2 net.c -- most of the network code
3 Copyright (C) 2014-2017 Guus Sliepen <guus@meshlink.io>
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License along
16 with this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 #include "connection.h"
27 #include "meshlink_internal.h"
35 static inline int min(int a, int b) {
41 Terminate a connection:
43 - Remove the edge representing this connection
45 - Check if we need to retry making an outgoing connection
47 void terminate_connection(meshlink_handle_t *mesh, connection_t *c, bool report) {
48 logger(mesh, MESHLINK_INFO, "Closing connection with %s", c->name);
50 c->status.active = false;
52 if(c->node && c->node->connection == c) {
53 c->node->connection = NULL;
58 send_del_edge(mesh, mesh->everyone, c->edge, 0);
61 edge_del(mesh, c->edge);
64 /* Run MST and SSSP algorithms */
68 /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
70 if(report && c->node && !c->node->status.reachable) {
72 e = lookup_edge(c->node, mesh->self);
75 send_del_edge(mesh, mesh->everyone, e, 0);
81 outgoing_t *outgoing = c->outgoing;
82 connection_del(mesh, c);
84 /* Check if this was our outgoing connection */
87 do_outgoing_connection(mesh, outgoing);
91 /* Clean up dead proxy processes */
93 while(waitpid(-1, NULL, WNOHANG) > 0);
99 Check if the other end is active.
100 If we have sent packets, but didn't receive any,
101 then possibly the other end is dead. We send a
102 PING request over the meta connection. If the other
103 end does not reply in time, we consider them dead
104 and close the connection.
106 static void timeout_handler(event_loop_t *loop, void *data) {
107 meshlink_handle_t *mesh = loop->data;
108 logger(mesh, MESHLINK_DEBUG, "timeout_handler()");
110 for list_each(connection_t, c, mesh->connections) {
111 // Also make sure that if outstanding key requests for the UDP counterpart of a connection has timed out, we restart it.
113 if(c->node->status.waitingforkey && c->node->last_req_key + mesh->pingtimeout <= mesh->loop.now.tv_sec) {
114 send_req_key(mesh, c->node);
118 if(c->last_ping_time + mesh->pingtimeout <= mesh->loop.now.tv_sec) {
119 if(c->status.active) {
120 if(c->status.pinged) {
121 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);
122 } else if(c->last_ping_time + mesh->pinginterval <= mesh->loop.now.tv_sec) {
129 if(c->status.connecting) {
130 logger(mesh, MESHLINK_WARNING, "Timeout while connecting to %s", c->name);
132 logger(mesh, MESHLINK_WARNING, "Timeout from %s during authentication", c->name);
136 terminate_connection(mesh, c, c->status.active);
140 timeout_set(&mesh->loop, data, &(struct timeval) {
141 mesh->pingtimeout, rand() % 100000
145 // devclass asc, last_successfull_connection desc
146 static int node_compare_devclass_asc_lsc_desc(const void *a, const void *b) {
147 const node_t *na = a, *nb = b;
149 if(na->devclass < nb->devclass) {
153 if(na->devclass > nb->devclass) {
157 if(na->last_successfull_connection == nb->last_successfull_connection) {
161 if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection) {
165 if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection) {
180 // last_successfull_connection desc
181 static int node_compare_lsc_desc(const void *a, const void *b) {
182 const node_t *na = a, *nb = b;
184 if(na->last_successfull_connection == nb->last_successfull_connection) {
188 if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection) {
192 if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection) {
208 static int node_compare_devclass_desc(const void *a, const void *b) {
209 const node_t *na = a, *nb = b;
211 if(na->devclass < nb->devclass) {
215 if(na->devclass > nb->devclass) {
237 // find the best one for initial connect
242 where dclass <= my.dclass and !connection and (timestamp - last_retry) > retry_timeout
243 order by dclass asc, last_connection desc
249 // 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
253 for i = BACKBONE to my.dclass
254 j += count(from connections where node.dclass = i)
258 where dclass = i and !connection and (timestamp - last_retry) > retry_timeout
259 order by last_connection desc
271 where dclass <= my.dclass and !reachable and (timestamp - last_retry) > retry_timeout
272 order by dclass asc, last_connection desc
284 // 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]
288 for i = BACKBONE to my.dclass
289 j += count(from connections where node.dclass = i)
293 where dclass >= i and outgoing_connection
301 // disconnect connections in case we have more than enough connections
306 where outgoing_connection
318 next (timeout, autoconnect)
325 static void periodic_handler(event_loop_t *loop, void *data) {
326 meshlink_handle_t *mesh = loop->data;
328 /* Check if there are too many contradicting ADD_EDGE and DEL_EDGE messages.
329 This usually only happens when another node has the same Name as this node.
330 If so, sleep for a short while to prevent a storm of contradicting messages.
333 if(mesh->contradicting_del_edge > 100 && mesh->contradicting_add_edge > 100) {
334 logger(mesh, MESHLINK_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", mesh->sleeptime);
335 usleep(mesh->sleeptime * 1000000LL);
336 mesh->sleeptime *= 2;
338 if(mesh->sleeptime < 0) {
339 mesh->sleeptime = 3600;
342 mesh->sleeptime /= 2;
344 if(mesh->sleeptime < 10) {
345 mesh->sleeptime = 10;
349 mesh->contradicting_add_edge = 0;
350 mesh->contradicting_del_edge = 0;
354 /* Check if we need to make or break connections. */
356 if(mesh->nodes->count > 1) {
358 logger(mesh, MESHLINK_DEBUG, "--- autoconnect begin ---");
360 int retry_timeout = min(mesh->nodes->count * 5, 60);
362 logger(mesh, MESHLINK_DEBUG, "* devclass = %d", mesh->devclass);
363 logger(mesh, MESHLINK_DEBUG, "* nodes = %d", mesh->nodes->count);
364 logger(mesh, MESHLINK_DEBUG, "* retry_timeout = %d", retry_timeout);
367 // connect disconnect nodes
369 node_t *connect_to = NULL;
370 node_t *disconnect_from = NULL;
375 unsigned int cur_connects = 0;
377 for list_each(connection_t, c, mesh->connections) {
378 if(c->status.active) {
383 logger(mesh, MESHLINK_DEBUG, "* cur_connects = %d", cur_connects);
384 logger(mesh, MESHLINK_DEBUG, "* outgoings = %d", mesh->outgoings->count);
386 // get min_connects and max_connects
388 assert(mesh->devclass >= 0 && mesh->devclass <= _DEV_CLASS_MAX);
390 unsigned int min_connects = dev_class_traits[mesh->devclass].min_connects;
391 unsigned int max_connects = dev_class_traits[mesh->devclass].max_connects;
393 logger(mesh, MESHLINK_DEBUG, "* min_connects = %d", min_connects);
394 logger(mesh, MESHLINK_DEBUG, "* max_connects = %d", max_connects);
397 // find the best one for initial connect
399 if(cur_connects < min_connects) {
400 splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
402 for splay_each(node_t, n, mesh->nodes) {
403 logger(mesh, MESHLINK_DEBUG, "* n->devclass = %d", n->devclass);
405 if(n != mesh->self && n->devclass <= mesh->devclass && !n->connection && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout)) {
406 splay_insert(nodes, n);
411 logger(mesh, MESHLINK_DEBUG, "* found best one for initial connect");
414 connect_to = (node_t *)nodes->head->data;
416 logger(mesh, MESHLINK_DEBUG, "* could not find node for initial connect");
419 splay_free_tree(nodes);
423 // find better nodes to connect to
425 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects) {
426 unsigned int connects = 0;
428 for(unsigned int devclass = 0; devclass <= mesh->devclass; ++devclass) {
429 for list_each(connection_t, c, mesh->connections) {
430 if(c->status.active && c->node && c->node->devclass == devclass) {
435 if(connects < min_connects) {
436 splay_tree_t *nodes = splay_alloc_tree(node_compare_lsc_desc, NULL);
438 for splay_each(node_t, n, mesh->nodes) {
439 if(n != mesh->self && n->devclass == devclass && !n->connection && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout)) {
440 splay_insert(nodes, n);
445 logger(mesh, MESHLINK_DEBUG, "* found better node");
446 connect_to = (node_t *)nodes->head->data;
448 splay_free_tree(nodes);
452 splay_free_tree(nodes);
459 logger(mesh, MESHLINK_DEBUG, "* could not find better nodes");
466 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects) {
467 splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
469 for splay_each(node_t, n, mesh->nodes) {
470 if(n != mesh->self && n->devclass <= mesh->devclass && !n->status.reachable && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout)) {
471 splay_insert(nodes, n);
476 logger(mesh, MESHLINK_DEBUG, "* try to heal partition");
477 connect_to = (node_t *)nodes->head->data;
479 logger(mesh, MESHLINK_DEBUG, "* could not find nodes for partition healing");
482 splay_free_tree(nodes);
488 if(connect_to && !connect_to->connection) {
489 connect_to->last_connect_try = time(NULL);
491 /* check if there is already a connection attempt to this node */
494 for list_each(outgoing_t, outgoing, mesh->outgoings) {
495 if(!strcmp(outgoing->name, connect_to->name)) {
502 logger(mesh, MESHLINK_DEBUG, "Autoconnecting to %s", connect_to->name);
503 outgoing_t *outgoing = xzalloc(sizeof(outgoing_t));
504 outgoing->mesh = mesh;
505 outgoing->name = xstrdup(connect_to->name);
506 list_insert_tail(mesh->outgoings, outgoing);
507 setup_outgoing_connection(mesh, outgoing);
509 logger(mesh, MESHLINK_DEBUG, "* skip autoconnect since it is an outgoing connection already");
514 // disconnect suboptimal outgoing connections
516 if(min_connects < cur_connects /*&& cur_connects <= max_connects*/) {
517 unsigned int connects = 0;
519 for(unsigned int devclass = 0; devclass <= mesh->devclass; ++devclass) {
520 for list_each(connection_t, c, mesh->connections) {
521 if(c->status.active && c->node && c->node->devclass == devclass) {
526 if(min_connects < connects) {
527 splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
529 for list_each(connection_t, c, mesh->connections) {
530 if(c->outgoing && c->node && c->node->devclass >= devclass) {
531 splay_insert(nodes, c->node);
536 logger(mesh, MESHLINK_DEBUG, "* disconnect suboptimal outgoing connection");
537 disconnect_from = (node_t *)nodes->head->data;
540 splay_free_tree(nodes);
545 if(!disconnect_from) {
546 logger(mesh, MESHLINK_DEBUG, "* no suboptimal outgoing connections");
551 // disconnect connections (too many connections)
553 if(!disconnect_from && max_connects < cur_connects) {
554 splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
556 for list_each(connection_t, c, mesh->connections) {
557 if(c->status.active && c->node) {
558 splay_insert(nodes, c->node);
563 logger(mesh, MESHLINK_DEBUG, "* disconnect connection (too many connections)");
566 disconnect_from = (node_t *)nodes->head->data;
568 logger(mesh, MESHLINK_DEBUG, "* no node we want to disconnect, even though we have too many connections");
571 splay_free_tree(nodes);
575 // perform disconnect
577 if(disconnect_from && disconnect_from->connection) {
578 logger(mesh, MESHLINK_DEBUG, "Autodisconnecting from %s", disconnect_from->connection->name);
579 list_delete(mesh->outgoings, disconnect_from->connection->outgoing);
580 disconnect_from->connection->outgoing = NULL;
581 terminate_connection(mesh, disconnect_from->connection, disconnect_from->connection->status.active);
587 logger(mesh, MESHLINK_DEBUG, "--- autoconnect end ---");
590 timeout_set(&mesh->loop, data, &(struct timeval) {
591 timeout, rand() % 100000
595 void handle_meta_connection_data(meshlink_handle_t *mesh, connection_t *c) {
596 if(!receive_meta(mesh, c)) {
597 terminate_connection(mesh, c, c->status.active);
602 void retry(meshlink_handle_t *mesh) {
603 /* Reset the reconnection timers for all outgoing connections */
604 for list_each(outgoing_t, outgoing, mesh->outgoings) {
605 outgoing->timeout = 0;
608 timeout_set(&mesh->loop, &outgoing->ev, &(struct timeval) {
613 /* Check for outgoing connections that are in progress, and reset their ping timers */
614 for list_each(connection_t, c, mesh->connections) {
615 if(c->outgoing && !c->node) {
616 c->last_ping_time = 0;
620 /* Kick the ping timeout handler */
621 timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval) {
627 this is where it all happens...
629 int main_loop(meshlink_handle_t *mesh) {
630 timeout_add(&mesh->loop, &mesh->pingtimer, timeout_handler, &mesh->pingtimer, &(struct timeval) {
631 mesh->pingtimeout, rand() % 100000
633 timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timeval) {
638 mesh->datafromapp.signum = 0;
639 signal_add(&(mesh->loop), &(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue, mesh, mesh->datafromapp.signum);
641 if(!event_loop_run(&(mesh->loop), &(mesh->mesh_mutex))) {
642 logger(mesh, MESHLINK_ERROR, "Error while waiting for input: %s", strerror(errno));
646 timeout_del(&mesh->loop, &mesh->periodictimer);
647 timeout_del(&mesh->loop, &mesh->pingtimer);