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) 2007-2009, The Tor Project, Inc. */
5 /* See LICENSE for licensing information */
9 * \brief Toplevel module. Handles signals, multiplexes between
10 * connections, implements main loop, and drives scheduled events.
17 #include <openssl/crypto.h>
21 void evdns_shutdown(int);
23 /********* PROTOTYPES **********/
25 static void dumpmemusage(int severity
);
26 static void dumpstats(int severity
); /* log stats */
27 static void conn_read_callback(int fd
, short event
, void *_conn
);
28 static void conn_write_callback(int fd
, short event
, void *_conn
);
29 static void signal_callback(int fd
, short events
, void *arg
);
30 static void second_elapsed_callback(int fd
, short event
, void *args
);
31 static int conn_close_if_marked(int i
);
32 static void connection_start_reading_from_linked_conn(connection_t
*conn
);
33 static int connection_should_read_from_linked_conn(connection_t
*conn
);
35 /********* START VARIABLES **********/
37 int global_read_bucket
; /**< Max number of bytes I can read this second. */
38 int global_write_bucket
; /**< Max number of bytes I can write this second. */
40 /** Max number of relayed (bandwidth class 1) bytes I can read this second. */
41 int global_relayed_read_bucket
;
42 /** Max number of relayed (bandwidth class 1) bytes I can write this second. */
43 int global_relayed_write_bucket
;
45 /** What was the read bucket before the last second_elapsed_callback() call?
46 * (used to determine how many bytes we've read). */
47 static int stats_prev_global_read_bucket
;
48 /** What was the write bucket before the last second_elapsed_callback() call?
49 * (used to determine how many bytes we've written). */
50 static int stats_prev_global_write_bucket
;
51 /* XXX we might want to keep stats about global_relayed_*_bucket too. Or not.*/
52 /** How many bytes have we read since we started the process? */
53 static uint64_t stats_n_bytes_read
= 0;
54 /** How many bytes have we written since we started the process? */
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 long do we let OR connections handshake before we decide that
100 * they are obsolete? */
101 #define TLS_HANDSHAKE_TIMEOUT (60)
103 /********* END VARIABLES ************/
105 /****************************************************************************
107 * This section contains accessors and other methods on the connection_array
108 * variables (which are global within this file and unavailable outside it).
110 ****************************************************************************/
112 /** Add <b>conn</b> to the array of connections that we can poll on. The
113 * connection's socket must be set; the connection starts out
114 * non-reading and non-writing.
117 connection_add(connection_t
*conn
)
120 tor_assert(conn
->s
>= 0 ||
122 (conn
->type
== CONN_TYPE_AP
&&
123 TO_EDGE_CONN(conn
)->is_dns_request
));
125 tor_assert(conn
->conn_array_index
== -1); /* can only connection_add once */
126 conn
->conn_array_index
= smartlist_len(connection_array
);
127 smartlist_add(connection_array
, conn
);
129 if (conn
->s
>= 0 || conn
->linked
) {
130 conn
->read_event
= tor_malloc_zero(sizeof(struct event
));
131 conn
->write_event
= tor_malloc_zero(sizeof(struct event
));
132 event_set(conn
->read_event
, conn
->s
, EV_READ
|EV_PERSIST
,
133 conn_read_callback
, conn
);
134 event_set(conn
->write_event
, conn
->s
, EV_WRITE
|EV_PERSIST
,
135 conn_write_callback
, conn
);
138 log_debug(LD_NET
,"new conn type %s, socket %d, address %s, n_conns %d.",
139 conn_type_to_string(conn
->type
), conn
->s
, conn
->address
,
140 smartlist_len(connection_array
));
145 /** Remove the connection from the global list, and remove the
146 * corresponding poll entry. Calling this function will shift the last
147 * connection (if any) into the position occupied by conn.
150 connection_remove(connection_t
*conn
)
157 log_debug(LD_NET
,"removing socket %d (type %s), n_conns now %d",
158 conn
->s
, conn_type_to_string(conn
->type
),
159 smartlist_len(connection_array
));
161 tor_assert(conn
->conn_array_index
>= 0);
162 current_index
= conn
->conn_array_index
;
163 connection_unregister_events(conn
); /* This is redundant, but cheap. */
164 if (current_index
== smartlist_len(connection_array
)-1) { /* at the end */
165 smartlist_del(connection_array
, current_index
);
169 /* replace this one with the one at the end */
170 smartlist_del(connection_array
, current_index
);
171 tmp
= smartlist_get(connection_array
, current_index
);
172 tmp
->conn_array_index
= current_index
;
177 /** If <b>conn</b> is an edge conn, remove it from the list
178 * of conn's on this circuit. If it's not on an edge,
179 * flush and send destroys for all circuits on this conn.
181 * Remove it from connection_array (if applicable) and
182 * from closeable_connection_list.
187 connection_unlink(connection_t
*conn
)
189 connection_about_to_close_connection(conn
);
190 if (conn
->conn_array_index
>= 0) {
191 connection_remove(conn
);
193 if (conn
->linked_conn
) {
194 conn
->linked_conn
->linked_conn
= NULL
;
195 if (! conn
->linked_conn
->marked_for_close
&&
196 conn
->linked_conn
->reading_from_linked_conn
)
197 connection_start_reading(conn
->linked_conn
);
198 conn
->linked_conn
= NULL
;
200 smartlist_remove(closeable_connection_lst
, conn
);
201 smartlist_remove(active_linked_connection_lst
, conn
);
202 if (conn
->type
== CONN_TYPE_EXIT
) {
203 assert_connection_edge_not_dns_pending(TO_EDGE_CONN(conn
));
205 if (conn
->type
== CONN_TYPE_OR
) {
206 if (!tor_digest_is_zero(TO_OR_CONN(conn
)->identity_digest
))
207 connection_or_remove_from_identity_map(TO_OR_CONN(conn
));
209 connection_free(conn
);
212 /** Schedule <b>conn</b> to be closed. **/
214 add_connection_to_closeable_list(connection_t
*conn
)
216 tor_assert(!smartlist_isin(closeable_connection_lst
, conn
));
217 tor_assert(conn
->marked_for_close
);
218 assert_connection_ok(conn
, time(NULL
));
219 smartlist_add(closeable_connection_lst
, conn
);
222 /** Return 1 if conn is on the closeable list, else return 0. */
224 connection_is_on_closeable_list(connection_t
*conn
)
226 return smartlist_isin(closeable_connection_lst
, conn
);
229 /** Return true iff conn is in the current poll array. */
231 connection_in_array(connection_t
*conn
)
233 return smartlist_isin(connection_array
, conn
);
236 /** Set <b>*array</b> to an array of all connections, and <b>*n</b>
237 * to the length of the array. <b>*array</b> and <b>*n</b> must not
241 get_connection_array(void)
243 if (!connection_array
)
244 connection_array
= smartlist_create();
245 return connection_array
;
248 /** Set the event mask on <b>conn</b> to <b>events</b>. (The event
249 * mask is a bitmask whose bits are EV_READ and EV_WRITE.)
252 connection_watch_events(connection_t
*conn
, short events
)
254 if (events
& EV_READ
)
255 connection_start_reading(conn
);
257 connection_stop_reading(conn
);
259 if (events
& EV_WRITE
)
260 connection_start_writing(conn
);
262 connection_stop_writing(conn
);
265 /** Return true iff <b>conn</b> is listening for read events. */
267 connection_is_reading(connection_t
*conn
)
271 return conn
->reading_from_linked_conn
||
272 (conn
->read_event
&& event_pending(conn
->read_event
, EV_READ
, NULL
));
275 /** Tell the main loop to stop notifying <b>conn</b> of any read events. */
277 connection_stop_reading(connection_t
*conn
)
280 tor_assert(conn
->read_event
);
283 conn
->reading_from_linked_conn
= 0;
284 connection_stop_reading_from_linked_conn(conn
);
286 if (event_del(conn
->read_event
))
287 log_warn(LD_NET
, "Error from libevent setting read event state for %d "
290 tor_socket_strerror(tor_socket_errno(conn
->s
)));
294 /** Tell the main loop to start notifying <b>conn</b> of any read events. */
296 connection_start_reading(connection_t
*conn
)
299 tor_assert(conn
->read_event
);
302 conn
->reading_from_linked_conn
= 1;
303 if (connection_should_read_from_linked_conn(conn
))
304 connection_start_reading_from_linked_conn(conn
);
306 if (event_add(conn
->read_event
, NULL
))
307 log_warn(LD_NET
, "Error from libevent setting read event state for %d "
310 tor_socket_strerror(tor_socket_errno(conn
->s
)));
314 /** Return true iff <b>conn</b> is listening for write events. */
316 connection_is_writing(connection_t
*conn
)
320 return conn
->writing_to_linked_conn
||
321 (conn
->write_event
&& event_pending(conn
->write_event
, EV_WRITE
, NULL
));
324 /** Tell the main loop to stop notifying <b>conn</b> of any write events. */
326 connection_stop_writing(connection_t
*conn
)
329 tor_assert(conn
->write_event
);
332 conn
->writing_to_linked_conn
= 0;
333 if (conn
->linked_conn
)
334 connection_stop_reading_from_linked_conn(conn
->linked_conn
);
336 if (event_del(conn
->write_event
))
337 log_warn(LD_NET
, "Error from libevent setting write event state for %d "
340 tor_socket_strerror(tor_socket_errno(conn
->s
)));
344 /** Tell the main loop to start notifying <b>conn</b> of any write events. */
346 connection_start_writing(connection_t
*conn
)
349 tor_assert(conn
->write_event
);
352 conn
->writing_to_linked_conn
= 1;
353 if (conn
->linked_conn
&&
354 connection_should_read_from_linked_conn(conn
->linked_conn
))
355 connection_start_reading_from_linked_conn(conn
->linked_conn
);
357 if (event_add(conn
->write_event
, NULL
))
358 log_warn(LD_NET
, "Error from libevent setting write event state for %d "
361 tor_socket_strerror(tor_socket_errno(conn
->s
)));
365 /** Return true iff <b>conn</b> is linked conn, and reading from the conn
366 * linked to it would be good and feasible. (Reading is "feasible" if the
367 * other conn exists and has data in its outbuf, and is "good" if we have our
368 * reading_from_linked_conn flag set and the other conn has its
369 * writing_to_linked_conn flag set.)*/
371 connection_should_read_from_linked_conn(connection_t
*conn
)
373 if (conn
->linked
&& conn
->reading_from_linked_conn
) {
374 if (! conn
->linked_conn
||
375 (conn
->linked_conn
->writing_to_linked_conn
&&
376 buf_datalen(conn
->linked_conn
->outbuf
)))
382 /** Helper: Tell the main loop to begin reading bytes into <b>conn</b> from
383 * its linked connection, if it is not doing so already. Called by
384 * connection_start_reading and connection_start_writing as appropriate. */
386 connection_start_reading_from_linked_conn(connection_t
*conn
)
389 tor_assert(conn
->linked
== 1);
391 if (!conn
->active_on_link
) {
392 conn
->active_on_link
= 1;
393 smartlist_add(active_linked_connection_lst
, conn
);
394 if (!called_loop_once
) {
395 /* This is the first event on the list; we won't be in LOOP_ONCE mode,
396 * so we need to make sure that the event_loop() actually exits at the
397 * end of its run through the current connections and
398 * lets us activate read events for linked connections. */
399 struct timeval tv
= { 0, 0 };
403 tor_assert(smartlist_isin(active_linked_connection_lst
, conn
));
407 /** Tell the main loop to stop reading bytes into <b>conn</b> from its linked
408 * connection, if is currently doing so. Called by connection_stop_reading,
409 * connection_stop_writing, and connection_read. */
411 connection_stop_reading_from_linked_conn(connection_t
*conn
)
414 tor_assert(conn
->linked
== 1);
416 if (conn
->active_on_link
) {
417 conn
->active_on_link
= 0;
418 /* FFFF We could keep an index here so we can smartlist_del
419 * cleanly. On the other hand, this doesn't show up on profiles,
420 * so let's leave it alone for now. */
421 smartlist_remove(active_linked_connection_lst
, conn
);
423 tor_assert(!smartlist_isin(active_linked_connection_lst
, conn
));
427 /** Close all connections that have been scheduled to get closed. */
429 close_closeable_connections(void)
432 for (i
= 0; i
< smartlist_len(closeable_connection_lst
); ) {
433 connection_t
*conn
= smartlist_get(closeable_connection_lst
, i
);
434 if (conn
->conn_array_index
< 0) {
435 connection_unlink(conn
); /* blow it away right now */
437 if (!conn_close_if_marked(conn
->conn_array_index
))
443 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
444 * some data to read. */
446 conn_read_callback(int fd
, short event
, void *_conn
)
448 connection_t
*conn
= _conn
;
452 log_debug(LD_NET
,"socket %d wants to read.",conn
->s
);
454 /* assert_connection_ok(conn, time(NULL)); */
456 if (connection_handle_read(conn
) < 0) {
457 if (!conn
->marked_for_close
) {
459 log_warn(LD_BUG
,"Unhandled error on read for %s connection "
461 conn_type_to_string(conn
->type
), conn
->s
);
462 tor_fragile_assert();
464 if (CONN_IS_EDGE(conn
))
465 connection_edge_end_errno(TO_EDGE_CONN(conn
));
466 connection_mark_for_close(conn
);
469 assert_connection_ok(conn
, time(NULL
));
471 if (smartlist_len(closeable_connection_lst
))
472 close_closeable_connections();
475 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
476 * some data to write. */
478 conn_write_callback(int fd
, short events
, void *_conn
)
480 connection_t
*conn
= _conn
;
484 LOG_FN_CONN(conn
, (LOG_DEBUG
, LD_NET
, "socket %d wants to write.",conn
->s
));
486 /* assert_connection_ok(conn, time(NULL)); */
488 if (connection_handle_write(conn
, 0) < 0) {
489 if (!conn
->marked_for_close
) {
490 /* this connection is broken. remove it. */
491 log_fn(LOG_WARN
,LD_BUG
,
492 "unhandled error on write for %s connection (fd %d); removing",
493 conn_type_to_string(conn
->type
), conn
->s
);
494 tor_fragile_assert();
495 if (CONN_IS_EDGE(conn
)) {
496 /* otherwise we cry wolf about duplicate close */
497 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
498 if (!edge_conn
->end_reason
)
499 edge_conn
->end_reason
= END_STREAM_REASON_INTERNAL
;
500 edge_conn
->edge_has_sent_end
= 1;
502 connection_close_immediate(conn
); /* So we don't try to flush. */
503 connection_mark_for_close(conn
);
506 assert_connection_ok(conn
, time(NULL
));
508 if (smartlist_len(closeable_connection_lst
))
509 close_closeable_connections();
512 /** If the connection at connection_array[i] is marked for close, then:
513 * - If it has data that it wants to flush, try to flush it.
514 * - If it _still_ has data to flush, and conn->hold_open_until_flushed is
515 * true, then leave the connection open and return.
516 * - Otherwise, remove the connection from connection_array and from
517 * all other lists, close it, and free it.
518 * Returns 1 if the connection was closed, 0 otherwise.
521 conn_close_if_marked(int i
)
527 conn
= smartlist_get(connection_array
, i
);
528 if (!conn
->marked_for_close
)
529 return 0; /* nothing to see here, move along */
531 assert_connection_ok(conn
, now
);
532 /* assert_all_pending_dns_resolves_ok(); */
534 log_debug(LD_NET
,"Cleaning up connection (fd %d).",conn
->s
);
535 if ((conn
->s
>= 0 || conn
->linked_conn
) && connection_wants_to_flush(conn
)) {
536 /* s == -1 means it's an incomplete edge connection, or that the socket
537 * has already been closed as unflushable. */
538 ssize_t sz
= connection_bucket_write_limit(conn
, now
);
539 if (!conn
->hold_open_until_flushed
)
541 "Conn (addr %s, fd %d, type %s, state %d) marked, but wants "
542 "to flush %d bytes. (Marked at %s:%d)",
543 escaped_safe_str(conn
->address
),
544 conn
->s
, conn_type_to_string(conn
->type
), conn
->state
,
545 (int)conn
->outbuf_flushlen
,
546 conn
->marked_for_close_file
, conn
->marked_for_close
);
547 if (conn
->linked_conn
) {
548 retval
= move_buf_to_buf(conn
->linked_conn
->inbuf
, conn
->outbuf
,
549 &conn
->outbuf_flushlen
);
551 /* The linked conn will notice that it has data when it notices that
553 connection_start_reading_from_linked_conn(conn
->linked_conn
);
555 log_debug(LD_GENERAL
, "Flushed last %d bytes from a linked conn; "
556 "%d left; flushlen %d; wants-to-flush==%d", retval
,
557 (int)buf_datalen(conn
->outbuf
),
558 (int)conn
->outbuf_flushlen
,
559 connection_wants_to_flush(conn
));
560 } else if (connection_speaks_cells(conn
)) {
561 if (conn
->state
== OR_CONN_STATE_OPEN
) {
562 retval
= flush_buf_tls(TO_OR_CONN(conn
)->tls
, conn
->outbuf
, sz
,
563 &conn
->outbuf_flushlen
);
565 retval
= -1; /* never flush non-open broken tls connections */
567 retval
= flush_buf(conn
->s
, conn
->outbuf
, sz
, &conn
->outbuf_flushlen
);
569 if (retval
>= 0 && /* Technically, we could survive things like
570 TLS_WANT_WRITE here. But don't bother for now. */
571 conn
->hold_open_until_flushed
&& connection_wants_to_flush(conn
)) {
573 LOG_FN_CONN(conn
, (LOG_INFO
,LD_NET
,
574 "Holding conn (fd %d) open for more flushing.",
576 conn
->timestamp_lastwritten
= now
; /* reset so we can flush more */
580 if (connection_wants_to_flush(conn
)) {
582 if (conn
->type
== CONN_TYPE_EXIT
||
583 (conn
->type
== CONN_TYPE_OR
&& server_mode(get_options())) ||
584 (conn
->type
== CONN_TYPE_DIR
&& conn
->purpose
== DIR_PURPOSE_SERVER
))
587 severity
= LOG_NOTICE
;
588 /* XXXX Maybe allow this to happen a certain amount per hour; it usually
590 log_fn(severity
, LD_NET
, "We stalled too much while trying to write %d "
591 "bytes to address %s. If this happens a lot, either "
592 "something is wrong with your network connection, or "
593 "something is wrong with theirs. "
594 "(fd %d, type %s, state %d, marked at %s:%d).",
595 (int)buf_datalen(conn
->outbuf
),
596 escaped_safe_str(conn
->address
), conn
->s
,
597 conn_type_to_string(conn
->type
), conn
->state
,
598 conn
->marked_for_close_file
,
599 conn
->marked_for_close
);
602 connection_unlink(conn
); /* unlink, remove, free */
606 /** We've just tried every dirserver we know about, and none of
607 * them were reachable. Assume the network is down. Change state
608 * so next time an application connection arrives we'll delay it
609 * and try another directory fetch. Kill off all the circuit_wait
610 * streams that are waiting now, since they will all timeout anyway.
613 directory_all_unreachable(time_t now
)
618 stats_n_seconds_working
=0; /* reset it */
620 while ((conn
= connection_get_by_type_state(CONN_TYPE_AP
,
621 AP_CONN_STATE_CIRCUIT_WAIT
))) {
622 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
624 "Is your network connection down? "
625 "Failing connection to '%s:%d'.",
626 safe_str(edge_conn
->socks_request
->address
),
627 edge_conn
->socks_request
->port
);
628 connection_mark_unattached_ap(edge_conn
,
629 END_STREAM_REASON_NET_UNREACHABLE
);
631 control_event_general_status(LOG_ERR
, "DIR_ALL_UNREACHABLE");
634 /** This function is called whenever we successfully pull down some new
635 * network statuses or server descriptors. */
637 directory_info_has_arrived(time_t now
, int from_cache
)
639 or_options_t
*options
= get_options();
641 if (!router_have_minimum_dir_info()) {
642 int quiet
= directory_too_idle_to_fetch_descriptors(options
, now
);
643 log(quiet
? LOG_INFO
: 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 (directory_fetches_from_authorities(options
))
650 update_router_descriptor_downloads(now
);
652 /* if we have enough dir info, then update our guard status with
653 * whatever we just learned. */
654 entry_guards_compute_status();
655 /* Don't even bother trying to get extrainfo until the rest of our
656 * directory info is up-to-date */
657 if (options
->DownloadExtraInfo
)
658 update_extrainfo_downloads(now
);
661 if (server_mode(options
) && !we_are_hibernating() && !from_cache
&&
662 (has_completed_circuit
|| !any_predicted_circuits(now
)))
663 consider_testing_reachability(1, 1);
666 /** Perform regular maintenance tasks for a single connection. This
667 * function gets run once per second per connection by run_scheduled_events.
670 run_connection_housekeeping(int i
, time_t now
)
673 connection_t
*conn
= smartlist_get(connection_array
, i
);
674 or_options_t
*options
= get_options();
675 or_connection_t
*or_conn
;
677 if (conn
->outbuf
&& !buf_datalen(conn
->outbuf
) && conn
->type
== CONN_TYPE_OR
)
678 TO_OR_CONN(conn
)->timestamp_lastempty
= now
;
680 if (conn
->marked_for_close
) {
681 /* nothing to do here */
685 /* Expire any directory connections that haven't been active (sent
686 * if a server or received if a client) for 5 min */
687 if (conn
->type
== CONN_TYPE_DIR
&&
688 ((DIR_CONN_IS_SERVER(conn
) &&
689 conn
->timestamp_lastwritten
+ DIR_CONN_MAX_STALL
< now
) ||
690 (!DIR_CONN_IS_SERVER(conn
) &&
691 conn
->timestamp_lastread
+ DIR_CONN_MAX_STALL
< now
))) {
692 log_info(LD_DIR
,"Expiring wedged directory conn (fd %d, purpose %d)",
693 conn
->s
, conn
->purpose
);
694 /* This check is temporary; it's to let us know whether we should consider
695 * parsing partial serverdesc responses. */
696 if (conn
->purpose
== DIR_PURPOSE_FETCH_SERVERDESC
&&
697 buf_datalen(conn
->inbuf
)>=1024) {
698 log_info(LD_DIR
,"Trying to extract information from wedged server desc "
700 connection_dir_reached_eof(TO_DIR_CONN(conn
));
702 connection_mark_for_close(conn
);
707 if (!connection_speaks_cells(conn
))
708 return; /* we're all done here, the rest is just for OR conns */
710 or_conn
= TO_OR_CONN(conn
);
712 if (or_conn
->is_bad_for_new_circs
&& !or_conn
->n_circuits
) {
713 /* It's bad for new circuits, and has no unmarked circuits on it:
716 "Expiring non-used OR connection to fd %d (%s:%d) [Too old].",
717 conn
->s
, conn
->address
, conn
->port
);
718 if (conn
->state
== OR_CONN_STATE_CONNECTING
)
719 connection_or_connect_failed(TO_OR_CONN(conn
),
720 END_OR_CONN_REASON_TIMEOUT
,
721 "Tor gave up on the connection");
722 connection_mark_for_close(conn
);
723 conn
->hold_open_until_flushed
= 1;
727 /* If we haven't written to an OR connection for a while, then either nuke
728 the connection or send a keepalive, depending. */
729 if (now
>= conn
->timestamp_lastwritten
+ options
->KeepalivePeriod
) {
730 routerinfo_t
*router
= router_get_by_digest(or_conn
->identity_digest
);
731 int maxCircuitlessPeriod
= options
->MaxCircuitDirtiness
*3/2;
732 if (!connection_state_is_open(conn
)) {
733 /* We never managed to actually get this connection open and happy. */
734 log_info(LD_OR
,"Expiring non-open OR connection to fd %d (%s:%d).",
735 conn
->s
,conn
->address
, conn
->port
);
736 connection_mark_for_close(conn
);
737 conn
->hold_open_until_flushed
= 1;
738 } else if (we_are_hibernating() && !or_conn
->n_circuits
&&
739 !buf_datalen(conn
->outbuf
)) {
740 /* We're hibernating, there's no circuits, and nothing to flush.*/
741 log_info(LD_OR
,"Expiring non-used OR connection to fd %d (%s:%d) "
742 "[Hibernating or exiting].",
743 conn
->s
,conn
->address
, conn
->port
);
744 connection_mark_for_close(conn
);
745 conn
->hold_open_until_flushed
= 1;
746 } else if (!clique_mode(options
) && !or_conn
->n_circuits
&&
747 now
>= or_conn
->timestamp_last_added_nonpadding
+
748 maxCircuitlessPeriod
&&
749 (!router
|| !server_mode(options
) ||
750 !router_is_clique_mode(router
))) {
751 log_info(LD_OR
,"Expiring non-used OR connection to fd %d (%s:%d) "
752 "[Not in clique mode].",
753 conn
->s
,conn
->address
, conn
->port
);
754 connection_mark_for_close(conn
);
755 conn
->hold_open_until_flushed
= 1;
757 now
>= or_conn
->timestamp_lastempty
+ options
->KeepalivePeriod
*10 &&
758 now
>= conn
->timestamp_lastwritten
+ options
->KeepalivePeriod
*10) {
759 log_fn(LOG_PROTOCOL_WARN
,LD_PROTOCOL
,
760 "Expiring stuck OR connection to fd %d (%s:%d). (%d bytes to "
761 "flush; %d seconds since last write)",
762 conn
->s
, conn
->address
, conn
->port
,
763 (int)buf_datalen(conn
->outbuf
),
764 (int)(now
-conn
->timestamp_lastwritten
));
765 connection_mark_for_close(conn
);
766 } else if (!buf_datalen(conn
->outbuf
)) {
767 /* either in clique mode, or we've got a circuit. send a padding cell. */
768 log_fn(LOG_DEBUG
,LD_OR
,"Sending keepalive to (%s:%d)",
769 conn
->address
, conn
->port
);
770 memset(&cell
,0,sizeof(cell_t
));
771 cell
.command
= CELL_PADDING
;
772 connection_or_write_cell_to_buf(&cell
, or_conn
);
777 /** Honor a NEWNYM request: make future requests unlinkability to past
780 signewnym_impl(time_t now
)
782 circuit_expire_all_dirty_circs();
783 addressmap_clear_transient();
784 time_of_last_signewnym
= now
;
785 signewnym_is_pending
= 0;
788 /** Perform regular maintenance tasks. This function gets run once per
789 * second by second_elapsed_callback().
792 run_scheduled_events(time_t now
)
794 static time_t last_rotated_x509_certificate
= 0;
795 static time_t time_to_check_v3_certificate
= 0;
796 static time_t time_to_check_listeners
= 0;
797 static time_t time_to_check_descriptor
= 0;
798 static time_t time_to_check_ipaddress
= 0;
799 static time_t time_to_shrink_memory
= 0;
800 static time_t time_to_try_getting_descriptors
= 0;
801 static time_t time_to_reset_descriptor_failures
= 0;
802 static time_t time_to_add_entropy
= 0;
803 static time_t time_to_write_hs_statistics
= 0;
804 static time_t time_to_write_bridge_status_file
= 0;
805 static time_t time_to_downrate_stability
= 0;
806 static time_t time_to_save_stability
= 0;
807 static time_t time_to_clean_caches
= 0;
808 static time_t time_to_recheck_bandwidth
= 0;
809 static time_t time_to_check_for_expired_networkstatus
= 0;
810 static time_t time_to_dump_geoip_stats
= 0;
811 static time_t time_to_retry_dns_init
= 0;
812 or_options_t
*options
= get_options();
816 /** 0. See if we've been asked to shut down and our timeout has
817 * expired; or if our bandwidth limits are exhausted and we
818 * should hibernate; or if it's time to wake up from hibernation.
820 consider_hibernation(now
);
822 /* 0b. If we've deferred a signewnym, make sure it gets handled
824 if (signewnym_is_pending
&&
825 time_of_last_signewnym
+ MAX_SIGNEWNYM_RATE
<= now
) {
826 log(LOG_INFO
, LD_CONTROL
, "Honoring delayed NEWNYM request");
830 /** 1a. Every MIN_ONION_KEY_LIFETIME seconds, rotate the onion keys,
831 * shut down and restart all cpuworkers, and update the directory if
834 if (server_mode(options
) &&
835 get_onion_key_set_at()+MIN_ONION_KEY_LIFETIME
< now
) {
836 log_info(LD_GENERAL
,"Rotating onion key.");
839 if (router_rebuild_descriptor(1)<0) {
840 log_info(LD_CONFIG
, "Couldn't rebuild router descriptor");
842 if (advertised_server_mode())
843 router_upload_dir_desc_to_dirservers(0);
846 if (time_to_try_getting_descriptors
< now
) {
847 update_router_descriptor_downloads(now
);
848 update_extrainfo_downloads(now
);
849 if (options
->UseBridges
)
850 fetch_bridge_descriptors(now
);
851 if (router_have_minimum_dir_info())
852 time_to_try_getting_descriptors
= now
+ LAZY_DESCRIPTOR_RETRY_INTERVAL
;
854 time_to_try_getting_descriptors
= now
+ GREEDY_DESCRIPTOR_RETRY_INTERVAL
;
857 if (time_to_reset_descriptor_failures
< now
) {
858 router_reset_descriptor_download_failures();
859 time_to_reset_descriptor_failures
=
860 now
+ DESCRIPTOR_FAILURE_RESET_INTERVAL
;
863 /** 1b. Every MAX_SSL_KEY_LIFETIME seconds, we change our TLS context. */
864 if (!last_rotated_x509_certificate
)
865 last_rotated_x509_certificate
= now
;
866 if (last_rotated_x509_certificate
+MAX_SSL_KEY_LIFETIME
< now
) {
867 log_info(LD_GENERAL
,"Rotating tls context.");
868 if (tor_tls_context_new(get_identity_key(), MAX_SSL_KEY_LIFETIME
) < 0) {
869 log_warn(LD_BUG
, "Error reinitializing TLS context");
870 /* XXX is it a bug here, that we just keep going? -RD */
872 last_rotated_x509_certificate
= now
;
873 /* We also make sure to rotate the TLS connections themselves if they've
874 * been up for too long -- but that's done via is_bad_for_new_circs in
875 * connection_run_housekeeping() above. */
878 if (time_to_add_entropy
< now
) {
879 if (time_to_add_entropy
) {
880 /* We already seeded once, so don't die on failure. */
883 /** How often do we add more entropy to OpenSSL's RNG pool? */
884 #define ENTROPY_INTERVAL (60*60)
885 time_to_add_entropy
= now
+ ENTROPY_INTERVAL
;
888 /** 1c. If we have to change the accounting interval or record
889 * bandwidth used in this accounting interval, do so. */
890 if (accounting_is_enabled(options
))
891 accounting_run_housekeeping(now
);
893 if (now
% 10 == 0 && (authdir_mode_tests_reachability(options
)) &&
894 !we_are_hibernating()) {
895 /* try to determine reachability of the other Tor relays */
896 dirserv_test_reachability(now
, 0);
899 /** 1d. Periodically, we discount older stability information so that new
900 * stability info counts more, and save the stability information to disk as
902 if (time_to_downrate_stability
< now
)
903 time_to_downrate_stability
= rep_hist_downrate_old_runs(now
);
904 if (authdir_mode_tests_reachability(options
)) {
905 if (time_to_save_stability
< now
) {
906 if (time_to_save_stability
&& rep_hist_record_mtbf_data(now
, 1)<0) {
907 log_warn(LD_GENERAL
, "Couldn't store mtbf data.");
909 #define SAVE_STABILITY_INTERVAL (30*60)
910 time_to_save_stability
= now
+ SAVE_STABILITY_INTERVAL
;
914 /* 1e. Periodically, if we're a v3 authority, we check whether our cert is
915 * close to expiring and warn the admin if it is. */
916 if (time_to_check_v3_certificate
< now
) {
917 v3_authority_check_key_expiry();
918 #define CHECK_V3_CERTIFICATE_INTERVAL (5*60)
919 time_to_check_v3_certificate
= now
+ CHECK_V3_CERTIFICATE_INTERVAL
;
922 /* 1f. Check whether our networkstatus has expired.
924 if (time_to_check_for_expired_networkstatus
< now
) {
925 networkstatus_t
*ns
= networkstatus_get_latest_consensus();
926 /*XXXX RD: This value needs to be the same as REASONABLY_LIVE_TIME in
927 * networkstatus_get_reasonably_live_consensus(), but that value is way
928 * way too high. Arma: is the bridge issue there resolved yet? -NM */
929 #define NS_EXPIRY_SLOP (24*60*60)
930 if (ns
&& ns
->valid_until
< now
+NS_EXPIRY_SLOP
&&
931 router_have_minimum_dir_info()) {
932 router_dir_info_changed();
934 #define CHECK_EXPIRED_NS_INTERVAL (2*60)
935 time_to_check_for_expired_networkstatus
= now
+ CHECK_EXPIRED_NS_INTERVAL
;
938 if (time_to_dump_geoip_stats
< now
) {
939 #define DUMP_GEOIP_STATS_INTERVAL (60*60);
940 if (time_to_dump_geoip_stats
)
942 time_to_dump_geoip_stats
= now
+ DUMP_GEOIP_STATS_INTERVAL
;
945 /* Remove old information from rephist and the rend cache. */
946 if (time_to_clean_caches
< now
) {
947 rep_history_clean(now
- options
->RephistTrackTime
);
949 rend_cache_clean_v2_descs_as_dir();
950 #define CLEAN_CACHES_INTERVAL (30*60)
951 time_to_clean_caches
= now
+ CLEAN_CACHES_INTERVAL
;
954 #define RETRY_DNS_INTERVAL (10*60)
955 /* If we're a server and initializing dns failed, retry periodically. */
956 if (time_to_retry_dns_init
< now
) {
957 time_to_retry_dns_init
= now
+ RETRY_DNS_INTERVAL
;
958 if (server_mode(options
) && has_dns_init_failed())
962 /** 2. Periodically, we consider force-uploading our descriptor
963 * (if we've passed our internal checks). */
965 /** How often do we check whether part of our router info has changed in a way
966 * that would require an upload? */
967 #define CHECK_DESCRIPTOR_INTERVAL (60)
968 /** How often do we (as a router) check whether our IP address has changed? */
969 #define CHECK_IPADDRESS_INTERVAL (15*60)
971 /* 2b. Once per minute, regenerate and upload the descriptor if the old
972 * one is inaccurate. */
973 if (time_to_check_descriptor
< now
) {
974 static int dirport_reachability_count
= 0;
975 time_to_check_descriptor
= now
+ CHECK_DESCRIPTOR_INTERVAL
;
976 check_descriptor_bandwidth_changed(now
);
977 if (time_to_check_ipaddress
< now
) {
978 time_to_check_ipaddress
= now
+ CHECK_IPADDRESS_INTERVAL
;
979 check_descriptor_ipaddress_changed(now
);
981 /** If our router descriptor ever goes this long without being regenerated
982 * because something changed, we force an immediate regenerate-and-upload. */
983 #define FORCE_REGENERATE_DESCRIPTOR_INTERVAL (18*60*60)
984 mark_my_descriptor_dirty_if_older_than(
985 now
- FORCE_REGENERATE_DESCRIPTOR_INTERVAL
);
986 consider_publishable_server(0);
987 /* also, check religiously for reachability, if it's within the first
988 * 20 minutes of our uptime. */
989 if (server_mode(options
) &&
990 (has_completed_circuit
|| !any_predicted_circuits(now
)) &&
991 !we_are_hibernating()) {
992 if (stats_n_seconds_working
< TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT
) {
993 consider_testing_reachability(1, dirport_reachability_count
==0);
994 if (++dirport_reachability_count
> 5)
995 dirport_reachability_count
= 0;
996 } else if (time_to_recheck_bandwidth
< now
) {
997 /* If we haven't checked for 12 hours and our bandwidth estimate is
998 * low, do another bandwidth test. This is especially important for
999 * bridges, since they might go long periods without much use. */
1000 routerinfo_t
*me
= router_get_my_routerinfo();
1001 if (time_to_recheck_bandwidth
&& me
&&
1002 me
->bandwidthcapacity
< me
->bandwidthrate
&&
1003 me
->bandwidthcapacity
< 51200) {
1004 reset_bandwidth_test();
1006 #define BANDWIDTH_RECHECK_INTERVAL (12*60*60)
1007 time_to_recheck_bandwidth
= now
+ BANDWIDTH_RECHECK_INTERVAL
;
1011 /* If any networkstatus documents are no longer recent, we need to
1012 * update all the descriptors' running status. */
1013 /* purge obsolete entries */
1014 networkstatus_v2_list_clean(now
);
1015 /* Remove dead routers. */
1016 routerlist_remove_old_routers();
1018 /* Also, once per minute, check whether we want to download any
1019 * networkstatus documents.
1021 update_networkstatus_downloads(now
);
1024 /** 2c. Let directory voting happen. */
1025 if (authdir_mode_v3(options
))
1026 dirvote_act(options
, now
);
1028 /** 3a. Every second, we examine pending circuits and prune the
1029 * ones which have been pending for more than a few seconds.
1030 * We do this before step 4, so it can try building more if
1031 * it's not comfortable with the number of available circuits.
1033 circuit_expire_building(now
);
1035 /** 3b. Also look at pending streams and prune the ones that 'began'
1036 * a long time ago but haven't gotten a 'connected' yet.
1037 * Do this before step 4, so we can put them back into pending
1038 * state to be picked up by the new circuit.
1040 connection_ap_expire_beginning();
1042 /** 3c. And expire connections that we've held open for too long.
1044 connection_expire_held_open();
1046 /** 3d. And every 60 seconds, we relaunch listeners if any died. */
1047 if (!we_are_hibernating() && time_to_check_listeners
< now
) {
1048 retry_all_listeners(NULL
, NULL
);
1049 time_to_check_listeners
= now
+60;
1052 /** 4. Every second, we try a new circuit if there are no valid
1053 * circuits. Every NewCircuitPeriod seconds, we expire circuits
1054 * that became dirty more than MaxCircuitDirtiness seconds ago,
1055 * and we make a new circ if there are no clean circuits.
1057 have_dir_info
= router_have_minimum_dir_info();
1058 if (have_dir_info
&& !we_are_hibernating())
1059 circuit_build_needed_circs(now
);
1061 /** 5. We do housekeeping for each connection... */
1062 connection_or_set_bad_connections();
1063 for (i
=0;i
<smartlist_len(connection_array
);i
++) {
1064 run_connection_housekeeping(i
, now
);
1066 if (time_to_shrink_memory
< now
) {
1067 SMARTLIST_FOREACH(connection_array
, connection_t
*, conn
, {
1069 buf_shrink(conn
->outbuf
);
1071 buf_shrink(conn
->inbuf
);
1074 buf_shrink_freelists(0);
1075 /** How often do we check buffers and pools for empty space that can be
1077 #define MEM_SHRINK_INTERVAL (60)
1078 time_to_shrink_memory
= now
+ MEM_SHRINK_INTERVAL
;
1081 /** 6. And remove any marked circuits... */
1082 circuit_close_all_marked();
1084 /** 7. And upload service descriptors if necessary. */
1085 if (has_completed_circuit
&& !we_are_hibernating()) {
1086 rend_consider_services_upload(now
);
1087 rend_consider_descriptor_republication();
1090 /** 8. and blow away any connections that need to die. have to do this now,
1091 * because if we marked a conn for close and left its socket -1, then
1092 * we'll pass it to poll/select and bad things will happen.
1094 close_closeable_connections();
1096 /** 8b. And if anything in our state is ready to get flushed to disk, we
1100 /** 9. and if we're a server, check whether our DNS is telling stories to
1102 if (server_mode(options
) && time_to_check_for_correct_dns
< now
) {
1103 if (!time_to_check_for_correct_dns
) {
1104 time_to_check_for_correct_dns
= now
+ 60 + crypto_rand_int(120);
1106 dns_launch_correctness_checks();
1107 time_to_check_for_correct_dns
= now
+ 12*3600 +
1108 crypto_rand_int(12*3600);
1112 /** 10. write hidden service usage statistic to disk */
1113 if (options
->HSAuthorityRecordStats
&& time_to_write_hs_statistics
< now
) {
1114 hs_usage_write_statistics_to_file(now
);
1115 #define WRITE_HSUSAGE_INTERVAL (30*60)
1116 time_to_write_hs_statistics
= now
+WRITE_HSUSAGE_INTERVAL
;
1118 /** 10b. write bridge networkstatus file to disk */
1119 if (options
->BridgeAuthoritativeDir
&&
1120 time_to_write_bridge_status_file
< now
) {
1121 networkstatus_dump_bridge_status_to_file(now
);
1122 #define BRIDGE_STATUSFILE_INTERVAL (30*60)
1123 time_to_write_bridge_status_file
= now
+BRIDGE_STATUSFILE_INTERVAL
;
1127 /** Libevent timer: used to invoke second_elapsed_callback() once per
1129 static struct event
*timeout_event
= NULL
;
1130 /** Number of libevent errors in the last second: we die if we get too many. */
1131 static int n_libevent_errors
= 0;
1133 /** Libevent callback: invoked once every second. */
1135 second_elapsed_callback(int fd
, short event
, void *args
)
1137 /* XXXX This could be sensibly refactored into multiple callbacks, and we
1138 * could use Libevent's timers for this rather than checking the current
1139 * time against a bunch of timeouts every second. */
1140 static struct timeval one_second
;
1141 static time_t current_second
= 0;
1143 size_t bytes_written
;
1145 int seconds_elapsed
;
1146 or_options_t
*options
= get_options();
1150 if (!timeout_event
) {
1151 timeout_event
= tor_malloc_zero(sizeof(struct event
));
1152 evtimer_set(timeout_event
, second_elapsed_callback
, NULL
);
1153 one_second
.tv_sec
= 1;
1154 one_second
.tv_usec
= 0;
1157 n_libevent_errors
= 0;
1159 /* log_fn(LOG_NOTICE, "Tick."); */
1161 update_approx_time(now
);
1163 /* the second has rolled over. check more stuff. */
1164 bytes_written
= stats_prev_global_write_bucket
- global_write_bucket
;
1165 bytes_read
= stats_prev_global_read_bucket
- global_read_bucket
;
1166 seconds_elapsed
= current_second
? (int)(now
- current_second
) : 0;
1167 stats_n_bytes_read
+= bytes_read
;
1168 stats_n_bytes_written
+= bytes_written
;
1169 if (accounting_is_enabled(options
) && seconds_elapsed
>= 0)
1170 accounting_add_bytes(bytes_read
, bytes_written
, seconds_elapsed
);
1171 control_event_bandwidth_used((uint32_t)bytes_read
,(uint32_t)bytes_written
);
1172 control_event_stream_bandwidth_used();
1174 if (seconds_elapsed
> 0)
1175 connection_bucket_refill(seconds_elapsed
, now
);
1176 stats_prev_global_read_bucket
= global_read_bucket
;
1177 stats_prev_global_write_bucket
= global_write_bucket
;
1179 if (server_mode(options
) &&
1180 !we_are_hibernating() &&
1181 seconds_elapsed
> 0 &&
1182 has_completed_circuit
&&
1183 stats_n_seconds_working
/ TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT
!=
1184 (stats_n_seconds_working
+seconds_elapsed
) /
1185 TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT
) {
1186 /* every 20 minutes, check and complain if necessary */
1187 routerinfo_t
*me
= router_get_my_routerinfo();
1188 if (me
&& !check_whether_orport_reachable())
1189 log_warn(LD_CONFIG
,"Your server (%s:%d) has not managed to confirm that "
1190 "its ORPort is reachable. Please check your firewalls, ports, "
1191 "address, /etc/hosts file, etc.",
1192 me
->address
, me
->or_port
);
1193 if (me
&& !check_whether_dirport_reachable())
1195 "Your server (%s:%d) has not managed to confirm that its "
1196 "DirPort is reachable. Please check your firewalls, ports, "
1197 "address, /etc/hosts file, etc.",
1198 me
->address
, me
->dir_port
);
1201 /** If more than this many seconds have elapsed, probably the clock
1202 * jumped: doesn't count. */
1203 #define NUM_JUMPED_SECONDS_BEFORE_WARN 100
1204 if (seconds_elapsed
< -NUM_JUMPED_SECONDS_BEFORE_WARN
||
1205 seconds_elapsed
>= NUM_JUMPED_SECONDS_BEFORE_WARN
) {
1206 circuit_note_clock_jumped(seconds_elapsed
);
1207 /* XXX if the time jumps *back* many months, do our events in
1208 * run_scheduled_events() recover? I don't think they do. -RD */
1209 } else if (seconds_elapsed
> 0)
1210 stats_n_seconds_working
+= seconds_elapsed
;
1212 run_scheduled_events(now
);
1214 current_second
= now
; /* remember which second it is, for next time */
1217 if (current_second
% 300 == 0) {
1218 rep_history_clean(current_second
- options
->RephistTrackTime
);
1219 dumpmemusage(get_min_log_level()<LOG_INFO
?
1220 get_min_log_level() : LOG_INFO
);
1224 if (evtimer_add(timeout_event
, &one_second
))
1226 "Error from libevent when setting one-second timeout event");
1230 /** Called when a possibly ignorable libevent error occurs; ensures that we
1231 * don't get into an infinite loop by ignoring too many errors from
1234 got_libevent_error(void)
1236 if (++n_libevent_errors
> 8) {
1237 log_err(LD_NET
, "Too many libevent errors in one second; dying");
1244 #define UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST (6*60*60)
1246 /** Called when our IP address seems to have changed. <b>at_interface</b>
1247 * should be true if we detected a change in our interface, and false if we
1248 * detected a change in our published address. */
1250 ip_address_changed(int at_interface
)
1252 int server
= server_mode(get_options());
1256 /* Okay, change our keys. */
1261 if (stats_n_seconds_working
> UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST
)
1262 reset_bandwidth_test();
1263 stats_n_seconds_working
= 0;
1264 router_reset_reachability();
1265 mark_my_descriptor_dirty();
1269 dns_servers_relaunch_checks();
1272 /** Forget what we've learned about the correctness of our DNS servers, and
1273 * start learning again. */
1275 dns_servers_relaunch_checks(void)
1277 if (server_mode(get_options())) {
1278 dns_reset_correctness_checks();
1279 time_to_check_for_correct_dns
= 0;
1283 /** Called when we get a SIGHUP: reload configuration files and keys,
1284 * retry all connections, and so on. */
1288 or_options_t
*options
= get_options();
1291 dmalloc_log_stats();
1292 dmalloc_log_changed(0, 1, 0, 0);
1295 log_notice(LD_GENERAL
,"Received reload signal (hup). Reloading config and "
1296 "resetting internal state.");
1297 if (accounting_is_enabled(options
))
1298 accounting_record_bandwidth_usage(time(NULL
), get_or_state());
1300 router_reset_warnings();
1301 routerlist_reset_warnings();
1302 addressmap_clear_transient();
1303 /* first, reload config variables, in case they've changed */
1304 if (options
->ReloadTorrcOnSIGHUP
) {
1305 /* no need to provide argc/v, they've been cached in init_from_config */
1306 if (options_init_from_torrc(0, NULL
) < 0) {
1307 log_err(LD_CONFIG
,"Reading config failed--see warnings above. "
1308 "For usage, try -h.");
1311 options
= get_options(); /* they have changed now */
1313 log_notice(LD_GENERAL
, "Not reloading config file: the controller told "
1316 if (authdir_mode_handles_descs(options
, -1)) {
1317 /* reload the approved-routers file */
1318 if (dirserv_load_fingerprint_file() < 0) {
1319 /* warnings are logged from dirserv_load_fingerprint_file() directly */
1320 log_info(LD_GENERAL
, "Error reloading fingerprints. "
1321 "Continuing with old list.");
1325 /* Rotate away from the old dirty circuits. This has to be done
1326 * after we've read the new options, but before we start using
1327 * circuits for directory fetches. */
1328 circuit_expire_all_dirty_circs();
1330 /* retry appropriate downloads */
1331 router_reset_status_download_failures();
1332 router_reset_descriptor_download_failures();
1333 update_networkstatus_downloads(time(NULL
));
1335 /* We'll retry routerstatus downloads in about 10 seconds; no need to
1336 * force a retry there. */
1338 if (server_mode(options
)) {
1339 /* Restart cpuworker and dnsworker processes, so they get up-to-date
1340 * configuration options. */
1341 cpuworkers_rotate();
1347 /** Tor main loop. */
1354 /* initialize dns resolve map, spawn workers if needed */
1355 if (dns_init() < 0) {
1356 if (get_options()->ServerDNSAllowBrokenConfig
)
1357 log_warn(LD_GENERAL
, "Couldn't set up any working nameservers. "
1358 "Network not up yet? Will try again soon.");
1360 log_err(LD_GENERAL
,"Error initializing dns subsystem; exiting. To "
1361 "retry instead, set the ServerDNSAllowBrokenResolvConf option.");
1367 /* load the private keys, if we're supposed to have them, and set up the
1369 if (! identity_key_is_set()) {
1370 if (init_keys() < 0) {
1371 log_err(LD_BUG
,"Error initializing keys; exiting");
1376 /* Set up the packed_cell_t memory pool. */
1379 /* Set up our buckets */
1380 connection_bucket_init();
1381 stats_prev_global_read_bucket
= global_read_bucket
;
1382 stats_prev_global_write_bucket
= global_write_bucket
;
1384 /* initialize the bootstrap status events to know we're starting up */
1385 control_event_bootstrap(BOOTSTRAP_STATUS_STARTING
, 0);
1387 if (trusted_dirs_reload_certs())
1389 if (router_reload_v2_networkstatus()) {
1392 if (router_reload_consensus_networkstatus()) {
1395 /* load the routers file, or assign the defaults. */
1396 if (router_reload_router_list()) {
1399 /* load the networkstatuses. (This launches a download for new routers as
1403 directory_info_has_arrived(now
, 1);
1405 if (authdir_mode_tests_reachability(get_options())) {
1406 /* the directory is already here, run startup things */
1407 dirserv_test_reachability(now
, 1);
1410 if (server_mode(get_options())) {
1411 /* launch cpuworkers. Need to do this *after* we've read the onion key. */
1415 /* set up once-a-second callback. */
1416 second_elapsed_callback(0,0,NULL
);
1419 if (nt_service_is_stopping())
1423 /* Make it easier to tell whether libevent failure is our fault or not. */
1426 /* All active linked conns should get their read events activated. */
1427 SMARTLIST_FOREACH(active_linked_connection_lst
, connection_t
*, conn
,
1428 event_active(conn
->read_event
, EV_READ
, 1));
1429 called_loop_once
= smartlist_len(active_linked_connection_lst
) ? 1 : 0;
1431 update_approx_time(time(NULL
));
1433 /* poll until we have an event, or the second ends, or until we have
1434 * some active linked connections to trigger events for. */
1435 loop_result
= event_loop(called_loop_once
? EVLOOP_ONCE
: 0);
1437 /* let catch() handle things like ^c, and otherwise don't worry about it */
1438 if (loop_result
< 0) {
1439 int e
= tor_socket_errno(-1);
1440 /* let the program survive things like ^z */
1441 if (e
!= EINTR
&& !ERRNO_IS_EINPROGRESS(e
)) {
1442 #ifdef HAVE_EVENT_GET_METHOD
1443 log_err(LD_NET
,"libevent call with %s failed: %s [%d]",
1444 event_get_method(), tor_socket_strerror(e
), e
);
1446 log_err(LD_NET
,"libevent call failed: %s [%d]",
1447 tor_socket_strerror(e
), e
);
1451 } else if (e
== EINVAL
) {
1452 log_warn(LD_NET
, "EINVAL from libevent: should you upgrade libevent?");
1453 if (got_libevent_error())
1457 if (ERRNO_IS_EINPROGRESS(e
))
1459 "libevent call returned EINPROGRESS? Please report.");
1460 log_debug(LD_NET
,"libevent call interrupted.");
1461 /* You can't trust the results of this poll(). Go back to the
1462 * top of the big for loop. */
1469 /** Used to implement the SIGNAL control command: if we accept
1470 * <b>the_signal</b> as a remote pseudo-signal, act on it. */
1471 /* We don't re-use catch() here because:
1472 * 1. We handle a different set of signals than those allowed in catch.
1473 * 2. Platforms without signal() are unlikely to define SIGfoo.
1474 * 3. The control spec is defined to use fixed numeric signal values
1475 * which just happen to match the Unix values.
1478 control_signal_act(int the_signal
)
1483 signal_callback(0,0,(void*)(uintptr_t)SIGHUP
);
1486 signal_callback(0,0,(void*)(uintptr_t)SIGINT
);
1489 signal_callback(0,0,(void*)(uintptr_t)SIGUSR1
);
1492 signal_callback(0,0,(void*)(uintptr_t)SIGUSR2
);
1495 signal_callback(0,0,(void*)(uintptr_t)SIGTERM
);
1498 signal_callback(0,0,(void*)(uintptr_t)SIGNEWNYM
);
1500 case SIGCLEARDNSCACHE
:
1501 signal_callback(0,0,(void*)(uintptr_t)SIGCLEARDNSCACHE
);
1504 log_warn(LD_BUG
, "Unrecognized signal number %d.", the_signal
);
1509 /** Libevent callback: invoked when we get a signal.
1512 signal_callback(int fd
, short events
, void *arg
)
1514 uintptr_t sig
= (uintptr_t)arg
;
1520 log_notice(LD_GENERAL
,"Catching signal TERM, exiting cleanly.");
1525 if (!server_mode(get_options())) { /* do it now */
1526 log_notice(LD_GENERAL
,"Interrupt: exiting cleanly.");
1530 hibernate_begin_shutdown();
1534 log_debug(LD_GENERAL
,"Caught SIGPIPE. Ignoring.");
1538 /* prefer to log it at INFO, but make sure we always see it */
1539 dumpstats(get_min_log_level()<LOG_INFO
? get_min_log_level() : LOG_INFO
);
1542 switch_logs_debug();
1543 log_debug(LD_GENERAL
,"Caught USR2, going to loglevel debug. "
1544 "Send HUP to change back.");
1548 log_warn(LD_CONFIG
,"Restart failed (config error?). Exiting.");
1555 while (waitpid(-1,NULL
,WNOHANG
) > 0) ; /* keep reaping until no more
1560 time_t now
= time(NULL
);
1561 if (time_of_last_signewnym
+ MAX_SIGNEWNYM_RATE
> now
) {
1562 signewnym_is_pending
= 1;
1563 log(LOG_NOTICE
, LD_CONTROL
,
1564 "Rate limiting NEWNYM request: delaying by %d second(s)",
1565 (int)(MAX_SIGNEWNYM_RATE
+time_of_last_signewnym
-now
));
1567 signewnym_impl(now
);
1571 case SIGCLEARDNSCACHE
:
1572 addressmap_clear_transient();
1577 extern uint64_t rephist_total_alloc
;
1578 extern uint32_t rephist_total_num
;
1581 * Write current memory usage information to the log.
1584 dumpmemusage(int severity
)
1586 connection_dump_buffer_mem_stats(severity
);
1587 log(severity
, LD_GENERAL
, "In rephist: "U64_FORMAT
" used by %d Tors.",
1588 U64_PRINTF_ARG(rephist_total_alloc
), rephist_total_num
);
1589 dump_routerlist_mem_usage(severity
);
1590 dump_cell_pool_usage(severity
);
1591 buf_dump_freelist_sizes(severity
);
1592 tor_log_mallinfo(severity
);
1595 /** Write all statistics to the log, with log level 'severity'. Called
1596 * in response to a SIGUSR1. */
1598 dumpstats(int severity
)
1600 time_t now
= time(NULL
);
1602 int rbuf_cap
, wbuf_cap
, rbuf_len
, wbuf_len
;
1604 log(severity
, LD_GENERAL
, "Dumping stats:");
1606 SMARTLIST_FOREACH(connection_array
, connection_t
*, conn
,
1608 int i
= conn_sl_idx
;
1609 log(severity
, LD_GENERAL
,
1610 "Conn %d (socket %d) type %d (%s), state %d (%s), created %d secs ago",
1611 i
, conn
->s
, conn
->type
, conn_type_to_string(conn
->type
),
1612 conn
->state
, conn_state_to_string(conn
->type
, conn
->state
),
1613 (int)(now
- conn
->timestamp_created
));
1614 if (!connection_is_listener(conn
)) {
1615 log(severity
,LD_GENERAL
,
1616 "Conn %d is to %s:%d.", i
,
1617 safe_str(conn
->address
), conn
->port
);
1618 log(severity
,LD_GENERAL
,
1619 "Conn %d: %d bytes waiting on inbuf (len %d, last read %d secs ago)",
1621 (int)buf_datalen(conn
->inbuf
),
1622 (int)buf_allocation(conn
->inbuf
),
1623 (int)(now
- conn
->timestamp_lastread
));
1624 log(severity
,LD_GENERAL
,
1625 "Conn %d: %d bytes waiting on outbuf "
1626 "(len %d, last written %d secs ago)",i
,
1627 (int)buf_datalen(conn
->outbuf
),
1628 (int)buf_allocation(conn
->outbuf
),
1629 (int)(now
- conn
->timestamp_lastwritten
));
1630 if (conn
->type
== CONN_TYPE_OR
) {
1631 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
1633 tor_tls_get_buffer_sizes(or_conn
->tls
, &rbuf_cap
, &rbuf_len
,
1634 &wbuf_cap
, &wbuf_len
);
1635 log(severity
, LD_GENERAL
,
1636 "Conn %d: %d/%d bytes used on OpenSSL read buffer; "
1637 "%d/%d bytes used on write buffer.",
1638 i
, rbuf_len
, rbuf_cap
, wbuf_len
, wbuf_cap
);
1642 circuit_dump_by_conn(conn
, severity
); /* dump info about all the circuits
1643 * using this conn */
1645 log(severity
, LD_NET
,
1646 "Cells processed: "U64_FORMAT
" padding\n"
1647 " "U64_FORMAT
" create\n"
1648 " "U64_FORMAT
" created\n"
1649 " "U64_FORMAT
" relay\n"
1650 " ("U64_FORMAT
" relayed)\n"
1651 " ("U64_FORMAT
" delivered)\n"
1652 " "U64_FORMAT
" destroy",
1653 U64_PRINTF_ARG(stats_n_padding_cells_processed
),
1654 U64_PRINTF_ARG(stats_n_create_cells_processed
),
1655 U64_PRINTF_ARG(stats_n_created_cells_processed
),
1656 U64_PRINTF_ARG(stats_n_relay_cells_processed
),
1657 U64_PRINTF_ARG(stats_n_relay_cells_relayed
),
1658 U64_PRINTF_ARG(stats_n_relay_cells_delivered
),
1659 U64_PRINTF_ARG(stats_n_destroy_cells_processed
));
1660 if (stats_n_data_cells_packaged
)
1661 log(severity
,LD_NET
,"Average packaged cell fullness: %2.3f%%",
1662 100*(U64_TO_DBL(stats_n_data_bytes_packaged
) /
1663 U64_TO_DBL(stats_n_data_cells_packaged
*RELAY_PAYLOAD_SIZE
)) );
1664 if (stats_n_data_cells_received
)
1665 log(severity
,LD_NET
,"Average delivered cell fullness: %2.3f%%",
1666 100*(U64_TO_DBL(stats_n_data_bytes_received
) /
1667 U64_TO_DBL(stats_n_data_cells_received
*RELAY_PAYLOAD_SIZE
)) );
1669 if (now
- time_of_process_start
>= 0)
1670 elapsed
= now
- time_of_process_start
;
1675 log(severity
, LD_NET
,
1676 "Average bandwidth: "U64_FORMAT
"/%d = %d bytes/sec reading",
1677 U64_PRINTF_ARG(stats_n_bytes_read
),
1679 (int) (stats_n_bytes_read
/elapsed
));
1680 log(severity
, LD_NET
,
1681 "Average bandwidth: "U64_FORMAT
"/%d = %d bytes/sec writing",
1682 U64_PRINTF_ARG(stats_n_bytes_written
),
1684 (int) (stats_n_bytes_written
/elapsed
));
1687 log(severity
, LD_NET
, "--------------- Dumping memory information:");
1688 dumpmemusage(severity
);
1690 rep_hist_dump_stats(now
,severity
);
1691 rend_service_dump_stats(severity
);
1692 dump_pk_ops(severity
);
1693 dump_distinct_digest_count(severity
);
1696 /** Called by exit() as we shut down the process.
1701 /* NOTE: If we ever daemonize, this gets called immediately. That's
1702 * okay for now, because we only use this on Windows. */
1708 /** Set up the signal handlers for either parent or child. */
1710 handle_signals(int is_parent
)
1712 #ifndef MS_WINDOWS /* do signal stuff only on Unix */
1714 static int signals
[] = {
1715 SIGINT
, /* do a controlled slow shutdown */
1716 SIGTERM
, /* to terminate now */
1717 SIGPIPE
, /* otherwise SIGPIPE kills us */
1718 SIGUSR1
, /* dump stats */
1719 SIGUSR2
, /* go to loglevel debug */
1720 SIGHUP
, /* to reload config, retry conns, etc */
1722 SIGXFSZ
, /* handle file-too-big resource exhaustion */
1724 SIGCHLD
, /* handle dns/cpu workers that exit */
1726 static struct event signal_events
[16]; /* bigger than it has to be. */
1728 for (i
= 0; signals
[i
] >= 0; ++i
) {
1729 signal_set(&signal_events
[i
], signals
[i
], signal_callback
,
1730 (void*)(uintptr_t)signals
[i
]);
1731 if (signal_add(&signal_events
[i
], NULL
))
1732 log_warn(LD_BUG
, "Error from libevent when adding event for signal %d",
1736 struct sigaction action
;
1737 action
.sa_flags
= 0;
1738 sigemptyset(&action
.sa_mask
);
1739 action
.sa_handler
= SIG_IGN
;
1740 sigaction(SIGINT
, &action
, NULL
);
1741 sigaction(SIGTERM
, &action
, NULL
);
1742 sigaction(SIGPIPE
, &action
, NULL
);
1743 sigaction(SIGUSR1
, &action
, NULL
);
1744 sigaction(SIGUSR2
, &action
, NULL
);
1745 sigaction(SIGHUP
, &action
, NULL
);
1747 sigaction(SIGXFSZ
, &action
, NULL
);
1750 #else /* MS windows */
1752 #endif /* signal stuff */
1755 /** Main entry point for the Tor command-line client.
1758 tor_init(int argc
, char *argv
[])
1762 time_of_process_start
= time(NULL
);
1763 if (!connection_array
)
1764 connection_array
= smartlist_create();
1765 if (!closeable_connection_lst
)
1766 closeable_connection_lst
= smartlist_create();
1767 if (!active_linked_connection_lst
)
1768 active_linked_connection_lst
= smartlist_create();
1769 /* Have the log set up with our application name. */
1770 tor_snprintf(buf
, sizeof(buf
), "Tor %s", get_version());
1771 log_set_application_name(buf
);
1772 /* Initialize the history structures. */
1774 /* Initialize the service cache. */
1776 addressmap_init(); /* Init the client dns cache. Do it always, since it's
1779 /* We search for the "quiet" option first, since it decides whether we
1780 * will log anything at all to the command line. */
1781 for (i
=1;i
<argc
;++i
) {
1782 if (!strcmp(argv
[i
], "--hush"))
1784 if (!strcmp(argv
[i
], "--quiet"))
1787 /* give it somewhere to log to initially */
1790 /* no initial logging */
1793 add_temp_log(LOG_WARN
);
1796 add_temp_log(LOG_NOTICE
);
1799 log(LOG_NOTICE
, LD_GENERAL
, "Tor v%s. This is experimental software. "
1800 "Do not rely on it for strong anonymity. (Running on %s)",get_version(),
1803 if (network_init()<0) {
1804 log_err(LD_BUG
,"Error initializing network; exiting.");
1807 atexit(exit_function
);
1809 if (options_init_from_torrc(argc
,argv
) < 0) {
1810 log_err(LD_CONFIG
,"Reading config failed--see warnings above.");
1816 log_warn(LD_GENERAL
,"You are running Tor as root. You don't need to, "
1817 "and you probably shouldn't.");
1820 if (crypto_global_init(get_options()->HardwareAccel
)) {
1821 log_err(LD_BUG
, "Unable to initialize OpenSSL. Exiting.");
1828 /** A lockfile structure, used to prevent two Tors from messing with the
1829 * data directory at once. If this variable is non-NULL, we're holding
1831 static tor_lockfile_t
*lockfile
= NULL
;
1833 /** Try to grab the lock file described in <b>options</b>, if we do not
1834 * already have it. If <b>err_if_locked</b> is true, warn if somebody else is
1835 * holding the lock, and exit if we can't get it after waiting. Otherwise,
1836 * return -1 if we can't get the lockfile. Return 0 on success.
1839 try_locking(or_options_t
*options
, int err_if_locked
)
1844 char *fname
= options_get_datadir_fname2_suffix(options
, "lock",NULL
,NULL
);
1845 int already_locked
= 0;
1846 tor_lockfile_t
*lf
= tor_lockfile_lock(fname
, 0, &already_locked
);
1849 if (err_if_locked
&& already_locked
) {
1851 log_warn(LD_GENERAL
, "It looks like another Tor process is running "
1852 "with the same data directory. Waiting 5 seconds to see "
1853 "if it goes away.");
1859 r
= try_locking(options
, 0);
1861 log_err(LD_GENERAL
, "No, it's still there. Exiting.");
1873 /** Return true iff we've successfully acquired the lock file. */
1877 return lockfile
!= NULL
;
1880 /** If we have successfully acquired the lock file, release it. */
1882 release_lockfile(void)
1885 tor_lockfile_unlock(lockfile
);
1890 /** Free all memory that we might have allocated somewhere.
1891 * If <b>postfork</b>, we are a worker process and we want to free
1892 * only the parts of memory that we won't touch. If !<b>postfork</b>,
1893 * Tor is shutting down and we should free everything.
1895 * Helps us find the real leaks with dmalloc and the like. Also valgrind
1896 * should then report 0 reachable in its leak report (in an ideal world --
1897 * in practice libevent, SSL, libc etc never quite free everything). */
1899 tor_free_all(int postfork
)
1906 routerlist_free_all();
1907 networkstatus_free_all();
1908 addressmap_free_all();
1910 rend_service_free_all();
1911 rend_cache_free_all();
1912 rend_service_authorization_free_all();
1913 rep_hist_free_all();
1914 hs_usage_free_all();
1916 clear_pending_onions();
1918 entry_guards_free_all();
1919 connection_free_all();
1920 buf_shrink_freelists(1);
1921 memarea_clear_freelist();
1925 policies_free_all();
1931 /* stuff in main.c */
1932 if (connection_array
)
1933 smartlist_free(connection_array
);
1934 if (closeable_connection_lst
)
1935 smartlist_free(closeable_connection_lst
);
1936 if (active_linked_connection_lst
)
1937 smartlist_free(active_linked_connection_lst
);
1938 tor_free(timeout_event
);
1942 /* Stuff in util.c and address.c*/
1945 esc_router_info(NULL
);
1946 logs_free_all(); /* free log strings. do this last so logs keep working. */
1950 /** Do whatever cleanup is necessary before shutting Tor down. */
1954 or_options_t
*options
= get_options();
1955 /* Remove our pid file. We don't care if there was an error when we
1956 * unlink, nothing we could do about it anyways. */
1957 if (options
->command
== CMD_RUN_TOR
) {
1958 time_t now
= time(NULL
);
1959 if (options
->PidFile
)
1960 unlink(options
->PidFile
);
1961 if (accounting_is_enabled(options
))
1962 accounting_record_bandwidth_usage(now
, get_or_state());
1963 or_state_mark_dirty(get_or_state(), 0); /* force an immediate save. */
1965 if (authdir_mode_tests_reachability(options
))
1966 rep_hist_record_mtbf_data(now
, 0);
1969 dmalloc_log_stats();
1971 tor_free_all(0); /* We could move tor_free_all back into the ifdef below
1972 later, if it makes shutdown unacceptably slow. But for
1973 now, leave it here: it's helped us catch bugs in the
1975 crypto_global_cleanup();
1977 dmalloc_log_unfreed();
1982 /** Read/create keys as needed, and echo our fingerprint to stdout. */
1984 do_list_fingerprint(void)
1986 char buf
[FINGERPRINT_LEN
+1];
1988 const char *nickname
= get_options()->Nickname
;
1989 if (!server_mode(get_options())) {
1991 "Clients don't have long-term identity keys. Exiting.\n");
1994 tor_assert(nickname
);
1995 if (init_keys() < 0) {
1996 log_err(LD_BUG
,"Error initializing keys; can't display fingerprint");
1999 if (!(k
= get_identity_key())) {
2000 log_err(LD_GENERAL
,"Error: missing identity key.");
2003 if (crypto_pk_get_fingerprint(k
, buf
, 1)<0) {
2004 log_err(LD_BUG
, "Error computing fingerprint");
2007 printf("%s %s\n", nickname
, buf
);
2011 /** Entry point for password hashing: take the desired password from
2012 * the command line, and print its salted hash to stdout. **/
2014 do_hash_password(void)
2018 char key
[S2K_SPECIFIER_LEN
+DIGEST_LEN
];
2020 crypto_rand(key
, S2K_SPECIFIER_LEN
-1);
2021 key
[S2K_SPECIFIER_LEN
-1] = (uint8_t)96; /* Hash 64 K of data. */
2022 secret_to_key(key
+S2K_SPECIFIER_LEN
, DIGEST_LEN
,
2023 get_options()->command_arg
, strlen(get_options()->command_arg
),
2025 base16_encode(output
, sizeof(output
), key
, sizeof(key
));
2026 printf("16:%s\n",output
);
2029 /** Main entry point for the Tor process. Called from main(). */
2030 /* This function is distinct from main() only so we can link main.c into
2031 * the unittest binary without conflicting with the unittests' main. */
2033 tor_main(int argc
, char *argv
[])
2036 update_approx_time(time(NULL
));
2041 /* Instruct OpenSSL to use our internal wrappers for malloc,
2042 realloc and free. */
2043 int r
= CRYPTO_set_mem_ex_functions(_tor_malloc
, _tor_realloc
, _tor_free
);
2050 result
= nt_service_parse_options(argc
, argv
, &done
);
2051 if (done
) return result
;
2054 if (tor_init(argc
, argv
)<0)
2056 switch (get_options()->command
) {
2059 nt_service_set_state(SERVICE_RUNNING
);
2061 result
= do_main_loop();
2063 case CMD_LIST_FINGERPRINT
:
2064 result
= do_list_fingerprint();
2066 case CMD_HASH_PASSWORD
:
2070 case CMD_VERIFY_CONFIG
:
2071 printf("Configuration was valid\n");
2074 case CMD_RUN_UNITTESTS
: /* only set by test.c */
2076 log_warn(LD_BUG
,"Illegal command number %d: internal error.",
2077 get_options()->command
);