src/or/main.c

   1 /* Copyright 2001,2002,2003 Roger Dingledine, Matej Pfajfar. */
   2 /* See LICENSE for licensing information */
   3 /* $Id$ */
   4
   5 /**
   6  * \file main.c
   7  * \brief Tor main loop and startup functions.
   8  **/
   9
  10 #include "or.h"
  11
  12 /********* PROTOTYPES **********/
  13
  14 static void dumpstats(int severity); /* log stats */
  15 static int init_from_config(int argc, char **argv);
  16
  17 /********* START VARIABLES **********/
  18
  19 /* declared in connection.c */
  20 extern char *conn_state_to_string[][_CONN_TYPE_MAX+1];
  21
  22 or_options_t options; /**< Command-line and config-file options. */
  23 int global_read_bucket; /**< Max number of bytes I can read this second. */
  24
  25 /** What was the read bucket before the last call to prepare_for_pool?
  26  * (used to determine how many bytes we've read). */
  27 static int stats_prev_global_read_bucket;
  28 /** How many bytes have we read since we started the process? */
  29 static uint64_t stats_n_bytes_read = 0;
  30 /** How many seconds have we been running? */
  31 static long stats_n_seconds_reading = 0;
  32
  33 /** Array of all open connections; each element corresponds to the element of
  34  * poll_array in the same position.  The first nfds elements are valid. */
  35 static connection_t *connection_array[MAXCONNECTIONS] =
  36         { NULL };
  37
  38 /** Array of pollfd objects for calls to poll(). */
  39 static struct pollfd poll_array[MAXCONNECTIONS];
  40
  41 static int nfds=0; /**< Number of connections currently active. */
  42
  43 #ifndef MS_WINDOWS /* do signal stuff only on unix */
  44 static int please_dumpstats=0; /**< Whether we should dump stats during the loop. */
  45 static int please_reset=0; /**< Whether we just got a sighup. */
  46 static int please_reap_children=0; /**< Whether we should waitpid for exited children. */
  47 #endif /* signal stuff */
  48
  49 /** We set this to 1 when we've fetched a dir, to know whether to complain
  50  * yet about unrecognized nicknames in entrynodes, exitnodes, etc.
  51  * Also, we don't try building circuits unless this is 1. */
  52 int has_fetched_directory=0;
  53
  54 /** We set this to 1 when we've opened a circuit, so we can print a log
  55  * entry to inform the user that Tor is working. */
  56 int has_completed_circuit=0;
  57
  58 /********* END VARIABLES ************/
  59
  60 /****************************************************************************
  61 *
  62 * This section contains accessors and other methods on the connection_array
  63 * and poll_array variables (which are global within this file and unavailable
  64 * outside it).
  65 *
  66 ****************************************************************************/
  67
  68 /** Add <b>conn</b> to the array of connections that we can poll on.  The
  69  * connection's socket must be set; the connection starts out
  70  * non-reading and non-writing.
  71  */
  72 int connection_add(connection_t *conn) {
  73   tor_assert(conn);
  74   tor_assert(conn->s >= 0);
  75
  76   if(nfds >= options.MaxConn-1) {
  77     log_fn(LOG_WARN,"failing because nfds is too high.");
  78     return -1;
  79   }
  80
  81   tor_assert(conn->poll_index == -1); /* can only connection_add once */
  82   conn->poll_index = nfds;
  83   connection_array[nfds] = conn;
  84
  85   poll_array[nfds].fd = conn->s;
  86
  87   /* zero these out here, because otherwise we'll inherit values from the previously freed one */
  88   poll_array[nfds].events = 0;
  89   poll_array[nfds].revents = 0;
  90
  91   nfds++;
  92
  93   log_fn(LOG_INFO,"new conn type %s, socket %d, nfds %d.",
  94       CONN_TYPE_TO_STRING(conn->type), conn->s, nfds);
  95
  96   return 0;
  97 }
  98
  99 /** Remove the connection from the global list, and remove the
 100  * corresponding poll entry.  Calling this function will shift the last
 101  * connection (if any) into the position occupied by conn.
 102  */
 103 int connection_remove(connection_t *conn) {
 104   int current_index;
 105
 106   tor_assert(conn);
 107   tor_assert(nfds>0);
 108
 109   log_fn(LOG_INFO,"removing socket %d (type %s), nfds now %d",
 110          conn->s, CONN_TYPE_TO_STRING(conn->type), nfds-1);
 111
 112   tor_assert(conn->poll_index >= 0);
 113   current_index = conn->poll_index;
 114   if(current_index == nfds-1) { /* this is the end */
 115     nfds--;
 116     return 0;
 117   }
 118
 119   /* replace this one with the one at the end */
 120   nfds--;
 121   poll_array[current_index].fd = poll_array[nfds].fd;
 122   poll_array[current_index].events = poll_array[nfds].events;
 123   poll_array[current_index].revents = poll_array[nfds].revents;
 124   connection_array[current_index] = connection_array[nfds];
 125   connection_array[current_index]->poll_index = current_index;
 126
 127   return 0;
 128 }
 129
 130 /** Set <b>*array</b> to an array of all connections, and <b>*n</b>
 131  * to the length of the array. <b>*array</b> and <b>*n</b> must not
 132  * be modified.
 133  */
 134 void get_connection_array(connection_t ***array, int *n) {
 135   *array = connection_array;
 136   *n = nfds;
 137 }
 138
 139 /** Set the event mask on <b>conn</b> to <b>events</b>.  (The form of
 140 * the event mask is as for poll().)
 141  */
 142 void connection_watch_events(connection_t *conn, short events) {
 143
 144   tor_assert(conn && conn->poll_index >= 0 && conn->poll_index < nfds);
 145
 146   poll_array[conn->poll_index].events = events;
 147 }
 148
 149 /** Return true iff <b>conn</b> is listening for read events. */
 150 int connection_is_reading(connection_t *conn) {
 151   tor_assert(conn && conn->poll_index >= 0);
 152   return poll_array[conn->poll_index].events & POLLIN;
 153 }
 154
 155 /** Tell the main loop to stop notifying <b>conn</b> of any read events. */
 156 void connection_stop_reading(connection_t *conn) {
 157   tor_assert(conn && conn->poll_index >= 0 && conn->poll_index < nfds);
 158
 159   log(LOG_DEBUG,"connection_stop_reading() called.");
 160   if(poll_array[conn->poll_index].events & POLLIN)
 161     poll_array[conn->poll_index].events -= POLLIN;
 162 }
 163
 164 /** Tell the main loop to start notifying <b>conn</b> of any read events. */
 165 void connection_start_reading(connection_t *conn) {
 166   tor_assert(conn && conn->poll_index >= 0 && conn->poll_index < nfds);
 167   poll_array[conn->poll_index].events |= POLLIN;
 168 }
 169
 170 /** Return true iff <b>conn</b> is listening for write events. */
 171 int connection_is_writing(connection_t *conn) {
 172   return poll_array[conn->poll_index].events & POLLOUT;
 173 }
 174
 175 /** Tell the main loop to stop notifying <b>conn</b> of any write events. */
 176 void connection_stop_writing(connection_t *conn) {
 177   tor_assert(conn && conn->poll_index >= 0 && conn->poll_index < nfds);
 178   if(poll_array[conn->poll_index].events & POLLOUT)
 179     poll_array[conn->poll_index].events -= POLLOUT;
 180 }
 181
 182 /** Tell the main loop to start notifying <b>conn</b> of any write events. */
 183 void connection_start_writing(connection_t *conn) {
 184   tor_assert(conn && conn->poll_index >= 0 && conn->poll_index < nfds);
 185   poll_array[conn->poll_index].events |= POLLOUT;
 186 }
 187
 188 /** Called when the connection at connection_array[i] has a read event,
 189  * or it has pending tls data waiting to be read: checks for validity,
 190  * catches numerous errors, and dispatches to connection_handle_read.
 191  */
 192 static void conn_read(int i) {
 193   connection_t *conn = connection_array[i];
 194
 195   if (conn->marked_for_close)
 196     return;
 197
 198   /* see http://www.greenend.org.uk/rjk/2001/06/poll.html for
 199    * discussion of POLLIN vs POLLHUP */
 200   if(!(poll_array[i].revents & (POLLIN|POLLHUP|POLLERR)))
 201     if(!connection_is_reading(conn) ||
 202        !connection_has_pending_tls_data(conn))
 203       return; /* this conn should not read */
 204
 205   log_fn(LOG_DEBUG,"socket %d wants to read.",conn->s);
 206
 207   assert_connection_ok(conn, time(NULL));
 208   assert_all_pending_dns_resolves_ok();
 209
 210   if(
 211     /* XXX does POLLHUP also mean it's definitely broken? */
 212 #ifdef MS_WINDOWS
 213     (poll_array[i].revents & POLLERR) ||
 214 #endif
 215     connection_handle_read(conn) < 0) {
 216       if (!conn->marked_for_close) {
 217         /* this connection is broken. remove it */
 218         /* XXX This shouldn't ever happen anymore. */
 219         /* XXX but it'll clearly happen on MS_WINDOWS from POLLERR, right? */
 220         log_fn(LOG_ERR,"Unhandled error on read for %s connection (fd %d); removing",
 221                CONN_TYPE_TO_STRING(conn->type), conn->s);
 222         connection_mark_for_close(conn,0);
 223       }
 224   }
 225   assert_connection_ok(conn, time(NULL));
 226   assert_all_pending_dns_resolves_ok();
 227 }
 228
 229 /** Called when the connection at connection_array[i] has a write event:
 230  * checks for validity, catches numerous errors, and dispatches to
 231  * connection_handle_write.
 232  */
 233 static void conn_write(int i) {
 234   connection_t *conn;
 235
 236   if(!(poll_array[i].revents & POLLOUT))
 237     return; /* this conn doesn't want to write */
 238
 239   conn = connection_array[i];
 240   log_fn(LOG_DEBUG,"socket %d wants to write.",conn->s);
 241   if (conn->marked_for_close)
 242     return;
 243
 244   assert_connection_ok(conn, time(NULL));
 245   assert_all_pending_dns_resolves_ok();
 246
 247   if (connection_handle_write(conn) < 0) {
 248     if (!conn->marked_for_close) {
 249       /* this connection is broken. remove it. */
 250       log_fn(LOG_WARN,"Unhandled error on read for %s connection (fd %d); removing",
 251              CONN_TYPE_TO_STRING(conn->type), conn->s);
 252       conn->has_sent_end = 1; /* otherwise we cry wolf about duplicate close */
 253       connection_mark_for_close(conn,0);
 254     }
 255   }
 256   assert_connection_ok(conn, time(NULL));
 257   assert_all_pending_dns_resolves_ok();
 258 }
 259
 260 /** If the connection at connection_array[i] is marked for close, then:
 261  *    - If it has data that it wants to flush, try to flush it.
 262  *    - If it _still_ has data to flush, and conn->hold_open_until_flushed is
 263  *      true, then leave the connection open and return.
 264  *    - Otherwise, remove the connection from connection_array and from
 265  *      all other lists, close it, and free it.
 266  * If we remove the connection, then call conn_closed_if_marked at the new
 267  * connection at position i.
 268  */
 269 static void conn_close_if_marked(int i) {
 270   connection_t *conn;
 271   int retval;
 272
 273   conn = connection_array[i];
 274   assert_connection_ok(conn, time(NULL));
 275   assert_all_pending_dns_resolves_ok();
 276   if(!conn->marked_for_close)
 277     return; /* nothing to see here, move along */
 278
 279   log_fn(LOG_INFO,"Cleaning up connection (fd %d).",conn->s);
 280   if(conn->s >= 0 && connection_wants_to_flush(conn)) {
 281     /* -1 means it's an incomplete edge connection, or that the socket
 282      * has already been closed as unflushable. */
 283     if(!conn->hold_open_until_flushed)
 284       log_fn(LOG_WARN,
 285         "Conn (fd %d, type %s, state %d) marked, but wants to flush %d bytes. "
 286         "(Marked at %s:%d)",
 287         conn->s, CONN_TYPE_TO_STRING(conn->type), conn->state,
 288         conn->outbuf_flushlen, conn->marked_for_close_file, conn->marked_for_close);
 289     if(connection_speaks_cells(conn)) {
 290       if(conn->state == OR_CONN_STATE_OPEN) {
 291         retval = flush_buf_tls(conn->tls, conn->outbuf, &conn->outbuf_flushlen);
 292         /* XXX actually, some non-zero results are maybe ok. which ones? */
 293       } else
 294         retval = -1; /* never flush non-open broken tls connections */
 295     } else {
 296       retval = flush_buf(conn->s, conn->outbuf, &conn->outbuf_flushlen);
 297     }
 298     if(retval >= 0 &&
 299        conn->hold_open_until_flushed && connection_wants_to_flush(conn)) {
 300       log_fn(LOG_INFO,"Holding conn (fd %d) open for more flushing.",conn->s);
 301       /* XXX should we reset timestamp_lastwritten here? */
 302       return;
 303     }
 304     if(connection_wants_to_flush(conn)) {
 305       log_fn(LOG_WARN,"Conn (fd %d, type %s, state %d) still wants to flush. Losing %d bytes! (Marked at %s:%d)",
 306              conn->s, CONN_TYPE_TO_STRING(conn->type), conn->state,
 307              (int)buf_datalen(conn->outbuf), conn->marked_for_close_file,
 308              conn->marked_for_close);
 309     }
 310   }
 311   /* if it's an edge conn, remove it from the list
 312    * of conn's on this circuit. If it's not on an edge,
 313    * flush and send destroys for all circuits on this conn
 314    */
 315   circuit_about_to_close_connection(conn);
 316   connection_remove(conn);
 317   if(conn->type == CONN_TYPE_EXIT) {
 318     assert_connection_edge_not_dns_pending(conn);
 319   }
 320   connection_free(conn);
 321   if(i<nfds) { /* we just replaced the one at i with a new one.
 322                   process it too. */
 323     conn_close_if_marked(i);
 324   }
 325 }
 326
 327 /** This function is called whenever we successfully pull down a directory */
 328 void directory_has_arrived(void) {
 329
 330   log_fn(LOG_INFO, "A directory has arrived.");
 331
 332   /* just for testing */
 333 //  directory_initiate_command(router_pick_directory_server(),
 334 //                             DIR_PURPOSE_FETCH_RENDDESC, "foo", 3);
 335
 336   has_fetched_directory=1;
 337
 338   if(options.ORPort) { /* connect to them all */
 339     router_retry_connections();
 340   }
 341 }
 342
 343 /** Perform regular maintenance tasks for a single connection.  This
 344  * function gets run once per second per connection by run_housekeeping.
 345  */
 346 static void run_connection_housekeeping(int i, time_t now) {
 347   cell_t cell;
 348   connection_t *conn = connection_array[i];
 349
 350   /* Expire any directory connections that haven't sent anything for 5 min */
 351   if(conn->type == CONN_TYPE_DIR &&
 352      !conn->marked_for_close &&
 353      conn->timestamp_lastwritten + 5*60 < now) {
 354     log_fn(LOG_WARN,"Expiring wedged directory conn (purpose %d)", conn->purpose);
 355     /* XXXX This next check may help isolate where the pesky EPIPE bug
 356      * really occurs. */
 357     if (connection_wants_to_flush(conn)) {
 358       flush_buf(conn->s, conn->outbuf, &conn->outbuf_flushlen);
 359     }
 360     connection_mark_for_close(conn,0);
 361     /* XXXX Does this next part make sense, really? */
 362     conn->hold_open_until_flushed = 1; /* give it a last chance */
 363     return;
 364   }
 365
 366   /* check connections to see whether we should send a keepalive, expire, or wait */
 367   if(!connection_speaks_cells(conn))
 368     return;
 369
 370   /* If we haven't written to an OR connection for a while, then either nuke
 371      the connection or send a keepalive, depending. */
 372   if(now >= conn->timestamp_lastwritten + options.KeepalivePeriod) {
 373     if((!options.ORPort && !circuit_get_by_conn(conn)) ||
 374        (!connection_state_is_open(conn))) {
 375       /* we're an onion proxy, with no circuits; or our handshake has expired. kill it. */
 376       log_fn(LOG_INFO,"Expiring connection to %d (%s:%d).",
 377              i,conn->address, conn->port);
 378       /* flush anything waiting, e.g. a destroy for a just-expired circ */
 379       connection_mark_for_close(conn,CLOSE_REASON_UNUSED_OR_CONN);
 380       conn->hold_open_until_flushed = 1;
 381     } else {
 382       /* either a full router, or we've got a circuit. send a padding cell. */
 383       log_fn(LOG_DEBUG,"Sending keepalive to (%s:%d)",
 384              conn->address, conn->port);
 385       memset(&cell,0,sizeof(cell_t));
 386       cell.command = CELL_PADDING;
 387       connection_or_write_cell_to_buf(&cell, conn);
 388     }
 389   }
 390 }
 391
 392 /** Perform regular maintenance tasks.  This function gets run once per
 393  * second by prepare_for_poll.
 394  */
 395 static void run_scheduled_events(time_t now) {
 396   static long time_to_fetch_directory = 0;
 397   static time_t last_uploaded_services = 0;
 398   static time_t last_rotated_certificate = 0;
 399   int i;
 400
 401
 402   /** 1a. Every MIN_ONION_KEY_LIFETIME seconds, rotate the onion keys,
 403    *  shut down and restart all cpuworkers, and update the directory if
 404    *  necessary.
 405    */
 406   if (options.ORPort && get_onion_key_set_at()+MIN_ONION_KEY_LIFETIME < now) {
 407     log_fn(LOG_INFO,"Rotating onion key.");
 408     rotate_onion_key();
 409     cpuworkers_rotate();
 410     if (router_rebuild_descriptor()<0) {
 411       log_fn(LOG_WARN, "Couldn't rebuild router descriptor");
 412     }
 413     router_upload_dir_desc_to_dirservers();
 414   }
 415
 416   /** 1b. Every MAX_SSL_KEY_LIFETIME seconds, we change our TLS context. */
 417   if (!last_rotated_certificate)
 418     last_rotated_certificate = now;
 419   if (options.ORPort && last_rotated_certificate+MAX_SSL_KEY_LIFETIME < now) {
 420     log_fn(LOG_INFO,"Rotating tls context.");
 421     if (tor_tls_context_new(get_identity_key(), 1, options.Nickname,
 422                             MAX_SSL_KEY_LIFETIME) < 0) {
 423       log_fn(LOG_WARN, "Error reinitializing TLS context");
 424     }
 425     last_rotated_certificate = now;
 426     /* XXXX We should rotate TLS connections as well; this code doesn't change
 427      * XXXX them at all. */
 428   }
 429
 430   /** 1c. Every DirFetchPostPeriod seconds, we get a new directory and upload
 431    *    our descriptor (if any). */
 432   if(time_to_fetch_directory < now) {
 433     /* it's time to fetch a new directory and/or post our descriptor */
 434     if(options.ORPort) {
 435       router_rebuild_descriptor();
 436       router_upload_dir_desc_to_dirservers();
 437     }
 438     if(!options.DirPort) {
 439       /* NOTE directory servers do not currently fetch directories.
 440        * Hope this doesn't bite us later. */
 441       directory_initiate_command(router_pick_directory_server(),
 442                                  DIR_PURPOSE_FETCH_DIR, NULL, 0);
 443     } else {
 444       /* We're a directory; dump any old descriptors. */
 445       dirserv_remove_old_servers();
 446     }
 447     /* Force an upload of our descriptors every DirFetchPostPeriod seconds. */
 448     rend_services_upload(1);
 449     last_uploaded_services = now;
 450     rend_cache_clean(); /* should this go elsewhere? */
 451     time_to_fetch_directory = now + options.DirFetchPostPeriod;
 452   }
 453
 454
 455   /** 2. Every second, we examine pending circuits and prune the
 456    *    ones which have been pending for more than a few seconds.
 457    *    We do this before step 3, so it can try building more if
 458    *    it's not comfortable with the number of available circuits.
 459    */
 460   circuit_expire_building(now);
 461
 462   /** 2b. Also look at pending streams and prune the ones that 'began'
 463    *     a long time ago but haven't gotten a 'connected' yet.
 464    *     Do this before step 3, so we can put them back into pending
 465    *     state to be picked up by the new circuit.
 466    */
 467   connection_ap_expire_beginning();
 468
 469
 470   /** 2c. And expire connections that we've held open for too long.
 471    */
 472   connection_expire_held_open();
 473
 474   /** 3. Every second, we try a new circuit if there are no valid
 475    *    circuits. Every NewCircuitPeriod seconds, we expire circuits
 476    *    that became dirty more than NewCircuitPeriod seconds ago,
 477    *    and we make a new circ if there are no clean circuits.
 478    */
 479   if(has_fetched_directory)
 480     circuit_build_needed_circs(now);
 481
 482   /** 4. We do housekeeping for each connection... */
 483   for(i=0;i<nfds;i++) {
 484     run_connection_housekeeping(i, now);
 485   }
 486
 487   /** 5. And remove any marked circuits... */
 488   circuit_close_all_marked();
 489
 490   /** 6. And upload service descriptors for any services whose intro points
 491    *    have changed in the last second. */
 492   if (last_uploaded_services < now-5) {
 493     rend_services_upload(0);
 494     last_uploaded_services = now;
 495   }
 496
 497   /** 7. and blow away any connections that need to die. have to do this now,
 498    * because if we marked a conn for close and left its socket -1, then
 499    * we'll pass it to poll/select and bad things will happen.
 500    */
 501   for(i=0;i<nfds;i++)
 502     conn_close_if_marked(i);
 503 }
 504
 505 /** Called every time we're about to call tor_poll.  Increments statistics,
 506  * and adjusts token buckets.  Returns the number of milliseconds to use for
 507  * the poll() timeout.
 508  */
 509 static int prepare_for_poll(void) {
 510   static long current_second = 0; /* from previous calls to gettimeofday */
 511   connection_t *conn;
 512   struct timeval now;
 513   int i;
 514
 515   tor_gettimeofday(&now);
 516
 517   /* Check how much bandwidth we've consumed, and increment the token
 518    * buckets. */
 519   stats_n_bytes_read += stats_prev_global_read_bucket-global_read_bucket;
 520   connection_bucket_refill(&now);
 521   stats_prev_global_read_bucket = global_read_bucket;
 522
 523   if(now.tv_sec > current_second) { /* the second has rolled over. check more stuff. */
 524
 525     ++stats_n_seconds_reading;
 526     assert_all_pending_dns_resolves_ok();
 527     run_scheduled_events(now.tv_sec);
 528     assert_all_pending_dns_resolves_ok();
 529
 530     current_second = now.tv_sec; /* remember which second it is, for next time */
 531   }
 532
 533   for(i=0;i<nfds;i++) {
 534     conn = connection_array[i];
 535     if(connection_has_pending_tls_data(conn) &&
 536        connection_is_reading(conn)) {
 537       log_fn(LOG_DEBUG,"sock %d has pending bytes.",conn->s);
 538       return 0; /* has pending bytes to read; don't let poll wait. */
 539     }
 540   }
 541
 542   return (1000 - (now.tv_usec / 1000)); /* how many milliseconds til the next second? */
 543 }
 544
 545 /** Configure the Tor process from the command line arguments and from the
 546  * configuration file.
 547  */
 548 static int init_from_config(int argc, char **argv) {
 549   /* read the configuration file. */
 550   if(getconfig(argc,argv,&options)) {
 551     log_fn(LOG_ERR,"Reading config failed. For usage, try -h.");
 552     return -1;
 553   }
 554   close_logs(); /* we'll close, then open with correct loglevel if necessary */
 555
 556   /* Setuid/setgid as appropriate */
 557   if(options.User || options.Group) {
 558     if(switch_id(options.User, options.Group) != 0) {
 559       return -1;
 560     }
 561   }
 562
 563   /* Start backgrounding the process, if requested. */
 564   if (options.RunAsDaemon) {
 565     start_daemon(options.DataDirectory);
 566   }
 567
 568   /* Configure the log(s) */
 569   if(!options.LogFile && !options.RunAsDaemon)
 570     add_stream_log(options.loglevel, "<stdout>", stdout);
 571   if(options.LogFile) {
 572     if (add_file_log(options.loglevel, options.LogFile) != 0) {
 573       /* opening the log file failed!  Use stderr and log a warning */
 574       add_stream_log(options.loglevel, "<stderr>", stderr);
 575       log_fn(LOG_WARN, "Cannot write to LogFile '%s': %s.", options.LogFile, strerror(errno));
 576     }
 577     log_fn(LOG_NOTICE, "Successfully opened LogFile '%s', redirecting output.",
 578            options.LogFile);
 579   }
 580   if(options.DebugLogFile) {
 581     if (add_file_log(LOG_DEBUG, options.DebugLogFile) != 0)
 582       log_fn(LOG_WARN, "Cannot write to DebugLogFile '%s': %s.", options.DebugLogFile, strerror(errno));
 583     log_fn(LOG_DEBUG, "Successfully opened DebugLogFile '%s'.", options.DebugLogFile);
 584   }
 585
 586   /* Set up our buckets */
 587   connection_bucket_init();
 588   stats_prev_global_read_bucket = global_read_bucket;
 589
 590   /* Finish backgrounding the process */
 591   if(options.RunAsDaemon) {
 592     /* XXXX Can we delay this any more? */
 593     finish_daemon();
 594   }
 595
 596   /* Write our pid to the pid file. if we do not have write permissions we
 597    * will log a warning */
 598   if(options.PidFile)
 599     write_pidfile(options.PidFile);
 600
 601   return 0;
 602 }
 603
 604 /** Called when we get a SIGHUP: reload configuration files and keys,
 605  * retry all connections, re-upload all descriptors, and so on. */
 606 static int do_hup(void) {
 607   char keydir[512];
 608
 609   log_fn(LOG_NOTICE,"Received sighup. Reloading config.");
 610   has_completed_circuit=0;
 611   /* first, reload config variables, in case they've changed */
 612   /* no need to provide argc/v, they've been cached inside init_from_config */
 613   if (init_from_config(0, NULL) < 0) {
 614     exit(1);
 615   }
 616   /* reload keys as needed for rendezvous services. */
 617   if (rend_service_load_keys()<0) {
 618     log_fn(LOG_ERR,"Error reloading rendezvous service keys");
 619     exit(1);
 620   }
 621   if(retry_all_connections() < 0) {
 622     log_fn(LOG_ERR,"Failed to bind one of the listener ports.");
 623     return -1;
 624   }
 625   if(options.DirPort) {
 626     /* reload the approved-routers file */
 627     sprintf(keydir,"%s/approved-routers", options.DataDirectory);
 628     log_fn(LOG_INFO,"Reloading approved fingerprints from %s...",keydir);
 629     if(dirserv_parse_fingerprint_file(keydir) < 0) {
 630       log_fn(LOG_WARN, "Error reloading fingerprints. Continuing with old list.");
 631     }
 632     /* Since we aren't fetching a directory, we won't retry rendezvous points
 633      * when it gets in.  Try again now. */
 634     rend_services_introduce();
 635   } else {
 636     /* fetch a new directory */
 637     directory_initiate_command(router_pick_directory_server(),
 638                                DIR_PURPOSE_FETCH_DIR, NULL, 0);
 639   }
 640   if(options.ORPort) {
 641     router_rebuild_descriptor();
 642     sprintf(keydir,"%s/router.desc", options.DataDirectory);
 643     log_fn(LOG_INFO,"Dumping descriptor to %s...",keydir);
 644     if (write_str_to_file(keydir, router_get_my_descriptor())) {
 645       return -1;
 646     }
 647   }
 648   return 0;
 649 }
 650
 651 /** Tor main loop. */
 652 static int do_main_loop(void) {
 653   int i;
 654   int timeout;
 655   int poll_result;
 656
 657   /* Initialize the history structures. */
 658   rep_hist_init();
 659   /* Intialize the service cache. */
 660   rend_cache_init();
 661
 662   /* load the private keys, if we're supposed to have them, and set up the
 663    * TLS context. */
 664   if (init_keys() < 0 || rend_service_load_keys() < 0) {
 665     log_fn(LOG_ERR,"Error initializing keys; exiting");
 666     return -1;
 667   }
 668
 669   /* load the routers file */
 670   if(options.RouterFile &&
 671      router_set_routerlist_from_file(options.RouterFile) < 0) {
 672     log_fn(LOG_ERR,"Error loading router list.");
 673     return -1;
 674   }
 675
 676   if(options.DirPort) { /* the directory is already here, run startup things */
 677     has_fetched_directory = 1;
 678     directory_has_arrived();
 679   }
 680
 681   if(options.ORPort) {
 682     cpu_init(); /* launch cpuworkers. Need to do this *after* we've read the onion key. */
 683   }
 684
 685   /* start up the necessary connections based on which ports are
 686    * non-zero. This is where we try to connect to all the other ORs,
 687    * and start the listeners.
 688    */
 689   if(retry_all_connections() < 0) {
 690     log_fn(LOG_ERR,"Failed to bind one of the listener ports.");
 691     return -1;
 692   }
 693
 694   for(;;) {
 695 #ifndef MS_WINDOWS /* do signal stuff only on unix */
 696     if(please_dumpstats) {
 697       /* prefer to log it at INFO, but make sure we always see it */
 698       dumpstats(options.loglevel>LOG_INFO ? options.loglevel : LOG_INFO);
 699       please_dumpstats = 0;
 700     }
 701     if(please_reset) {
 702       do_hup();
 703       please_reset = 0;
 704     }
 705     if(please_reap_children) {
 706       while(waitpid(-1,NULL,WNOHANG)) ; /* keep reaping until no more zombies */
 707       please_reap_children = 0;
 708     }
 709 #endif /* signal stuff */
 710
 711     timeout = prepare_for_poll();
 712
 713     /* poll until we have an event, or the second ends */
 714     poll_result = tor_poll(poll_array, nfds, timeout);
 715
 716     /* let catch() handle things like ^c, and otherwise don't worry about it */
 717     if(poll_result < 0) {
 718       /* let the program survive things like ^z */
 719       if(tor_socket_errno(-1) != EINTR) {
 720         log_fn(LOG_ERR,"poll failed: %s [%d]",
 721                tor_socket_strerror(tor_socket_errno(-1)),
 722                tor_socket_errno(-1));
 723         return -1;
 724       } else {
 725         log_fn(LOG_DEBUG,"poll interrupted.");
 726       }
 727     }
 728
 729     /* do all the reads and errors first, so we can detect closed sockets */
 730     for(i=0;i<nfds;i++)
 731       conn_read(i); /* this also marks broken connections */
 732
 733     /* then do the writes */
 734     for(i=0;i<nfds;i++)
 735       conn_write(i);
 736
 737     /* any of the conns need to be closed now? */
 738     for(i=0;i<nfds;i++)
 739       conn_close_if_marked(i);
 740
 741     /* refilling buckets and sending cells happens at the beginning of the
 742      * next iteration of the loop, inside prepare_for_poll()
 743      */
 744   }
 745 }
 746
 747 /** Unix signal handler. */
 748 static void catch(int the_signal) {
 749
 750 #ifndef MS_WINDOWS /* do signal stuff only on unix */
 751   switch(the_signal) {
 752 //    case SIGABRT:
 753     case SIGTERM:
 754     case SIGINT:
 755       log(LOG_ERR,"Catching signal %d, exiting cleanly.", the_signal);
 756       /* we don't care if there was an error when we unlink, nothing
 757          we could do about it anyways */
 758       if(options.PidFile)
 759         unlink(options.PidFile);
 760       exit(0);
 761     case SIGPIPE:
 762       log(LOG_WARN,"Bug: caught sigpipe. Ignoring.");
 763       break;
 764     case SIGHUP:
 765       please_reset = 1;
 766       break;
 767     case SIGUSR1:
 768       please_dumpstats = 1;
 769       break;
 770     case SIGCHLD:
 771       please_reap_children = 1;
 772       break;
 773     default:
 774       log(LOG_WARN,"Caught signal %d that we can't handle??", the_signal);
 775   }
 776 #endif /* signal stuff */
 777 }
 778
 779 /** Write all statistics to the log, with log level 'severity'.  Called
 780  * in response to a SIGUSR1. */
 781 static void dumpstats(int severity) {
 782   int i;
 783   connection_t *conn;
 784   time_t now = time(NULL);
 785
 786   log(severity, "Dumping stats:");
 787
 788   for(i=0;i<nfds;i++) {
 789     conn = connection_array[i];
 790     log(severity, "Conn %d (socket %d) type %d (%s), state %d (%s), created %d secs ago",
 791       i, conn->s, conn->type, CONN_TYPE_TO_STRING(conn->type),
 792       conn->state, conn_state_to_string[conn->type][conn->state], (int)(now - conn->timestamp_created));
 793     if(!connection_is_listener(conn)) {
 794       log(severity,"Conn %d is to '%s:%d'.",i,conn->address, conn->port);
 795       log(severity,"Conn %d: %d bytes waiting on inbuf (last read %d secs ago)",i,
 796              (int)buf_datalen(conn->inbuf),
 797              (int)(now - conn->timestamp_lastread));
 798       log(severity,"Conn %d: %d bytes waiting on outbuf (last written %d secs ago)",i,
 799              (int)buf_datalen(conn->outbuf), (int)(now - conn->timestamp_lastwritten));
 800     }
 801     circuit_dump_by_conn(conn, severity); /* dump info about all the circuits using this conn */
 802   }
 803   log(severity,
 804          "Cells processed: %10lu padding\n"
 805          "                 %10lu create\n"
 806          "                 %10lu created\n"
 807          "                 %10lu relay\n"
 808          "                        (%10lu relayed)\n"
 809          "                        (%10lu delivered)\n"
 810          "                 %10lu destroy",
 811          stats_n_padding_cells_processed,
 812          stats_n_create_cells_processed,
 813          stats_n_created_cells_processed,
 814          stats_n_relay_cells_processed,
 815          stats_n_relay_cells_relayed,
 816          stats_n_relay_cells_delivered,
 817          stats_n_destroy_cells_processed);
 818   if (stats_n_data_cells_packaged)
 819     log(severity,"Average packaged cell fullness: %2.3f%%",
 820            100*(((double)stats_n_data_bytes_packaged) /
 821                 (stats_n_data_cells_packaged*RELAY_PAYLOAD_SIZE)) );
 822   if (stats_n_data_cells_received)
 823     log(severity,"Average delivered cell fullness: %2.3f%%",
 824            100*(((double)stats_n_data_bytes_received) /
 825                 (stats_n_data_cells_received*RELAY_PAYLOAD_SIZE)) );
 826
 827   if (stats_n_seconds_reading)
 828     log(severity,"Average bandwidth used: %d bytes/sec",
 829            (int) (stats_n_bytes_read/stats_n_seconds_reading));
 830
 831   rep_hist_dump_stats(now,severity);
 832   rend_service_dump_stats(severity);
 833 }
 834
 835 /** Called before we make any calls to network-related functions.
 836  * (Some operating systems require their network libraries to be
 837  * initialized.) */
 838 int network_init(void)
 839 {
 840 #ifdef MS_WINDOWS
 841   /* This silly exercise is necessary before windows will allow gethostbyname to work.
 842    */
 843   WSADATA WSAData;
 844   int r;
 845   r = WSAStartup(0x101,&WSAData);
 846   if (r) {
 847     log_fn(LOG_WARN,"Error initializing windows network layer: code was %d",r);
 848     return -1;
 849   }
 850   /* XXXX We should call WSACleanup on exit, I think. */
 851 #endif
 852   return 0;
 853 }
 854
 855 /** Called by exit() as we shut down the process.
 856  */
 857 void exit_function(void)
 858 {
 859 #ifdef MS_WINDOWS
 860   WSACleanup();
 861 #endif
 862 }
 863
 864 /** Main entry point for the Tor command-line client.
 865  */
 866 int tor_main(int argc, char *argv[]) {
 867
 868   /* give it somewhere to log to initially */
 869   add_stream_log(LOG_INFO, "<stdout>", stdout);
 870   log_fn(LOG_NOTICE,"Tor v%s. This is experimental software. Do not use it if you need anonymity.",VERSION);
 871
 872   if (network_init()<0) {
 873     log_fn(LOG_ERR,"Error initializing network; exiting.");
 874     return 1;
 875   }
 876   atexit(exit_function);
 877
 878   if (init_from_config(argc,argv) < 0)
 879     return -1;
 880
 881 #ifndef MS_WINDOWS
 882   if(geteuid()==0)
 883     log_fn(LOG_WARN,"You are running Tor as root. You don't need to, and you probably shouldn't.");
 884 #endif
 885
 886   if(options.ORPort) { /* only spawn dns handlers if we're a router */
 887     dns_init(); /* initialize the dns resolve tree, and spawn workers */
 888   }
 889   if(options.SocksPort) {
 890     client_dns_init(); /* init the client dns cache */
 891   }
 892
 893 #ifndef MS_WINDOWS /* do signal stuff only on unix */
 894 {
 895   struct sigaction action;
 896   action.sa_flags = 0;
 897   sigemptyset(&action.sa_mask);
 898
 899   action.sa_handler = catch;
 900   sigaction(SIGINT,  &action, NULL);
 901   sigaction(SIGTERM, &action, NULL);
 902   sigaction(SIGPIPE, &action, NULL);
 903   sigaction(SIGUSR1, &action, NULL);
 904   sigaction(SIGHUP,  &action, NULL); /* to reload config, retry conns, etc */
 905   sigaction(SIGCHLD, &action, NULL); /* handle dns/cpu workers that exit */
 906 }
 907 #endif /* signal stuff */
 908
 909   crypto_global_init();
 910   crypto_seed_rng();
 911   do_main_loop();
 912   crypto_global_cleanup();
 913   return -1;
 914 }
 915
 916 /*
 917   Local Variables:
 918   mode:c
 919   indent-tabs-mode:nil
 920   c-basic-offset:2
 921   End:
 922 */