reload /etc/resolv.conf in SIGALRM handler
[tinc] / src / net.c
1 /*
2     net.c -- most of the network code
3     Copyright (C) 1998-2005 Ivo Timmermans,
4                   2000-2013 Guus Sliepen <guus@tinc-vpn.org>
5                   2006      Scott Lamb <slamb@slamb.org>
6                   2011      Loïc Grenié <loic.grenie@gmail.com>
7
8     This program is free software; you can redistribute it and/or modify
9     it under the terms of the GNU General Public License as published by
10     the Free Software Foundation; either version 2 of the License, or
11     (at your option) any later version.
12
13     This program is distributed in the hope that it will be useful,
14     but WITHOUT ANY WARRANTY; without even the implied warranty of
15     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16     GNU General Public License for more details.
17
18     You should have received a copy of the GNU General Public License along
19     with this program; if not, write to the Free Software Foundation, Inc.,
20     51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
21 */
22
23 #include "system.h"
24
25 #include "utils.h"
26 #include "conf.h"
27 #include "connection.h"
28 #include "device.h"
29 #include "graph.h"
30 #include "logger.h"
31 #include "meta.h"
32 #include "names.h"
33 #include "net.h"
34 #include "netutl.h"
35 #include "protocol.h"
36 #include "subnet.h"
37 #include "xalloc.h"
38
39 #ifdef HAVE_RESOLV_H
40 #include <resolv.h>
41 #endif
42
43 int contradicting_add_edge = 0;
44 int contradicting_del_edge = 0;
45 static int sleeptime = 10;
46 time_t last_config_check = 0;
47 static timeout_t pingtimer;
48 static timeout_t periodictimer;
49
50 /* Purge edges and subnets of unreachable nodes. Use carefully. */
51
52 void purge(void) {
53         logger(DEBUG_PROTOCOL, LOG_DEBUG, "Purging unreachable nodes");
54
55         /* Remove all edges and subnets owned by unreachable nodes. */
56
57         for splay_each(node_t, n, node_tree) {
58                 if(!n->status.reachable) {
59                         logger(DEBUG_SCARY_THINGS, LOG_DEBUG, "Purging node %s (%s)", n->name, n->hostname);
60
61                         for splay_each(subnet_t, s, n->subnet_tree) {
62                                 send_del_subnet(everyone, s);
63                                 if(!strictsubnets)
64                                         subnet_del(n, s);
65                         }
66
67                         for splay_each(edge_t, e, n->edge_tree) {
68                                 if(!tunnelserver)
69                                         send_del_edge(everyone, e);
70                                 edge_del(e);
71                         }
72                 }
73         }
74
75         /* Check if anyone else claims to have an edge to an unreachable node. If not, delete node. */
76
77         for splay_each(node_t, n, node_tree) {
78                 if(!n->status.reachable) {
79                         for splay_each(edge_t, e, edge_weight_tree)
80                                 if(e->to == n)
81                                         return;
82
83                         if(!autoconnect && (!strictsubnets || !n->subnet_tree->head))
84                                 /* in strictsubnets mode do not delete nodes with subnets */
85                                 node_del(n);
86                 }
87         }
88 }
89
90 /*
91   Terminate a connection:
92   - Mark it as inactive
93   - Remove the edge representing this connection
94   - Kill it with fire
95   - Check if we need to retry making an outgoing connection
96 */
97 void terminate_connection(connection_t *c, bool report) {
98         logger(DEBUG_CONNECTIONS, LOG_NOTICE, "Closing connection with %s (%s)", c->name, c->hostname);
99
100         c->status.active = false;
101
102         if(c->node && c->node->connection == c)
103                 c->node->connection = NULL;
104
105         if(c->edge) {
106                 if(report && !tunnelserver)
107                         send_del_edge(everyone, c->edge);
108
109                 edge_del(c->edge);
110                 c->edge = NULL;
111
112                 /* Run MST and SSSP algorithms */
113
114                 graph();
115
116                 /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
117
118                 if(report && !c->node->status.reachable) {
119                         edge_t *e;
120                         e = lookup_edge(c->node, myself);
121                         if(e) {
122                                 if(!tunnelserver)
123                                         send_del_edge(everyone, e);
124                                 edge_del(e);
125                         }
126                 }
127         }
128
129         outgoing_t *outgoing = c->outgoing;
130         connection_del(c);
131
132         /* Check if this was our outgoing connection */
133
134         if(outgoing)
135                 do_outgoing_connection(outgoing);
136
137 #ifndef HAVE_MINGW
138         /* Clean up dead proxy processes */
139
140         while(waitpid(-1, NULL, WNOHANG) > 0);
141 #endif
142 }
143
144 /*
145   Check if the other end is active.
146   If we have sent packets, but didn't receive any,
147   then possibly the other end is dead. We send a
148   PING request over the meta connection. If the other
149   end does not reply in time, we consider them dead
150   and close the connection.
151 */
152 static void timeout_handler(void *data) {
153         for list_each(connection_t, c, connection_list) {
154                 if(c->status.control)
155                         continue;
156
157                 if(c->last_ping_time + pingtimeout <= now.tv_sec) {
158                         if(c->status.active) {
159                                 if(c->status.pinged) {
160                                         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);
161                                 } else if(c->last_ping_time + pinginterval <= now.tv_sec) {
162                                         send_ping(c);
163                                         continue;
164                                 } else {
165                                         continue;
166                                 }
167                         } else {
168                                 if(c->status.connecting)
169                                         logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
170                                 else
171                                         logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
172                         }
173                         terminate_connection(c, c->status.active);
174                 }
175         }
176
177         timeout_set(data, &(struct timeval){pingtimeout, rand() % 100000});
178 }
179
180 static void periodic_handler(void *data) {
181         /* Check if there are too many contradicting ADD_EDGE and DEL_EDGE messages.
182            This usually only happens when another node has the same Name as this node.
183            If so, sleep for a short while to prevent a storm of contradicting messages.
184         */
185
186         if(contradicting_del_edge > 100 && contradicting_add_edge > 100) {
187                 logger(DEBUG_ALWAYS, LOG_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", sleeptime);
188                 usleep(sleeptime * 1000000LL);
189                 sleeptime *= 2;
190                 if(sleeptime < 0)
191                         sleeptime = 3600;
192         } else {
193                 sleeptime /= 2;
194                 if(sleeptime < 10)
195                         sleeptime = 10;
196         }
197
198         contradicting_add_edge = 0;
199         contradicting_del_edge = 0;
200
201         /* If AutoConnect is set, check if we need to make or break connections. */
202
203         if(autoconnect && node_tree->count > 1) {
204                 /* Count number of active connections */
205                 int nc = 0;
206                 for list_each(connection_t, c, connection_list) {
207                         if(c->status.active && !c->status.control)
208                                 nc++;
209                 }
210
211                 if(nc < 3) {
212                         /* Not enough active connections, try to add one.
213                            Choose a random node, if we don't have a connection to it,
214                            and we are not already trying to make one, create an
215                            outgoing connection to this node.
216                         */
217                         int r = rand() % node_tree->count;
218                         int i = 0;
219
220                         for splay_each(node_t, n, node_tree) {
221                                 if(i++ != r)
222                                         continue;
223
224                                 if(n->connection)
225                                         break;
226
227                                 bool found = false;
228
229                                 for list_each(outgoing_t, outgoing, outgoing_list) {
230                                         if(!strcmp(outgoing->name, n->name)) {
231                                                 found = true;
232                                                 break;
233                                         }
234                                 }
235
236                                 if(!found) {
237                                         logger(DEBUG_CONNECTIONS, LOG_INFO, "Autoconnecting to %s", n->name);
238                                         outgoing_t *outgoing = xzalloc(sizeof *outgoing);
239                                         outgoing->name = xstrdup(n->name);
240                                         list_insert_tail(outgoing_list, outgoing);
241                                         setup_outgoing_connection(outgoing);
242                                 }
243                                 break;
244                         }
245                 } else if(nc > 3) {
246                         /* Too many active connections, try to remove one.
247                            Choose a random outgoing connection to a node
248                            that has at least one other connection.
249                         */
250                         int r = rand() % nc;
251                         int i = 0;
252
253                         for list_each(connection_t, c, connection_list) {
254                                 if(!c->status.active || c->status.control)
255                                         continue;
256
257                                 if(i++ != r)
258                                         continue;
259
260                                 if(!c->outgoing || !c->node || c->node->edge_tree->count < 2)
261                                         break;
262
263                                 logger(DEBUG_CONNECTIONS, LOG_INFO, "Autodisconnecting from %s", c->name);
264                                 list_delete(outgoing_list, c->outgoing);
265                                 c->outgoing = NULL;
266                                 terminate_connection(c, c->status.active);
267                                 break;
268                         }
269                 }
270
271                 if(nc >= 3) {
272                         /* If we have enough active connections,
273                            remove any pending outgoing connections.
274                         */
275                         for list_each(outgoing_t, o, outgoing_list) {
276                                 bool found = false;
277                                 for list_each(connection_t, c, connection_list) {
278                                         if(c->outgoing == o) {
279                                                 found = true;
280                                                 break;
281                                         }
282                                 }
283                                 if(!found) {
284                                         logger(DEBUG_CONNECTIONS, LOG_INFO, "Cancelled outgoing connection to %s", o->name);
285                                         list_delete_node(outgoing_list, node);
286                                 }
287                         }
288                 }
289         }
290
291         timeout_set(data, &(struct timeval){5, rand() % 100000});
292 }
293
294 void handle_meta_connection_data(connection_t *c) {
295         if (!receive_meta(c)) {
296                 terminate_connection(c, c->status.active);
297                 return;
298         }
299 }
300
301 #ifndef HAVE_MINGW
302 static void sigterm_handler(void *data) {
303         logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum));
304         event_exit();
305 }
306
307 static void sighup_handler(void *data) {
308         logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum));
309         reopenlogger();
310         if(reload_configuration())
311                 exit(1);
312 }
313
314 static void sigalrm_handler(void *data) {
315         logger(DEBUG_ALWAYS, LOG_NOTICE, "Got %s signal", strsignal(((signal_t *)data)->signum));
316 #ifdef HAVE_DECL_RES_INIT
317         res_init();
318 #endif
319         retry();
320 }
321 #endif
322
323 int reload_configuration(void) {
324         char *fname = NULL;
325
326         /* Reread our own configuration file */
327
328         exit_configuration(&config_tree);
329         init_configuration(&config_tree);
330
331         if(!read_server_config()) {
332                 logger(DEBUG_ALWAYS, LOG_ERR, "Unable to reread configuration file.");
333                 return EINVAL;
334         }
335
336         read_config_options(config_tree, NULL);
337
338         xasprintf(&fname, "%s" SLASH "hosts" SLASH "%s", confbase, myself->name);
339         read_config_file(config_tree, fname);
340         free(fname);
341
342         /* Parse some options that are allowed to be changed while tinc is running */
343
344         setup_myself_reloadable();
345
346         /* If StrictSubnet is set, expire deleted Subnets and read new ones in */
347
348         if(strictsubnets) {
349                 for splay_each(subnet_t, subnet, subnet_tree)
350                         subnet->expires = 1;
351
352                 load_all_subnets();
353
354                 for splay_each(subnet_t, subnet, subnet_tree) {
355                         if(subnet->expires == 1) {
356                                 send_del_subnet(everyone, subnet);
357                                 if(subnet->owner->status.reachable)
358                                         subnet_update(subnet->owner, subnet, false);
359                                 subnet_del(subnet->owner, subnet);
360                         } else if(subnet->expires == -1) {
361                                 subnet->expires = 0;
362                         } else {
363                                 send_add_subnet(everyone, subnet);
364                                 if(subnet->owner->status.reachable)
365                                         subnet_update(subnet->owner, subnet, true);
366                         }
367                 }
368         } else { /* Only read our own subnets back in */
369                 for splay_each(subnet_t, subnet, myself->subnet_tree)
370                         if(!subnet->expires)
371                                 subnet->expires = 1;
372
373                 config_t *cfg = lookup_config(config_tree, "Subnet");
374
375                 while(cfg) {
376                         subnet_t *subnet, *s2;
377
378                         if(!get_config_subnet(cfg, &subnet))
379                                 continue;
380
381                         if((s2 = lookup_subnet(myself, subnet))) {
382                                 if(s2->expires == 1)
383                                         s2->expires = 0;
384
385                                 free_subnet(subnet);
386                         } else {
387                                 subnet_add(myself, subnet);
388                                 send_add_subnet(everyone, subnet);
389                                 subnet_update(myself, subnet, true);
390                         }
391
392                         cfg = lookup_config_next(config_tree, cfg);
393                 }
394
395                 for splay_each(subnet_t, subnet, myself->subnet_tree) {
396                         if(subnet->expires == 1) {
397                                 send_del_subnet(everyone, subnet);
398                                 subnet_update(myself, subnet, false);
399                                 subnet_del(myself, subnet);
400                         }
401                 }
402         }
403
404         /* Try to make outgoing connections */
405
406         try_outgoing_connections();
407
408         /* Close connections to hosts that have a changed or deleted host config file */
409
410         for list_each(connection_t, c, connection_list) {
411                 if(c->status.control)
412                         continue;
413
414                 xasprintf(&fname, "%s" SLASH "hosts" SLASH "%s", confbase, c->name);
415                 struct stat s;
416                 if(stat(fname, &s) || s.st_mtime > last_config_check) {
417                         logger(DEBUG_CONNECTIONS, LOG_INFO, "Host config file of %s has been changed", c->name);
418                         terminate_connection(c, c->status.active);
419                 }
420                 free(fname);
421         }
422
423         last_config_check = now.tv_sec;
424
425         return 0;
426 }
427
428 void retry(void) {
429         /* Reset the reconnection timers for all outgoing connections */
430         for list_each(outgoing_t, outgoing, outgoing_list) {
431                 outgoing->timeout = 0;
432                 if(outgoing->ev.cb)
433                         timeout_set(&outgoing->ev, &(struct timeval){0, 0});
434         }
435
436         /* Check for outgoing connections that are in progress, and reset their ping timers */
437         for list_each(connection_t, c, connection_list) {
438                 if(c->outgoing && !c->node)
439                         c->last_ping_time = 0;
440         }
441
442         /* Kick the ping timeout handler */
443         timeout_set(&pingtimer, &(struct timeval){0, 0});
444 }
445
446 /*
447   this is where it all happens...
448 */
449 int main_loop(void) {
450         timeout_add(&pingtimer, timeout_handler, &pingtimer, &(struct timeval){pingtimeout, rand() % 100000});
451         timeout_add(&periodictimer, periodic_handler, &periodictimer, &(struct timeval){pingtimeout, rand() % 100000});
452
453 #ifndef HAVE_MINGW
454         signal_t sighup = {0};
455         signal_t sigterm = {0};
456         signal_t sigquit = {0};
457         signal_t sigint = {0};
458         signal_t sigalrm = {0};
459
460         signal_add(&sighup, sighup_handler, &sighup, SIGHUP);
461         signal_add(&sigterm, sigterm_handler, &sigterm, SIGTERM);
462         signal_add(&sigquit, sigterm_handler, &sigquit, SIGQUIT);
463         signal_add(&sigint, sigterm_handler, &sigint, SIGINT);
464         signal_add(&sigalrm, sigalrm_handler, &sigalrm, SIGALRM);
465 #endif
466
467         if(!event_loop()) {
468                 logger(DEBUG_ALWAYS, LOG_ERR, "Error while waiting for input: %s", strerror(errno));
469                 return 1;
470         }
471
472 #ifndef HAVE_MINGW
473         signal_del(&sighup);
474         signal_del(&sigterm);
475         signal_del(&sigquit);
476         signal_del(&sigint);
477         signal_del(&sigalrm);
478 #endif
479
480         timeout_del(&periodictimer);
481         timeout_del(&pingtimer);
482
483         return 0;
484 }