backout regular_cleanup changes
[l2tpns.git] / l2tpns.c
index 2e0f8ca..e9f768a 100644 (file)
--- a/l2tpns.c
+++ b/l2tpns.c
@@ -1,8 +1,11 @@
 // L2TP Network Server
 // Adrian Kennard 2002
-// (c) Copyrigth 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd)
+// Copyright (c) 2003, 2004, 2005 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 1.73.2.17 2005/07/03 02:40:22 bodea Exp $";
+
 #include <arpa/inet.h>
 #include <assert.h>
 #include <errno.h>
 #include <sys/mman.h>
 #include <netdb.h>
 #include <netinet/in.h>
-#include <signal.h>
 #include <stdarg.h>
 #include <stdlib.h>
 #include <stdio.h>
-#define __USE_GNU
 #include <string.h>
+#include <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>
+#include <sched.h>
+#include <sys/sysinfo.h>
+#include <libcli.h>
+
 #include "md5.h"
 #include "l2tpns.h"
 #include "cluster.h"
 #include "constants.h"
 #include "control.h"
 #include "util.h"
+#include "tbf.h"
+
+#ifdef BGP
+#include "bgp.h"
+#endif /* BGP */
 
 // Globals
-struct configt *config = NULL; // all configuration
-int tapfd = -1;                        // tap interface file handle
+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 snoopfd = -1;              // UDP file handle for sending out intercept data
-int radfd = -1;                        // RADIUS requests file handle
+int *radfds = NULL;            // RADIUS requests file handles
 int ifrfd = -1;                        // File descriptor for routing, etc
 time_t basetime = 0;           // base clock
 char hostname[1000] = "";      // us.
-ipt myip = 0;                  // MY IP
-u16 tapmac[3];                 // MAC of tap interface
-int tapidx;                    // ifr_ifindex of tap device
-u32 sessionid = 0;             // session id for radius accounting
-int syslog_log = 0;            // are we logging to syslog
-FILE *log_stream = NULL;
-struct sockaddr_in snoop_addr = {0};
-extern int cluster_sockfd;
-unsigned long last_sid = 0;
-int handle_interface = 0;
-int clifd = 0;
-sessionidt *cli_session_kill = NULL;
-tunnelidt *cli_tunnel_kill = NULL;
-static void *ip_hash[256];
-unsigned long udp_tx = 0, udp_rx = 0, udp_rx_pkt = 0;
-unsigned long eth_tx = 0, eth_rx = 0, eth_rx_pkt = 0;
-unsigned int ip_pool_size = 0;
-time_t time_now;
-char time_now_string[64] = {0};
-char main_quit = 0;
-char *_program_name = NULL;
+static uint32_t sessionid = 0; // session id for radius accounting
+static int syslog_log = 0;     // are we logging to syslog
+static FILE *log_stream = NULL;        // file handle for direct logging (i.e. direct into file, not via syslog).
+extern int cluster_sockfd;     // Intra-cluster communications socket.
+uint32_t last_id = 0;          // Last used PPP SID. Can I kill this?? -- mo
+
+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.
+
+// 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.
+char main_quit = 0;                    // True if we're in the process of exiting.
 linked_list *loaded_plugins;
 linked_list *plugins[MAX_PLUGIN_TYPES];
 
 #define membersize(STRUCT, MEMBER) sizeof(((STRUCT *)0)->MEMBER)
-#define CONFIG(NAME, MEMBER, TYPE) { NAME, offsetof(struct configt, MEMBER), membersize(struct configt, MEMBER), TYPE }
+#define CONFIG(NAME, MEMBER, TYPE) { NAME, offsetof(configt, MEMBER), membersize(configt, MEMBER), TYPE }
 
-struct config_descriptt config_values[] = {
+config_descriptt config_values[] = {
        CONFIG("debug", debug, INT),
        CONFIG("log_file", log_filename, STRING),
+       CONFIG("pid_file", pid_file, STRING),
        CONFIG("l2tp_secret", l2tpsecret, STRING),
        CONFIG("primary_dns", default_dns1, IP),
        CONFIG("secondary_dns", default_dns2, IP),
        CONFIG("save_state", save_state, BOOL),
-       CONFIG("snoop_host", snoop_destination_host, IP),
-       CONFIG("snoop_port", snoop_destination_port, SHORT),
        CONFIG("primary_radius", radiusserver[0], IP),
        CONFIG("secondary_radius", radiusserver[1], IP),
+       CONFIG("primary_radius_port", radiusport[0], SHORT),
+       CONFIG("secondary_radius_port", radiusport[1], SHORT),
        CONFIG("radius_accounting", radius_accounting, BOOL),
        CONFIG("radius_secret", radiussecret, STRING),
        CONFIG("bind_address", bind_address, IP),
-       CONFIG("cluster_master", cluster_address, IP),
+       CONFIG("peer_address", peer_address, IP),
+       CONFIG("send_garp", send_garp, BOOL),
        CONFIG("throttle_speed", rl_rate, UNSIGNED_LONG),
+       CONFIG("throttle_buckets", num_tbfs, INT),
        CONFIG("accounting_dir", accounting_dir, STRING),
        CONFIG("setuid", target_uid, INT),
        CONFIG("dump_speed", dump_speed, BOOL),
+       CONFIG("cleanup_interval", cleanup_interval, INT),
+       CONFIG("multi_read_count", multi_read_count, INT),
+       CONFIG("scheduler_fifo", scheduler_fifo, BOOL),
+       CONFIG("lock_pages", lock_pages, BOOL),
+       CONFIG("icmp_rate", icmp_rate, INT),
+       CONFIG("packet_limit", max_packets, INT),
+       CONFIG("cluster_address", cluster_address, IP),
+       CONFIG("cluster_interface", cluster_interface, STRING),
+       CONFIG("cluster_hb_interval", cluster_hb_interval, INT),
+       CONFIG("cluster_hb_timeout", cluster_hb_timeout, INT),
+       CONFIG("cluster_master_min_adv", cluster_master_min_adv, INT),
        { NULL, 0, 0, 0 },
 };
 
-char *plugin_functions[] = {
+static char *plugin_functions[] = {
        NULL,
        "plugin_pre_auth",
        "plugin_post_auth",
@@ -113,35 +137,55 @@ char *plugin_functions[] = {
        "plugin_kill_session",
        "plugin_control",
        "plugin_radius_response",
+       "plugin_become_master",
+       "plugin_new_session_master",
 };
+
 #define max_plugin_functions (sizeof(plugin_functions) / sizeof(char *))
 
-tunnelt *tunnel = NULL;                // 1000 * 45 = 45000 = 45k
-sessiont *session = NULL;      // 5000 * 213 = 1065000 = 1 Mb
-radiust *radius = NULL;
-ippoolt *ip_address_pool = NULL;
-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.
+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();
-void tunnel_clean();
-tunnelidt new_tunnel();
-void update_config();
-
-// 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 free_ip_address(sessionidt s);
+static void dump_acct_info(int all);
+static void sighup_handler(int sig);
+static void sigalrm_handler(int sig);
+static void sigterm_handler(int sig);
+static void sigquit_handler(int sig);
+static void sigchild_handler(int sig);
+static void read_state(void);
+static void dump_state(void);
+static void build_chap_response(char *challenge, uint8_t id, uint16_t challenge_length, char **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);
+static tunnelidt new_tunnel(void);
+static int unhide_avp(uint8_t *avp, tunnelidt t, sessionidt s, uint16_t length);
+
+// on slaves, alow BGP to withdraw cleanly before exiting
+#define QUIT_DELAY     5
+
+// return internal time (10ths since process startup)
+static clockt now(void)
 {
        struct timeval t;
        gettimeofday(&t, 0);
@@ -149,16 +193,22 @@ clockt now(void)
 }
 
 // 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);
 }
 
-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[65535] = {0};
-       static char message2[65535] = {0};
+       static char message[65536] = {0};
+       static char message2[65536] = {0};
        va_list ap;
 
 #ifdef RINGBUFFER
@@ -171,7 +221,6 @@ 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);
@@ -187,7 +236,7 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
        {
                vsnprintf(message2, 65535, format, ap);
                snprintf(message, 65535, "%s %02d/%02d %s", time_now_string, t, s, message2);
-               fprintf(log_stream, message);
+               fprintf(log_stream, "%s", message);
        }
        else if (syslog_log)
        {
@@ -198,17 +247,17 @@ void _log(int level, ipt address, sessionidt s, tunnelidt t, const char *format,
        va_end(ap);
 }
 
-void _log_hex(int level, ipt address, sessionidt s, tunnelidt t, const char *title, const char *data, int maxsize)
+void _log_hex(int level, const char *title, const char *data, int maxsize)
 {
        int i, j;
-       unsigned const char *d = (unsigned const char *)data;
+       const uint8_t *d = (const uint8_t *) data;
 
        if (config->debug < level) return;
 
-       // No support for log_hex to syslog
+       // No support for _log_hex to syslog
        if (log_stream)
        {
-               log(level, address, s, t, "%s (%d bytes):\n", title, maxsize);
+               _log(level, 0, 0, "%s (%d bytes):\n", title, maxsize);
                setvbuf(log_stream, NULL, _IOFBF, 16384);
 
                for (i = 0; i < maxsize; )
@@ -251,80 +300,122 @@ void _log_hex(int level, ipt address, sessionidt s, tunnelidt t, const char *tit
 
 
 // Add a route
-void routeset(ipt ip, ipt mask, ipt gw, u8 add)
+//
+// This adds it to the routing table, advertises it
+// via 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 = config->tapdevice;
+       r.rt_dev = config->tundevice;
        r.rt_dst.sa_family = AF_INET;
-       *(u32 *) & (((struct sockaddr_in *) &r.rt_dst)->sin_addr.s_addr) = htonl(ip);
+       *(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 ? 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);
+       }
 }
 
-// Set up TAP interface
-void inittap(void)
+//
+// Set up TUN interface
+static void inittun(void)
 {
        struct ifreq ifr;
        struct sockaddr_in sin = {0};
        memset(&ifr, 0, sizeof(ifr));
        ifr.ifr_flags = IFF_TUN;
 
-       tapfd = open(TAPDEVICE, O_RDWR);
-       if (tapfd < 0)
+       tunfd = open(TUNDEVICE, O_RDWR);
+       if (tunfd < 0)
        {                          // fatal
-               log(0, 0, 0, 0, "Can't open %s: %s\n", TAPDEVICE, strerror(errno));
-               exit(-1);
+               LOG(0, 0, 0, "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(config->tapdevice));
-       strncpy(config->tapdevice, ifr.ifr_name, sizeof(config->tapdevice) - 1);
+       assert(strlen(ifr.ifr_name) < sizeof(config->tundevice));
+       strncpy(config->tundevice, ifr.ifr_name, sizeof(config->tundevice) - 1);
        ifrfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 
        sin.sin_family = AF_INET;
-       sin.sin_addr.s_addr = handle_interface ? config->bind_address : 0x01010101; // 1.1.1.1
+       sin.sin_addr.s_addr = config->bind_address ? config->bind_address : 0x01010101; // 1.1.1.1
        memcpy(&ifr.ifr_addr, &sin, sizeof(struct sockaddr));
 
        if (ioctl(ifrfd, SIOCSIFADDR, (void *) &ifr) < 0)
        {
-               perror("set tap addr");
-               exit( -1);
-       }
-       ifr.ifr_flags = IFF_UP;
-       if (ioctl(ifrfd, SIOCSIFFLAGS, (void *) &ifr) < 0)
-       {
-               perror("set tap flags");
-               exit( -1);
+               LOG(0, 0, 0, "Error setting tun address: %s\n", strerror(errno));
+               exit(1);
        }
-       if (ioctl(ifrfd, SIOCGIFHWADDR, (void *) &ifr) < 0)
+       /* Bump up the qlen to deal with bursts from the network */
+       ifr.ifr_qlen = 1000;
+       if (ioctl(ifrfd, SIOCSIFTXQLEN, (void *) &ifr) < 0)
        {
-               perror("get tap hwaddr");
-               exit( -1);
+               LOG(0, 0, 0, "Error setting tun queue length: %s\n", strerror(errno));
+               exit(1);
        }
-       memcpy(&tapmac, 2 + (u8 *) & ifr.ifr_hwaddr, 6);
-       if (ioctl(ifrfd, SIOCGIFINDEX, (void *) &ifr) < 0)
+       ifr.ifr_flags = IFF_UP;
+       if (ioctl(ifrfd, SIOCSIFFLAGS, (void *) &ifr) < 0)
        {
-               perror("get tap ifindex");
-               exit( -1);
+               LOG(0, 0, 0, "Error setting tun flags: %s\n", strerror(errno));
+               exit(1);
        }
-       tapidx = ifr.ifr_ifindex;
 }
 
 // set up UDP port
-void initudp(void)
+static void initudp(void)
 {
        int on = 1;
        struct sockaddr_in addr;
@@ -334,250 +425,255 @@ void initudp(void)
        addr.sin_family = AF_INET;
        addr.sin_port = htons(L2TPPORT);
        addr.sin_addr.s_addr = config->bind_address;
-       udpfd = socket(AF_INET, SOCK_DGRAM, UDP);
+       udpfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
        setsockopt(udpfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
+       {
+               int flags = fcntl(udpfd, F_GETFL, 0);
+               fcntl(udpfd, F_SETFL, flags | O_NONBLOCK);
+       }
        if (bind(udpfd, (void *) &addr, sizeof(addr)) < 0)
        {
-               perror("udp bind");
-               exit( -1);
+               LOG(0, 0, 0, "Error in UDP bind: %s\n", strerror(errno));
+               exit(1);
        }
-       snoopfd = socket(AF_INET, SOCK_DGRAM, UDP);
+       snoopfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
 
        // Control
        memset(&addr, 0, sizeof(addr));
        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));
        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);
        }
 }
 
-// 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 = (unsigned 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;
 
-       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 (int) (intptr_t) d[(size_t) *a];
+}
+
+sessionidt sessionbyip(in_addr_t ip)
+{
+       int s = lookup_ipmap(ip);
+       CSTAT(call_sessionbyip);
+
+       if (s > 0 && s < MAXSESSION && session[s].opened)
+               return (sessionidt) 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 = (unsigned 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)
 {
-       unsigned char *a = (unsigned char *) &ip;
-       char **d = (char **) ip_hash;
-       int i;
+       cache_ipmap(ip, 0);     // Assign it to the NULL session.
+}
 
-       for (i = 0; i < 3; i++)
+//
+// CLI list to dump current ipcache.
+//
+int cmd_show_ipcache(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       char **d = (char **) ip_hash, **e, **f, **g;
+       int i, j, k, l;
+       int count = 0;
+
+       if (CLI_HELP_REQUESTED)
+               return CLI_HELP_NO_ARGS;
+
+       cli_print(cli, "%7s %s", "Sess#", "IP Address");
+
+       for (i = 0; i < 256; ++i)
        {
-               if (!d[(size_t) a[i]]) return;
-               d = (char **) d[(size_t) a[i]];
+               if (!d[i])
+                       continue;
+               e = (char **) d[i];
+               for (j = 0; j < 256; ++j)
+               {
+                       if (!e[j])
+                               continue;
+                       f = (char **) e[j];
+                       for (k = 0; k < 256; ++k)
+                       {
+                               if (!f[k])
+                                       continue;
+                               g = (char **)f[k];
+                               for (l = 0; l < 256; ++l)
+                               {
+                                       if (!g[l])
+                                               continue;
+                                       cli_print(cli, "%7d %d.%d.%d.%d", (int) (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 users aren't authenticated, so the username is
 // reasonably useless. Ignore them to avoid incorrect actions
+//
+// This is VERY inefficent. Don't call it often. :)
+//
 sessionidt sessionbyuser(char *username)
 {
        int s;
-#ifdef STAT_CALLS
-       STAT(call_sessionbyuser);
-#endif
-       for (s = 1; s < MAXSESSION && (session[s].walled_garden || strncmp(session[s].user, username, 128)); s++);
-       if (s < MAXSESSION)
-               return s;
-       return 0;
-}
+       CSTAT(call_sessionbyuser);
 
-void send_garp(ipt ip)
-{
-    int s;
-    struct ifreq ifr;
-    unsigned char mac[6];
-
-    s = socket(PF_INET, SOCK_DGRAM, 0);
-    if (s < 0)
-    {
-      perror("socket");
-      exit(-1);
-    }
-    memset(&ifr, 0, sizeof(ifr));
-    strncpy(ifr.ifr_name, "eth0", sizeof(ifr.ifr_name) - 1);
-    if (ioctl(s, SIOCGIFHWADDR, &ifr) < 0)
-    {
-      perror("get eth0 hwaddr");
-      exit(-1);
-    }
-    memcpy(mac, &ifr.ifr_hwaddr.sa_data, 6*sizeof(char));
-    if (ioctl(s, SIOCGIFINDEX, &ifr) < 0)
-    {
-      perror("get eth0 ifindex");
-      exit(-1);
-    }
-    close(s);
-    sendarp(ifr.ifr_ifindex, mac, ip);
-}
+       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))
-       {
-               log(0, 0, 0, 0, "Unexpected ARP prot type %04X\n", ntohs(*(u16 *) (buf + 20)));
-               STAT(arp_errors);
-               return;
-       }
-       if (buf[22] != 6)
+       s = socket(PF_INET, SOCK_DGRAM, 0);
+       if (s < 0)
        {
-               log(0, 0, 0, 0, "Unexpected ARP hard len %d\n", buf[22]);
-               STAT(arp_errors);
+               LOG(0, 0, 0, "Error creating socket for GARP: %s\n", strerror(errno));
                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(call_tunnelsend);
+
+       if (!t)
+       {
+               static int backtrace_count = 0;
+               LOG(0, 0, t, "tunnelsend called with 0 as tunnel id\n");
+               STAT(tunnel_tx_errors);
+               log_backtrace(backtrace_count, 5)
+               return;
+       }
+
        if (!tunnel[t].ip)
        {
-               log(1, 0, 0, t, "Error sending data out tunnel: no remote endpoint (tunnel not set up)\n");
+               static int backtrace_count = 0;
+               LOG(1, 0, t, "Error sending data out tunnel: no remote endpoint (tunnel not set up)\n");
+               log_backtrace(backtrace_count, 5)
                STAT(tunnel_tx_errors);
                return;
        }
+
        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)
@@ -587,38 +683,58 @@ void tunnelsend(u8 * buf, u16 l, tunnelidt t)
                if (tunnel[t].try > 1)
                {
                        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);
 }
 
+//
+// Tiny helper function to write data to
+// the 'tun' device.
+//
+int tun_write(uint8_t * data, int size)
+{
+       return write(tunfd, data, size);
+}
+
 // process outgoing (to tunnel) IP
-void processipout(u8 * buf, int len)
+//
+static void processipout(uint8_t * buf, int len)
 {
        sessionidt s;
        sessiont *sp;
        tunnelidt t;
-       ipt ip;
-       u8 b[MAXETHER];
-#ifdef STAT_CALLS
-       STAT(call_processipout);
-#endif
-       if (len < 38)
+       in_addr_t ip;
+
+       char *data = buf;       // Keep a copy of the originals.
+       int size = len;
+
+       uint8_t b[MAXETHER + 20];
+
+       CSTAT(call_processipout);
+
+       if (len < MIN_IP_SIZE)
        {
-               log(1, 0, 0, 0, "Short IP, %d bytes\n", len);
-               STAT(tunnel_tx_errors);
+               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;
        }
 
@@ -627,90 +743,210 @@ void processipout(u8 * buf, int len)
        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: Sending ICMP host unreachable to %s\n", inet_toa(*(u32 *)(buf + 12)));
-               host_unreachable(*(u32 *)(buf + 12), *(u16 *)(buf + 4), ip, buf, (len < 64) ? 64 : len);
+               // Is this a packet for a session that doesn't exist?
+               static int rate = 0;    // Number of ICMP packets we've sent this second.
+               static int last = 0;    // Last time we reset the ICMP packet counter 'rate'.
+
+               if (last != time_now)
+               {
+                       last = time_now;
+                       rate = 0;
+               }
+
+               if (rate++ < config->icmp_rate) // Only send a max of icmp_rate per second.
+               {
+                       LOG(4, 0, 0, "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), ip, buf, (len < 64) ? 64 : len);
+               }
                return;
        }
        t = session[s].tunnel;
        sp = &session[s];
 
-       // Snooping this session, send it to ASIO
-       if (sp->snoop) snoop_send_packet(buf, len);
+       // 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;
+
+       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, t, s, PPPIP);
+               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);
+
+       sp->cout += len; // byte count
+       sp->total_cout += len; // byte count
+       sp->pout++;
+       udp_tx += len;
+       sess_local[s].cout += len;      // To send to master..
+}
+
+//
+// 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, t, s, PPPIP);
+               if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
-               sp->cout += len; // byte count
-               sp->total_cout += len; // byte count
-               sp->pout++;
-               udp_tx += len;
        }
+
+       // Snooping this session.
+       if (sp->snoop_ip && sp->snoop_port)
+               snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
+
+       sp->cout += len; // byte count
+       sp->total_cout += len; // byte count
+       sp->pout++;
+       udp_tx += len;
+       sess_local[s].cout += len;      // To send to master..
 }
 
 // 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)
+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, char *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)
@@ -722,42 +958,46 @@ 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, tunnelidt t, sessionidt s)
 {
-       *(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(s ? session[s].far : 0); // 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)
        {
@@ -766,37 +1006,121 @@ 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)
 {
-       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(call_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 %d: %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 (session[s].ip && !walled_garden && !session[s].die)
+       {
+               // RADIUS Stop message
+               uint16_t r = session[s].radius;
                if (!r)
                {
                        if (!(r = radiusnew(s)))
                        {
-                               log(1, 0, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");
-                               STAT(radius_overflow);
+                               LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");
                        }
                        else
                        {
@@ -805,92 +1129,155 @@ void sessionshutdown(sessionidt s, char *reason)
                                        radius[r].auth[n] = rand();
                        }
                }
+
                if (r && radius[r].state != RADIUSSTOP)
                        radiussend(r, RADIUSSTOP); // stop, if not already trying
+
+               // 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(s);
+               else
+                       free_ip_address(s);
        }
+
+       if (session[s].throttle_in || session[s].throttle_out) // Unthrottle if throttled.
+               throttle_session(s, 0, 0);
+
        {                            // Send CDN
                controlt *c = controlnew(14); // sending CDN
                control16(c, 1, 3, 1);    // result code (admin reasons - TBA make error, general error, add message
                control16(c, 14, s, 1);   // assigned session (our end)
                controladd(c, session[s].tunnel, s); // send the message
        }
+
+       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--;
+
        cluster_send_session(s);
 }
 
 void sendipcp(tunnelidt t, sessionidt s)
 {
-       u8 buf[MAXCONTROL];
-       u8 r = session[s].radius;
-       u8 *q;
-#ifdef STAT_CALLS
-       STAT(call_sendipcp);
-#endif
+       uint8_t buf[MAXCONTROL];
+       uint16_t r = session[s].radius;
+       uint8_t *q;
+
+       CSTAT(call_sendipcp);
+
        if (!r)
                r = radiusnew(s);
-       if (radius[r].state != RADIUSIPCP)
-       {
+
+       if (!r)
+       {
+               sessionshutdown(s, "No free RADIUS sessions for IPCP");
+               return;
+       }
+
+       if (radius[r].state != RADIUSIPCP)
+       {
                radius[r].state = RADIUSIPCP;
                radius[r].try = 0;
        }
+
        radius[r].retry = backoff(radius[r].try++);
        if (radius[r].try > 10)
        {
+               radiusclear(r, s);      // Clear radius session.
                sessionshutdown(s, "No reply on IPCP");
                return;
        }
-       q = makeppp(buf, 0, 0, t, s, PPPIPCP);
+
+       q = makeppp(buf,sizeof(buf), 0, 0, t, s, PPPIPCP);
+       if (!q) return;
+
        *q = ConfigReq;
-       q[1] = r;                    // ID, dont care, we only send one type of request
-       *(u16 *) (q + 2) = htons(10);
+       q[1] = r >> RADIUS_SHIFT;                    // ID, dont care, we only send one type of request
+       *(uint16_t *) (q + 2) = htons(10);
        q[4] = 3;
        q[5] = 6;
-       *(u32 *) (q + 6) = htonl(myip ? myip : session[s].ip); // send my IP (use theirs if I dont have one)
+       *(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
+       session[s].flags &= ~SF_IPCP_ACKED;     // Clear flag.
+}
+
+static void sessionclear(sessionidt s)
+{
+       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;
 }
 
 // kill a session now
 void sessionkill(sessionidt s, char *reason)
 {
-#ifdef STAT_CALLS
-       STAT(call_sessionkill);
-#endif
+
+       CSTAT(call_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 (%d)\n", session[s].next);
+               return;
+       }
+
+       session[s].die = TIME;
        sessionshutdown(s, reason);  // close radius/routes, etc.
        if (session[s].radius)
-               radiusclear(session[s].radius, 0); // cant send clean accounting data, session is killed
-       memset(&session[s], 0, sizeof(session[s]));
-       session[s].next = sessionfree;
-       sessionfree = s;
-       log(2, 0, s, session[s].tunnel, "Kill session %d: %s\n", s, reason);
+               radiusclear(session[s].radius, s); // cant send clean accounting data, session is killed
+
+       LOG(2, s, session[s].tunnel, "Kill session %d (%s): %s\n", s, session[s].user, reason);
+       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;
+}
+
 // 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(call_tunnelkill);
 
        tunnel[t].state = TUNNELDIE;
 
@@ -904,159 +1291,200 @@ void tunnelkill(tunnelidt t, char *reason)
                controlfree = c;
        }
        // kill sessions
-       for (s = 1; s < MAXSESSION; s++)
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                if (session[s].tunnel == t)
                        sessionkill(s, reason);
 
        // free tunnel
        tunnelclear(t);
+       LOG(1, 0, t, "Kill tunnel %d: %s\n", t, reason);
+       cli_tunnel_actions[s].action = 0;
        cluster_send_tunnel(t);
-       log(1, 0, 0, t, "Kill tunnel %d: %s\n", t, reason);
-       tunnel[t].die = 0;
-       tunnel[t].state = TUNNELFREE;
 }
 
 // shut down a tunnel cleanly
-void tunnelshutdown(tunnelidt t, char *reason)
+static void tunnelshutdown(tunnelidt t, char *reason)
 {
        sessionidt s;
-#ifdef STAT_CALLS
-       STAT(call_tunnelshutdown);
-#endif
+
+       CSTAT(call_tunnelshutdown);
+
        if (!tunnel[t].last || !tunnel[t].far || tunnel[t].state == TUNNELFREE)
        {
                // never set up, can immediately kill
                tunnelkill(t, reason);
                return;
        }
-       log(1, 0, 0, t, "Shutting down tunnel %d (%s)\n", t, reason);
+       LOG(1, 0, t, "Shutting down tunnel %d (%s)\n", t, reason);
 
        // close session
-       for (s = 1; s < MAXSESSION; s++)
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                if (session[s].tunnel == t)
-                       sessionkill(s, reason);
+                       sessionshutdown(s, reason);
 
        tunnel[t].state = TUNNELDIE;
-       tunnel[t].die = now() + 700; // Clean up in 70 seconds
+       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, 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
        }
 }
 
 // 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;
+       uint16_t l = len, t = 0, s = 0, ns = 0, nr = 0;
+       uint8_t *p = buf + 2;
+
+
+       CSTAT(call_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) != 2);
                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 (*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
+               uint16_t message = 0xFFFF;      // message type
+               uint8_t fatal = 0;
+               uint8_t mandatorymessage = 0;
+               uint8_t chap = 0;               // if CHAP being used
+               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)
+               int requestchap = 0;            // do we request PAP instead of original CHAP request?
+               char called[MAXTEL] = "";       // called number
+               char calling[MAXTEL] = "";      // calling number
+
+               if (!config->cluster_iam_master)
+               {
+                       master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                       return;
+               }
+
                if ((*buf & 0xCA) != 0xC8)
                {
-                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Bad control header %02X\n", *buf);
+                       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);
+
+               // check for duplicate tunnel open message
+               if (!t && ns == 0)
+               {
+                       int i;
+
+                               //
+                               // Is this a duplicate of the first packet? (SCCRQ)
+                               //
+                       for (i = 1; i <= config->cluster_highest_tunnelid ; ++i)
+                       {
+                               if (tunnel[i].state != TUNNELOPENING ||
+                                       tunnel[i].ip != ntohl(*(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 %d l-nr %d r-ns %d r-nr %d\n",
+                       l, tunnel[t].controlc, tunnel[t].ns, tunnel[t].nr, ns, nr);
+
                // if no tunnel specified, assign one
                if (!t)
                {
                        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;
                        }
                        tunnelclear(t);
-                       tunnel[t].ip = ntohl(*(ipt *) & addr->sin_addr);
+                       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 %d\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 %d, (%d is not the expected %d)\n",
+                               t, ns, tunnel[t].nr);
+
+                       if (l)  // Is this not a ZLB?
+                               controlnull(t);
+                       return;
                }
 
                // This is used to time out old tunnels
@@ -1065,29 +1493,20 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                // check sequence of this message
                {
                        int skip = tunnel[t].window; // track how many in-window packets are still in queue
-                       if (tunnel[t].controlc)
-                       {                  // some to clear maybe
-                               while (tunnel[t].controlc && (((tunnel[t].ns - tunnel[t].controlc) - nr) & 0x8000))
-                               {
-                                       controlt *c = tunnel[t].controls;
-                                       tunnel[t].controls = c->next;
-                                       tunnel[t].controlc--;
-                                       c->next = controlfree;
-                                       controlfree = c;
-                                       skip--;
-                                       tunnel[t].try = 0; // we have progress
-                               }
-                       }
-                       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
                        }
+
                        // receiver advance (do here so quoted correctly in any sends below)
-                       if (l) tunnel[t].nr++;
+                       if (l) tunnel[t].nr = (ns + 1);
                        if (skip < 0) skip = 0;
                        if (skip < tunnel[t].controlc)
                        {
@@ -1115,14 +1534,14 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                        // process AVPs
                        while (l && !(fatal & 0x80))
                        {
-                               u16 n = (ntohs(*(u16 *) p) & 0x3FF);
-                               u8 *b = p;
-                               u8 flags = *p;
-                               u16 mtype;
+                               uint16_t n = (ntohs(*(uint16_t *) p) & 0x3FF);
+                               uint8_t *b = p;
+                               uint8_t flags = *p;
+                               uint16_t mtype;
                                p += n;       // next
                                if (l < n)
                                {
-                                       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;
@@ -1133,159 +1552,169 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        // handle hidden AVPs
                                        if (!*config->l2tpsecret)
                                        {
-                                               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;
                                                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;
+                                               continue;
+                                       }
+                                       LOG(4, s, t, "Hidden AVP\n");
+                                       // Unhide the AVP
+                                       n = unhide_avp(b, t, s, n);
+                                       if (n == 0)
+                                       {
                                                fatal = flags;
                                                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);
+                                       LOG(1, s, t, "Unrecognised AVP flags %02X\n", *b);
                                        fatal = flags;
                                        continue; // next
                                }
                                b += 2;
-                               if (*(u16 *) (b))
+                               if (*(uint16_t *) (b))
                                {
-                                       log(2, ntohl(addr->sin_addr.s_addr), s, t, "Unknown AVP vendor %d\n", ntohs(*(u16 *) (b)));
+                                       LOG(2, s, t, "Unknown AVP vendor %d\n", ntohs(*(uint16_t *) (b)));
                                        fatal = flags;
                                        continue; // next
                                }
                                b += 2;
-                               mtype = ntohs(*(u16 *) (b));
+                               mtype = ntohs(*(uint16_t *) (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 %d (%s) len %d\n", mtype, avpnames[mtype], n);
                                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]);
+                                       message = ntohs(*(uint16_t *) b);
+                                       LOG(4, s, t, "   Message type = %d (%s)\n", *b, l2tp_message_types[message]);
                                        mandatorymessage = flags;
                                        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 (message == 4)
+                                               { /* StopCCN */
                                                        if (rescode <= MAX_STOPCCN_RESULT_CODE)
                                                                resdesc = stopccn_result_codes[rescode];
-                                               } else if (message == 14) { /* CDN */
+                                               }
+                                               else if (message == 14)
+                                               { /* CDN */
                                                        if (rescode <= MAX_CDN_RESULT_CODE)
                                                                resdesc = cdn_result_codes[rescode];
                                                }
 
-                                               log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Result Code %d: %s\n",
-                                                       rescode, resdesc);
-                                               if (n >= 4) {
-                                                       u16 errcode = ntohs(*(u16 *)(b + 2));
+                                               LOG(4, s, t, "   Result Code %d: %s\n", rescode, resdesc);
+                                               if (n >= 4)
+                                               {
+                                                       uint16_t errcode = ntohs(*(uint16_t *)(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);
-                                               }
-                                               if (n > 4) {
-                                                       /* %*s doesn't work?? */
-                                                       char *buf = (char *)strndup(b+4, n-4);
-                                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   Error String: %s\n",
-                                                               buf);
-                                                       free(buf);
+                                                       LOG(4, s, t, "   Error Code %d: %s\n", errcode, errdesc);
                                                }
+                                               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 = %d\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;
                                                        continue; // next
                                                }
                                        }
                                        break;
                                case 3:     // framing capabilities
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Framing capabilities\n");
+//                                     LOG(4, s, t, "Framing capabilities\n");
                                        break;
                                case 4:     // bearer capabilities
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Bearer capabilities\n");
+//                                     LOG(4, 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");
+//                                     LOG(4, s, t, "Tie breaker\n");
                                        continue;
                                case 6:     // firmware revision
-//                                     log(4, ntohl(addr->sin_addr.s_addr), s, t, "Firmware revision\n");
+//                                     LOG(4, 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);
+                                       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) - 1) ? sizeof(tunnel[t].vendor) - 1 : n);
+                                       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 = %d\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 = %d\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 = %d\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 = %d\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 = %d\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 = %d\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);
+                                       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);
+                                       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
                                        {
@@ -1294,13 +1723,12 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                memcpy(tmp, b, (n >= 30) ? 30 : n);
                                                session[s].tx_connect_speed = atol(tmp);
                                        }
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   TX connect speed <%d>\n",
-                                                       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
                                        {
@@ -1309,20 +1737,18 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                                memcpy(tmp, b, (n >= 30) ? 30 : n);
                                                session[s].rx_connect_speed = atol(tmp);
                                        }
-                                       log(4, ntohl(addr->sin_addr.s_addr), s, t, "   RX connect speed <%d>\n",
-                                                       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]);
+                                               uint16_t authtype = ntohs(*(uint16_t *)b);
+                                               LOG(4, s, t, "   Proxy Auth Type %d (%s)\n", authtype, authtypes[authtype]);
                                                requestchap = (authtype == 2);
                                                break;
                                        }
@@ -1330,21 +1756,20 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        {
                                                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",
+                                               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);
+                                               uint16_t authid = ntohs(*(uint16_t *)(b));
+                                               LOG(4, s, t, "   Proxy Auth ID (%d)\n", authid);
                                                if (session[s].radius)
                                                        radius[session[s].radius].id = authid;
                                                break;
@@ -1353,56 +1778,40 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        {
                                                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");
+                                               LOG(4, s, t, "   Proxy Auth Response\n");
                                                break;
                                        }
                                case 27:    // last send 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)
+                                                       if (*p == 5 && p[1] == 6) // Magic-Number
+                                                               amagic = ntohl(*(uint32_t *) (p + 2));
+                                                       else if (*p == 3 && p[1] == 5 && *(uint16_t *) (p + 2) == htons(PPPCHAP) && p[4] == 5) // Authentication-Protocol
                                                                chap = 1;
-                                                       else if (*p == 7)
+                                                       else if (*p == 7) // Protocol-Field-Compression
                                                                aflags |= SESSIONPFC;
-                                                       else if (*p == 8)
+                                                       else if (*p == 8) // Address-and-Control-Field-Compression
                                                                aflags |= SESSIONACFC;
                                                        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));
                                        memcpy(session[s].random_vector, b, n);
                                        session[s].random_vector_length = n;
                                        break;
                                default:
-                                       log(2, ntohl(addr->sin_addr.s_addr), s, t, "   Unknown AVP type %d\n", mtype);
+                                       LOG(2, s, t, "   Unknown AVP type %d\n", mtype);
                                        fatal = flags;
                                        continue; // next
                                }
@@ -1453,32 +1862,36 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        if (!sessionfree)
                                        {
                                                STAT(session_overflow);
-                                               tunnelshutdown(t, "No free sessions");
+                                               LOG(1, 0, t, "No free sessions\n");
+                                               return;
                                        }
                                        else
                                        {
-                                               u8 r;
+                                               uint16_t r;
                                                controlt *c;
 
                                                s = sessionfree;
                                                sessionfree = session[s].next;
                                                memset(&session[s], 0, sizeof(session[s]));
 
+                                               if (s > config->cluster_highest_sessionid)
+                                                       config->cluster_highest_sessionid = s;
+
                                                // make a RADIUS session
                                                if (!(r = radiusnew(s)))
                                                {
-                                                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "No free RADIUS sessions for ICRQ\n");
-                                                       sessionkill(s, "no free RADIUS sesions");
+                                                       LOG(1, s, t, "No free RADIUS sessions for ICRQ\n");
+                                                       sessionclear(s);
                                                        return;
                                                }
 
                                                c = controlnew(11); // sending ICRP
                                                session[s].id = sessionid++;
-                                               session[s].opened = time(NULL);
+                                               session[s].opened = time_now;
                                                session[s].tunnel = t;
                                                session[s].far = asession;
                                                session[s].last_packet = time_now;
-                                               log(3, ntohl(addr->sin_addr.s_addr), s, t, "New session (%d/%d)\n", tunnel[t].far, session[s].far);
+                                               LOG(3, s, t, "New session (%d/%d)\n", tunnel[t].far, session[s].far);
                                                control16(c, 14, s, 1); // assigned session
                                                controladd(c, t, s); // send the reply
                                                {
@@ -1497,9 +1910,10 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        // TBA
                                        break;
                                case 12:      // ICCN
+                                       if (amagic == 0) amagic = time_now;
                                        session[s].magic = amagic; // set magic number
-                                       session[s].flags = aflags; // set flags received
-                                       log(3, ntohl(addr->sin_addr.s_addr), s, t, "Magic %X Flags %X\n", amagic, aflags);
+                                       session[s].l2tp_flags = aflags; // set flags received
+                                       LOG(3, s, t, "Magic %X Flags %X\n", amagic, aflags);
                                        controlnull(t); // ack
                                        // In CHAP state, request PAP instead
                                        if (requestchap)
@@ -1510,14 +1924,14 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                                        sessionshutdown(s, "Closed (Received CDN)");
                                        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");
                                        else
-                                               log(1, ntohl(addr->sin_addr.s_addr), s, t, "Unknown message type %d\n", message);
+                                               LOG(1, s, t, "Unknown message type %d\n", message);
                                        break;
                                }
                        if (chapresponse) free(chapresponse);
@@ -1525,19 +1939,14 @@ 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 prot;
 
-               log_hex(5, "Receive Tunnel Data", p, l);
-               if (session[s].die)
-               {
-                       log(3, ntohl(addr->sin_addr.s_addr), s, t, "Session %d is closing. Don't process PPP packets\n", s);
-                       return;              // closing session, PPP not processed
-               }
+               LOG_HEX(5, "Receive Tunnel Data", p, l);
                if (l > 2 && p[0] == 0xFF && p[1] == 0x03)
                {                     // HDLC address header, discard
                        p += 2;
@@ -1545,7 +1954,7 @@ 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;
                }
@@ -1556,416 +1965,871 @@ void processudp(u8 * buf, int len, struct sockaddr_in *addr)
                }
                else
                {
-                       prot = ntohs(*(u16 *) p);
+                       prot = ntohs(*(uint16_t *) p);
                        p += 2;
                        l -= 2;
                }
+
+               if (s && !session[s].opened)    // Is something wrong??
+               {
+                       if (!config->cluster_iam_master)
+                       {
+                               // Pass it off to the master to deal with..
+                               master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                               return;
+                       }
+
+                       LOG(1, s, t, "UDP packet contains session which is not opened.  Dropping packet.\n");
+                       STAT(tunnel_rx_errors);
+                       return;
+               }
+
                if (prot == PPPPAP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processpap(t, s, p, l);
                }
                else if (prot == PPPCHAP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processchap(t, s, p, l);
                }
                else if (prot == PPPLCP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processlcp(t, s, p, l);
                }
                else if (prot == PPPIPCP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processipcp(t, s, p, l);
                }
                else if (prot == PPPCCP)
                {
                        session[s].last_packet = time_now;
+                       if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processccp(t, s, p, l);
                }
                else if (prot == PPPIP)
                {
+                       if (session[s].die)
+                       {
+                               LOG(4, s, t, "Session %d is closing.  Don't process PPP packets\n", s);
+                               return;              // closing session, PPP not processed
+                       }
+
                        session[s].last_packet = time_now;
+                       if (session[s].walled_garden && !config->cluster_iam_master)
+                       {
+                               master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
+                               return;
+                       }
+
                        processipin(t, s, p, l);
                }
                else
                {
                        STAT(tunnel_rx_errors);
-                       log(1, ntohl(addr->sin_addr.s_addr), s, t, "Unknown PPP protocol %04X\n", prot);
+                       LOG(1, s, t, "Unknown PPP protocol %04X\n", prot);
                }
        }
 }
 
-// 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(call_processtun);
+
        eth_rx_pkt++;
        eth_rx += len;
        if (len < 22)
        {
-               log(1, 0, 0, 0, "Short tap packet %d bytes\n", len);
-               STAT(tap_rx_errors);
+               LOG(1, 0, 0, "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)) // IP
                processipout(buf, len);
-       else
-       {
-               log(1, 0, 0, 0, "Unexpected tap packet %04X, %d bytes\n", ntohs(*(u16 *) (buf + 2)), len);
-       }
+       // Else discard.
 }
 
-// main loop - gets packets on tap or udp and processes them
-void mainloop(void)
+//
+// Maximum number of actions to complete.
+// This is to avoid sending out too many packets
+// at once.
+#define MAX_ACTIONS 500
+
+static int regular_cleanups(void)
 {
-       fd_set cr;
-       int cn;
-       u8 buf[65536];
-       struct timeval to;
-       clockt slow = now();       // occasional functions like session/tunnel expiry, tunnel hello, etc
-       clockt next_acct = slow + ACCT_TIME;
-       clockt next_cluster_ping = slow + 50;
-       to.tv_sec = 1;
-       to.tv_usec = 0;
-       log(4, 0, 0, 0, "Beginning of main loop. udpfd=%d, tapfd=%d, radfd=%d, cluster_sockfd=%d, controlfd=%d\n",
-                       udpfd, tapfd, radfd, cluster_sockfd, controlfd);
-
-       FD_ZERO(&cr);
-       FD_SET(udpfd, &cr);
-       FD_SET(tapfd, &cr);
-       FD_SET(radfd, &cr);
-       FD_SET(controlfd, &cr);
-       FD_SET(clifd, &cr);
-       if (cluster_sockfd) FD_SET(cluster_sockfd, &cr);
-       cn = udpfd;
-       if (cn < radfd) cn = radfd;
-       if (cn < tapfd) cn = tapfd;
-       if (cn < controlfd) cn = controlfd;
-       if (cn < clifd) cn = clifd;
-       if (cn < cluster_sockfd) cn = cluster_sockfd;
-
-       while (!main_quit)
-       {
-               fd_set r;
-               int n = cn;
+       static sessionidt s = 0;        // Next session to check for actions on.
+       tunnelidt t;
+       int count=0,i;
+       uint16_t r;
+       static clockt next_acct = 0;
+       static clockt next_shut_acct = 0;
+       int a;
 
-               if (config->reload_config)
+       LOG(3, 0, 0, "Begin regular cleanup\n");
+
+       for (r = 1; r < MAXRADIUS; r++)
+       {
+               if (!radius[r].state)
+                       continue;
+               if (radius[r].retry)
                {
-                       // Update the config state based on config settings
-                       update_config();
+                       if (radius[r].retry <= TIME)
+                               radiusretry(r);
+               } else
+                       radius[r].retry = backoff(radius[r].try+1);     // Is this really needed? --mo
+       }
+       for (t = 1; t <= config->cluster_highest_tunnelid; t++)
+       {
+               // check for expired tunnels
+               if (tunnel[t].die && tunnel[t].die <= TIME)
+               {
+                       STAT(tunnel_timeout);
+                       tunnelkill(t, "Expired");
+                       continue;
                }
-
-               memcpy(&r, &cr, sizeof(fd_set));
-               n = select(n + 1, &r, 0, 0, &to);
-               if (n < 0)
+               // check for message resend
+               if (tunnel[t].retry && tunnel[t].controlc)
                {
-                       if (errno != EINTR)
+                       // 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;
+                                       }
                        }
                }
-               else if (n)
+               // Send hello
+               if (tunnel[t].state == TUNNELOPEN && tunnel[t].lastrec < TIME + 600)
                {
-                       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);
-                       else if (FD_ISSET(clifd, &r))
-                       {
-                               struct sockaddr_in addr;
-                               int sockfd;
-                               int len = sizeof(addr);
+                       controlt *c = controlnew(6); // sending HELLO
+                       controladd(c, t, 0); // send the message
+                       LOG(3, 0, t, "Sending HELLO message\n");
+               }
 
-                               if ((sockfd = accept(clifd, (struct sockaddr *)&addr, &len)) <= 0)
-                               {
-                                       log(0, 0, 0, 0, "accept error: %s\n", strerror(errno));
-                                       continue;
-                               }
-                               else
-                               {
-                                       cli_do(sockfd);
-                                       close(sockfd);
-                               }
-                       }
-                       else
+               // 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(1, 0, 0, 0, "Main select() loop returned %d, but no fds have data waiting\n", n);
-                               continue;
+                               LOG(2, 0, t, "Dropping tunnel by CLI\n");
+                               tunnelshutdown(t, "Requested by administrator");
                        }
                }
-               else if (n == 0) {                       // handle timeouts
-                       clockt when = now();
-                       clockt best = when + 100; // default timeout
-                       sessionidt s;
-                       tunnelidt t;
-                       u8 r;
 
-                       for (r = 1; r < MAXRADIUS; r++)
-                               if (radius[r].state && radius[r].retry)
-                               {
-                                       if (radius[r].retry <= when)
-                                               radiusretry(r);
-                                       if (radius[r].retry && radius[r].retry < best)
-                                               best = radius[r].retry;
-                               }
-                       for (t = 1; t < MAXTUNNEL; t++)
+       }
+
+       count = 0;
+       for (i = 1; i <= config->cluster_highest_sessionid; i++)
+       {
+               s++;
+               if (s > config->cluster_highest_sessionid)
+                       s = 1;
+
+               if (!session[s].opened) // Session isn't in use
+                       continue;
+
+               if (!session[s].die && session[s].ip && !(session[s].flags & SF_IPCP_ACKED))
+               {
+                       // IPCP has not completed yet. Resend
+                       LOG(3, s, session[s].tunnel, "No ACK for initial IPCP ConfigReq... resending\n");
+                       sendipcp(session[s].tunnel, s);
+               }
+
+               // check for expired sessions
+               if (session[s].die && session[s].die <= TIME)
+               {
+                       sessionkill(s, "Expired");
+                       if (++count >= MAX_ACTIONS) break;
+                       continue;
+               }
+
+               // Drop sessions who have not responded within IDLE_TIMEOUT seconds
+               if (session[s].last_packet && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
+               {
+                       sessionshutdown(s, "No response to LCP ECHO requests");
+                       STAT(session_timeout);
+                       if (++count >= MAX_ACTIONS) break;
+                       continue;
+               }
+
+               // No data in IDLE_TIMEOUT seconds, send LCP ECHO
+               if (session[s].user[0] && (time_now - session[s].last_packet >= ECHO_TIMEOUT))
+               {
+                       uint8_t b[MAXCONTROL] = {0};
+
+                       uint8_t *q = makeppp(b, sizeof(b), 0, 0, session[s].tunnel, s, PPPLCP);
+                       if (!q) continue;
+
+                       *q = EchoReq;
+                       *(uint8_t *)(q + 1) = (time_now % 255); // ID
+                       *(uint16_t *)(q + 2) = htons(8); // Length
+                       *(uint32_t *)(q + 4) = 0; // Magic Number (not supported)
+
+                       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
+                       if (++count >= MAX_ACTIONS) break;
+               }
+
+               // Check for actions requested from the CLI
+               if ((a = cli_session_actions[s].action))
+               {
+                       int send = 0;
+
+                       cli_session_actions[s].action = 0;
+                       if (a & CLI_SESS_KILL)
                        {
-                               // check for expired tunnels
-                               if (tunnel[t].die && tunnel[t].die <= when)
-                               {
-                                       STAT(tunnel_timeout);
-                                       tunnelkill(t, "Expired");
-                                       continue;
-                               }
-                               // check for message resend
-                               if (tunnel[t].retry && tunnel[t].controlc)
-                               {
-                                       // resend pending messages as timeout on reply
-                                       if (tunnel[t].retry <= when)
-                                       {
-                                               controlt *c = tunnel[t].controls;
-                                               u8 w = tunnel[t].window;
-                                               tunnel[t].try++; // another try
-                                               if (tunnel[t].try > 5)
-                                                       tunnelkill(t, "Timeout on control message"); // game over
-                                               else
-                                                       while (c && w--)
-                                                       {
-                                                               tunnelsend(c->buf, c->length, t);
-                                                               c = c->next;
-                                                       }
-                                       }
-                                       if (tunnel[t].retry && tunnel[t].retry < best)
-                                               best = tunnel[t].retry;
-                               }
-                               // Send hello
-                               if (tunnel[t].state == TUNNELOPEN && tunnel[t].lastrec < when + 600)
-                               {
-                                       controlt *c = controlnew(6); // sending HELLO
-                                       controladd(c, t, 0); // send the message
-                                       log(3, tunnel[t].ip, 0, t, "Sending HELLO message\n");
-                               }
+                               LOG(2, s, session[s].tunnel, "Dropping session by CLI\n");
+                               sessionshutdown(s, "Requested by administrator");
+                               a = 0; // dead, no need to check for other actions
                        }
 
-                       // Check for sessions that have been killed from the CLI
-                       if (cli_session_kill[0])
+                       if (a & CLI_SESS_NOSNOOP)
                        {
-                               int i;
-                               for (i = 1; i < MAXSESSION && cli_session_kill[i]; i++)
-                               {
-                                       log(2, 0, cli_session_kill[i], 0, "Dropping session by CLI\n");
-                                       sessionshutdown(cli_session_kill[i], "Requested by administrator");
-                                       cli_session_kill[i] = 0;
-                               }
+                               LOG(2, s, session[s].tunnel, "Unsnooping session by CLI\n");
+                               session[s].snoop_ip = 0;
+                               session[s].snoop_port = 0;
+                               send++;
                        }
-                       // Check for tunnels that have been killed from the CLI
-                       if (cli_tunnel_kill[0])
+                       else if (a & CLI_SESS_SNOOP)
                        {
-                               int i;
-                               for (i = 1; i < MAXTUNNEL && cli_tunnel_kill[i]; i++)
-                               {
-                                       log(2, 0, cli_tunnel_kill[i], 0, "Dropping tunnel by CLI\n");
-                                       tunnelshutdown(cli_tunnel_kill[i], "Requested by administrator");
-                                       cli_tunnel_kill[i] = 0;
-                               }
+                               LOG(2, s, session[s].tunnel, "Snooping session by CLI (to %s:%d)\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;
+                               send++;
                        }
 
-                       for (s = 1; s < MAXSESSION; s++)
+                       if (a & CLI_SESS_NOTHROTTLE)
                        {
-                               // check for expired sessions
-                               if (session[s].die && session[s].die <= when)
-                               {
-                                       sessionkill(s, "Expired");
-                                       continue;
-                               }
-
-                               // Drop sessions who have not responded within IDLE_TIMEOUT seconds
-                               if (session[s].last_packet && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
-                               {
-                                       sessionkill(s, "No response to LCP ECHO requests");
-                                       STAT(session_timeout);
-                                       continue;
-                               }
-
-                               // 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);
-
-                                       *q = EchoReq;
-                                       *(u8 *)(q + 1) = (time_now % 255); // ID
-                                       *(u16 *)(q + 2) = htons(8); // Length
-                                       *(u32 *)(q + 4) = 0; // Magic Number (not supported)
-
-                                       log(4, session[s].ip, s, session[s].tunnel, "No data in %d seconds, sending LCP ECHO\n",
-                                                       time_now - session[s].last_packet);
-                                       tunnelsend(b, 24, session[s].tunnel); // send it
-                                       continue;
-                               }
+                               LOG(2, s, session[s].tunnel, "Un-throttling session by CLI\n");
+                               throttle_session(s, 0, 0);
+                               send++;
                        }
-                       if (config->accounting_dir && next_acct <= when)
+                       else if (a & CLI_SESS_THROTTLE)
                        {
-                               // Dump accounting data
-                               next_acct = when + ACCT_TIME;
-                               dump_acct_info();
+                               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);
+                               send++;
                        }
 
-                       if (cluster_sockfd && next_cluster_ping <= when)
+                       if (a & CLI_SESS_NOFILTER)
+                       {
+                               LOG(2, s, session[s].tunnel, "Un-filtering session by CLI\n");
+                               filter_session(s, 0, 0);
+                               send++;
+                       }
+                       else if (a & CLI_SESS_FILTER)
                        {
-                               // Dump accounting data
-                               next_cluster_ping = when + 50;
-                               cluster_send_message(config->cluster_address, config->bind_address, C_PING, hostname, strlen(hostname));
+                               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);
+                               send++;
                        }
 
-                       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));
+                       if (send)
+                               cluster_send_session(s);
+
+                       if (++count >= MAX_ACTIONS) break;
+               }
+       }
+
+       if (*config->accounting_dir)
+       {
+               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);
                }
        }
+
+       if (count >= MAX_ACTIONS)
+               return 1;       // Didn't finish!
+
+       LOG(3, 0, 0, "End regular cleanup (%d actions), next in %d seconds\n", count, config->cleanup_interval);
+       return 0;
 }
 
-// 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 time_t stopped_bgp = 0;
+       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 == MAP_FAILED)
+       if (!config->cluster_iam_master)
        {
-               log(0, 0, 0, 0, "Error doing mmap for _statistics: %s\n", strerror(errno));
-               exit(1);
+#ifdef BGP
+               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;
        }
-       config = mmap(NULL, sizeof(struct configt), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (config == MAP_FAILED)
+
+       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 configuration: %s\n", strerror(errno));
-               exit(1);
+               if (!tunnel[i].controlc)
+                       continue;
+
+               if (last_talked != TIME)
+               {
+                       LOG(2, 0, 0, "Tunnel %d still has un-acked control messages.\n", i);
+                       last_talked = TIME;
+               }
+               return 1;
        }
-       memset(config, 0, sizeof(struct configt));
-       time(&config->start_time);
-       strncpy(config->config_file, CONFIGFILE, sizeof(config->config_file) - 1);
-       tunnel = mmap(NULL, sizeof(tunnelt) * MAXTUNNEL, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (tunnel == MAP_FAILED)
+
+       // 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 tunnels: %s\n", strerror(errno));
-               exit(1);
+               LOG(1, 0, 0, "Giving up waiting for RADIUS to be empty.  Shutting down anyway.\n");
+               return 0;
+       }
+
+       for (i = 1; i < MAXRADIUS; i++)
+       {
+               if (radius[i].state == RADIUSNULL)
+                       continue;
+               if (radius[i].state == RADIUSWAIT)
+                       continue;
+
+               if (last_talked != TIME)
+               {
+                       LOG(2, 0, 0, "Radius session %d is still busy (sid %d)\n", i, radius[i].session);
+                       last_talked = TIME;
+               }
+               return 1;
+       }
+
+       return 0;
+}
+
+static fd_set readset;
+static int readset_n = 0;
+
+// main loop - gets packets on tun or udp and processes them
+static void mainloop(void)
+{
+       int i;
+       uint8_t buf[65536];
+       struct timeval to;
+       clockt next_cluster_ping = 0;   // send initial ping immediately
+       time_t next_clean = time_now + config->cleanup_interval;
+
+       LOG(4, 0, 0, "Beginning of main loop.  udpfd=%d, tunfd=%d, cluster_sockfd=%d, controlfd=%d\n",
+               udpfd, tunfd, cluster_sockfd, controlfd);
+
+       FD_ZERO(&readset);
+       FD_SET(udpfd, &readset);
+       FD_SET(tunfd, &readset);
+       FD_SET(controlfd, &readset);
+       FD_SET(clifd, &readset);
+       if (cluster_sockfd) FD_SET(cluster_sockfd, &readset);
+       readset_n = udpfd;
+       if (tunfd > readset_n)          readset_n = tunfd;
+       if (controlfd > readset_n)      readset_n = controlfd;
+       if (clifd > readset_n)          readset_n = clifd;
+       if (cluster_sockfd > readset_n) readset_n = cluster_sockfd;
+
+       while (!main_quit || still_busy())
+       {
+               fd_set r;
+               int n = readset_n;
+#ifdef BGP
+               fd_set w;
+               int bgp_set[BGP_NUM_PEERS];
+#endif /* BGP */
+
+               if (config->reload_config)
+               {
+                       // Update the config state based on config settings
+                       update_config();
+               }
+
+               memcpy(&r, &readset, sizeof(fd_set));
+               to.tv_sec = 0;
+               to.tv_usec = 100000; // 1/10th of a second.
+
+#ifdef BGP
+               FD_ZERO(&w);
+               for (i = 0; i < BGP_NUM_PEERS; i++)
+               {
+                       bgp_set[i] = bgp_select_state(&bgp_peers[i]);
+                       if (bgp_set[i] & 1)
+                       {
+                               FD_SET(bgp_peers[i].sock, &r);
+                               if (bgp_peers[i].sock > n)
+                                       n = bgp_peers[i].sock;
+                       }
+
+                       if (bgp_set[i] & 2)
+                       {
+                               FD_SET(bgp_peers[i].sock, &w);
+                               if (bgp_peers[i].sock > n)
+                                       n = bgp_peers[i].sock;
+                       }
+               }
+
+               n = select(n + 1, &r, &w, 0, &to);
+#else /* BGP */
+               n = select(n + 1, &r, 0, 0, &to);
+#endif /* BGP */
+
+               STAT(select_called);
+
+               TIME = now();
+               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));
+                       main_quit++;
+                       break;
+               }
+               else if (n)
+               {
+                       struct sockaddr_in addr;
+                       int alen, c, s;
+                       int udp_pkts = 0;
+                       int tun_pkts = 0;
+                       int cluster_pkts = 0;
+
+                       // nsctl commands
+                       if (FD_ISSET(controlfd, &r))
+                       {
+                               alen = sizeof(addr);
+                               processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr, alen);
+                               n--;
+                       }
+
+                       // RADIUS responses
+                       if (config->cluster_iam_master)
+                       {
+                               for (i = 0; i < config->num_radfds; i++)
+                               {
+                                       if (FD_ISSET(radfds[i], &r))
+                                       {
+                                               processrad(buf, recv(radfds[i], buf, sizeof(buf), 0), i);
+                                               n--;
+                                       }
+                               }
+                       }
+
+                       // CLI connections
+                       if (FD_ISSET(clifd, &r))
+                       {
+                               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--;
+                       }
+
+#ifdef BGP
+                       for (i = 0; i < BGP_NUM_PEERS; i++)
+                       {
+                               int isr = bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &r) : 0;
+                               int isw = bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &w) : 0;
+                               bgp_process(&bgp_peers[i], isr, isw);
+                               if (isr) n--;
+                               if (isw) n--;
+                       }
+#endif /* BGP */
+
+                       for (c = 0; n && c < config->multi_read_count; c++)
+                       {
+                               // L2TP
+                               if (FD_ISSET(udpfd, &r))
+                               {
+                                       alen = sizeof(addr);
+                                       if ((s = recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen)) > 0)
+                                       {
+                                               processudp(buf, s, &addr);
+                                               udp_pkts++;
+                                       }
+                                       else
+                                       {
+                                               FD_CLR(udpfd, &r);
+                                               n--;
+                                       }
+                               }
+
+                               // incoming IP
+                               if (FD_ISSET(tunfd, &r))
+                               {
+                                       if ((s = read(tunfd, buf, sizeof(buf))) > 0)
+                                       {
+                                               processtun(buf, s);
+                                               tun_pkts++;
+                                       }
+                                       else
+                                       {
+                                               FD_CLR(tunfd, &r);
+                                               n--;
+                                       }
+                               }
+
+                               // cluster
+                               if (FD_ISSET(cluster_sockfd, &r))
+                               {
+                                       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
+                                       {
+                                               FD_CLR(cluster_sockfd, &r);
+                                               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);
+                       }
+               }
+
+                       // Runs on every machine (master and slaves).
+               if (cluster_sockfd && next_cluster_ping <= TIME)
+               {
+                       // Check to see which of the cluster is still alive..
+
+                       cluster_send_ping(basetime);    // Only does anything if we're a slave
+                       cluster_check_master();         // ditto.
+
+                       cluster_heartbeat();            // Only does anything if we're a master.
+                       cluster_check_slaves();         // ditto.
+
+                       master_update_counts();         // If we're a slave, send our byte counters to our master.
+
+                       if (config->cluster_iam_master && !config->cluster_iam_uptodate)
+                               next_cluster_ping = TIME + 1; // out-of-date slaves, do fast updates
+                       else
+                               next_cluster_ping = TIME + config->cluster_hb_interval;
+               }
+
+                       // Run token bucket filtering queue..
+                       // Only run it every 1/10th of a second.
+                       // Runs on all machines both master and slave.
+               {
+                       static clockt last_run = 0;
+                       if (last_run != TIME)
+                       {
+                               last_run = TIME;
+                               tbf_run_timer();
+                       }
+               }
+
+               /* Handle timeouts. Make sure that this gets run anyway, even if there was
+                * something to read, else under load this will never actually run....
+                *
+                */
+               if (config->cluster_iam_master && next_clean <= time_now)
+               {
+                       if (regular_cleanups())
+                       {
+                               // Did it finish?
+                               next_clean = time_now + 1 ;     // Didn't finish. Check quickly.
+                       }
+                       else
+                       {
+                               next_clean = time_now + config->cleanup_interval; // Did. Move to next interval.
+                       }
+               }
+       }
+
+               // 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, "Clean 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
+               }
        }
-       session = mmap(NULL, sizeof(sessiont) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (session == MAP_FAILED)
+}
+
+// Init data structures
+static void initdata(int optdebug, char *optconfig)
+{
+       int i;
+
+       if (!(_statistics = shared_malloc(sizeof(struct Tstats))))
        {
-               log(0, 0, 0, 0, "Error doing mmap for sessions: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for _statistics: %s\n", strerror(errno));
                exit(1);
        }
-       radius = mmap(NULL, sizeof(radiust) * MAXRADIUS, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (radius == MAP_FAILED)
+       if (!(config = shared_malloc(sizeof(configt))))
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for configuration: %s\n", strerror(errno));
                exit(1);
        }
-       ip_address_pool = mmap(NULL, sizeof(ippoolt) * MAXIPPOOL, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (ip_address_pool == MAP_FAILED)
+       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_master_min_adv = 1;
+
+       if (!(tunnel = shared_malloc(sizeof(tunnelt) * MAXTUNNEL)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for tunnels: %s\n", strerror(errno));
                exit(1);
        }
-#ifdef RINGBUFFER
-       ringbuffer = mmap(NULL, sizeof(struct Tringbuffer), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (ringbuffer == MAP_FAILED)
+       if (!(session = shared_malloc(sizeof(sessiont) * MAXSESSION)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for radius: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for sessions: %s\n", strerror(errno));
                exit(1);
        }
-       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
-#endif
 
-       cli_session_kill = mmap(NULL, sizeof(sessionidt) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (cli_session_kill == MAP_FAILED)
+       if (!(sess_local = shared_malloc(sizeof(sessionlocalt) * MAXSESSION)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for cli session kill: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for sess_local: %s\n", strerror(errno));
                exit(1);
        }
-       memset(cli_session_kill, 0, sizeof(sessionidt) * MAXSESSION);
-       cli_tunnel_kill = mmap(NULL, sizeof(tunnelidt) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (cli_tunnel_kill == MAP_FAILED)
+
+       if (!(radius = shared_malloc(sizeof(radiust) * MAXRADIUS)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for cli tunnel kill: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for radius: %s\n", strerror(errno));
                exit(1);
        }
-       memset(cli_tunnel_kill, 0, sizeof(tunnelidt) * MAXSESSION);
 
-       filter_buckets = mmap(NULL, sizeof(tbft) * MAXSESSION, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, 0, 0);
-       if (filter_buckets == MAP_FAILED)
+       if (!(ip_address_pool = shared_malloc(sizeof(ippoolt) * MAXIPPOOL)))
        {
-               log(0, 0, 0, 0, "Error doing mmap for filter buckets: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error doing malloc for ip_address_pool: %s\n", strerror(errno));
                exit(1);
        }
-       memset(filter_buckets, 0, sizeof(tbft) * MAXSESSION);
+
+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);
+
+#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 (!(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(session, 0, sizeof(sessiont) * MAXSESSION);
        memset(radius, 0, sizeof(radiust) * MAXRADIUS);
        memset(ip_address_pool, 0, sizeof(ippoolt) * MAXIPPOOL);
-       for (i = 1; i < MAXSESSION - 1; i++)
+
+               // Put all the sessions on the free list marked as undefined.
+       for (i = 1; i < MAXSESSION; 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.
+
        if (!*hostname)
        {
-               char *p;
                // Grab my hostname unless it's been specified
                gethostname(hostname, sizeof(hostname));
-               {
-                       struct hostent *h = gethostbyname(hostname);
-                       if (h)
-                               myip = ntohl(*(u32 *) h->h_addr);
-               }
-
-               if ((p = strstr(hostname, ".optusnet.com.au"))) *p = 0;
+               stripdomain(hostname);
        }
+
        _statistics->start_time = _statistics->last_reset = time(NULL);
-}
 
-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");
+#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 */
 }
 
-int assign_ip_address(sessionidt s)
+static int assign_ip_address(sessionidt s)
 {
-       unsigned i;
+       uint32_t i;
        int best = -1;
-       clockt best_time = time_now;
+       time_t best_time = time_now;
        char *u = session[s].user;
        char reuse = 0;
 
-#ifdef STAT_CALLS
-       STAT(call_assign_ip_address);
-#endif
-       for (i = 0; i < ip_pool_size; i++)
+
+       CSTAT(call_assign_ip_address);
+
+       for (i = 1; i < ip_pool_size; i++)
        {
                if (!ip_address_pool[i].address || ip_address_pool[i].assigned)
                        continue;
@@ -1987,14 +2851,15 @@ int assign_ip_address(sessionidt s)
 
        if (best < 0)
        {
-               log(0, 0, s, session[s].tunnel, "assign_ip_address(): out of addresses\n");
+               LOG(0, s, session[s].tunnel, "assign_ip_address(): out of addresses\n");
                return 0;
        }
 
-       session[s].ip = ntohl(ip_address_pool[best].address);
+       session[s].ip = ip_address_pool[best].address;
        session[s].ip_pool_index = best;
        ip_address_pool[best].assigned = 1;
        ip_address_pool[best].last = time_now;
+       ip_address_pool[best].session = s;
        if (session[s].walled_garden)
                /* Don't track addresses of users in walled garden (note: this
                   means that their address isn't "sticky" even if they get
@@ -2004,61 +2869,176 @@ int assign_ip_address(sessionidt s)
                strncpy(ip_address_pool[best].user, u, sizeof(ip_address_pool[best].user) - 1);
 
        STAT(ip_allocated);
-       log(4, ip_address_pool[best].address, s, session[s].tunnel,
-               "assign_ip_address(): %s ip address %lu from pool\n", reuse ? "Reusing" : "Allocating", best);
+       LOG(4, s, session[s].tunnel, "assign_ip_address(): %s ip address %d from pool\n",
+               reuse ? "Reusing" : "Allocating", best);
 
        return 1;
 }
 
-void free_ip_address(sessionidt s)
+static void free_ip_address(sessionidt s)
 {
        int i = session[s].ip_pool_index;
 
-#ifdef STAT_CALLS
-       STAT(call_free_ip_address);
-#endif
+
+       CSTAT(call_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);
-       uncache_sessionid(session[s].ip);
+       cache_ipmap(session[s].ip, -i); // Change the mapping to point back to the ip pool index.
        session[s].ip = 0;
        ip_address_pool[i].assigned = 0;
+       ip_address_pool[i].session = 0;
        ip_address_pool[i].last = time_now;
 }
 
+//
+// 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 %d 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 %d has a pool IP that's not found in the pool! (%d)\n", i, ipid);
+                       ipid = -1;
+                       session[i].ip_pool_index = ipid;
+                       continue;
+               }
+
+               ip_address_pool[ipid].assigned = 1;
+               ip_address_pool[ipid].session = i;
+               ip_address_pool[ipid].last = time_now;
+               strncpy(ip_address_pool[ipid].user, session[i].user, sizeof(ip_address_pool[ipid].user) - 1);
+               session[i].ip_pool_index = ipid;
+               cache_ipmap(session[i].ip, i);  // Fix the ip map.
+       }
+}
+
+//
+// Fix the address pool to match a changed session.
+// (usually when the master sends us an update).
+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
-void initippool()
+static void initippool()
 {
        FILE *f;
-       char *buf, *p;
-       int pi = 0;
+       char *p;
+       char buf[4096];
        memset(ip_address_pool, 0, sizeof(ip_address_pool));
 
        if (!(f = fopen(IPPOOLFILE, "r")))
        {
-               log(0, 0, 0, 0, "Can't load pool file " IPPOOLFILE ": %s\n", strerror(errno));
-               exit(-1);
+               LOG(0, 0, 0, "Can't load pool file " IPPOOLFILE ": %s\n", strerror(errno));
+               exit(1);
        }
 
-       buf = (char *)malloc(4096);
-
-       while (pi < MAXIPPOOL && fgets(buf, 4096, f))
+       while (ip_pool_size < MAXIPPOOL && fgets(buf, 4096, f))
        {
-               char* pool = buf;
+               char *pool = buf;
+               buf[4095] = 0;  // Force it to be zero terminated/
+
                if (*buf == '#' || *buf == '\n')
                        continue; // Skip comments / blank lines
                if ((p = (char *)strrchr(buf, '\n'))) *p = 0;
                if ((p = (char *)strchr(buf, ':')))
                {
-                       ipt src;
+                       in_addr_t src;
                        *p = '\0';
                        src = inet_addr(buf);
                        if (src == INADDR_NONE)
                        {
-                               log(0, 0, 0, 0, "Invalid address pool IP %s", buf);
-                               exit(-1);
+                               LOG(0, 0, 0, "Invalid address pool IP %s\n", buf);
+                               exit(1);
                        }
                        // This entry is for a specific IP only
                        if (src != config->bind_address)
@@ -2070,209 +3050,260 @@ void initippool()
                {
                        // It's a range
                        int numbits = 0;
-                       unsigned long start = 0, end = 0, mask = 0, ip;
-                       struct rtentry r;
+                       in_addr_t start = 0, mask = 0;
 
-                       log(2, 0, 0, 0, "Adding IP address range %s\n", buf);
+                       LOG(2, 0, 0, "Adding IP address range %s\n", buf);
                        *p++ = 0;
                        if (!*p || !(numbits = atoi(p)))
                        {
-                               log(0, 0, 0, 0, "Invalid pool range %s/\n", buf, p);
+                               LOG(0, 0, 0, "Invalid pool range %s\n", buf);
                                continue;
                        }
-                       start = end = ntohl(inet_addr(pool));
-                       mask = (unsigned long)(pow(2, numbits) - 1) << (32 - numbits);
-                       start &= mask;
-                       end = start + (int)(pow(2, (32 - numbits))) - 1;
-                       for (ip = (start + 1); ip < end && pi < MAXIPPOOL; ip++)
-                       {
-                               if ((ip & 0xFF) == 0 || (ip & 0xFF) == 255)
-                                       continue;
-                               ip_address_pool[pi++].address = htonl(ip);
-                       }
+                       start = ntohl(inet_addr(pool));
+                       mask = (in_addr_t) (pow(2, numbits) - 1) << (32 - numbits);
 
                        // Add a static route for this pool
-                       log(5, 0, 0, 0, "Adding route for address pool %s/%d\n", inet_toa(htonl(start)), 32+mask);
-                       memset(&r, 0, sizeof(r));
-                       r.rt_dev = config->tapdevice;
-                       r.rt_dst.sa_family = AF_INET;
-                       *(u32 *) & (((struct sockaddr_in *) &r.rt_dst)->sin_addr.s_addr) = htonl(start);
-                       r.rt_genmask.sa_family = AF_INET;
-                       *(u32 *) & (((struct sockaddr_in *) &r.rt_genmask)->sin_addr.s_addr) = htonl(mask);
-                       r.rt_flags = (RTF_UP | RTF_STATIC);
-                       if (ioctl(ifrfd, SIOCADDRT, (void *) &r) < 0)
-                       {
-                               log(0, 0, 0, 0, "Error adding ip address pool route %s/%d: %s\n", inet_toa(start), mask, strerror(errno));
-                       }
+                       LOG(5, 0, 0, "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
-                       ip_address_pool[pi++].address = inet_addr(pool);
+                       add_to_ip_pool(inet_addr(pool), 0);
                }
        }
-
-       free(buf);
        fclose(f);
-       log(1, 0, 0, 0, "IP address pool is %d addresses\n", pi);
-       ip_pool_size = pi;
+       LOG(1, 0, 0, "IP address pool is %d addresses\n", ip_pool_size - 1);
 }
 
-void snoop_send_packet(char *packet, u16 size)
+void snoop_send_packet(char *packet, uint16_t size, in_addr_t destination, uint16_t port)
 {
-       if (!snoop_addr.sin_port || snoopfd <= 0 || size <= 0 || !packet)
+       struct sockaddr_in snoop_addr = {0};
+       if (!destination || !port || snoopfd <= 0 || size <= 0 || !packet)
                return;
 
-       log(5, 0, 0, 0, "Snooping packet at %p (%d bytes) to %s:%d\n", packet, size, inet_toa(snoop_addr.sin_addr.s_addr), htons(snoop_addr.sin_port));
+       snoop_addr.sin_family = AF_INET;
+       snoop_addr.sin_addr.s_addr = destination;
+       snoop_addr.sin_port = ntohs(port);
+
+       LOG(5, 0, 0, "Snooping %d byte packet to %s:%d\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, 0, "Error sending intercept packet: %s\n", strerror(errno));
+               LOG(0, 0, 0, "Error sending intercept packet: %s\n", strerror(errno));
+
        STAT(packets_snooped);
 }
 
-void dump_acct_info()
+static int dump_session(FILE **f, sessiont *s)
 {
-    char filename[1024];
-    char timestr[64];
-    time_t t = time(NULL);
-    int i;
-    FILE *f = NULL;
+       if (!s->opened || !s->ip || !(s->cin || s->cout) || !*s->user || s->walled_garden)
+               return 1;
 
-#ifdef STAT_CALLS
-       STAT(call_dump_acct_info);
-#endif
-    strftime(timestr, 64, "%Y%m%d%H%M%S", localtime(&t));
-    snprintf(filename, 1024, "%s/%s", config->accounting_dir, timestr);
+       if (!*f)
+       {
+               char filename[1024];
+               char timestr[64];
+               time_t now = time(NULL);
 
-    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)
+               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"
+                       "# time: %ld\n"
+                       "# uptime: %ld\n"
+                       "# format: username ip qos uptxoctets downrxoctets\n",
+                       hostname,
+                       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,                                      // uptxoctets
+               (uint32_t) s->cout);                                    // downrxoctets
+
+       s->pin = s->cin = 0;
+       s->pout = s->cout = 0;
+
+       return 1;
+}
+
+static void dump_acct_info(int all)
+{
+       int i;
+       FILE *f = NULL;
+
+
+       CSTAT(call_dump_acct_info);
+
+       if (shut_acct_n)
        {
-           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);
+               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 o;
+// 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));
+       signal(SIGALRM, sigalrm_handler);
+       siginterrupt(SIGALRM, 0);
+
+       initplugins();
+       initdata(optdebug, optconfig);
 
-       _program_name = strdup(argv[0]);
+       init_cli(hostname);
+       read_config_file();
+       init_tbf(config->num_tbfs);
 
+       LOG(0, 0, 0, "L2TPNS version " VERSION "\n");
+       LOG(0, 0, 0, "Copyright (c) 2003, 2004, 2005 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
-               setrlimit(RLIMIT_CORE, &rlim);
+               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");
        }
 
-       time(&basetime);             // start clock
-       // scan args
-
-       while ((o = getopt(argc, argv, "vc:h:a:")) >= 0)
+       if (config->scheduler_fifo)
        {
-               switch (o)
+               int ret;
+               struct sched_param params = {0};
+               params.sched_priority = 1;
+
+               if (get_nprocs() < 2)
                {
-                       case 'v':
-                               config->debug++;
-                               break;
-                       case 'c':
-                               strncpy(config->config_file, optarg, sizeof(config->config_file) - 1);
-                               break;
-                       case 'h':
-                               strncpy(hostname, optarg, 999);
-                               break;
-                       case 'a':
-                               myip = inet_addr(optarg);
-                               if (myip == INADDR_NONE) {
-                                       log(0, 0, 0, 0, "Invalid ip %s\n", optarg);
-                                       exit(-1);
-                               }
-                               config->bind_address = myip;
-                               handle_interface = 1;
-                               break;
-                       case '?':
-                       default:
-                               printf("Args are:\n\t-c <file>\tConfig file\n\t-h <hostname>\tForce hostname\n\t-a <address>\tUse specific address\n\t-v\t\tDebug\n");
-                               return (0);
-                               break;
+                       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;
+                       }
                }
        }
 
-       // Start the timer routine off
-       time(&time_now);
-       strftime(time_now_string, 64, "%Y-%m-%d %H:%M:%S", localtime(&time_now));
-
-       initiptables();
-       initplugins();
-       initdata();
-       init_cli();
-       read_config_file();
-       log(0, 0, 0, 0, "$Id: l2tpns.c,v 1.3 2004/03/05 00:22:06 fred_nerk Exp $\n(c) Copyright 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced\n");
+       /* Set up the cluster communications port. */
+       if (cluster_init() < 0)
+               exit(1);
 
-       /* Start up the cluster first, so that we don't have two machines with
-        * the same IP at once.
-        * This is still racy, but the second GARP should fix that
-        */
-       cluster_init(config->bind_address, 0);
-       cluster_send_message(config->cluster_address, config->bind_address, C_HELLO, hostname, strlen(hostname));
+#ifdef BGP
+       signal(SIGPIPE, SIG_IGN);
+       bgp_setup(config->as_number);
+       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 */
 
-       inittap();
-       log(1, 0, 0, 0, "Set up on interface %s\n", config->tapdevice);
+       inittun();
+       LOG(1, 0, 0, "Set up on interface %s\n", config->tundevice);
 
        initudp();
        initrad();
        initippool();
-       init_rl();
-       if (handle_interface) {
-               send_garp(config->bind_address);
-       }
-
-       // If NOSTATEFILE exists, we will ignore any updates from the cluster master for this execution
-       if (!unlink(NOSTATEFILE))
-           config->ignore_cluster_updates = 1;
 
        read_state();
 
-       signal(SIGALRM, sigalrm_handler);
        signal(SIGHUP, sighup_handler);
        signal(SIGTERM, sigterm_handler);
        signal(SIGINT, sigterm_handler);
        signal(SIGQUIT, sigquit_handler);
        signal(SIGCHLD, sigchild_handler);
-       signal(SIGSEGV, sigsegv_handler);
+
+       // Prevent us from getting paged out
+       if (config->lock_pages)
+       {
+               if (!mlockall(MCL_CURRENT))
+                       LOG(1, 0, 0, "Locking pages into memory\n");
+               else
+                       LOG(0, 0, 0, "Can't lock pages: %s\n", strerror(errno));
+       }
 
        alarm(1);
 
@@ -2281,10 +3312,21 @@ int main(int argc, char *argv[])
                setuid(config->target_uid);
 
        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;
 }
 
-void sighup_handler(int junk)
+static void sighup_handler(int sig)
 {
        if (log_stream && log_stream != stderr)
        {
@@ -2295,12 +3337,12 @@ void sighup_handler(int junk)
        read_config_file();
 }
 
-void sigalrm_handler(int junk)
+static void sigalrm_handler(int sig)
 {
        // Log current traffic stats
 
        snprintf(config->bandwidth, sizeof(config->bandwidth),
-               "UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%lu OUT:%lu",
+               "UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%u OUT:%u",
                (udp_rx / 1024.0 / 1024.0 * 8),
                (eth_tx / 1024.0 / 1024.0 * 8),
                (eth_rx / 1024.0 / 1024.0 * 8),
@@ -2317,7 +3359,7 @@ void sigalrm_handler(int junk)
 
        // Update the internal time counter
        time(&time_now);
-       strftime(time_now_string, 64, "%Y-%m-%d %H:%M:%S", localtime(&time_now));
+       strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
        alarm(1);
 
        {
@@ -2328,233 +3370,232 @@ void sigalrm_handler(int junk)
 
 }
 
-void sigterm_handler(int junk)
+static void sigterm_handler(int sig)
 {
-       log(1, 0, 0, 0, "Shutting down cleanly\n");
+       LOG(1, 0, 0, "Shutting down cleanly\n");
        if (config->save_state)
                dump_state();
+
        main_quit++;
 }
 
-void sigquit_handler(int junk)
+static void sigquit_handler(int sig)
 {
-       FILE *f;
        int i;
 
-       log(1, 0, 0, 0, "Shutting down without saving sessions\n");
-       for (i = 1; i < MAXSESSION; i++)
-       {
-               if (session[i].opened)
-                       sessionkill(i, "L2TPNS Closing");
-       }
-       for (i = 1; i < MAXTUNNEL; i++)
+       LOG(1, 0, 0, "Shutting down without saving sessions\n");
+
+       if (config->cluster_iam_master)
        {
-               if (tunnel[i].ip || tunnel[i].state)
-                       tunnelshutdown(i, "L2TPNS Closing");
+               for (i = 1; i < MAXSESSION; i++)
+               {
+                       if (session[i].opened)
+                               sessionkill(i, "L2TPNS Closing");
+               }
+               for (i = 1; i < MAXTUNNEL; i++)
+               {
+                       if (tunnel[i].ip || tunnel[i].state)
+                               tunnelshutdown(i, "L2TPNS Closing");
+               }
        }
 
-       cluster_send_goodbye();
-
-       // Touch a file which says not to reload the state
-       f = fopen(NOSTATEFILE, "w");
-       if (f) fclose(f);
-
        main_quit++;
 }
 
-void sigchild_handler(int signal)
+static void sigchild_handler(int sig)
 {
        while (waitpid(-1, NULL, WNOHANG) > 0)
            ;
 }
 
-void sigsegv_handler(int signal)
+static void read_state()
 {
-       log(0, 0, 0, 0, "----------------------------------------------\n");
-       log(0, 0, 0, 0, "- SEGFAULT!                                  -\n");
-       log(0, 0, 0, 0, "----------------------------------------------\n");
-       _exit(0);
-}
+       struct stat sb;
+       int i;
+       ippoolt itmp;
+       FILE *f;
+       char magic[sizeof(DUMP_MAGIC) - 1];
+       uint32_t buf[2];
 
-void read_state()
-{
-    struct stat sb;
-    int i;
-    ippoolt itmp;
-    FILE *f;
-    char magic[sizeof(DUMP_MAGIC)-1];
-    u32 buf[2];
+       if (!config->save_state)
+       {
+               unlink(STATEFILE);
+               return ;
+       }
 
-    if (!config->save_state)
-       return;
+       if (stat(STATEFILE, &sb) < 0)
+       {
+               unlink(STATEFILE);
+               return ;
+       }
+
+       if (sb.st_mtime < (time(NULL) - 60))
+       {
+               LOG(0, 0, 0, "State file is too old to read, ignoring\n");
+               unlink(STATEFILE);
+               return ;
+       }
 
-    // Ignore saved state if NOSTATEFILE exists
-    if (config->ignore_cluster_updates)
-    {
+       f = fopen(STATEFILE, "r");
        unlink(STATEFILE);
-       return;
-    }
 
-    if (stat(STATEFILE, &sb) < 0)
-       return;
+       if (!f)
+       {
+               LOG(0, 0, 0, "Can't read state file: %s\n", strerror(errno));
+               exit(1);
+       }
 
-    if (sb.st_mtime < (time(NULL) - 60))
-    {
-       log(0, 0, 0, 0, "State file is too old to read, ignoring\n");
-       unlink(STATEFILE);
-       return;
-    }
+       if (fread(magic, sizeof(magic), 1, f) != 1 || strncmp(magic, DUMP_MAGIC, sizeof(magic)))
+       {
+               LOG(0, 0, 0, "Bad state file magic\n");
+               exit(1);
+       }
 
-    f = fopen(STATEFILE, "r");
-    unlink(STATEFILE);
+       LOG(1, 0, 0, "Reading state information\n");
+       if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] > MAXIPPOOL || buf[1] != sizeof(ippoolt))
+       {
+               LOG(0, 0, 0, "Error/mismatch reading ip pool header from state file\n");
+               exit(1);
+       }
 
-    if (!f)
-    {
-       log(0, 0, 0, 0, "Can't read state file: %s\n", strerror(errno));
-       exit(1);
-    }
+       if (buf[0] > ip_pool_size)
+       {
+               LOG(0, 0, 0, "ip pool has shrunk!  state = %d, current = %d\n", buf[0], ip_pool_size);
+               exit(1);
+       }
 
-    if (fread(magic, sizeof(magic), 1, f) != 1 || strncmp(magic, DUMP_MAGIC, sizeof(magic)))
-    {
-       log(0, 0, 0, 0, "Bad state file magic\n");
-       exit(1);
-    }
+       LOG(2, 0, 0, "Loading %u ip addresses\n", buf[0]);
+       for (i = 0; i < buf[0]; i++)
+       {
+               if (fread(&itmp, sizeof(itmp), 1, f) != 1)
+               {
+                       LOG(0, 0, 0, "Error reading ip %d from state file: %s\n", i, strerror(errno));
+                       exit(1);
+               }
 
-    log(1, 0, 0, 0, "Reading state information\n");
-    if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] > MAXIPPOOL || buf[1] != sizeof(ippoolt))
-    {
-       log(0, 0, 0, 0, "Error/mismatch reading ip pool header from state file\n");
-       exit(1);
-    }
+               if (itmp.address != ip_address_pool[i].address)
+               {
+                       LOG(0, 0, 0, "Mismatched ip %d from state file: pool may only be extended\n", i);
+                       exit(1);
+               }
 
-    if (buf[0] > ip_pool_size)
-    {
-       log(0, 0, 0, 0, "ip pool has shrunk!  state = %d, current = %d\n", buf[0], ip_pool_size);
-       exit(1);
-    }
+               memcpy(&ip_address_pool[i], &itmp, sizeof(itmp));
+       }
 
-    log(2, 0, 0, 0, "Loading %u ip addresses\n", buf[0]);
-    for (i = 0; i < buf[0]; i++)
-    {
-       if (fread(&itmp, sizeof(itmp), 1, f) != 1)
+       if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXTUNNEL || buf[1] != sizeof(tunnelt))
        {
-           log(0, 0, 0, 0, "Error reading ip %d from state file: %s\n", i, strerror(errno));
-           exit(1);
+               LOG(0, 0, 0, "Error/mismatch reading tunnel header from state file\n");
+               exit(1);
        }
 
-       if (itmp.address != ip_address_pool[i].address)
+       LOG(2, 0, 0, "Loading %u tunnels\n", MAXTUNNEL);
+       if (fread(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL)
        {
-           log(0, 0, 0, 0, "Mismatched ip %d from state file: pool may only be extended\n", i);
-           exit(1);
+               LOG(0, 0, 0, "Error reading tunnel data from state file\n");
+               exit(1);
        }
 
-       memcpy(&ip_address_pool[i], &itmp, sizeof(itmp));
-    }
-
-    if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXTUNNEL || buf[1] != sizeof(tunnelt))
-    {
-       log(0, 0, 0, 0, "Error/mismatch reading tunnel header from state file\n");
-       exit(1);
-    }
+       for (i = 0; i < MAXTUNNEL; i++)
+       {
+               tunnel[i].controlc = 0;
+               tunnel[i].controls = NULL;
+               tunnel[i].controle = NULL;
+               if (*tunnel[i].hostname)
+                       LOG(3, 0, 0, "Created tunnel for %s\n", tunnel[i].hostname);
+       }
 
-    log(2, 0, 0, 0, "Loading %u tunnels\n", MAXTUNNEL);
-    if (fread(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL)
-    {
-       log(0, 0, 0, 0, "Error reading tunnel data from state file\n");
-       exit(1);
-    }
-
-    for (i = 0; i < MAXTUNNEL; i++)
-    {
-       tunnel[i].controlc = 0;
-       tunnel[i].controls = NULL;
-       tunnel[i].controle = NULL;
-       if (*tunnel[i].hostname)
-           log(3, 0, 0, 0, "Created tunnel for %s\n", tunnel[i].hostname);
-    }
-
-    if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXSESSION || buf[1] != sizeof(sessiont))
-    {
-       log(0, 0, 0, 0, "Error/mismatch reading session header from state file\n");
-       exit(1);
-    }
+       if (fread(buf, sizeof(buf), 1, f) != 1 || buf[0] != MAXSESSION || buf[1] != sizeof(sessiont))
+       {
+               LOG(0, 0, 0, "Error/mismatch reading session header from state file\n");
+               exit(1);
+       }
 
-    log(2, 0, 0, 0, "Loading %u sessions\n", MAXSESSION);
-    if (fread(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION)
-    {
-       log(0, 0, 0, 0, "Error reading session data from state file\n");
-       exit(1);
-    }
+       LOG(2, 0, 0, "Loading %u sessions\n", MAXSESSION);
+       if (fread(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION)
+       {
+               LOG(0, 0, 0, "Error reading session data from state file\n");
+               exit(1);
+       }
 
-    for (i = 0; i < MAXSESSION; i++)
-    {
-       session[i].tbf = 0;
-       session[i].throttle = 0;
-       if (session[i].opened)
+       for (i = 0; i < MAXSESSION; i++)
        {
-           log(2, 0, i, 0, "Loaded active session for user %s\n", session[i].user);
-           if (session[i].ip && session[i].ip != 0xFFFFFFFE)
-               sessionsetup(session[i].tunnel, i, 0);
+               session[i].tbf_in = 0;
+               session[i].tbf_out = 0;
+               if (session[i].opened)
+               {
+                       LOG(2, i, 0, "Loaded active session for user %s\n", session[i].user);
+                       if (session[i].ip)
+                               sessionsetup(session[i].tunnel, i);
+               }
        }
-    }
 
-    fclose(f);
-    log(0, 0, 0, 0, "Loaded saved state information\n");
+       fclose(f);
+       LOG(0, 0, 0, "Loaded saved state information\n");
 }
 
-void dump_state()
+static void dump_state()
 {
-    FILE *f;
-    u32 buf[2];
+       FILE *f;
+       uint32_t buf[2];
 
-    if (!config->save_state)
-       return;
+       if (!config->save_state)
+               return;
 
-    do {
-       if (!(f = fopen(STATEFILE, "w")))
-           break;
+       do
+       {
+               if (!(f = fopen(STATEFILE, "w")))
+                       break;
 
-       log(1, 0, 0, 0, "Dumping state information\n");
+               LOG(1, 0, 0, "Dumping state information\n");
 
-       if (fwrite(DUMP_MAGIC, sizeof(DUMP_MAGIC)-1, 1, f) != 1) break;
+               if (fwrite(DUMP_MAGIC, sizeof(DUMP_MAGIC) - 1, 1, f) != 1)
+                       break;
 
-       log(2, 0, 0, 0, "Dumping %u ip addresses\n", ip_pool_size);
-       buf[0] = ip_pool_size;
-       buf[1] = sizeof(ippoolt);
-       if (fwrite(buf, sizeof(buf), 1, f) != 1) break;
-       if (fwrite(ip_address_pool, sizeof(ippoolt), ip_pool_size, f) != ip_pool_size) break;
+               LOG(2, 0, 0, "Dumping %u ip addresses\n", ip_pool_size);
+               buf[0] = ip_pool_size;
+               buf[1] = sizeof(ippoolt);
+               if (fwrite(buf, sizeof(buf), 1, f) != 1)
+                       break;
+               if (fwrite(ip_address_pool, sizeof(ippoolt), ip_pool_size, f) != ip_pool_size)
+                       break;
 
-       log(2, 0, 0, 0, "Dumping %u tunnels\n", MAXTUNNEL);
-       buf[0] = MAXTUNNEL;
-       buf[1] = sizeof(tunnelt);
-       if (fwrite(buf, sizeof(buf), 1, f) != 1) break;
-       if (fwrite(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL) break;
+               LOG(2, 0, 0, "Dumping %u tunnels\n", MAXTUNNEL);
+               buf[0] = MAXTUNNEL;
+               buf[1] = sizeof(tunnelt);
+               if (fwrite(buf, sizeof(buf), 1, f) != 1)
+                       break;
+               if (fwrite(tunnel, sizeof(tunnelt), MAXTUNNEL, f) != MAXTUNNEL)
+                       break;
 
-       log(2, 0, 0, 0, "Dumping %u sessions\n", MAXSESSION);
-       buf[0] = MAXSESSION;
-       buf[1] = sizeof(sessiont);
-       if (fwrite(buf, sizeof(buf), 1, f) != 1) break;
-       if (fwrite(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION) break;
+               LOG(2, 0, 0, "Dumping %u sessions\n", MAXSESSION);
+               buf[0] = MAXSESSION;
+               buf[1] = sizeof(sessiont);
+               if (fwrite(buf, sizeof(buf), 1, f) != 1)
+                       break;
+               if (fwrite(session, sizeof(sessiont), MAXSESSION, f) != MAXSESSION)
+                       break;
 
-       if (fclose(f) == 0) return; // OK
-    } while (0);
+               if (fclose(f) == 0)
+                       return ; // OK
+       }
+       while (0);
 
-    log(0, 0, 0, 0, "Can't write state information: %s\n", strerror(errno));
-    unlink(STATEFILE);
+       LOG(0, 0, 0, "Can't write state information: %s\n", strerror(errno));
+       unlink(STATEFILE);
 }
 
-void build_chap_response(char *challenge, u8 id, u16 challenge_length, char **challenge_response)
+static void build_chap_response(char *challenge, uint8_t id, uint16_t challenge_length, char **challenge_response)
 {
        MD5_CTX ctx;
        *challenge_response = NULL;
 
        if (!*config->l2tpsecret)
        {
-               log(0, 0, 0, 0, "LNS requested CHAP authentication, but no l2tp secret is defined\n");
+               LOG(0, 0, 0, "LNS requested CHAP authentication, but no l2tp secret is defined\n");
                return;
        }
 
-       log(4, 0, 0, 0, "   Building challenge response for CHAP request\n");
+       LOG(4, 0, 0, "   Building challenge response for CHAP request\n");
 
        *challenge_response = (char *)calloc(17, 1);
 
@@ -2569,22 +3610,20 @@ void build_chap_response(char *challenge, u8 id, u16 challenge_length, char **ch
 
 static int facility_value(char *name)
 {
-    int i;
-    for (i = 0; facilitynames[i].c_name; i++)
-    {
-       if (strcmp(facilitynames[i].c_name, name) == 0)
-           return facilitynames[i].c_val;
-    }
-    return 0;
+       int i;
+       for (i = 0; facilitynames[i].c_name; i++)
+       {
+               if (strcmp(facilitynames[i].c_name, name) == 0)
+                       return facilitynames[i].c_val;
+       }
+       return 0;
 }
 
-void update_config()
+static void update_config()
 {
        int i;
-
-       snoop_addr.sin_family = AF_INET;
-       snoop_addr.sin_addr.s_addr = config->snoop_destination_host;
-       snoop_addr.sin_port = htons(config->snoop_destination_port);
+       static int timeout = 0;
+       static int interval = 0;
 
        // Update logging
        closelog();
@@ -2596,9 +3635,18 @@ void update_config()
        }
        if (*config->log_filename)
        {
-               if (strstr(config->log_filename, "file:") == config->log_filename)
+               if (strstr(config->log_filename, "syslog:") == config->log_filename)
+               {
+                       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 + 5), "a")))
+                       if ((log_stream = fopen((char *)(config->log_filename), "a")))
                        {
                                fseek(log_stream, 0, SEEK_END);
                                setbuf(log_stream, NULL);
@@ -2609,15 +3657,6 @@ void update_config()
                                setbuf(log_stream, NULL);
                        }
                }
-               else if (strstr(config->log_filename, "syslog:") == config->log_filename)
-               {
-                       char *p = config->log_filename + 7;
-                       if (*p)
-                       {
-                               openlog("l2tpns", LOG_PID, facility_value(p));
-                               syslog_log = 1;
-                       }
-               }
        }
        else
        {
@@ -2629,18 +3668,33 @@ void update_config()
        // Update radius
        config->numradiusservers = 0;
        for (i = 0; i < MAXRADSERVER; i++)
-               if (config->radiusserver[i]) config->numradiusservers++;
+               if (config->radiusserver[i])
+               {
+                       config->numradiusservers++;
+                       // Set radius port: if not set, take the port from the
+                       // first radius server.  For the first radius server,
+                       // take the #defined default value from l2tpns.h
+
+                       // test twice, In case someone works with
+                       // a secondary radius server without defining
+                       // a primary one, this will work even then.
+                       if (i>0 && !config->radiusport[i])
+                               config->radiusport[i] = config->radiusport[i-1];
+                       if (!config->radiusport[i])
+                               config->radiusport[i] = RADPORT;
+               }
 
        if (!config->numradiusservers)
-       {
-               log(0, 0, 0, 0, "No RADIUS servers defined!\n");
-       }
+               LOG(0, 0, 0, "No RADIUS servers defined!\n");
+
+       config->num_radfds = 1 << RADIUS_SHIFT;
 
        // Update plugins
        for (i = 0; i < MAXPLUGINS; i++)
        {
                if (strcmp(config->plugins[i], config->old_plugins[i]) == 0)
                        continue;
+
                if (*config->plugins[i])
                {
                        // Plugin added
@@ -2653,118 +3707,303 @@ void update_config()
                }
        }
        memcpy(config->old_plugins, config->plugins, sizeof(config->plugins));
+       if (!config->cleanup_interval) config->cleanup_interval = 10;
+       if (!config->multi_read_count) config->multi_read_count = 10;
+       if (!config->cluster_address) config->cluster_address = inet_addr(DEFAULT_MCAST_ADDR);
+       if (!*config->cluster_interface)
+               strncpy(config->cluster_interface, DEFAULT_MCAST_INTERFACE, sizeof(config->cluster_interface) - 1);
+
+       if (!config->cluster_hb_interval)
+               config->cluster_hb_interval = PING_INTERVAL;    // Heartbeat every 0.5 seconds.
+
+       if (!config->cluster_hb_timeout)
+               config->cluster_hb_timeout = HB_TIMEOUT;        // 10 missed heartbeat triggers an election.
+
+       if (interval != config->cluster_hb_interval || timeout != config->cluster_hb_timeout)
+       {
+               // Paranoia:  cluster_check_master() treats 2 x interval + 1 sec as
+               // late, ensure we're sufficiently larger than that
+               int t = 4 * config->cluster_hb_interval + 11;
+
+               if (config->cluster_hb_timeout < t)
+               {
+                       LOG(0, 0, 0, "Heartbeat timeout %d too low, adjusting to %d\n", config->cluster_hb_timeout, t);
+                       config->cluster_hb_timeout = t;
+               }
+
+               // Push timing changes to the slaves immediately if we're the master
+               if (config->cluster_iam_master)
+                       cluster_heartbeat();
+
+               interval = config->cluster_hb_interval;
+               timeout = config->cluster_hb_timeout;
+       }
+
+       // Write PID file
+       if (*config->pid_file == '/' && !config->wrote_pid)
+       {
+               FILE *f;
+               if ((f = fopen(config->pid_file, "w")))
+               {
+                       fprintf(f, "%d\n", getpid());
+                       fclose(f);
+                       config->wrote_pid = 1;
+               }
+               else
+               {
+                       LOG(0, 0, 0, "Can't write to PID file %s: %s\n", config->pid_file, strerror(errno));
+               }
+       }
+
        config->reload_config = 0;
 }
 
-void read_config_file()
+static void read_config_file()
 {
        FILE *f;
 
        if (!config->config_file) return;
-       if (!(f = fopen(config->config_file, "r"))) {
+       if (!(f = fopen(config->config_file, "r")))
+       {
                fprintf(stderr, "Can't open config file %s: %s\n", config->config_file, strerror(errno));
                return;
        }
 
-       log(3, 0, 0, 0, "Reading config file %s\n", config->config_file);
+       LOG(3, 0, 0, "Reading config file %s\n", config->config_file);
        cli_do_file(f);
-       log(3, 0, 0, 0, "Done reading config file\n");
+       LOG(3, 0, 0, "Done reading config file\n");
        fclose(f);
        update_config();
-       log_stream = NULL;
 }
 
-int sessionsetup(tunnelidt t, sessionidt s, u8 routes)
+int sessionsetup(tunnelidt t, sessionidt s)
 {
        // A session now exists, set it up
-       ipt ip;
+       in_addr_t 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;
+       int r;
+
+       CSTAT(call_sessionsetup);
+
+       LOG(3, s, t, "Doing session setup for session\n");
+
+       if (!session[s].ip)
+       {
+               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");
+                       return 0;
+               }
+               LOG(3, s, t, "   No IP allocated.  Assigned %s from pool\n",
+                       fmtaddr(htonl(session[s].ip), 0));
        }
 
+
        // Make sure this is right
        session[s].tunnel = t;
+
        // zap old sessions with same IP and/or username
-       // Don't kill walled garden sessions - doing so leads to a DoS
+       // Don't kill gardened sessions - doing so leads to a DoS
        // from someone who doesn't need to know the password
-       ip = session[s].ip;
-       user = session[s].user;
-       for (i = 1; i < MAXSESSION; i++)
        {
-               if (i == s) continue;
-               if (ip == session[i].ip) sessionkill(i, "Duplicate IP address");
-               if (!session[s].walled_garden && !session[i].walled_garden && strcasecmp(user, session[i].user) == 0)
-                       sessionkill(i, "Duplicate session for user");
+               ip = session[s].ip;
+               user = session[s].user;
+               for (i = 1; i <= config->cluster_highest_sessionid; 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 users");
+               }
+       }
+
+       {
+               int routed = 0;
+
+               // Add the route for this session.
+               for (r = 0; r < MAXROUTE && session[s].route[r].ip; r++)
+               {
+                       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.
+               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);
+       }
+
+       if (!session[s].unique_id)
+       {
+               // did this session just finish radius?
+               LOG(3, s, t, "Sending initial IPCP to client\n");
+               sendipcp(t, s);
+               session[s].unique_id = ++last_id;
+       }
+
+       // Run the plugin's against this new session.
+       {
+               struct param_new_session data = { &tunnel[t], &session[s] };
+               run_plugins(PLUGIN_NEW_SESSION, &data);
+       }
+
+       // Allocate TBFs if throttled
+       if (session[s].throttle_in || session[s].throttle_out)
+               throttle_session(s, session[s].throttle_in, session[s].throttle_out);
+
+       session[s].last_packet = time_now;
+
+       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...
+}
+
+//
+// This session just got dropped on us by the master or something.
+// Make sure our tables up up to date...
+//
+int load_session(sessionidt s, sessiont *new)
+{
+       int i;
+       int newip = 0;
+
+               // Sanity checks.
+       if (new->ip_pool_index >= MAXIPPOOL ||
+               new->tunnel >= MAXTUNNEL)
+       {
+               LOG(0, s, 0, "Strange session update received!\n");
+                       // FIXME! What to do here?
+               return 0;
+       }
+
+               //
+               // Ok. All sanity checks passed. Now we're committed to
+               // loading the new session.
+               //
+
+       session[s].tunnel = new->tunnel; // For logging in cache_ipmap
+
+       // 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)
+       {
+               int routed = 0;
+
+               // remove old routes...
+               for (i = 0; i < MAXROUTE && session[s].route[i].ip; i++)
+               {
+                       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);
+               }
+
+               // ...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);
+               }
+
+               routed = 0;
+
+               // 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++;
+
+                       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);
+               }
        }
 
-       if (routes)
+       // check filters
+       if (new->filter_in && (new->filter_in > MAXFILTER || !ip_filters[new->filter_in - 1].name[0]))
        {
-               if (session[s].route[routes].ip && session[s].route[routes].mask)
-               {
-                       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);
-               }
+               LOG(2, s, session[s].tunnel, "Dropping invalid input filter %d\n", (int) new->filter_in);
+               new->filter_in = 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);
 
+       if (new->filter_out && (new->filter_out > MAXFILTER || !ip_filters[new->filter_out - 1].name[0]))
        {
-               struct param_new_session data = { &tunnel[t], &session[s] };
-               run_plugins(PLUGIN_NEW_SESSION, &data);
+               LOG(2, s, session[s].tunnel, "Dropping invalid output filter %d\n", (int) new->filter_out);
+               new->filter_out = 0;
        }
 
-       if (!session[s].sid)
-               session[s].sid = ++last_sid;
-
-       cache_sessionid(htonl(session[s].ip), s);
+       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++;
+       }
 
-       cluster_send_session(s);
-       session[s].last_packet = time_now;
+       if (new->filter_out != session[s].filter_out)
        {
-               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);
+               if (session[s].filter_out) ip_filters[session[s].filter_out - 1].used--;
+               if (new->filter_out)       ip_filters[new->filter_out - 1].used++;
        }
 
-       return 1;       // RADIUS OK and IP allocated, done...
-}
+       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;
 
-#ifdef RINGBUFFER
-void ringbuffer_dump(FILE *stream)
-{
-       int i = ringbuffer->head;
+       // TEMP: old session struct used a uint32_t to define the throttle
+       // speed for both up/down, new uses a uint16_t for each.  Deal with
+       // sessions from an old master for migration.
+       if (new->throttle_out == 0 && new->tbf_out)
+               new->throttle_out = new->throttle_in;
 
-       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;
-       }
+       memcpy(&session[s], new, sizeof(session[s]));   // Copy over..
+
+               // Do fixups into address pool.
+       if (new->ip_pool_index != -1)
+               fix_address_pool(s);
+
+       return 1;
 }
-#endif
 
-void initplugins()
+static void initplugins()
 {
        int i;
 
@@ -2774,211 +4013,659 @@ void initplugins()
                plugins[i] = ll_init();
 }
 
-void add_plugin(char *plugin_name)
+static void *open_plugin(char *plugin_name, int load)
 {
-       void *p;
+       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);
+}
+
+// plugin callback to get a config value
+static void *getconfig(char *key, enum config_typet type)
+{
+       int i;
+
+       for (i = 0; config_values[i].key; i++)
+       {
+               if (!strcmp(config_values[i].key, key))
+               {
+                       if (config_values[i].type == type)
+                               return ((void *) config) + config_values[i].offset;
+
+                       LOG(1, 0, 0, "plugin requested config item \"%s\" expecting type %d, have type %d\n",
+                               key, type, config_values[i].type);
+
+                       return 0;
+               }
+       }
+
+       LOG(1, 0, 0, "plugin requested unknown config item \"%s\"\n", key);
+       return 0;
+}
+
+static int add_plugin(char *plugin_name)
+{
+       static struct pluginfuncs funcs = {
+               _log,
+               _log_hex,
+               fmtaddr,
+               sessionbyuser,
+               sessiontbysessionidt,
+               sessionidtbysessiont,
+               radiusnew,
+               radiussend,
+               getconfig,
+               sessionkill,
+               throttle_session,
+               cluster_send_session,
+       };
+
+       void *p = open_plugin(plugin_name, 1);
        int (*initfunc)(struct pluginfuncs *);
-       char path[256] = {0};
        int i;
-       struct pluginfuncs funcs;
-
-       funcs._log = _log;
-       funcs._log_hex = _log_hex;
-       funcs.inet_toa = inet_toa;
-       funcs.get_session_by_username = sessionbyuser;
-       funcs.get_session_by_id = sessiontbysessionidt;
-       funcs.get_id_by_session = sessionidtbysessiont;
-       funcs.sessionkill = sessionkill;
-       funcs.radiusnew = radiusnew;
-       funcs.radiussend = radiussend;
-
-       snprintf(path, 256, "%s/%s.so", LIBDIR, plugin_name);
-
-       log(2, 0, 0, 0, "Loading plugin from %s\n", path);
-       p = dlopen(path, RTLD_NOW);
+
        if (!p)
        {
-               log(1, 0, 0, 0, "   Plugin load failed: %s\n", dlerror());
-               return;
+               LOG(1, 0, 0, "   Plugin load failed: %s\n", dlerror());
+               return -1;
        }
 
        if (ll_contains(loaded_plugins, p))
        {
                dlclose(p);
-               return;
+               return 0; // already loaded
        }
 
        {
-               int *v = dlsym(p, "__plugin_api_version");
+               int *v = dlsym(p, "plugin_api_version");
                if (!v || *v != PLUGIN_API_VERSION)
                {
-                       log(1, 0, 0, 0, "   Plugin load failed: API version mismatch\n", dlerror());
+                       LOG(1, 0, 0, "   Plugin load failed: API version mismatch: %s\n", dlerror());
                        dlclose(p);
-                       return;
+                       return -1;
                }
        }
 
-       initfunc = dlsym(p, "plugin_init");
-       if (!initfunc)
+       if ((initfunc = dlsym(p, "plugin_init")))
        {
-               log(1, 0, 0, 0, "   Plugin load failed: function plugin_init() does not exist.\n", dlerror());
-               dlclose(p);
-               return;
+               if (!initfunc(&funcs))
+               {
+                       LOG(1, 0, 0, "   Plugin load failed: plugin_init() returned FALSE: %s\n", dlerror());
+                       dlclose(p);
+                       return -1;
+               }
        }
 
-       if (!initfunc(&funcs))
-       {
-               log(1, 0, 0, 0, "   Plugin load failed: plugin_init() returned FALSE.\n", dlerror());
-               dlclose(p);
-               return;
-       }
+       ll_push(loaded_plugins, p);
 
        for (i = 0; i < max_plugin_functions; i++)
        {
                void *x;
-               if (!plugin_functions[i]) continue;
-               if ((x = dlsym(p, plugin_functions[i])))
+               if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
                {
-                       log(3, 0, 0, 0, "   Supports function \"%s\"\n", plugin_functions[i]);
+                       LOG(3, 0, 0, "   Supports function \"%s\"\n", plugin_functions[i]);
                        ll_push(plugins[i], x);
                }
        }
-       log(2, 0, 0, 0, "   Loaded plugin %s\n", plugin_name);
+
+       LOG(2, 0, 0, "   Loaded plugin %s\n", plugin_name);
+       return 1;
 }
 
-void remove_plugin(char *plugin_name)
+static void run_plugin_done(void *plugin)
 {
-       void *p;
-       int (*donefunc)();
-       char path[256] = {0};
-       int i;
+       int (*donefunc)(void) = dlsym(plugin, "plugin_done");
 
-       snprintf(path, 256, "%s/%s.so", LIBDIR, plugin_name);
+       if (donefunc)
+               donefunc();
+}
 
-       log(2, 0, 0, 0, "Removing plugin %s\n", plugin_name);
-       // Get the existing pointer
-       p = dlopen(path, RTLD_LAZY);
-       if (!p) return;
+static int remove_plugin(char *plugin_name)
+{
+       void *p = open_plugin(plugin_name, 0);
+       int loaded = 0;
 
-       for (i = 0; i < max_plugin_functions; i++)
-       {
-               void *x;
-               if (!plugin_functions[i]) continue;
-               if ((x = dlsym(p, plugin_functions[i]))) ll_delete(plugins[i], x);
-       }
+       if (!p)
+               return -1;
 
        if (ll_contains(loaded_plugins, p))
        {
-               ll_delete(loaded_plugins, p);
+               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);
+               }
 
-               donefunc = dlsym(p, "plugin_done");
-               if (donefunc) donefunc();
+               ll_delete(loaded_plugins, p);
+               run_plugin_done(p);
+               loaded = 1;
        }
 
        dlclose(p);
-       dlclose(p);
-       log(2, 0, 0, 0, "Removed plugin %s\n", plugin_name);
+       LOG(2, 0, 0, "Removed plugin %s\n", plugin_name);
+       return loaded;
 }
 
 int run_plugins(int plugin_type, void *data)
 {
        int (*func)(void *data);
-       if (!plugins[plugin_type] || plugin_type > max_plugin_functions) return 1;
+
+       if (!plugins[plugin_type] || plugin_type > max_plugin_functions)
+               return PLUGIN_RET_ERROR;
 
        ll_reset(plugins[plugin_type]);
        while ((func = ll_next(plugins[plugin_type])))
        {
-               int rc;
-               rc = func(data);
-               if (rc == PLUGIN_RET_STOP) return 1;
-               if (rc == PLUGIN_RET_ERROR) return 0;
+               int r = func(data);
+
+               if (r != PLUGIN_RET_OK)
+                       return r; // stop here
        }
-       return 1;
+
+       return PLUGIN_RET_OK;
 }
 
-void processcontrol(u8 * buf, int len, struct sockaddr_in *addr)
+static void plugins_done()
 {
-       char *resp;
-       int l;
-       struct param_control param = { buf, len, ntohl(addr->sin_addr.s_addr), ntohs(addr->sin_port), NULL, 0, 0 };
+       void *p;
 
-       log(4, ntohl(addr->sin_addr.s_addr), 0, 0, "Received ");
-       if (log_stream)
-               dump_packet(buf, log_stream);
+       ll_reset(loaded_plugins);
+       while ((p = ll_next(loaded_plugins)))
+               run_plugin_done(p);
+}
+
+static void processcontrol(uint8_t * buf, int len, struct sockaddr_in *addr, int alen)
+{
+       struct nsctl request;
+       struct nsctl response;
+       int type = unpack_control(&request, buf, len);
+       int r;
+       void *p;
+
+       if (log_stream && config->debug >= 4)
+       {
+               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);
+               }
+       }
+
+       switch (type)
+       {
+       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;
+               }
+
+               break;
+
+       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;
+               }
+
+               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)
+                       {
+                               response.type = NSCTL_RES_ERR;
+                               response.argc = 1;
+                               response.argv[0] = param.additional
+                                       ? param.additional
+                                       : "error returned by plugin";
+                       }
+                       else if (r == PLUGIN_RET_NOTMASTER)
+                       {
+                               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)
+                               {
+                                       strcpy(msg + 23, fmtaddr(config->cluster_master_address, 0));
+                                       response.argv[0] = msg;
+                               }
+                               else
+                               {
+                                       response.argv[0] = "must be run on master: none elected";
+                               }
+                       }
+                       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
+                       {
+                               response.type = param.response;
+                               response.argc = 0;
+                               if (param.additional)
+                               {
+                                       response.argc = 1;
+                                       response.argv[0] = param.additional;
+                               }
+                       }
+               }
 
-       resp = calloc(1400, 1);
-       l = new_packet(PKT_RESP_ERROR, resp);
-       *(int *)(resp + 6) = *(int *)(buf + 6);
+               break;
 
-       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;
+       default:
+               response.type = NSCTL_RES_ERR;
+               response.argc = 1;
+               response.argv[0] = "error unpacking control packet";
+       }
 
-       if (param.type == PKT_LOAD_PLUGIN && param.data_length)
+       buf = calloc(NSCTL_MAX_PKT_SZ, 1);
+       if (!buf)
        {
-               add_plugin(param.data);
+               LOG(2, 0, 0, "Failed to allocate nsctl response\n");
+               return;
        }
-       else if (param.type == PKT_UNLOAD_PLUGIN && param.data_length)
+
+       r = pack_control(buf, NSCTL_MAX_PKT_SZ, response.type, response.argc, response.argv);
+       if (r > 0)
        {
-               remove_plugin(param.data);
+               sendto(controlfd, buf, r, 0, (const struct sockaddr *) addr, alen);
+               if (log_stream && config->debug >= 4)
+               {
+                       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);
+}
+
+static tunnelidt new_tunnel()
+{
+       tunnelidt i;
+       for (i = 1; i < MAXTUNNEL; i++)
+       {
+               if (tunnel[i].state == TUNNELFREE)
+               {
+                       LOG(4, 0, i, "Assigning tunnel ID %d\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;
+}
+
+//
+// We're becoming the master. Do any required setup..
+//
+// This is principally telling all the plugins that we're
+// now a master, and telling them about all the sessions
+// that are active too..
+//
+void become_master(void)
+{
+       int s, i;
+       run_plugins(PLUGIN_BECOME_MASTER, NULL);
+
+       // running a bunch of iptables commands is slow and can cause
+       // the master to drop tunnels on takeover--kludge around the
+       // problem by forking for the moment (note: race)
+       if (!fork_and_close())
+       {
+               for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
+               {
+                       if (!session[s].opened) // Not an in-use session.
+                               continue;
+
+                       run_plugins(PLUGIN_NEW_SESSION_MASTER, &session[s]);
+               }
+               exit(0);
+       }
+
+       // add radius fds
+       for (i = 0; i < config->num_radfds; i++)
+       {
+               FD_SET(radfds[i], &readset);
+               if (radfds[i] > readset_n)
+                       readset_n = radfds[i];
+       }
+}
+
+int cmd_show_hist_idle(struct cli_def *cli, char *command, char **argv, int argc)
+{
+       int s, i;
+       int count = 0;
+       int buckets[64];
+
+       if (CLI_HELP_REQUESTED)
+               return CLI_HELP_NO_ARGS;
+
+       time(&time_now);
+       for (i = 0; i < 64;++i) buckets[i] = 0;
+
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
+       {
+               int idle;
+               if (!session[s].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];
+       }
+
+       for (i = 0; i < 63; ++i)
        {
-               run_plugins(PLUGIN_CONTROL, &param);
+               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;
 
-       if (param.send_response)
+       for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
        {
-               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");
+               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)
+               {
+                       ++open;
+                       d >>= 1; // half.
+               }
+               ++count;
+               ++buckets[open];
        }
 
-       free(resp);
+       s = 1;
+       for (i = 0; i  < 30; ++i)
+       {
+               cli_print(cli, " < %8d seconds : %7.2f%% (%6d)", s, (double) buckets[i] * 100.0 / count , buckets[i]);
+               s <<= 1;
+       }
+       cli_print(cli, "%d total sessions open.", count);
+       return CLI_OK;
 }
 
-/*
- * HACK
- * Go through all of the tunnels and do some cleanups
+/* Unhide an avp.
+ *
+ * This unencodes the AVP using the L2TP CHAP secret and the
+ * previously stored random vector. It replaces the hidden data with
+ * the cleartext data and returns the length of the cleartext data
+ * (including the AVP "header" of 6 bytes).
+ *
+ * Based on code from rp-l2tpd by Roaring Penguin Software Inc.
  */
-void tunnel_clean()
+static int unhide_avp(uint8_t *avp, tunnelidt t, sessionidt s, uint16_t length)
 {
-       int i;
+       MD5_CTX ctx;
+       uint8_t *cursor;
+       uint8_t digest[16];
+       uint8_t working_vector[16];
+       uint16_t hidden_length;
+       uint8_t type[2];
+       size_t done, todo;
+       uint8_t *output;
+
+       // Find the AVP type.
+       type[0] = *(avp + 4);
+       type[1] = *(avp + 5);
+
+       // Line up with the hidden data
+       cursor = output = avp + 6;
+
+       // Compute initial pad
+       MD5Init(&ctx);
+       MD5Update(&ctx, type, 2);
+       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
+       MD5Update(&ctx, session[s].random_vector, session[s].random_vector_length);
+       MD5Final(digest, &ctx);
 
-       log(1, 0, 0, 0, "Cleaning tunnels array\n");
+       // Get hidden length
+       hidden_length = ((uint16_t) (digest[0] ^ cursor[0])) * 256 + (uint16_t) (digest[1] ^ cursor[1]);
 
-       for (i = 1; i < MAXTUNNEL; i++)
+       // Keep these for later use
+       working_vector[0] = *cursor;
+       working_vector[1] = *(cursor + 1);
+       cursor += 2;
+
+       if (hidden_length > length - 8)
+       {
+               LOG(1, s, t, "Hidden length %d too long in AVP of length %d\n", (int) hidden_length, (int) length);
+               return 0;
+       }
+
+       /* Decrypt remainder */
+       done = 2;
+       todo = hidden_length;
+       while (todo)
        {
-               if (!tunnel[i].ip
-                               || !*tunnel[i].hostname
-                               || (tunnel[i].state == TUNNELDIE && tunnel[i].die >= time_now))
+               working_vector[done] = *cursor;
+               *output = digest[done] ^ *cursor;
+               ++output;
+               ++cursor;
+               --todo;
+               ++done;
+               if (done == 16 && todo)
                {
-                       tunnelclear(i);
+                       // Compute new digest
+                       done = 0;
+                       MD5Init(&ctx);
+                       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
+                       MD5Update(&ctx, &working_vector, 16);
+                       MD5Final(digest, &ctx);
                }
        }
+
+       return hidden_length + 6;
 }
 
-void tunnelclear(tunnelidt t)
+static int ip_filter_port(ip_filter_portt *p, uint16_t port)
 {
-       if (!t) return;
-       memset(&tunnel[t], 0, sizeof(tunnel[t]));
-       tunnel[t].state = TUNNELFREE;
+       switch (p->op)
+       {
+       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 0;
 }
 
-tunnelidt new_tunnel()
+static int ip_filter_flag(uint8_t op, uint8_t sflags, uint8_t cflags, uint8_t flags)
 {
-       tunnelidt i;
-       for (i = 1; i < MAXTUNNEL; i++)
+       switch (op)
        {
-               if (tunnel[i].state == TUNNELFREE)
-               {
-                       log(4, 0, 0, i, "Assigning tunnel ID %d\n", i);
-                       return i;
-               }
+       case FILTER_FLAG_OP_ANY:
+               return (flags & sflags) || (~flags & cflags);
+
+       case FILTER_FLAG_OP_ALL:
+               return (flags & sflags) == sflags && (~flags & cflags) == cflags;
+
+       case FILTER_FLAG_OP_EST:
+               return (flags & (TCP_FLAG_ACK|TCP_FLAG_RST)) && (~flags & TCP_FLAG_SYN);
        }
-       log(0, 0, 0, 0, "Can't find a free tunnel! There shouldn't be this many in use!\n");
+
        return 0;
 }
 
+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))
+       {
+               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;
+               }
+       }
+
+       for (rule = ip_filters[filter].rules; rule->action; rule++)
+       {
+               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)
+               {
+                       if (!rule->frag || rule->action == FILTER_ACTION_DENY)
+                               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;
+       }
+
+       // default deny
+       return 0;
+}