIMEOUT))
+               {
+                       uint8_t b[MAXETHER];
+
+                       uint8_t *q = makeppp(b, sizeof(b), 0, 0, s, session[s].tunnel, PPPLCP, 1, 0, 0);
+                       if (!q) continue;
+
+                       *q = EchoReq;
+                       *(uint8_t *)(q + 1) = (time_now % 255); // ID
+                       *(uint16_t *)(q + 2) = htons(8); // Length
+                       *(uint32_t *)(q + 4) = session[s].ppp.lcp == Opened ? htonl(session[s].magic) : 0; // Magic Number
+
+                       LOG(4, 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
+                       sess_local[s].last_echo = time_now;
+                       s_actions++;
+               }
+
+               // 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, s, session[s].tunnel, "Dropping session by CLI\n");
+                               sessionshutdown(s, "Requested by administrator.", CDN_ADMIN_DISC, TERM_ADMIN_RESET);
+                               a = 0; // dead, no need to check for other actions
+                               s_actions++;
                        }
-                       if (accounting_dir && next_acct <= when)
+
+                       if (a & CLI_SESS_NOSNOOP)
                        {
-                               // Dump accounting data
-                               next_acct = when + ACCT_TIME;
-                               dump_acct_info();
+                               LOG(2, s, session[s].tunnel, "Unsnooping session by CLI\n");
+                               session[s].snoop_ip = 0;
+                               session[s].snoop_port = 0;
+                               s_actions++;
+                               send++;
+                       }
+                       else if (a & CLI_SESS_SNOOP)
+                       {
+                               LOG(2, s, session[s].tunnel, "Snooping session by CLI (to %s:%u)\n",
+                                   fmtaddr(cli_session_actions[s].snoop_ip, 0),
+                                   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;
+                               s_actions++;
+                               send++;
                        }
 
-                       if (cluster_sockfd && next_cluster_ping <= when)
+                       if (a & CLI_SESS_NOTHROTTLE)
                        {
-                               // Dump accounting data
-                               next_cluster_ping = when + 50;
-                               cluster_send_message(cluster_address, bind_address, C_PING, hostname, strlen(hostname));
+                               LOG(2, s, session[s].tunnel, "Un-throttling session by CLI\n");
+                               throttle_session(s, 0, 0);
+                               s_actions++;
+                               send++;
+                       }
+                       else if (a & CLI_SESS_THROTTLE)
+                       {
+                               LOG(2, 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);
+                               s_actions++;
+                               send++;
+                       }
+
+                       if (a & CLI_SESS_NOFILTER)
+                       {
+                               LOG(2, s, session[s].tunnel, "Un-filtering session by CLI\n");
+                               filter_session(s, 0, 0);
+                               s_actions++;
+                               send++;
+                       }
+                       else if (a & CLI_SESS_FILTER)
+                       {
+                               LOG(2, s, session[s].tunnel, "Filtering session by CLI (in=%d, out=%d)\n",
+                                   cli_session_actions[s].filter_in,
+                                   cli_session_actions[s].filter_out);
+
+                               filter_session(s, cli_session_actions[s].filter_in, cli_session_actions[s].filter_out);
+                               s_actions++;
+                               send++;
+                       }
+
+                       if (send)
+                               cluster_send_session(s);
+               }
+
+               // RADIUS interim accounting
+               if (config->radius_accounting && config->radius_interim > 0
+                   && session[s].ip && !session[s].walled_garden
+                   && !sess_local[s].radius // RADIUS already in progress
+                   && time_now - sess_local[s].last_interim >= config->radius_interim)
+               {
+                       int rad = radiusnew(s);
+                       if (!rad)
+                       {
+                               LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Interim message\n");
+                               STAT(radius_overflow);
+                               continue;
                        }
 
-                       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));
+                       LOG(3, s, session[s].tunnel, "Sending RADIUS Interim for %s (%u)\n",
+                               session[s].user, session[s].unique_id);
+
+                       radiussend(rad, RADIUSINTERIM);
+                       sess_local[s].last_interim = time_now;
+                       s_actions++;
                }
        }
+
+       LOG(4, 0, 0, "End regular cleanup: checked %d/%d/%d tunnels/radius/sessions; %d/%d/%d actions\n",
+               t_slice, r_slice, s_slice, t_actions, r_actions, s_actions);
 }
 
-// Init data structures
-void initdata(void)
+//
+// Are we in the middle of a tunnel update, or radius
+// requests??
+//
+static int still_busy(void)
 {
        int i;
+       static clockt last_talked = 0;
+       static clockt start_busy_wait = 0;
 
-       _statistics = mmap(NULL, sizeof(struct Tstats), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (_statistics <= 0)
+       if (!config->cluster_iam_master)
        {
-               log(0, 0, 0, 0, "Error doing mmap for _statistics: %s\n", strerror(errno));
-               exit(1);
+#ifdef BGP
+               static time_t stopped_bgp = 0;
+               if (bgp_configured)
+               {
+                       if (!stopped_bgp)
+                       {
+                               LOG(1, 0, 0, "Shutting down in %d seconds, stopping BGP...\n", QUIT_DELAY);
+
+                               for (i = 0; i < BGP_NUM_PEERS; i++)
+                                       if (bgp_peers[i].state == Established)
+                                               bgp_stop(&bgp_peers[i]);
+
+                               stopped_bgp = time_now;
+
+                               // we don't want to become master
+                               cluster_send_ping(0);
+
+                               return 1;
+                       }
+
+                       if (time_now < (stopped_bgp + QUIT_DELAY))
+                               return 1;
+               }
+#endif /* BGP */
+
+               return 0;
        }
-       tunnel = mmap(NULL, sizeof(tunnelt) * MAXTUNNEL, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (tunnel <= 0)
+
+       if (main_quit == QUIT_SHUTDOWN)
        {
-               log(0, 0, 0, 0, "Error doing mmap for tunnels: %s\n", strerror(errno));
-               exit(1);
+               static int dropped = 0;
+               if (!dropped)
+               {
+                       int i;
+
+                       LOG(1, 0, 0, "Dropping sessions and tunnels\n");
+                       for (i = 1; i < MAXTUNNEL; i++)
+                               if (tunnel[i].ip || tunnel[i].state)
+                                       tunnelshutdown(i, "L2TPNS Closing", 6, 0, 0);
+
+                       dropped = 1;
+               }
        }
-       session = mmap(NULL, sizeof(sessiont) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (session <= 0)
+
+       if (start_busy_wait == 0)
+               start_busy_wait = TIME;
+
+       for (i = config->cluster_highest_tunnelid ; i > 0 ; --i)
        {
-               log(0, 0, 0, 0, "Error doing mmap for sessions: %s\n", strerror(errno));
-               exit(1);
+               if (!tunnel[i].controlc)
+                       continue;
+
+               if (last_talked != TIME)
+               {
+                       LOG(2, 0, 0, "Tunnel %u still has un-acked control messages.\n", i);
+                       last_talked = TIME;
+               }
+               return 1;
        }
-       radius = mmap(NULL, sizeof(radiust) * MAXRADIUS, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (radius <= 0)
+
+       // We stop waiting for radius after BUSY_WAIT_TIME 1/10th seconds
+       if (abs(TIME - start_busy_wait) > BUSY_WAIT_TIME)
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
-               exit(1);
+               LOG(1, 0, 0, "Giving up waiting for RADIUS to be empty.  Shutting down anyway.\n");
+               return 0;
        }
-       ip_address_pool = mmap(NULL, sizeof(ippoolt) * MAXIPPOOL, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (ip_address_pool <= 0)
+
+       for (i = 1; i < MAXRADIUS; i++)
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
-               exit(1);
+               if (radius[i].state == RADIUSNULL)
+                       continue;
+               if (radius[i].state == RADIUSWAIT)
+                       continue;
+
+               if (last_talked != TIME)
+               {
+                       LOG(2, 0, 0, "Radius session %u is still busy (sid %u)\n", i, radius[i].session);
+                       last_talked = TIME;
+               }
+               return 1;
        }
-#ifdef RINGBUFFER
-       ringbuffer = mmap(NULL, sizeof(struct Tringbuffer), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (ringbuffer <= 0)
+
+       return 0;
+}
+
+#ifdef HAVE_EPOLL
+# include <sys/epoll.h>
+#else
+# define FAKE_EPOLL_IMPLEMENTATION /* include the functions */
+# include "fake_epoll.h"
+#endif
+
+// the base set of fds polled: cli, cluster, tun, udp, control, dae
+#define BASE_FDS       6
+
+// additional polled fds
+#ifdef BGP
+# define EXTRA_FDS     BGP_NUM_PEERS
+#else
+# define EXTRA_FDS     0
+#endif
+
+// main loop - gets packets on tun or udp and processes them
+static void mainloop(void)
+{
+       int i;
+       uint8_t buf[65536];
+       clockt next_cluster_ping = 0;   // send initial ping immediately
+       struct epoll_event events[BASE_FDS + RADIUS_FDS + EXTRA_FDS];
+       int maxevent = sizeof(events)/sizeof(*events);
+
+       if ((epollfd = epoll_create(maxevent)) < 0)
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               LOG(0, 0, 0, "epoll_create failed: %s\n", strerror(errno));
                exit(1);
        }
-       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
-#endif
 
-#ifdef HAVE_LIBCLI
-       cli_session_kill = mmap(NULL, sizeof(sessionidt) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (cli_session_kill <= 0)
+       LOG(4, 0, 0, "Beginning of main loop.  clifd=%d, cluster_sockfd=%d, tunfd=%d, udpfd=%d, controlfd=%d, daefd=%d\n",
+               clifd, cluster_sockfd, tunfd, udpfd, controlfd, daefd);
+
+       /* setup our fds to poll for input */
+       {
+               static struct event_data d[BASE_FDS];
+               struct epoll_event e;
+
+               e.events = EPOLLIN;
+               i = 0;
+
+               if (clifd >= 0)
+               {
+                       d[i].type = FD_TYPE_CLI;
+                       e.data.ptr = &d[i++];
+                       epoll_ctl(epollfd, EPOLL_CTL_ADD, clifd, &e);
+               }
+
+               d[i].type = FD_TYPE_CLUSTER;
+               e.data.ptr = &d[i++];
+               epoll_ctl(epollfd, EPOLL_CTL_ADD, cluster_sockfd, &e);
+
+               d[i].type = FD_TYPE_TUN;
+               e.data.ptr = &d[i++];
+               epoll_ctl(epollfd, EPOLL_CTL_ADD, tunfd, &e);
+
+               d[i].type = FD_TYPE_UDP;
+               e.data.ptr = &d[i++];
+               epoll_ctl(epollfd, EPOLL_CTL_ADD, udpfd, &e);
+
+               d[i].type = FD_TYPE_CONTROL;
+               e.data.ptr = &d[i++];
+               epoll_ctl(epollfd, EPOLL_CTL_ADD, controlfd, &e);
+
+               d[i].type = FD_TYPE_DAE;
+               e.data.ptr = &d[i++];
+               epoll_ctl(epollfd, EPOLL_CTL_ADD, daefd, &e);
+       }
+
+#ifdef BGP
+       signal(SIGPIPE, SIG_IGN);
+       bgp_setup(config->as_number);
+       if (config->bind_address)
+               bgp_add_route(config->bind_address, 0xffffffff);
+
+       for (i = 0; i < BGP_NUM_PEERS; i++)
+       {
+               if (config->neighbour[i].name[0])
+                       bgp_start(&bgp_peers[i], config->neighbour[i].name,
+                               config->neighbour[i].as, config->neighbour[i].keepalive,
+                               config->neighbour[i].hold, 0); /* 0 = routing disabled */
+       }
+#endif /* BGP */
+
+       while (!main_quit || still_busy())
+       {
+               int more = 0;
+               int n;
+
+
+               if (main_reload)
+               {
+                       main_reload = 0;
+                       read_config_file();
+                       config->reload_config++;
+               }
+
+               if (config->reload_config)
+               {
+                       config->reload_config = 0;
+                       update_config();
+               }
+
+#ifdef BGP
+               bgp_set_poll();
+#endif /* BGP */
+
+               n = epoll_wait(epollfd, events, maxevent, 100); // timeout 100ms (1/10th sec)
+               STAT(select_called);
+
+               TIME = now(NULL);
+               if (n < 0)
+               {
+                       if (errno == EINTR ||
+                           errno == ECHILD) // EINTR was clobbered by sigchild_handler()
+                               continue;
+
+                       LOG(0, 0, 0, "Error returned from select(): %s\n", strerror(errno));
+                       break; // exit
+               }
+
+               if (n)
+               {
+                       struct sockaddr_in addr;
+                       struct in_addr local;
+                       socklen_t alen;
+                       int c, s;
+                       int udp_ready = 0;
+                       int tun_ready = 0;
+                       int cluster_ready = 0;
+                       int udp_pkts = 0;
+                       int tun_pkts = 0;
+                       int cluster_pkts = 0;
+#ifdef BGP
+                       uint32_t bgp_events[BGP_NUM_PEERS];
+                       memset(bgp_events, 0, sizeof(bgp_events));
+#endif /* BGP */
+
+                       for (c = n, i = 0; i < c; i++)
+                       {
+                               struct event_data *d = events[i].data.ptr;
+
+                               switch (d->type)
+                               {
+                               case FD_TYPE_CLI: // CLI connections
+                               {
+                                       int cli;
+                                       
+                                       alen = sizeof(addr);
+                                       if ((cli = accept(clifd, (struct sockaddr *)&addr, &alen)) >= 0)
+                                       {
+                                               cli_do(cli);
+                                               close(cli);
+                                       }
+                                       else
+                                               LOG(0, 0, 0, "accept error: %s\n", strerror(errno));
+
+                                       n--;
+                                       break;
+                               }
+
+                               // these are handled below, with multiple interleaved reads
+                               case FD_TYPE_CLUSTER:   cluster_ready++; break;
+                               case FD_TYPE_TUN:       tun_ready++; break;
+                               case FD_TYPE_UDP:       udp_ready++; break;
+
+                               case FD_TYPE_CONTROL: // nsctl commands
+                                       alen = sizeof(addr);
+                                       s = recvfromto(controlfd, buf, sizeof(buf), MSG_WAITALL, (struct sockaddr *) &addr, &alen, &local);
+                                       if (s > 0) processcontrol(buf, s, &addr, alen, &local);
+                                       n--;
+                                       break;
+
+                               case FD_TYPE_DAE: // DAE requests
+                                       alen = sizeof(addr);
+                                       s = recvfromto(daefd, buf, sizeof(buf), MSG_WAITALL, (struct sockaddr *) &addr, &alen, &local);
+                                       if (s > 0) processdae(buf, s, &addr, alen, &local);
+                                       n--;
+                                       break;
+
+                               case FD_TYPE_RADIUS: // RADIUS response
+                                       alen = sizeof(addr);
+                                       s = recvfrom(radfds[d->index], buf, sizeof(buf), MSG_WAITALL, (struct sockaddr *) &addr, &alen);
+                                       if (s >= 0 && config->cluster_iam_master)
+                                       {
+                                               if (addr.sin_addr.s_addr == config->radiusserver[0] ||
+                                                   addr.sin_addr.s_addr == config->radiusserver[1])
+                                                       processrad(buf, s, d->index);
+                                               else
+                                                       LOG(3, 0, 0, "Dropping RADIUS packet from unknown source %s\n",
+                                                               fmtaddr(addr.sin_addr.s_addr, 0));
+                                       }
+
+                                       n--;
+                                       break;
+
+#ifdef BGP
+                               case FD_TYPE_BGP:
+                                       bgp_events[d->index] = events[i].events;
+                                       n--;
+                                       break;
+#endif /* BGP */
+
+                               default:
+                                       LOG(0, 0, 0, "Unexpected fd type returned from epoll_wait: %d\n", d->type);
+                               }
+                       }
+
+#ifdef BGP
+                       bgp_process(bgp_events);
+#endif /* BGP */
+
+                       for (c = 0; n && c < config->multi_read_count; c++)
+                       {
+                               // L2TP
+                               if (udp_ready)
+                               {
+                                       alen = sizeof(addr);
+                                       if ((s = recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen)) > 0)
+                                       {
+                                               processudp(buf, s, &addr);
+                                               udp_pkts++;
+                                       }
+                                       else
+                                       {
+                                               udp_ready = 0;
+                                               n--;
+                                       }
+                               }
+
+                               // incoming IP
+                               if (tun_ready)
+                               {
+                                       if ((s = read(tunfd, buf, sizeof(buf))) > 0)
+                                       {
+                                               processtun(buf, s);
+                                               tun_pkts++;
+                                       }
+                                       else
+                                       {
+                                               tun_ready = 0;
+                                               n--;
+                                       }
+                               }
+
+                               // cluster
+                               if (cluster_ready)
+                               {
+                                       alen = sizeof(addr);
+                                       if ((s = recvfrom(cluster_sockfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen)) > 0)
+                                       {
+                                               processcluster(buf, s, addr.sin_addr.s_addr);
+                                               cluster_pkts++;
+                                       }
+                                       else
+                                       {
+                                               cluster_ready = 0;
+                                               n--;
+                                       }
+                               }
+                       }
+
+                       if (udp_pkts > 1 || tun_pkts > 1 || cluster_pkts > 1)
+                               STAT(multi_read_used);
+
+                       if (c >= config->multi_read_count)
+                       {
+                               LOG(3, 0, 0, "Reached multi_read_count (%d); processed %d udp, %d tun and %d cluster packets\n",
+                                       config->multi_read_count, udp_pkts, tun_pkts, cluster_pkts);
+
+                               STAT(multi_read_exceeded);
+                               more++;
+                       }
+               }
+
+               if (time_changed)
+               {
+                       double Mbps = 1024.0 * 1024.0 / 8 * time_changed;
+
+                       // 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:%u OUT:%u",
+                               (udp_rx / Mbps), (eth_tx / Mbps), (eth_rx / Mbps), (udp_tx / Mbps),
+                               ((udp_tx + udp_rx + eth_tx + eth_rx) / Mbps),
+                               udp_rx_pkt / time_changed, eth_rx_pkt / time_changed);
+                
+                       udp_tx = udp_rx = 0;
+                       udp_rx_pkt = eth_rx_pkt = 0;
+                       eth_tx = eth_rx = 0;
+                       time_changed = 0;
+                
+                       if (config->dump_speed)
+                               printf("%s\n", config->bandwidth);
+                
+                       // Update the internal time counter
+                       strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
+                
+                       {
+                               // Run timer hooks
+                               struct param_timer p = { time_now };
+                               run_plugins(PLUGIN_TIMER, &p);
+                       }
+               }
+
+                       // Runs on every machine (master and slaves).
+               if (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
+                               next_cluster_ping = TIME + config->cluster_hb_interval;
+               }
+
+               if (!config->cluster_iam_master)
+                       continue;
+
+                       // Run token bucket filtering queue..
+                       // Only run it every 1/10th of a second.
+               {
+                       static clockt last_run = 0;
+                       if (last_run != TIME)
+                       {
+                               last_run = TIME;
+                               tbf_run_timer();
+                       }
+               }
+
+                       // Handle timeouts, retries etc.
+               {
+                       static double last_clean = 0;
+                       double this_clean;
+                       double diff;
+
+                       TIME = now(&this_clean);
+                       diff = this_clean - last_clean;
+
+                       // Run during idle time (after we've handled
+                       // all incoming packets) or every 1/10th sec
+                       if (!more || diff > 0.1)
+                       {
+                               regular_cleanups(diff);
+                               last_clean = this_clean;
+                       }
+               }
+
+               if (*config->accounting_dir)
+               {
+                       static clockt next_acct = 0;
+                       static clockt next_shut_acct = 0;
+
+                       if (next_acct <= TIME)
+                       {
+                               // Dump accounting data
+                               next_acct = TIME + ACCT_TIME;
+                               next_shut_acct = TIME + ACCT_SHUT_TIME;
+                               dump_acct_info(1);
+                       }
+                       else if (next_shut_acct <= TIME)
+                       {
+                               // Dump accounting data for shutdown sessions
+                               next_shut_acct = TIME + ACCT_SHUT_TIME;
+                               if (shut_acct_n)
+                                       dump_acct_info(0);
+                       }
+               }
+       }
+
+               // Are we the master and shutting down??
+       if (config->cluster_iam_master)
+               cluster_heartbeat(); // Flush any queued changes..
+
+               // Ok. Notify everyone we're shutting down. If we're
+               // the master, this will force an election.
+       cluster_send_ping(0);
+
+       //
+       // Important!!! We MUST not process any packets past this point!
+       LOG(1, 0, 0, "Shutdown complete\n");
+}
+
+static void stripdomain(char *host)
+{
+       char *p;
+
+       if ((p = strchr(host, '.')))
+       {
+               char *domain = 0;
+               char _domain[1024];
+
+               // strip off domain
+               FILE *resolv = fopen("/etc/resolv.conf", "r");
+               if (resolv)
+               {
+                       char buf[1024];
+                       char *b;
+
+                       while (fgets(buf, sizeof(buf), resolv))
+                       {
+                               if (strncmp(buf, "domain", 6) && strncmp(buf, "search", 6))
+                                       continue;
+
+                               if (!isspace(buf[6]))
+                                       continue;
+
+                               b = buf + 7;
+                               while (isspace(*b)) b++;
+
+                               if (*b)
+                               {
+                                       char *d = b;
+                                       while (*b && !isspace(*b)) b++;
+                                       *b = 0;
+                                       if (buf[0] == 'd') // domain is canonical
+                                       {
+                                               domain = d;
+                                               break;
+                                       }
+
+                                       // first search line
+                                       if (!domain)
+                                       {
+                                               // hold, may be subsequent domain line
+                                               strncpy(_domain, d, sizeof(_domain))[sizeof(_domain)-1] = 0;
+                                               domain = _domain;
+                                       }
+                               }
+                       }
+
+                       fclose(resolv);
+               }
+
+               if (domain)
+               {
+                       int hl = strlen(host);
+                       int dl = strlen(domain);
+                       if (dl < hl && host[hl - dl - 1] == '.' && !strcmp(host + hl - dl, domain))
+                               host[hl -dl - 1] = 0;
+               }
+               else
+               {
+                       *p = 0; // everything after first dot
+               }
+       }
+}
+
+// Init data structures
+static void initdata(int optdebug, char *optconfig)
+{
+       int i;
+
+       if (!(config = shared_malloc(sizeof(configt))))
+       {
+               fprintf(stderr, "Error doing malloc for configuration: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       memset(config, 0, sizeof(configt));
+       time(&config->start_time);
+       strncpy(config->config_file, optconfig, strlen(optconfig));
+       config->debug = optdebug;
+       config->num_tbfs = MAXTBFS;
+       config->rl_rate = 28; // 28kbps
+       config->cluster_mcast_ttl = 1;
+       config->cluster_master_min_adv = 1;
+       config->ppp_restart_time = 3;
+       config->ppp_max_configure = 10;
+       config->ppp_max_failure = 5;
+       strcpy(config->random_device, RANDOMDEVICE);
+
+       log_stream = stderr;
+
+#ifdef RINGBUFFER
+       if (!(ringbuffer = shared_malloc(sizeof(struct Tringbuffer))))
+       {
+               LOG(0, 0, 0, "Error doing malloc for ringbuffer: %s\n", strerror(errno));
+               exit(1);
+       }
+       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
+#endif
+
+       if (!(_statistics = shared_malloc(sizeof(struct Tstats))))
+       {
+               LOG(0, 0, 0, "Error doing malloc for _statistics: %s\n", strerror(errno));
+               exit(1);
+       }
+       if (!(tunnel = shared_malloc(sizeof(tunnelt) * MAXTUNNEL)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for tunnels: %s\n", strerror(errno));
+               exit(1);
+       }
+       if (!(bundle = shared_malloc(sizeof(bundlet) * MAXBUNDLE)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for bundles: %s\n", strerror(errno));
+               exit(1);
+       }
+       if (!(frag = shared_malloc(sizeof(fragmentationt) * MAXBUNDLE)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for fragmentations: %s\n", strerror(errno));
+               exit(1);
+       }
+       if (!(session = shared_malloc(sizeof(sessiont) * MAXSESSION)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for sessions: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       if (!(sess_local = shared_malloc(sizeof(sessionlocalt) * MAXSESSION)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for sess_local: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       if (!(radius = shared_malloc(sizeof(radiust) * MAXRADIUS)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for radius: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       if (!(ip_address_pool = shared_malloc(sizeof(ippoolt) * MAXIPPOOL)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for ip_address_pool: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       if (!(ip_filters = shared_malloc(sizeof(ip_filtert) * MAXFILTER)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for ip_filters: %s\n", strerror(errno));
+               exit(1);
+       }
+       memset(ip_filters, 0, sizeof(ip_filtert) * MAXFILTER);
+
+       if (!(cli_session_actions = shared_malloc(sizeof(struct cli_session_actions) * MAXSESSION)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for cli session actions: %s\n", strerror(errno));
+               exit(1);
+       }
+       memset(cli_session_actions, 0, sizeof(struct cli_session_actions) * MAXSESSION);
+
+       if (!(cli_tunnel_actions = shared_malloc(sizeof(struct cli_tunnel_actions) * MAXSESSION)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for cli tunnel actions: %s\n", strerror(errno));
+               exit(1);
+       }
+       memset(cli_tunnel_actions, 0, sizeof(struct cli_tunnel_actions) * MAXSESSION);
+
+       memset(tunnel, 0, sizeof(tunnelt) * MAXTUNNEL);
+       memset(bundle, 0, sizeof(bundlet) * MAXBUNDLE);
+       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; 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; i++)
+               tunnel[i].state = TUNNELUNDEF;  // mark it as not filled in.
+
+       for (i = 1; i < MAXBUNDLE; i++) {
+               bundle[i].state = BUNDLEUNDEF;
+       }
+
+       if (!*hostname)
+       {
+               // Grab my hostname unless it's been specified
+               gethostname(hostname, sizeof(hostname));
+               stripdomain(hostname);
+       }
+
+       _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, "Error doing malloc for bgp: %s\n", strerror(errno));
+               exit(1);
+       }
+#endif /* BGP */
+}
+
+static int assign_ip_address(sessionidt s)
+{
+       uint32_t i;
+       int best = -1;
+       time_t best_time = time_now;
+       char *u = session[s].user;
+       char reuse = 0;
+
+
+       CSTAT(assign_ip_address);
+
+       for (i = 1; i < ip_pool_size; i++)
+       {
+               if (!ip_address_pool[i].address || ip_address_pool[i].assigned)
+                       continue;
+
+               if (!session[s].walled_garden && ip_address_pool[i].user[0] && !strcmp(u, ip_address_pool[i].user))
+               {
+                       best = i;
+                       reuse = 1;
+                       break;
+               }
+
+               if (ip_address_pool[i].last < best_time)
+               {
+                       best = i;
+                       if (!(best_time = ip_address_pool[i].last))
+                               break; // never used, grab this one
+               }
+       }
+
+       if (best < 0)
+       {
+               LOG(0, s, session[s].tunnel, "assign_ip_address(): out of addresses\n");
+               return 0;
+       }
+
+       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
+                  un-gardened). */
+               ip_address_pool[best].user[0] = 0;
+       else
+               strncpy(ip_address_pool[best].user, u, sizeof(ip_address_pool[best].user) - 1);
+
+       STAT(ip_allocated);
+       LOG(4, s, session[s].tunnel, "assign_ip_address(): %s ip address %d from pool\n",
+               reuse ? "Reusing" : "Allocating", best);
+
+       return 1;
+}
+
+static void free_ip_address(sessionidt s)
+{
+       int i = session[s].ip_pool_index;
+
+
+       CSTAT(free_ip_address);
+
+       if (!session[s].ip)
+               return; // what the?
+
+       if (i < 0)      // Is this actually part of the ip pool?
+               i = 0;
+
+       STAT(ip_freed);
+       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;
+}
+
+//
+// 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].opened && session[i].ip))
+                       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, i, 0, "Session %u has an IP address (%s) that was marked static, but is in the pool (%d)!\n",
+                               i, fmtaddr(session[i].ip, 0), ipid);
+
+                       // Fall through and process it as part of the pool.
+               }
+
+
+               if (ipid > MAXIPPOOL || ipid < 0)
+               {
+                       LOG(0, i, 0, "Session %u 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).
+static 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.
+//
+static void add_to_ip_pool(in_addr_t addr, in_addr_t 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, "Overflowed IP pool adding %s\n", fmtaddr(htonl(addr), 0));
+                       return;
+               }
+       }
+}
+
+// Initialize the IP address pool
+static void initippool()
+{
+       FILE *f;
+       char *p;
+       char buf[4096];
+       memset(ip_address_pool, 0, sizeof(ip_address_pool));
+
+       if (!(f = fopen(IPPOOLFILE, "r")))
+       {
+               LOG(0, 0, 0, "Can't load pool file " IPPOOLFILE ": %s\n", strerror(errno));
+               exit(1);
+       }
+
+       while (ip_pool_size < MAXIPPOOL && fgets(buf, 4096, f))
+       {
+               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;
+               if ((p = (char *)strchr(buf, ':')))
+               {
+                       in_addr_t src;
+                       *p = '\0';
+                       src = inet_addr(buf);
+                       if (src == INADDR_NONE)
+                       {
+                               LOG(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)
+                               continue;
+                       *p = ':';
+                       pool = p+1;
+               }
+               if ((p = (char *)strchr(pool, '/')))
+               {
+                       // It's a range
+                       int numbits = 0;
+                       in_addr_t start = 0, mask = 0;
+
+                       LOG(2, 0, 0, "Adding IP address range %s\n", buf);
+                       *p++ = 0;
+                       if (!*p || !(numbits = atoi(p)))
+                       {
+                               LOG(0, 0, 0, "Invalid pool range %s\n", buf);
+                               continue;
+                       }
+                       start = ntohl(inet_addr(pool));
+                       mask = (in_addr_t) (pow(2, numbits) - 1) << (32 - numbits);
+
+                       // Add a static route for this pool
+                       LOG(5, 0, 0, "Adding route for address pool %s/%u\n",
+                               fmtaddr(htonl(start), 0), 32 + mask);
+
+                       routeset(0, start, mask, 0, 1);
+
+                       add_to_ip_pool(start, mask);
+               }
+               else
+               {
+                       // It's a single ip address
+                       add_to_ip_pool(ntohl(inet_addr(pool)), 0);
+               }
+       }
+       fclose(f);
+       LOG(1, 0, 0, "IP address pool is %d addresses\n", ip_pool_size - 1);
+}
+
+void snoop_send_packet(uint8_t *packet, uint16_t size, in_addr_t destination, uint16_t port)
+{
+       struct sockaddr_in snoop_addr = {0};
+       if (!destination || !port || snoopfd <= 0 || size <= 0 || !packet)
+               return;
+
+       snoop_addr.sin_family = AF_INET;
+       snoop_addr.sin_addr.s_addr = destination;
+       snoop_addr.sin_port = ntohs(port);
+
+       LOG(5, 0, 0, "Snooping %d byte packet to %s:%u\n", size,
+               fmtaddr(snoop_addr.sin_addr.s_addr, 0),
+               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, "Error sending intercept packet: %s\n", strerror(errno));
+
+       STAT(packets_snooped);
+}
+
+static int dump_session(FILE **f, sessiont *s)
+{
+       if (!s->opened || !s->ip || !(s->cin_delta || s->cout_delta) || !*s->user || s->walled_garden)
+               return 1;
+
+       if (!*f)
+       {
+               char filename[1024];
+               char timestr[64];
+               time_t now = time(NULL);
+
+               strftime(timestr, sizeof(timestr), "%Y%m%d%H%M%S", localtime(&now));
+               snprintf(filename, sizeof(filename), "%s/%s", config->accounting_dir, timestr);
+
+               if (!(*f = fopen(filename, "w")))
+               {
+                       LOG(0, 0, 0, "Can't write accounting info to %s: %s\n", filename, strerror(errno));
+                       return 0;
+               }
+
+               LOG(3, 0, 0, "Dumping accounting information to %s\n", filename);
+               fprintf(*f, "# dslwatch.pl dump file V1.01\n"
+                       "# host: %s\n"
+                       "# endpoint: %s\n"
+                       "# time: %ld\n"
+                       "# uptime: %ld\n"
+                       "# format: username ip qos uptxoctets downrxoctets\n",
+                       hostname,
+                       fmtaddr(config->bind_address ? config->bind_address : my_address, 0),
+                       now,
+                       now - basetime);
+       }
+
+       LOG(4, 0, 0, "Dumping accounting information for %s\n", s->user);
+       fprintf(*f, "%s %s %d %u %u\n",
+               s->user,                                                // username
+               fmtaddr(htonl(s->ip), 0),                               // ip
+               (s->throttle_in || s->throttle_out) ? 2 : 1,            // qos
+               (uint32_t) s->cin_delta,                                // uptxoctets
+               (uint32_t) s->cout_delta);                              // downrxoctets
+
+       s->cin_delta = s->cout_delta = 0;
+
+       return 1;
+}
+
+static void dump_acct_info(int all)
+{
+       int i;
+       FILE *f = NULL;
+
+
+       CSTAT(dump_acct_info);
+
+       if (shut_acct_n)
+       {
+               for (i = 0; i < shut_acct_n; i++)
+                       dump_session(&f, &shut_acct[i]);
+
+               shut_acct_n = 0;
+       }
+
+       if (all)
+               for (i = 1; i <= config->cluster_highest_sessionid; i++)
+                       dump_session(&f, &session[i]);
+
+       if (f)
+               fclose(f);
+}
+
+// Main program
+int main(int argc, char *argv[])
+{
+       int i;
+       int optdebug = 0;
+       char *optconfig = CONFIGFILE;
+
+       time(&basetime);             // start clock
+
+       // scan args
+       while ((i = getopt(argc, argv, "dvc:h:")) >= 0)
+       {
+               switch (i)
+               {
+               case 'd':
+                       if (fork()) exit(0);
+                       setsid();
+                       freopen("/dev/null", "r", stdin);
+                       freopen("/dev/null", "w", stdout);
+                       freopen("/dev/null", "w", stderr);
+                       break;
+               case 'v':
+                       optdebug++;
+                       break;
+               case 'c':
+                       optconfig = optarg;
+                       break;
+               case 'h':
+                       snprintf(hostname, sizeof(hostname), "%s", optarg);
+                       break;
+               default:
+                       printf("Args are:\n"
+                              "\t-d\t\tDetach from terminal\n"
+                              "\t-c <file>\tConfig file\n"
+                              "\t-h <hostname>\tForce hostname\n"
+                              "\t-v\t\tDebug\n");
+
+                       return (0);
+                       break;
+               }
+       }
+
+       // Start the timer routine off
+       time(&time_now);
+       strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
+
+       initplugins();
+       initdata(optdebug, optconfig);
+
+       init_cli(hostname);
+       read_config_file();
+       update_config();
+       init_tbf(config->num_tbfs);
+
+       LOG(0, 0, 0, "L2TPNS version " VERSION "\n");
+       LOG(0, 0, 0, "Copyright (c) 2003, 2004, 2005, 2006 Optus Internet Engineering\n");
+       LOG(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, "Can't set ulimit: %s\n", strerror(errno));
+
+               // Make core dumps go to /tmp
+               chdir("/tmp");
+       }
+
+       if (config->scheduler_fifo)
+       {
+               int ret;
+               struct sched_param params = {0};
+               params.sched_priority = 1;
+
+               if (get_nprocs() < 2)
+               {
+                       LOG(0, 0, 0, "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, "Using FIFO scheduler.  Say goodbye to any other processes running\n");
+                       }
+                       else
+                       {
+                               LOG(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() < 0)
+               exit(1);
+
+       inittun();
+       LOG(1, 0, 0, "Set up on interface %s\n", config->tundevice);
+
+       initudp();
+       initrad();
+       initippool();
+
+       // seed prng
+       {
+               unsigned seed = time_now ^ getpid();
+               LOG(4, 0, 0, "Seeding the pseudo random generator: %u\n", seed);
+               srand(seed);
+       }
+
+       signal(SIGHUP,  sighup_handler);
+       signal(SIGCHLD, sigchild_handler);
+       signal(SIGTERM, shutdown_handler);
+       signal(SIGINT,  shutdown_handler);
+       signal(SIGQUIT, shutdown_handler);
+
+       // Prevent us from getting paged out
+       if (config->lock_pages)
+       {
+               if (!mlockall(MCL_CURRENT))
+                       LOG(1, 0, 0, "Locking pages into memory\n");
+               else
+                       LOG(0, 0, 0, "Can't lock pages: %s\n", strerror(errno));
+       }
+
+       mainloop();
+
+       /* 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;
+}
+
+static void sighup_handler(int sig)
+{
+       main_reload++;
+}
+
+static void shutdown_handler(int sig)
+{
+       main_quit = (sig == SIGQUIT) ? QUIT_SHUTDOWN : QUIT_FAILOVER;
+}
+
+static void sigchild_handler(int sig)
+{
+       while (waitpid(-1, NULL, WNOHANG) > 0)
+           ;
+}
+
+static void build_chap_response(uint8_t *challenge, uint8_t id, uint16_t challenge_length, uint8_t **challenge_response)
+{
+       MD5_CTX ctx;
+       *challenge_response = NULL;
+
+       if (!*config->l2tp_secret)
+       {
+               LOG(0, 0, 0, "LNS requested CHAP authentication, but no l2tp secret is defined\n");
+               return;
+       }
+
+       LOG(4, 0, 0, "   Building challenge response for CHAP request\n");
+
+       *challenge_response = calloc(17, 1);
+
+       MD5_Init(&ctx);
+       MD5_Update(&ctx, &id, 1);
+       MD5_Update(&ctx, config->l2tp_secret, strlen(config->l2tp_secret));
+       MD5_Update(&ctx, challenge, challenge_length);
+       MD5_Final(*challenge_response, &ctx);
+
+       return;
+}
+
+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;
+}
+
+static void update_config()
+{
+       int i;
+       char *p;
+       static int timeout = 0;
+       static int interval = 0;
+
+       // Update logging
+       closelog();
+       syslog_log = 0;
+       if (log_stream)
+       {
+               if (log_stream != stderr)
+                       fclose(log_stream);
+
+               log_stream = NULL;
+       }
+
+       if (*config->log_filename)
+       {
+               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 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);
+                       }
+                       else
+                       {
+                               log_stream = stderr;
+                               setbuf(log_stream, NULL);
+                       }
+               }
+       }
+       else
+       {
+               log_stream = stderr;
+               setbuf(log_stream, NULL);
+       }
+
+#define L2TP_HDRS              (20+8+6+4)      // L2TP data encaptulation: ip + udp + l2tp (data) + ppp (inc hdlc)
+#define TCP_HDRS               (20+20)         // TCP encapsulation: ip + tcp
+
+       if (config->l2tp_mtu <= 0)              config->l2tp_mtu = 1500; // ethernet default
+       else if (config->l2tp_mtu < MINMTU)     config->l2tp_mtu = MINMTU;
+       else if (config->l2tp_mtu > MAXMTU)     config->l2tp_mtu = MAXMTU;
+
+       // reset MRU/MSS globals
+       MRU = config->l2tp_mtu - L2TP_HDRS;
+       if (MRU > PPPoE_MRU)
+               MRU = PPPoE_MRU;
+
+       MSS = MRU - TCP_HDRS;
+
+       // Update radius
+       config->numradiusservers = 0;
+       for (i = 0; i < MAXRADSERVER; i++)
+               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, "No RADIUS servers defined!\n");
+
+       // parse radius_authtypes_s
+       config->radius_authtypes = config->radius_authprefer = 0;
+       p = config->radius_authtypes_s;
+       while (p && *p)
+       {
+               char *s = strpbrk(p, " \t,");
+               int type = 0;
+
+               if (s)
+               {
+                       *s++ = 0;
+                       while (*s == ' ' || *s == '\t')
+                               s++;
+
+                       if (!*s)
+                               s = 0;
+               }
+
+               if (!strncasecmp("chap", p, strlen(p)))
+                       type = AUTHCHAP;
+               else if (!strncasecmp("pap", p, strlen(p)))
+                       type = AUTHPAP;
+               else
+                       LOG(0, 0, 0, "Invalid RADIUS authentication type \"%s\"\n", p);
+
+               config->radius_authtypes |= type;
+               if (!config->radius_authprefer)
+                       config->radius_authprefer = type;
+
+               p = s;
+       }
+
+       if (!config->radius_authtypes)
+       {
+               LOG(0, 0, 0, "Defaulting to PAP authentication\n");
+               config->radius_authtypes = config->radius_authprefer = AUTHPAP;
+       }
+
+       // normalise radius_authtypes_s
+       if (config->radius_authprefer == AUTHPAP)
        {
-               log(0, 0, 0, 0, "Error doing mmap for cli session kill: %s\n", strerror(errno));
-               exit(1);
+               strcpy(config->radius_authtypes_s, "pap");
+               if (config->radius_authtypes & AUTHCHAP)
+                       strcat(config->radius_authtypes_s, ", chap");
        }
-       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 <= 0)
+       else
        {
-               log(0, 0, 0, 0, "Error doing mmap for cli tunnel kill: %s\n", strerror(errno));
-               exit(1);
+               strcpy(config->radius_authtypes_s, "chap");
+               if (config->radius_authtypes & AUTHPAP)
+                       strcat(config->radius_authtypes_s, ", pap");
        }
-       memset(cli_tunnel_kill, 0, sizeof(tunnelidt) * MAXSESSION);
 
-       filter_buckets = mmap(NULL, sizeof(tbft) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (filter_buckets <= 0)
+       if (!config->radius_dae_port)
+               config->radius_dae_port = DAEPORT;
+
+       // re-initialise the random number source
+       initrandom(config->random_device);
+
+       // Update plugins
+       for (i = 0; i < MAXPLUGINS; i++)
        {
-               log(0, 0, 0, 0, "Error doing mmap for filter buckets: %s\n", strerror(errno));
-               exit(1);
+               if (strcmp(config->plugins[i], config->old_plugins[i]) == 0)
+                       continue;
+
+               if (*config->plugins[i])
+               {
+                       // Plugin added
+                       add_plugin(config->plugins[i]);
+               }
+               else if (*config->old_plugins[i])
+               {
+                       // Plugin removed
+                       remove_plugin(config->old_plugins[i]);
+               }
        }
-       memset(filter_buckets, 0, sizeof(tbft) * MAXSESSION);
 
-#endif
+       memcpy(config->old_plugins, config->plugins, sizeof(config->plugins));
+       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);
 
-       memset(tunnel, 0, sizeof(tunnelt) * MAXTUNNEL);
-       memset(session, 0, sizeof(sessiont) * MAXSESSION);
-       memset(radius, 0, sizeof(radiust) * MAXRADIUS);
-       memset(ip_address_pool, 0, sizeof(ippoolt) * MAXIPPOOL);
-       for (i = 1; i < MAXTUNNEL - 1; i++)
-               tunnel[i].next = i + 1;
-       tunnel[MAXTUNNEL - 1].next = 0;
-       tunnelfree = 1;
-       for (i = 1; i < MAXSESSION - 1; i++)
-               session[i].next = i + 1;
-       session[MAXSESSION - 1].next = 0;
-       sessionfree = 1;
-       for (i = 1; i < MAXRADIUS - 1; i++)
-               radius[i].next = i + 1;
-       radius[MAXRADIUS - 1].next = 0;
-       radiusfree = 1;
-       if (!*hostname)
+       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)
        {
-               // Grab my hostname unless it's been specified
-               gethostname(hostname, sizeof(hostname));
+               // 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)
                {
-                       struct hostent *h = gethostbyname(hostname);
-                       if (h)
-                               myip = ntohl(*(u32 *) h->h_addr);
+                       LOG(0, 0, 0, "Heartbeat timeout %d too low, adjusting to %d\n", config->cluster_hb_timeout, t);
+                       config->cluster_hb_timeout = t;
                }
-       }
-       _statistics->start_time = _statistics->last_reset = time(NULL);
 
-       // Start the timer routine off
-       time(&time_now);
-       strftime(time_now_string, 64, "%Y-%m-%d %H:%M:%S", localtime(&time_now));
-}
+               // Push timing changes to the slaves immediately if we're the master
+               if (config->cluster_iam_master)
+                       cluster_heartbeat();
 
-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");
-}
+               interval = config->cluster_hb_interval;
+               timeout = config->cluster_hb_timeout;
+       }
 
-ipt assign_ip_address()
-{
-       int c = 0;
-#ifdef STAT_CALLS
-       STAT(call_assign_ip_address);
-#endif
-       ip_pool_index++;
-       while (1)
+       // Write PID file
+       if (*config->pid_file == '/' && !config->wrote_pid)
        {
-               if (ip_pool_index >= ip_pool_size)
+               FILE *f;
+               if ((f = fopen(config->pid_file, "w")))
                {
-                       if (++c == 2)
-                               return 0;
-                       ip_pool_index = 0;
+                       fprintf(f, "%d\n", getpid());
+                       fclose(f);
+                       config->wrote_pid = 1;
                }
-               if (!ip_address_pool[ip_pool_index].assigned && ip_address_pool[ip_pool_index].address)
+               else
                {
-                       ip_address_pool[ip_pool_index].assigned = 1;
-                       log(4, ip_address_pool[ip_pool_index].address, 0, 0, "assign_ip_address(): Allocating ip address %lu from pool\n", ip_pool_index);
-                       STAT(ip_allocated);
-                       return ntohl(ip_address_pool[ip_pool_index].address);
+                       LOG(0, 0, 0, "Can't write to PID file %s: %s\n", config->pid_file, strerror(errno));
                }
-               ip_pool_index++;
        }
-       return 0;
 }
 
-void free_ip_address(ipt address)
+static void read_config_file()
 {
-       int i;
-       ipt a;
-#ifdef STAT_CALLS
-       STAT(call_free_ip_address);
-#endif
+       FILE *f;
 
-       a = ntohl(address);
-       for (i = 0; i <= ip_pool_size; i++)
+       if (!config->config_file) return;
+       if (!(f = fopen(config->config_file, "r")))
        {
-               if (ip_address_pool[i].address == a)
-               {
-                       STAT(ip_freed);
-                       ip_address_pool[i].assigned = 0;
-               }
+               fprintf(stderr, "Can't open config file %s: %s\n", config->config_file, strerror(errno));
+               return;
        }
-       uncache_sessionid(htonl(address));
+
+       LOG(3, 0, 0, "Reading config file %s\n", config->config_file);
+       cli_do_file(f);
+       LOG(3, 0, 0, "Done reading config file\n");
+       fclose(f);
 }
 
-// Initialize the IP address pool
-void initippool()
+int sessionsetup(sessionidt s, tunnelidt t)
 {
-       FILE *f;
-       char *buf, *p;
-       int pi = 0;
-       memset(ip_address_pool, 0, sizeof(ip_address_pool));
+       // A session now exists, set it up
+       in_addr_t ip;
+       char *user;
+       sessionidt i;
+       int r;
 
-       if (!(f = fopen(IPPOOLFILE, "r")))
+       CSTAT(sessionsetup);
+
+       LOG(3, s, t, "Doing session setup for session\n");
+
+       if (!session[s].ip)
        {
-               log(0, 0, 0, 0, "Can't load pool file " IPPOOLFILE ": %s\n", strerror(errno));
-               exit(-1);
+               assign_ip_address(s);
+               if (!session[s].ip)
+               {
+                       LOG(0, s, t, "   No IP allocated.  The IP address pool is FULL!\n");
+                       sessionshutdown(s, "No IP addresses available.", CDN_TRY_ANOTHER, TERM_SERVICE_UNAVAILABLE);
+                       return 0;
+               }
+               LOG(3, s, t, "   No IP allocated.  Assigned %s from pool\n",
+                       fmtaddr(htonl(session[s].ip), 0));
        }
 
-       buf = (char *)malloc(4096);
 
-       while (pi < MAXIPPOOL && fgets(buf, 4096, f))
+       // Make sure this is right
+       session[s].tunnel = t;
+
+       // Join a bundle if the MRRU option is accepted
+       if (session[s].mrru > 0 && !session[s].bundle)
        {
-               char* pool = buf;
-               if (*buf == '#' || *buf == '\n')
-                       continue; // Skip comments / blank lines
-               if ((p = (char *)strrchr(buf, '\n'))) *p = 0;
-               if ((p = (char *)strchr(buf, ':')))
+               LOG(3, s, t, "This session can be part of multilink bundle\n");
+               if (join_bundle(s))
+                       cluster_send_bundle(session[s].bundle);
+       }
+
+       // zap old sessions with same IP and/or username
+       // 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 <= config->cluster_highest_sessionid; i++)
                {
-                       ipt src;
-                       *p = '\0';
-                       src = inet_addr(buf);
-                       if (src == INADDR_NONE)
+                       if (i == s) continue;
+                       if (!session[s].opened) continue;
+                       if (ip == session[i].ip)
                        {
-                               log(0, 0, 0, 0, "Invalid address pool IP %s", buf);
-                               exit(-1);
+                               sessionkill(i, "Duplicate IP address");
+                               continue;
                        }
-                       // This entry is for a specific IP only
-                       if (src != bind_address)
+
+                       if (config->allow_duplicate_users)
                                continue;
-                       *p = ':';
-                       pool = p+1;
-               }
-               if ((p = (char *)strchr(pool, '/')))
-               {
-                       // It's a range
-                       int numbits = 0;
-                       unsigned long start = 0, end = 0, mask = 0, ip;
-                       struct rtentry r;
 
-                       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);
+                       if (session[s].walled_garden || session[i].walled_garden)
                                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);
-                       }
 
-                       // 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 = 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));
-                       }
+                       // Allow duplicate sessions for guest account.
+                       if (*config->guest_user && !strcasecmp(user, config->guest_user))
+                               continue;
+
+                       // Allow duplicate sessions for multilink ones of the same bundle.
+                       if (session[s].bundle && session[i].bundle && session[s].bundle == session[i].bundle)
+                               continue;
+
+                       // Drop the new session in case of duplicate sessionss, not the old one.
+                       if (!strcasecmp(user, session[i].user))
+                               sessionkill(i, "Duplicate session for users");
                }
-               else
+       }
+
+       {
+               int routed = 0;
+
+               // Add the route for this session.
+               for (r = 0; r < MAXROUTE && session[s].route[r].ip; r++)
                {
-                       // It's a single ip address
-                       ip_address_pool[pi++].address = inet_addr(pool);
+                       if ((session[s].ip & session[s].route[r].mask) ==
+                           (session[s].route[r].ip & session[s].route[r].mask))
+                               routed++;
+
+                       routeset(s, session[s].route[r].ip, session[s].route[r].mask, 0, 1);
+               }
+
+               // Static IPs need to be routed if not already
+               // convered by a Framed-Route.  Anything else is part
+               // of the IP address pool and is already routed, it
+               // just needs to be added to the IP cache.
+               // IPv6 route setup is done in ppp.c, when IPV6CP is acked.
+               if (session[s].ip_pool_index == -1) // static ip
+               {
+                       if (!routed) routeset(s, session[s].ip, 0, 0, 1);
                }
+               else
+                       cache_ipmap(session[s].ip, s);
        }
 
-       free(buf);
-       fclose(f);
-       log(1, 0, 0, 0, "IP address pool is %d addresses\n", pi);
-       ip_pool_size = pi;
-}
+       sess_local[s].lcp_authtype = 0; // RADIUS authentication complete
+       lcp_open(s, t); // transition to Network phase and send initial IPCP
 
-void snoop_send_packet(char *packet, u16 size)
-{
-       if (!snoop_addr.sin_port || snoopfd <= 0 || size <= 0 || !packet)
-               return;
+       // Run the plugin's against this new session.
+       {
+               struct param_new_session data = { &tunnel[t], &session[s] };
+               run_plugins(PLUGIN_NEW_SESSION, &data);
+       }
 
-       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);
-}
+       // Allocate TBFs if throttled
+       if (session[s].throttle_in || session[s].throttle_out)
+               throttle_session(s, session[s].throttle_in, session[s].throttle_out);
 
-void dump_acct_info()
-{
-    char filename[1024];
-    char timestr[64];
-    time_t t = time(NULL);
-    int i;
-    FILE *f = NULL;
+       session[s].last_packet = time_now;
 
-#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", 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)
-       {
-           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);
+       LOG(2, s, t, "Login by %s at %s from %s (%s)\n", session[s].user,
+               fmtaddr(htonl(session[s].ip), 0),
+               fmtaddr(htonl(tunnel[t].ip), 1), tunnel[t].hostname);
+
+       cluster_send_session(s);        // Mark it as dirty, and needing to the flooded to the cluster.
+
+       return 1;       // RADIUS OK and IP allocated, done...
 }
 
-// Main program
-int main(int argc, char *argv[])
+//
+// 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 o;
-
-       _program_name = strdup(argv[0]);
+       int i;
+       int newip = 0;
 
+               // Sanity checks.
+       if (new->ip_pool_index >= MAXIPPOOL ||
+               new->tunnel >= MAXTUNNEL)
        {
-               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");
+               LOG(0, s, 0, "Strange session update received!\n");
+                       // FIXME! What to do here?
+               return 0;
        }
 
-       time(&basetime);             // start clock
-       // scan args
+               //
+               // Ok. All sanity checks passed. Now we're committed to
+               // loading the new session.
+               //
+
+       session[s].tunnel = new->tunnel; // For logging in cache_ipmap
 
-       while ((o = getopt(argc, argv, "vc:f:h:a:")) >= 0)
+       // See if routes/ip cache need updating
+       if (new->ip != session[s].ip)
+               newip++;
+
+       for (i = 0; !newip && 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)
+                       newip++;
+
+       // needs update
+       if (newip)
        {
-               switch (o)
+               int routed = 0;
+
+               // remove old routes...
+               for (i = 0; i < MAXROUTE && session[s].route[i].ip; i++)
                {
-                       case 'v':
-                               debug++;
-                               break;
-                       case 'c':
-                               config_file = strdup(optarg);
-                               break;
-                       case 'f':
-                               log_filename = strdup(optarg);
-                               break;
-                       case 'h':
-                               strncpy(hostname, optarg, 1000);
-                               break;
-                       case 'a':
-                               myip = inet_addr(optarg);
-                               if (myip == INADDR_NONE) {
-                                       log(0, 0, 0, 0, "Invalid ip %s\n", optarg);
-                                       exit(-1);
-                               }
-                               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-f <file>\tLog File\n\t-v\t\tDebug\n");
-                               return (0);
-                               break;
+                       if ((session[s].ip & session[s].route[i].mask) ==
+                           (session[s].route[i].ip & session[s].route[i].mask))
+                               routed++;
+
+                       routeset(s, session[s].route[i].ip, session[s].route[i].mask, 0, 0);
                }
-       }
 
-       initiptables();
-       initplugins();
-       read_config_file();
-       initdata();
-       log(0, 0, 0, 0, "$Id: l2tpns.c,v 1.1 2003/12/16 07:07:39 fred_nerk Exp $\n(c) Copyright 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced\n");
+               // ...ip
+               if (session[s].ip)
+               {
+                       if (session[s].ip_pool_index == -1) // static IP
+                       {
+                               if (!routed) routeset(s, session[s].ip, 0, 0, 0);
+                       }
+                       else            // It's part of the IP pool, remove it manually.
+                               uncache_ipmap(session[s].ip);
+               }
 
-       /* 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(bind_address, 0);
-       cluster_send_message(cluster_address, bind_address, C_HELLO, hostname, strlen(hostname));
+               routed = 0;
 
-       inittap();
-       log(1, 0, 0, 0, "Set up on interface %s\n", tapdevice);
+               // add new routes...
+               for (i = 0; i < MAXROUTE && new->route[i].ip; i++)
+               {
+                       if ((new->ip & new->route[i].mask) ==
+                           (new->route[i].ip & new->route[i].mask))
+                               routed++;
 
-       initudp();
-       initrad();
-       initippool();
-       init_rl();
-       if (handle_interface) {
-               send_garp(bind_address);
+                       routeset(s, new->route[i].ip, new->route[i].mask, 0, 1);
+               }
+
+               // ...ip
+               if (new->ip)
+               {
+                       // If there's a new one, add it.
+                       if (new->ip_pool_index == -1)
+                       {
+                               if (!routed) routeset(s, new->ip, 0, 0, 1);
+                       }
+                       else
+                               cache_ipmap(new->ip, s);
+               }
        }
-       read_state();
 
-#ifdef HAVE_LIBCLI
-       init_cli();
-#endif
+       // check v6 routing
+       if (new->ipv6prefixlen && new->ppp.ipv6cp == Opened && session[s].ppp.ipv6cp != Opened)
+                   route6set(s, new->ipv6route, new->ipv6prefixlen, 1);
 
-       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);
-       if (debug)
+       // check filters
+       if (new->filter_in && (new->filter_in > MAXFILTER || !ip_filters[new->filter_in - 1].name[0]))
        {
-               int n;
-               for (n = 0; n < numradiusservers; n++)
-                       log(1, 0, 0, 0, "RADIUS to %s\n", inet_toa(htonl(radiusserver[n])));
+               LOG(2, s, session[s].tunnel, "Dropping invalid input filter %u\n", (int) new->filter_in);
+               new->filter_in = 0;
        }
 
-       alarm(1);
+       if (new->filter_out && (new->filter_out > MAXFILTER || !ip_filters[new->filter_out - 1].name[0]))
+       {
+               LOG(2, s, session[s].tunnel, "Dropping invalid output filter %u\n", (int) new->filter_out);
+               new->filter_out = 0;
+       }
 
-       // Drop privileges here
-       if (target_uid > 0 && geteuid() == 0)
-               setuid(target_uid);
+       if (new->filter_in != session[s].filter_in)
+       {
+               if (session[s].filter_in) ip_filters[session[s].filter_in - 1].used--;
+               if (new->filter_in)       ip_filters[new->filter_in - 1].used++;
+       }
 
-       mainloop();
+       if (new->filter_out != session[s].filter_out)
+       {
+               if (session[s].filter_out) ip_filters[session[s].filter_out - 1].used--;
+               if (new->filter_out)       ip_filters[new->filter_out - 1].used++;
+       }
 
-       if (l2tpsecret) free(l2tpsecret);
-       if (log_filename) free(log_filename);
-       if (snoop_destination_host) free(snoop_destination_host);
-       if (radiussecret) free(radiussecret);
+       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;
 
-       return 0;
-}
+       memcpy(&session[s], new, sizeof(session[s]));   // Copy over..
 
-void sighup_handler(int junk)
-{
-       if (log_stream != stderr)
-               fclose(log_stream);
+               // Do fixups into address pool.
+       if (new->ip_pool_index != -1)
+               fix_address_pool(s);
 
-       log_stream = NULL;
-       read_config_file();
+       return 1;
 }
 
-void sigalrm_handler(int junk)
+static void initplugins()
 {
-       // Log current traffic stats
-       if (dump_speed)
-       {
-               printf("UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%lu OUT:%lu\n",
-                       (udp_rx / 1024.0 / 1024.0 * 8),
-                       (eth_tx / 1024.0 / 1024.0 * 8),
-                       (eth_rx / 1024.0 / 1024.0 * 8),
-                       (udp_tx / 1024.0 / 1024.0 * 8),
-                       ((udp_tx + udp_rx + eth_tx + eth_rx) / 1024.0 / 1024.0 * 8),
-                       udp_rx_pkt, eth_rx_pkt);
-
-               udp_tx = udp_rx = 0;
-               udp_rx_pkt = eth_rx_pkt = 0;
-               eth_tx = eth_rx = 0;
-       }
-
-       // Update the internal time counter
-       time(&time_now);
-       strftime(time_now_string, 64, "%Y-%m-%d %H:%M:%S", localtime(&time_now));
-       alarm(1);
-
-       {
-               // Run timer hooks
-               struct param_timer p = { time_now };
-               run_plugins(PLUGIN_TIMER, &p);
-       }
+       int i;
 
+       loaded_plugins = ll_init();
+       // Initialize the plugins to nothing
+       for (i = 0; i < MAX_PLUGIN_TYPES; i++)
+               plugins[i] = ll_init();
 }
 
-void sigterm_handler(int junk)
+static void *open_plugin(char *plugin_name, int load)
 {
-       log(1, 0, 0, 0, "Shutting down cleanly\n");
-       if (config_save_state)
-               dump_state();
-       main_quit++;
+       char path[256] = "";
+
+       snprintf(path, 256, PLUGINDIR "/%s.so", plugin_name);
+       LOG(2, 0, 0, "%soading plugin from %s\n", load ? "L" : "Un-l", path);
+       return dlopen(path, RTLD_NOW);
 }
 
-void sigquit_handler(int junk)
+// plugin callback to get a config value
+static void *getconfig(char *key, enum config_typet type)
 {
        int i;
-       log(1, 0, 0, 0, "Shutting down without saving sessions\n");
-       for (i = 0; i < MAXSESSION; i++)
-       {
-               if (session[i].opened)
-                       sessionkill(i, "L2TPNS Closing");
-       }
-       for (i = 0; i < MAXTUNNEL; i++)
+
+       for (i = 0; config_values[i].key; i++)
        {
-               if (tunnel[i].ip)
-                       tunnelshutdown(i, "L2TPNS Closing");
-       }
-       main_quit++;
-}
+               if (!strcmp(config_values[i].key, key))
+               {
+                       if (config_values[i].type == type)
+                               return ((void *) config) + config_values[i].offset;
 
-void sigchild_handler(int signal)
-{
-       int status;
-       int pid;
+                       LOG(1, 0, 0, "plugin requested config item \"%s\" expecting type %d, have type %d\n",
+                               key, type, config_values[i].type);
 
-       pid = wait(&status);
-#ifdef HAVE_LIBCLI
-       status = (WIFEXITED(status)) ? WEXITSTATUS(status) : 0;
-       if (pid == cli_pid)
-       {
-               if (status == 0)
-                       log(3, 0, 0, 0, "CLI client closed connection\n");
-               else
-                       log(2, 0, 0, 0, "CLI child died with rc %d!\n", status);
+                       return 0;
+               }
        }
-#endif
-}
 
-void *backtrace_buffer[30] = {0};
-
-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);
+       LOG(1, 0, 0, "plugin requested unknown config item \"%s\"\n", key);
+       return 0;
 }
 
-void read_state()
+static int add_plugin(char *plugin_name)
 {
-       struct stat sb;
-       FILE *f;
-
-       if (!config_save_state) return;
-
-       if (stat(STATEFILE, &sb) < 0)
-               return;
+       static struct pluginfuncs funcs = {
+               _log,
+               _log_hex,
+               fmtaddr,
+               sessionbyuser,
+               sessiontbysessionidt,
+               sessionidtbysessiont,
+               radiusnew,
+               radiussend,
+               getconfig,
+               sessionshutdown,
+               sessionkill,
+               throttle_session,
+               cluster_send_session,
+       };
+
+       void *p = open_plugin(plugin_name, 1);
+       int (*initfunc)(struct pluginfuncs *);
+       int i;
 
-       if (sb.st_mtime < (time(NULL) - 60))
+       if (!p)
        {
-               log(0, 0, 0, 0, "State file is too old to read\n");
-               unlink(STATEFILE);
-               return;
+               LOG(1, 0, 0, "   Plugin load failed: %s\n", dlerror());
+               return -1;
        }
 
-       if (!(f = fopen(STATEFILE, "r")))
+       if (ll_contains(loaded_plugins, p))
        {
-               log(0, 0, 0, 0, "Can't read state file: %s\n", strerror(errno));
-               unlink(STATEFILE);
-               return;
+               dlclose(p);
+               return 0; // already loaded
        }
-       fseek(f, 0, 0);
 
-       log(1, 0, 0, 0, "Reading state information\n");
-       {
-               u32 i, numtunnels;
-               if (fread(&numtunnels, sizeof(numtunnels), 1, f) <= 0)
-               {
-                       log(0, 0, 0, 0, "Error reading saved state (tunnel count): %s\n", strerror(errno));
-                       fclose(f);
-                       unlink(STATEFILE);
-                       return;
-               }
-               log(2, 0, 0, 0, "Reading %lu tunnels\n", numtunnels);
-               fread(tunnel, sizeof(tunnelt), numtunnels, f);
-               tunnelfree = 0;
-               for (i = 0; i < numtunnels; 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);
-                               tunnelfree = i;
-                       }
-               }
-               tunnelfree++;
-       }
-       {
-               u32 i, numsessions;
-               if (fread(&numsessions, sizeof(numsessions), 1, f) <= 0)
-               {
-                       log(0, 0, 0, 0, "Error reading saved state (session count): %s\n", strerror(errno));
-                       fclose(f);
-                       unlink(STATEFILE);
-                       return;
-               }
-               log(2, 0, 0, 0, "Reading %lu sessions\n", numsessions);
-               if (fread(session, sizeof(sessiont), numsessions, f) < numsessions)
-               {
-                       log(0, 0, 0, 0, "Error reading saved state (%d sessions): %s\n", numsessions, strerror(errno));
-                       fclose(f);
-                       unlink(STATEFILE);
-                       return;
-               }
-               for (i = 0; i < numsessions; 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)
-                               {
-                                       int x;
-                                       sessionsetup(session[i].tunnel, i, 0);
-                                       for (x = 0; x < MAXIPPOOL && ip_address_pool[x].address; x++)
-                                       {
-                                               if (ip_address_pool[x].address == session[i].ip)
-                                               {
-                                                       ip_address_pool[x].assigned = 1;
-                                                       break;
-                                               }
-                                       }
-                               }
-                               else
-                               {
-                                       log(2, 0, i, 0, "No IP for session\n");
-                               }
-                       }
+       {
+               int *v = dlsym(p, "plugin_api_version");
+               if (!v || *v != PLUGIN_API_VERSION)
+               {
+                       LOG(1, 0, 0, "   Plugin load failed: API version mismatch: %s\n", dlerror());
+                       dlclose(p);
+                       return -1;
                }
-               for (i = 0; i < numsessions && session[i].opened; i++)
-                       sessionfree = session[i].next;
        }
-       fclose(f);
-       log(0, 0, 0, 0, "Loaded saved state information\n");
-       unlink(STATEFILE);
-}
 
-void dump_state()
-{
-       FILE *f;
+       if ((initfunc = dlsym(p, "plugin_init")))
+       {
+               if (!initfunc(&funcs))
+               {
+                       LOG(1, 0, 0, "   Plugin load failed: plugin_init() returned FALSE: %s\n", dlerror());
+                       dlclose(p);
+                       return -1;
+               }
+       }
 
-       if (!config_save_state) return;
+       ll_push(loaded_plugins, p);
 
-       if ((f = fopen(STATEFILE, "w")))
+       for (i = 0; i < max_plugin_functions; i++)
        {
-               u32 i;
-               log(1, 0, 0, 0, "Dumping state information\n");
+               void *x;
+               if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
+               {
+                       LOG(3, 0, 0, "   Supports function \"%s\"\n", plugin_functions[i]);
+                       ll_push(plugins[i], x);
+               }
+       }
 
-               i = MAXTUNNEL;
-               fwrite(&i, sizeof(i), 1, f);    // Number of tunnels
+       LOG(2, 0, 0, "   Loaded plugin %s\n", plugin_name);
+       return 1;
+}
 
-               log(2, 0, 0, 0, "Dumping %lu tunnels\n", i);
-               fwrite(tunnel, sizeof(tunnelt), MAXTUNNEL, f);
+static void run_plugin_done(void *plugin)
+{
+       int (*donefunc)(void) = dlsym(plugin, "plugin_done");
 
-               i = MAXSESSION;
-               fwrite(&i, sizeof(i), 1, f);    // Number of sessions
-               log(2, 0, 0, 0, "Dumping %lu sessions\n", i);
-               fwrite(session, sizeof(sessiont), MAXSESSION, f);
+       if (donefunc)
+               donefunc();
+}
 
-               fclose(f);
-       }
-       else
+static int remove_plugin(char *plugin_name)
+{
+       void *p = open_plugin(plugin_name, 0);
+       int loaded = 0;
+
+       if (!p)
+               return -1;
+
+       if (ll_contains(loaded_plugins, p))
        {
-               log(0, 0, 0, 0, "Can't write state information: %s\n", strerror(errno));
+               int i;
+               for (i = 0; i < max_plugin_functions; i++)
+               {
+                       void *x;
+                       if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
+                               ll_delete(plugins[i], x);
+               }
+
+               ll_delete(loaded_plugins, p);
+               run_plugin_done(p);
+               loaded = 1;
        }
-       return;
+
+       dlclose(p);
+       LOG(2, 0, 0, "Removed plugin %s\n", plugin_name);
+       return loaded;
 }
 
-void build_chap_response(char *challenge, u8 id, u16 challenge_length, char **challenge_response)
+int run_plugins(int plugin_type, void *data)
 {
-       MD5_CTX ctx;
-       *challenge_response = NULL;
+       int (*func)(void *data);
 
-       if (!l2tpsecret || !*l2tpsecret)
-       {
-               log(0, 0, 0, 0, "LNS requested CHAP authentication, but no l2tp secret is defined\n");
-               return;
-       }
+       if (!plugins[plugin_type] || plugin_type > max_plugin_functions)
+               return PLUGIN_RET_ERROR;
 
-       /*
-       if (challenge_length != 16)
+       ll_reset(plugins[plugin_type]);
+       while ((func = ll_next(plugins[plugin_type])))
        {
-               log(0, 0, 0, 0, "Challenge length != 16.\n");
-               return;
-       }
-       */
+               int r = func(data);
 
-       log(4, 0, 0, 0, "   Building challenge response for CHAP request\n");
+               if (r != PLUGIN_RET_OK)
+                       return r; // stop here
+       }
 
-       *challenge_response = (char *)calloc(17, 1);
+       return PLUGIN_RET_OK;
+}
 
-       MD5Init(&ctx);
-       MD5Update(&ctx, &id, 1);
-       MD5Update(&ctx, l2tpsecret, strlen(l2tpsecret));
-       MD5Update(&ctx, challenge, challenge_length);
-       MD5Final(*challenge_response, &ctx);
+static void plugins_done()
+{
+       void *p;
 
-       return;
+       ll_reset(loaded_plugins);
+       while ((p = ll_next(loaded_plugins)))
+               run_plugin_done(p);
 }
 
-void read_config_file()
+static void processcontrol(uint8_t *buf, int len, struct sockaddr_in *addr, int alen, struct in_addr *local)
 {
-       FILE *f;
-       char *buf;
-
-       if (!config_file) return;
-       if (!(f = fopen(config_file, "r"))) {
-               fprintf(stderr, "Can't open config file %s: %s\n", config_file, strerror(errno));
-               return;
-       }
+       struct nsctl request;
+       struct nsctl response;
+       int type = unpack_control(&request, buf, len);
+       int r;
+       void *p;
 
-       if (radiussecret)
+       if (log_stream && config->debug >= 4)
        {
-               free(radiussecret);
-               radiussecret = NULL;
+               if (type < 0)
+               {
+                       LOG(4, 0, 0, "Bogus control message from %s (%d)\n",
+                               fmtaddr(addr->sin_addr.s_addr, 0), type);
+               }
+               else
+               {
+                       LOG(4, 0, 0, "Received [%s] ", fmtaddr(addr->sin_addr.s_addr, 0));
+                       dump_control(&request, log_stream);
+               }
        }
 
-       if (l2tpsecret)
+       switch (type)
        {
-               free(l2tpsecret);
-               l2tpsecret = NULL;
-       }
+       case NSCTL_REQ_LOAD:
+               if (request.argc != 1)
+               {
+                       response.type = NSCTL_RES_ERR;
+                       response.argc = 1;
+                       response.argv[0] = "name of plugin required";
+               }
+               else if ((r = add_plugin(request.argv[0])) < 1)
+               {
+                       response.type = NSCTL_RES_ERR;
+                       response.argc = 1;
+                       response.argv[0] = !r
+                               ? "plugin already loaded"
+                               : "error loading plugin";
+               }
+               else
+               {
+                       response.type = NSCTL_RES_OK;
+                       response.argc = 0;
+               }
 
-       if (log_filename)
-       {
-               free(log_filename);
-               log_filename = NULL;
-       }
+               break;
 
-       if (snoop_destination_host)
-       {
-               free(snoop_destination_host);
-               snoop_destination_host = NULL;
-       }
+       case NSCTL_REQ_UNLOAD:
+               if (request.argc != 1)
+               {
+                       response.type = NSCTL_RES_ERR;
+                       response.argc = 1;
+                       response.argv[0] = "name of plugin required";
+               }
+               else if ((r = remove_plugin(request.argv[0])) < 1)
+               {
+                       response.type = NSCTL_RES_ERR;
+                       response.argc = 1;
+                       response.argv[0] = !r
+                               ? "plugin not loaded"
+                               : "plugin not found";
+               }
+               else
+               {
+                       response.type = NSCTL_RES_OK;
+                       response.argc = 0;
+               }
 
-       if (numradiusservers) {
-               int n;
-               for (n = 0; n < numradiusservers; n++)
-                       radiusserver[n] = 0;
-               numradiusservers = 0;
-       }
-
-       snoop_destination_port = 0L;
-       config_save_state = 0;
-       rl_rate = 0L;
-       debug = 1;
-       default_dns1 = default_dns2 = 0;
-       radius_accounting = 0;
-
-       buf = (char *)malloc(4096);
-
-       while (fgets(buf, 4096, f)) {
-               char *p, *t;
-
-               if (*buf == '#') continue;
-               if ((p = strchr(buf, '\n'))) *p = 0;
-               p = t = strchr(buf, '=');
-               if (!p) continue;
-               *p = 0; p++;
-               t--;
-               while (*p && *p == ' ') p++;
-               while (*t && *t == ' ') *t-- = 0;
-
-               if (strcmp(buf, "log file") == 0) {
-                       if (!log_filename)
-                               log_filename = strdup(p);
-               } else if (strcmp(buf, "l2tp secret") == 0) {
-                       if (!l2tpsecret)
-                               l2tpsecret = strdup(p);
-                       log(0, 0, 0, 0, "L2TP Secret is \"%s\"\n", l2tpsecret);
-               } else if (strcmp(buf, "radius secret") == 0) {
-                       if (!radiussecret)
-                               radiussecret = strdup(p);
-                       log(4, 0, 0, 0, "Radius Secret is \"%s\"\n", radiussecret);
-               } else if (strcmp(buf, "radius accounting") == 0) {
-                       radius_accounting = atoi(p);
-                       log(4, 0, 0, 0, "Radius Account is %s\n", radius_accounting ? "on" : "off");
-               } else if (strcmp(buf, "throttle rate") == 0) {
-                       rl_rate = atol(p);
-                       if (rl_rate == 0)
-                       {
-                               log(1, 0, 0, 0, "Disabled throttling.\n");
-                       }
-                       else
+               break;
+
+       case NSCTL_REQ_HELP:
+               response.type = NSCTL_RES_OK;
+               response.argc = 0;
+
+               ll_reset(loaded_plugins);
+               while ((p = ll_next(loaded_plugins)))
+               {
+                       char **help = dlsym(p, "plugin_control_help");
+                       while (response.argc < 0xff && help && *help)
+                               response.argv[response.argc++] = *help++;
+               }
+
+               break;
+
+       case NSCTL_REQ_CONTROL:
+               {
+                       struct param_control param = {
+                               config->cluster_iam_master,
+                               request.argc,
+                               request.argv,
+                               0,
+                               NULL,
+                       };
+
+                       int r = run_plugins(PLUGIN_CONTROL, &param);
+
+                       if (r == PLUGIN_RET_ERROR)
                        {
-                               log(1, 0, 0, 0, "Enabled throttling (rate is %lu kbits/s)\n", rl_rate);
-                       }
-               } else if (strcmp(buf, "debug") == 0) {
-                       debug = atoi(p);
-                       log(debug, 0, 0, 0, "Set debugging level to %d\n", debug);
-               } else if (strcmp(buf, "accounting dir") == 0) {
-                       accounting_dir = strdup(p);
-                       log(debug, 0, 0, 0, "Will dump accounting information to %s\n", accounting_dir);
-               } else if (strcmp(buf, "dns server") == 0) {
-                       unsigned long addr = 0;
-                       if (inet_aton(p, (struct in_addr *)&addr) < 0) {
-                               printf("Invalid DNS server %s\n", p);
-                               continue;
+                               response.type = NSCTL_RES_ERR;
+                               response.argc = 1;
+                               response.argv[0] = param.additional
+                                       ? param.additional
+                                       : "error returned by plugin";
                        }
-                       if (default_dns1 == 0)
-                               default_dns1 = addr;
-                       else if (default_dns2 == 0)
-                               default_dns2 = addr;
-               } else if (strcmp(buf, "radius server") == 0) {
-                       struct hostent *h = gethostbyname(p);
-                       if (h)
+                       else if (r == PLUGIN_RET_NOTMASTER)
                        {
-                               while (*h->h_addr_list)
-                               {
-                                       ipt ip = ntohl(*(u32 *) * h->h_addr_list);
-                                       if (numradiusservers < MAXRADSERVER)
-                                               radiusserver[numradiusservers++] = ip;
-                                       else
-                                               log(0, 0, 0, 0, "Too many RADIUS IPs\n");
-                                       h->h_addr_list++;
-                               }
-                       }
-                       else
-                       {        // may be IP?
-                               ipt ip = ntohl(inet_addr(p));
-                               if (ip && ip != 0xFFFFFFFF)
+                               static char msg[] = "must be run on master: 000.000.000.000";
+
+                               response.type = NSCTL_RES_ERR;
+                               response.argc = 1;
+                               if (config->cluster_master_address)
                                {
-                                       if (numradiusservers < MAXRADSERVER)
-                                               radiusserver[numradiusservers++] = ip;
-                                       else
-                                               log(0, 0, 0, 0, "Too many RADIUS IPs\n");
+                                       strcpy(msg + 23, fmtaddr(config->cluster_master_address, 0));
+                                       response.argv[0] = msg;
                                }
                                else
-                                       log(0, 0, 0, 0, "Unknown server %s\n", p);
-                       }
-               } else if (strcmp(buf, "snoop host") == 0) {
-                       snoop_destination_host = strdup(p);
-               } else if (strcmp(buf, "snoop port") == 0) {
-                       snoop_destination_port = atol(p);
-               } else if (strcmp(buf, "bind address") == 0) {
-                       if (!bind_address)
-                       {
-                               // Already overridden on the command line
-                               bind_address = inet_addr(p);
-                               handle_interface = 1;
-                       }
-               } else if (strcmp(buf, "dump speed") == 0) {
-                       dump_speed = atoi(p);
-               } else if (strcmp(buf, "setuid") == 0) {
-                       target_uid = atoi(p);
-               } else if (strcmp(buf, "cluster master") == 0) {
-                       struct hostent *h = gethostbyname(p);
-                       if (h)
-                       {
-                               if (*h->h_addr_list)
                                {
-                                       cluster_address = *(u32 *) *h->h_addr_list;
+                                       response.argv[0] = "must be run on master: none elected";
                                }
                        }
-                       else
-                       {        // may be IP?
-                               cluster_address = inet_addr(p);
+                       else if (!(param.response & NSCTL_RESPONSE))
+                       {
+                               response.type = NSCTL_RES_ERR;
+                               response.argc = 1;
+                               response.argv[0] = param.response
+                                       ? "unrecognised response value from plugin"
+                                       : "unhandled action";
                        }
-               } else if (strcmp(buf, "save state") == 0) {
-                       if (strcasecmp(p, "no") == 0) {
-                               config_save_state = 0;
-                       } else {
-                               config_save_state = 1;
+                       else
+                       {
+                               response.type = param.response;
+                               response.argc = 0;
+                               if (param.additional)
+                               {
+                                       response.argc = 1;
+                                       response.argv[0] = param.additional;
+                               }
                        }
-               } else if (strcmp(buf, "plugin") == 0) {
-                       add_plugin(p);
-               } else {
-                       struct param_config cp = { buf, p };
-                       int rc = run_plugins(PLUGIN_CONFIG, &cp);
-                       if (rc == 0) log(0, 0, 0, 0, "Unknown config directive \"%s\"\n", buf);
                }
+
+               break;
+
+       default:
+               response.type = NSCTL_RES_ERR;
+               response.argc = 1;
+               response.argv[0] = "error unpacking control packet";
        }
 
-       if (snoop_destination_host)
+       buf = calloc(NSCTL_MAX_PKT_SZ, 1);
+       if (!buf)
        {
-               if (inet_aton(snoop_destination_host, &snoop_addr.sin_addr))
-               {
-                       snoop_addr.sin_port = htons(snoop_destination_port);
-                       snoop_addr.sin_family = AF_INET;
-               }
-               else
+               LOG(2, 0, 0, "Failed to allocate nsctl response\n");
+               return;
+       }
+
+       r = pack_control(buf, NSCTL_MAX_PKT_SZ, response.type, response.argc, response.argv);
+       if (r > 0)
+       {
+               sendtofrom(controlfd, buf, r, 0, (const struct sockaddr *) addr, alen, local);
+               if (log_stream && config->debug >= 4)
                {
-                       log(0, 0, 0, 0, "Can't find address for snoop host %s\n", snoop_destination_host);
+                       LOG(4, 0, 0, "Sent [%s] ", fmtaddr(addr->sin_addr.s_addr, 0));
+                       dump_control(&response, log_stream);
                }
        }
+       else
+               LOG(2, 0, 0, "Failed to pack nsctl response for %s (%d)\n",
+                       fmtaddr(addr->sin_addr.s_addr, 0), r);
 
        free(buf);
-       fclose(f);
-       log(2, 0, 0, 0, "Done reading config file\n");
 }
 
-int sessionsetup(tunnelidt t, sessionidt s, u8 routes)
+static tunnelidt new_tunnel()
 {
-       // A session now exists, set it up
-       ipt ip;
-       char *user;
-       sessionidt i;
-#ifdef STAT_CALLS
-       STAT(call_sessionsetup);
-#endif
-       log(3, session[s].ip, s, t, "Doing session setup for session\n");
-       if (!session[s].ip) {
-               log(0, session[s].ip, s, t, "VERY VERY BAD! sessionsetup() called with no session[s].ip\n");
-               return 1;
-       }
-       if (session[s].ip == 0xFFFFFFFE)
+       tunnelidt i;
+       for (i = 1; i < MAXTUNNEL; i++)
        {
-               session[s].ip = assign_ip_address(); // Assign one from the pool;
-
-               log(2, session[s].ip, s, t, "IP assigned is a magic token. Assign address from pool: %s\n",
-                               inet_toa(htonl(session[s].ip)));
+               if (tunnel[i].state == TUNNELFREE)
+               {
+                       LOG(4, 0, i, "Assigning tunnel ID %u\n", i);
+                       if (i > config->cluster_highest_tunnelid)
+                               config->cluster_highest_tunnelid = i;
+                       return i;
+               }
        }
+       LOG(0, 0, 0, "Can't find a free tunnel! There shouldn't be this many in use!\n");
+       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
-       // from someone who doesn't need to know the password
-       ip = session[s].ip;
-       user = session[s].user;
-       for (i = 0; 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");
-       }
+//
+// 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;
+       static struct event_data d[RADIUS_FDS];
+       struct epoll_event e;
 
-       if (routes)
+       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())
        {
-               if (session[s].route[routes].ip && session[s].route[routes].mask)
+               for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                {
-                       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 (!session[s].opened) // Not an in-use session.
+                               continue;
+
+                       run_plugins(PLUGIN_NEW_SESSION_MASTER, &session[s]);
                }
+               exit(0);
        }
-       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 radius fds
+       e.events = EPOLLIN;
+       for (i = 0; i < RADIUS_FDS; i++)
        {
-               struct param_new_session data = { &tunnel[t], &session[s] };
-               run_plugins(PLUGIN_NEW_SESSION, &data);
-       }
-
-       session[s].sid = ++last_sid;
-       cache_sessionid(htonl(session[s].ip), s);
+               d[i].type = FD_TYPE_RADIUS;
+               d[i].index = i;
+               e.data.ptr = &d[i];
 
-       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)));
-               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);
+               epoll_ctl(epollfd, EPOLL_CTL_ADD, radfds[i], &e);
        }
-
-       return 1;       // RADIUS OK and IP allocated, done...
 }
 
-#ifdef RINGBUFFER
-void ringbuffer_dump(FILE *stream)
+int cmd_show_hist_idle(struct cli_def *cli, char *command, char **argv, int argc)
 {
-       int i = ringbuffer->head;
+       int s, i;
+       int count = 0;
+       int buckets[64];
 
-       while (i != ringbuffer->tail)
-       {
-               if (*ringbuffer->buffer[i].message)
-                       fprintf(stream, "%d-%s", ringbuffer->buffer[i].level, ringbuffer->buffer[i].message);
-               if (++i == ringbuffer->tail) break;
-               if (i == RINGBUFFER_SIZE) i = 0;
-       }
-}
-#endif
+       if (CLI_HELP_REQUESTED)
+               return CLI_HELP_NO_ARGS;
 
-void initplugins()
-{
-       int i;
-
-       loaded_plugins = ll_init();
-       // Initialize the plugins to nothing
-       for (i = 0; i < MAX_PLUGIN_TYPES; i++)
-               plugins[i] = ll_init();
-}
+       time(&time_now);
+       for (i = 0; i < 64;++i) buckets[i] = 0;
 
-void add_plugin(char *plugin_name)
-{
-       void *p;
-       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)
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
        {
-               log(1, 0, 0, 0, "   Plugin load failed: %s\n", dlerror());
-               return;
+               int idle;
+               if (!session[s].opened)
+                       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];
        }
 
-       if (ll_contains(loaded_plugins, p))
+       for (i = 0; i < 63; ++i)
        {
-               dlclose(p);
-               return;
+               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 *v = dlsym(p, "__plugin_api_version");
-               if (!v || *v != PLUGIN_API_VERSION)
+               int open = 0, d;
+               if (!session[s].opened)
+                       continue;
+
+               d = time_now - session[s].opened;
+               if (d < 0)
+                       d = 0;
+               while (d > 1 && open < 32)
                {
-                       log(1, 0, 0, 0, "   Plugin load failed: API version mismatch\n", dlerror());
-                       dlclose(p);
-                       return;
+                       ++open;
+                       d >>= 1; // half.
                }
+               ++count;
+               ++buckets[open];
        }
 
-       initfunc = dlsym(p, "plugin_init");
-       if (!initfunc)
-       {
-               log(1, 0, 0, 0, "   Plugin load failed: function plugin_init() does not exist.\n", dlerror());
-               dlclose(p);
-               return;
-       }
-
-       if (!initfunc(&funcs))
+       s = 1;
+       for (i = 0; i  < 30; ++i)
        {
-               log(1, 0, 0, 0, "   Plugin load failed: plugin_init() returned FALSE.\n", dlerror());
-               dlclose(p);
-               return;
+               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;
+}
 
-       for (i = 0; i < max_plugin_functions; i++)
+/* Unhide an avp.
+ *
+ * This unencodes the AVP using the L2TP secret and the previously
+ * stored random vector.  It overwrites the hidden data with the
+ * unhidden AVP subformat.
+ */
+static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vector, size_t vec_len)
+{
+       MD5_CTX ctx;
+       uint8_t digest[16];
+       uint8_t *last;
+       size_t d = 0;
+       uint16_t m = htons(type);
+
+       // Compute initial pad
+       MD5_Init(&ctx);
+       MD5_Update(&ctx, (unsigned char *) &m, 2);
+       MD5_Update(&ctx, config->l2tp_secret, strlen(config->l2tp_secret));
+       MD5_Update(&ctx, vector, vec_len);
+       MD5_Final(digest, &ctx);
+
+       // pointer to last decoded 16 octets
+       last = value;
+
+       while (len > 0)
        {
-               void *x;
-               if (!plugin_functions[i]) continue;
-               if ((x = dlsym(p, plugin_functions[i])))
+               // calculate a new pad based on the last decoded block
+               if (d >= sizeof(digest))
                {
-                       log(3, 0, 0, 0, "   Supports function \"%s\"\n", plugin_functions[i]);
-                       ll_push(plugins[i], x);
+                       MD5_Init(&ctx);
+                       MD5_Update(&ctx, config->l2tp_secret, strlen(config->l2tp_secret));
+                       MD5_Update(&ctx, last, sizeof(digest));
+                       MD5_Final(digest, &ctx);
+
+                       d = 0;
+                       last = value;
                }
+
+               *value++ ^= digest[d++];
+               len--;
        }
-       log(2, 0, 0, 0, "   Loaded plugin %s\n", plugin_name);
 }
 
-void remove_plugin(char *plugin_name)
+int find_filter(char const *name, size_t len)
 {
-       void *p;
-       int (*donefunc)();
-       char path[256] = {0};
+       int free = -1;
        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;
-
-       for (i = 0; i < max_plugin_functions; i++)
+       for (i = 0; i < MAXFILTER; i++)
        {
-               void *x;
-               if (!plugin_functions[i]) continue;
-               if ((x = dlsym(p, plugin_functions[i]))) ll_delete(plugins[i], x);
-       }
+               if (!*ip_filters[i].name)
+               {
+                       if (free < 0)
+                               free = i;
 
-       if (ll_contains(loaded_plugins, p))
-       {
-               ll_delete(loaded_plugins, p);
+                       continue;
+               }
 
-               donefunc = dlsym(p, "plugin_done");
-               if (donefunc) donefunc();
-       }
+               if (strlen(ip_filters[i].name) != len)
+                       continue;
 
-       dlclose(p);
-       dlclose(p);
-       log(2, 0, 0, 0, "Removed plugin %s\n", plugin_name);
+               if (!strncmp(ip_filters[i].name, name, len))
+                       return i;
+       }
+                       
+       return free;
 }
 
-int run_plugins(int plugin_type, void *data)
+static int ip_filter_port(ip_filter_portt *p, uint16_t port)
 {
-       int (*func)(void *data);
-       if (!plugins[plugin_type] || plugin_type > max_plugin_functions) return 1;
-
-       ll_reset(plugins[plugin_type]);
-       while ((func = ll_next(plugins[plugin_type])))
+       switch (p->op)
        {
-               int rc;
-               rc = func(data);
-               if (rc == PLUGIN_RET_STOP) return 1;
-               if (rc == PLUGIN_RET_ERROR) return 0;
+       case FILTER_PORT_OP_EQ:    return port == p->port;
+       case FILTER_PORT_OP_NEQ:   return port != p->port;
+       case FILTER_PORT_OP_GT:    return port > p->port;
+       case FILTER_PORT_OP_LT:    return port < p->port;
+       case FILTER_PORT_OP_RANGE: return port >= p->port && port <= p->port2;
        }
-       return 1;
+
+       return 0;
 }
 
-void processcontrol(u8 * buf, int len, struct sockaddr_in *addr)
+static int ip_filter_flag(uint8_t op, uint8_t sflags, uint8_t cflags, uint8_t flags)
 {
-       char *resp;
-       int l;
-       struct param_control param = { buf, len, ntohl(addr->sin_addr.s_addr), ntohs(addr->sin_port), NULL, 0, 0 };
+       switch (op)
+       {
+       case FILTER_FLAG_OP_ANY:
+               return (flags & sflags) || (~flags & cflags);
 
-       log(4, ntohl(addr->sin_addr.s_addr), 0, 0, "Received ");
-       dump_packet(buf, log_stream);
+       case FILTER_FLAG_OP_ALL:
+               return (flags & sflags) == sflags && (~flags & cflags) == cflags;
 
-       resp = calloc(1400, 1);
-       l = new_packet(PKT_RESP_ERROR, resp);
-       *(int *)(resp + 6) = *(int *)(buf + 6);
+       case FILTER_FLAG_OP_EST:
+               return (flags & (TCP_FLAG_ACK|TCP_FLAG_RST)) && (~flags & TCP_FLAG_SYN);
+       }
 
-       param.type = ntohs(*(short *)(buf + 2));
-       param.id = ntohl(*(int *)(buf + 6));
-       param.data_length = ntohs(*(short *)(buf + 4)) - 10;
-       param.data = (param.data_length > 0) ? (char *)(buf + 10) : NULL;
-       param.response = resp;
-       param.response_length = l;
+       return 0;
+}
 
-       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
+int ip_filter(uint8_t *buf, int len, uint8_t filter)
+{
+       uint16_t frag_offset;
+       uint8_t proto;
+       in_addr_t src_ip;
+       in_addr_t dst_ip;
+       uint16_t src_port = 0;
+       uint16_t dst_port = 0;
+       uint8_t flags = 0;
+       ip_filter_rulet *rule;
+
+       if (len < 20) // up to end of destination address
+               return 0;
+
+       if ((*buf >> 4) != 4) // IPv4
+               return 0;
+
+       frag_offset = ntohs(*(uint16_t *) (buf + 6)) & 0x1fff;
+       proto = buf[9];
+       src_ip = *(in_addr_t *) (buf + 12);
+       dst_ip = *(in_addr_t *) (buf + 16);
+
+       if (frag_offset == 0 && (proto == IPPROTO_TCP || proto == IPPROTO_UDP))
        {
-               run_plugins(PLUGIN_CONTROL, &param);
+               int l = (buf[0] & 0xf) * 4; // length of IP header
+               if (len < l + 4) // ports
+                       return 0;
+
+               src_port = ntohs(*(uint16_t *) (buf + l));
+               dst_port = ntohs(*(uint16_t *) (buf + l + 2));
+               if (proto == IPPROTO_TCP)
+               {
+                       if (len < l + 14) // flags
+                               return 0;
+
+                       flags = buf[l + 13] & 0x3f;
+               }
        }
 
-       if (param.send_response)
+       for (rule = ip_filters[filter].rules; rule->action; rule++)
        {
-               send_packet(controlfd, ntohl(addr->sin_addr.s_addr), ntohs(addr->sin_port), param.response, param.response_length);
-               log(4, ntohl(addr->sin_addr.s_addr), 0, 0, "Sent Control packet response\n");
+               if (rule->proto != IPPROTO_IP && proto != rule->proto)
+                       continue;
+
+               if (rule->src_wild != INADDR_BROADCAST &&
+                   (src_ip & ~rule->src_wild) != (rule->src_ip & ~rule->src_wild))
+                       continue;
+
+               if (rule->dst_wild != INADDR_BROADCAST &&
+                   (dst_ip & ~rule->dst_wild) != (rule->dst_ip & ~rule->dst_wild))
+                       continue;
+
+               if (frag_offset)
+               {
+                       // layer 4 deny rules are skipped
+                       if (rule->action == FILTER_ACTION_DENY &&
+                           (rule->src_ports.op || rule->dst_ports.op || rule->tcp_flag_op))
+                               continue;
+               }
+               else
+               {
+                       if (rule->frag)
+                               continue;
+
+                       if (proto == IPPROTO_TCP || proto == IPPROTO_UDP)
+                       {
+                               if (rule->src_ports.op && !ip_filter_port(&rule->src_ports, src_port))
+                                       continue;
+
+                               if (rule->dst_ports.op && !ip_filter_port(&rule->dst_ports, dst_port))
+                                       continue;
+
+                               if (proto == IPPROTO_TCP && rule->tcp_flag_op &&
+                                   !ip_filter_flag(rule->tcp_flag_op, rule->tcp_sflags, rule->tcp_cflags, flags))
+                                       continue;
+                       }
+               }
+
+               // matched
+               rule->counter++;
+               return rule->action == FILTER_ACTION_PERMIT;
        }
 
-       free(resp);
+       // default deny
+       return 0;
 }
-