X-Git-Url: http://git.meshlink.io/?a=blobdiff_plain;f=src%2Fnet.c;h=d8f7d642e4ac0307d1ef6f84c9156cf70d46f75d;hb=f2b21858a8e77663a02c2d586b21c5568fb435a0;hp=cae8629881e79ec3d3ebd620b7feebf1e84ca57a;hpb=1564f52f7643c554ec49a3e014bc9b9a8745c856;p=meshlink diff --git a/src/net.c b/src/net.c index cae86298..d8f7d642 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, - 2000-2013 Guus Sliepen - 2006 Scott Lamb - 2011 Loïc Grenié + Copyright (C) 2014 Guus Sliepen 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 @@ -27,53 +24,13 @@ #include "connection.h" #include "graph.h" #include "logger.h" +#include "meshlink_internal.h" #include "meta.h" #include "net.h" #include "netutl.h" #include "protocol.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; - -//TODO: move this to a better place -char *confbase; - -/* Purge edges of unreachable nodes. Use carefully. */ - -// TODO: remove -void purge(void) { - logger(DEBUG_PROTOCOL, LOG_DEBUG, "Purging unreachable nodes"); - - /* Remove all edges owned by unreachable nodes. */ - - for splay_each(node_t, n, node_tree) { - if(!n->status.reachable) { - logger(DEBUG_SCARY_THINGS, LOG_DEBUG, "Purging node %s (%s)", n->name, n->hostname); - - for splay_each(edge_t, e, n->edge_tree) { - if(!tunnelserver) - send_del_edge(everyone, e); - edge_del(e); - } - } - } - - /* Check if anyone else claims to have an edge to an unreachable node. If not, delete node. */ - - for splay_each(node_t, n, node_tree) { - if(!n->status.reachable) { - for splay_each(edge_t, e, edge_weight_tree) - if(e->to == n) - return; - } - } -} - /* Terminate a connection: - Mark it as inactive @@ -81,7 +38,7 @@ void purge(void) { - Kill it with fire - Check if we need to retry making an outgoing connection */ -void terminate_connection(connection_t *c, bool report) { +void terminate_connection(meshlink_handle_t *mesh, connection_t *c, bool report) { logger(DEBUG_CONNECTIONS, LOG_NOTICE, "Closing connection with %s (%s)", c->name, c->hostname); c->status.active = false; @@ -90,36 +47,35 @@ void terminate_connection(connection_t *c, bool report) { c->node->connection = NULL; if(c->edge) { - if(report && !tunnelserver) - send_del_edge(everyone, 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(everyone, e); - edge_del(e); + send_del_edge(mesh, mesh->everyone, e); + edge_del(mesh, e); } } } outgoing_t *outgoing = c->outgoing; - connection_del(c); + connection_del(mesh, c); /* Check if this was our outgoing connection */ if(outgoing) - do_outgoing_connection(outgoing); + do_outgoing_connection(mesh, outgoing); #ifndef HAVE_MINGW /* Clean up dead proxy processes */ @@ -136,14 +92,16 @@ 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(void *data) { - for list_each(connection_t, c, connection_list) { - if(c->last_ping_time + pingtimeout <= now.tv_sec) { +static void timeout_handler(event_loop_t *loop, void *data) { + meshlink_handle_t *mesh = loop->data; + + 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) { - 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); + logger(DEBUG_CONNECTIONS, LOG_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 { continue; @@ -154,79 +112,137 @@ static void timeout_handler(void *data) { else logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname); } - terminate_connection(c, c->status.active); + terminate_connection(mesh, c, c->status.active); + } + } + + timeout_set(&mesh->loop, data, &(struct timeval){mesh->pingtimeout, rand() % 100000}); +} + +/// 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(!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(DEBUG_CONNECTIONS, LOG_INFO, "Autoconnecting to %s", n->name); + outgoing_t *outgoing = xzalloc(sizeof *outgoing); + outgoing->name = xstrdup(n->name); + list_insert_tail(mesh->outgoings, outgoing); + setup_outgoing_connection(mesh, outgoing); } + break; } +} + +static bool found_random_node(int *i, int *r, node_t *n) { + if((*i)++ != *r) + return false; - timeout_set(data, &(struct timeval){pingtimeout, rand() % 100000}); + if(n->connection) + return false; + + return true; } -static void periodic_handler(void *data) { +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; + + return true; +} + +static void periodic_handler(event_loop_t *loop, void *data) { + meshlink_handle_t *mesh = loop->data; + /* 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(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); - sleeptime *= 2; - if(sleeptime < 0) - sleeptime = 3600; + if(mesh->contradicting_del_edge > 100 && mesh->contradicting_add_edge > 100) { + logger(DEBUG_ALWAYS, LOG_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 { - sleeptime /= 2; - if(sleeptime < 10) - sleeptime = 10; + mesh->sleeptime /= 2; + if(mesh->sleeptime < 10) + mesh->sleeptime = 10; } - contradicting_add_edge = 0; - contradicting_del_edge = 0; + mesh->contradicting_add_edge = 0; + mesh->contradicting_del_edge = 0; /* If AutoConnect is set, check if we need to make or break connections. */ - if(autoconnect && node_tree->count > 1) { + if(autoconnect && mesh->nodes->count > 1) { /* Count number of active connections */ int nc = 0; - for list_each(connection_t, c, connection_list) { + for list_each(connection_t, c, mesh->connections) { if(c->status.active) nc++; } + /* Count number of unreachable nodes */ + int num_unreachable = 0; + for splay_each(node_t, n, mesh->nodes) { + if(!n->status.reachable) + num_unreachable++; + } + 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 = xzalloc(sizeof *outgoing); - outgoing->name = xstrdup(n->name); - list_insert_tail(outgoing_list, outgoing); - setup_outgoing_connection(outgoing); - } - break; - } - } else if(nc > autoconnect) { + 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. @@ -234,7 +250,7 @@ static void periodic_handler(void *data) { int r = rand() % nc; int i = 0; - for list_each(connection_t, c, connection_list) { + for list_each(connection_t, c, mesh->connections) { if(!c->status.active) continue; @@ -245,9 +261,9 @@ static void periodic_handler(void *data) { break; logger(DEBUG_CONNECTIONS, LOG_INFO, "Autodisconnecting from %s", c->name); - list_delete(outgoing_list, c->outgoing); + list_delete(mesh->outgoings, c->outgoing); c->outgoing = NULL; - terminate_connection(c, c->status.active); + terminate_connection(mesh, c, c->status.active); break; } } @@ -255,10 +271,18 @@ static void periodic_handler(void *data) { if(nc >= autoconnect) { /* If we have enough active connections, remove any pending outgoing connections. + Do not remove pending connections to unreachable + nodes. */ - for list_each(outgoing_t, o, outgoing_list) { + 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, connection_list) { + for list_each(connection_t, c, mesh->connections) { if(c->outgoing == o) { found = true; break; @@ -266,138 +290,62 @@ static void periodic_handler(void *data) { } if(!found) { logger(DEBUG_CONNECTIONS, LOG_INFO, "Cancelled outgoing connection to %s", o->name); - list_delete_node(outgoing_list, node); + /* 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); } } } } - timeout_set(data, &(struct timeval){5, rand() % 100000}); + timeout_set(&mesh->loop, data, &(struct timeval){5, rand() % 100000}); } -void handle_meta_connection_data(connection_t *c) { - if (!receive_meta(c)) { - terminate_connection(c, c->status.active); +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; } } -#ifndef HAVE_MINGW -static void sigterm_handler(void *data) { - logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum)); - event_exit(); -} - -static void sighup_handler(void *data) { - logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum)); - 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 = NULL; - - /* Reread our own configuration file */ - - exit_configuration(&config_tree); - init_configuration(&config_tree); - - if(!read_server_config()) { - 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); - read_config_file(config_tree, fname); - free(fname); - - /* Parse some options that are allowed to be changed while tinc is running */ - - setup_myself_reloadable(); - - /* Try to make outgoing connections */ - - try_outgoing_connections(); - - /* Close connections to hosts that have a changed or deleted host config file */ - - for list_each(connection_t, c, connection_list) { - xasprintf(&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); - } - free(fname); - } - - last_config_check = now.tv_sec; - - return 0; -} - -void retry(void) { +void retry(meshlink_handle_t *mesh) { /* Reset the reconnection timers for all outgoing connections */ - for list_each(outgoing_t, outgoing, outgoing_list) { + for list_each(outgoing_t, outgoing, mesh->outgoings) { outgoing->timeout = 0; if(outgoing->ev.cb) - timeout_set(&outgoing->ev, &(struct timeval){0, 0}); + timeout_set(&mesh->loop, &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) { + 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(&pingtimer, &(struct timeval){0, 0}); + timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval){0, 0}); } /* this is where it all happens... */ -int main_loop(void) { - timeout_add(&pingtimer, timeout_handler, &pingtimer, &(struct timeval){pingtimeout, rand() % 100000}); - timeout_add(&periodictimer, periodic_handler, &periodictimer, &(struct timeval){pingtimeout, rand() % 100000}); +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){mesh->pingtimeout, rand() % 100000}); -#ifndef HAVE_MINGW - signal_t sighup = {0}; - signal_t sigterm = {0}; - signal_t sigquit = {0}; - signal_t sigint = {0}; - signal_t sigalrm = {0}; - - signal_add(&sighup, sighup_handler, &sighup, SIGHUP); - signal_add(&sigterm, sigterm_handler, &sigterm, SIGTERM); - signal_add(&sigquit, sigterm_handler, &sigquit, SIGQUIT); - signal_add(&sigint, sigterm_handler, &sigint, SIGINT); - signal_add(&sigalrm, sigalrm_handler, &sigalrm, SIGALRM); -#endif + //Add signal handler + mesh->datafromapp.signum = 0; + signal_add(&(mesh->loop),&(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue,mesh, mesh->datafromapp.signum); - if(!event_loop()) { + if(!event_loop_run(&mesh->loop)) { logger(DEBUG_ALWAYS, LOG_ERR, "Error while waiting for input: %s", strerror(errno)); return 1; } -#ifndef HAVE_MINGW - signal_del(&sighup); - signal_del(&sigterm); - signal_del(&sigquit); - signal_del(&sigint); - signal_del(&sigalrm); -#endif - - timeout_del(&periodictimer); - timeout_del(&pingtimer); + timeout_del(&mesh->loop, &mesh->periodictimer); + timeout_del(&mesh->loop, &mesh->pingtimer); return 0; }