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) 2007-2008, The Tor Project, Inc. */
5 /* See LICENSE for licensing information */
7 const char main_c_id
[] =
12 * \brief Toplevel module. Handles signals, multiplexes between
13 * connections, implements main loop, and drives scheduled events.
22 void evdns_shutdown(int);
24 /********* PROTOTYPES **********/
26 static void dumpmemusage(int severity
);
27 static void dumpstats(int severity
); /* log stats */
28 static void conn_read_callback(int fd
, short event
, void *_conn
);
29 static void conn_write_callback(int fd
, short event
, void *_conn
);
30 static void signal_callback(int fd
, short events
, void *arg
);
31 static void second_elapsed_callback(int fd
, short event
, void *args
);
32 static int conn_close_if_marked(int i
);
33 static void connection_start_reading_from_linked_conn(connection_t
*conn
);
34 static int connection_should_read_from_linked_conn(connection_t
*conn
);
36 /********* START VARIABLES **********/
38 int global_read_bucket
; /**< Max number of bytes I can read this second. */
39 int global_write_bucket
; /**< Max number of bytes I can write this second. */
41 /** Max number of relayed (bandwidth class 1) bytes I can read this second. */
42 int global_relayed_read_bucket
;
43 /** Max number of relayed (bandwidth class 1) bytes I can write this second. */
44 int global_relayed_write_bucket
;
46 /** What was the read bucket before the last call to prepare_for_pool?
47 * (used to determine how many bytes we've read). */
48 static int stats_prev_global_read_bucket
;
49 /** What was the write bucket before the last call to prepare_for_pool?
50 * (used to determine how many bytes we've written). */
51 static int stats_prev_global_write_bucket
;
52 /* XXX we might want to keep stats about global_relayed_*_bucket too. Or not.*/
53 /** How many bytes have we read/written since we started the process? */
54 static uint64_t stats_n_bytes_read
= 0;
55 static uint64_t stats_n_bytes_written
= 0;
56 /** What time did this process start up? */
57 time_t time_of_process_start
= 0;
58 /** How many seconds have we been running? */
59 long stats_n_seconds_working
= 0;
60 /** When do we next launch DNS wildcarding checks? */
61 static time_t time_to_check_for_correct_dns
= 0;
63 /** How often will we honor SIGNEWNYM requests? */
64 #define MAX_SIGNEWNYM_RATE 10
65 /** When did we last process a SIGNEWNYM request? */
66 static time_t time_of_last_signewnym
= 0;
67 /** Is there a signewnym request we're currently waiting to handle? */
68 static int signewnym_is_pending
= 0;
70 /** Smartlist of all open connections. */
71 static smartlist_t
*connection_array
= NULL
;
72 /** List of connections that have been marked for close and need to be freed
73 * and removed from connection_array. */
74 static smartlist_t
*closeable_connection_lst
= NULL
;
75 /** List of linked connections that are currently reading data into their
76 * inbuf from their partner's outbuf. */
77 static smartlist_t
*active_linked_connection_lst
= NULL
;
78 /** Flag: Set to true iff we entered the current libevent main loop via
79 * <b>loop_once</b>. If so, there's no need to trigger a loopexit in order
80 * to handle linked connections. */
81 static int called_loop_once
= 0;
83 /** We set this to 1 when we've opened a circuit, so we can print a log
84 * entry to inform the user that Tor is working. */
85 int has_completed_circuit
=0;
87 /** How often do we check for router descriptors that we should download
88 * when we have too little directory info? */
89 #define GREEDY_DESCRIPTOR_RETRY_INTERVAL (10)
90 /** How often do we check for router descriptors that we should download
91 * when we have enough directory info? */
92 #define LAZY_DESCRIPTOR_RETRY_INTERVAL (60)
93 /** How often do we 'forgive' undownloadable router descriptors and attempt
94 * to download them again? */
95 #define DESCRIPTOR_FAILURE_RESET_INTERVAL (60*60)
96 /** How long do we let a directory connection stall before expiring it? */
97 #define DIR_CONN_MAX_STALL (5*60)
99 /** How old do we let a connection to an OR get before deciding it's
101 #define TIME_BEFORE_OR_CONN_IS_OBSOLETE (60*60*24*7)
102 /** How long do we let OR connections handshake before we decide that
103 * they are obsolete? */
104 #define TLS_HANDSHAKE_TIMEOUT (60)
106 /********* END VARIABLES ************/
108 /****************************************************************************
110 * This section contains accessors and other methods on the connection_array
111 * variables (which are global within this file and unavailable outside it).
113 ****************************************************************************/
115 /** Add <b>conn</b> to the array of connections that we can poll on. The
116 * connection's socket must be set; the connection starts out
117 * non-reading and non-writing.
120 connection_add(connection_t
*conn
)
123 tor_assert(conn
->s
>= 0 ||
125 (conn
->type
== CONN_TYPE_AP
&&
126 TO_EDGE_CONN(conn
)->is_dns_request
));
128 tor_assert(conn
->conn_array_index
== -1); /* can only connection_add once */
129 conn
->conn_array_index
= smartlist_len(connection_array
);
130 smartlist_add(connection_array
, conn
);
132 if (conn
->s
>= 0 || conn
->linked
) {
133 conn
->read_event
= tor_malloc_zero(sizeof(struct event
));
134 conn
->write_event
= tor_malloc_zero(sizeof(struct event
));
135 event_set(conn
->read_event
, conn
->s
, EV_READ
|EV_PERSIST
,
136 conn_read_callback
, conn
);
137 event_set(conn
->write_event
, conn
->s
, EV_WRITE
|EV_PERSIST
,
138 conn_write_callback
, conn
);
141 log_debug(LD_NET
,"new conn type %s, socket %d, n_conns %d.",
142 conn_type_to_string(conn
->type
), conn
->s
,
143 smartlist_len(connection_array
));
148 /** Remove the connection from the global list, and remove the
149 * corresponding poll entry. Calling this function will shift the last
150 * connection (if any) into the position occupied by conn.
153 connection_remove(connection_t
*conn
)
160 log_debug(LD_NET
,"removing socket %d (type %s), n_conns now %d",
161 conn
->s
, conn_type_to_string(conn
->type
),
162 smartlist_len(connection_array
));
164 tor_assert(conn
->conn_array_index
>= 0);
165 current_index
= conn
->conn_array_index
;
166 connection_unregister_events(conn
); /* This is redundant, but cheap. */
167 if (current_index
== smartlist_len(connection_array
)-1) { /* at the end */
168 smartlist_del(connection_array
, current_index
);
172 /* replace this one with the one at the end */
173 smartlist_del(connection_array
, current_index
);
174 tmp
= smartlist_get(connection_array
, current_index
);
175 tmp
->conn_array_index
= current_index
;
180 /** If <b>conn</b> is an edge conn, remove it from the list
181 * of conn's on this circuit. If it's not on an edge,
182 * flush and send destroys for all circuits on this conn.
184 * Remove it from connection_array (if applicable) and
185 * from closeable_connection_list.
190 connection_unlink(connection_t
*conn
)
192 connection_about_to_close_connection(conn
);
193 if (conn
->conn_array_index
>= 0) {
194 connection_remove(conn
);
196 if (conn
->linked_conn
) {
197 conn
->linked_conn
->linked_conn
= NULL
;
198 if (! conn
->linked_conn
->marked_for_close
&&
199 conn
->linked_conn
->reading_from_linked_conn
)
200 connection_start_reading(conn
->linked_conn
);
201 conn
->linked_conn
= NULL
;
203 smartlist_remove(closeable_connection_lst
, conn
);
204 smartlist_remove(active_linked_connection_lst
, conn
);
205 if (conn
->type
== CONN_TYPE_EXIT
) {
206 assert_connection_edge_not_dns_pending(TO_EDGE_CONN(conn
));
208 if (conn
->type
== CONN_TYPE_OR
) {
209 if (!tor_digest_is_zero(TO_OR_CONN(conn
)->identity_digest
))
210 connection_or_remove_from_identity_map(TO_OR_CONN(conn
));
212 connection_free(conn
);
215 /** Schedule <b>conn</b> to be closed. **/
217 add_connection_to_closeable_list(connection_t
*conn
)
219 tor_assert(!smartlist_isin(closeable_connection_lst
, conn
));
220 tor_assert(conn
->marked_for_close
);
221 assert_connection_ok(conn
, time(NULL
));
222 smartlist_add(closeable_connection_lst
, conn
);
225 /** Return 1 if conn is on the closeable list, else return 0. */
227 connection_is_on_closeable_list(connection_t
*conn
)
229 return smartlist_isin(closeable_connection_lst
, conn
);
232 /** Return true iff conn is in the current poll array. */
234 connection_in_array(connection_t
*conn
)
236 return smartlist_isin(connection_array
, conn
);
239 /** Set <b>*array</b> to an array of all connections, and <b>*n</b>
240 * to the length of the array. <b>*array</b> and <b>*n</b> must not
244 get_connection_array(void)
246 return connection_array
;
249 /** Set the event mask on <b>conn</b> to <b>events</b>. (The event
250 * mask is a bitmask whose bits are EV_READ and EV_WRITE.)
253 connection_watch_events(connection_t
*conn
, short events
)
255 if (events
& EV_READ
)
256 connection_start_reading(conn
);
258 connection_stop_reading(conn
);
260 if (events
& EV_WRITE
)
261 connection_start_writing(conn
);
263 connection_stop_writing(conn
);
266 /** Return true iff <b>conn</b> is listening for read events. */
268 connection_is_reading(connection_t
*conn
)
272 return conn
->reading_from_linked_conn
||
273 (conn
->read_event
&& event_pending(conn
->read_event
, EV_READ
, NULL
));
276 /** Tell the main loop to stop notifying <b>conn</b> of any read events. */
278 connection_stop_reading(connection_t
*conn
)
281 tor_assert(conn
->read_event
);
284 conn
->reading_from_linked_conn
= 0;
285 connection_stop_reading_from_linked_conn(conn
);
287 if (event_del(conn
->read_event
))
288 log_warn(LD_NET
, "Error from libevent setting read event state for %d "
291 tor_socket_strerror(tor_socket_errno(conn
->s
)));
295 /** Tell the main loop to start notifying <b>conn</b> of any read events. */
297 connection_start_reading(connection_t
*conn
)
300 tor_assert(conn
->read_event
);
303 conn
->reading_from_linked_conn
= 1;
304 if (connection_should_read_from_linked_conn(conn
))
305 connection_start_reading_from_linked_conn(conn
);
307 if (event_add(conn
->read_event
, NULL
))
308 log_warn(LD_NET
, "Error from libevent setting read event state for %d "
311 tor_socket_strerror(tor_socket_errno(conn
->s
)));
315 /** Return true iff <b>conn</b> is listening for write events. */
317 connection_is_writing(connection_t
*conn
)
321 return conn
->writing_to_linked_conn
||
322 (conn
->write_event
&& event_pending(conn
->write_event
, EV_WRITE
, NULL
));
325 /** Tell the main loop to stop notifying <b>conn</b> of any write events. */
327 connection_stop_writing(connection_t
*conn
)
330 tor_assert(conn
->write_event
);
333 conn
->writing_to_linked_conn
= 0;
334 if (conn
->linked_conn
)
335 connection_stop_reading_from_linked_conn(conn
->linked_conn
);
337 if (event_del(conn
->write_event
))
338 log_warn(LD_NET
, "Error from libevent setting write event state for %d "
341 tor_socket_strerror(tor_socket_errno(conn
->s
)));
345 /** Tell the main loop to start notifying <b>conn</b> of any write events. */
347 connection_start_writing(connection_t
*conn
)
350 tor_assert(conn
->write_event
);
353 conn
->writing_to_linked_conn
= 1;
354 if (conn
->linked_conn
&&
355 connection_should_read_from_linked_conn(conn
->linked_conn
))
356 connection_start_reading_from_linked_conn(conn
->linked_conn
);
358 if (event_add(conn
->write_event
, NULL
))
359 log_warn(LD_NET
, "Error from libevent setting write event state for %d "
362 tor_socket_strerror(tor_socket_errno(conn
->s
)));
366 /** Return true iff <b>conn</b> is linked conn, and reading from the conn
367 * linked to it would be good and feasible. (Reading is "feasible" if the
368 * other conn exists and has data in its outbuf, and is "good" if we have our
369 * reading_from_linked_conn flag set and the other conn has its
370 * writing_to_linked_conn flag set.)*/
372 connection_should_read_from_linked_conn(connection_t
*conn
)
374 if (conn
->linked
&& conn
->reading_from_linked_conn
) {
375 if (! conn
->linked_conn
||
376 (conn
->linked_conn
->writing_to_linked_conn
&&
377 buf_datalen(conn
->linked_conn
->outbuf
)))
383 /** Helper: Tell the main loop to begin reading bytes into <b>conn</b> from
384 * its linked connection, if it is not doing so already. Called by
385 * connection_start_reading and connection_start_writing as appropriate. */
387 connection_start_reading_from_linked_conn(connection_t
*conn
)
390 tor_assert(conn
->linked
== 1);
392 if (!conn
->active_on_link
) {
393 conn
->active_on_link
= 1;
394 smartlist_add(active_linked_connection_lst
, conn
);
395 if (!called_loop_once
) {
396 /* This is the first event on the list; we won't be in LOOP_ONCE mode,
397 * so we need to make sure that the event_loop() actually exits at the
398 * end of its run through the current connections and
399 * lets us activate read events for linked connections. */
400 struct timeval tv
= { 0, 0 };
404 tor_assert(smartlist_isin(active_linked_connection_lst
, conn
));
408 /** Tell the main loop to stop reading bytes into <b>conn</b> from its linked
409 * connection, if is currently doing so. Called by connection_stop_reading,
410 * connection_stop_writing, and connection_read. */
412 connection_stop_reading_from_linked_conn(connection_t
*conn
)
415 tor_assert(conn
->linked
== 1);
417 if (conn
->active_on_link
) {
418 conn
->active_on_link
= 0;
419 /* FFFF We could keep an index here so we can smartlist_del
420 * cleanly. On the other hand, this doesn't show up on profiles,
421 * so let's leave it alone for now. */
422 smartlist_remove(active_linked_connection_lst
, conn
);
424 tor_assert(!smartlist_isin(active_linked_connection_lst
, conn
));
428 /** Close all connections that have been scheduled to get closed. */
430 close_closeable_connections(void)
433 for (i
= 0; i
< smartlist_len(closeable_connection_lst
); ) {
434 connection_t
*conn
= smartlist_get(closeable_connection_lst
, i
);
435 if (conn
->conn_array_index
< 0) {
436 connection_unlink(conn
); /* blow it away right now */
438 if (!conn_close_if_marked(conn
->conn_array_index
))
444 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
445 * some data to read. */
447 conn_read_callback(int fd
, short event
, void *_conn
)
449 connection_t
*conn
= _conn
;
453 log_debug(LD_NET
,"socket %d wants to read.",conn
->s
);
455 assert_connection_ok(conn
, time(NULL
));
457 if (connection_handle_read(conn
) < 0) {
458 if (!conn
->marked_for_close
) {
460 log_warn(LD_BUG
,"Unhandled error on read for %s connection "
462 conn_type_to_string(conn
->type
), conn
->s
);
463 tor_fragile_assert();
465 if (CONN_IS_EDGE(conn
))
466 connection_edge_end_errno(TO_EDGE_CONN(conn
));
467 connection_mark_for_close(conn
);
470 assert_connection_ok(conn
, time(NULL
));
472 if (smartlist_len(closeable_connection_lst
))
473 close_closeable_connections();
476 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
477 * some data to write. */
479 conn_write_callback(int fd
, short events
, void *_conn
)
481 connection_t
*conn
= _conn
;
485 LOG_FN_CONN(conn
, (LOG_DEBUG
, LD_NET
, "socket %d wants to write.",conn
->s
));
487 assert_connection_ok(conn
, time(NULL
));
489 if (connection_handle_write(conn
, 0) < 0) {
490 if (!conn
->marked_for_close
) {
491 /* this connection is broken. remove it. */
492 log_fn(LOG_WARN
,LD_BUG
,
493 "unhandled error on write for %s connection (fd %d); removing",
494 conn_type_to_string(conn
->type
), conn
->s
);
495 tor_fragile_assert();
496 if (CONN_IS_EDGE(conn
)) {
497 /* otherwise we cry wolf about duplicate close */
498 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
499 if (!edge_conn
->end_reason
)
500 edge_conn
->end_reason
= END_STREAM_REASON_INTERNAL
;
501 conn
->edge_has_sent_end
= 1;
503 connection_close_immediate(conn
); /* So we don't try to flush. */
504 connection_mark_for_close(conn
);
507 assert_connection_ok(conn
, time(NULL
));
509 if (smartlist_len(closeable_connection_lst
))
510 close_closeable_connections();
513 /** If the connection at connection_array[i] is marked for close, then:
514 * - If it has data that it wants to flush, try to flush it.
515 * - If it _still_ has data to flush, and conn->hold_open_until_flushed is
516 * true, then leave the connection open and return.
517 * - Otherwise, remove the connection from connection_array and from
518 * all other lists, close it, and free it.
519 * Returns 1 if the connection was closed, 0 otherwise.
522 conn_close_if_marked(int i
)
528 conn
= smartlist_get(connection_array
, i
);
529 if (!conn
->marked_for_close
)
530 return 0; /* nothing to see here, move along */
532 assert_connection_ok(conn
, now
);
533 assert_all_pending_dns_resolves_ok();
535 log_debug(LD_NET
,"Cleaning up connection (fd %d).",conn
->s
);
536 if ((conn
->s
>= 0 || conn
->linked_conn
) && connection_wants_to_flush(conn
)) {
537 /* s == -1 means it's an incomplete edge connection, or that the socket
538 * has already been closed as unflushable. */
539 int sz
= connection_bucket_write_limit(conn
, now
);
540 if (!conn
->hold_open_until_flushed
)
542 "Conn (addr %s, fd %d, type %s, state %d) marked, but wants "
543 "to flush %d bytes. (Marked at %s:%d)",
544 escaped_safe_str(conn
->address
),
545 conn
->s
, conn_type_to_string(conn
->type
), conn
->state
,
546 (int)conn
->outbuf_flushlen
,
547 conn
->marked_for_close_file
, conn
->marked_for_close
);
548 if (conn
->linked_conn
) {
549 retval
= move_buf_to_buf(conn
->linked_conn
->inbuf
, conn
->outbuf
,
550 &conn
->outbuf_flushlen
);
552 /* The linked conn will notice that it has data when it notices that
554 connection_start_reading_from_linked_conn(conn
->linked_conn
);
556 log_debug(LD_GENERAL
, "Flushed last %d bytes from a linked conn; "
557 "%d left; flushlen %d; wants-to-flush==%d", retval
,
558 (int)buf_datalen(conn
->outbuf
),
559 (int)conn
->outbuf_flushlen
,
560 connection_wants_to_flush(conn
));
561 } else if (connection_speaks_cells(conn
)) {
562 if (conn
->state
== OR_CONN_STATE_OPEN
) {
563 retval
= flush_buf_tls(TO_OR_CONN(conn
)->tls
, conn
->outbuf
, sz
,
564 &conn
->outbuf_flushlen
);
566 retval
= -1; /* never flush non-open broken tls connections */
568 retval
= flush_buf(conn
->s
, conn
->outbuf
, sz
, &conn
->outbuf_flushlen
);
570 if (retval
>= 0 && /* Technically, we could survive things like
571 TLS_WANT_WRITE here. But don't bother for now. */
572 conn
->hold_open_until_flushed
&& connection_wants_to_flush(conn
)) {
574 LOG_FN_CONN(conn
, (LOG_INFO
,LD_NET
,
575 "Holding conn (fd %d) open for more flushing.",
577 conn
->timestamp_lastwritten
= now
; /* reset so we can flush more */
581 if (connection_wants_to_flush(conn
)) {
583 if (conn
->type
== CONN_TYPE_EXIT
||
584 (conn
->type
== CONN_TYPE_OR
&& server_mode(get_options())) ||
585 (conn
->type
== CONN_TYPE_DIR
&& conn
->purpose
== DIR_PURPOSE_SERVER
))
588 severity
= LOG_NOTICE
;
589 /* XXXX Maybe allow this to happen a certain amount per hour; it usually
591 log_fn(severity
, LD_NET
, "We stalled too much while trying to write %d "
592 "bytes to address %s. If this happens a lot, either "
593 "something is wrong with your network connection, or "
594 "something is wrong with theirs. "
595 "(fd %d, type %s, state %d, marked at %s:%d).",
596 (int)buf_datalen(conn
->outbuf
),
597 escaped_safe_str(conn
->address
), conn
->s
,
598 conn_type_to_string(conn
->type
), conn
->state
,
599 conn
->marked_for_close_file
,
600 conn
->marked_for_close
);
603 connection_unlink(conn
); /* unlink, remove, free */
607 /** We've just tried every dirserver we know about, and none of
608 * them were reachable. Assume the network is down. Change state
609 * so next time an application connection arrives we'll delay it
610 * and try another directory fetch. Kill off all the circuit_wait
611 * streams that are waiting now, since they will all timeout anyway.
614 directory_all_unreachable(time_t now
)
619 stats_n_seconds_working
=0; /* reset it */
621 while ((conn
= connection_get_by_type_state(CONN_TYPE_AP
,
622 AP_CONN_STATE_CIRCUIT_WAIT
))) {
623 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
625 "Is your network connection down? "
626 "Failing connection to '%s:%d'.",
627 safe_str(edge_conn
->socks_request
->address
),
628 edge_conn
->socks_request
->port
);
629 connection_mark_unattached_ap(edge_conn
,
630 END_STREAM_REASON_NET_UNREACHABLE
);
632 control_event_general_status(LOG_ERR
, "DIR_ALL_UNREACHABLE");
635 /** This function is called whenever we successfully pull down some new
636 * network statuses or server descriptors. */
638 directory_info_has_arrived(time_t now
, int from_cache
)
640 or_options_t
*options
= get_options();
642 if (!router_have_minimum_dir_info()) {
643 log(LOG_NOTICE
, LD_DIR
,
644 "I learned some more directory information, but not enough to "
645 "build a circuit: %s", get_dir_info_status_string());
646 update_router_descriptor_downloads(now
);
649 /* if we have enough dir info, then update our guard status with
650 * whatever we just learned. */
651 entry_guards_compute_status();
652 /* Don't even bother trying to get extrainfo until the rest of our
653 * directory info is up-to-date */
654 if (options
->DownloadExtraInfo
)
655 update_extrainfo_downloads(now
);
658 if (server_mode(options
) && !we_are_hibernating() && !from_cache
&&
659 (has_completed_circuit
|| !any_predicted_circuits(now
)))
660 consider_testing_reachability(1, 1);
663 /** Perform regular maintenance tasks for a single connection. This
664 * function gets run once per second per connection by run_scheduled_events.
667 run_connection_housekeeping(int i
, time_t now
)
670 connection_t
*conn
= smartlist_get(connection_array
, i
);
671 or_options_t
*options
= get_options();
672 or_connection_t
*or_conn
;
674 if (conn
->outbuf
&& !buf_datalen(conn
->outbuf
) && conn
->type
== CONN_TYPE_OR
)
675 TO_OR_CONN(conn
)->timestamp_lastempty
= now
;
677 if (conn
->marked_for_close
) {
678 /* nothing to do here */
682 /* Expire any directory connections that haven't been active (sent
683 * if a server or received if a client) for 5 min */
684 if (conn
->type
== CONN_TYPE_DIR
&&
685 ((DIR_CONN_IS_SERVER(conn
) &&
686 conn
->timestamp_lastwritten
+ DIR_CONN_MAX_STALL
< now
) ||
687 (!DIR_CONN_IS_SERVER(conn
) &&
688 conn
->timestamp_lastread
+ DIR_CONN_MAX_STALL
< now
))) {
689 log_info(LD_DIR
,"Expiring wedged directory conn (fd %d, purpose %d)",
690 conn
->s
, conn
->purpose
);
691 /* This check is temporary; it's to let us know whether we should consider
692 * parsing partial serverdesc responses. */
693 if (conn
->purpose
== DIR_PURPOSE_FETCH_SERVERDESC
&&
694 buf_datalen(conn
->inbuf
)>=1024) {
695 log_info(LD_DIR
,"Trying to extract information from wedged server desc "
697 connection_dir_reached_eof(TO_DIR_CONN(conn
));
699 connection_mark_for_close(conn
);
704 if (!connection_speaks_cells(conn
))
705 return; /* we're all done here, the rest is just for OR conns */
707 or_conn
= TO_OR_CONN(conn
);
709 if (!conn
->or_is_obsolete
) {
710 if (conn
->timestamp_created
+ TIME_BEFORE_OR_CONN_IS_OBSOLETE
< now
) {
712 "Marking OR conn to %s:%d obsolete (fd %d, %d secs old).",
713 conn
->address
, conn
->port
, conn
->s
,
714 (int)(now
- conn
->timestamp_created
));
715 conn
->or_is_obsolete
= 1;
717 or_connection_t
*best
=
718 connection_or_get_by_identity_digest(or_conn
->identity_digest
);
719 if (best
&& best
!= or_conn
&&
720 (conn
->state
== OR_CONN_STATE_OPEN
||
721 now
> conn
->timestamp_created
+ TLS_HANDSHAKE_TIMEOUT
)) {
722 /* We only mark as obsolete connections that already are in
723 * OR_CONN_STATE_OPEN, i.e. that have finished their TLS handshaking.
724 * This is necessary because authorities judge whether a router is
725 * reachable based on whether they were able to TLS handshake with it
726 * recently. Without this check we would expire connections too
727 * early for router->last_reachable to be updated.
730 "Marking duplicate conn to %s:%d obsolete "
731 "(fd %d, %d secs old).",
732 conn
->address
, conn
->port
, conn
->s
,
733 (int)(now
- conn
->timestamp_created
));
734 conn
->or_is_obsolete
= 1;
739 if (conn
->or_is_obsolete
&& !or_conn
->n_circuits
) {
740 /* no unmarked circs -- mark it now */
742 "Expiring non-used OR connection to fd %d (%s:%d) [Obsolete].",
743 conn
->s
, conn
->address
, conn
->port
);
744 connection_mark_for_close(conn
);
745 conn
->hold_open_until_flushed
= 1;
749 /* If we haven't written to an OR connection for a while, then either nuke
750 the connection or send a keepalive, depending. */
751 if (now
>= conn
->timestamp_lastwritten
+ options
->KeepalivePeriod
) {
752 routerinfo_t
*router
= router_get_by_digest(or_conn
->identity_digest
);
753 int maxCircuitlessPeriod
= options
->MaxCircuitDirtiness
*3/2;
754 if (!connection_state_is_open(conn
)) {
755 /* We never managed to actually get this connection open and happy. */
756 log_info(LD_OR
,"Expiring non-open OR connection to fd %d (%s:%d).",
757 conn
->s
,conn
->address
, conn
->port
);
758 connection_mark_for_close(conn
);
759 conn
->hold_open_until_flushed
= 1;
760 } else if (we_are_hibernating() && !or_conn
->n_circuits
&&
761 !buf_datalen(conn
->outbuf
)) {
762 /* We're hibernating, there's no circuits, and nothing to flush.*/
763 log_info(LD_OR
,"Expiring non-used OR connection to fd %d (%s:%d) "
764 "[Hibernating or exiting].",
765 conn
->s
,conn
->address
, conn
->port
);
766 connection_mark_for_close(conn
);
767 conn
->hold_open_until_flushed
= 1;
768 } else if (!clique_mode(options
) && !or_conn
->n_circuits
&&
769 now
>= or_conn
->timestamp_last_added_nonpadding
+
770 maxCircuitlessPeriod
&&
771 (!router
|| !server_mode(options
) ||
772 !router_is_clique_mode(router
))) {
773 log_info(LD_OR
,"Expiring non-used OR connection to fd %d (%s:%d) "
774 "[Not in clique mode].",
775 conn
->s
,conn
->address
, conn
->port
);
776 connection_mark_for_close(conn
);
777 conn
->hold_open_until_flushed
= 1;
779 now
>= or_conn
->timestamp_lastempty
+ options
->KeepalivePeriod
*10 &&
780 now
>= conn
->timestamp_lastwritten
+ options
->KeepalivePeriod
*10) {
781 log_fn(LOG_PROTOCOL_WARN
,LD_PROTOCOL
,
782 "Expiring stuck OR connection to fd %d (%s:%d). (%d bytes to "
783 "flush; %d seconds since last write)",
784 conn
->s
, conn
->address
, conn
->port
,
785 (int)buf_datalen(conn
->outbuf
),
786 (int)(now
-conn
->timestamp_lastwritten
));
787 connection_mark_for_close(conn
);
788 } else if (!buf_datalen(conn
->outbuf
)) {
789 /* either in clique mode, or we've got a circuit. send a padding cell. */
790 log_fn(LOG_DEBUG
,LD_OR
,"Sending keepalive to (%s:%d)",
791 conn
->address
, conn
->port
);
792 memset(&cell
,0,sizeof(cell_t
));
793 cell
.command
= CELL_PADDING
;
794 connection_or_write_cell_to_buf(&cell
, or_conn
);
799 /** Honor a NEWNYM request: make future requests unlinkability to past
802 signewnym_impl(time_t now
)
804 circuit_expire_all_dirty_circs();
805 addressmap_clear_transient();
806 time_of_last_signewnym
= now
;
807 signewnym_is_pending
= 0;
810 /** Perform regular maintenance tasks. This function gets run once per
811 * second by prepare_for_poll.
814 run_scheduled_events(time_t now
)
816 static time_t time_to_fetch_directory
= 0;
817 static time_t time_to_fetch_running_routers
= 0;
818 static time_t last_rotated_x509_certificate
= 0;
819 static time_t time_to_check_v3_certificate
= 0;
820 static time_t time_to_check_listeners
= 0;
821 static time_t time_to_check_descriptor
= 0;
822 static time_t time_to_check_ipaddress
= 0;
823 static time_t time_to_shrink_memory
= 0;
824 static time_t time_to_try_getting_descriptors
= 0;
825 static time_t time_to_reset_descriptor_failures
= 0;
826 static time_t time_to_add_entropy
= 0;
827 static time_t time_to_write_hs_statistics
= 0;
828 static time_t time_to_write_bridge_status_file
= 0;
829 static time_t time_to_downrate_stability
= 0;
830 static time_t time_to_save_stability
= 0;
831 static time_t time_to_clean_caches
= 0;
832 static time_t time_to_recheck_bandwidth
= 0;
833 static time_t time_to_check_for_expired_networkstatus
= 0;
834 or_options_t
*options
= get_options();
838 /** 0. See if we've been asked to shut down and our timeout has
839 * expired; or if our bandwidth limits are exhausted and we
840 * should hibernate; or if it's time to wake up from hibernation.
842 consider_hibernation(now
);
844 /* 0b. If we've deferred a signewnym, make sure it gets handled
846 if (signewnym_is_pending
&&
847 time_of_last_signewnym
+ MAX_SIGNEWNYM_RATE
<= now
) {
848 log(LOG_INFO
, LD_CONTROL
, "Honoring delayed NEWNYM request");
852 /** 1a. Every MIN_ONION_KEY_LIFETIME seconds, rotate the onion keys,
853 * shut down and restart all cpuworkers, and update the directory if
856 if (server_mode(options
) &&
857 get_onion_key_set_at()+MIN_ONION_KEY_LIFETIME
< now
) {
858 log_info(LD_GENERAL
,"Rotating onion key.");
861 if (router_rebuild_descriptor(1)<0) {
862 log_info(LD_CONFIG
, "Couldn't rebuild router descriptor");
864 if (advertised_server_mode())
865 router_upload_dir_desc_to_dirservers(0);
868 if (time_to_try_getting_descriptors
< now
) {
869 update_router_descriptor_downloads(now
);
870 update_extrainfo_downloads(now
);
871 if (options
->UseBridges
)
872 fetch_bridge_descriptors(now
);
873 if (router_have_minimum_dir_info())
874 time_to_try_getting_descriptors
= now
+ LAZY_DESCRIPTOR_RETRY_INTERVAL
;
876 time_to_try_getting_descriptors
= now
+ GREEDY_DESCRIPTOR_RETRY_INTERVAL
;
879 if (time_to_reset_descriptor_failures
< now
) {
880 router_reset_descriptor_download_failures();
881 time_to_reset_descriptor_failures
=
882 now
+ DESCRIPTOR_FAILURE_RESET_INTERVAL
;
885 /** 1b. Every MAX_SSL_KEY_LIFETIME seconds, we change our TLS context. */
886 if (!last_rotated_x509_certificate
)
887 last_rotated_x509_certificate
= now
;
888 if (last_rotated_x509_certificate
+MAX_SSL_KEY_LIFETIME
< now
) {
889 log_info(LD_GENERAL
,"Rotating tls context.");
890 if (tor_tls_context_new(get_identity_key(), MAX_SSL_KEY_LIFETIME
) < 0) {
891 log_warn(LD_BUG
, "Error reinitializing TLS context");
892 /* XXX is it a bug here, that we just keep going? -RD */
894 last_rotated_x509_certificate
= now
;
895 /* We also make sure to rotate the TLS connections themselves if they've
896 * been up for too long -- but that's done via or_is_obsolete in
897 * connection_run_housekeeping() above. */
900 if (time_to_add_entropy
< now
) {
901 if (time_to_add_entropy
) {
902 /* We already seeded once, so don't die on failure. */
905 /** How often do we add more entropy to OpenSSL's RNG pool? */
906 #define ENTROPY_INTERVAL (60*60)
907 time_to_add_entropy
= now
+ ENTROPY_INTERVAL
;
910 /** 1c. If we have to change the accounting interval or record
911 * bandwidth used in this accounting interval, do so. */
912 if (accounting_is_enabled(options
))
913 accounting_run_housekeeping(now
);
915 if (now
% 10 == 0 && (authdir_mode_tests_reachability(options
)) &&
916 !we_are_hibernating()) {
917 /* try to determine reachability of the other Tor relays */
918 dirserv_test_reachability(now
, 0);
921 /** 1d. Periodically, we discount older stability information so that new
922 * stability info counts more, and save the stability information to disk as
924 if (time_to_downrate_stability
< now
)
925 time_to_downrate_stability
= rep_hist_downrate_old_runs(now
);
926 if (authdir_mode_tests_reachability(options
)) {
927 if (time_to_save_stability
< now
) {
928 if (time_to_save_stability
&& rep_hist_record_mtbf_data()<0) {
929 log_warn(LD_GENERAL
, "Couldn't store mtbf data.");
931 #define SAVE_STABILITY_INTERVAL (30*60)
932 time_to_save_stability
= now
+ SAVE_STABILITY_INTERVAL
;
936 /* 1e. Periodicaly, if we're a v3 authority, we check whether our cert is
937 * close to expiring and warn the admin if it is. */
938 if (time_to_check_v3_certificate
< now
) {
939 v3_authority_check_key_expiry();
940 #define CHECK_V3_CERTIFICATE_INTERVAL (5*60)
941 time_to_check_v3_certificate
= now
+ CHECK_V3_CERTIFICATE_INTERVAL
;
944 /* 1f. Check whether our networkstatus has expired.
946 if (time_to_check_for_expired_networkstatus
< now
) {
947 networkstatus_t
*ns
= networkstatus_get_latest_consensus();
948 /*XXXX020 this value needs to be the same as REASONABLY_LIVE_TIME in
949 * networkstatus_get_reasonably_live_consensus(), but that value is way
950 * way too high. Arma: is the bridge issue there resolved yet? -NM */
951 #define NS_EXPIRY_SLOP (24*60*60)
952 if (ns
&& ns
->valid_until
< now
+NS_EXPIRY_SLOP
&&
953 router_have_minimum_dir_info()) {
954 router_dir_info_changed();
956 #define CHECK_EXPIRED_NS_INTERVAL (2*60)
957 time_to_check_for_expired_networkstatus
= now
+ CHECK_EXPIRED_NS_INTERVAL
;
960 /** 2. Periodically, we consider getting a new directory, getting a
961 * new running-routers list, and/or force-uploading our descriptor
962 * (if we've passed our internal checks). */
963 if (time_to_fetch_directory
< now
) {
964 /* Only caches actually need to fetch v1 directories now. */
965 if (directory_fetches_dir_info_early(options
) &&
966 !authdir_mode_v1(options
) && any_trusted_dir_is_v1_authority() &&
967 !should_delay_dir_fetches(options
))
968 directory_get_from_dirserver(DIR_PURPOSE_FETCH_DIR
,
969 ROUTER_PURPOSE_GENERAL
, NULL
, 1);
970 /** How often do we (as a cache) fetch a new V1 directory? */
971 #define V1_DIR_FETCH_PERIOD (12*60*60)
972 time_to_fetch_directory
= now
+ V1_DIR_FETCH_PERIOD
;
975 /* Caches need to fetch running_routers; directory clients don't. */
976 if (time_to_fetch_running_routers
< now
) {
977 if (directory_fetches_dir_info_early(options
) &&
978 !authdir_mode_v1(options
) && any_trusted_dir_is_v1_authority() &&
979 !should_delay_dir_fetches(options
))
980 directory_get_from_dirserver(DIR_PURPOSE_FETCH_RUNNING_LIST
,
981 ROUTER_PURPOSE_GENERAL
, NULL
, 1);
982 /** How often do we (as a cache) fetch a new V1 runningrouters document? */
983 #define V1_RUNNINGROUTERS_FETCH_PERIOD (12*60*60)
984 time_to_fetch_running_routers
= now
+ V1_RUNNINGROUTERS_FETCH_PERIOD
;
987 /* Remove old information from rephist and the rend cache. */
988 if (time_to_clean_caches
< now
) {
989 rep_history_clean(now
- options
->RephistTrackTime
);
991 rend_cache_clean_v2_descs_as_dir();
992 #define CLEAN_CACHES_INTERVAL (30*60)
993 time_to_clean_caches
= now
+ CLEAN_CACHES_INTERVAL
;
996 /** How often do we check whether part of our router info has changed in a way
997 * that would require an upload? */
998 #define CHECK_DESCRIPTOR_INTERVAL (60)
999 /** How often do we (as a router) check whether our IP address has changed? */
1000 #define CHECK_IPADDRESS_INTERVAL (15*60)
1002 /* 2b. Once per minute, regenerate and upload the descriptor if the old
1003 * one is inaccurate. */
1004 if (time_to_check_descriptor
< now
) {
1005 static int dirport_reachability_count
= 0;
1006 time_to_check_descriptor
= now
+ CHECK_DESCRIPTOR_INTERVAL
;
1007 check_descriptor_bandwidth_changed(now
);
1008 if (time_to_check_ipaddress
< now
) {
1009 time_to_check_ipaddress
= now
+ CHECK_IPADDRESS_INTERVAL
;
1010 check_descriptor_ipaddress_changed(now
);
1012 /** If our router descriptor ever goes this long without being regenerated
1013 * because something changed, we force an immediate regenerate-and-upload. */
1014 #define FORCE_REGENERATE_DESCRIPTOR_INTERVAL (18*60*60)
1015 mark_my_descriptor_dirty_if_older_than(
1016 now
- FORCE_REGENERATE_DESCRIPTOR_INTERVAL
);
1017 consider_publishable_server(0);
1018 /* also, check religiously for reachability, if it's within the first
1019 * 20 minutes of our uptime. */
1020 if (server_mode(options
) &&
1021 (has_completed_circuit
|| !any_predicted_circuits(now
)) &&
1022 !we_are_hibernating()) {
1023 if (stats_n_seconds_working
< TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT
) {
1024 consider_testing_reachability(1, dirport_reachability_count
==0);
1025 if (++dirport_reachability_count
> 5)
1026 dirport_reachability_count
= 0;
1027 } else if (time_to_recheck_bandwidth
< now
) {
1028 /* If we haven't checked for 12 hours and our bandwidth estimate is
1029 * low, do another bandwidth test. This is especially important for
1030 * bridges, since they might go long periods without much use. */
1031 routerinfo_t
*me
= router_get_my_routerinfo();
1032 if (time_to_recheck_bandwidth
&& me
&&
1033 me
->bandwidthcapacity
< me
->bandwidthrate
&&
1034 me
->bandwidthcapacity
< 51200) {
1035 reset_bandwidth_test();
1037 #define BANDWIDTH_RECHECK_INTERVAL (12*60*60)
1038 time_to_recheck_bandwidth
= now
+ BANDWIDTH_RECHECK_INTERVAL
;
1042 /* If any networkstatus documents are no longer recent, we need to
1043 * update all the descriptors' running status. */
1044 /* purge obsolete entries */
1045 networkstatus_v2_list_clean(now
);
1046 /* Remove dead routers. */
1047 routerlist_remove_old_routers();
1049 /* Also, once per minute, check whether we want to download any
1050 * networkstatus documents.
1052 update_networkstatus_downloads(now
);
1055 /** 2c. Let directory voting happen. */
1056 if (authdir_mode_v3(options
))
1057 dirvote_act(options
, now
);
1059 /** 3a. Every second, we examine pending circuits and prune the
1060 * ones which have been pending for more than a few seconds.
1061 * We do this before step 4, so it can try building more if
1062 * it's not comfortable with the number of available circuits.
1064 circuit_expire_building(now
);
1066 /** 3b. Also look at pending streams and prune the ones that 'began'
1067 * a long time ago but haven't gotten a 'connected' yet.
1068 * Do this before step 4, so we can put them back into pending
1069 * state to be picked up by the new circuit.
1071 connection_ap_expire_beginning();
1073 /** 3c. And expire connections that we've held open for too long.
1075 connection_expire_held_open();
1077 /** 3d. And every 60 seconds, we relaunch listeners if any died. */
1078 if (!we_are_hibernating() && time_to_check_listeners
< now
) {
1079 retry_all_listeners(NULL
, NULL
);
1080 time_to_check_listeners
= now
+60;
1083 /** 4. Every second, we try a new circuit if there are no valid
1084 * circuits. Every NewCircuitPeriod seconds, we expire circuits
1085 * that became dirty more than MaxCircuitDirtiness seconds ago,
1086 * and we make a new circ if there are no clean circuits.
1088 have_dir_info
= router_have_minimum_dir_info();
1089 if (have_dir_info
&& !we_are_hibernating())
1090 circuit_build_needed_circs(now
);
1092 /** 5. We do housekeeping for each connection... */
1093 for (i
=0;i
<smartlist_len(connection_array
);i
++) {
1094 run_connection_housekeeping(i
, now
);
1096 if (time_to_shrink_memory
< now
) {
1097 SMARTLIST_FOREACH(connection_array
, connection_t
*, conn
, {
1099 buf_shrink(conn
->outbuf
);
1101 buf_shrink(conn
->inbuf
);
1104 buf_shrink_freelists(0);
1105 /** How often do we check buffers and pools for empty space that can be
1107 #define MEM_SHRINK_INTERVAL (60)
1108 time_to_shrink_memory
= now
+ MEM_SHRINK_INTERVAL
;
1111 /** 6. And remove any marked circuits... */
1112 circuit_close_all_marked();
1114 /** 7. And upload service descriptors if necessary. */
1115 if (has_completed_circuit
&& !we_are_hibernating())
1116 rend_consider_services_upload(now
);
1118 /** 8. and blow away any connections that need to die. have to do this now,
1119 * because if we marked a conn for close and left its socket -1, then
1120 * we'll pass it to poll/select and bad things will happen.
1122 close_closeable_connections();
1124 /** 8b. And if anything in our state is ready to get flushed to disk, we
1128 /** 9. and if we're a server, check whether our DNS is telling stories to
1130 if (server_mode(options
) && time_to_check_for_correct_dns
< now
) {
1131 if (!time_to_check_for_correct_dns
) {
1132 time_to_check_for_correct_dns
= now
+ 60 + crypto_rand_int(120);
1134 dns_launch_correctness_checks();
1135 time_to_check_for_correct_dns
= now
+ 12*3600 +
1136 crypto_rand_int(12*3600);
1140 /** 10. write hidden service usage statistic to disk */
1141 if (options
->HSAuthorityRecordStats
&& time_to_write_hs_statistics
< now
) {
1142 hs_usage_write_statistics_to_file(now
);
1143 #define WRITE_HSUSAGE_INTERVAL (30*60)
1144 time_to_write_hs_statistics
= now
+WRITE_HSUSAGE_INTERVAL
;
1146 /** 10b. write bridge networkstatus file to disk */
1147 if (options
->BridgeAuthoritativeDir
&&
1148 time_to_write_bridge_status_file
< now
) {
1149 networkstatus_dump_bridge_status_to_file(now
);
1150 #define BRIDGE_STATUSFILE_INTERVAL (30*60)
1151 time_to_write_bridge_status_file
= now
+BRIDGE_STATUSFILE_INTERVAL
;
1155 /** Libevent timer: used to invoke second_elapsed_callback() once per
1157 static struct event
*timeout_event
= NULL
;
1158 /** Number of libevent errors in the last second: we die if we get too many. */
1159 static int n_libevent_errors
= 0;
1161 /** Libevent callback: invoked once every second. */
1163 second_elapsed_callback(int fd
, short event
, void *args
)
1165 /* XXXX This could be sensibly refactored into multiple callbacks, and we
1166 * could use libevent's timers for this rather than checking the current
1167 * time against a bunch of timeouts every second. */
1168 static struct timeval one_second
;
1169 static long current_second
= 0;
1171 size_t bytes_written
;
1173 int seconds_elapsed
;
1174 or_options_t
*options
= get_options();
1178 if (!timeout_event
) {
1179 timeout_event
= tor_malloc_zero(sizeof(struct event
));
1180 evtimer_set(timeout_event
, second_elapsed_callback
, NULL
);
1181 one_second
.tv_sec
= 1;
1182 one_second
.tv_usec
= 0;
1185 n_libevent_errors
= 0;
1187 /* log_fn(LOG_NOTICE, "Tick."); */
1188 tor_gettimeofday(&now
);
1190 /* the second has rolled over. check more stuff. */
1191 bytes_written
= stats_prev_global_write_bucket
- global_write_bucket
;
1192 bytes_read
= stats_prev_global_read_bucket
- global_read_bucket
;
1193 seconds_elapsed
= current_second
? (int)(now
.tv_sec
- current_second
) : 0;
1194 stats_n_bytes_read
+= bytes_read
;
1195 stats_n_bytes_written
+= bytes_written
;
1196 if (accounting_is_enabled(options
) && seconds_elapsed
>= 0)
1197 accounting_add_bytes(bytes_read
, bytes_written
, seconds_elapsed
);
1198 control_event_bandwidth_used((uint32_t)bytes_read
,(uint32_t)bytes_written
);
1199 control_event_stream_bandwidth_used();
1201 if (seconds_elapsed
> 0)
1202 connection_bucket_refill(seconds_elapsed
, now
.tv_sec
);
1203 stats_prev_global_read_bucket
= global_read_bucket
;
1204 stats_prev_global_write_bucket
= global_write_bucket
;
1206 if (server_mode(options
) &&
1207 !we_are_hibernating() &&
1208 seconds_elapsed
> 0 &&
1209 has_completed_circuit
&&
1210 stats_n_seconds_working
/ TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT
!=
1211 (stats_n_seconds_working
+seconds_elapsed
) /
1212 TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT
) {
1213 /* every 20 minutes, check and complain if necessary */
1214 routerinfo_t
*me
= router_get_my_routerinfo();
1215 if (me
&& !check_whether_orport_reachable())
1216 log_warn(LD_CONFIG
,"Your server (%s:%d) has not managed to confirm that "
1217 "its ORPort is reachable. Please check your firewalls, ports, "
1218 "address, /etc/hosts file, etc.",
1219 me
->address
, me
->or_port
);
1220 if (me
&& !check_whether_dirport_reachable())
1222 "Your server (%s:%d) has not managed to confirm that its "
1223 "DirPort is reachable. Please check your firewalls, ports, "
1224 "address, /etc/hosts file, etc.",
1225 me
->address
, me
->dir_port
);
1228 /** If more than this many seconds have elapsed, probably the clock
1229 * jumped: doesn't count. */
1230 #define NUM_JUMPED_SECONDS_BEFORE_WARN 100
1231 if (seconds_elapsed
< -NUM_JUMPED_SECONDS_BEFORE_WARN
||
1232 seconds_elapsed
>= NUM_JUMPED_SECONDS_BEFORE_WARN
) {
1233 circuit_note_clock_jumped(seconds_elapsed
);
1234 /* XXX if the time jumps *back* many months, do our events in
1235 * run_scheduled_events() recover? I don't think they do. -RD */
1236 } else if (seconds_elapsed
> 0)
1237 stats_n_seconds_working
+= seconds_elapsed
;
1239 run_scheduled_events(now
.tv_sec
);
1241 current_second
= now
.tv_sec
; /* remember which second it is, for next time */
1244 if (current_second
% 300 == 0) {
1245 rep_history_clean(current_second
- options
->RephistTrackTime
);
1246 dumpmemusage(get_min_log_level()<LOG_INFO
?
1247 get_min_log_level() : LOG_INFO
);
1251 if (evtimer_add(timeout_event
, &one_second
))
1253 "Error from libevent when setting one-second timeout event");
1257 /** Called when a possibly ignorable libevent error occurs; ensures that we
1258 * don't get into an infinite loop by ignoring too many errors from
1261 got_libevent_error(void)
1263 if (++n_libevent_errors
> 8) {
1264 log_err(LD_NET
, "Too many libevent errors in one second; dying");
1271 #define UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST (6*60*60)
1273 /** Called when our IP address seems to have changed. <b>at_interface</b>
1274 * should be true if we detected a change in our interface, and false if we
1275 * detected a change in our published address. */
1277 ip_address_changed(int at_interface
)
1279 int server
= server_mode(get_options());
1283 /* Okay, change our keys. */
1288 if (stats_n_seconds_working
> UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST
)
1289 reset_bandwidth_test();
1290 stats_n_seconds_working
= 0;
1291 router_reset_reachability();
1292 mark_my_descriptor_dirty();
1296 dns_servers_relaunch_checks();
1299 /** Forget what we've learned about the correctness of our DNS servers, and
1300 * start learning again. */
1302 dns_servers_relaunch_checks(void)
1304 if (server_mode(get_options())) {
1305 dns_reset_correctness_checks();
1306 time_to_check_for_correct_dns
= 0;
1310 /** Called when we get a SIGHUP: reload configuration files and keys,
1311 * retry all connections, and so on. */
1315 or_options_t
*options
= get_options();
1318 dmalloc_log_stats();
1319 dmalloc_log_changed(0, 1, 0, 0);
1322 log_notice(LD_GENERAL
,"Received reload signal (hup). Reloading config.");
1323 if (accounting_is_enabled(options
))
1324 accounting_record_bandwidth_usage(time(NULL
), get_or_state());
1326 router_reset_warnings();
1327 routerlist_reset_warnings();
1328 addressmap_clear_transient();
1329 /* first, reload config variables, in case they've changed */
1330 /* no need to provide argc/v, they've been cached inside init_from_config */
1331 if (options_init_from_torrc(0, NULL
) < 0) {
1332 log_err(LD_CONFIG
,"Reading config failed--see warnings above. "
1333 "For usage, try -h.");
1336 options
= get_options(); /* they have changed now */
1337 if (authdir_mode_handles_descs(options
, -1)) {
1338 /* reload the approved-routers file */
1339 if (dirserv_load_fingerprint_file() < 0) {
1340 /* warnings are logged from dirserv_load_fingerprint_file() directly */
1341 log_info(LD_GENERAL
, "Error reloading fingerprints. "
1342 "Continuing with old list.");
1346 /* Rotate away from the old dirty circuits. This has to be done
1347 * after we've read the new options, but before we start using
1348 * circuits for directory fetches. */
1349 circuit_expire_all_dirty_circs();
1351 /* retry appropriate downloads */
1352 router_reset_status_download_failures();
1353 router_reset_descriptor_download_failures();
1354 update_networkstatus_downloads(time(NULL
));
1356 /* We'll retry routerstatus downloads in about 10 seconds; no need to
1357 * force a retry there. */
1359 if (server_mode(options
)) {
1360 /* Restart cpuworker and dnsworker processes, so they get up-to-date
1361 * configuration options. */
1362 cpuworkers_rotate();
1368 /** Tor main loop. */
1375 /* initialize dns resolve map, spawn workers if needed */
1376 if (dns_init() < 0) {
1377 log_err(LD_GENERAL
,"Error initializing dns subsystem; exiting");
1383 /* load the private keys, if we're supposed to have them, and set up the
1385 if (! identity_key_is_set()) {
1386 if (init_keys() < 0) {
1387 log_err(LD_BUG
,"Error initializing keys; exiting");
1392 /* Set up the packed_cell_t memory pool. */
1395 /* Set up our buckets */
1396 connection_bucket_init();
1397 stats_prev_global_read_bucket
= global_read_bucket
;
1398 stats_prev_global_write_bucket
= global_write_bucket
;
1400 if (trusted_dirs_reload_certs())
1402 if (router_reload_v2_networkstatus()) {
1405 if (router_reload_consensus_networkstatus()) {
1408 /* load the routers file, or assign the defaults. */
1409 if (router_reload_router_list()) {
1412 /* load the networkstatuses. (This launches a download for new routers as
1416 directory_info_has_arrived(now
, 1);
1418 if (authdir_mode_tests_reachability(get_options())) {
1419 /* the directory is already here, run startup things */
1420 dirserv_test_reachability(now
, 1);
1423 if (server_mode(get_options())) {
1424 /* launch cpuworkers. Need to do this *after* we've read the onion key. */
1428 /* set up once-a-second callback. */
1429 second_elapsed_callback(0,0,NULL
);
1432 if (nt_service_is_stopping())
1436 /* Make it easier to tell whether libevent failure is our fault or not. */
1439 /* All active linked conns should get their read events activated. */
1440 SMARTLIST_FOREACH(active_linked_connection_lst
, connection_t
*, conn
,
1441 event_active(conn
->read_event
, EV_READ
, 1));
1442 called_loop_once
= smartlist_len(active_linked_connection_lst
) ? 1 : 0;
1444 /* poll until we have an event, or the second ends, or until we have
1445 * some active linked connections to trigger events for. */
1446 loop_result
= event_loop(called_loop_once
? EVLOOP_ONCE
: 0);
1448 /* let catch() handle things like ^c, and otherwise don't worry about it */
1449 if (loop_result
< 0) {
1450 int e
= tor_socket_errno(-1);
1451 /* let the program survive things like ^z */
1452 if (e
!= EINTR
&& !ERRNO_IS_EINPROGRESS(e
)) {
1453 #ifdef HAVE_EVENT_GET_METHOD
1454 log_err(LD_NET
,"libevent call with %s failed: %s [%d]",
1455 event_get_method(), tor_socket_strerror(e
), e
);
1457 log_err(LD_NET
,"libevent call failed: %s [%d]",
1458 tor_socket_strerror(e
), e
);
1462 } else if (e
== EINVAL
) {
1463 log_warn(LD_NET
, "EINVAL from libevent: should you upgrade libevent?");
1464 if (got_libevent_error())
1468 if (ERRNO_IS_EINPROGRESS(e
))
1470 "libevent call returned EINPROGRESS? Please report.");
1471 log_debug(LD_NET
,"libevent call interrupted.");
1472 /* You can't trust the results of this poll(). Go back to the
1473 * top of the big for loop. */
1480 /** Used to implement the SIGNAL control command: if we accept
1481 * <b>the_signal</b> as a remote pseudo-signal, act on it. */
1482 /* We don't re-use catch() here because:
1483 * 1. We handle a different set of signals than those allowed in catch.
1484 * 2. Platforms without signal() are unlikely to define SIGfoo.
1485 * 3. The control spec is defined to use fixed numeric signal values
1486 * which just happen to match the unix values.
1489 control_signal_act(int the_signal
)
1494 signal_callback(0,0,(void*)(uintptr_t)SIGHUP
);
1497 signal_callback(0,0,(void*)(uintptr_t)SIGINT
);
1500 signal_callback(0,0,(void*)(uintptr_t)SIGUSR1
);
1503 signal_callback(0,0,(void*)(uintptr_t)SIGUSR2
);
1506 signal_callback(0,0,(void*)(uintptr_t)SIGTERM
);
1509 signal_callback(0,0,(void*)(uintptr_t)SIGNEWNYM
);
1511 case SIGCLEARDNSCACHE
:
1512 signal_callback(0,0,(void*)(uintptr_t)SIGCLEARDNSCACHE
);
1515 log_warn(LD_BUG
, "Unrecognized signal number %d.", the_signal
);
1520 /** Libevent callback: invoked when we get a signal.
1523 signal_callback(int fd
, short events
, void *arg
)
1525 uintptr_t sig
= (uintptr_t)arg
;
1531 log_notice(LD_GENERAL
,"Catching signal TERM, exiting cleanly.");
1536 if (!server_mode(get_options())) { /* do it now */
1537 log_notice(LD_GENERAL
,"Interrupt: exiting cleanly.");
1541 hibernate_begin_shutdown();
1545 log_debug(LD_GENERAL
,"Caught sigpipe. Ignoring.");
1549 /* prefer to log it at INFO, but make sure we always see it */
1550 dumpstats(get_min_log_level()<LOG_INFO
? get_min_log_level() : LOG_INFO
);
1553 switch_logs_debug();
1554 log_debug(LD_GENERAL
,"Caught USR2, going to loglevel debug. "
1555 "Send HUP to change back.");
1559 log_warn(LD_CONFIG
,"Restart failed (config error?). Exiting.");
1566 while (waitpid(-1,NULL
,WNOHANG
) > 0) ; /* keep reaping until no more
1571 time_t now
= time(NULL
);
1572 if (time_of_last_signewnym
+ MAX_SIGNEWNYM_RATE
> now
) {
1573 signewnym_is_pending
= 1;
1574 log(LOG_NOTICE
, LD_CONTROL
,
1575 "Rate limiting NEWNYM request: delaying by %d second(s)",
1576 (int)(MAX_SIGNEWNYM_RATE
+time_of_last_signewnym
-now
));
1578 signewnym_impl(now
);
1582 case SIGCLEARDNSCACHE
:
1583 addressmap_clear_transient();
1588 extern uint64_t rephist_total_alloc
;
1589 extern uint32_t rephist_total_num
;
1592 * Write current memory usage information to the log.
1595 dumpmemusage(int severity
)
1597 connection_dump_buffer_mem_stats(severity
);
1598 log(severity
, LD_GENERAL
, "In rephist: "U64_FORMAT
" used by %d Tors.",
1599 U64_PRINTF_ARG(rephist_total_alloc
), rephist_total_num
);
1600 dump_routerlist_mem_usage(severity
);
1601 dump_cell_pool_usage(severity
);
1602 buf_dump_freelist_sizes(severity
);
1603 tor_log_mallinfo(severity
);
1606 /** Write all statistics to the log, with log level 'severity'. Called
1607 * in response to a SIGUSR1. */
1609 dumpstats(int severity
)
1611 time_t now
= time(NULL
);
1614 log(severity
, LD_GENERAL
, "Dumping stats:");
1616 SMARTLIST_FOREACH(connection_array
, connection_t
*, conn
,
1618 int i
= conn_sl_idx
;
1619 log(severity
, LD_GENERAL
,
1620 "Conn %d (socket %d) type %d (%s), state %d (%s), created %d secs ago",
1621 i
, conn
->s
, conn
->type
, conn_type_to_string(conn
->type
),
1622 conn
->state
, conn_state_to_string(conn
->type
, conn
->state
),
1623 (int)(now
- conn
->timestamp_created
));
1624 if (!connection_is_listener(conn
)) {
1625 log(severity
,LD_GENERAL
,
1626 "Conn %d is to %s:%d.", i
,
1627 safe_str(conn
->address
), conn
->port
);
1628 log(severity
,LD_GENERAL
,
1629 "Conn %d: %d bytes waiting on inbuf (len %d, last read %d secs ago)",
1631 (int)buf_datalen(conn
->inbuf
),
1632 (int)buf_allocation(conn
->inbuf
),
1633 (int)(now
- conn
->timestamp_lastread
));
1634 log(severity
,LD_GENERAL
,
1635 "Conn %d: %d bytes waiting on outbuf "
1636 "(len %d, last written %d secs ago)",i
,
1637 (int)buf_datalen(conn
->outbuf
),
1638 (int)buf_allocation(conn
->outbuf
),
1639 (int)(now
- conn
->timestamp_lastwritten
));
1641 circuit_dump_by_conn(conn
, severity
); /* dump info about all the circuits
1642 * using this conn */
1644 log(severity
, LD_NET
,
1645 "Cells processed: "U64_FORMAT
" padding\n"
1646 " "U64_FORMAT
" create\n"
1647 " "U64_FORMAT
" created\n"
1648 " "U64_FORMAT
" relay\n"
1649 " ("U64_FORMAT
" relayed)\n"
1650 " ("U64_FORMAT
" delivered)\n"
1651 " "U64_FORMAT
" destroy",
1652 U64_PRINTF_ARG(stats_n_padding_cells_processed
),
1653 U64_PRINTF_ARG(stats_n_create_cells_processed
),
1654 U64_PRINTF_ARG(stats_n_created_cells_processed
),
1655 U64_PRINTF_ARG(stats_n_relay_cells_processed
),
1656 U64_PRINTF_ARG(stats_n_relay_cells_relayed
),
1657 U64_PRINTF_ARG(stats_n_relay_cells_delivered
),
1658 U64_PRINTF_ARG(stats_n_destroy_cells_processed
));
1659 if (stats_n_data_cells_packaged
)
1660 log(severity
,LD_NET
,"Average packaged cell fullness: %2.3f%%",
1661 100*(U64_TO_DBL(stats_n_data_bytes_packaged
) /
1662 U64_TO_DBL(stats_n_data_cells_packaged
*RELAY_PAYLOAD_SIZE
)) );
1663 if (stats_n_data_cells_received
)
1664 log(severity
,LD_NET
,"Average delivered cell fullness: %2.3f%%",
1665 100*(U64_TO_DBL(stats_n_data_bytes_received
) /
1666 U64_TO_DBL(stats_n_data_cells_received
*RELAY_PAYLOAD_SIZE
)) );
1668 if (now
- time_of_process_start
>= 0)
1669 elapsed
= now
- time_of_process_start
;
1674 log(severity
, LD_NET
,
1675 "Average bandwidth: "U64_FORMAT
"/%d = %d bytes/sec reading",
1676 U64_PRINTF_ARG(stats_n_bytes_read
),
1678 (int) (stats_n_bytes_read
/elapsed
));
1679 log(severity
, LD_NET
,
1680 "Average bandwidth: "U64_FORMAT
"/%d = %d bytes/sec writing",
1681 U64_PRINTF_ARG(stats_n_bytes_written
),
1683 (int) (stats_n_bytes_written
/elapsed
));
1686 log(severity
, LD_NET
, "--------------- Dumping memory information:");
1687 dumpmemusage(severity
);
1689 rep_hist_dump_stats(now
,severity
);
1690 rend_service_dump_stats(severity
);
1691 dump_pk_ops(severity
);
1692 dump_distinct_digest_count(severity
);
1695 /** Called by exit() as we shut down the process.
1700 /* NOTE: If we ever daemonize, this gets called immediately. That's
1701 * okay for now, because we only use this on Windows. */
1707 /** Set up the signal handlers for either parent or child. */
1709 handle_signals(int is_parent
)
1711 #ifndef MS_WINDOWS /* do signal stuff only on unix */
1713 static int signals
[] = {
1714 SIGINT
, /* do a controlled slow shutdown */
1715 SIGTERM
, /* to terminate now */
1716 SIGPIPE
, /* otherwise sigpipe kills us */
1717 SIGUSR1
, /* dump stats */
1718 SIGUSR2
, /* go to loglevel debug */
1719 SIGHUP
, /* to reload config, retry conns, etc */
1721 SIGXFSZ
, /* handle file-too-big resource exhaustion */
1723 SIGCHLD
, /* handle dns/cpu workers that exit */
1725 static struct event signal_events
[16]; /* bigger than it has to be. */
1727 for (i
= 0; signals
[i
] >= 0; ++i
) {
1728 signal_set(&signal_events
[i
], signals
[i
], signal_callback
,
1729 (void*)(uintptr_t)signals
[i
]);
1730 if (signal_add(&signal_events
[i
], NULL
))
1731 log_warn(LD_BUG
, "Error from libevent when adding event for signal %d",
1735 struct sigaction action
;
1736 action
.sa_flags
= 0;
1737 sigemptyset(&action
.sa_mask
);
1738 action
.sa_handler
= SIG_IGN
;
1739 sigaction(SIGINT
, &action
, NULL
);
1740 sigaction(SIGTERM
, &action
, NULL
);
1741 sigaction(SIGPIPE
, &action
, NULL
);
1742 sigaction(SIGUSR1
, &action
, NULL
);
1743 sigaction(SIGUSR2
, &action
, NULL
);
1744 sigaction(SIGHUP
, &action
, NULL
);
1746 sigaction(SIGXFSZ
, &action
, NULL
);
1749 #else /* MS windows */
1751 #endif /* signal stuff */
1754 /** Main entry point for the Tor command-line client.
1757 tor_init(int argc
, char *argv
[])
1761 time_of_process_start
= time(NULL
);
1762 if (!connection_array
)
1763 connection_array
= smartlist_create();
1764 if (!closeable_connection_lst
)
1765 closeable_connection_lst
= smartlist_create();
1766 if (!active_linked_connection_lst
)
1767 active_linked_connection_lst
= smartlist_create();
1768 /* Have the log set up with our application name. */
1769 tor_snprintf(buf
, sizeof(buf
), "Tor %s", get_version());
1770 log_set_application_name(buf
);
1771 /* Initialize threading. */
1773 /* Initialize the history structures. */
1775 /* Initialize the service cache. */
1777 addressmap_init(); /* Init the client dns cache. Do it always, since it's
1780 /* We search for the "quiet" option first, since it decides whether we
1781 * will log anything at all to the command line. */
1782 for (i
=1;i
<argc
;++i
) {
1783 if (!strcmp(argv
[i
], "--quiet"))
1787 /* give it somewhere to log to initially */
1791 log(LOG_NOTICE
, LD_GENERAL
, "Tor v%s. This is experimental software. "
1792 "Do not rely on it for strong anonymity. (Running on %s)",get_version(),
1795 if (network_init()<0) {
1796 log_err(LD_BUG
,"Error initializing network; exiting.");
1799 atexit(exit_function
);
1801 if (options_init_from_torrc(argc
,argv
) < 0) {
1802 log_err(LD_CONFIG
,"Reading config failed--see warnings above.");
1808 log_warn(LD_GENERAL
,"You are running Tor as root. You don't need to, "
1809 "and you probably shouldn't.");
1812 crypto_global_init(get_options()->HardwareAccel
);
1813 if (crypto_seed_rng()) {
1814 log_err(LD_BUG
, "Unable to seed random number generator. Exiting.");
1821 /** Free all memory that we might have allocated somewhere.
1822 * If <b>postfork</b>, we are a worker process and we want to free
1823 * only the parts of memory that we won't touch. If !<b>postfork</b>,
1824 * Tor is shutting down and we should free everything.
1826 * Helps us find the real leaks with dmalloc and the like. Also valgrind
1827 * should then report 0 reachable in its leak report (in an ideal world --
1828 * in practice libevent, ssl, libc etc never quite free everything). */
1830 tor_free_all(int postfork
)
1837 routerlist_free_all();
1838 networkstatus_free_all();
1839 addressmap_free_all();
1840 set_exit_redirects(NULL
); /* free the registered exit redirects */
1842 rend_service_free_all();
1843 rend_cache_free_all();
1844 rep_hist_free_all();
1845 hs_usage_free_all();
1847 clear_pending_onions();
1849 entry_guards_free_all();
1850 connection_free_all();
1851 buf_shrink_freelists(1);
1852 policies_free_all();
1859 /* stuff in main.c */
1860 smartlist_free(connection_array
);
1861 smartlist_free(closeable_connection_lst
);
1862 smartlist_free(active_linked_connection_lst
);
1863 tor_free(timeout_event
);
1864 /* Stuff in util.c */
1867 esc_router_info(NULL
);
1868 logs_free_all(); /* free log strings. do this last so logs keep working. */
1872 /** Do whatever cleanup is necessary before shutting Tor down. */
1876 or_options_t
*options
= get_options();
1877 /* Remove our pid file. We don't care if there was an error when we
1878 * unlink, nothing we could do about it anyways. */
1879 if (options
->command
== CMD_RUN_TOR
) {
1880 if (options
->PidFile
)
1881 unlink(options
->PidFile
);
1882 if (accounting_is_enabled(options
))
1883 accounting_record_bandwidth_usage(time(NULL
), get_or_state());
1884 or_state_mark_dirty(get_or_state(), 0); /* force an immediate save. */
1885 or_state_save(time(NULL
));
1886 if (authdir_mode_tests_reachability(options
))
1887 rep_hist_record_mtbf_data();
1890 dmalloc_log_stats();
1892 tor_free_all(0); /* We could move tor_free_all back into the ifdef below
1893 later, if it makes shutdown unacceptably slow. But for
1894 now, leave it here: it's helped us catch bugs in the
1896 crypto_global_cleanup();
1898 dmalloc_log_unfreed();
1903 /** Read/create keys as needed, and echo our fingerprint to stdout. */
1905 do_list_fingerprint(void)
1907 char buf
[FINGERPRINT_LEN
+1];
1909 const char *nickname
= get_options()->Nickname
;
1910 if (!server_mode(get_options())) {
1912 "Clients don't have long-term identity keys. Exiting.\n");
1915 tor_assert(nickname
);
1916 if (init_keys() < 0) {
1917 log_err(LD_BUG
,"Error initializing keys; can't display fingerprint");
1920 if (!(k
= get_identity_key())) {
1921 log_err(LD_GENERAL
,"Error: missing identity key.");
1924 if (crypto_pk_get_fingerprint(k
, buf
, 1)<0) {
1925 log_err(LD_BUG
, "Error computing fingerprint");
1928 printf("%s %s\n", nickname
, buf
);
1932 /** Entry point for password hashing: take the desired password from
1933 * the command line, and print its salted hash to stdout. **/
1935 do_hash_password(void)
1939 char key
[S2K_SPECIFIER_LEN
+DIGEST_LEN
];
1941 crypto_rand(key
, S2K_SPECIFIER_LEN
-1);
1942 key
[S2K_SPECIFIER_LEN
-1] = (uint8_t)96; /* Hash 64 K of data. */
1943 secret_to_key(key
+S2K_SPECIFIER_LEN
, DIGEST_LEN
,
1944 get_options()->command_arg
, strlen(get_options()->command_arg
),
1946 base16_encode(output
, sizeof(output
), key
, sizeof(key
));
1947 printf("16:%s\n",output
);
1951 #include <openssl/crypto.h>
1953 _tor_dmalloc_free(void *p
)
1959 /** Main entry point for the Tor process. Called from main(). */
1960 /* This function is distinct from main() only so we can link main.c into
1961 * the unittest binary without conflicting with the unittests' main. */
1963 tor_main(int argc
, char *argv
[])
1967 int r
= CRYPTO_set_mem_ex_functions(_tor_malloc
, _tor_realloc
,
1969 log_notice(LD_CONFIG
, "Set up dmalloc; returned %d", r
);
1975 result
= nt_service_parse_options(argc
, argv
, &done
);
1976 if (done
) return result
;
1979 if (tor_init(argc
, argv
)<0)
1981 switch (get_options()->command
) {
1984 nt_service_set_state(SERVICE_RUNNING
);
1986 result
= do_main_loop();
1988 case CMD_LIST_FINGERPRINT
:
1989 result
= do_list_fingerprint();
1991 case CMD_HASH_PASSWORD
:
1995 case CMD_VERIFY_CONFIG
:
1996 printf("Configuration was valid\n");
1999 case CMD_RUN_UNITTESTS
: /* only set by test.c */
2001 log_warn(LD_BUG
,"Illegal command number %d: internal error.",
2002 get_options()->command
);