time_t last_config_check = 0;
static timeout_t pingtimer;
static timeout_t periodictimer;
+static struct timeval last_periodic_run_time;
/* Purge edges and subnets of unreachable nodes. Use carefully. */
and close the connection.
*/
static void timeout_handler(void *data) {
+
+ bool close_all_connections = false;
+
+ /*
+ timeout_hanlder will start after 30 seconds from start of tincd
+ hold information about the elapsed time since last time the handler
+ has been run
+ */
+ long sleep_time = now.tv_sec - last_periodic_run_time.tv_sec;
+ /*
+ It seems that finding sane default value is harder than expected
+ Since we send every second a UDP packet to make holepunching work
+ And default UDP state expire on firewalls is between 15-30 seconds
+ we drop all connections after 60 Seconds - UDPDiscoveryTimeout=30
+ by default
+ */
+ if (sleep_time > 2 * udp_discovery_timeout) {
+ logger(DEBUG_ALWAYS, LOG_ERR, "Awaking from dead after %ld seconds of sleep", sleep_time);
+ /*
+ Do not send any packets to tinc after we wake up.
+ The other node probably closed our connection but we still
+ are holding context information to them. This may happen on
+ laptops or any other hardware which can be suspended for some time.
+ Sending any data to node that wasn't expecting it will produce
+ annoying and misleading errors on the other side about failed signature
+ verification and or about missing sptps context
+ */
+ close_all_connections = true;
+ }
+ last_periodic_run_time = now;
+
for list_each(connection_t, c, connection_list) {
if(c->status.control)
continue;
+ if(close_all_connections) {
+ logger(DEBUG_ALWAYS, LOG_ERR, "Forcing connection close after sleep time %s (%s)", c->name, c->hostname);
+ terminate_connection(c, c->edge);
+ continue;
+ }
+
if(c->last_ping_time + pingtimeout <= now.tv_sec) {
if(c->edge) {
try_tx(c->node, false);
if(c->status.pinged) {
- logger(DEBUG_CONNECTIONS, LOG_INFO, "%s (%s) didn't respond to PING in %ld seconds", c->name, c->hostname, (long)now.tv_sec - c->last_ping_time);
+ logger(DEBUG_CONNECTIONS, LOG_INFO, "%s (%s) didn't respond to PING in %ld seconds", c->name, c->hostname, (long)(now.tv_sec - c->last_ping_time));
} else if(c->last_ping_time + pinginterval <= now.tv_sec) {
send_ping(c);
continue;
and we are not already trying to make one, create an
outgoing connection to this node.
*/
- int r = rand() % node_tree->count;
- int i = 0;
+ int count = 0;
+ for splay_each(node_t, n, node_tree) {
+ if(n == myself || n->connection || !(n->status.has_address || n->status.reachable))
+ continue;
+ count++;
+ }
+
+ if(!count)
+ goto end;
+
+ int r = rand() % count;
for splay_each(node_t, n, node_tree) {
- if(i++ != r)
+ if(n == myself || n->connection || !(n->status.has_address || n->status.reachable))
continue;
- if(n->connection)
- break;
+ if(r--)
+ continue;
bool found = false;
list_insert_tail(outgoing_list, outgoing);
setup_outgoing_connection(outgoing);
}
+
break;
}
} else if(nc > 3) {
}
}
+end:
timeout_set(data, &(struct timeval){5, rand() % 100000});
}
for splay_each(subnet_t, subnet, subnet_tree)
if (subnet->owner)
subnet->expires = 1;
+ }
- load_all_subnets();
+ for splay_each(node_t, n, node_tree)
+ n->status.has_address = false;
+ load_all_nodes();
+
+ if(strictsubnets) {
for splay_each(subnet_t, subnet, subnet_tree) {
if (!subnet->owner)
continue;
this is where it all happens...
*/
int main_loop(void) {
+ last_periodic_run_time = now;
timeout_add(&pingtimer, timeout_handler, &pingtimer, &(struct timeval){pingtimeout, rand() % 100000});
- timeout_add(&periodictimer, periodic_handler, &periodictimer, &(struct timeval){pingtimeout, rand() % 100000});
+ timeout_add(&periodictimer, periodic_handler, &periodictimer, &(struct timeval){0, 0});
#ifndef HAVE_MINGW
signal_t sighup = {0};