set MTU on tunnel interface so the kernel will re-fragment large packets to within MRU
[l2tpns.git] / l2tpns.c
index 1f47934..39f72d7 100644 (file)
--- a/l2tpns.c
+++ b/l2tpns.c
@@ -4,7 +4,7 @@
 // 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 1.90 2005/04/18 05:07:20 bodea Exp $";
+char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.150 2005/11/17 07:35:35 bodea Exp $";
 
 #include <arpa/inet.h>
 #include <assert.h>
@@ -52,7 +52,7 @@ char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.90 2005/04/18 05:07:20 bodea Exp
 
 #ifdef BGP
 #include "bgp.h"
-#endif /* BGP */
+#endif
 
 // Globals
 configt *config = NULL;                // all configuration
@@ -60,19 +60,25 @@ 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
-static int rand_fd = -1;       // Random data source
+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 = NULL;        // file handle for direct logging (i.e. direct into file, not via syslog).
-extern int cluster_sockfd;     // Intra-cluster communications socket.
+static FILE *log_stream = 0;   // file handle for direct logging (i.e. direct into file, not via syslog).
 uint32_t last_id = 0;          // Unique ID for radius accounting
 
+// calculated from config->l2tp_mtu
+uint16_t MRU = 0;              // PPP MRU
+uint16_t MSS = 0;              // TCP MSS
+
 struct cli_session_actions *cli_session_actions = NULL;        // Pending session changes requested by CLI
 struct cli_tunnel_actions *cli_tunnel_actions = NULL;  // Pending tunnel changes required by CLI
 
@@ -90,7 +96,9 @@ 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 char main_quit = 0;             // True if we're in the process of exiting.
+int time_changed = 0;                  // time_now changed
+char main_quit = 0;                    // True if we're in the process of exiting.
+char main_reload = 0;                  // Re-load pending
 linked_list *loaded_plugins;
 linked_list *plugins[MAX_PLUGIN_TYPES];
 
@@ -102,7 +110,11 @@ config_descriptt config_values[] = {
        CONFIG("log_file", log_filename, STRING),
        CONFIG("pid_file", pid_file, STRING),
        CONFIG("random_device", random_device, STRING),
-       CONFIG("l2tp_secret", l2tpsecret, 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),
@@ -110,8 +122,11 @@ config_descriptt config_values[] = {
        CONFIG("primary_radius_port", radiusport[0], SHORT),
        CONFIG("secondary_radius_port", radiusport[1], SHORT),
        CONFIG("radius_accounting", radius_accounting, BOOL),
+       CONFIG("radius_interim", radius_interim, INT),
        CONFIG("radius_secret", radiussecret, STRING),
        CONFIG("radius_authtypes", radius_authtypes_s, STRING),
+       CONFIG("radius_dae_port", radius_dae_port, SHORT),
+       CONFIG("allow_duplicate_users", allow_duplicate_users, BOOL),
        CONFIG("bind_address", bind_address, IPv4),
        CONFIG("peer_address", peer_address, IPv4),
        CONFIG("send_garp", send_garp, BOOL),
@@ -120,7 +135,6 @@ config_descriptt config_values[] = {
        CONFIG("accounting_dir", accounting_dir, STRING),
        CONFIG("setuid", target_uid, INT),
        CONFIG("dump_speed", dump_speed, BOOL),
-       CONFIG("cleanup_interval", cleanup_interval, INT),
        CONFIG("multi_read_count", multi_read_count, INT),
        CONFIG("scheduler_fifo", scheduler_fifo, BOOL),
        CONFIG("lock_pages", lock_pages, BOOL),
@@ -128,8 +142,10 @@ config_descriptt config_values[] = {
        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 },
 };
@@ -145,6 +161,8 @@ static char *plugin_functions[] = {
        "plugin_kill_session",
        "plugin_control",
        "plugin_radius_response",
+       "plugin_radius_reset",
+       "plugin_radius_account",
        "plugin_become_master",
        "plugin_new_session_master",
 };
@@ -160,7 +178,7 @@ 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.
+ip_filtert *ip_filters = NULL;         // Array of named filters.
 static controlt *controlfree = 0;
 struct Tstats *_statistics = NULL;
 #ifdef RINGBUFFER
@@ -173,11 +191,9 @@ static void cache_ipv6map(struct in6_addr ip, int prefixlen, int s);
 static void free_ip_address(sessionidt s);
 static void dump_acct_info(int all);
 static void sighup_handler(int sig);
-static void sigalrm_handler(int sig);
-static void sigterm_handler(int sig);
-static void sigquit_handler(int sig);
+static void shutdown_handler(int sig);
 static void sigchild_handler(int sig);
-static void build_chap_response(char *challenge, uint8_t id, uint16_t challenge_length, char **challenge_response);
+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);
@@ -188,11 +204,25 @@ static void processcontrol(uint8_t *buf, int len, struct sockaddr_in *addr, int
 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);
 
-// return internal time (10ths since process startup)
-static clockt now(void)
+// 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;
 }
 
@@ -202,7 +232,7 @@ static clockt now(void)
 clockt backoff(uint8_t try)
 {
        if (try > 5) try = 5;                  // max backoff
-       return now() + 10 * (1 << try);
+       return now(NULL) + 10 * (1 << try);
 }
 
 
@@ -245,10 +275,10 @@ void _log(int level, sessionidt s, tunnelidt t, const char *format, ...)
        va_end(ap);
 }
 
-void _log_hex(int level, const char *title, const char *data, int maxsize)
+void _log_hex(int level, const char *title, const uint8_t *data, int maxsize)
 {
        int i, j;
-       const uint8_t *d = (const uint8_t *) data;
+       const uint8_t *d = data;
 
        if (config->debug < level) return;
 
@@ -296,6 +326,16 @@ void _log_hex(int level, const char *title, const char *data, int maxsize)
        }
 }
 
+// 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)
 {
@@ -306,7 +346,8 @@ static void initrandom(char *source)
                return;
 
        // close previous source, if any
-       if (rand_fd >= 0) close(rand_fd);
+       if (rand_fd >= 0)
+               close(rand_fd);
 
        rand_fd = -1;
 
@@ -323,13 +364,6 @@ static void initrandom(char *source)
                                        path, strerror(errno));
                }
        }
-
-       // no source: seed prng
-       {
-               unsigned seed = time_now ^ getpid();
-               LOG(4, 0, 0, "Seeding the pseudo random generator: %u\n", seed);
-               srand(seed);
-       }
 }
 
 // fill buffer with random data
@@ -350,7 +384,7 @@ void random_data(uint8_t *buf, int len)
                                        strerror(errno));
 
                                // fall back to rand()
-                               initrandom(0);
+                               initrandom(NULL);
                        }
 
                        n = 0;
@@ -520,6 +554,13 @@ static void inittun(void)
                LOG(0, 0, 0, "Error setting tun queue length: %s\n", strerror(errno));
                exit(1);
        }
+       /* set MTU to modem MRU + 4 (tun header) */
+       ifr.ifr_mtu = MRU + 4;
+       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)
        {
@@ -534,7 +575,7 @@ static void inittun(void)
        tunidx = ifr.ifr_ifindex;
 
        // Only setup IPv6 on the tun device if we have a configured prefix
-       if (config->ipv6_prefix.s6_addr[0] > 0) {
+       if (config->ipv6_prefix.s6_addr[0]) {
                ifr6fd = socket(PF_INET6, SOCK_DGRAM, 0);
 
                // Link local address is FE80::1
@@ -564,7 +605,7 @@ static void inittun(void)
        }
 }
 
-// set up UDP port
+// set up UDP ports
 static void initudp(void)
 {
        int on = 1;
@@ -586,7 +627,6 @@ static void initudp(void)
                LOG(0, 0, 0, "Error in UDP bind: %s\n", strerror(errno));
                exit(1);
        }
-       snoopfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
 
        // Control
        memset(&addr, 0, sizeof(addr));
@@ -599,6 +639,21 @@ static void initudp(void)
                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));
+       if (bind(daefd, (void *) &addr, sizeof(addr)) < 0)
+       {
+               LOG(0, 0, 0, "Error in DAE bind: %s\n", strerror(errno));
+               exit(1);
+       }
+
+       // Intercept
+       snoopfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
 }
 
 //
@@ -671,8 +726,11 @@ sessionidt sessionbyipv6(struct in6_addr ip)
        CSTAT(sessionbyipv6);
 
        if (!memcmp(&config->ipv6_prefix, &ip, 8) ||
-               (ip.s6_addr[0] == 0xFE && ip.s6_addr[1] == 0x80 &&
-                (ip.s6_addr16[1] == ip.s6_addr16[2] == ip.s6_addr16[3] == 0))) {
+               (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);
@@ -911,7 +969,7 @@ void tunnelsend(uint8_t * buf, uint16_t l, tunnelidt t)
        {
                tunnel[t].last = time_now; // control message sent
                tunnel[t].retry = backoff(tunnel[t].try); // when to resend
-               if (tunnel[t].try > 1)
+               if (tunnel[t].try)
                {
                        STAT(tunnel_retries);
                        LOG(3, 0, t, "Control message resend try %d\n", tunnel[t].try);
@@ -940,16 +998,71 @@ 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));
+}
+
 // process outgoing (to tunnel) IP
 //
-static void processipout(uint8_t * buf, int len)
+static void processipout(uint8_t *buf, int len)
 {
        sessionidt s;
        sessiont *sp;
        tunnelidt t;
        in_addr_t ip;
 
-       char *data = buf;       // Keep a copy of the originals.
+       uint8_t *data = buf;    // Keep a copy of the originals.
        int size = len;
 
        uint8_t b[MAXETHER + 20];
@@ -996,7 +1109,8 @@ static void processipout(uint8_t * buf, int len)
                if (rate++ < config->icmp_rate) // Only send a max of icmp_rate per second.
                {
                        LOG(4, 0, 0, "IP: Sending ICMP host unreachable to %s\n", fmtaddr(*(in_addr_t *)(buf + 12), 0));
-                       host_unreachable(*(in_addr_t *)(buf + 12), *(uint16_t *)(buf + 4), ip, buf, (len < 64) ? 64 : len);
+                       host_unreachable(*(in_addr_t *)(buf + 12), *(uint16_t *)(buf + 4),
+                               config->bind_address ? config->bind_address : my_address, buf, len);
                }
                return;
        }
@@ -1046,6 +1160,14 @@ static void processipout(uint8_t * buf, int len)
        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?
@@ -1055,7 +1177,8 @@ static void processipout(uint8_t * buf, int len)
                        master_throttle_packet(sp->tbf_out, data, size);
                return;
        }
-       else if (sp->walled_garden && !config->cluster_iam_master)
+
+       if (sp->walled_garden && !config->cluster_iam_master)
        {
                // We are walled-gardening this
                master_garden_packet(s, data, size);
@@ -1066,7 +1189,7 @@ static void processipout(uint8_t * buf, int len)
 
        // Add on L2TP header
        {
-               uint8_t *p = makeppp(b, sizeof(b), buf, len, t, s, PPPIP);
+               uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIP);
                if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
        }
@@ -1075,11 +1198,13 @@ static void processipout(uint8_t * buf, int len)
        if (sp->snoop_ip && sp->snoop_port)
                snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
 
-       sp->cout += len; // byte count
-       sp->total_cout += len; // byte count
+       increment_counter(&sp->cout, &sp->cout_wrap, len); // byte count
+       sp->cout_delta += len;
        sp->pout++;
        udp_tx += len;
+
        sess_local[s].cout += len;      // To send to master..
+       sess_local[s].pout++;
 }
 
 // process outgoing (to tunnel) IPv6
@@ -1092,7 +1217,7 @@ static void processipv6out(uint8_t * buf, int len)
        in_addr_t ip;
        struct in6_addr ip6;
 
-       char *data = buf;       // Keep a copy of the originals.
+       uint8_t *data = buf;    // Keep a copy of the originals.
        int size = len;
 
        uint8_t b[MAXETHER + 20];
@@ -1175,7 +1300,7 @@ static void processipv6out(uint8_t * buf, int len)
 
        // Add on L2TP header
        {
-               uint8_t *p = makeppp(b, sizeof(b), buf, len, t, s, PPPIPV6);
+               uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIPV6);
                if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
        }
@@ -1184,11 +1309,13 @@ static void processipv6out(uint8_t * buf, int len)
        if (sp->snoop_ip && sp->snoop_port)
                snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
 
-       sp->cout += len; // byte count
-       sp->total_cout += len; // byte count
+       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++;
 }
 
 //
@@ -1225,7 +1352,7 @@ static void send_ipout(sessionidt s, uint8_t *buf, int len)
 
        // Add on L2TP header
        {
-               uint8_t *p = makeppp(b, sizeof(b),  buf, len, t, s, PPPIP);
+               uint8_t *p = makeppp(b, sizeof(b), buf, len, s, t, PPPIP);
                if (!p) return;
                tunnelsend(b, len + (p-b), t); // send it...
        }
@@ -1234,11 +1361,13 @@ static void send_ipout(sessionidt s, uint8_t *buf, int len)
        if (sp->snoop_ip && sp->snoop_port)
                snoop_send_packet(buf, len, sp->snoop_ip, sp->snoop_port);
 
-       sp->cout += len; // byte count
-       sp->total_cout += len; // byte count
+       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)
@@ -1275,7 +1404,7 @@ static void controls(controlt * c, uint16_t avp, char *val, uint8_t m)
 }
 
 // add a binary AVP
-static void controlb(controlt * c, uint16_t avp, char *val, unsigned int len, uint8_t m)
+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);
@@ -1322,11 +1451,11 @@ static void controlnull(tunnelidt t)
 }
 
 // add a control message to a tunnel, and send if within window
-static void controladd(controlt * c, tunnelidt t, sessionidt s)
+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(s ? session[s].far : 0); // session
+       *(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
@@ -1394,7 +1523,7 @@ void throttle_session(sessionidt s, int rate_in, int rate_out)
 }
 
 // add/remove filters from session (-1 = no change)
-static void filter_session(sessionidt s, int filter_in, int filter_out)
+void filter_session(sessionidt s, int filter_in, int filter_out)
 {
        if (!session[s].opened)
                return; // No-one home.
@@ -1455,22 +1584,15 @@ void sessionshutdown(sessionidt s, char *reason, int result, int error)
        if (session[s].ip && !walled_garden && !session[s].die)
        {
                // RADIUS Stop message
-               uint16_t r = session[s].radius;
-               if (!r)
+               uint16_t r = radiusnew(s);
+               if (r)
                {
-                       if (!(r = radiusnew(s)))
-                       {
-                               LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");
-                               STAT(radius_overflow);
-                       }
-                       else
-                       {
-                               random_data(radius[r].auth, sizeof(radius[r].auth));
-                       }
+                       // stop, if not already trying
+                       if (radius[r].state != RADIUSSTOP)
+                               radiussend(r, RADIUSSTOP);
                }
-
-               if (r && radius[r].state != RADIUSSTOP)
-                       radiussend(r, RADIUSSTOP); // stop, if not already trying
+               else
+                       LOG(1, s, session[s].tunnel, "No free RADIUS sessions for Stop message\n");
 
                // Save counters to dump to accounting file
                if (*config->accounting_dir && shut_acct_n < sizeof(shut_acct) / sizeof(*shut_acct))
@@ -1500,7 +1622,7 @@ void sessionshutdown(sessionidt s, char *reason, int result, int error)
                        free_ip_address(s);
 
                // unroute IPv6, if setup
-               if (session[s].flags & SF_IPV6_ROUTED)
+               if (session[s].ppp.ipv6cp == Opened && session[s].ipv6prefixlen)
                        route6set(s, session[s].ipv6route, session[s].ipv6prefixlen, 0);
        }
 
@@ -1512,7 +1634,7 @@ void sessionshutdown(sessionidt s, char *reason, int result, int error)
                controlt *c = controlnew(14); // sending CDN
                if (error)
                {
-                       char buf[4];
+                       uint8_t buf[4];
                        *(uint16_t *) buf     = htons(result);
                        *(uint16_t *) (buf+2) = htons(error);
                        controlb(c, 1, buf, 4, 1);
@@ -1521,7 +1643,7 @@ void sessionshutdown(sessionidt s, char *reason, int result, int error)
                        control16(c, 1, result, 1);
 
                control16(c, 14, s, 1);   // assigned session (our end)
-               controladd(c, session[s].tunnel, s); // send the message
+               controladd(c, session[s].far, session[s].tunnel); // send the message
        }
 
        if (!session[s].die)
@@ -1531,72 +1653,81 @@ void sessionshutdown(sessionidt s, char *reason, int result, int error)
        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(tunnelidt t, sessionidt s)
+void sendipcp(sessionidt s, tunnelidt t)
 {
-       uint8_t buf[MAXCONTROL];
-       uint16_t r = session[s].radius;
+       uint8_t buf[MAXETHER];
        uint8_t *q;
 
        CSTAT(sendipcp);
+       LOG(3, s, t, "IPCP: send ConfigReq\n");
 
-       if (!r)
-               r = radiusnew(s);
-
-       if (radius[r].state != RADIUSIPCP)
-       {
-               radius[r].state = RADIUSIPCP;
-               radius[r].try = 0;
-       }
-
-       radius[r].retry = backoff(radius[r].try++);
-       if (radius[r].try > 10)
+       if (!session[s].unique_id)
        {
-               radiusclear(r, s);      // Clear radius session.
-               sessionshutdown(s, "No reply on IPCP", 3, 0);
-               return;
+               if (!++last_id) ++last_id; // skip zero
+               session[s].unique_id = last_id;
        }
 
-       q = makeppp(buf,sizeof(buf), 0, 0, t, s, PPPIPCP);
+       q = makeppp(buf, sizeof(buf), 0, 0, s, t, PPPIPCP);
        if (!q) return;
 
        *q = ConfigReq;
-       q[1] = r << RADIUS_SHIFT;                    // ID, dont care, we only send one type of request
-       *(uint16_t *) (q + 2) = htons(10);
-       q[4] = 3;
-       q[5] = 6;
+       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
-       session[s].flags &= ~SF_IPCP_ACKED;     // Clear flag.
-
-       // If we have an IPv6 prefix length configured, assume we should
-       // try to negotiate an IPv6 session as well. Unless we've had a
-       // (N)ACK for IPV6CP.
-       if (config->ipv6_prefix.s6_addr[0] > 0 && 
-                       !(session[s].flags & SF_IPV6CP_ACKED) &&
-                       !(session[s].flags & SF_IPV6_NACKED))
-       {
-               q = makeppp(buf,sizeof(buf), 0, 0, t, s, PPPIPV6CP);
-               if (!q) return;
-
-               *q = ConfigReq;
-               q[1] = r << RADIUS_SHIFT;               // ID, don't care, we
-                                                       // only send one type
-                                                       // of request
-               *(uint16_t *) (q + 2) = htons(14);
-               q[4] = 1;
-               q[5] = 10;
-               *(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, 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);
+       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
@@ -1616,16 +1747,11 @@ void sessionkill(sessionidt s, char *reason)
 
        session[s].die = TIME;
        sessionshutdown(s, reason, 3, 0);  // close radius/routes, etc.
-       if (session[s].radius)
-               radiusclear(session[s].radius, s); // cant send clean accounting data, session is killed
+       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);
-
-       memset(&session[s], 0, sizeof(session[s]));
-       session[s].tunnel = T_FREE;     // Mark it as free.
-       session[s].next = sessionfree;
-       sessionfree = s;
-       cli_session_actions[s].action = 0;
+       sessionclear(s);
        cluster_send_session(s);
 }
 
@@ -1685,7 +1811,7 @@ static void tunnelshutdown(tunnelidt t, char *reason, int result, int error, cha
        // close session
        for (s = 1; s <= config->cluster_highest_sessionid ; ++s)
                if (session[s].tunnel == t)
-                       sessionshutdown(s, reason, 3, 0);
+                       sessionshutdown(s, reason, 0, 0);
 
        tunnel[t].state = TUNNELDIE;
        tunnel[t].die = TIME + 700; // Clean up in 70 seconds
@@ -1696,7 +1822,7 @@ static void tunnelshutdown(tunnelidt t, char *reason, int result, int error, cha
                controlt *c = controlnew(4);    // sending StopCCN
                if (error)
                {
-                       char buf[64];
+                       uint8_t buf[64];
                        int l = 4;
                        *(uint16_t *) buf     = htons(result);
                        *(uint16_t *) (buf+2) = htons(error);
@@ -1716,14 +1842,14 @@ static void tunnelshutdown(tunnelidt t, char *reason, int result, int error, cha
                        control16(c, 1, result, 1);
 
                control16(c, 9, t, 1);          // assigned tunnel (our end)
-               controladd(c, t, 0);            // send the message
+               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)
+void processudp(uint8_t *buf, int len, struct sockaddr_in *addr)
 {
-       char *chapresponse = NULL;
+       uint8_t *chapresponse = NULL;
        uint16_t l = len, t = 0, s = 0, ns = 0, nr = 0;
        uint8_t *p = buf + 2;
 
@@ -1787,17 +1913,20 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                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;
-               uint8_t chap = 0;               // if CHAP being used
                uint16_t asession = 0;          // assigned session
                uint32_t amagic = 0;            // magic number
                uint8_t aflags = 0;             // flags from last LCF
                uint16_t version = 0x0100;      // protocol version (we handle 0.0 as well and send that back just in case)
-               int requestchap = 0;            // do we request PAP instead of original CHAP request?
                char called[MAXTEL] = "";       // called number
                char calling[MAXTEL] = "";      // calling number
 
@@ -1876,9 +2005,6 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                        return;
                }
 
-               // This is used to time out old tunnels
-               tunnel[t].lastrec = time_now;
-
                // check sequence of this message
                {
                        int skip = tunnel[t].window; // track how many in-window packets are still in queue
@@ -1968,7 +2094,7 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        uint16_t orig_len;
 
                                        // handle hidden AVPs
-                                       if (!*config->l2tpsecret)
+                                       if (!*config->l2tp_secret)
                                        {
                                                LOG(1, s, t, "Hidden AVP requested, but no L2TP secret.\n");
                                                fatal = flags;
@@ -1996,8 +2122,6 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                                continue;
                                        }
 
-                                       LOG(4, s, t, "Hidden AVP\n");
-
                                        // Unhide the AVP
                                        unhide_value(b, n, mtype, session[s].random_vector, session[s].random_vector_length);
 
@@ -2018,13 +2142,15 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        n = orig_len;
                                }
 
-                               LOG(4, s, t, "   AVP %d (%s) len %d\n", mtype, avp_name(mtype), n);
+                               LOG(4, s, t, "   AVP %d (%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 = %d (%s)\n", *b, l2tp_message_type(message));
+                                       LOG(4, s, t, "   Message type = %d (%s)\n", *b, l2tp_code(message));
                                        break;
                                case 1:     // result code
                                        {
@@ -2032,18 +2158,18 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                                const char* resdesc = "(unknown)";
                                                if (message == 4)
                                                { /* StopCCN */
-                                                       resdesc = stopccn_result_code(rescode);
+                                                       resdesc = l2tp_stopccn_result_code(rescode);
                                                }
                                                else if (message == 14)
                                                { /* CDN */
-                                                       resdesc = cdn_result_code(rescode);
+                                                       resdesc = l2tp_cdn_result_code(rescode);
                                                }
 
                                                LOG(4, s, t, "   Result Code %d: %s\n", rescode, resdesc);
                                                if (n >= 4)
                                                {
                                                        uint16_t errcode = ntohs(*(uint16_t *)(b + 2));
-                                                       LOG(4, s, t, "   Error Code %d: %s\n", errcode, error_code(errcode));
+                                                       LOG(4, s, t, "   Error Code %d: %s\n", errcode, l2tp_error_code(errcode));
                                                }
                                                if (n > 4)
                                                        LOG(4, s, t, "   Error String: %.*s\n", n-4, b+4);
@@ -2080,14 +2206,14 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
 //                                     LOG(4, s, t, "Firmware revision\n");
                                        break;
                                case 7:     // host name
-                                       memset(tunnel[t].hostname, 0, 128);
-                                       memcpy(tunnel[t].hostname, b, (n >= 127) ? 127 : n);
+                                       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) - 1) ? sizeof(tunnel[t].vendor) - 1 : n);
+                                       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
@@ -2127,13 +2253,13 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        // TBA
                                        break;
                                case 21:    // called number
-                                       memset(called, 0, MAXTEL);
-                                       memcpy(called, b, (n >= MAXTEL) ? (MAXTEL-1) : n);
+                                       memset(called, 0, sizeof(called));
+                                       memcpy(called, b, (n < sizeof(called)) ? n : sizeof(called) - 1);
                                        LOG(4, s, t, "   Called <%s>\n", called);
                                        break;
                                case 22:    // calling number
-                                       memset(calling, 0, MAXTEL);
-                                       memcpy(calling, b, (n >= MAXTEL) ? (MAXTEL-1) : n);
+                                       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
@@ -2146,8 +2272,9 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        else
                                        {
                                                // AS5300s send connect speed as a string
-                                               char tmp[30] = {0};
-                                               memcpy(tmp, b, (n >= 30) ? 30 : n);
+                                               char tmp[30];
+                                               memset(tmp, 0, sizeof(tmp));
+                                               memcpy(tmp, b, (n < sizeof(tmp)) ? n : sizeof(tmp) - 1);
                                                session[s].tx_connect_speed = atol(tmp);
                                        }
                                        LOG(4, s, t, "   TX connect speed <%u>\n", session[s].tx_connect_speed);
@@ -2160,8 +2287,9 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        else
                                        {
                                                // AS5300s send connect speed as a string
-                                               char tmp[30] = {0};
-                                               memcpy(tmp, b, (n >= 30) ? 30 : n);
+                                               char tmp[30];
+                                               memset(tmp, 0, sizeof(tmp));
+                                               memcpy(tmp, b, (n < sizeof(tmp)) ? n : sizeof(tmp) - 1);
                                                session[s].rx_connect_speed = atol(tmp);
                                        }
                                        LOG(4, s, t, "   RX connect speed <%u>\n", session[s].rx_connect_speed);
@@ -2175,21 +2303,20 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                case 29:    // Proxy Authentication Type
                                        {
                                                uint16_t atype = ntohs(*(uint16_t *)b);
-                                               LOG(4, s, t, "   Proxy Auth Type %d (%s)\n", atype, auth_type(atype));
-                                               requestchap = (atype == 2);
+                                               LOG(4, s, t, "   Proxy Auth Type %d (%s)\n", atype, ppp_auth_type(atype));
                                                break;
                                        }
                                case 30:    // Proxy Authentication Name
                                        {
-                                               char authname[64] = {0};
-                                               memcpy(authname, b, (n > 63) ? 63 : n);
+                                               char authname[64];
+                                               memset(authname, 0, sizeof(authname));
+                                               memcpy(authname, b, (n < sizeof(authname)) ? n : sizeof(authname) - 1);
                                                LOG(4, s, t, "   Proxy Auth Name (%s)\n",
                                                        authname);
                                                break;
                                        }
                                case 31:    // Proxy Authentication Challenge
                                        {
-                                               memcpy(radius[session[s].radius].auth, b, 16);
                                                LOG(4, s, t, "   Proxy Auth Challenge\n");
                                                break;
                                        }
@@ -2197,30 +2324,22 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        {
                                                uint16_t authid = ntohs(*(uint16_t *)(b));
                                                LOG(4, s, t, "   Proxy Auth ID (%d)\n", authid);
-                                               if (session[s].radius)
-                                                       radius[session[s].radius].id = authid;
                                                break;
                                        }
                                case 33:    // Proxy Authentication Response
-                                       {
-                                               char authresp[64] = {0};
-                                               memcpy(authresp, b, (n > 63) ? 63 : n);
-                                               LOG(4, s, t, "   Proxy Auth Response\n");
-                                               break;
-                                       }
-                               case 27:    // last send lcp
+                                       LOG(4, s, t, "   Proxy Auth Response\n");
+                                       break;
+                               case 27:    // last sent lcp
                                        {        // find magic number
                                                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 == 3 && p[1] == 5 && *(uint16_t *) (p + 2) == htons(PPPCHAP) && p[4] == 5) // Authentication-Protocol
-                                                               chap = 1;
                                                        else if (*p == 7) // Protocol-Field-Compression
-                                                               aflags |= SESSIONPFC;
+                                                               aflags |= SESSION_PFC;
                                                        else if (*p == 8) // Address-and-Control-Field-Compression
-                                                               aflags |= SESSIONACFC;
+                                                               aflags |= SESSION_ACFC;
                                                        p += p[1];
                                                }
                                        }
@@ -2234,6 +2353,8 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                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;
@@ -2256,16 +2377,21 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                switch (message)
                                {
                                case 1:       // SCCRQ - Start Control Connection Request
+                                       tunnel[t].state = TUNNELOPENING;
+                                       if (main_quit != QUIT_SHUTDOWN)
                                        {
                                                controlt *c = controlnew(2); // sending SCCRP
                                                control16(c, 2, version, 1); // protocol version
                                                control32(c, 3, 3, 1); // framing
-                                               controls(c, 7, tunnel[t].hostname, 1); // host name (TBA)
+                                               controls(c, 7, hostname, 1); // host name
                                                if (chapresponse) controlb(c, 13, chapresponse, 16, 1); // Challenge response
                                                control16(c, 9, t, 1); // assigned tunnel
-                                               controladd(c, t, s); // send the resply
+                                               controladd(c, 0, t); // send the resply
+                                       }
+                                       else
+                                       {
+                                               tunnelshutdown(t, "Shutting down", 6, 0, 0);
                                        }
-                                       tunnel[t].state = TUNNELOPENING;
                                        break;
                                case 2:       // SCCRP
                                        tunnel[t].state = TUNNELOPEN;
@@ -2291,16 +2417,9 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                        // TBA
                                        break;
                                case 10:      // ICRQ
-                                       if (!sessionfree)
+                                       if (sessionfree && main_quit != QUIT_SHUTDOWN)
                                        {
-                                               STAT(session_overflow);
-                                               LOG(1, 0, t, "No free sessions");
-                                               return;
-                                       }
-                                       else
-                                       {
-                                               uint16_t r;
-                                               controlt *c;
+                                               controlt *c = controlnew(11); // ICRP
 
                                                s = sessionfree;
                                                sessionfree = session[s].next;
@@ -2309,47 +2428,58 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                                if (s > config->cluster_highest_sessionid)
                                                        config->cluster_highest_sessionid = s;
 
-                                               // make a RADIUS session
-                                               if (!(r = radiusnew(s)))
-                                               {
-                                                       LOG(1, s, t, "No free RADIUS sessions for ICRQ\n");
-                                                       sessionkill(s, "no free RADIUS sesions");
-                                                       return;
-                                               }
-
-                                               c = controlnew(11); // sending ICRP
                                                session[s].opened = time_now;
                                                session[s].tunnel = t;
                                                session[s].far = asession;
                                                session[s].last_packet = time_now;
                                                LOG(3, s, t, "New session (%d/%d)\n", tunnel[t].far, session[s].far);
                                                control16(c, 14, s, 1); // assigned session
-                                               controladd(c, t, s); // send the reply
+                                               controladd(c, asession, t); // send the reply
 
-                                               // Generate a random challenge
-                                               random_data(radius[r].auth, sizeof(radius[r].auth));
-                                               strncpy(radius[r].calling, calling, sizeof(radius[r].calling) - 1);
                                                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;
                                        }
-                                       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].l2tp_flags = aflags; // set flags received
-                                       LOG(3, s, t, "Magic %X Flags %X\n", amagic, aflags);
+                                       session[s].flags = aflags; // set flags received
+                                       session[s].mru = PPPMTU; // default
                                        controlnull(t); // ack
-                                       // In CHAP state, request PAP instead
-                                       if (requestchap)
-                                               initlcp(t, s);
+
+                                       // start LCP
+                                       sess_local[s].lcp_authtype = config->radius_authprefer;
+                                       sess_local[s].ppp_mru = MRU;
+                                       sendlcp(s, t);
+                                       change_state(s, lcp, RequestSent);
                                        break;
+
                                case 14:      // CDN
                                        controlnull(t); // ack
-                                       sessionshutdown(s, "Closed (Received CDN)", 0, 0);
+                                       sessionshutdown(s, "Closed (Received CDN).", 0, 0);
                                        break;
                                case 0xFFFF:
                                        LOG(1, s, t, "Missing message type\n");
@@ -2372,7 +2502,7 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
        }
        else
        {                          // data
-               uint16_t prot;
+               uint16_t proto;
 
                LOG_HEX(5, "Receive Tunnel Data", p, l);
                if (l > 2 && p[0] == 0xFF && p[1] == 0x03)
@@ -2388,12 +2518,12 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                }
                if (*p & 1)
                {
-                       prot = *p++;
+                       proto = *p++;
                        l--;
                }
                else
                {
-                       prot = ntohs(*(uint16_t *) p);
+                       proto = ntohs(*(uint16_t *) p);
                        p += 2;
                        l -= 2;
                }
@@ -2413,50 +2543,43 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                        return;
                }
 
-               if (prot == PPPPAP)
+               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(t, s, p, l);
+                       processpap(s, t, p, l);
                }
-               else if (prot == PPPCHAP)
+               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(t, s, p, l);
+                       processchap(s, t, p, l);
                }
-               else if (prot == PPPLCP)
+               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(t, s, p, l);
+                       processlcp(s, t, p, l);
                }
-               else if (prot == PPPIPCP)
+               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(t, s, p, l);
+                       processipcp(s, t, p, l);
                }
-               else if (prot == PPPIPV6CP)
+               else if (proto == PPPIPV6CP && config->ipv6_prefix.s6_addr[0])
                {
-                       if (config->ipv6_prefix.s6_addr[0] > 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(t, s, p, l);
-                       }
-                       else
-                       {
-                               LOG(1, s, t, "IPv6 not configured; ignoring IPv6CP\n");
-                       }
+                       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 (prot == PPPCCP)
+               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(t, s, p, l);
+                       processccp(s, t, p, l);
                }
-               else if (prot == PPPIP)
+               else if (proto == PPPIP)
                {
                        if (session[s].die)
                        {
@@ -2471,15 +2594,10 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                return;
                        }
 
-                       processipin(t, s, p, l);
+                       processipin(s, t, p, l);
                }
-               else if (prot == PPPIPV6)
+               else if (proto == PPPIPV6 && config->ipv6_prefix.s6_addr[0])
                {
-                       if (!config->ipv6_prefix.s6_addr[0] > 0)
-                       {
-                               LOG(1, s, t, "IPv6 not configured; yet received IPv6 packet. Ignoring.\n");
-                               return;
-                       }
                        if (session[s].die)
                        {
                                LOG(4, s, t, "Session %d is closing.  Don't process PPP packets\n", s);
@@ -2493,12 +2611,38 @@ void processudp(uint8_t * buf, int len, struct sockaddr_in *addr)
                                return;
                        }
 
-                       processipv6in(t, s, p, l);
+                       processipv6in(s, t, p, l);
+               }
+               else if (session[s].ppp.lcp == Opened)
+               {
+                       uint8_t buf[MAXETHER];
+                       uint8_t *q;
+                       int mru = session[s].mru;
+                       if (mru > sizeof(buf)) mru = sizeof(buf);
+
+                       l += 6;
+                       if (l > mru) l = mru;
+
+                       q = makeppp(buf, sizeof(buf), 0, 0, s, t, PPPLCP);
+                       if (!q) return;
+
+                       *q = ProtocolRej;
+                       *(q + 1) = ++sess_local[s].lcp_ident;
+                       *(uint16_t *)(q + 2) = htons(l);
+                       *(uint16_t *)(q + 4) = htons(proto);
+                       memcpy(q + 6, p, l - 6);
+
+                       if (proto == PPPIPV6CP)
+                               LOG(3, s, t, "LCP: send ProtocolRej (IPV6CP: not configured)\n");
+                       else
+                               LOG(2, s, t, "LCP: sent ProtocolRej (0x%04X: unsupported)\n", proto);
+
+                       tunnelsend(buf, l + (q - buf), t);
                }
                else
                {
-                       STAT(tunnel_rx_errors);
-                       LOG(1, s, t, "Unknown PPP protocol %04X\n", prot);
+                       LOG(2, s, t, "Unknown PPP protocol 0x%04X received in LCP %s state\n",
+                               proto, ppp_state(session[s].ppp.lcp));
                }
        }
 }
@@ -2524,48 +2668,67 @@ static void processtun(uint8_t * buf, int len)
        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] > 0)
+           && config->ipv6_prefix.s6_addr[0])
                processipv6out(buf, len);
 
        // Else discard.
 }
 
-//
-// Maximum number of actions to complete.
-// This is to avoid sending out too many packets
-// at once.
-#define MAX_ACTIONS 500
-
-static int regular_cleanups(void)
+// 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)
 {
-       static sessionidt s = 0;        // Next session to check for actions on.
-       tunnelidt t;
-       int count=0,i;
-       uint16_t r;
-       static clockt next_acct = 0;
-       static clockt next_shut_acct = 0;
+       // 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;
 
-       LOG(3, 0, 0, "Begin regular cleanup\n");
+       // 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;
 
-       for (r = 1; r < MAXRADIUS; r++)
-       {
-               if (!radius[r].state)
-                       continue;
-               if (radius[r].retry)
-               {
-                       if (radius[r].retry <= TIME)
-                               radiusretry(r);
-               } else
-                       radius[r].retry = backoff(radius[r].try+1);     // Is this really needed? --mo
-       }
-       for (t = 1; t <= config->cluster_highest_tunnelid; t++)
+       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
@@ -2585,14 +2748,17 @@ static int regular_cleanups(void)
                                                tunnelsend(c->buf, c->length, t);
                                                c = c->next;
                                        }
+
+                               t_actions++;
                        }
                }
                // Send hello
-               if (tunnel[t].state == TUNNELOPEN && tunnel[t].lastrec < TIME + 600)
+               if (tunnel[t].state == TUNNELOPEN && !tunnel[t].controlc && (time_now - tunnel[t].lastrec) > 60)
                {
                        controlt *c = controlnew(6); // sending HELLO
-                       controladd(c, t, 0); // send the message
+                       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
@@ -2603,13 +2769,28 @@ static int regular_cleanups(void)
                        {
                                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++;
+               }
        }
 
-       count = 0;
-       for (i = 1; i <= config->cluster_highest_sessionid; i++)
+       for (i = 0; i < s_slice; i++)
        {
                s++;
                if (s > config->cluster_highest_sessionid)
@@ -2618,36 +2799,144 @@ static int regular_cleanups(void)
                if (!session[s].opened) // Session isn't in use
                        continue;
 
-               if (!session[s].die && session[s].ip && !(session[s].flags & SF_IPCP_ACKED))
+               // check for expired sessions
+               if (session[s].die)
                {
-                       // IPCP has not completed yet. Resend
-                       LOG(3, s, session[s].tunnel, "No ACK for initial IPCP ConfigReq... resending\n");
-                       sendipcp(session[s].tunnel, s);
+                       if (session[s].die <= TIME)
+                       {
+                               sessionkill(s, "Expired");
+                               s_actions++;
+                       }
+                       continue;
                }
 
-               // check for expired sessions
-               if (session[s].die && session[s].die <= TIME)
+               // PPP timeouts
+               if (sess_local[s].lcp.restart <= time_now)
                {
-                       sessionkill(s, "Expired");
-                       if (++count >= MAX_ACTIONS) break;
-                       continue;
+                       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.", 3, 0);
+                                       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.", 3, 0);
+                                       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", 3, 0);
+                       sessionshutdown(s, "No response to LCP ECHO requests.", 3, 0);
                        STAT(session_timeout);
-                       if (++count >= MAX_ACTIONS) break;
+                       s_actions++;
                        continue;
                }
 
                // No data in ECHO_TIMEOUT seconds, send LCP ECHO
-               if (session[s].user[0] && (time_now - session[s].last_packet >= ECHO_TIMEOUT))
+               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[MAXCONTROL] = {0};
+                       uint8_t b[MAXETHER];
 
-                       uint8_t *q = makeppp(b, sizeof(b), 0, 0, session[s].tunnel, s, PPPLCP);
+                       uint8_t *q = makeppp(b, sizeof(b), 0, 0, s, session[s].tunnel, PPPLCP);
                        if (!q) continue;
 
                        *q = EchoReq;
@@ -2658,7 +2947,8 @@ static int regular_cleanups(void)
                        LOG(4, s, session[s].tunnel, "No data in %d seconds, sending LCP ECHO\n",
                                        (int)(time_now - session[s].last_packet));
                        tunnelsend(b, 24, session[s].tunnel); // send it
-                       if (++count >= MAX_ACTIONS) break;
+                       sess_local[s].last_echo = time_now;
+                       s_actions++;
                }
 
                // Check for actions requested from the CLI
@@ -2670,8 +2960,9 @@ static int regular_cleanups(void)
                        if (a & CLI_SESS_KILL)
                        {
                                LOG(2, s, session[s].tunnel, "Dropping session by CLI\n");
-                               sessionshutdown(s, "Requested by administrator", 3, 0);
+                               sessionshutdown(s, "Requested by administrator.", 3, 0);
                                a = 0; // dead, no need to check for other actions
+                               s_actions++;
                        }
 
                        if (a & CLI_SESS_NOSNOOP)
@@ -2679,6 +2970,7 @@ static int regular_cleanups(void)
                                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)
@@ -2689,6 +2981,7 @@ static int regular_cleanups(void)
 
                                session[s].snoop_ip = cli_session_actions[s].snoop_ip;
                                session[s].snoop_port = cli_session_actions[s].snoop_port;
+                               s_actions++;
                                send++;
                        }
 
@@ -2696,6 +2989,7 @@ static int regular_cleanups(void)
                        {
                                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)
@@ -2705,6 +2999,7 @@ static int regular_cleanups(void)
                                    cli_session_actions[s].throttle_out);
 
                                throttle_session(s, cli_session_actions[s].throttle_in, cli_session_actions[s].throttle_out);
+                               s_actions++;
                                send++;
                        }
 
@@ -2712,6 +3007,7 @@ static int regular_cleanups(void)
                        {
                                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)
@@ -2721,42 +3017,41 @@ static int regular_cleanups(void)
                                    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);
-
-                       if (++count >= MAX_ACTIONS) break;
                }
-       }
 
-       if (*config->accounting_dir)
-       {
-               if (next_acct <= TIME)
-               {
-                       // Dump accounting data
-                       next_acct = TIME + ACCT_TIME;
-                       next_shut_acct = TIME + ACCT_SHUT_TIME;
-                       dump_acct_info(1);
-               }
-               else if (next_shut_acct <= TIME)
+               // 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)
                {
-                       // Dump accounting data for shutdown sessions
-                       next_shut_acct = TIME + ACCT_SHUT_TIME;
-                       if (shut_acct_n)
-                               dump_acct_info(0);
+                       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++;
                }
        }
 
-       if (count >= MAX_ACTIONS)
-               return 1;       // Didn't finish!
-
-       LOG(3, 0, 0, "End regular cleanup (%d actions), next in %d seconds\n", count, config->cleanup_interval);
-       return 0;
+       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??
@@ -2766,6 +3061,53 @@ static int still_busy(void)
        int i;
        static clockt last_talked = 0;
        static clockt start_busy_wait = 0;
+
+       if (!config->cluster_iam_master)
+       {
+#ifdef BGP
+               static time_t stopped_bgp = 0;
+               if (bgp_configured)
+               {
+                       if (!stopped_bgp)
+                       {
+                               LOG(1, 0, 0, "Shutting down in %d seconds, stopping BGP...\n", QUIT_DELAY);
+
+                               for (i = 0; i < BGP_NUM_PEERS; i++)
+                                       if (bgp_peers[i].state == Established)
+                                               bgp_stop(&bgp_peers[i]);
+
+                               stopped_bgp = time_now;
+
+                               // we don't want to become master
+                               cluster_send_ping(0);
+
+                               return 1;
+                       }
+
+                       if (time_now < (stopped_bgp + QUIT_DELAY))
+                               return 1;
+               }
+#endif /* BGP */
+
+               return 0;
+       }
+
+       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;
 
@@ -2807,80 +3149,116 @@ static int still_busy(void)
        return 0;
 }
 
-static fd_set readset;
-static int readset_n = 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];
-       struct timeval to;
        clockt next_cluster_ping = 0;   // send initial ping immediately
-       time_t next_clean = time_now + config->cleanup_interval;
-
-       LOG(4, 0, 0, "Beginning of main loop.  udpfd=%d, tunfd=%d, cluster_sockfd=%d, controlfd=%d\n",
-               udpfd, tunfd, cluster_sockfd, controlfd);
-
-       FD_ZERO(&readset);
-       FD_SET(udpfd, &readset);
-       FD_SET(tunfd, &readset);
-       FD_SET(controlfd, &readset);
-       FD_SET(clifd, &readset);
-       if (cluster_sockfd) FD_SET(cluster_sockfd, &readset);
-       readset_n = udpfd;
-       if (tunfd > readset_n)          readset_n = tunfd;
-       if (controlfd > readset_n)      readset_n = controlfd;
-       if (clifd > readset_n)          readset_n = clifd;
-       if (cluster_sockfd > readset_n) readset_n = cluster_sockfd;
+       struct epoll_event events[BASE_FDS + RADIUS_FDS + EXTRA_FDS];
+       int maxevent = sizeof(events)/sizeof(*events);
 
-       while (!main_quit || still_busy())
+       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 */
        {
-               fd_set r;
-               int n = readset_n;
+               static struct event_data d[BASE_FDS];
+               struct epoll_event e;
+
+               e.events = EPOLLIN;
+               i = 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
-               fd_set w;
-               int bgp_set[BGP_NUM_PEERS];
+       signal(SIGPIPE, SIG_IGN);
+       bgp_setup(config->as_number);
+       if (config->bind_address)
+               bgp_add_route(config->bind_address, 0xffffffff);
+
+       for (i = 0; i < BGP_NUM_PEERS; i++)
+       {
+               if (config->neighbour[i].name[0])
+                       bgp_start(&bgp_peers[i], config->neighbour[i].name,
+                               config->neighbour[i].as, config->neighbour[i].keepalive,
+                               config->neighbour[i].hold, 0); /* 0 = routing disabled */
+       }
 #endif /* BGP */
 
-               if (config->reload_config)
-               {
-                       // Update the config state based on config settings
-                       update_config();
-               }
+       while (!main_quit || still_busy())
+       {
+               int more = 0;
+               int n;
 
-               memcpy(&r, &readset, sizeof(fd_set));
-               to.tv_sec = 0;
-               to.tv_usec = 100000; // 1/10th of a second.
 
-#ifdef BGP
-               FD_ZERO(&w);
-               for (i = 0; i < BGP_NUM_PEERS; i++)
+               if (main_reload)
                {
-                       bgp_set[i] = bgp_select_state(&bgp_peers[i]);
-                       if (bgp_set[i] & 1)
-                       {
-                               FD_SET(bgp_peers[i].sock, &r);
-                               if (bgp_peers[i].sock > n)
-                                       n = bgp_peers[i].sock;
-                       }
+                       main_reload = 0;
+                       read_config_file();
+                       config->reload_config++;
+               }
 
-                       if (bgp_set[i] & 2)
-                       {
-                               FD_SET(bgp_peers[i].sock, &w);
-                               if (bgp_peers[i].sock > n)
-                                       n = bgp_peers[i].sock;
-                       }
+               if (config->reload_config)
+               {
+                       config->reload_config = 0;
+                       update_config();
                }
 
-               n = select(n + 1, &r, &w, 0, &to);
-#else /* BGP */
-               n = select(n + 1, &r, 0, 0, &to);
+#ifdef BGP
+               bgp_set_poll();
 #endif /* BGP */
 
+               n = epoll_wait(epollfd, events, maxevent, 100); // timeout 100ms (1/10th sec)
                STAT(select_called);
 
-               TIME = now();
+               TIME = now(NULL);
                if (n < 0)
                {
                        if (errno == EINTR ||
@@ -2888,70 +3266,92 @@ static void mainloop(void)
                                continue;
 
                        LOG(0, 0, 0, "Error returned from select(): %s\n", strerror(errno));
-                       main_quit++;
-                       break;
+                       break; // exit
                }
-               else if (n)
+
+               if (n)
                {
                        struct sockaddr_in addr;
-                       int alen, c, s;
+                       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 */
 
-                       // nsctl commands
-                       if (FD_ISSET(controlfd, &r))
-                       {
-                               alen = sizeof(addr);
-                               processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr, alen);
-                               n--;
-                       }
-
-                       // RADIUS responses
-                       if (config->cluster_iam_master)
+                       for (c = n, i = 0; i < c; i++)
                        {
-                               for (i = 0; i < config->num_radfds; i++)
+                               struct event_data *d = events[i].data.ptr;
+                               switch (d->type)
                                {
-                                       if (FD_ISSET(radfds[i], &r))
+                               case FD_TYPE_CLI: // CLI connections
+                               {
+                                       int cli;
+                                       
+                                       alen = sizeof(addr);
+                                       if ((cli = accept(clifd, (struct sockaddr *)&addr, &alen)) >= 0)
                                        {
-                                               processrad(buf, recv(radfds[i], buf, sizeof(buf), 0), i);
-                                               n--;
+                                               cli_do(cli);
+                                               close(cli);
                                        }
-                               }
-                       }
+                                       else
+                                               LOG(0, 0, 0, "accept error: %s\n", strerror(errno));
 
-                       // CLI connections
-                       if (FD_ISSET(clifd, &r))
-                       {
-                               int cli;
-                               
-                               alen = sizeof(addr);
-                               if ((cli = accept(clifd, (struct sockaddr *)&addr, &alen)) >= 0)
-                               {
-                                       cli_do(cli);
-                                       close(cli);
+                                       n--;
+                                       break;
                                }
-                               else
-                                       LOG(0, 0, 0, "accept error: %s\n", strerror(errno));
 
-                               n--;
-                       }
+                               // 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);
+                                       processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr, alen);
+                                       n--;
+                                       break;
+
+                               case FD_TYPE_DAE: // DAE requests
+                                       alen = sizeof(addr);
+                                       processdae(buf, recvfrom(daefd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr, alen);
+                                       n--;
+                                       break;
+
+                               case FD_TYPE_RADIUS: // RADIUS response
+                                       s = recv(radfds[d->index], buf, sizeof(buf), 0);
+                                       if (s >= 0 && config->cluster_iam_master)
+                                               processrad(buf, s, d->index);
+
+                                       n--;
+                                       break;
 
 #ifdef BGP
-                       for (i = 0; i < BGP_NUM_PEERS; i++)
-                       {
-                               int isr = bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &r) : 0;
-                               int isw = bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &w) : 0;
-                               bgp_process(&bgp_peers[i], isr, isw);
-                               if (isr) n--;
-                               if (isw) n--;
+                               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 (FD_ISSET(udpfd, &r))
+                               if (udp_ready)
                                {
                                        alen = sizeof(addr);
                                        if ((s = recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen)) > 0)
@@ -2961,13 +3361,13 @@ static void mainloop(void)
                                        }
                                        else
                                        {
-                                               FD_CLR(udpfd, &r);
+                                               udp_ready = 0;
                                                n--;
                                        }
                                }
 
                                // incoming IP
-                               if (FD_ISSET(tunfd, &r))
+                               if (tun_ready)
                                {
                                        if ((s = read(tunfd, buf, sizeof(buf))) > 0)
                                        {
@@ -2976,13 +3376,13 @@ static void mainloop(void)
                                        }
                                        else
                                        {
-                                               FD_CLR(tunfd, &r);
+                                               tun_ready = 0;
                                                n--;
                                        }
                                }
 
                                // cluster
-                               if (FD_ISSET(cluster_sockfd, &r))
+                               if (cluster_ready)
                                {
                                        alen = sizeof(addr);
                                        if ((s = recvfrom(cluster_sockfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen)) > 0)
@@ -2992,7 +3392,7 @@ static void mainloop(void)
                                        }
                                        else
                                        {
-                                               FD_CLR(cluster_sockfd, &r);
+                                               cluster_ready = 0;
                                                n--;
                                        }
                                }
@@ -3007,11 +3407,41 @@ static void mainloop(void)
                                        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 (cluster_sockfd && next_cluster_ping <= TIME)
+               if (next_cluster_ping <= TIME)
                {
                        // Check to see which of the cluster is still alive..
 
@@ -3029,9 +3459,11 @@ static void mainloop(void)
                                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.
-                       // Runs on all machines both master and slave.
                {
                        static clockt last_run = 0;
                        if (last_run != TIME)
@@ -3041,20 +3473,42 @@ static void mainloop(void)
                        }
                }
 
-               /* Handle timeouts. Make sure that this gets run anyway, even if there was
-                * something to read, else under load this will never actually run....
-                *
-                */
-               if (config->cluster_iam_master && next_clean <= time_now)
+                       // Handle timeouts, retries etc.
                {
-                       if (regular_cleanups())
+                       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)
                        {
-                               // Did it finish?
-                               next_clean = time_now + 1 ;     // Didn't finish. Check quickly.
+                               regular_cleanups(diff);
+                               last_clean = this_clean;
                        }
-                       else
+               }
+
+               if (*config->accounting_dir)
+               {
+                       static clockt next_acct = 0;
+                       static clockt next_shut_acct = 0;
+
+                       if (next_acct <= TIME)
                        {
-                               next_clean = time_now + config->cleanup_interval; // Did. Move to next interval.
+                               // 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);
                        }
                }
        }
@@ -3069,6 +3523,7 @@ static void mainloop(void)
 
        //
        // Important!!! We MUST not process any packets past this point!
+       LOG(1, 0, 0, "Shutdown complete\n");
 }
 
 static void stripdomain(char *host)
@@ -3141,24 +3596,41 @@ static void initdata(int optdebug, char *optconfig)
 {
        int i;
 
-       if (!(_statistics = shared_malloc(sizeof(struct Tstats))))
-       {
-               LOG(0, 0, 0, "Error doing malloc for _statistics: %s\n", strerror(errno));
-               exit(1);
-       }
        if (!(config = shared_malloc(sizeof(configt))))
        {
-               LOG(0, 0, 0, "Error doing malloc for configuration: %s\n", strerror(errno));
+               fprintf(stderr, "Error doing malloc for configuration: %s\n", strerror(errno));
                exit(1);
        }
+
        memset(config, 0, sizeof(configt));
        time(&config->start_time);
        strncpy(config->config_file, optconfig, strlen(optconfig));
        config->debug = optdebug;
        config->num_tbfs = MAXTBFS;
        config->rl_rate = 28; // 28kbps
+       config->cluster_mcast_ttl = 1;
+       config->cluster_master_min_adv = 1;
+       config->ppp_restart_time = 3;
+       config->ppp_max_configure = 10;
+       config->ppp_max_failure = 5;
        strcpy(config->random_device, RANDOMDEVICE);
 
+       log_stream = stderr;
+
+#ifdef RINGBUFFER
+       if (!(ringbuffer = shared_malloc(sizeof(struct Tringbuffer))))
+       {
+               LOG(0, 0, 0, "Error doing malloc for ringbuffer: %s\n", strerror(errno));
+               exit(1);
+       }
+       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
+#endif
+
+       if (!(_statistics = shared_malloc(sizeof(struct Tstats))))
+       {
+               LOG(0, 0, 0, "Error doing malloc for _statistics: %s\n", strerror(errno));
+               exit(1);
+       }
        if (!(tunnel = shared_malloc(sizeof(tunnelt) * MAXTUNNEL)))
        {
                LOG(0, 0, 0, "Error doing malloc for tunnels: %s\n", strerror(errno));
@@ -3195,15 +3667,6 @@ static void initdata(int optdebug, char *optconfig)
        }
        memset(ip_filters, 0, sizeof(ip_filtert) * MAXFILTER);
 
-#ifdef RINGBUFFER
-       if (!(ringbuffer = shared_malloc(sizeof(struct Tringbuffer))))
-       {
-               LOG(0, 0, 0, "Error doing malloc for ringbuffer: %s\n", strerror(errno));
-               exit(1);
-       }
-       memset(ringbuffer, 0, sizeof(struct Tringbuffer));
-#endif
-
        if (!(cli_session_actions = shared_malloc(sizeof(struct cli_session_actions) * MAXSESSION)))
        {
                LOG(0, 0, 0, "Error doing malloc for cli session actions: %s\n", strerror(errno));
@@ -3517,7 +3980,7 @@ static void initippool()
        LOG(1, 0, 0, "IP address pool is %d addresses\n", ip_pool_size - 1);
 }
 
-void snoop_send_packet(char *packet, uint16_t size, in_addr_t destination, uint16_t port)
+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)
@@ -3539,7 +4002,7 @@ void snoop_send_packet(char *packet, uint16_t size, in_addr_t destination, uint1
 
 static int dump_session(FILE **f, sessiont *s)
 {
-       if (!s->opened || !s->ip || !(s->cin || s->cout) || !*s->user || s->walled_garden)
+       if (!s->opened || !s->ip || !(s->cin_delta || s->cout_delta) || !*s->user || s->walled_garden)
                return 1;
 
        if (!*f)
@@ -3560,10 +4023,12 @@ static int dump_session(FILE **f, sessiont *s)
                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);
        }
@@ -3573,11 +4038,10 @@ static int dump_session(FILE **f, sessiont *s)
                s->user,                                                // username
                fmtaddr(htonl(s->ip), 0),                               // ip
                (s->throttle_in || s->throttle_out) ? 2 : 1,            // qos
-               (uint32_t) s->cin,                                      // uptxoctets
-               (uint32_t) s->cout);                                    // downrxoctets
+               (uint32_t) s->cin_delta,                                // uptxoctets
+               (uint32_t) s->cout_delta);                              // downrxoctets
 
-       s->pin = s->cin = 0;
-       s->pout = s->cout = 0;
+       s->cin_delta = s->cout_delta = 0;
 
        return 1;
 }
@@ -3651,14 +4115,13 @@ int main(int argc, char *argv[])
        // Start the timer routine off
        time(&time_now);
        strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
-       signal(SIGALRM, sigalrm_handler);
-       siginterrupt(SIGALRM, 0);
 
        initplugins();
        initdata(optdebug, optconfig);
 
        init_cli(hostname);
        read_config_file();
+       update_config();
        init_tbf(config->num_tbfs);
 
        LOG(0, 0, 0, "L2TPNS version " VERSION "\n");
@@ -3705,19 +4168,6 @@ int main(int argc, char *argv[])
        if (cluster_init() < 0)
                exit(1);
 
-#ifdef BGP
-       signal(SIGPIPE, SIG_IGN);
-       bgp_setup(config->as_number);
-       bgp_add_route(config->bind_address, 0xffffffff);
-       for (i = 0; i < BGP_NUM_PEERS; i++)
-       {
-               if (config->neighbour[i].name[0])
-                       bgp_start(&bgp_peers[i], config->neighbour[i].name,
-                               config->neighbour[i].as, config->neighbour[i].keepalive,
-                               config->neighbour[i].hold, 0); /* 0 = routing disabled */
-       }
-#endif /* BGP */
-
        inittun();
        LOG(1, 0, 0, "Set up on interface %s\n", config->tundevice);
 
@@ -3725,11 +4175,18 @@ int main(int argc, char *argv[])
        initrad();
        initippool();
 
-       signal(SIGHUP, sighup_handler);
-       signal(SIGTERM, sigterm_handler);
-       signal(SIGINT, sigterm_handler);
-       signal(SIGQUIT, sigquit_handler);
+       // 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)
@@ -3740,22 +4197,12 @@ int main(int argc, char *argv[])
                        LOG(0, 0, 0, "Can't lock pages: %s\n", strerror(errno));
        }
 
-       alarm(1);
-
        // Drop privileges here
        if (config->target_uid > 0 && geteuid() == 0)
                setuid(config->target_uid);
 
        mainloop();
 
-#ifdef BGP
-       /* try to shut BGP down cleanly; with luck the sockets will be
-          writable since we're out of the select */
-       for (i = 0; i < BGP_NUM_PEERS; i++)
-               if (bgp_peers[i].state == Established)
-                       bgp_stop(&bgp_peers[i]);
-#endif /* BGP */
-
        /* remove plugins (so cleanup code gets run) */
        plugins_done();
 
@@ -3771,75 +4218,12 @@ int main(int argc, char *argv[])
 
 static void sighup_handler(int sig)
 {
-       if (log_stream && log_stream != stderr)
-       {
-               fclose(log_stream);
-               log_stream = NULL;
-       }
-
-       read_config_file();
-}
-
-static void sigalrm_handler(int sig)
-{
-       // Log current traffic stats
-
-       snprintf(config->bandwidth, sizeof(config->bandwidth),
-               "UDP-ETH:%1.0f/%1.0f  ETH-UDP:%1.0f/%1.0f  TOTAL:%0.1f   IN:%u OUT:%u",
-               (udp_rx / 1024.0 / 1024.0 * 8),
-               (eth_tx / 1024.0 / 1024.0 * 8),
-               (eth_rx / 1024.0 / 1024.0 * 8),
-               (udp_tx / 1024.0 / 1024.0 * 8),
-               ((udp_tx + udp_rx + eth_tx + eth_rx) / 1024.0 / 1024.0 * 8),
-               udp_rx_pkt, eth_rx_pkt);
-
-       udp_tx = udp_rx = 0;
-       udp_rx_pkt = eth_rx_pkt = 0;
-       eth_tx = eth_rx = 0;
-
-       if (config->dump_speed)
-               printf("%s\n", config->bandwidth);
-
-       // Update the internal time counter
-       time(&time_now);
-       strftime(time_now_string, sizeof(time_now_string), "%Y-%m-%d %H:%M:%S", localtime(&time_now));
-       alarm(1);
-
-       {
-               // Run timer hooks
-               struct param_timer p = { time_now };
-               run_plugins(PLUGIN_TIMER, &p);
-       }
-
+       main_reload++;
 }
 
-static void sigterm_handler(int sig)
+static void shutdown_handler(int sig)
 {
-       LOG(1, 0, 0, "Shutting down cleanly\n");
-       main_quit++;
-}
-
-static void sigquit_handler(int sig)
-{
-       int i;
-
-       LOG(1, 0, 0, "Shutting down without saving sessions\n");
-
-       if (config->cluster_iam_master)
-       {
-               for (i = 1; i < MAXSESSION; i++)
-               {
-                       if (session[i].opened)
-                               sessionkill(i, "L2TPNS Closing");
-               }
-               for (i = 1; i < MAXTUNNEL; i++)
-               {
-                       if (tunnel[i].ip || tunnel[i].state)
-                               tunnelshutdown(i, "L2TPNS Closing", 6, 0, 0);
-               }
-       }
-
-       main_quit++;
+       main_quit = (sig == SIGQUIT) ? QUIT_SHUTDOWN : QUIT_FAILOVER;
 }
 
 static void sigchild_handler(int sig)
@@ -3848,12 +4232,12 @@ static void sigchild_handler(int sig)
            ;
 }
 
-static void build_chap_response(char *challenge, uint8_t id, uint16_t challenge_length, char **challenge_response)
+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->l2tpsecret)
+       if (!*config->l2tp_secret)
        {
                LOG(0, 0, 0, "LNS requested CHAP authentication, but no l2tp secret is defined\n");
                return;
@@ -3861,13 +4245,13 @@ static void build_chap_response(char *challenge, uint8_t id, uint16_t challenge_
 
        LOG(4, 0, 0, "   Building challenge response for CHAP request\n");
 
-       *challenge_response = (char *)calloc(17, 1);
+       *challenge_response = calloc(17, 1);
 
-       MD5Init(&ctx);
-       MD5Update(&ctx, &id, 1);
-       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
-       MD5Update(&ctx, challenge, challenge_length);
-       MD5Final(*challenge_response, &ctx);
+       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;
 }
@@ -3895,7 +4279,9 @@ static void update_config()
        syslog_log = 0;
        if (log_stream)
        {
-               fclose(log_stream);
+               if (log_stream != stderr)
+                       fclose(log_stream);
+
                log_stream = NULL;
        }
 
@@ -3930,6 +4316,17 @@ static void update_config()
                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 = PPPMTU;
+       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;
+       MSS = MRU - TCP_HDRS;
+
        // Update radius
        config->numradiusservers = 0;
        for (i = 0; i < MAXRADSERVER; i++)
@@ -3943,7 +4340,7 @@ static void update_config()
                        // 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])
+                       if (i > 0 && !config->radiusport[i])
                                config->radiusport[i] = config->radiusport[i-1];
                        if (!config->radiusport[i])
                                config->radiusport[i] = RADPORT;
@@ -3952,12 +4349,10 @@ static void update_config()
        if (!config->numradiusservers)
                LOG(0, 0, 0, "No RADIUS servers defined!\n");
 
-       config->num_radfds = 2 << RADIUS_SHIFT;
-
        // parse radius_authtypes_s
        config->radius_authtypes = config->radius_authprefer = 0;
        p = config->radius_authtypes_s;
-       while (*p)
+       while (p && *p)
        {
                char *s = strpbrk(p, " \t,");
                int type = 0;
@@ -3977,11 +4372,13 @@ static void update_config()
                else if (!strncasecmp("pap", p, strlen(p)))
                        type = AUTHPAP;
                else
-                       LOG(0, 0, 0, "Invalid RADIUS authentication type \"%s\"", p);
+                       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)
@@ -4004,6 +4401,9 @@ static void update_config()
                        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);
 
@@ -4026,7 +4426,6 @@ static void update_config()
        }
 
        memcpy(config->old_plugins, config->plugins, sizeof(config->plugins));
-       if (!config->cleanup_interval) config->cleanup_interval = 10;
        if (!config->multi_read_count) config->multi_read_count = 10;
        if (!config->cluster_address) config->cluster_address = inet_addr(DEFAULT_MCAST_ADDR);
        if (!*config->cluster_interface)
@@ -4073,8 +4472,6 @@ static void update_config()
                        LOG(0, 0, 0, "Can't write to PID file %s: %s\n", config->pid_file, strerror(errno));
                }
        }
-
-       config->reload_config = 0;
 }
 
 static void read_config_file()
@@ -4092,10 +4489,9 @@ static void read_config_file()
        cli_do_file(f);
        LOG(3, 0, 0, "Done reading config file\n");
        fclose(f);
-       update_config();
 }
 
-int sessionsetup(tunnelidt t, sessionidt s)
+int sessionsetup(sessionidt s, tunnelidt t)
 {
        // A session now exists, set it up
        in_addr_t ip;
@@ -4113,7 +4509,7 @@ int sessionsetup(tunnelidt t, sessionidt 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", 2, 7);
+                       sessionshutdown(s, "No IP addresses available.", 2, 7); // try another
                        return 0;
                }
                LOG(3, s, t, "   No IP allocated.  Assigned %s from pool\n",
@@ -4133,8 +4529,16 @@ int sessionsetup(tunnelidt t, sessionidt s)
                for (i = 1; i <= config->cluster_highest_sessionid; i++)
                {
                        if (i == s) continue;
-                       if (ip == session[i].ip) sessionkill(i, "Duplicate IP address");
-                       if (!session[s].walled_garden && !session[i].walled_garden && strcasecmp(user, session[i].user) == 0)
+                       if (!session[s].opened) 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;
+                       if (!strcasecmp(user, session[i].user))
                                sessionkill(i, "Duplicate session for users");
                }
        }
@@ -4165,13 +4569,8 @@ int sessionsetup(tunnelidt t, sessionidt s)
                        cache_ipmap(session[s].ip, s);
        }
 
-       if (!session[s].unique_id)
-       {
-               // did this session just finish radius?
-               LOG(3, s, t, "Sending initial IPCP to client\n");
-               sendipcp(t, s);
-               session[s].unique_id = ++last_id;
-       }
+       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.
        {
@@ -4280,7 +4679,7 @@ int load_session(sessionidt s, sessiont *new)
        }
 
        // check v6 routing
-       if (new->flags & SF_IPV6_ROUTED && !(session[s].flags & SF_IPV6_ROUTED))
+       if (new->ipv6prefixlen && new->ppp.ipv6cp == Opened && session[s].ppp.ipv6cp != Opened)
                    route6set(s, new->ipv6route, new->ipv6prefixlen, 1);
 
        // check filters
@@ -4312,12 +4711,6 @@ int load_session(sessionidt s, sessiont *new)
                                        // for walking the sessions to forward byte counts to the master.
                config->cluster_highest_sessionid = s;
 
-       // TEMP: old session struct used a uint32_t to define the throttle
-       // speed for both up/down, new uses a uint16_t for each.  Deal with
-       // sessions from an old master for migration.
-       if (new->throttle_out == 0 && new->tbf_out)
-               new->throttle_out = new->throttle_in;
-
        memcpy(&session[s], new, sizeof(session[s]));   // Copy over..
 
                // Do fixups into address pool.
@@ -4381,6 +4774,7 @@ static int add_plugin(char *plugin_name)
                radiusnew,
                radiussend,
                getconfig,
+               sessionshutdown,
                sessionkill,
                throttle_session,
                cluster_send_session,
@@ -4502,7 +4896,7 @@ static void plugins_done()
                run_plugin_done(p);
 }
 
-static void processcontrol(uint8_t * buf, int len, struct sockaddr_in *addr, int alen)
+static void processcontrol(uint8_t *buf, int len, struct sockaddr_in *addr, int alen)
 {
        struct nsctl request;
        struct nsctl response;
@@ -4701,6 +5095,9 @@ static tunnelidt new_tunnel()
 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
@@ -4719,11 +5116,14 @@ void become_master(void)
        }
 
        // add radius fds
-       for (i = 0; i < config->num_radfds; i++)
+       e.events = EPOLLIN;
+       for (i = 0; i < RADIUS_FDS; i++)
        {
-               FD_SET(radfds[i], &readset);
-               if (radfds[i] > readset_n)
-                       readset_n = radfds[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);
        }
 }
 
@@ -4817,14 +5217,14 @@ static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vec
        uint8_t digest[16];
        uint8_t *last;
        size_t d = 0;
+       uint16_t m = htons(type);
 
        // Compute initial pad
-       MD5Init(&ctx);
-       MD5Update(&ctx, (uint8_t) (type >> 8) & 0xff, 1);
-       MD5Update(&ctx, (uint8_t)  type       & 0xff, 1);
-       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
-       MD5Update(&ctx, vector, vec_len);
-       MD5Final(digest, &ctx);
+       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;
@@ -4834,10 +5234,10 @@ static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vec
                // calculate a new pad based on the last decoded block
                if (d >= sizeof(digest))
                {
-                       MD5Init(&ctx);
-                       MD5Update(&ctx, config->l2tpsecret, strlen(config->l2tpsecret));
-                       MD5Update(&ctx, last, sizeof(digest));
-                       MD5Final(digest, &ctx);
+                       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;
@@ -4848,6 +5248,31 @@ static void unhide_value(uint8_t *value, size_t len, uint16_t type, uint8_t *vec
        }
 }
 
+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)
@@ -4933,7 +5358,9 @@ int ip_filter(uint8_t *buf, int len, uint8_t filter)
 
                if (frag_offset)
                {
-                       if (!rule->frag || rule->action == FILTER_ACTION_DENY)
+                       // 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