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 /*
  35   Terminate a connection:
  36   - Mark it as inactive
  37   - Remove the edge representing this connection
  38   - Kill it with fire
  39   - Check if we need to retry making an outgoing connection
  40 */
  41 void terminate_connection(meshlink_handle_t *mesh, connection_t *c, bool report) {
  42         logger(DEBUG_CONNECTIONS, LOG_NOTICE, "Closing connection with %s (%s)", c->name, c->hostname);
  43
  44         c->status.active = false;
  45
  46         if(c->node && c->node->connection == c)
  47                 c->node->connection = NULL;
  48
  49         if(c->edge) {
  50                 if(report)
  51                         send_del_edge(mesh, mesh->everyone, c->edge);
  52
  53                 edge_del(mesh, c->edge);
  54                 c->edge = NULL;
  55
  56                 /* Run MST and SSSP algorithms */
  57
  58                 graph(mesh);
  59
  60                 /* If the node is not reachable anymore but we remember it had an edge to us, clean it up */
  61
  62                 if(report && !c->node->status.reachable) {
  63                         edge_t *e;
  64                         e = lookup_edge(c->node, mesh->self);
  65                         if(e) {
  66                                 send_del_edge(mesh, mesh->everyone, e);
  67                                 edge_del(mesh, e);
  68                         }
  69                 }
  70         }
  71
  72         outgoing_t *outgoing = c->outgoing;
  73         connection_del(mesh, c);
  74
  75         /* Check if this was our outgoing connection */
  76
  77         if(outgoing)
  78                 do_outgoing_connection(mesh, outgoing);
  79
  80 #ifndef HAVE_MINGW
  81         /* Clean up dead proxy processes */
  82
  83         while(waitpid(-1, NULL, WNOHANG) > 0);
  84 #endif
  85 }
  86
  87 /*
  88   Check if the other end is active.
  89   If we have sent packets, but didn't receive any,
  90   then possibly the other end is dead. We send a
  91   PING request over the meta connection. If the other
  92   end does not reply in time, we consider them dead
  93   and close the connection.
  94 */
  95 static void timeout_handler(event_loop_t *loop, void *data) {
  96         meshlink_handle_t *mesh = loop->data;
  97
  98         for list_each(connection_t, c, mesh->connections) {
  99                 if(c->last_ping_time + mesh->pingtimeout <= mesh->loop.now.tv_sec) {
 100                         if(c->status.active) {
 101                                 if(c->status.pinged) {
 102                                         logger(DEBUG_CONNECTIONS, LOG_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);
 103                                 } else if(c->last_ping_time + mesh->pinginterval <= mesh->loop.now.tv_sec) {
 104                                         send_ping(mesh, c);
 105                                         continue;
 106                                 } else {
 107                                         continue;
 108                                 }
 109                         } else {
 110                                 if(c->status.connecting)
 111                                         logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout while connecting to %s (%s)", c->name, c->hostname);
 112                                 else
 113                                         logger(DEBUG_CONNECTIONS, LOG_WARNING, "Timeout from %s (%s) during authentication", c->name, c->hostname);
 114                         }
 115                         terminate_connection(mesh, c, c->status.active);
 116                 }
 117         }
 118
 119         timeout_set(&mesh->loop, data, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 120 }
 121
 122 static void periodic_handler(event_loop_t *loop, void *data) {
 123         meshlink_handle_t *mesh = loop->data;
 124
 125         /* Check if there are too many contradicting ADD_EDGE and DEL_EDGE messages.
 126            This usually only happens when another node has the same Name as this node.
 127            If so, sleep for a short while to prevent a storm of contradicting messages.
 128         */
 129
 130         if(mesh->contradicting_del_edge > 100 && mesh->contradicting_add_edge > 100) {
 131                 logger(DEBUG_ALWAYS, LOG_WARNING, "Possible node with same Name as us! Sleeping %d seconds.", mesh->sleeptime);
 132                 usleep(mesh->sleeptime * 1000000LL);
 133                 mesh->sleeptime *= 2;
 134                 if(mesh->sleeptime < 0)
 135                         mesh->sleeptime = 3600;
 136         } else {
 137                 mesh->sleeptime /= 2;
 138                 if(mesh->sleeptime < 10)
 139                         mesh->sleeptime = 10;
 140         }
 141
 142         mesh->contradicting_add_edge = 0;
 143         mesh->contradicting_del_edge = 0;
 144
 145         /* If AutoConnect is set, check if we need to make or break connections. */
 146
 147         if(autoconnect && mesh->nodes->count > 1) {
 148                 /* Count number of active connections */
 149                 int nc = 0;
 150                 for list_each(connection_t, c, mesh->connections) {
 151                         if(c->status.active)
 152                                 nc++;
 153                 }
 154
 155                 /* Count number of unreachable nodes */
 156                 int num_unreachable = 0;
 157                 for splay_each(node_t, n, mesh->nodes) {
 158                         if(!n->status.reachable)
 159                                 num_unreachable++;
 160                 }
 161
 162                 if(nc < autoconnect) {
 163                         /* Not enough active connections, try to add one.
 164                            Choose a random node, if we don't have a connection to it,
 165                            and we are not already trying to make one, create an
 166                            outgoing connection to this node.
 167                         */
 168                         int r = rand() % mesh->nodes->count;
 169                         int i = 0;
 170
 171                         for splay_each(node_t, n, mesh->nodes) {
 172                                 if(i++ != r)
 173                                         continue;
 174
 175                                 if(n->connection)
 176                                         break;
 177
 178                                 bool found = false;
 179
 180                                 for list_each(outgoing_t, outgoing, mesh->outgoings) {
 181                                         if(!strcmp(outgoing->name, n->name)) {
 182                                                 found = true;
 183                                                 break;
 184                                         }
 185                                 }
 186
 187                                 if(!found) {
 188                                         //TODO: if the node is blacklisted the connection will not happen, but
 189                                         //the user will read this debug message "Autoconnecting to %s" that is misleading
 190                                         logger(DEBUG_CONNECTIONS, LOG_INFO, "Autoconnecting to %s", n->name);
 191                                         outgoing_t *outgoing = xzalloc(sizeof *outgoing);
 192                                         outgoing->name = xstrdup(n->name);
 193                                         list_insert_tail(mesh->outgoings, outgoing);
 194                                         setup_outgoing_connection(mesh, outgoing);
 195                                 }
 196                                 break;
 197                         }
 198                 } else if (num_unreachable > 0) {
 199                         /* Min number of connections established. Now try
 200                            to connect to some unreachable nodes to attempt
 201                            to heal possible partitions.
 202                         */
 203                         int r = rand() % num_unreachable;
 204                         int i = 0;
 205
 206                         for splay_each(node_t, n, mesh->nodes) {
 207                                 if(n->status.reachable)
 208                                         continue;
 209
 210                                 if(i++ != r)
 211                                         continue;
 212
 213                                 if(n->connection)
 214                                         break;
 215
 216                                 bool found = false;
 217
 218                                 for list_each(outgoing_t, outgoing, mesh->outgoings) {
 219                                         if(!strcmp(outgoing->name, n->name)) {
 220                                                 found = true;
 221                                                 break;
 222                                         }
 223                                 }
 224
 225                                 if(!found) {
 226                                         //TODO: if the node is blacklisted the connection will not happen, but
 227                                         //the user will read this debug message "Autoconnecting to %s" that is misleading
 228                                         logger(DEBUG_CONNECTIONS, LOG_INFO, "Autoconnecting to %s", n->name);
 229                                         outgoing_t *outgoing = xzalloc(sizeof *outgoing);
 230                                         outgoing->name = xstrdup(n->name);
 231                                         list_insert_tail(mesh->outgoings, outgoing);
 232                                         setup_outgoing_connection(mesh, outgoing);
 233                                 }
 234                                 break;
 235                         }
 236                 }
 237                 if(nc > autoconnect) {
 238                         /* Too many active connections, try to remove one.
 239                            Choose a random outgoing connection to a node
 240                            that has at least one other connection.
 241                         */
 242                         int r = rand() % nc;
 243                         int i = 0;
 244
 245                         for list_each(connection_t, c, mesh->connections) {
 246                                 if(!c->status.active)
 247                                         continue;
 248
 249                                 if(i++ != r)
 250                                         continue;
 251
 252                                 if(!c->outgoing || !c->node || c->node->edge_tree->count < 2)
 253                                         break;
 254
 255                                 logger(DEBUG_CONNECTIONS, LOG_INFO, "Autodisconnecting from %s", c->name);
 256                                 list_delete(mesh->outgoings, c->outgoing);
 257                                 c->outgoing = NULL;
 258                                 terminate_connection(mesh, c, c->status.active);
 259                                 break;
 260                         }
 261                 }
 262
 263                 if(nc >= autoconnect) {
 264                         /* If we have enough active connections,
 265                            remove any pending outgoing connections.
 266                            Do not remove pending connections to unreachable
 267                            nodes.
 268                         */
 269                         node_t *o_node = NULL;
 270                         for list_each(outgoing_t, o, mesh->outgoings) {
 271                                 o_node = lookup_node(mesh, o->name);
 272                                 /* o_node is NULL if it is not part of the graph yet */
 273                                 if(!o_node || !o_node->status.reachable)
 274                                         continue;
 275
 276                                 bool found = false;
 277                                 for list_each(connection_t, c, mesh->connections) {
 278                                         if(c->outgoing == o) {
 279                                                 found = true;
 280                                                 break;
 281                                         }
 282                                 }
 283                                 if(!found) {
 284                                         logger(DEBUG_CONNECTIONS, LOG_INFO, "Cancelled outgoing connection to %s", o->name);
 285                                         /* The node variable is leaked in from using the list_each macro.
 286                                            The o variable could be used, but using node directly
 287                                            is more efficient.
 288                                         */
 289                                         list_delete_node(mesh->outgoings, node);
 290                                 }
 291                         }
 292                 }
 293         }
 294
 295         timeout_set(&mesh->loop, data, &(struct timeval){5, rand() % 100000});
 296 }
 297
 298 void handle_meta_connection_data(meshlink_handle_t *mesh, connection_t *c) {
 299         if (!receive_meta(mesh, c)) {
 300                 terminate_connection(mesh, c, c->status.active);
 301                 return;
 302         }
 303 }
 304
 305 void retry(meshlink_handle_t *mesh) {
 306         /* Reset the reconnection timers for all outgoing connections */
 307         for list_each(outgoing_t, outgoing, mesh->outgoings) {
 308                 outgoing->timeout = 0;
 309                 if(outgoing->ev.cb)
 310                         timeout_set(&mesh->loop, &outgoing->ev, &(struct timeval){0, 0});
 311         }
 312
 313         /* Check for outgoing connections that are in progress, and reset their ping timers */
 314         for list_each(connection_t, c, mesh->connections) {
 315                 if(c->outgoing && !c->node)
 316                         c->last_ping_time = 0;
 317         }
 318
 319         /* Kick the ping timeout handler */
 320         timeout_set(&mesh->loop, &mesh->pingtimer, &(struct timeval){0, 0});
 321 }
 322
 323 /*
 324   this is where it all happens...
 325 */
 326 int main_loop(meshlink_handle_t *mesh) {
 327         timeout_add(&mesh->loop, &mesh->pingtimer, timeout_handler, &mesh->pingtimer, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 328         timeout_add(&mesh->loop, &mesh->periodictimer, periodic_handler, &mesh->periodictimer, &(struct timeval){mesh->pingtimeout, rand() % 100000});
 329
 330         //Add signal handler
 331         mesh->datafromapp.signum = 0;
 332         signal_add(&(mesh->loop),&(mesh->datafromapp), (signal_cb_t)meshlink_send_from_queue,mesh, mesh->datafromapp.signum);
 333
 334         if(!event_loop_run(&mesh->loop)) {
 335                 logger(DEBUG_ALWAYS, LOG_ERR, "Error while waiting for input: %s", strerror(errno));
 336                 return 1;
 337         }
 338
 339         timeout_del(&mesh->loop, &mesh->periodictimer);
 340         timeout_del(&mesh->loop, &mesh->pingtimer);
 341
 342         return 0;
 343 }