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

   1 /*
   2     net.c -- most of the network code
   3     Copyright (C) 2014 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 #include <assert.h>
  35
  36 static const int min(int a, int b) {
  37         return a < b ? a : b;
  38 }
  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 (%s)", c->name, c->hostname);
  49
  50         c->status.active = false;
  51
  52         if(c->node && c->node->connection == c)
  53                 c->node->connection = NULL;
  54
  55         if(c->edge) {
  56                 if(report)
  57                         send_del_edge(mesh, mesh->everyone, c->edge);
  58
  59                 edge_del(mesh, c->edge);
  60                 c->edge = NULL;
  61
  62                 /* Run MST and SSSP algorithms */
  63
  64                 graph(mesh);
  65
  66                 /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
  67
  68                 if(report && !c->node->status.reachable) {
  69                         edge_t *e;
  70                         e = lookup_edge(c->node, mesh->self);
  71                         if(e) {
  72                                 send_del_edge(mesh, mesh->everyone, e);
  73                                 edge_del(mesh, e);
  74                         }
  75                 }
  76         }
  77
  78         outgoing_t *outgoing = c->outgoing;
  79         connection_del(mesh, c);
  80
  81         /* Check if this was our outgoing connection */
  82
  83         if(outgoing)
  84                 do_outgoing_connection(mesh, outgoing);
  85
  86 #ifndef HAVE_MINGW
  87         /* Clean up dead proxy processes */
  88
  89         while(waitpid(-1, NULL, WNOHANG) > 0);
  90 #endif
  91 }
  92
  93 /*
  94   Check if the other end is active.
  95   If we have sent packets, but didn't receive any,
  96   then possibly the other end is dead. We send a
  97   PING request over the meta connection. If the other
  98   end does not reply in time, we consider them dead
  99   and close the connection.
 100 */
 101 static void timeout_handler(event_loop_t *loop, void *data) {
 102         meshlink_handle_t *mesh = loop->data;
 103         logger(mesh, MESHLINK_DEBUG, "timeout_handler()");
 104
 105         for list_each(connection_t, c, mesh->connections) {
 106                 // Also make sure that if outstanding key requests for the UDP counterpart of a connection has timed out, we restart it.
 107                 if(c->node) {
 108                         if(c->node->status.waitingforkey && c->node->last_req_key + mesh->pingtimeout <= mesh->loop.now.tv_sec)
 109                                 send_req_key(mesh, c->node);
 110                 }
 111                 if(c->last_ping_time + mesh->pingtimeout <= mesh->loop.now.tv_sec) {
 112                         if(c->status.active) {
 113                                 if(c->status.pinged) {
 114                                         logger(mesh, MESHLINK_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);
 115                                 } else if(c->last_ping_time + mesh->pinginterval <= mesh->loop.now.tv_sec) {
 116                                         send_ping(mesh, c);
 117                                         continue;
 118                                 } else {
 119                                         continue;
 120                                 }
 121                         } else {
 122                                 if(c->status.connecting)
 123                                         logger(mesh, MESHLINK_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
 124                                 else
 125                                         logger(mesh, MESHLINK_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
 126                         }
 127                         terminate_connection(mesh, c, c->status.active);
 128                 }
 129         }
 130
 131         timeout_set(&mesh->loop, data, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 132 }
 133
 134 // devclass asc, last_successfull_connection desc
 135 static int node_compare_devclass_asc_lsc_desc(const void *a, const void *b)
 136 {
 137         const node_t *na = a, *nb = b;
 138
 139         if(na->devclass < nb->devclass)
 140                 { return -1; }
 141
 142         if(na->devclass > nb->devclass)
 143                 { return 1; }
 144
 145         if(na->last_successfull_connection == nb->last_successfull_connection)
 146                 return 0;
 147
 148         if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection)
 149                 return -1;
 150
 151         if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection)
 152                 return 1;
 153
 154         if(na < nb)
 155                 return -1;
 156
 157         if(na > nb)
 158                 return 1;
 159
 160         return 0;
 161 }
 162
 163 // last_successfull_connection desc
 164 static int node_compare_lsc_desc(const void *a, const void *b)
 165 {
 166         const node_t *na = a, *nb = b;
 167
 168         if(na->last_successfull_connection == nb->last_successfull_connection)
 169                 return 0;
 170
 171         if(na->last_successfull_connection == 0 || na->last_successfull_connection > nb->last_successfull_connection)
 172                 return -1;
 173
 174         if(nb->last_successfull_connection == 0 || na->last_successfull_connection < nb->last_successfull_connection)
 175                 return 1;
 176
 177         if(na < nb)
 178                 return -1;
 179
 180         if(na > nb)
 181                 return 1;
 182
 183         return 0;
 184 }
 185
 186 // devclass desc
 187 static int node_compare_devclass_desc(const void *a, const void *b)
 188 {
 189         const node_t *na = a, *nb = b;
 190
 191         if(na->devclass < nb->devclass)
 192                 { return -1; }
 193
 194         if(na->devclass > nb->devclass)
 195                 { return 1; }
 196
 197         if(na < nb)
 198                 return -1;
 199
 200         if(na > nb)
 201                 return 1;
 202
 203         return 0;
 204 }
 205
 206
 207 /*
 208
 209 autoconnect()
 210 {
 211         timeout = 5
 212
 213         // find the best one for initial connect
 214
 215         if cur < min
 216                 newcon =
 217                         first from nodes
 218                                 where dclass <= my.dclass and !connection and (timestamp - last_retry) > retry_timeout
 219                                 order by dclass asc, last_connection desc
 220                 if newcon
 221                         timeout = 0
 222                         goto connect
 223
 224
 225         // 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
 226
 227         if min <= cur < max
 228                 j = 0
 229                 for i = BACKBONE to my.dclass
 230                         j += count(from connections where node.dclass = i)
 231                         if j < min
 232                                 newcon =
 233                                         first from nodes
 234                                                 where dclass = i and !connection and (timestamp - last_retry) > retry_timeout
 235                                                 order by last_connection desc
 236                                 if newcon
 237                                         goto connect
 238                         else
 239                                 break
 240
 241
 242         // heal partitions
 243
 244         if min <= cur < max
 245                 newcon =
 246                         first from nodes
 247                                 where dclass <= my.dclass and !reachable and (timestamp - last_retry) > retry_timeout
 248                                 order by dclass asc, last_connection desc
 249                 if newcon
 250                         goto connect
 251
 252
 253         // connect
 254
 255 connect:
 256         if newcon
 257                 connect newcon
 258
 259
 260         // 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]
 261
 262         if min < cur <= max
 263                 j = 0
 264                 for i = BACKBONE to my.dclass
 265                         j += count(from connections where node.dclass = i)
 266                         if min < j
 267                                 delcon =
 268                                         first from nodes
 269                                                 where dclass >= i and outgoing_connection
 270                                                 order by dclass desc
 271                                 if disconnect
 272                                         goto disconnect
 273                                 else
 274                                         break
 275
 276
 277         // disconnect connections in case we have more than enough connections
 278
 279         if max < cur
 280                 delcon =
 281                         first from nodes
 282                                 where outgoing_connection
 283                                 order by dclass desc
 284                 goto disconnect
 285
 286         // disconnect
 287
 288 disconnect
 289         if delcon
 290                 disconnect delcon
 291
 292
 293         // next iteration
 294         next (timeout, autoconnect)
 295
 296 }
 297
 298 */
 299
 300
 301 static void periodic_handler(event_loop_t *loop, void *data) {
 302         meshlink_handle_t *mesh = loop->data;
 303
 304         /* Check if there are too many contradicting ADD_EDGE and DEL_EDGE messages.
 305            This usually only happens when another node has the same Name as this node.
 306            If so, sleep for a short while to prevent a storm of contradicting messages.
 307         */
 308
 309         if(mesh->contradicting_del_edge > 100 && mesh->contradicting_add_edge > 100) {
 310                 logger(mesh, MESHLINK_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", mesh->sleeptime);
 311                 usleep(mesh->sleeptime * 1000000LL);
 312                 mesh->sleeptime *= 2;
 313                 if(mesh->sleeptime < 0)
 314                         mesh->sleeptime = 3600;
 315         } else {
 316                 mesh->sleeptime /= 2;
 317                 if(mesh->sleeptime < 10)
 318                         mesh->sleeptime = 10;
 319         }
 320
 321         mesh->contradicting_add_edge = 0;
 322         mesh->contradicting_del_edge = 0;
 323
 324         int timeout = 5;
 325
 326         /* Check if we need to make or break connections. */
 327
 328         if(mesh->nodes->count > 1) {
 329
 330                 logger(mesh, MESHLINK_INFO, "--- autoconnect begin ---");
 331
 332                 int retry_timeout = min(mesh->nodes->count * 5, 60);
 333
 334                 logger(mesh, MESHLINK_INFO, "* devclass = %d", mesh->devclass);
 335                 logger(mesh, MESHLINK_INFO, "* nodes = %d", mesh->nodes->count);
 336                 logger(mesh, MESHLINK_INFO, "* retry_timeout = %d", retry_timeout);
 337
 338
 339                 // connect disconnect nodes
 340
 341                 node_t* connect_to = NULL;
 342                 node_t* disconnect_from = NULL;
 343
 344
 345                 // get cur_connects
 346
 347                 int cur_connects = 0;
 348
 349                 for list_each(connection_t, c, mesh->connections)
 350                 {
 351                         if(!c->status.remove_unused)
 352                         {
 353                                 cur_connects += 1;
 354                         }
 355                 }
 356
 357                 logger(mesh, MESHLINK_INFO, "* cur_connects = %d", cur_connects);
 358                 logger(mesh, MESHLINK_INFO, "* outgoings = %d", mesh->outgoings->count);
 359
 360                 // get min_connects and max_connects
 361
 362                 assert(mesh->devclass >= 0 && mesh->devclass <= _DEV_CLASS_MAX);
 363
 364                 int min_connects = dev_class_traits[mesh->devclass].min_connects;
 365                 int max_connects = dev_class_traits[mesh->devclass].max_connects;
 366
 367                 logger(mesh, MESHLINK_INFO, "* min_connects = %d", min_connects);
 368                 logger(mesh, MESHLINK_INFO, "* max_connects = %d", max_connects);
 369
 370
 371                 // find the best one for initial connect
 372
 373                 if(cur_connects < min_connects)
 374                 {
 375                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
 376
 377                         for splay_each(node_t, n, mesh->nodes)
 378                         {
 379                                 logger(mesh, MESHLINK_INFO, "* n->devclass = %d", n->devclass);
 380                                 if(n != mesh->self && n->devclass <= mesh->devclass && !n->connection && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout))
 381                                         { splay_insert(nodes, n); }
 382                         }
 383
 384                         if(nodes->head)
 385                         {
 386                                 logger(mesh, MESHLINK_INFO, "* found best one for initial connect");
 387
 388                                 //timeout = 0;
 389                                 connect_to = (node_t*)nodes->head->data;
 390                         }
 391                         else
 392                                 { logger(mesh, MESHLINK_INFO, "* could not find node for initial connect"); }
 393
 394                         splay_free_tree(nodes);
 395                 }
 396
 397
 398                 // find better nodes to connect to
 399
 400                 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects)
 401                 {
 402                         unsigned int connects = 0;
 403
 404                         for(int devclass = 0; devclass <= mesh->devclass; ++devclass)
 405                         {
 406                                 for list_each(connection_t, c, mesh->connections)
 407                                 {
 408                                         if(!c->status.remove_unused && c->node && c->node->devclass == devclass)
 409                                                 { connects += 1; }
 410                                 }
 411
 412                                 if( connects < min_connects )
 413                                 {
 414                                         splay_tree_t *nodes = splay_alloc_tree(node_compare_lsc_desc, NULL);
 415
 416                                         for splay_each(node_t, n, mesh->nodes)
 417                                         {
 418                                                 if(n != mesh->self && n->devclass == devclass && !n->connection && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout))
 419                                                         { splay_insert(nodes, n); }
 420                                         }
 421
 422                                         if(nodes->head)
 423                                         {
 424                                                 logger(mesh, MESHLINK_INFO, "* found better node");
 425                                                 connect_to = (node_t*)nodes->head->data;
 426
 427                                                 splay_free_tree(nodes);
 428                                                 break;
 429                                         }
 430
 431                                         splay_free_tree(nodes);
 432                                 }
 433                                 else
 434                                         { break; }
 435                         }
 436
 437                         if(!connect_to)
 438                                 { logger(mesh, MESHLINK_INFO, "* could not find better nodes"); }
 439                 }
 440
 441
 442                 // heal partitions
 443
 444                 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects)
 445                 {
 446                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_lsc_desc, NULL);
 447
 448                         for splay_each(node_t, n, mesh->nodes)
 449                         {
 450                                 if(n != mesh->self && n->devclass <= mesh->devclass && !n->status.reachable && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout))
 451                                         { splay_insert(nodes, n); }
 452                         }
 453
 454                         if(nodes->head)
 455                         {
 456                                 logger(mesh, MESHLINK_INFO, "* try to heal partition");
 457                                 connect_to = (node_t*)nodes->head->data;
 458                         }
 459                         else
 460                                 { logger(mesh, MESHLINK_INFO, "* could not find nodes for partition healing"); }
 461
 462                         splay_free_tree(nodes);
 463                 }
 464
 465
 466                 // perform connect
 467
 468                 if(connect_to && !connect_to->connection)
 469                 {
 470                         connect_to->last_connect_try = time(NULL);
 471
 472                         /* check if there is already a connection attempt to this node */
 473                         bool found = false;
 474                         for list_each(outgoing_t, outgoing, mesh->outgoings)
 475                         {
 476                                 if(!strcmp(outgoing->name, connect_to->name))
 477                                 {
 478                                         found = true;
 479                                         break;
 480                                 }
 481                         }
 482
 483                         if(!found)
 484                         {
 485                                 logger(mesh, MESHLINK_INFO, "Autoconnecting to %s", connect_to->name);
 486                                 outgoing_t *outgoing = xzalloc(sizeof(outgoing_t));
 487                                 outgoing->mesh = mesh;
 488                                 outgoing->name = xstrdup(connect_to->name);
 489                                 list_insert_tail(mesh->outgoings, outgoing);
 490                                 setup_outgoing_connection(mesh, outgoing);
 491                         }
 492                         else
 493                                 { logger(mesh, MESHLINK_INFO, "* skip autoconnect since it is an outgoing connection already"); }
 494                 }
 495
 496
 497                 // disconnect suboptimal outgoing connections
 498
 499                 if(min_connects < cur_connects && cur_connects <= max_connects)
 500                 {
 501                         unsigned int connects = 0;
 502
 503                         for(int devclass = 0; devclass <= mesh->devclass; ++devclass)
 504                         {
 505                                 for list_each(connection_t, c, mesh->connections)
 506                                 {
 507                                         if(!c->status.remove_unused && c->node && c->node->devclass == devclass)
 508                                                 { connects += 1; }
 509                                 }
 510
 511                                 if( min_connects < connects )
 512                                 {
 513                                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 514
 515                                         for list_each(connection_t, c, mesh->connections)
 516                                         {
 517                                                 if(!c->status.remove_unused && c->outgoing && c->node && c->node->devclass >= devclass)
 518                                                         { splay_insert(nodes, c->node); }
 519                                         }
 520
 521                                         if(nodes->head)
 522                                         {
 523                                                 logger(mesh, MESHLINK_INFO, "* disconnect suboptimal outgoing connection");
 524                                                 disconnect_from = (node_t*)nodes->head->data;
 525                                         }
 526
 527                                         splay_free_tree(nodes);
 528                                         break;
 529                                 }
 530                         }
 531
 532                         if(!disconnect_from)
 533                                 { logger(mesh, MESHLINK_INFO, "* no suboptimal outgoing connections"); }
 534                 }
 535
 536
 537                 // disconnect connections (too many connections)
 538
 539                 if(!disconnect_from && max_connects < cur_connects)
 540                 {
 541                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 542
 543                         for list_each(connection_t, c, mesh->connections)
 544                         {
 545                                 if(!c->status.remove_unused && c->node)
 546                                         { splay_insert(nodes, c->node); }
 547                         }
 548
 549                         if(nodes->head)
 550                         {
 551                                 logger(mesh, MESHLINK_INFO, "* disconnect connection (too many connections)");
 552
 553                                 //timeout = 0;
 554                                 disconnect_from = (node_t*)nodes->head->data;
 555                         }
 556                         else
 557                                 { logger(mesh, MESHLINK_INFO, "* no node we want to disconnect, even though we have too many connections"); }
 558
 559                         splay_free_tree(nodes);
 560                 }
 561
 562
 563                 // perform disconnect
 564
 565                 if(disconnect_from && disconnect_from->connection)
 566                 {
 567                         logger(mesh, MESHLINK_INFO, "Autodisconnecting from %s", disconnect_from->connection->name);
 568                         list_delete(mesh->outgoings, disconnect_from->connection->outgoing);
 569                         disconnect_from->connection->outgoing = NULL;
 570                         terminate_connection(mesh, disconnect_from->connection, disconnect_from->connection->status.active);
 571                 }
 572
 573
 574                 // done!
 575
 576                 logger(mesh, MESHLINK_INFO, "--- autoconnect end ---");
 577         }
 578
 579         timeout_set(&mesh->loop, data, &(struct timeval){timeout, rand() % 100000});
 580 }
 581
 582 void handle_meta_connection_data(meshlink_handle_t *mesh, connection_t *c) {
 583         if (!receive_meta(mesh, c)) {
 584                 terminate_connection(mesh, c, c->status.active);
 585                 return;
 586         }
 587 }
 588
 589 void retry(meshlink_handle_t *mesh) {
 590         /* Reset the reconnection timers for all outgoing connections */
 591         for list_each(outgoing_t, outgoing, mesh->outgoings) {
 592                 outgoing->timeout = 0;
 593                 if(outgoing->ev.cb)
 594                         timeout_set(&mesh->loop, &outgoing->ev, &(struct timeval){0, 0});
 595         }
 596
 597         /* Check for outgoing connections that are in progress, and reset their ping timers */
 598         for list_each(connection_t, c, mesh->connections) {
 599                 if(c->outgoing && !c->node)
 600                         c->last_ping_time = 0;
 601         }
 602
 603         /* Kick the ping timeout handler */
 604         timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval){0, 0});
 605 }
 606
 607 /*
 608   this is where it all happens...
 609 */
 610 int main_loop(meshlink_handle_t *mesh) {
 611         timeout_add(&mesh->loop, &mesh->pingtimer, timeout_handler, &mesh->pingtimer, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 612         timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timeval){0, 0});
 613
 614         //Add signal handler
 615         mesh->datafromapp.signum = 0;
 616         signal_add(&(mesh->loop),&(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue,mesh, mesh->datafromapp.signum);
 617
 618         if(!event_loop_run(&(mesh->loop), &(mesh->mesh_mutex))) {
 619                 logger(mesh, MESHLINK_ERROR, "Error while waiting for input: %s", strerror(errno));
 620                 return 1;
 621         }
 622
 623         timeout_del(&mesh->loop, &mesh->periodictimer);
 624         timeout_del(&mesh->loop, &mesh->pingtimer);
 625
 626         return 0;
 627 }