Fix compile warning on windows
[tor.git] / src / or / main.c
blob7bae59ce06bdb97edffa4b524b2e779bdfeb0ea2
1 /* Copyright (c) 2001 Matej Pfajfar.
2 * Copyright (c) 2001-2004, Roger Dingledine.
3 * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
4 * Copyright (c) 2007-2011, The Tor Project, Inc. */
5 /* See LICENSE for licensing information */
7 /**
8 * \file main.c
9 * \brief Toplevel module. Handles signals, multiplexes between
10 * connections, implements main loop, and drives scheduled events.
11 **/
13 #define MAIN_PRIVATE
14 #include "or.h"
15 #include "buffers.h"
16 #include "circuitbuild.h"
17 #include "circuitlist.h"
18 #include "circuituse.h"
19 #include "command.h"
20 #include "config.h"
21 #include "connection.h"
22 #include "connection_edge.h"
23 #include "connection_or.h"
24 #include "control.h"
25 #include "cpuworker.h"
26 #include "directory.h"
27 #include "dirserv.h"
28 #include "dirvote.h"
29 #include "dns.h"
30 #include "dnsserv.h"
31 #include "geoip.h"
32 #include "hibernate.h"
33 #include "main.h"
34 #include "microdesc.h"
35 #include "networkstatus.h"
36 #include "ntmain.h"
37 #include "onion.h"
38 #include "policies.h"
39 #include "relay.h"
40 #include "rendclient.h"
41 #include "rendcommon.h"
42 #include "rendservice.h"
43 #include "rephist.h"
44 #include "router.h"
45 #include "routerlist.h"
46 #include "routerparse.h"
47 #ifdef USE_DMALLOC
48 #include <dmalloc.h>
49 #include <openssl/crypto.h>
50 #endif
51 #include "memarea.h"
53 #ifdef HAVE_EVENT2_EVENT_H
54 #include <event2/event.h>
55 #else
56 #include <event.h>
57 #endif
59 void evdns_shutdown(int);
61 /********* PROTOTYPES **********/
63 static void dumpmemusage(int severity);
64 static void dumpstats(int severity); /* log stats */
65 static void conn_read_callback(int fd, short event, void *_conn);
66 static void conn_write_callback(int fd, short event, void *_conn);
67 static void second_elapsed_callback(periodic_timer_t *timer, void *args);
68 static int conn_close_if_marked(int i);
69 static void connection_start_reading_from_linked_conn(connection_t *conn);
70 static int connection_should_read_from_linked_conn(connection_t *conn);
72 /********* START VARIABLES **********/
74 int global_read_bucket; /**< Max number of bytes I can read this second. */
75 int global_write_bucket; /**< Max number of bytes I can write this second. */
77 /** Max number of relayed (bandwidth class 1) bytes I can read this second. */
78 int global_relayed_read_bucket;
79 /** Max number of relayed (bandwidth class 1) bytes I can write this second. */
80 int global_relayed_write_bucket;
82 /** What was the read bucket before the last second_elapsed_callback() call?
83 * (used to determine how many bytes we've read). */
84 static int stats_prev_global_read_bucket;
85 /** What was the write bucket before the last second_elapsed_callback() call?
86 * (used to determine how many bytes we've written). */
87 static int stats_prev_global_write_bucket;
88 /* XXX we might want to keep stats about global_relayed_*_bucket too. Or not.*/
89 /** How many bytes have we read since we started the process? */
90 static uint64_t stats_n_bytes_read = 0;
91 /** How many bytes have we written since we started the process? */
92 static uint64_t stats_n_bytes_written = 0;
93 /** What time did this process start up? */
94 time_t time_of_process_start = 0;
95 /** How many seconds have we been running? */
96 long stats_n_seconds_working = 0;
97 /** When do we next launch DNS wildcarding checks? */
98 static time_t time_to_check_for_correct_dns = 0;
100 /** How often will we honor SIGNEWNYM requests? */
101 #define MAX_SIGNEWNYM_RATE 10
102 /** When did we last process a SIGNEWNYM request? */
103 static time_t time_of_last_signewnym = 0;
104 /** Is there a signewnym request we're currently waiting to handle? */
105 static int signewnym_is_pending = 0;
107 /** Smartlist of all open connections. */
108 static smartlist_t *connection_array = NULL;
109 /** List of connections that have been marked for close and need to be freed
110 * and removed from connection_array. */
111 static smartlist_t *closeable_connection_lst = NULL;
112 /** List of linked connections that are currently reading data into their
113 * inbuf from their partner's outbuf. */
114 static smartlist_t *active_linked_connection_lst = NULL;
115 /** Flag: Set to true iff we entered the current libevent main loop via
116 * <b>loop_once</b>. If so, there's no need to trigger a loopexit in order
117 * to handle linked connections. */
118 static int called_loop_once = 0;
120 /** We set this to 1 when we've opened a circuit, so we can print a log
121 * entry to inform the user that Tor is working. We set it to 0 when
122 * we think the fact that we once opened a circuit doesn't mean we can do so
123 * any longer (a big time jump happened, when we notice our directory is
124 * heinously out-of-date, etc.
126 int can_complete_circuit=0;
128 /** How often do we check for router descriptors that we should download
129 * when we have too little directory info? */
130 #define GREEDY_DESCRIPTOR_RETRY_INTERVAL (10)
131 /** How often do we check for router descriptors that we should download
132 * when we have enough directory info? */
133 #define LAZY_DESCRIPTOR_RETRY_INTERVAL (60)
134 /** How often do we 'forgive' undownloadable router descriptors and attempt
135 * to download them again? */
136 #define DESCRIPTOR_FAILURE_RESET_INTERVAL (60*60)
137 /** How long do we let a directory connection stall before expiring it? */
138 #define DIR_CONN_MAX_STALL (5*60)
140 /** How long do we let OR connections handshake before we decide that
141 * they are obsolete? */
142 #define TLS_HANDSHAKE_TIMEOUT (60)
144 /********* END VARIABLES ************/
146 /****************************************************************************
148 * This section contains accessors and other methods on the connection_array
149 * variables (which are global within this file and unavailable outside it).
151 ****************************************************************************/
153 /** Add <b>conn</b> to the array of connections that we can poll on. The
154 * connection's socket must be set; the connection starts out
155 * non-reading and non-writing.
158 connection_add(connection_t *conn)
160 tor_assert(conn);
161 tor_assert(conn->s >= 0 ||
162 conn->linked ||
163 (conn->type == CONN_TYPE_AP &&
164 TO_EDGE_CONN(conn)->is_dns_request));
166 tor_assert(conn->conn_array_index == -1); /* can only connection_add once */
167 conn->conn_array_index = smartlist_len(connection_array);
168 smartlist_add(connection_array, conn);
170 if (conn->s >= 0 || conn->linked) {
171 conn->read_event = tor_event_new(tor_libevent_get_base(),
172 conn->s, EV_READ|EV_PERSIST, conn_read_callback, conn);
173 conn->write_event = tor_event_new(tor_libevent_get_base(),
174 conn->s, EV_WRITE|EV_PERSIST, conn_write_callback, conn);
177 log_debug(LD_NET,"new conn type %s, socket %d, address %s, n_conns %d.",
178 conn_type_to_string(conn->type), conn->s, conn->address,
179 smartlist_len(connection_array));
181 return 0;
184 /** Tell libevent that we don't care about <b>conn</b> any more. */
185 void
186 connection_unregister_events(connection_t *conn)
188 if (conn->read_event) {
189 if (event_del(conn->read_event))
190 log_warn(LD_BUG, "Error removing read event for %d", conn->s);
191 tor_free(conn->read_event);
193 if (conn->write_event) {
194 if (event_del(conn->write_event))
195 log_warn(LD_BUG, "Error removing write event for %d", conn->s);
196 tor_free(conn->write_event);
198 if (conn->dns_server_port) {
199 dnsserv_close_listener(conn);
203 /** Remove the connection from the global list, and remove the
204 * corresponding poll entry. Calling this function will shift the last
205 * connection (if any) into the position occupied by conn.
208 connection_remove(connection_t *conn)
210 int current_index;
211 connection_t *tmp;
213 tor_assert(conn);
215 log_debug(LD_NET,"removing socket %d (type %s), n_conns now %d",
216 conn->s, conn_type_to_string(conn->type),
217 smartlist_len(connection_array));
219 tor_assert(conn->conn_array_index >= 0);
220 current_index = conn->conn_array_index;
221 connection_unregister_events(conn); /* This is redundant, but cheap. */
222 if (current_index == smartlist_len(connection_array)-1) { /* at the end */
223 smartlist_del(connection_array, current_index);
224 return 0;
227 /* replace this one with the one at the end */
228 smartlist_del(connection_array, current_index);
229 tmp = smartlist_get(connection_array, current_index);
230 tmp->conn_array_index = current_index;
232 return 0;
235 /** If <b>conn</b> is an edge conn, remove it from the list
236 * of conn's on this circuit. If it's not on an edge,
237 * flush and send destroys for all circuits on this conn.
239 * Remove it from connection_array (if applicable) and
240 * from closeable_connection_list.
242 * Then free it.
244 static void
245 connection_unlink(connection_t *conn)
247 connection_about_to_close_connection(conn);
248 if (conn->conn_array_index >= 0) {
249 connection_remove(conn);
251 if (conn->linked_conn) {
252 conn->linked_conn->linked_conn = NULL;
253 if (! conn->linked_conn->marked_for_close &&
254 conn->linked_conn->reading_from_linked_conn)
255 connection_start_reading(conn->linked_conn);
256 conn->linked_conn = NULL;
258 smartlist_remove(closeable_connection_lst, conn);
259 smartlist_remove(active_linked_connection_lst, conn);
260 if (conn->type == CONN_TYPE_EXIT) {
261 assert_connection_edge_not_dns_pending(TO_EDGE_CONN(conn));
263 if (conn->type == CONN_TYPE_OR) {
264 if (!tor_digest_is_zero(TO_OR_CONN(conn)->identity_digest))
265 connection_or_remove_from_identity_map(TO_OR_CONN(conn));
267 connection_free(conn);
270 /** Schedule <b>conn</b> to be closed. **/
271 void
272 add_connection_to_closeable_list(connection_t *conn)
274 tor_assert(!smartlist_isin(closeable_connection_lst, conn));
275 tor_assert(conn->marked_for_close);
276 assert_connection_ok(conn, time(NULL));
277 smartlist_add(closeable_connection_lst, conn);
280 /** Return 1 if conn is on the closeable list, else return 0. */
282 connection_is_on_closeable_list(connection_t *conn)
284 return smartlist_isin(closeable_connection_lst, conn);
287 /** Return true iff conn is in the current poll array. */
289 connection_in_array(connection_t *conn)
291 return smartlist_isin(connection_array, conn);
294 /** Set <b>*array</b> to an array of all connections, and <b>*n</b>
295 * to the length of the array. <b>*array</b> and <b>*n</b> must not
296 * be modified.
298 smartlist_t *
299 get_connection_array(void)
301 if (!connection_array)
302 connection_array = smartlist_create();
303 return connection_array;
306 /** Set the event mask on <b>conn</b> to <b>events</b>. (The event
307 * mask is a bitmask whose bits are READ_EVENT and WRITE_EVENT)
309 void
310 connection_watch_events(connection_t *conn, watchable_events_t events)
312 if (events & READ_EVENT)
313 connection_start_reading(conn);
314 else
315 connection_stop_reading(conn);
317 if (events & WRITE_EVENT)
318 connection_start_writing(conn);
319 else
320 connection_stop_writing(conn);
323 /** Return true iff <b>conn</b> is listening for read events. */
325 connection_is_reading(connection_t *conn)
327 tor_assert(conn);
329 return conn->reading_from_linked_conn ||
330 (conn->read_event && event_pending(conn->read_event, EV_READ, NULL));
333 /** Tell the main loop to stop notifying <b>conn</b> of any read events. */
334 void
335 connection_stop_reading(connection_t *conn)
337 tor_assert(conn);
338 tor_assert(conn->read_event);
340 if (conn->linked) {
341 conn->reading_from_linked_conn = 0;
342 connection_stop_reading_from_linked_conn(conn);
343 } else {
344 if (event_del(conn->read_event))
345 log_warn(LD_NET, "Error from libevent setting read event state for %d "
346 "to unwatched: %s",
347 conn->s,
348 tor_socket_strerror(tor_socket_errno(conn->s)));
352 /** Tell the main loop to start notifying <b>conn</b> of any read events. */
353 void
354 connection_start_reading(connection_t *conn)
356 tor_assert(conn);
357 tor_assert(conn->read_event);
359 if (conn->linked) {
360 conn->reading_from_linked_conn = 1;
361 if (connection_should_read_from_linked_conn(conn))
362 connection_start_reading_from_linked_conn(conn);
363 } else {
364 if (event_add(conn->read_event, NULL))
365 log_warn(LD_NET, "Error from libevent setting read event state for %d "
366 "to watched: %s",
367 conn->s,
368 tor_socket_strerror(tor_socket_errno(conn->s)));
372 /** Return true iff <b>conn</b> is listening for write events. */
374 connection_is_writing(connection_t *conn)
376 tor_assert(conn);
378 return conn->writing_to_linked_conn ||
379 (conn->write_event && event_pending(conn->write_event, EV_WRITE, NULL));
382 /** Tell the main loop to stop notifying <b>conn</b> of any write events. */
383 void
384 connection_stop_writing(connection_t *conn)
386 tor_assert(conn);
387 tor_assert(conn->write_event);
389 if (conn->linked) {
390 conn->writing_to_linked_conn = 0;
391 if (conn->linked_conn)
392 connection_stop_reading_from_linked_conn(conn->linked_conn);
393 } else {
394 if (event_del(conn->write_event))
395 log_warn(LD_NET, "Error from libevent setting write event state for %d "
396 "to unwatched: %s",
397 conn->s,
398 tor_socket_strerror(tor_socket_errno(conn->s)));
402 /** Tell the main loop to start notifying <b>conn</b> of any write events. */
403 void
404 connection_start_writing(connection_t *conn)
406 tor_assert(conn);
407 tor_assert(conn->write_event);
409 if (conn->linked) {
410 conn->writing_to_linked_conn = 1;
411 if (conn->linked_conn &&
412 connection_should_read_from_linked_conn(conn->linked_conn))
413 connection_start_reading_from_linked_conn(conn->linked_conn);
414 } else {
415 if (event_add(conn->write_event, NULL))
416 log_warn(LD_NET, "Error from libevent setting write event state for %d "
417 "to watched: %s",
418 conn->s,
419 tor_socket_strerror(tor_socket_errno(conn->s)));
423 /** Return true iff <b>conn</b> is linked conn, and reading from the conn
424 * linked to it would be good and feasible. (Reading is "feasible" if the
425 * other conn exists and has data in its outbuf, and is "good" if we have our
426 * reading_from_linked_conn flag set and the other conn has its
427 * writing_to_linked_conn flag set.)*/
428 static int
429 connection_should_read_from_linked_conn(connection_t *conn)
431 if (conn->linked && conn->reading_from_linked_conn) {
432 if (! conn->linked_conn ||
433 (conn->linked_conn->writing_to_linked_conn &&
434 buf_datalen(conn->linked_conn->outbuf)))
435 return 1;
437 return 0;
440 /** Helper: Tell the main loop to begin reading bytes into <b>conn</b> from
441 * its linked connection, if it is not doing so already. Called by
442 * connection_start_reading and connection_start_writing as appropriate. */
443 static void
444 connection_start_reading_from_linked_conn(connection_t *conn)
446 tor_assert(conn);
447 tor_assert(conn->linked == 1);
449 if (!conn->active_on_link) {
450 conn->active_on_link = 1;
451 smartlist_add(active_linked_connection_lst, conn);
452 if (!called_loop_once) {
453 /* This is the first event on the list; we won't be in LOOP_ONCE mode,
454 * so we need to make sure that the event_base_loop() actually exits at
455 * the end of its run through the current connections and lets us
456 * activate read events for linked connections. */
457 struct timeval tv = { 0, 0 };
458 tor_event_base_loopexit(tor_libevent_get_base(), &tv);
460 } else {
461 tor_assert(smartlist_isin(active_linked_connection_lst, conn));
465 /** Tell the main loop to stop reading bytes into <b>conn</b> from its linked
466 * connection, if is currently doing so. Called by connection_stop_reading,
467 * connection_stop_writing, and connection_read. */
468 void
469 connection_stop_reading_from_linked_conn(connection_t *conn)
471 tor_assert(conn);
472 tor_assert(conn->linked == 1);
474 if (conn->active_on_link) {
475 conn->active_on_link = 0;
476 /* FFFF We could keep an index here so we can smartlist_del
477 * cleanly. On the other hand, this doesn't show up on profiles,
478 * so let's leave it alone for now. */
479 smartlist_remove(active_linked_connection_lst, conn);
480 } else {
481 tor_assert(!smartlist_isin(active_linked_connection_lst, conn));
485 /** Close all connections that have been scheduled to get closed. */
486 static void
487 close_closeable_connections(void)
489 int i;
490 for (i = 0; i < smartlist_len(closeable_connection_lst); ) {
491 connection_t *conn = smartlist_get(closeable_connection_lst, i);
492 if (conn->conn_array_index < 0) {
493 connection_unlink(conn); /* blow it away right now */
494 } else {
495 if (!conn_close_if_marked(conn->conn_array_index))
496 ++i;
501 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
502 * some data to read. */
503 static void
504 conn_read_callback(int fd, short event, void *_conn)
506 connection_t *conn = _conn;
507 (void)fd;
508 (void)event;
510 log_debug(LD_NET,"socket %d wants to read.",conn->s);
512 /* assert_connection_ok(conn, time(NULL)); */
514 if (connection_handle_read(conn) < 0) {
515 if (!conn->marked_for_close) {
516 #ifndef MS_WINDOWS
517 log_warn(LD_BUG,"Unhandled error on read for %s connection "
518 "(fd %d); removing",
519 conn_type_to_string(conn->type), conn->s);
520 tor_fragile_assert();
521 #endif
522 if (CONN_IS_EDGE(conn))
523 connection_edge_end_errno(TO_EDGE_CONN(conn));
524 connection_mark_for_close(conn);
527 assert_connection_ok(conn, time(NULL));
529 if (smartlist_len(closeable_connection_lst))
530 close_closeable_connections();
533 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
534 * some data to write. */
535 static void
536 conn_write_callback(int fd, short events, void *_conn)
538 connection_t *conn = _conn;
539 (void)fd;
540 (void)events;
542 LOG_FN_CONN(conn, (LOG_DEBUG, LD_NET, "socket %d wants to write.",conn->s));
544 /* assert_connection_ok(conn, time(NULL)); */
546 if (connection_handle_write(conn, 0) < 0) {
547 if (!conn->marked_for_close) {
548 /* this connection is broken. remove it. */
549 log_fn(LOG_WARN,LD_BUG,
550 "unhandled error on write for %s connection (fd %d); removing",
551 conn_type_to_string(conn->type), conn->s);
552 tor_fragile_assert();
553 if (CONN_IS_EDGE(conn)) {
554 /* otherwise we cry wolf about duplicate close */
555 edge_connection_t *edge_conn = TO_EDGE_CONN(conn);
556 if (!edge_conn->end_reason)
557 edge_conn->end_reason = END_STREAM_REASON_INTERNAL;
558 edge_conn->edge_has_sent_end = 1;
560 connection_close_immediate(conn); /* So we don't try to flush. */
561 connection_mark_for_close(conn);
564 assert_connection_ok(conn, time(NULL));
566 if (smartlist_len(closeable_connection_lst))
567 close_closeable_connections();
570 /** If the connection at connection_array[i] is marked for close, then:
571 * - If it has data that it wants to flush, try to flush it.
572 * - If it _still_ has data to flush, and conn->hold_open_until_flushed is
573 * true, then leave the connection open and return.
574 * - Otherwise, remove the connection from connection_array and from
575 * all other lists, close it, and free it.
576 * Returns 1 if the connection was closed, 0 otherwise.
578 static int
579 conn_close_if_marked(int i)
581 connection_t *conn;
582 int retval;
583 time_t now;
585 conn = smartlist_get(connection_array, i);
586 if (!conn->marked_for_close)
587 return 0; /* nothing to see here, move along */
588 now = time(NULL);
589 assert_connection_ok(conn, now);
590 /* assert_all_pending_dns_resolves_ok(); */
592 log_debug(LD_NET,"Cleaning up connection (fd %d).",conn->s);
593 if ((conn->s >= 0 || conn->linked_conn) && connection_wants_to_flush(conn)) {
594 /* s == -1 means it's an incomplete edge connection, or that the socket
595 * has already been closed as unflushable. */
596 ssize_t sz = connection_bucket_write_limit(conn, now);
597 if (!conn->hold_open_until_flushed)
598 log_info(LD_NET,
599 "Conn (addr %s, fd %d, type %s, state %d) marked, but wants "
600 "to flush %d bytes. (Marked at %s:%d)",
601 escaped_safe_str_client(conn->address),
602 conn->s, conn_type_to_string(conn->type), conn->state,
603 (int)conn->outbuf_flushlen,
604 conn->marked_for_close_file, conn->marked_for_close);
605 if (conn->linked_conn) {
606 retval = move_buf_to_buf(conn->linked_conn->inbuf, conn->outbuf,
607 &conn->outbuf_flushlen);
608 if (retval >= 0) {
609 /* The linked conn will notice that it has data when it notices that
610 * we're gone. */
611 connection_start_reading_from_linked_conn(conn->linked_conn);
613 log_debug(LD_GENERAL, "Flushed last %d bytes from a linked conn; "
614 "%d left; flushlen %d; wants-to-flush==%d", retval,
615 (int)buf_datalen(conn->outbuf),
616 (int)conn->outbuf_flushlen,
617 connection_wants_to_flush(conn));
618 } else if (connection_speaks_cells(conn)) {
619 if (conn->state == OR_CONN_STATE_OPEN) {
620 retval = flush_buf_tls(TO_OR_CONN(conn)->tls, conn->outbuf, sz,
621 &conn->outbuf_flushlen);
622 } else
623 retval = -1; /* never flush non-open broken tls connections */
624 } else {
625 retval = flush_buf(conn->s, conn->outbuf, sz, &conn->outbuf_flushlen);
627 if (retval >= 0 && /* Technically, we could survive things like
628 TLS_WANT_WRITE here. But don't bother for now. */
629 conn->hold_open_until_flushed && connection_wants_to_flush(conn)) {
630 if (retval > 0) {
631 LOG_FN_CONN(conn, (LOG_INFO,LD_NET,
632 "Holding conn (fd %d) open for more flushing.",
633 conn->s));
634 conn->timestamp_lastwritten = now; /* reset so we can flush more */
636 return 0;
638 if (connection_wants_to_flush(conn)) {
639 int severity;
640 if (conn->type == CONN_TYPE_EXIT ||
641 (conn->type == CONN_TYPE_OR && server_mode(get_options())) ||
642 (conn->type == CONN_TYPE_DIR && conn->purpose == DIR_PURPOSE_SERVER))
643 severity = LOG_INFO;
644 else
645 severity = LOG_NOTICE;
646 /* XXXX Maybe allow this to happen a certain amount per hour; it usually
647 * is meaningless. */
648 log_fn(severity, LD_NET, "We stalled too much while trying to write %d "
649 "bytes to address %s. If this happens a lot, either "
650 "something is wrong with your network connection, or "
651 "something is wrong with theirs. "
652 "(fd %d, type %s, state %d, marked at %s:%d).",
653 (int)buf_datalen(conn->outbuf),
654 escaped_safe_str_client(conn->address),
655 conn->s, conn_type_to_string(conn->type), conn->state,
656 conn->marked_for_close_file,
657 conn->marked_for_close);
660 connection_unlink(conn); /* unlink, remove, free */
661 return 1;
664 /** We've just tried every dirserver we know about, and none of
665 * them were reachable. Assume the network is down. Change state
666 * so next time an application connection arrives we'll delay it
667 * and try another directory fetch. Kill off all the circuit_wait
668 * streams that are waiting now, since they will all timeout anyway.
670 void
671 directory_all_unreachable(time_t now)
673 connection_t *conn;
674 (void)now;
676 stats_n_seconds_working=0; /* reset it */
678 while ((conn = connection_get_by_type_state(CONN_TYPE_AP,
679 AP_CONN_STATE_CIRCUIT_WAIT))) {
680 edge_connection_t *edge_conn = TO_EDGE_CONN(conn);
681 log_notice(LD_NET,
682 "Is your network connection down? "
683 "Failing connection to '%s:%d'.",
684 safe_str_client(edge_conn->socks_request->address),
685 edge_conn->socks_request->port);
686 connection_mark_unattached_ap(edge_conn,
687 END_STREAM_REASON_NET_UNREACHABLE);
689 control_event_general_status(LOG_ERR, "DIR_ALL_UNREACHABLE");
692 /** This function is called whenever we successfully pull down some new
693 * network statuses or server descriptors. */
694 void
695 directory_info_has_arrived(time_t now, int from_cache)
697 or_options_t *options = get_options();
699 if (!router_have_minimum_dir_info()) {
700 int quiet = directory_too_idle_to_fetch_descriptors(options, now);
701 log(quiet ? LOG_INFO : LOG_NOTICE, LD_DIR,
702 "I learned some more directory information, but not enough to "
703 "build a circuit: %s", get_dir_info_status_string());
704 update_router_descriptor_downloads(now);
705 return;
706 } else {
707 if (directory_fetches_from_authorities(options))
708 update_router_descriptor_downloads(now);
710 /* if we have enough dir info, then update our guard status with
711 * whatever we just learned. */
712 entry_guards_compute_status(options, now);
713 /* Don't even bother trying to get extrainfo until the rest of our
714 * directory info is up-to-date */
715 if (options->DownloadExtraInfo)
716 update_extrainfo_downloads(now);
719 if (server_mode(options) && !we_are_hibernating() && !from_cache &&
720 (can_complete_circuit || !any_predicted_circuits(now)))
721 consider_testing_reachability(1, 1);
724 /** How long do we wait before killing OR connections with no circuits?
725 * In Tor versions up to 0.2.1.25 and 0.2.2.12-alpha, we waited 15 minutes
726 * before cancelling these connections, which caused fast relays to accrue
727 * many many idle connections. Hopefully 3 minutes is low enough that
728 * it kills most idle connections, without being so low that we cause
729 * clients to bounce on and off.
731 #define IDLE_OR_CONN_TIMEOUT 180
733 /** Perform regular maintenance tasks for a single connection. This
734 * function gets run once per second per connection by run_scheduled_events.
736 static void
737 run_connection_housekeeping(int i, time_t now)
739 cell_t cell;
740 connection_t *conn = smartlist_get(connection_array, i);
741 or_options_t *options = get_options();
742 or_connection_t *or_conn;
743 int past_keepalive =
744 now >= conn->timestamp_lastwritten + options->KeepalivePeriod;
746 if (conn->outbuf && !buf_datalen(conn->outbuf) && conn->type == CONN_TYPE_OR)
747 TO_OR_CONN(conn)->timestamp_lastempty = now;
749 if (conn->marked_for_close) {
750 /* nothing to do here */
751 return;
754 /* Expire any directory connections that haven't been active (sent
755 * if a server or received if a client) for 5 min */
756 if (conn->type == CONN_TYPE_DIR &&
757 ((DIR_CONN_IS_SERVER(conn) &&
758 conn->timestamp_lastwritten + DIR_CONN_MAX_STALL < now) ||
759 (!DIR_CONN_IS_SERVER(conn) &&
760 conn->timestamp_lastread + DIR_CONN_MAX_STALL < now))) {
761 log_info(LD_DIR,"Expiring wedged directory conn (fd %d, purpose %d)",
762 conn->s, conn->purpose);
763 /* This check is temporary; it's to let us know whether we should consider
764 * parsing partial serverdesc responses. */
765 if (conn->purpose == DIR_PURPOSE_FETCH_SERVERDESC &&
766 buf_datalen(conn->inbuf)>=1024) {
767 log_info(LD_DIR,"Trying to extract information from wedged server desc "
768 "download.");
769 connection_dir_reached_eof(TO_DIR_CONN(conn));
770 } else {
771 connection_mark_for_close(conn);
773 return;
776 if (!connection_speaks_cells(conn))
777 return; /* we're all done here, the rest is just for OR conns */
779 /* If we haven't written to an OR connection for a while, then either nuke
780 the connection or send a keepalive, depending. */
782 or_conn = TO_OR_CONN(conn);
783 tor_assert(conn->outbuf);
785 if (or_conn->is_bad_for_new_circs && !or_conn->n_circuits) {
786 /* It's bad for new circuits, and has no unmarked circuits on it:
787 * mark it now. */
788 log_info(LD_OR,
789 "Expiring non-used OR connection to fd %d (%s:%d) [Too old].",
790 conn->s, conn->address, conn->port);
791 if (conn->state == OR_CONN_STATE_CONNECTING)
792 connection_or_connect_failed(TO_OR_CONN(conn),
793 END_OR_CONN_REASON_TIMEOUT,
794 "Tor gave up on the connection");
795 connection_mark_for_close(conn);
796 conn->hold_open_until_flushed = 1;
797 } else if (!connection_state_is_open(conn)) {
798 if (past_keepalive) {
799 /* We never managed to actually get this connection open and happy. */
800 log_info(LD_OR,"Expiring non-open OR connection to fd %d (%s:%d).",
801 conn->s,conn->address, conn->port);
802 connection_mark_for_close(conn);
804 } else if (we_are_hibernating() && !or_conn->n_circuits &&
805 !buf_datalen(conn->outbuf)) {
806 /* We're hibernating, there's no circuits, and nothing to flush.*/
807 log_info(LD_OR,"Expiring non-used OR connection to fd %d (%s:%d) "
808 "[Hibernating or exiting].",
809 conn->s,conn->address, conn->port);
810 connection_mark_for_close(conn);
811 conn->hold_open_until_flushed = 1;
812 } else if (!or_conn->n_circuits &&
813 now >= or_conn->timestamp_last_added_nonpadding +
814 IDLE_OR_CONN_TIMEOUT) {
815 log_info(LD_OR,"Expiring non-used OR connection to fd %d (%s:%d) "
816 "[idle %d].", conn->s,conn->address, conn->port,
817 (int)(now - or_conn->timestamp_last_added_nonpadding));
818 connection_mark_for_close(conn);
819 conn->hold_open_until_flushed = 1;
820 } else if (
821 now >= or_conn->timestamp_lastempty + options->KeepalivePeriod*10 &&
822 now >= conn->timestamp_lastwritten + options->KeepalivePeriod*10) {
823 log_fn(LOG_PROTOCOL_WARN,LD_PROTOCOL,
824 "Expiring stuck OR connection to fd %d (%s:%d). (%d bytes to "
825 "flush; %d seconds since last write)",
826 conn->s, conn->address, conn->port,
827 (int)buf_datalen(conn->outbuf),
828 (int)(now-conn->timestamp_lastwritten));
829 connection_mark_for_close(conn);
830 } else if (past_keepalive && !buf_datalen(conn->outbuf)) {
831 /* send a padding cell */
832 log_fn(LOG_DEBUG,LD_OR,"Sending keepalive to (%s:%d)",
833 conn->address, conn->port);
834 memset(&cell,0,sizeof(cell_t));
835 cell.command = CELL_PADDING;
836 connection_or_write_cell_to_buf(&cell, or_conn);
840 /** Honor a NEWNYM request: make future requests unlinkable to past
841 * requests. */
842 static void
843 signewnym_impl(time_t now)
845 or_options_t *options = get_options();
846 if (!proxy_mode(options)) {
847 log_info(LD_CONTROL, "Ignoring SIGNAL NEWNYM because client functionality "
848 "is disabled.");
849 return;
852 circuit_expire_all_dirty_circs();
853 addressmap_clear_transient();
854 rend_cache_purge();
855 rend_client_cancel_descriptor_fetches();
856 time_of_last_signewnym = now;
857 signewnym_is_pending = 0;
860 /** Perform regular maintenance tasks. This function gets run once per
861 * second by second_elapsed_callback().
863 static void
864 run_scheduled_events(time_t now)
866 static time_t last_rotated_x509_certificate = 0;
867 static time_t time_to_check_v3_certificate = 0;
868 static time_t time_to_check_listeners = 0;
869 static time_t time_to_check_descriptor = 0;
870 static time_t time_to_check_ipaddress = 0;
871 static time_t time_to_shrink_memory = 0;
872 static time_t time_to_try_getting_descriptors = 0;
873 static time_t time_to_reset_descriptor_failures = 0;
874 static time_t time_to_add_entropy = 0;
875 static time_t time_to_write_bridge_status_file = 0;
876 static time_t time_to_downrate_stability = 0;
877 static time_t time_to_save_stability = 0;
878 static time_t time_to_clean_caches = 0;
879 static time_t time_to_recheck_bandwidth = 0;
880 static time_t time_to_check_for_expired_networkstatus = 0;
881 static time_t time_to_write_stats_files = 0;
882 static time_t time_to_write_bridge_stats = 0;
883 static time_t time_to_launch_reachability_tests = 0;
884 static int should_init_bridge_stats = 1;
885 static time_t time_to_retry_dns_init = 0;
886 or_options_t *options = get_options();
887 int is_server = server_mode(options);
888 int i;
889 int have_dir_info;
891 /** 0. See if we've been asked to shut down and our timeout has
892 * expired; or if our bandwidth limits are exhausted and we
893 * should hibernate; or if it's time to wake up from hibernation.
895 consider_hibernation(now);
897 /* 0b. If we've deferred a signewnym, make sure it gets handled
898 * eventually. */
899 if (signewnym_is_pending &&
900 time_of_last_signewnym + MAX_SIGNEWNYM_RATE <= now) {
901 log(LOG_INFO, LD_CONTROL, "Honoring delayed NEWNYM request");
902 signewnym_impl(now);
905 /* 0c. If we've deferred log messages for the controller, handle them now */
906 flush_pending_log_callbacks();
908 /** 1a. Every MIN_ONION_KEY_LIFETIME seconds, rotate the onion keys,
909 * shut down and restart all cpuworkers, and update the directory if
910 * necessary.
912 if (is_server &&
913 get_onion_key_set_at()+MIN_ONION_KEY_LIFETIME < now) {
914 log_info(LD_GENERAL,"Rotating onion key.");
915 rotate_onion_key();
916 cpuworkers_rotate();
917 if (router_rebuild_descriptor(1)<0) {
918 log_info(LD_CONFIG, "Couldn't rebuild router descriptor");
920 if (advertised_server_mode())
921 router_upload_dir_desc_to_dirservers(0);
924 if (time_to_try_getting_descriptors < now) {
925 update_router_descriptor_downloads(now);
926 update_extrainfo_downloads(now);
927 if (options->UseBridges)
928 fetch_bridge_descriptors(options, now);
929 if (router_have_minimum_dir_info())
930 time_to_try_getting_descriptors = now + LAZY_DESCRIPTOR_RETRY_INTERVAL;
931 else
932 time_to_try_getting_descriptors = now + GREEDY_DESCRIPTOR_RETRY_INTERVAL;
935 if (time_to_reset_descriptor_failures < now) {
936 router_reset_descriptor_download_failures();
937 time_to_reset_descriptor_failures =
938 now + DESCRIPTOR_FAILURE_RESET_INTERVAL;
941 /** 1b. Every MAX_SSL_KEY_LIFETIME seconds, we change our TLS context. */
942 if (!last_rotated_x509_certificate)
943 last_rotated_x509_certificate = now;
944 if (last_rotated_x509_certificate+MAX_SSL_KEY_LIFETIME < now) {
945 log_info(LD_GENERAL,"Rotating tls context.");
946 if (tor_tls_context_init(public_server_mode(options),
947 get_tlsclient_identity_key(),
948 is_server ? get_server_identity_key() : NULL,
949 MAX_SSL_KEY_LIFETIME) < 0) {
950 log_warn(LD_BUG, "Error reinitializing TLS context");
951 /* XXX is it a bug here, that we just keep going? -RD */
953 last_rotated_x509_certificate = now;
954 /* We also make sure to rotate the TLS connections themselves if they've
955 * been up for too long -- but that's done via is_bad_for_new_circs in
956 * connection_run_housekeeping() above. */
959 if (time_to_add_entropy < now) {
960 if (time_to_add_entropy) {
961 /* We already seeded once, so don't die on failure. */
962 crypto_seed_rng(0);
964 /** How often do we add more entropy to OpenSSL's RNG pool? */
965 #define ENTROPY_INTERVAL (60*60)
966 time_to_add_entropy = now + ENTROPY_INTERVAL;
969 /** 1c. If we have to change the accounting interval or record
970 * bandwidth used in this accounting interval, do so. */
971 if (accounting_is_enabled(options))
972 accounting_run_housekeeping(now);
974 if (time_to_launch_reachability_tests < now &&
975 (authdir_mode_tests_reachability(options)) &&
976 !we_are_hibernating()) {
977 time_to_launch_reachability_tests = now + REACHABILITY_TEST_INTERVAL;
978 /* try to determine reachability of the other Tor relays */
979 dirserv_test_reachability(now);
982 /** 1d. Periodically, we discount older stability information so that new
983 * stability info counts more, and save the stability information to disk as
984 * appropriate. */
985 if (time_to_downrate_stability < now)
986 time_to_downrate_stability = rep_hist_downrate_old_runs(now);
987 if (authdir_mode_tests_reachability(options)) {
988 if (time_to_save_stability < now) {
989 if (time_to_save_stability && rep_hist_record_mtbf_data(now, 1)<0) {
990 log_warn(LD_GENERAL, "Couldn't store mtbf data.");
992 #define SAVE_STABILITY_INTERVAL (30*60)
993 time_to_save_stability = now + SAVE_STABILITY_INTERVAL;
997 /* 1e. Periodically, if we're a v3 authority, we check whether our cert is
998 * close to expiring and warn the admin if it is. */
999 if (time_to_check_v3_certificate < now) {
1000 v3_authority_check_key_expiry();
1001 #define CHECK_V3_CERTIFICATE_INTERVAL (5*60)
1002 time_to_check_v3_certificate = now + CHECK_V3_CERTIFICATE_INTERVAL;
1005 /* 1f. Check whether our networkstatus has expired.
1007 if (time_to_check_for_expired_networkstatus < now) {
1008 networkstatus_t *ns = networkstatus_get_latest_consensus();
1009 /*XXXX RD: This value needs to be the same as REASONABLY_LIVE_TIME in
1010 * networkstatus_get_reasonably_live_consensus(), but that value is way
1011 * way too high. Arma: is the bridge issue there resolved yet? -NM */
1012 #define NS_EXPIRY_SLOP (24*60*60)
1013 if (ns && ns->valid_until < now+NS_EXPIRY_SLOP &&
1014 router_have_minimum_dir_info()) {
1015 router_dir_info_changed();
1017 #define CHECK_EXPIRED_NS_INTERVAL (2*60)
1018 time_to_check_for_expired_networkstatus = now + CHECK_EXPIRED_NS_INTERVAL;
1021 /* 1g. Check whether we should write statistics to disk.
1023 if (time_to_write_stats_files < now) {
1024 #define CHECK_WRITE_STATS_INTERVAL (60*60)
1025 time_t next_time_to_write_stats_files = (time_to_write_stats_files > 0 ?
1026 time_to_write_stats_files : now) + CHECK_WRITE_STATS_INTERVAL;
1027 if (options->CellStatistics) {
1028 time_t next_write =
1029 rep_hist_buffer_stats_write(time_to_write_stats_files);
1030 if (next_write && next_write < next_time_to_write_stats_files)
1031 next_time_to_write_stats_files = next_write;
1033 if (options->DirReqStatistics) {
1034 time_t next_write = geoip_dirreq_stats_write(time_to_write_stats_files);
1035 if (next_write && next_write < next_time_to_write_stats_files)
1036 next_time_to_write_stats_files = next_write;
1038 if (options->EntryStatistics) {
1039 time_t next_write = geoip_entry_stats_write(time_to_write_stats_files);
1040 if (next_write && next_write < next_time_to_write_stats_files)
1041 next_time_to_write_stats_files = next_write;
1043 if (options->ExitPortStatistics) {
1044 time_t next_write = rep_hist_exit_stats_write(time_to_write_stats_files);
1045 if (next_write && next_write < next_time_to_write_stats_files)
1046 next_time_to_write_stats_files = next_write;
1048 time_to_write_stats_files = next_time_to_write_stats_files;
1051 /* 1h. Check whether we should write bridge statistics to disk.
1053 if (should_record_bridge_info(options)) {
1054 if (time_to_write_bridge_stats < now) {
1055 if (should_init_bridge_stats) {
1056 /* (Re-)initialize bridge statistics. */
1057 geoip_bridge_stats_init(now);
1058 time_to_write_bridge_stats = now + WRITE_STATS_INTERVAL;
1059 should_init_bridge_stats = 0;
1060 } else {
1061 /* Possibly write bridge statistics to disk and ask when to write
1062 * them next time. */
1063 time_to_write_bridge_stats = geoip_bridge_stats_write(
1064 time_to_write_bridge_stats);
1067 } else if (!should_init_bridge_stats) {
1068 /* Bridge mode was turned off. Ensure that stats are re-initialized
1069 * next time bridge mode is turned on. */
1070 should_init_bridge_stats = 1;
1073 /* Remove old information from rephist and the rend cache. */
1074 if (time_to_clean_caches < now) {
1075 rep_history_clean(now - options->RephistTrackTime);
1076 rend_cache_clean();
1077 rend_cache_clean_v2_descs_as_dir();
1078 if (authdir_mode_v3(options))
1079 microdesc_cache_rebuild(NULL, 0);
1080 #define CLEAN_CACHES_INTERVAL (30*60)
1081 time_to_clean_caches = now + CLEAN_CACHES_INTERVAL;
1084 #define RETRY_DNS_INTERVAL (10*60)
1085 /* If we're a server and initializing dns failed, retry periodically. */
1086 if (time_to_retry_dns_init < now) {
1087 time_to_retry_dns_init = now + RETRY_DNS_INTERVAL;
1088 if (server_mode(options) && has_dns_init_failed())
1089 dns_init();
1092 /** 2. Periodically, we consider force-uploading our descriptor
1093 * (if we've passed our internal checks). */
1095 /** How often do we check whether part of our router info has changed in a way
1096 * that would require an upload? */
1097 #define CHECK_DESCRIPTOR_INTERVAL (60)
1098 /** How often do we (as a router) check whether our IP address has changed? */
1099 #define CHECK_IPADDRESS_INTERVAL (15*60)
1101 /* 2b. Once per minute, regenerate and upload the descriptor if the old
1102 * one is inaccurate. */
1103 if (time_to_check_descriptor < now) {
1104 static int dirport_reachability_count = 0;
1105 time_to_check_descriptor = now + CHECK_DESCRIPTOR_INTERVAL;
1106 check_descriptor_bandwidth_changed(now);
1107 if (time_to_check_ipaddress < now) {
1108 time_to_check_ipaddress = now + CHECK_IPADDRESS_INTERVAL;
1109 check_descriptor_ipaddress_changed(now);
1111 /** If our router descriptor ever goes this long without being regenerated
1112 * because something changed, we force an immediate regenerate-and-upload. */
1113 #define FORCE_REGENERATE_DESCRIPTOR_INTERVAL (18*60*60)
1114 mark_my_descriptor_dirty_if_older_than(
1115 now - FORCE_REGENERATE_DESCRIPTOR_INTERVAL);
1116 consider_publishable_server(0);
1117 /* also, check religiously for reachability, if it's within the first
1118 * 20 minutes of our uptime. */
1119 if (server_mode(options) &&
1120 (can_complete_circuit || !any_predicted_circuits(now)) &&
1121 !we_are_hibernating()) {
1122 if (stats_n_seconds_working < TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT) {
1123 consider_testing_reachability(1, dirport_reachability_count==0);
1124 if (++dirport_reachability_count > 5)
1125 dirport_reachability_count = 0;
1126 } else if (time_to_recheck_bandwidth < now) {
1127 /* If we haven't checked for 12 hours and our bandwidth estimate is
1128 * low, do another bandwidth test. This is especially important for
1129 * bridges, since they might go long periods without much use. */
1130 routerinfo_t *me = router_get_my_routerinfo();
1131 if (time_to_recheck_bandwidth && me &&
1132 me->bandwidthcapacity < me->bandwidthrate &&
1133 me->bandwidthcapacity < 51200) {
1134 reset_bandwidth_test();
1136 #define BANDWIDTH_RECHECK_INTERVAL (12*60*60)
1137 time_to_recheck_bandwidth = now + BANDWIDTH_RECHECK_INTERVAL;
1141 /* If any networkstatus documents are no longer recent, we need to
1142 * update all the descriptors' running status. */
1143 /* purge obsolete entries */
1144 networkstatus_v2_list_clean(now);
1145 /* Remove dead routers. */
1146 routerlist_remove_old_routers();
1148 /* Also, once per minute, check whether we want to download any
1149 * networkstatus documents.
1151 update_networkstatus_downloads(now);
1154 /** 2c. Let directory voting happen. */
1155 if (authdir_mode_v3(options))
1156 dirvote_act(options, now);
1158 /** 3a. Every second, we examine pending circuits and prune the
1159 * ones which have been pending for more than a few seconds.
1160 * We do this before step 4, so it can try building more if
1161 * it's not comfortable with the number of available circuits.
1163 /* XXXX022 If our circuit build timeout is much lower than a second, maybe
1164 we should do this more often? */
1165 circuit_expire_building();
1167 /** 3b. Also look at pending streams and prune the ones that 'began'
1168 * a long time ago but haven't gotten a 'connected' yet.
1169 * Do this before step 4, so we can put them back into pending
1170 * state to be picked up by the new circuit.
1172 connection_ap_expire_beginning();
1174 /** 3c. And expire connections that we've held open for too long.
1176 connection_expire_held_open();
1178 /** 3d. And every 60 seconds, we relaunch listeners if any died. */
1179 if (!we_are_hibernating() && time_to_check_listeners < now) {
1180 retry_all_listeners(NULL, NULL);
1181 time_to_check_listeners = now+60;
1184 /** 4. Every second, we try a new circuit if there are no valid
1185 * circuits. Every NewCircuitPeriod seconds, we expire circuits
1186 * that became dirty more than MaxCircuitDirtiness seconds ago,
1187 * and we make a new circ if there are no clean circuits.
1189 have_dir_info = router_have_minimum_dir_info();
1190 if (have_dir_info && !we_are_hibernating())
1191 circuit_build_needed_circs(now);
1193 /* every 10 seconds, but not at the same second as other such events */
1194 if (now % 10 == 5)
1195 circuit_expire_old_circuits_serverside(now);
1197 /** 5. We do housekeeping for each connection... */
1198 connection_or_set_bad_connections(NULL, 0);
1199 for (i=0;i<smartlist_len(connection_array);i++) {
1200 run_connection_housekeeping(i, now);
1202 if (time_to_shrink_memory < now) {
1203 SMARTLIST_FOREACH(connection_array, connection_t *, conn, {
1204 if (conn->outbuf)
1205 buf_shrink(conn->outbuf);
1206 if (conn->inbuf)
1207 buf_shrink(conn->inbuf);
1209 clean_cell_pool();
1210 buf_shrink_freelists(0);
1211 /** How often do we check buffers and pools for empty space that can be
1212 * deallocated? */
1213 #define MEM_SHRINK_INTERVAL (60)
1214 time_to_shrink_memory = now + MEM_SHRINK_INTERVAL;
1217 /** 6. And remove any marked circuits... */
1218 circuit_close_all_marked();
1220 /** 7. And upload service descriptors if necessary. */
1221 if (can_complete_circuit && !we_are_hibernating()) {
1222 rend_consider_services_upload(now);
1223 rend_consider_descriptor_republication();
1226 /** 8. and blow away any connections that need to die. have to do this now,
1227 * because if we marked a conn for close and left its socket -1, then
1228 * we'll pass it to poll/select and bad things will happen.
1230 close_closeable_connections();
1232 /** 8b. And if anything in our state is ready to get flushed to disk, we
1233 * flush it. */
1234 or_state_save(now);
1236 /** 9. and if we're a server, check whether our DNS is telling stories to
1237 * us. */
1238 if (is_server && time_to_check_for_correct_dns < now) {
1239 if (!time_to_check_for_correct_dns) {
1240 time_to_check_for_correct_dns = now + 60 + crypto_rand_int(120);
1241 } else {
1242 dns_launch_correctness_checks();
1243 time_to_check_for_correct_dns = now + 12*3600 +
1244 crypto_rand_int(12*3600);
1248 /** 10b. write bridge networkstatus file to disk */
1249 if (options->BridgeAuthoritativeDir &&
1250 time_to_write_bridge_status_file < now) {
1251 networkstatus_dump_bridge_status_to_file(now);
1252 #define BRIDGE_STATUSFILE_INTERVAL (30*60)
1253 time_to_write_bridge_status_file = now+BRIDGE_STATUSFILE_INTERVAL;
1257 /** Timer: used to invoke second_elapsed_callback() once per second. */
1258 static periodic_timer_t *second_timer = NULL;
1259 /** Number of libevent errors in the last second: we die if we get too many. */
1260 static int n_libevent_errors = 0;
1262 /** Libevent callback: invoked once every second. */
1263 static void
1264 second_elapsed_callback(periodic_timer_t *timer, void *arg)
1266 /* XXXX This could be sensibly refactored into multiple callbacks, and we
1267 * could use Libevent's timers for this rather than checking the current
1268 * time against a bunch of timeouts every second. */
1269 static time_t current_second = 0;
1270 time_t now;
1271 size_t bytes_written;
1272 size_t bytes_read;
1273 int seconds_elapsed;
1274 or_options_t *options = get_options();
1275 (void)timer;
1276 (void)arg;
1278 n_libevent_errors = 0;
1280 /* log_notice(LD_GENERAL, "Tick."); */
1281 now = time(NULL);
1282 update_approx_time(now);
1284 /* the second has rolled over. check more stuff. */
1285 bytes_written = stats_prev_global_write_bucket - global_write_bucket;
1286 bytes_read = stats_prev_global_read_bucket - global_read_bucket;
1287 seconds_elapsed = current_second ? (int)(now - current_second) : 0;
1288 stats_n_bytes_read += bytes_read;
1289 stats_n_bytes_written += bytes_written;
1290 if (accounting_is_enabled(options) && seconds_elapsed >= 0)
1291 accounting_add_bytes(bytes_read, bytes_written, seconds_elapsed);
1292 control_event_bandwidth_used((uint32_t)bytes_read,(uint32_t)bytes_written);
1293 control_event_stream_bandwidth_used();
1295 if (seconds_elapsed > 0)
1296 connection_bucket_refill(seconds_elapsed, now);
1297 stats_prev_global_read_bucket = global_read_bucket;
1298 stats_prev_global_write_bucket = global_write_bucket;
1300 if (server_mode(options) &&
1301 !we_are_hibernating() &&
1302 seconds_elapsed > 0 &&
1303 can_complete_circuit &&
1304 stats_n_seconds_working / TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT !=
1305 (stats_n_seconds_working+seconds_elapsed) /
1306 TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT) {
1307 /* every 20 minutes, check and complain if necessary */
1308 routerinfo_t *me = router_get_my_routerinfo();
1309 if (me && !check_whether_orport_reachable()) {
1310 log_warn(LD_CONFIG,"Your server (%s:%d) has not managed to confirm that "
1311 "its ORPort is reachable. Please check your firewalls, ports, "
1312 "address, /etc/hosts file, etc.",
1313 me->address, me->or_port);
1314 control_event_server_status(LOG_WARN,
1315 "REACHABILITY_FAILED ORADDRESS=%s:%d",
1316 me->address, me->or_port);
1319 if (me && !check_whether_dirport_reachable()) {
1320 log_warn(LD_CONFIG,
1321 "Your server (%s:%d) has not managed to confirm that its "
1322 "DirPort is reachable. Please check your firewalls, ports, "
1323 "address, /etc/hosts file, etc.",
1324 me->address, me->dir_port);
1325 control_event_server_status(LOG_WARN,
1326 "REACHABILITY_FAILED DIRADDRESS=%s:%d",
1327 me->address, me->dir_port);
1331 /** If more than this many seconds have elapsed, probably the clock
1332 * jumped: doesn't count. */
1333 #define NUM_JUMPED_SECONDS_BEFORE_WARN 100
1334 if (seconds_elapsed < -NUM_JUMPED_SECONDS_BEFORE_WARN ||
1335 seconds_elapsed >= NUM_JUMPED_SECONDS_BEFORE_WARN) {
1336 circuit_note_clock_jumped(seconds_elapsed);
1337 /* XXX if the time jumps *back* many months, do our events in
1338 * run_scheduled_events() recover? I don't think they do. -RD */
1339 } else if (seconds_elapsed > 0)
1340 stats_n_seconds_working += seconds_elapsed;
1342 run_scheduled_events(now);
1344 current_second = now; /* remember which second it is, for next time */
1347 #ifndef MS_WINDOWS
1348 /** Called when a possibly ignorable libevent error occurs; ensures that we
1349 * don't get into an infinite loop by ignoring too many errors from
1350 * libevent. */
1351 static int
1352 got_libevent_error(void)
1354 if (++n_libevent_errors > 8) {
1355 log_err(LD_NET, "Too many libevent errors in one second; dying");
1356 return -1;
1358 return 0;
1360 #endif
1362 #define UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST (6*60*60)
1364 /** Called when our IP address seems to have changed. <b>at_interface</b>
1365 * should be true if we detected a change in our interface, and false if we
1366 * detected a change in our published address. */
1367 void
1368 ip_address_changed(int at_interface)
1370 int server = server_mode(get_options());
1372 if (at_interface) {
1373 if (! server) {
1374 /* Okay, change our keys. */
1375 init_keys();
1377 } else {
1378 if (server) {
1379 if (stats_n_seconds_working > UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST)
1380 reset_bandwidth_test();
1381 stats_n_seconds_working = 0;
1382 router_reset_reachability();
1383 mark_my_descriptor_dirty();
1387 dns_servers_relaunch_checks();
1390 /** Forget what we've learned about the correctness of our DNS servers, and
1391 * start learning again. */
1392 void
1393 dns_servers_relaunch_checks(void)
1395 if (server_mode(get_options())) {
1396 dns_reset_correctness_checks();
1397 time_to_check_for_correct_dns = 0;
1401 /** Called when we get a SIGHUP: reload configuration files and keys,
1402 * retry all connections, and so on. */
1403 static int
1404 do_hup(void)
1406 or_options_t *options = get_options();
1408 #ifdef USE_DMALLOC
1409 dmalloc_log_stats();
1410 dmalloc_log_changed(0, 1, 0, 0);
1411 #endif
1413 log_notice(LD_GENERAL,"Received reload signal (hup). Reloading config and "
1414 "resetting internal state.");
1415 if (accounting_is_enabled(options))
1416 accounting_record_bandwidth_usage(time(NULL), get_or_state());
1418 router_reset_warnings();
1419 routerlist_reset_warnings();
1420 addressmap_clear_transient();
1421 /* first, reload config variables, in case they've changed */
1422 if (options->ReloadTorrcOnSIGHUP) {
1423 /* no need to provide argc/v, they've been cached in init_from_config */
1424 if (options_init_from_torrc(0, NULL) < 0) {
1425 log_err(LD_CONFIG,"Reading config failed--see warnings above. "
1426 "For usage, try -h.");
1427 return -1;
1429 options = get_options(); /* they have changed now */
1430 } else {
1431 log_notice(LD_GENERAL, "Not reloading config file: the controller told "
1432 "us not to.");
1434 if (authdir_mode_handles_descs(options, -1)) {
1435 /* reload the approved-routers file */
1436 if (dirserv_load_fingerprint_file() < 0) {
1437 /* warnings are logged from dirserv_load_fingerprint_file() directly */
1438 log_info(LD_GENERAL, "Error reloading fingerprints. "
1439 "Continuing with old list.");
1443 /* Rotate away from the old dirty circuits. This has to be done
1444 * after we've read the new options, but before we start using
1445 * circuits for directory fetches. */
1446 circuit_expire_all_dirty_circs();
1448 /* retry appropriate downloads */
1449 router_reset_status_download_failures();
1450 router_reset_descriptor_download_failures();
1451 update_networkstatus_downloads(time(NULL));
1453 /* We'll retry routerstatus downloads in about 10 seconds; no need to
1454 * force a retry there. */
1456 if (server_mode(options)) {
1457 /* Restart cpuworker and dnsworker processes, so they get up-to-date
1458 * configuration options. */
1459 cpuworkers_rotate();
1460 dns_reset();
1462 return 0;
1465 /** Tor main loop. */
1466 /* static */ int
1467 do_main_loop(void)
1469 int loop_result;
1470 time_t now;
1472 /* initialize dns resolve map, spawn workers if needed */
1473 if (dns_init() < 0) {
1474 if (get_options()->ServerDNSAllowBrokenConfig)
1475 log_warn(LD_GENERAL, "Couldn't set up any working nameservers. "
1476 "Network not up yet? Will try again soon.");
1477 else {
1478 log_err(LD_GENERAL,"Error initializing dns subsystem; exiting. To "
1479 "retry instead, set the ServerDNSAllowBrokenResolvConf option.");
1483 handle_signals(1);
1485 /* load the private keys, if we're supposed to have them, and set up the
1486 * TLS context. */
1487 if (! client_identity_key_is_set()) {
1488 if (init_keys() < 0) {
1489 log_err(LD_BUG,"Error initializing keys; exiting");
1490 return -1;
1494 /* Set up the packed_cell_t memory pool. */
1495 init_cell_pool();
1497 /* Set up our buckets */
1498 connection_bucket_init();
1499 stats_prev_global_read_bucket = global_read_bucket;
1500 stats_prev_global_write_bucket = global_write_bucket;
1502 /* initialize the bootstrap status events to know we're starting up */
1503 control_event_bootstrap(BOOTSTRAP_STATUS_STARTING, 0);
1505 if (trusted_dirs_reload_certs()) {
1506 log_warn(LD_DIR,
1507 "Couldn't load all cached v3 certificates. Starting anyway.");
1509 if (router_reload_v2_networkstatus()) {
1510 return -1;
1512 if (router_reload_consensus_networkstatus()) {
1513 return -1;
1515 /* load the routers file, or assign the defaults. */
1516 if (router_reload_router_list()) {
1517 return -1;
1519 /* load the networkstatuses. (This launches a download for new routers as
1520 * appropriate.)
1522 now = time(NULL);
1523 directory_info_has_arrived(now, 1);
1525 if (server_mode(get_options())) {
1526 /* launch cpuworkers. Need to do this *after* we've read the onion key. */
1527 cpu_init();
1530 /* set up once-a-second callback. */
1531 if (! second_timer) {
1532 struct timeval one_second;
1533 one_second.tv_sec = 1;
1534 one_second.tv_usec = 0;
1536 second_timer = periodic_timer_new(tor_libevent_get_base(),
1537 &one_second,
1538 second_elapsed_callback,
1539 NULL);
1540 tor_assert(second_timer);
1543 for (;;) {
1544 if (nt_service_is_stopping())
1545 return 0;
1547 #ifndef MS_WINDOWS
1548 /* Make it easier to tell whether libevent failure is our fault or not. */
1549 errno = 0;
1550 #endif
1551 /* All active linked conns should get their read events activated. */
1552 SMARTLIST_FOREACH(active_linked_connection_lst, connection_t *, conn,
1553 event_active(conn->read_event, EV_READ, 1));
1554 called_loop_once = smartlist_len(active_linked_connection_lst) ? 1 : 0;
1556 update_approx_time(time(NULL));
1558 /* poll until we have an event, or the second ends, or until we have
1559 * some active linked connections to trigger events for. */
1560 loop_result = event_base_loop(tor_libevent_get_base(),
1561 called_loop_once ? EVLOOP_ONCE : 0);
1563 /* let catch() handle things like ^c, and otherwise don't worry about it */
1564 if (loop_result < 0) {
1565 int e = tor_socket_errno(-1);
1566 /* let the program survive things like ^z */
1567 if (e != EINTR && !ERRNO_IS_EINPROGRESS(e)) {
1568 log_err(LD_NET,"libevent call with %s failed: %s [%d]",
1569 tor_libevent_get_method(), tor_socket_strerror(e), e);
1570 return -1;
1571 #ifndef MS_WINDOWS
1572 } else if (e == EINVAL) {
1573 log_warn(LD_NET, "EINVAL from libevent: should you upgrade libevent?");
1574 if (got_libevent_error())
1575 return -1;
1576 #endif
1577 } else {
1578 if (ERRNO_IS_EINPROGRESS(e))
1579 log_warn(LD_BUG,
1580 "libevent call returned EINPROGRESS? Please report.");
1581 log_debug(LD_NET,"libevent call interrupted.");
1582 /* You can't trust the results of this poll(). Go back to the
1583 * top of the big for loop. */
1584 continue;
1590 #ifndef MS_WINDOWS /* Only called when we're willing to use signals */
1591 /** Libevent callback: invoked when we get a signal.
1593 static void
1594 signal_callback(int fd, short events, void *arg)
1596 uintptr_t sig = (uintptr_t)arg;
1597 (void)fd;
1598 (void)events;
1600 process_signal(sig);
1602 #endif
1604 /** Do the work of acting on a signal received in <b>sig</b> */
1605 void
1606 process_signal(uintptr_t sig)
1608 switch (sig)
1610 case SIGTERM:
1611 log_notice(LD_GENERAL,"Catching signal TERM, exiting cleanly.");
1612 tor_cleanup();
1613 exit(0);
1614 break;
1615 case SIGINT:
1616 if (!server_mode(get_options())) { /* do it now */
1617 log_notice(LD_GENERAL,"Interrupt: exiting cleanly.");
1618 tor_cleanup();
1619 exit(0);
1621 hibernate_begin_shutdown();
1622 break;
1623 #ifdef SIGPIPE
1624 case SIGPIPE:
1625 log_debug(LD_GENERAL,"Caught SIGPIPE. Ignoring.");
1626 break;
1627 #endif
1628 case SIGUSR1:
1629 /* prefer to log it at INFO, but make sure we always see it */
1630 dumpstats(get_min_log_level()<LOG_INFO ? get_min_log_level() : LOG_INFO);
1631 break;
1632 case SIGUSR2:
1633 switch_logs_debug();
1634 log_debug(LD_GENERAL,"Caught USR2, going to loglevel debug. "
1635 "Send HUP to change back.");
1636 break;
1637 case SIGHUP:
1638 if (do_hup() < 0) {
1639 log_warn(LD_CONFIG,"Restart failed (config error?). Exiting.");
1640 tor_cleanup();
1641 exit(1);
1643 break;
1644 #ifdef SIGCHLD
1645 case SIGCHLD:
1646 while (waitpid(-1,NULL,WNOHANG) > 0) ; /* keep reaping until no more
1647 zombies */
1648 break;
1649 #endif
1650 case SIGNEWNYM: {
1651 time_t now = time(NULL);
1652 if (time_of_last_signewnym + MAX_SIGNEWNYM_RATE > now) {
1653 signewnym_is_pending = 1;
1654 log(LOG_NOTICE, LD_CONTROL,
1655 "Rate limiting NEWNYM request: delaying by %d second(s)",
1656 (int)(MAX_SIGNEWNYM_RATE+time_of_last_signewnym-now));
1657 } else {
1658 signewnym_impl(now);
1660 break;
1662 case SIGCLEARDNSCACHE:
1663 addressmap_clear_transient();
1664 break;
1668 extern uint64_t rephist_total_alloc;
1669 extern uint32_t rephist_total_num;
1672 * Write current memory usage information to the log.
1674 static void
1675 dumpmemusage(int severity)
1677 connection_dump_buffer_mem_stats(severity);
1678 log(severity, LD_GENERAL, "In rephist: "U64_FORMAT" used by %d Tors.",
1679 U64_PRINTF_ARG(rephist_total_alloc), rephist_total_num);
1680 dump_routerlist_mem_usage(severity);
1681 dump_cell_pool_usage(severity);
1682 dump_dns_mem_usage(severity);
1683 buf_dump_freelist_sizes(severity);
1684 tor_log_mallinfo(severity);
1687 /** Write all statistics to the log, with log level <b>severity</b>. Called
1688 * in response to a SIGUSR1. */
1689 static void
1690 dumpstats(int severity)
1692 time_t now = time(NULL);
1693 time_t elapsed;
1694 size_t rbuf_cap, wbuf_cap, rbuf_len, wbuf_len;
1696 log(severity, LD_GENERAL, "Dumping stats:");
1698 SMARTLIST_FOREACH(connection_array, connection_t *, conn,
1700 int i = conn_sl_idx;
1701 log(severity, LD_GENERAL,
1702 "Conn %d (socket %d) type %d (%s), state %d (%s), created %d secs ago",
1703 i, conn->s, conn->type, conn_type_to_string(conn->type),
1704 conn->state, conn_state_to_string(conn->type, conn->state),
1705 (int)(now - conn->timestamp_created));
1706 if (!connection_is_listener(conn)) {
1707 log(severity,LD_GENERAL,
1708 "Conn %d is to %s:%d.", i,
1709 safe_str_client(conn->address),
1710 conn->port);
1711 log(severity,LD_GENERAL,
1712 "Conn %d: %d bytes waiting on inbuf (len %d, last read %d secs ago)",
1714 (int)buf_datalen(conn->inbuf),
1715 (int)buf_allocation(conn->inbuf),
1716 (int)(now - conn->timestamp_lastread));
1717 log(severity,LD_GENERAL,
1718 "Conn %d: %d bytes waiting on outbuf "
1719 "(len %d, last written %d secs ago)",i,
1720 (int)buf_datalen(conn->outbuf),
1721 (int)buf_allocation(conn->outbuf),
1722 (int)(now - conn->timestamp_lastwritten));
1723 if (conn->type == CONN_TYPE_OR) {
1724 or_connection_t *or_conn = TO_OR_CONN(conn);
1725 if (or_conn->tls) {
1726 tor_tls_get_buffer_sizes(or_conn->tls, &rbuf_cap, &rbuf_len,
1727 &wbuf_cap, &wbuf_len);
1728 log(severity, LD_GENERAL,
1729 "Conn %d: %d/%d bytes used on OpenSSL read buffer; "
1730 "%d/%d bytes used on write buffer.",
1731 i, (int)rbuf_len, (int)rbuf_cap, (int)wbuf_len, (int)wbuf_cap);
1735 circuit_dump_by_conn(conn, severity); /* dump info about all the circuits
1736 * using this conn */
1738 log(severity, LD_NET,
1739 "Cells processed: "U64_FORMAT" padding\n"
1740 " "U64_FORMAT" create\n"
1741 " "U64_FORMAT" created\n"
1742 " "U64_FORMAT" relay\n"
1743 " ("U64_FORMAT" relayed)\n"
1744 " ("U64_FORMAT" delivered)\n"
1745 " "U64_FORMAT" destroy",
1746 U64_PRINTF_ARG(stats_n_padding_cells_processed),
1747 U64_PRINTF_ARG(stats_n_create_cells_processed),
1748 U64_PRINTF_ARG(stats_n_created_cells_processed),
1749 U64_PRINTF_ARG(stats_n_relay_cells_processed),
1750 U64_PRINTF_ARG(stats_n_relay_cells_relayed),
1751 U64_PRINTF_ARG(stats_n_relay_cells_delivered),
1752 U64_PRINTF_ARG(stats_n_destroy_cells_processed));
1753 if (stats_n_data_cells_packaged)
1754 log(severity,LD_NET,"Average packaged cell fullness: %2.3f%%",
1755 100*(U64_TO_DBL(stats_n_data_bytes_packaged) /
1756 U64_TO_DBL(stats_n_data_cells_packaged*RELAY_PAYLOAD_SIZE)) );
1757 if (stats_n_data_cells_received)
1758 log(severity,LD_NET,"Average delivered cell fullness: %2.3f%%",
1759 100*(U64_TO_DBL(stats_n_data_bytes_received) /
1760 U64_TO_DBL(stats_n_data_cells_received*RELAY_PAYLOAD_SIZE)) );
1762 if (now - time_of_process_start >= 0)
1763 elapsed = now - time_of_process_start;
1764 else
1765 elapsed = 0;
1767 if (elapsed) {
1768 log(severity, LD_NET,
1769 "Average bandwidth: "U64_FORMAT"/%d = %d bytes/sec reading",
1770 U64_PRINTF_ARG(stats_n_bytes_read),
1771 (int)elapsed,
1772 (int) (stats_n_bytes_read/elapsed));
1773 log(severity, LD_NET,
1774 "Average bandwidth: "U64_FORMAT"/%d = %d bytes/sec writing",
1775 U64_PRINTF_ARG(stats_n_bytes_written),
1776 (int)elapsed,
1777 (int) (stats_n_bytes_written/elapsed));
1780 log(severity, LD_NET, "--------------- Dumping memory information:");
1781 dumpmemusage(severity);
1783 rep_hist_dump_stats(now,severity);
1784 rend_service_dump_stats(severity);
1785 dump_pk_ops(severity);
1786 dump_distinct_digest_count(severity);
1789 /** Called by exit() as we shut down the process.
1791 static void
1792 exit_function(void)
1794 /* NOTE: If we ever daemonize, this gets called immediately. That's
1795 * okay for now, because we only use this on Windows. */
1796 #ifdef MS_WINDOWS
1797 WSACleanup();
1798 #endif
1801 /** Set up the signal handlers for either parent or child. */
1802 void
1803 handle_signals(int is_parent)
1805 #ifndef MS_WINDOWS /* do signal stuff only on Unix */
1806 int i;
1807 static const int signals[] = {
1808 SIGINT, /* do a controlled slow shutdown */
1809 SIGTERM, /* to terminate now */
1810 SIGPIPE, /* otherwise SIGPIPE kills us */
1811 SIGUSR1, /* dump stats */
1812 SIGUSR2, /* go to loglevel debug */
1813 SIGHUP, /* to reload config, retry conns, etc */
1814 #ifdef SIGXFSZ
1815 SIGXFSZ, /* handle file-too-big resource exhaustion */
1816 #endif
1817 SIGCHLD, /* handle dns/cpu workers that exit */
1818 -1 };
1819 static struct event *signal_events[16]; /* bigger than it has to be. */
1820 if (is_parent) {
1821 for (i = 0; signals[i] >= 0; ++i) {
1822 signal_events[i] = tor_evsignal_new(
1823 tor_libevent_get_base(), signals[i], signal_callback,
1824 (void*)(uintptr_t)signals[i]);
1825 if (event_add(signal_events[i], NULL))
1826 log_warn(LD_BUG, "Error from libevent when adding event for signal %d",
1827 signals[i]);
1829 } else {
1830 struct sigaction action;
1831 action.sa_flags = 0;
1832 sigemptyset(&action.sa_mask);
1833 action.sa_handler = SIG_IGN;
1834 sigaction(SIGINT, &action, NULL);
1835 sigaction(SIGTERM, &action, NULL);
1836 sigaction(SIGPIPE, &action, NULL);
1837 sigaction(SIGUSR1, &action, NULL);
1838 sigaction(SIGUSR2, &action, NULL);
1839 sigaction(SIGHUP, &action, NULL);
1840 #ifdef SIGXFSZ
1841 sigaction(SIGXFSZ, &action, NULL);
1842 #endif
1844 #else /* MS windows */
1845 (void)is_parent;
1846 #endif /* signal stuff */
1849 /** Main entry point for the Tor command-line client.
1851 /* static */ int
1852 tor_init(int argc, char *argv[])
1854 char buf[256];
1855 int i, quiet = 0;
1856 time_of_process_start = time(NULL);
1857 if (!connection_array)
1858 connection_array = smartlist_create();
1859 if (!closeable_connection_lst)
1860 closeable_connection_lst = smartlist_create();
1861 if (!active_linked_connection_lst)
1862 active_linked_connection_lst = smartlist_create();
1863 /* Have the log set up with our application name. */
1864 tor_snprintf(buf, sizeof(buf), "Tor %s", get_version());
1865 log_set_application_name(buf);
1866 /* Initialize the history structures. */
1867 rep_hist_init();
1868 /* Initialize the service cache. */
1869 rend_cache_init();
1870 addressmap_init(); /* Init the client dns cache. Do it always, since it's
1871 * cheap. */
1873 /* We search for the "quiet" option first, since it decides whether we
1874 * will log anything at all to the command line. */
1875 for (i=1;i<argc;++i) {
1876 if (!strcmp(argv[i], "--hush"))
1877 quiet = 1;
1878 if (!strcmp(argv[i], "--quiet"))
1879 quiet = 2;
1881 /* give it somewhere to log to initially */
1882 switch (quiet) {
1883 case 2:
1884 /* no initial logging */
1885 break;
1886 case 1:
1887 add_temp_log(LOG_WARN);
1888 break;
1889 default:
1890 add_temp_log(LOG_NOTICE);
1893 log(LOG_NOTICE, LD_GENERAL, "Tor v%s. This is experimental software. "
1894 "Do not rely on it for strong anonymity. (Running on %s)",get_version(),
1895 get_uname());
1897 if (network_init()<0) {
1898 log_err(LD_BUG,"Error initializing network; exiting.");
1899 return -1;
1901 atexit(exit_function);
1903 if (options_init_from_torrc(argc,argv) < 0) {
1904 log_err(LD_CONFIG,"Reading config failed--see warnings above.");
1905 return -1;
1908 #ifndef MS_WINDOWS
1909 if (geteuid()==0)
1910 log_warn(LD_GENERAL,"You are running Tor as root. You don't need to, "
1911 "and you probably shouldn't.");
1912 #endif
1914 if (crypto_global_init(get_options()->HardwareAccel,
1915 get_options()->AccelName,
1916 get_options()->AccelDir)) {
1917 log_err(LD_BUG, "Unable to initialize OpenSSL. Exiting.");
1918 return -1;
1921 return 0;
1924 /** A lockfile structure, used to prevent two Tors from messing with the
1925 * data directory at once. If this variable is non-NULL, we're holding
1926 * the lockfile. */
1927 static tor_lockfile_t *lockfile = NULL;
1929 /** Try to grab the lock file described in <b>options</b>, if we do not
1930 * already have it. If <b>err_if_locked</b> is true, warn if somebody else is
1931 * holding the lock, and exit if we can't get it after waiting. Otherwise,
1932 * return -1 if we can't get the lockfile. Return 0 on success.
1935 try_locking(or_options_t *options, int err_if_locked)
1937 if (lockfile)
1938 return 0;
1939 else {
1940 char *fname = options_get_datadir_fname2_suffix(options, "lock",NULL,NULL);
1941 int already_locked = 0;
1942 tor_lockfile_t *lf = tor_lockfile_lock(fname, 0, &already_locked);
1943 tor_free(fname);
1944 if (!lf) {
1945 if (err_if_locked && already_locked) {
1946 int r;
1947 log_warn(LD_GENERAL, "It looks like another Tor process is running "
1948 "with the same data directory. Waiting 5 seconds to see "
1949 "if it goes away.");
1950 #ifndef WIN32
1951 sleep(5);
1952 #else
1953 Sleep(5000);
1954 #endif
1955 r = try_locking(options, 0);
1956 if (r<0) {
1957 log_err(LD_GENERAL, "No, it's still there. Exiting.");
1958 exit(0);
1960 return r;
1962 return -1;
1964 lockfile = lf;
1965 return 0;
1969 /** Return true iff we've successfully acquired the lock file. */
1971 have_lockfile(void)
1973 return lockfile != NULL;
1976 /** If we have successfully acquired the lock file, release it. */
1977 void
1978 release_lockfile(void)
1980 if (lockfile) {
1981 tor_lockfile_unlock(lockfile);
1982 lockfile = NULL;
1986 /** Free all memory that we might have allocated somewhere.
1987 * If <b>postfork</b>, we are a worker process and we want to free
1988 * only the parts of memory that we won't touch. If !<b>postfork</b>,
1989 * Tor is shutting down and we should free everything.
1991 * Helps us find the real leaks with dmalloc and the like. Also valgrind
1992 * should then report 0 reachable in its leak report (in an ideal world --
1993 * in practice libevent, SSL, libc etc never quite free everything). */
1994 void
1995 tor_free_all(int postfork)
1997 if (!postfork) {
1998 evdns_shutdown(1);
2000 geoip_free_all();
2001 dirvote_free_all();
2002 routerlist_free_all();
2003 networkstatus_free_all();
2004 addressmap_free_all();
2005 dirserv_free_all();
2006 rend_service_free_all();
2007 rend_cache_free_all();
2008 rend_service_authorization_free_all();
2009 rep_hist_free_all();
2010 dns_free_all();
2011 clear_pending_onions();
2012 circuit_free_all();
2013 entry_guards_free_all();
2014 connection_free_all();
2015 buf_shrink_freelists(1);
2016 memarea_clear_freelist();
2017 microdesc_free_all();
2018 if (!postfork) {
2019 config_free_all();
2020 router_free_all();
2021 policies_free_all();
2023 free_cell_pool();
2024 if (!postfork) {
2025 tor_tls_free_all();
2027 /* stuff in main.c */
2029 smartlist_free(connection_array);
2030 smartlist_free(closeable_connection_lst);
2031 smartlist_free(active_linked_connection_lst);
2032 periodic_timer_free(second_timer);
2033 if (!postfork) {
2034 release_lockfile();
2036 /* Stuff in util.c and address.c*/
2037 if (!postfork) {
2038 escaped(NULL);
2039 esc_router_info(NULL);
2040 logs_free_all(); /* free log strings. do this last so logs keep working. */
2044 /** Do whatever cleanup is necessary before shutting Tor down. */
2045 void
2046 tor_cleanup(void)
2048 or_options_t *options = get_options();
2049 /* Remove our pid file. We don't care if there was an error when we
2050 * unlink, nothing we could do about it anyways. */
2051 if (options->command == CMD_RUN_TOR) {
2052 time_t now = time(NULL);
2053 if (options->PidFile)
2054 unlink(options->PidFile);
2055 if (accounting_is_enabled(options))
2056 accounting_record_bandwidth_usage(now, get_or_state());
2057 or_state_mark_dirty(get_or_state(), 0); /* force an immediate save. */
2058 or_state_save(now);
2059 if (authdir_mode_tests_reachability(options))
2060 rep_hist_record_mtbf_data(now, 0);
2062 #ifdef USE_DMALLOC
2063 dmalloc_log_stats();
2064 #endif
2065 tor_free_all(0); /* We could move tor_free_all back into the ifdef below
2066 later, if it makes shutdown unacceptably slow. But for
2067 now, leave it here: it's helped us catch bugs in the
2068 past. */
2069 crypto_global_cleanup();
2070 #ifdef USE_DMALLOC
2071 dmalloc_log_unfreed();
2072 dmalloc_shutdown();
2073 #endif
2076 /** Read/create keys as needed, and echo our fingerprint to stdout. */
2077 /* static */ int
2078 do_list_fingerprint(void)
2080 char buf[FINGERPRINT_LEN+1];
2081 crypto_pk_env_t *k;
2082 const char *nickname = get_options()->Nickname;
2083 if (!server_mode(get_options())) {
2084 log_err(LD_GENERAL,
2085 "Clients don't have long-term identity keys. Exiting.\n");
2086 return -1;
2088 tor_assert(nickname);
2089 if (init_keys() < 0) {
2090 log_err(LD_BUG,"Error initializing keys; can't display fingerprint");
2091 return -1;
2093 if (!(k = get_server_identity_key())) {
2094 log_err(LD_GENERAL,"Error: missing identity key.");
2095 return -1;
2097 if (crypto_pk_get_fingerprint(k, buf, 1)<0) {
2098 log_err(LD_BUG, "Error computing fingerprint");
2099 return -1;
2101 printf("%s %s\n", nickname, buf);
2102 return 0;
2105 /** Entry point for password hashing: take the desired password from
2106 * the command line, and print its salted hash to stdout. **/
2107 /* static */ void
2108 do_hash_password(void)
2111 char output[256];
2112 char key[S2K_SPECIFIER_LEN+DIGEST_LEN];
2114 crypto_rand(key, S2K_SPECIFIER_LEN-1);
2115 key[S2K_SPECIFIER_LEN-1] = (uint8_t)96; /* Hash 64 K of data. */
2116 secret_to_key(key+S2K_SPECIFIER_LEN, DIGEST_LEN,
2117 get_options()->command_arg, strlen(get_options()->command_arg),
2118 key);
2119 base16_encode(output, sizeof(output), key, sizeof(key));
2120 printf("16:%s\n",output);
2123 #if defined (WINCE)
2125 find_flashcard_path(PWCHAR path, size_t size)
2127 WIN32_FIND_DATA d = {0};
2128 HANDLE h = NULL;
2130 if (!path)
2131 return -1;
2133 h = FindFirstFlashCard(&d);
2134 if (h == INVALID_HANDLE_VALUE)
2135 return -1;
2137 if (wcslen(d.cFileName) == 0) {
2138 FindClose(h);
2139 return -1;
2142 wcsncpy(path,d.cFileName,size);
2143 FindClose(h);
2144 return 0;
2146 #endif
2148 /** Main entry point for the Tor process. Called from main(). */
2149 /* This function is distinct from main() only so we can link main.c into
2150 * the unittest binary without conflicting with the unittests' main. */
2152 tor_main(int argc, char *argv[])
2154 int result = 0;
2155 #if defined (WINCE)
2156 WCHAR path [MAX_PATH] = {0};
2157 WCHAR fullpath [MAX_PATH] = {0};
2158 PWCHAR p = NULL;
2159 FILE* redir = NULL;
2160 FILE* redirdbg = NULL;
2162 // this is to facilitate debugging by opening
2163 // a file on a folder shared by the wm emulator.
2164 // if no flashcard (real or emulated) is present,
2165 // log files will be written in the root folder
2166 if (find_flashcard_path(path,MAX_PATH) == -1)
2168 redir = _wfreopen( L"\\stdout.log", L"w", stdout );
2169 redirdbg = _wfreopen( L"\\stderr.log", L"w", stderr );
2170 } else {
2171 swprintf(fullpath,L"\\%s\\tor",path);
2172 CreateDirectory(fullpath,NULL);
2174 swprintf(fullpath,L"\\%s\\tor\\stdout.log",path);
2175 redir = _wfreopen( fullpath, L"w", stdout );
2177 swprintf(fullpath,L"\\%s\\tor\\stderr.log",path);
2178 redirdbg = _wfreopen( fullpath, L"w", stderr );
2180 #endif
2182 #ifdef MS_WINDOWS
2183 /* Call SetProcessDEPPolicy to permanently enable DEP.
2184 The function will not resolve on earlier versions of Windows,
2185 and failure is not dangerous. */
2186 HMODULE hMod = GetModuleHandleA("Kernel32.dll");
2187 if (hMod) {
2188 typedef BOOL (WINAPI *PSETDEP)(DWORD);
2189 PSETDEP setdeppolicy = (PSETDEP)GetProcAddress(hMod,
2190 "SetProcessDEPPolicy");
2191 if (setdeppolicy) setdeppolicy(1); /* PROCESS_DEP_ENABLE */
2193 #endif
2195 update_approx_time(time(NULL));
2196 tor_threads_init();
2197 init_logging();
2198 #ifdef USE_DMALLOC
2200 /* Instruct OpenSSL to use our internal wrappers for malloc,
2201 realloc and free. */
2202 int r = CRYPTO_set_mem_ex_functions(_tor_malloc, _tor_realloc, _tor_free);
2203 tor_assert(r);
2205 #endif
2206 #ifdef NT_SERVICE
2208 int done = 0;
2209 result = nt_service_parse_options(argc, argv, &done);
2210 if (done) return result;
2212 #endif
2213 if (tor_init(argc, argv)<0)
2214 return -1;
2215 switch (get_options()->command) {
2216 case CMD_RUN_TOR:
2217 #ifdef NT_SERVICE
2218 nt_service_set_state(SERVICE_RUNNING);
2219 #endif
2220 result = do_main_loop();
2221 break;
2222 case CMD_LIST_FINGERPRINT:
2223 result = do_list_fingerprint();
2224 break;
2225 case CMD_HASH_PASSWORD:
2226 do_hash_password();
2227 result = 0;
2228 break;
2229 case CMD_VERIFY_CONFIG:
2230 printf("Configuration was valid\n");
2231 result = 0;
2232 break;
2233 case CMD_RUN_UNITTESTS: /* only set by test.c */
2234 default:
2235 log_warn(LD_BUG,"Illegal command number %d: internal error.",
2236 get_options()->command);
2237 result = -1;
2239 tor_cleanup();
2240 return result;