git.meshlink.io Git - meshlink/blob - src/net.c

   1 /*
   2     net.c -- most of the network code
   3     Copyright (C) 2014-2017 Guus Sliepen <guus@meshlink.io>
   4
   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.
   9
  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.
  14
  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.
  18 */
  19
  20 #include "system.h"
  21
  22 #include "utils.h"
  23 #include "conf.h"
  24 #include "connection.h"
  25 #include "graph.h"
  26 #include "logger.h"
  27 #include "meshlink_internal.h"
  28 #include "meta.h"
  29 #include "net.h"
  30 #include "netutl.h"
  31 #include "protocol.h"
  32 #include "xalloc.h"
  33
  34 #if !defined(min)
  35 static inline int min(int a, int b) {
  36         return a < b ? a : b;
  37 }
  38 #endif
  39
  40 /*
  41   Terminate a connection:
  42   - Mark it as inactive
  43   - Remove the edge representing this connection
  44   - Kill it with fire
  45   - Check if we need to retry making an outgoing connection
  46 */
  47 void terminate_connection(meshlink_handle_t *mesh, connection_t *c, bool report) {
  48         logger(mesh, MESHLINK_INFO, "Closing connection with %s", c->name);
  49
  50         c->status.active = false;
  51
  52         if(c->node && c->node->connection == c) {
  53                 c->node->connection = NULL;
  54         }
  55
  56         if(c->edge) {
  57                 if(report) {
  58                         send_del_edge(mesh, mesh->everyone, c->edge, 0);
  59                 }
  60
  61                 edge_del(mesh, c->edge);
  62                 c->edge = NULL;
  63
  64                 /* Run MST and SSSP algorithms */
  65
  66                 graph(mesh);
  67
  68                 /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
  69
  70                 if(report && c->node && !c->node->status.reachable) {
  71                         edge_t *e;
  72                         e = lookup_edge(c->node, mesh->self);
  73
  74                         if(e) {
  75                                 send_del_edge(mesh, mesh->everyone, e, 0);
  76                                 edge_del(mesh, e);
  77                         }
  78                 }
  79         }
  80
  81         outgoing_t *outgoing = c->outgoing;
  82         connection_del(mesh, c);
  83
  84         /* Check if this was our outgoing connection */
  85
  86         if(outgoing) {
  87                 do_outgoing_connection(mesh, outgoing);
  88         }
  89
  90 #ifndef HAVE_MINGW
  91         /* Clean up dead proxy processes */
  92
  93         while(waitpid(-1, NULL, WNOHANG) > 0);
  94
  95 #endif
  96 }
  97
  98 /*
  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.
 105 */
 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()");
 109
 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.
 112                 if(c->node) {
 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);
 115                         }
 116                 }
 117
 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) {
 123                                         send_ping(mesh, c);
 124                                         continue;
 125                                 } else {
 126                                         continue;
 127                                 }
 128                         } else {
 129                                 if(c->status.connecting) {
 130                                         logger(mesh, MESHLINK_WARNING, "Timeout while connecting to %s", c->name);
 131                                 } else {
 132                                         logger(mesh, MESHLINK_WARNING, "Timeout from %s during authentication", c->name);
 133                                 }
 134                         }
 135
 136                         terminate_connection(mesh, c, c->status.active);
 137                 }
 138         }
 139
 140         timeout_set(&mesh->loop, data, &(struct timeval) {
 141                 mesh->pingtimeout, rand() % 100000
 142         });
 143 }
 144
 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;
 148
 149         if(na->devclass < nb->devclass) {
 150                 return -1;
 151         }
 152
 153         if(na->devclass > nb->devclass) {
 154                 return 1;
 155         }
 156
 157         if(na->last_successfull_connection == nb->last_successfull_connection) {
 158                 return 0;
 159         }
 160
 161         if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection) {
 162                 return -1;
 163         }
 164
 165         if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection) {
 166                 return 1;
 167         }
 168
 169         if(na < nb) {
 170                 return -1;
 171         }
 172
 173         if(na > nb) {
 174                 return 1;
 175         }
 176
 177         return 0;
 178 }
 179
 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;
 183
 184         if(na->last_successfull_connection == nb->last_successfull_connection) {
 185                 return 0;
 186         }
 187
 188         if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection) {
 189                 return -1;
 190         }
 191
 192         if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection) {
 193                 return 1;
 194         }
 195
 196         if(na < nb) {
 197                 return -1;
 198         }
 199
 200         if(na > nb) {
 201                 return 1;
 202         }
 203
 204         return 0;
 205 }
 206
 207 // devclass desc
 208 static int node_compare_devclass_desc(const void *a, const void *b) {
 209         const node_t *na = a, *nb = b;
 210
 211         if(na->devclass < nb->devclass) {
 212                 return -1;
 213         }
 214
 215         if(na->devclass > nb->devclass) {
 216                 return 1;
 217         }
 218
 219         if(na < nb) {
 220                 return -1;
 221         }
 222
 223         if(na > nb) {
 224                 return 1;
 225         }
 226
 227         return 0;
 228 }
 229
 230
 231 /*
 232
 233 autoconnect()
 234 {
 235         timeout = 5
 236
 237         // find the best one for initial connect
 238
 239         if cur < min
 240                 newcon =
 241                         first from nodes
 242                                 where dclass <= my.dclass and !connection and (timestamp - last_retry) > retry_timeout
 243                                 order by dclass asc, last_connection desc
 244                 if newcon
 245                         timeout = 0
 246                         goto connect
 247
 248
 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
 250
 251         if min <= cur < max
 252                 j = 0
 253                 for i = BACKBONE to my.dclass
 254                         j += count(from connections where node.dclass = i)
 255                         if j < min
 256                                 newcon =
 257                                         first from nodes
 258                                                 where dclass = i and !connection and (timestamp - last_retry) > retry_timeout
 259                                                 order by last_connection desc
 260                                 if newcon
 261                                         goto connect
 262                         else
 263                                 break
 264
 265
 266         // heal partitions
 267
 268         if min <= cur < max
 269                 newcon =
 270                         first from nodes
 271                                 where dclass <= my.dclass and !reachable and (timestamp - last_retry) > retry_timeout
 272                                 order by dclass asc, last_connection desc
 273                 if newcon
 274                         goto connect
 275
 276
 277         // connect
 278
 279 connect:
 280         if newcon
 281                 connect newcon
 282
 283
 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]
 285
 286         if min < cur <= max
 287                 j = 0
 288                 for i = BACKBONE to my.dclass
 289                         j += count(from connections where node.dclass = i)
 290                         if min < j
 291                                 delcon =
 292                                         first from nodes
 293                                                 where dclass >= i and outgoing_connection
 294                                                 order by dclass desc
 295                                 if disconnect
 296                                         goto disconnect
 297                                 else
 298                                         break
 299
 300
 301         // disconnect connections in case we have more than enough connections
 302
 303         if max < cur
 304                 delcon =
 305                         first from nodes
 306                                 where outgoing_connection
 307                                 order by dclass desc
 308                 goto disconnect
 309
 310         // disconnect
 311
 312 disconnect
 313         if delcon
 314                 disconnect delcon
 315
 316
 317         // next iteration
 318         next (timeout, autoconnect)
 319
 320 }
 321
 322 */
 323
 324
 325 static void periodic_handler(event_loop_t *loop, void *data) {
 326         meshlink_handle_t *mesh = loop->data;
 327
 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.
 331         */
 332
 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;
 337
 338                 if(mesh->sleeptime < 0) {
 339                         mesh->sleeptime = 3600;
 340                 }
 341         } else {
 342                 mesh->sleeptime /= 2;
 343
 344                 if(mesh->sleeptime < 10) {
 345                         mesh->sleeptime = 10;
 346                 }
 347         }
 348
 349         mesh->contradicting_add_edge = 0;
 350         mesh->contradicting_del_edge = 0;
 351
 352         int timeout = 5;
 353
 354         /* Check if we need to make or break connections. */
 355
 356         if(mesh->nodes->count > 1) {
 357
 358                 logger(mesh, MESHLINK_DEBUG, "--- autoconnect begin ---");
 359
 360                 int retry_timeout = min(mesh->nodes->count * 5, 60);
 361
 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);
 365
 366
 367                 // connect disconnect nodes
 368
 369                 node_t *connect_to = NULL;
 370                 node_t *disconnect_from = NULL;
 371
 372
 373                 // get cur_connects
 374
 375                 unsigned int cur_connects = 0;
 376
 377                 for list_each(connection_t, c, mesh->connections) {
 378                         if(c->status.active) {
 379                                 cur_connects += 1;
 380                         }
 381                 }
 382
 383                 logger(mesh, MESHLINK_DEBUG, "* cur_connects = %d", cur_connects);
 384                 logger(mesh, MESHLINK_DEBUG, "* outgoings = %d", mesh->outgoings->count);
 385
 386                 // get min_connects and max_connects
 387
 388                 assert(mesh->devclass >= 0 && mesh->devclass <= _DEV_CLASS_MAX);
 389
 390                 unsigned int min_connects = dev_class_traits[mesh->devclass].min_connects;
 391                 unsigned int max_connects = dev_class_traits[mesh->devclass].max_connects;
 392
 393                 logger(mesh, MESHLINK_DEBUG, "* min_connects = %d", min_connects);
 394                 logger(mesh, MESHLINK_DEBUG, "* max_connects = %d", max_connects);
 395
 396
 397                 // find the best one for initial connect
 398
 399                 if(cur_connects < min_connects) {
 400                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
 401
 402                         for splay_each(node_t, n, mesh->nodes) {
 403                                 logger(mesh, MESHLINK_DEBUG, "* n->devclass = %d", n->devclass);
 404
 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);
 407                                 }
 408                         }
 409
 410                         if(nodes->head) {
 411                                 logger(mesh, MESHLINK_DEBUG, "* found best one for initial connect");
 412
 413                                 //timeout = 0;
 414                                 connect_to = (node_t *)nodes->head->data;
 415                         } else {
 416                                 logger(mesh, MESHLINK_DEBUG, "* could not find node for initial connect");
 417                         }
 418
 419                         splay_free_tree(nodes);
 420                 }
 421
 422
 423                 // find better nodes to connect to
 424
 425                 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects) {
 426                         unsigned int connects = 0;
 427
 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) {
 431                                                 connects += 1;
 432                                         }
 433                                 }
 434
 435                                 if(connects < min_connects) {
 436                                         splay_tree_t *nodes = splay_alloc_tree(node_compare_lsc_desc, NULL);
 437
 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);
 441                                                 }
 442                                         }
 443
 444                                         if(nodes->head) {
 445                                                 logger(mesh, MESHLINK_DEBUG, "* found better node");
 446                                                 connect_to = (node_t *)nodes->head->data;
 447
 448                                                 splay_free_tree(nodes);
 449                                                 break;
 450                                         }
 451
 452                                         splay_free_tree(nodes);
 453                                 } else {
 454                                         break;
 455                                 }
 456                         }
 457
 458                         if(!connect_to) {
 459                                 logger(mesh, MESHLINK_DEBUG, "* could not find better nodes");
 460                         }
 461                 }
 462
 463
 464                 // heal partitions
 465
 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);
 468
 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);
 472                                 }
 473                         }
 474
 475                         if(nodes->head) {
 476                                 logger(mesh, MESHLINK_DEBUG, "* try to heal partition");
 477                                 connect_to = (node_t *)nodes->head->data;
 478                         } else {
 479                                 logger(mesh, MESHLINK_DEBUG, "* could not find nodes for partition healing");
 480                         }
 481
 482                         splay_free_tree(nodes);
 483                 }
 484
 485
 486                 // perform connect
 487
 488                 if(connect_to && !connect_to->connection) {
 489                         connect_to->last_connect_try = time(NULL);
 490
 491                         /* check if there is already a connection attempt to this node */
 492                         bool found = false;
 493
 494                         for list_each(outgoing_t, outgoing, mesh->outgoings) {
 495                                 if(!strcmp(outgoing->name, connect_to->name)) {
 496                                         found = true;
 497                                         break;
 498                                 }
 499                         }
 500
 501                         if(!found) {
 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);
 508                         } else {
 509                                 logger(mesh, MESHLINK_DEBUG, "* skip autoconnect since it is an outgoing connection already");
 510                         }
 511                 }
 512
 513
 514                 // disconnect suboptimal outgoing connections
 515
 516                 if(min_connects < cur_connects /*&& cur_connects <= max_connects*/) {
 517                         unsigned int connects = 0;
 518
 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) {
 522                                                 connects += 1;
 523                                         }
 524                                 }
 525
 526                                 if(min_connects < connects) {
 527                                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 528
 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);
 532                                                 }
 533                                         }
 534
 535                                         if(nodes->head) {
 536                                                 logger(mesh, MESHLINK_DEBUG, "* disconnect suboptimal outgoing connection");
 537                                                 disconnect_from = (node_t *)nodes->head->data;
 538                                         }
 539
 540                                         splay_free_tree(nodes);
 541                                         break;
 542                                 }
 543                         }
 544
 545                         if(!disconnect_from) {
 546                                 logger(mesh, MESHLINK_DEBUG, "* no suboptimal outgoing connections");
 547                         }
 548                 }
 549
 550
 551                 // disconnect connections (too many connections)
 552
 553                 if(!disconnect_from && max_connects < cur_connects) {
 554                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 555
 556                         for list_each(connection_t, c, mesh->connections) {
 557                                 if(c->status.active && c->node) {
 558                                         splay_insert(nodes, c->node);
 559                                 }
 560                         }
 561
 562                         if(nodes->head) {
 563                                 logger(mesh, MESHLINK_DEBUG, "* disconnect connection (too many connections)");
 564
 565                                 //timeout = 0;
 566                                 disconnect_from = (node_t *)nodes->head->data;
 567                         } else {
 568                                 logger(mesh, MESHLINK_DEBUG, "* no node we want to disconnect, even though we have too many connections");
 569                         }
 570
 571                         splay_free_tree(nodes);
 572                 }
 573
 574
 575                 // perform disconnect
 576
 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);
 582                 }
 583
 584
 585                 // done!
 586
 587                 logger(mesh, MESHLINK_DEBUG, "--- autoconnect end ---");
 588         }
 589
 590         timeout_set(&mesh->loop, data, &(struct timeval) {
 591                 timeout, rand() % 100000
 592         });
 593 }
 594
 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);
 598                 return;
 599         }
 600 }
 601
 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;
 606
 607                 if(outgoing->ev.cb)
 608                         timeout_set(&mesh->loop, &outgoing->ev, &(struct timeval) {
 609                         0, 0
 610                 });
 611         }
 612
 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;
 617                 }
 618         }
 619
 620         /* Kick the ping timeout handler */
 621         timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval) {
 622                 0, 0
 623         });
 624 }
 625
 626 /*
 627   this is where it all happens...
 628 */
 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
 632         });
 633         timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timeval) {
 634                 0, 0
 635         });
 636
 637         //Add signal handler
 638         mesh->datafromapp.signum = 0;
 639         signal_add(&(mesh->loop), &(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue, mesh, mesh->datafromapp.signum);
 640
 641         if(!event_loop_run(&(mesh->loop), &(mesh->mesh_mutex))) {
 642                 logger(mesh, MESHLINK_ERROR, "Error while waiting for input: %s", strerror(errno));
 643                 return 1;
 644         }
 645
 646         timeout_del(&mesh->loop, &mesh->periodictimer);
 647         timeout_del(&mesh->loop, &mesh->pingtimer);
 648
 649         return 0;
 650 }