c5e5707b9ae415538e3db79ca8cc21dae0c8f90b
[tinc] / src / net.c
1 /*
2     net.c -- most of the network code
3     Copyright (C) 1998-2005 Ivo Timmermans,
4                   2000-2012 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 <openssl/rand.h>
26
27 #include "utils.h"
28 #include "avl_tree.h"
29 #include "conf.h"
30 #include "connection.h"
31 #include "device.h"
32 #include "event.h"
33 #include "graph.h"
34 #include "logger.h"
35 #include "meta.h"
36 #include "net.h"
37 #include "netutl.h"
38 #include "process.h"
39 #include "protocol.h"
40 #include "route.h"
41 #include "subnet.h"
42 #include "xalloc.h"
43
44 bool do_purge = false;
45 volatile bool running = false;
46 #ifdef HAVE_PSELECT
47 bool graph_dump = false;
48 #endif
49
50 time_t now = 0;
51 int contradicting_add_edge = 0;
52 int contradicting_del_edge = 0;
53 static int sleeptime = 10;
54
55 /* Purge edges and subnets of unreachable nodes. Use carefully. */
56
57 static void purge(void) {
58         avl_node_t *nnode, *nnext, *enode, *enext, *snode, *snext;
59         node_t *n;
60         edge_t *e;
61         subnet_t *s;
62
63         ifdebug(PROTOCOL) logger(LOG_DEBUG, "Purging unreachable nodes");
64
65         /* Remove all edges and subnets owned by unreachable nodes. */
66
67         for(nnode = node_tree->head; nnode; nnode = nnext) {
68                 nnext = nnode->next;
69                 n = nnode->data;
70
71                 if(!n->status.reachable) {
72                         ifdebug(SCARY_THINGS) logger(LOG_DEBUG, "Purging node %s (%s)", n->name,
73                                            n->hostname);
74
75                         for(snode = n->subnet_tree->head; snode; snode = snext) {
76                                 snext = snode->next;
77                                 s = snode->data;
78                                 send_del_subnet(everyone, s);
79                                 if(!strictsubnets)
80                                         subnet_del(n, s);
81                         }
82
83                         for(enode = n->edge_tree->head; enode; enode = enext) {
84                                 enext = enode->next;
85                                 e = enode->data;
86                                 if(!tunnelserver)
87                                         send_del_edge(everyone, e);
88                                 edge_del(e);
89                         }
90                 }
91         }
92
93         /* Check if anyone else claims to have an edge to an unreachable node. If not, delete node. */
94
95         for(nnode = node_tree->head; nnode; nnode = nnext) {
96                 nnext = nnode->next;
97                 n = nnode->data;
98
99                 if(!n->status.reachable) {
100                         for(enode = edge_weight_tree->head; enode; enode = enext) {
101                                 enext = enode->next;
102                                 e = enode->data;
103
104                                 if(e->to == n)
105                                         break;
106                         }
107
108                         if(!enode && (!strictsubnets || !n->subnet_tree->head))
109                                 /* in strictsubnets mode do not delete nodes with subnets */
110                                 node_del(n);
111                 }
112         }
113 }
114
115 /*
116   put all file descriptors in an fd_set array
117   While we're at it, purge stuff that needs to be removed.
118 */
119 static int build_fdset(fd_set *readset, fd_set *writeset) {
120         avl_node_t *node, *next;
121         connection_t *c;
122         int i, max = 0;
123
124         FD_ZERO(readset);
125         FD_ZERO(writeset);
126
127         for(node = connection_tree->head; node; node = next) {
128                 next = node->next;
129                 c = node->data;
130
131                 if(c->status.remove) {
132                         connection_del(c);
133                         if(!connection_tree->head)
134                                 purge();
135                 } else {
136                         FD_SET(c->socket, readset);
137                         if(c->outbuflen > 0 || c->status.connecting)
138                                 FD_SET(c->socket, writeset);
139                         if(c->socket > max)
140                                 max = c->socket;
141                 }
142         }
143
144         for(i = 0; i < listen_sockets; i++) {
145                 FD_SET(listen_socket[i].tcp, readset);
146                 if(listen_socket[i].tcp > max)
147                         max = listen_socket[i].tcp;
148                 FD_SET(listen_socket[i].udp, readset);
149                 if(listen_socket[i].udp > max)
150                         max = listen_socket[i].udp;
151         }
152
153         if(device_fd >= 0)
154                 FD_SET(device_fd, readset);
155         if(device_fd > max)
156                 max = device_fd;
157         
158         return max;
159 }
160
161 /*
162   Terminate a connection:
163   - Close the socket
164   - Remove associated edge and tell other connections about it if report = true
165   - Check if we need to retry making an outgoing connection
166   - Deactivate the host
167 */
168 void terminate_connection(connection_t *c, bool report) {
169         if(c->status.remove)
170                 return;
171
172         ifdebug(CONNECTIONS) logger(LOG_NOTICE, "Closing connection with %s (%s)",
173                            c->name, c->hostname);
174
175         c->status.remove = true;
176         c->status.active = false;
177
178         if(c->node)
179                 c->node->connection = NULL;
180
181         if(c->socket)
182                 closesocket(c->socket);
183
184         if(c->edge) {
185                 if(report && !tunnelserver)
186                         send_del_edge(everyone, c->edge);
187
188                 edge_del(c->edge);
189
190                 /* Run MST and SSSP algorithms */
191
192                 graph();
193
194                 /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
195
196                 if(report && !c->node->status.reachable) {
197                         edge_t *e;
198                         e = lookup_edge(c->node, myself);
199                         if(e) {
200                                 if(!tunnelserver)
201                                         send_del_edge(everyone, e);
202                                 edge_del(e);
203                         }
204                 }
205         }
206
207         free_connection_partially(c);
208
209         /* Check if this was our outgoing connection */
210
211         if(c->outgoing) {
212                 c->status.remove = false;
213                 do_outgoing_connection(c);      
214         }
215
216 #ifndef HAVE_MINGW
217         /* Clean up dead proxy processes */
218
219         while(waitpid(-1, NULL, WNOHANG) > 0);
220 #endif
221 }
222
223 /*
224   Check if the other end is active.
225   If we have sent packets, but didn't receive any,
226   then possibly the other end is dead. We send a
227   PING request over the meta connection. If the other
228   end does not reply in time, we consider them dead
229   and close the connection.
230 */
231 static void check_dead_connections(void) {
232         avl_node_t *node, *next;
233         connection_t *c;
234
235         for(node = connection_tree->head; node; node = next) {
236                 next = node->next;
237                 c = node->data;
238
239                 if(c->last_ping_time + pingtimeout <= now) {
240                         if(c->status.active) {
241                                 if(c->status.pinged) {
242                                         ifdebug(CONNECTIONS) logger(LOG_INFO, "%s (%s) didn't respond to PING in %ld seconds",
243                                                            c->name, c->hostname, (long)now - c->last_ping_time);
244                                         c->status.timeout = true;
245                                         terminate_connection(c, true);
246                                 } else if(c->last_ping_time + pinginterval <= now) {
247                                         send_ping(c);
248                                 }
249                         } else {
250                                 if(c->status.remove) {
251                                         logger(LOG_WARNING, "Old connection_t for %s (%s) status %04x still lingering, deleting...",
252                                                    c->name, c->hostname, bitfield_to_int(&c->status, sizeof c->status));
253                                         connection_del(c);
254                                         continue;
255                                 }
256                                 ifdebug(CONNECTIONS) logger(LOG_WARNING, "Timeout from %s (%s) during authentication",
257                                                    c->name, c->hostname);
258                                 if(c->status.connecting) {
259                                         c->status.connecting = false;
260                                         closesocket(c->socket);
261                                         do_outgoing_connection(c);
262                                 } else {
263                                         terminate_connection(c, false);
264                                 }
265                         }
266                 }
267
268                 if(c->outbuflen > 0 && c->last_flushed_time + pingtimeout <= now) {
269                         if(c->status.active) {
270                                 ifdebug(CONNECTIONS) logger(LOG_INFO,
271                                                 "%s (%s) could not flush for %ld seconds (%d bytes remaining)",
272                                                 c->name, c->hostname, (long)now - c->last_flushed_time, c->outbuflen);
273                                 c->status.timeout = true;
274                                 terminate_connection(c, true);
275                         }
276                 }
277         }
278 }
279
280 /*
281   check all connections to see if anything
282   happened on their sockets
283 */
284 static void check_network_activity(fd_set * readset, fd_set * writeset) {
285         connection_t *c;
286         avl_node_t *node;
287         int result, i;
288         socklen_t len = sizeof(result);
289         vpn_packet_t packet;
290         static int errors = 0;
291
292         /* check input from kernel */
293         if(device_fd >= 0 && FD_ISSET(device_fd, readset)) {
294                 if(devops.read(&packet)) {
295                         if(packet.len) {
296                                 errors = 0;
297                                 packet.priority = 0;
298                                 route(myself, &packet);
299                         }
300                 } else {
301                         usleep(errors * 50000);
302                         errors++;
303                         if(errors > 10) {
304                                 logger(LOG_ERR, "Too many errors from %s, exiting!", device);
305                                 running = false;
306                         }
307                 }
308         }
309
310         /* check meta connections */
311         for(node = connection_tree->head; node; node = node->next) {
312                 c = node->data;
313
314                 if(c->status.remove)
315                         continue;
316
317                 if(FD_ISSET(c->socket, writeset)) {
318                         if(c->status.connecting) {
319                                 c->status.connecting = false;
320                                 getsockopt(c->socket, SOL_SOCKET, SO_ERROR, (void *)&result, &len);
321
322                                 if(!result)
323                                         finish_connecting(c);
324                                 else {
325                                         ifdebug(CONNECTIONS) logger(LOG_DEBUG,
326                                                            "Error while connecting to %s (%s): %s",
327                                                            c->name, c->hostname, sockstrerror(result));
328                                         closesocket(c->socket);
329                                         do_outgoing_connection(c);
330                                         continue;
331                                 }
332                         }
333
334                         if(!flush_meta(c)) {
335                                 terminate_connection(c, c->status.active);
336                                 continue;
337                         }
338                 }
339
340                 if(FD_ISSET(c->socket, readset)) {
341                         if(!receive_meta(c)) {
342                                 terminate_connection(c, c->status.active);
343                                 continue;
344                         }
345                 }
346         }
347
348         for(i = 0; i < listen_sockets; i++) {
349                 if(FD_ISSET(listen_socket[i].udp, readset))
350                         handle_incoming_vpn_data(i);
351
352                 if(FD_ISSET(listen_socket[i].tcp, readset))
353                         handle_new_meta_connection(listen_socket[i].tcp);
354         }
355 }
356
357 /*
358   this is where it all happens...
359 */
360 int main_loop(void) {
361         fd_set readset, writeset;
362 #ifdef HAVE_PSELECT
363         struct timespec tv;
364         sigset_t omask, block_mask;
365         time_t next_event;
366 #else
367         struct timeval tv;
368 #endif
369         int r, maxfd;
370         time_t last_ping_check, last_config_check, last_graph_dump;
371         event_t *event;
372
373         last_ping_check = now;
374         last_config_check = now;
375         last_graph_dump = now;
376         
377         srand(now);
378
379 #ifdef HAVE_PSELECT
380         if(lookup_config(config_tree, "GraphDumpFile"))
381                 graph_dump = true;
382         /* Block SIGHUP & SIGALRM */
383         sigemptyset(&block_mask);
384         sigaddset(&block_mask, SIGHUP);
385         sigaddset(&block_mask, SIGALRM);
386         sigprocmask(SIG_BLOCK, &block_mask, &omask);
387 #endif
388
389         running = true;
390
391         while(running) {
392 #ifdef HAVE_PSELECT
393                 next_event = last_ping_check + pingtimeout;
394                 if(graph_dump && next_event > last_graph_dump + 60)
395                         next_event = last_graph_dump + 60;
396
397                 if((event = peek_next_event()) && next_event > event->time)
398                         next_event = event->time;
399
400                 if(next_event <= now)
401                         tv.tv_sec = 0;
402                 else
403                         tv.tv_sec = next_event - now;
404                 tv.tv_nsec = 0;
405 #else
406                 tv.tv_sec = 1;
407                 tv.tv_usec = 0;
408 #endif
409
410                 maxfd = build_fdset(&readset, &writeset);
411
412 #ifdef HAVE_MINGW
413                 LeaveCriticalSection(&mutex);
414 #endif
415 #ifdef HAVE_PSELECT
416                 r = pselect(maxfd + 1, &readset, &writeset, NULL, &tv, &omask);
417 #else
418                 r = select(maxfd + 1, &readset, &writeset, NULL, &tv);
419 #endif
420                 now = time(NULL);
421 #ifdef HAVE_MINGW
422                 EnterCriticalSection(&mutex);
423 #endif
424
425                 if(r < 0) {
426                         if(!sockwouldblock(sockerrno)) {
427                                 logger(LOG_ERR, "Error while waiting for input: %s", sockstrerror(sockerrno));
428                                 dump_connections();
429                                 return 1;
430                         }
431                 }
432
433                 if(r > 0)
434                         check_network_activity(&readset, &writeset);
435
436                 if(do_purge) {
437                         purge();
438                         do_purge = false;
439                 }
440
441                 /* Let's check if everybody is still alive */
442
443                 if(last_ping_check + pingtimeout <= now) {
444                         check_dead_connections();
445                         last_ping_check = now;
446
447                         if(routing_mode == RMODE_SWITCH)
448                                 age_subnets();
449
450                         age_past_requests();
451
452                         /* Should we regenerate our key? */
453
454                         if(keyexpires <= now) {
455                                 avl_node_t *node;
456                                 node_t *n;
457
458                                 ifdebug(STATUS) logger(LOG_INFO, "Expiring symmetric keys");
459
460                                 for(node = node_tree->head; node; node = node->next) {
461                                         n = node->data;
462                                         if(n->inkey) {
463                                                 free(n->inkey);
464                                                 n->inkey = NULL;
465                                         }
466                                 }
467
468                                 send_key_changed();
469                                 keyexpires = now + keylifetime;
470                         }
471
472                         /* Detect ADD_EDGE/DEL_EDGE storms that are caused when
473                          * two tinc daemons with the same name are on the VPN.
474                          * If so, sleep a while. If this happens multiple times
475                          * in a row, sleep longer. */
476
477                         if(contradicting_del_edge > 100 && contradicting_add_edge > 100) {
478                                 logger(LOG_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", sleeptime);
479                                 usleep(sleeptime * 1000000LL);
480                                 sleeptime *= 2;
481                                 if(sleeptime < 0)
482                                         sleeptime = 3600;
483                         } else {
484                                 sleeptime /= 2;
485                                 if(sleeptime < 10)
486                                         sleeptime = 10;
487                         }
488
489                         contradicting_add_edge = 0;
490                         contradicting_del_edge = 0;
491                 }
492
493                 if(sigalrm) {
494                         avl_node_t *node;
495                         logger(LOG_INFO, "Flushing event queue");
496                         expire_events();
497                         for(node = connection_tree->head; node; node = node->next) {
498                                 connection_t *c = node->data;
499                                 if(c->status.active)
500                                         send_ping(c);
501                         }
502                         sigalrm = false;
503                 }
504
505                 while((event = get_expired_event())) {
506                         event->handler(event->data);
507                         free_event(event);
508                 }
509
510                 if(sighup) {
511                         connection_t *c;
512                         avl_node_t *node, *next;
513                         char *fname;
514                         struct stat s;
515                         
516                         sighup = false;
517
518                         reopenlogger();
519                         
520                         /* Reread our own configuration file */
521
522                         exit_configuration(&config_tree);
523                         init_configuration(&config_tree);
524
525                         if(!read_server_config()) {
526                                 logger(LOG_ERR, "Unable to reread configuration file, exitting.");
527                                 return 1;
528                         }
529
530                         /* Cancel non-active outgoing connections */
531
532                         for(node = connection_tree->head; node; node = next) {
533                                 next = node->next;
534                                 c = node->data;
535
536                                 c->outgoing = NULL;
537
538                                 if(c->status.connecting) {
539                                         terminate_connection(c, false);
540                                         connection_del(c);
541                                 }
542                         }
543
544                         /* Wipe list of outgoing connections */
545
546                         for(list_node_t *node = outgoing_list->head; node; node = node->next) {
547                                 outgoing_t *outgoing = node->data;
548
549                                 if(outgoing->event)
550                                         event_del(outgoing->event);
551                         }
552
553                         list_delete_list(outgoing_list);
554
555                         /* Close connections to hosts that have a changed or deleted host config file */
556                         
557                         for(node = connection_tree->head; node; node = node->next) {
558                                 c = node->data;
559                                 
560                                 xasprintf(&fname, "%s/hosts/%s", confbase, c->name);
561                                 if(stat(fname, &s) || s.st_mtime > last_config_check)
562                                         terminate_connection(c, c->status.active);
563                                 free(fname);
564                         }
565
566                         last_config_check = now;
567
568                         /* If StrictSubnet is set, expire deleted Subnets and read new ones in */
569
570                         if(strictsubnets) {
571                                 subnet_t *subnet;
572
573                                 for(node = subnet_tree->head; node; node = node->next) {
574                                         subnet = node->data;
575                                         subnet->expires = 1;
576                                 }
577
578                                 load_all_subnets();
579
580                                 for(node = subnet_tree->head; node; node = next) {
581                                         next = node->next;
582                                         subnet = node->data;
583                                         if(subnet->expires == 1) {
584                                                 send_del_subnet(everyone, subnet);
585                                                 if(subnet->owner->status.reachable)
586                                                         subnet_update(subnet->owner, subnet, false);
587                                                 subnet_del(subnet->owner, subnet);
588                                         } else if(subnet->expires == -1) {
589                                                 subnet->expires = 0;
590                                         } else {
591                                                 send_add_subnet(everyone, subnet);
592                                                 if(subnet->owner->status.reachable)
593                                                         subnet_update(subnet->owner, subnet, true);
594                                         }
595                                 }
596                         }
597
598                         /* Try to make outgoing connections */
599                         
600                         try_outgoing_connections();
601                 }
602                 
603                 /* Dump graph if wanted every 60 seconds*/
604
605                 if(last_graph_dump + 60 <= now) {
606                         dump_graph();
607                         last_graph_dump = now;
608                 }
609         }
610
611 #ifdef HAVE_PSELECT
612         /* Restore SIGHUP & SIGALARM mask */
613         sigprocmask(SIG_SETMASK, &omask, NULL);
614 #endif
615
616         return 0;
617 }