X-Git-Url: http://git.sameswireless.fr/l2tpns.git/blobdiff_plain/bc5b25832a728dea4346782d84d63128035bcfaf..9f41772ca94e7c6eb89910a0f09e98efb8dca644:/cluster.c diff --git a/cluster.c b/cluster.c index 7a83858..645403c 100644 --- a/cluster.c +++ b/cluster.c @@ -1,8 +1,12 @@ // L2TPNS Clustering Stuff -char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004/11/29 02:17:17 bodea Exp $"; +char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.11 2005/05/30 02:55:40 bodea Exp $"; #include +#include +#include +#include +#include #include #include #include @@ -13,10 +17,6 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004/11/29 02:17:17 bodea Ex #include #include #include -#include -#include -#include -#include #include #include "l2tpns.h" @@ -40,7 +40,7 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004/11/29 02:17:17 bodea Ex // Module variables. int cluster_sockfd = 0; // The filedescriptor for the cluster communications port. -ipt my_address = 0; // The network address of my ethernet port. +in_addr_t 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. @@ -60,15 +60,15 @@ static struct { // we can re-transmit if needed. static struct { - u32 peer; - time_t basetime; - clockt timestamp; - int uptodate; + in_addr_t peer; + uint32_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); +static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize); +static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize); // // Create a listening socket @@ -81,7 +81,7 @@ int cluster_init() struct sockaddr_in interface_addr; struct ip_mreq mreq; struct ifreq ifr; - int opt = 0; + int opt; config->cluster_undefined_sessions = MAXSESSION-1; config->cluster_undefined_tunnels = MAXTUNNEL-1; @@ -99,6 +99,9 @@ int cluster_init() addr.sin_addr.s_addr = INADDR_ANY; setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr)); + opt = fcntl(cluster_sockfd, F_GETFL, 0); + fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK); + if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0) { LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno)); @@ -129,7 +132,7 @@ int cluster_init() return -1; } - if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0) + if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0) { LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno)); return -1; @@ -174,22 +177,22 @@ static int cluster_send_data(void *data, int datalen) // 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) +static void add_type(char **p, int type, int more, char *data, int size) { - * ( (u32*)(*p) ) = type; - *p += sizeof(u32); + *((uint32_t *) (*p)) = type; + *p += sizeof(uint32_t); - * ( (u32*)(*p) ) = more; - *p += sizeof(u32); + *((uint32_t *)(*p)) = more; + *p += sizeof(uint32_t); if (data && size > 0) { memcpy(*p, data, size); - (*p) += size; + *p += size; } } // advertise our presence via BGP or gratuitous ARP -static void advertise(void) +static void advertise_routes(void) { #ifdef BGP if (bgp_configured) @@ -200,6 +203,15 @@ static void advertise(void) send_garp(config->bind_address); // Start taking traffic. } +// withdraw our routes (BGP only) +static void withdraw_routes(void) +{ +#ifdef BGP + if (bgp_configured) + bgp_enable_routing(0); +#endif /* BGP */ +} + static void cluster_uptodate(void) { if (config->cluster_iam_uptodate) @@ -211,14 +223,14 @@ static void cluster_uptodate(void) config->cluster_iam_uptodate = 1; LOG(0, 0, 0, "Now uptodate with master.\n"); - advertise(); + advertise_routes(); } // // Send a unicast UDP packet to a peer with 'data' as the // contents. // -static int peer_send_data(u32 peer, char * data, int size) +static int peer_send_data(in_addr_t peer, char *data, int size) { struct sockaddr_in addr = {0}; @@ -246,10 +258,10 @@ static int peer_send_data(u32 peer, char * data, int size) // // 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) +static int peer_send_message(in_addr_t peer, int type, int more, char *data, int size) { char buf[65536]; // Vast overkill. - char * p = buf; + char *p = buf; LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size); add_type(&p, type, more, data, size); @@ -263,7 +275,7 @@ static int peer_send_message(u32 peer, int type, int more, char * data, int size // 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) +int master_forward_packet(char *data, int size, in_addr_t addr, int port) { char buf[65536]; // Vast overkill. char *p = buf; @@ -274,12 +286,11 @@ int master_forward_packet(char *data, int size, u32 addr, int port) LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size); STAT(c_forwarded); - add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) ); + 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) ); - + return peer_send_data(config->cluster_master_address, buf, (p - buf)); } // @@ -334,7 +345,7 @@ int master_garden_packet(sessionidt s, char *data, int size) // 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) +static void send_heartbeat(int seq, char *data, int size) { int i; @@ -403,18 +414,18 @@ void master_update_counts(void) 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) + if (!sess_local[walk_session_number].cin && !sess_local[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; + b[c].in = sess_local[walk_session_number].cin; + b[c].out = sess_local[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; + sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0; } if (!c) // Didn't find any that changes. Get out of here! @@ -423,7 +434,7 @@ void master_update_counts(void) // Forward the data to the master. LOG(4, 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); + peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c); return; } @@ -454,17 +465,22 @@ void cluster_check_slaves(void) continue; // Shutdown peer! Skip them. if (peers[i].uptodate) - have_peers = 1; - - if (!peers[i].uptodate) + have_peers++; + else 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 */ + // in a cluster, withdraw/add routes when we get a peer/lose peers + if (have_peers != had_peers) + { + if (had_peers < config->cluster_master_min_adv && + have_peers >= config->cluster_master_min_adv) + withdraw_routes(); + + else if (had_peers >= config->cluster_master_min_adv && + have_peers < config->cluster_master_min_adv) + advertise_routes(); + } } // @@ -477,20 +493,22 @@ void cluster_check_master(void) int last_free = 0; clockt t = TIME; static int probed = 0; + int have_peers; 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 + // multicast group, try unicasting probes 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 (config->cluster_master_address + && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11)) { - if (!probed && config->cluster_master_address) + if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval))) { - probed = 1; + probed = TIME; LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n", 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval))); @@ -508,7 +526,11 @@ void cluster_check_master(void) LOG(0, 0, 0, "Master timed out! Holding election...\n"); - for (i = 0; i < num_peers; i++) + // In the process of shutting down, can't be master + if (main_quit) + return; + + for (i = have_peers = 0; i < num_peers; i++) { if ((peers[i].timestamp + config->cluster_hb_timeout) < t) continue; // Stale peer! Skip them. @@ -526,6 +548,9 @@ void cluster_check_master(void) LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0)); return; // They'll win the election. Wait for them to come up. } + + if (peers[i].uptodate) + have_peers++; } // Wow. it's been ages since I last heard a heartbeat @@ -537,6 +562,11 @@ void cluster_check_master(void) LOG(0, 0, 0, "I am declaring myself the master!\n"); + if (have_peers < config->cluster_master_min_adv) + advertise_routes(); + else + withdraw_routes(); + if (config->cluster_seq_number == -1) config->cluster_seq_number = 0; @@ -568,38 +598,42 @@ void cluster_check_master(void) ++count; } - if (session[i].tunnel == T_FREE) { // Unused session. Add to free list. + if (!session[i].opened) { // Unused session. Add to free list. + memset(&session[i], 0, sizeof(session[i])); + session[i].tunnel = T_FREE; session[last_free].next = i; session[i].next = 0; last_free = i; + continue; } - // Reset all the idle timeouts.. + // Reset idle timeouts.. session[i].last_packet = time_now; + // Reset die relative to our uptime rather than the old master's + if (session[i].die) session[i].die = TIME; + // 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; + session[i].cin += sess_local[i].cin; + session[i].cout += sess_local[i].cout; + session[i].total_cin += sess_local[i].cin; + session[i].total_cout += sess_local[i].cout; - sess_count[i].cin = sess_count[i].cout = 0; + sess_local[i].cin = sess_local[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; + 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. + last_id = high_unique_id; // Keep track of the highest used session ID. become_master(); @@ -613,9 +647,6 @@ void cluster_check_master(void) 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. @@ -647,12 +678,12 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig config->cluster_undefined_sessions = 0; for (i = 1 ; i < MAXSESSION; ++i) { if (i > highsession) { - session[i].tunnel = 0; // Defined. + if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined. continue; } - if (session[i].tunnel != T_UNDEF) - continue; - ++config->cluster_undefined_sessions; + + if (session[i].tunnel == T_UNDEF) + ++config->cluster_undefined_sessions; } // Clear out defined tunnels, counting the number of @@ -660,12 +691,12 @@ static void cluster_check_sessions(int highsession, int freesession_ptr, int hig config->cluster_undefined_tunnels = 0; for (i = 1 ; i < MAXTUNNEL; ++i) { if (i > hightunnel) { - tunnel[i].state = TUNNELFREE; // Defined. + if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined. continue; } - if (tunnel[i].state != TUNNELUNDEF) - continue; - ++config->cluster_undefined_tunnels; + + if (tunnel[i].state == TUNNELUNDEF) + ++config->cluster_undefined_tunnels; } @@ -685,9 +716,9 @@ 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; + uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case. + uint8_t *d = (uint8_t *) &session[id]; + uint8_t *orig = d; int size; size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) ); @@ -695,19 +726,19 @@ static int hb_add_type(char **p, int type, int id) // 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); + 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)); + (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; + uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case. + uint8_t *d = (uint8_t *) &tunnel[id]; + uint8_t *orig = d; int size; size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) ); @@ -721,7 +752,7 @@ static int hb_add_type(char **p, int type, int id) // Failed to compress : Fall through. } case C_TUNNEL: add_type(p, C_TUNNEL, id, - (char*) &tunnel[id], sizeof(tunnelt)); + (char *) &tunnel[id], sizeof(tunnelt)); break; default: LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type); @@ -744,6 +775,8 @@ void cluster_heartbeat() if (!config->cluster_iam_master) // Only the master does this. return; + config->cluster_table_version += config->cluster_num_changes; + // Fill out the heartbeat header. memset(&h, 0, sizeof(h)); @@ -759,15 +792,16 @@ void cluster_heartbeat() h.size_tunn = sizeof(tunnelt); h.interval = config->cluster_hb_interval; h.timeout = config->cluster_hb_timeout; + h.table_version = config->cluster_table_version; - add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h)); + 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, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff); + LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff)); kill(0, SIGTERM); exit(1); } @@ -775,7 +809,7 @@ void cluster_heartbeat() // // 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) ) { + while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) { if (!walk_session_number) // session #0 isn't valid. ++walk_session_number; @@ -795,7 +829,7 @@ void cluster_heartbeat() // 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) ) { + while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) { if (!walk_tunnel_number) // tunnel #0 isn't valid. ++walk_tunnel_number; @@ -812,14 +846,16 @@ void cluster_heartbeat() // // Did we do something wrong? if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer? - LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff); + LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff)); kill(0, SIGTERM); exit(1); } - LOG(3, 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)); + LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes " + "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n", + HB_VERSION, h.seq, h.table_version, config->cluster_num_changes, + count, tcount, config->cluster_highest_sessionid, + config->cluster_highest_tunnelid, (int) (p - buff)); config->cluster_num_changes = 0; @@ -879,26 +915,35 @@ int cluster_send_tunnel(int tid) // 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) +static int cluster_catchup_slave(int seq, in_addr_t slave) { int s; int diff; LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq); + if (!config->cluster_iam_master) { + LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n", + fmtaddr(config->cluster_master_address, 0)); + + peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0); + return 0; + } 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, "A slaved asked for message %d when our seq number is %d. Killing it.\n", + LOG(0, 0, 0, "A slave 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. } + LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) ); + // Now resend every packet that it missed, in order. while (seq != config->cluster_seq_number) { - s = seq%HB_HISTORY_SIZE; + s = seq % HB_HISTORY_SIZE; if (seq != past_hearts[s].seq) { LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n", fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number); @@ -914,10 +959,10 @@ static 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. // -static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size) +static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size) { int i; - u32 clusterid; + in_addr_t clusterid; pingt p; // Allow for backward compatability. @@ -927,8 +972,8 @@ static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size) if (size > sizeof(p)) size = sizeof(p); - memset( (void*) &p, 0, sizeof(p) ); - memcpy( (void*) &p, (void*) pp, size); + memset( (void *) &p, 0, sizeof(p) ); + memcpy( (void *) &p, (void *) pp, size); clusterid = p.addr; if (clusterid != config->bind_address) @@ -951,8 +996,10 @@ static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size) } // Is this the master shutting down?? - if (peer == config->cluster_master_address && !basetime) { - LOG(3, 0, 0, "Master %s shutting down...\n", fmtaddr(config->cluster_master_address, 0)); + if (peer == config->cluster_master_address) { + LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0), + basetime ? "has restarted!" : "shutting down..."); + config->cluster_master_address = 0; config->cluster_last_hb = 0; // Force an election. cluster_check_master(); @@ -993,16 +1040,30 @@ static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size) return 1; } +// A slave responds with C_MASTER when it gets a message which should have gone to a master. +static int cluster_set_master(in_addr_t peer, in_addr_t master) +{ + if (config->cluster_iam_master) // Sanity... + return 0; + + LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0), + fmtaddr(master, 1)); + + config->cluster_master_address = master; + cluster_check_master(); + return 0; +} + /* 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) +static int cluster_handle_bytes(char *data, int size) { - bytest * b; + bytest *b; - b = (bytest*) data; + b = (bytest *) data; LOG(3, 0, 0, "Got byte counter update (size %d)\n", size); @@ -1020,7 +1081,9 @@ static int cluster_handle_bytes(char * data, int size) session[b->sid].cin += b->in; session[b->sid].cout += b->out; - session[b->sid].last_packet = time_now; // Reset idle timer! + + if (b->in) + session[b->sid].last_packet = time_now; // Reset idle timer! size -= sizeof(*b); ++b; @@ -1035,7 +1098,7 @@ static int cluster_handle_bytes(char * data, int size) // // Handle receiving a session structure in a heartbeat packet. // -static int cluster_recv_session(int more , u8 * p) +static int cluster_recv_session(int more, uint8_t *p) { if (more >= MAXSESSION) { LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n"); @@ -1050,7 +1113,7 @@ static int cluster_recv_session(int more , u8 * p) } } - load_session(more, (sessiont*) p); // Copy session into session table.. + load_session(more, (sessiont *) p); // Copy session into session table.. LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions); @@ -1060,7 +1123,7 @@ static int cluster_recv_session(int more , u8 * p) return 0; } -static int cluster_recv_tunnel(int more, u8 *p) +static int cluster_recv_tunnel(int more, uint8_t *p) { if (more >= MAXTUNNEL) { LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n"); @@ -1096,18 +1159,20 @@ static int cluster_recv_tunnel(int more, u8 *p) // // Process a heartbeat.. // -static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr) +// v3: added interval, timeout +// v4: added table_version +static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr) { - heartt * h; + heartt *h; int s = size - (p-data); int i, type; -#if HB_VERSION != 3 +#if HB_VERSION != 4 # error "need to update cluster_process_heartbeat()" #endif - // we handle version 2+ - if (more < 2 || more > HB_VERSION) { + // we handle versions 3 through 4 + if (more < 3 || more > HB_VERSION) { LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more); return -1; // Ignore it?? } @@ -1116,7 +1181,7 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 if (s < sizeof(*h)) goto shortpacket; - h = (heartt*) p; + h = (heartt *) p; p += sizeof(*h); s -= sizeof(*h); @@ -1126,33 +1191,88 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 if (config->cluster_iam_master) { // Sanity... // Note that this MUST match the election process above! - LOG(0, 0, 0, "I just got a packet claiming to be from a master but _I_ am the master!\n"); + LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0)); if (!h->basetime) { - LOG(0, 0, 0, "Heartbeat from addr %s with zero basetime!\n", fmtaddr(addr, 0)); + LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n"); return -1; // Skip it. } + + if (more >= 4) { + if (h->table_version > config->cluster_table_version) { + LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n", + h->table_version, config->cluster_table_version); + + kill(0, SIGTERM); + exit(1); + } + if (h->table_version < config->cluster_table_version) + return -1; + } + if (basetime > h->basetime) { - LOG(0, 0, 0, "They're (%s) an older master than me so I'm gone!\n", fmtaddr(addr, 0)); + LOG(0, 0, 0, "They're an older master than me so I'm gone!\n"); kill(0, SIGTERM); exit(1); } - if (basetime == h->basetime && my_address < addr) { // Tie breaker. + + if (basetime < h->basetime) + return -1; + + if (my_address < addr) { // Tie breaker. LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n"); kill(0, SIGTERM); exit(1); } + + // + // Send it a unicast heartbeat to see give it a chance to die. + // NOTE: It's actually safe to do seq-number - 1 without checking + // for wrap around. + // + cluster_catchup_slave(config->cluster_seq_number - 1, addr); + return -1; // Skip it. } + // + // Try and guard against a stray master appearing. + // + // Ignore heartbeats received from another master before the + // timeout (less a smidgen) for the old master has elapsed. + // + // Note that after a clean failover, the cluster_master_address + // is cleared, so this doesn't run. + // + if (config->cluster_master_address && addr != config->cluster_master_address + && (config->cluster_last_hb + config->cluster_hb_timeout - 11) > TIME) { + LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n", + fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1), + 0.1 * (TIME - config->cluster_last_hb)); + return -1; // ignore + } + 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, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number); + static int lastseen_seq = 0; + static time_t lastseen_time = 0; - peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0); + // limit to once per second for a particular seq# + int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time); + + LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n", + h->seq, config->cluster_seq_number, + ask ? "Asking for resend" : "Ignoring"); + + if (ask) + { + lastseen_seq = config->cluster_seq_number; + lastseen_time = time_now; + 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!! @@ -1173,43 +1293,40 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 // 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) { - if (h->interval != config->cluster_hb_interval) - { - LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n", - h->interval, config->cluster_hb_interval); + LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n", + 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, "Master set heartbeat timeout to %u (was %u)\n", - h->timeout, config->cluster_hb_timeout); + if (h->timeout != config->cluster_hb_timeout) + { + LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n", + h->timeout, config->cluster_hb_timeout); - config->cluster_hb_timeout = h->timeout; - } + config->cluster_hb_timeout = h->timeout; } // Ok. process the packet... while ( s > 0) { - type = * ((u32*) p); - p += sizeof(u32); - s -= sizeof(u32); + type = *((uint32_t *) p); + p += sizeof(uint32_t); + s -= sizeof(uint32_t); - more = * ((u32*) p); - p += sizeof(u32); - s -= sizeof(u32); + more = *((uint32_t *) p); + p += sizeof(uint32_t); + s -= sizeof(uint32_t); switch (type) { case C_CSESSION: { // Compressed session structure. - u8 c [ sizeof(sessiont) + 2]; + uint8_t c[ sizeof(sessiont) + 2]; int size; - u8 * orig_p = p; + uint8_t *orig_p = p; - size = rle_decompress((u8 **) &p, s, c, sizeof(c) ); + size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) ); s -= (p - orig_p); if (size != sizeof(sessiont) ) { // Ouch! Very very bad! @@ -1232,11 +1349,11 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 break; case C_CTUNNEL: { // Compressed tunnel structure. - u8 c [ sizeof(tunnelt) + 2]; + uint8_t c[ sizeof(tunnelt) + 2]; int size; - u8 * orig_p = p; + uint8_t *orig_p = p; - size = rle_decompress( (u8 **) &p, s, c, sizeof(c) ); + size = rle_decompress((uint8_t **) &p, s, c, sizeof(c)); s -= (p - orig_p); if (size != sizeof(tunnelt) ) { // Ouch! Very very bad! @@ -1273,6 +1390,7 @@ static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 } config->cluster_last_hb = TIME; // Successfully received a heartbeat! + config->cluster_table_version = h->table_version; return 0; shortpacket: @@ -1284,10 +1402,10 @@ shortpacket: // We got a packet on the cluster port! // Handle pings, lastseens, and heartbeats! // -int processcluster(char * data, int size, u32 addr) +int processcluster(char *data, int size, in_addr_t addr) { int type, more; - char * p = data; + char *p = data; int s = size; if (addr == my_address) @@ -1301,26 +1419,29 @@ int processcluster(char * data, int size, u32 addr) if (s < 8) goto shortpacket; - type = * ((u32*) p); - p += sizeof(u32); - s -= sizeof(u32); + type = *((uint32_t *) p); + p += sizeof(uint32_t); + s -= sizeof(uint32_t); - more = * ((u32*) p); - p += sizeof(u32); - s -= sizeof(u32); + more = *((uint32_t *) p); + p += sizeof(uint32_t); + s -= sizeof(uint32_t); switch (type) { - case C_PING: // Update the peers table. - return cluster_add_peer(addr, more, (pingt*)p, s); + 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). + case C_MASTER: // Our master is wrong + return cluster_set_master(addr, more); + + 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; + a.sin_port = *(int *) p; s -= sizeof(int); p += sizeof(int); @@ -1354,6 +1475,11 @@ int processcluster(char * data, int size, u32 addr) return 0; case C_BYTES: + if (!config->cluster_iam_master) { + LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr, 0)); + return -1; + } + return cluster_handle_bytes(p, s); case C_KILL: // The master asked us to die!? (usually because we're too out of date). @@ -1414,10 +1540,12 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc) : "Not defined", 0.1 * (TIME - config->cluster_last_hb)); cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No"); + cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version); 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, "Table version # : %" PRIu64, config->cluster_table_version); 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); @@ -1428,7 +1556,7 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc) 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", fmtaddr(peers[i].peer, 0), + cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0), peers[i].basetime, TIME - peers[i].timestamp); } return CLI_OK; @@ -1453,11 +1581,11 @@ int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc) // // 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) +static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize) { int count; int orig_dsize = dsize; - u8 * x,*src; + uint8_t *x, *src; src = *src_p; while (ssize > 0 && dsize > 2) { @@ -1499,11 +1627,11 @@ static int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize) // 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) +static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize) { int count; int orig_dsize = dsize; - char * src = *src_p; + 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.