Stop BGP session even when we are master

[l2tpns.git] / l2tpns.c

// L2TP Network Server
// Adrian Kennard 2002
// Copyright (c) 2003, 2004, 2005, 2006 Optus Internet Engineering
// Copyright (c) 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced
// vim: sw=8 ts=8

char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.176 2011/01/20 12:48:40 bodea Exp $";

#include <arpa/inet.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <linux/if_tun.h>
#define SYSLOG_NAMES
#include <syslog.h>
#include <malloc.h>
#include <math.h>
#include <net/route.h>
#include <sys/mman.h>
#include <netdb.h>
#include <netinet/in.h>
#include <netinet/ip6.h>
#include <stdarg.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <ctype.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/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 "plugin.h"
#include "ll.h"
#include "constants.h"
#include "control.h"
#include "util.h"
#include "tbf.h"

#ifdef BGP
#include "bgp.h"
#endif

// Globals
configt *config = NULL;         // all configuration
int tunfd = -1;                 // tun interface file handle. (network device)
int udpfd = -1;                 // UDP file handle
int controlfd = -1;             // Control signal handle
int clifd = -1;                 // Socket listening for CLI connections.
int daefd = -1;                 // Socket listening for DAE connections.
int snoopfd = -1;               // UDP file handle for sending out intercept data
int *radfds = NULL;             // RADIUS requests file handles
int ifrfd = -1;                 // File descriptor for routing, etc
int ifr6fd = -1;                // File descriptor for IPv6 routing, etc
int rand_fd = -1;               // Random data source
int cluster_sockfd = -1;        // Intra-cluster communications socket.
int epollfd = -1;               // event polling
time_t basetime = 0;            // base clock
char hostname[1000] = "";       // us.
static int tunidx;              // ifr_ifindex of tun device
static int syslog_log = 0;      // are we logging to syslog
static FILE *log_stream = 0;    // file handle for direct logging (i.e. direct into file, not via syslog).
uint32_t last_id = 0;           // Unique ID for radius accounting
// Guest change
char guest_users[10][32];       // Array of guest users
int guest_accounts_num = 0;     // Number of guest users

// calculated from config->l2tp_mtu
uint16_t MRU = 0;               // PPP MRU
uint16_t MSS = 0;               // TCP MSS

struct cli_session_actions *cli_session_actions = NULL; // Pending session changes requested by CLI
struct cli_tunnel_actions *cli_tunnel_actions = NULL;   // Pending tunnel changes required by CLI

union iphash {
        sessionidt sess;
        union iphash *idx;
} ip_hash[256];                 // Mapping from IP address to session structures.

struct ipv6radix {
        sessionidt sess;
        struct ipv6radix *branch;
} ipv6_hash[256];               // Mapping from IPv6 address to session structures.

// Traffic counters.
static uint32_t udp_rx = 0, udp_rx_pkt = 0, udp_tx = 0;
static uint32_t eth_rx = 0, eth_rx_pkt = 0;
uint32_t eth_tx = 0;

static uint32_t ip_pool_size = 1;       // Size of the pool of addresses used for dynamic address allocation.
time_t time_now = 0;                    // Current time in seconds since epoch.
static char time_now_string[64] = {0};  // Current time as a string.
static int time_changed = 0;            // time_now changed
char main_quit = 0;                     // True if we're in the process of exiting.
static char main_reload = 0;            // Re-load pending
linked_list *loaded_plugins;
linked_list *plugins[MAX_PLUGIN_TYPES];

#define membersize(STRUCT, MEMBER) sizeof(((STRUCT *)0)->MEMBER)
#define CONFIG(NAME, MEMBER, TYPE) { NAME, offsetof(configt, MEMBER), membersize(configt, MEMBER), TYPE }

config_descriptt config_values[] = {
        CONFIG("debug", debug, INT),
        CONFIG("log_file", log_filename, STRING),
        CONFIG("pid_file", pid_file, STRING),
        CONFIG("random_device", random_device, STRING),
        CONFIG("l2tp_secret", l2tp_secret, STRING),
        CONFIG("l2tp_mtu", l2tp_mtu, INT),
        CONFIG("ppp_restart_time", ppp_restart_time, INT),
        CONFIG("ppp_max_configure", ppp_max_configure, INT),
        CONFIG("ppp_max_failure", ppp_max_failure, INT),
        CONFIG("primary_dns", default_dns1, IPv4),
        CONFIG("secondary_dns", default_dns2, IPv4),
        CONFIG("primary_radius", radiusserver[0], IPv4),
        CONFIG("secondary_radius", radiusserver[1], IPv4),
        CONFIG("primary_radius_port", radiusport[0], SHORT),
        CONFIG("secondary_radius_port", radiusport[1], SHORT),
        CONFIG("radius_accounting", radius_accounting, BOOL),
        CONFIG("radius_interim", radius_interim, INT),
        CONFIG("radius_secret", radiussecret, STRING),
        CONFIG("radius_authtypes", radius_authtypes_s, STRING),
        CONFIG("radius_dae_port", radius_dae_port, SHORT),
        CONFIG("radius_bind_min", radius_bind_min, SHORT),
        CONFIG("radius_bind_max", radius_bind_max, SHORT),
        CONFIG("allow_duplicate_users", allow_duplicate_users, BOOL),
        CONFIG("kill_timedout_sessions", kill_timedout_sessions, BOOL),
        CONFIG("guest_account", guest_user, STRING),
        CONFIG("bind_address", bind_address, IPv4),
        CONFIG("peer_address", peer_address, IPv4),
        CONFIG("send_garp", send_garp, BOOL),
        CONFIG("throttle_speed", rl_rate, UNSIGNED_LONG),
        CONFIG("throttle_buckets", num_tbfs, INT),
        CONFIG("accounting_dir", accounting_dir, STRING),
        CONFIG("dump_speed", dump_speed, BOOL),
        CONFIG("multi_read_count", multi_read_count, INT),
        CONFIG("scheduler_fifo", scheduler_fifo, BOOL),
        CONFIG("lock_pages", lock_pages, BOOL),
        CONFIG("icmp_rate", icmp_rate, INT),
        CONFIG("packet_limit", max_packets, INT),
        CONFIG("cluster_address", cluster_address, IPv4),
        CONFIG("cluster_interface", cluster_interface, STRING),
        CONFIG("cluster_mcast_ttl", cluster_mcast_ttl, INT),
        CONFIG("cluster_hb_interval", cluster_hb_interval, INT),
        CONFIG("cluster_hb_timeout", cluster_hb_timeout, INT),
        CONFIG("cluster_master_min_adv", cluster_master_min_adv, INT),
        CONFIG("ipv6_prefix", ipv6_prefix, IPv6),
        { NULL, 0, 0, 0 },
};

static char *plugin_functions[] = {
        NULL,
        "plugin_pre_auth",
        "plugin_post_auth",
        "plugin_timer",
        "plugin_new_session",
        "plugin_kill_session",
        "plugin_control",
        "plugin_radius_response",
        "plugin_radius_reset",
        "plugin_radius_account",
        "plugin_become_master",
        "plugin_new_session_master",
};

#define max_plugin_functions (sizeof(plugin_functions) / sizeof(char *))

// Counters for shutdown sessions
static sessiont shut_acct[8192];
static sessionidt shut_acct_n = 0;

tunnelt *tunnel = NULL;                 // Array of tunnel structures.
bundlet *bundle = NULL;                 // Array of bundle structures.
fragmentationt *frag = NULL;            // Array of fragmentation structures.
sessiont *session = NULL;               // Array of session structures.
sessionlocalt *sess_local = NULL;       // Array of local per-session counters.
radiust *radius = NULL;                 // Array of radius structures.
ippoolt *ip_address_pool = NULL;        // Array of dynamic IP addresses.
ip_filtert *ip_filters = NULL;          // Array of named filters.
static controlt *controlfree = 0;
struct Tstats *_statistics = NULL;
#ifdef RINGBUFFER
struct Tringbuffer *ringbuffer = NULL;
#endif

static void cache_ipmap(in_addr_t ip, sessionidt s);
static void uncache_ipmap(in_addr_t ip);
static void cache_ipv6map(struct in6_addr ip, int prefixlen, sessionidt s);
static void free_ip_address(sessionidt s);
static void dump_acct_info(int all);
static void sighup_handler(int sig);
static void shutdown_handler(int sig);
static void sigchild_handler(int sig);
static void build_chap_response(uint8_t *challenge, uint8_t id, uint16_t challenge_length, uint8_t **challenge_response);
static void update_config(void);
static void read_config_file(void);
static void initplugins(void);
static int add_plugin(char *plugin_name);
static int remove_plugin(char *plugin_name);
static void plugins_done(void);
static void processcontrol(uint8_t *buf, int len, struct sockaddr_in *addr, int alen, struct in_addr *local);
static tunnelidt new_tunnel(void);
static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vector, size_t vec_len);
static void bundleclear(bundleidt b);

// on slaves, alow BGP to withdraw cleanly before exiting
#define QUIT_DELAY      5

// quit actions (master)
#define QUIT_FAILOVER   1 // SIGTERM: exit when all control messages have been acked (for cluster failover)
#define QUIT_SHUTDOWN   2 // SIGQUIT: shutdown sessions/tunnels, reject new connections

// return internal time (10ths since process startup), set f if given
// as a side-effect sets time_now, and time_changed
static clockt now(double *f)
{
        struct timeval t;
        gettimeofday(&t, 0);
        if (f) *f = t.tv_sec + t.tv_usec / 1000000.0;
        if (t.tv_sec != time_now)
        {
            time_now = t.tv_sec;
            time_changed++;
        }
        return (t.tv_sec - basetime) * 10 + t.tv_usec / 100000 + 1;
}

// work out a retry time based on try number
// 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(NULL) + 10 * (1 << try);
}


//
// Log a debug message.  Typically called via the LOG macro
//
void _log(int level, sessionidt s, tunnelidt t, const char *format, ...)
{
        static char message[65536] = {0};
        va_list ap;

#ifdef RINGBUFFER
        if (ringbuffer)
        {
                if (++ringbuffer->tail >= RINGBUFFER_SIZE)
                        ringbuffer->tail = 0;
                if (ringbuffer->tail == ringbuffer->head)
                        if (++ringbuffer->head >= RINGBUFFER_SIZE)
                                ringbuffer->head = 0;

                ringbuffer->buffer[ringbuffer->tail].level = level;
                ringbuffer->buffer[ringbuffer->tail].session = s;
                ringbuffer->buffer[ringbuffer->tail].tunnel = t;
                va_start(ap, format);
                vsnprintf(ringbuffer->buffer[ringbuffer->tail].message, 4095, format, ap);
                va_end(ap);
        }
#endif

        if (config->debug < level) return;

        va_start(ap, format);
        vsnprintf(message, sizeof(message), format, ap);

        if (log_stream)
                fprintf(log_stream, "%s %02d/%02d %s", time_now_string, t, s, message);
        else if (syslog_log)
                syslog(level + 2, "%02d/%02d %s", t, s, message); // We don't need LOG_EMERG or LOG_ALERT

        va_end(ap);
}

void _log_hex(int level, const char *title, const uint8_t *data, int maxsize)
{
        int i, j;
        const uint8_t *d = data;

        if (config->debug < level) return;

        // No support for _log_hex to syslog
        if (log_stream)
        {
                _log(level, 0, 0, "%s (%d bytes):\n", title, maxsize);
                setvbuf(log_stream, NULL, _IOFBF, 16384);

                for (i = 0; i < maxsize; )
                {
                        fprintf(log_stream, "%4X: ", i);
                        for (j = i; j < maxsize && j < (i + 16); j++)
                        {
                                fprintf(log_stream, "%02X ", d[j]);
                                if (j == i + 7)
                                        fputs(": ", log_stream);
                        }

                        for (; j < i + 16; j++)
                        {
                                fputs("   ", log_stream);
                                if (j == i + 7)
                                        fputs(": ", log_stream);
                        }

                        fputs("  ", log_stream);
                        for (j = i; j < maxsize && j < (i + 16); j++)
                        {
                                if (d[j] >= 0x20 && d[j] < 0x7f && d[j] != 0x20)
                                        fputc(d[j], log_stream);
                                else
                                        fputc('.', log_stream);

                                if (j == i + 7)
                                        fputs("  ", log_stream);
                        }

                        i = j;
                        fputs("\n", log_stream);
                }

                fflush(log_stream);
                setbuf(log_stream, NULL);
        }
}

// update a counter, accumulating 2^32 wraps
void increment_counter(uint32_t *counter, uint32_t *wrap, uint32_t delta)
{
        uint32_t new = *counter + delta;
        if (new < *counter)
                (*wrap)++;

        *counter = new;
}

// initialise the random generator
static void initrandom(char *source)
{
        static char path[sizeof(config->random_device)] = "*undefined*";

        // reinitialise only if we are forced to do so or if the config has changed
        if (source && !strncmp(path, source, sizeof(path)))
                return;

        // close previous source, if any
        if (rand_fd >= 0)
                close(rand_fd);

        rand_fd = -1;

        if (source)
        {
                // register changes
                snprintf(path, sizeof(path), "%s", source);

                if (*path == '/')
                {
                        rand_fd = open(path, O_RDONLY|O_NONBLOCK);
                        if (rand_fd < 0)
                                LOG(0, 0, 0, "Error opening the random device %s: %s\n",
                                        path, strerror(errno));
                }
        }
}

// fill buffer with random data
void random_data(uint8_t *buf, int len)
{
        int n = 0;

        CSTAT(random_data);
        if (rand_fd >= 0)
        {
                n = read(rand_fd, buf, len);
                if (n >= len) return;
                if (n < 0)
                {
                        if (errno != EAGAIN)
                        {
                                LOG(0, 0, 0, "Error reading from random source: %s\n",
                                        strerror(errno));

                                // fall back to rand()
                                initrandom(NULL);
                        }

                        n = 0;
                }
        }

        // append missing data
        while (n < len)
                // not using the low order bits from the prng stream
                buf[n++] = (rand() >> 4) & 0xff;
}

// Add a route
//
// 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->tundevice;
        r.rt_dst.sa_family = AF_INET;
        *(uint32_t *) & (((struct sockaddr_in *) &r.rt_dst)->sin_addr.s_addr) = htonl(ip);
        r.rt_gateway.sa_family = AF_INET;
        *(uint32_t *) & (((struct sockaddr_in *) &r.rt_gateway)->sin_addr.s_addr) = htonl(gw);
        r.rt_genmask.sa_family = AF_INET;
        *(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 if (mask == 0xffffffff)
                r.rt_flags |= RTF_HOST;

        LOG(1, s, 0, "Route %s %s/%s%s%s\n", add ? "add" : "del",
            fmtaddr(htonl(ip), 0), fmtaddr(htonl(mask), 1),
            gw ? " via" : "", gw ? fmtaddr(htonl(gw), 2) : "");

        if (ioctl(ifrfd, add ? SIOCADDRT : SIOCDELRT, (void *) &r) < 0)
                LOG(0, 0, 0, "routeset() error in ioctl: %s\n", strerror(errno));

#ifdef BGP
        if (add)
                bgp_add_route(htonl(ip), htonl(mask));
        else
                bgp_del_route(htonl(ip), htonl(mask));
#endif /* BGP */

                // Add/Remove the IPs to the 'sessionbyip' cache.
                // Note that we add the zero address in the case of
                // a network route. Roll on CIDR.

                // Note that 's == 0' implies this is the address pool.
                // We still cache it here, because it will pre-fill
                // the malloc'ed tree.

        if (s)
        {
                if (!add)       // Are we deleting a route?
                        s = 0;  // Caching the session as '0' is the same as uncaching.

                for (i = ip; (i&mask) == (ip&mask) ; ++i)
                        cache_ipmap(i, s);
        }
}

void route6set(sessionidt s, struct in6_addr ip, int prefixlen, int add)
{
        struct in6_rtmsg rt;
        char ipv6addr[INET6_ADDRSTRLEN];

        if (ifr6fd < 0)
        {
                LOG(0, 0, 0, "Asked to set IPv6 route, but IPv6 not setup.\n");
                return;
        }

        memset(&rt, 0, sizeof(rt));

        memcpy(&rt.rtmsg_dst, &ip, sizeof(struct in6_addr));
        rt.rtmsg_dst_len = prefixlen;
        rt.rtmsg_metric = 1;
        rt.rtmsg_flags = RTF_UP;
        rt.rtmsg_ifindex = tunidx;

        LOG(1, 0, 0, "Route %s %s/%d\n",
            add ? "add" : "del",
            inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN),
            prefixlen);

        if (ioctl(ifr6fd, add ? SIOCADDRT : SIOCDELRT, (void *) &rt) < 0)
                LOG(0, 0, 0, "route6set() error in ioctl: %s\n",
                                strerror(errno));

        // FIXME: need to add BGP routing (RFC2858)

        if (s)
        {
                if (!add)       // Are we deleting a route?
                        s = 0;  // Caching the session as '0' is the same as uncaching.

                cache_ipv6map(ip, prefixlen, s);
        }
        
        return;
}

// defined in linux/ipv6.h, but tricky to include from user-space
// TODO: move routing to use netlink rather than ioctl
struct in6_ifreq {
        struct in6_addr ifr6_addr;
        __u32 ifr6_prefixlen;
        unsigned int ifr6_ifindex;
};

//
// Set up TUN interface
static void inittun(void)
{
        struct ifreq ifr;
        struct in6_ifreq ifr6;
        struct sockaddr_in sin = {0};
        memset(&ifr, 0, sizeof(ifr));
        ifr.ifr_flags = IFF_TUN;

        tunfd = open(TUNDEVICE, O_RDWR);
        if (tunfd < 0)
        {                          // fatal
                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(tunfd, TUNSETIFF, (void *) &ifr) < 0)
        {
                LOG(0, 0, 0, "Can't set tun interface: %s\n", strerror(errno));
                exit(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 = 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)
        {
                LOG(0, 0, 0, "Error setting tun address: %s\n", strerror(errno));
                exit(1);
        }
        /* Bump up the qlen to deal with bursts from the network */
        ifr.ifr_qlen = 1000;
        if (ioctl(ifrfd, SIOCSIFTXQLEN, (void *) &ifr) < 0)
        {
                LOG(0, 0, 0, "Error setting tun queue length: %s\n", strerror(errno));
                exit(1);
        }
        /* set MTU to modem MRU */
        ifr.ifr_mtu = MRU;
        if (ioctl(ifrfd, SIOCSIFMTU, (void *) &ifr) < 0)
        {
                LOG(0, 0, 0, "Error setting tun MTU: %s\n", strerror(errno));
                exit(1);
        }
        ifr.ifr_flags = IFF_UP;
        if (ioctl(ifrfd, SIOCSIFFLAGS, (void *) &ifr) < 0)
        {
                LOG(0, 0, 0, "Error setting tun flags: %s\n", strerror(errno));
                exit(1);
        }
        if (ioctl(ifrfd, SIOCGIFINDEX, (void *) &ifr) < 0)
        {
                LOG(0, 0, 0, "Error getting tun ifindex: %s\n", strerror(errno));
                exit(1);
        }
        tunidx = ifr.ifr_ifindex;

        // Only setup IPv6 on the tun device if we have a configured prefix
        if (config->ipv6_prefix.s6_addr[0]) {
                ifr6fd = socket(PF_INET6, SOCK_DGRAM, 0);

                // Link local address is FE80::1
                memset(&ifr6.ifr6_addr, 0, sizeof(ifr6.ifr6_addr));
                ifr6.ifr6_addr.s6_addr[0] = 0xFE;
                ifr6.ifr6_addr.s6_addr[1] = 0x80;
                ifr6.ifr6_addr.s6_addr[15] = 1;
                ifr6.ifr6_prefixlen = 64;
                ifr6.ifr6_ifindex = ifr.ifr_ifindex;
                if (ioctl(ifr6fd, SIOCSIFADDR, (void *) &ifr6) < 0)
                {
                        LOG(0, 0, 0, "Error setting tun IPv6 link local address:"
                                " %s\n", strerror(errno));
                }

                // Global address is prefix::1
                memset(&ifr6.ifr6_addr, 0, sizeof(ifr6.ifr6_addr));
                ifr6.ifr6_addr = config->ipv6_prefix;
                ifr6.ifr6_addr.s6_addr[15] = 1;
                ifr6.ifr6_prefixlen = 64;
                ifr6.ifr6_ifindex = ifr.ifr_ifindex;
                if (ioctl(ifr6fd, SIOCSIFADDR, (void *) &ifr6) < 0)
                {
                        LOG(0, 0, 0, "Error setting tun IPv6 global address: %s\n",
                                strerror(errno));
                }
        }
}

// set up UDP ports
static void initudp(void)
{
        int on = 1;
        struct sockaddr_in addr;

        // Tunnel
        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
        addr.sin_port = htons(L2TPPORT);
        addr.sin_addr.s_addr = config->bind_address;
        udpfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
        setsockopt(udpfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
        {
                int flags = fcntl(udpfd, F_GETFL, 0);
                fcntl(udpfd, F_SETFL, flags | O_NONBLOCK);
        }
        if (bind(udpfd, (struct sockaddr *) &addr, sizeof(addr)) < 0)
        {
                LOG(0, 0, 0, "Error in UDP bind: %s\n", strerror(errno));
                exit(1);
        }

        // Control
        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
        addr.sin_port = htons(NSCTL_PORT);
        controlfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
        setsockopt(controlfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
        setsockopt(controlfd, SOL_IP, IP_PKTINFO, &on, sizeof(on)); // recvfromto
        if (bind(controlfd, (struct sockaddr *) &addr, sizeof(addr)) < 0)
        {
                LOG(0, 0, 0, "Error in control bind: %s\n", strerror(errno));
                exit(1);
        }

        // Dynamic Authorization Extensions to RADIUS
        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
        addr.sin_port = htons(config->radius_dae_port);
        daefd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
        setsockopt(daefd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
        setsockopt(daefd, SOL_IP, IP_PKTINFO, &on, sizeof(on)); // recvfromto
        if (bind(daefd, (struct sockaddr *) &addr, sizeof(addr)) < 0)
        {
                LOG(0, 0, 0, "Error in DAE bind: %s\n", strerror(errno));
                exit(1);
        }

        // Intercept
        snoopfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
}

//
// Find session by IP, < 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 sessionidt lookup_ipmap(in_addr_t ip)
{
        uint8_t *a = (uint8_t *) &ip;
        union iphash *h = ip_hash;

        if (!(h = h[*a++].idx)) return 0;
        if (!(h = h[*a++].idx)) return 0;
        if (!(h = h[*a++].idx)) return 0;

        return h[*a].sess;
}

static sessionidt lookup_ipv6map(struct in6_addr ip)
{
        struct ipv6radix *curnode;
        int i;
        int s;
        char ipv6addr[INET6_ADDRSTRLEN];

        curnode = &ipv6_hash[ip.s6_addr[0]];
        i = 1;
        s = curnode->sess;

        while (s == 0 && i < 15 && curnode->branch != NULL)
        {
                curnode = &curnode->branch[ip.s6_addr[i]];
                s = curnode->sess;
                i++;
        }

        LOG(4, s, session[s].tunnel, "Looking up address %s and got %d\n",
                        inet_ntop(AF_INET6, &ip, ipv6addr,
                                INET6_ADDRSTRLEN),
                        s);

        return s;
}

sessionidt sessionbyip(in_addr_t ip)
{
        sessionidt s = lookup_ipmap(ip);
        CSTAT(sessionbyip);

        if (s > 0 && s < MAXSESSION && session[s].opened)
                return s;

        return 0;
}

sessionidt sessionbyipv6(struct in6_addr ip)
{
        sessionidt s;
        CSTAT(sessionbyipv6);

        if (!memcmp(&config->ipv6_prefix, &ip, 8) ||
                (ip.s6_addr[0] == 0xFE &&
                 ip.s6_addr[1] == 0x80 &&
                 ip.s6_addr16[1] == 0 &&
                 ip.s6_addr16[2] == 0 &&
                 ip.s6_addr16[3] == 0)) {
                s = lookup_ipmap(*(in_addr_t *) &ip.s6_addr[8]);
        } else {
                s = lookup_ipv6map(ip);
        }

        if (s > 0 && s < MAXSESSION && session[s].opened)
                return s;

        return 0;
}

//
// 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, sessionidt s)
{
        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;
        union iphash *h = ip_hash;
        int i;

        for (i = 0; i < 3; i++)
        {
                if (!(h[a[i]].idx || (h[a[i]].idx = calloc(256, sizeof(union iphash)))))
                        return;

                h = h[a[i]].idx;
        }

        h[a[3]].sess = 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.
}

static void uncache_ipmap(in_addr_t ip)
{
        cache_ipmap(ip, 0);     // Assign it to the NULL session.
}

static void cache_ipv6map(struct in6_addr ip, int prefixlen, sessionidt s)
{
        int i;
        int bytes;
        struct ipv6radix *curnode;
        char ipv6addr[INET6_ADDRSTRLEN];

        curnode = &ipv6_hash[ip.s6_addr[0]];

        bytes = prefixlen >> 3;
        i = 1;
        while (i < bytes) {
                if (curnode->branch == NULL)
                {
                        if (!(curnode->branch = calloc(256,
                                        sizeof (struct ipv6radix))))
                                return;
                }
                curnode = &curnode->branch[ip.s6_addr[i]];
                i++;
        }

        curnode->sess = s;

        if (s > 0)
                LOG(4, s, session[s].tunnel, "Caching ip address %s/%d\n",
                                inet_ntop(AF_INET6, &ip, ipv6addr, 
                                        INET6_ADDRSTRLEN),
                                prefixlen);
        else if (s == 0)
                LOG(4, 0, 0, "Un-caching ip address %s/%d\n",
                                inet_ntop(AF_INET6, &ip, ipv6addr, 
                                        INET6_ADDRSTRLEN),
                                prefixlen);
}

//
// CLI list to dump current ipcache.
//
int cmd_show_ipcache(struct cli_def *cli, char *command, char **argv, int argc)
{
        union iphash *d = 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[i].idx)
                        continue;

                e = d[i].idx;
                for (j = 0; j < 256; ++j)
                {
                        if (!e[j].idx)
                                continue;

                        f = e[j].idx;
                        for (k = 0; k < 256; ++k)
                        {
                                if (!f[k].idx)
                                        continue;

                                g = f[k].idx;
                                for (l = 0; l < 256; ++l)
                                {
                                        if (!g[l].sess)
                                                continue;

                                        cli_print(cli, "%7d %d.%d.%d.%d", g[l].sess, i, j, k, l);
                                        ++count;
                                }
                        }
                }
        }
        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;
        CSTAT(sessionbyuser);

        for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
        {
                if (!session[s].opened)
                        continue;

                if (session[s].walled_garden)
                        continue;               // Skip walled garden users.

                if (!strncmp(session[s].user, username, 128))
                        return s;

        }
        return 0;       // Not found.
}

void send_garp(in_addr_t ip)
{
        int s;
        struct ifreq ifr;
        uint8_t mac[6];

        s = socket(PF_INET, SOCK_DGRAM, 0);
        if (s < 0)
        {
                LOG(0, 0, 0, "Error creating socket for GARP: %s\n", strerror(errno));
                return;
        }
        memset(&ifr, 0, sizeof(ifr));
        strncpy(ifr.ifr_name, "eth0", sizeof(ifr.ifr_name) - 1);
        if (ioctl(s, SIOCGIFHWADDR, &ifr) < 0)
        {
                LOG(0, 0, 0, "Error getting eth0 hardware address for GARP: %s\n", strerror(errno));
                close(s);
                return;
        }
        memcpy(mac, &ifr.ifr_hwaddr.sa_data, 6*sizeof(char));
        if (ioctl(s, SIOCGIFINDEX, &ifr) < 0)
        {
                LOG(0, 0, 0, "Error getting eth0 interface index for GARP: %s\n", strerror(errno));
                close(s);
                return;
        }
        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(uint8_t * buf, uint16_t l, tunnelidt t)
{
        struct sockaddr_in addr;

        CSTAT(tunnelsend);

        if (!t)
        {
                LOG(0, 0, t, "tunnelsend called with 0 as tunnel id\n");
                STAT(tunnel_tx_errors);
                return;
        }

        if (!tunnel[t].ip)
        {
                LOG(1, 0, t, "Error sending data out tunnel: no remote endpoint (tunnel not set up)\n");
                STAT(tunnel_tx_errors);
                return;
        }

        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
        *(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) *(uint16_t *) (buf + ((*buf & 0x40) ? 10 : 8)) = htons(tunnel[t].nr);

        // If this is a control message, deal with retries
        if (*buf & 0x80)
        {
                tunnel[t].last = time_now; // control message sent
                tunnel[t].retry = backoff(tunnel[t].try); // when to resend
                if (tunnel[t].try)
                {
                        STAT(tunnel_retries);
                        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, 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);
        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);
}

// adjust tcp mss to avoid fragmentation (called only for tcp packets with syn set)
void adjust_tcp_mss(sessionidt s, tunnelidt t, uint8_t *buf, int len, uint8_t *tcp)
{
        int d = (tcp[12] >> 4) * 4;
        uint8_t *mss = 0;
        uint8_t *opts;
        uint8_t *data;
        uint16_t orig;
        uint32_t sum;

        if ((tcp[13] & 0x3f) & ~(TCP_FLAG_SYN|TCP_FLAG_ACK)) // only want SYN and SYN,ACK
                return;

        if (tcp + d > buf + len) // short?
                return;

        opts = tcp + 20;
        data = tcp + d;

        while (opts < data)
        {
                if (*opts == 2 && opts[1] == 4) // mss option (2), length 4
                {
                        mss = opts + 2;
                        if (mss + 2 > data) return; // short?
                        break;
                }

                if (*opts == 0) return; // end of options
                if (*opts == 1 || !opts[1]) // no op (one byte), or no length (prevent loop)
                        opts++;
                else
                        opts += opts[1]; // skip over option
        }

        if (!mss) return; // not found
        orig = ntohs(*(uint16_t *) mss);

        if (orig <= MSS) return; // mss OK

        LOG(5, s, t, "TCP: %s:%u -> %s:%u SYN%s: adjusted mss from %u to %u\n",
                fmtaddr(*(in_addr_t *) (buf + 12), 0), ntohs(*(uint16_t *) tcp),
                fmtaddr(*(in_addr_t *) (buf + 16), 1), ntohs(*(uint16_t *) (tcp + 2)),
                (tcp[13] & TCP_FLAG_ACK) ? ",ACK" : "", orig, MSS);

        // set mss
        *(int16_t *) mss = htons(MSS);

        // adjust checksum (see rfc1141)
        sum = orig + (~MSS & 0xffff);
        sum += ntohs(*(uint16_t *) (tcp + 16));
        sum = (sum & 0xffff) + (sum >> 16);
        *(uint16_t *) (tcp + 16) = htons(sum + (sum >> 16));
}

void processmpframe(sessionidt s, tunnelidt t, uint8_t *p, uint16_t l, uint8_t extra)
{
        uint16_t proto;
        if (extra) {
                // Skip the four extra bytes
                p += 4;
                l -= 4;
        }

        if (*p & 1)
        {
                proto = *p++;
                l--;
        }
        else
        {
                proto = ntohs(*(uint16_t *) p);
                p += 2;
                l -= 2;
        }
        if (proto == PPPIP)
        {
                if (session[s].die)
                {
                        LOG(4, s, t, "MPPP: Session %d is closing.  Don't process PPP packets\n", s);
                        return;              // closing session, PPP not processed
                }
                session[s].last_packet = session[s].last_data = time_now;
                processipin(s, t, p, l);
        }
        else if (proto == PPPIPV6 && config->ipv6_prefix.s6_addr[0])
        {
                if (session[s].die)
                {
                        LOG(4, s, t, "MPPP: Session %d is closing.  Don't process PPP packets\n", s);
                        return;              // closing session, PPP not processed
                }

                session[s].last_packet = session[s].last_data = time_now;
                processipv6in(s, t, p, l);
        }
        else if (proto == PPPIPCP)
        {
                session[s].last_packet = session[s].last_data = time_now;
                processipcp(s, t, p, l);
        }
        else if (proto == PPPCCP)
        {
                session[s].last_packet = session[s].last_data = time_now;
                processccp(s, t, p, l);
        }
        else
        {
                LOG(2, s, t, "MPPP: Unsupported MP protocol 0x%04X received\n",proto);
        }
}

static void update_session_out_stat(sessionidt s, sessiont *sp, int len)
{
        increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
        sp->cout_delta += len;
        sp->pout++;
        sp->last_data = time_now;

        sess_local[s].cout += len;      // To send to master..
        sess_local[s].pout++;
}

// process outgoing (to tunnel) IP
//
static void processipout(uint8_t *buf, int len)
{
        sessionidt s;
        sessiont *sp;
        tunnelidt t;
        in_addr_t ip;

        uint8_t *data = buf;    // Keep a copy of the originals.
        int size = len;

        uint8_t fragbuf[MAXETHER + 20];

        CSTAT(processipout);

        if (len < MIN_IP_SIZE)
        {
                LOG(1, 0, 0, "Short IP, %d bytes\n", len);
                STAT(tun_rx_errors);
                return;
        }
        if (len >= MAXETHER)
        {
                LOG(1, 0, 0, "Oversize IP packet %d bytes\n", len);
                STAT(tun_rx_errors);
                return;
        }

        // Skip the tun header
        buf += 4;
        len -= 4;

        // Got an IP header now
        if (*(uint8_t *)(buf) >> 4 != 4)
        {
                LOG(1, 0, 0, "IP: Don't understand anything except IPv4\n");
                return;
        }

        ip = *(uint32_t *)(buf + 16);
        if (!(s = sessionbyip(ip)))
        {
                // Is this a packet for a session that doesn't exist?
                static int rate = 0;    // Number of ICMP packets we've sent this second.
                static int last = 0;    // Last time we reset the ICMP packet counter 'rate'.

                if (last != time_now)
                {
                        last = time_now;
                        rate = 0;
                }

                if (rate++ < config->icmp_rate) // Only send a max of icmp_rate per second.
                {
                        LOG(4, 0, 0, "IP: Sending ICMP host unreachable to %s\n", fmtaddr(*(in_addr_t *)(buf + 12), 0));
                        host_unreachable(*(in_addr_t *)(buf + 12), *(uint16_t *)(buf + 4),
                                config->bind_address ? config->bind_address : my_address, buf, len);
                }
                return;
        }

        t = session[s].tunnel;
        if (len > session[s].mru || (session[s].mrru && len > session[s].mrru))
        {
                LOG(3, s, t, "Packet size more than session MRU\n");
                return;
        }

        sp = &session[s];

        // DoS prevention: enforce a maximum number of packets per 0.1s for a session
        if (config->max_packets > 0)
        {
                if (sess_local[s].last_packet_out == TIME)
                {
                        int max = config->max_packets;

                        // All packets for throttled sessions are handled by the
                        // master, so further limit by using the throttle rate.
                        // A bit of a kludge, since throttle rate is in kbps,
                        // but should still be generous given our average DSL
                        // packet size is 200 bytes: a limit of 28kbps equates
                        // to around 180 packets per second.
                        if (!config->cluster_iam_master && sp->throttle_out && sp->throttle_out < max)
                                max = sp->throttle_out;

                        if (++sess_local[s].packets_out > max)
                        {
                                sess_local[s].packets_dropped++;
                                return;
                        }
                }
                else
                {
                        if (sess_local[s].packets_dropped)
                        {
                                INC_STAT(tun_rx_dropped, sess_local[s].packets_dropped);
                                LOG(3, s, t, "Dropped %u/%u packets to %s for %suser %s\n",
                                        sess_local[s].packets_dropped, sess_local[s].packets_out,
                                        fmtaddr(ip, 0), sp->throttle_out ? "throttled " : "",
                                        sp->user);
                        }

                        sess_local[s].last_packet_out = TIME;
                        sess_local[s].packets_out = 1;
                        sess_local[s].packets_dropped = 0;
                }
        }

        // run access-list if any
        if (session[s].filter_out && !ip_filter(buf, len, session[s].filter_out - 1))
                return;

        // adjust MSS on SYN and SYN,ACK packets with options
        if ((ntohs(*(uint16_t *) (buf + 6)) & 0x1fff) == 0 && buf[9] == IPPROTO_TCP) // first tcp fragment
        {
                int ihl = (buf[0] & 0xf) * 4; // length of IP header
                if (len >= ihl + 20 && (buf[ihl + 13] & TCP_FLAG_SYN) && ((buf[ihl + 12] >> 4) > 5))
                        adjust_tcp_mss(s, t, buf, len, buf + ihl);
        }

        if (sp->tbf_out)
        {
                // Are we throttling this session?
                if (config->cluster_iam_master)
                        tbf_queue_packet(sp->tbf_out, data, size);
                else
                        master_throttle_packet(sp->tbf_out, data, size);
                return;
        }

        if (sp->walled_garden && !config->cluster_iam_master)
        {
                // We are walled-gardening this
                master_garden_packet(s, data, size);
                return;
        }

        // Add on L2TP header
        {
                bundleidt bid = 0;
                if(session[s].bundle != 0 && bundle[session[s].bundle].num_of_links > 1)
                {
                        bid = session[s].bundle;
                        s = bundle[bid].members[bundle[bid].current_ses = ++bundle[bid].current_ses % bundle[bid].num_of_links];
                        t = session[s].tunnel;
                        sp = &session[s];
                        LOG(4, s, t, "MPPP: (1)Session number becomes: %d\n", s);
                        if(len > MINFRAGLEN)
                        {
                                // Partition the packet to "bundle[b].num_of_links" fragments
                                bundlet *b = &bundle[bid];
                                uint32_t num_of_links = b->num_of_links;
                                uint32_t fraglen = len / num_of_links;
                                fraglen = (fraglen > session[s].mru ? session[s].mru : fraglen);
                                uint32_t last_fraglen = fraglen + len % num_of_links;
                                last_fraglen = (last_fraglen > session[s].mru ? len % num_of_links : last_fraglen);
                                uint32_t remain = len;

                                // send the first packet
                                uint8_t *p = makeppp(fragbuf, sizeof(fragbuf), buf, fraglen, s, t, PPPIP, 0, bid, MP_BEGIN);
                                if (!p) return;
                                tunnelsend(fragbuf, fraglen + (p-fragbuf), t); // send it...
                                // statistics
                                update_session_out_stat(s, sp, fraglen);
                                remain -= fraglen;
                                while (remain > last_fraglen)
                                { 
                                        s = b->members[b->current_ses = ++b->current_ses % num_of_links];
                                        t = session[s].tunnel;
                                        sp = &session[s];
                                        LOG(4, s, t, "MPPP: (2)Session number becomes: %d\n", s);
                                        p = makeppp(fragbuf, sizeof(fragbuf), buf+(len - remain), fraglen, s, t, PPPIP, 0, bid, 0);
                                        if (!p) return;
                                        tunnelsend(fragbuf, fraglen + (p-fragbuf), t); // send it...
                                        update_session_out_stat(s, sp, fraglen);
                                        remain -= fraglen;
                                }
                                // send the last fragment
                                s = b->members[b->current_ses = ++b->current_ses % num_of_links];
                                t = session[s].tunnel;
                                sp = &session[s];
                                LOG(4, s, t, "MPPP: (2)Session number becomes: %d\n", s);
                                p = makeppp(fragbuf, sizeof(fragbuf), buf+(len - remain), remain, s, t, PPPIP, 0, bid, MP_END);
                                if (!p) return;
                                tunnelsend(fragbuf, remain + (p-fragbuf), t); // send it...
                                update_session_out_stat(s, sp, remain);
                                if (remain != last_fraglen)
                                        LOG(3, s, t, "PROCESSIPOUT ERROR REMAIN != LAST_FRAGLEN, %d != %d\n", remain, last_fraglen);
                        }
                        else {
                                // Send it as one frame
                                uint8_t *p = makeppp(fragbuf, sizeof(fragbuf), buf, len, s, t, PPPIP, 0, bid, MP_BOTH_BITS);
                                if (!p) return;
                                tunnelsend(fragbuf, len + (p-fragbuf), t); // send it...
                                LOG(4, s, t, "MPPP: packet sent as one frame\n");
                                update_session_out_stat(s, sp, len);
                        }
                }
                else
                {
                        uint8_t *p = makeppp(fragbuf, sizeof(fragbuf), buf, len, s, t, PPPIP, 0, 0, 0);
                        if (!p) return;
                        tunnelsend(fragbuf, len + (p-fragbuf), t); // send it...
                        update_session_out_stat(s, sp, len);
                }
        }

        // 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);

        udp_tx += len;
}

// process outgoing (to tunnel) IPv6
//
static void processipv6out(uint8_t * buf, int len)
{
        sessionidt s;
        sessiont *sp;
        tunnelidt t;
        in_addr_t ip;
        struct in6_addr ip6;

        uint8_t *data = buf;    // Keep a copy of the originals.
        int size = len;

        uint8_t b[MAXETHER + 20];

        CSTAT(processipv6out);

        if (len < MIN_IP_SIZE)
        {
                LOG(1, 0, 0, "Short IPv6, %d bytes\n", len);
                STAT(tunnel_tx_errors);
                return;
        }
        if (len >= MAXETHER)
        {
                LOG(1, 0, 0, "Oversize IPv6 packet %d bytes\n", len);
                STAT(tunnel_tx_errors);
                return;
        }

        // Skip the tun header
        buf += 4;
        len -= 4;

        // Got an IP header now
        if (*(uint8_t *)(buf) >> 4 != 6)
        {
                LOG(1, 0, 0, "IP: Don't understand anything except IPv6\n");
                return;
        }

        ip6 = *(struct in6_addr *)(buf+24);
        s = sessionbyipv6(ip6);

        if (s == 0)
        {
                ip = *(uint32_t *)(buf + 32);
                s = sessionbyip(ip);
        }
        
        if (s == 0)
        {
                // Is this a packet for a session that doesn't exist?
                static int rate = 0;    // Number of ICMP packets we've sent this second.
                static int last = 0;    // Last time we reset the ICMP packet counter 'rate'.

                if (last != time_now)
                {
                        last = time_now;
                        rate = 0;
                }

                if (rate++ < config->icmp_rate) // Only send a max of icmp_rate per second.
                {
                        // FIXME: Should send icmp6 host unreachable
                }
                return;
        }
        if (session[s].bundle && bundle[session[s].bundle].num_of_links > 1)
        {
                bundleidt bid = session[s].bundle;
                s = bundle[bid].members[bundle[bid].current_ses = ++bundle[bid].current_ses % bundle[bid].num_of_links];
                LOG(3, s, session[s].tunnel, "MPPP: Session number becomes: %u\n", s);
        }
        t = session[s].tunnel;
        sp = &session[s];
        sp->last_data = time_now;

        // FIXME: add DoS prevention/filters?

        if (sp->tbf_out)
        {
                // Are we throttling this session?
                if (config->cluster_iam_master)
                        tbf_queue_packet(sp->tbf_out, data, size);
                else
                        master_throttle_packet(sp->tbf_out, data, size);
                return;
        }
        else if (sp->walled_garden && !config->cluster_iam_master)
        {
                // We are walled-gardening this
                master_garden_packet(s, data, size);
                return;
        }

        LOG(5, s, t, "Ethernet -> Tunnel (%d bytes)\n", len);

        // Add on L2TP header
        {
                uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIPV6, 0, 0, 0);
                if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
        }

        // Snooping this session, send it to intercept box
        if (sp->snoop_ip && sp->snoop_port)
                snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);

        increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
        sp->cout_delta += len;
        sp->pout++;
        udp_tx += len;

        sess_local[s].cout += len;      // To send to master..
        sess_local[s].pout++;
}

//
// Helper routine for the TBF filters.
// Used to send queued data in to the user!
//
static void send_ipout(sessionidt s, uint8_t *buf, int len)
{
        sessiont *sp;
        tunnelidt t;
        in_addr_t ip;

        uint8_t b[MAXETHER + 20];

        if (len < 0 || len > MAXETHER)
        {
                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
        {
                uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIP, 0, 0, 0);
                if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
        }

        // Snooping this session.
        if (sp->snoop_ip && sp->snoop_port)
                snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);

        increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
        sp->cout_delta += len;
        sp->pout++;
        udp_tx += len;

        sess_local[s].cout += len;      // To send to master..
        sess_local[s].pout++;
}

// add an AVP (16 bit)
static void control16(controlt * c, uint16_t avp, uint16_t val, uint8_t m)
{
        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)
static void control32(controlt * c, uint16_t avp, uint32_t val, uint8_t m)
{
        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 (string)
static void controls(controlt * c, uint16_t avp, char *val, uint8_t m)
{
        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
static void controlb(controlt * c, uint16_t avp, uint8_t *val, unsigned int len, uint8_t m)
{
        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
static controlt *controlnew(uint16_t mtype)
{
        controlt *c;
        if (!controlfree)
                c = malloc(sizeof(controlt));
        else
        {
                c = controlfree;
                controlfree = c->next;
        }
        assert(c);
        c->next = 0;
        *(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
// (ZLB send).
static void controlnull(tunnelidt t)
{
        uint8_t buf[12];
        if (tunnel[t].controlc) // Messages queued; They will carry the ack.
                return;

        *(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
static void controladd(controlt *c, sessionidt far, tunnelidt t)
{
        *(uint16_t *) (c->buf + 2) = htons(c->length); // length
        *(uint16_t *) (c->buf + 4) = htons(tunnel[t].far); // tunnel
        *(uint16_t *) (c->buf + 6) = htons(far); // session
        *(uint16_t *) (c->buf + 8) = htons(tunnel[t].ns); // sequence
        tunnel[t].ns++;              // advance sequence
        // link in message in to queue
        if (tunnel[t].controlc)
                tunnel[t].controle->next = c;
        else
                tunnel[t].controls = c;

        tunnel[t].controle = c;
        tunnel[t].controlc++;

        // send now if space in window
        if (tunnel[t].controlc <= tunnel[t].window)
        {
                tunnel[t].try = 0;      // first send
                tunnelsend(c->buf, c->length, t);
        }
}

//
// 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 const *reason, int cdn_result, int cdn_error, int term_cause)
{
        int walled_garden = session[s].walled_garden;
        bundleidt b = session[s].bundle;
        //delete routes only for last session in bundle (in case of MPPP)
        int del_routes = !b || (bundle[b].num_of_links == 1);

        CSTAT(sessionshutdown);

        if (!session[s].opened)
        {
                LOG(3, s, session[s].tunnel, "Called sessionshutdown on an unopened session.\n");
                return;                   // not a live session
        }

        if (!session[s].die)
        {
                struct param_kill_session data = { &tunnel[session[s].tunnel], &session[s] };
                LOG(2, s, session[s].tunnel, "Shutting down session %u: %s\n", s, reason);
                run_plugins(PLUGIN_KILL_SESSION, &data);
                session[s].die = TIME + 150; // Clean up in 15 seconds
        }

        if (session[s].ip && !walled_garden && !session[s].die)
        {
                // RADIUS Stop message
                uint16_t r = radiusnew(s);
                if (r)
                {
                        // stop, if not already trying
                        if (radius[r].state != RADIUSSTOP)
                        {
                                radius[r].term_cause = term_cause;
                                radius[r].term_msg = reason;
                                radiussend(r, RADIUSSTOP);
                        }
                }
                else
                        LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");

                // Save counters to dump to accounting file
                if (*config->accounting_dir && shut_acct_n < sizeof(shut_acct) / sizeof(*shut_acct))
                        memcpy(&shut_acct[shut_acct_n++], &session[s], sizeof(session[s]));
        }

        if (session[s].ip)
        {                          // IP allocated, clear and unroute
                int r;
                int routed = 0;
                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++;

                        if (del_routes) 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 && del_routes) routeset(s, session[s].ip, 0, 0, 0);
                        session[s].ip = 0;
                }
                else
                        free_ip_address(s);

                // unroute IPv6, if setup
                if (session[s].ppp.ipv6cp == Opened && session[s].ipv6prefixlen && del_routes)
                        route6set(s, session[s].ipv6route, session[s].ipv6prefixlen, 0);
                
                if (b) 
                {
                        // This session was part of a bundle
                        bundle[b].num_of_links--;
                        LOG(3, s, 0, "MPPP: Dropping member link: %d from bundle %d\n",s,b);
                        if(bundle[b].num_of_links == 0) 
                        {
                                bundleclear(b);
                                LOG(3, s, 0, "MPPP: Kill bundle: %d (No remaing member links)\n",b);
                        }
                        else 
                        {
                                // Adjust the members array to accomodate the new change
                                uint8_t mem_num = 0;
                                // It should be here num_of_links instead of num_of_links-1 (previous instruction "num_of_links--")
                                if(bundle[b].members[bundle[b].num_of_links] != s) 
                                {
                                        uint8_t ml;
                                        for(ml = 0; ml<bundle[b].num_of_links; ml++)
                                                if(bundle[b].members[ml] == s)
                                                {
                                                        mem_num = ml;
                                                        break;
                                                }
                                        bundle[b].members[mem_num] = bundle[b].members[bundle[b].num_of_links];
                                        LOG(3, s, 0, "MPPP: Adjusted member links array\n");
                                }
                        }
                        cluster_send_bundle(b);
                }
        }

        if (session[s].throttle_in || session[s].throttle_out) // Unthrottle if throttled.
                throttle_session(s, 0, 0);

        if (cdn_result)
        {                            // Send CDN
                controlt *c = controlnew(14); // sending CDN
                if (cdn_error)
                {
                        uint8_t buf[4];
                        *(uint16_t *) buf     = htons(cdn_result);
                        *(uint16_t *) (buf+2) = htons(cdn_error);
                        controlb(c, 1, buf, 4, 1);
                }
                else
                        control16(c, 1, cdn_result, 1);

                control16(c, 14, s, 1);   // assigned session (our end)
                controladd(c, session[s].far, session[s].tunnel); // send the message
        }

        // update filter refcounts
        if (session[s].filter_in) ip_filters[session[s].filter_in - 1].used--;
        if (session[s].filter_out) ip_filters[session[s].filter_out - 1].used--;

        // clear PPP state
        memset(&session[s].ppp, 0, sizeof(session[s].ppp));
        sess_local[s].lcp.restart = 0;
        sess_local[s].ipcp.restart = 0;
        sess_local[s].ipv6cp.restart = 0;
        sess_local[s].ccp.restart = 0;

        cluster_send_session(s);
}

void sendipcp(sessionidt s, tunnelidt t)
{
        uint8_t buf[MAXETHER];
        uint8_t *q;

        CSTAT(sendipcp);
        LOG(3, s, t, "IPCP: send ConfigReq\n");

        if (!session[s].unique_id)
        {
                if (!++last_id) ++last_id; // skip zero
                session[s].unique_id = last_id;
        }

        q = makeppp(buf, sizeof(buf), 0, 0, s, t, PPPIPCP, 0, 0, 0);
        if (!q) return;

        *q = ConfigReq;
        q[1] = session[s].unique_id & 0xf;      // ID, dont care, we only send one type of request
        *(uint16_t *) (q + 2) = htons(10);      // packet length
        q[4] = 3;                               // ip address option
        q[5] = 6;                               // option length
        *(in_addr_t *) (q + 6) = config->peer_address ? config->peer_address :
                                 config->bind_address ? config->bind_address :
                                 my_address; // send my IP

        tunnelsend(buf, 10 + (q - buf), t); // send it
        restart_timer(s, ipcp);
}

void sendipv6cp(sessionidt s, tunnelidt t)
{
        uint8_t buf[MAXETHER];
        uint8_t *q;

        CSTAT(sendipv6cp);
        LOG(3, s, t, "IPV6CP: send ConfigReq\n");

        q = makeppp(buf, sizeof(buf), 0, 0, s, t, PPPIPV6CP, 0, 0, 0);
        if (!q) return;

        *q = ConfigReq;
        q[1] = session[s].unique_id & 0xf;      // ID, don't care, we
                                                // only send one type
                                                // of request
        *(uint16_t *) (q + 2) = htons(14);
        q[4] = 1;                               // interface identifier option
        q[5] = 10;                              // option length
        *(uint32_t *) (q + 6) = 0;              // We'll be prefix::1
        *(uint32_t *) (q + 10) = 0;
        q[13] = 1;

        tunnelsend(buf, 14 + (q - buf), t);     // send it
        restart_timer(s, ipv6cp);
}

static void sessionclear(sessionidt s)
{
        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)
{
        CSTAT(sessionkill);

        if (!session[s].opened) // not alive
                return;

        if (session[s].next)
        {
                LOG(0, s, session[s].tunnel, "Tried to kill a session with next pointer set (%u)\n", session[s].next);
                return;
        }

        if (!session[s].die)
                sessionshutdown(s, reason, CDN_ADMIN_DISC, TERM_ADMIN_RESET);  // close radius/routes, etc.

        if (sess_local[s].radius)
                radiusclear(sess_local[s].radius, s); // cant send clean accounting data, session is killed

        LOG(2, s, session[s].tunnel, "Kill session %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;
}

static void bundleclear(bundleidt b)
{
        if (!b) return;
        memset(&bundle[b], 0, sizeof(bundle[b]));
        bundle[b].state = BUNDLEFREE;
}

// kill a tunnel now
static void tunnelkill(tunnelidt t, char *reason)
{
        sessionidt s;
        controlt *c;

        CSTAT(tunnelkill);

        tunnel[t].state = TUNNELDIE;

        // free control messages
        while ((c = tunnel[t].controls))
        {
                controlt * n = c->next;
                tunnel[t].controls = n;
                tunnel[t].controlc--;
                c->next = controlfree;
                controlfree = c;
        }
        // kill sessions
        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 %u: %s\n", t, reason);
        cli_tunnel_actions[t].action = 0;
        cluster_send_tunnel(t);
}

// shut down a tunnel cleanly
static void tunnelshutdown(tunnelidt t, char *reason, int result, int error, char *msg)
{
        sessionidt s;

        CSTAT(tunnelshutdown);

        if (!tunnel[t].last || !tunnel[t].far || tunnel[t].state == TUNNELFREE)
        {
                // never set up, can immediately kill
                tunnelkill(t, reason);
                return;
        }
        LOG(1, 0, t, "Shutting down tunnel %u (%s)\n", t, reason);

        // close session
        for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                if (session[s].tunnel == t)
                        sessionshutdown(s, reason, CDN_NONE, TERM_ADMIN_RESET);

        tunnel[t].state = TUNNELDIE;
        tunnel[t].die = TIME + 700; // Clean up in 70 seconds
        cluster_send_tunnel(t);
        // TBA - should we wait for sessions to stop?
        if (result) 
        {
                controlt *c = controlnew(4);    // sending StopCCN
                if (error)
                {
                        uint8_t buf[64];
                        int l = 4;
                        *(uint16_t *) buf     = htons(result);
                        *(uint16_t *) (buf+2) = htons(error);
                        if (msg)
                        {
                                int m = strlen(msg);
                                if (m + 4 > sizeof(buf))
                                    m = sizeof(buf) - 4;

                                memcpy(buf+4, msg, m);
                                l += m;
                        }

                        controlb(c, 1, buf, l, 1);
                }
                else
                        control16(c, 1, result, 1);

                control16(c, 9, t, 1);          // assigned tunnel (our end)
                controladd(c, 0, t);            // send the message
        }
}

// read and process packet on tunnel (UDP)
void processudp(uint8_t *buf, int len, struct sockaddr_in *addr)
{
        uint8_t *chapresponse = NULL;
        uint16_t l = len, t = 0, s = 0, ns = 0, nr = 0;
        uint8_t *p = buf + 2;


        CSTAT(processudp);

        udp_rx += len;
        udp_rx_pkt++;
        LOG_HEX(5, "UDP Data", buf, len);
        STAT(tunnel_rx_packets);
        INC_STAT(tunnel_rx_bytes, len);
        if (len < 6)
        {
                LOG(1, 0, 0, "Short UDP, %d bytes\n", len);
                STAT(tunnel_rx_errors);
                return;
        }
        if ((buf[1] & 0x0F) != 2)
        {
                LOG(1, 0, 0, "Bad L2TP ver %d\n", buf[1] & 0x0F);
                STAT(tunnel_rx_errors);
                return;
        }
        if (*buf & 0x40)
        {                          // length
                l = ntohs(*(uint16_t *) p);
                p += 2;
        }
        t = ntohs(*(uint16_t *) p);
        p += 2;
        s = ntohs(*(uint16_t *) p);
        p += 2;
        if (s >= MAXSESSION)
        {
                LOG(1, s, t, "Received UDP packet with invalid session ID\n");
                STAT(tunnel_rx_errors);
                return;
        }
        if (t >= MAXTUNNEL)
        {
                LOG(1, s, t, "Received UDP packet with invalid tunnel ID\n");
                STAT(tunnel_rx_errors);
                return;
        }
        if (*buf & 0x08)
        {                          // ns/nr
                ns = ntohs(*(uint16_t *) p);
                p += 2;
                nr = ntohs(*(uint16_t *) p);
                p += 2;
        }
        if (*buf & 0x02)
        {                          // offset
                uint16_t o = ntohs(*(uint16_t *) p);
                p += o + 2;
        }
        if ((p - buf) > l)
        {
                LOG(1, s, t, "Bad length %d>%d\n", (int) (p - buf), l);
                STAT(tunnel_rx_errors);
                return;
        }
        l -= (p - buf);

        // used to time out old tunnels
        if (t && tunnel[t].state == TUNNELOPEN)
                tunnel[t].lastrec = time_now;

        if (*buf & 0x80)
        {                          // control
                uint16_t message = 0xFFFF;      // message type
                uint8_t fatal = 0;
                uint8_t mandatory = 0;
                uint16_t asession = 0;          // assigned session
                uint32_t amagic = 0;            // magic number
                uint8_t aflags = 0;             // flags from last LCF
                uint16_t version = 0x0100;      // protocol version (we handle 0.0 as well and send that back just in case)
                char called[MAXTEL] = "";       // called number
                char calling[MAXTEL] = "";      // calling number

                if (!config->cluster_iam_master)
                {
                        master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port);
                        return;
                }

                // control messages must have bits 0x80|0x40|0x08
                // (type, length and sequence) set, and bits 0x02|0x01
                // (offset and priority) clear
                if ((*buf & 0xCB) != 0xC8)
                {
                        LOG(1, s, t, "Bad control header %02X\n", *buf);
                        STAT(tunnel_rx_errors);
                        return;
                }

                // 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 %u l-nr %u r-ns %u r-nr %u\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, 0, 0, "No more tunnels\n");
                                STAT(tunnel_overflow);
                                return;
                        }
                        tunnelclear(t);
                        tunnel[t].ip = ntohl(*(in_addr_t *) & addr->sin_addr);
                        tunnel[t].port = ntohs(addr->sin_port);
                        tunnel[t].window = 4; // default window
                        STAT(tunnel_created);
                        LOG(1, 0, t, "   New tunnel from %s:%u ID %u\n",
                                fmtaddr(htonl(tunnel[t].ip), 0), tunnel[t].port, t);
                }

                        // If the 'ns' just received is not the 'nr' we're
                        // expecting, just send an ack and drop it.
                        //
                        // if 'ns' is less, then we got a retransmitted packet.
                        // if 'ns' is greater than missed a packet. Either way
                        // we should ignore it.
                if (ns != tunnel[t].nr)
                {
                        // is this the sequence we were expecting?
                        STAT(tunnel_rx_errors);
                        LOG(1, 0, t, "   Out of sequence tunnel %u, (%u is not the expected %u)\n",
                                t, ns, tunnel[t].nr);

                        if (l)  // Is this not a ZLB?
                                controlnull(t);
                        return;
                }

                // check sequence of this message
                {
                        int skip = tunnel[t].window; // track how many in-window packets are still in queue
                                // some to clear maybe?
                        while (tunnel[t].controlc > 0 && (((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
                        }

                        // receiver advance (do here so quoted correctly in any sends below)
                        if (l) tunnel[t].nr = (ns + 1);
                        if (skip < 0) skip = 0;
                        if (skip < tunnel[t].controlc)
                        {
                                // some control packets can now be sent that were previous stuck out of window
                                int tosend = tunnel[t].window - skip;
                                controlt *c = tunnel[t].controls;
                                while (c && skip)
                                {
                                        c = c->next;
                                        skip--;
                                }
                                while (c && tosend)
                                {
                                        tunnel[t].try = 0; // first send
                                        tunnelsend(c->buf, c->length, t);
                                        c = c->next;
                                        tosend--;
                                }
                        }
                        if (!tunnel[t].controlc)
                                tunnel[t].retry = 0; // caught up
                }
                if (l)
                {                     // if not a null message
                        int result = 0;
                        int error = 0;
                        char *msg = 0;

                        // Default disconnect cause/message on receipt of CDN.  Set to
                        // more specific value from attribute 1 (result code) or 46
                        // (disconnect cause) if present below.
                        int disc_cause_set = 0;
                        int disc_cause = TERM_NAS_REQUEST;
                        char const *disc_reason = "Closed (Received CDN).";

                        // process AVPs
                        while (l && !(fatal & 0x80)) // 0x80 = mandatory AVP
                        {
                                uint16_t n = (ntohs(*(uint16_t *) p) & 0x3FF);
                                uint8_t *b = p;
                                uint8_t flags = *p;
                                uint16_t mtype;

                                if (n > l)
                                {
                                        LOG(1, s, t, "Invalid length in AVP\n");
                                        STAT(tunnel_rx_errors);
                                        return;
                                }
                                p += n;       // next
                                l -= n;
                                if (flags & 0x3C) // reserved bits, should be clear
                                {
                                        LOG(1, s, t, "Unrecognised AVP flags %02X\n", *b);
                                        fatal = flags;
                                        result = 2; // general error
                                        error = 3; // reserved field non-zero
                                        msg = 0;
                                        continue; // next
                                }
                                b += 2;
                                if (*(uint16_t *) (b))
                                {
                                        LOG(2, s, t, "Unknown AVP vendor %u\n", ntohs(*(uint16_t *) (b)));
                                        fatal = flags;
                                        result = 2; // general error
                                        error = 6; // generic vendor-specific error
                                        msg = "unsupported vendor-specific";
                                        continue; // next
                                }
                                b += 2;
                                mtype = ntohs(*(uint16_t *) (b));
                                b += 2;
                                n -= 6;

                                if (flags & 0x40)
                                {
                                        uint16_t orig_len;

                                        // handle hidden AVPs
                                        if (!*config->l2tp_secret)
                                        {
                                                LOG(1, s, t, "Hidden AVP requested, but no L2TP secret.\n");
                                                fatal = flags;
                                                result = 2; // general error
                                                error = 6; // generic vendor-specific error
                                                msg = "secret not specified";
                                                continue;
                                        }
                                        if (!session[s].random_vector_length)
                                        {
                                                LOG(1, s, t, "Hidden AVP requested, but no random vector.\n");
                                                fatal = flags;
                                                result = 2; // general error
                                                error = 6; // generic
                                                msg = "no random vector";
                                                continue;
                                        }
                                        if (n < 8)
                                        {
                                                LOG(2, s, t, "Short hidden AVP.\n");
                                                fatal = flags;
                                                result = 2; // general error
                                                error = 2; // length is wrong
                                                msg = 0;
                                                continue;
                                        }

                                        // Unhide the AVP
                                        unhide_value(b, n, mtype, session[s].random_vector, session[s].random_vector_length);

                                        orig_len = ntohs(*(uint16_t *) b);
                                        if (orig_len > n + 2)
                                        {
                                                LOG(1, s, t, "Original length %d too long in hidden AVP of length %d; wrong secret?\n",
                                                    orig_len, n);

                                                fatal = flags;
                                                result = 2; // general error
                                                error = 2; // length is wrong
                                                msg = 0;
                                                continue;
                                        }

                                        b += 2;
                                        n = orig_len;
                                }

                                LOG(4, s, t, "   AVP %u (%s) len %d%s%s\n", mtype, l2tp_avp_name(mtype), n,
                                        flags & 0x40 ? ", hidden" : "", flags & 0x80 ? ", mandatory" : "");

                                switch (mtype)
                                {
                                case 0:     // message type
                                        message = ntohs(*(uint16_t *) b);
                                        mandatory = flags & 0x80;
                                        LOG(4, s, t, "   Message type = %u (%s)\n", *b, l2tp_code(message));
                                        break;
                                case 1:     // result code
                                        {
                                                uint16_t rescode = ntohs(*(uint16_t *) b);
                                                char const *resdesc = "(unknown)";
                                                char const *errdesc = NULL;
                                                int cause = 0;

                                                if (message == 4)
                                                { /* StopCCN */
                                                        resdesc = l2tp_stopccn_result_code(rescode);
                                                        cause = TERM_LOST_SERVICE;
                                                }
                                                else if (message == 14)
                                                { /* CDN */
                                                        resdesc = l2tp_cdn_result_code(rescode);
                                                        if (rescode == 1)
                                                                cause = TERM_LOST_CARRIER;
                                                        else
                                                                cause = TERM_ADMIN_RESET;
                                                }

                                                LOG(4, s, t, "   Result Code %u: %s\n", rescode, resdesc);
                                                if (n >= 4)
                                                {
                                                        uint16_t errcode = ntohs(*(uint16_t *)(b + 2));
                                                        errdesc = l2tp_error_code(errcode);
                                                        LOG(4, s, t, "   Error Code %u: %s\n", errcode, errdesc);
                                                }
                                                if (n > 4)
                                                        LOG(4, s, t, "   Error String: %.*s\n", n-4, b+4);

                                                if (cause && disc_cause_set < mtype) // take cause from attrib 46 in preference
                                                {
                                                        disc_cause_set = mtype;
                                                        disc_reason = errdesc ? errdesc : resdesc;
                                                        disc_cause = cause;
                                                }

                                                break;
                                        }
                                        break;
                                case 2:     // protocol version
                                        {
                                                version = ntohs(*(uint16_t *) (b));
                                                LOG(4, s, t, "   Protocol version = %u\n", version);
                                                if (version && version != 0x0100)
                                                {   // allow 0.0 and 1.0
                                                        LOG(1, s, t, "   Bad protocol version %04X\n", version);
                                                        fatal = flags;
                                                        result = 5; // unspported protocol version
                                                        error = 0x0100; // supported version
                                                        msg = 0;
                                                        continue; // next
                                                }
                                        }
                                        break;
                                case 3:     // framing capabilities
                                        break;
                                case 4:     // bearer capabilities
                                        break;
                                case 5:         // tie breaker
                                        // We never open tunnels, so we don't care about tie breakers
                                        continue;
                                case 6:     // firmware revision
                                        break;
                                case 7:     // host name
                                        memset(tunnel[t].hostname, 0, sizeof(tunnel[t].hostname));
                                        memcpy(tunnel[t].hostname, b, (n < sizeof(tunnel[t].hostname)) ? n : sizeof(tunnel[t].hostname) - 1);
                                        LOG(4, s, t, "   Tunnel hostname = \"%s\"\n", tunnel[t].hostname);
                                        // TBA - to send to RADIUS
                                        break;
                                case 8:     // vendor name
                                        memset(tunnel[t].vendor, 0, sizeof(tunnel[t].vendor));
                                        memcpy(tunnel[t].vendor, b, (n < sizeof(tunnel[t].vendor)) ? n : sizeof(tunnel[t].vendor) - 1);
                                        LOG(4, s, t, "   Vendor name = \"%s\"\n", tunnel[t].vendor);
                                        break;
                                case 9:     // assigned tunnel
                                        tunnel[t].far = ntohs(*(uint16_t *) (b));
                                        LOG(4, s, t, "   Remote tunnel id = %u\n", tunnel[t].far);
                                        break;
                                case 10:    // rx window
                                        tunnel[t].window = ntohs(*(uint16_t *) (b));
                                        if (!tunnel[t].window)
                                                tunnel[t].window = 1; // window of 0 is silly
                                        LOG(4, s, t, "   rx window = %u\n", tunnel[t].window);
                                        break;
                                case 11:        // Challenge
                                        {
                                                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(*(uint16_t *) (b));
                                        LOG(4, s, t, "   assigned session = %u\n", asession);
                                        break;
                                case 15:    // call serial number
                                        LOG(4, s, t, "   call serial number = %u\n", ntohl(*(uint32_t *)b));
                                        break;
                                case 18:    // bearer type
                                        LOG(4, s, t, "   bearer type = %u\n", ntohl(*(uint32_t *)b));
                                        // TBA - for RADIUS
                                        break;
                                case 19:    // framing type
                                        LOG(4, s, t, "   framing type = %u\n", ntohl(*(uint32_t *)b));
                                        // TBA
                                        break;
                                case 21:    // called number
                                        memset(called, 0, sizeof(called));
                                        memcpy(called, b, (n < sizeof(called)) ? n : sizeof(called) - 1);
                                        LOG(4, s, t, "   Called <%s>\n", called);
                                        break;
                                case 22:    // calling number
                                        memset(calling, 0, sizeof(calling));
                                        memcpy(calling, b, (n < sizeof(calling)) ? n : sizeof(calling) - 1);
                                        LOG(4, s, t, "   Calling <%s>\n", calling);
                                        break;
                                case 23:    // subtype
                                        break;
                                case 24:    // tx connect speed
                                        if (n == 4)
                                        {
                                                session[s].tx_connect_speed = ntohl(*(uint32_t *)b);
                                        }
                                        else
                                        {
                                                // AS5300s send connect speed as a string
                                                char tmp[30];
                                                memset(tmp, 0, sizeof(tmp));
                                                memcpy(tmp, b, (n < sizeof(tmp)) ? n : sizeof(tmp) - 1);
                                                session[s].tx_connect_speed = atol(tmp);
                                        }
                                        LOG(4, 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(*(uint32_t *)b);
                                        }
                                        else
                                        {
                                                // AS5300s send connect speed as a string
                                                char tmp[30];
                                                memset(tmp, 0, sizeof(tmp));
                                                memcpy(tmp, b, (n < sizeof(tmp)) ? n : sizeof(tmp) - 1);
                                                session[s].rx_connect_speed = atol(tmp);
                                        }
                                        LOG(4, s, t, "   RX connect speed <%u>\n", session[s].rx_connect_speed);
                                        break;
                                case 25:    // Physical Channel ID
                                        {
                                                uint32_t tmp = ntohl(*(uint32_t *) b);
                                                LOG(4, s, t, "   Physical Channel ID <%X>\n", tmp);
                                                break;
                                        }
                                case 29:    // Proxy Authentication Type
                                        {
                                                uint16_t atype = ntohs(*(uint16_t *)b);
                                                LOG(4, s, t, "   Proxy Auth Type %u (%s)\n", atype, ppp_auth_type(atype));
                                                break;
                                        }
                                case 30:    // Proxy Authentication Name
                                        {
                                                char authname[64];
                                                memset(authname, 0, sizeof(authname));
                                                memcpy(authname, b, (n < sizeof(authname)) ? n : sizeof(authname) - 1);
                                                LOG(4, s, t, "   Proxy Auth Name (%s)\n",
                                                        authname);
                                                break;
                                        }
                                case 31:    // Proxy Authentication Challenge
                                        {
                                                LOG(4, s, t, "   Proxy Auth Challenge\n");
                                                break;
                                        }
                                case 32:    // Proxy Authentication ID
                                        {
                                                uint16_t authid = ntohs(*(uint16_t *)(b));
                                                LOG(4, s, t, "   Proxy Auth ID (%u)\n", authid);
                                                break;
                                        }
                                case 33:    // Proxy Authentication Response
                                        LOG(4, s, t, "   Proxy Auth Response\n");
                                        break;
                                case 27:    // last sent lcp
                                        {        // find magic number
                                                uint8_t *p = b, *e = p + n;
                                                while (p + 1 < e && p[1] && p + p[1] <= e)
                                                {
                                                        if (*p == 5 && p[1] == 6) // Magic-Number
                                                                amagic = ntohl(*(uint32_t *) (p + 2));
                                                        else if (*p == 7) // Protocol-Field-Compression
                                                                aflags |= SESSION_PFC;
                                                        else if (*p == 8) // Address-and-Control-Field-Compression
                                                                aflags |= SESSION_ACFC;
                                                        p += p[1];
                                                }
                                        }
                                        break;
                                case 28:    // last recv lcp confreq
                                        break;
                                case 26:    // Initial Received LCP CONFREQ
                                        break;
                                case 39:    // seq required - we control it as an LNS anyway...
                                        break;
                                case 36:    // Random Vector
                                        LOG(4, s, t, "   Random Vector received.  Enabled AVP Hiding.\n");
                                        memset(session[s].random_vector, 0, sizeof(session[s].random_vector));
                                        if (n > sizeof(session[s].random_vector))
                                                n = sizeof(session[s].random_vector);
                                        memcpy(session[s].random_vector, b, n);
                                        session[s].random_vector_length = n;
                                        break;
                                case 46:    // ppp disconnect cause
                                        if (n >= 5)
                                        {
                                                uint16_t code = ntohs(*(uint16_t *) b);
                                                uint16_t proto = ntohs(*(uint16_t *) (b + 2));
                                                uint8_t dir = *(b + 4);

                                                LOG(4, s, t, "   PPP disconnect cause "
                                                        "(code=%u, proto=%04X, dir=%u, msg=\"%.*s\")\n",
                                                        code, proto, dir, n - 5, b + 5);

                                                disc_cause_set = mtype;

                                                switch (code)
                                                {
                                                case 1: // admin disconnect
                                                        disc_cause = TERM_ADMIN_RESET;
                                                        disc_reason = "Administrative disconnect";
                                                        break;
                                                case 3: // lcp terminate
                                                        if (dir != 2) break; // 1=peer (LNS), 2=local (LAC)
                                                        disc_cause = TERM_USER_REQUEST;
                                                        disc_reason = "Normal disconnection";
                                                        break;
                                                case 4: // compulsory encryption unavailable
                                                        if (dir != 1) break; // 1=refused by peer, 2=local
                                                        disc_cause = TERM_USER_ERROR;
                                                        disc_reason = "Compulsory encryption refused";
                                                        break;
                                                case 5: // lcp: fsm timeout
                                                        disc_cause = TERM_PORT_ERROR;
                                                        disc_reason = "LCP: FSM timeout";
                                                        break;
                                                case 6: // lcp: no recognisable lcp packets received
                                                        disc_cause = TERM_PORT_ERROR;
                                                        disc_reason = "LCP: no recognisable LCP packets";
                                                        break;
                                                case 7: // lcp: magic-no error (possibly looped back)
                                                        disc_cause = TERM_PORT_ERROR;
                                                        disc_reason = "LCP: magic-no error (possible loop)";
                                                        break;
                                                case 8: // lcp: echo request timeout
                                                        disc_cause = TERM_PORT_ERROR;
                                                        disc_reason = "LCP: echo request timeout";
                                                        break;
                                                case 13: // auth: fsm timeout
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = "Authentication: FSM timeout";
                                                        break;
                                                case 15: // auth: unacceptable auth protocol
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = "Unacceptable authentication protocol";
                                                        break;
                                                case 16: // auth: authentication failed
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = "Authentication failed";
                                                        break;
                                                case 17: // ncp: fsm timeout
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = "NCP: FSM timeout";
                                                        break;
                                                case 18: // ncp: no ncps available
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = "NCP: no NCPs available";
                                                        break;
                                                case 19: // ncp: failure to converge on acceptable address
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = (dir == 1)
                                                                ? "NCP: too many Configure-Naks received from peer"
                                                                : "NCP: too many Configure-Naks sent to peer";
                                                        break;
                                                case 20: // ncp: user not permitted to use any address
                                                        disc_cause = TERM_SERVICE_UNAVAILABLE;
                                                        disc_reason = (dir == 1)
                                                                ? "NCP: local link address not acceptable to peer"
                                                                : "NCP: remote link address not acceptable";
                                                        break;
                                                }
                                        }
                                        break;
                                default:
                                        {
                                                static char e[] = "unknown AVP 0xXXXX";
                                                LOG(2, s, t, "   Unknown AVP type %u\n", mtype);
                                                fatal = flags;
                                                result = 2; // general error
                                                error = 8; // unknown mandatory AVP
                                                sprintf((msg = e) + 14, "%04x", mtype);
                                                continue; // next
                                        }
                                }
                        }
                        // process message
                        if (fatal & 0x80)
                                tunnelshutdown(t, "Invalid mandatory AVP", result, error, msg);
                        else
                                switch (message)
                                {
                                case 1:       // SCCRQ - Start Control Connection Request
                                        tunnel[t].state = TUNNELOPENING;
                                        if (main_quit != QUIT_SHUTDOWN)
                                        {
                                                controlt *c = controlnew(2); // sending SCCRP
                                                control16(c, 2, version, 1); // protocol version
                                                control32(c, 3, 3, 1); // framing
                                                controls(c, 7, hostname, 1); // host name
                                                if (chapresponse) controlb(c, 13, chapresponse, 16, 1); // Challenge response
                                                control16(c, 9, t, 1); // assigned tunnel
                                                controladd(c, 0, t); // send the resply
                                        }
                                        else
                                        {
                                                tunnelshutdown(t, "Shutting down", 6, 0, 0);
                                        }
                                        break;
                                case 2:       // SCCRP
                                        tunnel[t].state = TUNNELOPEN;
                                        tunnel[t].lastrec = time_now;
                                        break;
                                case 3:       // SCCN
                                        tunnel[t].state = TUNNELOPEN;
                                        tunnel[t].lastrec = time_now;
                                        controlnull(t); // ack
                                        break;
                                case 4:       // StopCCN
                                        controlnull(t); // ack
                                        tunnelshutdown(t, "Stopped", 0, 0, 0); // Shut down cleanly
                                        break;
                                case 6:       // HELLO
                                        controlnull(t); // simply ACK
                                        break;
                                case 7:       // OCRQ
                                        // TBA
                                        break;
                                case 8:       // OCRO
                                        // TBA
                                        break;
                                case 9:       // OCCN
                                        // TBA
                                        break;
                                case 10:      // ICRQ
                                        if (sessionfree && main_quit != QUIT_SHUTDOWN)
                                        {
                                                controlt *c = controlnew(11); // ICRP

                                                s = sessionfree;
                                                sessionfree = session[s].next;
                                                memset(&session[s], 0, sizeof(session[s]));

                                                if (s > config->cluster_highest_sessionid)
                                                        config->cluster_highest_sessionid = s;

                                                session[s].opened = time_now;
                                                session[s].tunnel = t;
                                                session[s].far = asession;
                                                session[s].last_packet = session[s].last_data = time_now;
                                                LOG(3, s, t, "New session (%u/%u)\n", tunnel[t].far, session[s].far);
                                                control16(c, 14, s, 1); // assigned session
                                                controladd(c, asession, t); // send the reply

                                                strncpy(session[s].called, called, sizeof(session[s].called) - 1);
                                                strncpy(session[s].calling, calling, sizeof(session[s].calling) - 1);

                                                session[s].ppp.phase = Establish;
                                                session[s].ppp.lcp = Starting;

                                                STAT(session_created);
                                                break;
                                        }

                                        {
                                                controlt *c = controlnew(14); // CDN
                                                if (!sessionfree)
                                                {
                                                        STAT(session_overflow);
                                                        LOG(1, 0, t, "No free sessions\n");
                                                        control16(c, 1, 4, 0); // temporary lack of resources
                                                }
                                                else
                                                        control16(c, 1, 2, 7); // shutting down, try another

                                                controladd(c, asession, t); // send the message
                                        }
                                        return;
                                case 11:      // ICRP
                                        // TBA
                                        break;
                                case 12:      // ICCN
                                        if (amagic == 0) amagic = time_now;
                                        session[s].magic = amagic; // set magic number
                                        session[s].flags = aflags; // set flags received
                                        session[s].mru = PPPoE_MRU; // default
                                        controlnull(t); // ack

                                        // start LCP
                                        sess_local[s].lcp_authtype = config->radius_authprefer;
                                        sess_local[s].ppp_mru = MRU;

                                        // Set multilink options before sending initial LCP packet
                                        sess_local[s].mp_mrru = 1614;
                                        sess_local[s].mp_epdis = config->bind_address ? config->bind_address : my_address;

                                        sendlcp(s, t);
                                        change_state(s, lcp, RequestSent);
                                        break;

                                case 14:      // CDN
                                        controlnull(t); // ack
                                        sessionshutdown(s, disc_reason, CDN_NONE, disc_cause);
                                        break;
                                case 0xFFFF:
                                        LOG(1, s, t, "Missing message type\n");
                                        break;
                                default:
                                        STAT(tunnel_rx_errors);
                                        if (mandatory)
                                                tunnelshutdown(t, "Unknown message type", 2, 6, "unknown message type");
                                        else
                                                LOG(1, s, t, "Unknown message type %u\n", message);
                                        break;
                                }
                        if (chapresponse) free(chapresponse);
                        cluster_send_tunnel(t);
                }
                else
                {
                        LOG(4, s, t, "   Got a ZLB ack\n");
                }
        }
        else
        {                          // data
                uint16_t proto;

                LOG_HEX(5, "Receive Tunnel Data", p, l);
                if (l > 2 && p[0] == 0xFF && p[1] == 0x03)
                {                     // HDLC address header, discard
                        p += 2;
                        l -= 2;
                }
                if (l < 2)
                {
                        LOG(1, s, t, "Short ppp length %d\n", l);
                        STAT(tunnel_rx_errors);
                        return;
                }
                if (*p & 1)
                {
                        proto = *p++;
                        l--;
                }
                else
                {
                        proto = 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 (proto == PPPPAP)
                {
                        session[s].last_packet = time_now;
                        if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processpap(s, t, p, l);
                }
                else if (proto == PPPCHAP)
                {
                        session[s].last_packet = time_now;
                        if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processchap(s, t, p, l);
                }
                else if (proto == PPPLCP)
                {
                        session[s].last_packet = time_now;
                        if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processlcp(s, t, p, l);
                }
                else if (proto == PPPIPCP)
                {
                        session[s].last_packet = time_now;
                        if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processipcp(s, t, p, l);
                }
                else if (proto == PPPIPV6CP && config->ipv6_prefix.s6_addr[0])
                {
                        session[s].last_packet = time_now;
                        if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processipv6cp(s, t, p, l);
                }
                else if (proto == PPPCCP)
                {
                        session[s].last_packet = time_now;
                        if (!config->cluster_iam_master) { master_forward_packet(buf, len, addr->sin_addr.s_addr, addr->sin_port); return; }
                        processccp(s, t, p, l);
                }
                else if (proto == PPPIP)
                {
                        if (session[s].die)
                        {
                                LOG(4, s, t, "Session %u is closing.  Don't process PPP packets\n", s);
                                return;              // closing session, PPP not processed
                        }

                        session[s].last_packet = session[s].last_data = 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(s, t, p, l);
                }
                else if (proto == PPPMP)
                {
                        if (session[s].die)
                        {
                                LOG(4, s, t, "Session %u is closing.  Don't process PPP packets\n", s);
                                return;              // closing session, PPP not processed
                        }

                        session[s].last_packet = session[s].last_data = 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;
                        }

                        processmpin(s, t, p, l);
                }
                else if (proto == PPPIPV6 && config->ipv6_prefix.s6_addr[0])
                {
                        if (session[s].die)
                        {
                                LOG(4, s, t, "Session %u is closing.  Don't process PPP packets\n", s);
                                return;              // closing session, PPP not processed
                        }

                        session[s].last_packet = session[s].last_data = 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;
                        }

                        processipv6in(s, t, p, l);
                }
                else if (session[s].ppp.lcp == Opened)
                {
                        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; }
                        protoreject(s, t, p, l, proto);
                }
                else
                {
                        LOG(2, s, t, "Unknown PPP protocol 0x%04X received in LCP %s state\n",
                                proto, ppp_state(session[s].ppp.lcp));
                }
        }
}

// read and process packet on tun
static void processtun(uint8_t * buf, int len)
{
        LOG_HEX(5, "Receive TUN Data", buf, len);
        STAT(tun_rx_packets);
        INC_STAT(tun_rx_bytes, len);

        CSTAT(processtun);

        eth_rx_pkt++;
        eth_rx += len;
        if (len < 22)
        {
                LOG(1, 0, 0, "Short tun packet %d bytes\n", len);
                STAT(tun_rx_errors);
                return;
        }

        if (*(uint16_t *) (buf + 2) == htons(PKTIP)) // IPv4
                processipout(buf, len);
        else if (*(uint16_t *) (buf + 2) == htons(PKTIPV6) // IPV6
            && config->ipv6_prefix.s6_addr[0])
                processipv6out(buf, len);

        // Else discard.
}

// Handle retries, timeouts.  Runs every 1/10th sec, want to ensure
// that we look at the whole of the tunnel, radius and session tables
// every second
static void regular_cleanups(double period)
{
        // Next tunnel, radius and session to check for actions on.
        static tunnelidt t = 0;
        static int r = 0;
        static sessionidt s = 0;

        int t_actions = 0;
        int r_actions = 0;
        int s_actions = 0;

        int t_slice;
        int r_slice;
        int s_slice;

        int i;
        int a;

        // divide up tables into slices based on the last run
        t_slice = config->cluster_highest_tunnelid  * period;
        r_slice = (MAXRADIUS - 1)                   * period;
        s_slice = config->cluster_highest_sessionid * period;

        if (t_slice < 1)
            t_slice = 1;
        else if (t_slice > config->cluster_highest_tunnelid)
            t_slice = config->cluster_highest_tunnelid;

        if (r_slice < 1)
            r_slice = 1;
        else if (r_slice > (MAXRADIUS - 1))
            r_slice = MAXRADIUS - 1;

        if (s_slice < 1)
            s_slice = 1;
        else if (s_slice > config->cluster_highest_sessionid)
            s_slice = config->cluster_highest_sessionid;

        LOG(4, 0, 0, "Begin regular cleanup (last %f seconds ago)\n", period);

        for (i = 0; i < t_slice; i++)
        {
                t++;
                if (t > config->cluster_highest_tunnelid)
                        t = 1;

                // check for expired tunnels
                if (tunnel[t].die && tunnel[t].die <= TIME)
                {
                        STAT(tunnel_timeout);
                        tunnelkill(t, "Expired");
                        t_actions++;
                        continue;
                }
                // check for message resend
                if (tunnel[t].retry && tunnel[t].controlc)
                {
                        // resend pending messages as timeout on reply
                        if (tunnel[t].retry <= TIME)
                        {
                                controlt *c = tunnel[t].controls;
                                uint16_t w = tunnel[t].window;
                                tunnel[t].try++; // another try
                                if (tunnel[t].try > 5)
                                        tunnelkill(t, "Timeout on control message"); // game over
                                else
                                        while (c && w--)
                                        {
                                                tunnelsend(c->buf, c->length, t);
                                                c = c->next;
                                        }

                                t_actions++;
                        }
                }
                // Send hello
                if (tunnel[t].state == TUNNELOPEN && !tunnel[t].controlc && (time_now - tunnel[t].lastrec) > 60)
                {
                        controlt *c = controlnew(6); // sending HELLO
                        controladd(c, 0, t); // send the message
                        LOG(3, 0, t, "Sending HELLO message\n");
                        t_actions++;
                }

                // Check for tunnel changes requested from the CLI
                if ((a = cli_tunnel_actions[t].action))
                {
                        cli_tunnel_actions[t].action = 0;
                        if (a & CLI_TUN_KILL)
                        {
                                LOG(2, 0, t, "Dropping tunnel by CLI\n");
                                tunnelshutdown(t, "Requested by administrator", 1, 0, 0);
                                t_actions++;
                        }
                }
        }

        for (i = 0; i < r_slice; i++)
        {
                r++;
                if (r >= MAXRADIUS)
                        r = 1;

                if (!radius[r].state)
                        continue;

                if (radius[r].retry <= TIME)
                {
                        radiusretry(r);
                        r_actions++;
                }
        }

        for (i = 0; i < s_slice; i++)
        {
                s++;
                if (s > config->cluster_highest_sessionid)
                        s = 1;

                if (!session[s].opened) // Session isn't in use
                        continue;

                // check for expired sessions
                if (session[s].die)
                {
                        if (session[s].die <= TIME)
                        {
                                sessionkill(s, "Expired");
                                s_actions++;
                        }
                        continue;
                }

                // PPP timeouts
                if (sess_local[s].lcp.restart <= time_now)
                {
                        int next_state = session[s].ppp.lcp;
                        switch (session[s].ppp.lcp)
                        {
                        case RequestSent:
                        case AckReceived:
                                next_state = RequestSent;

                        case AckSent:
                                if (sess_local[s].lcp.conf_sent < config->ppp_max_configure)
                                {
                                        LOG(3, s, session[s].tunnel, "No ACK for LCP ConfigReq... resending\n");
                                        sendlcp(s, session[s].tunnel);
                                        change_state(s, lcp, next_state);
                                }
                                else
                                {
                                        sessionshutdown(s, "No response to LCP ConfigReq.", CDN_ADMIN_DISC, TERM_LOST_SERVICE);
                                        STAT(session_timeout);
                                }

                                s_actions++;
                        }

                        if (session[s].die)
                                continue;
                }

                if (sess_local[s].ipcp.restart <= time_now)
                {
                        int next_state = session[s].ppp.ipcp;
                        switch (session[s].ppp.ipcp)
                        {
                        case RequestSent:
                        case AckReceived:
                                next_state = RequestSent;

                        case AckSent:
                                if (sess_local[s].ipcp.conf_sent < config->ppp_max_configure)
                                {
                                        LOG(3, s, session[s].tunnel, "No ACK for IPCP ConfigReq... resending\n");
                                        sendipcp(s, session[s].tunnel);
                                        change_state(s, ipcp, next_state);
                                }
                                else
                                {
                                        sessionshutdown(s, "No response to IPCP ConfigReq.", CDN_ADMIN_DISC, TERM_LOST_SERVICE);
                                        STAT(session_timeout);
                                }

                                s_actions++;
                        }

                        if (session[s].die)
                                continue;
                }

                if (sess_local[s].ipv6cp.restart <= time_now)
                {
                        int next_state = session[s].ppp.ipv6cp;
                        switch (session[s].ppp.ipv6cp)
                        {
                        case RequestSent:
                        case AckReceived:
                                next_state = RequestSent;

                        case AckSent:
                                if (sess_local[s].ipv6cp.conf_sent < config->ppp_max_configure)
                                {
                                        LOG(3, s, session[s].tunnel, "No ACK for IPV6CP ConfigReq... resending\n");
                                        sendipv6cp(s, session[s].tunnel);
                                        change_state(s, ipv6cp, next_state);
                                }
                                else
                                {
                                        LOG(3, s, session[s].tunnel, "No ACK for IPV6CP ConfigReq\n");
                                        change_state(s, ipv6cp, Stopped);
                                }

                                s_actions++;
                        }
                }

                if (sess_local[s].ccp.restart <= time_now)
                {
                        int next_state = session[s].ppp.ccp;
                        switch (session[s].ppp.ccp)
                        {
                        case RequestSent:
                        case AckReceived:
                                next_state = RequestSent;

                        case AckSent:
                                if (sess_local[s].ccp.conf_sent < config->ppp_max_configure)
                                {
                                        LOG(3, s, session[s].tunnel, "No ACK for CCP ConfigReq... resending\n");
                                        sendccp(s, session[s].tunnel);
                                        change_state(s, ccp, next_state);
                                }
                                else
                                {
                                        LOG(3, s, session[s].tunnel, "No ACK for CCP ConfigReq\n");
                                        change_state(s, ccp, Stopped);
                                }

                                s_actions++;
                        }
                }

                // Drop sessions who have not responded within IDLE_TIMEOUT seconds
                if (session[s].last_packet && (time_now - session[s].last_packet >= IDLE_TIMEOUT))
                {
                        sessionshutdown(s, "No response to LCP ECHO requests.", CDN_ADMIN_DISC, TERM_LOST_SERVICE);
                        STAT(session_timeout);
                        s_actions++;
                        continue;
                }

                // No data in ECHO_TIMEOUT seconds, send LCP ECHO
                if (session[s].ppp.phase >= Establish && (time_now - session[s].last_packet >= ECHO_TIMEOUT) &&
                        (time_now - sess_local[s].last_echo >= ECHO_TIMEOUT))
                {
                        uint8_t b[MAXETHER];

                        uint8_t *q = makeppp(b, sizeof(b), 0, 0, s, session[s].tunnel, PPPLCP, 1, 0, 0);
                        if (!q) continue;

                        *q = EchoReq;
                        *(uint8_t *)(q + 1) = (time_now % 255); // ID
                        *(uint16_t *)(q + 2) = htons(8); // Length
                        *(uint32_t *)(q + 4) = session[s].ppp.lcp == Opened ? htonl(session[s].magic) : 0; // Magic Number

                        LOG(4, s, session[s].tunnel, "No data in %d seconds, sending LCP ECHO\n",
                                        (int)(time_now - session[s].last_packet));
                        tunnelsend(b, 24, session[s].tunnel); // send it
                        sess_local[s].last_echo = time_now;
                        s_actions++;
                }

                // Drop sessions who have reached session_timeout seconds
                if (session[s].session_timeout)
                {
                        bundleidt bid = session[s].bundle;
                        if (bid)
                        {
                                if (time_now - bundle[bid].last_check >= 1)
                                {
                                        bundle[bid].online_time += (time_now - bundle[bid].last_check) * bundle[bid].num_of_links;
                                        bundle[bid].last_check = time_now;
                                        if (bundle[bid].online_time >= session[s].session_timeout)
                                        {
                                                int ses;
                                                for (ses = bundle[bid].num_of_links - 1; ses >= 0; ses--)
                                                {
                                                        sessionshutdown(bundle[bid].members[ses], "Session timeout", CDN_ADMIN_DISC, TERM_SESSION_TIMEOUT);
                                                        s_actions++;
                                                        continue;
                                                }
                                        }
                                }
                        }
                        else if (time_now - session[s].opened >= session[s].session_timeout)
                        {
                                sessionshutdown(s, "Session timeout", CDN_ADMIN_DISC, TERM_SESSION_TIMEOUT);
                                s_actions++;
                                continue;
                        }
                }

                // Drop sessions who have reached idle_timeout seconds
                if (session[s].last_data && session[s].idle_timeout && (time_now - session[s].last_data >= session[s].idle_timeout))
                {
                        sessionshutdown(s, "Idle Timeout Reached", CDN_ADMIN_DISC, TERM_IDLE_TIMEOUT);
                        STAT(session_timeout);
                        s_actions++;
                        continue;
                }

                // Check for actions requested from the CLI
                if ((a = cli_session_actions[s].action))
                {
                        int send = 0;

                        cli_session_actions[s].action = 0;
                        if (a & CLI_SESS_KILL)
                        {
                                LOG(2, s, session[s].tunnel, "Dropping session by CLI\n");
                                sessionshutdown(s, "Requested by administrator.", CDN_ADMIN_DISC, TERM_ADMIN_RESET);
                                a = 0; // dead, no need to check for other actions
                                s_actions++;
                        }

                        if (a & CLI_SESS_NOSNOOP)
                        {
                                LOG(2, s, session[s].tunnel, "Unsnooping session by CLI\n");
                                session[s].snoop_ip = 0;
                                session[s].snoop_port = 0;
                                s_actions++;
                                send++;
                        }
                        else if (a & CLI_SESS_SNOOP)
                        {
                                LOG(2, s, session[s].tunnel, "Snooping session by CLI (to %s:%u)\n",
                                    fmtaddr(cli_session_actions[s].snoop_ip, 0),
                                    cli_session_actions[s].snoop_port);

                                session[s].snoop_ip = cli_session_actions[s].snoop_ip;
                                session[s].snoop_port = cli_session_actions[s].snoop_port;
                                s_actions++;
                                send++;
                        }

                        if (a & CLI_SESS_NOTHROTTLE)
                        {
                                LOG(2, s, session[s].tunnel, "Un-throttling session by CLI\n");
                                throttle_session(s, 0, 0);
                                s_actions++;
                                send++;
                        }
                        else if (a & CLI_SESS_THROTTLE)
                        {
                                LOG(2, s, session[s].tunnel, "Throttling session by CLI (to %dkb/s up and %dkb/s down)\n",
                                    cli_session_actions[s].throttle_in,
                                    cli_session_actions[s].throttle_out);

                                throttle_session(s, cli_session_actions[s].throttle_in, cli_session_actions[s].throttle_out);
                                s_actions++;
                                send++;
                        }

                        if (a & CLI_SESS_NOFILTER)
                        {
                                LOG(2, s, session[s].tunnel, "Un-filtering session by CLI\n");
                                filter_session(s, 0, 0);
                                s_actions++;
                                send++;
                        }
                        else if (a & CLI_SESS_FILTER)
                        {
                                LOG(2, s, session[s].tunnel, "Filtering session by CLI (in=%d, out=%d)\n",
                                    cli_session_actions[s].filter_in,
                                    cli_session_actions[s].filter_out);

                                filter_session(s, cli_session_actions[s].filter_in, cli_session_actions[s].filter_out);
                                s_actions++;
                                send++;
                        }

                        if (send)
                                cluster_send_session(s);
                }

                // RADIUS interim accounting
                if (config->radius_accounting && config->radius_interim > 0
                    && session[s].ip && !session[s].walled_garden
                    && !sess_local[s].radius // RADIUS already in progress
                    && time_now - sess_local[s].last_interim >= config->radius_interim
                    && session[s].flags & SESSION_STARTED)
                {
                        int rad = radiusnew(s);
                        if (!rad)
                        {
                                LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Interim message\n");
                                STAT(radius_overflow);
                                continue;
                        }

                        LOG(3, s, session[s].tunnel, "Sending RADIUS Interim for %s (%u)\n",
                                session[s].user, session[s].unique_id);

                        radiussend(rad, RADIUSINTERIM);
                        sess_local[s].last_interim = time_now;
                        s_actions++;
                }
        }

        LOG(4, 0, 0, "End regular cleanup: checked %d/%d/%d tunnels/radius/sessions; %d/%d/%d actions\n",
                t_slice, r_slice, s_slice, t_actions, r_actions, s_actions);
}

//
// Are we in the middle of a tunnel update, or radius
// requests??
//
static int still_busy(void)
{
        int i;
        static clockt last_talked = 0;
        static clockt start_busy_wait = 0;

#ifdef BGP
        static time_t stopped_bgp = 0;
        if (bgp_configured)
        {
                if (!stopped_bgp)
                {
                        LOG(1, 0, 0, "Shutting down in %d seconds, stopping BGP...\n", QUIT_DELAY);

                        for (i = 0; i < BGP_NUM_PEERS; i++)
                                if (bgp_peers[i].state == Established)
                                        bgp_stop(&bgp_peers[i]);

                        stopped_bgp = time_now;

                        if (!config->cluster_iam_master)
                        {
                                // we don't want to become master
                                cluster_send_ping(0);

                                return 1;
                        }
                }

                if (!config->cluster_iam_master && time_now < (stopped_bgp + QUIT_DELAY))
                        return 1;
        }
#endif /* BGP */

        if (!config->cluster_iam_master)
                return 0;

        if (main_quit == QUIT_SHUTDOWN)
        {
                static int dropped = 0;
                if (!dropped)
                {
                        int i;

                        LOG(1, 0, 0, "Dropping sessions and tunnels\n");
                        for (i = 1; i < MAXTUNNEL; i++)
                                if (tunnel[i].ip || tunnel[i].state)
                                        tunnelshutdown(i, "L2TPNS Closing", 6, 0, 0);

                        dropped = 1;
                }
        }

        if (start_busy_wait == 0)
                start_busy_wait = TIME;

        for (i = config->cluster_highest_tunnelid ; i > 0 ; --i)
        {
                if (!tunnel[i].controlc)
                        continue;

                if (last_talked != TIME)
                {
                        LOG(2, 0, 0, "Tunnel %u still has un-acked control messages.\n", i);
                        last_talked = TIME;
                }
                return 1;
        }

        // We stop waiting for radius after BUSY_WAIT_TIME 1/10th seconds
        if (abs(TIME - start_busy_wait) > BUSY_WAIT_TIME)
        {
                LOG(1, 0, 0, "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 %u is still busy (sid %u)\n", i, radius[i].session);
                        last_talked = TIME;
                }
                return 1;
        }

        return 0;
}

#ifdef HAVE_EPOLL
# include <sys/epoll.h>
#else
# define FAKE_EPOLL_IMPLEMENTATION /* include the functions */
# include "fake_epoll.h"
#endif

// the base set of fds polled: cli, cluster, tun, udp, control, dae
#define BASE_FDS        6

// additional polled fds
#ifdef BGP
# define EXTRA_FDS      BGP_NUM_PEERS
#else
# define EXTRA_FDS      0
#endif

// main loop - gets packets on tun or udp and processes them
static void mainloop(void)
{
        int i;
        uint8_t buf[65536];
        clockt next_cluster_ping = 0;   // send initial ping immediately
        struct epoll_event events[BASE_FDS + RADIUS_FDS + EXTRA_FDS];
        int maxevent = sizeof(events)/sizeof(*events);

        if ((epollfd = epoll_create(maxevent)) < 0)
        {
                LOG(0, 0, 0, "epoll_create failed: %s\n", strerror(errno));
                exit(1);
        }

        LOG(4, 0, 0, "Beginning of main loop.  clifd=%d, cluster_sockfd=%d, tunfd=%d, udpfd=%d, controlfd=%d, daefd=%d\n",
                clifd, cluster_sockfd, tunfd, udpfd, controlfd, daefd);

        /* setup our fds to poll for input */
        {
                static struct event_data d[BASE_FDS];
                struct epoll_event e;

                e.events = EPOLLIN;
                i = 0;

                if (clifd >= 0)
                {
                        d[i].type = FD_TYPE_CLI;
                        e.data.ptr = &d[i++];
                        epoll_ctl(epollfd, EPOLL_CTL_ADD, clifd, &e);
                }

                d[i].type = FD_TYPE_CLUSTER;
                e.data.ptr = &d[i++];
                epoll_ctl(epollfd, EPOLL_CTL_ADD, cluster_sockfd, &e);

                d[i].type = FD_TYPE_TUN;
                e.data.ptr = &d[i++];
                epoll_ctl(epollfd, EPOLL_CTL_ADD, tunfd, &e);

                d[i].type = FD_TYPE_UDP;
                e.data.ptr = &d[i++];
                epoll_ctl(epollfd, EPOLL_CTL_ADD, udpfd, &e);

                d[i].type = FD_TYPE_CONTROL;
                e.data.ptr = &d[i++];
                epoll_ctl(epollfd, EPOLL_CTL_ADD, controlfd, &e);

                d[i].type = FD_TYPE_DAE;
                e.data.ptr = &d[i++];
                epoll_ctl(epollfd, EPOLL_CTL_ADD, daefd, &e);
        }

#ifdef BGP
        signal(SIGPIPE, SIG_IGN);
        bgp_setup(config->as_number);
        if (config->bind_address)
                bgp_add_route(config->bind_address, 0xffffffff);

        for (i = 0; i < BGP_NUM_PEERS; i++)
        {
                if (config->neighbour[i].name[0])
                        bgp_start(&bgp_peers[i], config->neighbour[i].name,
                                config->neighbour[i].as, config->neighbour[i].keepalive,
                                config->neighbour[i].hold, config->neighbour[i].update_source,
                                0); /* 0 = routing disabled */
        }
#endif /* BGP */

        while (!main_quit || still_busy())
        {
                int more = 0;
                int n;


                if (main_reload)
                {
                        main_reload = 0;
                        read_config_file();
                        config->reload_config++;
                }

                if (config->reload_config)
                {
                        config->reload_config = 0;
                        update_config();
                }

#ifdef BGP
                bgp_set_poll();
#endif /* BGP */

                n = epoll_wait(epollfd, events, maxevent, 100); // timeout 100ms (1/10th sec)
                STAT(select_called);

                TIME = now(NULL);
                if (n < 0)
                {
                        if (errno == EINTR ||
                            errno == ECHILD) // EINTR was clobbered by sigchild_handler()
                                continue;

                        LOG(0, 0, 0, "Error returned from select(): %s\n", strerror(errno));
                        break; // exit
                }

                if (n)
                {
                        struct sockaddr_in addr;
                        struct in_addr local;
                        socklen_t alen;
                        int c, s;
                        int udp_ready = 0;
                        int tun_ready = 0;
                        int cluster_ready = 0;
                        int udp_pkts = 0;
                        int tun_pkts = 0;
                        int cluster_pkts = 0;
#ifdef BGP
                        uint32_t bgp_events[BGP_NUM_PEERS];
                        memset(bgp_events, 0, sizeof(bgp_events));
#endif /* BGP */

                        for (c = n, i = 0; i < c; i++)
                        {
                                struct event_data *d = events[i].data.ptr;

                                switch (d->type)
                                {
                                case FD_TYPE_CLI: // CLI connections
                                {
                                        int cli;
                                        
                                        alen = sizeof(addr);
                                        if ((cli = accept(clifd, (struct sockaddr *)&addr, &alen)) >= 0)
                                        {
                                                cli_do(cli);
                                                close(cli);
                                        }
                                        else
                                                LOG(0, 0, 0, "accept error: %s\n", strerror(errno));

                                        n--;
                                        break;
                                }

                                // these are handled below, with multiple interleaved reads
                                case FD_TYPE_CLUSTER:   cluster_ready++; break;
                                case FD_TYPE_TUN:       tun_ready++; break;
                                case FD_TYPE_UDP:       udp_ready++; break;

                                case FD_TYPE_CONTROL: // nsctl commands
                                        alen = sizeof(addr);
                                        s = recvfromto(controlfd, buf, sizeof(buf), MSG_WAITALL, (struct sockaddr *) &addr, &alen, &local);
                                        if (s > 0) processcontrol(buf, s, &addr, alen, &local);
                                        n--;
                                        break;

                                case FD_TYPE_DAE: // DAE requests
                                        alen = sizeof(addr);
                                        s = recvfromto(daefd, buf, sizeof(buf), MSG_WAITALL, (struct sockaddr *) &addr, &alen, &local);
                                        if (s > 0) processdae(buf, s, &addr, alen, &local);
                                        n--;
                                        break;

                                case FD_TYPE_RADIUS: // RADIUS response
                                        alen = sizeof(addr);
                                        s = recvfrom(radfds[d->index], buf, sizeof(buf), MSG_WAITALL, (struct sockaddr *) &addr, &alen);
                                        if (s >= 0 && config->cluster_iam_master)
                                        {
                                                if (addr.sin_addr.s_addr == config->radiusserver[0] ||
                                                    addr.sin_addr.s_addr == config->radiusserver[1])
                                                        processrad(buf, s, d->index);
                                                else
                                                        LOG(3, 0, 0, "Dropping RADIUS packet from unknown source %s\n",
                                                                fmtaddr(addr.sin_addr.s_addr, 0));
                                        }

                                        n--;
                                        break;

#ifdef BGP
                                case FD_TYPE_BGP:
                                        bgp_events[d->index] = events[i].events;
                                        n--;
                                        break;
#endif /* BGP */

                                default:
                                        LOG(0, 0, 0, "Unexpected fd type returned from epoll_wait: %d\n", d->type);
                                }
                        }

#ifdef BGP
                        bgp_process(bgp_events);
#endif /* BGP */

                        for (c = 0; n && c < config->multi_read_count; c++)
                        {
                                // L2TP
                                if (udp_ready)
                                {
                                        alen = sizeof(addr);
                                        if ((s = recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen)) > 0)
                                        {
                                                processudp(buf, s, &addr);
                                                udp_pkts++;
                                        }
                                        else
                                        {
                                                udp_ready = 0;
                                                n--;
                                        }
                                }

                                // incoming IP
                                if (tun_ready)
                                {
                                        if ((s = read(tunfd, buf, sizeof(buf))) > 0)
                                        {
                                                processtun(buf, s);
                                                tun_pkts++;
                                        }
                                        else
                                        {
                                                tun_ready = 0;
                                                n--;
                                        }
                                }

                                // cluster
                                if (cluster_ready)
                                {
                                        alen = sizeof(addr);
                                        if ((s = recvfrom(cluster_sockfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen)) > 0)
                                        {
                                                processcluster(buf, s, addr.sin_addr.s_addr);
                                                cluster_pkts++;
                                        }
                                        else
                                        {
                                                cluster_ready = 0;
                                                n--;
                                        }
                                }
                        }

                        if (udp_pkts > 1 || tun_pkts > 1 || cluster_pkts > 1)
                                STAT(multi_read_used);

                        if (c >= config->multi_read_count)
                        {
                                LOG(3, 0, 0, "Reached multi_read_count (%d); processed %d udp, %d tun and %d cluster packets\n",
                                        config->multi_read_count, udp_pkts, tun_pkts, cluster_pkts);

                                STAT(multi_read_exceeded);
                                more++;
                        }
                }

                if (time_changed)
                {
                        double Mbps = 1024.0 * 1024.0 / 8 * time_changed;

                        // Log current traffic stats
                        snprintf(config->bandwidth, sizeof(config->bandwidth),
                                "UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%u OUT:%u",
                                (udp_rx / Mbps), (eth_tx / Mbps), (eth_rx / Mbps), (udp_tx / Mbps),
                                ((udp_tx + udp_rx + eth_tx + eth_rx) / Mbps),
                                udp_rx_pkt / time_changed, eth_rx_pkt / time_changed);
                 
                        udp_tx = udp_rx = 0;
                        udp_rx_pkt = eth_rx_pkt = 0;
                        eth_tx = eth_rx = 0;
                        time_changed = 0;
                 
                        if (config->dump_speed)
                                printf("%s\n", config->bandwidth);
                 
                        // Update the internal time counter
                        strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
                 
                        {
                                // Run timer hooks
                                struct param_timer p = { time_now };
                                run_plugins(PLUGIN_TIMER, &p);
                        }
                }

                        // Runs on every machine (master and slaves).
                if (next_cluster_ping <= TIME)
                {
                        // Check to see which of the cluster is still alive..

                        cluster_send_ping(basetime);    // Only does anything if we're a slave
                        cluster_check_master();         // ditto.

                        cluster_heartbeat();            // Only does anything if we're a master.
                        cluster_check_slaves();         // ditto.

                        master_update_counts();         // If we're a slave, send our byte counters to our master.

                        if (config->cluster_iam_master && !config->cluster_iam_uptodate)
                                next_cluster_ping = TIME + 1; // out-of-date slaves, do fast updates
                        else
                                next_cluster_ping = TIME + config->cluster_hb_interval;
                }

                if (!config->cluster_iam_master)
                        continue;

                        // Run token bucket filtering queue..
                        // Only run it every 1/10th of a second.
                {
                        static clockt last_run = 0;
                        if (last_run != TIME)
                        {
                                last_run = TIME;
                                tbf_run_timer();
                        }
                }

                        // Handle timeouts, retries etc.
                {
                        static double last_clean = 0;
                        double this_clean;
                        double diff;

                        TIME = now(&this_clean);
                        diff = this_clean - last_clean;

                        // Run during idle time (after we've handled
                        // all incoming packets) or every 1/10th sec
                        if (!more || diff > 0.1)
                        {
                                regular_cleanups(diff);
                                last_clean = this_clean;
                        }
                }

                if (*config->accounting_dir)
                {
                        static clockt next_acct = 0;
                        static clockt next_shut_acct = 0;

                        if (next_acct <= TIME)
                        {
                                // Dump accounting data
                                next_acct = TIME + ACCT_TIME;
                                next_shut_acct = TIME + ACCT_SHUT_TIME;
                                dump_acct_info(1);
                        }
                        else if (next_shut_acct <= TIME)
                        {
                                // Dump accounting data for shutdown sessions
                                next_shut_acct = TIME + ACCT_SHUT_TIME;
                                if (shut_acct_n)
                                        dump_acct_info(0);
                        }
                }
        }

                // Are we the master and shutting down??
        if (config->cluster_iam_master)
                cluster_heartbeat(); // Flush any queued changes..

                // Ok. Notify everyone we're shutting down. If we're
                // the master, this will force an election.
        cluster_send_ping(0);

        //
        // Important!!! We MUST not process any packets past this point!
        LOG(1, 0, 0, "Shutdown complete\n");
}

static void stripdomain(char *host)
{
        char *p;

        if ((p = strchr(host, '.')))
        {
                char *domain = 0;
                char _domain[1024];

                // strip off domain
                FILE *resolv = fopen("/etc/resolv.conf", "r");
                if (resolv)
                {
                        char buf[1024];
                        char *b;

                        while (fgets(buf, sizeof(buf), resolv))
                        {
                                if (strncmp(buf, "domain", 6) && strncmp(buf, "search", 6))
                                        continue;

                                if (!isspace(buf[6]))
                                        continue;

                                b = buf + 7;
                                while (isspace(*b)) b++;

                                if (*b)
                                {
                                        char *d = b;
                                        while (*b && !isspace(*b)) b++;
                                        *b = 0;
                                        if (buf[0] == 'd') // domain is canonical
                                        {
                                                domain = d;
                                                break;
                                        }

                                        // first search line
                                        if (!domain)
                                        {
                                                // hold, may be subsequent domain line
                                                strncpy(_domain, d, sizeof(_domain))[sizeof(_domain)-1] = 0;
                                                domain = _domain;
                                        }
                                }
                        }

                        fclose(resolv);
                }

                if (domain)
                {
                        int hl = strlen(host);
                        int dl = strlen(domain);
                        if (dl < hl && host[hl - dl - 1] == '.' && !strcmp(host + hl - dl, domain))
                                host[hl -dl - 1] = 0;
                }
                else
                {
                        *p = 0; // everything after first dot
                }
        }
}

// Init data structures
static void initdata(int optdebug, char *optconfig)
{
        int i;

        if (!(config = shared_malloc(sizeof(configt))))
        {
                fprintf(stderr, "Error doing malloc for configuration: %s\n", strerror(errno));
                exit(1);
        }

        memset(config, 0, sizeof(configt));
        time(&config->start_time);
        strncpy(config->config_file, optconfig, strlen(optconfig));
        config->debug = optdebug;
        config->num_tbfs = MAXTBFS;
        config->rl_rate = 28; // 28kbps
        config->cluster_mcast_ttl = 1;
        config->cluster_master_min_adv = 1;
        config->ppp_restart_time = 3;
        config->ppp_max_configure = 10;
        config->ppp_max_failure = 5;
        config->kill_timedout_sessions = 1;
        strcpy(config->random_device, RANDOMDEVICE);

        log_stream = stderr;

#ifdef RINGBUFFER
        if (!(ringbuffer = shared_malloc(sizeof(struct Tringbuffer))))
        {
                LOG(0, 0, 0, "Error doing malloc for ringbuffer: %s\n", strerror(errno));
                exit(1);
        }
        memset(ringbuffer, 0, sizeof(struct Tringbuffer));
#endif

        if (!(_statistics = shared_malloc(sizeof(struct Tstats))))
        {
                LOG(0, 0, 0, "Error doing malloc for _statistics: %s\n", strerror(errno));
                exit(1);
        }
        if (!(tunnel = shared_malloc(sizeof(tunnelt) * MAXTUNNEL)))
        {
                LOG(0, 0, 0, "Error doing malloc for tunnels: %s\n", strerror(errno));
                exit(1);
        }
        if (!(bundle = shared_malloc(sizeof(bundlet) * MAXBUNDLE)))
        {
                LOG(0, 0, 0, "Error doing malloc for bundles: %s\n", strerror(errno));
                exit(1);
        }
        if (!(frag = shared_malloc(sizeof(fragmentationt) * MAXBUNDLE)))
        {
                LOG(0, 0, 0, "Error doing malloc for fragmentations: %s\n", strerror(errno));
                exit(1);
        }
        if (!(session = shared_malloc(sizeof(sessiont) * MAXSESSION)))
        {
                LOG(0, 0, 0, "Error doing malloc for sessions: %s\n", strerror(errno));
                exit(1);
        }

        if (!(sess_local = shared_malloc(sizeof(sessionlocalt) * MAXSESSION)))
        {
                LOG(0, 0, 0, "Error doing malloc for sess_local: %s\n", strerror(errno));
                exit(1);
        }

        if (!(radius = shared_malloc(sizeof(radiust) * MAXRADIUS)))
        {
                LOG(0, 0, 0, "Error doing malloc for radius: %s\n", strerror(errno));
                exit(1);
        }

        if (!(ip_address_pool = shared_malloc(sizeof(ippoolt) * MAXIPPOOL)))
        {
                LOG(0, 0, 0, "Error doing malloc for ip_address_pool: %s\n", strerror(errno));
                exit(1);
        }

        if (!(ip_filters = shared_malloc(sizeof(ip_filtert) * MAXFILTER)))
        {
                LOG(0, 0, 0, "Error doing malloc for ip_filters: %s\n", strerror(errno));
                exit(1);
        }
        memset(ip_filters, 0, sizeof(ip_filtert) * MAXFILTER);

        if (!(cli_session_actions = shared_malloc(sizeof(struct cli_session_actions) * MAXSESSION)))
        {
                LOG(0, 0, 0, "Error doing malloc for cli session actions: %s\n", strerror(errno));
                exit(1);
        }
        memset(cli_session_actions, 0, sizeof(struct cli_session_actions) * MAXSESSION);

        if (!(cli_tunnel_actions = shared_malloc(sizeof(struct cli_tunnel_actions) * MAXSESSION)))
        {
                LOG(0, 0, 0, "Error doing malloc for cli tunnel actions: %s\n", strerror(errno));
                exit(1);
        }
        memset(cli_tunnel_actions, 0, sizeof(struct cli_tunnel_actions) * MAXSESSION);

        memset(tunnel, 0, sizeof(tunnelt) * MAXTUNNEL);
        memset(bundle, 0, sizeof(bundlet) * MAXBUNDLE);
        memset(session, 0, sizeof(sessiont) * MAXSESSION);
        memset(radius, 0, sizeof(radiust) * MAXRADIUS);
        memset(ip_address_pool, 0, sizeof(ippoolt) * MAXIPPOOL);

                // Put all the sessions on the free list marked as undefined.
        for (i = 1; i < MAXSESSION; i++)
        {
                session[i].next = i + 1;
                session[i].tunnel = T_UNDEF;    // mark it as not filled in.
        }
        session[MAXSESSION - 1].next = 0;
        sessionfree = 1;

                // Mark all the tunnels as undefined (waiting to be filled in by a download).
        for (i = 1; i < MAXTUNNEL; i++)
                tunnel[i].state = TUNNELUNDEF;  // mark it as not filled in.

        for (i = 1; i < MAXBUNDLE; i++) {
                bundle[i].state = BUNDLEUNDEF;
        }

        if (!*hostname)
        {
                // Grab my hostname unless it's been specified
                gethostname(hostname, sizeof(hostname));
                stripdomain(hostname);
        }

        _statistics->start_time = _statistics->last_reset = time(NULL);

#ifdef BGP
        if (!(bgp_peers = shared_malloc(sizeof(struct bgp_peer) * BGP_NUM_PEERS)))
        {
                LOG(0, 0, 0, "Error doing malloc for bgp: %s\n", strerror(errno));
                exit(1);
        }
#endif /* BGP */
}

static int assign_ip_address(sessionidt s)
{
        uint32_t i;
        int best = -1;
        time_t best_time = time_now;
        char *u = session[s].user;
        char reuse = 0;


        CSTAT(assign_ip_address);

        for (i = 1; i < ip_pool_size; i++)
        {
                if (!ip_address_pool[i].address || ip_address_pool[i].assigned)
                        continue;

                if (!session[s].walled_garden && ip_address_pool[i].user[0] && !strcmp(u, ip_address_pool[i].user))
                {
                        best = i;
                        reuse = 1;
                        break;
                }

                if (ip_address_pool[i].last < best_time)
                {
                        best = i;
                        if (!(best_time = ip_address_pool[i].last))
                                break; // never used, grab this one
                }
        }

        if (best < 0)
        {
                LOG(0, s, session[s].tunnel, "assign_ip_address(): out of addresses\n");
                return 0;
        }

        session[s].ip = ip_address_pool[best].address;
        session[s].ip_pool_index = best;
        ip_address_pool[best].assigned = 1;
        ip_address_pool[best].last = time_now;
        ip_address_pool[best].session = s;
        if (session[s].walled_garden)
                /* Don't track addresses of users in walled garden (note: this
                   means that their address isn't "sticky" even if they get
                   un-gardened). */
                ip_address_pool[best].user[0] = 0;
        else
                strncpy(ip_address_pool[best].user, u, sizeof(ip_address_pool[best].user) - 1);

        STAT(ip_allocated);
        LOG(4, s, session[s].tunnel, "assign_ip_address(): %s ip address %d from pool\n",
                reuse ? "Reusing" : "Allocating", best);

        return 1;
}

static void free_ip_address(sessionidt s)
{
        int i = session[s].ip_pool_index;


        CSTAT(free_ip_address);

        if (!session[s].ip)
                return; // what the?

        if (i < 0)      // Is this actually part of the ip pool?
                i = 0;

        STAT(ip_freed);
        cache_ipmap(session[s].ip, -i); // Change the mapping to point back to the ip pool index.
        session[s].ip = 0;
        ip_address_pool[i].assigned = 0;
        ip_address_pool[i].session = 0;
        ip_address_pool[i].last = time_now;
}

//
// Fsck the address pool against the session table.
// Normally only called when we become a master.
//
// This isn't perfect: We aren't keep tracking of which
// users used to have an IP address.
//
void rebuild_address_pool(void)
{
        int i;

                //
                // Zero the IP pool allocation, and build
                // a map from IP address to pool index.
        for (i = 1; i < MAXIPPOOL; ++i)
        {
                ip_address_pool[i].assigned = 0;
                ip_address_pool[i].session = 0;
                if (!ip_address_pool[i].address)
                        continue;

                cache_ipmap(ip_address_pool[i].address, -i);    // Map pool IP to pool index.
        }

        for (i = 0; i < MAXSESSION; ++i)
        {
                int ipid;
                if (!(session[i].opened && session[i].ip))
                        continue;

                ipid = - lookup_ipmap(htonl(session[i].ip));

                if (session[i].ip_pool_index < 0)
                {
                        // Not allocated out of the pool.
                        if (ipid < 1)                   // Not found in the pool either? good.
                                continue;

                        LOG(0, i, 0, "Session %u has an IP address (%s) that was marked static, but is in the pool (%d)!\n",
                                i, fmtaddr(session[i].ip, 0), ipid);

                        // Fall through and process it as part of the pool.
                }


                if (ipid > MAXIPPOOL || ipid < 0)
                {
                        LOG(0, i, 0, "Session %u has a pool IP that's not found in the pool! (%d)\n", i, ipid);
                        ipid = -1;
                        session[i].ip_pool_index = ipid;
                        continue;
                }

                ip_address_pool[ipid].assigned = 1;
                ip_address_pool[ipid].session = i;
                ip_address_pool[ipid].last = time_now;
                strncpy(ip_address_pool[ipid].user, session[i].user, sizeof(ip_address_pool[ipid].user) - 1);
                session[i].ip_pool_index = ipid;
                cache_ipmap(session[i].ip, i);  // Fix the ip map.
        }
}

//
// Fix the address pool to match a changed session.
// (usually when the master sends us an update).
static void fix_address_pool(int sid)
{
        int ipid;

        ipid = session[sid].ip_pool_index;

        if (ipid > ip_pool_size)
                return;         // Ignore it. rebuild_address_pool will fix it up.

        if (ip_address_pool[ipid].address != session[sid].ip)
                return;         // Just ignore it. rebuild_address_pool will take care of it.

        ip_address_pool[ipid].assigned = 1;
        ip_address_pool[ipid].session = sid;
        ip_address_pool[ipid].last = time_now;
        strncpy(ip_address_pool[ipid].user, session[sid].user, sizeof(ip_address_pool[ipid].user) - 1);
}

//
// Add a block of addresses to the IP pool to hand out.
//
static void add_to_ip_pool(in_addr_t addr, in_addr_t mask)
{
        int i;
        if (mask == 0)
                mask = 0xffffffff;      // Host route only.

        addr &= mask;

        if (ip_pool_size >= MAXIPPOOL)  // Pool is full!
                return ;

        for (i = addr ;(i & mask) == addr; ++i)
        {
                if ((i & 0xff) == 0 || (i&0xff) == 255)
                        continue;       // Skip 0 and broadcast addresses.

                ip_address_pool[ip_pool_size].address = i;
                ip_address_pool[ip_pool_size].assigned = 0;
                ++ip_pool_size;
                if (ip_pool_size >= MAXIPPOOL)
                {
                        LOG(0, 0, 0, "Overflowed IP pool adding %s\n", fmtaddr(htonl(addr), 0));
                        return;
                }
        }
}

// Initialize the IP address pool
static void initippool()
{
        FILE *f;
        char *p;
        char buf[4096];
        memset(ip_address_pool, 0, sizeof(ip_address_pool));

        if (!(f = fopen(IPPOOLFILE, "r")))
        {
                LOG(0, 0, 0, "Can't load pool file " IPPOOLFILE ": %s\n", strerror(errno));
                exit(1);
        }

        while (ip_pool_size < MAXIPPOOL && fgets(buf, 4096, f))
        {
                char *pool = buf;
                buf[4095] = 0;  // Force it to be zero terminated/

                if (*buf == '#' || *buf == '\n')
                        continue; // Skip comments / blank lines
                if ((p = (char *)strrchr(buf, '\n'))) *p = 0;
                if ((p = (char *)strchr(buf, ':')))
                {
                        in_addr_t src;
                        *p = '\0';
                        src = inet_addr(buf);
                        if (src == INADDR_NONE)
                        {
                                LOG(0, 0, 0, "Invalid address pool IP %s\n", buf);
                                exit(1);
                        }
                        // This entry is for a specific IP only
                        if (src != config->bind_address)
                                continue;
                        *p = ':';
                        pool = p+1;
                }
                if ((p = (char *)strchr(pool, '/')))
                {
                        // It's a range
                        int numbits = 0;
                        in_addr_t start = 0, mask = 0;

                        LOG(2, 0, 0, "Adding IP address range %s\n", buf);
                        *p++ = 0;
                        if (!*p || !(numbits = atoi(p)))
                        {
                                LOG(0, 0, 0, "Invalid pool range %s\n", buf);
                                continue;
                        }
                        start = ntohl(inet_addr(pool));
                        mask = (in_addr_t) (pow(2, numbits) - 1) << (32 - numbits);

                        // Add a static route for this pool
                        LOG(5, 0, 0, "Adding route for address pool %s/%u\n",
                                fmtaddr(htonl(start), 0), 32 + mask);

                        routeset(0, start, mask, 0, 1);

                        add_to_ip_pool(start, mask);
                }
                else
                {
                        // It's a single ip address
                        add_to_ip_pool(ntohl(inet_addr(pool)), 0);
                }
        }
        fclose(f);
        LOG(1, 0, 0, "IP address pool is %d addresses\n", ip_pool_size - 1);
}

void snoop_send_packet(uint8_t *packet, uint16_t size, in_addr_t destination, uint16_t port)
{
        struct sockaddr_in snoop_addr = {0};
        if (!destination || !port || snoopfd <= 0 || size <= 0 || !packet)
                return;

        snoop_addr.sin_family = AF_INET;
        snoop_addr.sin_addr.s_addr = destination;
        snoop_addr.sin_port = ntohs(port);

        LOG(5, 0, 0, "Snooping %d byte packet to %s:%u\n", size,
                fmtaddr(snoop_addr.sin_addr.s_addr, 0),
                htons(snoop_addr.sin_port));

        if (sendto(snoopfd, packet, size, MSG_DONTWAIT | MSG_NOSIGNAL, (void *) &snoop_addr, sizeof(snoop_addr)) < 0)
                LOG(0, 0, 0, "Error sending intercept packet: %s\n", strerror(errno));

        STAT(packets_snooped);
}

static int dump_session(FILE **f, sessiont *s)
{
        if (!s->opened || !s->ip || !(s->cin_delta || s->cout_delta) || !*s->user || s->walled_garden)
                return 1;

        if (!*f)
        {
                char filename[1024];
                char timestr[64];
                time_t now = time(NULL);

                strftime(timestr, sizeof(timestr), "%Y%m%d%H%M%S", localtime(&now));
                snprintf(filename, sizeof(filename), "%s/%s", config->accounting_dir, timestr);

                if (!(*f = fopen(filename, "w")))
                {
                        LOG(0, 0, 0, "Can't write accounting info to %s: %s\n", filename, strerror(errno));
                        return 0;
                }

                LOG(3, 0, 0, "Dumping accounting information to %s\n", filename);
                fprintf(*f, "# dslwatch.pl dump file V1.01\n"
                        "# host: %s\n"
                        "# endpoint: %s\n"
                        "# time: %ld\n"
                        "# uptime: %ld\n"
                        "# format: username ip qos uptxoctets downrxoctets\n",
                        hostname,
                        fmtaddr(config->bind_address ? config->bind_address : my_address, 0),
                        now,
                        now - basetime);
        }

        LOG(4, 0, 0, "Dumping accounting information for %s\n", s->user);
        fprintf(*f, "%s %s %d %u %u\n",
                s->user,                                                // username
                fmtaddr(htonl(s->ip), 0),                               // ip
                (s->throttle_in || s->throttle_out) ? 2 : 1,            // qos
                (uint32_t) s->cin_delta,                                // uptxoctets
                (uint32_t) s->cout_delta);                              // downrxoctets

        s->cin_delta = s->cout_delta = 0;

        return 1;
}

static void dump_acct_info(int all)
{
        int i;
        FILE *f = NULL;


        CSTAT(dump_acct_info);

        if (shut_acct_n)
        {
                for (i = 0; i < shut_acct_n; i++)
                        dump_session(&f, &shut_acct[i]);

                shut_acct_n = 0;
        }

        if (all)
                for (i = 1; i <= config->cluster_highest_sessionid; i++)
                        dump_session(&f, &session[i]);

        if (f)
                fclose(f);
}

// Main program
int main(int argc, char *argv[])
{
        int i;
        int optdebug = 0;
        char *optconfig = CONFIGFILE;

        time(&basetime);             // start clock

        // scan args
        while ((i = getopt(argc, argv, "dvc:h:")) >= 0)
        {
                switch (i)
                {
                case 'd':
                        if (fork()) exit(0);
                        setsid();
                        freopen("/dev/null", "r", stdin);
                        freopen("/dev/null", "w", stdout);
                        freopen("/dev/null", "w", stderr);
                        break;
                case 'v':
                        optdebug++;
                        break;
                case 'c':
                        optconfig = optarg;
                        break;
                case 'h':
                        snprintf(hostname, sizeof(hostname), "%s", optarg);
                        break;
                default:
                        printf("Args are:\n"
                               "\t-d\t\tDetach from terminal\n"
                               "\t-c <file>\tConfig file\n"
                               "\t-h <hostname>\tForce hostname\n"
                               "\t-v\t\tDebug\n");

                        return (0);
                        break;
                }
        }

        // Start the timer routine off
        time(&time_now);
        strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));

        initplugins();
        initdata(optdebug, optconfig);

        init_cli(hostname);
        read_config_file();
        update_config();
        init_tbf(config->num_tbfs);

        LOG(0, 0, 0, "L2TPNS version " VERSION "\n");
        LOG(0, 0, 0, "Copyright (c) 2003, 2004, 2005, 2006 Optus Internet Engineering\n");
        LOG(0, 0, 0, "Copyright (c) 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced\n");
        {
                struct rlimit rlim;
                rlim.rlim_cur = RLIM_INFINITY;
                rlim.rlim_max = RLIM_INFINITY;
                // Remove the maximum core size
                if (setrlimit(RLIMIT_CORE, &rlim) < 0)
                        LOG(0, 0, 0, "Can't set ulimit: %s\n", strerror(errno));

                // Make core dumps go to /tmp
                chdir("/tmp");
        }

        if (config->scheduler_fifo)
        {
                int ret;
                struct sched_param params = {0};
                params.sched_priority = 1;

                if (get_nprocs() < 2)
                {
                        LOG(0, 0, 0, "Not using FIFO scheduler, there is only 1 processor in the system.\n");
                        config->scheduler_fifo = 0;
                }
                else
                {
                        if ((ret = sched_setscheduler(0, SCHED_FIFO, &params)) == 0)
                        {
                                LOG(1, 0, 0, "Using FIFO scheduler.  Say goodbye to any other processes running\n");
                        }
                        else
                        {
                                LOG(0, 0, 0, "Error setting scheduler to FIFO: %s\n", strerror(errno));
                                config->scheduler_fifo = 0;
                        }
                }
        }

        /* Set up the cluster communications port. */
        if (cluster_init() < 0)
                exit(1);

        inittun();
        LOG(1, 0, 0, "Set up on interface %s\n", config->tundevice);

        initudp();
        initrad();
        initippool();

        // seed prng
        {
                unsigned seed = time_now ^ getpid();
                LOG(4, 0, 0, "Seeding the pseudo random generator: %u\n", seed);
                srand(seed);
        }

        signal(SIGHUP,  sighup_handler);
        signal(SIGCHLD, sigchild_handler);
        signal(SIGTERM, shutdown_handler);
        signal(SIGINT,  shutdown_handler);
        signal(SIGQUIT, shutdown_handler);

        // Prevent us from getting paged out
        if (config->lock_pages)
        {
                if (!mlockall(MCL_CURRENT))
                        LOG(1, 0, 0, "Locking pages into memory\n");
                else
                        LOG(0, 0, 0, "Can't lock pages: %s\n", strerror(errno));
        }

        mainloop();

        /* remove plugins (so cleanup code gets run) */
        plugins_done();

        // Remove the PID file if we wrote it
        if (config->wrote_pid && *config->pid_file == '/')
                unlink(config->pid_file);

        /* kill CLI children */
        signal(SIGTERM, SIG_IGN);
        kill(0, SIGTERM);
        return 0;
}

static void sighup_handler(int sig)
{
        main_reload++;
}

static void shutdown_handler(int sig)
{
        main_quit = (sig == SIGQUIT) ? QUIT_SHUTDOWN : QUIT_FAILOVER;
}

static void sigchild_handler(int sig)
{
        while (waitpid(-1, NULL, WNOHANG) > 0)
            ;
}

static void build_chap_response(uint8_t *challenge, uint8_t id, uint16_t challenge_length, uint8_t **challenge_response)
{
        MD5_CTX ctx;
        *challenge_response = NULL;

        if (!*config->l2tp_secret)
        {
                LOG(0, 0, 0, "LNS requested CHAP authentication, but no l2tp secret is defined\n");
                return;
        }

        LOG(4, 0, 0, "   Building challenge response for CHAP request\n");

        *challenge_response = calloc(17, 1);

        MD5_Init(&ctx);
        MD5_Update(&ctx, &id, 1);
        MD5_Update(&ctx, config->l2tp_secret, strlen(config->l2tp_secret));
        MD5_Update(&ctx, challenge, challenge_length);
        MD5_Final(*challenge_response, &ctx);

        return;
}

static int facility_value(char *name)
{
        int i;
        for (i = 0; facilitynames[i].c_name; i++)
        {
                if (strcmp(facilitynames[i].c_name, name) == 0)
                        return facilitynames[i].c_val;
        }
        return 0;
}

static void update_config()
{
        int i;
        char *p;
        static int timeout = 0;
        static int interval = 0;

        // Update logging
        closelog();
        syslog_log = 0;
        if (log_stream)
        {
                if (log_stream != stderr)
                        fclose(log_stream);

                log_stream = NULL;
        }

        if (*config->log_filename)
        {
                if (strstr(config->log_filename, "syslog:") == config->log_filename)
                {
                        char *p = config->log_filename + 7;
                        if (*p)
                        {
                                openlog("l2tpns", LOG_PID, facility_value(p));
                                syslog_log = 1;
                        }
                }
                else if (strchr(config->log_filename, '/') == config->log_filename)
                {
                        if ((log_stream = fopen((char *)(config->log_filename), "a")))
                        {
                                fseek(log_stream, 0, SEEK_END);
                                setbuf(log_stream, NULL);
                        }
                        else
                        {
                                log_stream = stderr;
                                setbuf(log_stream, NULL);
                        }
                }
        }
        else
        {
                log_stream = stderr;
                setbuf(log_stream, NULL);
        }

#define L2TP_HDRS               (20+8+6+4)      // L2TP data encaptulation: ip + udp + l2tp (data) + ppp (inc hdlc)
#define TCP_HDRS                (20+20)         // TCP encapsulation: ip + tcp

        if (config->l2tp_mtu <= 0)              config->l2tp_mtu = 1500; // ethernet default
        else if (config->l2tp_mtu < MINMTU)     config->l2tp_mtu = MINMTU;
        else if (config->l2tp_mtu > MAXMTU)     config->l2tp_mtu = MAXMTU;

        // reset MRU/MSS globals
        MRU = config->l2tp_mtu - L2TP_HDRS;
        if (MRU > PPPoE_MRU)
                MRU = PPPoE_MRU;

        MSS = MRU - TCP_HDRS;

        // Update radius
        config->numradiusservers = 0;
        for (i = 0; i < MAXRADSERVER; i++)
                if (config->radiusserver[i])
                {
                        config->numradiusservers++;
                        // Set radius port: if not set, take the port from the
                        // first radius server.  For the first radius server,
                        // take the #defined default value from l2tpns.h

                        // test twice, In case someone works with
                        // a secondary radius server without defining
                        // a primary one, this will work even then.
                        if (i > 0 && !config->radiusport[i])
                                config->radiusport[i] = config->radiusport[i-1];
                        if (!config->radiusport[i])
                                config->radiusport[i] = RADPORT;
                }

        if (!config->numradiusservers)
                LOG(0, 0, 0, "No RADIUS servers defined!\n");

        // parse radius_authtypes_s
        config->radius_authtypes = config->radius_authprefer = 0;
        p = config->radius_authtypes_s;
        while (p && *p)
        {
                char *s = strpbrk(p, " \t,");
                int type = 0;

                if (s)
                {
                        *s++ = 0;
                        while (*s == ' ' || *s == '\t')
                                s++;

                        if (!*s)
                                s = 0;
                }

                if (!strncasecmp("chap", p, strlen(p)))
                        type = AUTHCHAP;
                else if (!strncasecmp("pap", p, strlen(p)))
                        type = AUTHPAP;
                else
                        LOG(0, 0, 0, "Invalid RADIUS authentication type \"%s\"\n", p);

                config->radius_authtypes |= type;
                if (!config->radius_authprefer)
                        config->radius_authprefer = type;

                p = s;
        }

        if (!config->radius_authtypes)
        {
                LOG(0, 0, 0, "Defaulting to PAP authentication\n");
                config->radius_authtypes = config->radius_authprefer = AUTHPAP;
        }

        // normalise radius_authtypes_s
        if (config->radius_authprefer == AUTHPAP)
        {
                strcpy(config->radius_authtypes_s, "pap");
                if (config->radius_authtypes & AUTHCHAP)
                        strcat(config->radius_authtypes_s, ", chap");
        }
        else
        {
                strcpy(config->radius_authtypes_s, "chap");
                if (config->radius_authtypes & AUTHPAP)
                        strcat(config->radius_authtypes_s, ", pap");
        }

        if (!config->radius_dae_port)
                config->radius_dae_port = DAEPORT;

        // re-initialise the random number source
        initrandom(config->random_device);

        // Update plugins
        for (i = 0; i < MAXPLUGINS; i++)
        {
                if (strcmp(config->plugins[i], config->old_plugins[i]) == 0)
                        continue;

                if (*config->plugins[i])
                {
                        // Plugin added
                        add_plugin(config->plugins[i]);
                }
                else if (*config->old_plugins[i])
                {
                        // Plugin removed
                        remove_plugin(config->old_plugins[i]);
                }
        }

        // Guest change
        guest_accounts_num = 0;
        char *p2 = config->guest_user;
        while (p2 && *p2)
        {
                char *s = strpbrk(p2, " \t,");
                if (s)
                {
                        *s++ = 0;
                        while (*s == ' ' || *s == '\t')
                                s++;

                        if (!*s)
                                s = 0;
                }

                strcpy(guest_users[guest_accounts_num], p2);
                LOG(1, 0, 0, "Guest account[%d]: %s\n", guest_accounts_num, guest_users[guest_accounts_num]);
                guest_accounts_num++;
                p2 = s;
        }
        // Rebuild the guest_user array
        strcpy(config->guest_user, "");
        int ui = 0;
        for (ui=0; ui<guest_accounts_num; ui++)
        {
                strcat(config->guest_user, guest_users[ui]);
                if (ui<guest_accounts_num-1)
                {
                        strcat(config->guest_user, ",");
                }
        }


        memcpy(config->old_plugins, config->plugins, sizeof(config->plugins));
        if (!config->multi_read_count) config->multi_read_count = 10;
        if (!config->cluster_address) config->cluster_address = inet_addr(DEFAULT_MCAST_ADDR);
        if (!*config->cluster_interface)
                strncpy(config->cluster_interface, DEFAULT_MCAST_INTERFACE, sizeof(config->cluster_interface) - 1);

        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));
                }
        }
}

static void read_config_file()
{
        FILE *f;

        if (!config->config_file) return;
        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, "Reading config file %s\n", config->config_file);
        cli_do_file(f);
        LOG(3, 0, 0, "Done reading config file\n");
        fclose(f);
}

int sessionsetup(sessionidt s, tunnelidt t)
{
        // A session now exists, set it up
        in_addr_t ip;
        char *user;
        sessionidt i;
        int r;

        CSTAT(sessionsetup);

        LOG(3, s, t, "Doing session setup for session\n");

        // Join a bundle if the MRRU option is accepted
        if(session[s].mrru > 0 && session[s].bundle == 0)
        {
                LOG(3, s, t, "This session can be part of multilink bundle\n");
                if (join_bundle(s) > 0)
                        cluster_send_bundle(session[s].bundle);
                else
                {
                        LOG(0, s, t, "MPPP: Mismaching mssf option with other sessions in bundle\n");
                        sessionshutdown(s, "Mismaching mssf option.", CDN_NONE, TERM_SERVICE_UNAVAILABLE);
                        return 0;
                }
        }

        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.", CDN_TRY_ANOTHER, TERM_SERVICE_UNAVAILABLE);
                        return 0;
                }
                LOG(3, s, t, "   No IP allocated.  Assigned %s from pool\n",
                        fmtaddr(htonl(session[s].ip), 0));
        }


        // Make sure this is right
        session[s].tunnel = t;

        // zap old sessions with same IP and/or username
        // Don't kill gardened sessions - doing so leads to a DoS
        // from someone who doesn't need to know the password
        {
                ip = session[s].ip;
                user = session[s].user;
                for (i = 1; i <= config->cluster_highest_sessionid; i++)
                {
                        if (i == s) continue;
                        if (!session[s].opened) continue;
                        // Allow duplicate sessions for multilink ones of the same bundle.
                        if (session[s].bundle && session[i].bundle && session[s].bundle == session[i].bundle)
                                continue;
                        if (ip == session[i].ip)
                        {
                                sessionkill(i, "Duplicate IP address");
                                continue;
                        }

                        if (config->allow_duplicate_users) continue;
                        if (session[s].walled_garden || session[i].walled_garden) continue;
                        // Guest change
                        int found = 0;
                        int gu;
                        for (gu = 0; gu < guest_accounts_num; gu++)
                        {
                                if (!strcasecmp(user, guest_users[gu]))
                                {
                                        found = 1;
                                        break;
                                }
                        }
                        if (found) continue;

                        // Drop the new session in case of duplicate sessionss, not the old one.
                        if (!strcasecmp(user, session[i].user))
                                sessionkill(i, "Duplicate session for users");
                }
        }

        // no need to set a route for the same IP address of the bundle
        if (!session[s].bundle || (bundle[session[s].bundle].num_of_links == 1))
        {
                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.
                // IPv6 route setup is done in ppp.c, when IPV6CP is acked.
                if (session[s].ip_pool_index == -1) // static ip
                {
                        if (!routed) routeset(s, session[s].ip, 0, 0, 1);
                }
                else
                        cache_ipmap(session[s].ip, s);
        }

        sess_local[s].lcp_authtype = 0; // RADIUS authentication complete
        lcp_open(s, t); // transition to Network phase and send initial IPCP

        // 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 = session[s].last_data = 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);
                }
        }

        // check v6 routing
        if (new->ipv6prefixlen && new->ppp.ipv6cp == Opened && session[s].ppp.ipv6cp != Opened)
                    route6set(s, new->ipv6route, new->ipv6prefixlen, 1);

        // check filters
        if (new->filter_in && (new->filter_in > MAXFILTER || !ip_filters[new->filter_in - 1].name[0]))
        {
                LOG(2, s, session[s].tunnel, "Dropping invalid input filter %u\n", (int) new->filter_in);
                new->filter_in = 0;
        }

        if (new->filter_out && (new->filter_out > MAXFILTER || !ip_filters[new->filter_out - 1].name[0]))
        {
                LOG(2, s, session[s].tunnel, "Dropping invalid output filter %u\n", (int) new->filter_out);
                new->filter_out = 0;
        }

        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++;
        }

        if (new->filter_out != session[s].filter_out)
        {
                if (session[s].filter_out) ip_filters[session[s].filter_out - 1].used--;
                if (new->filter_out)       ip_filters[new->filter_out - 1].used++;
        }

        if (new->tunnel && s > config->cluster_highest_sessionid)       // Maintain this in the slave. It's used
                                        // for walking the sessions to forward byte counts to the master.
                config->cluster_highest_sessionid = s;

        memcpy(&session[s], new, sizeof(session[s]));   // Copy over..

                // Do fixups into address pool.
        if (new->ip_pool_index != -1)
                fix_address_pool(s);

        return 1;
}

static void initplugins()
{
        int i;

        loaded_plugins = ll_init();
        // Initialize the plugins to nothing
        for (i = 0; i < MAX_PLUGIN_TYPES; i++)
                plugins[i] = ll_init();
}

static void *open_plugin(char *plugin_name, int load)
{
        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,
                sessionshutdown,
                sessionkill,
                throttle_session,
                cluster_send_session,
        };

        void *p = open_plugin(plugin_name, 1);
        int (*initfunc)(struct pluginfuncs *);
        int i;

        if (!p)
        {
                LOG(1, 0, 0, "   Plugin load failed: %s\n", dlerror());
                return -1;
        }

        if (ll_contains(loaded_plugins, p))
        {
                dlclose(p);
                return 0; // already loaded
        }

        {
                int *v = dlsym(p, "plugin_api_version");
                if (!v || *v != PLUGIN_API_VERSION)
                {
                        LOG(1, 0, 0, "   Plugin load failed: API version mismatch: %s\n", dlerror());
                        dlclose(p);
                        return -1;
                }
        }

        if ((initfunc = dlsym(p, "plugin_init")))
        {
                if (!initfunc(&funcs))
                {
                        LOG(1, 0, 0, "   Plugin load failed: plugin_init() returned FALSE: %s\n", dlerror());
                        dlclose(p);
                        return -1;
                }
        }

        ll_push(loaded_plugins, p);

        for (i = 0; i < max_plugin_functions; i++)
        {
                void *x;
                if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
                {
                        LOG(3, 0, 0, "   Supports function \"%s\"\n", plugin_functions[i]);
                        ll_push(plugins[i], x);
                }
        }

        LOG(2, 0, 0, "   Loaded plugin %s\n", plugin_name);
        return 1;
}

static void run_plugin_done(void *plugin)
{
        int (*donefunc)(void) = dlsym(plugin, "plugin_done");

        if (donefunc)
                donefunc();
}

static int remove_plugin(char *plugin_name)
{
        void *p = open_plugin(plugin_name, 0);
        int loaded = 0;

        if (!p)
                return -1;

        if (ll_contains(loaded_plugins, p))
        {
                int i;
                for (i = 0; i < max_plugin_functions; i++)
                {
                        void *x;
                        if (plugin_functions[i] && (x = dlsym(p, plugin_functions[i])))
                                ll_delete(plugins[i], x);
                }

                ll_delete(loaded_plugins, p);
                run_plugin_done(p);
                loaded = 1;
        }

        dlclose(p);
        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 PLUGIN_RET_ERROR;

        ll_reset(plugins[plugin_type]);
        while ((func = ll_next(plugins[plugin_type])))
        {
                int r = func(data);

                if (r != PLUGIN_RET_OK)
                        return r; // stop here
        }

        return PLUGIN_RET_OK;
}

static void plugins_done()
{
        void *p;

        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 in_addr *local)
{
        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;
                                }
                        }
                }

                break;

        default:
                response.type = NSCTL_RES_ERR;
                response.argc = 1;
                response.argv[0] = "error unpacking control packet";
        }

        buf = calloc(NSCTL_MAX_PKT_SZ, 1);
        if (!buf)
        {
                LOG(2, 0, 0, "Failed to allocate nsctl response\n");
                return;
        }

        r = pack_control(buf, NSCTL_MAX_PKT_SZ, response.type, response.argc, response.argv);
        if (r > 0)
        {
                sendtofrom(controlfd, buf, r, 0, (const struct sockaddr *) addr, alen, local);
                if (log_stream && config->debug >= 4)
                {
                        LOG(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 %u\n", i);
                        if (i > config->cluster_highest_tunnelid)
                                config->cluster_highest_tunnelid = i;
                        return i;
                }
        }
        LOG(0, 0, 0, "Can't find a free tunnel! There shouldn't be this many in use!\n");
        return 0;
}

//
// We're becoming the master. Do any required setup..
//
// This is principally telling all the plugins that we're
// now a master, and telling them about all the sessions
// that are active too..
//
void become_master(void)
{
        int s, i;
        static struct event_data d[RADIUS_FDS];
        struct epoll_event e;

        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
        e.events = EPOLLIN;
        for (i = 0; i < RADIUS_FDS; i++)
        {
                d[i].type = FD_TYPE_RADIUS;
                d[i].index = i;
                e.data.ptr = &d[i];

                epoll_ctl(epollfd, EPOLL_CTL_ADD, radfds[i], &e);
        }
}

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_data;
                idle /= 5 ; // In multiples of 5 seconds.
                if (idle < 0)
                        idle = 0;
                if (idle > 63)
                        idle = 63;

                ++count;
                ++buckets[idle];
        }

        for (i = 0; i < 63; ++i)
        {
                cli_print(cli, "%3d seconds  : %7.2f%% (%6d)", i * 5, (double) buckets[i] * 100.0 / count , buckets[i]);
        }
        cli_print(cli, "lots of secs : %7.2f%% (%6d)", (double) buckets[63] * 100.0 / count , buckets[i]);
        cli_print(cli, "%d total sessions open.", count);
        return CLI_OK;
}

int cmd_show_hist_open(struct cli_def *cli, char *command, char **argv, int argc)
{
        int s, i;
        int count = 0;
        int buckets[64];

        if (CLI_HELP_REQUESTED)
                return CLI_HELP_NO_ARGS;

        time(&time_now);
        for (i = 0; i < 64;++i) buckets[i] = 0;

        for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
        {
                int open = 0, d;
                if (!session[s].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];
        }

        s = 1;
        for (i = 0; i  < 30; ++i)
        {
                cli_print(cli, " < %8d seconds : %7.2f%% (%6d)", s, (double) buckets[i] * 100.0 / count , buckets[i]);
                s <<= 1;
        }
        cli_print(cli, "%d total sessions open.", count);
        return CLI_OK;
}

/* Unhide an avp.
 *
 * This unencodes the AVP using the L2TP secret and the previously
 * stored random vector.  It overwrites the hidden data with the
 * unhidden AVP subformat.
 */
static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vector, size_t vec_len)
{
        MD5_CTX ctx;
        uint8_t digest[16];
        uint8_t *last;
        size_t d = 0;
        uint16_t m = htons(type);

        // Compute initial pad
        MD5_Init(&ctx);
        MD5_Update(&ctx, (unsigned char *) &m, 2);
        MD5_Update(&ctx, config->l2tp_secret, strlen(config->l2tp_secret));
        MD5_Update(&ctx, vector, vec_len);
        MD5_Final(digest, &ctx);

        // pointer to last decoded 16 octets
        last = value;

        while (len > 0)
        {
                // calculate a new pad based on the last decoded block
                if (d >= sizeof(digest))
                {
                        MD5_Init(&ctx);
                        MD5_Update(&ctx, config->l2tp_secret, strlen(config->l2tp_secret));
                        MD5_Update(&ctx, last, sizeof(digest));
                        MD5_Final(digest, &ctx);

                        d = 0;
                        last = value;
                }

                *value++ ^= digest[d++];
                len--;
        }
}

int find_filter(char const *name, size_t len)
{
        int free = -1;
        int i;

        for (i = 0; i < MAXFILTER; i++)
        {
                if (!*ip_filters[i].name)
                {
                        if (free < 0)
                                free = i;

                        continue;
                }

                if (strlen(ip_filters[i].name) != len)
                        continue;

                if (!strncmp(ip_filters[i].name, name, len))
                        return i;
        }
                        
        return free;
}

static int ip_filter_port(ip_filter_portt *p, uint16_t port)
{
        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;
}

static int ip_filter_flag(uint8_t op, uint8_t sflags, uint8_t cflags, uint8_t flags)
{
        switch (op)
        {
        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);
        }

        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)
                {
                        // layer 4 deny rules are skipped
                        if (rule->action == FILTER_ACTION_DENY &&
                            (rule->src_ports.op || rule->dst_ports.op || rule->tcp_flag_op))
                                continue;
                }
                else
                {
                        if (rule->frag)
                                continue;

                        if (proto == IPPROTO_TCP || proto == IPPROTO_UDP)
                        {
                                if (rule->src_ports.op && !ip_filter_port(&rule->src_ports, src_port))
                                        continue;

                                if (rule->dst_ports.op && !ip_filter_port(&rule->dst_ports, dst_port))
                                        continue;

                                if (proto == IPPROTO_TCP && rule->tcp_flag_op &&
                                    !ip_filter_flag(rule->tcp_flag_op, rule->tcp_sflags, rule->tcp_cflags, flags))
                                        continue;
                        }
                }

                // matched
                rule->counter++;
                return rule->action == FILTER_ACTION_PERMIT;
        }

        // default deny
        return 0;
}