Don't process C_LASTSEEN unless we're a master (otherwise a crashed
[l2tpns.git] / cluster.c
index 29d8ca2..d7a74fe 100644 (file)
--- a/cluster.c
+++ b/cluster.c
@@ -1,8 +1,12 @@
 // L2TPNS Clustering Stuff
 
-char const *cvs_id_cluster = "$Id: cluster.c,v 1.25 2004-12-16 08:49:53 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.36 2005-05-07 10:14:33 bodea Exp $";
 
 #include <stdio.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <unistd.h>
+#include <inttypes.h>
 #include <sys/file.h>
 #include <sys/stat.h>
 #include <sys/socket.h>
@@ -13,12 +17,7 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.25 2004-12-16 08:49:53 bodea Ex
 #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 <inttypes.h>
 
 #include "l2tpns.h"
 #include "cluster.h"
@@ -39,11 +38,12 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.25 2004-12-16 08:49:53 bodea Ex
  */
 
 // Module variables.
-int cluster_sockfd = 0;                // The filedescriptor for the cluster communications port.
+int cluster_sockfd = 0;                        // The filedescriptor for the cluster communications 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.
+int forked = 0;                                // Sanity check: CLI must not diddle with heartbeat table
 
 #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!
@@ -62,7 +62,7 @@ static struct {
 
 static struct {
        in_addr_t peer;
-       time_t basetime;
+       uint32_t basetime;
        clockt timestamp;
        int uptodate;
 } peers[CLUSTER_MAX_SIZE];     // List of all the peers we've heard from.
@@ -406,18 +406,18 @@ 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!
@@ -571,38 +571,42 @@ 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.
+               sess_local[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();
 
@@ -650,12 +654,12 @@ 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
@@ -663,12 +667,12 @@ 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;
        }
 
 
@@ -867,6 +871,11 @@ int cluster_send_session(int sid)
                return -1;
        }
 
+       if (forked) {
+               LOG(0, sid, 0, "cluster_send_session called from child process!\n");
+               return -1;
+       }
+
        return type_changed(C_CSESSION, sid);
 }
 
@@ -1028,7 +1037,9 @@ 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;
@@ -1178,9 +1189,22 @@ static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t
        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;
+
+               // limit to once per second for a particular seq#
+               int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
 
-               peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+               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!!
 
@@ -1340,6 +1364,11 @@ int processcluster(char *data, int size, in_addr_t addr)
                return cluster_add_peer(addr, more, (pingt *) p, s);
 
        case C_LASTSEEN:        // Catch up a slave (slave missed a packet).
+               if (!config->cluster_iam_master) { // huh?
+                       LOG(0, 0, 0, "I'm not the master, but I got a C_LASTSEEN from %s?\n", fmtaddr(addr, 0));
+                       return -1;
+               }
+
                return cluster_catchup_slave(more, addr);
 
        case C_FORWARD: { // Forwarded control packet. pass off to processudp.
@@ -1440,12 +1469,12 @@ 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 #  : %llu", config->cluster_table_version);
+               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 #  : %llu", config->cluster_table_version);
+               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);
@@ -1456,7 +1485,7 @@ 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;