X-Git-Url: http://git.meshlink.io/?a=blobdiff_plain;f=src%2Fnet.c;h=9d48a283f368e3346dcef49762e51b7540cab029;hb=1da7f28315be17a4ed854cdceb97a870f16a511e;hp=f9020b3b9aa8d4ab771167fa3cb5760ebe6f4211;hpb=574b380dfc75ef13ee4accba1f2416165c58a5a2;p=meshlink

diff --git a/src/net.c b/src/net.c
index f9020b3b..9d48a283 100644
--- a/src/net.c
+++ b/src/net.c
@@ -1,9 +1,6 @@
 /*
     net.c -- most of the network code
-    Copyright (C) 1998-2005 Ivo Timmermans,
-                  2000-2011 Guus Sliepen <guus@tinc-vpn.org>
-                  2006      Scott Lamb <slamb@slamb.org>
-		  2011      Loïc Grenié <loic.grenie@gmail.com>
+    Copyright (C) 2014 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
@@ -23,129 +20,72 @@
 #include "system.h"
 
 #include "utils.h"
-#include "splay_tree.h"
 #include "conf.h"
 #include "connection.h"
-#include "device.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 "subnet.h"
 #include "xalloc.h"
 
-int contradicting_add_edge = 0;
-int contradicting_del_edge = 0;
-static int sleeptime = 10;
-
-/* Purge edges and subnets of unreachable nodes. Use carefully. */
-
-void purge(void) {
-	splay_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;
-				send_del_subnet(broadcast, s);
-				if(!strictsubnets)
-					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;
-
-				if(e->to == n)
-					break;
-			}
-
-			if(!enode && (!strictsubnets || !n->subnet_tree->head))
-				/* in strictsubnets mode do not delete nodes with subnets */
-				node_del(n);
-		}
-	}
+static const int min(int a, int b) {
+	return a < b ? a : b;
 }
 
 /*
   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) {
-	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 (%s)", c->name, c->hostname);
 
 	c->status.active = false;
 
-	if(c->node)
+	if(c->node && c->node->connection == c)
 		c->node->connection = NULL;
 
 	if(c->edge) {
-		if(report && !tunnelserver)
-			send_del_edge(broadcast, c->edge);
+		if(report)
+			send_del_edge(mesh, mesh->everyone, c->edge);
 
-		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) {
 			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);
+				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);
+	if(outgoing)
+		do_outgoing_connection(mesh, outgoing);
 
-	connection_del(c);
+#ifndef HAVE_MINGW
+	/* Clean up dead proxy processes */
+
+	while(waitpid(-1, NULL, WNOHANG) > 0);
+#endif
 }
 
 /*
@@ -156,237 +96,266 @@ void terminate_connection(connection_t *c, bool report) {
   end does not reply in time, we consider them dead
   and close the connection.
 */
-static void timeout_handler(int fd, short events, void *event) {
-	splay_node_t *node, *next;
-	connection_t *c;
-	time_t now = time(NULL);
+static void timeout_handler(event_loop_t *loop, void *data) {
+	meshlink_handle_t *mesh = loop->data;
 
-	for(node = connection_tree->head; node; node = next) {
-		next = node->next;
-		c = node->data;
-
-		if(c->last_ping_time + pingtimeout <= now) {
+	for list_each(connection_t, c, mesh->connections) {
+		if(c->last_ping_time + mesh->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);
-					terminate_connection(c, true);
+					logger(mesh, MESHLINK_INFO, "%s (%s) didn't respond to PING in %ld seconds", c->name, c->hostname, (long)mesh->loop.now.tv_sec - c->last_ping_time);
+				} else if(c->last_ping_time + mesh->pinginterval <= mesh->loop.now.tv_sec) {
+					send_ping(mesh, c);
 					continue;
-				} else if(c->last_ping_time + pinginterval <= now) {
-					send_ping(c);
-				}
-			} else {
-				if(c->status.connecting) {
-					ifdebug(CONNECTIONS)
-						logger(LOG_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
-					c->status.connecting = false;
-					closesocket(c->socket);
-					do_outgoing_connection(c);
 				} else {
-					ifdebug(CONNECTIONS) logger(LOG_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
-					terminate_connection(c, false);
 					continue;
 				}
+			} else {
+				if(c->status.connecting)
+					logger(mesh, MESHLINK_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
+				else
+					logger(mesh, MESHLINK_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
 			}
+			terminate_connection(mesh, c, c->status.active);
 		}
 	}
 
-	if(contradicting_del_edge > 100 && contradicting_add_edge > 100) {
-		logger(LOG_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", sleeptime);
-		usleep(sleeptime * 1000000LL);
-		sleeptime *= 2;
-		if(sleeptime < 0)
-			sleeptime = 3600;
-	} else {
-		sleeptime /= 2;
-		if(sleeptime < 10)
-			sleeptime = 10;
-	}
-
-	contradicting_add_edge = 0;
-	contradicting_del_edge = 0;
-
-	event_add(event, &(struct timeval){pingtimeout, 0});
+	timeout_set(&mesh->loop, data, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 }
 
-void handle_meta_connection_data(int fd, short events, void *data) {
-	connection_t *c = data;
-	int result;
-	socklen_t len = sizeof result;
-
-	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);
-			return;
+/// Utility function to establish connections based on condition check
+/** The function iterates over all nodes, but skips those that do
+ *  not pass the condition check.
+ *  
+ *  The condition check function is passed
+ *  a pointer to a random number r between 0 and rand_modulo, a pointer to the
+ *  current node index i, and the node pointer n. This function should return true
+ *  if a connection attempt to the node should be made.
+ *  
+ *  @param mesh		A pointer to the mesh structure
+ *  @param rand_modulo	Random index is selected between 0 and rand_modulo
+ *  @cond_check		A function pointer. This function should return true
+ *  			if a connection attempt to the node should be made
+ */
+static void cond_add_connection(meshlink_handle_t *mesh, int rand_modulo, bool (*cond_check)(int*, int*, node_t*)) {
+	int r = rand() % rand_modulo;
+	int i = 0;
+
+	for splay_each(node_t, n, mesh->nodes) {
+		/* skip nodes that do not pass condition check */
+		if(!(*cond_check)(&i, &r, n))
+			continue;
+
+		/* check if there is already a connection attempt to this node */
+		bool found = false;
+		for list_each(outgoing_t, outgoing, mesh->outgoings) {
+			if(!strcmp(outgoing->name, n->name)) {
+				found = true;
+				break;
+			}
 		}
-	}
 
-	if (!receive_meta(c)) {
-		terminate_connection(c, c->status.active);
-		return;
+		if(!found) {
+			//TODO: if the node is blacklisted the connection will not happen, but
+			//the user will read this debug message "Autoconnecting to %s" that is misleading
+			logger(mesh, MESHLINK_INFO, "Autoconnecting to %s", n->name);
+			outgoing_t *outgoing = xzalloc(sizeof *outgoing);
+			outgoing->mesh = mesh;
+			outgoing->name = xstrdup(n->name);
+			list_insert_tail(mesh->outgoings, outgoing);
+			setup_outgoing_connection(mesh, outgoing);
+		}
+		break;
 	}
 }
 
-static void sigterm_handler(int signal, short events, void *data) {
-	logger(LOG_NOTICE, "Got %s signal", strsignal(signal));
-	event_loopexit(NULL);
-}
+static bool found_random_node(int *i, int *r, node_t *n) {
+	if((*i)++ != *r)
+		return false;
 
-static void sighup_handler(int signal, short events, void *data) {
-	logger(LOG_NOTICE, "Got %s signal", strsignal(signal));
-	reopenlogger();
-	reload_configuration();
+	if(n->connection)
+		return false;
+	
+	return true;
 }
 
-static void sigalrm_handler(int signal, short events, void *data) {
-	logger(LOG_NOTICE, "Got %s signal", strsignal(signal));
-	retry();
-}
+static bool found_random_unreachable_node(int *i, int *r, node_t *n) {
+	if(n->status.reachable)
+		return false;
+	
+	if((*i)++ != *r)
+		return false;
+
+	if(n->connection)
+		return false;
 
-int reload_configuration(void) {
-	connection_t *c;
-	splay_node_t *node, *next;
-	char *fname;
-	struct stat s;
-	static time_t last_config_check = 0;
+	return true;
+}
 
-	/* Reread our own configuration file */
+static void periodic_handler(event_loop_t *loop, void *data) {
+	meshlink_handle_t *mesh = loop->data;
 
-	exit_configuration(&config_tree);
-	init_configuration(&config_tree);
+	/* 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.
+	*/
 
-	if(!read_server_config()) {
-		logger(LOG_ERR, "Unable to reread configuration file, exitting.");
-		event_loopexit(NULL);
-		return EINVAL;
+	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);
+		usleep(mesh->sleeptime * 1000000LL);
+		mesh->sleeptime *= 2;
+		if(mesh->sleeptime < 0)
+			mesh->sleeptime = 3600;
+	} else {
+		mesh->sleeptime /= 2;
+		if(mesh->sleeptime < 10)
+			mesh->sleeptime = 10;
 	}
 
-	/* Close connections to hosts that have a changed or deleted host config file */
-	
-	for(node = connection_tree->head; node; node = next) {
-		c = node->data;
-		next = node->next;
-		
-		if(c->outgoing) {
-			free(c->outgoing->name);
-			if(c->outgoing->ai)
-				freeaddrinfo(c->outgoing->ai);
-			free(c->outgoing);
-			c->outgoing = NULL;
+	mesh->contradicting_add_edge = 0;
+	mesh->contradicting_del_edge = 0;
+
+	int timeout = 5;
+
+	/* If AutoConnect is set, check if we need to make or break connections. */
+
+	if(autoconnect && mesh->nodes->count > 1) {
+		/* Count number of active connections */
+		int nc = 0;
+		for list_each(connection_t, c, mesh->connections) {
+			if(c->status.active)
+				nc++;
 		}
-		
-		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);
-	}
 
-	last_config_check = time(NULL);
+		/* Count number of unreachable nodes */
+		int num_unreachable = 0;
+		for splay_each(node_t, n, mesh->nodes) {
+			if(!n->status.reachable)
+				num_unreachable++;
+		}
 
-	/* If StrictSubnet is set, expire deleted Subnets and read new ones in */
+		if(nc < autoconnect) {
+			/* Not enough active connections, try to add one.
+			   Choose a random node, if we don't have a connection to it,
+			   and we are not already trying to make one, create an
+			   outgoing connection to this node.
+			*/
+			cond_add_connection(mesh, mesh->nodes->count, &found_random_node);
+		} else if(num_unreachable > 0) {
+			/* Min number of connections established. Now try
+			   to connect to some unreachable nodes to attempt
+			   to heal possible partitions.
+			*/
+			cond_add_connection(mesh, num_unreachable, &found_random_unreachable_node);
+		}
+		
+		if(nc > autoconnect) {
+			/* Too many active connections, try to remove one.
+			   Choose a random outgoing connection to a node
+			   that has at least one other connection.
+			*/
+			int r = rand() % nc;
+			int i = 0;
+
+			for list_each(connection_t, c, mesh->connections) {
+				if(!c->status.active)
+					continue;
 
-	if(strictsubnets) {
-		subnet_t *subnet;
+				if(i++ != r)
+					continue;
 
+				if(!c->outgoing || !c->node || c->node->edge_tree->count < 2)
+					break;
 
-		for(node = subnet_tree->head; node; node = node->next) {
-			subnet = node->data;
-			subnet->expires = 1;
+				logger(mesh, MESHLINK_INFO, "Autodisconnecting from %s", c->name);
+				list_delete(mesh->outgoings, c->outgoing);
+				c->outgoing = NULL;
+				terminate_connection(mesh, c, c->status.active);
+				break;
+			}
 		}
 
-		load_all_subnets();
-
-		for(node = subnet_tree->head; node; node = next) {
-			next = node->next;
-			subnet = node->data;
-			if(subnet->expires == 1) {
-				send_del_subnet(broadcast, subnet);
-				if(subnet->owner->status.reachable)
-					subnet_update(subnet->owner, subnet, false);
-				subnet_del(subnet->owner, subnet);
-			} else if(subnet->expires == -1) {
-				subnet->expires = 0;
-			} else {
-				send_add_subnet(broadcast, subnet);
-				if(subnet->owner->status.reachable)
-					subnet_update(subnet->owner, subnet, true);
+		if(nc >= autoconnect) {
+			/* If we have enough active connections,
+			   remove any pending outgoing connections.
+			   Do not remove pending connections to unreachable
+			   nodes.
+			*/
+			node_t *o_node = NULL;
+			for list_each(outgoing_t, o, mesh->outgoings) {
+				o_node = lookup_node(mesh, o->name);
+				/* o_node is NULL if it is not part of the graph yet */
+				if(!o_node || !o_node->status.reachable)
+					continue;
+
+				bool found = false;
+				for list_each(connection_t, c, mesh->connections) {
+					if(c->outgoing == o) {
+						found = true;
+						break;
+					}
+				}
+				if(!found) {
+					logger(mesh, MESHLINK_INFO, "Cancelled outgoing connection to %s", o->name);
+					/* The node variable is leaked in from using the list_each macro.
+					   The o variable could be used, but using node directly
+					   is more efficient.
+					*/
+					list_delete_node(mesh->outgoings, node);
+				}
 			}
 		}
+
+		if (nc + mesh->outgoings->count < min(autoconnect, mesh->nodes->count - 1))
+			timeout = 0;
 	}
 
-	/* Try to make outgoing connections */
-	
-	try_outgoing_connections();
+	timeout_set(&mesh->loop, data, &(struct timeval){timeout, rand() % 100000});
+}
 
-	return 0;
+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(void) {
-	connection_t *c;
-	splay_node_t *node;
+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 timeval){0, 0});
+	}
 
-	for(node = connection_tree->head; node; node = node->next) {
-		c = node->data;
-		
-		if(c->outgoing && !c->node) {
-			if(timeout_initialized(&c->outgoing->ev))
-				event_del(&c->outgoing->ev);
-			if(c->status.connecting)
-				close(c->socket);
-			c->outgoing->timeout = 0;
-			do_outgoing_connection(c);
-		}
+	/* Check for outgoing connections that are in progress, and reset their ping timers */
+	for list_each(connection_t, c, mesh->connections) {
+		if(c->outgoing && !c->node)
+			c->last_ping_time = 0;
 	}
+
+	/* Kick the ping timeout handler */
+	timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval){0, 0});
 }
 
 /*
   this is where it all happens...
 */
-int main_loop(void) {
-	struct event timeout_event;
+int main_loop(meshlink_handle_t *mesh) {
+	timeout_add(&mesh->loop, &mesh->pingtimer, timeout_handler, &mesh->pingtimer, &(struct timeval){mesh->pingtimeout, rand() % 100000});
+	timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timeval){0, 0});
 
-	timeout_set(&timeout_event, timeout_handler, &timeout_event);
-	event_add(&timeout_event, &(struct timeval){pingtimeout, 0});
+	//Add signal handler
+	mesh->datafromapp.signum = 0;
+	signal_add(&(mesh->loop),&(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue,mesh, mesh->datafromapp.signum);
 
-#ifndef HAVE_MINGW
-	struct event sighup_event;
-	struct event sigterm_event;
-	struct event sigquit_event;
-	struct event sigalrm_event;
-
-	signal_set(&sighup_event, SIGHUP, sighup_handler, NULL);
-	signal_add(&sighup_event, NULL);
-	signal_set(&sigterm_event, SIGTERM, sigterm_handler, NULL);
-	signal_add(&sigterm_event, NULL);
-	signal_set(&sigquit_event, SIGQUIT, sigterm_handler, NULL);
-	signal_add(&sigquit_event, NULL);
-	signal_set(&sigalrm_event, SIGALRM, sigalrm_handler, NULL);
-	signal_add(&sigalrm_event, NULL);
-#endif
-
-	if(event_loop(0) < 0) {
-		logger(LOG_ERR, "Error while waiting for input: %s", strerror(errno));
+	if(!event_loop_run(&mesh->loop)) {
+		logger(mesh, MESHLINK_ERROR, "Error while waiting for input: %s", strerror(errno));
 		return 1;
 	}
 
-#ifndef HAVE_MINGW
-	signal_del(&sighup_event);
-	signal_del(&sigterm_event);
-	signal_del(&sigquit_event);
-	signal_del(&sigalrm_event);
-#endif
-
-	event_del(&timeout_event);
+	timeout_del(&mesh->loop, &mesh->periodictimer);
+	timeout_del(&mesh->loop, &mesh->pingtimer);
 
 	return 0;
 }