X-Git-Url: http://git.sameswireless.fr/l2tpns.git/blobdiff_plain/def1dfd3ccaafe0b50d1bd0c24a04aa990414ebe..1eb066f4db51b1bae4ed9ade5ac51311fb78468f:/cluster.c diff --git a/cluster.c b/cluster.c index a8c9310..2c8da60 100644 --- a/cluster.c +++ b/cluster.c @@ -1,6 +1,6 @@ // L2TPNS Clustering Stuff -char const *cvs_id_cluster = "$Id: cluster.c,v 1.7 2004/07/07 09:09:53 bodea Exp $"; +char const *cvs_id_cluster = "$Id: cluster.c,v 1.13 2004/10/29 04:25:40 bodea Exp $"; #include #include @@ -66,7 +66,6 @@ static struct { 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 have_peers; // At least one up to date peer int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize); int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize); @@ -78,65 +77,68 @@ int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize); // int cluster_init() { - struct sockaddr_in addr; - struct sockaddr_in interface_addr; - struct ip_mreq mreq; - struct ifreq ifr; - int opt = 0; + 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; + config->cluster_undefined_sessions = MAXSESSION-1; + config->cluster_undefined_tunnels = MAXTUNNEL-1; - if (!config->cluster_address) - return 0; - if (!*config->cluster_interface) - return 0; + if (!config->cluster_address) + return 0; + if (!*config->cluster_interface) + return 0; - cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); + 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)); + 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; - } + 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; - } + 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; + 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; + // 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)); + 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_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; - } + 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; + config->cluster_last_hb = TIME; + config->cluster_seq_number = -1; - return cluster_sockfd; + return cluster_sockfd; } @@ -147,24 +149,24 @@ int cluster_init() int cluster_send_data(void *data, int datalen) { - struct sockaddr_in addr = {0}; + struct sockaddr_in addr = {0}; - if (!cluster_sockfd) return -1; - if (!config->cluster_address) return 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; + 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); + 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; - } + 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; + return 0; } // @@ -213,27 +215,27 @@ void cluster_uptodate(void) // int peer_send_data(u32 peer, char * data, int size) { - struct sockaddr_in addr = {0}; + struct sockaddr_in addr = {0}; - if (!cluster_sockfd) return -1; - if (!config->cluster_address) return 0; + if (!cluster_sockfd) return -1; + if (!config->cluster_address) return 0; - if (!peer) // Odd?? - return -1; + if (!peer) // Odd?? + return -1; - addr.sin_addr.s_addr = peer; - addr.sin_port = htons(CLUSTERPORT); - addr.sin_family = AF_INET; + addr.sin_addr.s_addr = peer; + addr.sin_port = htons(CLUSTERPORT); + addr.sin_family = AF_INET; - log_hex(5, "Peer send", data, size); + 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; - } + 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; + return 0; } // @@ -254,12 +256,12 @@ int peer_send_message(u32 peer, int type, int more, char * data, int size) // Forward a state changing packet to the master. // // The master just processes the payload as if it had -// received it off the tap device. +// received it off the tun device. // -int master_forward_packet(char * data, int size, u32 addr, int port) +int master_forward_packet(char *data, int size, u32 addr, int port) { char buf[65536]; // Vast overkill. - char * p = buf; + char *p = buf; if (!config->cluster_master_address) // No election has been held yet. Just skip it. return -1; @@ -282,10 +284,10 @@ int master_forward_packet(char * data, int size, u32 addr, int port) // token bucket queue, and lets normal processing take care // of it. // -int master_throttle_packet(int tbfid, char * data, int size) +int master_throttle_packet(int tbfid, char *data, int size) { char buf[65536]; // Vast overkill. - char * p = buf; + char *p = buf; if (!config->cluster_master_address) // No election has been held yet. Just skip it. return -1; @@ -329,25 +331,19 @@ int master_garden_packet(sessionidt s, char *data, int size) // static void send_heartbeat(int seq, char * data, int size) { - int i; - static int last_seq = -1; - - if (last_seq != -1 && (seq != (last_seq+1)%HB_MAX_SEQ) ) { - log(0,0,0,0, "FATAL: Sequence number skipped! (%d != %d)\n", - seq, last_seq); - } - last_seq = seq; - - 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); + 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); } // @@ -375,12 +371,12 @@ void cluster_send_ping(time_t basetime) // // Walk the session counters looking for non-zero ones to send -// to the master. We send up to 100 of them at one time. -// We examine a maximum of 2000 sessions. +// 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 (400) +#define MAX_B_RECS (600) void master_update_counts(void) { int i, c; @@ -426,56 +422,90 @@ void master_update_counts(void) 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 + // master lost all slaves, need to handle traffic ourself + if (bgp_configured && had_peers && !have_peers) + bgp_enable_routing(1); + else if (bgp_configured && !had_peers && have_peers) + bgp_enable_routing(0); +#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_sid = 0; + int i, count, tcount, high_unique_id = 0; int last_free = 0; - int had_peers = have_peers; clockt t = TIME; static int probed = 0; - if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout)) + 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 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 (config->cluster_master_address - && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11)) + if (!probed && config->cluster_master_address) { - if (!probed) - { - probed = 1; - log(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n", - 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 (!config->cluster_iam_master) - return; // Everything's ok. return. + 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))); - // Master needs to check peer state + 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; } - config->cluster_last_hb = TIME + 1; + if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout)) + return; // Everything's ok! - if (config->cluster_iam_master) - config->cluster_iam_uptodate = 1; // cleared in loop below - else - log(0,0,0,0, "Master timed out! Holding election...\n"); + config->cluster_last_hb = TIME + 1; // Just the one election thanks. + log(0,0,0,0, "Master timed out! Holding election...\n"); - for (i = have_peers = 0; i < num_peers; i++) + for (i = 0; i < num_peers; i++) { if ((peers[i].timestamp + config->cluster_hb_timeout) < t) continue; // Stale peer! Skip them. @@ -483,17 +513,6 @@ void cluster_check_master(void) if (!peers[i].basetime) continue; // Shutdown peer! Skip them. - if (peers[i].uptodate) - have_peers = 1; - - if (config->cluster_iam_master) - { - if (!peers[i].uptodate) - config->cluster_iam_uptodate = 0; // Start fast heartbeats - - continue; - } - 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. @@ -506,16 +525,6 @@ void cluster_check_master(void) } } - if (config->cluster_iam_master) // If we're the master, we've already won - { -#ifdef BGP - // master lost all slaves, need to handle traffic ourself - if (bgp_configured && had_peers && !have_peers) - bgp_enable_routing(1); -#endif /* BGP */ - return; - } - // 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!!! @@ -525,11 +534,6 @@ void cluster_check_master(void) log(0,0,0,0, "I am declaring myself the master!\n"); -#ifdef BGP - if (bgp_configured && have_peers) - bgp_enable_routing(0); /* stop handling traffic */ -#endif /* BGP */ - if (config->cluster_seq_number == -1) config->cluster_seq_number = 0; @@ -553,7 +557,7 @@ void cluster_check_master(void) // Re-string the free list. // Find the ID of the highest session. last_free = 0; - high_sid = 0; + high_unique_id = 0; config->cluster_highest_sessionid = 0; for (i = 0, count = 0; i < MAXSESSION; ++i) { if (session[i].tunnel == T_UNDEF) { @@ -580,8 +584,8 @@ void cluster_check_master(void) session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date. - if (session[i].sid >= high_sid) // This is different to the index into the session table!!! - high_sid = session[i].sid+1; + 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. @@ -599,7 +603,7 @@ void cluster_check_master(void) } session[last_free].next = 0; // End of chain. - last_sid = high_sid; // Keep track of the highest used session ID. + last_id = high_unique_id; // Keep track of the highest used session ID. become_master(); @@ -611,7 +615,7 @@ void cluster_check_master(void) config->cluster_undefined_sessions = 0; config->cluster_undefined_tunnels = 0; - config->cluster_iam_uptodate = 1; // assume all peers are up-to-date + config->cluster_iam_uptodate = 1; // assume all peers are up-to-date // FIXME. We need to fix up the tunnel control message // queue here! There's a number of other variables we @@ -814,7 +818,7 @@ void cluster_heartbeat() 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", + 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)); @@ -897,13 +901,8 @@ int cluster_catchup_slave(int seq, u32 slave) while (seq != config->cluster_seq_number) { s = seq%HB_HISTORY_SIZE; if (seq != past_hearts[s].seq) { - int i; 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); - - for (i = 0; i < HB_HISTORY_SIZE; ++i) { - log(0,0,0,0, "\tentry %3d: seq %d (size %d)\n", i, past_hearts[s].seq, past_hearts[s].size); - } return -1; // What to do here!? } peer_send_data(slave, past_hearts[s].data, past_hearts[s].size); @@ -916,12 +915,23 @@ int cluster_catchup_slave(int seq, u32 slave) // We've heard from another peer! Add it to the list // that we select from at election time. // -int cluster_add_peer(u32 peer, time_t basetime, pingt *p) +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); - clusterid = p->addr; + memset( (void*) &p, 0, sizeof(p) ); + memcpy( (void*) &p, (void*) pp, size); + + clusterid = p.addr; if (clusterid != config->bind_address) { // Is this for us? @@ -937,12 +947,13 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p) // This peer already exists. Just update the timestamp. peers[i].basetime = basetime; peers[i].timestamp = TIME; - peers[i].uptodate = !p->undef; + peers[i].uptodate = !p.undef; break; } - // Is this the master shutting down?? + // 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(); @@ -973,28 +984,13 @@ int cluster_add_peer(u32 peer, time_t basetime, pingt *p) peers[i].peer = peer; peers[i].basetime = basetime; peers[i].timestamp = TIME; - peers[i].uptodate = !p->undef; + 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); } - if (peers[i].uptodate) - { -#ifdef BGP - /* drop routes if we've now got a peer */ - if (config->cluster_iam_master && bgp_configured && !have_peers) - bgp_enable_routing(0); -#endif /* BGP */ - have_peers = 1; - } - else if (config->cluster_iam_master) - { - config->cluster_iam_uptodate = 0; // increase heart-rate... - } - - return 1; } @@ -1183,17 +1179,17 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 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); + h->interval, config->cluster_hb_interval); - config->cluster_hb_interval = h->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); + h->timeout, config->cluster_hb_timeout); - config->cluster_hb_timeout = h->timeout; + config->cluster_hb_timeout = h->timeout; } } @@ -1316,7 +1312,7 @@ int processcluster(char * data, int size, u32 addr) switch (type) { case C_PING: // Update the peers table. - return cluster_add_peer(addr, more, (pingt*)p); + 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); @@ -1393,7 +1389,7 @@ int processcluster(char * data, int size, u32 addr) return 0; shortpacket: - log(0,0,0,0, "I got an cluster heartbeat packet! This means I'm probably out of sync!!\n"); + log(0,0,0,0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n"); return -1; } @@ -1406,17 +1402,17 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc) if (CLI_HELP_REQUESTED) return CLI_HELP_NO_ARGS; - cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" ); + 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) { + 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, "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);