X-Git-Url: http://git.meshlink.io/?a=blobdiff_plain;f=src%2Fnet.c;h=8a63615d658b6321ac1df565dc708be177037f59;hb=8c8b506da9ce578fe6e6e406a91bbd21af13f7a3;hp=feec8d6be238c51552c26635e0acd8cdbce9c90d;hpb=40d91ff619a6ea24a2a35c9d934bcc6bace27e24;p=meshlink

diff --git a/src/net.c b/src/net.c
index feec8d6b..2ee8aee4 100644
--- a/src/net.c
+++ b/src/net.c
@@ -1,8 +1,6 @@
 /*
     net.c -- most of the network code
-    Copyright (C) 1998-2005 Ivo Timmermans,
-                  2000-2010 Guus Sliepen <guus@tinc-vpn.org>
-                  2006      Scott Lamb <slamb@slamb.org>
+    Copyright (C) 2014-2017 Guus Sliepen <guus@meshlink.io>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -21,193 +19,79 @@
 
 #include "system.h"
 
-#include <openssl/rand.h>
-
 #include "utils.h"
-#include "avl_tree.h"
 #include "conf.h"
 #include "connection.h"
-#include "device.h"
-#include "event.h"
 #include "graph.h"
 #include "logger.h"
+#include "meshlink_internal.h"
 #include "meta.h"
 #include "net.h"
 #include "netutl.h"
-#include "process.h"
 #include "protocol.h"
-#include "route.h"
-#include "subnet.h"
+#include "sptps.h"
 #include "xalloc.h"
 
-bool do_purge = false;
-volatile bool running = false;
-
-time_t now = 0;
-
-/* Purge edges and subnets of unreachable nodes. Use carefully. */
-
-static void purge(void) {
-	avl_node_t *nnode, *nnext, *enode, *enext, *snode, *snext;
-	node_t *n;
-	edge_t *e;
-	subnet_t *s;
-
-	ifdebug(PROTOCOL) logger(LOG_DEBUG, "Purging unreachable nodes");
-
-	/* Remove all edges and subnets owned by unreachable nodes. */
-
-	for(nnode = node_tree->head; nnode; nnode = nnext) {
-		nnext = nnode->next;
-		n = nnode->data;
-
-		if(!n->status.reachable) {
-			ifdebug(SCARY_THINGS) logger(LOG_DEBUG, "Purging node %s (%s)", n->name,
-					   n->hostname);
-
-			for(snode = n->subnet_tree->head; snode; snode = snext) {
-				snext = snode->next;
-				s = snode->data;
-				if(!tunnelserver)
-					send_del_subnet(broadcast, s);
-				subnet_del(n, s);
-			}
-
-			for(enode = n->edge_tree->head; enode; enode = enext) {
-				enext = enode->next;
-				e = enode->data;
-				if(!tunnelserver)
-					send_del_edge(broadcast, e);
-				edge_del(e);
-			}
-		}
-	}
-
-	/* Check if anyone else claims to have an edge to an unreachable node. If not, delete node. */
-
-	for(nnode = node_tree->head; nnode; nnode = nnext) {
-		nnext = nnode->next;
-		n = nnode->data;
-
-		if(!n->status.reachable) {
-			for(enode = edge_weight_tree->head; enode; enode = enext) {
-				enext = enode->next;
-				e = enode->data;
+#include <assert.h>
 
-				if(e->to == n)
-					break;
-			}
-
-			if(!enode)
-				node_del(n);
-		}
-	}
+#if !defined(min)
+static inline int min(int a, int b) {
+	return a < b ? a : b;
 }
+#endif
 
-/*
-  put all file descriptors in an fd_set array
-  While we're at it, purge stuff that needs to be removed.
-*/
-static int build_fdset(fd_set *readset, fd_set *writeset) {
-	avl_node_t *node, *next;
-	connection_t *c;
-	int i, max = 0;
-
-	FD_ZERO(readset);
-	FD_ZERO(writeset);
-
-	for(node = connection_tree->head; node; node = next) {
-		next = node->next;
-		c = node->data;
-
-		if(c->status.remove) {
-			connection_del(c);
-			if(!connection_tree->head)
-				purge();
-		} else {
-			FD_SET(c->socket, readset);
-			if(c->outbuflen > 0)
-				FD_SET(c->socket, writeset);
-			if(c->socket > max)
-				max = c->socket;
-		}
-	}
-
-	for(i = 0; i < listen_sockets; i++) {
-		FD_SET(listen_socket[i].tcp, readset);
-		if(listen_socket[i].tcp > max)
-			max = listen_socket[i].tcp;
-		FD_SET(listen_socket[i].udp, readset);
-		if(listen_socket[i].udp > max)
-			max = listen_socket[i].udp;
-	}
-
-	if(device_fd >= 0)
-		FD_SET(device_fd, readset);
-	if(device_fd > max)
-		max = device_fd;
-	
-	return max;
-}
+static const int default_timeout = 5;
+static const int default_interval = 60;
 
 /*
   Terminate a connection:
-  - Close the socket
-  - Remove associated edge and tell other connections about it if report = true
+  - Mark it as inactive
+  - Remove the edge representing this connection
+  - Kill it with fire
   - Check if we need to retry making an outgoing connection
-  - Deactivate the host
 */
-void terminate_connection(connection_t *c, bool report) {
-	if(c->status.remove)
-		return;
-
-	ifdebug(CONNECTIONS) logger(LOG_NOTICE, "Closing connection with %s (%s)",
-			   c->name, c->hostname);
+void terminate_connection(meshlink_handle_t *mesh, connection_t *c, bool report) {
+	logger(mesh, MESHLINK_INFO, "Closing connection with %s", c->name);
 
-	c->status.remove = true;
 	c->status.active = false;
 
-	if(c->node)
+	if(c->node && c->node->connection == c) {
 		c->node->connection = NULL;
-
-	if(c->socket)
-		closesocket(c->socket);
+	}
 
 	if(c->edge) {
-		if(report && !tunnelserver)
-			send_del_edge(broadcast, c->edge);
+		if(report) {
+			send_del_edge(mesh, mesh->everyone, c->edge, 0);
+		}
 
-		edge_del(c->edge);
+		edge_del(mesh, c->edge);
+		c->edge = NULL;
 
 		/* Run MST and SSSP algorithms */
 
-		graph();
+		graph(mesh);
 
 		/* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
 
-		if(report && !c->node->status.reachable) {
+		if(report && c->node && !c->node->status.reachable) {
 			edge_t *e;
-			e = lookup_edge(c->node, myself);
+			e = lookup_edge(c->node, mesh->self);
+
 			if(e) {
-				if(!tunnelserver)
-					send_del_edge(broadcast, e);
-				edge_del(e);
+				send_del_edge(mesh, mesh->everyone, e, 0);
+				edge_del(mesh, e);
 			}
 		}
 	}
 
+	outgoing_t *outgoing = c->outgoing;
+	connection_del(mesh, c);
+
 	/* Check if this was our outgoing connection */
 
-	if(c->outgoing) {
-		retry_outgoing(c->outgoing);
-		c->outgoing = NULL;
+	if(outgoing) {
+		do_outgoing_connection(mesh, outgoing);
 	}
-
-	free(c->outbuf);
-	c->outbuf = NULL;
-	c->outbuflen = 0;
-	c->outbufsize = 0;
-	c->outbufstart = 0;
 }
 
 /*
@@ -218,288 +102,635 @@ void terminate_connection(connection_t *c, bool report) {
   end does not reply in time, we consider them dead
   and close the connection.
 */
-static void check_dead_connections(void) {
-	avl_node_t *node, *next;
-	connection_t *c;
+static void timeout_handler(event_loop_t *loop, void *data) {
+	assert(data);
+
+	meshlink_handle_t *mesh = loop->data;
+	logger(mesh, MESHLINK_DEBUG, "timeout_handler()");
+
+	for list_each(connection_t, c, mesh->connections) {
+		int pingtimeout = c->node ? mesh->dev_class_traits[c->node->devclass].pingtimeout : default_timeout;
+		int pinginterval = c->node ? mesh->dev_class_traits[c->node->devclass].pinginterval : default_interval;
 
-	for(node = connection_tree->head; node; node = next) {
-		next = node->next;
-		c = node->data;
+		if(c->outgoing && c->outgoing->timeout < 5) {
+			pingtimeout = 1;
+		}
+
+		// Also make sure that if outstanding key requests for the UDP counterpart of a connection has timed out, we restart it.
+		if(c->node) {
+			if(c->node->status.waitingforkey && c->node->last_req_key + pingtimeout <= mesh->loop.now.tv_sec) {
+				send_req_key(mesh, c->node);
+			}
+		}
+
+		if(c->status.active && c->last_key_renewal + 3600 < mesh->loop.now.tv_sec) {
+			if(!sptps_force_kex(&c->sptps)) {
+				logger(mesh, MESHLINK_ERROR, "SPTPS key renewal for connection with %s failed", c->name);
+				terminate_connection(mesh, c, true);
+				continue;
+			} else {
+				c->last_key_renewal = mesh->loop.now.tv_sec;
+			}
+		}
 
-		if(c->last_ping_time + pingtimeout < now) {
+		if(c->last_ping_time + pingtimeout <= mesh->loop.now.tv_sec) {
 			if(c->status.active) {
 				if(c->status.pinged) {
-					ifdebug(CONNECTIONS) logger(LOG_INFO, "%s (%s) didn't respond to PING in %ld seconds",
-							   c->name, c->hostname, now - c->last_ping_time);
-					c->status.timeout = true;
-					terminate_connection(c, true);
-				} else if(c->last_ping_time + pinginterval < now) {
-					send_ping(c);
-				}
-			} else {
-				if(c->status.remove) {
-					logger(LOG_WARNING, "Old connection_t for %s (%s) status %04x still lingering, deleting...",
-						   c->name, c->hostname, bitfield_to_int(&c->status, sizeof c->status));
-					connection_del(c);
+					logger(mesh, MESHLINK_INFO, "%s didn't respond to PING in %ld seconds", c->name, (long)mesh->loop.now.tv_sec - c->last_ping_time);
+				} else if(c->last_ping_time + pinginterval <= mesh->loop.now.tv_sec) {
+					send_ping(mesh, c);
+					continue;
+				} else {
 					continue;
 				}
-				ifdebug(CONNECTIONS) logger(LOG_WARNING, "Timeout from %s (%s) during authentication",
-						   c->name, c->hostname);
+			} else {
 				if(c->status.connecting) {
-					c->status.connecting = false;
-					closesocket(c->socket);
-					do_outgoing_connection(c);
+					logger(mesh, MESHLINK_WARNING, "Timeout while connecting to %s", c->name);
 				} else {
-					terminate_connection(c, false);
+					logger(mesh, MESHLINK_WARNING, "Timeout from %s during authentication", c->name);
 				}
 			}
-		}
 
-		if(c->outbuflen > 0 && c->last_flushed_time + pingtimeout < now) {
-			if(c->status.active) {
-				ifdebug(CONNECTIONS) logger(LOG_INFO,
-						"%s (%s) could not flush for %ld seconds (%d bytes remaining)",
-						c->name, c->hostname, now - c->last_flushed_time, c->outbuflen);
-				c->status.timeout = true;
-				terminate_connection(c, true);
-			}
+			terminate_connection(mesh, c, c->status.active);
 		}
 	}
+
+	timeout_set(&mesh->loop, data, &(struct timespec) {
+		1, prng(mesh, TIMER_FUDGE)
+	});
 }
 
-/*
-  check all connections to see if anything
-  happened on their sockets
-*/
-static void check_network_activity(fd_set * readset, fd_set * writeset) {
-	connection_t *c;
-	avl_node_t *node;
-	int result, i;
-	socklen_t len = sizeof(result);
-	vpn_packet_t packet;
-
-	/* check input from kernel */
-	if(device_fd >= 0 && FD_ISSET(device_fd, readset)) {
-		if(read_packet(&packet)) {
-			packet.priority = 0;
-			route(myself, &packet);
-		}
+// devclass asc, last_successfull_connection desc
+static int node_compare_devclass_asc_lsc_desc(const void *a, const void *b) {
+	const node_t *na = a, *nb = b;
+
+	if(na->devclass < nb->devclass) {
+		return -1;
 	}
 
-	/* check meta connections */
-	for(node = connection_tree->head; node; node = node->next) {
-		c = node->data;
+	if(na->devclass > nb->devclass) {
+		return 1;
+	}
 
-		if(c->status.remove)
-			continue;
+	if(na->last_successfull_connection == nb->last_successfull_connection) {
+		return 0;
+	}
 
-		if(FD_ISSET(c->socket, readset)) {
-			if(c->status.connecting) {
-				c->status.connecting = false;
-				getsockopt(c->socket, SOL_SOCKET, SO_ERROR, &result, &len);
-
-				if(!result)
-					finish_connecting(c);
-				else {
-					ifdebug(CONNECTIONS) logger(LOG_DEBUG,
-							   "Error while connecting to %s (%s): %s",
-							   c->name, c->hostname, sockstrerror(result));
-					closesocket(c->socket);
-					do_outgoing_connection(c);
-					continue;
-				}
-			}
+	if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection) {
+		return -1;
+	}
 
-			if(!receive_meta(c)) {
-				terminate_connection(c, c->status.active);
-				continue;
-			}
-		}
+	if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection) {
+		return 1;
+	}
 
-		if(FD_ISSET(c->socket, writeset)) {
-			if(!flush_meta(c)) {
-				terminate_connection(c, c->status.active);
-				continue;
-			}
-		}
+	if(na < nb) {
+		return -1;
+	}
+
+	if(na > nb) {
+		return 1;
 	}
 
-	for(i = 0; i < listen_sockets; i++) {
-		if(FD_ISSET(listen_socket[i].udp, readset))
-			handle_incoming_vpn_data(listen_socket[i].udp);
+	return 0;
+}
+
+// last_successfull_connection desc
+static int node_compare_lsc_desc(const void *a, const void *b) {
+	const node_t *na = a, *nb = b;
 
-		if(FD_ISSET(listen_socket[i].tcp, readset))
-			handle_new_meta_connection(listen_socket[i].tcp);
+	if(na->last_successfull_connection == nb->last_successfull_connection) {
+		return 0;
 	}
+
+	if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection) {
+		return -1;
+	}
+
+	if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection) {
+		return 1;
+	}
+
+	if(na < nb) {
+		return -1;
+	}
+
+	if(na > nb) {
+		return 1;
+	}
+
+	return 0;
 }
 
+// devclass desc
+static int node_compare_devclass_desc(const void *a, const void *b) {
+	const node_t *na = a, *nb = b;
+
+	if(na->devclass < nb->devclass) {
+		return -1;
+	}
+
+	if(na->devclass > nb->devclass) {
+		return 1;
+	}
+
+	if(na < nb) {
+		return -1;
+	}
+
+	if(na > nb) {
+		return 1;
+	}
+
+	return 0;
+}
+
+
 /*
-  this is where it all happens...
+
+autoconnect()
+{
+        timeout = 5
+
+        // find the best one for initial connect
+
+        if cur < min
+                newcon =
+                        first from nodes
+                                where dclass <= my.dclass and !connection and (timestamp - last_retry) > retry_timeout
+                                order by dclass asc, last_connection desc
+                if newcon
+                        timeout = 0
+                        goto connect
+
+
+        // find better nodes to connect to: in case we have less than min connections within [BACKBONE, i] and there are nodes which we are not connected to within the range
+
+        if min <= cur < max
+                j = 0
+                for i = BACKBONE to my.dclass
+                        j += count(from connections where node.dclass = i)
+                        if j < min
+                                newcon =
+                                        first from nodes
+                                                where dclass = i and !connection and (timestamp - last_retry) > retry_timeout
+                                                order by last_connection desc
+                                if newcon
+                                        goto connect
+                        else
+                                break
+
+
+        // heal partitions
+
+        if min <= cur < max
+                newcon =
+                        first from nodes
+                                where dclass <= my.dclass and !reachable and (timestamp - last_retry) > retry_timeout
+                                order by dclass asc, last_connection desc
+                if newcon
+                        goto connect
+
+
+        // connect
+
+connect:
+        if newcon
+                connect newcon
+
+
+        // disconnect outgoing connections in case we have more than min connections within [BACKBONE, i] and there are nodes which we are connected to within the range [i, PORTABLE]
+
+        if min < cur <= max
+                j = 0
+                for i = BACKBONE to my.dclass
+                        j += count(from connections where node.dclass = i)
+                        if min < j
+                                delcon =
+                                        first from nodes
+                                                where dclass >= i and outgoing_connection
+                                                order by dclass desc
+                                if disconnect
+                                        goto disconnect
+                                else
+                                        break
+
+
+        // disconnect connections in case we have more than enough connections
+
+        if max < cur
+                delcon =
+                        first from nodes
+                                where outgoing_connection
+                                order by dclass desc
+                goto disconnect
+
+        // disconnect
+
+disconnect
+        if delcon
+                disconnect delcon
+
+
+        // next iteration
+        next (timeout, autoconnect)
+
+}
+
 */
-int main_loop(void) {
-	fd_set readset, writeset;
-	struct timeval tv;
-	int r, maxfd;
-	time_t last_ping_check, last_config_check, last_graph_dump;
-	event_t *event;
 
-	last_ping_check = now;
-	last_config_check = now;
-	last_graph_dump = now;
-	
-	srand(now);
 
-	running = true;
+static void periodic_handler(event_loop_t *loop, void *data) {
+	meshlink_handle_t *mesh = loop->data;
 
-	while(running) {
-		now = time(NULL);
+	/* Check if there are too many contradicting ADD_EDGE and DEL_EDGE messages.
+	   This usually only happens when another node has the same Name as this node.
+	   If so, sleep for a short while to prevent a storm of contradicting messages.
+	*/
 
-	//	tv.tv_sec = 1 + (rand() & 7);	/* Approx. 5 seconds, randomized to prevent global synchronisation effects */
-		tv.tv_sec = 1;
-		tv.tv_usec = 0;
+	if(mesh->contradicting_del_edge > 100 && mesh->contradicting_add_edge > 100) {
+		logger(mesh, MESHLINK_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", mesh->sleeptime);
+		struct timespec ts = {mesh->sleeptime, 0};
+		clock_nanosleep(CLOCK_MONOTONIC, 0, &ts, NULL);
+		mesh->sleeptime *= 2;
 
-		maxfd = build_fdset(&readset, &writeset);
+		if(mesh->sleeptime < 0) {
+			mesh->sleeptime = 3600;
+		}
+	} else {
+		mesh->sleeptime /= 2;
+
+		if(mesh->sleeptime < 10) {
+			mesh->sleeptime = 10;
+		}
+	}
+
+	mesh->contradicting_add_edge = 0;
+	mesh->contradicting_del_edge = 0;
+
+	int timeout = default_timeout;
+
+	/* Check if we need to make or break connections. */
+
+	if(mesh->nodes->count > 1) {
+
+		logger(mesh, MESHLINK_DEBUG, "--- autoconnect begin ---");
+
+		int retry_timeout = min(mesh->nodes->count * default_timeout, 60);
+
+		logger(mesh, MESHLINK_DEBUG, "* devclass = %d", mesh->devclass);
+		logger(mesh, MESHLINK_DEBUG, "* nodes = %d", mesh->nodes->count);
+		logger(mesh, MESHLINK_DEBUG, "* retry_timeout = %d", retry_timeout);
+
+
+		// connect disconnect nodes
+
+		node_t *connect_to = NULL;
+		node_t *disconnect_from = NULL;
 
-#ifdef HAVE_MINGW
-		LeaveCriticalSection(&mutex);
-#endif
-		r = select(maxfd + 1, &readset, &writeset, NULL, &tv);
-#ifdef HAVE_MINGW
-		EnterCriticalSection(&mutex);
-#endif
 
-		if(r < 0) {
-			if(!sockwouldblock(sockerrno)) {
-				logger(LOG_ERR, "Error while waiting for input: %s", sockstrerror(sockerrno));
-				dump_connections();
-				return 1;
+		// get cur_connects
+
+		unsigned int cur_connects = 0;
+
+		for list_each(connection_t, c, mesh->connections) {
+			if(c->status.active) {
+				cur_connects += 1;
 			}
 		}
 
-		if(r > 0)
-			check_network_activity(&readset, &writeset);
+		logger(mesh, MESHLINK_DEBUG, "* cur_connects = %d", cur_connects);
+		logger(mesh, MESHLINK_DEBUG, "* outgoings = %d", mesh->outgoings->count);
+
+		// get min_connects and max_connects
+
+		unsigned int min_connects = mesh->dev_class_traits[mesh->devclass].min_connects;
+		unsigned int max_connects = mesh->dev_class_traits[mesh->devclass].max_connects;
+
+		logger(mesh, MESHLINK_DEBUG, "* min_connects = %d", min_connects);
+		logger(mesh, MESHLINK_DEBUG, "* max_connects = %d", max_connects);
+
+		// find the best one for initial connect
+
+		if(cur_connects < min_connects) {
+			splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
+
+			for splay_each(node_t, n, mesh->nodes) {
+				logger(mesh, MESHLINK_DEBUG, "* %s->devclass = %d", n->name, n->devclass);
+
+				if(n != mesh->self && n->devclass <= mesh->devclass && !n->connection && !n->status.blacklisted && (n->last_connect_try == 0 || (mesh->loop.now.tv_sec - n->last_connect_try) > retry_timeout)) {
+					splay_insert(nodes, n);
+				}
+			}
+
+			if(nodes->head) {
+				//timeout = 0;
+				connect_to = (node_t *)nodes->head->data;
+
+				logger(mesh, MESHLINK_DEBUG, "* found best one for initial connect: %s", connect_to->name);
+			} else {
+				logger(mesh, MESHLINK_DEBUG, "* could not find node for initial connect");
+			}
 
-		if(do_purge) {
-			purge();
-			do_purge = false;
+			splay_delete_tree(nodes);
 		}
 
-		/* Let's check if everybody is still alive */
 
-		if(last_ping_check + pingtimeout < now) {
-			check_dead_connections();
-			last_ping_check = now;
+		// find better nodes to connect to
 
-			if(routing_mode == RMODE_SWITCH)
-				age_subnets();
+		if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects) {
+			unsigned int connects = 0;
 
-			age_past_requests();
+			for(dev_class_t devclass = 0; devclass <= mesh->devclass; ++devclass) {
+				for list_each(connection_t, c, mesh->connections) {
+					if(c->status.active && c->node && c->node->devclass == devclass) {
+						connects += 1;
+					}
+				}
 
-			/* Should we regenerate our key? */
+				if(connects < min_connects) {
+					splay_tree_t *nodes = splay_alloc_tree(node_compare_lsc_desc, NULL);
 
-			if(keyexpires < now) {
-				avl_node_t *node;
-				node_t *n;
+					for splay_each(node_t, n, mesh->nodes) {
+						if(n != mesh->self && n->devclass == devclass && !n->connection && !n->status.blacklisted && (n->last_connect_try == 0 || (mesh->loop.now.tv_sec - n->last_connect_try) > retry_timeout)) {
+							splay_insert(nodes, n);
+						}
+					}
 
-				ifdebug(STATUS) logger(LOG_INFO, "Expiring symmetric keys");
+					if(nodes->head) {
+						logger(mesh, MESHLINK_DEBUG, "* found better node");
+						connect_to = (node_t *)nodes->head->data;
 
-				for(node = node_tree->head; node; node = node->next) {
-					n = node->data;
-					if(n->inkey) {
-						free(n->inkey);
-						n->inkey = NULL;
+						splay_delete_tree(nodes);
+						break;
 					}
+
+					splay_delete_tree(nodes);
+				} else {
+					break;
 				}
+			}
 
-				send_key_changed(broadcast, myself);
-				keyexpires = now + keylifetime;
+			if(!connect_to) {
+				logger(mesh, MESHLINK_DEBUG, "* could not find better nodes");
 			}
 		}
 
-		if(sigalrm) {
-			avl_node_t *node;
-			logger(LOG_INFO, "Flushing event queue");
-			expire_events();
-			for(node = connection_tree->head; node; node = node->next) {
-				connection_t *c = node->data;
-				send_ping(c);
+
+		// heal partitions
+
+		if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects) {
+			splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
+
+			for splay_each(node_t, n, mesh->nodes) {
+				if(n != mesh->self && n->devclass <= mesh->devclass && !n->status.reachable && !n->status.blacklisted && (n->last_connect_try == 0 || (mesh->loop.now.tv_sec - n->last_connect_try) > retry_timeout)) {
+					splay_insert(nodes, n);
+				}
+			}
+
+			if(nodes->head) {
+				logger(mesh, MESHLINK_DEBUG, "* try to heal partition");
+				connect_to = (node_t *)nodes->head->data;
+			} else {
+				logger(mesh, MESHLINK_DEBUG, "* could not find nodes for partition healing");
 			}
-			sigalrm = false;
+
+			splay_delete_tree(nodes);
 		}
 
-		while((event = get_expired_event())) {
-			event->handler(event->data);
-			free_event(event);
+
+		// perform connect
+
+		if(connect_to && !connect_to->connection) {
+			connect_to->last_connect_try = mesh->loop.now.tv_sec;
+			logger(mesh, MESHLINK_DEBUG, "Autoconnect trying to connect to %s", connect_to->name);
+
+			/* check if there is already a connection attempt to this node */
+			bool skip = false;
+
+			for list_each(outgoing_t, outgoing, mesh->outgoings) {
+				if(outgoing->node == connect_to) {
+					logger(mesh, MESHLINK_DEBUG, "* skip autoconnect since it is an outgoing connection already");
+					skip = true;
+					break;
+				}
+			}
+
+			if(!connect_to->status.reachable && !node_read_public_key(mesh, connect_to)) {
+				logger(mesh, MESHLINK_DEBUG, "* skip autoconnect since we don't know this node's public key");
+				skip = true;
+			}
+
+			if(!skip) {
+				logger(mesh, MESHLINK_DEBUG, "Autoconnecting to %s", connect_to->name);
+				outgoing_t *outgoing = xzalloc(sizeof(outgoing_t));
+				outgoing->node = connect_to;
+				list_insert_tail(mesh->outgoings, outgoing);
+				setup_outgoing_connection(mesh, outgoing);
+			}
 		}
 
-		if(sighup) {
-			connection_t *c;
-			avl_node_t *node, *next;
-			char *fname;
-			struct stat s;
-			
-			sighup = false;
-			
-			/* Reread our own configuration file */
-
-			exit_configuration(&config_tree);
-			init_configuration(&config_tree);
-
-			if(!read_server_config()) {
-				logger(LOG_ERR, "Unable to reread configuration file, exitting.");
-				return 1;
+
+		// disconnect suboptimal outgoing connections
+
+		if(min_connects < cur_connects /*&& cur_connects <= max_connects*/) {
+			unsigned int connects = 0;
+
+			for(dev_class_t devclass = 0; devclass <= mesh->devclass; ++devclass) {
+				for list_each(connection_t, c, mesh->connections) {
+					if(c->status.active && c->node && c->node->devclass == devclass) {
+						connects += 1;
+					}
+				}
+
+				if(min_connects < connects) {
+					splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
+
+					for list_each(connection_t, c, mesh->connections) {
+						if(c->outgoing && c->node && c->node->devclass >= devclass) {
+							splay_insert(nodes, c->node);
+						}
+					}
+
+					if(nodes->head) {
+						logger(mesh, MESHLINK_DEBUG, "* disconnect suboptimal outgoing connection");
+						disconnect_from = (node_t *)nodes->head->data;
+					}
+
+					splay_delete_tree(nodes);
+					break;
+				}
+			}
+
+			if(!disconnect_from) {
+				logger(mesh, MESHLINK_DEBUG, "* no suboptimal outgoing connections");
 			}
+		}
 
-			/* Cancel non-active outgoing connections */
 
-			for(node = connection_tree->head; node; node = next) {
-				next = node->next;
-				c = node->data;
+		// disconnect connections (too many connections)
 
-				c->outgoing = NULL;
+		if(!disconnect_from && max_connects < cur_connects) {
+			splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 
-				if(c->status.connecting) {
-					terminate_connection(c, false);
-					connection_del(c);
+			for list_each(connection_t, c, mesh->connections) {
+				if(c->status.active && c->node) {
+					splay_insert(nodes, c->node);
 				}
 			}
 
-			/* Wipe list of outgoing connections */
+			if(nodes->head) {
+				logger(mesh, MESHLINK_DEBUG, "* disconnect connection (too many connections)");
+
+				//timeout = 0;
+				disconnect_from = (node_t *)nodes->head->data;
+			} else {
+				logger(mesh, MESHLINK_DEBUG, "* no node we want to disconnect, even though we have too many connections");
+			}
+
+			splay_delete_tree(nodes);
+		}
+
+
+		// perform disconnect
+
+		if(disconnect_from && disconnect_from->connection) {
+			logger(mesh, MESHLINK_DEBUG, "Autodisconnecting from %s", disconnect_from->connection->name);
+			list_delete(mesh->outgoings, disconnect_from->connection->outgoing);
+			disconnect_from->connection->outgoing = NULL;
+			terminate_connection(mesh, disconnect_from->connection, disconnect_from->connection->status.active);
+		}
+
+		// reduce timeout if we don't have enough connections + outgoings
+		if(cur_connects + mesh->outgoings->count < 3) {
+			timeout = 1;
+		}
+
+		// done!
 
-			for(list_node_t *node = outgoing_list->head; node; node = node->next) {
-				outgoing_t *outgoing = node->data;
+		logger(mesh, MESHLINK_DEBUG, "--- autoconnect end ---");
+	}
 
-				if(outgoing->event)
-					event_del(outgoing->event);
+	for splay_each(node_t, n, mesh->nodes) {
+		if(n->status.dirty) {
+			if(!node_write_config(mesh, n)) {
+				logger(mesh, MESHLINK_DEBUG, "Could not update %s", n->name);
 			}
 
-			list_delete_list(outgoing_list);
-
-			/* Close connections to hosts that have a changed or deleted host config file */
-			
-			for(node = connection_tree->head; node; node = node->next) {
-				c = node->data;
-				
-				xasprintf(&fname, "%s/hosts/%s", confbase, c->name);
-				if(stat(fname, &s) || s.st_mtime > last_config_check)
-					terminate_connection(c, c->status.active);
-				free(fname);
+			n->status.dirty = false;
+		}
+
+		if(n->status.validkey && n->last_req_key + 3600 < mesh->loop.now.tv_sec) {
+			logger(mesh, MESHLINK_DEBUG, "SPTPS key renewal for node %s", n->name);
+
+			if(!sptps_force_kex(&n->sptps)) {
+				logger(mesh, MESHLINK_ERROR, "SPTPS key renewal for node %s failed", n->name);
+				n->status.validkey = false;
+				sptps_stop(&n->sptps);
+				n->status.waitingforkey = false;
+				n->last_req_key = 0;
+			} else {
+				n->last_req_key = mesh->loop.now.tv_sec;
 			}
+		}
+	}
+
+	timeout_set(&mesh->loop, data, &(struct timespec) {
+		timeout, prng(mesh, TIMER_FUDGE)
+	});
+}
 
-			last_config_check = now;
+void handle_meta_connection_data(meshlink_handle_t *mesh, connection_t *c) {
+	if(!receive_meta(mesh, c)) {
+		terminate_connection(mesh, c, c->status.active);
+		return;
+	}
+}
+
+void retry(meshlink_handle_t *mesh) {
+	/* Reset the reconnection timers for all outgoing connections */
+	for list_each(outgoing_t, outgoing, mesh->outgoings) {
+		outgoing->timeout = 0;
+
+		if(outgoing->ev.cb) {
+			timeout_set(&mesh->loop, &outgoing->ev, &(struct timespec) {
+				0, 0
+			});
+		}
+	}
+
+	/* For active connections, check if their addresses are still valid.
+	 * If yes, reset their ping timers, otherwise terminate them. */
+	for list_each(connection_t, c, mesh->connections) {
+		if(!c->status.active) {
+			continue;
+		}
+
+		if(!c->status.pinged) {
+			c->last_ping_time = 0;
+		}
+
+		sockaddr_t sa;
+		socklen_t salen = sizeof(sa);
+
+		if(getsockname(c->socket, &sa.sa, &salen)) {
+			continue;
+		}
 
-			/* Try to make outgoing connections */
-			
-			try_outgoing_connections();
+		switch(sa.sa.sa_family) {
+		case AF_INET:
+			sa.in.sin_port = 0;
+			break;
+
+		case AF_INET6:
+			sa.in6.sin6_port = 0;
+			break;
+
+		default:
+			continue;
+		}
+
+		int sock = socket(sa.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
+
+		if(sock != -1) {
+			continue;
 		}
-		
-		/* Dump graph if wanted every 60 seconds*/
 
-		if(last_graph_dump + 60 < now) {
-			dump_graph();
-			last_graph_dump = now;
+		if(bind(sock, &sa.sa, salen) && errno == EADDRNOTAVAIL) {
+			logger(mesh, MESHLINK_DEBUG, "Local address for connection to %s no longer valid, terminating", c->name);
+			terminate_connection(mesh, c, c->status.active);
 		}
+
+		closesocket(sock);
 	}
 
-	return 0;
+	/* Kick the ping timeout handler */
+	timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timespec) {
+		0, 0
+	});
+}
+
+/*
+  this is where it all happens...
+*/
+void main_loop(meshlink_handle_t *mesh) {
+	timeout_add(&mesh->loop, &mesh->pingtimer, timeout_handler, &mesh->pingtimer, &(struct timespec) {
+		1, prng(mesh, TIMER_FUDGE)
+	});
+	timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timespec) {
+		0, 0
+	});
+
+	//Add signal handler
+	mesh->datafromapp.signum = 0;
+	signal_add(&mesh->loop, &mesh->datafromapp, meshlink_send_from_queue, mesh, mesh->datafromapp.signum);
+
+	if(!event_loop_run(&mesh->loop, &mesh->mutex)) {
+		logger(mesh, MESHLINK_ERROR, "Error while waiting for input: %s", strerror(errno));
+		call_error_cb(mesh, MESHLINK_ENETWORK);
+	}
+
+	signal_del(&mesh->loop, &mesh->datafromapp);
+	timeout_del(&mesh->loop, &mesh->periodictimer);
+	timeout_del(&mesh->loop, &mesh->pingtimer);
 }