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_connect_try desc
 135 static int node_compare_devclass_asc_last_connect_try_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_connect_try == nb->last_connect_try)
 146                 return 0;
 147
 148         if(nb->last_connect_try == 0 || na->last_connect_try < nb->last_connect_try)
 149                 return -1;
 150
 151         if(na->last_connect_try == 0 || na->last_connect_try > nb->last_connect_try)
 152                 return 1;
 153
 154         return 0;
 155 }
 156
 157 // last_connect_try desc
 158 static int node_compare_last_connect_try_desc(const void *a, const void *b)
 159 {
 160         const node_t *na = a, *nb = b;
 161
 162         if(na->last_connect_try == nb->last_connect_try)
 163                 return 0;
 164
 165         if(nb->last_connect_try == 0 || na->last_connect_try < nb->last_connect_try)
 166                 return -1;
 167
 168         if(na->last_connect_try == 0 || na->last_connect_try > nb->last_connect_try)
 169                 return 1;
 170
 171         return 0;
 172 }
 173
 174 // devclass desc
 175 static int node_compare_devclass_desc(const void *a, const void *b)
 176 {
 177         const node_t *na = a, *nb = b;
 178
 179         if(na->devclass < nb->devclass)
 180                 { return -1; }
 181
 182         if(na->devclass > nb->devclass)
 183                 { return 1; }
 184
 185         return 0;
 186 }
 187
 188
 189 /*
 190
 191
 192 autoconnect()
 193 {
 194         timeout = 5
 195
 196         // find the best one for initial connect
 197
 198         if cur < min
 199                 newcon =
 200                         first from nodes
 201                                 where dclass <= my.dclass and !connection and (timestamp - last_retry) > retry_timeout
 202                                 order by dclass asc, last_connection desc
 203                 if newcon
 204                         timeout = 0
 205                         goto connect
 206
 207
 208         // 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
 209
 210         if min <= cur < max
 211                 j = 0
 212                 for i = BACKBONE to my.dclass
 213                         j += count(from connections where node.dclass = i)
 214                         if j < min
 215                                 newcon =
 216                                         first from nodes
 217                                                 where dclass = i and !connection and (timestamp - last_retry) > retry_timeout
 218                                                 order by last_connection desc
 219                                 if newcon
 220                                         goto connect
 221                         else
 222                                 break
 223
 224
 225         // heal partitions
 226
 227         if min <= cur < max
 228                 newcon =
 229                         first from nodes
 230                                 where dclass <= my.dclass and !reachable and (timestamp - last_retry) > retry_timeout
 231                                 order by dclass asc, last_connection desc
 232                 if newcon
 233                         goto connect
 234
 235
 236         // connect
 237
 238 connect:
 239         if newcon
 240                 connect newcon
 241
 242
 243         // 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]
 244
 245         if min < cur <= max
 246                 j = 0
 247                 for i = BACKBONE to my.dclass
 248                         j += count(from connections where node.dclass = i)
 249                         if min < j
 250                                 delcon =
 251                                         first from nodes
 252                                                 where dclass >= i and outgoing_connection
 253                                                 order by dclass desc
 254                                 if disconnect
 255                                         goto disconnect
 256                                 else
 257                                         break
 258
 259
 260         // disconnect connections in case we have more than enough connections
 261
 262         if max < cur
 263                 delcon =
 264                         first from nodes
 265                                 where outgoing_connection
 266                                 order by dclass desc
 267                 goto disconnect
 268
 269         // disconnect
 270
 271 disconnect
 272         if delcon
 273                 disconnect delcon
 274
 275
 276         // next iteration
 277         next (timeout, autoconnect)
 278
 279 }
 280
 281
 282  */
 283
 284
 285 static void periodic_handler(event_loop_t *loop, void *data) {
 286         meshlink_handle_t *mesh = loop->data;
 287
 288         /* Check if there are too many contradicting ADD_EDGE and DEL_EDGE messages.
 289            This usually only happens when another node has the same Name as this node.
 290            If so, sleep for a short while to prevent a storm of contradicting messages.
 291         */
 292
 293         if(mesh->contradicting_del_edge > 100 && mesh->contradicting_add_edge > 100) {
 294                 logger(mesh, MESHLINK_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", mesh->sleeptime);
 295                 usleep(mesh->sleeptime * 1000000LL);
 296                 mesh->sleeptime *= 2;
 297                 if(mesh->sleeptime < 0)
 298                         mesh->sleeptime = 3600;
 299         } else {
 300                 mesh->sleeptime /= 2;
 301                 if(mesh->sleeptime < 10)
 302                         mesh->sleeptime = 10;
 303         }
 304
 305         mesh->contradicting_add_edge = 0;
 306         mesh->contradicting_del_edge = 0;
 307
 308         int timeout = 5;
 309
 310         /* Check if we need to make or break connections. */
 311
 312         if(mesh->nodes->count > 1) {
 313
 314                 logger(mesh, MESHLINK_INFO, "--- autoconnect begin ---");
 315
 316
 317                 int retry_timeout = min(mesh->nodes->count * 5, 60);
 318
 319                 // connect disconnect nodes
 320
 321                 node_t* connect_to = NULL;
 322                 node_t* disconnect_from = NULL;
 323
 324
 325                 // get cur_connects
 326
 327                 int cur_connects = 0;
 328
 329                 for list_each(connection_t, c, mesh->connections)
 330                 {
 331                         if(!c->status.remove_unused)
 332                         {
 333                                 cur_connects += 1;
 334                         }
 335                 }
 336
 337                 logger(mesh, MESHLINK_INFO, "* cur_connects = %d", cur_connects);
 338
 339
 340                 // get min_connects and max_connects
 341
 342                 assert(mesh->devclass >= 0 && mesh->devclass <= _DEV_CLASS_MAX);
 343
 344                 int min_connects = dev_class_traits[mesh->devclass].min_connects;
 345                 int max_connects = dev_class_traits[mesh->devclass].max_connects;
 346
 347                 logger(mesh, MESHLINK_INFO, "* min_connects = %d", min_connects);
 348                 logger(mesh, MESHLINK_INFO, "* max_connects = %d", max_connects);
 349
 350
 351                 // find the best one for initial connect
 352
 353                 if(cur_connects < min_connects)
 354                 {
 355                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_last_connect_try_desc, NULL);
 356
 357                         for splay_each(node_t, n, mesh->nodes)
 358                         {
 359                                 if(n->devclass <= mesh->devclass && !n->connection && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout))
 360                                         { splay_insert(nodes, n); }
 361                         }
 362
 363                         if(nodes->head)
 364                         {
 365                                 logger(mesh, MESHLINK_INFO, "* found best one for initial connect");
 366
 367                                 //timeout = 0;
 368                                 connect_to = (node_t*)nodes->head->data;
 369                         }
 370
 371                         splay_free_tree(nodes);
 372                 }
 373
 374
 375                 // find better nodes to connect to
 376
 377                 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects)
 378                 {
 379                         unsigned int connects = 0;
 380
 381                         for(int devclass = 0; devclass <= mesh->devclass; ++devclass)
 382                         {
 383                                 for list_each(connection_t, c, mesh->connections)
 384                                 {
 385                                         if(!c->status.remove_unused && c->node && c->node->devclass == devclass)
 386                                                 { connects += 1; }
 387                                 }
 388
 389                                 if( connects < min_connects )
 390                                 {
 391                                         splay_tree_t *nodes = splay_alloc_tree(node_compare_last_connect_try_desc, NULL);
 392
 393                                         for splay_each(node_t, n, mesh->nodes)
 394                                         {
 395                                                 if(n->devclass == devclass && !n->connection && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout))
 396                                                         { splay_insert(nodes, n); }
 397                                         }
 398
 399                                         if(nodes->head)
 400                                         {
 401                                                 logger(mesh, MESHLINK_INFO, "* found better node");
 402                                                 connect_to = (node_t*)nodes->head->data;
 403
 404                                                 splay_free_tree(nodes);
 405                                                 break;
 406                                         }
 407
 408                                         splay_free_tree(nodes);
 409                                 }
 410                                 else
 411                                         { break; }
 412                         }
 413                 }
 414
 415
 416                 // heal partitions
 417
 418                 if(!connect_to && min_connects <= cur_connects && cur_connects < max_connects)
 419                 {
 420                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_asc_last_connect_try_desc, NULL);
 421
 422                         for splay_each(node_t, n, mesh->nodes)
 423                         {
 424                                 if(n->devclass <= mesh->devclass && !n->status.reachable && (n->last_connect_try == 0 || (time(NULL) - n->last_connect_try) > retry_timeout))
 425                                         { splay_insert(nodes, n); }
 426                         }
 427
 428                         if(nodes->head)
 429                         {
 430                                 logger(mesh, MESHLINK_INFO, "* try to heal partition");
 431                                 connect_to = (node_t*)nodes->head->data;
 432                         }
 433
 434                         splay_free_tree(nodes);
 435                 }
 436
 437
 438                 // perform connect
 439
 440                 if(connect_to && !connect_to->connection)
 441                 {
 442                         /* check if there is already a connection attempt to this node */
 443                         bool found = false;
 444                         for list_each(outgoing_t, outgoing, mesh->outgoings) {
 445                                 if(!strcmp(outgoing->name, connect_to->name)) {
 446                                         found = true;
 447                                         break;
 448                                 }
 449                         }
 450
 451                         if(!found)
 452                         {
 453                                 logger(mesh, MESHLINK_INFO, "Autoconnecting to %s", connect_to->name);
 454                                 outgoing_t *outgoing = xzalloc(sizeof(outgoing_t));
 455                                 outgoing->mesh = mesh;
 456                                 outgoing->name = xstrdup(connect_to->name);
 457                                 list_insert_tail(mesh->outgoings, outgoing);
 458                                 setup_outgoing_connection(mesh, outgoing);
 459                         }
 460                 }
 461
 462
 463                 // disconnect suboptimal outgoing connections
 464
 465                 if(min_connects < cur_connects && cur_connects <= max_connects)
 466                 {
 467                         unsigned int connects = 0;
 468
 469                         for(int devclass = 0; devclass <= mesh->devclass; ++devclass)
 470                         {
 471                                 for list_each(connection_t, c, mesh->connections)
 472                                 {
 473                                         if(!c->status.remove_unused && c->node && c->node->devclass == devclass)
 474                                                 { connects += 1; }
 475                                 }
 476
 477                                 if( min_connects < connects )
 478                                 {
 479                                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 480
 481                                         for list_each(connection_t, c, mesh->connections)
 482                                         {
 483                                                 if(!c->status.remove_unused && c->outgoing && c->node && c->node->devclass >= devclass)
 484                                                         { splay_insert(nodes, c->node); }
 485                                         }
 486
 487                                         if(nodes->head)
 488                                         {
 489                                                 logger(mesh, MESHLINK_INFO, "* disconnect suboptimal outgoing connection");
 490                                                 disconnect_from = (node_t*)nodes->head->data;
 491                                         }
 492
 493                                         splay_free_tree(nodes);
 494                                         break;
 495                                 }
 496                         }
 497                 }
 498
 499
 500                 // disconnect connections (too many connections)
 501
 502                 if(!disconnect_from && max_connects < cur_connects)
 503                 {
 504                         splay_tree_t *nodes = splay_alloc_tree(node_compare_devclass_desc, NULL);
 505
 506                         for list_each(connection_t, c, mesh->connections)
 507                         {
 508                                 if(!c->status.remove_unused && c->node)
 509                                         { splay_insert(nodes, c->node); }
 510                         }
 511
 512                         if(nodes->head)
 513                         {
 514                                 logger(mesh, MESHLINK_INFO, "* disconnect connection (too many connections");
 515
 516                                 //timeout = 0;
 517                                 disconnect_from = (node_t*)nodes->head->data;
 518                         }
 519
 520                         splay_free_tree(nodes);
 521                 }
 522
 523
 524                 // perform disconnect
 525
 526                 if(disconnect_from && disconnect_from->connection)
 527                 {
 528                         logger(mesh, MESHLINK_INFO, "Autodisconnecting from %s", disconnect_from->connection->name);
 529                         list_delete(mesh->outgoings, disconnect_from->connection->outgoing);
 530                         disconnect_from->connection->outgoing = NULL;
 531                         terminate_connection(mesh, disconnect_from->connection, disconnect_from->connection->status.active);
 532                 }
 533
 534
 535                 // done!
 536
 537                 logger(mesh, MESHLINK_INFO, "--- autoconnect end ---");
 538         }
 539
 540         timeout_set(&mesh->loop, data, &(struct timeval){timeout, rand() % 100000});
 541 }
 542
 543 void handle_meta_connection_data(meshlink_handle_t *mesh, connection_t *c) {
 544         if (!receive_meta(mesh, c)) {
 545                 terminate_connection(mesh, c, c->status.active);
 546                 return;
 547         }
 548 }
 549
 550 void retry(meshlink_handle_t *mesh) {
 551         /* Reset the reconnection timers for all outgoing connections */
 552         for list_each(outgoing_t, outgoing, mesh->outgoings) {
 553                 outgoing->timeout = 0;
 554                 if(outgoing->ev.cb)
 555                         timeout_set(&mesh->loop, &outgoing->ev, &(struct timeval){0, 0});
 556         }
 557
 558         /* Check for outgoing connections that are in progress, and reset their ping timers */
 559         for list_each(connection_t, c, mesh->connections) {
 560                 if(c->outgoing && !c->node)
 561                         c->last_ping_time = 0;
 562         }
 563
 564         /* Kick the ping timeout handler */
 565         timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval){0, 0});
 566 }
 567
 568 /*
 569   this is where it all happens...
 570 */
 571 int main_loop(meshlink_handle_t *mesh) {
 572         timeout_add(&mesh->loop, &mesh->pingtimer, timeout_handler, &mesh->pingtimer, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 573         timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timeval){0, 0});
 574
 575         //Add signal handler
 576         mesh->datafromapp.signum = 0;
 577         signal_add(&(mesh->loop),&(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue,mesh, mesh->datafromapp.signum);
 578
 579         if(!event_loop_run(&(mesh->loop), &(mesh->mesh_mutex))) {
 580                 logger(mesh, MESHLINK_ERROR, "Error while waiting for input: %s", strerror(errno));
 581                 return 1;
 582         }
 583
 584         timeout_del(&mesh->loop, &mesh->periodictimer);
 585         timeout_del(&mesh->loop, &mesh->pingtimer);
 586
 587         return 0;
 588 }