1 /* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
2 * Copyright (c) 2007-2011, The Tor Project, Inc. */
3 /* See LICENSE for licensing information */
7 * \brief Basic history and "reputation" functionality to remember
8 * which servers have worked in the past, how much bandwidth we've
9 * been using, which ports we tend to want, and so on; further,
10 * exit port statistics and cell statistics.
14 #include "circuitlist.h"
15 #include "circuituse.h"
17 #include "networkstatus.h"
20 #include "routerlist.h"
23 static void bw_arrays_init(void);
24 static void predicted_ports_init(void);
26 /** Total number of bytes currently allocated in fields used by rephist.c. */
27 uint64_t rephist_total_alloc
=0;
28 /** Number of or_history_t objects currently allocated. */
29 uint32_t rephist_total_num
=0;
31 /** If the total weighted run count of all runs for a router ever falls
32 * below this amount, the router can be treated as having 0 MTBF. */
33 #define STABILITY_EPSILON 0.0001
34 /** Value by which to discount all old intervals for MTBF purposes. This
35 * is compounded every STABILITY_INTERVAL. */
36 #define STABILITY_ALPHA 0.95
37 /** Interval at which to discount all old intervals for MTBF purposes. */
38 #define STABILITY_INTERVAL (12*60*60)
39 /* (This combination of ALPHA, INTERVAL, and EPSILON makes it so that an
40 * interval that just ended counts twice as much as one that ended a week ago,
41 * 20X as much as one that ended a month ago, and routers that have had no
42 * uptime data for about half a year will get forgotten.) */
44 /** History of an OR-\>OR link. */
45 typedef struct link_history_t
{
46 /** When did we start tracking this list? */
48 /** When did we most recently note a change to this link */
50 /** How many times did extending from OR1 to OR2 succeed? */
51 unsigned long n_extend_ok
;
52 /** How many times did extending from OR1 to OR2 fail? */
53 unsigned long n_extend_fail
;
56 /** History of an OR. */
57 typedef struct or_history_t
{
58 /** When did we start tracking this OR? */
60 /** When did we most recently note a change to this OR? */
62 /** How many times did we successfully connect? */
63 unsigned long n_conn_ok
;
64 /** How many times did we try to connect and fail?*/
65 unsigned long n_conn_fail
;
66 /** How many seconds have we been connected to this OR before
69 /** How many seconds have we been unable to connect to this OR before
71 unsigned long downtime
;
72 /** If nonzero, we have been connected since this time. */
74 /** If nonzero, we have been unable to connect since this time. */
77 /** The address at which we most recently connected to this OR
79 tor_addr_t last_reached_addr
;
81 /** The port at which we most recently connected to this OR successfully */
82 uint16_t last_reached_port
;
84 /* === For MTBF tracking: */
85 /** Weighted sum total of all times that this router has been online.
87 unsigned long weighted_run_length
;
88 /** If the router is now online (according to stability-checking rules),
89 * when did it come online? */
91 /** Sum of weights for runs in weighted_run_length. */
92 double total_run_weights
;
93 /* === For fractional uptime tracking: */
94 time_t start_of_downtime
;
95 unsigned long weighted_uptime
;
96 unsigned long total_weighted_time
;
98 /** Map from hex OR2 identity digest to a link_history_t for the link
99 * from this OR to OR2. */
100 digestmap_t
*link_history_map
;
103 /** When did we last multiply all routers' weighted_run_length and
104 * total_run_weights by STABILITY_ALPHA? */
105 static time_t stability_last_downrated
= 0;
108 static time_t started_tracking_stability
= 0;
110 /** Map from hex OR identity digest to or_history_t. */
111 static digestmap_t
*history_map
= NULL
;
113 /** Return the or_history_t for the OR with identity digest <b>id</b>,
114 * creating it if necessary. */
115 static or_history_t
*
116 get_or_history(const char* id
)
120 if (tor_mem_is_zero(id
, DIGEST_LEN
))
123 hist
= digestmap_get(history_map
, id
);
125 hist
= tor_malloc_zero(sizeof(or_history_t
));
126 rephist_total_alloc
+= sizeof(or_history_t
);
128 hist
->link_history_map
= digestmap_new();
129 hist
->since
= hist
->changed
= time(NULL
);
130 tor_addr_make_unspec(&hist
->last_reached_addr
);
131 digestmap_set(history_map
, id
, hist
);
136 /** Return the link_history_t for the link from the first named OR to
137 * the second, creating it if necessary. (ORs are identified by
140 static link_history_t
*
141 get_link_history(const char *from_id
, const char *to_id
)
143 or_history_t
*orhist
;
144 link_history_t
*lhist
;
145 orhist
= get_or_history(from_id
);
148 if (tor_mem_is_zero(to_id
, DIGEST_LEN
))
150 lhist
= (link_history_t
*) digestmap_get(orhist
->link_history_map
, to_id
);
152 lhist
= tor_malloc_zero(sizeof(link_history_t
));
153 rephist_total_alloc
+= sizeof(link_history_t
);
154 lhist
->since
= lhist
->changed
= time(NULL
);
155 digestmap_set(orhist
->link_history_map
, to_id
, lhist
);
160 /** Helper: free storage held by a single link history entry. */
162 _free_link_history(void *val
)
164 rephist_total_alloc
-= sizeof(link_history_t
);
168 /** Helper: free storage held by a single OR history entry. */
170 free_or_history(void *_hist
)
172 or_history_t
*hist
= _hist
;
173 digestmap_free(hist
->link_history_map
, _free_link_history
);
174 rephist_total_alloc
-= sizeof(or_history_t
);
179 /** Update an or_history_t object <b>hist</b> so that its uptime/downtime
180 * count is up-to-date as of <b>when</b>.
183 update_or_history(or_history_t
*hist
, time_t when
)
186 if (hist
->up_since
) {
187 tor_assert(!hist
->down_since
);
188 hist
->uptime
+= (when
- hist
->up_since
);
189 hist
->up_since
= when
;
190 } else if (hist
->down_since
) {
191 hist
->downtime
+= (when
- hist
->down_since
);
192 hist
->down_since
= when
;
196 /** Initialize the static data structures for tracking history. */
200 history_map
= digestmap_new();
202 predicted_ports_init();
205 /** Helper: note that we are no longer connected to the router with history
206 * <b>hist</b>. If <b>failed</b>, the connection failed; otherwise, it was
207 * closed correctly. */
209 mark_or_down(or_history_t
*hist
, time_t when
, int failed
)
211 if (hist
->up_since
) {
212 hist
->uptime
+= (when
- hist
->up_since
);
215 if (failed
&& !hist
->down_since
) {
216 hist
->down_since
= when
;
220 /** Helper: note that we are connected to the router with history
223 mark_or_up(or_history_t
*hist
, time_t when
)
225 if (hist
->down_since
) {
226 hist
->downtime
+= (when
- hist
->down_since
);
227 hist
->down_since
= 0;
229 if (!hist
->up_since
) {
230 hist
->up_since
= when
;
234 /** Remember that an attempt to connect to the OR with identity digest
235 * <b>id</b> failed at <b>when</b>.
238 rep_hist_note_connect_failed(const char* id
, time_t when
)
241 hist
= get_or_history(id
);
245 mark_or_down(hist
, when
, 1);
246 hist
->changed
= when
;
249 /** Remember that an attempt to connect to the OR with identity digest
250 * <b>id</b> succeeded at <b>when</b>.
253 rep_hist_note_connect_succeeded(const char* id
, time_t when
)
256 hist
= get_or_history(id
);
260 mark_or_up(hist
, when
);
261 hist
->changed
= when
;
264 /** Remember that we intentionally closed our connection to the OR
265 * with identity digest <b>id</b> at <b>when</b>.
268 rep_hist_note_disconnect(const char* id
, time_t when
)
271 hist
= get_or_history(id
);
274 mark_or_down(hist
, when
, 0);
275 hist
->changed
= when
;
278 /** Remember that our connection to the OR with identity digest
279 * <b>id</b> had an error and stopped working at <b>when</b>.
282 rep_hist_note_connection_died(const char* id
, time_t when
)
286 /* If conn has no identity, it didn't complete its handshake, or something
287 * went wrong. Ignore it.
291 hist
= get_or_history(id
);
294 mark_or_down(hist
, when
, 1);
295 hist
->changed
= when
;
298 /** We have just decided that this router with identity digest <b>id</b> is
299 * reachable, meaning we will give it a "Running" flag for the next while. */
301 rep_hist_note_router_reachable(const char *id
, const tor_addr_t
*at_addr
,
302 const uint16_t at_port
, time_t when
)
304 or_history_t
*hist
= get_or_history(id
);
306 char tbuf
[ISO_TIME_LEN
+1];
307 int addr_changed
, port_changed
;
310 tor_assert((!at_addr
&& !at_port
) || (at_addr
&& at_port
));
312 addr_changed
= at_addr
&&
313 tor_addr_compare(at_addr
, &hist
->last_reached_addr
, CMP_EXACT
) != 0;
314 port_changed
= at_port
&& at_port
!= hist
->last_reached_port
;
316 if (!started_tracking_stability
)
317 started_tracking_stability
= time(NULL
);
318 if (!hist
->start_of_run
) {
319 hist
->start_of_run
= when
;
322 if (hist
->start_of_downtime
) {
325 format_local_iso_time(tbuf
, hist
->start_of_downtime
);
326 log_info(LD_HIST
, "Router %s is now Running; it had been down since %s.",
327 hex_str(id
, DIGEST_LEN
), tbuf
);
329 log_info(LD_HIST
, " (Paradoxically, it was already Running too.)");
331 down_length
= when
- hist
->start_of_downtime
;
332 hist
->total_weighted_time
+= down_length
;
333 hist
->start_of_downtime
= 0;
334 } else if (addr_changed
|| port_changed
) {
335 /* If we're reachable, but the address changed, treat this as some
337 int penalty
= get_options()->TestingTorNetwork
? 240 : 3600;
340 if ((ns
= networkstatus_get_latest_consensus())) {
341 int fresh_interval
= (int)(ns
->fresh_until
- ns
->valid_after
);
342 int live_interval
= (int)(ns
->valid_until
- ns
->valid_after
);
343 /* on average, a descriptor addr change takes .5 intervals to make it
344 * into a consensus, and half a liveness period to make it to
346 penalty
= (int)(fresh_interval
+ live_interval
) / 2;
348 format_local_iso_time(tbuf
, hist
->start_of_run
);
349 log_info(LD_HIST
,"Router %s still seems Running, but its address appears "
350 "to have changed since the last time it was reachable. I'm "
351 "going to treat it as having been down for %d seconds",
352 hex_str(id
, DIGEST_LEN
), penalty
);
353 rep_hist_note_router_unreachable(id
, when
-penalty
);
354 rep_hist_note_router_reachable(id
, NULL
, 0, when
);
356 format_local_iso_time(tbuf
, hist
->start_of_run
);
358 log_debug(LD_HIST
, "Router %s is still Running; it has been Running "
359 "since %s", hex_str(id
, DIGEST_LEN
), tbuf
);
361 log_info(LD_HIST
,"Router %s is now Running; it was previously untracked",
362 hex_str(id
, DIGEST_LEN
));
365 tor_addr_copy(&hist
->last_reached_addr
, at_addr
);
367 hist
->last_reached_port
= at_port
;
370 /** We have just decided that this router is unreachable, meaning
371 * we are taking away its "Running" flag. */
373 rep_hist_note_router_unreachable(const char *id
, time_t when
)
375 or_history_t
*hist
= get_or_history(id
);
376 char tbuf
[ISO_TIME_LEN
+1];
378 if (!started_tracking_stability
)
379 started_tracking_stability
= time(NULL
);
382 if (hist
->start_of_run
) {
383 /*XXXX We could treat failed connections differently from failed
384 * connect attempts. */
385 long run_length
= when
- hist
->start_of_run
;
386 format_local_iso_time(tbuf
, hist
->start_of_run
);
388 hist
->total_run_weights
+= 1.0;
389 hist
->start_of_run
= 0;
390 if (run_length
< 0) {
391 unsigned long penalty
= -run_length
;
392 #define SUBTRACT_CLAMPED(var, penalty) \
393 do { (var) = (var) < (penalty) ? 0 : (var) - (penalty); } while (0)
395 SUBTRACT_CLAMPED(hist
->weighted_run_length
, penalty
);
396 SUBTRACT_CLAMPED(hist
->weighted_uptime
, penalty
);
398 hist
->weighted_run_length
+= run_length
;
399 hist
->weighted_uptime
+= run_length
;
400 hist
->total_weighted_time
+= run_length
;
403 log_info(LD_HIST
, "Router %s is now non-Running: it had previously been "
404 "Running since %s. Its total weighted uptime is %lu/%lu.",
405 hex_str(id
, DIGEST_LEN
), tbuf
, hist
->weighted_uptime
,
406 hist
->total_weighted_time
);
408 if (!hist
->start_of_downtime
) {
409 hist
->start_of_downtime
= when
;
412 log_info(LD_HIST
, "Router %s is now non-Running; it was previously "
413 "untracked.", hex_str(id
, DIGEST_LEN
));
416 format_local_iso_time(tbuf
, hist
->start_of_downtime
);
418 log_info(LD_HIST
, "Router %s is still non-Running; it has been "
419 "non-Running since %s.", hex_str(id
, DIGEST_LEN
), tbuf
);
424 /** Helper: Discount all old MTBF data, if it is time to do so. Return
425 * the time at which we should next discount MTBF data. */
427 rep_hist_downrate_old_runs(time_t now
)
429 digestmap_iter_t
*orhist_it
;
436 history_map
= digestmap_new();
437 if (!stability_last_downrated
)
438 stability_last_downrated
= now
;
439 if (stability_last_downrated
+ STABILITY_INTERVAL
> now
)
440 return stability_last_downrated
+ STABILITY_INTERVAL
;
442 /* Okay, we should downrate the data. By how much? */
443 while (stability_last_downrated
+ STABILITY_INTERVAL
< now
) {
444 stability_last_downrated
+= STABILITY_INTERVAL
;
445 alpha
*= STABILITY_ALPHA
;
448 log_info(LD_HIST
, "Discounting all old stability info by a factor of %lf",
451 /* Multiply every w_r_l, t_r_w pair by alpha. */
452 for (orhist_it
= digestmap_iter_init(history_map
);
453 !digestmap_iter_done(orhist_it
);
454 orhist_it
= digestmap_iter_next(history_map
,orhist_it
)) {
455 digestmap_iter_get(orhist_it
, &digest1
, &hist_p
);
458 hist
->weighted_run_length
=
459 (unsigned long)(hist
->weighted_run_length
* alpha
);
460 hist
->total_run_weights
*= alpha
;
462 hist
->weighted_uptime
= (unsigned long)(hist
->weighted_uptime
* alpha
);
463 hist
->total_weighted_time
= (unsigned long)
464 (hist
->total_weighted_time
* alpha
);
467 return stability_last_downrated
+ STABILITY_INTERVAL
;
470 /** Helper: Return the weighted MTBF of the router with history <b>hist</b>. */
472 get_stability(or_history_t
*hist
, time_t when
)
474 long total
= hist
->weighted_run_length
;
475 double total_weights
= hist
->total_run_weights
;
477 if (hist
->start_of_run
) {
478 /* We're currently in a run. Let total and total_weights hold the values
479 * they would hold if the current run were to end now. */
480 total
+= (when
-hist
->start_of_run
);
481 total_weights
+= 1.0;
483 if (total_weights
< STABILITY_EPSILON
) {
484 /* Round down to zero, and avoid divide-by-zero. */
488 return total
/ total_weights
;
491 /** Return the total amount of time we've been observing, with each run of
492 * time downrated by the appropriate factor. */
494 get_total_weighted_time(or_history_t
*hist
, time_t when
)
496 long total
= hist
->total_weighted_time
;
497 if (hist
->start_of_run
) {
498 total
+= (when
- hist
->start_of_run
);
499 } else if (hist
->start_of_downtime
) {
500 total
+= (when
- hist
->start_of_downtime
);
505 /** Helper: Return the weighted percent-of-time-online of the router with
506 * history <b>hist</b>. */
508 get_weighted_fractional_uptime(or_history_t
*hist
, time_t when
)
510 long total
= hist
->total_weighted_time
;
511 long up
= hist
->weighted_uptime
;
513 if (hist
->start_of_run
) {
514 long run_length
= (when
- hist
->start_of_run
);
517 } else if (hist
->start_of_downtime
) {
518 total
+= (when
- hist
->start_of_downtime
);
522 /* Avoid calling anybody's uptime infinity (which should be impossible if
523 * the code is working), or NaN (which can happen for any router we haven't
524 * observed up or down yet). */
528 return ((double) up
) / total
;
531 /** Return an estimated MTBF for the router whose identity digest is
532 * <b>id</b>. Return 0 if the router is unknown. */
534 rep_hist_get_stability(const char *id
, time_t when
)
536 or_history_t
*hist
= get_or_history(id
);
540 return get_stability(hist
, when
);
543 /** Return an estimated percent-of-time-online for the router whose identity
544 * digest is <b>id</b>. Return 0 if the router is unknown. */
546 rep_hist_get_weighted_fractional_uptime(const char *id
, time_t when
)
548 or_history_t
*hist
= get_or_history(id
);
552 return get_weighted_fractional_uptime(hist
, when
);
555 /** Return a number representing how long we've known about the router whose
556 * digest is <b>id</b>. Return 0 if the router is unknown.
558 * Be careful: this measure increases monotonically as we know the router for
559 * longer and longer, but it doesn't increase linearly.
562 rep_hist_get_weighted_time_known(const char *id
, time_t when
)
564 or_history_t
*hist
= get_or_history(id
);
568 return get_total_weighted_time(hist
, when
);
571 /** Return true if we've been measuring MTBFs for long enough to
572 * pronounce on Stability. */
574 rep_hist_have_measured_enough_stability(void)
576 /* XXXX021 This doesn't do so well when we change our opinion
577 * as to whether we're tracking router stability. */
578 return started_tracking_stability
< time(NULL
) - 4*60*60;
581 /** Remember that we successfully extended from the OR with identity
582 * digest <b>from_id</b> to the OR with identity digest
586 rep_hist_note_extend_succeeded(const char *from_id
, const char *to_id
)
588 link_history_t
*hist
;
589 /* log_fn(LOG_WARN, "EXTEND SUCCEEDED: %s->%s",from_name,to_name); */
590 hist
= get_link_history(from_id
, to_id
);
594 hist
->changed
= time(NULL
);
597 /** Remember that we tried to extend from the OR with identity digest
598 * <b>from_id</b> to the OR with identity digest <b>to_name</b>, but
602 rep_hist_note_extend_failed(const char *from_id
, const char *to_id
)
604 link_history_t
*hist
;
605 /* log_fn(LOG_WARN, "EXTEND FAILED: %s->%s",from_name,to_name); */
606 hist
= get_link_history(from_id
, to_id
);
609 ++hist
->n_extend_fail
;
610 hist
->changed
= time(NULL
);
613 /** Log all the reliability data we have remembered, with the chosen
617 rep_hist_dump_stats(time_t now
, int severity
)
619 digestmap_iter_t
*lhist_it
;
620 digestmap_iter_t
*orhist_it
;
621 const char *name1
, *name2
, *digest1
, *digest2
;
622 char hexdigest1
[HEX_DIGEST_LEN
+1];
623 or_history_t
*or_history
;
624 link_history_t
*link_history
;
625 void *or_history_p
, *link_history_p
;
630 unsigned long upt
, downt
;
633 rep_history_clean(now
- get_options()->RephistTrackTime
);
635 log(severity
, LD_HIST
, "--------------- Dumping history information:");
637 for (orhist_it
= digestmap_iter_init(history_map
);
638 !digestmap_iter_done(orhist_it
);
639 orhist_it
= digestmap_iter_next(history_map
,orhist_it
)) {
642 digestmap_iter_get(orhist_it
, &digest1
, &or_history_p
);
643 or_history
= (or_history_t
*) or_history_p
;
645 if ((r
= router_get_by_digest(digest1
)))
649 base16_encode(hexdigest1
, sizeof(hexdigest1
), digest1
, DIGEST_LEN
);
650 update_or_history(or_history
, now
);
651 upt
= or_history
->uptime
;
652 downt
= or_history
->downtime
;
653 s
= get_stability(or_history
, now
);
656 uptime
= ((double)upt
) / (upt
+downt
);
660 log(severity
, LD_HIST
,
661 "OR %s [%s]: %ld/%ld good connections; uptime %ld/%ld sec (%.2f%%); "
662 "wmtbf %lu:%02lu:%02lu",
664 or_history
->n_conn_ok
, or_history
->n_conn_fail
+or_history
->n_conn_ok
,
665 upt
, upt
+downt
, uptime
*100.0,
666 stability
/3600, (stability
/60)%60, stability
%60);
668 if (!digestmap_isempty(or_history
->link_history_map
)) {
669 strlcpy(buffer
, " Extend attempts: ", sizeof(buffer
));
670 len
= strlen(buffer
);
671 for (lhist_it
= digestmap_iter_init(or_history
->link_history_map
);
672 !digestmap_iter_done(lhist_it
);
673 lhist_it
= digestmap_iter_next(or_history
->link_history_map
,
675 digestmap_iter_get(lhist_it
, &digest2
, &link_history_p
);
676 if ((r
= router_get_by_digest(digest2
)))
681 link_history
= (link_history_t
*) link_history_p
;
683 ret
= tor_snprintf(buffer
+len
, 2048-len
, "%s(%ld/%ld); ", name2
,
684 link_history
->n_extend_ok
,
685 link_history
->n_extend_ok
+link_history
->n_extend_fail
);
691 log(severity
, LD_HIST
, "%s", buffer
);
696 /** Remove history info for routers/links that haven't changed since
700 rep_history_clean(time_t before
)
702 int authority
= authdir_mode(get_options());
703 or_history_t
*or_history
;
704 link_history_t
*link_history
;
705 void *or_history_p
, *link_history_p
;
706 digestmap_iter_t
*orhist_it
, *lhist_it
;
709 orhist_it
= digestmap_iter_init(history_map
);
710 while (!digestmap_iter_done(orhist_it
)) {
712 digestmap_iter_get(orhist_it
, &d1
, &or_history_p
);
713 or_history
= or_history_p
;
715 remove
= authority
? (or_history
->total_run_weights
< STABILITY_EPSILON
&&
716 !or_history
->start_of_run
)
717 : (or_history
->changed
< before
);
719 orhist_it
= digestmap_iter_next_rmv(history_map
, orhist_it
);
720 free_or_history(or_history
);
723 for (lhist_it
= digestmap_iter_init(or_history
->link_history_map
);
724 !digestmap_iter_done(lhist_it
); ) {
725 digestmap_iter_get(lhist_it
, &d2
, &link_history_p
);
726 link_history
= link_history_p
;
727 if (link_history
->changed
< before
) {
728 lhist_it
= digestmap_iter_next_rmv(or_history
->link_history_map
,
730 rephist_total_alloc
-= sizeof(link_history_t
);
731 tor_free(link_history
);
734 lhist_it
= digestmap_iter_next(or_history
->link_history_map
,lhist_it
);
736 orhist_it
= digestmap_iter_next(history_map
, orhist_it
);
740 /** Write MTBF data to disk. Return 0 on success, negative on failure.
742 * If <b>missing_means_down</b>, then if we're about to write an entry
743 * that is still considered up but isn't in our routerlist, consider it
746 rep_hist_record_mtbf_data(time_t now
, int missing_means_down
)
748 char time_buf
[ISO_TIME_LEN
+1];
750 digestmap_iter_t
*orhist_it
;
754 open_file_t
*open_file
= NULL
;
758 char *filename
= get_datadir_fname("router-stability");
759 f
= start_writing_to_stdio_file(filename
, OPEN_FLAGS_REPLACE
|O_TEXT
, 0600,
767 * FormatLine *KeywordLine Data
769 * FormatLine = "format 1" NL
770 * KeywordLine = Keyword SP Arguments NL
771 * Data = "data" NL *RouterMTBFLine "." NL
772 * RouterMTBFLine = Fingerprint SP WeightedRunLen SP
773 * TotalRunWeights [SP S=StartRunTime] NL
775 #define PUT(s) STMT_BEGIN if (fputs((s),f)<0) goto err; STMT_END
776 #define PRINTF(args) STMT_BEGIN if (fprintf args <0) goto err; STMT_END
780 format_iso_time(time_buf
, time(NULL
));
781 PRINTF((f
, "stored-at %s\n", time_buf
));
783 if (started_tracking_stability
) {
784 format_iso_time(time_buf
, started_tracking_stability
);
785 PRINTF((f
, "tracked-since %s\n", time_buf
));
787 if (stability_last_downrated
) {
788 format_iso_time(time_buf
, stability_last_downrated
);
789 PRINTF((f
, "last-downrated %s\n", time_buf
));
794 /* XXX Nick: now bridge auths record this for all routers too.
795 * Should we make them record it only for bridge routers? -RD
796 * Not for 0.2.0. -NM */
797 for (orhist_it
= digestmap_iter_init(history_map
);
798 !digestmap_iter_done(orhist_it
);
799 orhist_it
= digestmap_iter_next(history_map
,orhist_it
)) {
800 char dbuf
[HEX_DIGEST_LEN
+1];
801 const char *t
= NULL
;
802 digestmap_iter_get(orhist_it
, &digest
, &or_history_p
);
803 hist
= (or_history_t
*) or_history_p
;
805 base16_encode(dbuf
, sizeof(dbuf
), digest
, DIGEST_LEN
);
807 if (missing_means_down
&& hist
->start_of_run
&&
808 !router_get_by_digest(digest
)) {
809 /* We think this relay is running, but it's not listed in our
810 * routerlist. Somehow it fell out without telling us it went
811 * down. Complain and also correct it. */
813 "Relay '%s' is listed as up in rephist, but it's not in "
814 "our routerlist. Correcting.", dbuf
);
815 rep_hist_note_router_unreachable(digest
, now
);
818 PRINTF((f
, "R %s\n", dbuf
));
819 if (hist
->start_of_run
> 0) {
820 format_iso_time(time_buf
, hist
->start_of_run
);
823 PRINTF((f
, "+MTBF %lu %.5lf%s%s\n",
824 hist
->weighted_run_length
, hist
->total_run_weights
,
825 t
? " S=" : "", t
? t
: ""));
827 if (hist
->start_of_downtime
> 0) {
828 format_iso_time(time_buf
, hist
->start_of_downtime
);
831 PRINTF((f
, "+WFU %lu %lu%s%s\n",
832 hist
->weighted_uptime
, hist
->total_weighted_time
,
833 t
? " S=" : "", t
? t
: ""));
841 return finish_writing_to_file(open_file
);
843 abort_writing_to_file(open_file
);
847 /** Format the current tracked status of the router in <b>hist</b> at time
848 * <b>now</b> for analysis; return it in a newly allocated string. */
850 rep_hist_format_router_status(or_history_t
*hist
, time_t now
)
852 char sor_buf
[ISO_TIME_LEN
+1];
853 char sod_buf
[ISO_TIME_LEN
+1];
856 int up
= 0, down
= 0;
859 if (hist
->start_of_run
) {
860 format_iso_time(sor_buf
, hist
->start_of_run
);
863 if (hist
->start_of_downtime
) {
864 format_iso_time(sod_buf
, hist
->start_of_downtime
);
868 wfu
= get_weighted_fractional_uptime(hist
, now
);
869 mtbf
= get_stability(hist
, now
);
874 " weighted-time %lu\n"
875 " weighted-uptime %lu\n"
877 " weighted-run-length %lu\n"
878 " total-run-weights %lf\n",
879 up
?"uptime-started ":"", up
?sor_buf
:"", up
?" UTC\n":"",
880 down
?"downtime-started ":"", down
?sod_buf
:"", down
?" UTC\n":"",
882 hist
->total_weighted_time
,
883 hist
->weighted_uptime
,
885 hist
->weighted_run_length
,
886 hist
->total_run_weights
891 /** The last stability analysis document that we created, or NULL if we never
892 * have created one. */
893 static char *last_stability_doc
= NULL
;
894 /** The last time we created a stability analysis document, or 0 if we never
895 * have created one. */
896 static time_t built_last_stability_doc_at
= 0;
897 /** Shortest allowable time between building two stability documents. */
898 #define MAX_STABILITY_DOC_BUILD_RATE (3*60)
900 /** Return a pointer to a NUL-terminated document describing our view of the
901 * stability of the routers we've been tracking. Return NULL on failure. */
903 rep_hist_get_router_stability_doc(time_t now
)
907 if (built_last_stability_doc_at
+ MAX_STABILITY_DOC_BUILD_RATE
> now
)
908 return last_stability_doc
;
913 tor_free(last_stability_doc
);
914 chunks
= smartlist_create();
916 if (rep_hist_have_measured_enough_stability()) {
917 smartlist_add(chunks
, tor_strdup("we-have-enough-measurements\n"));
919 smartlist_add(chunks
, tor_strdup("we-do-not-have-enough-measurements\n"));
922 DIGESTMAP_FOREACH(history_map
, id
, or_history_t
*, hist
) {
924 char dbuf
[BASE64_DIGEST_LEN
+1];
925 char header_buf
[512];
927 digest_to_base64(dbuf
, id
);
928 ri
= router_get_by_digest(id
);
930 char *ip
= tor_dup_ip(ri
->addr
);
931 char tbuf
[ISO_TIME_LEN
+1];
932 format_iso_time(tbuf
, ri
->cache_info
.published_on
);
933 tor_snprintf(header_buf
, sizeof(header_buf
),
936 "relevant-flags %s%s%s\n"
937 "declared-uptime %ld\n",
938 dbuf
, ri
->nickname
, ip
,
940 ri
->is_running
? "Running " : "",
941 ri
->is_valid
? "Valid " : "",
942 ri
->is_hibernating
? "Hibernating " : "",
946 tor_snprintf(header_buf
, sizeof(header_buf
),
947 "router %s {no descriptor}\n", dbuf
);
949 smartlist_add(chunks
, tor_strdup(header_buf
));
950 info
= rep_hist_format_router_status(hist
, now
);
952 smartlist_add(chunks
, info
);
954 } DIGESTMAP_FOREACH_END
;
956 result
= smartlist_join_strings(chunks
, "", 0, NULL
);
957 SMARTLIST_FOREACH(chunks
, char *, cp
, tor_free(cp
));
958 smartlist_free(chunks
);
960 last_stability_doc
= result
;
961 built_last_stability_doc_at
= time(NULL
);
965 /** Helper: return the first j >= i such that !strcmpstart(sl[j], prefix) and
966 * such that no line sl[k] with i <= k < j starts with "R ". Return -1 if no
967 * such line exists. */
969 find_next_with(smartlist_t
*sl
, int i
, const char *prefix
)
971 for ( ; i
< smartlist_len(sl
); ++i
) {
972 const char *line
= smartlist_get(sl
, i
);
973 if (!strcmpstart(line
, prefix
))
975 if (!strcmpstart(line
, "R "))
981 /** How many bad times has parse_possibly_bad_iso_time parsed? */
982 static int n_bogus_times
= 0;
983 /** Parse the ISO-formatted time in <b>s</b> into *<b>time_out</b>, but
984 * rounds any pre-1970 date to Jan 1, 1970. */
986 parse_possibly_bad_iso_time(const char *s
, time_t *time_out
)
990 strlcpy(b
, s
, sizeof(b
));
992 year
= (int)tor_parse_long(b
, 10, 0, INT_MAX
, NULL
, NULL
);
998 return parse_iso_time(s
, time_out
);
1001 /** We've read a time <b>t</b> from a file stored at <b>stored_at</b>, which
1002 * says we started measuring at <b>started_measuring</b>. Return a new number
1003 * that's about as much before <b>now</b> as <b>t</b> was before
1006 static INLINE
time_t
1007 correct_time(time_t t
, time_t now
, time_t stored_at
, time_t started_measuring
)
1009 if (t
< started_measuring
- 24*60*60*365)
1011 else if (t
< started_measuring
)
1012 return started_measuring
;
1013 else if (t
> stored_at
)
1016 long run_length
= stored_at
- t
;
1017 t
= now
- run_length
;
1018 if (t
< started_measuring
)
1019 t
= started_measuring
;
1024 /** Load MTBF data from disk. Returns 0 on success or recoverable error, -1
1027 rep_hist_load_mtbf_data(time_t now
)
1029 /* XXXX won't handle being called while history is already populated. */
1031 const char *line
= NULL
;
1033 time_t last_downrated
= 0, stored_at
= 0, tracked_since
= 0;
1034 time_t latest_possible_start
= now
;
1038 char *filename
= get_datadir_fname("router-stability");
1039 char *d
= read_file_to_str(filename
, RFTS_IGNORE_MISSING
, NULL
);
1043 lines
= smartlist_create();
1044 smartlist_split_string(lines
, d
, "\n", SPLIT_SKIP_SPACE
, 0);
1049 const char *firstline
;
1050 if (smartlist_len(lines
)>4) {
1051 firstline
= smartlist_get(lines
, 0);
1052 if (!strcmpstart(firstline
, "format "))
1053 format
= tor_parse_long(firstline
+strlen("format "),
1054 10, -1, LONG_MAX
, NULL
, NULL
);
1057 if (format
!= 1 && format
!= 2) {
1059 "Unrecognized format in mtbf history file. Skipping.");
1062 for (i
= 1; i
< smartlist_len(lines
); ++i
) {
1063 line
= smartlist_get(lines
, i
);
1064 if (!strcmp(line
, "data"))
1066 if (!strcmpstart(line
, "last-downrated ")) {
1067 if (parse_iso_time(line
+strlen("last-downrated "), &last_downrated
)<0)
1068 log_warn(LD_HIST
,"Couldn't parse downrate time in mtbf "
1071 if (!strcmpstart(line
, "stored-at ")) {
1072 if (parse_iso_time(line
+strlen("stored-at "), &stored_at
)<0)
1073 log_warn(LD_HIST
,"Couldn't parse stored time in mtbf "
1076 if (!strcmpstart(line
, "tracked-since ")) {
1077 if (parse_iso_time(line
+strlen("tracked-since "), &tracked_since
)<0)
1078 log_warn(LD_HIST
,"Couldn't parse started-tracking time in mtbf "
1082 if (last_downrated
> now
)
1083 last_downrated
= now
;
1084 if (tracked_since
> now
)
1085 tracked_since
= now
;
1088 log_warn(LD_HIST
, "No stored time recorded.");
1092 if (line
&& !strcmp(line
, "data"))
1097 for (; i
< smartlist_len(lines
); ++i
) {
1098 char digest
[DIGEST_LEN
];
1099 char hexbuf
[HEX_DIGEST_LEN
+1];
1100 char mtbf_timebuf
[ISO_TIME_LEN
+1];
1101 char wfu_timebuf
[ISO_TIME_LEN
+1];
1102 time_t start_of_run
= 0;
1103 time_t start_of_downtime
= 0;
1104 int have_mtbf
= 0, have_wfu
= 0;
1107 long wt_uptime
= 0, total_wt_time
= 0;
1110 line
= smartlist_get(lines
, i
);
1111 if (!strcmp(line
, "."))
1114 mtbf_timebuf
[0] = '\0';
1115 wfu_timebuf
[0] = '\0';
1118 n
= sscanf(line
, "%40s %ld %lf S=%10s %8s",
1119 hexbuf
, &wrl
, &trw
, mtbf_timebuf
, mtbf_timebuf
+11);
1120 if (n
!= 3 && n
!= 5) {
1121 log_warn(LD_HIST
, "Couldn't scan line %s", escaped(line
));
1127 int mtbf_idx
, wfu_idx
;
1128 if (strcmpstart(line
, "R ") || strlen(line
) < 2+HEX_DIGEST_LEN
)
1130 strlcpy(hexbuf
, line
+2, sizeof(hexbuf
));
1131 mtbf_idx
= find_next_with(lines
, i
+1, "+MTBF ");
1132 wfu_idx
= find_next_with(lines
, i
+1, "+WFU ");
1133 if (mtbf_idx
>= 0) {
1134 const char *mtbfline
= smartlist_get(lines
, mtbf_idx
);
1135 n
= sscanf(mtbfline
, "+MTBF %lu %lf S=%10s %8s",
1136 &wrl
, &trw
, mtbf_timebuf
, mtbf_timebuf
+11);
1137 if (n
== 2 || n
== 4) {
1140 log_warn(LD_HIST
, "Couldn't scan +MTBF line %s",
1145 const char *wfuline
= smartlist_get(lines
, wfu_idx
);
1146 n
= sscanf(wfuline
, "+WFU %lu %lu S=%10s %8s",
1147 &wt_uptime
, &total_wt_time
,
1148 wfu_timebuf
, wfu_timebuf
+11);
1149 if (n
== 2 || n
== 4) {
1152 log_warn(LD_HIST
, "Couldn't scan +WFU line %s", escaped(wfuline
));
1160 if (base16_decode(digest
, DIGEST_LEN
, hexbuf
, HEX_DIGEST_LEN
) < 0) {
1161 log_warn(LD_HIST
, "Couldn't hex string %s", escaped(hexbuf
));
1164 hist
= get_or_history(digest
);
1169 if (mtbf_timebuf
[0]) {
1170 mtbf_timebuf
[10] = ' ';
1171 if (parse_possibly_bad_iso_time(mtbf_timebuf
, &start_of_run
)<0)
1172 log_warn(LD_HIST
, "Couldn't parse time %s",
1173 escaped(mtbf_timebuf
));
1175 hist
->start_of_run
= correct_time(start_of_run
, now
, stored_at
,
1177 if (hist
->start_of_run
< latest_possible_start
+ wrl
)
1178 latest_possible_start
= hist
->start_of_run
- wrl
;
1180 hist
->weighted_run_length
= wrl
;
1181 hist
->total_run_weights
= trw
;
1184 if (wfu_timebuf
[0]) {
1185 wfu_timebuf
[10] = ' ';
1186 if (parse_possibly_bad_iso_time(wfu_timebuf
, &start_of_downtime
)<0)
1187 log_warn(LD_HIST
, "Couldn't parse time %s", escaped(wfu_timebuf
));
1190 hist
->start_of_downtime
= correct_time(start_of_downtime
, now
, stored_at
,
1192 hist
->weighted_uptime
= wt_uptime
;
1193 hist
->total_weighted_time
= total_wt_time
;
1195 if (strcmp(line
, "."))
1196 log_warn(LD_HIST
, "Truncated MTBF file.");
1198 if (tracked_since
< 86400*365) /* Recover from insanely early value. */
1199 tracked_since
= latest_possible_start
;
1201 stability_last_downrated
= last_downrated
;
1202 started_tracking_stability
= tracked_since
;
1208 SMARTLIST_FOREACH(lines
, char *, cp
, tor_free(cp
));
1209 smartlist_free(lines
);
1213 /** For how many seconds do we keep track of individual per-second bandwidth
1215 #define NUM_SECS_ROLLING_MEASURE 10
1216 /** How large are the intervals for which we track and report bandwidth use? */
1217 /* XXXX Watch out! Before Tor 0.2.2.21-alpha, using any other value here would
1218 * generate an unparseable state file. */
1219 #define NUM_SECS_BW_SUM_INTERVAL (15*60)
1220 /** How far in the past do we remember and publish bandwidth use? */
1221 #define NUM_SECS_BW_SUM_IS_VALID (24*60*60)
1222 /** How many bandwidth usage intervals do we remember? (derived) */
1223 #define NUM_TOTALS (NUM_SECS_BW_SUM_IS_VALID/NUM_SECS_BW_SUM_INTERVAL)
1225 /** Structure to track bandwidth use, and remember the maxima for a given
1228 typedef struct bw_array_t
{
1229 /** Observation array: Total number of bytes transferred in each of the last
1230 * NUM_SECS_ROLLING_MEASURE seconds. This is used as a circular array. */
1231 uint64_t obs
[NUM_SECS_ROLLING_MEASURE
];
1232 int cur_obs_idx
; /**< Current position in obs. */
1233 time_t cur_obs_time
; /**< Time represented in obs[cur_obs_idx] */
1234 uint64_t total_obs
; /**< Total for all members of obs except
1235 * obs[cur_obs_idx] */
1236 uint64_t max_total
; /**< Largest value that total_obs has taken on in the
1237 * current period. */
1238 uint64_t total_in_period
; /**< Total bytes transferred in the current
1241 /** When does the next period begin? */
1243 /** Where in 'maxima' should the maximum bandwidth usage for the current
1244 * period be stored? */
1246 /** How many values in maxima/totals have been set ever? */
1248 /** Circular array of the maximum
1249 * bandwidth-per-NUM_SECS_ROLLING_MEASURE usage for the last
1250 * NUM_TOTALS periods */
1251 uint64_t maxima
[NUM_TOTALS
];
1252 /** Circular array of the total bandwidth usage for the last NUM_TOTALS
1254 uint64_t totals
[NUM_TOTALS
];
1257 /** Shift the current period of b forward by one. */
1259 commit_max(bw_array_t
*b
)
1261 /* Store total from current period. */
1262 b
->totals
[b
->next_max_idx
] = b
->total_in_period
;
1263 /* Store maximum from current period. */
1264 b
->maxima
[b
->next_max_idx
++] = b
->max_total
;
1265 /* Advance next_period and next_max_idx */
1266 b
->next_period
+= NUM_SECS_BW_SUM_INTERVAL
;
1267 if (b
->next_max_idx
== NUM_TOTALS
)
1268 b
->next_max_idx
= 0;
1269 if (b
->num_maxes_set
< NUM_TOTALS
)
1271 /* Reset max_total. */
1273 /* Reset total_in_period. */
1274 b
->total_in_period
= 0;
1277 /** Shift the current observation time of 'b' forward by one second. */
1279 advance_obs(bw_array_t
*b
)
1284 /* Calculate the total bandwidth for the last NUM_SECS_ROLLING_MEASURE
1285 * seconds; adjust max_total as needed.*/
1286 total
= b
->total_obs
+ b
->obs
[b
->cur_obs_idx
];
1287 if (total
> b
->max_total
)
1288 b
->max_total
= total
;
1290 nextidx
= b
->cur_obs_idx
+1;
1291 if (nextidx
== NUM_SECS_ROLLING_MEASURE
)
1294 b
->total_obs
= total
- b
->obs
[nextidx
];
1296 b
->cur_obs_idx
= nextidx
;
1298 if (++b
->cur_obs_time
>= b
->next_period
)
1302 /** Add <b>n</b> bytes to the number of bytes in <b>b</b> for second
1305 add_obs(bw_array_t
*b
, time_t when
, uint64_t n
)
1307 /* Don't record data in the past. */
1308 if (when
<b
->cur_obs_time
)
1310 /* If we're currently adding observations for an earlier second than
1311 * 'when', advance b->cur_obs_time and b->cur_obs_idx by an
1312 * appropriate number of seconds, and do all the other housekeeping */
1313 while (when
>b
->cur_obs_time
) {
1314 /* Doing this one second at a time is potentially inefficient, if we start
1315 with a state file that is very old. Fortunately, it doesn't seem to
1316 show up in profiles, so we can just ignore it for now. */
1320 b
->obs
[b
->cur_obs_idx
] += n
;
1321 b
->total_in_period
+= n
;
1324 /** Allocate, initialize, and return a new bw_array. */
1330 b
= tor_malloc_zero(sizeof(bw_array_t
));
1331 rephist_total_alloc
+= sizeof(bw_array_t
);
1333 b
->cur_obs_time
= start
;
1334 b
->next_period
= start
+ NUM_SECS_BW_SUM_INTERVAL
;
1338 /** Recent history of bandwidth observations for read operations. */
1339 static bw_array_t
*read_array
= NULL
;
1340 /** Recent history of bandwidth observations for write operations. */
1341 static bw_array_t
*write_array
= NULL
;
1342 /** Recent history of bandwidth observations for read operations for the
1343 directory protocol. */
1344 static bw_array_t
*dir_read_array
= NULL
;
1345 /** Recent history of bandwidth observations for write operations for the
1346 directory protocol. */
1347 static bw_array_t
*dir_write_array
= NULL
;
1349 /** Set up [dir-]read_array and [dir-]write_array, freeing them if they
1352 bw_arrays_init(void)
1354 tor_free(read_array
);
1355 tor_free(write_array
);
1356 tor_free(dir_read_array
);
1357 tor_free(dir_write_array
);
1358 read_array
= bw_array_new();
1359 write_array
= bw_array_new();
1360 dir_read_array
= bw_array_new();
1361 dir_write_array
= bw_array_new();
1364 /** We read <b>num_bytes</b> more bytes in second <b>when</b>.
1366 * Add num_bytes to the current running total for <b>when</b>.
1368 * <b>when</b> can go back to time, but it's safe to ignore calls
1369 * earlier than the latest <b>when</b> you've heard of.
1372 rep_hist_note_bytes_written(size_t num_bytes
, time_t when
)
1374 /* Maybe a circular array for recent seconds, and step to a new point
1375 * every time a new second shows up. Or simpler is to just to have
1376 * a normal array and push down each item every second; it's short.
1378 /* When a new second has rolled over, compute the sum of the bytes we've
1379 * seen over when-1 to when-1-NUM_SECS_ROLLING_MEASURE, and stick it
1380 * somewhere. See rep_hist_bandwidth_assess() below.
1382 add_obs(write_array
, when
, num_bytes
);
1385 /** We wrote <b>num_bytes</b> more bytes in second <b>when</b>.
1386 * (like rep_hist_note_bytes_written() above)
1389 rep_hist_note_bytes_read(size_t num_bytes
, time_t when
)
1391 /* if we're smart, we can make this func and the one above share code */
1392 add_obs(read_array
, when
, num_bytes
);
1395 /** We wrote <b>num_bytes</b> more directory bytes in second <b>when</b>.
1396 * (like rep_hist_note_bytes_written() above)
1399 rep_hist_note_dir_bytes_written(size_t num_bytes
, time_t when
)
1401 add_obs(dir_write_array
, when
, num_bytes
);
1404 /** We read <b>num_bytes</b> more directory bytes in second <b>when</b>.
1405 * (like rep_hist_note_bytes_written() above)
1408 rep_hist_note_dir_bytes_read(size_t num_bytes
, time_t when
)
1410 add_obs(dir_read_array
, when
, num_bytes
);
1413 /** Helper: Return the largest value in b->maxima. (This is equal to the
1414 * most bandwidth used in any NUM_SECS_ROLLING_MEASURE period for the last
1415 * NUM_SECS_BW_SUM_IS_VALID seconds.)
1418 find_largest_max(bw_array_t
*b
)
1423 for (i
=0; i
<NUM_TOTALS
; ++i
) {
1424 if (b
->maxima
[i
]>max
)
1430 /** Find the largest sums in the past NUM_SECS_BW_SUM_IS_VALID (roughly)
1431 * seconds. Find one sum for reading and one for writing. They don't have
1432 * to be at the same time.
1434 * Return the smaller of these sums, divided by NUM_SECS_ROLLING_MEASURE.
1437 rep_hist_bandwidth_assess(void)
1440 r
= find_largest_max(read_array
);
1441 w
= find_largest_max(write_array
);
1443 return (int)(U64_TO_DBL(w
)/NUM_SECS_ROLLING_MEASURE
);
1445 return (int)(U64_TO_DBL(r
)/NUM_SECS_ROLLING_MEASURE
);
1448 /** Print the bandwidth history of b (either [dir-]read_array or
1449 * [dir-]write_array) into the buffer pointed to by buf. The format is
1450 * simply comma separated numbers, from oldest to newest.
1452 * It returns the number of bytes written.
1455 rep_hist_fill_bandwidth_history(char *buf
, size_t len
, const bw_array_t
*b
)
1459 or_options_t
*options
= get_options();
1462 if (b
->num_maxes_set
<= b
->next_max_idx
) {
1463 /* We haven't been through the circular array yet; time starts at i=0.*/
1466 /* We've been around the array at least once. The next i to be
1467 overwritten is the oldest. */
1468 i
= b
->next_max_idx
;
1471 if (options
->RelayBandwidthRate
) {
1472 /* We don't want to report that we used more bandwidth than the max we're
1473 * willing to relay; otherwise everybody will know how much traffic
1474 * we used ourself. */
1475 cutoff
= options
->RelayBandwidthRate
* NUM_SECS_BW_SUM_INTERVAL
;
1477 cutoff
= UINT64_MAX
;
1480 for (n
=0; n
<b
->num_maxes_set
; ++n
,++i
) {
1482 if (i
>= NUM_TOTALS
)
1484 tor_assert(i
< NUM_TOTALS
);
1485 /* Round the bandwidth used down to the nearest 1k. */
1486 total
= b
->totals
[i
] & ~0x3ff;
1490 if (n
==(b
->num_maxes_set
-1))
1491 tor_snprintf(cp
, len
-(cp
-buf
), U64_FORMAT
, U64_PRINTF_ARG(total
));
1493 tor_snprintf(cp
, len
-(cp
-buf
), U64_FORMAT
",", U64_PRINTF_ARG(total
));
1499 /** Allocate and return lines for representing this server's bandwidth
1500 * history in its descriptor.
1503 rep_hist_get_bandwidth_lines(void)
1506 char t
[ISO_TIME_LEN
+1];
1508 bw_array_t
*b
= NULL
;
1509 const char *desc
= NULL
;
1512 /* opt [dirreq-](read|write)-history yyyy-mm-dd HH:MM:SS (n s) n,n,n... */
1513 len
= (67+21*NUM_TOTALS
)*4;
1514 buf
= tor_malloc_zero(len
);
1520 desc
= "write-history";
1524 desc
= "read-history";
1527 b
= dir_write_array
;
1528 desc
= "dirreq-write-history";
1532 desc
= "dirreq-read-history";
1536 format_iso_time(t
, b
->next_period
-NUM_SECS_BW_SUM_INTERVAL
);
1537 tor_snprintf(cp
, len
-(cp
-buf
), "%s %s (%d s) ",
1538 desc
, t
, NUM_SECS_BW_SUM_INTERVAL
);
1540 cp
+= rep_hist_fill_bandwidth_history(cp
, len
-(cp
-buf
), b
);
1541 strlcat(cp
, "\n", len
-(cp
-buf
));
1547 /** Write a single bw_array_t into the Values, Ends, Interval, and Maximum
1548 * entries of an or_state_t. */
1550 rep_hist_update_bwhist_state_section(or_state_t
*state
,
1551 const bw_array_t
*b
,
1552 smartlist_t
**s_values
,
1553 smartlist_t
**s_maxima
,
1561 SMARTLIST_FOREACH(*s_values
, char *, val
, tor_free(val
));
1562 smartlist_free(*s_values
);
1565 SMARTLIST_FOREACH(*s_maxima
, char *, val
, tor_free(val
));
1566 smartlist_free(*s_maxima
);
1568 if (! server_mode(get_options())) {
1569 /* Clients don't need to store bandwidth history persistently;
1570 * force these values to the defaults. */
1571 /* FFFF we should pull the default out of config.c's state table,
1572 * so we don't have two defaults. */
1573 if (*s_begins
!= 0 || *s_interval
!= 900) {
1574 time_t now
= time(NULL
);
1575 time_t save_at
= get_options()->AvoidDiskWrites
? now
+3600 : now
+600;
1576 or_state_mark_dirty(state
, save_at
);
1580 *s_values
= smartlist_create();
1581 *s_maxima
= smartlist_create();
1584 *s_begins
= b
->next_period
;
1585 *s_interval
= NUM_SECS_BW_SUM_INTERVAL
;
1587 *s_values
= smartlist_create();
1588 *s_maxima
= smartlist_create();
1589 /* Set i to first position in circular array */
1590 i
= (b
->num_maxes_set
<= b
->next_max_idx
) ? 0 : b
->next_max_idx
;
1591 for (j
=0; j
< b
->num_maxes_set
; ++j
,++i
) {
1593 if (i
>= NUM_TOTALS
)
1595 tor_asprintf(&cp
, U64_FORMAT
, U64_PRINTF_ARG(b
->totals
[i
] & ~0x3ff));
1596 smartlist_add(*s_values
, cp
);
1597 maxval
= b
->maxima
[i
] / NUM_SECS_ROLLING_MEASURE
;
1598 tor_asprintf(&cp
, U64_FORMAT
, U64_PRINTF_ARG(maxval
& ~0x3ff));
1599 smartlist_add(*s_maxima
, cp
);
1601 tor_asprintf(&cp
, U64_FORMAT
, U64_PRINTF_ARG(b
->total_in_period
& ~0x3ff));
1602 smartlist_add(*s_values
, cp
);
1603 tor_asprintf(&cp
, U64_FORMAT
, U64_PRINTF_ARG(b
->max_total
& ~0x3ff));
1604 smartlist_add(*s_maxima
, cp
);
1607 /** Update <b>state</b> with the newest bandwidth history. */
1609 rep_hist_update_state(or_state_t
*state
)
1611 #define UPDATE(arrname,st) \
1612 rep_hist_update_bwhist_state_section(state,\
1614 &state->BWHistory ## st ## Values, \
1615 &state->BWHistory ## st ## Maxima, \
1616 &state->BWHistory ## st ## Ends, \
1617 &state->BWHistory ## st ## Interval)
1619 UPDATE(write_array
, Write
);
1620 UPDATE(read_array
, Read
);
1621 UPDATE(dir_write_array
, DirWrite
);
1622 UPDATE(dir_read_array
, DirRead
);
1624 if (server_mode(get_options())) {
1625 or_state_mark_dirty(state
, time(NULL
)+(2*3600));
1630 /** Load a single bw_array_t from its Values, Ends, Maxima, and Interval
1631 * entries in an or_state_t. */
1633 rep_hist_load_bwhist_state_section(bw_array_t
*b
,
1634 const smartlist_t
*s_values
,
1635 const smartlist_t
*s_maxima
,
1636 const time_t s_begins
,
1637 const int s_interval
)
1639 time_t now
= time(NULL
);
1645 int have_maxima
= (smartlist_len(s_values
) == smartlist_len(s_maxima
));
1647 if (s_values
&& s_begins
>= now
- NUM_SECS_BW_SUM_INTERVAL
*NUM_TOTALS
) {
1648 start
= s_begins
- s_interval
*(smartlist_len(s_values
));
1651 b
->cur_obs_time
= start
;
1652 b
->next_period
= start
+ NUM_SECS_BW_SUM_INTERVAL
;
1653 SMARTLIST_FOREACH_BEGIN(s_values
, const char *, cp
) {
1654 const char *maxstr
= NULL
;
1655 v
= tor_parse_uint64(cp
, 10, 0, UINT64_MAX
, &ok
, NULL
);
1657 maxstr
= smartlist_get(s_maxima
, cp_sl_idx
);
1658 mv
= tor_parse_uint64(maxstr
, 10, 0, UINT64_MAX
, &ok_m
, NULL
);
1659 mv
*= NUM_SECS_ROLLING_MEASURE
;
1661 /* No maxima known; guess average rate to be conservative. */
1662 mv
= v
/ s_interval
;
1666 log_notice(LD_HIST
, "Could not parse value '%s' into a number.'",cp
);
1668 if (maxstr
&& !ok_m
) {
1670 log_notice(LD_HIST
, "Could not parse maximum '%s' into a number.'",
1675 add_obs(b
, start
, v
);
1677 /* This will result in some fairly choppy history if s_interval
1678 * is notthe same as NUM_SECS_BW_SUM_INTERVAL. XXXX */
1679 start
+= s_interval
;
1681 } SMARTLIST_FOREACH_END(cp
);
1684 /* Clean up maxima and observed */
1685 for (i
=0; i
<NUM_SECS_ROLLING_MEASURE
; ++i
) {
1693 /** Set bandwidth history from our saved state. */
1695 rep_hist_load_state(or_state_t
*state
, char **err
)
1699 /* Assert they already have been malloced */
1700 tor_assert(read_array
&& write_array
);
1701 tor_assert(dir_read_array
&& dir_write_array
);
1703 #define LOAD(arrname,st) \
1704 if (rep_hist_load_bwhist_state_section( \
1706 state->BWHistory ## st ## Values, \
1707 state->BWHistory ## st ## Maxima, \
1708 state->BWHistory ## st ## Ends, \
1709 state->BWHistory ## st ## Interval)<0) \
1712 LOAD(write_array
, Write
);
1713 LOAD(read_array
, Read
);
1714 LOAD(dir_write_array
, DirWrite
);
1715 LOAD(dir_read_array
, DirRead
);
1719 *err
= tor_strdup("Parsing of bandwidth history values failed");
1720 /* and create fresh arrays */
1727 /*********************************************************************/
1729 /** A list of port numbers that have been used recently. */
1730 static smartlist_t
*predicted_ports_list
=NULL
;
1731 /** The corresponding most recently used time for each port. */
1732 static smartlist_t
*predicted_ports_times
=NULL
;
1734 /** We just got an application request for a connection with
1735 * port <b>port</b>. Remember it for the future, so we can keep
1736 * some circuits open that will exit to this port.
1739 add_predicted_port(time_t now
, uint16_t port
)
1741 /* XXXX we could just use uintptr_t here, I think. */
1742 uint16_t *tmp_port
= tor_malloc(sizeof(uint16_t));
1743 time_t *tmp_time
= tor_malloc(sizeof(time_t));
1746 rephist_total_alloc
+= sizeof(uint16_t) + sizeof(time_t);
1747 smartlist_add(predicted_ports_list
, tmp_port
);
1748 smartlist_add(predicted_ports_times
, tmp_time
);
1751 /** Initialize whatever memory and structs are needed for predicting
1752 * which ports will be used. Also seed it with port 80, so we'll build
1753 * circuits on start-up.
1756 predicted_ports_init(void)
1758 predicted_ports_list
= smartlist_create();
1759 predicted_ports_times
= smartlist_create();
1760 add_predicted_port(time(NULL
), 80); /* add one to kickstart us */
1763 /** Free whatever memory is needed for predicting which ports will
1767 predicted_ports_free(void)
1769 rephist_total_alloc
-= smartlist_len(predicted_ports_list
)*sizeof(uint16_t);
1770 SMARTLIST_FOREACH(predicted_ports_list
, char *, cp
, tor_free(cp
));
1771 smartlist_free(predicted_ports_list
);
1772 rephist_total_alloc
-= smartlist_len(predicted_ports_times
)*sizeof(time_t);
1773 SMARTLIST_FOREACH(predicted_ports_times
, char *, cp
, tor_free(cp
));
1774 smartlist_free(predicted_ports_times
);
1777 /** Remember that <b>port</b> has been asked for as of time <b>now</b>.
1778 * This is used for predicting what sorts of streams we'll make in the
1779 * future and making exit circuits to anticipate that.
1782 rep_hist_note_used_port(time_t now
, uint16_t port
)
1788 tor_assert(predicted_ports_list
);
1789 tor_assert(predicted_ports_times
);
1791 if (!port
) /* record nothing */
1794 for (i
= 0; i
< smartlist_len(predicted_ports_list
); ++i
) {
1795 tmp_port
= smartlist_get(predicted_ports_list
, i
);
1796 tmp_time
= smartlist_get(predicted_ports_times
, i
);
1797 if (*tmp_port
== port
) {
1802 /* it's not there yet; we need to add it */
1803 add_predicted_port(now
, port
);
1806 /** For this long after we've seen a request for a given port, assume that
1807 * we'll want to make connections to the same port in the future. */
1808 #define PREDICTED_CIRCS_RELEVANCE_TIME (60*60)
1810 /** Return a pointer to the list of port numbers that
1811 * are likely to be asked for in the near future.
1813 * The caller promises not to mess with it.
1816 rep_hist_get_predicted_ports(time_t now
)
1822 tor_assert(predicted_ports_list
);
1823 tor_assert(predicted_ports_times
);
1825 /* clean out obsolete entries */
1826 for (i
= 0; i
< smartlist_len(predicted_ports_list
); ++i
) {
1827 tmp_time
= smartlist_get(predicted_ports_times
, i
);
1828 if (*tmp_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
< now
) {
1829 tmp_port
= smartlist_get(predicted_ports_list
, i
);
1830 log_debug(LD_CIRC
, "Expiring predicted port %d", *tmp_port
);
1831 smartlist_del(predicted_ports_list
, i
);
1832 smartlist_del(predicted_ports_times
, i
);
1833 rephist_total_alloc
-= sizeof(uint16_t)+sizeof(time_t);
1839 return predicted_ports_list
;
1842 /** The user asked us to do a resolve. Rather than keeping track of
1843 * timings and such of resolves, we fake it for now by treating
1844 * it the same way as a connection to port 80. This way we will continue
1845 * to have circuits lying around if the user only uses Tor for resolves.
1848 rep_hist_note_used_resolve(time_t now
)
1850 rep_hist_note_used_port(now
, 80);
1853 /** The last time at which we needed an internal circ. */
1854 static time_t predicted_internal_time
= 0;
1855 /** The last time we needed an internal circ with good uptime. */
1856 static time_t predicted_internal_uptime_time
= 0;
1857 /** The last time we needed an internal circ with good capacity. */
1858 static time_t predicted_internal_capacity_time
= 0;
1860 /** Remember that we used an internal circ at time <b>now</b>. */
1862 rep_hist_note_used_internal(time_t now
, int need_uptime
, int need_capacity
)
1864 predicted_internal_time
= now
;
1866 predicted_internal_uptime_time
= now
;
1868 predicted_internal_capacity_time
= now
;
1871 /** Return 1 if we've used an internal circ recently; else return 0. */
1873 rep_hist_get_predicted_internal(time_t now
, int *need_uptime
,
1876 if (!predicted_internal_time
) { /* initialize it */
1877 predicted_internal_time
= now
;
1878 predicted_internal_uptime_time
= now
;
1879 predicted_internal_capacity_time
= now
;
1881 if (predicted_internal_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
< now
)
1882 return 0; /* too long ago */
1883 if (predicted_internal_uptime_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
>= now
)
1885 // Always predict that we need capacity.
1890 /** Any ports used lately? These are pre-seeded if we just started
1891 * up or if we're running a hidden service. */
1893 any_predicted_circuits(time_t now
)
1895 return smartlist_len(predicted_ports_list
) ||
1896 predicted_internal_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
>= now
;
1899 /** Return 1 if we have no need for circuits currently, else return 0. */
1901 rep_hist_circbuilding_dormant(time_t now
)
1903 if (any_predicted_circuits(now
))
1906 /* see if we'll still need to build testing circuits */
1907 if (server_mode(get_options()) &&
1908 (!check_whether_orport_reachable() || !circuit_enough_testing_circs()))
1910 if (!check_whether_dirport_reachable())
1916 /** Structure to track how many times we've done each public key operation. */
1918 /** How many directory objects have we signed? */
1919 unsigned long n_signed_dir_objs
;
1920 /** How many routerdescs have we signed? */
1921 unsigned long n_signed_routerdescs
;
1922 /** How many directory objects have we verified? */
1923 unsigned long n_verified_dir_objs
;
1924 /** How many routerdescs have we verified */
1925 unsigned long n_verified_routerdescs
;
1926 /** How many onionskins have we encrypted to build circuits? */
1927 unsigned long n_onionskins_encrypted
;
1928 /** How many onionskins have we decrypted to do circuit build requests? */
1929 unsigned long n_onionskins_decrypted
;
1930 /** How many times have we done the TLS handshake as a client? */
1931 unsigned long n_tls_client_handshakes
;
1932 /** How many times have we done the TLS handshake as a server? */
1933 unsigned long n_tls_server_handshakes
;
1934 /** How many PK operations have we done as a hidden service client? */
1935 unsigned long n_rend_client_ops
;
1936 /** How many PK operations have we done as a hidden service midpoint? */
1937 unsigned long n_rend_mid_ops
;
1938 /** How many PK operations have we done as a hidden service provider? */
1939 unsigned long n_rend_server_ops
;
1940 } pk_op_counts
= {0,0,0,0,0,0,0,0,0,0,0};
1942 /** Increment the count of the number of times we've done <b>operation</b>. */
1944 note_crypto_pk_op(pk_op_t operation
)
1949 pk_op_counts
.n_signed_dir_objs
++;
1952 pk_op_counts
.n_signed_routerdescs
++;
1955 pk_op_counts
.n_verified_dir_objs
++;
1958 pk_op_counts
.n_verified_routerdescs
++;
1961 pk_op_counts
.n_onionskins_encrypted
++;
1964 pk_op_counts
.n_onionskins_decrypted
++;
1966 case TLS_HANDSHAKE_C
:
1967 pk_op_counts
.n_tls_client_handshakes
++;
1969 case TLS_HANDSHAKE_S
:
1970 pk_op_counts
.n_tls_server_handshakes
++;
1973 pk_op_counts
.n_rend_client_ops
++;
1976 pk_op_counts
.n_rend_mid_ops
++;
1979 pk_op_counts
.n_rend_server_ops
++;
1982 log_warn(LD_BUG
, "Unknown pk operation %d", operation
);
1986 /** Log the number of times we've done each public/private-key operation. */
1988 dump_pk_ops(int severity
)
1990 log(severity
, LD_HIST
,
1991 "PK operations: %lu directory objects signed, "
1992 "%lu directory objects verified, "
1993 "%lu routerdescs signed, "
1994 "%lu routerdescs verified, "
1995 "%lu onionskins encrypted, "
1996 "%lu onionskins decrypted, "
1997 "%lu client-side TLS handshakes, "
1998 "%lu server-side TLS handshakes, "
1999 "%lu rendezvous client operations, "
2000 "%lu rendezvous middle operations, "
2001 "%lu rendezvous server operations.",
2002 pk_op_counts
.n_signed_dir_objs
,
2003 pk_op_counts
.n_verified_dir_objs
,
2004 pk_op_counts
.n_signed_routerdescs
,
2005 pk_op_counts
.n_verified_routerdescs
,
2006 pk_op_counts
.n_onionskins_encrypted
,
2007 pk_op_counts
.n_onionskins_decrypted
,
2008 pk_op_counts
.n_tls_client_handshakes
,
2009 pk_op_counts
.n_tls_server_handshakes
,
2010 pk_op_counts
.n_rend_client_ops
,
2011 pk_op_counts
.n_rend_mid_ops
,
2012 pk_op_counts
.n_rend_server_ops
);
2015 /*** Exit port statistics ***/
2017 /* Some constants */
2018 /** To what multiple should byte numbers be rounded up? */
2019 #define EXIT_STATS_ROUND_UP_BYTES 1024
2020 /** To what multiple should stream counts be rounded up? */
2021 #define EXIT_STATS_ROUND_UP_STREAMS 4
2022 /** Number of TCP ports */
2023 #define EXIT_STATS_NUM_PORTS 65536
2024 /** Top n ports that will be included in exit stats. */
2025 #define EXIT_STATS_TOP_N_PORTS 10
2027 /* The following data structures are arrays and no fancy smartlists or maps,
2028 * so that all write operations can be done in constant time. This comes at
2029 * the price of some memory (1.25 MB) and linear complexity when writing
2030 * stats for measuring relays. */
2031 /** Number of bytes read in current period by exit port */
2032 static uint64_t *exit_bytes_read
= NULL
;
2033 /** Number of bytes written in current period by exit port */
2034 static uint64_t *exit_bytes_written
= NULL
;
2035 /** Number of streams opened in current period by exit port */
2036 static uint32_t *exit_streams
= NULL
;
2038 /** Start time of exit stats or 0 if we're not collecting exit stats. */
2039 static time_t start_of_exit_stats_interval
;
2041 /** Initialize exit port stats. */
2043 rep_hist_exit_stats_init(time_t now
)
2045 start_of_exit_stats_interval
= now
;
2046 exit_bytes_read
= tor_malloc_zero(EXIT_STATS_NUM_PORTS
*
2048 exit_bytes_written
= tor_malloc_zero(EXIT_STATS_NUM_PORTS
*
2050 exit_streams
= tor_malloc_zero(EXIT_STATS_NUM_PORTS
*
2054 /** Reset counters for exit port statistics. */
2056 rep_hist_reset_exit_stats(time_t now
)
2058 start_of_exit_stats_interval
= now
;
2059 memset(exit_bytes_read
, 0, EXIT_STATS_NUM_PORTS
* sizeof(uint64_t));
2060 memset(exit_bytes_written
, 0, EXIT_STATS_NUM_PORTS
* sizeof(uint64_t));
2061 memset(exit_streams
, 0, EXIT_STATS_NUM_PORTS
* sizeof(uint32_t));
2064 /** Stop collecting exit port stats in a way that we can re-start doing
2065 * so in rep_hist_exit_stats_init(). */
2067 rep_hist_exit_stats_term(void)
2069 start_of_exit_stats_interval
= 0;
2070 tor_free(exit_bytes_read
);
2071 tor_free(exit_bytes_written
);
2072 tor_free(exit_streams
);
2075 /** Helper for qsort: compare two ints. */
2077 _compare_int(const void *x
, const void *y
)
2079 return (*(int*)x
- *(int*)y
);
2082 /** Return a newly allocated string containing the exit port statistics
2083 * until <b>now</b>, or NULL if we're not collecting exit stats. */
2085 rep_hist_format_exit_stats(time_t now
)
2087 int i
, j
, top_elements
= 0, cur_min_idx
= 0, cur_port
;
2088 uint64_t top_bytes
[EXIT_STATS_TOP_N_PORTS
];
2089 int top_ports
[EXIT_STATS_TOP_N_PORTS
];
2090 uint64_t cur_bytes
= 0, other_read
= 0, other_written
= 0,
2091 total_read
= 0, total_written
= 0;
2092 uint32_t total_streams
= 0, other_streams
= 0;
2094 smartlist_t
*written_strings
, *read_strings
, *streams_strings
;
2095 char *written_string
, *read_string
, *streams_string
;
2096 char t
[ISO_TIME_LEN
+1];
2099 if (!start_of_exit_stats_interval
)
2100 return NULL
; /* Not initialized. */
2102 /* Go through all ports to find the n ports that saw most written and
2105 * Invariant: at the end of the loop for iteration i,
2106 * total_read is the sum of all exit_bytes_read[0..i]
2107 * total_written is the sum of all exit_bytes_written[0..i]
2108 * total_stream is the sum of all exit_streams[0..i]
2110 * top_elements = MAX(EXIT_STATS_TOP_N_PORTS,
2111 * #{j | 0 <= j <= i && volume(i) > 0})
2113 * For all 0 <= j < top_elements,
2115 * 0 <= top_ports[j] <= 65535
2116 * top_bytes[j] = volume(top_ports[j])
2118 * There is no j in 0..i and k in 0..top_elements such that:
2119 * volume(j) > top_bytes[k] AND j is not in top_ports[0..top_elements]
2121 * There is no j!=cur_min_idx in 0..top_elements such that:
2122 * top_bytes[j] < top_bytes[cur_min_idx]
2124 * where volume(x) == exit_bytes_read[x]+exit_bytes_written[x]
2126 * Worst case: O(EXIT_STATS_NUM_PORTS * EXIT_STATS_TOP_N_PORTS)
2128 for (i
= 1; i
< EXIT_STATS_NUM_PORTS
; i
++) {
2129 total_read
+= exit_bytes_read
[i
];
2130 total_written
+= exit_bytes_written
[i
];
2131 total_streams
+= exit_streams
[i
];
2132 cur_bytes
= exit_bytes_read
[i
] + exit_bytes_written
[i
];
2133 if (cur_bytes
== 0) {
2136 if (top_elements
< EXIT_STATS_TOP_N_PORTS
) {
2137 top_bytes
[top_elements
] = cur_bytes
;
2138 top_ports
[top_elements
++] = i
;
2139 } else if (cur_bytes
> top_bytes
[cur_min_idx
]) {
2140 top_bytes
[cur_min_idx
] = cur_bytes
;
2141 top_ports
[cur_min_idx
] = i
;
2146 for (j
= 1; j
< top_elements
; j
++) {
2147 if (top_bytes
[j
] < top_bytes
[cur_min_idx
]) {
2153 /* Add observations of top ports to smartlists. */
2154 written_strings
= smartlist_create();
2155 read_strings
= smartlist_create();
2156 streams_strings
= smartlist_create();
2157 other_read
= total_read
;
2158 other_written
= total_written
;
2159 other_streams
= total_streams
;
2160 /* Sort the ports; this puts them out of sync with top_bytes, but we
2161 * won't be using top_bytes again anyway */
2162 qsort(top_ports
, top_elements
, sizeof(int), _compare_int
);
2163 for (j
= 0; j
< top_elements
; j
++) {
2164 cur_port
= top_ports
[j
];
2165 if (exit_bytes_written
[cur_port
] > 0) {
2166 uint64_t num
= round_uint64_to_next_multiple_of(
2167 exit_bytes_written
[cur_port
],
2168 EXIT_STATS_ROUND_UP_BYTES
);
2171 tor_asprintf(&buf
, "%d="U64_FORMAT
, cur_port
, U64_PRINTF_ARG(num
));
2172 smartlist_add(written_strings
, buf
);
2173 other_written
-= exit_bytes_written
[cur_port
];
2175 if (exit_bytes_read
[cur_port
] > 0) {
2176 uint64_t num
= round_uint64_to_next_multiple_of(
2177 exit_bytes_read
[cur_port
],
2178 EXIT_STATS_ROUND_UP_BYTES
);
2181 tor_asprintf(&buf
, "%d="U64_FORMAT
, cur_port
, U64_PRINTF_ARG(num
));
2182 smartlist_add(read_strings
, buf
);
2183 other_read
-= exit_bytes_read
[cur_port
];
2185 if (exit_streams
[cur_port
] > 0) {
2186 uint32_t num
= round_uint32_to_next_multiple_of(
2187 exit_streams
[cur_port
],
2188 EXIT_STATS_ROUND_UP_STREAMS
);
2190 tor_asprintf(&buf
, "%d=%u", cur_port
, num
);
2191 smartlist_add(streams_strings
, buf
);
2192 other_streams
-= exit_streams
[cur_port
];
2196 /* Add observations of other ports in a single element. */
2197 other_written
= round_uint64_to_next_multiple_of(other_written
,
2198 EXIT_STATS_ROUND_UP_BYTES
);
2199 other_written
/= 1024;
2201 tor_asprintf(&buf
, "other="U64_FORMAT
, U64_PRINTF_ARG(other_written
));
2202 smartlist_add(written_strings
, buf
);
2203 other_read
= round_uint64_to_next_multiple_of(other_read
,
2204 EXIT_STATS_ROUND_UP_BYTES
);
2207 tor_asprintf(&buf
, "other="U64_FORMAT
, U64_PRINTF_ARG(other_read
));
2208 smartlist_add(read_strings
, buf
);
2209 other_streams
= round_uint32_to_next_multiple_of(other_streams
,
2210 EXIT_STATS_ROUND_UP_STREAMS
);
2212 tor_asprintf(&buf
, "other=%u", other_streams
);
2213 smartlist_add(streams_strings
, buf
);
2215 /* Join all observations in single strings. */
2216 written_string
= smartlist_join_strings(written_strings
, ",", 0, NULL
);
2217 read_string
= smartlist_join_strings(read_strings
, ",", 0, NULL
);
2218 streams_string
= smartlist_join_strings(streams_strings
, ",", 0, NULL
);
2219 SMARTLIST_FOREACH(written_strings
, char *, cp
, tor_free(cp
));
2220 SMARTLIST_FOREACH(read_strings
, char *, cp
, tor_free(cp
));
2221 SMARTLIST_FOREACH(streams_strings
, char *, cp
, tor_free(cp
));
2222 smartlist_free(written_strings
);
2223 smartlist_free(read_strings
);
2224 smartlist_free(streams_strings
);
2226 /* Put everything together. */
2227 format_iso_time(t
, now
);
2228 tor_asprintf(&result
, "exit-stats-end %s (%d s)\n"
2229 "exit-kibibytes-written %s\n"
2230 "exit-kibibytes-read %s\n"
2231 "exit-streams-opened %s\n",
2232 t
, (unsigned) (now
- start_of_exit_stats_interval
),
2236 tor_free(written_string
);
2237 tor_free(read_string
);
2238 tor_free(streams_string
);
2242 /** If 24 hours have passed since the beginning of the current exit port
2243 * stats period, write exit stats to $DATADIR/stats/exit-stats (possibly
2244 * overwriting an existing file) and reset counters. Return when we would
2245 * next want to write exit stats or 0 if we never want to write. */
2247 rep_hist_exit_stats_write(time_t now
)
2249 char *statsdir
= NULL
, *filename
= NULL
, *str
= NULL
;
2251 if (!start_of_exit_stats_interval
)
2252 return 0; /* Not initialized. */
2253 if (start_of_exit_stats_interval
+ WRITE_STATS_INTERVAL
> now
)
2254 goto done
; /* Not ready to write. */
2256 log_info(LD_HIST
, "Writing exit port statistics to disk.");
2258 /* Generate history string. */
2259 str
= rep_hist_format_exit_stats(now
);
2261 /* Reset counters. */
2262 rep_hist_reset_exit_stats(now
);
2264 /* Try to write to disk. */
2265 statsdir
= get_datadir_fname("stats");
2266 if (check_private_dir(statsdir
, CPD_CREATE
) < 0) {
2267 log_warn(LD_HIST
, "Unable to create stats/ directory!");
2270 filename
= get_datadir_fname2("stats", "exit-stats");
2271 if (write_str_to_file(filename
, str
, 0) < 0)
2272 log_warn(LD_HIST
, "Unable to write exit port statistics to disk!");
2278 return start_of_exit_stats_interval
+ WRITE_STATS_INTERVAL
;
2281 /** Note that we wrote <b>num_written</b> bytes and read <b>num_read</b>
2282 * bytes to/from an exit connection to <b>port</b>. */
2284 rep_hist_note_exit_bytes(uint16_t port
, size_t num_written
,
2287 if (!start_of_exit_stats_interval
)
2288 return; /* Not initialized. */
2289 exit_bytes_written
[port
] += num_written
;
2290 exit_bytes_read
[port
] += num_read
;
2291 log_debug(LD_HIST
, "Written %lu bytes and read %lu bytes to/from an "
2292 "exit connection to port %d.",
2293 (unsigned long)num_written
, (unsigned long)num_read
, port
);
2296 /** Note that we opened an exit stream to <b>port</b>. */
2298 rep_hist_note_exit_stream_opened(uint16_t port
)
2300 if (!start_of_exit_stats_interval
)
2301 return; /* Not initialized. */
2302 exit_streams
[port
]++;
2303 log_debug(LD_HIST
, "Opened exit stream to port %d", port
);
2306 /*** cell statistics ***/
2308 /** Start of the current buffer stats interval or 0 if we're not
2309 * collecting buffer statistics. */
2310 static time_t start_of_buffer_stats_interval
;
2312 /** Initialize buffer stats. */
2314 rep_hist_buffer_stats_init(time_t now
)
2316 start_of_buffer_stats_interval
= now
;
2319 typedef struct circ_buffer_stats_t
{
2320 uint32_t processed_cells
;
2321 double mean_num_cells_in_queue
;
2322 double mean_time_cells_in_queue
;
2323 uint32_t local_circ_id
;
2324 } circ_buffer_stats_t
;
2327 smartlist_t
*circuits_for_buffer_stats
= NULL
;
2329 /** Remember cell statistics for circuit <b>circ</b> at time
2330 * <b>end_of_interval</b> and reset cell counters in case the circuit
2331 * remains open in the next measurement interval. */
2333 rep_hist_buffer_stats_add_circ(circuit_t
*circ
, time_t end_of_interval
)
2335 circ_buffer_stats_t
*stat
;
2336 time_t start_of_interval
;
2337 int interval_length
;
2338 or_circuit_t
*orcirc
;
2339 if (CIRCUIT_IS_ORIGIN(circ
))
2341 orcirc
= TO_OR_CIRCUIT(circ
);
2342 if (!orcirc
->processed_cells
)
2344 if (!circuits_for_buffer_stats
)
2345 circuits_for_buffer_stats
= smartlist_create();
2346 start_of_interval
= circ
->timestamp_created
>
2347 start_of_buffer_stats_interval
?
2348 circ
->timestamp_created
:
2349 start_of_buffer_stats_interval
;
2350 interval_length
= (int) (end_of_interval
- start_of_interval
);
2351 stat
= tor_malloc_zero(sizeof(circ_buffer_stats_t
));
2352 stat
->processed_cells
= orcirc
->processed_cells
;
2353 /* 1000.0 for s -> ms; 2.0 because of app-ward and exit-ward queues */
2354 stat
->mean_num_cells_in_queue
= interval_length
== 0 ? 0.0 :
2355 (double) orcirc
->total_cell_waiting_time
/
2356 (double) interval_length
/ 1000.0 / 2.0;
2357 stat
->mean_time_cells_in_queue
=
2358 (double) orcirc
->total_cell_waiting_time
/
2359 (double) orcirc
->processed_cells
;
2360 smartlist_add(circuits_for_buffer_stats
, stat
);
2361 orcirc
->total_cell_waiting_time
= 0;
2362 orcirc
->processed_cells
= 0;
2365 /** Sorting helper: return -1, 1, or 0 based on comparison of two
2366 * circ_buffer_stats_t */
2368 _buffer_stats_compare_entries(const void **_a
, const void **_b
)
2370 const circ_buffer_stats_t
*a
= *_a
, *b
= *_b
;
2371 if (a
->processed_cells
< b
->processed_cells
)
2373 else if (a
->processed_cells
> b
->processed_cells
)
2379 /** Stop collecting cell stats in a way that we can re-start doing so in
2380 * rep_hist_buffer_stats_init(). */
2382 rep_hist_buffer_stats_term(void)
2384 start_of_buffer_stats_interval
= 0;
2385 if (!circuits_for_buffer_stats
)
2386 circuits_for_buffer_stats
= smartlist_create();
2387 SMARTLIST_FOREACH(circuits_for_buffer_stats
, circ_buffer_stats_t
*,
2388 stat
, tor_free(stat
));
2389 smartlist_clear(circuits_for_buffer_stats
);
2392 /** Write buffer statistics to $DATADIR/stats/buffer-stats and return when
2393 * we would next want to write exit stats. */
2395 rep_hist_buffer_stats_write(time_t now
)
2397 char *statsdir
= NULL
, *filename
= NULL
;
2398 char written
[ISO_TIME_LEN
+1];
2399 open_file_t
*open_file
= NULL
;
2402 int processed_cells
[SHARES
], circs_in_share
[SHARES
],
2403 number_of_circuits
, i
;
2404 double queued_cells
[SHARES
], time_in_queue
[SHARES
];
2405 smartlist_t
*str_build
= smartlist_create();
2406 char *str
= NULL
, *buf
=NULL
;
2409 if (!start_of_buffer_stats_interval
)
2410 return 0; /* Not initialized. */
2411 if (start_of_buffer_stats_interval
+ WRITE_STATS_INTERVAL
> now
)
2412 goto done
; /* Not ready to write */
2414 /* add current circuits to stats */
2415 for (circ
= _circuit_get_global_list(); circ
; circ
= circ
->next
)
2416 rep_hist_buffer_stats_add_circ(circ
, now
);
2417 /* calculate deciles */
2418 memset(processed_cells
, 0, SHARES
* sizeof(int));
2419 memset(circs_in_share
, 0, SHARES
* sizeof(int));
2420 memset(queued_cells
, 0, SHARES
* sizeof(double));
2421 memset(time_in_queue
, 0, SHARES
* sizeof(double));
2422 if (!circuits_for_buffer_stats
)
2423 circuits_for_buffer_stats
= smartlist_create();
2424 smartlist_sort(circuits_for_buffer_stats
,
2425 _buffer_stats_compare_entries
);
2426 number_of_circuits
= smartlist_len(circuits_for_buffer_stats
);
2427 if (number_of_circuits
< 1) {
2428 log_info(LD_HIST
, "Attempt to write cell statistics to disk failed. "
2429 "We haven't seen a single circuit to report about.");
2433 SMARTLIST_FOREACH_BEGIN(circuits_for_buffer_stats
,
2434 circ_buffer_stats_t
*, stat
)
2436 int share
= i
++ * SHARES
/ number_of_circuits
;
2437 processed_cells
[share
] += stat
->processed_cells
;
2438 queued_cells
[share
] += stat
->mean_num_cells_in_queue
;
2439 time_in_queue
[share
] += stat
->mean_time_cells_in_queue
;
2440 circs_in_share
[share
]++;
2442 SMARTLIST_FOREACH_END(stat
);
2443 /* clear buffer stats history */
2444 SMARTLIST_FOREACH(circuits_for_buffer_stats
, circ_buffer_stats_t
*,
2445 stat
, tor_free(stat
));
2446 smartlist_clear(circuits_for_buffer_stats
);
2448 statsdir
= get_datadir_fname("stats");
2449 if (check_private_dir(statsdir
, CPD_CREATE
) < 0)
2451 filename
= get_datadir_fname2("stats", "buffer-stats");
2452 out
= start_writing_to_stdio_file(filename
, OPEN_FLAGS_APPEND
,
2456 format_iso_time(written
, now
);
2457 if (fprintf(out
, "cell-stats-end %s (%d s)\n", written
,
2458 (unsigned) (now
- start_of_buffer_stats_interval
)) < 0)
2460 for (i
= 0; i
< SHARES
; i
++) {
2461 tor_asprintf(&buf
,"%d", !circs_in_share
[i
] ? 0 :
2462 processed_cells
[i
] / circs_in_share
[i
]);
2463 smartlist_add(str_build
, buf
);
2465 str
= smartlist_join_strings(str_build
, ",", 0, NULL
);
2466 if (fprintf(out
, "cell-processed-cells %s\n", str
) < 0)
2469 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2470 smartlist_clear(str_build
);
2471 for (i
= 0; i
< SHARES
; i
++) {
2472 tor_asprintf(&buf
, "%.2f", circs_in_share
[i
] == 0 ? 0.0 :
2473 queued_cells
[i
] / (double) circs_in_share
[i
]);
2474 smartlist_add(str_build
, buf
);
2476 str
= smartlist_join_strings(str_build
, ",", 0, NULL
);
2477 if (fprintf(out
, "cell-queued-cells %s\n", str
) < 0)
2480 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2481 smartlist_clear(str_build
);
2482 for (i
= 0; i
< SHARES
; i
++) {
2483 tor_asprintf(&buf
, "%.0f", circs_in_share
[i
] == 0 ? 0.0 :
2484 time_in_queue
[i
] / (double) circs_in_share
[i
]);
2485 smartlist_add(str_build
, buf
);
2487 str
= smartlist_join_strings(str_build
, ",", 0, NULL
);
2488 if (fprintf(out
, "cell-time-in-queue %s\n", str
) < 0)
2491 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2492 smartlist_free(str_build
);
2494 if (fprintf(out
, "cell-circuits-per-decile %d\n",
2495 (number_of_circuits
+ SHARES
- 1) / SHARES
) < 0)
2497 finish_writing_to_file(open_file
);
2499 start_of_buffer_stats_interval
= now
;
2502 abort_writing_to_file(open_file
);
2506 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2507 smartlist_free(str_build
);
2511 return start_of_buffer_stats_interval
+ WRITE_STATS_INTERVAL
;
2514 /** Free all storage held by the OR/link history caches, by the
2515 * bandwidth history arrays, by the port history, or by statistics . */
2517 rep_hist_free_all(void)
2519 digestmap_free(history_map
, free_or_history
);
2520 tor_free(read_array
);
2521 tor_free(write_array
);
2522 tor_free(last_stability_doc
);
2523 tor_free(exit_bytes_read
);
2524 tor_free(exit_bytes_written
);
2525 tor_free(exit_streams
);
2526 built_last_stability_doc_at
= 0;
2527 predicted_ports_free();