Releasing 1.1pre15.
[tinc] / src / net.c
index 3d17d3c..1cb467d 100644 (file)
--- a/src/net.c
+++ b/src/net.c
@@ -1,7 +1,7 @@
 /*
     net.c -- most of the network code
     Copyright (C) 1998-2005 Ivo Timmermans,
-                  2000-2013 Guus Sliepen <guus@tinc-vpn.org>
+                  2000-2017 Guus Sliepen <guus@tinc-vpn.org>
                   2006      Scott Lamb <slamb@slamb.org>
                   2011      Loïc Grenié <loic.grenie@gmail.com>
 
@@ -22,7 +22,7 @@
 
 #include "system.h"
 
-#include "utils.h"
+#include "autoconnect.h"
 #include "conf.h"
 #include "connection.h"
 #include "device.h"
 #include "names.h"
 #include "net.h"
 #include "netutl.h"
-#include "process.h"
 #include "protocol.h"
 #include "subnet.h"
+#include "utils.h"
 #include "xalloc.h"
 
 int contradicting_add_edge = 0;
 int contradicting_del_edge = 0;
 static int sleeptime = 10;
 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. */
 
@@ -92,31 +95,31 @@ void purge(void) {
 void terminate_connection(connection_t *c, bool report) {
        logger(DEBUG_CONNECTIONS, LOG_NOTICE, "Closing connection with %s (%s)", c->name, c->hostname);
 
-       c->status.active = false;
-
-       if(c->node && c->node->connection == c)
-               c->node->connection = NULL;
+       if(c->node) {
+               if(c->node->connection == c)
+                       c->node->connection = NULL;
 
-       if(c->edge) {
-               if(report && !tunnelserver)
-                       send_del_edge(everyone, c->edge);
+               if(c->edge) {
+                       if(report && !tunnelserver)
+                               send_del_edge(everyone, c->edge);
 
-               edge_del(c->edge);
-               c->edge = NULL;
+                       edge_del(c->edge);
+                       c->edge = NULL;
 
-               /* Run MST and SSSP algorithms */
+                       /* Run MST and SSSP algorithms */
 
-               graph();
+                       graph();
 
-               /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
+                       /* 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);
-                       if(e) {
-                               if(!tunnelserver)
-                                       send_del_edge(everyone, e);
-                               edge_del(e);
+                       if(report && !c->node->status.reachable) {
+                               edge_t *e;
+                               e = lookup_edge(c->node, myself);
+                               if(e) {
+                                       if(!tunnelserver)
+                                               send_del_edge(everyone, e);
+                                       edge_del(e);
+                               }
                        }
                }
        }
@@ -128,6 +131,12 @@ void terminate_connection(connection_t *c, bool report) {
 
        if(outgoing)
                do_outgoing_connection(outgoing);
+
+#ifndef HAVE_MINGW
+       /* Clean up dead proxy processes */
+
+       while(waitpid(-1, NULL, WNOHANG) > 0);
+#endif
 }
 
 /*
@@ -139,31 +148,79 @@ void terminate_connection(connection_t *c, bool report) {
   and close the connection.
 */
 static void timeout_handler(void *data) {
+
+       bool close_all_connections = false;
+
+       /*
+                timeout_handler 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) {
+               // control connections (eg. tinc ctl) do not have any timeout
                if(c->status.control)
                        continue;
 
-               if(c->last_ping_time + pingtimeout <= now.tv_sec) {
-                       if(c->status.active) {
-                               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);
-                               } else if(c->last_ping_time + pinginterval <= now.tv_sec) {
-                                       send_ping(c);
-                                       continue;
-                               } else {
-                                       continue;
-                               }
-                       } else {
-                               if(c->status.connecting)
-                                       logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
-                               else
-                                       logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
-                       }
-                       terminate_connection(c, c->status.active);
+               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;
                }
+
+               // Bail out early if we haven't reached the ping timeout for this node yet
+               if(c->last_ping_time + pingtimeout > now.tv_sec)
+                       continue;
+
+               // timeout during connection establishing
+               if(!c->edge) {
+                       if(c->status.connecting)
+                               logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
+                       else
+                               logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
+
+                       terminate_connection(c, c->edge);
+                       continue;
+               }
+
+               // helps in UDP holepunching
+               try_tx(c->node, false);
+
+               // timeout during ping
+               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));
+                       terminate_connection(c, c->edge);
+                       continue;
+               }
+
+               // check whether we need to send a new ping
+               if(c->last_ping_time + pinginterval <= now.tv_sec)
+                       send_ping(c);
        }
 
-       timeout_set(data, &(struct timeval){pingtimeout, rand() % 100000});
+       timeout_set(data, &(struct timeval){1, rand() % 100000});
 }
 
 static void periodic_handler(void *data) {
@@ -174,7 +231,7 @@ static void periodic_handler(void *data) {
 
        if(contradicting_del_edge > 100 && contradicting_add_edge > 100) {
                logger(DEBUG_ALWAYS, LOG_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", sleeptime);
-               usleep(sleeptime * 1000000LL);
+               nanosleep(&(struct timespec){sleeptime, 0}, NULL);
                sleeptime *= 2;
                if(sleeptime < 0)
                        sleeptime = 3600;
@@ -189,121 +246,20 @@ static void periodic_handler(void *data) {
 
        /* If AutoConnect is set, check if we need to make or break connections. */
 
-       if(autoconnect && node_tree->count > 1) {
-               /* Count number of active connections */
-               int nc = 0;
-               for list_each(connection_t, c, connection_list) {
-                       if(c->status.active && !c->status.control)
-                               nc++;
-               }
-
-               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.
-                       */
-                       int r = rand() % node_tree->count;
-                       int i = 0;
-
-                       for splay_each(node_t, n, node_tree) {
-                               if(i++ != r)
-                                       continue;
-
-                               if(n->connection)
-                                       break;
-
-                               bool found = false;
-
-                               for list_each(outgoing_t, outgoing, outgoing_list) {
-                                       if(!strcmp(outgoing->name, n->name)) {
-                                               found = true;
-                                               break;
-                                       }
-                               }
-
-                               if(!found) {
-                                       logger(DEBUG_CONNECTIONS, LOG_INFO, "Autoconnecting to %s", n->name);
-                                       outgoing_t *outgoing = xmalloc_and_zero(sizeof *outgoing);
-                                       outgoing->name = xstrdup(n->name);
-                                       list_insert_tail(outgoing_list, outgoing);
-                                       setup_outgoing_connection(outgoing);
-                               }
-                               break;
-                       }
-               } else 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, connection_list) {
-                               if(!c->status.active || c->status.control)
-                                       continue;
-
-                               if(i++ != r)
-                                       continue;
-
-                               if(!c->outgoing || !c->node || c->node->edge_tree->count < 2)
-                                       break;
-
-                               logger(DEBUG_CONNECTIONS, LOG_INFO, "Autodisconnecting from %s", c->name);
-                               list_delete(outgoing_list, c->outgoing);
-                               c->outgoing = NULL;
-                               terminate_connection(c, c->status.active);
-                               break;
-                       }
-               }
-
-               if(nc >= autoconnect) {
-                       /* If we have enough active connections,
-                          remove any pending outgoing connections.
-                       */
-                       for list_each(outgoing_t, o, outgoing_list) {
-                               bool found = false;
-                               for list_each(connection_t, c, connection_list) {
-                                       if(c->outgoing == o) {
-                                               found = true;
-                                               break;
-                                       }
-                               }
-                               if(!found) {
-                                       logger(DEBUG_CONNECTIONS, LOG_INFO, "Cancelled outgoing connection to %s", o->name);
-                                       list_delete_node(outgoing_list, node);
-                               }
-                       }
-               }
-       }
+       if(autoconnect && node_tree->count > 1)
+               do_autoconnect();
 
        timeout_set(data, &(struct timeval){5, rand() % 100000});
 }
 
 void handle_meta_connection_data(connection_t *c) {
-       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 {
-                       logger(DEBUG_CONNECTIONS, LOG_DEBUG, "Error while connecting to %s (%s): %s", c->name, c->hostname, sockstrerror(result));
-                       terminate_connection(c, false);
-                       return;
-               }
-       }
-
        if (!receive_meta(c)) {
-               terminate_connection(c, c->status.active);
+               terminate_connection(c, c->edge);
                return;
        }
 }
 
+#ifndef HAVE_MINGW
 static void sigterm_handler(void *data) {
        logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum));
        event_exit();
@@ -312,16 +268,18 @@ static void sigterm_handler(void *data) {
 static void sighup_handler(void *data) {
        logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum));
        reopenlogger();
-       reload_configuration();
+       if(reload_configuration())
+               exit(1);
 }
 
 static void sigalrm_handler(void *data) {
        logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum));
        retry();
 }
+#endif
 
 int reload_configuration(void) {
-       char *fname;
+       char fname[PATH_MAX];
 
        /* Reread our own configuration file */
 
@@ -329,16 +287,14 @@ int reload_configuration(void) {
        init_configuration(&config_tree);
 
        if(!read_server_config()) {
-               logger(DEBUG_ALWAYS, LOG_ERR, "Unable to reread configuration file, exitting.");
-               event_exit();
+               logger(DEBUG_ALWAYS, LOG_ERR, "Unable to reread configuration file.");
                return EINVAL;
        }
 
        read_config_options(config_tree, NULL);
 
-       xasprintf(&fname, "%s" SLASH "hosts" SLASH "%s", confbase, myself->name);
+       snprintf(fname, sizeof fname, "%s" SLASH "hosts" SLASH "%s", confbase, myself->name);
        read_config_file(config_tree, fname);
-       free(fname);
 
        /* Parse some options that are allowed to be changed while tinc is running */
 
@@ -348,11 +304,19 @@ int reload_configuration(void) {
 
        if(strictsubnets) {
                for splay_each(subnet_t, subnet, subnet_tree)
-                       subnet->expires = 1;
+                       if (subnet->owner)
+                               subnet->expires = 1;
+       }
+
+       for splay_each(node_t, n, node_tree)
+               n->status.has_address = false;
 
-               load_all_subnets();
+       load_all_nodes();
 
+       if(strictsubnets) {
                for splay_each(subnet_t, subnet, subnet_tree) {
+                       if (!subnet->owner)
+                               continue;
                        if(subnet->expires == 1) {
                                send_del_subnet(everyone, subnet);
                                if(subnet->owner->status.reachable)
@@ -412,41 +376,44 @@ int reload_configuration(void) {
                if(c->status.control)
                        continue;
 
-               xasprintf(&fname, "%s" SLASH "hosts" SLASH "%s", confbase, c->name);
+               snprintf(fname, sizeof fname, "%s" SLASH "hosts" SLASH "%s", confbase, c->name);
                struct stat s;
                if(stat(fname, &s) || s.st_mtime > last_config_check) {
                        logger(DEBUG_CONNECTIONS, LOG_INFO, "Host config file of %s has been changed", c->name);
-                       terminate_connection(c, c->status.active);
+                       terminate_connection(c, c->edge);
                }
-               free(fname);
        }
 
-       last_config_check = time(NULL);
+       last_config_check = now.tv_sec;
 
        return 0;
 }
 
 void retry(void) {
+       /* Reset the reconnection timers for all outgoing connections */
+       for list_each(outgoing_t, outgoing, outgoing_list) {
+               outgoing->timeout = 0;
+               if(outgoing->ev.cb)
+                       timeout_set(&outgoing->ev, &(struct timeval){0, 0});
+       }
+
+       /* Check for outgoing connections that are in progress, and reset their ping timers */
        for list_each(connection_t, c, connection_list) {
-               if(c->outgoing && !c->node) {
-                       timeout_del(&c->outgoing->ev);
-                       if(c->status.connecting)
-                               close(c->socket);
-                       c->outgoing->timeout = 0;
-                       terminate_connection(c, c->status.active);
-               }
+               if(c->outgoing && !c->node)
+                       c->last_ping_time = 0;
        }
+
+       /* Kick the ping timeout handler */
+       timeout_set(&pingtimer, &(struct timeval){0, 0});
 }
 
 /*
   this is where it all happens...
 */
 int main_loop(void) {
-       timeout_t pingtimer = {{0}};
-       timeout_t periodictimer = {{0}};
-
+       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};
@@ -463,15 +430,16 @@ int main_loop(void) {
 #endif
 
        if(!event_loop()) {
-               logger(DEBUG_ALWAYS, LOG_ERR, "Error while waiting for input: %s", strerror(errno));
+               logger(DEBUG_ALWAYS, LOG_ERR, "Error while waiting for input: %s", sockstrerror(sockerrno));
                return 1;
        }
 
 #ifndef HAVE_MINGW
        signal_del(&sighup);
-       signal_del(&sigalrm);
-       signal_del(&sigquit);
        signal_del(&sigterm);
+       signal_del(&sigquit);
+       signal_del(&sigint);
+       signal_del(&sigalrm);
 #endif
 
        timeout_del(&periodictimer);