- Ignore gateway address in Framed-Route (from Jonathan McDowell).

[l2tpns.git] / cluster.c

// L2TPNS Clustering Stuff

char const *cvs_id_cluster = "$Id: cluster.c,v 1.18 2004/11/16 07:54:32 bodea Exp $";

#include <stdio.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <sys/ioctl.h>
#include <net/if.h>
#include <string.h>
#include <malloc.h>
#include <errno.h>
#include <stdlib.h>
#include <stdarg.h>
#include <unistd.h>
#include <stdio.h>
#include <libcli.h>

#include "l2tpns.h"
#include "cluster.h"
#include "util.h"
#include "tbf.h"

#ifdef BGP
#include "bgp.h"
#endif
/*
 * All cluster packets have the same format.
 *
 * One or more instances of
 *      a 32 bit 'type' id.
 *      a 32 bit 'extra' data dependant on the 'type'.
 *      zero or more bytes of structure data, dependant on the type.
 *
 */

// Module variables.
int cluster_sockfd = 0;         // The filedescriptor for the cluster communications port.

ipt my_address = 0;             // The network address of my ethernet port.
static int walk_session_number = 0;     // The next session to send when doing the slow table walk.
static int walk_tunnel_number = 0;      // The next tunnel to send when doing the slow table walk.

#define MAX_HEART_SIZE (8192)   // Maximum size of heartbeat packet. Must be less than max IP packet size :)
#define MAX_CHANGES  (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2)     // Assumes a session is the biggest type!

static struct {
        int type;
        int id;
} cluster_changes[MAX_CHANGES]; // Queue of changed structures that need to go out when next heartbeat.

static struct {
        int seq;
        int size;
        char data[MAX_HEART_SIZE];
} past_hearts[HB_HISTORY_SIZE]; // Ring buffer of heartbeats that we've recently sent out. Needed so
                                // we can re-transmit if needed.

static struct {
        u32 peer;
        time_t  basetime;
        clockt  timestamp;
        int     uptodate;
} peers[CLUSTER_MAX_SIZE];      // List of all the peers we've heard from.
static int num_peers;           // Number of peers in list.

static int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
static int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);

//
// Create a listening socket
//
// This joins the cluster multi-cast group.
//
int cluster_init()
{
        struct sockaddr_in addr;
        struct sockaddr_in interface_addr;
        struct ip_mreq mreq;
        struct ifreq   ifr;
        int opt = 0;

        config->cluster_undefined_sessions = MAXSESSION-1;
        config->cluster_undefined_tunnels = MAXTUNNEL-1;

        if (!config->cluster_address)
                return 0;
        if (!*config->cluster_interface)
                return 0;

        cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);

        memset(&addr, 0, sizeof(addr));
        addr.sin_family = AF_INET;
        addr.sin_port = htons(CLUSTERPORT);
        addr.sin_addr.s_addr = INADDR_ANY;
        setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));

        if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
        {
                LOG(0, 0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
                return -1;
        }

        strcpy(ifr.ifr_name, config->cluster_interface);
        if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0)
        {
                LOG(0, 0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
                return -1;
        }

        memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr));
        my_address = interface_addr.sin_addr.s_addr;

        // Join multicast group.
        mreq.imr_multiaddr.s_addr = config->cluster_address;
        mreq.imr_interface = interface_addr.sin_addr;


        opt = 0;        // Turn off multicast loopback.
        setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));

        if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0)
        {
                LOG(0, 0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
                return -1;
        }

        if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
        {
                LOG(0, 0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
                return -1;
        }

        config->cluster_last_hb = TIME;
        config->cluster_seq_number = -1;

        return cluster_sockfd;
}


//
// Send a chunk of data to the entire cluster (usually via the multicast
// address ).
//

static int cluster_send_data(void *data, int datalen)
{
        struct sockaddr_in addr = {0};

        if (!cluster_sockfd) return -1;
        if (!config->cluster_address) return 0;

        addr.sin_addr.s_addr = config->cluster_address;
        addr.sin_port = htons(CLUSTERPORT);
        addr.sin_family = AF_INET;

        LOG(5,0,0,0, "Cluster send data: %d bytes\n", datalen);

        if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
        {
                LOG(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
                return -1;
        }

        return 0;
}

//
// Add a chunk of data to a heartbeat packet.
// Maintains the format. Assumes that the caller
// has passed in a big enough buffer!
//
static void add_type(char ** p, int type, int more, char * data, int size)
{
        * ( (u32*)(*p) ) = type;
        *p += sizeof(u32);

        * ( (u32*)(*p) ) = more;
        *p += sizeof(u32);

        if (data && size > 0) {
                memcpy(*p, data, size);
                (*p) += size;
        }
}

// advertise our presence via BGP or gratuitous ARP
static void advertise(void)
{
#ifdef BGP
        if (bgp_configured)
                bgp_enable_routing(1);
        else
#endif /* BGP */
                if (config->send_garp)
                        send_garp(config->bind_address);        // Start taking traffic.
}

static void cluster_uptodate(void)
{
        if (config->cluster_iam_uptodate)
                return;

        if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
                return;

        config->cluster_iam_uptodate = 1;

        LOG(0,0,0,0, "Now uptodate with master.\n");
        advertise();
}

//
// Send a unicast UDP packet to a peer with 'data' as the
// contents.
//
static int peer_send_data(u32 peer, char * data, int size)
{
        struct sockaddr_in addr = {0};

        if (!cluster_sockfd) return -1;
        if (!config->cluster_address) return 0;

        if (!peer)      // Odd??
                return -1;

        addr.sin_addr.s_addr = peer;
        addr.sin_port = htons(CLUSTERPORT);
        addr.sin_family = AF_INET;

        LOG_HEX(5, "Peer send", data, size);

        if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
        {
                LOG(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
                return -1;
        }

        return 0;
}

//
// Send a structured message to a peer with a single element of type 'type'.
//
static int peer_send_message(u32 peer, int type, int more, char * data, int size)
{
        char buf[65536];        // Vast overkill.
        char * p = buf;

        LOG(4,0,0,0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
        add_type(&p, type, more, data, size);

        return peer_send_data(peer, buf, (p-buf) );
}

//
// Forward a state changing packet to the master.
//
// The master just processes the payload as if it had
// received it off the tun device.
//
int master_forward_packet(char *data, int size, u32 addr, int port)
{
        char buf[65536];        // Vast overkill.
        char *p = buf;

        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;

        LOG(4,0,0,0,    "Forwarding packet from %s to master (size %d)\n", inet_toa(addr), size);

        STAT(c_forwarded);
        add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) );
        memcpy(p, data, size);
        p += size;

        return peer_send_data(config->cluster_master_address, buf, (p-buf) );

}

//
// Forward a throttled packet to the master for handling.
//
// The master just drops the packet into the appropriate
// token bucket queue, and lets normal processing take care
// of it.
//
int master_throttle_packet(int tbfid, char *data, int size)
{
        char buf[65536];        // Vast overkill.
        char *p = buf;

        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;

        LOG(4,0,0,0,    "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);

        add_type(&p, C_THROTTLE, tbfid, data, size);

        return peer_send_data(config->cluster_master_address, buf, (p-buf) );

}

//
// Forward a walled garden packet to the master for handling.
//
// The master just writes the packet straight to the tun
// device (where is will normally loop through the
// firewall rules, and come back in on the tun device)
//
// (Note that this must be called with the tun header
// as the start of the data).
int master_garden_packet(sessionidt s, char *data, int size)
{
        char buf[65536];        // Vast overkill.
        char *p = buf;

        if (!config->cluster_master_address) // No election has been held yet. Just skip it.
                return -1;

        LOG(4,0,0,0,    "Walled garden packet to master (size %d)\n", size);

        add_type(&p, C_GARDEN, s, data, size);

        return peer_send_data(config->cluster_master_address, buf, (p-buf));

}

//
// Send a chunk of data as a heartbeat..
// We save it in the history buffer as we do so.
//
static void send_heartbeat(int seq, char * data, int size)
{
        int i;

        if (size > sizeof(past_hearts[0].data))
        {
                LOG(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
                kill(0, SIGTERM);
                exit(1);
        }
        i = seq % HB_HISTORY_SIZE;
        past_hearts[i].seq = seq;
        past_hearts[i].size = size;
        memcpy(&past_hearts[i].data, data, size);       // Save it.
        cluster_send_data(data, size);
}

//
// Send an 'i am alive' message to every machine in the cluster.
//
void cluster_send_ping(time_t basetime)
{
        char buff[100 + sizeof(pingt)];
        char *p = buff;
        pingt x;

        if (config->cluster_iam_master && basetime)             // We're heartbeating so no need to ping.
                return;

        LOG(5,0,0,0, "Sending cluster ping...\n");

        x.ver = 1;
        x.addr = config->bind_address;
        x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
        x.basetime = basetime;

        add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
        cluster_send_data(buff, (p-buff) );
}

//
// Walk the session counters looking for non-zero ones to send
// to the master. We send up to 600 of them at one time.
// We examine a maximum of 3000 sessions.
// (50k max session should mean that we normally
// examine the entire session table every 25 seconds).

#define MAX_B_RECS (600)
void master_update_counts(void)
{
        int i, c;
        bytest b[MAX_B_RECS+1];

        if (config->cluster_iam_master)         // Only happens on the slaves.
                return;

        if (!config->cluster_master_address)            // If we don't have a master, skip it for a while.
                return;

        i = MAX_B_RECS * 5; // Examine max 2000 sessions;
        if (config->cluster_highest_sessionid > i)
                i = config->cluster_highest_sessionid;

        for ( c = 0; i > 0 ; --i) {
                        // Next session to look at.
                walk_session_number++;
                if ( walk_session_number > config->cluster_highest_sessionid)
                        walk_session_number = 1;

                if (!sess_count[walk_session_number].cin && !sess_count[walk_session_number].cout)
                        continue; // Unused. Skip it.

                b[c].sid = walk_session_number;
                b[c].in = sess_count[walk_session_number].cin;
                b[c].out = sess_count[walk_session_number].cout;

                if (++c > MAX_B_RECS)   // Send a max of 400 elements in a packet.
                        break;

                        // Reset counters.
                sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
        }

        if (!c)         // Didn't find any that changes. Get out of here!
                return;


                        // Forward the data to the master.
        LOG(4,0,0,0, "Sending byte counters to master (%d elements)\n", c);
        peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
        return;
}

//
// On the master, check how our slaves are going. If
// one of them's not up-to-date we'll heartbeat faster.
// If we don't have any of them, then we need to turn
// on our own packet handling!
//
void cluster_check_slaves(void)
{
        int i;
        static int have_peers = 0;
        int had_peers = have_peers;
        clockt t = TIME;

        if (!config->cluster_iam_master)
                return;         // Only runs on the master...

        config->cluster_iam_uptodate = 1;       // cleared in loop below

        for (i = have_peers = 0; i < num_peers; i++)
        {
                if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
                        continue;       // Stale peer! Skip them.

                if (!peers[i].basetime)
                        continue;       // Shutdown peer! Skip them.

                if (peers[i].uptodate)
                        have_peers = 1;

                if (!peers[i].uptodate)
                        config->cluster_iam_uptodate = 0; // Start fast heartbeats
        }

#ifdef BGP
        // in a cluster, withdraw/add routes when we get a peer/lose all peers
        if (bgp_configured && have_peers != had_peers)
                bgp_enable_routing(!have_peers);
#endif /* BGP */
}

//
// Check that we have a master. If it's been too
// long since we heard from a master then hold an election.
//
void cluster_check_master(void)
{
        int i, count, tcount, high_unique_id = 0;
        int last_free = 0;
        clockt t = TIME;
        static int probed = 0;

        if (config->cluster_iam_master)
                return;         // Only runs on the slaves...

        // If the master is late (missed 2 hearbeats by a second and a
        // hair) it may be that the switch has dropped us from the
        // multicast group, try unicasting one probe to the master
        // which will hopefully respond with a unicast heartbeat that
        // will allow us to limp along until the querier next runs.
        if (TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
        {
                if (!probed && config->cluster_master_address)
                {
                        probed = 1;
                        LOG(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
                                0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));

                        peer_send_message(config->cluster_master_address,
                                C_LASTSEEN, config->cluster_seq_number, NULL, 0);
                }
        } else {        // We got a recent heartbeat; reset the probe flag.
                probed = 0;
        }

        if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
                return; // Everything's ok!

        config->cluster_last_hb = TIME + 1;     // Just the one election thanks.

        LOG(0,0,0,0, "Master timed out! Holding election...\n");

        for (i = 0; i < num_peers; i++)
        {
                if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
                        continue;       // Stale peer! Skip them.

                if (!peers[i].basetime)
                        continue;       // Shutdown peer! Skip them.

                if (peers[i].basetime < basetime) {
                        LOG(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
                        return;         // They'll win the election. Get out of here.
                }

                if (peers[i].basetime == basetime &&
                        peers[i].peer > my_address) {
                        LOG(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
                        return;         // They'll win the election. Wait for them to come up.
                }
        }

                // Wow. it's been ages since I last heard a heartbeat
                // and I'm better than an of my peers so it's time
                // to become a master!!!

        config->cluster_iam_master = 1;
        config->cluster_master_address = 0;

        LOG(0,0,0,0, "I am declaring myself the master!\n");

        if (config->cluster_seq_number == -1)
                config->cluster_seq_number = 0;

                //
                // Go through and mark all the tunnels as defined.
                // Count the highest used tunnel number as well.
                //
        config->cluster_highest_tunnelid = 0;
        for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
                if (tunnel[i].state == TUNNELUNDEF)
                        tunnel[i].state = TUNNELFREE;

                if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
                        config->cluster_highest_tunnelid = i;
        }

                //
                // Go through and mark all the sessions as being defined.
                // reset the idle timeouts.
                // add temporary byte counters to permanent ones.
                // Re-string the free list.
                // Find the ID of the highest session.
        last_free = 0;
        high_unique_id = 0;
        config->cluster_highest_sessionid = 0;
        for (i = 0, count = 0; i < MAXSESSION; ++i) {
                if (session[i].tunnel == T_UNDEF) {
                        session[i].tunnel = T_FREE;
                        ++count;
                }

                if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
                        session[last_free].next = i;
                        session[i].next = 0;
                        last_free = i;
                }

                        // Reset all the idle timeouts..
                session[i].last_packet = time_now;

                        // Accumulate un-sent byte counters.
                session[i].cin += sess_count[i].cin;
                session[i].cout += sess_count[i].cout;
                session[i].total_cin += sess_count[i].cin;
                session[i].total_cout += sess_count[i].cout;

                sess_count[i].cin = sess_count[i].cout = 0;

                session[i].radius = 0;  // Reset authentication as the radius blocks aren't up to date.

                if (session[i].unique_id >= high_unique_id)     // This is different to the index into the session table!!!
                        high_unique_id = session[i].unique_id+1;


                session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
                throttle_session(i, session[i].throttle_in, session[i].throttle_out);

                if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
                        config->cluster_highest_sessionid = i;
        }

        session[last_free].next = 0;    // End of chain.
        last_id = high_unique_id;               // Keep track of the highest used session ID.

        become_master();

        rebuild_address_pool();

                // If we're not the very first master, this is a big issue!
        if(count>0)
                LOG(0,0,0,0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);

        config->cluster_undefined_sessions = 0;
        config->cluster_undefined_tunnels = 0;
        config->cluster_iam_uptodate = 1; // assume all peers are up-to-date

        if (!num_peers) // lone master
                advertise();

        // FIXME. We need to fix up the tunnel control message
        // queue here! There's a number of other variables we
        // should also update.
}


//
// Check that our session table is validly matching what the
// master has in mind.
//
// In particular, if we have too many sessions marked 'undefined'
// we fix it up here, and we ensure that the 'first free session'
// pointer is valid.
//
static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
{
        int i;

        sessionfree = freesession_ptr;  // Keep the freesession ptr valid.

        if (config->cluster_iam_uptodate)
                return;

        if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
                return;

                // Clear out defined sessions, counting the number of
                // undefs remaining.
        config->cluster_undefined_sessions = 0;
        for (i = 1 ; i < MAXSESSION; ++i) {
                if (i > highsession) {
                        session[i].tunnel = 0; // Defined.
                        continue;
                }
                if (session[i].tunnel != T_UNDEF)
                        continue;
                ++config->cluster_undefined_sessions;
        }

                // Clear out defined tunnels, counting the number of
                // undefs remaining.
        config->cluster_undefined_tunnels = 0;
        for (i = 1 ; i < MAXTUNNEL; ++i) {
                if (i > hightunnel) {
                        tunnel[i].state = TUNNELFREE; // Defined.
                        continue;
                }
                if (tunnel[i].state != TUNNELUNDEF)
                        continue;
                ++config->cluster_undefined_tunnels;
        }


        if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
                LOG(2,0,0,0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
                        config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
                return;
        }

                // Are we up to date?

        if (!config->cluster_iam_uptodate)
                cluster_uptodate();
}

static int hb_add_type(char **p, int type, int id)
{
        switch (type) {
                case C_CSESSION: { // Compressed C_SESSION.
                        u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
                        u8 *d = (u8 *) &session[id];
                        u8 *orig = d;
                        int size;

                        size = rle_compress( &d,  sizeof(sessiont), c, sizeof(c) );

                                // Did we compress the full structure, and is the size actually
                                // reduced??
                        if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
                                add_type(p, C_CSESSION, id, (char*) c, size);
                                break;
                        }
                        // Failed to compress : Fall through.
                }
                case C_SESSION: add_type(p, C_SESSION, id,
                        (char*) &session[id], sizeof(sessiont));
                        break;

                case C_CTUNNEL: { // Compressed C_TUNNEL
                        u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
                        u8 *d = (u8 *) &tunnel[id];
                        u8 *orig = d;
                        int size;

                        size = rle_compress( &d,  sizeof(tunnelt), c, sizeof(c) );

                                // Did we compress the full structure, and is the size actually
                                // reduced??
                        if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
                                add_type(p, C_CTUNNEL, id, c, size);
                                break;
                        }
                        // Failed to compress : Fall through.
                }
                case C_TUNNEL: add_type(p, C_TUNNEL, id,
                        (char*) &tunnel[id], sizeof(tunnelt));
                        break;
                default:
                        LOG(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
                        kill(0, SIGTERM);
                        exit(1);
        }
        return 0;
}

//
// Send a heartbeat, incidently sending out any queued changes..
//
void cluster_heartbeat()
{
        int i, count = 0, tcount = 0;
        char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
        heartt h;
        char *p = buff;

        if (!config->cluster_iam_master)        // Only the master does this.
                return;

        // Fill out the heartbeat header.
        memset(&h, 0, sizeof(h));

        h.version = HB_VERSION;
        h.seq = config->cluster_seq_number;
        h.basetime = basetime;
        h.clusterid = config->bind_address;     // Will this do??
        h.basetime = basetime;
        h.highsession = config->cluster_highest_sessionid;
        h.freesession = sessionfree;
        h.hightunnel = config->cluster_highest_tunnelid;
        h.size_sess = sizeof(sessiont);         // Just in case.
        h.size_tunn = sizeof(tunnelt);
        h.interval = config->cluster_hb_interval;
        h.timeout  = config->cluster_hb_timeout;

        add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h));

        for (i = 0; i < config->cluster_num_changes; ++i) {
                hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
        }

        if (p > (buff + sizeof(buff))) {        // Did we somehow manage to overun the buffer?
                LOG(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
                kill(0, SIGTERM);
                exit(1);
        }

                //
                // Fill out the packet with sessions from the session table...
                // (not forgetting to leave space so we can get some tunnels in too )
        while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {

                if (!walk_session_number)       // session #0 isn't valid.
                        ++walk_session_number;

                if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
                        break;

                hb_add_type(&p, C_CSESSION, walk_session_number);
                walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1);    // +1 avoids divide by zero.

                ++count;                        // Count the number of extra sessions we're sending.
        }

                //
                // Fill out the packet with tunnels from the tunnel table...
                // This effectively means we walk the tunnel table more quickly
                // than the session table. This is good because stuffing up a 
                // tunnel is a much bigger deal than stuffing up a session.
                //
        while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {

                if (!walk_tunnel_number)        // tunnel #0 isn't valid.
                        ++walk_tunnel_number;

                if (tcount >= config->cluster_highest_tunnelid)
                        break;

                hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
                walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1);       // +1 avoids divide by zero.

                ++tcount;
        }

                //
                // Did we do something wrong?
        if (p > (buff + sizeof(buff))) {        // Did we somehow manage to overun the buffer?
                LOG(0,0,0,0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
                kill(0, SIGTERM);
                exit(1);
        }

        LOG(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
                        h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
                        config->cluster_highest_tunnelid, (p-buff));

        config->cluster_num_changes = 0;

        send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.

        config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
}

//
// A structure of type 'type' has changed; Add it to the queue to send.
//
static int type_changed(int type, int id)
{
        int i;

        for (i = 0 ; i < config->cluster_num_changes ; ++i)
                if ( cluster_changes[i].id == id &&
                        cluster_changes[i].type == type)
                        return 0;       // Already marked for change.

        cluster_changes[i].type = type;
        cluster_changes[i].id = id;
        ++config->cluster_num_changes;

        if (config->cluster_num_changes > MAX_CHANGES)
                cluster_heartbeat(); // flush now

        return 1;
}


// A particular session has been changed!
int cluster_send_session(int sid)
{
        if (!config->cluster_iam_master) {
                LOG(0,0,sid,0, "I'm not a master, but I just tried to change a session!\n");
                return -1;
        }

        return type_changed(C_CSESSION, sid);
}

// A particular tunnel has been changed!
int cluster_send_tunnel(int tid)
{
        if (!config->cluster_iam_master) {
                LOG(0,0,0,tid, "I'm not a master, but I just tried to change a tunnel!\n");
                return -1;
        }

        return type_changed(C_CTUNNEL, tid);
}


//
// We're a master, and a slave has just told us that it's
// missed a packet. We'll resend it every packet since
// the last one it's seen.
//
static int cluster_catchup_slave(int seq, u32 slave)
{
        int s;
        int diff;

        LOG(1,0,0,0, "Slave %s sent LASTSEEN with seq %d\n", inet_toa(slave), seq);

        diff = config->cluster_seq_number - seq;        // How many packet do we need to send?
        if (diff < 0)
                diff += HB_MAX_SEQ;

        if (diff >= HB_HISTORY_SIZE) {  // Ouch. We don't have the packet to send it!
                LOG(0,0,0,0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
                        seq, config->cluster_seq_number);
                return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
        }

                // Now resend every packet that it missed, in order.
        while (seq != config->cluster_seq_number) {
                s = seq%HB_HISTORY_SIZE;
                if (seq != past_hearts[s].seq) {
                        LOG(0,0,0,0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
                                inet_toa(slave), seq, past_hearts[s].seq, s, config->cluster_seq_number);
                        return -1;      // What to do here!?
                }
                peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
                seq = (seq+1)%HB_MAX_SEQ;       // Increment to next seq number.
        }
        return 0; // All good!
}

//
// We've heard from another peer! Add it to the list
// that we select from at election time.
//
static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
{
        int i;
        u32 clusterid;
        pingt p;

        // Allow for backward compatability.
        // Just the ping packet into a new structure to allow
        // for the possibility that we might have received
        // more or fewer elements than we were expecting.
        if (size > sizeof(p))
                size = sizeof(p);

        memset( (void*) &p, 0, sizeof(p) );
        memcpy( (void*) &p, (void*) pp, size);

        clusterid = p.addr;
        if (clusterid != config->bind_address)
        {
                // Is this for us?
                LOG(4,0,0,0, "Skipping ping from %s (different cluster)\n", inet_toa(peer));
                return 0;
        }

        for (i = 0; i < num_peers ; ++i)
        {
                if (peers[i].peer != peer)
                        continue;

                // This peer already exists. Just update the timestamp.
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;
                peers[i].uptodate = !p.undef;
                break;
        }

        // Is this the master shutting down??
        if (peer == config->cluster_master_address && !basetime) {
                LOG(3,0,0,0, "Master %s shutting down...\n", inet_toa(config->cluster_master_address));
                config->cluster_master_address = 0;
                config->cluster_last_hb = 0; // Force an election.
                cluster_check_master();
                return 0;
        }

        if (i >= num_peers)
        {
                LOG(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));

                // Not found. Is there a stale slot to re-use?
                for (i = 0; i < num_peers ; ++i)
                {
                        if (!peers[i].basetime) // Shutdown
                                break;

                        if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
                                break;
                }

                if (i >= CLUSTER_MAX_SIZE)
                {
                        // Too many peers!!
                        LOG(0,0,0,0, "Tried to add %s as a peer, but I already have %d of them!\n", inet_toa(peer), i);
                        return -1;
                }

                peers[i].peer = peer;
                peers[i].basetime = basetime;
                peers[i].timestamp = TIME;
                peers[i].uptodate = !p.undef;
                if (i == num_peers)
                        ++num_peers;

                LOG(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer), num_peers);
        }

        return 1;
}

/* Handle the slave updating the byte counters for the master. */
//
// Note that we don't mark the session as dirty; We rely on
// the slow table walk to propogate this back out to the slaves.
//
static int cluster_handle_bytes(char * data, int size)
{
        bytest * b;

        b = (bytest*) data;

        LOG(3,0,0,0, "Got byte counter update (size %d)\n", size);

                                /* Loop around, adding the byte
                                counts to each of the sessions. */

        while (size >= sizeof(*b) ) {
                if (b->sid > MAXSESSION) {
                        LOG(0,0,0,0, "Got C_BYTES with session #%d!\n", b->sid);
                        return -1; /* Abort processing */
                }

                session[b->sid].total_cin += b->in;
                session[b->sid].total_cout += b->out;

                session[b->sid].cin += b->in;
                session[b->sid].cout += b->out;
                session[b->sid].last_packet = time_now; // Reset idle timer!

                size -= sizeof(*b);
                ++b;
        }

        if (size != 0)
                LOG(0,0,0,0, "Got C_BYTES with %d bytes of trailing junk!\n", size);

        return size;
}

//
// Handle receiving a session structure in a heartbeat packet.
//
static int cluster_recv_session(int more , u8 * p)
{
        if (more >= MAXSESSION) {
                LOG(0,0,0,0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
                return -1;
        }

        if (session[more].tunnel == T_UNDEF) {
                if (config->cluster_iam_uptodate) { // Sanity.
                        LOG(0,0,0,0, "I thought I was uptodate but I just found an undefined session!\n");
                } else {
                        --config->cluster_undefined_sessions;
                }
        }

        load_session(more, (sessiont*) p);      // Copy session into session table..

        LOG(5,0,more,0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);

        if (!config->cluster_iam_uptodate)
                cluster_uptodate();     // Check to see if we're up to date.

        return 0;
}

static int cluster_recv_tunnel(int more, u8 *p)
{
        if (more >= MAXTUNNEL) {
                LOG(0,0,0,0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
                return -1;
        }

        if (tunnel[more].state == TUNNELUNDEF) {
                if (config->cluster_iam_uptodate) { // Sanity.
                        LOG(0,0,0,0, "I thought I was uptodate but I just found an undefined tunnel!\n");
                } else {
                        --config->cluster_undefined_tunnels;
                }
        }

        memcpy(&tunnel[more], p, sizeof(tunnel[more]) );

                //
                // Clear tunnel control messages. These are dynamically allocated.
                // If we get unlucky, this may cause the tunnel to drop!
                //
        tunnel[more].controls = tunnel[more].controle = NULL;
        tunnel[more].controlc = 0;

        LOG(5,0,0,more, "Received tunnel update\n");

        if (!config->cluster_iam_uptodate)
                cluster_uptodate();     // Check to see if we're up to date.

        return 0;
}


//
// Process a heartbeat..
//
static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
{
        heartt * h;
        int s = size - (p-data);
        int i, type;

#if HB_VERSION != 3
# error "need to update cluster_process_heartbeat()"
#endif

        // we handle version 2+
        if (more < 2 || more > HB_VERSION) {
                LOG(0,0,0,0, "Received a heartbeat version that I don't support (%d)!\n", more);
                return -1; // Ignore it??
        }

                // Ok. It's a heartbeat packet from a cluster master!
        if (s < sizeof(*h))
                goto shortpacket;

        h = (heartt*) p;
        p += sizeof(*h);
        s -= sizeof(*h);

        if (h->clusterid != config->bind_address)
                return -1;      // It's not part of our cluster.

        if (config->cluster_iam_master) {       // Sanity...
                                // Note that this MUST match the election process above!

                LOG(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
                if (!h->basetime) {
                        LOG(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr) );
                        return -1; // Skip it.
                }
                if (basetime > h->basetime) {
                        LOG(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr));
                        kill(0, SIGTERM);
                        exit(1);
                }
                if (basetime == h->basetime && my_address < addr) { // Tie breaker.
                        LOG(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
                        kill(0, SIGTERM);
                        exit(1);
                }
                return -1; // Skip it.
        }

        if (config->cluster_seq_number == -1)   // Don't have one. Just align to the master...
                config->cluster_seq_number = h->seq;

        config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!

        if (config->cluster_seq_number != h->seq) {     // Out of sequence heartbeat!
                LOG(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);

                peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);

                config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!

                        // Just drop the packet. The master will resend it as part of the catchup.

                return 0;
        }
                // Save the packet in our buffer.
                // This is needed in case we become the master.
        config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
        i = h->seq % HB_HISTORY_SIZE;
        past_hearts[i].seq = h->seq;
        past_hearts[i].size = size;
        memcpy(&past_hearts[i].data, data, size);       // Save it.


                        // Check that we don't have too many undefined sessions, and
                        // that the free session pointer is correct.
        cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);

        if (more > 2) // reserved section of heartt was not initialized prior to v3
        {
                if (h->interval != config->cluster_hb_interval)
                {
                        LOG(2, 0, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
                                h->interval, config->cluster_hb_interval);

                        config->cluster_hb_interval = h->interval;
                }

                if (h->timeout != config->cluster_hb_timeout)
                {
                        LOG(2, 0, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
                                h->timeout, config->cluster_hb_timeout);

                        config->cluster_hb_timeout = h->timeout;
                }
        }

                // Ok. process the packet...
        while ( s > 0) {

                type = * ((u32*) p);
                p += sizeof(u32);
                s -= sizeof(u32);

                more = * ((u32*) p);
                p += sizeof(u32);
                s -= sizeof(u32);

                switch (type) {
                        case C_CSESSION: { // Compressed session structure.
                                u8 c [ sizeof(sessiont) + 2];
                                int size;
                                u8 * orig_p = p;

                                size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
                                s -= (p - orig_p);

                                if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
                                        LOG(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }

                                cluster_recv_session(more, c);
                                break;
                        }
                        case C_SESSION:
                                if ( s < sizeof(session[more]))
                                        goto shortpacket;

                                cluster_recv_session(more, p);

                                p += sizeof(session[more]);
                                s -= sizeof(session[more]);
                                break;

                        case C_CTUNNEL: { // Compressed tunnel structure.
                                u8 c [ sizeof(tunnelt) + 2];
                                int size;
                                u8 * orig_p = p;

                                size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
                                s -= (p - orig_p);

                                if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
                                        LOG(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
                                                // Now what? Should exit! No-longer up to date!
                                        break;
                                }

                                cluster_recv_tunnel(more, c);
                                break;

                        }
                        case C_TUNNEL:
                                if ( s < sizeof(tunnel[more]))
                                        goto shortpacket;

                                cluster_recv_tunnel(more, p);

                                p += sizeof(tunnel[more]);
                                s -= sizeof(tunnel[more]);
                                break;
                        default:
                                LOG(0,0,0,0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
                                return -1; // can't process any more of the packet!!
                }
        }
        if (config->cluster_master_address != addr)
        {
                char *str;
                str = strdup(inet_toa(config->cluster_master_address));
                LOG(0,0,0,0, "My master just changed from %s to %s!\n", str, inet_toa(addr));
                if (str) free(str);
        }

        config->cluster_master_address = addr;
        config->cluster_last_hb = TIME; // Successfully received a heartbeat!
        return 0;

shortpacket:
        LOG(0,0,0,0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
        return -1;
}

//
// We got a packet on the cluster port!
// Handle pings, lastseens, and heartbeats!
//
int processcluster(char * data, int size, u32 addr)
{
        int type, more;
        char * p = data;
        int s = size;

        if (addr == my_address)
                return -1;      // Ignore it. Something looped back the multicast!

        LOG(5,0,0,0, "Process cluster: %d bytes from %s\n", size, inet_toa(addr));

        if (s <= 0)     // Any data there??
                return -1;

        if (s < 8)
                goto shortpacket;

        type = * ((u32*) p);
        p += sizeof(u32);
        s -= sizeof(u32);

        more = * ((u32*) p);
        p += sizeof(u32);
        s -= sizeof(u32);

        switch (type) {
        case C_PING:    // Update the peers table.
                return cluster_add_peer(addr, more, (pingt*)p, s);

        case C_LASTSEEN:        // Catch up a slave (slave missed a packet).
                return cluster_catchup_slave(more, addr);

        case C_FORWARD: { // Forwarded control packet. pass off to processudp.
                struct sockaddr_in a;
                a.sin_addr.s_addr = more;

                a.sin_port = * (int*) p;
                s -= sizeof(int);
                p += sizeof(int);

                if (!config->cluster_iam_master) { // huh?
                        LOG(0,0,0,0, "I'm not the master, but I got a C_FORWARD from %s?\n", inet_toa(addr));
                        return -1;
                }

                LOG(4,0,0,0, "Got a forwarded packet... (%s:%d)\n", inet_toa(more), a.sin_port);
                STAT(recv_forward);
                processudp(p, s, &a);
                return 0;
        }
        case C_THROTTLE: {      // Receive a forwarded packet from a slave.
                if (!config->cluster_iam_master) {
                        LOG(0,0,0,0, "I'm not the master, but I got a C_THROTTLE from %s?\n", inet_toa(addr));
                        return -1;
                }

                tbf_queue_packet(more, p, s);   // The TBF id tells wether it goes in or out.
                return 0;
        }
        case C_GARDEN:
                // Receive a walled garden packet from a slave.
                if (!config->cluster_iam_master) {
                        LOG(0,0,0,0, "I'm not the master, but I got a C_GARDEN from %s?\n", inet_toa(addr));
                        return -1;
                }

                tun_write(p, s);
                return 0;

        case C_BYTES:
                return cluster_handle_bytes(p, s);

        case C_KILL:    // The master asked us to die!? (usually because we're too out of date).
                if (config->cluster_iam_master) {
                        LOG(0,0,0,0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", inet_toa(addr), more);
                        return -1;
                }
                if (more != config->cluster_seq_number) {
                        LOG(0,0,0,0, "The master asked us to die but the seq number didn't match!?\n");
                        return -1;
                }

                if (addr != config->cluster_master_address) {
                        LOG(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
                                inet_toa(addr), config->cluster_master_address);
                        // We can only warn about it. The master might really have switched!
                }

                LOG(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
                kill(0, SIGTERM);
                exit(0);        // Lets be paranoid;
                return -1;              // Just signalling the compiler.

        case C_HEARTBEAT:
                LOG(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));
                return cluster_process_heartbeat(data, size, more, p, addr);

        default:
                LOG(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type);
                return -1;
        }
        return 0;

shortpacket:
        LOG(0,0,0,0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
        return -1;
}

//====================================================================================================

int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
{
        int i;

        if (CLI_HELP_REQUESTED)
                return CLI_HELP_NO_ARGS;

        cli_print(cli, "Cluster status   : %s", config->cluster_iam_master ? "Master" : "Slave" );
        cli_print(cli, "My address       : %s", inet_toa(my_address));
        cli_print(cli, "VIP address      : %s", inet_toa(config->bind_address));
        cli_print(cli, "Multicast address: %s", inet_toa(config->cluster_address));
        cli_print(cli, "Multicast i'face : %s", config->cluster_interface);

        if (!config->cluster_iam_master) {
                cli_print(cli, "My master        : %s (last heartbeat %.1f seconds old)",
                        config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
                        0.1 * (TIME - config->cluster_last_hb));
                cli_print(cli, "Uptodate         : %s", config->cluster_iam_uptodate ? "Yes" : "No");
                cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
                cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
                cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
        } else {
                cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
                cli_print(cli, "Highest session  : %d", config->cluster_highest_sessionid);
                cli_print(cli, "Highest tunnel   : %d", config->cluster_highest_tunnelid);
                cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
        }
        cli_print(cli, "%d peers.", num_peers);

        if (num_peers)
                cli_print(cli, "%20s  %10s %8s", "Address", "Basetime", "Age");
        for (i = 0; i < num_peers; ++i) {
                cli_print(cli, "%20s  %10d %8d", inet_toa(peers[i].peer),
                        peers[i].basetime, TIME - peers[i].timestamp);
        }
        return CLI_OK;
}

//
// Simple run-length-encoding compression.
// Format is
//      1 byte < 128 = count of non-zero bytes following.       // Not legal to be zero.
//      n non-zero bytes;
// or
//      1 byte > 128 = (count - 128) run of zero bytes.         //
//   repeat.
//   count == 0 indicates end of compressed stream.
//
// Compress from 'src' into 'dst'. return number of bytes
// used from 'dst'.
// Updates *src_p to indicate 1 past last bytes used.
//
// We could get an extra byte in the zero runs by storing (count-1)
// but I'm playing it safe.
//
// Worst case is a 50% expansion in space required (trying to
// compress { 0x00, 0x01 } * N )
static int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
{
        int count;
        int orig_dsize = dsize;
        u8 * x,*src;
        src = *src_p;

        while (ssize > 0 && dsize > 2) {
                count = 0;
                x = dst++; --dsize;     // Reserve space for count byte..

                if (*src) {             // Copy a run of non-zero bytes.
                        while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
                                *dst++ = *src++;
                                --dsize; --ssize;
                                ++count;
                        }
                        *x = count;     // Store number of non-zero bytes. Guarenteed to be non-zero!

                } else {                // Compress a run of zero bytes.
                        while (*src == 0 && count < 127 && ssize > 0) {
                                ++src;
                                --ssize;
                                ++count;
                        }
                        *x = count | 0x80 ;
                }
        }

        *dst++ = 0x0; // Add Stop byte.
        --dsize;

        *src_p = src;
        return (orig_dsize - dsize);
}

//
// Decompress the buffer into **p.
// 'psize' is the size of the decompression buffer available.
//
// Returns the number of bytes decompressed.
//
// Decompresses from '*src_p' into 'dst'.
// Return the number of dst bytes used.
// Updates the 'src_p' pointer to point to the
// first un-used byte.
static int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
{
        int count;
        int orig_dsize = dsize;
        char * src = *src_p;

        while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
                count = *src++; --ssize;  // get the count byte from the source.
                if (count == 0x0)       // End marker reached? If so, finish.
                        break;

                if (count & 0x80) {     // Decompress a run of zeros
                        for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
                                *dst++ = 0x0;
                                --dsize;
                        }
                } else {                // Copy run of non-zero bytes.
                        for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
                                *dst++ = *src++;
                                --ssize; --dsize;
                        }
                }
        }
        *src_p = src;
        return (orig_dsize - dsize);
}